lustre/mgs/mgs_nids.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  *
  32  * Copyright (c) 2011, Whamcloud, Inc.
  33  */
  34 /*
  35  * This file is part of Lustre, http://www.lustre.org/
  36  * Lustre is a trademark of Sun Microsystems, Inc.
  37  *
  38  * lustre/mgs/mgs_nids.c
  39  *
  40  * NID table management for lustre.
  41  *
  42  * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
  43  */
  44
  45 #ifndef EXPORT_SYMTAB
  46 #define EXPORT_SYMTAB
  47 #endif
  48 #define DEBUG_SUBSYSTEM S_MGS
  49 #define D_MGS D_CONFIG
  50
  51 #ifdef __KERNEL__
  52 #include <linux/module.h>
  53 #include <linux/pagemap.h>
  54 #include <linux/fs.h>
  55 #endif
  56
  57 #include <obd.h>
  58 #include <obd_lov.h>
  59 #include <obd_class.h>
  60 #include <lustre_log.h>
  61 #include <obd_ost.h>
  62 #include <libcfs/list.h>
  63 #include <linux/lvfs.h>
  64 #include <lustre_fsfilt.h>
  65 #include <lustre_disk.h>
  66 #include <lustre_param.h>
  67 #include "mgs_internal.h"
  68
  69 static unsigned int ir_timeout;
  70
  71 static int nidtbl_is_sane(struct mgs_nidtbl *tbl)
  72 {
  73         struct mgs_nidtbl_target *tgt;
  74         int version = 0;
  75
  76         LASSERT(cfs_mutex_is_locked(&tbl->mn_lock));
  77         cfs_list_for_each_entry(tgt, &tbl->mn_targets, mnt_list) {
  78                 if (!tgt->mnt_version)
  79                         continue;
  80
  81                 if (version >= tgt->mnt_version)
  82                         return 0;
  83
  84                 version = tgt->mnt_version;
  85         }
  86         return 1;
  87 }
  88
  89 /**
  90  * Fetch nidtbl entries whose version are not less than @version
  91  * nidtbl entries will be packed in @pages by @unit_size units - entries
  92  * shouldn't cross unit boundaries.
  93  */
  94 static int mgs_nidtbl_read(struct obd_device *unused, struct mgs_nidtbl *tbl,
  95                            struct mgs_config_res *res, cfs_page_t **pages,
  96                            int nrpages, int units_total, int unit_size)
  97 {
  98         struct mgs_nidtbl_target *tgt;
  99         struct mgs_nidtbl_entry  *entry;
 100         struct mgs_nidtbl_entry  *last_in_unit = NULL;
 101         struct mgs_target_info   *mti;
 102         __u64 version = res->mcr_offset;
 103         bool nobuf = false;
 104         void *buf = NULL;
 105         int bytes_in_unit = 0;
 106         int units_in_page = 0;
 107         int index = 0;
 108         int rc = 0;
 109         ENTRY;
 110
 111         /* make sure unit_size is power 2 */
 112         LASSERT((unit_size & (unit_size - 1)) == 0);
 113         LASSERT(nrpages << CFS_PAGE_SHIFT >= units_total * unit_size);
 114
 115         cfs_mutex_lock(&tbl->mn_lock);
 116         LASSERT(nidtbl_is_sane(tbl));
 117
 118         /* no more entries ? */
 119         if (version > tbl->mn_version) {
 120                 version = tbl->mn_version;
 121                 goto out;
 122         }
 123
 124         /* iterate over all targets to compose a bitmap by the type of llog.
 125          * If the llog is for MDTs, llog entries for OSTs will be returned;
 126          * otherwise, it's for clients, then llog entries for both OSTs and
 127          * MDTs will be returned.
