xref: /titanic_41/usr/src/uts/common/fs/zfs/zfs_sa.c (revision 0521c5ebd0c906e1574175b64bd238405444c004)
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 
31 /*
32  * ZPL attribute registration table.
33  * Order of attributes doesn't matter
34  * a unique value will be assigned for each
35  * attribute that is file system specific
36  *
37  * This is just the set of ZPL attributes that this
38  * version of ZFS deals with natively.  The file system
39  * could have other attributes stored in files, but they will be
40  * ignored.  The SA framework will preserve them, just that
41  * this version of ZFS won't change or delete them.
42  */
43 
44 sa_attr_reg_t zfs_attr_table[ZPL_END+1] = {
45 	{"ZPL_ATIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 0},
46 	{"ZPL_MTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 1},
47 	{"ZPL_CTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 2},
48 	{"ZPL_CRTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 3},
49 	{"ZPL_GEN", sizeof (uint64_t), SA_UINT64_ARRAY, 4},
50 	{"ZPL_MODE", sizeof (uint64_t), SA_UINT64_ARRAY, 5},
51 	{"ZPL_SIZE", sizeof (uint64_t), SA_UINT64_ARRAY, 6},
52 	{"ZPL_PARENT", sizeof (uint64_t), SA_UINT64_ARRAY, 7},
53 	{"ZPL_LINKS", sizeof (uint64_t), SA_UINT64_ARRAY, 8},
54 	{"ZPL_XATTR", sizeof (uint64_t), SA_UINT64_ARRAY, 9},
55 	{"ZPL_RDEV", sizeof (uint64_t), SA_UINT64_ARRAY, 10},
56 	{"ZPL_FLAGS", sizeof (uint64_t), SA_UINT64_ARRAY, 11},
57 	{"ZPL_UID", sizeof (uint64_t), SA_UINT64_ARRAY, 12},
58 	{"ZPL_GID", sizeof (uint64_t), SA_UINT64_ARRAY, 13},
59 	{"ZPL_PAD", sizeof (uint64_t) * 4, SA_UINT64_ARRAY, 14},
60 	{"ZPL_ZNODE_ACL", 88, SA_UINT8_ARRAY, 15},
61 	{"ZPL_DACL_COUNT", sizeof (uint64_t), SA_UINT64_ARRAY, 0},
62 	{"ZPL_SYMLINK", 0, SA_UINT8_ARRAY, 0},
63 	{"ZPL_SCANSTAMP", 32, SA_UINT8_ARRAY, 0},
64 	{"ZPL_DACL_ACES", 0, SA_ACL, 0},
65 	{NULL, 0, 0, 0}
66 };
67 
68 #ifdef _KERNEL
69 
70 int
zfs_sa_readlink(znode_t * zp,uio_t * uio)71 zfs_sa_readlink(znode_t *zp, uio_t *uio)
72 {
73 	dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
74 	size_t bufsz;
75 	int error;
76 
77 	bufsz = zp->z_size;
78 	if (bufsz + ZFS_OLD_ZNODE_PHYS_SIZE <= db->db_size) {
79 		error = uiomove((caddr_t)db->db_data +
80 		    ZFS_OLD_ZNODE_PHYS_SIZE,
81 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
82 	} else {
83 		dmu_buf_t *dbp;
84 		if ((error = dmu_buf_hold(zp->z_zfsvfs->z_os, zp->z_id,
85 		    0, FTAG, &dbp, DMU_READ_NO_PREFETCH)) == 0) {
86 			error = uiomove(dbp->db_data,
87 			    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
88 			dmu_buf_rele(dbp, FTAG);
89 		}
90 	}
91 	return (error);
92 }
93 
94 void
zfs_sa_symlink(znode_t * zp,char * link,int len,dmu_tx_t * tx)95 zfs_sa_symlink(znode_t *zp, char *link, int len, dmu_tx_t *tx)
96 {
97 	dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
98 
99 	if (ZFS_OLD_ZNODE_PHYS_SIZE + len <= dmu_bonus_max()) {
100 		VERIFY(dmu_set_bonus(db,
101 		    len + ZFS_OLD_ZNODE_PHYS_SIZE, tx) == 0);
102 		if (len) {
103 			bcopy(link, (caddr_t)db->db_data +
104 			    ZFS_OLD_ZNODE_PHYS_SIZE, len);
105 		}
106 	} else {
107 		dmu_buf_t *dbp;
108 
109 		zfs_grow_blocksize(zp, len, tx);
110 		VERIFY(0 == dmu_buf_hold(zp->z_zfsvfs->z_os,
111 		    zp->z_id, 0, FTAG, &dbp, DMU_READ_NO_PREFETCH));
112 
113 		dmu_buf_will_dirty(dbp, tx);
114 
115 		ASSERT3U(len, <=, dbp->db_size);
116 		bcopy(link, dbp->db_data, len);
117 		dmu_buf_rele(dbp, FTAG);
118 	}
119 }
120 
121 void
zfs_sa_get_scanstamp(znode_t * zp,xvattr_t * xvap)122 zfs_sa_get_scanstamp(znode_t *zp, xvattr_t *xvap)
123 {
124 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
125 	xoptattr_t *xoap;
126 
127 	ASSERT(MUTEX_HELD(&zp->z_lock));
128 	VERIFY((xoap = xva_getxoptattr(xvap)) != NULL);
129 	if (zp->z_is_sa) {
130 		if (sa_lookup(zp->z_sa_hdl, SA_ZPL_SCANSTAMP(zfsvfs),
131 		    &xoap->xoa_av_scanstamp,
132 		    sizeof (xoap->xoa_av_scanstamp)) != 0)
133 			return;
134 	} else {
135 		dmu_object_info_t doi;
136 		dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
137 		int len;
138 
139 		if (!(zp->z_pflags & ZFS_BONUS_SCANSTAMP))
140 			return;
141 
142 		sa_object_info(zp->z_sa_hdl, &doi);
143 		len = sizeof (xoap->xoa_av_scanstamp) +
144 		    ZFS_OLD_ZNODE_PHYS_SIZE;
145 
146 		if (len <= doi.doi_bonus_size) {
147 			(void) memcpy(xoap->xoa_av_scanstamp,
148 			    (caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
149 			    sizeof (xoap->xoa_av_scanstamp));
150 		}
151 	}
152 	XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
153 }
154 
155 void
zfs_sa_set_scanstamp(znode_t * zp,xvattr_t * xvap,dmu_tx_t * tx)156 zfs_sa_set_scanstamp(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
157 {
158 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
159 	xoptattr_t *xoap;
160 
161 	ASSERT(MUTEX_HELD(&zp->z_lock));
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 no performance difference.
