Print this page
NEX-9200 Improve the scalability of attribute locking in zfs_zget
Reviewed by: Joyce McIntosh <joyce.mcintosh@nexenta.com>
Reviewed by: Sanjay Nadkarni <sanjay.nadkarni@nexenta.com>
6385 Fix unlocking order in zfs_zget
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Dan Kimmel <dan.kimmel@delphix.com>
Reviewed by: Andriy Gapon <avg@freebsd.org>
Approved by: Robert Mustacchi <rm@joyent.com>
5961 Fix stack overflow in zfs_create_fs
Reviewed by: George Wilson <george@delphix.com>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Approved by: Dan McDonald <danmcd@omniti.com>
4370 avoid transmitting holes during zfs send
4371 DMU code clean up
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: Josef 'Jeff' Sipek <jeffpc@josefsipek.net>
Approved by: Garrett D'Amore <garrett@damore.org>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/zfs_znode.c
+++ new/usr/src/uts/common/fs/zfs/zfs_znode.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 23 * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
24 24 * Copyright (c) 2014 Integros [integros.com]
25 25 */
26 26
27 27 /* Portions Copyright 2007 Jeremy Teo */
28 28
29 29 #ifdef _KERNEL
30 30 #include <sys/types.h>
31 31 #include <sys/param.h>
32 32 #include <sys/time.h>
33 33 #include <sys/systm.h>
34 34 #include <sys/sysmacros.h>
35 35 #include <sys/resource.h>
36 36 #include <sys/mntent.h>
37 37 #include <sys/mkdev.h>
38 38 #include <sys/u8_textprep.h>
39 39 #include <sys/dsl_dataset.h>
40 40 #include <sys/vfs.h>
41 41 #include <sys/vfs_opreg.h>
42 42 #include <sys/vnode.h>
43 43 #include <sys/file.h>
44 44 #include <sys/kmem.h>
45 45 #include <sys/errno.h>
46 46 #include <sys/unistd.h>
47 47 #include <sys/mode.h>
48 48 #include <sys/atomic.h>
49 49 #include <vm/pvn.h>
50 50 #include "fs/fs_subr.h"
51 51 #include <sys/zfs_dir.h>
52 52 #include <sys/zfs_acl.h>
53 53 #include <sys/zfs_ioctl.h>
54 54 #include <sys/zfs_rlock.h>
55 55 #include <sys/zfs_fuid.h>
56 56 #include <sys/dnode.h>
57 57 #include <sys/fs/zfs.h>
58 58 #include <sys/kidmap.h>
59 59 #endif /* _KERNEL */
60 60
61 61 #include <sys/dmu.h>
62 62 #include <sys/dmu_objset.h>
63 63 #include <sys/refcount.h>
64 64 #include <sys/stat.h>
65 65 #include <sys/zap.h>
66 66 #include <sys/zfs_znode.h>
67 67 #include <sys/sa.h>
68 68 #include <sys/zfs_sa.h>
69 69 #include <sys/zfs_stat.h>
70 70
71 71 #include "zfs_prop.h"
72 72 #include "zfs_comutil.h"
73 73
74 74 /*
75 75 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only
76 76 * turned on when DEBUG is also defined.
77 77 */
78 78 #ifdef DEBUG
79 79 #define ZNODE_STATS
80 80 #endif /* DEBUG */
81 81
82 82 #ifdef ZNODE_STATS
83 83 #define ZNODE_STAT_ADD(stat) ((stat)++)
84 84 #else
85 85 #define ZNODE_STAT_ADD(stat) /* nothing */
86 86 #endif /* ZNODE_STATS */
87 87
88 88 /*
89 89 * Functions needed for userland (ie: libzpool) are not put under
90 90 * #ifdef_KERNEL; the rest of the functions have dependencies
91 91 * (such as VFS logic) that will not compile easily in userland.
92 92 */
93 93 #ifdef _KERNEL
94 94 /*
95 95 * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to
96 96 * be freed before it can be safely accessed.
97 97 */
98 98 krwlock_t zfsvfs_lock;
99 99
100 100 static kmem_cache_t *znode_cache = NULL;
101 101
102 102 /*ARGSUSED*/
103 103 static void
104 104 znode_evict_error(dmu_buf_t *dbuf, void *user_ptr)
105 105 {
106 106 /*
107 107 * We should never drop all dbuf refs without first clearing
108 108 * the eviction callback.
109 109 */
110 110 panic("evicting znode %p\n", user_ptr);
111 111 }
112 112
113 113 /*ARGSUSED*/
114 114 static int
115 115 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
116 116 {
117 117 znode_t *zp = buf;
118 118
119 119 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
120 120
121 121 zp->z_vnode = vn_alloc(kmflags);
122 122 if (zp->z_vnode == NULL) {
123 123 return (-1);
124 124 }
125 125 ZTOV(zp)->v_data = zp;
126 126
127 127 list_link_init(&zp->z_link_node);
128 128
129 129 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
130 130 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL);
131 131 rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL);
132 132 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
133 133
134 134 mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL);
135 135 avl_create(&zp->z_range_avl, zfs_range_compare,
136 136 sizeof (rl_t), offsetof(rl_t, r_node));
137 137
138 138 zp->z_dirlocks = NULL;
139 139 zp->z_acl_cached = NULL;
140 140 zp->z_moved = 0;
141 141 return (0);
142 142 }
143 143
144 144 /*ARGSUSED*/
145 145 static void
146 146 zfs_znode_cache_destructor(void *buf, void *arg)
147 147 {
148 148 znode_t *zp = buf;
149 149
150 150 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
151 151 ASSERT(ZTOV(zp)->v_data == zp);
152 152 vn_free(ZTOV(zp));
153 153 ASSERT(!list_link_active(&zp->z_link_node));
154 154 mutex_destroy(&zp->z_lock);
155 155 rw_destroy(&zp->z_parent_lock);
156 156 rw_destroy(&zp->z_name_lock);
157 157 mutex_destroy(&zp->z_acl_lock);
158 158 avl_destroy(&zp->z_range_avl);
159 159 mutex_destroy(&zp->z_range_lock);
160 160
161 161 ASSERT(zp->z_dirlocks == NULL);
162 162 ASSERT(zp->z_acl_cached == NULL);
163 163 }
164 164
165 165 #ifdef ZNODE_STATS
166 166 static struct {
167 167 uint64_t zms_zfsvfs_invalid;
168 168 uint64_t zms_zfsvfs_recheck1;
169 169 uint64_t zms_zfsvfs_unmounted;
170 170 uint64_t zms_zfsvfs_recheck2;
171 171 uint64_t zms_obj_held;
172 172 uint64_t zms_vnode_locked;
173 173 uint64_t zms_not_only_dnlc;
174 174 } znode_move_stats;
175 175 #endif /* ZNODE_STATS */
176 176
177 177 static void
178 178 zfs_znode_move_impl(znode_t *ozp, znode_t *nzp)
179 179 {
180 180 vnode_t *vp;
181 181
182 182 /* Copy fields. */
183 183 nzp->z_zfsvfs = ozp->z_zfsvfs;
184 184
185 185 /* Swap vnodes. */
186 186 vp = nzp->z_vnode;
187 187 nzp->z_vnode = ozp->z_vnode;
188 188 ozp->z_vnode = vp; /* let destructor free the overwritten vnode */
189 189 ZTOV(ozp)->v_data = ozp;
190 190 ZTOV(nzp)->v_data = nzp;
191 191
192 192 nzp->z_id = ozp->z_id;
193 193 ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */
194 194 ASSERT(avl_numnodes(&ozp->z_range_avl) == 0);
195 195 nzp->z_unlinked = ozp->z_unlinked;
196 196 nzp->z_atime_dirty = ozp->z_atime_dirty;
197 197 nzp->z_zn_prefetch = ozp->z_zn_prefetch;
198 198 nzp->z_blksz = ozp->z_blksz;
199 199 nzp->z_seq = ozp->z_seq;
200 200 nzp->z_mapcnt = ozp->z_mapcnt;
201 201 nzp->z_gen = ozp->z_gen;
202 202 nzp->z_sync_cnt = ozp->z_sync_cnt;
203 203 nzp->z_is_sa = ozp->z_is_sa;
204 204 nzp->z_sa_hdl = ozp->z_sa_hdl;
205 205 bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2);
206 206 nzp->z_links = ozp->z_links;
207 207 nzp->z_size = ozp->z_size;
208 208 nzp->z_pflags = ozp->z_pflags;
209 209 nzp->z_uid = ozp->z_uid;
210 210 nzp->z_gid = ozp->z_gid;
211 211 nzp->z_mode = ozp->z_mode;
212 212
213 213 /*
214 214 * Since this is just an idle znode and kmem is already dealing with
215 215 * memory pressure, release any cached ACL.
216 216 */
217 217 if (ozp->z_acl_cached) {
218 218 zfs_acl_free(ozp->z_acl_cached);
219 219 ozp->z_acl_cached = NULL;
220 220 }
221 221
222 222 sa_set_userp(nzp->z_sa_hdl, nzp);
223 223
224 224 /*
225 225 * Invalidate the original znode by clearing fields that provide a
226 226 * pointer back to the znode. Set the low bit of the vfs pointer to
227 227 * ensure that zfs_znode_move() recognizes the znode as invalid in any
228 228 * subsequent callback.
229 229 */
230 230 ozp->z_sa_hdl = NULL;
231 231 POINTER_INVALIDATE(&ozp->z_zfsvfs);
232 232
233 233 /*
234 234 * Mark the znode.
235 235 */
236 236 nzp->z_moved = 1;
237 237 ozp->z_moved = (uint8_t)-1;
238 238 }
239 239
240 240 /*ARGSUSED*/
241 241 static kmem_cbrc_t
242 242 zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg)
243 243 {
244 244 znode_t *ozp = buf, *nzp = newbuf;
245 245 zfsvfs_t *zfsvfs;
246 246 vnode_t *vp;
247 247
248 248 /*
249 249 * The znode is on the file system's list of known znodes if the vfs
250 250 * pointer is valid. We set the low bit of the vfs pointer when freeing
251 251 * the znode to invalidate it, and the memory patterns written by kmem
252 252 * (baddcafe and deadbeef) set at least one of the two low bits. A newly
253 253 * created znode sets the vfs pointer last of all to indicate that the
254 254 * znode is known and in a valid state to be moved by this function.
255 255 */
256 256 zfsvfs = ozp->z_zfsvfs;
257 257 if (!POINTER_IS_VALID(zfsvfs)) {
258 258 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid);
259 259 return (KMEM_CBRC_DONT_KNOW);
260 260 }
261 261
262 262 /*
263 263 * Close a small window in which it's possible that the filesystem could
264 264 * be unmounted and freed, and zfsvfs, though valid in the previous
265 265 * statement, could point to unrelated memory by the time we try to
266 266 * prevent the filesystem from being unmounted.
267 267 */
268 268 rw_enter(&zfsvfs_lock, RW_WRITER);
269 269 if (zfsvfs != ozp->z_zfsvfs) {
270 270 rw_exit(&zfsvfs_lock);
271 271 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1);
272 272 return (KMEM_CBRC_DONT_KNOW);
273 273 }
274 274
275 275 /*
276 276 * If the znode is still valid, then so is the file system. We know that
277 277 * no valid file system can be freed while we hold zfsvfs_lock, so we
278 278 * can safely ensure that the filesystem is not and will not be
279 279 * unmounted. The next statement is equivalent to ZFS_ENTER().
280 280 */
281 281 rrm_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG);
282 282 if (zfsvfs->z_unmounted) {
283 283 ZFS_EXIT(zfsvfs);
284 284 rw_exit(&zfsvfs_lock);
285 285 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted);
286 286 return (KMEM_CBRC_DONT_KNOW);
287 287 }
288 288 rw_exit(&zfsvfs_lock);
289 289
290 290 mutex_enter(&zfsvfs->z_znodes_lock);
291 291 /*
292 292 * Recheck the vfs pointer in case the znode was removed just before
293 293 * acquiring the lock.
294 294 */
295 295 if (zfsvfs != ozp->z_zfsvfs) {
296 296 mutex_exit(&zfsvfs->z_znodes_lock);
297 297 ZFS_EXIT(zfsvfs);
298 298 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2);
299 299 return (KMEM_CBRC_DONT_KNOW);
300 300 }
301 301
302 302 /*
303 303 * At this point we know that as long as we hold z_znodes_lock, the
304 304 * znode cannot be freed and fields within the znode can be safely
305 305 * accessed. Now, prevent a race with zfs_zget().
306 306 */
307 307 if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) {
308 308 mutex_exit(&zfsvfs->z_znodes_lock);
309 309 ZFS_EXIT(zfsvfs);
310 310 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held);
311 311 return (KMEM_CBRC_LATER);
312 312 }
313 313
314 314 vp = ZTOV(ozp);
315 315 if (mutex_tryenter(&vp->v_lock) == 0) {
316 316 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
317 317 mutex_exit(&zfsvfs->z_znodes_lock);
318 318 ZFS_EXIT(zfsvfs);
319 319 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked);
320 320 return (KMEM_CBRC_LATER);
321 321 }
322 322
323 323 /* Only move znodes that are referenced _only_ by the DNLC. */
324 324 if (vp->v_count != 1 || !vn_in_dnlc(vp)) {
325 325 mutex_exit(&vp->v_lock);
326 326 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
327 327 mutex_exit(&zfsvfs->z_znodes_lock);
328 328 ZFS_EXIT(zfsvfs);
329 329 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc);
330 330 return (KMEM_CBRC_LATER);
331 331 }
332 332
333 333 /*
334 334 * The znode is known and in a valid state to move. We're holding the
335 335 * locks needed to execute the critical section.
336 336 */
337 337 zfs_znode_move_impl(ozp, nzp);
338 338 mutex_exit(&vp->v_lock);
339 339 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
340 340
341 341 list_link_replace(&ozp->z_link_node, &nzp->z_link_node);
342 342 mutex_exit(&zfsvfs->z_znodes_lock);
343 343 ZFS_EXIT(zfsvfs);
344 344
345 345 return (KMEM_CBRC_YES);
346 346 }
347 347
348 348 void
349 349 zfs_znode_init(void)
350 350 {
351 351 /*
352 352 * Initialize zcache
353 353 */
354 354 rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL);
355 355 ASSERT(znode_cache == NULL);
356 356 znode_cache = kmem_cache_create("zfs_znode_cache",
357 357 sizeof (znode_t), 0, zfs_znode_cache_constructor,
358 358 zfs_znode_cache_destructor, NULL, NULL, NULL, 0);
359 359 kmem_cache_set_move(znode_cache, zfs_znode_move);
360 360 }
361 361
362 362 void
363 363 zfs_znode_fini(void)
364 364 {
365 365 /*
366 366 * Cleanup vfs & vnode ops
367 367 */
368 368 zfs_remove_op_tables();
369 369
370 370 /*
371 371 * Cleanup zcache
372 372 */
373 373 if (znode_cache)
374 374 kmem_cache_destroy(znode_cache);
375 375 znode_cache = NULL;
376 376 rw_destroy(&zfsvfs_lock);
377 377 }
378 378
379 379 struct vnodeops *zfs_dvnodeops;
380 380 struct vnodeops *zfs_fvnodeops;
381 381 struct vnodeops *zfs_symvnodeops;
382 382 struct vnodeops *zfs_xdvnodeops;
383 383 struct vnodeops *zfs_evnodeops;
384 384 struct vnodeops *zfs_sharevnodeops;
385 385
386 386 void
387 387 zfs_remove_op_tables()
388 388 {
389 389 /*
390 390 * Remove vfs ops
391 391 */
392 392 ASSERT(zfsfstype);
393 393 (void) vfs_freevfsops_by_type(zfsfstype);
394 394 zfsfstype = 0;
395 395
396 396 /*
397 397 * Remove vnode ops
398 398 */
399 399 if (zfs_dvnodeops)
400 400 vn_freevnodeops(zfs_dvnodeops);
401 401 if (zfs_fvnodeops)
402 402 vn_freevnodeops(zfs_fvnodeops);
403 403 if (zfs_symvnodeops)
404 404 vn_freevnodeops(zfs_symvnodeops);
405 405 if (zfs_xdvnodeops)
406 406 vn_freevnodeops(zfs_xdvnodeops);
407 407 if (zfs_evnodeops)
408 408 vn_freevnodeops(zfs_evnodeops);
409 409 if (zfs_sharevnodeops)
410 410 vn_freevnodeops(zfs_sharevnodeops);
411 411
412 412 zfs_dvnodeops = NULL;
413 413 zfs_fvnodeops = NULL;
414 414 zfs_symvnodeops = NULL;
415 415 zfs_xdvnodeops = NULL;
416 416 zfs_evnodeops = NULL;
417 417 zfs_sharevnodeops = NULL;
418 418 }
419 419
420 420 extern const fs_operation_def_t zfs_dvnodeops_template[];
421 421 extern const fs_operation_def_t zfs_fvnodeops_template[];
422 422 extern const fs_operation_def_t zfs_xdvnodeops_template[];
423 423 extern const fs_operation_def_t zfs_symvnodeops_template[];
424 424 extern const fs_operation_def_t zfs_evnodeops_template[];
425 425 extern const fs_operation_def_t zfs_sharevnodeops_template[];
426 426
427 427 int
428 428 zfs_create_op_tables()
429 429 {
430 430 int error;
431 431
432 432 /*
433 433 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs()
434 434 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv).
435 435 * In this case we just return as the ops vectors are already set up.
436 436 */
437 437 if (zfs_dvnodeops)
438 438 return (0);
439 439
440 440 error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template,
441 441 &zfs_dvnodeops);
442 442 if (error)
443 443 return (error);
444 444
445 445 error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template,
446 446 &zfs_fvnodeops);
447 447 if (error)
448 448 return (error);
449 449
450 450 error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template,
451 451 &zfs_symvnodeops);
452 452 if (error)
453 453 return (error);
454 454
455 455 error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template,
456 456 &zfs_xdvnodeops);
457 457 if (error)
458 458 return (error);
459 459
460 460 error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template,
461 461 &zfs_evnodeops);
462 462 if (error)
463 463 return (error);
464 464
465 465 error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template,
466 466 &zfs_sharevnodeops);
467 467
468 468 return (error);
469 469 }
470 470
471 471 int
472 472 zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
473 473 {
474 474 zfs_acl_ids_t acl_ids;
475 475 vattr_t vattr;
476 476 znode_t *sharezp;
477 477 vnode_t *vp;
478 478 znode_t *zp;
479 479 int error;
480 480
481 481 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
482 482 vattr.va_type = VDIR;
483 483 vattr.va_mode = S_IFDIR|0555;
484 484 vattr.va_uid = crgetuid(kcred);
485 485 vattr.va_gid = crgetgid(kcred);
486 486
487 487 sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP);
488 488 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs));
489 489 sharezp->z_moved = 0;
490 490 sharezp->z_unlinked = 0;
491 491 sharezp->z_atime_dirty = 0;
492 492 sharezp->z_zfsvfs = zfsvfs;
493 493 sharezp->z_is_sa = zfsvfs->z_use_sa;
494 494
495 495 vp = ZTOV(sharezp);
496 496 vn_reinit(vp);
497 497 vp->v_type = VDIR;
498 498
499 499 VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr,
500 500 kcred, NULL, &acl_ids));
501 501 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids);
502 502 ASSERT3P(zp, ==, sharezp);
503 503 ASSERT(!vn_in_dnlc(ZTOV(sharezp))); /* not valid to move */
504 504 POINTER_INVALIDATE(&sharezp->z_zfsvfs);
505 505 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
506 506 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx);
507 507 zfsvfs->z_shares_dir = sharezp->z_id;
508 508
509 509 zfs_acl_ids_free(&acl_ids);
510 510 ZTOV(sharezp)->v_count = 0;
511 511 sa_handle_destroy(sharezp->z_sa_hdl);
512 512 kmem_cache_free(znode_cache, sharezp);
513 513
514 514 return (error);
515 515 }
516 516
517 517 /*
518 518 * define a couple of values we need available
519 519 * for both 64 and 32 bit environments.
