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.
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.
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]
22 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
25 #include <sys/types.h>
26 #include <sys/param.h>
27 #include <sys/vnode.h>
29 #include <sys/zfs_acl.h>
30 #include <sys/zfs_sa.h>
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
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.
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 {"ZPL_DXATTR", 0, SA_UINT8_ARRAY, 0},
72 zfs_sa_readlink(znode_t *zp, uio_t *uio)
74 dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
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);
85 if ((error = dmu_buf_hold(ZTOZSB(zp)->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);
96 zfs_sa_symlink(znode_t *zp, char *link, int len, dmu_tx_t *tx)
98 dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
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);
104 bcopy(link, (caddr_t)db->db_data +
105 ZFS_OLD_ZNODE_PHYS_SIZE, len);
110 zfs_grow_blocksize(zp, len, tx);
111 VERIFY(0 == dmu_buf_hold(ZTOZSB(zp)->z_os,
112 zp->z_id, 0, FTAG, &dbp, DMU_READ_NO_PREFETCH));
114 dmu_buf_will_dirty(dbp, tx);
116 ASSERT3U(len, <=, dbp->db_size);
117 bcopy(link, dbp->db_data, len);
118 dmu_buf_rele(dbp, FTAG);
123 zfs_sa_get_scanstamp(znode_t *zp, xvattr_t *xvap)
125 zfs_sb_t *zsb = ZTOZSB(zp);
128 ASSERT(MUTEX_HELD(&zp->z_lock));
129 VERIFY((xoap = xva_getxoptattr(xvap)) != NULL);
131 if (sa_lookup(zp->z_sa_hdl, SA_ZPL_SCANSTAMP(zsb),
132 &xoap->xoa_av_scanstamp,
133 sizeof (xoap->xoa_av_scanstamp)) != 0)
136 dmu_object_info_t doi;
137 dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
140 if (!(zp->z_pflags & ZFS_BONUS_SCANSTAMP))
143 sa_object_info(zp->z_sa_hdl, &doi);
144 len = sizeof (xoap->xoa_av_scanstamp) +
145 ZFS_OLD_ZNODE_PHYS_SIZE;
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));
153 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
157 zfs_sa_set_scanstamp(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
159 zfs_sb_t *zsb = ZTOZSB(zp);
162 ASSERT(MUTEX_HELD(&zp->z_lock));
163 VERIFY((xoap = xva_getxoptattr(xvap)) != NULL);
165 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SCANSTAMP(zsb),
166 &xoap->xoa_av_scanstamp,
167 sizeof (xoap->xoa_av_scanstamp), tx));
169 dmu_object_info_t doi;
170 dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
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));
181 zp->z_pflags |= ZFS_BONUS_SCANSTAMP;
182 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_FLAGS(zsb),
183 &zp->z_pflags, sizeof (uint64_t), tx));
188 zfs_sa_get_xattr(znode_t *zp)
190 zfs_sb_t *zsb = ZTOZSB(zp);
196 ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock));
197 ASSERT(!zp->z_xattr_cached);
200 error = sa_handle_get(zsb->z_os, zp->z_id, NULL, SA_HDL_PRIVATE, &sa);
204 error = sa_size(sa, SA_ZPL_DXATTR(zsb), &size);
206 sa_handle_destroy(sa);
209 return nvlist_alloc(&zp->z_xattr_cached,
210 NV_UNIQUE_NAME, KM_SLEEP);
215 obj = sa_spill_alloc(KM_SLEEP);
217 error = sa_lookup(sa, SA_ZPL_DXATTR(zsb), obj, size);
219 error = nvlist_unpack(obj, size, &zp->z_xattr_cached, KM_SLEEP);
222 sa_handle_destroy(sa);
228 zfs_sa_set_xattr(znode_t *zp)
230 zfs_sb_t *zsb = ZTOZSB(zp);
237 ASSERT(RW_WRITE_HELD(&zp->z_xattr_lock));
238 ASSERT(zp->z_xattr_cached);
241 error = nvlist_size(zp->z_xattr_cached, &size, NV_ENCODE_XDR);
245 obj = sa_spill_alloc(KM_SLEEP);
247 error = nvlist_pack(zp->z_xattr_cached, &obj, &size,
248 NV_ENCODE_XDR, KM_SLEEP);
253 * A private SA handle must be used to ensure we can drop the hold
254 * on the spill block prior to calling dmu_tx_commit(). If we call
255 * dmu_tx_commit() before sa_handle_destroy(), then our hold will
256 * trigger a copy of the buffer at txg sync time. This is done to
257 * prevent data from leaking in to the syncing txg. As a result
258 * the original dirty spill block will be remain dirty in the arc
259 * while the copy is written and laundered.
261 error = sa_handle_get(zsb->z_os, zp->z_id, NULL, SA_HDL_PRIVATE, &sa);
265 tx = dmu_tx_create(zsb->z_os);
266 dmu_tx_hold_sa_create(tx, size);
267 dmu_tx_hold_sa(tx, sa, B_TRUE);
269 error = dmu_tx_assign(tx, TXG_WAIT);
272 sa_handle_destroy(sa);
274 error = sa_update(sa, SA_ZPL_DXATTR(zsb), obj, size, tx);
275 sa_handle_destroy(sa);
288 * I'm not convinced we should do any of this upgrade.
289 * since the SA code can read both old/new znode formats
290 * with probably little to know performance difference.
292 * All new files will be created with the new format.
