lustre_patchless_compat.h \
lustre_quota.h \
lustre_req_layout.h \
+ lustre_scrub.h \
lustre_sec.h \
lustre_swab.h \
lustre_update.h \
LUSTRE_SEQ_SUPER_WIDTH = ((1ULL << 30ULL) * LUSTRE_SEQ_META_WIDTH)
};
-enum {
- /** 2^6 FIDs for OI containers */
- OSD_OI_FID_OID_BITS = 6,
- /** reserve enough FIDs in case we want more in the future */
- OSD_OI_FID_OID_BITS_MAX = 10,
-};
-
/** special OID for local objects */
enum local_oid {
/** \see fld_mod_init */
--- /dev/null
+/*
+ * GPL HEADER START
+ *
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.gnu.org/licenses/gpl-2.0.html
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 2017, Intel Corporation.
+ */
+/*
+ * lustre/include/lustre_scrub.h
+ *
+ * Shared definitions and declarations for Lustre OI scrub.
+ *
+ * Author: Fan Yong <fan.yong@intel.com>
+ */
+
+#ifndef _LUSTRE_SCRUB_H
+# define _LUSTRE_SCRUB_H
+
+#include <dt_object.h>
+#include <lustre_net.h>
+
+#define OSD_OI_FID_OID_BITS_MAX 10
+#define OSD_OI_FID_NR_MAX (1UL << OSD_OI_FID_OID_BITS_MAX)
+#define SCRUB_OI_BITMAP_SIZE (OSD_OI_FID_NR_MAX >> 3)
+
+#define SCRUB_MAGIC_V1 0x4C5FD252
+#define SCRUB_CHECKPOINT_INTERVAL 60
+#define SCRUB_WINDOW_SIZE 1024
+
+#define HALF_SEC msecs_to_jiffies(MSEC_PER_SEC >> 1)
+
+enum scrub_next_status {
+ /* exit current loop and process next group */
+ SCRUB_NEXT_BREAK = 1,
+
+ /* skip current object and process next bit */
+ SCRUB_NEXT_CONTINUE = 2,
+
+ /* exit all the loops */
+ SCRUB_NEXT_EXIT = 3,
+
+ /* wait for free cache slot */
+ SCRUB_NEXT_WAIT = 4,
+
+ /* simulate system crash during OI scrub */
+ SCRUB_NEXT_CRASH = 5,
+
+ /* simulate failure during OI scrub */
+ SCRUB_NEXT_FATAL = 6,
+
+ /* new created object, no scrub on it */
+ SCRUB_NEXT_NOSCRUB = 7,
+
+ /* the object has no FID-in-LMA */
+ SCRUB_NEXT_NOLMA = 8,
+
+ /* for OST-object */
+ SCRUB_NEXT_OSTOBJ = 9,
+
+ /* old OST-object, no LMA or no FID-on-OST flags in LMA */
+ SCRUB_NEXT_OSTOBJ_OLD = 10,
+};
+
+enum scrub_local_file_flags {
+ SLFF_SCAN_SUBITEMS = 0x0001,
+ SLFF_HIDE_FID = 0x0002,
+ SLFF_SHOW_NAME = 0x0004,
+ SLFF_NO_OI = 0x0008,
+ SLFF_IDX_IN_FID = 0x0010,
+};
+
+enum scrub_status {
+ /* The scrub file is new created, for new MDT, upgrading from old disk,
+ * or re-creating the scrub file manually. */
+ SS_INIT = 0,
+
+ /* The scrub is checking/repairing the OI files. */
+ SS_SCANNING = 1,
+
+ /* The scrub checked/repaired the OI files successfully. */
+ SS_COMPLETED = 2,
+
+ /* The scrub failed to check/repair the OI files. */
+ SS_FAILED = 3,
+
+ /* The scrub is stopped manually, the OI files may be inconsistent. */
+ SS_STOPPED = 4,
+
+ /* The scrub is paused automatically when umount. */
+ SS_PAUSED = 5,
+
+ /* The scrub crashed during the scanning, should be restarted. */
+ SS_CRASHED = 6,
+};
+
+enum scrub_flags {
+ /* OI files have been recreated, OI mappings should be re-inserted. */
+ SF_RECREATED = 0x0000000000000001ULL,
+
+ /* OI files are invalid, should be rebuild ASAP */
+ SF_INCONSISTENT = 0x0000000000000002ULL,
+
+ /* OI scrub is triggered automatically. */
+ SF_AUTO = 0x0000000000000004ULL,
+
+ /* The device is upgraded from 1.8 format. */
+ SF_UPGRADE = 0x0000000000000008ULL,
+};
+
+enum scrub_param {
+ /* Exit when fail. */
+ SP_FAILOUT = 0x0001,
+
+ /* Check only without repairing. */
+ SP_DRYRUN = 0x0002,
+};
+
+enum scrub_start {
+ /* Set failout flag. */
+ SS_SET_FAILOUT = 0x00000001,
+
+ /* Clear failout flag. */
+ SS_CLEAR_FAILOUT = 0x00000002,
+
+ /* Reset scrub start position. */
+ SS_RESET = 0x00000004,
+
+ /* Trigger full scrub automatically. */
+ SS_AUTO_FULL = 0x00000008,
+
+ /* Trigger partial scrub automatically. */
+ SS_AUTO_PARTIAL = 0x00000010,
+
+ /* Set dryrun flag. */
+ SS_SET_DRYRUN = 0x00000020,
+
+ /* Clear dryrun flag. */
+ SS_CLEAR_DRYRUN = 0x00000040,
+};
+
+enum osd_lf_flags {
+ OLF_SCAN_SUBITEMS = 0x0001,
+ OLF_HIDE_FID = 0x0002,
+ OLF_SHOW_NAME = 0x0004,
+ OLF_NO_OI = 0x0008,
+ OLF_IDX_IN_FID = 0x0010,
+};
+
+struct scrub_file {
+ /* 128-bit uuid for volume. */
+ __u8 sf_uuid[16];
+
+ /* See 'enum scrub_flags'. */
+ __u64 sf_flags;
+
+ /* The scrub magic. */
+ __u32 sf_magic;
+
+ /* See 'enum scrub_status'. */
+ __u16 sf_status;
+
+ /* See 'enum scrub_param'. */
+ __u16 sf_param;
+
+ /* The time for the last OI scrub completed. */
+ __u64 sf_time_last_complete;
+
+ /* The time for the latest OI scrub ran. */
+ __u64 sf_time_latest_start;
+
+ /* The time for the last OI scrub checkpoint. */
+ __u64 sf_time_last_checkpoint;
+
+ /* The position for the latest OI scrub started from. */
+ __u64 sf_pos_latest_start;
+
+ /* The position for the last OI scrub checkpoint. */
+ __u64 sf_pos_last_checkpoint;
+
+ /* The position for the first should be updated object. */
+ __u64 sf_pos_first_inconsistent;
+
+ /* How many objects have been checked. */
+ __u64 sf_items_checked;
+
+ /* How many objects have been updated. */
+ __u64 sf_items_updated;
+
+ /* How many objects failed to be processed. */
+ __u64 sf_items_failed;
+
+ /* How many prior objects have been updated during scanning. */
+ __u64 sf_items_updated_prior;
+
+ /* How many objects marked as LDISKFS_STATE_LUSTRE_NOSCRUB. */
+ __u64 sf_items_noscrub;
+
+ /* How many IGIF objects. */
+ __u64 sf_items_igif;
+
+ /* How long the OI scrub has run. */
+ __u32 sf_run_time;
+
+ /* How many completed OI scrub ran on the device. */
+ __u32 sf_success_count;
+
+ /* How many OI files. */
+ __u16 sf_oi_count;
+
+ /* Keep the flags after scrub reset. See 'enum scrub_internal_flags' */
+ __u16 sf_internal_flags;
+
+ __u32 sf_reserved_1;
+ __u64 sf_reserved_2[16];
+
+ /* Bitmap for OI files recreated case. */
+ __u8 sf_oi_bitmap[SCRUB_OI_BITMAP_SIZE];
+};
+
+struct lustre_scrub {
+ struct lu_object_header os_obj_header;
+ /* Object for the scrub file. */
+ struct dt_object *os_obj;
+
+ struct ptlrpc_thread os_thread;
+ struct list_head os_inconsistent_items;
+
+ /* write lock for scrub prep/update/post/checkpoint,
+ * read lock for scrub dump. */
+ struct rw_semaphore os_rwsem;
+ spinlock_t os_lock;
+
+ /* Scrub file in memory. */
+ struct scrub_file os_file;
+
+ /* Buffer for scrub file load/store. */
+ struct scrub_file os_file_disk;
+
+ const char *os_name;
+
+ /* The time for last checkpoint, jiffies */
+ cfs_time_t os_time_last_checkpoint;
+
+ /* The time for next checkpoint, jiffies */
+ cfs_time_t os_time_next_checkpoint;
+
+ /* How many objects have been checked since last checkpoint. */
+ __u64 os_new_checked;
+ __u64 os_pos_current;
+ __u32 os_start_flags;
+ unsigned int os_in_prior:1, /* process inconsistent item
+ * found by RPC prior */
+ os_waiting:1, /* Waiting for scan window. */
+ os_full_speed:1, /* run w/o speed limit */
+ os_paused:1, /* The scrub is paused. */
+ os_convert_igif:1,
+ os_partial_scan:1,
+ os_in_join:1,
+ os_full_scrub:1;
+};
+
+void scrub_file_init(struct lustre_scrub *scrub, __u8 *uuid);
+void scrub_file_reset(struct lustre_scrub *scrub, __u8 *uuid, __u64 flags);
+int scrub_file_load(const struct lu_env *env, struct lustre_scrub *scrub);
+int scrub_file_store(const struct lu_env *env, struct lustre_scrub *scrub);
+int scrub_checkpoint(const struct lu_env *env, struct lustre_scrub *scrub);
+int scrub_start(int (*threadfn)(void *data), struct lustre_scrub *scrub,
+ void *data, __u32 flags);
+void scrub_stop(struct lustre_scrub *scrub);
+void scrub_dump(struct seq_file *m, struct lustre_scrub *scrub);
+
+static inline const char *osd_scrub2name(struct lustre_scrub *scrub)
+{
+ return scrub->os_name;
+}
+#endif /* _LUSTRE_SCRUB_H */
@SERVER_TRUE@obdclass-all-objs += lu_ucred.o
@SERVER_TRUE@obdclass-all-objs += md_attrs.o
@SERVER_TRUE@obdclass-all-objs += obd_mount_server.o
+@SERVER_TRUE@obdclass-all-objs += scrub.o
obdclass-objs := $(obdclass-linux-objs) $(obdclass-all-objs)
@SERVER_FALSE@EXTRA_DIST += lu_ucred.c
@SERVER_FALSE@EXTRA_DIST += md_attrs.c
@SERVER_FALSE@EXTRA_DIST += obd_mount_server.c
+@SERVER_FALSE@EXTRA_DIST += scrub.c
@INCLUDE_RULES@
--- /dev/null
+/*
+ * GPL HEADER START
+ *
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.gnu.org/licenses/gpl-2.0.html
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 2017, Intel Corporation.
+ */
+/*
+ * lustre/obdclass/scrub.c
+ *
+ * The OI scrub is used for checking and (re)building Object Index files
+ * that are usually backend special. Here are some general scrub related
+ * functions that can be shared by different backends for OI scrub.
