xref: /freebsd/sys/contrib/openzfs/module/zfs/zfs_sa.c (revision e2eeea75eb8b6dd50c1298067a0655880d186734)
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) 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 
25 #include <sys/zfs_context.h>
26 #include <sys/vnode.h>
27 #include <sys/sa.h>
28 #include <sys/zfs_acl.h>
29 #include <sys/zfs_sa.h>
30 #include <sys/dmu_objset.h>
31 #include <sys/sa_impl.h>
32 
33 /*
34  * ZPL attribute registration table.
35  * Order of attributes doesn't matter
36  * a unique value will be assigned for each
37  * attribute that is file system specific
38  *
39  * This is just the set of ZPL attributes that this
40  * version of ZFS deals with natively.  The file system
41  * could have other attributes stored in files, but they will be
42  * ignored.  The SA framework will preserve them, just that
43  * this version of ZFS won't change or delete them.
44  */
45 
46 sa_attr_reg_t zfs_attr_table[ZPL_END+1] = {
47 	{"ZPL_ATIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 0},
48 	{"ZPL_MTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 1},
49 	{"ZPL_CTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 2},
50 	{"ZPL_CRTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 3},
51 	{"ZPL_GEN", sizeof (uint64_t), SA_UINT64_ARRAY, 4},
52 	{"ZPL_MODE", sizeof (uint64_t), SA_UINT64_ARRAY, 5},
53 	{"ZPL_SIZE", sizeof (uint64_t), SA_UINT64_ARRAY, 6},
54 	{"ZPL_PARENT", sizeof (uint64_t), SA_UINT64_ARRAY, 7},
55 	{"ZPL_LINKS", sizeof (uint64_t), SA_UINT64_ARRAY, 8},
56 	{"ZPL_XATTR", sizeof (uint64_t), SA_UINT64_ARRAY, 9},
57 	{"ZPL_RDEV", sizeof (uint64_t), SA_UINT64_ARRAY, 10},
58 	{"ZPL_FLAGS", sizeof (uint64_t), SA_UINT64_ARRAY, 11},
59 	{"ZPL_UID", sizeof (uint64_t), SA_UINT64_ARRAY, 12},
60 	{"ZPL_GID", sizeof (uint64_t), SA_UINT64_ARRAY, 13},
61 	{"ZPL_PAD", sizeof (uint64_t) * 4, SA_UINT64_ARRAY, 14},
62 	{"ZPL_ZNODE_ACL", 88, SA_UINT8_ARRAY, 15},
63 	{"ZPL_DACL_COUNT", sizeof (uint64_t), SA_UINT64_ARRAY, 0},
64 	{"ZPL_SYMLINK", 0, SA_UINT8_ARRAY, 0},
65 	{"ZPL_SCANSTAMP", 32, SA_UINT8_ARRAY, 0},
66 	{"ZPL_DACL_ACES", 0, SA_ACL, 0},
67 	{"ZPL_DXATTR", 0, SA_UINT8_ARRAY, 0},
68 	{"ZPL_PROJID", sizeof (uint64_t), SA_UINT64_ARRAY, 0},
69 	{NULL, 0, 0, 0}
70 };
71 
72 #ifdef _KERNEL
73 int
74 zfs_sa_readlink(znode_t *zp, uio_t *uio)
75 {
76 	dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
77 	size_t bufsz;
78 	int error;
79 
80 	bufsz = zp->z_size;
81 	if (bufsz + ZFS_OLD_ZNODE_PHYS_SIZE <= db->db_size) {
82 		error = uiomove((caddr_t)db->db_data +
83 		    ZFS_OLD_ZNODE_PHYS_SIZE,
84 		    MIN((size_t)bufsz, uio_resid(uio)), UIO_READ, uio);
85 	} else {
86 		dmu_buf_t *dbp;
87 		if ((error = dmu_buf_hold(ZTOZSB(zp)->z_os, zp->z_id,
88 		    0, FTAG, &dbp, DMU_READ_NO_PREFETCH)) == 0) {
89 			error = uiomove(dbp->db_data,
90 			    MIN((size_t)bufsz, uio_resid(uio)), UIO_READ, uio);
91 			dmu_buf_rele(dbp, FTAG);
92 		}
93 	}
94 	return (error);
95 }
96 
97 void
98 zfs_sa_symlink(znode_t *zp, char *link, int len, dmu_tx_t *tx)
99 {
100 	dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
101 
102 	if (ZFS_OLD_ZNODE_PHYS_SIZE + len <= dmu_bonus_max()) {
