xref: /titanic_51/usr/src/uts/common/fs/zfs/zfs_sa.c (revision 8f798d3afbe38d59cc0a708261dbb729f1b6b209)
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 2010 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <sys/types.h>
27 #include <sys/param.h>
28 #include <sys/vnode.h>
29 #include <sys/sa.h>
30 #include <sys/zfs_acl.h>
31 #include <sys/zfs_sa.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 	{NULL, 0, 0, 0}
68 };
69 
70 #ifdef _KERNEL
71 
72 int
73 zfs_sa_readlink(znode_t *zp, uio_t *uio)
74 {
75 	dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
76 	size_t bufsz;
77 	int error;
78 
79 	bufsz = zp->z_size;
80 	if (bufsz + ZFS_OLD_ZNODE_PHYS_SIZE <= db->db_size) {
81 		error = uiomove((caddr_t)db->db_data +
82 		    ZFS_OLD_ZNODE_PHYS_SIZE,
83 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
84 	} else {
85 		dmu_buf_t *dbp;
86 		if ((error = dmu_buf_hold(zp->z_zfsvfs->z_os, zp->z_id,
87 		    0, FTAG, &dbp)) == 0) {
88 			error = uiomove(dbp->db_data,
89 			    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
90 			dmu_buf_rele(dbp, FTAG);
91 		}
92 	}
93 	return (error);
94 }
95 
96 void
97 zfs_sa_symlink(znode_t *zp, char *link, int len, dmu_tx_t *tx)
98 {
99 	dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
100 
101 	if (ZFS_OLD_ZNODE_PHYS_SIZE + len <= dmu_bonus_max()) {
102 		VERIFY(dmu_set_bonus(db,
103 		    len + ZFS_OLD_ZNODE_PHYS_SIZE, tx) == 0);
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));
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 	VERIFY((xoap = xva_getxoptattr(xvap)) != NULL);
130 	if (zp->z_is_sa) {
131 		if (sa_lookup(zp->z_sa_hdl, SA_ZPL_SCANSTAMP(zfsvfs),
132 		    &xoap->xoa_av_scanstamp,
133 		    sizeof (xoap->xoa_av_scanstamp)) != 0)
134 			return;
135 	} else {
136 		dmu_object_info_t doi;
137 		dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
138 		int len;
139 
140 		if (!(zp->z_pflags & ZFS_BONUS_SCANSTAMP))
141 			return;
142 
143 		sa_object_info(zp->z_sa_hdl, &doi);
144 		len = sizeof (xoap->xoa_av_scanstamp) +
145 		    ZFS_OLD_ZNODE_PHYS_SIZE;
146 
147 		if (len <= doi.doi_bonus_size) {
148 			(void) memcpy(xoap->xoa_av_scanstamp,
149 			    (caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
150 			    sizeof (xoap->xoa_av_scanstamp));
151 		}
152 	}
153 	XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
154 }
155 
156 void
157 zfs_sa_set_scanstamp(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
158 {
159 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
160 	xoptattr_t *xoap;
161 
162 	VERIFY((xoap = xva_getxoptattr(xvap)) != NULL);
163 	if (zp->z_is_sa)
164 		VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SCANSTAMP(zfsvfs),
165 		    &xoap->xoa_av_scanstamp,
166 		    sizeof (xoap->xoa_av_scanstamp), tx));
167 	else {
168 		dmu_object_info_t doi;
169 		dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
170 		int len;
171 
172 		sa_object_info(zp->z_sa_hdl, &doi);
173 		len = sizeof (xoap->xoa_av_scanstamp) +
174 		    ZFS_OLD_ZNODE_PHYS_SIZE;
175 		if (len > doi.doi_bonus_size)
176 			VERIFY(dmu_set_bonus(db, len, tx) == 0);
177 		(void) memcpy((caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
178 		    xoap->xoa_av_scanstamp, sizeof (xoap->xoa_av_scanstamp));
179 
180 		zp->z_pflags |= ZFS_BONUS_SCANSTAMP;
181 		VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
182 		    &zp->z_pflags, sizeof (uint64_t), tx));
183 	}
184 }
185 
186 /*
187  * I'm not convinced we should do any of this upgrade.
188  * since the SA code can read both old/new znode formats
189  * with probably little to know performance difference.
190  *
191  * All new files will be created with the new format.
192  */
193 
194 void
195 zfs_sa_upgrade(sa_handle_t *hdl, dmu_tx_t *tx)
196 {
197 	dmu_buf_t *db = sa_get_db(hdl);
198 	znode_t *zp = sa_get_userdata(hdl);
199 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
200 	sa_bulk_attr_t bulk[20];
201 	int count = 0;
202 	sa_bulk_attr_t sa_attrs[20] = { 0 };
203 	zfs_acl_locator_cb_t locate = { 0 };
204 	uint64_t uid, gid, mode, rdev, xattr, parent;
205 	uint64_t crtime[2], mtime[2], ctime[2];
206 	zfs_acl_phys_t znode_acl;
207 	char *slink = NULL;
208 	char scanstamp[AV_SCANSTAMP_SZ];
209 
210 	/*
211 	 * No upgrade if ACL isn't cached
212 	 * since we won't know which locks are held
213 	 * and ready the ACL would require special "locked"
214 	 * interfaces that would be messy
215 	 */
216 	if (zp->z_acl_cached == NULL)
217 		return;
218 
219 	/* First do a bulk query of the attributes that aren't cached */
220 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
221 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
222 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
223 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
224 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
225 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_XATTR(zfsvfs), NULL, &xattr, 8);
226 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, &rdev, 8);
227 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
228 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
229 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
230 	    &znode_acl, 88);
231 
232 	if (sa_bulk_lookup_locked(hdl, bulk, count) != 0)
233 		return;
234 
235 
236 	/*
237 	 * While the order here doesn't matter its best to try and organize
238 	 * it is such a way to pick up an already existing layout number
239 	 */
240 	count = 0;
241 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
242 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SIZE(zfsvfs), NULL,
243 	    &zp->z_size, 8);
244 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_GEN(zfsvfs),
245 	    NULL, &zp->z_gen, 8);
246 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
247 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
248 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_PARENT(zfsvfs),
249 	    NULL, &parent, 8);
250 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_FLAGS(zfsvfs), NULL,
251 	    &zp->z_pflags, 8);
252 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_ATIME(zfsvfs), NULL,
253 	    zp->z_atime, 16);
254 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_MTIME(zfsvfs), NULL,
255 	    &mtime, 16);
256 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_CTIME(zfsvfs), NULL,
257 	    &ctime, 16);
258 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_CRTIME(zfsvfs), NULL,
259 	    &crtime, 16);
260 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_LINKS(zfsvfs), NULL,
261 	    &zp->z_links, 8);
262 	if (zp->z_vnode->v_type == VBLK || zp->z_vnode->v_type == VCHR)
263 		SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_RDEV(zfsvfs), NULL,
264 		    &rdev, 8);
265 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
266 	    &zp->z_acl_cached->z_acl_count, 8);
267 
268 	if (zp->z_acl_cached->z_version < ZFS_ACL_VERSION_FUID)
269 		zfs_acl_xform(zp, zp->z_acl_cached, CRED());
270 
271 	locate.cb_aclp = zp->z_acl_cached;
272 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_DACL_ACES(zfsvfs),
273 	    zfs_acl_data_locator, &locate, zp->z_acl_cached->z_acl_bytes);
274 	if (xattr)
275 		SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_RDEV(zfsvfs),
276 		    NULL, &rdev, 8);
277 
278 	/*
279 	 * is it a symlink ?
280 	 *
281 	 * this will probably never be exercised since we won't
282 	 * have the cached ACL.
283 	 */
284 	if (ZTOV(zp)->v_type == VLNK) {
285 		slink = kmem_zalloc(zp->z_size + 1, KM_SLEEP);
286 		if (zp->z_size + ZFS_OLD_ZNODE_PHYS_SIZE)
287 			bcopy((caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
288 			    slink, zp->z_size);
289 		else {
290 			dmu_buf_t *dbp;
291 			if (dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0,
292 			    FTAG, &dbp))
293 				return;
294 			bcopy(dbp->db_data, slink, zp->z_size);
295 			dmu_buf_rele(dbp, FTAG);
296 		}
297 		SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SYMLINK(zfsvfs),
298 		    NULL, slink, zp->z_size);
299 	}
300 
301 	/* if scanstamp then add scanstamp */
302 
303 	if (zp->z_pflags & ZFS_BONUS_SCANSTAMP) {
304 		bcopy((caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
305 		    scanstamp, AV_SCANSTAMP_SZ);
306 		SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SCANSTAMP(zfsvfs),
307 		    NULL, scanstamp, AV_SCANSTAMP_SZ);
308 		zp->z_pflags &= ~ZFS_BONUS_SCANSTAMP;
309 	}
310 
311 	VERIFY(dmu_set_bonustype(db, DMU_OT_SA, tx) == 0);
312 	VERIFY(sa_replace_all_by_template_locked(hdl, sa_attrs,
313 	    count, tx) == 0);
314 	if (znode_acl.z_acl_extern_obj)
315 		VERIFY(0 == dmu_object_free(zfsvfs->z_os,
316 		    znode_acl.z_acl_extern_obj, tx));
317 
318 	zp->z_is_sa = B_TRUE;
319 	if (slink)
320 		kmem_free(slink, zp->z_size + 1);
321 
322 }
323 
324 void
325 zfs_sa_upgrade_txholds(dmu_tx_t *tx, znode_t *zp)
326 {
327 	if (!zp->z_zfsvfs->z_use_sa || zp->z_is_sa)
328 		return;
329 
330 	ASSERT(!zp->z_is_sa);
331 
332 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
333 
334 	if (ZFS_EXTERNAL_ACL(zp)) {
335 		dmu_tx_hold_free(tx, ZFS_EXTERNAL_ACL(zp), 0,
336 		    DMU_OBJECT_END);
337 	}
338 }
339 
340 #endif
341