 128          */
 129         cfs_list_for_each_entry(tgt, &tbl->mn_targets, mnt_list) {
 130                 int entry_len = sizeof(*entry);
 131
 132                 if (tgt->mnt_version < version)
 133                         continue;
 134
 135                 /* write target recover information */
 136                 mti  = &tgt->mnt_mti;
 137                 LASSERT(mti->mti_nid_count < MTI_NIDS_MAX);
 138                 entry_len += mti->mti_nid_count * sizeof(lnet_nid_t);
 139
 140                 if (entry_len > unit_size) {
 141                         CWARN("nidtbl: too large entry: entry length %d,"
 142                               "unit size: %d\n", entry_len, unit_size);
 143                         GOTO(out, rc = -EOVERFLOW);
 144                 }
 145
 146                 if (bytes_in_unit < entry_len) {
 147                         if (units_total == 0) {
 148                                 nobuf = true;
 149                                 break;
 150                         }
 151
 152                         /* check if we need to consume remaining bytes. */
 153                         if (last_in_unit != NULL && bytes_in_unit) {
 154                                 /* entry has been swapped. */
 155                                 __swab32s(&last_in_unit->mne_length);
 156                                 last_in_unit->mne_length += bytes_in_unit;
 157                                 __swab32s(&last_in_unit->mne_length);
 158                                 rc  += bytes_in_unit;
 159                                 buf += bytes_in_unit;
 160                                 last_in_unit = NULL;
 161                         }
 162                         LASSERT((rc & (unit_size - 1)) == 0);
 163
 164                         if (units_in_page == 0) {
 165                                 /* allocate a new page */
 166                                 pages[index] = cfs_alloc_page(CFS_ALLOC_STD);
 167                                 if (pages[index] == NULL) {
 168                                         rc = -ENOMEM;
 169                                         break;
 170                                 }
 171
 172                                 /* destroy previous map */
 173                                 if (index > 0)
 174                                         cfs_kunmap(pages[index - 1]);
 175
 176                                 /* reassign buffer */
 177                                 buf = cfs_kmap(pages[index]);
 178                                 ++index;
 179
 180                                 units_in_page = CFS_PAGE_SIZE / unit_size;
 181                                 LASSERT(units_in_page > 0);
 182                         }
 183
 184                         /* allocate an unit */
 185                         LASSERT(((long)buf & (unit_size - 1)) == 0);
 186                         bytes_in_unit = unit_size;
 187                         --units_in_page;
 188                         --units_total;
 189                 }
 190
 191                 /* fill in entry. */
 192                 entry = (struct mgs_nidtbl_entry *)buf;
 193                 entry->mne_version   = tgt->mnt_version;
 194                 entry->mne_instance  = mti->mti_instance;
 195                 entry->mne_index     = mti->mti_stripe_index;
 196                 entry->mne_length    = entry_len;
 197                 entry->mne_type      = tgt->mnt_type;
 198                 entry->mne_nid_type  = 0;
 199                 entry->mne_nid_size  = sizeof(lnet_nid_t);
 200                 entry->mne_nid_count = mti->mti_nid_count;
 201                 memcpy(entry->u.nids, mti->mti_nids,
 202                        mti->mti_nid_count * sizeof(lnet_nid_t));
 203                 lustre_swab_mgs_nidtbl_entry(entry);
 204
 205                 version = tgt->mnt_version;
 206                 rc     += entry_len;
 207                 buf    += entry_len;
 208
 209                 bytes_in_unit -= entry_len;
 210                 last_in_unit   = entry;
 211
 212                 CDEBUG(D_MGS, "fsname %s, entry size %d, pages %d/%d/%d/%d.\n",
 213                        tbl->mn_fsdb->fsdb_name, entry_len,
 214                        bytes_in_unit, index, nrpages, units_total);
 215         }
 216         if (index > 0)
 217                 cfs_kunmap(pages[index - 1]);
 218 out:
 219         LASSERT(version <= tbl->mn_version);
 220         res->mcr_size = tbl->mn_version;
 221         res->mcr_offset = nobuf ? version : tbl->mn_version;