520 520 */
521 521 #ifndef NBITSMINOR64
522 522 #define NBITSMINOR64 32
523 523 #endif
524 524 #ifndef MAXMAJ64
525 525 #define MAXMAJ64 0xffffffffUL
526 526 #endif
527 527 #ifndef MAXMIN64
528 528 #define MAXMIN64 0xffffffffUL
529 529 #endif
530 530
531 531 /*
532 532 * Create special expldev for ZFS private use.
533 533 * Can't use standard expldev since it doesn't do
534 534 * what we want. The standard expldev() takes a
535 535 * dev32_t in LP64 and expands it to a long dev_t.
536 536 * We need an interface that takes a dev32_t in ILP32
537 537 * and expands it to a long dev_t.
538 538 */
539 539 static uint64_t
540 540 zfs_expldev(dev_t dev)
541 541 {
542 542 #ifndef _LP64
543 543 major_t major = (major_t)dev >> NBITSMINOR32 & MAXMAJ32;
544 544 return (((uint64_t)major << NBITSMINOR64) |
545 545 ((minor_t)dev & MAXMIN32));
546 546 #else
547 547 return (dev);
548 548 #endif
549 549 }
550 550
551 551 /*
552 552 * Special cmpldev for ZFS private use.
553 553 * Can't use standard cmpldev since it takes
554 554 * a long dev_t and compresses it to dev32_t in
555 555 * LP64. We need to do a compaction of a long dev_t
556 556 * to a dev32_t in ILP32.
557 557 */
558 558 dev_t
559 559 zfs_cmpldev(uint64_t dev)
560 560 {
561 561 #ifndef _LP64
562 562 minor_t minor = (minor_t)dev & MAXMIN64;
563 563 major_t major = (major_t)(dev >> NBITSMINOR64) & MAXMAJ64;
564 564
565 565 if (major > MAXMAJ32 || minor > MAXMIN32)
566 566 return (NODEV32);
567 567
568 568 return (((dev32_t)major << NBITSMINOR32) | minor);
569 569 #else
570 570 return (dev);
571 571 #endif
572 572 }
573 573
574 574 static void
575 575 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
576 576 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
577 577 {
578 578 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs));
579 579 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)));
580 580
581 581 mutex_enter(&zp->z_lock);
582 582
583 583 ASSERT(zp->z_sa_hdl == NULL);
584 584 ASSERT(zp->z_acl_cached == NULL);
585 585 if (sa_hdl == NULL) {
586 586 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
587 587 SA_HDL_SHARED, &zp->z_sa_hdl));
588 588 } else {
589 589 zp->z_sa_hdl = sa_hdl;
590 590 sa_set_userp(sa_hdl, zp);
591 591 }
592 592
593 593 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
594 594
595 595 /*
596 596 * Slap on VROOT if we are the root znode
597 597 */
598 598 if (zp->z_id == zfsvfs->z_root)
599 599 ZTOV(zp)->v_flag |= VROOT;
600 600
601 601 mutex_exit(&zp->z_lock);
602 602 vn_exists(ZTOV(zp));
603 603 }
604 604
605 605 void
606 606 zfs_znode_dmu_fini(znode_t *zp)
607 607 {
608 608 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) ||
609 609 zp->z_unlinked ||
610 610 RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock));
611 611
612 612 sa_handle_destroy(zp->z_sa_hdl);
613 613 zp->z_sa_hdl = NULL;
614 614 }
615 615
616 616 /*
617 617 * Construct a new znode/vnode and intialize.
618 618 *
619 619 * This does not do a call to dmu_set_user() that is
620 620 * up to the caller to do, in case you don't want to
621 621 * return the znode
622 622 */
623 623 static znode_t *
624 624 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
625 625 dmu_object_type_t obj_type, sa_handle_t *hdl)
626 626 {
627 627 znode_t *zp;
628 628 vnode_t *vp;
629 629 uint64_t mode;
630 630 uint64_t parent;
631 631 sa_bulk_attr_t bulk[9];
632 632 int count = 0;
633 633
634 634 zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
635 635
636 636 ASSERT(zp->z_dirlocks == NULL);
637 637 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
638 638 zp->z_moved = 0;
639 639
640 640 /*
641 641 * Defer setting z_zfsvfs until the znode is ready to be a candidate for
642 642 * the zfs_znode_move() callback.
643 643 */
644 644 zp->z_sa_hdl = NULL;
645 645 zp->z_unlinked = 0;
646 646 zp->z_atime_dirty = 0;
647 647 zp->z_mapcnt = 0;
648 648 zp->z_id = db->db_object;
649 649 zp->z_blksz = blksz;
650 650 zp->z_seq = 0x7A4653;
651 651 zp->z_sync_cnt = 0;
652 652
653 653 vp = ZTOV(zp);
654 654 vn_reinit(vp);
655 655
656 656 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
657 657
658 658 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
659 659 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8);
660 660 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
661 661 &zp->z_size, 8);
662 662 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
663 663 &zp->z_links, 8);
664 664 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
665 665 &zp->z_pflags, 8);
666 666 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
667 667 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
668 668 &zp->z_atime, 16);
669 669 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
670 670 &zp->z_uid, 8);
671 671 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
672 672 &zp->z_gid, 8);
673 673
674 674 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) {
675 675 if (hdl == NULL)
676 676 sa_handle_destroy(zp->z_sa_hdl);
677 677 kmem_cache_free(znode_cache, zp);
678 678 return (NULL);
679 679 }
680 680
681 681 zp->z_mode = mode;
682 682 vp->v_vfsp = zfsvfs->z_parent->z_vfs;
683 683
684 684 vp->v_type = IFTOVT((mode_t)mode);
685 685
686 686 switch (vp->v_type) {
687 687 case VDIR:
688 688 if (zp->z_pflags & ZFS_XATTR) {
689 689 vn_setops(vp, zfs_xdvnodeops);
690 690 vp->v_flag |= V_XATTRDIR;
691 691 } else {
692 692 vn_setops(vp, zfs_dvnodeops);
693 693 }
694 694 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */
695 695 break;
696 696 case VBLK:
697 697 case VCHR:
698 698 {
699 699 uint64_t rdev;
700 700 VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs),
701 701 &rdev, sizeof (rdev)) == 0);
702 702
703 703 vp->v_rdev = zfs_cmpldev(rdev);
704 704 }
705 705 /*FALLTHROUGH*/
706 706 case VFIFO:
707 707 case VSOCK:
708 708 case VDOOR:
709 709 vn_setops(vp, zfs_fvnodeops);
710 710 break;
711 711 case VREG:
712 712 vp->v_flag |= VMODSORT;
713 713 if (parent == zfsvfs->z_shares_dir) {
714 714 ASSERT(zp->z_uid == 0 && zp->z_gid == 0);
715 715 vn_setops(vp, zfs_sharevnodeops);
716 716 } else {
717 717 vn_setops(vp, zfs_fvnodeops);
718 718 }
719 719 break;
720 720 case VLNK:
721 721 vn_setops(vp, zfs_symvnodeops);
722 722 break;
723 723 default:
724 724 vn_setops(vp, zfs_evnodeops);
725 725 break;
726 726 }
727 727
728 728 mutex_enter(&zfsvfs->z_znodes_lock);
729 729 list_insert_tail(&zfsvfs->z_all_znodes, zp);
730 730 membar_producer();
731 731 /*
732 732 * Everything else must be valid before assigning z_zfsvfs makes the
733 733 * znode eligible for zfs_znode_move().
734 734 */
735 735 zp->z_zfsvfs = zfsvfs;
736 736 mutex_exit(&zfsvfs->z_znodes_lock);
737 737
738 738 VFS_HOLD(zfsvfs->z_vfs);
739 739 return (zp);
740 740 }
741 741
742 742 static uint64_t empty_xattr;
743 743 static uint64_t pad[4];
744 744 static zfs_acl_phys_t acl_phys;
745 745 /*
746 746 * Create a new DMU object to hold a zfs znode.
747 747 *
748 748 * IN: dzp - parent directory for new znode
749 749 * vap - file attributes for new znode
750 750 * tx - dmu transaction id for zap operations
751 751 * cr - credentials of caller
752 752 * flag - flags:
753 753 * IS_ROOT_NODE - new object will be root
754 754 * IS_XATTR - new object is an attribute
755 755 * bonuslen - length of bonus buffer
756 756 * setaclp - File/Dir initial ACL
757 757 * fuidp - Tracks fuid allocation.
758 758 *
759 759 * OUT: zpp - allocated znode
760 760 *
761 761 */
762 762 void
763 763 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
764 764 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
765 765 {
766 766 uint64_t crtime[2], atime[2], mtime[2], ctime[2];
767 767 uint64_t mode, size, links, parent, pflags;
768 768 uint64_t dzp_pflags = 0;
769 769 uint64_t rdev = 0;
770 770 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
771 771 dmu_buf_t *db;
772 772 timestruc_t now;
773 773 uint64_t gen, obj;
774 774 int bonuslen;
775 775 sa_handle_t *sa_hdl;
776 776 dmu_object_type_t obj_type;
777 777 sa_bulk_attr_t sa_attrs[ZPL_END];
778 778 int cnt = 0;
779 779 zfs_acl_locator_cb_t locate = { 0 };
780 780
781 781 ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
782 782
783 783 if (zfsvfs->z_replay) {
784 784 obj = vap->va_nodeid;
785 785 now = vap->va_ctime; /* see zfs_replay_create() */
786 786 gen = vap->va_nblocks; /* ditto */
787 787 } else {
788 788 obj = 0;
789 789 gethrestime(&now);
790 790 gen = dmu_tx_get_txg(tx);
791 791 }
792 792
793 793 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
794 794 bonuslen = (obj_type == DMU_OT_SA) ?
