Print this page
3749 zfs event processing should work on R/O root filesystems
Submitted by:   Justin Gibbs <justing@spectralogic.com>


   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 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  25  * Copyright (c) 2012 by Delphix. All rights reserved.
  26  */
  27 
  28 #include <sys/spa.h>

  29 #include <sys/spa_impl.h>
  30 #include <sys/nvpair.h>
  31 #include <sys/uio.h>
  32 #include <sys/fs/zfs.h>
  33 #include <sys/vdev_impl.h>
  34 #include <sys/zfs_ioctl.h>
  35 #include <sys/utsname.h>
  36 #include <sys/systeminfo.h>
  37 #include <sys/sunddi.h>
  38 #include <sys/zfeature.h>
  39 #ifdef _KERNEL
  40 #include <sys/kobj.h>
  41 #include <sys/zone.h>
  42 #endif
  43 
  44 /*
  45  * Pool configuration repository.
  46  *
  47  * Pool configuration is stored as a packed nvlist on the filesystem.  By
  48  * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot


 123                 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
 124                         continue;
 125 
 126                 VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
 127 
 128                 if (spa_lookup(nvpair_name(nvpair)) != NULL)
 129                         continue;
 130                 (void) spa_add(nvpair_name(nvpair), child, NULL);
 131         }
 132         mutex_exit(&spa_namespace_lock);
 133 
 134         nvlist_free(nvlist);
 135 
 136 out:
 137         if (buf != NULL)
 138                 kmem_free(buf, fsize);
 139 
 140         kobj_close_file(file);
 141 }
 142 
 143 static void
 144 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
 145 {
 146         size_t buflen;
 147         char *buf;
 148         vnode_t *vp;
 149         int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
 150         char *temp;

 151 
 152         /*
 153          * If the nvlist is empty (NULL), then remove the old cachefile.
 154          */
 155         if (nvl == NULL) {
 156                 (void) vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
 157                 return;
 158         }
 159 
 160         /*
 161          * Pack the configuration into a buffer.
 162          */
 163         VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
 164 
 165         buf = kmem_alloc(buflen, KM_SLEEP);
 166         temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
 167 
 168         VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
 169             KM_SLEEP) == 0);
 170 
 171         /*
 172          * Write the configuration to disk.  We need to do the traditional
 173          * 'write to temporary file, sync, move over original' to make sure we
 174          * always have a consistent view of the data.
 175          */
 176         (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
 177 
 178         if (vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0) == 0) {
 179                 if (vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
 180                     0, RLIM64_INFINITY, kcred, NULL) == 0 &&
 181                     VOP_FSYNC(vp, FSYNC, kcred, NULL) == 0) {
 182                         (void) vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
 183                 }


 184                 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
 185                 VN_RELE(vp);
 186         }
 187 
 188         (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
 189 
 190         kmem_free(buf, buflen);
 191         kmem_free(temp, MAXPATHLEN);

 192 }
 193 
 194 /*
 195  * Synchronize pool configuration to disk.  This must be called with the
 196  * namespace lock held.
 197  */
 198 void
 199 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
 200 {
 201         spa_config_dirent_t *dp, *tdp;
 202         nvlist_t *nvl;


 203 
 204         ASSERT(MUTEX_HELD(&spa_namespace_lock));
 205 
 206         if (rootdir == NULL || !(spa_mode_global & FWRITE))
 207                 return;
 208 
 209         /*
 210          * Iterate over all cachefiles for the pool, past or present.  When the
 211          * cachefile is changed, the new one is pushed onto this list, allowing
 212          * us to update previous cachefiles that no longer contain this pool.
 213          */

 214         for (dp = list_head(&target->spa_config_list); dp != NULL;
 215             dp = list_next(&target->spa_config_list, dp)) {
 216                 spa_t *spa = NULL;
 217                 if (dp->scd_path == NULL)
 218                         continue;
 219 
 220                 /*
 221                  * Iterate over all pools, adding any matching pools to 'nvl'.
 222                  */
 223                 nvl = NULL;
 224                 while ((spa = spa_next(spa)) != NULL) {
 225                         /*
 226                          * Skip over our own pool if we're about to remove
 227                          * ourselves from the spa namespace or any pool that
 228                          * is readonly. Since we cannot guarantee that a
 229                          * readonly pool would successfully import upon reboot,
 230                          * we don't allow them to be written to the cache file.
 231                          */
 232                         if ((spa == target && removing) ||
 233                             !spa_writeable(spa))
 234                                 continue;
 235 
 236                         mutex_enter(&spa->spa_props_lock);
 237                         tdp = list_head(&spa->spa_config_list);
 238                         if (spa->spa_config == NULL ||
 239                             tdp->scd_path == NULL ||
 240                             strcmp(tdp->scd_path, dp->scd_path) != 0) {
 241                                 mutex_exit(&spa->spa_props_lock);
 242                                 continue;
 243                         }
 244 
 245                         if (nvl == NULL)
 246                                 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
 247                                     KM_SLEEP) == 0);
 248 
 249                         VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
 250                             spa->spa_config) == 0);
 251                         mutex_exit(&spa->spa_props_lock);
 252                 }
 253 
 254                 spa_config_write(dp, nvl);


 255                 nvlist_free(nvl);
 256         }
 257 


















 258         /*
 259          * Remove any config entries older than the current one.
 260          */
 261         dp = list_head(&target->spa_config_list);
 262         while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
 263                 list_remove(&target->spa_config_list, tdp);
 264                 if (tdp->scd_path != NULL)
 265                         spa_strfree(tdp->scd_path);
 266                 kmem_free(tdp, sizeof (spa_config_dirent_t));
 267         }
 268 
 269         spa_config_generation++;
 270 
 271         if (postsysevent)
 272                 spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
 273 }
 274 
 275 /*
 276  * Sigh.  Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
 277  * and we don't want to allow the local zone to see all the pools anyway.




   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 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  25  * Copyright (c) 2012 by Delphix. All rights reserved.
  26  */
  27 
  28 #include <sys/spa.h>
  29 #include <sys/fm/fs/zfs.h>
  30 #include <sys/spa_impl.h>
  31 #include <sys/nvpair.h>
  32 #include <sys/uio.h>
  33 #include <sys/fs/zfs.h>
  34 #include <sys/vdev_impl.h>
  35 #include <sys/zfs_ioctl.h>
  36 #include <sys/utsname.h>
  37 #include <sys/systeminfo.h>
  38 #include <sys/sunddi.h>
  39 #include <sys/zfeature.h>
  40 #ifdef _KERNEL
  41 #include <sys/kobj.h>
  42 #include <sys/zone.h>
  43 #endif
  44 
  45 /*
  46  * Pool configuration repository.
  47  *
  48  * Pool configuration is stored as a packed nvlist on the filesystem.  By
  49  * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot


 124                 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
 125                         continue;
 126 
 127                 VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
 128 
 129                 if (spa_lookup(nvpair_name(nvpair)) != NULL)
 130                         continue;
 131                 (void) spa_add(nvpair_name(nvpair), child, NULL);
 132         }
 133         mutex_exit(&spa_namespace_lock);
 134 
 135         nvlist_free(nvlist);
 136 
 137 out:
 138         if (buf != NULL)
 139                 kmem_free(buf, fsize);
 140 
 141         kobj_close_file(file);
 142 }
 143 
 144 static int
 145 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
 146 {
 147         size_t buflen;
 148         char *buf;
 149         vnode_t *vp;
 150         int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
 151         char *temp;
 152         int err;
 153 
 154         /*
 155          * If the nvlist is empty (NULL), then remove the old cachefile.
 156          */
 157         if (nvl == NULL) {
 158                 err = vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
 159                 return (err);
 160         }
 161 
 162         /*
 163          * Pack the configuration into a buffer.
 164          */
 165         VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
 166 
 167         buf = kmem_alloc(buflen, KM_SLEEP);
 168         temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
 169 
 170         VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
 171             KM_SLEEP) == 0);
 172 
 173         /*
 174          * Write the configuration to disk.  We need to do the traditional
 175          * 'write to temporary file, sync, move over original' to make sure we
 176          * always have a consistent view of the data.
 177          */
 178         (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
 179 
 180         err = vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0);
 181         if (err == 0) {
 182                 err = vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
 183                     0, RLIM64_INFINITY, kcred, NULL);
 184                 if (err == 0)
 185                         err = VOP_FSYNC(vp, FSYNC, kcred, NULL);
 186                 if (err == 0)
 187                         err = vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
 188                 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
 189                 VN_RELE(vp);
 190         }
 191 
 192         (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
 193 
 194         kmem_free(buf, buflen);
 195         kmem_free(temp, MAXPATHLEN);
 196         return (err);
 197 }
 198 
 199 /*
 200  * Synchronize pool configuration to disk.  This must be called with the
 201  * namespace lock held.
 202  */
 203 void
 204 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
 205 {
 206         spa_config_dirent_t *dp, *tdp;
 207         nvlist_t *nvl;
 208         boolean_t ccw_failure;
 209         int error;
 210 
 211         ASSERT(MUTEX_HELD(&spa_namespace_lock));
 212 
 213         if (rootdir == NULL || !(spa_mode_global & FWRITE))
 214                 return;
 215 
 216         /*
 217          * Iterate over all cachefiles for the pool, past or present.  When the
 218          * cachefile is changed, the new one is pushed onto this list, allowing
 219          * us to update previous cachefiles that no longer contain this pool.
 220          */
 221         ccw_failure = B_FALSE;
 222         for (dp = list_head(&target->spa_config_list); dp != NULL;
 223             dp = list_next(&target->spa_config_list, dp)) {
 224                 spa_t *spa = NULL;
 225                 if (dp->scd_path == NULL)
 226                         continue;
 227 
 228                 /*
 229                  * Iterate over all pools, adding any matching pools to 'nvl'.
 230                  */
 231                 nvl = NULL;
 232                 while ((spa = spa_next(spa)) != NULL) {
 233                         /*
 234                          * Skip over our own pool if we're about to remove
 235                          * ourselves from the spa namespace or any pool that
 236                          * is readonly. Since we cannot guarantee that a
 237                          * readonly pool would successfully import upon reboot,
 238                          * we don't allow them to be written to the cache file.
 239                          */
 240                         if ((spa == target && removing) ||
 241                             !spa_writeable(spa))
 242                                 continue;
 243 
 244                         mutex_enter(&spa->spa_props_lock);
 245                         tdp = list_head(&spa->spa_config_list);
 246                         if (spa->spa_config == NULL ||
 247                             tdp->scd_path == NULL ||
 248                             strcmp(tdp->scd_path, dp->scd_path) != 0) {
 249                                 mutex_exit(&spa->spa_props_lock);
 250                                 continue;
 251                         }
 252 
 253                         if (nvl == NULL)
 254                                 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
 255                                     KM_SLEEP) == 0);
 256 
 257                         VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
 258                             spa->spa_config) == 0);
 259                         mutex_exit(&spa->spa_props_lock);
 260                 }
 261 
 262                 error = spa_config_write(dp, nvl);
 263                 if (error != 0)
 264                         ccw_failure = B_TRUE;
 265                 nvlist_free(nvl);
 266         }
 267 
 268         if (ccw_failure) {
 269                 /*
 270                  * Keep trying so that configuration data is
 271                  * written if/when any temporary filesystem
 272                  * resource issues are resolved.
 273                  */
 274                 target->spa_ccw_fail_time = ddi_get_lbolt64();
 275                 spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
 276                 zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
 277                     target, NULL, NULL, 0, 0);
 278         } else {
 279                 /*
 280                  * Do not rate limit future attempts to update
 281                  * the config cache.
 282                  */
 283                 target->spa_ccw_fail_time = 0;
 284         }
 285 
 286         /*
 287          * Remove any config entries older than the current one.
 288          */
 289         dp = list_head(&target->spa_config_list);
 290         while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
 291                 list_remove(&target->spa_config_list, tdp);
 292                 if (tdp->scd_path != NULL)
 293                         spa_strfree(tdp->scd_path);
 294                 kmem_free(tdp, sizeof (spa_config_dirent_t));
 295         }
 296 
 297         spa_config_generation++;
 298 
 299         if (postsysevent)
 300                 spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
 301 }
 302 
 303 /*
 304  * Sigh.  Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
 305  * and we don't want to allow the local zone to see all the pools anyway.