 222         cfs_mutex_unlock(&tbl->mn_lock);
 223         LASSERT(ergo(version == 1, rc == 0)); /* get the log first time */
 224
 225         CDEBUG(D_MGS, "Read IR logs %s return with %d, version %llu\n",
 226                tbl->mn_fsdb->fsdb_name, rc, version);
 227         RETURN(rc);
 228 }
 229
 230 static int nidtbl_update_version(struct obd_device *obd, struct mgs_nidtbl *tbl)
 231 {
 232         struct lvfs_run_ctxt saved;
 233         struct file         *file = NULL;
 234         char                 filename[sizeof(MGS_NIDTBL_DIR) + 9];
 235         u64                  version;
 236         loff_t               off = 0;
 237         int                  rc;
 238         ENTRY;
 239
 240         LASSERT(cfs_mutex_is_locked(&tbl->mn_lock));
 241         LASSERT(sizeof(filename) < 32);
 242
 243         sprintf(filename, "%s/%s",
 244                 MGS_NIDTBL_DIR, tbl->mn_fsdb->fsdb_name);
 245
 246         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 247
 248         file = l_filp_open(filename, O_RDWR|O_CREAT, 0660);
 249         if (!IS_ERR(file)) {
 250                 version = cpu_to_le64(tbl->mn_version);
 251                 rc = lustre_fwrite(file, &version, sizeof(version), &off);
 252                 if (rc == sizeof(version))
 253                         rc = 0;
 254                 filp_close(file, 0);
 255                 fsfilt_sync(obd, obd->u.mgs.mgs_sb);
 256         } else {
 257                 rc = PTR_ERR(file);
 258         }
 259
 260         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 261         RETURN(rc);
 262 }
 263
 264 #define MGS_NIDTBL_VERSION_INIT 2
 265
 266 static int nidtbl_read_version(struct obd_device *obd, struct mgs_nidtbl *tbl)
 267 {
 268         struct lvfs_run_ctxt saved;
 269         struct file         *file = NULL;
 270         char                 filename[sizeof(MGS_NIDTBL_DIR) + 9];
 271         u64                  version;
 272         loff_t               off = 0;
 273         int                  rc;
 274         ENTRY;
 275
 276         LASSERT(cfs_mutex_is_locked(&tbl->mn_lock));
 277         LASSERT(sizeof(filename) < 32);
 278
 279         sprintf(filename, "%s/%s",
 280                 MGS_NIDTBL_DIR, tbl->mn_fsdb->fsdb_name);
 281
 282         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 283
 284         file = l_filp_open(filename, O_RDONLY, 0);
 285         if (!IS_ERR(file)) {
 286                 rc = lustre_fread(file, &version, sizeof(version), &off);
 287                 if (rc == sizeof(version))
 288                         rc = cpu_to_le64(version);
 289                 else if (rc == 0)
 290                         rc = MGS_NIDTBL_VERSION_INIT;
 291                 else
 292                         CERROR("read version file %s error %d\n", filename, rc);
 293                 filp_close(file, 0);
 294         } else {
 295                 rc = PTR_ERR(file);
 296                 if (rc == -ENOENT)
 297                         rc = MGS_NIDTBL_VERSION_INIT;
 298         }
 299
 300         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 301         RETURN(rc);
 302 }
 303
 304 static int mgs_nidtbl_write(struct fs_db *fsdb, struct mgs_target_info *mti)
 305 {
 306         struct mgs_nidtbl        *tbl;
 307         struct mgs_nidtbl_target *tgt;
 308         bool found = false;
 309         int type   = mti->mti_flags & LDD_F_SV_TYPE_MASK;
 310         int rc     = 0;
 311         ENTRY;
 312
 313         type &= ~LDD_F_SV_TYPE_MGS;
 314         LASSERT(type != 0);
 315
 316         tbl = &fsdb->fsdb_nidtbl;
 317         cfs_mutex_lock(&tbl->mn_lock);
 318         cfs_list_for_each_entry(tgt, &tbl->mn_targets, mnt_list) {
 319                 struct mgs_target_info *info = &tgt->mnt_mti;
 320                 if (type == tgt->mnt_type &&
 321                     mti->mti_stripe_index == info->mti_stripe_index) {
 322                         found = true;
 323                         break;
 324                 }
 325         }
 326         if (!found) {
 327                 OBD_ALLOC_PTR(tgt);
 328                 if (tgt == NULL)
 329                         GOTO(out, rc = -ENOMEM);
 330
 331                 CFS_INIT_LIST_HEAD(&tgt->mnt_list);
 332                 tgt->mnt_fs      = tbl;
 333                 tgt->mnt_version = 0;       /* 0 means invalid */
 334                 tgt->mnt_type    = type;
 335
 336                 ++tbl->mn_nr_targets;
 337         }
 338