795 795 DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE;
796 796
797 797 /*
798 798 * Create a new DMU object.
799 799 */
800 800 /*
801 801 * There's currently no mechanism for pre-reading the blocks that will
802 802 * be needed to allocate a new object, so we accept the small chance
803 803 * that there will be an i/o error and we will fail one of the
804 804 * assertions below.
805 805 */
806 806 if (vap->va_type == VDIR) {
807 807 if (zfsvfs->z_replay) {
808 808 VERIFY0(zap_create_claim_norm(zfsvfs->z_os, obj,
809 809 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
810 810 obj_type, bonuslen, tx));
811 811 } else {
812 812 obj = zap_create_norm(zfsvfs->z_os,
813 813 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
814 814 obj_type, bonuslen, tx);
815 815 }
816 816 } else {
817 817 if (zfsvfs->z_replay) {
818 818 VERIFY0(dmu_object_claim(zfsvfs->z_os, obj,
819 819 DMU_OT_PLAIN_FILE_CONTENTS, 0,
820 820 obj_type, bonuslen, tx));
821 821 } else {
822 822 obj = dmu_object_alloc(zfsvfs->z_os,
823 823 DMU_OT_PLAIN_FILE_CONTENTS, 0,
824 824 obj_type, bonuslen, tx);
825 825 }
826 826 }
827 827
828 828 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
829 829 VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
830 830
831 831 /*
832 832 * If this is the root, fix up the half-initialized parent pointer
833 833 * to reference the just-allocated physical data area.
834 834 */
835 835 if (flag & IS_ROOT_NODE) {
836 836 dzp->z_id = obj;
837 837 } else {
838 838 dzp_pflags = dzp->z_pflags;
839 839 }
840 840
841 841 /*
842 842 * If parent is an xattr, so am I.
843 843 */
844 844 if (dzp_pflags & ZFS_XATTR) {
845 845 flag |= IS_XATTR;
846 846 }
847 847
848 848 if (zfsvfs->z_use_fuids)
849 849 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
850 850 else
851 851 pflags = 0;
852 852
853 853 if (vap->va_type == VDIR) {
854 854 size = 2; /* contents ("." and "..") */
855 855 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1;
856 856 } else {
857 857 size = links = 0;
858 858 }
859 859
860 860 if (vap->va_type == VBLK || vap->va_type == VCHR) {
861 861 rdev = zfs_expldev(vap->va_rdev);
862 862 }
863 863
864 864 parent = dzp->z_id;
865 865 mode = acl_ids->z_mode;
866 866 if (flag & IS_XATTR)
867 867 pflags |= ZFS_XATTR;
868 868
869 869 /*
870 870 * No execs denied will be deterimed when zfs_mode_compute() is called.
871 871 */
872 872 pflags |= acl_ids->z_aclp->z_hints &
873 873 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
874 874 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
875 875
876 876 ZFS_TIME_ENCODE(&now, crtime);
877 877 ZFS_TIME_ENCODE(&now, ctime);
878 878
879 879 if (vap->va_mask & AT_ATIME) {
880 880 ZFS_TIME_ENCODE(&vap->va_atime, atime);
881 881 } else {
882 882 ZFS_TIME_ENCODE(&now, atime);
883 883 }
884 884
885 885 if (vap->va_mask & AT_MTIME) {
886 886 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
887 887 } else {
888 888 ZFS_TIME_ENCODE(&now, mtime);
889 889 }
890 890
891 891 /* Now add in all of the "SA" attributes */
892 892 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
893 893 &sa_hdl));
894 894
895 895 /*
896 896 * Setup the array of attributes to be replaced/set on the new file
897 897 *
898 898 * order for DMU_OT_ZNODE is critical since it needs to be constructed
899 899 * in the old znode_phys_t format. Don't change this ordering
900 900 */
901 901
902 902 if (obj_type == DMU_OT_ZNODE) {
903 903 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
904 904 NULL, &atime, 16);
905 905 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
906 906 NULL, &mtime, 16);
907 907 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
908 908 NULL, &ctime, 16);
909 909 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
910 910 NULL, &crtime, 16);
911 911 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
912 912 NULL, &gen, 8);
913 913 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
914 914 NULL, &mode, 8);
915 915 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
916 916 NULL, &size, 8);
917 917 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
918 918 NULL, &parent, 8);
919 919 } else {
920 920 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
921 921 NULL, &mode, 8);
922 922 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
923 923 NULL, &size, 8);
924 924 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
925 925 NULL, &gen, 8);
926 926 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
927 927 &acl_ids->z_fuid, 8);
928 928 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
929 929 &acl_ids->z_fgid, 8);
930 930 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
931 931 NULL, &parent, 8);
932 932 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
933 933 NULL, &pflags, 8);
934 934 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
935 935 NULL, &atime, 16);
936 936 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
937 937 NULL, &mtime, 16);
938 938 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
939 939 NULL, &ctime, 16);
940 940 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
941 941 NULL, &crtime, 16);
942 942 }
943 943
944 944 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
945 945
946 946 if (obj_type == DMU_OT_ZNODE) {
947 947 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
948 948 &empty_xattr, 8);
949 949 }
950 950 if (obj_type == DMU_OT_ZNODE ||
951 951 (vap->va_type == VBLK || vap->va_type == VCHR)) {
952 952 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
953 953 NULL, &rdev, 8);
954 954
955 955 }
956 956 if (obj_type == DMU_OT_ZNODE) {
957 957 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
958 958 NULL, &pflags, 8);
959 959 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
960 960 &acl_ids->z_fuid, 8);
961 961 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
962 962 &acl_ids->z_fgid, 8);
963 963 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
964 964 sizeof (uint64_t) * 4);
965 965 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
966 966 &acl_phys, sizeof (zfs_acl_phys_t));
967 967 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
968 968 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
969 969 &acl_ids->z_aclp->z_acl_count, 8);
970 970 locate.cb_aclp = acl_ids->z_aclp;
971 971 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
972 972 zfs_acl_data_locator, &locate,
973 973 acl_ids->z_aclp->z_acl_bytes);
974 974 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
975 975 acl_ids->z_fuid, acl_ids->z_fgid);
976 976 }
977 977
978 978 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
979 979
980 980 if (!(flag & IS_ROOT_NODE)) {
981 981 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
982 982 ASSERT(*zpp != NULL);
983 983 } else {
984 984 /*
985 985 * If we are creating the root node, the "parent" we
986 986 * passed in is the znode for the root.
987 987 */
988 988 *zpp = dzp;
989 989
990 990 (*zpp)->z_sa_hdl = sa_hdl;
991 991 }
992 992
993 993 (*zpp)->z_pflags = pflags;
994 994 (*zpp)->z_mode = mode;
995 995
996 996 if (vap->va_mask & AT_XVATTR)
997 997 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx);
998 998
999 999 if (obj_type == DMU_OT_ZNODE ||
1000 1000 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
1001 1001 VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx));
1002 1002 }
1003 1003 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1004 1004 }
1005 1005
1006 1006 /*
1007 1007 * Update in-core attributes. It is assumed the caller will be doing an
1008 1008 * sa_bulk_update to push the changes out.
