lustre/mdd/mdd_lfsck.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   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.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2012 Whamcloud, Inc.
  24  */
  25 /*
  26  * lustre/mdd/mdd_lfsck.c
  27  *
  28  * Top-level entry points into mdd module
  29  *
  30  * LFSCK controller, which scans the whole device through low layer
  31  * iteration APIs, drives all lfsck compeonents, controls the speed.
  32  *
  33  * Author: Fan Yong <yong.fan@whamcloud.com>
  34  */
  35
  36 #ifndef EXPORT_SYMTAB
  37 # define EXPORT_SYMTAB
  38 #endif
  39 #define DEBUG_SUBSYSTEM S_MDS
  40
  41 #include <lustre/lustre_idl.h>
  42 #include <lustre_fid.h>
  43
  44 #include "mdd_internal.h"
  45
  46 static inline char *mdd_lfsck2name(struct md_lfsck *lfsck)
  47 {
  48         struct mdd_device *mdd;
  49
  50         mdd = container_of0(lfsck, struct mdd_device, mdd_lfsck);
  51         return mdd2obd_dev(mdd)->obd_name;
  52 }
  53
  54 void mdd_lfsck_set_speed(struct md_lfsck *lfsck, __u32 limit)
  55 {
  56         spin_lock(&lfsck->ml_lock);
  57         lfsck->ml_speed_limit = limit;
  58         if (limit != LFSCK_SPEED_NO_LIMIT) {
  59                 if (limit > CFS_HZ) {
  60                         lfsck->ml_sleep_rate = limit / CFS_HZ;
  61                         lfsck->ml_sleep_jif = 1;
  62                 } else {
  63                         lfsck->ml_sleep_rate = 1;
  64                         lfsck->ml_sleep_jif = CFS_HZ / limit;
  65                 }
  66         } else {
  67                 lfsck->ml_sleep_jif = 0;
  68                 lfsck->ml_sleep_rate = 0;
  69         }
  70         spin_unlock(&lfsck->ml_lock);
  71 }
  72
  73 static void mdd_lfsck_control_speed(struct md_lfsck *lfsck)
  74 {
  75         struct ptlrpc_thread *thread = &lfsck->ml_thread;
  76         struct l_wait_info    lwi;
  77
  78         if (lfsck->ml_sleep_jif > 0 &&
  79             lfsck->ml_new_scanned >= lfsck->ml_sleep_rate) {
  80                 spin_lock(&lfsck->ml_lock);
  81                 if (likely(lfsck->ml_sleep_jif > 0 &&
  82                            lfsck->ml_new_scanned >= lfsck->ml_sleep_rate)) {
  83                         lwi = LWI_TIMEOUT_INTR(lfsck->ml_sleep_jif, NULL,
  84                                                LWI_ON_SIGNAL_NOOP, NULL);
  85                         spin_unlock(&lfsck->ml_lock);
  86
  87                         l_wait_event(thread->t_ctl_waitq,
  88                                      !thread_is_running(thread),
  89                                      &lwi);
  90                         lfsck->ml_new_scanned = 0;
  91                 } else {
  92                         spin_unlock(&lfsck->ml_lock);
  93                 }
  94         }
  95 }
  96
  97 static int mdd_lfsck_main(void *args)
  98 {
  99         struct lu_env            env;
 100         struct md_lfsck         *lfsck  = (struct md_lfsck *)args;
 101         struct ptlrpc_thread    *thread = &lfsck->ml_thread;
 102         struct dt_object        *obj    = lfsck->ml_it_obj;
 103         const struct dt_it_ops  *iops   = &obj->do_index_ops->dio_it;
 104         struct dt_it            *di;
 105         struct lu_fid           *fid;
 106         int                      rc;
 107         ENTRY;
 108
 109         cfs_daemonize("lfsck");
 110         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
 111         if (rc != 0) {
 112                 CERROR("%s: LFSCK, fail to init env, rc = %d\n",
 113                        mdd_lfsck2name(lfsck), rc);
 114                 GOTO(noenv, rc);
 115         }
 116
 117         di = iops->init(&env, obj, lfsck->ml_args, BYPASS_CAPA);
 118         if (IS_ERR(di)) {
 119                 rc = PTR_ERR(di);
 120                 CERROR("%s: LFSCK, fail to init iteration, rc = %d\n",
 121                        mdd_lfsck2name(lfsck), rc);
 122                 GOTO(fini_env, rc);
 123         }
 124
 125         CDEBUG(D_LFSCK, "LFSCK: flags = 0x%x, pid = %d\n",
 126                lfsck->ml_args, cfs_curproc_pid());
 127
 128         /* XXX: Prepare before wakeup the sponsor.
