Whamcloud - gitweb
LU-4927 scrub: rebuild stale OST OI mapping
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_scrub.c
index 1804daa..d830e22 100644 (file)
@@ -94,30 +94,26 @@ static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
                                     int ops, bool force,
                                     enum oi_check_flags flags)
 {
-       const struct lu_env *env = info->oti_env;
-       struct thandle      *th;
-       struct osd_thandle  *oh;
-       int                  rc;
+       handle_t *th;
+       int       rc;
        ENTRY;
 
        if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
                RETURN(0);
 
-       th = dt_trans_create(env, &dev->od_dt_dev);
-       if (IS_ERR(th))
-               RETURN(PTR_ERR(th));
-
-       oh = container_of0(th, struct osd_thandle, ot_super);
-       LASSERT(oh->ot_handle == NULL);
+       /* DTO_INDEX_INSERT is enough for other two ops:
+        * delete/update, but save stack. */
+       th = ldiskfs_journal_start_sb(osd_sb(dev),
+                               osd_dto_credits_noquota[DTO_INDEX_INSERT]);
+       if (IS_ERR(th)) {
+               rc = PTR_ERR(th);
+               CERROR("%s: fail to start trans for scrub %d: rc = %d\n",
+                      osd_name(dev), ops, rc);
+               RETURN(rc);
+       }
 
        switch (ops) {
        case DTO_INDEX_UPDATE:
-               osd_trans_declare_op(env, oh, OSD_OT_UPDATE,
-                                    osd_dto_credits_noquota[DTO_INDEX_UPDATE]);
-               rc = dt_trans_start_local(env, &dev->od_dt_dev, th);
-               if (rc != 0)
-                       GOTO(stop, rc);
-
                rc = osd_oi_update(info, dev, fid, id, th, flags);
                if (unlikely(rc == -ENOENT)) {
                        /* Some unlink thread may removed the OI mapping. */
@@ -125,12 +121,6 @@ static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
                }
                break;
        case DTO_INDEX_INSERT:
-               osd_trans_declare_op(env, oh, OSD_OT_INSERT,
-                                    osd_dto_credits_noquota[DTO_INDEX_INSERT]);
-               rc = dt_trans_start_local(env, &dev->od_dt_dev, th);
-               if (rc != 0)
-                       GOTO(stop, rc);
-
                rc = osd_oi_insert(info, dev, fid, id, th, flags);
                if (unlikely(rc == -EEXIST)) {
                        rc = 1;
@@ -168,12 +158,6 @@ static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
                }
                break;
        case DTO_INDEX_DELETE:
-               osd_trans_declare_op(env, oh, OSD_OT_DELETE,
-                                    osd_dto_credits_noquota[DTO_INDEX_DELETE]);
-               rc = dt_trans_start_local(env, &dev->od_dt_dev, th);
-               if (rc != 0)
-                       GOTO(stop, rc);
-
                rc = osd_oi_delete(info, dev, fid, th, flags);
                if (rc == -ENOENT) {
                        /* It is normal that the unlink thread has removed the
@@ -186,10 +170,7 @@ static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
                break;
        }
 
-       GOTO(stop, rc);
-
-stop:
-       dt_trans_stop(env, &dev->od_dt_dev, th);
+       ldiskfs_journal_stop(th);
        return rc;
 }
 
@@ -505,7 +486,7 @@ osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
                (val == SCRUB_NEXT_OSTOBJ ||
                 val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
        if (rc != 0) {
-               if (rc != -ENOENT)
+               if (rc != -ENOENT && rc != -ESTALE)
                        GOTO(out, rc);
 
 iget:
@@ -827,36 +808,31 @@ static int osd_scrub_get_fid(struct osd_thread_info *info,
        rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
        if (rc == 0) {
                has_lma = true;
-               if (lma->lma_compat & LMAC_NOT_IN_OI) {
-                       ldiskfs_set_inode_state(inode,
-                                               LDISKFS_STATE_LUSTRE_NO_OI);
+               if (lma->lma_compat & LMAC_NOT_IN_OI ||
+                   lma->lma_incompat & LMAI_AGENT)
                        return SCRUB_NEXT_CONTINUE;
-               }
 