1009 1009 */
1010 1010 void
1011 1011 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
1012 1012 {
1013 1013 xoptattr_t *xoap;
1014 1014
1015 1015 xoap = xva_getxoptattr(xvap);
1016 1016 ASSERT(xoap);
1017 1017
1018 1018 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
1019 1019 uint64_t times[2];
1020 1020 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
1021 1021 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs),
1022 1022 ×, sizeof (times), tx);
1023 1023 XVA_SET_RTN(xvap, XAT_CREATETIME);
1024 1024 }
1025 1025 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
1026 1026 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
1027 1027 zp->z_pflags, tx);
1028 1028 XVA_SET_RTN(xvap, XAT_READONLY);
1029 1029 }
1030 1030 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
1031 1031 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
1032 1032 zp->z_pflags, tx);
1033 1033 XVA_SET_RTN(xvap, XAT_HIDDEN);
1034 1034 }
1035 1035 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
1036 1036 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
1037 1037 zp->z_pflags, tx);
1038 1038 XVA_SET_RTN(xvap, XAT_SYSTEM);
1039 1039 }
1040 1040 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
1041 1041 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
1042 1042 zp->z_pflags, tx);
1043 1043 XVA_SET_RTN(xvap, XAT_ARCHIVE);
1044 1044 }
1045 1045 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
1046 1046 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
1047 1047 zp->z_pflags, tx);
1048 1048 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
1049 1049 }
1050 1050 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
1051 1051 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
1052 1052 zp->z_pflags, tx);
1053 1053 XVA_SET_RTN(xvap, XAT_NOUNLINK);
1054 1054 }
1055 1055 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
1056 1056 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
1057 1057 zp->z_pflags, tx);
1058 1058 XVA_SET_RTN(xvap, XAT_APPENDONLY);
1059 1059 }
1060 1060 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
1061 1061 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
1062 1062 zp->z_pflags, tx);
1063 1063 XVA_SET_RTN(xvap, XAT_NODUMP);
1064 1064 }
1065 1065 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
1066 1066 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
1067 1067 zp->z_pflags, tx);
1068 1068 XVA_SET_RTN(xvap, XAT_OPAQUE);
1069 1069 }
1070 1070 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
1071 1071 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
1072 1072 xoap->xoa_av_quarantined, zp->z_pflags, tx);
1073 1073 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
1074 1074 }
1075 1075 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
1076 1076 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
1077 1077 zp->z_pflags, tx);
1078 1078 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
1079 1079 }
1080 1080 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
1081 1081 zfs_sa_set_scanstamp(zp, xvap, tx);
1082 1082 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
1083 1083 }
1084 1084 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
1085 1085 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
1086 1086 zp->z_pflags, tx);
1087 1087 XVA_SET_RTN(xvap, XAT_REPARSE);
1088 1088 }
1089 1089 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
1090 1090 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
1091 1091 zp->z_pflags, tx);
1092 1092 XVA_SET_RTN(xvap, XAT_OFFLINE);
1093 1093 }
1094 1094 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
1095 1095 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
1096 1096 zp->z_pflags, tx);
1097 1097 XVA_SET_RTN(xvap, XAT_SPARSE);
1098 1098 }
1099 1099 }
1100 1100
1101 1101 int
1102 1102 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
1103 1103 {
1104 1104 dmu_object_info_t doi;
1105 1105 dmu_buf_t *db;
1106 1106 znode_t *zp;
1107 1107 int err;
1108 1108 sa_handle_t *hdl;
1109 1109
1110 1110 *zpp = NULL;
1111 1111
1112 1112 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1113 1113
1114 1114 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1115 1115 if (err) {
1116 1116 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1117 1117 return (err);
1118 1118 }
1119 1119
1120 1120 dmu_object_info_from_db(db, &doi);
1121 1121 if (doi.doi_bonus_type != DMU_OT_SA &&
1122 1122 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1123 1123 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1124 1124 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1125 1125 sa_buf_rele(db, NULL);
1126 1126 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1127 1127 return (SET_ERROR(EINVAL));
1128 1128 }
1129 1129
1130 1130 hdl = dmu_buf_get_user(db);
1131 1131 if (hdl != NULL) {
1132 1132 zp = sa_get_userdata(hdl);
1133 1133
1134 1134
1135 1135 /*
1136 1136 * Since "SA" does immediate eviction we
1137 1137 * should never find a sa handle that doesn't
1138 1138 * know about the znode.
1139 1139 */
1140 1140
1141 1141 ASSERT3P(zp, !=, NULL);
1142 1142
1143 1143 mutex_enter(&zp->z_lock);
1144 1144 ASSERT3U(zp->z_id, ==, obj_num);
1145 1145 if (zp->z_unlinked) {
1146 1146 err = SET_ERROR(ENOENT);
1147 1147 } else {
1148 1148 VN_HOLD(ZTOV(zp));
1149 1149 *zpp = zp;
1150 1150 err = 0;
1151 1151 }
1152 1152 mutex_exit(&zp->z_lock);
1153 1153 sa_buf_rele(db, NULL);
1154 1154 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1155 1155 return (err);
1156 1156 }
1157 1157
1158 1158 /*
1159 1159 * Not found create new znode/vnode
1160 1160 * but only if file exists.
1161 1161 *
1162 1162 * There is a small window where zfs_vget() could
1163 1163 * find this object while a file create is still in
1164 1164 * progress. This is checked for in zfs_znode_alloc()
1165 1165 *
1166 1166 * if zfs_znode_alloc() fails it will drop the hold on the
1167 1167 * bonus buffer.
1168 1168 */
1169 1169 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1170 1170 doi.doi_bonus_type, NULL);
1171 1171 if (zp == NULL) {
1172 1172 err = SET_ERROR(ENOENT);
1173 1173 } else {
1174 1174 *zpp = zp;
1175 1175 }
1176 1176 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1177 1177 return (err);
1178 1178 }
1179 1179
1180 1180 int
1181 1181 zfs_rezget(znode_t *zp)
1182 1182 {
1183 1183 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1184 1184 dmu_object_info_t doi;
1185 1185 dmu_buf_t *db;
1186 1186 uint64_t obj_num = zp->z_id;
1187 1187 uint64_t mode;
1188 1188 sa_bulk_attr_t bulk[8];
1189 1189 int err;
1190 1190 int count = 0;
1191 1191 uint64_t gen;
1192 1192
1193 1193 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1194 1194
1195 1195 mutex_enter(&zp->z_acl_lock);
1196 1196 if (zp->z_acl_cached) {
1197 1197 zfs_acl_free(zp->z_acl_cached);
1198 1198 zp->z_acl_cached = NULL;
1199 1199 }
1200 1200
1201 1201 mutex_exit(&zp->z_acl_lock);
1202 1202 ASSERT(zp->z_sa_hdl == NULL);
1203 1203 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1204 1204 if (err) {
1205 1205 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1206 1206 return (err);
1207 1207 }
1208 1208
1209 1209 dmu_object_info_from_db(db, &doi);
1210 1210 if (doi.doi_bonus_type != DMU_OT_SA &&
1211 1211 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1212 1212 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1213 1213 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1214 1214 sa_buf_rele(db, NULL);
1215 1215 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1216 1216 return (SET_ERROR(EINVAL));
1217 1217 }
1218 1218
1219 1219 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1220 1220
1221 1221 /* reload cached values */
1222 1222 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1223 1223 &gen, sizeof (gen));
1224 1224 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1225 1225 &zp->z_size, sizeof (zp->z_size));
1226 1226 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1227 1227 &zp->z_links, sizeof (zp->z_links));
1228 1228 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1229 1229 &zp->z_pflags, sizeof (zp->z_pflags));
1230 1230 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1231 1231 &zp->z_atime, sizeof (zp->z_atime));
1232 1232 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1233 1233 &zp->z_uid, sizeof (zp->z_uid));
1234 1234 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1235 1235 &zp->z_gid, sizeof (zp->z_gid));
1236 1236 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1237 1237 &mode, sizeof (mode));
1238 1238
1239 1239 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1240 1240 zfs_znode_dmu_fini(zp);
1241 1241 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1242 1242 return (SET_ERROR(EIO));
1243 1243 }
1244 1244
1245 1245 zp->z_mode = mode;
1246 1246
1247 1247 if (gen != zp->z_gen) {
1248 1248 zfs_znode_dmu_fini(zp);
1249 1249 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1250 1250 return (SET_ERROR(EIO));
1251 1251 }
1252 1252
1253 1253 zp->z_blksz = doi.doi_data_block_size;
1254 1254
1255 1255 /*
1256 1256 * If the file has zero links, then it has been unlinked on the send
1257 1257 * side and it must be in the received unlinked set.
1258 1258 * We call zfs_znode_dmu_fini() now to prevent any accesses to the
1259 1259 * stale data and to prevent automatical removal of the file in
1260 1260 * zfs_zinactive(). The file will be removed either when it is removed
1261 1261 * on the send side and the next incremental stream is received or
1262 1262 * when the unlinked set gets processed.
1263 1263 */
1264 1264 zp->z_unlinked = (zp->z_links == 0);
1265 1265 if (zp->z_unlinked)
1266 1266 zfs_znode_dmu_fini(zp);
1267 1267
1268 1268 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1269 1269
1270 1270 return (0);
1271 1271 }
1272 1272
1273 1273 void
1274 1274 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1275 1275 {
1276 1276 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1277 1277 objset_t *os = zfsvfs->z_os;
1278 1278 uint64_t obj = zp->z_id;
1279 1279 uint64_t acl_obj = zfs_external_acl(zp);
1280 1280
1281 1281 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
1282 1282 if (acl_obj) {
1283 1283 VERIFY(!zp->z_is_sa);
1284 1284 VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1285 1285 }
1286 1286 VERIFY(0 == dmu_object_free(os, obj, tx));
1287 1287 zfs_znode_dmu_fini(zp);
1288 1288 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1289 1289 zfs_znode_free(zp);
1290 1290 }
1291 1291
1292 1292 void
1293 1293 zfs_zinactive(znode_t *zp)
1294 1294 {
1295 1295 vnode_t *vp = ZTOV(zp);
1296 1296 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1297 1297 uint64_t z_id = zp->z_id;
1298 1298
1299 1299 ASSERT(zp->z_sa_hdl);
1300 1300
1301 1301 /*
1302 1302 * Don't allow a zfs_zget() while were trying to release this znode
1303 1303 */
1304 1304 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
1305 1305
1306 1306 mutex_enter(&zp->z_lock);
1307 1307 mutex_enter(&vp->v_lock);
1308 1308 VN_RELE_LOCKED(vp);
1309 1309 if (vp->v_count > 0 || vn_has_cached_data(vp)) {
1310 1310 /*
1311 1311 * If the hold count is greater than zero, somebody has
1312 1312 * obtained a new reference on this znode while we were
1313 1313 * processing it here, so we are done. If we still have
1314 1314 * mapped pages then we are also done, since we don't
1315 1315 * want to inactivate the znode until the pages get pushed.
1316 1316 *
1317 1317 * XXX - if vn_has_cached_data(vp) is true, but count == 0,
1318 1318 * this seems like it would leave the znode hanging with
1319 1319 * no chance to go inactive...
1320 1320 */
1321 1321 mutex_exit(&vp->v_lock);
1322 1322 mutex_exit(&zp->z_lock);
1323 1323 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1324 1324 return;
1325 1325 }
1326 1326 mutex_exit(&vp->v_lock);
1327 1327
1328 1328 /*
1329 1329 * If this was the last reference to a file with no links, remove
1330 1330 * the file from the file system unless the file system is mounted
1331 1331 * read-only. That can happen, for example, if the file system was
1332 1332 * originally read-write, the file was opened, then unlinked and
1333 1333 * the file system was made read-only before the file was finally
1334 1334 * closed. The file will remain in the unlinked set.
