Print this page
3752 want more verifiable dbuf user eviction
Submitted by: Justin Gibbs <justing@spectralogic.com>
Submitted by: Will Andrews <willa@spectralogic.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/zfs_sa.c
+++ new/usr/src/uts/common/fs/zfs/zfs_sa.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
|
↓ open down ↓ |
14 lines elided |
↑ open up ↑ |
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
23 23 */
24 24
25 +#include <sys/zfs_context.h>
25 26 #include <sys/types.h>
26 27 #include <sys/param.h>
27 28 #include <sys/vnode.h>
28 29 #include <sys/sa.h>
29 30 #include <sys/zfs_acl.h>
30 31 #include <sys/zfs_sa.h>
31 32
32 33 /*
33 34 * ZPL attribute registration table.
34 35 * Order of attributes doesn't matter
35 36 * a unique value will be assigned for each
36 37 * attribute that is file system specific
37 38 *
38 39 * This is just the set of ZPL attributes that this
39 40 * version of ZFS deals with natively. The file system
40 41 * could have other attributes stored in files, but they will be
41 42 * ignored. The SA framework will preserve them, just that
42 43 * this version of ZFS won't change or delete them.
43 44 */
44 45
45 46 sa_attr_reg_t zfs_attr_table[ZPL_END+1] = {
46 47 {"ZPL_ATIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 0},
47 48 {"ZPL_MTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 1},
48 49 {"ZPL_CTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 2},
49 50 {"ZPL_CRTIME", sizeof (uint64_t) * 2, SA_UINT64_ARRAY, 3},
50 51 {"ZPL_GEN", sizeof (uint64_t), SA_UINT64_ARRAY, 4},
51 52 {"ZPL_MODE", sizeof (uint64_t), SA_UINT64_ARRAY, 5},
52 53 {"ZPL_SIZE", sizeof (uint64_t), SA_UINT64_ARRAY, 6},
53 54 {"ZPL_PARENT", sizeof (uint64_t), SA_UINT64_ARRAY, 7},
54 55 {"ZPL_LINKS", sizeof (uint64_t), SA_UINT64_ARRAY, 8},
55 56 {"ZPL_XATTR", sizeof (uint64_t), SA_UINT64_ARRAY, 9},
56 57 {"ZPL_RDEV", sizeof (uint64_t), SA_UINT64_ARRAY, 10},
57 58 {"ZPL_FLAGS", sizeof (uint64_t), SA_UINT64_ARRAY, 11},
58 59 {"ZPL_UID", sizeof (uint64_t), SA_UINT64_ARRAY, 12},
59 60 {"ZPL_GID", sizeof (uint64_t), SA_UINT64_ARRAY, 13},
60 61 {"ZPL_PAD", sizeof (uint64_t) * 4, SA_UINT64_ARRAY, 14},
61 62 {"ZPL_ZNODE_ACL", 88, SA_UINT8_ARRAY, 15},
62 63 {"ZPL_DACL_COUNT", sizeof (uint64_t), SA_UINT64_ARRAY, 0},
63 64 {"ZPL_SYMLINK", 0, SA_UINT8_ARRAY, 0},
64 65 {"ZPL_SCANSTAMP", 32, SA_UINT8_ARRAY, 0},
65 66 {"ZPL_DACL_ACES", 0, SA_ACL, 0},
66 67 {NULL, 0, 0, 0}
67 68 };
68 69
69 70 #ifdef _KERNEL
70 71
71 72 int
72 73 zfs_sa_readlink(znode_t *zp, uio_t *uio)
73 74 {
74 75 dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
75 76 size_t bufsz;
76 77 int error;
77 78
78 79 bufsz = zp->z_size;
79 80 if (bufsz + ZFS_OLD_ZNODE_PHYS_SIZE <= db->db_size) {
80 81 error = uiomove((caddr_t)db->db_data +
81 82 ZFS_OLD_ZNODE_PHYS_SIZE,
82 83 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
83 84 } else {
84 85 dmu_buf_t *dbp;
85 86 if ((error = dmu_buf_hold(zp->z_zfsvfs->z_os, zp->z_id,
86 87 0, FTAG, &dbp, DMU_READ_NO_PREFETCH)) == 0) {
87 88 error = uiomove(dbp->db_data,
88 89 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
89 90 dmu_buf_rele(dbp, FTAG);
90 91 }
91 92 }
92 93 return (error);
93 94 }
94 95
95 96 void
96 97 zfs_sa_symlink(znode_t *zp, char *link, int len, dmu_tx_t *tx)
97 98 {
98 99 dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
99 100
100 101 if (ZFS_OLD_ZNODE_PHYS_SIZE + len <= dmu_bonus_max()) {
101 102 VERIFY(dmu_set_bonus(db,
102 103 len + ZFS_OLD_ZNODE_PHYS_SIZE, tx) == 0);
103 104 if (len) {
104 105 bcopy(link, (caddr_t)db->db_data +
105 106 ZFS_OLD_ZNODE_PHYS_SIZE, len);
106 107 }
107 108 } else {
108 109 dmu_buf_t *dbp;
109 110
110 111 zfs_grow_blocksize(zp, len, tx);
111 112 VERIFY(0 == dmu_buf_hold(zp->z_zfsvfs->z_os,
112 113 zp->z_id, 0, FTAG, &dbp, DMU_READ_NO_PREFETCH));
113 114
114 115 dmu_buf_will_dirty(dbp, tx);