 339         tgt->mnt_version = ++tbl->mn_version;
 340         tgt->mnt_mti     = *mti;
 341
 342         cfs_list_move_tail(&tgt->mnt_list, &tbl->mn_targets);
 343
 344         rc = nidtbl_update_version(fsdb->fsdb_obd, tbl);
 345         EXIT;
 346
 347 out:
 348         cfs_mutex_unlock(&tbl->mn_lock);
 349         if (rc)
 350                 CERROR("Write NID table version for file system %s error %d\n",
 351                        fsdb->fsdb_name, rc);
 352         return rc;
 353 }
 354
 355 static void mgs_nidtbl_fini_fs(struct fs_db *fsdb)
 356 {
 357         struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl;
 358         CFS_LIST_HEAD(head);
 359
 360         cfs_mutex_lock(&tbl->mn_lock);
 361         tbl->mn_nr_targets = 0;
 362         cfs_list_splice_init(&tbl->mn_targets, &head);
 363         cfs_mutex_unlock(&tbl->mn_lock);
 364
 365         while (!cfs_list_empty(&head)) {
 366                 struct mgs_nidtbl_target *tgt;
 367                 tgt = list_entry(head.next, struct mgs_nidtbl_target, mnt_list);
 368                 cfs_list_del(&tgt->mnt_list);
 369                 OBD_FREE_PTR(tgt);
 370         }
 371 }
 372
 373 static int mgs_nidtbl_init_fs(struct fs_db *fsdb)
 374 {
 375         struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl;
 376
 377         CFS_INIT_LIST_HEAD(&tbl->mn_targets);
 378         cfs_mutex_init(&tbl->mn_lock);
 379         tbl->mn_nr_targets = 0;
 380         tbl->mn_fsdb = fsdb;
 381         cfs_mutex_lock(&tbl->mn_lock);
 382         tbl->mn_version = nidtbl_read_version(fsdb->fsdb_obd, tbl);
 383         cfs_mutex_unlock(&tbl->mn_lock);
 384         CDEBUG(D_MGS, "IR: current version is %llu\n", tbl->mn_version);
 385
 386         return 0;
 387 }
 388
 389 /* --------- Imperative Recovery relies on nidtbl stuff ------- */
 390 void mgs_ir_notify_complete(struct fs_db *fsdb)
 391 {
 392         struct timeval tv;
 393         cfs_duration_t delta;
 394
 395         cfs_atomic_set(&fsdb->fsdb_notify_phase, 0);
 396
 397         /* do statistic */
 398         fsdb->fsdb_notify_count++;
 399         delta = cfs_time_sub(cfs_time_current(), fsdb->fsdb_notify_start);
 400         fsdb->fsdb_notify_total += delta;
 401         if (delta > fsdb->fsdb_notify_max)
 402                 fsdb->fsdb_notify_max = delta;
 403
 404         cfs_duration_usec(delta, &tv);
 405         CDEBUG(D_MGS, "Revoke recover lock of %s completed after %ld.%06lds\n",
 406                fsdb->fsdb_name, tv.tv_sec, tv.tv_usec);
 407 }
 408
 409 static int mgs_ir_notify(void *arg)
 410 {
 411         struct fs_db      *fsdb   = arg;
 412         struct ldlm_res_id resid;
 413
 414         char name[sizeof(fsdb->fsdb_name) + 20];
 415
 416         LASSERTF(sizeof(name) < 32, "name is too large to be in stack.\n");
 417         sprintf(name, "mgs_%s_notify", fsdb->fsdb_name);
 418         cfs_daemonize(name);
 419
 420         cfs_complete(&fsdb->fsdb_notify_comp);
 421
 422         set_user_nice(current, -2);
 423
 424         mgc_fsname2resid(fsdb->fsdb_name, &resid, CONFIG_T_RECOVER);
 425         while (1) {
 426                 struct l_wait_info   lwi = { 0 };
 427
 428                 l_wait_event(fsdb->fsdb_notify_waitq,
 429                              fsdb->fsdb_notify_stop ||
 430                              cfs_atomic_read(&fsdb->fsdb_notify_phase),
 431                              &lwi);
 432                 if (fsdb->fsdb_notify_stop)
 433                         break;
 434
 435                 CDEBUG(D_MGS, "%s woken up, phase is %d\n",
 436                        name, cfs_atomic_read(&fsdb->fsdb_notify_phase));
 437
 438                 fsdb->fsdb_notify_start = cfs_time_current();
 439                 mgs_revoke_lock(fsdb->fsdb_obd, fsdb, CONFIG_T_RECOVER);
 440         }
 441
 442         cfs_complete(&fsdb->fsdb_notify_comp);
 443         return 0;
 444 }
 445
 446 int mgs_ir_init_fs(struct obd_device *obd, struct fs_db *fsdb)
 447 {
 448         struct mgs_obd *mgs = &obd->u.mgs;
 449         int rc;
 450
 451         if (!ir_timeout)
 452                 ir_timeout = OBD_IR_MGS_TIMEOUT;
 453
 454         fsdb->fsdb_ir_state = IR_FULL;
 455         if (cfs_time_before(cfs_time_current_sec(),
 456                             mgs->mgs_start_time + ir_timeout))
 457                 fsdb->fsdb_ir_state = IR_STARTUP;
 458         fsdb->fsdb_nonir_clients = 0;
 459         CFS_INIT_LIST_HEAD(&fsdb->fsdb_clients);
 460
 461         /* start notify thread */