+ *
+ * Author: Fan Yong <fan.yong@intel.com>
+ */
+
+#define DEBUG_SUBSYSTEM S_LFSCK
+
+#include <linux/kthread.h>
+#include <lustre_scrub.h>
+#include <lustre_lib.h>
+
+static inline struct dt_device *scrub_obj2dev(struct dt_object *obj)
+{
+ return container_of0(obj->do_lu.lo_dev, struct dt_device, dd_lu_dev);
+}
+
+static void scrub_file_to_cpu(struct scrub_file *des, struct scrub_file *src)
+{
+ memcpy(des->sf_uuid, src->sf_uuid, 16);
+ des->sf_flags = le64_to_cpu(src->sf_flags);
+ des->sf_magic = le32_to_cpu(src->sf_magic);
+ des->sf_status = le16_to_cpu(src->sf_status);
+ des->sf_param = le16_to_cpu(src->sf_param);
+ des->sf_time_last_complete =
+ le64_to_cpu(src->sf_time_last_complete);
+ des->sf_time_latest_start =
+ le64_to_cpu(src->sf_time_latest_start);
+ des->sf_time_last_checkpoint =
+ le64_to_cpu(src->sf_time_last_checkpoint);
+ des->sf_pos_latest_start =
+ le64_to_cpu(src->sf_pos_latest_start);
+ des->sf_pos_last_checkpoint =
+ le64_to_cpu(src->sf_pos_last_checkpoint);
+ des->sf_pos_first_inconsistent =
+ le64_to_cpu(src->sf_pos_first_inconsistent);
+ des->sf_items_checked =
+ le64_to_cpu(src->sf_items_checked);
+ des->sf_items_updated =
+ le64_to_cpu(src->sf_items_updated);
+ des->sf_items_failed =
+ le64_to_cpu(src->sf_items_failed);
+ des->sf_items_updated_prior =
+ le64_to_cpu(src->sf_items_updated_prior);
+ des->sf_run_time = le32_to_cpu(src->sf_run_time);
+ des->sf_success_count = le32_to_cpu(src->sf_success_count);
+ des->sf_oi_count = le16_to_cpu(src->sf_oi_count);
+ des->sf_internal_flags = le16_to_cpu(src->sf_internal_flags);
+ memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
+}
+
+static void scrub_file_to_le(struct scrub_file *des, struct scrub_file *src)
+{
+ memcpy(des->sf_uuid, src->sf_uuid, 16);
+ des->sf_flags = cpu_to_le64(src->sf_flags);
+ des->sf_magic = cpu_to_le32(src->sf_magic);
+ des->sf_status = cpu_to_le16(src->sf_status);
+ des->sf_param = cpu_to_le16(src->sf_param);
+ des->sf_time_last_complete =
+ cpu_to_le64(src->sf_time_last_complete);
+ des->sf_time_latest_start =
+ cpu_to_le64(src->sf_time_latest_start);
+ des->sf_time_last_checkpoint =
+ cpu_to_le64(src->sf_time_last_checkpoint);
+ des->sf_pos_latest_start =
+ cpu_to_le64(src->sf_pos_latest_start);
+ des->sf_pos_last_checkpoint =
+ cpu_to_le64(src->sf_pos_last_checkpoint);
+ des->sf_pos_first_inconsistent =
+ cpu_to_le64(src->sf_pos_first_inconsistent);
+ des->sf_items_checked =
+ cpu_to_le64(src->sf_items_checked);
+ des->sf_items_updated =
+ cpu_to_le64(src->sf_items_updated);
+ des->sf_items_failed =
+ cpu_to_le64(src->sf_items_failed);
+ des->sf_items_updated_prior =
+ cpu_to_le64(src->sf_items_updated_prior);
+ des->sf_run_time = cpu_to_le32(src->sf_run_time);
+ des->sf_success_count = cpu_to_le32(src->sf_success_count);
+ des->sf_oi_count = cpu_to_le16(src->sf_oi_count);
+ des->sf_internal_flags = cpu_to_le16(src->sf_internal_flags);
+ memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
+}
+
+void scrub_file_init(struct lustre_scrub *scrub, __u8 *uuid)
+{
+ struct scrub_file *sf = &scrub->os_file;
+
+ memset(sf, 0, sizeof(*sf));
+ memcpy(sf->sf_uuid, uuid, 16);
+ sf->sf_magic = SCRUB_MAGIC_V1;
+ sf->sf_status = SS_INIT;
+}
+EXPORT_SYMBOL(scrub_file_init);
+
+void scrub_file_reset(struct lustre_scrub *scrub, __u8 *uuid, __u64 flags)
+{
+ struct scrub_file *sf = &scrub->os_file;
+
+ CDEBUG(D_LFSCK, "%s: reset OI scrub file, old flags = "
+ "%#llx, add flags = %#llx\n",
+ scrub->os_name, sf->sf_flags, flags);
+
+ memcpy(sf->sf_uuid, uuid, 16);
+ sf->sf_status = SS_INIT;
+ sf->sf_flags |= flags;
+ sf->sf_flags &= ~SF_AUTO;
+ sf->sf_run_time = 0;
+ sf->sf_time_latest_start = 0;
+ sf->sf_time_last_checkpoint = 0;
+ sf->sf_pos_latest_start = 0;
+ sf->sf_pos_last_checkpoint = 0;
+ sf->sf_pos_first_inconsistent = 0;
+ sf->sf_items_checked = 0;
+ sf->sf_items_updated = 0;
+ sf->sf_items_failed = 0;
+ sf->sf_items_noscrub = 0;
+ sf->sf_items_igif = 0;
+ if (!scrub->os_in_join)
+ sf->sf_items_updated_prior = 0;
+}
+EXPORT_SYMBOL(scrub_file_reset);
+
+int scrub_file_load(const struct lu_env *env, struct lustre_scrub *scrub)
+{
+ struct scrub_file *sf = &scrub->os_file;
+ struct lu_buf buf = {
+ .lb_buf = &scrub->os_file_disk,
+ .lb_len = sizeof(scrub->os_file_disk)
+ };
+ loff_t pos = 0;
+ int rc;
+
+ rc = dt_read(env, scrub->os_obj, &buf, &pos);
+ /* failure */
+ if (rc < 0) {
+ CERROR("%s: fail to load scrub file: rc = %d\n",
+ scrub->os_name, rc);
+ return rc;
+ }
+
+ /* empty */
+ if (!rc)
+ return -ENOENT;
+
+ /* corrupted */
+ if (rc < buf.lb_len) {
+ CDEBUG(D_LFSCK, "%s: fail to load scrub file, "
+ "expected = %d: rc = %d\n",
+ scrub->os_name, (int)buf.lb_len, rc);
+ return -EFAULT;
+ }
+
+ scrub_file_to_cpu(sf, &scrub->os_file_disk);
+ if (sf->sf_magic != SCRUB_MAGIC_V1) {
+ CDEBUG(D_LFSCK, "%s: invalid scrub magic 0x%x != 0x%x\n",
+ scrub->os_name, sf->sf_magic, SCRUB_MAGIC_V1);
+ return -EFAULT;
+ }
+
+ return 0;
+}
+EXPORT_SYMBOL(scrub_file_load);
+
+int scrub_file_store(const struct lu_env *env, struct lustre_scrub *scrub)
+{
+ struct scrub_file *sf = &scrub->os_file_disk;
+ struct dt_object *obj = scrub->os_obj;
+ struct dt_device *dev = scrub_obj2dev(obj);
+ struct lu_buf buf = {
+ .lb_buf = sf,
+ .lb_len = sizeof(*sf)
+ };
+ struct thandle *th;
+ loff_t pos = 0;
+ int rc;
+ ENTRY;
+
+ /* Skip store under rdonly mode. */
+ if (dev->dd_rdonly)
+ RETURN(0);
+
+ scrub_file_to_le(sf, &scrub->os_file);
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ GOTO(log, rc = PTR_ERR(th));
+
+ rc = dt_declare_record_write(env, obj, &buf, pos, th);
+ if (rc)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, th);
+ if (rc)
+ GOTO(stop, rc);
+
+ rc = dt_record_write(env, obj, &buf, &pos, th);
+
+ GOTO(stop, rc);
+
+stop:
+ dt_trans_stop(env, dev, th);
+
+log:
+ if (rc)
+ CERROR("%s: store scrub file: rc = %d\n",
+ scrub->os_name, rc);
+ else
+ CDEBUG(D_LFSCK, "%s: store scrub file: rc = %d\n",
+ scrub->os_name, rc);
+
+ scrub->os_time_last_checkpoint = cfs_time_current();
+ scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
+ cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
+ return rc;
+}
+EXPORT_SYMBOL(scrub_file_store);
+
+int scrub_checkpoint(const struct lu_env *env, struct lustre_scrub *scrub)
+{
+ struct scrub_file *sf = &scrub->os_file;
+ int rc;
+
+ if (likely(cfs_time_before(cfs_time_current(),
+ scrub->os_time_next_checkpoint) ||
+ scrub->os_new_checked == 0))
+ return 0;
+
+ CDEBUG(D_LFSCK, "%s: OI scrub checkpoint at pos %llu\n",
+ scrub->os_name, scrub->os_pos_current);
+
+ down_write(&scrub->os_rwsem);
+ sf->sf_items_checked += scrub->os_new_checked;
+ scrub->os_new_checked = 0;
+ sf->sf_pos_last_checkpoint = scrub->os_pos_current;
+ sf->sf_time_last_checkpoint = cfs_time_current_sec();
+ sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
+ scrub->os_time_last_checkpoint);
+ rc = scrub_file_store(env, scrub);
+ up_write(&scrub->os_rwsem);
+
+ return rc;
+}
+EXPORT_SYMBOL(scrub_checkpoint);
+
+int scrub_start(int (*threadfn)(void *data), struct lustre_scrub *scrub,
+ void *data, __u32 flags)
+{
+ struct ptlrpc_thread *thread = &scrub->os_thread;
+ struct l_wait_info lwi = { 0 };
+ struct task_struct *task;
+ int rc;
+ ENTRY;
+
+again:
+ /* os_lock: sync status between stop and scrub thread */
+ spin_lock(&scrub->os_lock);
+ if (thread_is_running(thread)) {
+ spin_unlock(&scrub->os_lock);
+ RETURN(-EALREADY);
+ }
+
+ if (unlikely(thread_is_stopping(thread))) {
+ spin_unlock(&scrub->os_lock);
+ l_wait_event(thread->t_ctl_waitq,
+ thread_is_stopped(thread),
+ &lwi);
+ goto again;
+ }
+ spin_unlock(&scrub->os_lock);
+
+ if (scrub->os_file.sf_status == SS_COMPLETED) {
+ if (!(flags & SS_SET_FAILOUT))
+ flags |= SS_CLEAR_FAILOUT;
+
+ if (!(flags & SS_SET_DRYRUN))
+ flags |= SS_CLEAR_DRYRUN;
+
+ flags |= SS_RESET;
+ }
+
+ scrub->os_start_flags = flags;
+ thread_set_flags(thread, 0);
+ task = kthread_run(threadfn, data, "OI_scrub");
+ if (IS_ERR(task)) {
+ rc = PTR_ERR(task);
+ CERROR("%s: cannot start iteration thread: rc = %d\n",
+ scrub->os_name, rc);
+ RETURN(rc);
+ }
+
+ l_wait_event(thread->t_ctl_waitq,
+ thread_is_running(thread) || thread_is_stopped(thread),
+ &lwi);
+
+ RETURN(0);
+}
+EXPORT_SYMBOL(scrub_start);
+
+void scrub_stop(struct lustre_scrub *scrub)
+{
+ struct ptlrpc_thread *thread = &scrub->os_thread;
+ struct l_wait_info lwi = { 0 };
+
+ /* os_lock: sync status between stop and scrub thread */
+ spin_lock(&scrub->os_lock);
+ if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
+ thread_set_flags(thread, SVC_STOPPING);
+ spin_unlock(&scrub->os_lock);
+ wake_up_all(&thread->t_ctl_waitq);
+ l_wait_event(thread->t_ctl_waitq,
+ thread_is_stopped(thread),
+ &lwi);
+ /* Do not skip the last lock/unlock, which can guarantee that
+ * the caller cannot return until the OI scrub thread exit. */
+ spin_lock(&scrub->os_lock);
+ }
+ spin_unlock(&scrub->os_lock);
+}
+EXPORT_SYMBOL(scrub_stop);
+
+const char *scrub_status_names[] = {
+ "init",
+ "scanning",
+ "completed",
+ "failed",
+ "stopped",
+ "paused",
+ "crashed",
+ NULL
+};
+
+const char *scrub_flags_names[] = {
+ "recreated",
+ "inconsistent",
+ "auto",
+ "upgrade",
+ NULL
+};
+
+const char *scrub_param_names[] = {
+ "failout",
+ "dryrun",
+ NULL
+};
+
+static void scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
+ const char *prefix)
+{
+ int flag;
+ int i;
+
+ seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
+
+ for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
+ if (flag & bits) {
+ bits &= ~flag;
+ seq_printf(m, "%s%c", names[i],
+ bits != 0 ? ',' : '\n');
+ }
+ }
+}
+
+static void scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
+{
+ if (time != 0)
+ seq_printf(m, "%s: %llu seconds\n", prefix,
+ cfs_time_current_sec() - time);
+ else
+ seq_printf(m, "%s: N/A\n", prefix);
+}
+
+static void scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
+{
+ if (pos != 0)
+ seq_printf(m, "%s: %llu\n", prefix, pos);
+ else
+ seq_printf(m, "%s: N/A\n", prefix);
+}
+
+void scrub_dump(struct seq_file *m, struct lustre_scrub *scrub)
+{
+ struct scrub_file *sf = &scrub->os_file;
+ __u64 checked;
+ __u64 speed;
+
+ down_read(&scrub->os_rwsem);
+ 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]);
+
+ scrub_bits_dump(m, sf->sf_flags, scrub_flags_names, "flags");
+
+ scrub_bits_dump(m, sf->sf_param, scrub_param_names, "param");
+
+ scrub_time_dump(m, sf->sf_time_last_complete,
+ "time_since_last_completed");
+
+ scrub_time_dump(m, sf->sf_time_latest_start,
+ "time_since_latest_start");
+
+ scrub_time_dump(m, sf->sf_time_last_checkpoint,
+ "time_since_last_checkpoint");
+
+ scrub_pos_dump(m, sf->sf_pos_latest_start,