115 116
116 117 ASSERT3U(len, <=, dbp->db_size);
117 118 bcopy(link, dbp->db_data, len);
118 119 dmu_buf_rele(dbp, FTAG);
119 120 }
120 121 }
121 122
122 123 void
123 124 zfs_sa_get_scanstamp(znode_t *zp, xvattr_t *xvap)
124 125 {
125 126 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
126 127 xoptattr_t *xoap;
127 128
128 129 ASSERT(MUTEX_HELD(&zp->z_lock));
129 130 VERIFY((xoap = xva_getxoptattr(xvap)) != NULL);
130 131 if (zp->z_is_sa) {
131 132 if (sa_lookup(zp->z_sa_hdl, SA_ZPL_SCANSTAMP(zfsvfs),
132 133 &xoap->xoa_av_scanstamp,
133 134 sizeof (xoap->xoa_av_scanstamp)) != 0)
134 135 return;
135 136 } else {
136 137 dmu_object_info_t doi;
137 138 dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
138 139 int len;
139 140
140 141 if (!(zp->z_pflags & ZFS_BONUS_SCANSTAMP))
141 142 return;
142 143
143 144 sa_object_info(zp->z_sa_hdl, &doi);
144 145 len = sizeof (xoap->xoa_av_scanstamp) +
145 146 ZFS_OLD_ZNODE_PHYS_SIZE;
146 147
147 148 if (len <= doi.doi_bonus_size) {
148 149 (void) memcpy(xoap->xoa_av_scanstamp,
149 150 (caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
150 151 sizeof (xoap->xoa_av_scanstamp));
151 152 }
152 153 }
153 154 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
154 155 }
155 156
156 157 void
157 158 zfs_sa_set_scanstamp(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
158 159 {
159 160 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
160 161 xoptattr_t *xoap;
161 162
162 163 ASSERT(MUTEX_HELD(&zp->z_lock));
163 164 VERIFY((xoap = xva_getxoptattr(xvap)) != NULL);
164 165 if (zp->z_is_sa)
165 166 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SCANSTAMP(zfsvfs),
166 167 &xoap->xoa_av_scanstamp,
167 168 sizeof (xoap->xoa_av_scanstamp), tx));
168 169 else {
169 170 dmu_object_info_t doi;
170 171 dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
171 172 int len;
172 173
173 174 sa_object_info(zp->z_sa_hdl, &doi);
174 175 len = sizeof (xoap->xoa_av_scanstamp) +
175 176 ZFS_OLD_ZNODE_PHYS_SIZE;
176 177 if (len > doi.doi_bonus_size)
177 178 VERIFY(dmu_set_bonus(db, len, tx) == 0);
178 179 (void) memcpy((caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
179 180 xoap->xoa_av_scanstamp, sizeof (xoap->xoa_av_scanstamp));
180 181
181 182 zp->z_pflags |= ZFS_BONUS_SCANSTAMP;
182 183 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
183 184 &zp->z_pflags, sizeof (uint64_t), tx));
184 185 }
185 186 }
186 187
187 188 /*
188 189 * I'm not convinced we should do any of this upgrade.
189 190 * since the SA code can read both old/new znode formats
190 191 * with probably little to know performance difference.
191 192 *
192 193 * All new files will be created with the new format.
193 194 */
194 195
195 196 void
196 197 zfs_sa_upgrade(sa_handle_t *hdl, dmu_tx_t *tx)
197 198 {
198 199 dmu_buf_t *db = sa_get_db(hdl);
199 200 znode_t *zp = sa_get_userdata(hdl);
200 201 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
201 202 sa_bulk_attr_t bulk[20];
202 203 int count = 0;
203 204 sa_bulk_attr_t sa_attrs[20] = { 0 };
204 205 zfs_acl_locator_cb_t locate = { 0 };
205 206 uint64_t uid, gid, mode, rdev, xattr, parent;
206 207 uint64_t crtime[2], mtime[2], ctime[2];
207 208 zfs_acl_phys_t znode_acl;
208 209 char scanstamp[AV_SCANSTAMP_SZ];
209 210 boolean_t drop_lock = B_FALSE;
210 211
211 212 /*
212 213 * No upgrade if ACL isn't cached
213 214 * since we won't know which locks are held
214 215 * and ready the ACL would require special "locked"
215 216 * interfaces that would be messy
216 217 */
217 218 if (zp->z_acl_cached == NULL || ZTOV(zp)->v_type == VLNK)
218 219 return;
219 220
220 221 /*
221 222 * If the z_lock is held and we aren't the owner
222 223 * the just return since we don't want to deadlock
223 224 * trying to update the status of z_is_sa. This
224 225 * file can then be upgraded at a later time.