 462         fsdb->fsdb_obd = obd;
 463         cfs_atomic_set(&fsdb->fsdb_notify_phase, 0);
 464         cfs_waitq_init(&fsdb->fsdb_notify_waitq);
 465         cfs_init_completion(&fsdb->fsdb_notify_comp);
 466         rc = cfs_create_thread(mgs_ir_notify, fsdb, CFS_DAEMON_FLAGS);
 467         if (rc > 0)
 468                 cfs_wait_for_completion(&fsdb->fsdb_notify_comp);
 469         else
 470                 CERROR("Start notify thread error %d\n", rc);
 471
 472         mgs_nidtbl_init_fs(fsdb);
 473         return 0;
 474 }
 475
 476 void mgs_ir_fini_fs(struct obd_device *obd, struct fs_db *fsdb)
 477 {
 478         if (cfs_test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags))
 479                 return;
 480
 481         mgs_fsc_cleanup_by_fsdb(fsdb);
 482
 483         mgs_nidtbl_fini_fs(fsdb);
 484
 485         LASSERT(cfs_list_empty(&fsdb->fsdb_clients));
 486
 487         fsdb->fsdb_notify_stop = 1;
 488         cfs_waitq_signal(&fsdb->fsdb_notify_waitq);
 489         cfs_wait_for_completion(&fsdb->fsdb_notify_comp);
 490 }
 491
 492 /* caller must have held fsdb_sem */
 493 static inline void ir_state_graduate(struct fs_db *fsdb)
 494 {
 495         struct mgs_obd *mgs = &fsdb->fsdb_obd->u.mgs;
 496
 497         if (fsdb->fsdb_ir_state == IR_STARTUP) {
 498                 if (cfs_time_before(mgs->mgs_start_time + ir_timeout,
 499                                     cfs_time_current_sec())) {
 500                         fsdb->fsdb_ir_state = IR_FULL;
 501                         if (fsdb->fsdb_nonir_clients)
 502                                 fsdb->fsdb_ir_state = IR_PARTIAL;
 503                 }
 504         }
 505 }
 506
 507 int mgs_ir_update(struct obd_device *obd, struct mgs_target_info *mti)
 508 {
 509         struct fs_db *fsdb;
 510         bool notify = true;
 511         int rc;
 512
 513         if (mti->mti_instance == 0)
 514                 return -EINVAL;
 515
 516         rc = mgs_find_or_make_fsdb(obd, mti->mti_fsname, &fsdb);
 517         if (rc)
 518                 return rc;
 519
 520         rc = mgs_nidtbl_write(fsdb, mti);
 521         if (rc)
 522                 return rc;
 523
 524         /* check ir state */
 525         cfs_down(&fsdb->fsdb_sem);
 526         ir_state_graduate(fsdb);
 527         switch (fsdb->fsdb_ir_state) {
 528         case IR_FULL:
 529                 mti->mti_flags |= LDD_F_IR_CAPABLE;
 530                 break;
 531         case IR_DISABLED:
 532                 notify = false;
 533         case IR_STARTUP:
 534         case IR_PARTIAL:
 535                 break;
 536         default:
 537                 LBUG();
 538         }
 539         cfs_up(&fsdb->fsdb_sem);
 540
 541         LASSERT(ergo(mti->mti_flags & LDD_F_IR_CAPABLE, notify));
 542         if (notify) {
 543                 CDEBUG(D_MGS, "Try to revoke recover lock of %s\n",
 544                        fsdb->fsdb_name);
 545                 cfs_atomic_inc(&fsdb->fsdb_notify_phase);
 546                 cfs_waitq_signal(&fsdb->fsdb_notify_waitq);
 547         }
 548         return 0;
 549 }
 550
 551 /* NID table can be cached by two entities: Clients and MDTs */
 552 enum {
 553         IR_CLIENT  = 1,
 554         IR_MDT     = 2
 555 };
 556
 557 static int delogname(char *logname, char *fsname, int *typ)
 558 {
 559         char *ptr;
 560         int   type;
 561         int   len;
 562
 563         ptr = strrchr(logname, '-');
 564         if (ptr == NULL)
 565                 return -EINVAL;
 566
 567         /* decouple file system name. The llog name may be:
 568          * - "prefix-fsname", prefix is "cliir" or "mdtir"
 569          */
 570         if (strncmp(ptr, "-mdtir", 6) == 0)
 571                 type = IR_MDT;
 572         else if (strncmp(ptr, "-cliir", 6) == 0)
 573                 type = IR_CLIENT;
 574         else
 575                 return -EINVAL;
 576
 577         len = ptr - logname;
 578         if (len == 0)
 579                 return -EINVAL;
 580
 581         memcpy(fsname, logname, len);
 582         fsname[len] = 0;
 583         if (typ)
 584                 *typ = type;
 585         return 0;
 586 }
 587
 588 int mgs_get_ir_logs(struct ptlrpc_request *req)
 589 {
 590         struct obd_device *obd = req->rq_export->exp_obd;
 591         struct fs_db      *fsdb;
 592         struct mgs_config_body  *body;
 593         struct mgs_config_res   *res;
 594         struct ptlrpc_bulk_desc *desc;
 595         struct l_wait_info lwi;