                *fid = lma->lma_self_fid;
-               if (fid_is_internal(&lma->lma_self_fid)) {
-                       if (!scrub)
-                               rc = SCRUB_NEXT_CONTINUE;
-                       return rc;
-               }
-
                if (!scrub)
                        return 0;
 
-               if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
-                       return 0;
-
                if (lma->lma_compat & LMAC_FID_ON_OST)
                        return SCRUB_NEXT_OSTOBJ;
 
-               if (fid_is_idif(fid) || fid_is_last_id(fid))
+               if (fid_is_idif(fid))
                        return SCRUB_NEXT_OSTOBJ_OLD;
 
-               if (lma->lma_incompat & LMAI_AGENT)
-                       return SCRUB_NEXT_CONTINUE;
+               /* For local object. */
+               if (fid_is_internal(fid))
+                       return 0;
 
-               /* Here, it may be MDT-object, or may be 2.4 OST-object.
-                * Fall through. */
+               /* For external visible MDT-object with non-normal FID. */
+               if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
+                       return 0;
+
+               /* For the object with normal FID, it may be MDT-object,
+                * or may be 2.4 OST-object, need further distinguish.
+                * Fall through to next section. */
        }
 
        if (rc == -ENODATA || rc == 0) {
@@ -930,11 +906,6 @@ static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
                RETURN(rc);
        }
 
-       /* If the inode has no OI mapping, then it is special locally used,
-        * should be invisible to OI scrub or up layer LFSCK. */
-       if (ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI))
-               GOTO(put, rc = SCRUB_NEXT_CONTINUE);
-
        if (scrub &&
            ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
                /* Only skip it for the first OI scrub accessing. */
@@ -1328,6 +1299,8 @@ static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
                               loff_t offset, __u64 ino, unsigned d_type);
 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
                           loff_t offset, __u64 ino, unsigned d_type);
+static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
+                          loff_t offset, __u64 ino, unsigned d_type);
 
 static int
 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
@@ -1438,6 +1411,10 @@ static const struct osd_lf_map osd_lf_maps[] = {
        { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
                OLF_SHOW_NAME, NULL, NULL },
 
+       /* SLAVE_LOG, llog for destroy slave stripes of striped dir */
+       { "SLAVE_LOG", { FID_SEQ_LOCAL_FILE, SLAVE_LLOG_CATALOGS_OID, 0 },
+              OLF_SHOW_NAME, NULL, NULL },
+
        /* lost+found */
        { "lost+found", { 0, 0, 0 }, OLF_SCAN_SUBITEMS | OLF_NO_OI,
                osd_ios_general_scan, osd_ios_lf_fill },
@@ -1445,6 +1422,19 @@ static const struct osd_lf_map osd_lf_maps[] = {
        { NULL, { 0, 0, 0 }, 0, NULL, NULL }
 };
 
+/* Add the new introduced files under .lustre/ in the list in the future. */
+static const struct osd_lf_map osd_dl_maps[] = {
+       /* .lustre/fid */
+       { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
+               NULL, NULL },
+
+       /* .lustre/lost+found */
+       { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
+               NULL, NULL },
+
+       { NULL, { 0, 0, 0 }, 0, NULL, NULL }
+};
+
 struct osd_ios_item {
        cfs_list_t       oii_list;
        struct dentry   *oii_dentry;
@@ -1453,6 +1443,10 @@ struct osd_ios_item {
 };
 
 struct osd_ios_filldir_buf {
+#ifdef HAVE_DIR_CONTEXT
+       /* please keep it as first member */
+       struct dir_context       ctx;
+#endif
        struct osd_thread_info  *oifb_info;
        struct osd_device       *oifb_dev;
        struct dentry           *oifb_dentry;
@@ -1668,6 +1662,42 @@ static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
        RETURN(rc);
 }
 