1335 1335 */
1336 1336 if (zp->z_unlinked) {
1337 1337 ASSERT(!zfsvfs->z_issnap);
1338 1338 if ((zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) == 0) {
1339 1339 mutex_exit(&zp->z_lock);
1340 1340 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1341 1341 zfs_rmnode(zp);
1342 1342 return;
1343 1343 }
1344 1344 }
1345 1345
1346 1346 mutex_exit(&zp->z_lock);
1347 1347 zfs_znode_dmu_fini(zp);
1348 1348 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1349 1349 zfs_znode_free(zp);
1350 1350 }
1351 1351
1352 1352 void
1353 1353 zfs_znode_free(znode_t *zp)
1354 1354 {
1355 1355 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1356 1356
1357 1357 vn_invalid(ZTOV(zp));
1358 1358
1359 1359 ASSERT(ZTOV(zp)->v_count == 0);
1360 1360
1361 1361 mutex_enter(&zfsvfs->z_znodes_lock);
1362 1362 POINTER_INVALIDATE(&zp->z_zfsvfs);
1363 1363 list_remove(&zfsvfs->z_all_znodes, zp);
1364 1364 mutex_exit(&zfsvfs->z_znodes_lock);
1365 1365
1366 1366 if (zp->z_acl_cached) {
1367 1367 zfs_acl_free(zp->z_acl_cached);
1368 1368 zp->z_acl_cached = NULL;
1369 1369 }
1370 1370
1371 1371 kmem_cache_free(znode_cache, zp);
1372 1372
1373 1373 VFS_RELE(zfsvfs->z_vfs);
1374 1374 }
1375 1375
1376 1376 void
1377 1377 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1378 1378 uint64_t ctime[2], boolean_t have_tx)
1379 1379 {
1380 1380 timestruc_t now;
1381 1381
1382 1382 gethrestime(&now);
1383 1383
1384 1384 if (have_tx) { /* will sa_bulk_update happen really soon? */
1385 1385 zp->z_atime_dirty = 0;
1386 1386 zp->z_seq++;
1387 1387 } else {
1388 1388 zp->z_atime_dirty = 1;
1389 1389 }
1390 1390
1391 1391 if (flag & AT_ATIME) {
1392 1392 ZFS_TIME_ENCODE(&now, zp->z_atime);
1393 1393 }
1394 1394
1395 1395 if (flag & AT_MTIME) {
1396 1396 ZFS_TIME_ENCODE(&now, mtime);
1397 1397 if (zp->z_zfsvfs->z_use_fuids) {
1398 1398 zp->z_pflags |= (ZFS_ARCHIVE |
1399 1399 ZFS_AV_MODIFIED);
1400 1400 }
1401 1401 }
1402 1402
1403 1403 if (flag & AT_CTIME) {
1404 1404 ZFS_TIME_ENCODE(&now, ctime);
1405 1405 if (zp->z_zfsvfs->z_use_fuids)
1406 1406 zp->z_pflags |= ZFS_ARCHIVE;
1407 1407 }
1408 1408 }
1409 1409
1410 1410 /*
1411 1411 * Grow the block size for a file.
1412 1412 *
1413 1413 * IN: zp - znode of file to free data in.
1414 1414 * size - requested block size
1415 1415 * tx - open transaction.
1416 1416 *
1417 1417 * NOTE: this function assumes that the znode is write locked.
1418 1418 */
1419 1419 void
1420 1420 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1421 1421 {
1422 1422 int error;
1423 1423 u_longlong_t dummy;
1424 1424
1425 1425 if (size <= zp->z_blksz)
1426 1426 return;
1427 1427 /*
1428 1428 * If the file size is already greater than the current blocksize,
1429 1429 * we will not grow. If there is more than one block in a file,
1430 1430 * the blocksize cannot change.
1431 1431 */
1432 1432 if (zp->z_blksz && zp->z_size > zp->z_blksz)
1433 1433 return;
1434 1434
1435 1435 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id,
1436 1436 size, 0, tx);
1437 1437
1438 1438 if (error == ENOTSUP)
1439 1439 return;
1440 1440 ASSERT0(error);
1441 1441
1442 1442 /* What blocksize did we actually get? */
1443 1443 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1444 1444 }
1445 1445
1446 1446 /*
1447 1447 * This is a dummy interface used when pvn_vplist_dirty() should *not*
1448 1448 * be calling back into the fs for a putpage(). E.g.: when truncating
1449 1449 * a file, the pages being "thrown away* don't need to be written out.
1450 1450 */
1451 1451 /* ARGSUSED */
1452 1452 static int
1453 1453 zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
1454 1454 int flags, cred_t *cr)
1455 1455 {
1456 1456 ASSERT(0);
1457 1457 return (0);
1458 1458 }
1459 1459
1460 1460 /*
1461 1461 * Increase the file length
1462 1462 *
1463 1463 * IN: zp - znode of file to free data in.
1464 1464 * end - new end-of-file
1465 1465 *
1466 1466 * RETURN: 0 on success, error code on failure
1467 1467 */
1468 1468 static int
1469 1469 zfs_extend(znode_t *zp, uint64_t end)
1470 1470 {
1471 1471 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1472 1472 dmu_tx_t *tx;
1473 1473 rl_t *rl;
1474 1474 uint64_t newblksz;
1475 1475 int error;
1476 1476
1477 1477 /*
1478 1478 * We will change zp_size, lock the whole file.
1479 1479 */
1480 1480 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1481 1481
1482 1482 /*
1483 1483 * Nothing to do if file already at desired length.
1484 1484 */
1485 1485 if (end <= zp->z_size) {
1486 1486 zfs_range_unlock(rl);
1487 1487 return (0);
1488 1488 }
1489 1489 tx = dmu_tx_create(zfsvfs->z_os);
1490 1490 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1491 1491 zfs_sa_upgrade_txholds(tx, zp);
1492 1492 if (end > zp->z_blksz &&
1493 1493 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1494 1494 /*
1495 1495 * We are growing the file past the current block size.
1496 1496 */
1497 1497 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) {
1498 1498 /*
1499 1499 * File's blocksize is already larger than the
1500 1500 * "recordsize" property. Only let it grow to
1501 1501 * the next power of 2.
1502 1502 */
1503 1503 ASSERT(!ISP2(zp->z_blksz));
1504 1504 newblksz = MIN(end, 1 << highbit64(zp->z_blksz));
1505 1505 } else {
1506 1506 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz);
1507 1507 }
1508 1508 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1509 1509 } else {
1510 1510 newblksz = 0;
1511 1511 }
1512 1512
1513 1513 error = dmu_tx_assign(tx, TXG_WAIT);
1514 1514 if (error) {
1515 1515 dmu_tx_abort(tx);
1516 1516 zfs_range_unlock(rl);
1517 1517 return (error);
1518 1518 }
1519 1519
1520 1520 if (newblksz)
1521 1521 zfs_grow_blocksize(zp, newblksz, tx);
1522 1522
1523 1523 zp->z_size = end;
1524 1524
1525 1525 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs),
1526 1526 &zp->z_size, sizeof (zp->z_size), tx));
1527 1527
1528 1528 zfs_range_unlock(rl);
1529 1529
1530 1530 dmu_tx_commit(tx);
1531 1531
1532 1532 return (0);
1533 1533 }
1534 1534
1535 1535 /*
1536 1536 * Free space in a file.
1537 1537 *
1538 1538 * IN: zp - znode of file to free data in.
1539 1539 * off - start of section to free.
1540 1540 * len - length of section to free.
1541 1541 *
1542 1542 * RETURN: 0 on success, error code on failure
1543 1543 */
1544 1544 static int
1545 1545 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1546 1546 {
1547 1547 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1548 1548 rl_t *rl;
1549 1549 int error;
1550 1550
1551 1551 /*
1552 1552 * Lock the range being freed.
1553 1553 */
1554 1554 rl = zfs_range_lock(zp, off, len, RL_WRITER);
1555 1555
1556 1556 /*
1557 1557 * Nothing to do if file already at desired length.
1558 1558 */
1559 1559 if (off >= zp->z_size) {
1560 1560 zfs_range_unlock(rl);
1561 1561 return (0);
1562 1562 }
1563 1563
1564 1564 if (off + len > zp->z_size)
1565 1565 len = zp->z_size - off;
1566 1566
1567 1567 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1568 1568
1569 1569 zfs_range_unlock(rl);
1570 1570
1571 1571 return (error);
1572 1572 }
1573 1573
1574 1574 /*
1575 1575 * Truncate a file
1576 1576 *
1577 1577 * IN: zp - znode of file to free data in.
1578 1578 * end - new end-of-file.
1579 1579 *
1580 1580 * RETURN: 0 on success, error code on failure
1581 1581 */
1582 1582 static int
1583 1583 zfs_trunc(znode_t *zp, uint64_t end)
1584 1584 {
1585 1585 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1586 1586 vnode_t *vp = ZTOV(zp);
1587 1587 dmu_tx_t *tx;
1588 1588 rl_t *rl;
1589 1589 int error;
1590 1590 sa_bulk_attr_t bulk[2];
1591 1591 int count = 0;
1592 1592
1593 1593 /*
1594 1594 * We will change zp_size, lock the whole file.
1595 1595 */
1596 1596 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1597 1597
1598 1598 /*
1599 1599 * Nothing to do if file already at desired length.