 129          *      Each lfsck component should call iops->get() API with
 130          *      every bookmark, then low layer module can decide the
 131          *      start point for current iteration. */
 132
 133         spin_lock(&lfsck->ml_lock);
 134         thread_set_flags(thread, SVC_RUNNING);
 135         spin_unlock(&lfsck->ml_lock);
 136         cfs_waitq_broadcast(&thread->t_ctl_waitq);
 137
 138         /* Call iops->load() to finish the choosing start point. */
 139         rc = iops->load(&env, di, 0);
 140         if (rc != 0)
 141                 GOTO(out, rc);
 142
 143         CDEBUG(D_LFSCK, "LFSCK: iteration start: pos = %s\n",
 144                (char *)iops->key(&env, di));
 145
 146         lfsck->ml_new_scanned = 0;
 147         fid = &mdd_env_info(&env)->mti_fid;
 148         while (rc == 0) {
 149                 iops->rec(&env, di, (struct dt_rec *)fid, 0);
 150
 151                 /* XXX: here, perform LFSCK when some LFSCK component(s)
 152                  *      introduced in the future. */
 153                 lfsck->ml_new_scanned++;
 154
 155                 /* XXX: here, make checkpoint when some LFSCK component(s)
 156                  *      introduced in the future. */
 157
 158                 /* Rate control. */
 159                 mdd_lfsck_control_speed(lfsck);
 160                 if (unlikely(!thread_is_running(thread)))
 161                         GOTO(out, rc = 0);
 162
 163                 rc = iops->next(&env, di);
 164         }
 165
 166         GOTO(out, rc);
 167
 168 out:
 169         if (lfsck->ml_paused) {
 170                 /* XXX: It is hack here: if the lfsck is still running when MDS
 171                  *      umounts, it should be restarted automatically after MDS
 172                  *      remounts up.
 173                  *
 174                  *      To support that, we need to record the lfsck status in
 175                  *      the lfsck on-disk bookmark file. But now, there is not
 176                  *      lfsck component under the lfsck framework. To avoid to
 177                  *      introduce unnecessary bookmark incompatibility issues,
 178                  *      we write nothing to the lfsck bookmark file now.
 179                  *
 180                  *      Instead, we will reuse dt_it_ops::put() method to notify
 181                  *      low layer iterator to process such case.
 182                  *
 183                  *      It is just temporary solution, and will be replaced when
 184                  *      some lfsck component is introduced in the future. */
 185                 iops->put(&env, di);
 186                 CDEBUG(D_LFSCK, "LFSCK: iteration pasued: pos = %s, rc = %d\n",
 187                        (char *)iops->key(&env, di), rc);
 188         } else {
 189                 CDEBUG(D_LFSCK, "LFSCK: iteration stop: pos = %s, rc = %d\n",
 190                        (char *)iops->key(&env, di), rc);
 191         }
 192         iops->fini(&env, di);
 193
 194 fini_env:
 195         lu_env_fini(&env);
 196
 197 noenv:
 198         spin_lock(&lfsck->ml_lock);
 199         thread_set_flags(thread, SVC_STOPPED);
 200         cfs_waitq_broadcast(&thread->t_ctl_waitq);
 201         spin_unlock(&lfsck->ml_lock);
 202         return rc;
 203 }
 204
 205 int mdd_lfsck_start(const struct lu_env *env, struct md_lfsck *lfsck,
 206                     struct lfsck_start *start)
 207 {
 208         struct ptlrpc_thread *thread  = &lfsck->ml_thread;
 209         struct l_wait_info    lwi     = { 0 };
 210         int                   rc      = 0;
 211         __u16                 valid   = 0;
 212         __u16                 flags   = 0;
 213         ENTRY;
 214
 215         if (lfsck->ml_it_obj == NULL)
 216                 RETURN(-ENOTSUPP);
 217
 218         mutex_lock(&lfsck->ml_mutex);
 219         spin_lock(&lfsck->ml_lock);
 220         if (thread_is_running(thread)) {
 221                 spin_unlock(&lfsck->ml_lock);
 222                 mutex_unlock(&lfsck->ml_mutex);
 223                 RETURN(-EALREADY);
 224         }
 225
 226         spin_unlock(&lfsck->ml_lock);
 227         if (start->ls_valid & LSV_SPEED_LIMIT)
 228                 mdd_lfsck_set_speed(lfsck, start->ls_speed_limit);
 229
 230         if (start->ls_valid & LSV_ERROR_HANDLE) {
 231                 valid |= DOIV_ERROR_HANDLE;
 232                 if (start->ls_flags & LPF_FAILOUT)
 233                         flags |= DOIF_FAILOUT;
 234         }
 235
 236         /* XXX: 1. low layer does not care 'dryrun'.
