Whamcloud - gitweb
LU-7428 osd: set device read-only correctly
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_handler.c
index 060e16f..60855df 100644 (file)
 #include <lustre_linkea.h>
 
 int ldiskfs_pdo = 1;
-CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
-                "ldiskfs with parallel directory operations");
+module_param(ldiskfs_pdo, int, 0644);
+MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
 
 int ldiskfs_track_declares_assert;
-CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
-               "LBUG during tracking of declares");
+module_param(ldiskfs_track_declares_assert, int, 0644);
+MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
 
 /* Slab to allocate dynlocks */
 struct kmem_cache *dynlock_cachep;
@@ -779,9 +779,9 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
        struct osd_scrub       *scrub;
        struct scrub_file      *sf;
        int                     result;
-       int                     saved  = 0;
-       bool                    cached  = true;
-       bool                    triggered = false;
+       int                     rc1     = 0;
+       bool                    cached  = true;
+       bool                    remote  = false;
        ENTRY;
 
        LINVRNT(osd_invariant(obj));
@@ -847,29 +847,6 @@ iget:
                if (result == -EREMCHG) {
 
 trigger:
-                       if (unlikely(triggered))
-                               GOTO(out, result = saved);
-
-                       triggered = true;
-                       if (thread_is_running(&scrub->os_thread)) {
-                               result = -EINPROGRESS;
-                       } else if (!dev->od_noscrub) {
-                               result = osd_scrub_start(dev, SS_AUTO_FULL |
-                                       SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
-                               LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
-                                             "for "DFID", rc = %d [1]\n",
-                                             osd_name(dev), PFID(fid), result);
-                               if (result == 0 || result == -EALREADY)
-                                       result = -EINPROGRESS;
-                               else
-                                       result = -EREMCHG;
-                       } else {
-                               result = -EREMCHG;
-                       }
-
-                       if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD))
-                               GOTO(out, result);
-
                        /* We still have chance to get the valid inode: for the
                         * object which is referenced by remote name entry, the
                         * object on the local MDT will be linked under the dir
@@ -885,18 +862,54 @@ trigger:
                         * only happened for the RPC from other MDT during the
                         * OI scrub, or for the client side RPC with FID only,
                         * such as FID to path, or from old connected client. */
-                       saved = result;
-                       result = osd_lookup_in_remote_parent(info, dev,
-                                                            fid, id);
-                       if (result == 0) {
-                               cached = true;
-                               goto iget;
+                       if (!remote &&
+                           !fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
+                               rc1 = osd_lookup_in_remote_parent(info, dev,
+                                                                 fid, id);
+                               if (rc1 == 0) {
+                                       remote = true;
+                                       cached = true;
+                                       goto iget;
+                               }
                        }
 
-                       result = saved;
+                       if (thread_is_running(&scrub->os_thread)) {
+                               if (remote) {
+                                       osd_add_oi_cache(info, dev, id, fid);
+                                       osd_oii_insert(dev, oic, true);
+                               } else {
+                                       result = -EINPROGRESS;
+                               }
+                       } else if (!dev->od_noscrub) {
+                               __u32 flags = SS_CLEAR_DRYRUN |
+                                             SS_CLEAR_FAILOUT;
+
+                               flags |= (remote ? SS_AUTO_PARTIAL :
+                                                  SS_AUTO_FULL);
+                               rc1 = osd_scrub_start(dev, flags);
+                               LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
+                                             "for the "DFID" with flags 0x%x,"
+                                             " rc = %d\n", osd_name(dev),
+                                             PFID(fid), flags, rc1);
+                               if (rc1 == 0 || rc1 == -EALREADY) {
+                                       result = -EINPROGRESS;
+                                       if (remote) {
+                                               osd_add_oi_cache(info, dev, id,
+                                                                fid);
+                                               osd_oii_insert(dev, oic, true);
+                                       }
+                               } else {
+                                       result = -EREMCHG;
+                               }
+                       } else {
+                               result = -EREMCHG;
+                       }
                }
 
                GOTO(out, result);
+       } else if (remote) {
+               result = 0;
+               goto trigger;
        }
 
        obj->oo_inode = inode;
@@ -1213,6 +1226,11 @@ static void osd_trans_commit_cb(struct super_block *sb,
        OBD_FREE_PTR(oh);
 }
 
+#ifndef HAVE_SB_START_WRITE
+# define sb_start_write(sb) do {} while (0)
+# define sb_end_write(sb) do {} while (0)
+#endif
+
 static struct thandle *osd_trans_create(const struct lu_env *env,
                                        struct dt_device *d)
 {
@@ -1225,7 +1243,8 @@ static struct thandle *osd_trans_create(const struct lu_env *env,
        /* on pending IO in this thread should left from prev. request */
        LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
 
-       th = ERR_PTR(-ENOMEM);
+       sb_start_write(osd_sb(osd_dt_dev(d)));
+
        OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
        if (oh != NULL) {
                oh->ot_quota_trans = &oti->oti_quota_trans;
@@ -1245,6 +1264,9 @@ static struct thandle *osd_trans_create(const struct lu_env *env,
                       sizeof(oti->oti_declare_ops_cred));
                memset(oti->oti_declare_ops_used, 0,
                       sizeof(oti->oti_declare_ops_used));
+       } else {
+               sb_end_write(osd_sb(osd_dt_dev(d)));
+               th = ERR_PTR(-ENOMEM);
        }
        RETURN(th);
 }
@@ -1486,6 +1508,8 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
        if (unlikely(remove_agents != 0))
                osd_process_scheduled_agent_removals(env, osd);
 
+       sb_end_write(osd_sb(osd));
+
        RETURN(rc);
 }
 
@@ -1761,6 +1785,12 @@ static int osd_ro(const struct lu_env *env, struct dt_device *d)
 #ifdef HAVE_DEV_SET_RDONLY
        CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
 
+       if (sb->s_op->freeze_fs) {
+               rc = sb->s_op->freeze_fs(sb);
+               if (rc)
+                       goto out;
+       }
+
        if (jdev && (jdev != dev)) {
                CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
                       (long)jdev);
@@ -1768,10 +1798,16 @@ static int osd_ro(const struct lu_env *env, struct dt_device *d)
        }
        CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
        dev_set_rdonly(dev);
-#else
-       CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
-              osd_dt_dev(d)->od_svname, (long)dev, rc);
+
+       if (sb->s_op->unfreeze_fs)
+               sb->s_op->unfreeze_fs(sb);
+
+out:
 #endif
+       if (rc)
+               CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
+                      osd_dt_dev(d)->od_svname, (long)dev, rc);
+
        RETURN(rc);
 }
 
@@ -2610,7 +2646,8 @@ static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
        osd_trans_exec_op(env, th, OSD_OT_INSERT);
 
        osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
-       rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
+       rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
+                          OI_CHECK_FLD, NULL);
        osd_trans_exec_check(env, th, OSD_OT_INSERT);
 
        return rc;
@@ -6227,7 +6264,7 @@ static int osd_mount(const struct lu_env *env,
        struct osd_thread_info  *info = osd_oti_get(env);
        struct lu_fid           *fid = &info->oti_fid;
        struct inode            *inode;
-       int                      rc = 0, force_over_128tb = 0;
+       int                      rc = 0, force_over_256tb = 0;
         ENTRY;
 
        if (o->od_mnt != NULL)
@@ -6251,10 +6288,17 @@ static int osd_mount(const struct lu_env *env,
                RETURN(-EINVAL);
        }
 #endif
-       if (opts != NULL && strstr(opts, "force_over_128tb") != NULL)
-               force_over_128tb = 1;
+       if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
+               CWARN("force_over_128tb option is depricated."
+                     "Filesystems less then 256TB can be created without any"
+                     "force options. Use force_over_256tb option for"
+                     "filesystems greather then 256TB.\n");
+       }
 
-       __page = alloc_page(GFP_IOFS);
+       if (opts != NULL && strstr(opts, "force_over_256tb") != NULL)
+               force_over_256tb = 1;
+
+       __page = alloc_page(GFP_KERNEL);
        if (__page == NULL)
                GOTO(out, rc = -ENOMEM);
        page = (unsigned long)page_address(__page);
@@ -6271,7 +6315,12 @@ static int osd_mount(const struct lu_env *env,
                        "noextents",
                        /* strip out option we processed in osd */
                        "bigendian_extents",
-                       "force_over_128tb",
+#if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(3,0,53,0)
+#warning "remove force_over_128 option"
+#else
+                       "force_over_128tb (deprecated)",
+#endif
+                       "force_over_256tb",
                        NULL
                };
                strcat(options, opts);
@@ -6317,11 +6366,11 @@ static int osd_mount(const struct lu_env *env,
                GOTO(out, rc);
        }
 
-       if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) > (8ULL << 32) &&
-           force_over_128tb == 0) {
+       if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) > (64ULL << 30) &&
+           force_over_256tb == 0) {
                CERROR("%s: device %s LDISKFS does not support filesystems "
-                      "greater than 128TB and can cause data corruption. "
-                      "Use \"force_over_128tb\" mount option to override.\n",
+                      "greater than 256TB and can cause data corruption. "
+                      "Use \"force_over_256tb\" mount option to override.\n",
                       name, dev);
                GOTO(out, rc = -EINVAL);
        }