103 		VERIFY0(dmu_set_bonus(db, len + ZFS_OLD_ZNODE_PHYS_SIZE, tx));
104 		if (len) {
105 			bcopy(link, (caddr_t)db->db_data +
106 			    ZFS_OLD_ZNODE_PHYS_SIZE, len);
107 		}
108 	} else {
109 		dmu_buf_t *dbp;
110 
111 		zfs_grow_blocksize(zp, len, tx);
112 		VERIFY0(dmu_buf_hold(ZTOZSB(zp)->z_os, zp->z_id, 0, FTAG, &dbp,
113 		    DMU_READ_NO_PREFETCH));
114 
115 		dmu_buf_will_dirty(dbp, tx);
116 
117 		ASSERT3U(len, <=, dbp->db_size);
118 		bcopy(link, dbp->db_data, len);
119 		dmu_buf_rele(dbp, FTAG);
120 	}
121 }
122 
123 void
124 zfs_sa_get_scanstamp(znode_t *zp, xvattr_t *xvap)
125 {
126 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
127 	xoptattr_t *xoap;
128 
129 	ASSERT(MUTEX_HELD(&zp->z_lock));
130 	VERIFY((xoap = xva_getxoptattr(xvap)) != NULL);
131 	if (zp->z_is_sa) {
132 		if (sa_lookup(zp->z_sa_hdl, SA_ZPL_SCANSTAMP(zfsvfs),
133 		    &xoap->xoa_av_scanstamp,
134 		    sizeof (xoap->xoa_av_scanstamp)) != 0)
135 			return;
136 	} else {
137 		dmu_object_info_t doi;
138 		dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
139 		int len;
140 
141 		if (!(zp->z_pflags & ZFS_BONUS_SCANSTAMP))
142 			return;
143 
144 		sa_object_info(zp->z_sa_hdl, &doi);
145 		len = sizeof (xoap->xoa_av_scanstamp) +
146 		    ZFS_OLD_ZNODE_PHYS_SIZE;
147 
148 		if (len <= doi.doi_bonus_size) {
149 			(void) memcpy(xoap->xoa_av_scanstamp,
150 			    (caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
151 			    sizeof (xoap->xoa_av_scanstamp));
152 		}
153 	}
154 	XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
155 }
156 
157 void
158 zfs_sa_set_scanstamp(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
159 {
160 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
161 	xoptattr_t *xoap;
162 
163 	ASSERT(MUTEX_HELD(&zp->z_lock));
164 	VERIFY((xoap = xva_getxoptattr(xvap)) != NULL);
165 	if (zp->z_is_sa)
166 		VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SCANSTAMP(zfsvfs),
167 		    &xoap->xoa_av_scanstamp,
168 		    sizeof (xoap->xoa_av_scanstamp), tx));
169 	else {
170 		dmu_object_info_t doi;
171 		dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
172 		int len;
173 
174 		sa_object_info(zp->z_sa_hdl, &doi);
175 		len = sizeof (xoap->xoa_av_scanstamp) +
176 		    ZFS_OLD_ZNODE_PHYS_SIZE;
177 		if (len > doi.doi_bonus_size)
178 			VERIFY(dmu_set_bonus(db, len, tx) == 0);
179 		(void) memcpy((caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
180 		    xoap->xoa_av_scanstamp, sizeof (xoap->xoa_av_scanstamp));
181 
182 		zp->z_pflags |= ZFS_BONUS_SCANSTAMP;
183 		VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
184 		    &zp->z_pflags, sizeof (uint64_t), tx));
185 	}
186 }
187 
188 int
189 zfs_sa_get_xattr(znode_t *zp)
190 {
191 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
192 	char *obj;
193 	int size;
194 	int error;
195 
196 	ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock));
197 	ASSERT(!zp->z_xattr_cached);
198 	ASSERT(zp->z_is_sa);
199 
200 	error = sa_size(zp->z_sa_hdl, SA_ZPL_DXATTR(zfsvfs), &size);
201 	if (error) {
202 		if (error == ENOENT)
203 			return nvlist_alloc(&zp->z_xattr_cached,
204 			    NV_UNIQUE_NAME, KM_SLEEP);
205 		else
206 			return (error);
207 	}
208 
209 	obj = vmem_alloc(size, KM_SLEEP);
210 
211 	error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DXATTR(zfsvfs), obj, size);
212 	if (error == 0)
213 		error = nvlist_unpack(obj, size, &zp->z_xattr_cached, KM_SLEEP);
214 
215 	vmem_free(obj, size);
216 
217 	return (error);
218 }
219 
220 int
221 zfs_sa_set_xattr(znode_t *zp)
222 {
223 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
224 	dmu_tx_t *tx;
225 	char *obj;
226 	size_t size;
227 	int error;
228 
229 	ASSERT(RW_WRITE_HELD(&zp->z_xattr_lock));
230 	ASSERT(zp->z_xattr_cached);
231 	ASSERT(zp->z_is_sa);
232 
233 	error = nvlist_size(zp->z_xattr_cached, &size, NV_ENCODE_XDR);
234 	if ((error == 0) && (size > SA_ATTR_MAX_LEN))
235 		error = SET_ERROR(EFBIG);
236 	if (error)
237 		goto out;
238 
239 	obj = vmem_alloc(size, KM_SLEEP);
240 
241 	error = nvlist_pack(zp->z_xattr_cached, &obj, &size,
242 	    NV_ENCODE_XDR, KM_SLEEP);
243 	if (error)
244 		goto out_free;
245 
246 	tx = dmu_tx_create(zfsvfs->z_os);
247 	dmu_tx_hold_sa_create(tx, size);
248 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
249 
250 	error = dmu_tx_assign(tx, TXG_WAIT);
251 	if (error) {
252 		dmu_tx_abort(tx);
253 	} else {
254 		int count = 0;
255 		sa_bulk_attr_t bulk[2];
256 		uint64_t ctime[2];
257 
258 		zfs_tstamp_update_setup(zp, STATE_CHANGED, NULL, ctime);
259 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DXATTR(zfsvfs),
260 		    NULL, obj, size);
261 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
262 		    NULL, &ctime, 16);
263 		VERIFY0(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
264 
265 		dmu_tx_commit(tx);
266 	}
267 out_free:
268 	vmem_free(obj, size);
269 out:
270 	return (error);
271 }
272 
273 /*
274  * I'm not convinced we should do any of this upgrade.
275  * since the SA code can read both old/new znode formats
276  * with probably little to no performance difference.
277  *
278  * All new files will be created with the new format.
279  */
280 
281 void
282 zfs_sa_upgrade(sa_handle_t *hdl, dmu_tx_t *tx)
283 {
284 	dmu_buf_t *db = sa_get_db(hdl);
285 	znode_t *zp = sa_get_userdata(hdl);
286 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
287 	int count = 0;
288 	sa_bulk_attr_t *bulk, *sa_attrs;
289 	zfs_acl_locator_cb_t locate = { 0 };
290 	uint64_t uid, gid, mode, rdev, xattr, parent, tmp_gen;
291 	uint64_t crtime[2], mtime[2], ctime[2], atime[2];
292 	uint64_t links;
293 	zfs_acl_phys_t znode_acl;
294 	char scanstamp[AV_SCANSTAMP_SZ];
295 	boolean_t drop_lock = B_FALSE;
296 
297 	/*
298 	 * No upgrade if ACL isn't cached
299 	 * since we won't know which locks are held
300 	 * and ready the ACL would require special "locked"
301 	 * interfaces that would be messy
302 	 */
303 	if (zp->z_acl_cached == NULL || Z_ISLNK(ZTOTYPE(zp)))
304 		return;
305 
306 	/*
307 	 * If the z_lock is held and we aren't the owner
308 	 * the just return since we don't want to deadlock
309 	 * trying to update the status of z_is_sa.  This
310 	 * file can then be upgraded at a later time.
311 	 *
312 	 * Otherwise, we know we are doing the
313 	 * sa_update() that caused us to enter this function.
314 	 */
315 	if (MUTEX_NOT_HELD(&zp->z_lock)) {
316 		if (mutex_tryenter(&zp->z_lock) == 0)
317 			return;
318 		else
319 			drop_lock = B_TRUE;
320 	}
321 
322 	/* First do a bulk query of the attributes that aren't cached */
323 	bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
324 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
325 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
326 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
327 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
328 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
329 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
330 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_XATTR(zfsvfs), NULL, &xattr, 8);
331 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, &rdev, 8);
332 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
333 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
334 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &tmp_gen, 8);
335 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
336 	    &znode_acl, 88);
337 
338 	if (sa_bulk_lookup_locked(hdl, bulk, count) != 0)
339 		goto done;
340 
341 	if (dmu_objset_projectquota_enabled(hdl->sa_os) &&
342 	    !(zp->z_pflags & ZFS_PROJID)) {
343 		zp->z_pflags |= ZFS_PROJID;
344 		zp->z_projid = ZFS_DEFAULT_PROJID;
345 	}
346 
347 	/*
348 	 * While the order here doesn't matter its best to try and organize
349 	 * it is such a way to pick up an already existing layout number
350 	 */
351 	count = 0;
352 	sa_attrs = kmem_zalloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
353 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
354 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SIZE(zfsvfs), NULL,
355 	    &zp->z_size, 8);
356 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_GEN(zfsvfs),
357 	    NULL, &tmp_gen, 8);
358 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
359 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
360 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_PARENT(zfsvfs),
361 	    NULL, &parent, 8);
362 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_FLAGS(zfsvfs), NULL,
363 	    &zp->z_pflags, 8);
364 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_ATIME(zfsvfs), NULL,
365 	    &atime, 16);
366 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_MTIME(zfsvfs), NULL,
367 	    &mtime, 16);
368 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_CTIME(zfsvfs), NULL,
369 	    &ctime, 16);
370 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_CRTIME(zfsvfs), NULL,
371 	    &crtime, 16);
372 	links = ZTONLNK(zp);
373 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_LINKS(zfsvfs), NULL,
374 	    &links, 8);
375 	if (dmu_objset_projectquota_enabled(hdl->sa_os))
376 		SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_PROJID(zfsvfs), NULL,
377 		    &zp->z_projid, 8);
378 	if (Z_ISBLK(ZTOTYPE(zp)) || Z_ISCHR(ZTOTYPE(zp)))
379 		SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_RDEV(zfsvfs), NULL,
380 		    &rdev, 8);
381 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
382 	    &zp->z_acl_cached->z_acl_count, 8);
383 
384 	if (zp->z_acl_cached->z_version < ZFS_ACL_VERSION_FUID)
385 		zfs_acl_xform(zp, zp->z_acl_cached, CRED());
386 
387 	locate.cb_aclp = zp->z_acl_cached;
388 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_DACL_ACES(zfsvfs),
389 	    zfs_acl_data_locator, &locate, zp->z_acl_cached->z_acl_bytes);
390 
391 	if (xattr)
392 		SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_XATTR(zfsvfs),
393 		    NULL, &xattr, 8);
394 
395 	/* if scanstamp then add scanstamp */
396 
397 	if (zp->z_pflags & ZFS_BONUS_SCANSTAMP) {
398 		bcopy((caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
399 		    scanstamp, AV_SCANSTAMP_SZ);
400 		SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SCANSTAMP(zfsvfs),
401 		    NULL, scanstamp, AV_SCANSTAMP_SZ);
402 		zp->z_pflags &= ~ZFS_BONUS_SCANSTAMP;
403 	}
404 
405 	VERIFY(dmu_set_bonustype(db, DMU_OT_SA, tx) == 0);
406 	VERIFY(sa_replace_all_by_template_locked(hdl, sa_attrs,
407 	    count, tx) == 0);
408 	if (znode_acl.z_acl_extern_obj)
409 		VERIFY(0 == dmu_object_free(zfsvfs->z_os,
410 		    znode_acl.z_acl_extern_obj, tx));
411 
412 	zp->z_is_sa = B_TRUE;
413 	kmem_free(sa_attrs, sizeof (sa_bulk_attr_t) * ZPL_END);
414 done:
415 	kmem_free(bulk, sizeof (sa_bulk_attr_t) * ZPL_END);
416 	if (drop_lock)
417 		mutex_exit(&zp->z_lock);
418 }
419 
420 void
421 zfs_sa_upgrade_txholds(dmu_tx_t *tx, znode_t *zp)
422 {
423 	if (!ZTOZSB(zp)->z_use_sa || zp->z_is_sa)
424 		return;
425 
426 
427 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
428 
429 	if (zfs_external_acl(zp)) {
430 		dmu_tx_hold_free(tx, zfs_external_acl(zp), 0,
431 		    DMU_OBJECT_END);
432 	}
433 }
434 
435 EXPORT_SYMBOL(zfs_attr_table);
436 EXPORT_SYMBOL(zfs_sa_readlink);
437 EXPORT_SYMBOL(zfs_sa_symlink);
438 EXPORT_SYMBOL(zfs_sa_get_scanstamp);
439 EXPORT_SYMBOL(zfs_sa_set_scanstamp);
440 EXPORT_SYMBOL(zfs_sa_get_xattr);
441 EXPORT_SYMBOL(zfs_sa_set_xattr);
442 EXPORT_SYMBOL(zfs_sa_upgrade);
443 EXPORT_SYMBOL(zfs_sa_upgrade_txholds);
444 
445 #endif
446