+static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
+                          loff_t offset, __u64 ino, unsigned d_type)
+{
+       struct osd_ios_filldir_buf *fill_buf = buf;
+       struct osd_device          *dev      = fill_buf->oifb_dev;
+       const struct osd_lf_map    *map;
+       struct dentry              *child;
+       int                         rc       = 0;
+       ENTRY;
+
+       /* skip any '.' started names */
+       if (name[0] == '.')
+               RETURN(0);
+
+       for (map = osd_dl_maps; map->olm_name != NULL; map++) {
+               if (strlen(map->olm_name) != namelen)
+                       continue;
+
+               if (strncmp(map->olm_name, name, namelen) == 0)
+                       break;
+       }
+
+       if (map->olm_name == NULL)
+               RETURN(0);
+
+       child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
+       if (IS_ERR(child))
+               RETURN(PTR_ERR(child));
+
+       rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
+                             &map->olm_fid, map->olm_flags);
+       dput(child);
+
+       RETURN(rc);
+}
+
 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
                             loff_t offset, __u64 ino, unsigned d_type)
 {
@@ -1712,7 +1742,13 @@ static int
 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
                     struct dentry *dentry, filldir_t filldir)
 {
-       struct osd_ios_filldir_buf    buf   = { info, dev, dentry };
+       struct osd_ios_filldir_buf    buf   = {
+#ifdef HAVE_DIR_CONTEXT
+                                               .ctx.actor = filldir,
+#endif
+                                               .oifb_info = info,
+                                               .oifb_dev = dev,
+                                               .oifb_dentry = dentry };
        struct file                  *filp  = &info->oti_it_ea.oie_file;
        struct inode                 *inode = dentry->d_inode;
        const struct file_operations *fops  = inode->i_fop;
@@ -1729,7 +1765,13 @@ osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
        filp->private_data = NULL;
        set_file_inode(filp, inode);
 
+#ifdef HAVE_DIR_CONTEXT
+       buf.ctx.pos = filp->f_pos;
+       rc = fops->iterate(filp, &buf.ctx);
+       filp->f_pos = buf.ctx.pos;
+#else
        rc = fops->readdir(filp, &buf, filldir);
+#endif
        fops->release(inode, filp);
 
        RETURN(rc);
@@ -1801,6 +1843,9 @@ osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
                 * to the solution 2). */
                rc = osd_ios_scan_one(info, dev, child->d_inode,
                                      &LU_DOT_LUSTRE_FID, 0);
+               if (rc == 0)
+                       rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
+                                             osd_ios_dl_fill);
                dput(child);
        }
 
@@ -1949,6 +1994,7 @@ static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
        struct osd_scrub     *scrub  = &dev->od_scrub;
        struct ptlrpc_thread *thread = &scrub->os_thread;
        struct l_wait_info    lwi    = { 0 };
+       struct task_struct   *task;
        int                   rc;
        ENTRY;
 
@@ -1972,9 +2018,10 @@ again:
 
        scrub->os_start_flags = flags;
        thread_set_flags(thread, 0);
-       rc = PTR_ERR(kthread_run(osd_scrub_main, dev, "OI_scrub"));
-       if (IS_ERR_VALUE(rc)) {
-               CERROR("%.16s: cannot start iteration thread, rc = %d\n",
+       task = kthread_run(osd_scrub_main, dev, "OI_scrub");
+       if (IS_ERR(task)) {
+               rc = PTR_ERR(task);
+               CERROR("%.16s: cannot start iteration thread: rc = %d\n",
                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
                RETURN(rc);
        }
@@ -2060,28 +2107,27 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
        spin_lock_init(&scrub->os_lock);
        CFS_INIT_LIST_HEAD(&scrub->os_inconsistent_items);
 