 596         char               fsname[16];
 597         long               bufsize;
 598         int                unit_size;
 599
 600         int                type;
 601         int                rc = 0;
 602         int                i;
 603         int                bytes;
 604         int                page_count;
 605         int                nrpages;
 606         cfs_page_t       **pages = NULL;
 607         ENTRY;
 608
 609         body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
 610         if (body == NULL)
 611                 RETURN(-EINVAL);
 612
 613         if (body->mcb_type != CONFIG_T_RECOVER)
 614                 RETURN(-EINVAL);
 615
 616         rc = delogname(body->mcb_name, fsname, &type);
 617         if (rc)
 618                 RETURN(rc);
 619
 620         rc = mgs_find_or_make_fsdb(obd, fsname, &fsdb);
 621         if (rc)
 622                 GOTO(out, rc);
 623
 624         bufsize = body->mcb_units << body->mcb_bits;
 625         nrpages = (bufsize + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
 626         if (nrpages > PTLRPC_MAX_BRW_PAGES)
 627                 RETURN(-EINVAL);
 628
 629         CDEBUG(D_MGS, "Reading IR log %s bufsize %ld.\n",
 630                body->mcb_name, bufsize);
 631
 632         OBD_ALLOC(pages, sizeof(*pages) * nrpages);
 633         if (pages == NULL)
 634                 RETURN(-ENOMEM);
 635
 636         rc = req_capsule_server_pack(&req->rq_pill);
 637         if (rc)
 638                 GOTO(out, rc);
 639
 640         res = req_capsule_server_get(&req->rq_pill, &RMF_MGS_CONFIG_RES);
 641         if (res == NULL)
 642                 GOTO(out, rc = -EINVAL);
 643
 644         res->mcr_offset = body->mcb_offset;
 645         unit_size = min_t(int, 1 << body->mcb_bits, CFS_PAGE_SIZE);
 646         bytes = mgs_nidtbl_read(obd, &fsdb->fsdb_nidtbl, res, pages, nrpages,
 647                                 bufsize / unit_size, unit_size);
 648         if (bytes < 0)
 649                 GOTO(out, rc = bytes);
 650
 651         /* start bulk transfer */
 652         page_count = (bytes + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
 653         LASSERT(page_count <= nrpages);
 654         desc = ptlrpc_prep_bulk_exp(req, page_count,
 655                                     BULK_PUT_SOURCE, MGS_BULK_PORTAL);
 656         if (desc == NULL)
 657                 GOTO(out, rc = -ENOMEM);
 658
 659         for (i = 0; i < page_count && bytes > 0; i++) {
 660                 ptlrpc_prep_bulk_page(desc, pages[i], 0,
 661                                 min_t(int, bytes, CFS_PAGE_SIZE));
 662                 bytes -= CFS_PAGE_SIZE;
 663         }
 664
 665         rc = target_bulk_io(req->rq_export, desc, &lwi);
 666         ptlrpc_free_bulk(desc);
 667
 668 out:
 669         if (pages) {
 670                 for (i = 0; i < nrpages; i++) {
 671                         if (pages[i] == NULL)
 672                                 break;
 673                         cfs_free_page(pages[i]);
 674                 }
 675                 OBD_FREE(pages, sizeof(*pages) * nrpages);
 676         }
 677         return rc;
 678 }
 679
 680 static int lprocfs_ir_set_state(struct fs_db *fsdb, const char *buf)
 681 {
 682         const char *strings[] = IR_STRINGS;
 683         int         state = -1;
 684         int         i;
 685
 686         for (i = 0; i < ARRAY_SIZE(strings); i++) {
 687                 if (strcmp(strings[i], buf) == 0) {
 688                         state = i;
 689                         break;
 690                 }
 691         }
 692         if (state < 0)
 693                 return -EINVAL;
 694
 695         CDEBUG(D_MGS, "change fsr state of %s from %s to %s\n",
 696                fsdb->fsdb_name, strings[fsdb->fsdb_ir_state], strings[state]);
 697         cfs_down(&fsdb->fsdb_sem);
 698         if (state == IR_FULL && fsdb->fsdb_nonir_clients)
 699                 state = IR_PARTIAL;
 700         fsdb->fsdb_ir_state = state;
 701         cfs_up(&fsdb->fsdb_sem);
 702
 703         return 0;
 704 }
 705
 706 static int lprocfs_ir_set_timeout(struct fs_db *fsdb, const char *buf)
 707 {
 708         return -EINVAL;
 709 }
 710
 711 static int lprocfs_ir_clear_stats(struct fs_db *fsdb, const char *buf)
 712 {
 713         if (*buf)
 714                 return -EINVAL;
 715
 716         fsdb->fsdb_notify_total = 0;
 717         fsdb->fsdb_notify_max   = 0;
 718         fsdb->fsdb_notify_count = 0;
 719         return 0;
 720 }
 721
 722 static struct lproc_ir_cmd {
 723         char *name;
 724         int   namelen;
 725         int (*handler)(struct fs_db *, const char *);
 726 } ir_cmds[] = {
 727         { "state=",   6, lprocfs_ir_set_state },
 728         { "timeout=", 8, lprocfs_ir_set_timeout },
 729         { "0",        1, lprocfs_ir_clear_stats }
 730 };
 731
 732 int lprocfs_wr_ir_state(struct file *file, const char *buffer,
 733                          unsigned long count, void *data)
 734 {
 735         struct fs_db *fsdb = data;
 736         char *kbuf;
 737         char *ptr;
 738         int rc = 0;
 739
 740         if (count > CFS_PAGE_SIZE)
 741                 return -EINVAL;
 742
 743         OBD_ALLOC(kbuf, count + 1);
 744         if (kbuf == NULL)
 745                 return -ENOMEM;
 746
 747         if (copy_from_user(kbuf, buffer, count)) {
 748                 OBD_FREE(kbuf, count);
 749                 return -EFAULT;
 750         }
 751
 752         kbuf[count] = 0; /* buffer is supposed to end with 0 */
 753         if (kbuf[count - 1] == '\n')
 754                 kbuf[count - 1] = 0;
 755         ptr = kbuf;
 756
 757         /* fsname=<file system name> must be the 1st entry */
 758         while (ptr != NULL) {
 759                 char *tmpptr;
 760                 int i;
 761
 762                 tmpptr = strchr(ptr, ';');
 763                 if (tmpptr)
 764                         *tmpptr++ = 0;
 765
 766                 rc = -EINVAL;
 767                 for (i = 0; i < ARRAY_SIZE(ir_cmds); i++) {
 768                         struct lproc_ir_cmd *cmd;
 769                         int cmdlen;
 770
 771                         cmd    = &ir_cmds[i];
 772                         cmdlen = cmd->namelen;
 773                         if (strncmp(cmd->name, ptr, cmdlen) == 0) {
 774                                 ptr += cmdlen;
 775                                 rc = cmd->handler(fsdb, ptr);
 776                                 break;
 777                         }
 778                 }
 779                 if (rc)
 780                         break;
 781
 782                 ptr = tmpptr;
 783         }
 784         if (rc)
 785                 CERROR("Unable to process command: %s(%d)\n", ptr, rc);
 786         OBD_FREE(kbuf, count + 1);
 787         return rc ?: count;
 788 }
 789
 790 int lprocfs_rd_ir_state(struct seq_file *seq, void *data)
 791 {
 792         struct fs_db      *fsdb = data;
 793         struct mgs_nidtbl *tbl  = &fsdb->fsdb_nidtbl;
 794         const char        *ir_strings[] = IR_STRINGS;
 795         struct timeval     tv_max;
 796         struct timeval     tv;
 797
 798         /* mgs_live_seq_show() already holds fsdb_sem. */
 799         ir_state_graduate(fsdb);
 800
 801         seq_printf(seq, "\nimperative_recovery_state:\n");
 802         seq_printf(seq,
 803                    "    state: %s\n"
 804                    "    nonir_clients: %d\n"
 805                    "    nidtbl_version: %lld\n",
 806                    ir_strings[fsdb->fsdb_ir_state], fsdb->fsdb_nonir_clients,
 807                    tbl->mn_version);
 808
 809         cfs_duration_usec(fsdb->fsdb_notify_total, &tv);
 810         cfs_duration_usec(fsdb->fsdb_notify_max, &tv_max);
 811
 812         seq_printf(seq, "    notify_duration_total: %lu.%06lu\n"
 813                         "    notify_duation_max: %lu.%06lu\n"
 814                         "    notify_count: %u\n",
 815                    tv.tv_sec, tv.tv_usec,
 816                    tv_max.tv_sec, tv_max.tv_usec,
 817                    fsdb->fsdb_notify_count);
 818
 819         return 0;
 820 }
 821
 822 int lprocfs_rd_ir_timeout(char *page, char **start, off_t off, int count,
 823                           int *eof, void *data)
 824 {
 825         *eof = 1;
 826         return snprintf(page, count, "%d\n", ir_timeout);
 827 }
 828
 829 int lprocfs_wr_ir_timeout(struct file *file, const char *buffer,
 830                           unsigned long count, void *data)
 831 {
 832         return lprocfs_wr_uint(file, buffer, count, &ir_timeout);
 833 }
 834
 835 /* --------------- Handle non IR support clients --------------- */
 836 /* attach a lustre file system to an export */
 837 int mgs_fsc_attach(struct obd_export *exp, char *fsname)
 838 {
 839         struct mgs_export_data *data = &exp->u.eu_mgs_data;
 840         struct obd_device *obd       = exp->exp_obd;
 841         struct fs_db      *fsdb;
 842         struct mgs_fsc    *fsc     = NULL;
 843         struct mgs_fsc    *new_fsc = NULL;
 844         bool               found   = false;
 845         int                rc;
 846         ENTRY;
 847
 848         rc = mgs_find_or_make_fsdb(obd, fsname, &fsdb);
 849         if (rc)
 850                 RETURN(rc);
 851
 852         /* allocate a new fsc in case we need it in spinlock. */
 853         OBD_ALLOC_PTR(new_fsc);
 854         if (new_fsc == NULL)
 855                 RETURN(-ENOMEM);
 856
 857         CFS_INIT_LIST_HEAD(&new_fsc->mfc_export_list);
 858         CFS_INIT_LIST_HEAD(&new_fsc->mfc_fsdb_list);
 859         new_fsc->mfc_fsdb       = fsdb;
 860         new_fsc->mfc_export     = class_export_get(exp);
 861         new_fsc->mfc_ir_capable =
 862                         !!(exp->exp_connect_flags & OBD_CONNECT_IMP_RECOV);
 863
 864         rc = -EEXIST;
 865         cfs_down(&fsdb->fsdb_sem);
 866
 867         /* tend to find it in export list because this list is shorter. */
 868         cfs_spin_lock(&data->med_lock);
 869         cfs_list_for_each_entry(fsc, &data->med_clients, mfc_export_list) {
 870                 if (strcmp(fsname, fsc->mfc_fsdb->fsdb_name) == 0) {
 871                         found = true;
 872                         break;
 873                 }
 874         }
 875         if (!found) {
 876                 fsc = new_fsc;
 877                 new_fsc = NULL;
 878
 879                 /* add it into export list. */
 880                 cfs_list_add(&fsc->mfc_export_list, &data->med_clients);
 881
 882                 /* add into fsdb list. */
 883                 cfs_list_add(&fsc->mfc_fsdb_list, &fsdb->fsdb_clients);
 884                 if (!fsc->mfc_ir_capable) {
 885                         ++fsdb->fsdb_nonir_clients;
 886                         if (fsdb->fsdb_ir_state == IR_FULL)
 887                                 fsdb->fsdb_ir_state = IR_PARTIAL;
 888                 }
 889                 rc = 0;
 890         }
 891         cfs_spin_unlock(&data->med_lock);
 892         cfs_up(&fsdb->fsdb_sem);
 893
 894         if (new_fsc) {
 895                 class_export_put(new_fsc->mfc_export);
 896                 OBD_FREE_PTR(new_fsc);
 897         }
 898         RETURN(rc);
 899 }
 900
 901 void mgs_fsc_cleanup(struct obd_export *exp)
 902 {
 903         struct mgs_export_data *data = &exp->u.eu_mgs_data;
 904         struct mgs_fsc *fsc, *tmp;
 905         CFS_LIST_HEAD(head);
 906
 907         cfs_spin_lock(&data->med_lock);
 908         cfs_list_splice_init(&data->med_clients, &head);
 909         cfs_spin_unlock(&data->med_lock);
 910
 911         cfs_list_for_each_entry_safe(fsc, tmp, &head, mfc_export_list) {
 912                 struct fs_db *fsdb = fsc->mfc_fsdb;
 913
 914                 LASSERT(fsc->mfc_export == exp);
 915
 916                 cfs_down(&fsdb->fsdb_sem);
 917                 cfs_list_del_init(&fsc->mfc_fsdb_list);
 918                 if (fsc->mfc_ir_capable == 0) {
 919                         --fsdb->fsdb_nonir_clients;
 920                         LASSERT(fsdb->fsdb_ir_state != IR_FULL);
 921                         if (fsdb->fsdb_nonir_clients == 0 &&
 922                             fsdb->fsdb_ir_state == IR_PARTIAL)
 923                                 fsdb->fsdb_ir_state = IR_FULL;
 924                 }
 925                 cfs_up(&fsdb->fsdb_sem);
 926                 cfs_list_del_init(&fsc->mfc_export_list);
 927                 class_export_put(fsc->mfc_export);
 928                 OBD_FREE_PTR(fsc);
 929         }
 930 }
 931
 932 /* must be called with fsdb->fsdb_sem held */
 933 void mgs_fsc_cleanup_by_fsdb(struct fs_db *fsdb)
 934 {
 935         struct mgs_fsc *fsc, *tmp;
 936
 937         cfs_list_for_each_entry_safe(fsc, tmp, &fsdb->fsdb_clients,
 938                                      mfc_fsdb_list) {
 939                 struct mgs_export_data *data = &fsc->mfc_export->u.eu_mgs_data;
 940
 941                 LASSERT(fsdb == fsc->mfc_fsdb);
 942                 cfs_list_del_init(&fsc->mfc_fsdb_list);
 943
 944                 cfs_spin_lock(&data->med_lock);
 945                 cfs_list_del_init(&fsc->mfc_export_list);
 946                 cfs_spin_unlock(&data->med_lock);
 947                 class_export_put(fsc->mfc_export);
 948                 OBD_FREE_PTR(fsc);
 949         }
 950
 951         fsdb->fsdb_nonir_clients = 0;
 952         if (fsdb->fsdb_ir_state == IR_PARTIAL)
 953                 fsdb->fsdb_ir_state = IR_FULL;
 954 }