 237          *      2. will process 'ls_active' when introduces LFSCK for layout
 238          *         consistency, DNE consistency, and so on in the future. */
 239         start->ls_active = 0;
 240
 241         if (start->ls_flags & LPF_RESET)
 242                 flags |= DOIF_RESET;
 243
 244         if (start->ls_active != 0)
 245                 flags |= DOIF_OUTUSED;
 246
 247         lfsck->ml_args = (flags << DT_OTABLE_IT_FLAGS_SHIFT) | valid;
 248         thread_set_flags(thread, 0);
 249         rc = cfs_create_thread(mdd_lfsck_main, lfsck, 0);
 250         if (rc < 0)
 251                 CERROR("%s: cannot start LFSCK thread, rc = %d\n",
 252                        mdd_lfsck2name(lfsck), rc);
 253         else
 254                 l_wait_event(thread->t_ctl_waitq,
 255                              thread_is_running(thread) ||
 256                              thread_is_stopped(thread),
 257                              &lwi);
 258         mutex_unlock(&lfsck->ml_mutex);
 259
 260         RETURN(rc < 0 ? rc : 0);
 261 }
 262
 263 int mdd_lfsck_stop(const struct lu_env *env, struct md_lfsck *lfsck)
 264 {
 265         struct ptlrpc_thread *thread = &lfsck->ml_thread;
 266         struct l_wait_info    lwi    = { 0 };
 267         ENTRY;
 268
 269         mutex_lock(&lfsck->ml_mutex);
 270         spin_lock(&lfsck->ml_lock);
 271         if (thread_is_init(thread) || thread_is_stopped(thread)) {
 272                 spin_unlock(&lfsck->ml_lock);
 273                 mutex_unlock(&lfsck->ml_mutex);
 274                 RETURN(-EALREADY);
 275         }
 276
 277         thread_set_flags(thread, SVC_STOPPING);
 278         spin_unlock(&lfsck->ml_lock);
 279
 280         cfs_waitq_broadcast(&thread->t_ctl_waitq);
 281         l_wait_event(thread->t_ctl_waitq,
 282                      thread_is_stopped(thread),
 283                      &lwi);
 284         mutex_unlock(&lfsck->ml_mutex);
 285
 286         RETURN(0);
 287 }
 288
 289 const char lfsck_bookmark_name[] = "lfsck_bookmark";
 290
 291 static const struct lu_fid lfsck_it_fid = { .f_seq = FID_SEQ_LOCAL_FILE,
 292                                             .f_oid = OTABLE_IT_OID,
 293                                             .f_ver = 0 };
 294
 295 int mdd_lfsck_setup(const struct lu_env *env, struct mdd_device *mdd)
 296 {
 297         struct md_lfsck  *lfsck = &mdd->mdd_lfsck;
 298         struct dt_object *obj;
 299         int               rc;
 300
 301         memset(lfsck, 0, sizeof(*lfsck));
 302         lfsck->ml_version = LFSCK_VERSION_V1;
 303         cfs_waitq_init(&lfsck->ml_thread.t_ctl_waitq);
 304         mutex_init(&lfsck->ml_mutex);
 305         spin_lock_init(&lfsck->ml_lock);
 306
 307         obj = dt_store_open(env, mdd->mdd_child, "", lfsck_bookmark_name,
 308                             &mdd_env_info(env)->mti_fid);
 309         if (IS_ERR(obj))
 310                 return PTR_ERR(obj);
 311
 312         lfsck->ml_bookmark_obj = obj;
 313
 314         obj = dt_locate(env, mdd->mdd_child, &lfsck_it_fid);
 315         if (IS_ERR(obj))
 316                 return PTR_ERR(obj);
 317
 318         rc = obj->do_ops->do_index_try(env, obj, &dt_otable_features);
 319         if (rc != 0) {
 320                 lu_object_put(env, &obj->do_lu);
 321                 if (rc == -ENOTSUPP)
 322                         rc = 0;
 323                 return rc;
 324         }
 325
 326         lfsck->ml_it_obj = obj;
 327
 328         return 0;
 329 }
 330
 331 void mdd_lfsck_cleanup(const struct lu_env *env, struct mdd_device *mdd)
 332 {
 333         struct md_lfsck *lfsck = &mdd->mdd_lfsck;
 334
 335         if (lfsck->ml_it_obj != NULL) {
 336                 lfsck->ml_paused = 1;
 337                 mdd_lfsck_stop(env, lfsck);
 338                 lu_object_put(env, &lfsck->ml_it_obj->do_lu);
 339                 lfsck->ml_it_obj = NULL;
 340         }
 341
 342         if (lfsck->ml_bookmark_obj != NULL) {
 343                 lu_object_put(env, &lfsck->ml_bookmark_obj->do_lu);
 344                 lfsck->ml_bookmark_obj = NULL;
 345         }
 346 }