lustre/obdclass/lprocfs_status_server.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, 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).
  15  *
  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
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2014, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * lustre/obdclass/lprocfs_status_server.c
  32  */
  33
  34 #define DEBUG_SUBSYSTEM S_CLASS
  35
  36 #include <linux/kobject.h>
  37 #include <linux/sysfs.h>
  38
  39 #include <obd_class.h>
  40 #include <lprocfs_status.h>
  41 #include <lustre_nodemap.h>
  42
  43 #define MAX_STRING_SIZE 128
  44
  45 struct dentry *ldebugfs_add_symlink(const char *name, const char *target,
  46                                     const char *format, ...)
  47 {
  48         struct dentry *entry = NULL;
  49         struct dentry *parent;
  50         struct qstr dname;
  51         va_list ap;
  52         char *dest;
  53
  54         if (!target || !format)
  55                 return NULL;
  56
  57         dname.name = target;
  58         dname.len = strlen(dname.name);
  59         dname.hash = ll_full_name_hash(debugfs_lustre_root,
  60                                        dname.name, dname.len);
  61         parent = d_lookup(debugfs_lustre_root, &dname);
  62         if (!parent)
  63                 return NULL;
  64
  65         OBD_ALLOC_WAIT(dest, MAX_STRING_SIZE + 1);
  66         if (!dest)
  67                 goto no_entry;
  68
  69         va_start(ap, format);
  70         vsnprintf(dest, MAX_STRING_SIZE, format, ap);
  71         va_end(ap);
  72
  73         entry = debugfs_create_symlink(name, parent, dest);
  74
  75         OBD_FREE(dest, MAX_STRING_SIZE + 1);
  76 no_entry:
  77         dput(parent);
  78         return entry;
  79 }
  80 EXPORT_SYMBOL(ldebugfs_add_symlink);
  81
  82 int lprocfs_recovery_stale_clients_seq_show(struct seq_file *m, void *data)
  83 {
  84         struct obd_device *obd = m->private;
  85         struct obd_export *exp, *n;
  86         int connected;
  87
  88         if (!obd->obd_recovering ||
  89             atomic_read(&obd->obd_connected_clients) >=
  90             atomic_read(&obd->obd_max_recoverable_clients))
  91                 /* not in recovery */
  92                 return 0;
  93
  94         spin_lock(&obd->obd_dev_lock);
  95         list_for_each_entry_safe(exp, n, &obd->obd_exports, exp_obd_chain) {
  96                 /* don't count self-export as client */
  97                 if (obd_uuid_equals(&exp->exp_client_uuid,
  98                                     &exp->exp_obd->obd_uuid))
  99                         continue;
 100
 101                 /* don't count clients which have no slot in last_rcvd
 102                  * (e.g. lightweight connection)
 103                  */
 104                 if (exp->exp_target_data.ted_lr_idx == -1)
 105                         continue;
 106
 107                 connected = !exp->exp_failed && (exp->exp_conn_cnt > 0);
 108
 109                 if (!connected)
 110                         seq_printf(m, "%s\n", exp->exp_client_uuid.uuid);
 111         }
 112         spin_unlock(&obd->obd_dev_lock);
 113
 114         return 0;
 115 }
 116 EXPORT_SYMBOL(lprocfs_recovery_stale_clients_seq_show);
 117
 118 #ifdef CONFIG_PROC_FS
 119 #define BUFLEN (UUID_MAX + 5)
 120
 121 ssize_t
 122 lprocfs_evict_client_seq_write(struct file *file, const char __user *buffer,
 123                                size_t count, loff_t *off)
 124 {
 125         struct seq_file *m = file->private_data;
 126         struct obd_device *obd = m->private;
 127         char *tmpbuf, *kbuf;
 128
 129         OBD_ALLOC(kbuf, BUFLEN);
 130         if (kbuf == NULL)
 131                 return -ENOMEM;
 132
 133         /*
 134          * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
 135          * bytes into kbuf, to ensure that the string is NUL-terminated.
 136          * UUID_MAX should include a trailing NUL already.
 137          */
 138         if (copy_from_user(kbuf, buffer,
 139                            min_t(unsigned long, BUFLEN - 1, count))) {
 140                 count = -EFAULT;
 141                 goto out;
 142         }
 143         tmpbuf = skip_spaces(kbuf);
 144         tmpbuf = strsep(&tmpbuf, " \t\n\f\v\r");
 145         class_incref(obd, __func__, current);
 146
 147         if (strncmp(tmpbuf, "nid:", 4) == 0)
 148                 obd_export_evict_by_nid(obd, tmpbuf + 4);
 149         else if (strncmp(tmpbuf, "uuid:", 5) == 0)
 150                 obd_export_evict_by_uuid(obd, tmpbuf + 5);
 151         else
 152                 obd_export_evict_by_uuid(obd, tmpbuf);
 153
 154         class_decref(obd, __func__, current);
 155
 156 out:
 157         OBD_FREE(kbuf, BUFLEN);
 158         return count;
 159 }
 160 EXPORT_SYMBOL(lprocfs_evict_client_seq_write);
 161
 162 #undef BUFLEN
 163
 164 ssize_t eviction_count_show(struct kobject *kobj, struct attribute *attr,
 165                          char *buf)
 166 {
 167         struct obd_device *obd = container_of(kobj, struct obd_device,
 168                                               obd_kset.kobj);
 169
 170         return scnprintf(buf, PAGE_SIZE, "%u\n",
 171                          atomic_read(&obd->obd_eviction_count));
 172 }
 173 EXPORT_SYMBOL(eviction_count_show);
 174
 175 ssize_t num_exports_show(struct kobject *kobj, struct attribute *attr,
 176                          char *buf)
 177 {
 178         struct obd_device *obd = container_of(kobj, struct obd_device,
 179                                               obd_kset.kobj);
 180
 181         return scnprintf(buf, PAGE_SIZE, "%u\n", obd->obd_num_exports);
 182 }
 183 EXPORT_SYMBOL(num_exports_show);
 184
 185 ssize_t grant_check_threshold_show(struct kobject *kobj, struct attribute *attr,
 186                                    char *buf)
 187 {
 188         struct obd_device *obd = container_of(kobj, struct obd_device,
 189                                               obd_kset.kobj);
 190
 191         return scnprintf(buf, PAGE_SIZE, "%d\n",
 192                          obd->obd_grant_check_threshold);
 193 }
 194 EXPORT_SYMBOL(grant_check_threshold_show);
 195
 196 ssize_t grant_check_threshold_store(struct kobject *kobj,
 197                                     struct attribute *attr,
 198                                     const char *buffer, size_t count)
 199 {
 200         struct obd_device *obd = container_of(kobj, struct obd_device,
 201                                               obd_kset.kobj);
 202         int val;
 203         int rc;
 204
 205         rc = kstrtoint(buffer, 10, &val);
 206         if (rc)
 207                 return rc;
 208
 209         if (val < 0)
 210                 return -EINVAL;
 211         obd->obd_grant_check_threshold = val;
 212         return count;
 213 }
 214 EXPORT_SYMBOL(grant_check_threshold_store);
 215
 216 static int obd_export_flags2str(struct obd_export *exp, struct seq_file *m)
 217 {
 218         bool first = true;
 219
 220         flag2str(exp, failed);
 221         flag2str(exp, in_recovery);
 222         flag2str(exp, disconnected);
 223         flag2str(exp, connecting);
 224         flag2str(exp, no_recovery);
 225
 226         return 0;
 227 }
 228
 229 static int
 230 lprocfs_exp_print_export_seq(struct obd_export *exp, void *cb_data)
 231 {
 232         struct seq_file         *m = cb_data;
 233         struct obd_device       *obd;
 234         struct obd_connect_data *ocd;
 235
 236         LASSERT(exp != NULL);
 237         if (exp->exp_nid_stats == NULL)
 238                 goto out;
 239         obd = exp->exp_obd;
 240         ocd = &exp->exp_connect_data;
 241
 242         seq_printf(m, "%s:\n"
 243                    "    name: %s\n"
 244                    "    client: %s\n"
 245                    "    connect_flags: [ ",
 246                    obd_uuid2str(&exp->exp_client_uuid),
 247                    obd->obd_name,
 248                    obd_export_nid2str(exp));
 249         obd_connect_seq_flags2str(m, ocd->ocd_connect_flags,
 250                                   ocd->ocd_connect_flags2, ", ");
 251         seq_printf(m, " ]\n");
 252         obd_connect_data_seqprint(m, ocd);
 253         seq_printf(m, "    export_flags: [ ");
 254         obd_export_flags2str(exp, m);
 255         seq_printf(m, " ]\n");
 256
 257         if (obd->obd_type &&
 258             strcmp(obd->obd_type->typ_name, "obdfilter") == 0) {
 259                 struct filter_export_data *fed = &exp->exp_filter_data;
 260
 261                 seq_printf(m, "    grant:\n");
 262                 seq_printf(m, "       granted: %ld\n",
 263                         fed->fed_ted.ted_grant);
 264                 seq_printf(m, "       dirty: %ld\n",
 265                         fed->fed_ted.ted_dirty);
 266                 seq_printf(m, "       pending: %ld\n",
 267                         fed->fed_ted.ted_pending);
 268         }
 269
 270 out:
 271         return 0;
 272 }
 273
 274 /**
 275  * RPC connections are composed of an import and an export. Using the
 276  * lctl utility we can extract important information about the state.
 277  * The lprocfs_exp_export_seq_show routine displays the state information
 278  * for the export.
 279  *
 280  * \param[in] m         seq file
 281  * \param[in] data      unused
 282  *
 283  * \retval              0 on success
 284  *
 285  * The format of the export state information is like:
 286  * a793e354-49c0-aa11-8c4f-a4f2b1a1a92b:
 287  *     name: MGS
 288  *     client: 10.211.55.10@tcp
 289  *     connect_flags: [ version, barrier, adaptive_timeouts, ... ]
 290  *     connect_data:
 291  *        flags: 0x2000011005002020
 292  *        instance: 0
 293  *        target_version: 2.10.51.0
 294  *        export_flags: [ ... ]
 295  *
 296  */
 297 static int lprocfs_exp_export_seq_show(struct seq_file *m, void *data)
 298 {
 299         struct nid_stat *stats = m->private;
 300
 301         return obd_nid_export_for_each(stats->nid_obd, &stats->nid,
 302                                        lprocfs_exp_print_export_seq, m);
 303 }
 304 LPROC_SEQ_FOPS_RO(lprocfs_exp_export);
 305
 306 static void lprocfs_free_client_stats(struct nid_stat *client_stat)
 307 {
 308         CDEBUG(D_CONFIG, "stat %p - data %p/%p\n", client_stat,
 309                client_stat->nid_proc, client_stat->nid_stats);
 310
 311         LASSERTF(atomic_read(&client_stat->nid_exp_ref_count) == 0,
 312                  "nid %s:count %d\n", libcfs_nidstr(&client_stat->nid),
 313                  atomic_read(&client_stat->nid_exp_ref_count));
 314
 315         if (client_stat->nid_proc)
 316                 lprocfs_remove(&client_stat->nid_proc);
 317
 318         if (client_stat->nid_stats)
 319                 lprocfs_stats_free(&client_stat->nid_stats);
 320
 321         if (client_stat->nid_ldlm_stats)
 322                 lprocfs_stats_free(&client_stat->nid_ldlm_stats);
 323
 324         OBD_FREE_PTR(client_stat);
 325 }
 326
 327 void lprocfs_free_per_client_stats(struct obd_device *obd)
 328 {
 329         struct cfs_hash *hash = obd->obd_nid_stats_hash;
 330         struct nid_stat *stat;
 331         ENTRY;
 332
 333         /* we need extra list - because hash_exit called to early */
 334         /* not need locking because all clients is died */
 335         while (!list_empty(&obd->obd_nid_stats)) {
 336                 stat = list_first_entry(&obd->obd_nid_stats,
 337                                         struct nid_stat, nid_list);
 338                 list_del_init(&stat->nid_list);
 339                 cfs_hash_del(hash, &stat->nid, &stat->nid_hash);
 340                 lprocfs_free_client_stats(stat);
 341         }
 342         EXIT;
 343 }
 344 EXPORT_SYMBOL(lprocfs_free_per_client_stats);
 345
 346 static int
 347 lprocfs_exp_print_nodemap_seq(struct obd_export *exp, void *cb_data)
 348 {
 349         struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap;
 350         struct seq_file *m = cb_data;
 351
 352         if (nodemap)
 353                 seq_printf(m, "%s\n", nodemap->nm_name);
 354         return 0;
 355 }
 356
 357 static int
 358 lprocfs_exp_nodemap_seq_show(struct seq_file *m, void *data)
 359 {
 360         struct nid_stat *stats = m->private;
 361
 362         return obd_nid_export_for_each(stats->nid_obd, &stats->nid,
 363                                        lprocfs_exp_print_nodemap_seq, m);
 364 }
 365 LPROC_SEQ_FOPS_RO(lprocfs_exp_nodemap);
 366
 367 static int
 368 lprocfs_exp_print_uuid_seq(struct obd_export *exp, void *cb_data)
 369 {
 370         struct seq_file *m = cb_data;
 371
 372         if (exp->exp_nid_stats)
 373                 seq_printf(m, "%s\n", obd_uuid2str(&exp->exp_client_uuid));
 374         return 0;
 375 }
 376
 377 static int lprocfs_exp_uuid_seq_show(struct seq_file *m, void *data)
 378 {
 379         struct nid_stat *stats = m->private;
 380
 381         return obd_nid_export_for_each(stats->nid_obd, &stats->nid,
 382                                        lprocfs_exp_print_uuid_seq, m);
 383 }
 384 LPROC_SEQ_FOPS_RO(lprocfs_exp_uuid);
 385
 386 #define HASH_NAME_LEN   16
 387
 388 static void ldebugfs_rhash_seq_show(const char *name, struct rhashtable *ht,
 389                                     struct seq_file *m)
 390 {
 391         unsigned int max_size = ht->p.max_size ? ht->p.max_size : UINT_MAX;
 392         struct bucket_table *tbl;
 393         int dist[8] = { 0, };
 394         int maxdep = 0;
 395         int i;
 396
 397         rcu_read_lock();
 398         tbl = rht_dereference(ht->tbl, ht);
 399         for (i = 0; i < tbl->size; i++) {
 400                 struct rhash_head *pos;
 401                 int count = 0;
 402
 403                 rht_for_each(pos, tbl, i)
 404                         count++;
 405
 406                 if (count)
 407                         maxdep = max(maxdep, count);
 408
 409                 dist[min(fls(count), 7)]++;
 410         }
 411
 412         seq_printf(m, "%-*s %5d %5d %10u %d.%03d 0.300 0.750 0x%03x %7d %7d %7d ",
 413                    HASH_NAME_LEN, name, tbl->size, ht->p.min_size, max_size,
 414                    atomic_read(&ht->nelems) / tbl->size,
 415                    atomic_read(&ht->nelems) * 1000 / tbl->size,
 416                    ht->p.automatic_shrinking, 0,
 417                    atomic_read(&ht->nelems), maxdep);
 418         rcu_read_unlock();
 419
 420         for (i = 0; i < 8; i++)
 421                 seq_printf(m, "%d%c",  dist[i], (i == 7) ? '\n' : '/');
 422 }
 423
 424 static int
 425 lprocfs_exp_print_hash_seq(struct obd_export *exp, void *cb_data)
 426
 427 {
 428         struct obd_device *obd = exp->exp_obd;
 429         struct seq_file *m = cb_data;
 430
 431         if (exp->exp_lock_hash != NULL) {
 432                 seq_printf(m, "%-*s   cur   min        max theta t-min t-max flags rehash   count distribution\n",
 433                            HASH_NAME_LEN, "name");
 434                 ldebugfs_rhash_seq_show("NID_HASH", &obd->obd_nid_hash.ht, m);
 435         }
 436         return 0;
 437 }
 438
 439 static int lprocfs_exp_hash_seq_show(struct seq_file *m, void *data)
 440 {
 441         struct nid_stat *stats = m->private;
 442
 443         return obd_nid_export_for_each(stats->nid_obd, &stats->nid,
 444                                        lprocfs_exp_print_hash_seq, m);
 445 }
 446 LPROC_SEQ_FOPS_RO(lprocfs_exp_hash);
 447
 448 static int lprocfs_exp_print_replydata_seq(struct obd_export *exp,
 449                                            void *cb_data)
 450
 451 {
 452         struct seq_file *m = cb_data;
 453         struct tg_export_data *ted = &exp->exp_target_data;
 454
 455         seq_printf(m, "reply_cnt: %d\n"
 456                    "reply_max: %d\n"
 457                    "reply_released_by_xid: %d\n"
 458                    "reply_released_by_tag: %d\n\n",
 459                    ted->ted_reply_cnt,
 460                    ted->ted_reply_max,
 461                    ted->ted_release_xid,
 462                    ted->ted_release_tag);
 463         return 0;
 464 }
 465
 466 static int lprocfs_exp_replydata_seq_show(struct seq_file *m, void *data)
 467 {
 468         struct nid_stat *stats = m->private;
 469
 470         return obd_nid_export_for_each(stats->nid_obd, &stats->nid,
 471                                        lprocfs_exp_print_replydata_seq, m);
 472 }
 473 LPROC_SEQ_FOPS_RO(lprocfs_exp_replydata);
 474
 475 static int lprocfs_exp_print_fmd_count_seq(struct obd_export *exp,
 476                                            void *cb_data)
 477 {
 478         struct seq_file *m = cb_data;
 479         struct tg_export_data *ted = &exp->exp_target_data;
 480
 481         seq_printf(m, "%d\n", ted->ted_fmd_count);
 482
 483         return 0;
 484 }
 485
 486 static int lprocfs_exp_fmd_count_seq_show(struct seq_file *m, void *data)
 487 {
 488         struct nid_stat *stats = m->private;
 489
 490         return obd_nid_export_for_each(stats->nid_obd, &stats->nid,
 491                                        lprocfs_exp_print_fmd_count_seq, m);
 492 }
 493 LPROC_SEQ_FOPS_RO(lprocfs_exp_fmd_count);
 494
 495 int lprocfs_nid_stats_clear_seq_show(struct seq_file *m, void *data)
 496 {
 497         seq_puts(m, "Write into this file to clear all nid stats and stale nid entries\n");
 498         return 0;
 499 }
 500 EXPORT_SYMBOL(lprocfs_nid_stats_clear_seq_show);
 501
 502 static int lprocfs_nid_stats_clear_write_cb(void *obj, void *data)
 503 {
 504         struct nid_stat *stat = obj;
 505         ENTRY;
 506
 507         CDEBUG(D_INFO, "refcnt %d\n", atomic_read(&stat->nid_exp_ref_count));
 508         if (atomic_read(&stat->nid_exp_ref_count) == 1) {
 509                 /* object has only hash references. */
 510                 spin_lock(&stat->nid_obd->obd_nid_lock);
 511                 list_move(&stat->nid_list, data);
 512                 spin_unlock(&stat->nid_obd->obd_nid_lock);
 513                 RETURN(1);
 514         }
 515         /* we has reference to object - only clear data*/
 516         if (stat->nid_stats)
 517                 lprocfs_stats_clear(stat->nid_stats);
 518
 519         RETURN(0);
 520 }
 521
 522 ssize_t
 523 lprocfs_nid_stats_clear_seq_write(struct file *file, const char __user *buffer,
 524                                         size_t count, loff_t *off)
 525 {
 526         struct seq_file *m = file->private_data;
 527         struct obd_device *obd = m->private;
 528         struct nid_stat *client_stat;
 529         LIST_HEAD(free_list);
 530
 531         cfs_hash_cond_del(obd->obd_nid_stats_hash,
 532                           lprocfs_nid_stats_clear_write_cb, &free_list);
 533
 534         while (!list_empty(&free_list)) {
 535                 client_stat = list_first_entry(&free_list, struct nid_stat,
 536                                                nid_list);
 537                 list_del_init(&client_stat->nid_list);
 538                 lprocfs_free_client_stats(client_stat);
 539         }
 540         return count;
 541 }
 542 EXPORT_SYMBOL(lprocfs_nid_stats_clear_seq_write);
 543
 544 int lprocfs_exp_setup(struct obd_export *exp, struct lnet_nid *nid)
 545 {
 546         struct nid_stat *new_stat, *old_stat;
 547         struct obd_device *obd = NULL;
 548         struct proc_dir_entry *entry;
 549         char nidstr[LNET_NIDSTR_SIZE];
 550         int rc = 0;
 551         ENTRY;
 552
 553         if (!exp || !exp->exp_obd || !exp->exp_obd->obd_proc_exports_entry ||
 554             !exp->exp_obd->obd_nid_stats_hash)
 555                 RETURN(-EINVAL);
 556
 557         /* not test against zero because eric say:
 558          * You may only test nid against another nid, or LNET_NID_ANY.
 559          * Anything else is nonsense.*/
 560         if (nid == NULL || LNET_NID_IS_ANY(nid))
 561                 RETURN(-EALREADY);
 562
 563         libcfs_nidstr_r(nid, nidstr, sizeof(nidstr));
 564
 565         spin_lock(&exp->exp_lock);
 566         if (exp->exp_nid_stats != NULL) {
 567                 spin_unlock(&exp->exp_lock);
 568                 RETURN(-EALREADY);
 569         }
 570         spin_unlock(&exp->exp_lock);
 571
 572         obd = exp->exp_obd;
 573
 574         CDEBUG(D_CONFIG, "using hash %p\n", obd->obd_nid_stats_hash);
 575
 576         OBD_ALLOC_PTR(new_stat);
 577         if (new_stat == NULL)
 578                 RETURN(-ENOMEM);
 579
 580         new_stat->nid = *nid;
 581         new_stat->nid_obd = exp->exp_obd;
 582         /* we need set default refcount to 1 to balance obd_disconnect */
 583         atomic_set(&new_stat->nid_exp_ref_count, 1);
 584
 585         old_stat = cfs_hash_findadd_unique(obd->obd_nid_stats_hash,
 586                                            &new_stat->nid,
 587                                            &new_stat->nid_hash);
 588         CDEBUG(D_INFO, "Found stats %p for nid %s - ref %d\n",
 589                old_stat, nidstr, atomic_read(&old_stat->nid_exp_ref_count));
 590
 591         /* Return -EALREADY here so that we know that the /proc
 592          * entry already has been created */
 593         if (old_stat != new_stat) {
 594                 spin_lock(&exp->exp_lock);
 595                 if (exp->exp_nid_stats) {
 596                         LASSERT(exp->exp_nid_stats == old_stat);
 597                         nidstat_putref(exp->exp_nid_stats);
 598                 }
 599                 exp->exp_nid_stats = old_stat;
 600                 spin_unlock(&exp->exp_lock);
 601                 GOTO(destroy_new, rc = -EALREADY);
 602         }
 603         /* not found - create */
 604         new_stat->nid_proc = lprocfs_register(nidstr,
 605                                               obd->obd_proc_exports_entry,
 606                                               NULL, NULL);
 607
 608         if (IS_ERR(new_stat->nid_proc)) {
 609                 rc = PTR_ERR(new_stat->nid_proc);
 610                 new_stat->nid_proc = NULL;
 611                 CERROR("%s: cannot create proc entry for export %s: rc = %d\n",
 612                        obd->obd_name, nidstr, rc);
 613                 GOTO(destroy_new_ns, rc);
 614         }
 615
 616         entry = lprocfs_add_simple(new_stat->nid_proc, "nodemap", new_stat,
 617                                    &lprocfs_exp_nodemap_fops);
 618         if (IS_ERR(entry)) {
 619                 rc = PTR_ERR(entry);
 620                 CWARN("%s: error adding the nodemap file: rc = %d\n",
 621                       obd->obd_name, rc);
 622                 GOTO(destroy_new_ns, rc);
 623         }
 624
 625         entry = lprocfs_add_simple(new_stat->nid_proc, "uuid", new_stat,
 626                                    &lprocfs_exp_uuid_fops);
 627         if (IS_ERR(entry)) {
 628                 rc = PTR_ERR(entry);
 629                 CWARN("%s: error adding the NID stats file: rc = %d\n",
 630                       obd->obd_name, rc);
 631                 GOTO(destroy_new_ns, rc);
 632         }
 633
 634         entry = lprocfs_add_simple(new_stat->nid_proc, "hash", new_stat,
 635                                    &lprocfs_exp_hash_fops);
 636         if (IS_ERR(entry)) {
 637                 rc = PTR_ERR(entry);
 638                 CWARN("%s: error adding the hash file: rc = %d\n",
 639                       obd->obd_name, rc);
 640                 GOTO(destroy_new_ns, rc);
 641         }
 642
 643         entry = lprocfs_add_simple(new_stat->nid_proc, "export",
 644                                    new_stat, &lprocfs_exp_export_fops);
 645         if (IS_ERR(entry)) {
 646                 rc = PTR_ERR(entry);
 647                 CWARN("%s: error adding the export file: rc = %d\n",
 648                       obd->obd_name, rc);
 649                 GOTO(destroy_new_ns, rc);
 650         }
 651
 652         entry = lprocfs_add_simple(new_stat->nid_proc, "reply_data", new_stat,
 653                                    &lprocfs_exp_replydata_fops);
 654         if (IS_ERR(entry)) {
 655                 rc = PTR_ERR(entry);
 656                 CWARN("%s: error adding the reply_data file: rc = %d\n",
 657                       obd->obd_name, rc);
 658                 GOTO(destroy_new_ns, rc);
 659         }
 660
 661         entry = lprocfs_add_simple(new_stat->nid_proc, "fmd_count", new_stat,
 662                                    &lprocfs_exp_fmd_count_fops);
 663         if (IS_ERR(entry)) {
 664                 rc = PTR_ERR(entry);
 665                 CWARN("%s: error adding the fmd_count file: rc = %d\n",
 666                       obd->obd_name, rc);
 667                 GOTO(destroy_new_ns, rc);
 668         }
 669
 670         spin_lock(&exp->exp_lock);
 671         exp->exp_nid_stats = new_stat;
 672         spin_unlock(&exp->exp_lock);
 673
 674         /* protect competitive add to list, not need locking on destroy */
 675         spin_lock(&obd->obd_nid_lock);
 676         list_add(&new_stat->nid_list, &obd->obd_nid_stats);
 677         spin_unlock(&obd->obd_nid_lock);
 678
 679         RETURN(0);
 680
 681 destroy_new_ns:
 682         if (new_stat->nid_proc != NULL)
 683                 lprocfs_remove(&new_stat->nid_proc);
 684         cfs_hash_del(obd->obd_nid_stats_hash, &new_stat->nid,
 685                      &new_stat->nid_hash);
 686
 687 destroy_new:
 688         nidstat_putref(new_stat);
 689         OBD_FREE_PTR(new_stat);
 690         RETURN(rc);
 691 }
 692 EXPORT_SYMBOL(lprocfs_exp_setup);
 693
 694 int lprocfs_exp_cleanup(struct obd_export *exp)
 695 {
 696         struct nid_stat *stat = exp->exp_nid_stats;
 697
 698         if (!stat || !exp->exp_obd)
 699                 RETURN(0);
 700
 701         nidstat_putref(exp->exp_nid_stats);
 702         exp->exp_nid_stats = NULL;
 703
 704         return 0;
 705 }
 706
 707 int lprocfs_alloc_obd_stats(struct obd_device *obd, unsigned int num_stats)
 708 {
 709         struct lprocfs_stats *stats;
 710         int rc;
 711
 712         LASSERT(obd->obd_stats == NULL);
 713         LASSERT(obd->obd_proc_entry != NULL);
 714
 715         stats = lprocfs_stats_alloc(num_stats, 0);
 716         if (stats == NULL)
 717                 return -ENOMEM;
 718
 719         rc = lprocfs_stats_register(obd->obd_proc_entry, "stats", stats);
 720         if (rc < 0)
 721                 lprocfs_stats_free(&stats);
 722         else
 723                 obd->obd_stats = stats;
 724
 725         return rc;
 726 }
 727 EXPORT_SYMBOL(lprocfs_alloc_obd_stats);
 728
 729 void lprocfs_free_obd_stats(struct obd_device *obd)
 730 {
 731         if (obd->obd_stats)
 732                 lprocfs_stats_free(&obd->obd_stats);
 733 }
 734 EXPORT_SYMBOL(lprocfs_free_obd_stats);
 735
 736 static void display_brw_stats(struct seq_file *seq, const char *name,
 737                               const char *units, struct obd_hist_pcpu *read,
 738                               struct obd_hist_pcpu *write, bool scale)
 739 {
 740         unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
 741         unsigned int i;
 742
 743         seq_printf(seq, "\n%26s read      |     write\n", " ");
 744         seq_printf(seq, "%-22s %-5s %% cum %% |  %-11s %% cum %%\n",
 745                    name, units, units);
 746
 747         read_tot = lprocfs_oh_sum_pcpu(read);
 748         write_tot = lprocfs_oh_sum_pcpu(write);
 749
 750         if (!read_tot && !write_tot)
 751                 return;
 752
 753         for (i = 0; i < OBD_HIST_MAX; i++) {
 754                 r = lprocfs_oh_counter_pcpu(read, i);
 755                 w = lprocfs_oh_counter_pcpu(write, i);
 756                 read_cum += r;
 757                 write_cum += w;
 758                 if (read_cum == 0 && write_cum == 0)
 759                         continue;
 760
 761                 if (!scale)
 762                         seq_printf(seq, "%u", i);
 763                 else if (i < 10)
 764                         seq_printf(seq, "%lu", BIT(i));
 765                 else if (i < 20)
 766                         seq_printf(seq, "%luK", BIT(i - 10));
 767                 else
 768                         seq_printf(seq, "%luM", BIT(i - 20));
 769
 770                 seq_printf(seq, ":\t\t%10lu %3u %3u   | %4lu %3u %3u\n",
 771                            r, pct(r, read_tot), pct(read_cum, read_tot),
 772                            w, pct(w, write_tot), pct(write_cum, write_tot));
 773
 774                 if (read_cum == read_tot && write_cum == write_tot)
 775                         break;
 776         }
 777 }
 778
 779 static const struct brw_stats_props brw_props[] = {
 780         { .bsp_name     = "pages per bulk r/w",
 781           .bsp_units    = "rpcs",
 782           .bsp_scale    = true                          },
 783         { .bsp_name     = "discontiguous pages",
 784           .bsp_units    = "rpcs",
 785           .bsp_scale    = false                         },
 786         { .bsp_name     = "discontiguous blocks",