+ "latest_start_position");
+
+ scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
+ "last_checkpoint_position");
+
+ scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
+ "first_failure_position");
+
+ checked = sf->sf_items_checked + scrub->os_new_checked;
+ seq_printf(m, "checked: %llu\n"
+ "%s: %llu\n"
+ "failed: %llu\n"
+ "prior_%s: %llu\n"
+ "noscrub: %llu\n"
+ "igif: %llu\n"
+ "success_count: %u\n",
+ checked,
+ sf->sf_param & SP_DRYRUN ? "inconsistent" : "updated",
+ sf->sf_items_updated, sf->sf_items_failed,
+ sf->sf_param & SP_DRYRUN ? "inconsistent" : "updated",
+ sf->sf_items_updated_prior, sf->sf_items_noscrub,
+ sf->sf_items_igif, sf->sf_success_count);
+
+ speed = checked;
+ if (thread_is_running(&scrub->os_thread)) {
+ cfs_duration_t duration = cfs_time_current() -
+ scrub->os_time_last_checkpoint;
+ __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
+ MSEC_PER_SEC);
+ __u32 rtime = sf->sf_run_time +
+ cfs_duration_sec(duration + HALF_SEC);
+
+ if (duration != 0)
+ do_div(new_checked, duration);
+ if (rtime != 0)
+ do_div(speed, rtime);
+ seq_printf(m, "run_time: %u seconds\n"
+ "average_speed: %llu objects/sec\n"
+ "real-time_speed: %llu objects/sec\n"
+ "current_position: %llu\n"
+ "scrub_in_prior: %s\n"
+ "scrub_full_speed: %s\n"
+ "partial_scan: %s\n",
+ rtime, speed, new_checked, scrub->os_pos_current,
+ scrub->os_in_prior ? "yes" : "no",
+ scrub->os_full_speed ? "yes" : "no",
+ scrub->os_partial_scan ? "yes" : "no");
+ } else {
+ if (sf->sf_run_time != 0)
+ do_div(speed, sf->sf_run_time);
+ seq_printf(m, "run_time: %u seconds\n"
+ "average_speed: %llu objects/sec\n"
+ "real-time_speed: N/A\n"
+ "current_position: N/A\n",
+ sf->sf_run_time, speed);
+ }
+
+ up_read(&scrub->os_rwsem);
+}
+EXPORT_SYMBOL(scrub_dump);
OBD_ALLOC_PTR(mo);
if (mo != NULL) {
struct lu_object *l;
-
- l = &mo->oo_dt.do_lu;
- dt_object_init(&mo->oo_dt, NULL, d);
+ struct lu_object_header *h;
+
+ l = &mo->oo_dt.do_lu;
+ h = &osd_dev(d)->od_scrub.os_scrub.os_obj_header;
+ if (unlikely(fid_is_zero(&h->loh_fid))) {
+ /* For the OI_scrub object during OSD device init. */
+ lu_object_header_init(h);
+ lu_object_init(l, h, d);
+ lu_object_add_top(h, l);
+ } else {
+ dt_object_init(&mo->oo_dt, NULL, d);
+ }
mo->oo_dt.do_ops = &osd_obj_ops;
- l->lo_ops = &osd_lu_obj_ops;
+ l->lo_ops = &osd_lu_obj_ops;
init_rwsem(&mo->oo_sem);
init_rwsem(&mo->oo_ext_idx_sem);
spin_lock_init(&mo->oo_guard);
}
if (IS_ERR(inode)) {
- if (dev->od_scrub.os_file.sf_flags & SF_INCONSISTENT)
+ if (dev->od_scrub.os_scrub.os_file.sf_flags &
+ SF_INCONSISTENT)
/* It still can be the case 2, but we cannot
* distinguish it from the case 1. So return
* -EREMCHG to block current operation until
struct osd_idmap_cache *oic;
struct osd_inode_id *id;
struct inode *inode = NULL;
- struct osd_scrub *scrub;
+ struct lustre_scrub *scrub;
struct scrub_file *sf;
__u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
__u32 saved_ino;
LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
dev = osd_dev(ldev);
- scrub = &dev->od_scrub;
+ scrub = &dev->od_scrub.os_scrub;
sf = &scrub->os_file;
info = osd_oti_get(env);
LASSERT(info);
}
join:
- rc1 = osd_scrub_start(dev, flags);
+ rc1 = osd_scrub_start(env, dev, flags);
LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
"0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
if (rc1 && rc1 != -EALREADY)
}
result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
- obj->oo_dt.do_body_ops = &osd_body_ops_new;
+ if (unlikely(l->lo_header ==
+ &osd_obj2dev(obj)->od_scrub.os_scrub.os_obj_header)) {
+ /* For the OI_scrub object during OSD device init. */
+ l->lo_header->loh_attr |= LOHA_EXISTS;
+ obj->oo_dt.do_body_ops = &osd_body_ops;
+ } else {
+ obj->oo_dt.do_body_ops = &osd_body_ops_new;
+ }
if (result == 0 && obj->oo_inode != NULL) {
struct osd_thread_info *oti = osd_oti_get(env);
struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
*/
static void osd_object_free(const struct lu_env *env, struct lu_object *l)
{
- struct osd_object *obj = osd_obj(l);
+ struct osd_object *obj = osd_obj(l);
+ struct lu_object_header *h = NULL;
- LINVRNT(osd_invariant(obj));
+ LINVRNT(osd_invariant(obj));
+
+ if (unlikely(l->lo_header ==
+ &osd_obj2dev(obj)->od_scrub.os_scrub.os_obj_header))
+ h = l->lo_header;
osd_oxc_fini(obj);
- dt_object_fini(&obj->oo_dt);
- if (obj->oo_hl_head != NULL)
- ldiskfs_htree_lock_head_free(obj->oo_hl_head);
- OBD_FREE_PTR(obj);
+ dt_object_fini(&obj->oo_dt);
+ if (obj->oo_hl_head != NULL)
+ ldiskfs_htree_lock_head_free(obj->oo_hl_head);
+ OBD_FREE_PTR(obj);
+ if (unlikely(h))
+ lu_object_header_fini(h);
}
/*
osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
- if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
+ if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
+ !osd_obj2dev(obj)->od_is_ost) {
struct osd_thread_info *oti = osd_oti_get(env);
const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
struct lu_fid *fid1 = &oti->oti_fid;
osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
struct osd_idmap_cache *oic)
{
- struct osd_scrub *scrub = &dev->od_scrub;
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
struct lu_fid *fid = &oic->oic_fid;
struct osd_inode_id *id = &oic->oic_lid;
struct inode *inode = NULL;
- int once = 0;
+ int once = 0;
bool insert;
int rc;
ENTRY;
if (!fid_is_norm(fid) && !fid_is_igif(fid))
RETURN(0);
+ if (dev->od_noscrub && !thread_is_running(&scrub->os_thread))
+ RETURN(0);
+
if (scrub->os_pos_current > id->oii_ino)
RETURN(0);
}
if (!dev->od_noscrub && ++once == 1) {
- rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
- SS_CLEAR_FAILOUT);
+ rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
+ SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
"%s: trigger partial OI scrub for RPC inconsistency "
"checking FID "DFID": rc = %d\n",
osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
rc = osd_consistency_check(oti, dev, oic);
- if (rc != 0)
+ if (rc == -ENOENT)
fid_zero(&oic->oic_fid);
+ else
+ /* Other error should not affect lookup result. */
+ rc = 0;
} else {
rc = PTR_ERR(bh);
}
ent->oied_namelen);
if (rc == -ENOENT ||
(rc == -ENODATA &&
- !(dev->od_scrub.os_file.sf_flags & SF_UPGRADE))) {
+ !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
/* linkEA does not recognize the dirent entry,
* it may because the dirent entry corruption
* and points to other's inode. */
}
osd_fid_fini(env, o);
+ osd_scrub_cleanup(env, o);
RETURN(0);
}
osd_shutdown(env, o);
osd_procfs_fini(o);
- osd_scrub_cleanup(env, o);
osd_obj_map_fini(o);
osd_umount(env, o);
#define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
struct osd_inconsistent_item {
- /* link into osd_scrub::os_inconsistent_items,
- * protected by osd_scrub::os_lock. */
+ /* link into lustre_scrub::os_inconsistent_items,
+ * protected by lustre_scrub::os_lock. */
struct list_head oii_list;
/* The right FID <=> ino#/gen mapping. */
int ooc_consumer_idx;
/* How many items in ooc_cache. */
- int ooc_cached_items;
+ __u64 ooc_cached_items;
/* Position for up layer LFSCK iteration pre-loading. */
- __u32 ooc_pos_preload;
+ __u64 ooc_pos_preload;
};
struct osd_otable_it {
const struct lu_fid *fid, const struct osd_inode_id *id,
handle_t *th);
-void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags);
-int osd_scrub_file_store(struct osd_scrub *scrub);
char *osd_lf_fid2name(const struct lu_fid *fid);
-int osd_scrub_start(struct osd_device *dev, __u32 flags);
+int osd_scrub_start(const struct lu_env *env, struct osd_device *dev,
+ __u32 flags);
int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
int insert);
int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
struct osd_inode_id *id);
-int osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
+void osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
u64 seq, struct lu_seq_range *range);
if (unlikely(dev->od_mnt == NULL))
return -EINPROGRESS;
- return osd_scrub_dump(m, dev);
+ osd_scrub_dump(m, dev);
+ return 0;
}
LPROC_SEQ_FOPS_RO(ldiskfs_osd_oi_scrub);
/* fid_cpu_to_be() */
#include <lustre_fid.h>
#include <dt_object.h>
+#include <lustre_scrub.h>
#include "osd_oi.h"
/* osd_lookup(), struct osd_thread_info */
#include "osd_internal.h"
-#include "osd_scrub.h"
-static unsigned int osd_oi_count = OSD_OI_FID_NR;
+unsigned int osd_oi_count = OSD_OI_FID_NR;
module_param(osd_oi_count, int, 0444);
MODULE_PARM_DESC(osd_oi_count, "Number of Object Index containers to be created, it's only valid for new filesystem.");
-/** to serialize concurrent OI index initialization */
-static struct mutex oi_init_lock;
-
static struct dt_index_features oi_feat = {
.dif_flags = DT_IND_UPDATE,
.dif_recsize_min = sizeof(struct osd_inode_id),
return PTR_ERR(inode);
}
+ if (osd->od_dt_dev.dd_rdonly)
+ RETURN(-EROFS);
+
jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, 100);
if (IS_ERR(jh))
return PTR_ERR(jh);
if (IS_ERR(inode))
RETURN(PTR_ERR(inode));
- /* 'What the @fid is' is not imporatant, because these objects
- * have no OI mappings, and only are visible inside the OSD.*/
- lu_igif_build(&info->oti_fid, inode->i_ino, inode->i_generation);
- rc = osd_ea_fid_set(info, inode, &info->oti_fid, LMAC_NOT_IN_OI, 0);
- if (rc != 0)
- GOTO(out_inode, rc);
+ if (!osd->od_dt_dev.dd_rdonly) {
+ /* 'What the @fid is' is not imporatant, because these objects
+ * have no OI mappings, and only are visible inside the OSD.*/
+ lu_igif_build(&info->oti_fid, inode->i_ino,
+ inode->i_generation);
+ rc = osd_ea_fid_set(info, inode, &info->oti_fid,
+ LMAC_NOT_IN_OI, 0);
+ if (rc)
+ GOTO(out_inode, rc);
+ }
OBD_ALLOC_PTR(oi);
if (oi == NULL)
osd_oi_table_open(struct osd_thread_info *info, struct osd_device *osd,
struct osd_oi **oi_table, unsigned oi_count, bool create)
{
- struct scrub_file *sf = &osd->od_scrub.os_file;
+ struct scrub_file *sf = &osd->od_scrub.os_scrub.os_file;
int count = 0;
int rc = 0;
int i;
int rc;
int i;
- for (i = 0; i < osd->od_scrub.os_file.sf_oi_count; i++) {
+ if (osd->od_dt_dev.dd_rdonly)
+ RETURN(-EROFS);
+
+ for (i = 0; i < OSD_OI_FID_NR_MAX; i++) {
namelen = snprintf(name, sizeof(name), "%s.%d",
OSD_OI_NAME_BASE, i);
rc = osd_remove_oi_one(osd_sb(osd)->s_root, name, namelen);
int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd,
bool restored)
{
- struct osd_scrub *scrub = &osd->od_scrub;
+ struct lustre_scrub *scrub = &osd->od_scrub.os_scrub;
struct scrub_file *sf = &scrub->os_file;
- struct osd_oi **oi;
- int rc;
+ struct osd_oi **oi;
+ int count;
+ int rc;
ENTRY;
+ if (unlikely(sf->sf_oi_count & (sf->sf_oi_count - 1)) != 0) {
+ LCONSOLE_WARN("%s: Invalid OI count in scrub file %d\n",
+ osd_dev2name(osd), sf->sf_oi_count);
+ sf->sf_oi_count = 0;
+ }
+
if (restored) {
rc = osd_remove_ois(info, osd);
- if (rc != 0)
- return rc;
+ if (rc)
+ RETURN(rc);
}
OBD_ALLOC(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
if (oi == NULL)
RETURN(-ENOMEM);
- mutex_lock(&oi_init_lock);
/* try to open existing multiple OIs first */
- rc = osd_oi_table_open(info, osd, oi, sf->sf_oi_count, false);
- if (rc < 0)
- GOTO(out, rc);
-
- if (rc > 0) {
- if (rc == sf->sf_oi_count || sf->sf_oi_count == 0)
- GOTO(out, rc);
+ count = osd_oi_table_open(info, osd, oi, sf->sf_oi_count, false);
+ if (count < 0)
+ GOTO(out, rc = count);
+
+ if (count > 0) {
+ if (count == sf->sf_oi_count)
+ GOTO(out, rc = count);
+
+ if (sf->sf_oi_count == 0) {
+ if (likely((count & (count - 1)) == 0))
+ GOTO(out, rc = count);
+
+ LCONSOLE_WARN("%s: invalid oi count %d, remove them, "
+ "then set it to %d\n", osd_dev2name(osd),
+ count, osd_oi_count);
+ osd_oi_table_put(info, oi, count);
+ rc = osd_remove_ois(info, osd);
+ if (rc)
+ GOTO(out, rc);
+
+ sf->sf_oi_count = osd_oi_count;
+ }
- osd_scrub_file_reset(scrub,
- LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
- SF_RECREATED);
- osd_oi_count = sf->sf_oi_count;
+ scrub_file_reset(scrub, LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
+ SF_RECREATED);
+ count = sf->sf_oi_count;
goto create;
}
/* if previous failed then try found single OI from old filesystem */
rc = osd_oi_open(info, osd, OSD_OI_NAME_BASE, &oi[0], false);
if (rc == 0) { /* found single OI from old filesystem */
+ count = 1;
ldiskfs_clear_bit(0, sf->sf_oi_bitmap);
if (sf->sf_success_count == 0)
/* XXX: There is one corner case that if the OI_scrub
* and restored after former upgrading from 1.8
* to 2.x. Fortunately, the osd_fid_lookup()can
* verify the inode to decrease the risk. */
- osd_scrub_file_reset(scrub,
- LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
- SF_UPGRADE);
+ scrub_file_reset(scrub,
+ LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
+ SF_UPGRADE);
GOTO(out, rc = 1);
} else if (rc != -ENOENT) {
CERROR("%s: can't open %s: rc = %d\n",
if (sf->sf_oi_count > 0) {
int i;
+ count = sf->sf_oi_count;
memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
- for (i = 0; i < osd_oi_count; i++)
+ for (i = 0; i < count; i++)
ldiskfs_set_bit(i, sf->sf_oi_bitmap);
- osd_scrub_file_reset(scrub,
- LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
- SF_RECREATED);
+ scrub_file_reset(scrub, LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
+ SF_RECREATED);
+ } else {
+ count = sf->sf_oi_count = osd_oi_count;
}
- sf->sf_oi_count = osd_oi_count;
create:
- rc = osd_scrub_file_store(scrub);
+ rc = scrub_file_store(info->oti_env, scrub);
if (rc < 0) {
- osd_oi_table_put(info, oi, sf->sf_oi_count);
+ osd_oi_table_put(info, oi, count);
GOTO(out, rc);
}
/* No OIs exist, new filesystem, create OI objects */
- rc = osd_oi_table_open(info, osd, oi, osd_oi_count, true);
- LASSERT(ergo(rc >= 0, rc == osd_oi_count));
+ rc = osd_oi_table_open(info, osd, oi, count, true);
+ LASSERT(ergo(rc >= 0, rc == count));
GOTO(out, rc);
if (rc < 0) {
OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
} else {
- LASSERT((rc & (rc - 1)) == 0);
+ LASSERTF((rc & (rc - 1)) == 0, "Invalid OI count %d\n", rc);
+
osd->od_oi_table = oi;
osd->od_oi_count = rc;
if (sf->sf_oi_count != rc) {
sf->sf_oi_count = rc;
- rc = osd_scrub_file_store(scrub);
+ rc = scrub_file_store(info->oti_env, scrub);
if (rc < 0) {
- osd_oi_table_put(info, oi, sf->sf_oi_count);
+ osd_oi_table_put(info, oi, count);
OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
}
} else {
}
}
- mutex_unlock(&oi_init_lock);
return rc;
}
void osd_oi_fini(struct osd_thread_info *info, struct osd_device *osd)
{
- if (unlikely(osd->od_oi_table == NULL))
+ if (unlikely(!osd->od_oi_table))
return;
- osd_oi_table_put(info, osd->od_oi_table, osd->od_oi_count);
+ osd_oi_table_put(info, osd->od_oi_table, osd->od_oi_count);
- OBD_FREE(osd->od_oi_table,
- sizeof(*(osd->od_oi_table)) * OSD_OI_FID_NR_MAX);
- osd->od_oi_table = NULL;
+ OBD_FREE(osd->od_oi_table,
+ sizeof(*(osd->od_oi_table)) * OSD_OI_FID_NR_MAX);
+ osd->od_oi_table = NULL;
}
static inline int fid_is_fs_root(const struct lu_fid *fid)
int osd_oi_mod_init(void)
{
- if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
- osd_oi_count = OSD_OI_FID_NR;
+ if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
+ osd_oi_count = OSD_OI_FID_NR;
- if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
- LCONSOLE_WARN("Round up oi_count %d to power2 %d\n",
- osd_oi_count, size_roundup_power2(osd_oi_count));
- osd_oi_count = size_roundup_power2(osd_oi_count);
- }
+ if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
+ LCONSOLE_WARN("Round up oi_count %d to power2 %d\n",
+ osd_oi_count, size_roundup_power2(osd_oi_count));
+ osd_oi_count = size_roundup_power2(osd_oi_count);
+ }
- mutex_init(&oi_init_lock);
- return 0;
+ return 0;
}
#include <lu_object.h>
#include <md_object.h>
-#define OSD_OI_FID_NR (1UL << OSD_OI_FID_OID_BITS)
-#define OSD_OI_FID_NR_MAX (1UL << OSD_OI_FID_OID_BITS_MAX)
-
-#define OSD_OII_NOGEN (0)
+#define OSD_OI_FID_OID_BITS 6
+#define OSD_OI_FID_NR (1UL << OSD_OI_FID_OID_BITS)
+#define OSD_OII_NOGEN (0)
struct lu_fid;
struct osd_thread_info;
OI_KNOWN_ON_OST = 0x00000002,
};
+extern unsigned int osd_oi_count;
+
int osd_oi_mod_init(void);
int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd,
bool restored);
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 only,
* as published by the Free Software Foundation.
-
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * GNU General Public License version 2 for more details. A copy is
- * included in the COPYING file that accompanied this code.
-
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
+ *
* You should have received a copy of the GNU General Public License
- * along with this program; if not, write to the Free Software
- * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ * version 2 along with this program; If not, see
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
#include <lustre_disk.h>
#include <dt_object.h>
#include <linux/xattr.h>
+#include <lustre_scrub.h>
+#include <lustre_nodemap.h>
#include "osd_internal.h"
#include "osd_oi.h"
#include "osd_scrub.h"
-#define HALF_SEC msecs_to_jiffies(MSEC_PER_SEC >> 1)
-
#define OSD_OTABLE_MAX_HASH 0x00000000ffffffffULL
-#define SCRUB_NEXT_BREAK 1 /* exit current loop and process next group */
-#define SCRUB_NEXT_CONTINUE 2 /* skip current object and process next bit */
-#define SCRUB_NEXT_EXIT 3 /* exit all the loops */
-#define SCRUB_NEXT_WAIT 4 /* wait for free cache slot */
-#define SCRUB_NEXT_CRASH 5 /* simulate system crash during OI scrub */
-#define SCRUB_NEXT_FATAL 6 /* simulate failure during OI scrub */
-#define SCRUB_NEXT_NOSCRUB 7 /* new created object, no scrub on it */
-#define SCRUB_NEXT_NOLMA 8 /* the inode has no FID-in-LMA */
-#define SCRUB_NEXT_OSTOBJ 9 /* for OST-object */
-#define SCRUB_NEXT_OSTOBJ_OLD 10 /* old OST-object, no LMA or no FID-on-OST
- * flags in LMA */
-
-/* misc functions */
-
-static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
-{
- return container_of0(scrub, struct osd_device, od_scrub);
-}
-
-static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
-{
- return osd_sb(osd_scrub2dev(scrub));
-}
-
-static inline int osd_scrub_has_window(struct osd_scrub *scrub,
+static inline int osd_scrub_has_window(struct lustre_scrub *scrub,
struct osd_otable_cache *ooc)
{
return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
}
-static inline const char *osd_scrub2name(struct osd_scrub *scrub)
-{
- return osd_dev2name(osd_scrub2dev(scrub));
-}
-
/**
* update/insert/delete the specified OI mapping (@fid @id) according to the ops
*
int rc;
ENTRY;
- if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
+ if (dev->od_scrub.os_scrub.os_file.sf_param & SP_DRYRUN && !force)
RETURN(0);
/* DTO_INDEX_INSERT is enough for other two ops:
RETURN(rc);
}
-/* OI_scrub file ops */
-
-static void osd_scrub_file_to_cpu(struct scrub_file *des,
- struct scrub_file *src)
-{
- memcpy(des->sf_uuid, src->sf_uuid, 16);
- des->sf_flags = le64_to_cpu(src->sf_flags);
- des->sf_magic = le32_to_cpu(src->sf_magic);
- des->sf_status = le16_to_cpu(src->sf_status);
- des->sf_param = le16_to_cpu(src->sf_param);
- des->sf_time_last_complete =
- le64_to_cpu(src->sf_time_last_complete);
- des->sf_time_latest_start =
- le64_to_cpu(src->sf_time_latest_start);
- des->sf_time_last_checkpoint =
- le64_to_cpu(src->sf_time_last_checkpoint);
- des->sf_pos_latest_start =
- le64_to_cpu(src->sf_pos_latest_start);
- des->sf_pos_last_checkpoint =
- le64_to_cpu(src->sf_pos_last_checkpoint);
- des->sf_pos_first_inconsistent =
- le64_to_cpu(src->sf_pos_first_inconsistent);
- des->sf_items_checked =
- le64_to_cpu(src->sf_items_checked);
- des->sf_items_updated =
- le64_to_cpu(src->sf_items_updated);
- des->sf_items_failed =
- le64_to_cpu(src->sf_items_failed);
- des->sf_items_updated_prior =
- le64_to_cpu(src->sf_items_updated_prior);
- des->sf_run_time = le32_to_cpu(src->sf_run_time);
- des->sf_success_count = le32_to_cpu(src->sf_success_count);
- des->sf_oi_count = le16_to_cpu(src->sf_oi_count);
- des->sf_internal_flags = le16_to_cpu(src->sf_internal_flags);
- memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
-}
-
-static void osd_scrub_file_to_le(struct scrub_file *des,
- struct scrub_file *src)
-{
- memcpy(des->sf_uuid, src->sf_uuid, 16);
- des->sf_flags = cpu_to_le64(src->sf_flags);
- des->sf_magic = cpu_to_le32(src->sf_magic);
- des->sf_status = cpu_to_le16(src->sf_status);
- des->sf_param = cpu_to_le16(src->sf_param);
- des->sf_time_last_complete =
- cpu_to_le64(src->sf_time_last_complete);
- des->sf_time_latest_start =
- cpu_to_le64(src->sf_time_latest_start);
- des->sf_time_last_checkpoint =
- cpu_to_le64(src->sf_time_last_checkpoint);
- des->sf_pos_latest_start =
- cpu_to_le64(src->sf_pos_latest_start);
- des->sf_pos_last_checkpoint =
- cpu_to_le64(src->sf_pos_last_checkpoint);
- des->sf_pos_first_inconsistent =
- cpu_to_le64(src->sf_pos_first_inconsistent);
- des->sf_items_checked =
- cpu_to_le64(src->sf_items_checked);
- des->sf_items_updated =
- cpu_to_le64(src->sf_items_updated);
- des->sf_items_failed =
- cpu_to_le64(src->sf_items_failed);
- des->sf_items_updated_prior =
- cpu_to_le64(src->sf_items_updated_prior);
- des->sf_run_time = cpu_to_le32(src->sf_run_time);
- des->sf_success_count = cpu_to_le32(src->sf_success_count);
- des->sf_oi_count = cpu_to_le16(src->sf_oi_count);
- des->sf_internal_flags = cpu_to_le16(src->sf_internal_flags);
- memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
-}
-
-static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
-{
- struct scrub_file *sf = &scrub->os_file;
-
- memset(sf, 0, sizeof(*sf));
- memcpy(sf->sf_uuid, uuid, 16);
- sf->sf_magic = SCRUB_MAGIC_V1;
- sf->sf_status = SS_INIT;
-}
-
-void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
-{
- struct scrub_file *sf = &scrub->os_file;
-
- CDEBUG(D_LFSCK, "%s: reset OI scrub file, old flags = "
- "%#llx, add flags = %#llx\n",
- osd_scrub2name(scrub), sf->sf_flags, flags);
-
- memcpy(sf->sf_uuid, uuid, 16);
- sf->sf_status = SS_INIT;
- sf->sf_flags |= flags;
- sf->sf_flags &= ~SF_AUTO;
- sf->sf_run_time = 0;
- sf->sf_time_latest_start = 0;
- sf->sf_time_last_checkpoint = 0;
- sf->sf_pos_latest_start = 0;
- sf->sf_pos_last_checkpoint = 0;
- sf->sf_pos_first_inconsistent = 0;
- sf->sf_items_checked = 0;
- sf->sf_items_updated = 0;
- sf->sf_items_failed = 0;
- if (!scrub->os_in_join)
- sf->sf_items_updated_prior = 0;
-
- sf->sf_items_noscrub = 0;
- sf->sf_items_igif = 0;
-}
-
-static int osd_scrub_file_load(struct osd_scrub *scrub)
-{
- loff_t pos = 0;
- int len = sizeof(scrub->os_file_disk);
- int rc;
-
- rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
- if (rc == len) {
- struct scrub_file *sf = &scrub->os_file;
-
- osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
- if (sf->sf_magic != SCRUB_MAGIC_V1) {
- CDEBUG(D_LFSCK, "%s: invalid scrub magic "
- "0x%x != 0x%x\n", osd_scrub2name(scrub),
- sf->sf_magic, SCRUB_MAGIC_V1);
- /* Process it as new scrub file. */
- rc = -ENOENT;
- } else {
- rc = 0;
- }
- } else if (rc != 0) {
- CDEBUG(D_LFSCK, "%s: fail to load scrub file, "
- "expected = %d: rc = %d\n",
- osd_scrub2name(scrub), len, rc);
- if (rc > 0)
- rc = -EFAULT;
- } else {
- /* return -ENOENT for empty scrub file case. */
- rc = -ENOENT;
- }
-
- return rc;
-}
-
-int osd_scrub_file_store(struct osd_scrub *scrub)
-{
- struct osd_device *dev;
- handle_t *jh;
- loff_t pos = 0;
- int len = sizeof(scrub->os_file_disk);
- int credits;
- int rc;
-
- dev = container_of0(scrub, struct osd_device, od_scrub);
- credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
- osd_dto_credits_noquota[DTO_WRITE_BLOCK];
- jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
- if (IS_ERR(jh)) {
- rc = PTR_ERR(jh);
- CDEBUG(D_LFSCK, "%s: fail to start trans for scrub store: "
- "rc = %d\n", osd_scrub2name(scrub), rc);
- return rc;
- }
-
- osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
- rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
- len, 0, &pos, jh);
- ldiskfs_journal_stop(jh);
- if (rc != 0)
- CDEBUG(D_LFSCK, "%s: fail to store scrub file, "
- "expected = %d: rc = %d\n",
- osd_scrub2name(scrub), len, rc);
-
- scrub->os_time_last_checkpoint = cfs_time_current();
- scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
- cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
- return rc;
-}
-
static int
osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
struct inode *inode, const struct lu_fid *fid)
bool reset = false;
ENTRY;
- if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
+ if (dev->od_scrub.os_scrub.os_file.sf_param & SP_DRYRUN)
RETURN(0);
if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
struct osd_idmap_cache *oic, int val)
{
- struct osd_scrub *scrub = &dev->od_scrub;
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
struct scrub_file *sf = &scrub->os_file;
struct lu_fid *fid = &oic->oic_fid;
struct osd_inode_id *lid = &oic->oic_lid;
RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
}
-/* OI scrub APIs */
-
-static int osd_scrub_prep(struct osd_device *dev)
+static int osd_scrub_prep(const struct lu_env *env, struct osd_device *dev)
{
- struct osd_scrub *scrub = &dev->od_scrub;
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
struct ptlrpc_thread *thread = &scrub->os_thread;
struct scrub_file *sf = &scrub->os_file;
__u32 flags = scrub->os_start_flags;
}
if (flags & SS_RESET)
- osd_scrub_file_reset(scrub,
+ scrub_file_reset(scrub,
LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
if (flags & SS_AUTO_FULL) {
sf->sf_time_latest_start = cfs_time_current_sec();
sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
- rc = osd_scrub_file_store(scrub);
+ rc = scrub_file_store(env, scrub);
if (rc == 0) {
spin_lock(&scrub->os_lock);
thread_set_flags(thread, SVC_RUNNING);
RETURN(rc);
}
-static int osd_scrub_checkpoint(struct osd_scrub *scrub)
-{
- struct scrub_file *sf = &scrub->os_file;
- int rc;
-
- if (likely(cfs_time_before(cfs_time_current(),
- scrub->os_time_next_checkpoint) ||
- scrub->os_new_checked == 0))
- return 0;
-
- down_write(&scrub->os_rwsem);
- sf->sf_items_checked += scrub->os_new_checked;
- scrub->os_new_checked = 0;
- sf->sf_pos_last_checkpoint = scrub->os_pos_current;
- sf->sf_time_last_checkpoint = cfs_time_current_sec();
- sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
- scrub->os_time_last_checkpoint);
- rc = osd_scrub_file_store(scrub);
- up_write(&scrub->os_rwsem);
-
- return rc;
-}
-
-static int osd_scrub_post(struct osd_scrub *scrub, int result)
+static int osd_scrub_post(const struct lu_env *env, struct osd_device *dev,
+ int result)
{
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
struct scrub_file *sf = &scrub->os_file;
int rc;
ENTRY;
- CDEBUG(D_LFSCK, "%s: OI scrub post, result = %d\n",
+ CDEBUG(D_LFSCK, "%s: OI scrub post with result = %d\n",
osd_scrub2name(scrub), result);
down_write(&scrub->os_rwsem);
}
sf->sf_time_last_checkpoint = cfs_time_current_sec();
if (result > 0) {
- struct osd_device *dev =
- container_of0(scrub, struct osd_device, od_scrub);
-
dev->od_igif_inoi = 1;
dev->od_check_ff = 0;
sf->sf_status = SS_COMPLETED;
}
sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
scrub->os_time_last_checkpoint);
- rc = osd_scrub_file_store(scrub);
+ rc = scrub_file_store(env, scrub);
up_write(&scrub->os_rwsem);
RETURN(rc < 0 ? rc : result);
struct osd_idmap_cache *oic,
bool *noslot, int rc);
-static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
+static int osd_iit_next(struct osd_iit_param *param, __u64 *pos)
{
__u32 offset;
return rc;
if (!has_lma) {
- if (dev->od_scrub.os_convert_igif) {
+ if (dev->od_scrub.os_scrub.os_convert_igif) {
lu_igif_build(fid, inode->i_ino,
inode->i_generation);
if (scrub)
struct osd_iit_param *param,
struct osd_idmap_cache **oic, const bool noslot)
{
- struct osd_scrub *scrub = &dev->od_scrub;
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
struct ptlrpc_thread *thread = &scrub->os_thread;
- struct lu_fid *fid;
- struct osd_inode_id *lid;
- int rc;
+ struct lu_fid *fid;
+ struct osd_inode_id *lid;
+ int rc;
if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
struct l_wait_info lwi;
if (rc != 0)
return rc;
- *oic = &scrub->os_oic;
+ *oic = &dev->od_scrub.os_oic;
fid = &(*oic)->oic_fid;
lid = &(*oic)->oic_lid;
rc = osd_iit_iget(info, dev, fid, lid,
struct osd_idmap_cache **oic, const bool noslot)
{
struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
- struct osd_scrub *scrub = &dev->od_scrub;
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
struct ptlrpc_thread *thread = &scrub->os_thread;
int rc;
}
static inline int
-osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
+osd_scrub_wakeup(struct lustre_scrub *scrub, struct osd_otable_it *it)
{
spin_lock(&scrub->os_lock);
if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
struct osd_iit_param *param,
struct osd_idmap_cache *oic, bool *noslot, int rc)
{
- struct l_wait_info lwi = { 0 };
- struct osd_scrub *scrub = &dev->od_scrub;
- struct scrub_file *sf = &scrub->os_file;
- struct ptlrpc_thread *thread = &scrub->os_thread;
- struct osd_otable_it *it = dev->od_otable_it;
- struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL;
+ struct l_wait_info lwi = { 0 };
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
+ struct scrub_file *sf = &scrub->os_file;
+ struct ptlrpc_thread *thread = &scrub->os_thread;
+ struct osd_otable_it *it = dev->od_otable_it;
+ struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL;
switch (rc) {
case SCRUB_NEXT_NOSCRUB:
return rc;
}
- rc = osd_scrub_checkpoint(scrub);
- if (rc != 0) {
- CDEBUG(D_LFSCK, "%s: fail to checkpoint, pos = %u: "
+ rc = scrub_checkpoint(info->oti_env, scrub);
+ if (rc) {
+ CDEBUG(D_LFSCK, "%s: fail to checkpoint, pos = %llu: "
"rc = %d\n", osd_scrub2name(scrub),
scrub->os_pos_current, rc);
/* Continue, as long as the scrub itself can go ahead. */
#define SCRUB_IT_ALL 1
#define SCRUB_IT_CRASH 2
-static void osd_scrub_join(struct osd_device *dev, __u32 flags,
- bool inconsistent)
+static void osd_scrub_join(const struct lu_env *env, struct osd_device *dev,
+ __u32 flags, bool inconsistent)
{
- struct osd_scrub *scrub = &dev->od_scrub;
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
struct ptlrpc_thread *thread = &scrub->os_thread;
struct scrub_file *sf = &scrub->os_file;
int rc;
sf->sf_param &= ~SP_DRYRUN;
if (flags & SS_RESET) {
- osd_scrub_file_reset(scrub,
- LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
+ scrub_file_reset(scrub, LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
inconsistent ? SF_INCONSISTENT : 0);
sf->sf_status = SS_SCANNING;
}
sf->sf_time_latest_start = cfs_time_current_sec();
sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
- rc = osd_scrub_file_store(scrub);
- if (rc != 0)
- CDEBUG(D_LFSCK, "%s: fail to store scrub file when join "
- "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
+ rc = scrub_file_store(env, scrub);
spin_lock(&scrub->os_lock);
scrub->os_waiting = 0;
wake_up_all(&thread->t_ctl_waitq);
up_write(&scrub->os_rwsem);
+ CDEBUG(D_LFSCK, "%s: joined in the OI scrub with flag %u: rc = %d\n",
+ osd_scrub2name(scrub), flags, rc);
+
EXIT;
}
static int osd_inode_iteration(struct osd_thread_info *info,
struct osd_device *dev, __u32 max, bool preload)
{
- struct osd_scrub *scrub = &dev->od_scrub;
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
struct ptlrpc_thread *thread = &scrub->os_thread;
- struct scrub_file *sf = &scrub->os_file;
- osd_iit_next_policy next;
- osd_iit_exec_policy exec;
- __u32 *pos;
- __u32 *count;
+ struct scrub_file *sf = &scrub->os_file;
+ osd_iit_next_policy next;
+ osd_iit_exec_policy exec;
+ __u64 *pos;
+ __u64 *count;
struct osd_iit_param *param;
- struct l_wait_info lwi = { 0 };
- __u32 limit =
- le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count);
- int rc;
- bool noslot = true;
+ struct l_wait_info lwi = { 0 };
+ __u32 limit;
+ int rc;
+ bool noslot = true;
ENTRY;
if (preload)
goto full;
- param = &scrub->os_iit_param;
+ param = &dev->od_scrub.os_iit_param;
memset(param, 0, sizeof(*param));
param->sb = osd_sb(dev);
if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
scrub->os_full_scrub) {
- osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
- true);
+ osd_scrub_join(info->oti_env, dev,
+ SS_AUTO_FULL | SS_RESET, true);
goto full;
}
* mappings during the partial scan,
* then scan the device completely. */
if (used < dev->od_full_scrub_ratio) {
- osd_scrub_join(dev,
+ osd_scrub_join(info->oti_env, dev,
SS_AUTO_FULL | SS_RESET, true);
goto full;
}
}
rc = 0;
+ limit = le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count);
while (*pos <= limit && *count < max) {
struct ldiskfs_group_desc *desc;
bool next_group = false;
static int osd_otable_it_preload(const struct lu_env *env,
struct osd_otable_it *it)
{
- struct osd_device *dev = it->ooi_dev;
- struct osd_scrub *scrub = &dev->od_scrub;
+ struct osd_device *dev = it->ooi_dev;
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
struct osd_otable_cache *ooc = &it->ooi_cache;
int rc;
ENTRY;
static int osd_scrub_main(void *args)
{
- struct lu_env env;
- struct osd_device *dev = (struct osd_device *)args;
- struct osd_scrub *scrub = &dev->od_scrub;
+ struct lu_env env;
+ struct osd_device *dev = (struct osd_device *)args;
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
struct ptlrpc_thread *thread = &scrub->os_thread;
- int rc;
+ int rc;
ENTRY;
- rc = lu_env_init(&env, LCT_LOCAL);
+ rc = lu_env_init(&env, LCT_LOCAL | LCT_DT_THREAD);
if (rc != 0) {
CDEBUG(D_LFSCK, "%s: OI scrub fail to init env: rc = %d\n",
osd_scrub2name(scrub), rc);
GOTO(noenv, rc);
}
- rc = osd_scrub_prep(dev);
+ rc = osd_scrub_prep(&env, dev);
if (rc != 0) {
CDEBUG(D_LFSCK, "%s: OI scrub fail to scrub prep: rc = %d\n",
osd_scrub2name(scrub), rc);
scrub->os_pos_current = ooc->ooc_pos_preload;
}
- CDEBUG(D_LFSCK, "%s: OI scrub start, flags = 0x%x, pos = %u\n",
+ CDEBUG(D_LFSCK, "%s: OI scrub start, flags = 0x%x, pos = %llu\n",
osd_scrub2name(scrub), scrub->os_start_flags,
scrub->os_pos_current);
rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
- if (unlikely(rc == SCRUB_IT_CRASH))
+ if (unlikely(rc == SCRUB_IT_CRASH)) {
+ spin_lock(&scrub->os_lock);
+ thread_set_flags(&scrub->os_thread, SVC_STOPPING);
+ spin_unlock(&scrub->os_lock);
GOTO(out, rc = -EINVAL);
+ }
+
GOTO(post, rc);
post:
- rc = osd_scrub_post(scrub, rc);
- CDEBUG(D_LFSCK, "%s: OI scrub: stop, pos = %u: rc = %d\n",
+ rc = osd_scrub_post(&env, dev, rc);
+ CDEBUG(D_LFSCK, "%s: OI scrub: stop, pos = %llu: rc = %d\n",
osd_scrub2name(scrub), scrub->os_pos_current, rc);
out:
struct osd_inconsistent_item *oii;
oii = list_entry(scrub->os_inconsistent_items.next,
- struct osd_inconsistent_item, oii_list);
+ struct osd_inconsistent_item, oii_list);
list_del_init(&oii->oii_list);
OBD_FREE_PTR(oii);
}
osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
struct dentry *dentry, filldir_t filldir);
-enum osd_lf_flags {
- OLF_SCAN_SUBITEMS = 0x0001,
- OLF_HIDE_FID = 0x0002,
- OLF_SHOW_NAME = 0x0004,
- OLF_NO_OI = 0x0008,
- OLF_IDX_IN_FID = 0x0010,
-};
-
struct osd_lf_map {
char *olm_name;
struct lu_fid olm_fid;
/* LFSCK */
{
.olm_name = LFSCK_DIR,
+ .olm_flags = OLF_SCAN_SUBITEMS,
.olm_namelen = sizeof(LFSCK_DIR) - 1,
.olm_scandir = osd_ios_general_scan,
.olm_filldir = osd_ios_varfid_fill,
.olm_filldir = osd_ios_lf_fill,
},
+ /* hsm_actions */
+ {
+ .olm_name = HSM_ACTIONS,
+ },
+
+ /* nodemap */
+ {
+ .olm_name = LUSTRE_NODEMAP_NAME,
+ },
+
{
.olm_name = NULL
}
},
.olm_namelen = sizeof("fid") - 1,
},
+
/* .lustre/lost+found */
{
.olm_name = "lost+found",
},
.olm_namelen = sizeof("lost+found") - 1,
},
+
{
.olm_name = NULL
}
struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
struct osd_inode_id *id = &info->oti_id;
struct osd_inode_id *id2 = &info->oti_id2;
- struct osd_scrub *scrub = &dev->od_scrub;
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
struct scrub_file *sf = &scrub->os_file;
struct lu_fid tfid;
int rc;
RETURN(0);
if (!(sf->sf_flags & SF_INCONSISTENT)) {
- osd_scrub_file_reset(scrub,
- LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
- SF_INCONSISTENT);
- rc = osd_scrub_file_store(scrub);
+ scrub_file_reset(scrub, LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
+ SF_INCONSISTENT);
+ rc = scrub_file_store(info->oti_env, scrub);
if (rc != 0)
RETURN(rc);
}
osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
struct dentry *dentry, filldir_t filldir)
{
- struct osd_scrub *scrub = &dev->od_scrub;
- struct scrub_file *sf = &scrub->os_file;
- struct dentry *child;
- int rc;
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
+ struct scrub_file *sf = &scrub->os_file;
+ struct dentry *child;
+ int rc;
ENTRY;
/* It is existing MDT0 device. We only allow the case of object without
if (rc == -ENOENT) {
/* It is 1.8 MDT device. */
if (!(sf->sf_flags & SF_UPGRADE)) {
- osd_scrub_file_reset(scrub,
+ scrub_file_reset(scrub,
LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
SF_UPGRADE);
sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
- rc = osd_scrub_file_store(scrub);
+ rc = scrub_file_store(info->oti_env, scrub);
} else {
rc = 0;
}
osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
struct dentry *dentry, filldir_t filldir)
{
- struct osd_scrub *scrub = &dev->od_scrub;
- struct scrub_file *sf = &scrub->os_file;
- struct dentry *child;
- int rc;
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
+ struct scrub_file *sf = &scrub->os_file;
+ struct dentry *child;
+ int rc;
ENTRY;
if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
- rc = osd_scrub_file_store(scrub);
+ rc = scrub_file_store(info->oti_env, scrub);
if (rc != 0)
RETURN(rc);
}
/* OI scrub start/stop */
-static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
+int osd_scrub_start(const struct lu_env *env, 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;
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
+ int rc;
ENTRY;
if (dev->od_dt_dev.dd_rdonly)
RETURN(-EROFS);
- /* os_lock: sync status between stop and scrub thread */
- spin_lock(&scrub->os_lock);
-
-again:
- if (thread_is_running(thread)) {
- spin_unlock(&scrub->os_lock);
- if (!(scrub->os_file.sf_flags & SF_AUTO ||
- scrub->os_partial_scan) ||
- (flags & SS_AUTO_PARTIAL))
- RETURN(-EALREADY);
-
- osd_scrub_join(dev, flags, false);
- spin_lock(&scrub->os_lock);
- if (!thread_is_running(thread))
- goto again;
-
- spin_unlock(&scrub->os_lock);
- RETURN(0);
- }
-
- if (unlikely(thread_is_stopping(thread))) {
- spin_unlock(&scrub->os_lock);
- l_wait_event(thread->t_ctl_waitq,
- thread_is_stopped(thread),
- &lwi);
- spin_lock(&scrub->os_lock);
- goto again;
- }
- spin_unlock(&scrub->os_lock);
-
- if (scrub->os_file.sf_status == SS_COMPLETED) {
- if (!(flags & SS_SET_FAILOUT))
- flags |= SS_CLEAR_FAILOUT;
-
- if (!(flags & SS_SET_DRYRUN))
- flags |= SS_CLEAR_DRYRUN;
-
- flags |= SS_RESET;
- }
-
- scrub->os_start_flags = flags;
- thread_set_flags(thread, 0);
- task = kthread_run(osd_scrub_main, dev, "OI_scrub");
- if (IS_ERR(task)) {
- rc = PTR_ERR(task);
- CERROR("%s: cannot start iteration thread: rc = %d\n",
- osd_scrub2name(scrub), rc);
- RETURN(rc);
- }
-
- l_wait_event(thread->t_ctl_waitq,
- thread_is_running(thread) || thread_is_stopped(thread),
- &lwi);
-
- RETURN(0);
-}
-
-int osd_scrub_start(struct osd_device *dev, __u32 flags)
-{
- int rc;
- ENTRY;
-
/* od_otable_mutex: prevent curcurrent start/stop */
mutex_lock(&dev->od_otable_mutex);
- rc = do_osd_scrub_start(dev, flags);
+ rc = scrub_start(osd_scrub_main, scrub, dev, flags);
+ if (rc == -EALREADY) {
+ rc = 0;
+ if ((scrub->os_file.sf_flags & SF_AUTO ||
+ scrub->os_partial_scan) &&
+ !(flags & SS_AUTO_PARTIAL))
+ osd_scrub_join(env, dev, flags, false);
+ }
mutex_unlock(&dev->od_otable_mutex);
- RETURN(rc == -EALREADY ? 0 : rc);
-}
-
-static void do_osd_scrub_stop(struct osd_scrub *scrub)
-{
- struct ptlrpc_thread *thread = &scrub->os_thread;
- struct l_wait_info lwi = { 0 };
-
- /* os_lock: sync status between stop and scrub thread */
- spin_lock(&scrub->os_lock);
- if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
- thread_set_flags(thread, SVC_STOPPING);
- spin_unlock(&scrub->os_lock);
- wake_up_all(&thread->t_ctl_waitq);
- l_wait_event(thread->t_ctl_waitq,
- thread_is_stopped(thread),
- &lwi);
- /* Do not skip the last lock/unlock, which can guarantee that
- * the caller cannot return until the OI scrub thread exit. */
- spin_lock(&scrub->os_lock);
- }
- spin_unlock(&scrub->os_lock);
+ RETURN(rc);
}
static void osd_scrub_stop(struct osd_device *dev)
{
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
+
/* od_otable_mutex: prevent curcurrent start/stop */
mutex_lock(&dev->od_otable_mutex);
- dev->od_scrub.os_paused = 1;
- do_osd_scrub_stop(&dev->od_scrub);
+ scrub->os_paused = 1;
+ scrub_stop(scrub);
mutex_unlock(&dev->od_otable_mutex);
}
int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
{
- struct osd_thread_info *info = osd_oti_get(env);
- struct osd_scrub *scrub = &dev->od_scrub;
- struct lvfs_run_ctxt *ctxt = &scrub->os_ctxt;
- struct scrub_file *sf = &scrub->os_file;
- struct super_block *sb = osd_sb(dev);
- struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es;
- struct lvfs_run_ctxt saved;
- struct file *filp;
- struct inode *inode;
- struct lu_fid *fid = &info->oti_fid;
- bool dirty = false;
- bool restored = false;
- int rc = 0;
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
+ struct lvfs_run_ctxt *ctxt = &dev->od_scrub.os_ctxt;
+ struct scrub_file *sf = &scrub->os_file;
+ struct super_block *sb = osd_sb(dev);
+ struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es;
+ struct lvfs_run_ctxt saved;
+ struct file *filp;
+ struct inode *inode;
+ struct lu_fid *fid = &info->oti_fid;
+ struct lu_object_conf conf;
+ struct dt_object *obj;
+ bool dirty = false;
+ bool restored = false;
+ int rc = 0;
ENTRY;
- memset(scrub, 0, sizeof(*scrub));
+ memset(&dev->od_scrub, 0, sizeof(struct osd_scrub));
OBD_SET_CTXT_MAGIC(ctxt);
ctxt->pwdmnt = dev->od_mnt;
ctxt->pwd = dev->od_mnt->mnt_root;
init_rwsem(&scrub->os_rwsem);
spin_lock_init(&scrub->os_lock);
INIT_LIST_HEAD(&scrub->os_inconsistent_items);
+ scrub->os_name = osd_name(dev);
push_ctxt(&saved, ctxt);
- filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
+ filp = filp_open(osd_scrub_name, O_RDWR |
+ (dev->od_dt_dev.dd_rdonly ? 0 : O_CREAT), 0644);
if (IS_ERR(filp)) {
pop_ctxt(&saved, ctxt);
RETURN(PTR_ERR(filp));
}
inode = file_inode(filp);
- /* '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, NULL);
- pop_ctxt(&saved, ctxt);
- RETURN(rc);
+ if (!dev->od_dt_dev.dd_rdonly) {
+ /* '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) {
+ filp_close(filp, NULL);
+ pop_ctxt(&saved, ctxt);
+ RETURN(rc);
+ }
}
- scrub->os_inode = igrab(inode);
+ igrab(inode);
filp_close(filp, NULL);
pop_ctxt(&saved, ctxt);
- rc = osd_scrub_file_load(scrub);
- if (rc == -ENOENT) {
- osd_scrub_file_init(scrub, es->s_uuid);
+ conf.loc_flags = LOC_F_NEW;
+ obj = lu2dt(lu_object_find_slice(env, osd2lu_dev(dev), fid, &conf));
+ if (IS_ERR_OR_NULL(obj)) {
+ iput(inode);
+ RETURN(obj == NULL ? -ENOENT : PTR_ERR(obj));
+ }
+
+ osd_dt_obj(obj)->oo_inode = inode;
+ scrub->os_obj = obj;
+ rc = scrub_file_load(env, scrub);
+ if (rc == -ENOENT || rc == -EFAULT) {
+ scrub_file_init(scrub, es->s_uuid);
/* If the "/O" dir does not exist when mount (indicated by
* osd_device::od_maybe_new), neither for the "/OI_scrub",
* then it is quite probably that the device is a new one,
* need to check "filter_fid_old" and to convert it to
* "filter_fid" for each object, and all the IGIF should
* have their FID mapping in OI files already. */
- if (dev->od_maybe_new)
+ if (dev->od_maybe_new && rc == -ENOENT)
sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
dirty = true;
- } else if (rc != 0) {
- GOTO(cleanup_inode, rc);
+ } else if (rc < 0) {
+ GOTO(cleanup_obj, rc);
} else {
if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
struct obd_uuid *old_uuid;
} else {
class_uuid_unparse(sf->sf_uuid, old_uuid);
class_uuid_unparse(es->s_uuid, new_uuid);
- CERROR("%s: UUID has been changed from "
- "%s to %s\n", osd_dev2name(dev),
+ CDEBUG(D_LFSCK, "%s: UUID has been changed "
+ "from %s to %s\n", osd_dev2name(dev),
old_uuid->uuid, new_uuid->uuid);
}
- osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
+ scrub_file_reset(scrub, es->s_uuid, SF_INCONSISTENT);
dirty = true;
restored = true;
if (old_uuid != NULL)
sf->sf_status = SS_CRASHED;
dirty = true;
}
+
+ if ((sf->sf_oi_count & (sf->sf_oi_count - 1)) != 0) {
+ LCONSOLE_WARN("%s: invalid oi count %d, set it to %d\n",
+ osd_dev2name(dev), sf->sf_oi_count,
+ osd_oi_count);
+ sf->sf_oi_count = osd_oi_count;
+ dirty = true;
+ }
}
if (sf->sf_pos_last_checkpoint != 0)
scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
if (dirty) {
- rc = osd_scrub_file_store(scrub);
- if (rc != 0)
- GOTO(cleanup_inode, rc);
+ rc = scrub_file_store(env, scrub);
+ if (rc)
+ GOTO(cleanup_obj, rc);
}
/* Initialize OI files. */
rc = osd_oi_init(info, dev, restored);
if (rc < 0)
- GOTO(cleanup_inode, rc);
+ GOTO(cleanup_obj, rc);
- rc = osd_initial_OI_scrub(info, dev);
- if (rc != 0)
- GOTO(cleanup_oi, rc);
+ if (!dev->od_dt_dev.dd_rdonly) {
+ rc = osd_initial_OI_scrub(info, dev);
+ if (rc)
+ GOTO(cleanup_oi, rc);
+ }
if (sf->sf_flags & SF_UPGRADE ||
!(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
(sf->sf_status == SS_INIT &&
sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
SF_UPGRADE))))
- rc = osd_scrub_start(dev, SS_AUTO_FULL);
+ rc = osd_scrub_start(env, dev, SS_AUTO_FULL);
if (rc != 0)
GOTO(cleanup_oi, rc);
RETURN(0);
cleanup_oi:
osd_oi_fini(info, dev);
-cleanup_inode:
- iput(scrub->os_inode);
- scrub->os_inode = NULL;
+cleanup_obj:
+ dt_object_put_nocache(env, scrub->os_obj);
+ scrub->os_obj = NULL;
return rc;
}
void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
{
- struct osd_scrub *scrub = &dev->od_scrub;
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
LASSERT(dev->od_otable_it == NULL);
- if (scrub->os_inode != NULL) {
+ if (scrub->os_obj != NULL) {
osd_scrub_stop(dev);
- iput(scrub->os_inode);
- scrub->os_inode = NULL;
+ dt_object_put_nocache(env, scrub->os_obj);
+ scrub->os_obj = NULL;
}
if (dev->od_oi_table != NULL)
osd_oi_fini(osd_oti_get(env), dev);
enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
- struct osd_scrub *scrub = &dev->od_scrub;
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
struct osd_otable_it *it;
__u32 start = 0;
int rc;
start |= SS_CLEAR_DRYRUN;
}
- rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
- if (rc < 0 && rc != -EALREADY) {
+ rc = scrub_start(osd_scrub_main, scrub, dev, start & ~SS_AUTO_PARTIAL);
+ if (rc == -EALREADY) {
+ it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
+ } else if (rc < 0) {
dev->od_otable_it = NULL;
OBD_FREE_PTR(it);
- GOTO(out, it = ERR_PTR(rc));
+ it = ERR_PTR(rc);
+ } else {
+ /* We have to start from the begining. */
+ it->ooi_cache.ooc_pos_preload =
+ LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
}
- it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
-
GOTO(out, it);
out:
/* od_otable_mutex: prevent curcurrent init/fini */
mutex_lock(&dev->od_otable_mutex);
- do_osd_scrub_stop(&dev->od_scrub);
+ scrub_stop(&dev->od_scrub.os_scrub);
LASSERT(dev->od_otable_it == it);
dev->od_otable_it = NULL;
}
static inline int
-osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
+osd_otable_it_wakeup(struct lustre_scrub *scrub, struct osd_otable_it *it)
{
spin_lock(&scrub->os_lock);
if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
{
struct osd_otable_it *it = (struct osd_otable_it *)di;
struct osd_device *dev = it->ooi_dev;
- struct osd_scrub *scrub = &dev->od_scrub;
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
struct osd_otable_cache *ooc = &it->ooi_cache;
struct ptlrpc_thread *thread = &scrub->os_thread;
struct l_wait_info lwi = { 0 };
struct osd_otable_it *it = (struct osd_otable_it *)di;
struct osd_device *dev = it->ooi_dev;
struct osd_otable_cache *ooc = &it->ooi_cache;
- struct osd_scrub *scrub = &dev->od_scrub;
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
struct osd_iit_param *param = &it->ooi_iit_param;
int rc;
ENTRY;
int insert)
{
struct osd_inconsistent_item *oii;
- struct osd_scrub *scrub = &dev->od_scrub;
- struct ptlrpc_thread *thread = &scrub->os_thread;
- int wakeup = 0;
+ struct osd_scrub *oscrub = &dev->od_scrub;
+ struct lustre_scrub *lscrub = &oscrub->os_scrub;
+ struct ptlrpc_thread *thread = &lscrub->os_thread;
+ int wakeup = 0;
ENTRY;
OBD_ALLOC_PTR(oii);
oii->oii_cache = *oic;
oii->oii_insert = insert;
- if (scrub->os_partial_scan) {
+ if (lscrub->os_partial_scan) {
__u64 now = cfs_time_current_sec();
/* If there haven't been errors in a long time,
* decay old count until either the errors are
* gone or we reach the current interval. */
- while (unlikely(scrub->os_bad_oimap_count > 0 &&
- scrub->os_bad_oimap_time +
+ while (unlikely(oscrub->os_bad_oimap_count > 0 &&
+ oscrub->os_bad_oimap_time +
SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
- scrub->os_bad_oimap_count >>= 1;
- scrub->os_bad_oimap_time +=
+ oscrub->os_bad_oimap_count >>= 1;
+ oscrub->os_bad_oimap_time +=
SCRUB_BAD_OIMAP_DECAY_INTERVAL;
}
- scrub->os_bad_oimap_time = now;
- if (++scrub->os_bad_oimap_count >
+ oscrub->os_bad_oimap_time = now;
+ if (++oscrub->os_bad_oimap_count >
dev->od_full_scrub_threshold_rate)
- scrub->os_full_scrub = 1;
+ lscrub->os_full_scrub = 1;
}
- spin_lock(&scrub->os_lock);
+ spin_lock(&lscrub->os_lock);
if (unlikely(!thread_is_running(thread))) {
- spin_unlock(&scrub->os_lock);
+ spin_unlock(&lscrub->os_lock);
OBD_FREE_PTR(oii);
RETURN(-EAGAIN);
}
- if (list_empty(&scrub->os_inconsistent_items))
+ if (list_empty(&lscrub->os_inconsistent_items))
wakeup = 1;
- list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
- spin_unlock(&scrub->os_lock);
+ list_add_tail(&oii->oii_list, &lscrub->os_inconsistent_items);
+ spin_unlock(&lscrub->os_lock);
if (wakeup != 0)
wake_up_all(&thread->t_ctl_waitq);
int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
struct osd_inode_id *id)
{
- struct osd_scrub *scrub = &dev->od_scrub;
+ struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
struct osd_inconsistent_item *oii;
ENTRY;
RETURN(-ENOENT);
}
-/* OI scrub dump */
-
-static const char *scrub_status_names[] = {
- "init",
- "scanning",
- "completed",
- "failed",
- "stopped",
- "paused",
- "crashed",
- NULL
-};
-
-static const char *scrub_flags_names[] = {
- "recreated",
- "inconsistent",
- "auto",
- "upgrade",
- NULL
-};
-
-static const char *scrub_param_names[] = {
- "failout",
- "dryrun",
- NULL
-};
-
-static void scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
- const char *prefix)
-{
- int flag;
- int i;
-
- seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
-
- for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
- if (flag & bits) {
- bits &= ~flag;
- seq_printf(m, "%s%c", names[i],
- bits != 0 ? ',' : '\n');
- }
- }
-}
-
-static void scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
-{
- if (time != 0)
- seq_printf(m, "%s: %llu seconds\n", prefix,
- cfs_time_current_sec() - time);
- else
- seq_printf(m, "%s: N/A\n", prefix);
-}
-
-static void scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
+void osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
{
- if (pos != 0)
- seq_printf(m, "%s: %llu\n", prefix, pos);
- else
- seq_printf(m, "%s: N/A\n", prefix);
-}
+ struct osd_scrub *scrub = &dev->od_scrub;
-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;
-
- down_read(&scrub->os_rwsem);
- 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]);
-
- scrub_bits_dump(m, sf->sf_flags, scrub_flags_names, "flags");
-
- scrub_bits_dump(m, sf->sf_param, scrub_param_names, "param");
-
- scrub_time_dump(m, sf->sf_time_last_complete,
- "time_since_last_completed");
-
- scrub_time_dump(m, sf->sf_time_latest_start,
- "time_since_latest_start");
-
- scrub_time_dump(m, sf->sf_time_last_checkpoint,
- "time_since_last_checkpoint");
-
- scrub_pos_dump(m, sf->sf_pos_latest_start,
- "latest_start_position");
-
- scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
- "last_checkpoint_position");
-
- scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
- "first_failure_position");
-
- checked = sf->sf_items_checked + scrub->os_new_checked;
- seq_printf(m, "checked: %llu\n"
- "%s: %llu\n"
- "failed: %llu\n"
- "prior_%s: %llu\n"
- "noscrub: %llu\n"
- "igif: %llu\n"
- "success_count: %u\n",
- checked,
- sf->sf_param & SP_DRYRUN ? "inconsistent" : "updated",
- sf->sf_items_updated, sf->sf_items_failed,
- sf->sf_param & SP_DRYRUN ? "inconsistent" : "updated",
- sf->sf_items_updated_prior, sf->sf_items_noscrub,
- sf->sf_items_igif, sf->sf_success_count);
-
- speed = checked;
- if (thread_is_running(&scrub->os_thread)) {
- cfs_duration_t duration = cfs_time_current() -
- scrub->os_time_last_checkpoint;
- __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
- MSEC_PER_SEC);
- __u32 rtime = sf->sf_run_time +
- cfs_duration_sec(duration + HALF_SEC);
-
- if (duration != 0)
- do_div(new_checked, duration);
- if (rtime != 0)
- do_div(speed, rtime);
- seq_printf(m, "run_time: %u seconds\n"
- "average_speed: %llu objects/sec\n"
- "real-time_speed: %llu objects/sec\n"
- "current_position: %u\n"
- "lf_scanned: %llu\n"
- "lf_%s: %llu\n"
- "lf_failed: %llu\n",
- rtime, speed, new_checked, scrub->os_pos_current,
- scrub->os_lf_scanned,
- sf->sf_param & SP_DRYRUN ?
- "inconsistent" : "repaired",
- scrub->os_lf_repaired,
- scrub->os_lf_failed);
- seq_printf(m, "inodes_per_group: %lu\n"
- "current_iit_group: %u\n"
- "current_iit_base: %u\n"
- "current_iit_offset: %u\n"
- "scrub_in_prior: %s\n"
- "scrub_full_speed: %s\n"
- "partial_scan: %s\n",
- LDISKFS_INODES_PER_GROUP(osd_sb(dev)),
- scrub->os_iit_param.bg,
- scrub->os_iit_param.gbase,
- scrub->os_iit_param.offset,
- scrub->os_in_prior ? "yes" : "no",
- scrub->os_full_speed ? "yes" : "no",
- scrub->os_partial_scan ? "yes" : "no");
- } else {
- if (sf->sf_run_time != 0)
- do_div(speed, sf->sf_run_time);
- seq_printf(m, "run_time: %u seconds\n"
- "average_speed: %llu objects/sec\n"
- "real-time_speed: N/A\n"
- "current_position: N/A\n"
- "lf_scanned: %llu\n"
- "lf_%s: %llu\n"
- "lf_failed: %llu\n",
- sf->sf_run_time, speed, scrub->os_lf_scanned,
- sf->sf_param & SP_DRYRUN ?
- "inconsistent" : "repaired",
- scrub->os_lf_repaired, scrub->os_lf_failed);
- }
-
- up_read(&scrub->os_rwsem);
- return 0;
+ scrub_dump(m, &scrub->os_scrub);
+ seq_printf(m, "lf_scanned: %llu\n"
+ "lf_%s: %llu\n"
+ "lf_failed: %llu\n",
+ scrub->os_lf_scanned,
+ scrub->os_scrub.os_file.sf_param & SP_DRYRUN ?
+ "inconsistent" : "repaired",
+ scrub->os_lf_repaired,
+ scrub->os_lf_failed);
}
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 only,
* as published by the Free Software Foundation.
-
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * GNU General Public License version 2 for more details. A copy is
- * included in the COPYING file that accompanied this code.
-
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
+ *
* You should have received a copy of the GNU General Public License
- * along with this program; if not, write to the Free Software
- * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ * version 2 along with this program; If not, see
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
/*
* lustre/osd-ldiskfs/osd_scrub.h
*
- * Shared definitions and declarations for OI scrub.
+ * Definitions and declarations for ldiskfs backend OI scrub.
*
* Author: Fan Yong <yong.fan@whamcloud.com>
*/
#ifndef _OSD_SCRUB_H
# define _OSD_SCRUB_H
+#include <lustre_scrub.h>
#include "osd_oi.h"
-#define SCRUB_MAGIC_V1 0x4C5FD252
-#define SCRUB_CHECKPOINT_INTERVAL 60
-#define SCRUB_OI_BITMAP_SIZE (OSD_OI_FID_NR_MAX >> 3)
-#define SCRUB_WINDOW_SIZE 1024
-
-enum scrub_status {
- /* The scrub file is new created, for new MDT, upgrading from old disk,
- * or re-creating the scrub file manually. */
- SS_INIT = 0,
-
- /* The scrub is checking/repairing the OI files. */
- SS_SCANNING = 1,
-
- /* The scrub checked/repaired the OI files successfully. */
- SS_COMPLETED = 2,
-
- /* The scrub failed to check/repair the OI files. */
- SS_FAILED = 3,
-
- /* The scrub is stopped manually, the OI files may be inconsistent. */
- SS_STOPPED = 4,
-
- /* The scrub is paused automatically when umount. */
- SS_PAUSED = 5,
-
- /* The scrub crashed during the scanning, should be restarted. */
- SS_CRASHED = 6,
-};
-
-enum scrub_flags {
- /* OI files have been recreated, OI mappings should be re-inserted. */
- SF_RECREATED = 0x0000000000000001ULL,
-
- /* OI files are invalid, should be rebuild ASAP */
- SF_INCONSISTENT = 0x0000000000000002ULL,
-
- /* OI scrub is triggered automatically. */
- SF_AUTO = 0x0000000000000004ULL,
-
- /* The device is upgraded from 1.8 format. */
- SF_UPGRADE = 0x0000000000000008ULL,
-};
-
-enum scrub_param {
- /* Exit when fail. */
- SP_FAILOUT = 0x0001,
-
- /* Check only without repairing. */
- SP_DRYRUN = 0x0002,
-};
-
-enum scrub_start {
- /* Set failout flag. */
- SS_SET_FAILOUT = 0x00000001,
-
- /* Clear failout flag. */
- SS_CLEAR_FAILOUT = 0x00000002,
-
- /* Reset scrub start position. */
- SS_RESET = 0x00000004,
-
- /* Trigger full scrub automatically. */
- SS_AUTO_FULL = 0x00000008,
-
- /* Trigger partial scrub automatically. */
- SS_AUTO_PARTIAL = 0x00000010,
-
- /* Set dryrun flag. */
- SS_SET_DRYRUN = 0x00000020,
-
- /* Clear dryrun flag. */
- SS_CLEAR_DRYRUN = 0x00000040,
-};
-
/* The flags here are only used inside OSD, NOT be visible by dump(). */
enum scrub_internal_flags {
/* This is a new formatted device. */
SIF_NO_HANDLE_OLD_FID = 0x0001,
};
-struct scrub_file {
- /* 128-bit uuid for volume. */
- __u8 sf_uuid[16];
-
- /* See 'enum scrub_flags'. */
- __u64 sf_flags;
-
- /* The scrub magic. */
- __u32 sf_magic;
-
- /* See 'enum scrub_status'. */
- __u16 sf_status;
-
- /* See 'enum scrub_param'. */
- __u16 sf_param;
-
- /* The time for the last OI scrub completed. */
- __u64 sf_time_last_complete;
-
- /* The time for the latest OI scrub ran. */
- __u64 sf_time_latest_start;
-
- /* The time for the last OI scrub checkpoint. */
- __u64 sf_time_last_checkpoint;
-
- /* The position for the latest OI scrub started from. */
- __u64 sf_pos_latest_start;
-
- /* The position for the last OI scrub checkpoint. */
- __u64 sf_pos_last_checkpoint;
-
- /* The position for the first should be updated object. */
- __u64 sf_pos_first_inconsistent;
-
- /* How many objects have been checked. */
- __u64 sf_items_checked;
-
- /* How many objects have been updated. */
- __u64 sf_items_updated;
-
- /* How many objects failed to be processed. */
- __u64 sf_items_failed;
-
- /* How many prior objects have been updated during scanning. */
- __u64 sf_items_updated_prior;
-
- /* How many objects marked as LDISKFS_STATE_LUSTRE_NOSCRUB. */
- __u64 sf_items_noscrub;
-
- /* How many IGIF objects. */
- __u64 sf_items_igif;
-
- /* How long the OI scrub has run. */
- __u32 sf_run_time;
-
- /* How many completed OI scrub ran on the device. */
- __u32 sf_success_count;
-
- /* How many OI files. */
- __u16 sf_oi_count;
-
- /* Keep the flags after scrub reset. See 'enum scrub_internal_flags' */
- __u16 sf_internal_flags;
-
- __u32 sf_reserved_1;
- __u64 sf_reserved_2[16];
-
- /* Bitmap for OI files recreated case. */
- __u8 sf_oi_bitmap[SCRUB_OI_BITMAP_SIZE];
-};
-
struct osd_iit_param {
struct super_block *sb;
struct buffer_head *bitmap;
};
struct osd_scrub {
+ struct lustre_scrub os_scrub;
struct lvfs_run_ctxt os_ctxt;
- struct ptlrpc_thread os_thread;
struct osd_idmap_cache os_oic;
struct osd_iit_param os_iit_param;
- struct list_head os_inconsistent_items;
-
- /* write lock for scrub prep/update/post/checkpoint,
- * read lock for scrub dump. */
- struct rw_semaphore os_rwsem;
- spinlock_t os_lock;
-
- /* Scrub file in memory. */
- struct scrub_file os_file;
-
- /* Buffer for scrub file load/store. */
- struct scrub_file os_file_disk;
-
- /* Inode for the scrub file. */
- struct inode *os_inode;
-
- /* The time for last checkpoint, jiffies */
- cfs_time_t os_time_last_checkpoint;
-
- /* The time for next checkpoint, jiffies */
- cfs_time_t os_time_next_checkpoint;
/* statistics for /lost+found are in ram only, it will be reset
* when each time the device remount. */
/* How many objects failed to be processed during initial OI scrub. */
__u64 os_lf_failed;
- /* How many objects have been checked since last checkpoint. */
- __u32 os_new_checked;
- __u32 os_pos_current;
- __u32 os_start_flags;
- unsigned int os_in_prior:1, /* process inconsistent item
- * found by RPC prior */
- os_waiting:1, /* Waiting for scan window. */
- os_full_speed:1, /* run w/o speed limit */
- os_paused:1, /* The scrub is paused. */
- os_convert_igif:1,
- os_partial_scan:1,
- os_in_join:1,
- os_full_scrub:1;
__u64 os_bad_oimap_count;
__u64 os_bad_oimap_time;
};
#include <sys/dsl_prop.h>
#include <sys/sa_impl.h>
#include <sys/txg.h>
+#include <lustre_scrub.h>
#define OSD_OI_FID_NR (1UL << 7)
-#define OSD_OI_FID_NR_MAX (1UL << OSD_OI_FID_OID_BITS_MAX)
unsigned int osd_oi_count = OSD_OI_FID_NR;
scrub_check_repaired() {
local error_id=$1
local expected=$2
+ local dryrun=$3
local actual
local n
for n in $(seq $MDSCOUNT); do
- actual=$(do_facet mds$n $LCTL get_param -n \
- osd-ldiskfs.$(facet_svc mds$n).oi_scrub |
- awk '/^updated/ { print $2 }')
+ if [ $dryrun -eq 1 ]; then
+ actual=$(do_facet mds$n $LCTL get_param -n \
+ osd-ldiskfs.$(facet_svc mds$n).oi_scrub |
+ awk '/^inconsistent:/ { print $2 }')
+ else
+ actual=$(do_facet mds$n $LCTL get_param -n \
+ osd-ldiskfs.$(facet_svc mds$n).oi_scrub |
+ awk '/^updated:/ { print $2 }')
+ fi
if [ $expected -eq 0 -a $actual -ne 0 ]; then
error "($error_id) Expected no repaired on mds$n, but" \
scrub_check_status 6 completed
scrub_check_flags 7 recreated,inconsistent
scrub_check_params 8 dryrun
- scrub_check_repaired 9 20
+ scrub_check_repaired 9 20 1
# run under dryrun mode again
if [ $server_version -lt $(version_code 2.5.58) ]; then
scrub_check_status 11 completed
scrub_check_flags 12 recreated,inconsistent
scrub_check_params 13 dryrun
- scrub_check_repaired 14 20
+ scrub_check_repaired 14 20 1
# run under normal mode
#
scrub_check_status 16 completed
scrub_check_flags 17 ""
scrub_check_params 18 ""
- scrub_check_repaired 19 20
+ scrub_check_repaired 19 20 0
# run under normal mode again
if [ $server_version -lt $(version_code 2.5.58) ]; then
scrub_check_status 21 completed
scrub_check_flags 22 ""
scrub_check_params 23 ""
- scrub_check_repaired 24 0
+ scrub_check_repaired 24 0 0
}
run_test 15 "Dryrun mode OI scrub"