xref: /illumos-gate/usr/src/uts/common/fs/zfs/zfs_sa.c (revision 0250c53ad267726f2438e3c6556199a0bbf588a2)
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_PROJID", sizeof (uint64_t), SA_UINT64_ARRAY, 0},
68 	{NULL, 0, 0, 0}
69 };
70 
71 #ifdef _KERNEL
72 
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->uio_resid), UIO_READ, uio);
85 	} else {
86 		dmu_buf_t *dbp;
87 		if ((error = dmu_buf_hold(zp->z_zfsvfs->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->uio_resid), 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 		VERIFY(0 == dmu_buf_hold(zp->z_zfsvfs->z_os,
113 		    zp->z_id, 0, FTAG, &dbp, 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 = zp->z_zfsvfs;
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 = zp->z_zfsvfs;
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 /*
189  * I'm not convinced we should do any of this upgrade.
190  * since the SA code can read both old/new znode formats
191  * with probably little to no performance difference.
192  *
193  * All new files will be created with the new format.
194  */
195 
196 void
197 zfs_sa_upgrade(sa_handle_t *hdl, dmu_tx_t *tx)
198 {
199 	dmu_buf_t *db = sa_get_db(hdl);
200 	znode_t *zp = sa_get_userdata(hdl);
201 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
202 	sa_bulk_attr_t bulk[22];
203 	int count = 0;
204 	sa_bulk_attr_t sa_attrs[20] = { 0 };
205 	zfs_acl_locator_cb_t locate = { 0 };
206 	uint64_t uid, gid, mode, rdev, xattr, parent;
207 	uint64_t crtime[2], mtime[2], ctime[2];
208 	zfs_acl_phys_t znode_acl;
209 	char scanstamp[AV_SCANSTAMP_SZ];
210 	boolean_t drop_lock = B_FALSE;
211 
212 	/*
213 	 * No upgrade if ACL isn't cached
214 	 * since we won't know which locks are held
215 	 * and ready the ACL would require special "locked"
216 	 * interfaces that would be messy
217 	 */
218 	if (zp->z_acl_cached == NULL || ZTOV(zp)->v_type == VLNK)
219 		return;
220 
221 	/*
222 	 * If the z_lock is held and we aren't the owner
223 	 * the just return since we don't want to deadlock
224 	 * trying to update the status of z_is_sa.  This
225 	 * file can then be upgraded at a later time.
226 	 *
227 	 * Otherwise, we know we are doing the
228 	 * sa_update() that caused us to enter this function.
229 	 */
230 	if (mutex_owner(&zp->z_lock) != curthread) {
231 		if (mutex_tryenter(&zp->z_lock) == 0)
232 			return;
233 		else
234 			drop_lock = B_TRUE;
235 	}
236 
237 	/* First do a bulk query of the attributes that aren't cached */
238 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
239 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
240 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
241 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
242 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
243 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_XATTR(zfsvfs), NULL, &xattr, 8);
244 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, &rdev, 8);
245 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
246 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
247 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
248 	    &znode_acl, 88);
249 
250 	if (sa_bulk_lookup_locked(hdl, bulk, count) != 0)
251 		goto done;
252 
253 	if (dmu_objset_projectquota_enabled(hdl->sa_os) &&
254 	    !(zp->z_pflags & ZFS_PROJID)) {
255 		zp->z_pflags |= ZFS_PROJID;
256 		zp->z_projid = ZFS_DEFAULT_PROJID;
257 	}
258 
259 	/*
260 	 * While the order here doesn't matter its best to try and organize
261 	 * it is such a way to pick up an already existing layout number
262 	 */
263 	count = 0;
264 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
265 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SIZE(zfsvfs), NULL,
266 	    &zp->z_size, 8);
267 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_GEN(zfsvfs),
268 	    NULL, &zp->z_gen, 8);
269 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
270 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
271 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_PARENT(zfsvfs),
272 	    NULL, &parent, 8);
273 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_FLAGS(zfsvfs), NULL,
274 	    &zp->z_pflags, 8);
275 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_ATIME(zfsvfs), NULL,
276 	    zp->z_atime, 16);
277 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_MTIME(zfsvfs), NULL,
278 	    &mtime, 16);
279 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_CTIME(zfsvfs), NULL,
280 	    &ctime, 16);
281 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_CRTIME(zfsvfs), NULL,
282 	    &crtime, 16);
283 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_LINKS(zfsvfs), NULL,
284 	    &zp->z_links, 8);
285 	if (dmu_objset_projectquota_enabled(hdl->sa_os))
286 		SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_PROJID(zfsvfs), NULL,
287 		    &zp->z_projid, 8);
288 	if (zp->z_vnode->v_type == VBLK || zp->z_vnode->v_type == VCHR)
289 		SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_RDEV(zfsvfs), NULL,
290 		    &rdev, 8);
291 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
292 	    &zp->z_acl_cached->z_acl_count, 8);
293 
294 	if (zp->z_acl_cached->z_version < ZFS_ACL_VERSION_FUID)
295 		zfs_acl_xform(zp, zp->z_acl_cached, CRED());
296 
297 	locate.cb_aclp = zp->z_acl_cached;
298 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_DACL_ACES(zfsvfs),
299 	    zfs_acl_data_locator, &locate, zp->z_acl_cached->z_acl_bytes);
300 
301 	if (xattr)
302 		SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_XATTR(zfsvfs),
303 		    NULL, &xattr, 8);
304 
305 	/* if scanstamp then add scanstamp */
306 
307 	if (zp->z_pflags & ZFS_BONUS_SCANSTAMP) {
308 		bcopy((caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
309 		    scanstamp, AV_SCANSTAMP_SZ);
310 		SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SCANSTAMP(zfsvfs),
311 		    NULL, scanstamp, AV_SCANSTAMP_SZ);
312 		zp->z_pflags &= ~ZFS_BONUS_SCANSTAMP;
313 	}
314 
315 	VERIFY(dmu_set_bonustype(db, DMU_OT_SA, tx) == 0);
316 	VERIFY(sa_replace_all_by_template_locked(hdl, sa_attrs,
317 	    count, tx) == 0);
318 	if (znode_acl.z_acl_extern_obj)
319 		VERIFY(0 == dmu_object_free(zfsvfs->z_os,
320 		    znode_acl.z_acl_extern_obj, tx));
321 
322 	zp->z_is_sa = B_TRUE;
323 done:
324 	if (drop_lock)
325 		mutex_exit(&zp->z_lock);
326 }
327 
328 void
329 zfs_sa_upgrade_txholds(dmu_tx_t *tx, znode_t *zp)
330 {
331 	if (!zp->z_zfsvfs->z_use_sa || zp->z_is_sa)
332 		return;
333 
334 
335 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
336 
337 	if (zfs_external_acl(zp)) {
338 		dmu_tx_hold_free(tx, zfs_external_acl(zp), 0,
339 		    DMU_OBJECT_END);
340 	}
341 }
342 
343 #endif
344