-       push_ctxt(&saved, ctxt, NULL);
+       push_ctxt(&saved, ctxt);
        filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
        if (IS_ERR(filp)) {
-               pop_ctxt(&saved, ctxt, NULL);
+               pop_ctxt(&saved, ctxt);
                RETURN(PTR_ERR(filp));
        }
 
        inode = filp->f_dentry->d_inode;
-       ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
        /* 'What the @fid is' is not imporatant, because the object
         * has no OI mapping, and only is visible inside the OSD.*/
        lu_igif_build(fid, inode->i_ino, inode->i_generation);
        rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
        if (rc != 0) {
                filp_close(filp, 0);
-               pop_ctxt(&saved, ctxt, NULL);
+               pop_ctxt(&saved, ctxt);
                RETURN(rc);
        }
 
        scrub->os_inode = igrab(inode);
        filp_close(filp, 0);
-       pop_ctxt(&saved, ctxt, NULL);
+       pop_ctxt(&saved, ctxt);
 
        rc = osd_scrub_file_load(scrub);
        if (rc == -ENOENT) {
@@ -2527,134 +2573,112 @@ static const char *scrub_param_names[] = {
        NULL
 };
 
-static int scrub_bits_dump(char **buf, int *len, int bits, const char *names[],
+static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
                           const char *prefix)
 {
-       int save = *len;
        int flag;
        int rc;
        int i;
 
-       rc = snprintf(*buf, *len, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
-       if (rc <= 0)
-               return -ENOSPC;
+       rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
+       if (rc < 0)
+               return rc;
 
-       *buf += rc;
-       *len -= rc;
        for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
                if (flag & bits) {
                        bits &= ~flag;
-                       rc = snprintf(*buf, *len, "%s%c", names[i],
-                                     bits != 0 ? ',' : '\n');
-                       if (rc <= 0)
-                               return -ENOSPC;
-
-                       *buf += rc;
-                       *len -= rc;
+                       rc = seq_printf(m, "%s%c", names[i],
+                                       bits != 0 ? ',' : '\n');
+                       if (rc < 0)
+                               return rc;
                }
        }
-       return save - *len;
+       return 0;
 }
 
-static int scrub_time_dump(char **buf, int *len, __u64 time, const char *prefix)
+static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
 {
        int rc;
 
        if (time != 0)
-               rc = snprintf(*buf, *len, "%s: "LPU64" seconds\n", prefix,
+               rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
                              cfs_time_current_sec() - time);
        else
-               rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
-       if (rc <= 0)
-               return -ENOSPC;
-
-       *buf += rc;
-       *len -= rc;
+               rc = seq_printf(m, "%s: N/A\n", prefix);
        return rc;
 }
 
-static int scrub_pos_dump(char **buf, int *len, __u64 pos, const char *prefix)
+static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
 {
        int rc;
 
        if (pos != 0)
-               rc = snprintf(*buf, *len, "%s: "LPU64"\n", prefix, pos);
+               rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
        else
-               rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
-       if (rc <= 0)
-               return -ENOSPC;
-
-       *buf += rc;
-       *len -= rc;
+               rc = seq_printf(m, "%s: N/A\n", prefix);
        return rc;
 }
 
-int osd_scrub_dump(struct osd_device *dev, char *buf, int len)
+int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
 {
        struct osd_scrub  *scrub   = &dev->od_scrub;
        struct scrub_file *sf      = &scrub->os_file;
        __u64              checked;
        __u64              speed;
-       int                save    = len;
-       int                ret     = -ENOSPC;
        int                rc;
 
        down_read(&scrub->os_rwsem);
-       rc = snprintf(buf, len,
-                     "name: OI_scrub\n"
-                     "magic: 0x%x\n"
-                     "oi_files: %d\n"
-                     "status: %s\n",
-                     sf->sf_magic, (int)sf->sf_oi_count,
-                     scrub_status_names[sf->sf_status]);
-       if (rc <= 0)
+       rc = seq_printf(m, "name: OI_scrub\n"
+                       "magic: 0x%x\n"
+                       "oi_files: %d\n"
+                       "status: %s\n",
+                       sf->sf_magic, (int)sf->sf_oi_count,
+                       scrub_status_names[sf->sf_status]);
+       if (rc < 0)
                goto out;
 
-       buf += rc;
-       len -= rc;
-       rc = scrub_bits_dump(&buf, &len, sf->sf_flags, scrub_flags_names,
+       rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
                             "flags");
        if (rc < 0)
                goto out;
 