296 zfs_sa_upgrade(sa_handle_t *hdl, dmu_tx_t *tx)
298 dmu_buf_t *db = sa_get_db(hdl);
299 znode_t *zp = sa_get_userdata(hdl);
300 zfs_sb_t *zsb = ZTOZSB(zp);
302 sa_bulk_attr_t *bulk, *sa_attrs;
303 zfs_acl_locator_cb_t locate = { 0 };
304 uint64_t uid, gid, mode, rdev, xattr, parent;
305 uint64_t crtime[2], mtime[2], ctime[2];
306 zfs_acl_phys_t znode_acl;
307 char scanstamp[AV_SCANSTAMP_SZ];
308 boolean_t drop_lock = B_FALSE;
311 * No upgrade if ACL isn't cached
312 * since we won't know which locks are held
313 * and ready the ACL would require special "locked"
314 * interfaces that would be messy
316 if (zp->z_acl_cached == NULL || S_ISLNK(ZTOI(zp)->i_mode))
320 * If the z_lock is held and we aren't the owner
321 * the just return since we don't want to deadlock
322 * trying to update the status of z_is_sa. This
323 * file can then be upgraded at a later time.
325 * Otherwise, we know we are doing the
326 * sa_update() that caused us to enter this function.
328 if (mutex_owner(&zp->z_lock) != curthread) {
329 if (mutex_tryenter(&zp->z_lock) == 0)
335 /* First do a bulk query of the attributes that aren't cached */
336 bulk = kmem_alloc(sizeof(sa_bulk_attr_t) * 20, KM_SLEEP);
337 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL, &mtime, 16);
338 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zsb), NULL, &ctime, 16);
339 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zsb), NULL, &crtime, 16);
340 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zsb), NULL, &mode, 8);
341 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zsb), NULL, &parent, 8);
342 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_XATTR(zsb), NULL, &xattr, 8);
343 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zsb), NULL, &rdev, 8);
344 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zsb), NULL, &uid, 8);
345 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zsb), NULL, &gid, 8);
346 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zsb), NULL,
349 if (sa_bulk_lookup_locked(hdl, bulk, count) != 0) {
350 kmem_free(bulk, sizeof(sa_bulk_attr_t) * 20);
355 * While the order here doesn't matter its best to try and organize
356 * it is such a way to pick up an already existing layout number
359 sa_attrs = kmem_zalloc(sizeof(sa_bulk_attr_t) * 20, KM_SLEEP);
360 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_MODE(zsb), NULL, &mode, 8);
361 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SIZE(zsb), NULL,
363 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_GEN(zsb),
364 NULL, &zp->z_gen, 8);
365 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_UID(zsb), NULL, &uid, 8);
366 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_GID(zsb), NULL, &gid, 8);
367 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_PARENT(zsb),
369 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_FLAGS(zsb), NULL,
371 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_ATIME(zsb), NULL,
373 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_MTIME(zsb), NULL,
375 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_CTIME(zsb), NULL,
377 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_CRTIME(zsb), NULL,
379 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_LINKS(zsb), NULL,
381 if (S_ISBLK(ZTOI(zp)->i_mode) || S_ISCHR(ZTOI(zp)->i_mode))
382 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_RDEV(zsb), NULL,
384 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_DACL_COUNT(zsb), NULL,
385 &zp->z_acl_cached->z_acl_count, 8);
387 if (zp->z_acl_cached->z_version < ZFS_ACL_VERSION_FUID)
388 zfs_acl_xform(zp, zp->z_acl_cached, CRED());
390 locate.cb_aclp = zp->z_acl_cached;
391 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_DACL_ACES(zsb),
392 zfs_acl_data_locator, &locate, zp->z_acl_cached->z_acl_bytes);
395 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_XATTR(zsb),
398 /* if scanstamp then add scanstamp */
400 if (zp->z_pflags & ZFS_BONUS_SCANSTAMP) {
401 bcopy((caddr_t)db->db_data + ZFS_OLD_ZNODE_PHYS_SIZE,
402 scanstamp, AV_SCANSTAMP_SZ);
403 SA_ADD_BULK_ATTR(sa_attrs, count, SA_ZPL_SCANSTAMP(zsb),
404 NULL, scanstamp, AV_SCANSTAMP_SZ);
405 zp->z_pflags &= ~ZFS_BONUS_SCANSTAMP;
408 VERIFY(dmu_set_bonustype(db, DMU_OT_SA, tx) == 0);
409 VERIFY(sa_replace_all_by_template_locked(hdl, sa_attrs,
411 if (znode_acl.z_acl_extern_obj)
412 VERIFY(0 == dmu_object_free(zsb->z_os,
413 znode_acl.z_acl_extern_obj, tx));
415 zp->z_is_sa = B_TRUE;
416 kmem_free(sa_attrs, sizeof(sa_bulk_attr_t) * 20);
417 kmem_free(bulk, sizeof(sa_bulk_attr_t) * 20);
420 mutex_exit(&zp->z_lock);
424 zfs_sa_upgrade_txholds(dmu_tx_t *tx, znode_t *zp)
426 if (!ZTOZSB(zp)->z_use_sa || zp->z_is_sa)
430 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
432 if (zfs_external_acl(zp)) {
433 dmu_tx_hold_free(tx, zfs_external_acl(zp), 0,
438 EXPORT_SYMBOL(zfs_attr_table);
439 EXPORT_SYMBOL(zfs_sa_readlink);
440 EXPORT_SYMBOL(zfs_sa_symlink);
441 EXPORT_SYMBOL(zfs_sa_get_scanstamp);
442 EXPORT_SYMBOL(zfs_sa_set_scanstamp);
443 EXPORT_SYMBOL(zfs_sa_get_xattr);
444 EXPORT_SYMBOL(zfs_sa_set_xattr);
445 EXPORT_SYMBOL(zfs_sa_upgrade);
446 EXPORT_SYMBOL(zfs_sa_upgrade_txholds);