 787           .bsp_units    = "rpcs",
 788           .bsp_scale    = false                         },
 789         { .bsp_name     = "disk fragmented I/Os",
 790           .bsp_units    = "ios",
 791           .bsp_scale    = false                         },
 792         { .bsp_name     = "disk I/Os in flight",
 793           .bsp_units    = "ios",
 794           .bsp_scale    = false                         },
 795         { .bsp_name     = "I/O time (1/1000s)",
 796           .bsp_units    = "ios",
 797           .bsp_scale    = true                          },
 798         { .bsp_name     = "disk I/O size",
 799           .bsp_units    = "ios",
 800           .bsp_scale    = true                          },
 801         { .bsp_name     = "block maps msec",
 802           .bsp_units    = "maps",
 803           .bsp_scale    = true,                         },
 804 };
 805
 806 static int brw_stats_seq_show(struct seq_file *seq, void *v)
 807 {
 808         struct brw_stats *brw_stats = seq->private;
 809         int i;
 810
 811         /* this sampling races with updates */
 812         lprocfs_stats_header(seq, ktime_get_real(), brw_stats->bs_init, 25,
 813                              ":", true, "");
 814
 815         for (i = 0; i < ARRAY_SIZE(brw_stats->bs_props); i++) {
 816                 if (!brw_stats->bs_props[i].bsp_name)
 817                         continue;
 818
 819                 display_brw_stats(seq, brw_stats->bs_props[i].bsp_name,
 820                                   brw_stats->bs_props[i].bsp_units,
 821                                   &brw_stats->bs_hist[i * 2],
 822                                   &brw_stats->bs_hist[i * 2 + 1],
 823                                   brw_stats->bs_props[i].bsp_scale);
 824         }
 825
 826         return 0;
 827 }
 828
 829 static ssize_t brw_stats_seq_write(struct file *file,
 830                                    const char __user *buf,
 831                                    size_t len, loff_t *off)
 832 {
 833         struct seq_file *seq = file->private_data;
 834         struct brw_stats *brw_stats = seq->private;
 835         int i;
 836
 837         for (i = 0; i < BRW_RW_STATS_NUM; i++)
 838                 lprocfs_oh_clear_pcpu(&brw_stats->bs_hist[i]);
 839         brw_stats->bs_init = ktime_get_real();
 840
 841         return len;
 842 }
 843
 844 LDEBUGFS_SEQ_FOPS(brw_stats);
 845
 846 int lprocfs_init_brw_stats(struct brw_stats *brw_stats)
 847 {
 848         int i, result;
 849
 850         for (i = 0; i < BRW_RW_STATS_NUM; i++) {
 851                 result = lprocfs_oh_alloc_pcpu(&brw_stats->bs_hist[i]);
 852                 if (result)
 853                         break;
 854         }
 855
 856         return result;
 857 }
 858 EXPORT_SYMBOL(lprocfs_init_brw_stats);
 859
 860 void lprocfs_fini_brw_stats(struct brw_stats *brw_stats)
 861 {
 862         int i;
 863
 864         for (i = 0; i < BRW_RW_STATS_NUM; i++)
 865                 lprocfs_oh_release_pcpu(&brw_stats->bs_hist[i]);
 866 }
 867 EXPORT_SYMBOL(lprocfs_fini_brw_stats);
 868
 869 void ldebugfs_register_osd_stats(struct dentry *parent,
 870                                  struct brw_stats *brw_stats,
 871                                  struct lprocfs_stats *stats)
 872 {
 873         int i;
 874
 875         LASSERT(brw_stats);
 876         brw_stats->bs_init = ktime_get_real();
 877         for (i = 0; i < BRW_RW_STATS_NUM; i++) {
 878                 struct brw_stats_props *props = brw_stats->bs_props;
 879
 880                 if (i % 2) {
 881                         props[i / 2].bsp_name = brw_props[i / 2].bsp_name;
 882                         props[i / 2].bsp_units = brw_props[i / 2].bsp_units;
 883                         props[i / 2].bsp_scale = brw_props[i / 2].bsp_scale;
 884                 }
 885         }
 886
 887         if (!parent)
 888                 return;
 889
 890         debugfs_create_file("brw_stats", 0644, parent, brw_stats,
 891                             &brw_stats_fops);
 892
 893         if (stats)
 894                 debugfs_create_file("stats", 0644, parent, stats,
 895                                     &ldebugfs_stats_seq_fops);
 896 }
 897 EXPORT_SYMBOL(ldebugfs_register_osd_stats);
 898
 899 int lprocfs_hash_seq_show(struct seq_file *m, void *data)
 900 {
 901         struct obd_device *obd = m->private;
 902
 903         if (obd == NULL)
 904                 return 0;
 905
 906         /* header for rhashtable state */
 907         seq_printf(m, "%-*s   cur   min        max theta t-min t-max flags  rehash   count  maxdep distribution\n",
 908                    HASH_NAME_LEN, "name");