-       rc = scrub_bits_dump(&buf, &len, sf->sf_param, scrub_param_names,
+       rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,
                             "param");
        if (rc < 0)
                goto out;
 
-       rc = scrub_time_dump(&buf, &len, sf->sf_time_last_complete,
+       rc = scrub_time_dump(m, sf->sf_time_last_complete,
                             "time_since_last_completed");
        if (rc < 0)
                goto out;
 
-       rc = scrub_time_dump(&buf, &len, sf->sf_time_latest_start,
+       rc = scrub_time_dump(m, sf->sf_time_latest_start,
                             "time_since_latest_start");
        if (rc < 0)
                goto out;
 
-       rc = scrub_time_dump(&buf, &len, sf->sf_time_last_checkpoint,
+       rc = scrub_time_dump(m, sf->sf_time_last_checkpoint,
                             "time_since_last_checkpoint");
        if (rc < 0)
                goto out;
 
-       rc = scrub_pos_dump(&buf, &len, sf->sf_pos_latest_start,
+       rc = scrub_pos_dump(m, sf->sf_pos_latest_start,
                            "latest_start_position");
        if (rc < 0)
                goto out;
 
-       rc = scrub_pos_dump(&buf, &len, sf->sf_pos_last_checkpoint,
+       rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
                            "last_checkpoint_position");
        if (rc < 0)
                goto out;
 
-       rc = scrub_pos_dump(&buf, &len, sf->sf_pos_first_inconsistent,
+       rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
                            "first_failure_position");
        if (rc < 0)
                goto out;
 
        checked = sf->sf_items_checked + scrub->os_new_checked;
-       rc = snprintf(buf, len,
-                     "checked: "LPU64"\n"
+       rc = seq_printf(m, "checked: "LPU64"\n"
                      "updated: "LPU64"\n"
                      "failed: "LPU64"\n"
                      "prior_updated: "LPU64"\n"
@@ -2664,11 +2688,9 @@ int osd_scrub_dump(struct osd_device *dev, char *buf, int len)
                      checked, sf->sf_items_updated, sf->sf_items_failed,
                      sf->sf_items_updated_prior, sf->sf_items_noscrub,
                      sf->sf_items_igif, sf->sf_success_count);
-       if (rc <= 0)
+       if (rc < 0)
                goto out;
 
-       buf += rc;
-       len -= rc;
        speed = checked;
        if (thread_is_running(&scrub->os_thread)) {
                cfs_duration_t duration = cfs_time_current() -
@@ -2681,8 +2703,7 @@ int osd_scrub_dump(struct osd_device *dev, char *buf, int len)
                        do_div(new_checked, duration);
                if (rtime != 0)
                        do_div(speed, rtime);
-               rc = snprintf(buf, len,
-                             "run_time: %u seconds\n"
+               rc = seq_printf(m, "run_time: %u seconds\n"
                              "average_speed: "LPU64" objects/sec\n"
                              "real-time_speed: "LPU64" objects/sec\n"
                              "current_position: %u\n"
@@ -2695,8 +2716,7 @@ int osd_scrub_dump(struct osd_device *dev, char *buf, int len)
        } else {
                if (sf->sf_run_time != 0)
                        do_div(speed, sf->sf_run_time);
-               rc = snprintf(buf, len,
-                             "run_time: %u seconds\n"
+               rc = seq_printf(m, "run_time: %u seconds\n"
                              "average_speed: "LPU64" objects/sec\n"
                              "real-time_speed: N/A\n"
                              "current_position: N/A\n"
@@ -2706,14 +2726,8 @@ int osd_scrub_dump(struct osd_device *dev, char *buf, int len)
                              sf->sf_run_time, speed, scrub->os_lf_scanned,
                              scrub->os_lf_repaired, scrub->os_lf_failed);
        }
-       if (rc <= 0)
-               goto out;
-
-       buf += rc;
-       len -= rc;
-       ret = save - len;
 
 out:
        up_read(&scrub->os_rwsem);
-       return ret;
+       return (rc < 0 ? -ENOSPC : 0);
 }