1600 1600 */
1601 1601 if (end >= zp->z_size) {
1602 1602 zfs_range_unlock(rl);
1603 1603 return (0);
1604 1604 }
1605 1605
1606 1606 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, -1);
1607 1607 if (error) {
1608 1608 zfs_range_unlock(rl);
1609 1609 return (error);
1610 1610 }
1611 1611 tx = dmu_tx_create(zfsvfs->z_os);
1612 1612 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1613 1613 zfs_sa_upgrade_txholds(tx, zp);
1614 1614 dmu_tx_mark_netfree(tx);
1615 1615 error = dmu_tx_assign(tx, TXG_WAIT);
1616 1616 if (error) {
1617 1617 dmu_tx_abort(tx);
1618 1618 zfs_range_unlock(rl);
1619 1619 return (error);
1620 1620 }
1621 1621
1622 1622 zp->z_size = end;
1623 1623 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1624 1624 NULL, &zp->z_size, sizeof (zp->z_size));
1625 1625
1626 1626 if (end == 0) {
1627 1627 zp->z_pflags &= ~ZFS_SPARSE;
1628 1628 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1629 1629 NULL, &zp->z_pflags, 8);
1630 1630 }
1631 1631 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0);
1632 1632
1633 1633 dmu_tx_commit(tx);
1634 1634
1635 1635 /*
1636 1636 * Clear any mapped pages in the truncated region. This has to
1637 1637 * happen outside of the transaction to avoid the possibility of
1638 1638 * a deadlock with someone trying to push a page that we are
1639 1639 * about to invalidate.
1640 1640 */
1641 1641 if (vn_has_cached_data(vp)) {
1642 1642 page_t *pp;
1643 1643 uint64_t start = end & PAGEMASK;
1644 1644 int poff = end & PAGEOFFSET;
1645 1645
1646 1646 if (poff != 0 && (pp = page_lookup(vp, start, SE_SHARED))) {
1647 1647 /*
1648 1648 * We need to zero a partial page.
1649 1649 */
1650 1650 pagezero(pp, poff, PAGESIZE - poff);
1651 1651 start += PAGESIZE;
1652 1652 page_unlock(pp);
1653 1653 }
1654 1654 error = pvn_vplist_dirty(vp, start, zfs_no_putpage,
1655 1655 B_INVAL | B_TRUNC, NULL);
1656 1656 ASSERT(error == 0);
1657 1657 }
1658 1658
1659 1659 zfs_range_unlock(rl);
1660 1660
1661 1661 return (0);
1662 1662 }
1663 1663
1664 1664 /*
1665 1665 * Free space in a file
1666 1666 *
1667 1667 * IN: zp - znode of file to free data in.
1668 1668 * off - start of range
1669 1669 * len - end of range (0 => EOF)
1670 1670 * flag - current file open mode flags.
1671 1671 * log - TRUE if this action should be logged
1672 1672 *
1673 1673 * RETURN: 0 on success, error code on failure
1674 1674 */
1675 1675 int
1676 1676 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1677 1677 {
1678 1678 vnode_t *vp = ZTOV(zp);
1679 1679 dmu_tx_t *tx;
1680 1680 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1681 1681 zilog_t *zilog = zfsvfs->z_log;
1682 1682 uint64_t mode;
1683 1683 uint64_t mtime[2], ctime[2];
1684 1684 sa_bulk_attr_t bulk[3];
1685 1685 int count = 0;
1686 1686 int error;
1687 1687
1688 1688 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1689 1689 sizeof (mode))) != 0)
1690 1690 return (error);
1691 1691
1692 1692 if (off > zp->z_size) {
1693 1693 error = zfs_extend(zp, off+len);
1694 1694 if (error == 0 && log)
1695 1695 goto log;
1696 1696 else
1697 1697 return (error);
1698 1698 }
1699 1699
1700 1700 /*
1701 1701 * Check for any locks in the region to be freed.
1702 1702 */
1703 1703
1704 1704 if (MANDLOCK(vp, (mode_t)mode)) {
1705 1705 uint64_t length = (len ? len : zp->z_size - off);
1706 1706 if (error = chklock(vp, FWRITE, off, length, flag, NULL))
1707 1707 return (error);
1708 1708 }
1709 1709
1710 1710 if (len == 0) {
1711 1711 error = zfs_trunc(zp, off);
1712 1712 } else {
1713 1713 if ((error = zfs_free_range(zp, off, len)) == 0 &&
1714 1714 off + len > zp->z_size)
1715 1715 error = zfs_extend(zp, off+len);
1716 1716 }
1717 1717 if (error || !log)
1718 1718 return (error);
1719 1719 log:
1720 1720 tx = dmu_tx_create(zfsvfs->z_os);
1721 1721 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1722 1722 zfs_sa_upgrade_txholds(tx, zp);
1723 1723 error = dmu_tx_assign(tx, TXG_WAIT);
1724 1724 if (error) {
1725 1725 dmu_tx_abort(tx);
1726 1726 return (error);
1727 1727 }
1728 1728
1729 1729 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1730 1730 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1731 1731 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1732 1732 NULL, &zp->z_pflags, 8);
1733 1733 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
1734 1734 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1735 1735 ASSERT(error == 0);
1736 1736
1737 1737 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1738 1738
1739 1739 dmu_tx_commit(tx);
1740 1740 return (0);
1741 1741 }
|
↓ open down ↓ |
1741 lines elided |
↑ open up ↑ |
1742 1742
1743 1743 void
1744 1744 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1745 1745 {
1746 1746 uint64_t moid, obj, sa_obj, version;
1747 1747 uint64_t sense = ZFS_CASE_SENSITIVE;
1748 1748 uint64_t norm = 0;
1749 1749 nvpair_t *elem;
1750 1750 int error;
1751 1751 int i;
1752 + int size = spa_get_obj_mtx_sz(dmu_objset_spa(os));
1752 1753 znode_t *rootzp = NULL;
1753 1754 zfsvfs_t *zfsvfs;
1754 1755 vnode_t *vp;
1755 1756 vattr_t vattr;
1756 1757 znode_t *zp;
1757 1758 zfs_acl_ids_t acl_ids;
1758 1759
1759 1760 /*
1760 1761 * First attempt to create master node.
1761 1762 */
1762 1763 /*
1763 1764 * In an empty objset, there are no blocks to read and thus
1764 1765 * there can be no i/o errors (which we assert below).
1765 1766 */
1766 1767 moid = MASTER_NODE_OBJ;
1767 1768 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1768 1769 DMU_OT_NONE, 0, tx);
1769 1770 ASSERT(error == 0);
1770 1771
1771 1772 /*
1772 1773 * Set starting attributes.
1773 1774 */
1774 1775 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1775 1776 elem = NULL;
1776 1777 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1777 1778 /* For the moment we expect all zpl props to be uint64_ts */
1778 1779 uint64_t val;
1779 1780 char *name;
1780 1781
1781 1782 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1782 1783 VERIFY(nvpair_value_uint64(elem, &val) == 0);
1783 1784 name = nvpair_name(elem);
1784 1785 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1785 1786 if (val < version)
1786 1787 version = val;
1787 1788 } else {
1788 1789 error = zap_update(os, moid, name, 8, 1, &val, tx);
1789 1790 }
1790 1791 ASSERT(error == 0);
1791 1792 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1792 1793 norm = val;
1793 1794 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1794 1795 sense = val;
1795 1796 }
1796 1797 ASSERT(version != 0);
1797 1798 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1798 1799
1799 1800 /*
1800 1801 * Create zap object used for SA attribute registration
1801 1802 */
1802 1803
1803 1804 if (version >= ZPL_VERSION_SA) {
1804 1805 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1805 1806 DMU_OT_NONE, 0, tx);
1806 1807 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1807 1808 ASSERT(error == 0);
1808 1809 } else {
1809 1810 sa_obj = 0;
1810 1811 }
1811 1812 /*
1812 1813 * Create a delete queue.
1813 1814 */
1814 1815 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1815 1816
1816 1817 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1817 1818 ASSERT(error == 0);
1818 1819
1819 1820 /*
1820 1821 * Create root znode. Create minimal znode/vnode/zfsvfs
1821 1822 * to allow zfs_mknode to work.
1822 1823 */
1823 1824 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
1824 1825 vattr.va_type = VDIR;
1825 1826 vattr.va_mode = S_IFDIR|0755;
1826 1827 vattr.va_uid = crgetuid(cr);
1827 1828 vattr.va_gid = crgetgid(cr);
1828 1829
1829 1830 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP);
1830 1831 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs));
1831 1832 rootzp->z_moved = 0;
1832 1833 rootzp->z_unlinked = 0;
1833 1834 rootzp->z_atime_dirty = 0;
1834 1835 rootzp->z_is_sa = USE_SA(version, os);
1835 1836
1836 1837 vp = ZTOV(rootzp);
1837 1838 vn_reinit(vp);
1838 1839 vp->v_type = VDIR;
1839 1840
1840 1841 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1841 1842 zfsvfs->z_os = os;
1842 1843 zfsvfs->z_parent = zfsvfs;
1843 1844 zfsvfs->z_version = version;
1844 1845 zfsvfs->z_use_fuids = USE_FUIDS(version, os);
1845 1846 zfsvfs->z_use_sa = USE_SA(version, os);
1846 1847 zfsvfs->z_norm = norm;
1847 1848
1848 1849 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1849 1850 &zfsvfs->z_attr_table);
1850 1851
1851 1852 ASSERT(error == 0);
1852 1853
1853 1854 /*
|
↓ open down ↓ |
92 lines elided |
↑ open up ↑ |
1854 1855 * Fold case on file systems that are always or sometimes case
1855 1856 * insensitive.
1856 1857 */
1857 1858 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1858 1859 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1859 1860
1860 1861 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1861 1862 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1862 1863 offsetof(znode_t, z_link_node));
1863 1864
1864 - for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1865 + zfsvfs->z_hold_mtx_sz = size;
1866 + zfsvfs->z_hold_mtx = kmem_zalloc(sizeof (kmutex_t) * size, KM_SLEEP);
1867 + for (i = 0; i != size; i++)
1865 1868 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1866 1869
1867 1870 rootzp->z_zfsvfs = zfsvfs;
1868 1871 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1869 1872 cr, NULL, &acl_ids));
1870 1873 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1871 1874 ASSERT3P(zp, ==, rootzp);
1872 1875 ASSERT(!vn_in_dnlc(ZTOV(rootzp))); /* not valid to move */
1873 1876 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1874 1877 ASSERT(error == 0);
1875 1878 zfs_acl_ids_free(&acl_ids);
1876 1879 POINTER_INVALIDATE(&rootzp->z_zfsvfs);
1877 1880
1878 1881 ZTOV(rootzp)->v_count = 0;
1879 1882 sa_handle_destroy(rootzp->z_sa_hdl);
|
↓ open down ↓ |
5 lines elided |
↑ open up ↑ |
1880 1883 kmem_cache_free(znode_cache, rootzp);
1881 1884
1882 1885 /*
1883 1886 * Create shares directory
1884 1887 */
1885 1888
1886 1889 error = zfs_create_share_dir(zfsvfs, tx);
1887 1890
1888 1891 ASSERT(error == 0);
1889 1892
1890 - for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1893 + for (i = 0; i != size; i++)
1891 1894 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1895 +
1896 + kmem_free(zfsvfs->z_hold_mtx, sizeof (kmutex_t) * size);
1892 1897 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1893 1898 }
1894 1899
1895 1900 #endif /* _KERNEL */
1896 1901
1897 1902 static int
1898 1903 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
1899 1904 {
1900 1905 uint64_t sa_obj = 0;
1901 1906 int error;
1902 1907
1903 1908 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
1904 1909 if (error != 0 && error != ENOENT)
1905 1910 return (error);
1906 1911
1907 1912 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
1908 1913 return (error);
1909 1914 }
1910 1915
1911 1916 static int
1912 1917 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
1913 1918 dmu_buf_t **db, void *tag)
1914 1919 {
1915 1920 dmu_object_info_t doi;
1916 1921 int error;
1917 1922
1918 1923 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
1919 1924 return (error);
1920 1925
1921 1926 dmu_object_info_from_db(*db, &doi);
1922 1927 if ((doi.doi_bonus_type != DMU_OT_SA &&
1923 1928 doi.doi_bonus_type != DMU_OT_ZNODE) ||
1924 1929 doi.doi_bonus_type == DMU_OT_ZNODE &&
1925 1930 doi.doi_bonus_size < sizeof (znode_phys_t)) {
1926 1931 sa_buf_rele(*db, tag);
1927 1932 return (SET_ERROR(ENOTSUP));
1928 1933 }
1929 1934
1930 1935 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
1931 1936 if (error != 0) {
1932 1937 sa_buf_rele(*db, tag);
1933 1938 return (error);
1934 1939 }
1935 1940
1936 1941 return (0);
1937 1942 }
1938 1943
1939 1944 void
1940 1945 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
1941 1946 {
1942 1947 sa_handle_destroy(hdl);
1943 1948 sa_buf_rele(db, tag);
1944 1949 }
1945 1950
1946 1951 /*
1947 1952 * Given an object number, return its parent object number and whether
1948 1953 * or not the object is an extended attribute directory.
1949 1954 */
1950 1955 static int
1951 1956 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table,
1952 1957 uint64_t *pobjp, int *is_xattrdir)
1953 1958 {
1954 1959 uint64_t parent;
1955 1960 uint64_t pflags;
1956 1961 uint64_t mode;
1957 1962 uint64_t parent_mode;
1958 1963 sa_bulk_attr_t bulk[3];
1959 1964 sa_handle_t *sa_hdl;
1960 1965 dmu_buf_t *sa_db;
1961 1966 int count = 0;
1962 1967 int error;
1963 1968
1964 1969 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
1965 1970 &parent, sizeof (parent));
1966 1971 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
1967 1972 &pflags, sizeof (pflags));
1968 1973 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
1969 1974 &mode, sizeof (mode));
1970 1975
1971 1976 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
1972 1977 return (error);
1973 1978
1974 1979 /*
1975 1980 * When a link is removed its parent pointer is not changed and will
1976 1981 * be invalid. There are two cases where a link is removed but the
1977 1982 * file stays around, when it goes to the delete queue and when there
1978 1983 * are additional links.
1979 1984 */
1980 1985 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG);
1981 1986 if (error != 0)
1982 1987 return (error);
1983 1988
1984 1989 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode));
1985 1990 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
1986 1991 if (error != 0)
1987 1992 return (error);
1988 1993
1989 1994 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
1990 1995
1991 1996 /*
1992 1997 * Extended attributes can be applied to files, directories, etc.
1993 1998 * Otherwise the parent must be a directory.
1994 1999 */
1995 2000 if (!*is_xattrdir && !S_ISDIR(parent_mode))
1996 2001 return (SET_ERROR(EINVAL));
1997 2002
1998 2003 *pobjp = parent;
1999 2004
2000 2005 return (0);
2001 2006 }
2002 2007
2003 2008 /*
2004 2009 * Given an object number, return some zpl level statistics
2005 2010 */
2006 2011 static int
2007 2012 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
2008 2013 zfs_stat_t *sb)
2009 2014 {
2010 2015 sa_bulk_attr_t bulk[4];
2011 2016 int count = 0;
2012 2017
2013 2018 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2014 2019 &sb->zs_mode, sizeof (sb->zs_mode));
2015 2020 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
2016 2021 &sb->zs_gen, sizeof (sb->zs_gen));
2017 2022 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
2018 2023 &sb->zs_links, sizeof (sb->zs_links));
2019 2024 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
2020 2025 &sb->zs_ctime, sizeof (sb->zs_ctime));
2021 2026
2022 2027 return (sa_bulk_lookup(hdl, bulk, count));
2023 2028 }
2024 2029
2025 2030 static int
2026 2031 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
2027 2032 sa_attr_type_t *sa_table, char *buf, int len)
2028 2033 {
2029 2034 sa_handle_t *sa_hdl;
2030 2035 sa_handle_t *prevhdl = NULL;
2031 2036 dmu_buf_t *prevdb = NULL;
2032 2037 dmu_buf_t *sa_db = NULL;
2033 2038 char *path = buf + len - 1;
2034 2039 int error;
2035 2040
2036 2041 *path = '\0';
2037 2042 sa_hdl = hdl;
2038 2043
2039 2044 for (;;) {
2040 2045 uint64_t pobj;
2041 2046 char component[MAXNAMELEN + 2];
2042 2047 size_t complen;
2043 2048 int is_xattrdir;
2044 2049
2045 2050 if (prevdb)
2046 2051 zfs_release_sa_handle(prevhdl, prevdb, FTAG);
2047 2052
2048 2053 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj,
2049 2054 &is_xattrdir)) != 0)
2050 2055 break;
2051 2056
2052 2057 if (pobj == obj) {
2053 2058 if (path[0] != '/')
2054 2059 *--path = '/';
2055 2060 break;
2056 2061 }
2057 2062
2058 2063 component[0] = '/';
2059 2064 if (is_xattrdir) {
2060 2065 (void) sprintf(component + 1, "<xattrdir>");
2061 2066 } else {
2062 2067 error = zap_value_search(osp, pobj, obj,
2063 2068 ZFS_DIRENT_OBJ(-1ULL), component + 1);
2064 2069 if (error != 0)
2065 2070 break;
2066 2071 }
2067 2072
2068 2073 complen = strlen(component);
2069 2074 path -= complen;
2070 2075 ASSERT(path >= buf);
2071 2076 bcopy(component, path, complen);
2072 2077 obj = pobj;
2073 2078
2074 2079 if (sa_hdl != hdl) {
2075 2080 prevhdl = sa_hdl;
2076 2081 prevdb = sa_db;
2077 2082 }
2078 2083 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
2079 2084 if (error != 0) {
2080 2085 sa_hdl = prevhdl;
2081 2086 sa_db = prevdb;
2082 2087 break;
2083 2088 }
2084 2089 }
2085 2090
2086 2091 if (sa_hdl != NULL && sa_hdl != hdl) {
2087 2092 ASSERT(sa_db != NULL);
2088 2093 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2089 2094 }
2090 2095
2091 2096 if (error == 0)
2092 2097 (void) memmove(buf, path, buf + len - path);
2093 2098
2094 2099 return (error);
2095 2100 }
2096 2101
2097 2102 int
2098 2103 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
2099 2104 {
2100 2105 sa_attr_type_t *sa_table;
2101 2106 sa_handle_t *hdl;
2102 2107 dmu_buf_t *db;
2103 2108 int error;
2104 2109
2105 2110 error = zfs_sa_setup(osp, &sa_table);
2106 2111 if (error != 0)
2107 2112 return (error);
2108 2113
2109 2114 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2110 2115 if (error != 0)
2111 2116 return (error);
2112 2117
2113 2118 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2114 2119
2115 2120 zfs_release_sa_handle(hdl, db, FTAG);
2116 2121 return (error);
2117 2122 }
2118 2123
2119 2124 int
2120 2125 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
2121 2126 char *buf, int len)
2122 2127 {
2123 2128 char *path = buf + len - 1;
2124 2129 sa_attr_type_t *sa_table;
2125 2130 sa_handle_t *hdl;
2126 2131 dmu_buf_t *db;
2127 2132 int error;
2128 2133
2129 2134 *path = '\0';
2130 2135
2131 2136 error = zfs_sa_setup(osp, &sa_table);
2132 2137 if (error != 0)
2133 2138 return (error);
2134 2139
2135 2140 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2136 2141 if (error != 0)
2137 2142 return (error);
2138 2143
2139 2144 error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2140 2145 if (error != 0) {
2141 2146 zfs_release_sa_handle(hdl, db, FTAG);
2142 2147 return (error);
2143 2148 }
2144 2149
2145 2150 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2146 2151
2147 2152 zfs_release_sa_handle(hdl, db, FTAG);
2148 2153 return (error);
2149 2154 }
|
↓ open down ↓ |
248 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX