xref: /titanic_52/usr/src/uts/common/fs/zfs/zfs_sa.c (revision e4f5a11d4a234623168c1558fcdf4341e11769e1)
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/types.h>
26 #include <sys/param.h>
27 #include <sys/vnode.h>
28 #include <sys/sa.h>
29 #include <sys/zfs_acl.h>
30 #include <sys/zfs_sa.h>
31 
32 /*
33  * ZPL attribute registration table.
34  * Order of attributes doesn't matter
35  * a unique value will be assigned for each
36  * attribute that is file system specific
37  *
38  * This is just the set of ZPL attributes that this
39  * version of ZFS deals with natively.  The file system
40  * could have other attributes stored in files, but they will be
41  * ignored.  The SA framework will preserve them, just that
42  * this version of ZFS won't change or delete them.
43  */
44 
45 sa_attr_reg_t zfs_attr_table[ZPL_END+1] = {
46 	{"ZPL_ATIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 0},
47 	{"ZPL_MTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 1},
48 	{"ZPL_CTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 2},
49 	{"ZPL_CRTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 3},
50 	{"ZPL_GEN", sizeof (uint64_t), SA_UINT64_ARRAY, 4},
51 	{"ZPL_MODE", sizeof (uint64_t), SA_UINT64_ARRAY, 5},
52 	{"ZPL_SIZE", sizeof (uint64_t), SA_UINT64_ARRAY, 6},
53 	{"ZPL_PARENT", sizeof (uint64_t), SA_UINT64_ARRAY, 7},
54 	{"ZPL_LINKS", sizeof (uint64_t), SA_UINT64_ARRAY, 8},
55 	{"ZPL_XATTR", sizeof (uint64_t), SA_UINT64_ARRAY, 9},
56 	{"ZPL_RDEV", sizeof (uint64_t), SA_UINT64_ARRAY, 10},
57 	{"ZPL_FLAGS", sizeof (uint64_t), SA_UINT64_ARRAY, 11},
58 	{"ZPL_UID", sizeof (uint64_t), SA_UINT64_ARRAY, 12},
59 	{"ZPL_GID", sizeof (uint64_t), SA_UINT64_ARRAY, 13},
60 	{"ZPL_PAD", sizeof (uint64_t) * 4, SA_UINT64_ARRAY, 14},
61 	{"ZPL_ZNODE_ACL", 88, SA_UINT8_ARRAY, 15},
62 	{"ZPL_DACL_COUNT", sizeof (uint64_t), SA_UINT64_ARRAY, 0},
63 	{"ZPL_SYMLINK", 0, SA_UINT8_ARRAY, 0},
64 	{"ZPL_SCANSTAMP", 32, SA_UINT8_ARRAY, 0},
65 	{"ZPL_DACL_ACES", 0, SA_ACL, 0},
66 	{NULL, 0, 0, 0}
67 };
68 
69 #ifdef _KERNEL
70 
71 int
72 zfs_sa_readlink(znode_t *zp, uio_t *uio)
73 {
74 	dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
75 	size_t bufsz;
76 	int error;
77 
78 	bufsz = zp->z_size;
79 	if (bufsz + ZFS_OLD_ZNODE_PHYS_SIZE <= db->db_size) {
80 		error = uiomove((caddr_t)db->db_data +
81 		    ZFS_OLD_ZNODE_PHYS_SIZE,
82 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
83 	} else {
84 		dmu_buf_t *dbp;
85 		if ((error = dmu_buf_hold(zp->z_zfsvfs->z_os, zp->z_id,
86 		    0, FTAG, &dbp, DMU_READ_NO_PREFETCH)) == 0) {
87 			error = uiomove(dbp->db_data,
88 			    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
89 			dmu_buf_rele(dbp, FTAG);
90 		}
91 	}
92 	return (error);
93 }
94 
95 void
96 zfs_sa_symlink(znode_t *zp, char *link, int len, dmu_tx_t *tx)
97 {
98 	dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
99 
100 	if (ZFS_OLD_ZNODE_PHYS_SIZE + len <= dmu_bonus_max()) {
101 		VERIFY(dmu_set_bonus(db,
102 		    len + ZFS_OLD_ZNODE_PHYS_SIZE, tx) == 0);
103 		if (len) {
104 			bcopy(link, (caddr_t)db->db_data +
105 			    ZFS_OLD_ZNODE_PHYS_SIZE, len);
106 		}
107 	} else {
108 		dmu_buf_t *dbp;
109 
110 		zfs_grow_blocksize(zp, len, tx);
111 		VERIFY(0 == dmu_buf_hold(zp->z_zfsvfs->z_os,
112 		    zp->z_id, 0, FTAG, &dbp, DMU_READ_NO_PREFETCH));
113 
114 		dmu_buf_will_dirty(dbp, tx);
115 
116 		ASSERT3U(len, <=, dbp->db_size);
117 		bcopy(link, dbp->db_data, len);
118 		dmu_buf_rele(dbp, FTAG);
119 	}
120 }
121 
122 void
123 zfs_sa_get_scanstamp(znode_t *zp, xvattr_t *xvap)
124 {
125 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
126 	xoptattr_t *xoap;
127 
128 	ASSERT(MUTEX_HELD(&zp->z_lock));
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 	ASSERT(MUTEX_HELD(&zp->z_lock));
163 	VERIFY((xoap = xva_getxoptattr(xvap)) != NULL);
164 	if (zp->z_is_sa)
165 		VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SCANSTAMP(zfsvfs),
166 		    &xoap->xoa_av_scanstamp,
167 		    sizeof (xoap->xoa_av_scanstamp), tx));
168 	else {
169 		dmu_object_info_t doi;
170 		dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
171 		int len;
172 
173 		sa_object_info(zp->z_sa_hdl, &doi);
174 		len = sizeof (xoap->xoa_av_scanstamp) +
175 		    ZFS_OLD_ZNODE_PHYS_SIZE;
176 		if (len > doi.doi_bonus_size)
177 			VERIFY(dmu_set_bonus(db, len, tx) == 0);
178 		(void) memcpy((caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
179 		    xoap->xoa_av_scanstamp, sizeof (xoap->xoa_av_scanstamp));
180 
181 		zp->z_pflags |= ZFS_BONUS_SCANSTAMP;
182 		VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
183 		    &zp->z_pflags, sizeof (uint64_t), tx));
184 	}
185 }
186 
187 /*
188  * I'm not convinced we should do any of this upgrade.
189  * since the SA code can read both old/new znode formats
190  * with probably little to know performance difference.
191  *
192  * All new files will be created with the new format.
193  */
194 
195 void
196 zfs_sa_upgrade(sa_handle_t *hdl, dmu_tx_t *tx)
197 {
198 	dmu_buf_t *db = sa_get_db(hdl);
199 	znode_t *zp = sa_get_userdata(hdl);
200 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
201 	sa_bulk_attr_t bulk[20];
202 	int count = 0;
203 	sa_bulk_attr_t sa_attrs[20] = { 0 };
204 	zfs_acl_locator_cb_t locate = { 0 };
205 	uint64_t uid, gid, mode, rdev, xattr, parent;
206 	uint64_t crtime[2], mtime[2], ctime[2];
207 	zfs_acl_phys_t znode_acl;
208 	char scanstamp[AV_SCANSTAMP_SZ];
209 	boolean_t drop_lock = B_FALSE;
210 
211 	/*
212 	 * No upgrade if ACL isn't cached
213 	 * since we won't know which locks are held
214 	 * and ready the ACL would require special "locked"
215 	 * interfaces that would be messy
216 	 */
217 	if (zp->z_acl_cached == NULL || ZTOV(zp)->v_type == VLNK)
218 		return;
219 
220 	/*
221 	 * If the z_lock is held and we aren't the owner
222 	 * the just return since we don't want to deadlock
223 	 * trying to update the status of z_is_sa.  This
224 	 * file can then be upgraded at a later time.
225 	 *
226 	 * Otherwise, we know we are doing the
227 	 * sa_update() that caused us to enter this function.
228 	 */
229 	if (mutex_owner(&zp->z_lock) != curthread) {
230 		if (mutex_tryenter(&zp->z_lock) == 0)
231 			return;
232 		else
233 			drop_lock = B_TRUE;
234 	}
235 
236 	/* First do a bulk query of the attributes that aren't cached */
237 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
238 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
239 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
240 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
241 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
242 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_XATTR(zfsvfs), NULL, &xattr, 8);
243 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, &rdev, 8);
244 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
245 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
246 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
247 	    &znode_acl, 88);
248 
249 	if (sa_bulk_lookup_locked(hdl, bulk, count) != 0)
250 		goto done;
251 
252 
253 	/*
254 	 * While the order here doesn't matter its best to try and organize
255 	 * it is such a way to pick up an already existing layout number
256 	 */
257 	count = 0;
258 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
259 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SIZE(zfsvfs), NULL,
260 	    &zp->z_size, 8);
261 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_GEN(zfsvfs),
262 	    NULL, &zp->z_gen, 8);
263 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
264 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
265 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_PARENT(zfsvfs),
266 	    NULL, &parent, 8);
267 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_FLAGS(zfsvfs), NULL,
268 	    &zp->z_pflags, 8);
269 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_ATIME(zfsvfs), NULL,
270 	    zp->z_atime, 16);
271 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_MTIME(zfsvfs), NULL,
272 	    &mtime, 16);
273 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_CTIME(zfsvfs), NULL,
274 	    &ctime, 16);
275 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_CRTIME(zfsvfs), NULL,
276 	    &crtime, 16);
277 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_LINKS(zfsvfs), NULL,
278 	    &zp->z_links, 8);
279 	if (zp->z_vnode->v_type == VBLK || zp->z_vnode->v_type == VCHR)
280 		SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_RDEV(zfsvfs), NULL,
281 		    &rdev, 8);
282 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
283 	    &zp->z_acl_cached->z_acl_count, 8);
284 
285 	if (zp->z_acl_cached->z_version < ZFS_ACL_VERSION_FUID)
286 		zfs_acl_xform(zp, zp->z_acl_cached, CRED());
287 
288 	locate.cb_aclp = zp->z_acl_cached;
289 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_DACL_ACES(zfsvfs),
290 	    zfs_acl_data_locator, &locate, zp->z_acl_cached->z_acl_bytes);
291 
292 	if (xattr)
293 		SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_XATTR(zfsvfs),
294 		    NULL, &xattr, 8);
295 
296 	/* if scanstamp then add scanstamp */
297 
298 	if (zp->z_pflags & ZFS_BONUS_SCANSTAMP) {
299 		bcopy((caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
300 		    scanstamp, AV_SCANSTAMP_SZ);
301 		SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SCANSTAMP(zfsvfs),
302 		    NULL, scanstamp, AV_SCANSTAMP_SZ);
303 		zp->z_pflags &= ~ZFS_BONUS_SCANSTAMP;
304 	}
305 
306 	VERIFY(dmu_set_bonustype(db, DMU_OT_SA, tx) == 0);
307 	VERIFY(sa_replace_all_by_template_locked(hdl, sa_attrs,
308 	    count, tx) == 0);
309 	if (znode_acl.z_acl_extern_obj)
310 		VERIFY(0 == dmu_object_free(zfsvfs->z_os,
311 		    znode_acl.z_acl_extern_obj, tx));
312 
313 	zp->z_is_sa = B_TRUE;
314 done:
315 	if (drop_lock)
316 		mutex_exit(&zp->z_lock);
317 }
318 
319 void
320 zfs_sa_upgrade_txholds(dmu_tx_t *tx, znode_t *zp)
321 {
322 	if (!zp->z_zfsvfs->z_use_sa || zp->z_is_sa)
323 		return;
324 
325 
326 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
327 
328 	if (zfs_external_acl(zp)) {
329 		dmu_tx_hold_free(tx, zfs_external_acl(zp), 0,
330 		    DMU_OBJECT_END);
331 	}
332 }
333 
334 #endif
335