190  *
191  * All new files will be created with the new format.
192  */
193 
194 void
zfs_sa_upgrade(sa_handle_t * hdl,dmu_tx_t * tx)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 scanstamp[AV_SCANSTAMP_SZ];
208 	boolean_t drop_lock = B_FALSE;
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 || ZTOV(zp)->v_type == VLNK)
217 		return;
218 
219 	/*
220 	 * If the z_lock is held and we aren't the owner
221 	 * the just return since we don't want to deadlock
222 	 * trying to update the status of z_is_sa.  This
223 	 * file can then be upgraded at a later time.
224 	 *
225 	 * Otherwise, we know we are doing the
226 	 * sa_update() that caused us to enter this function.
227 	 */
228 	if (mutex_owner(&zp->z_lock) != curthread) {
229 		if (mutex_tryenter(&zp->z_lock) == 0)
230 			return;
231 		else
232 			drop_lock = B_TRUE;
233 	}
234 
235 	/* First do a bulk query of the attributes that aren't cached */
236 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
237 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
238 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
239 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
240 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
241 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_XATTR(zfsvfs), NULL, &xattr, 8);
242 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, &rdev, 8);
243 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
244 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
245 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
246 	    &znode_acl, 88);
247 
248 	if (sa_bulk_lookup_locked(hdl, bulk, count) != 0)
249 		goto done;
250 
251 
252 	/*
253 	 * While the order here doesn't matter its best to try and organize
254 	 * it is such a way to pick up an already existing layout number
255 	 */
256 	count = 0;
257 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
258 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SIZE(zfsvfs), NULL,
259 	    &zp->z_size, 8);
260 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_GEN(zfsvfs),
261 	    NULL, &zp->z_gen, 8);
262 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
263 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
264 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_PARENT(zfsvfs),
265 	    NULL, &parent, 8);
266 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_FLAGS(zfsvfs), NULL,
267 	    &zp->z_pflags, 8);
268 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_ATIME(zfsvfs), NULL,
269 	    zp->z_atime, 16);
270 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_MTIME(zfsvfs), NULL,
271 	    &mtime, 16);
272 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_CTIME(zfsvfs), NULL,
273 	    &ctime, 16);
274 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_CRTIME(zfsvfs), NULL,
275 	    &crtime, 16);
276 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_LINKS(zfsvfs), NULL,
277 	    &zp->z_links, 8);
278 	if (zp->z_vnode->v_type == VBLK || zp->z_vnode->v_type == VCHR)
279 		SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_RDEV(zfsvfs), NULL,
280 		    &rdev, 8);
281 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
282 	    &zp->z_acl_cached->z_acl_count, 8);
283 
284 	if (zp->z_acl_cached->z_version < ZFS_ACL_VERSION_FUID)
285 		zfs_acl_xform(zp, zp->z_acl_cached, CRED());
286 
287 	locate.cb_aclp = zp->z_acl_cached;
288 	SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_DACL_ACES(zfsvfs),
289 	    zfs_acl_data_locator, &locate, zp->z_acl_cached->z_acl_bytes);
290 
291 	if (xattr)
292 		SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_XATTR(zfsvfs),
293 		    NULL, &xattr, 8);
294 
295 	/* if scanstamp then add scanstamp */
296 
297 	if (zp->z_pflags & ZFS_BONUS_SCANSTAMP) {
298 		bcopy((caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
299 		    scanstamp, AV_SCANSTAMP_SZ);
300 		SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SCANSTAMP(zfsvfs),
301 		    NULL, scanstamp, AV_SCANSTAMP_SZ);
302 		zp->z_pflags &= ~ZFS_BONUS_SCANSTAMP;
303 	}
304 
305 	VERIFY(dmu_set_bonustype(db, DMU_OT_SA, tx) == 0);
306 	VERIFY(sa_replace_all_by_template_locked(hdl, sa_attrs,
307 	    count, tx) == 0);
308 	if (znode_acl.z_acl_extern_obj)
309 		VERIFY(0 == dmu_object_free(zfsvfs->z_os,
310 		    znode_acl.z_acl_extern_obj, tx));
311 
312 	zp->z_is_sa = B_TRUE;
313 done:
314 	if (drop_lock)
315 		mutex_exit(&zp->z_lock);
316 }
317 
318 void
zfs_sa_upgrade_txholds(dmu_tx_t * tx,znode_t * zp)319 zfs_sa_upgrade_txholds(dmu_tx_t *tx, znode_t *zp)
320 {
321 	if (!zp->z_zfsvfs->z_use_sa || zp->z_is_sa)
322 		return;
323 
324 
325 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
326 
327 	if (zfs_external_acl(zp)) {
328 		dmu_tx_hold_free(tx, zfs_external_acl(zp), 0,
329 		    DMU_OBJECT_END);
330 	}
331 }
332 
333 #endif
334