225 226 *
226 227 * Otherwise, we know we are doing the
227 228 * sa_update() that caused us to enter this function.
228 229 */
229 230 if (mutex_owner(&zp->z_lock) != curthread) {
230 231 if (mutex_tryenter(&zp->z_lock) == 0)
231 232 return;
232 233 else
233 234 drop_lock = B_TRUE;
234 235 }
235 236
236 237 /* First do a bulk query of the attributes that aren't cached */
237 238 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
238 239 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
239 240 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
240 241 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
241 242 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
242 243 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_XATTR(zfsvfs), NULL, &xattr, 8);
243 244 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, &rdev, 8);
244 245 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
245 246 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
246 247 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
247 248 &znode_acl, 88);
248 249
249 250 if (sa_bulk_lookup_locked(hdl, bulk, count) != 0)
250 251 goto done;
251 252
252 253
253 254 /*
254 255 * While the order here doesn't matter its best to try and organize
255 256 * it is such a way to pick up an already existing layout number
256 257 */
257 258 count = 0;
258 259 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
259 260 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SIZE(zfsvfs), NULL,
260 261 &zp->z_size, 8);
261 262 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_GEN(zfsvfs),
262 263 NULL, &zp->z_gen, 8);
263 264 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_UID(zfsvfs), NULL, &uid, 8);
264 265 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_GID(zfsvfs), NULL, &gid, 8);
265 266 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_PARENT(zfsvfs),
266 267 NULL, &parent, 8);
267 268 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_FLAGS(zfsvfs), NULL,
268 269 &zp->z_pflags, 8);
269 270 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_ATIME(zfsvfs), NULL,
270 271 zp->z_atime, 16);
271 272 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_MTIME(zfsvfs), NULL,
272 273 &mtime, 16);
273 274 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_CTIME(zfsvfs), NULL,
274 275 &ctime, 16);
275 276 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_CRTIME(zfsvfs), NULL,
276 277 &crtime, 16);
277 278 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_LINKS(zfsvfs), NULL,
278 279 &zp->z_links, 8);
279 280 if (zp->z_vnode->v_type == VBLK || zp->z_vnode->v_type == VCHR)
280 281 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_RDEV(zfsvfs), NULL,
281 282 &rdev, 8);
282 283 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
283 284 &zp->z_acl_cached->z_acl_count, 8);
284 285
285 286 if (zp->z_acl_cached->z_version < ZFS_ACL_VERSION_FUID)
286 287 zfs_acl_xform(zp, zp->z_acl_cached, CRED());
287 288
288 289 locate.cb_aclp = zp->z_acl_cached;
289 290 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_DACL_ACES(zfsvfs),
290 291 zfs_acl_data_locator, &locate, zp->z_acl_cached->z_acl_bytes);
291 292
292 293 if (xattr)
293 294 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_XATTR(zfsvfs),
294 295 NULL, &xattr, 8);
295 296
296 297 /* if scanstamp then add scanstamp */
297 298
298 299 if (zp->z_pflags & ZFS_BONUS_SCANSTAMP) {
299 300 bcopy((caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
300 301 scanstamp, AV_SCANSTAMP_SZ);
301 302 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SCANSTAMP(zfsvfs),
302 303 NULL, scanstamp, AV_SCANSTAMP_SZ);
303 304 zp->z_pflags &= ~ZFS_BONUS_SCANSTAMP;
304 305 }
305 306
306 307 VERIFY(dmu_set_bonustype(db, DMU_OT_SA, tx) == 0);
307 308 VERIFY(sa_replace_all_by_template_locked(hdl, sa_attrs,
308 309 count, tx) == 0);
309 310 if (znode_acl.z_acl_extern_obj)
310 311 VERIFY(0 == dmu_object_free(zfsvfs->z_os,
311 312 znode_acl.z_acl_extern_obj, tx));
312 313
313 314 zp->z_is_sa = B_TRUE;
314 315 done:
315 316 if (drop_lock)
316 317 mutex_exit(&zp->z_lock);
317 318 }
318 319
319 320 void
320 321 zfs_sa_upgrade_txholds(dmu_tx_t *tx, znode_t *zp)
321 322 {
322 323 if (!zp->z_zfsvfs->z_use_sa || zp->z_is_sa)
323 324 return;
324 325
325 326
326 327 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
327 328
328 329 if (zfs_external_acl(zp)) {
329 330 dmu_tx_hold_free(tx, zfs_external_acl(zp), 0,
330 331 DMU_OBJECT_END);
331 332 }
332 333 }
333 334
334 335 #endif
|
↓ open down ↓ |
300 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX