4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2017, Intel Corporation.
26 * lustre/include/lustre_scrub.h
28 * Shared definitions and declarations for Lustre OI scrub.
30 * Author: Fan Yong <fan.yong@intel.com>
33 #ifndef _LUSTRE_SCRUB_H
34 # define _LUSTRE_SCRUB_H
36 #include <dt_object.h>
37 #include <lustre_net.h>
39 #define OSD_OI_FID_OID_BITS_MAX 10
40 #define OSD_OI_FID_NR_MAX (1UL << OSD_OI_FID_OID_BITS_MAX)
41 #define SCRUB_OI_BITMAP_SIZE (OSD_OI_FID_NR_MAX >> 3)
42 #define PFID_STRIPE_IDX_BITS 16
43 #define PFID_STRIPE_COUNT_MASK ((1 << PFID_STRIPE_IDX_BITS) - 1)
45 #define SCRUB_MAGIC_V1 0x4C5FD252
46 #define SCRUB_CHECKPOINT_INTERVAL 60
47 #define SCRUB_WINDOW_SIZE 1024
49 #define HALF_SEC msecs_to_jiffies(MSEC_PER_SEC >> 1)
51 enum scrub_next_status {
52 /* exit current loop and process next group */
55 /* skip current object and process next bit */
56 SCRUB_NEXT_CONTINUE = 2,
58 /* exit all the loops */
61 /* wait for free cache slot */
64 /* simulate system crash during OI scrub */
67 /* simulate failure during OI scrub */
70 /* new created object, no scrub on it */
71 SCRUB_NEXT_NOSCRUB = 7,
73 /* the object has no FID-in-LMA */
77 SCRUB_NEXT_OSTOBJ = 9,
79 /* old OST-object, no LMA or no FID-on-OST flags in LMA */
80 SCRUB_NEXT_OSTOBJ_OLD = 10,
83 enum scrub_local_file_flags {
84 SLFF_SCAN_SUBITEMS = 0x0001,
85 SLFF_HIDE_FID = 0x0002,
86 SLFF_SHOW_NAME = 0x0004,
88 SLFF_IDX_IN_FID = 0x0010,
92 /* The scrub file is new created, for new MDT, upgrading from old disk,
93 * or re-creating the scrub file manually. */
96 /* The scrub is checking/repairing the OI files. */
99 /* The scrub checked/repaired the OI files successfully. */
102 /* The scrub failed to check/repair the OI files. */
105 /* The scrub is stopped manually, the OI files may be inconsistent. */
108 /* The scrub is paused automatically when umount. */
111 /* The scrub crashed during the scanning, should be restarted. */
116 /* OI files have been recreated, OI mappings should be re-inserted. */
117 SF_RECREATED = 0x0000000000000001ULL,
119 /* OI files are invalid, should be rebuild ASAP */
120 SF_INCONSISTENT = 0x0000000000000002ULL,
122 /* OI scrub is triggered automatically. */
123 SF_AUTO = 0x0000000000000004ULL,
125 /* The device is upgraded from 1.8 format. */
126 SF_UPGRADE = 0x0000000000000008ULL,
130 /* Exit when fail. */
133 /* Check only without repairing. */
138 /* Set failout flag. */
139 SS_SET_FAILOUT = 0x00000001,
141 /* Clear failout flag. */
142 SS_CLEAR_FAILOUT = 0x00000002,
144 /* Reset scrub start position. */
145 SS_RESET = 0x00000004,
147 /* Trigger full scrub automatically. */
148 SS_AUTO_FULL = 0x00000008,
150 /* Trigger partial scrub automatically. */
151 SS_AUTO_PARTIAL = 0x00000010,
153 /* Set dryrun flag. */
154 SS_SET_DRYRUN = 0x00000020,
156 /* Clear dryrun flag. */
157 SS_CLEAR_DRYRUN = 0x00000040,
161 OLF_SCAN_SUBITEMS = 0x0001,
162 OLF_HIDE_FID = 0x0002,
163 OLF_SHOW_NAME = 0x0004,
165 OLF_IDX_IN_FID = 0x0010,
168 /* There are some overhead to detect OI inconsistency automatically
169 * during normal RPC handling. We do not want to always auto detect
170 * OI inconsistency especailly when OI scrub just done recently.
172 * The 'auto_scrub' defines the time (united as second) interval to
173 * enable auto detect OI inconsistency since last OI scurb done. */
175 /* Disable auto scrub. */
178 /* 1 second is too short interval, it is almost equal to always auto
179 * detect inconsistent OI, usually used for test. */
182 /* Enable auto detect OI inconsistency one month (60 * 60 * 24 * 30)
183 * after last OI scrub. */
184 AS_DEFAULT = 2592000LL,
188 /* 128-bit uuid for volume. */
191 /* See 'enum scrub_flags'. */
194 /* The scrub magic. */
197 /* See 'enum scrub_status'. */
200 /* See 'enum scrub_param'. */
203 /* The time for the last OI scrub completed. */
204 __u64 sf_time_last_complete;
206 /* The time for the latest OI scrub ran. */
207 __u64 sf_time_latest_start;
209 /* The time for the last OI scrub checkpoint. */
210 __u64 sf_time_last_checkpoint;
212 /* The position for the latest OI scrub started from. */
213 __u64 sf_pos_latest_start;
215 /* The position for the last OI scrub checkpoint. */
216 __u64 sf_pos_last_checkpoint;
218 /* The position for the first should be updated object. */
219 __u64 sf_pos_first_inconsistent;
221 /* How many objects have been checked. */
222 __u64 sf_items_checked;
224 /* How many objects have been updated. */
225 __u64 sf_items_updated;
227 /* How many objects failed to be processed. */
228 __u64 sf_items_failed;
230 /* How many prior objects have been updated during scanning. */
231 __u64 sf_items_updated_prior;
233 /* How many objects marked as LDISKFS_STATE_LUSTRE_NOSCRUB. */
234 __u64 sf_items_noscrub;
236 /* How many IGIF objects. */
239 /* How long the OI scrub has run. */
242 /* How many completed OI scrub ran on the device. */
243 __u32 sf_success_count;
245 /* How many OI files. */
248 /* Keep the flags after scrub reset. See 'enum scrub_internal_flags' */
249 __u16 sf_internal_flags;
252 __u64 sf_reserved_2[16];
254 /* Bitmap for OI files recreated case. */
255 __u8 sf_oi_bitmap[SCRUB_OI_BITMAP_SIZE];
258 struct lustre_scrub {
259 struct lu_object_header os_obj_header;
260 /* Object for the scrub file. */
261 struct dt_object *os_obj;
263 struct ptlrpc_thread os_thread;
264 struct list_head os_inconsistent_items;
266 /* write lock for scrub prep/update/post/checkpoint,
267 * read lock for scrub dump. */
268 struct rw_semaphore os_rwsem;
271 /* Scrub file in memory. */
272 struct scrub_file os_file;
274 /* Buffer for scrub file load/store. */
275 struct scrub_file os_file_disk;
279 /* The time for last checkpoint, jiffies */
280 cfs_time_t os_time_last_checkpoint;
282 /* The time for next checkpoint, jiffies */
283 cfs_time_t os_time_next_checkpoint;
285 /* How many objects have been checked since last checkpoint. */
286 __u64 os_new_checked;
287 __u64 os_pos_current;
288 __u32 os_start_flags;
289 unsigned int os_in_prior:1, /* process inconsistent item
290 * found by RPC prior */
291 os_waiting:1, /* Waiting for scan window. */
292 os_full_speed:1, /* run w/o speed limit */
293 os_paused:1, /* The scrub is paused. */
300 void scrub_file_init(struct lustre_scrub *scrub, __u8 *uuid);
301 void scrub_file_reset(struct lustre_scrub *scrub, __u8 *uuid, __u64 flags);
302 int scrub_file_load(const struct lu_env *env, struct lustre_scrub *scrub);
303 int scrub_file_store(const struct lu_env *env, struct lustre_scrub *scrub);
304 int scrub_checkpoint(const struct lu_env *env, struct lustre_scrub *scrub);
305 int scrub_start(int (*threadfn)(void *data), struct lustre_scrub *scrub,
306 void *data, __u32 flags);
307 void scrub_stop(struct lustre_scrub *scrub);
308 void scrub_dump(struct seq_file *m, struct lustre_scrub *scrub);
310 static inline const char *osd_scrub2name(struct lustre_scrub *scrub)
312 return scrub->os_name;
314 #endif /* _LUSTRE_SCRUB_H */