 909         ldebugfs_rhash_seq_show("UUID_HASH", &obd->obd_uuid_hash, m);
 910         ldebugfs_rhash_seq_show("NID_HASH", &obd->obd_nid_hash.ht, m);
 911
 912         cfs_hash_debug_header(m);
 913         cfs_hash_debug_str(obd->obd_nid_stats_hash, m);
 914         return 0;
 915 }
 916 EXPORT_SYMBOL(lprocfs_hash_seq_show);
 917
 918 int lprocfs_recovery_status_seq_show(struct seq_file *m, void *data)
 919 {
 920         struct obd_device *obd = m->private;
 921         struct target_distribute_txn_data *tdtd;
 922
 923         LASSERT(obd != NULL);
 924
 925         seq_printf(m, "status: ");
 926         if (atomic_read(&obd->obd_max_recoverable_clients) == 0) {
 927                 seq_printf(m, "INACTIVE\n");
 928                 goto out;
 929         }
 930
 931         /* There is gap between client data read from storage and setting
 932          * obd_recovering so check obd_recovery_end as well to make sure
 933          * recovery is really finished
 934          */
 935         if (obd->obd_recovery_end > 0 && !obd->obd_recovering) {
 936                 seq_printf(m, "COMPLETE\n");
 937                 seq_printf(m, "recovery_start: %lld\n",
 938                            (s64)ktime_get_real_seconds() -
 939                            (ktime_get_seconds() - obd->obd_recovery_start));
 940                 seq_printf(m, "recovery_duration: %lld\n",
 941                            obd->obd_recovery_end ?
 942                            obd->obd_recovery_end - obd->obd_recovery_start :
 943                            ktime_get_seconds() - obd->obd_recovery_start);
 944                 /* Number of clients that have completed recovery */
 945                 seq_printf(m, "completed_clients: %d/%d\n",
 946                            atomic_read(&obd->obd_max_recoverable_clients) -
 947                            obd->obd_stale_clients,
 948                            atomic_read(&obd->obd_max_recoverable_clients));
 949                 seq_printf(m, "replayed_requests: %d\n",
 950                            obd->obd_replayed_requests);
 951                 seq_printf(m, "last_transno: %lld\n",
 952                            obd->obd_next_recovery_transno - 1);
 953                 seq_printf(m, "VBR: %s\n", obd->obd_version_recov ?
 954                            "ENABLED" : "DISABLED");
 955                 seq_printf(m, "IR: %s\n", obd->obd_no_ir ?
 956                            "DISABLED" : "ENABLED");
 957                 goto out;
 958         }
 959
 960         tdtd = obd2obt(obd)->obt_lut->lut_tdtd;
 961         if (tdtd && tdtd->tdtd_show_update_logs_retrievers) {
 962                 char *buf;
 963                 int size = 0;
 964                 int count = 0;
 965
 966                 buf = tdtd->tdtd_show_update_logs_retrievers(
 967                         tdtd->tdtd_show_retrievers_cbdata,
 968                         &size, &count);
 969                 if (count > 0) {
 970                         seq_printf(m, "WAITING\n");
 971                         seq_printf(m, "non-ready MDTs: %s\n",
 972                                    buf ? buf : "unknown (not enough RAM)");
 973                         seq_printf(m, "recovery_start: %lld\n",
 974                                    (s64)ktime_get_real_seconds() -
 975                                    (ktime_get_seconds() -
 976                                     obd->obd_recovery_start));
 977                         seq_printf(m, "time_waited: %lld\n",
 978                                    (s64)(ktime_get_seconds() -
 979                                          obd->obd_recovery_start));
 980                 }
 981
 982                 if (buf != NULL)
 983                         OBD_FREE(buf, size);
 984
 985                 if (likely(count > 0))
 986                         goto out;
 987         }
 988
 989         /* recovery won't start until the clients connect */
 990         if (obd->obd_recovery_start == 0) {
 991                 seq_printf(m, "WAITING_FOR_CLIENTS\n");
 992                 goto out;
 993         }
 994
 995         seq_printf(m, "RECOVERING\n");
 996         seq_printf(m, "recovery_start: %lld\n", (s64)ktime_get_real_seconds() -
 997                    (ktime_get_seconds() - obd->obd_recovery_start));
 998         seq_printf(m, "time_remaining: %lld\n",
 999                    ktime_get_seconds() >=
1000                    obd->obd_recovery_start +
1001                    obd->obd_recovery_timeout ? 0 :
1002                    (s64)(obd->obd_recovery_start +
1003                          obd->obd_recovery_timeout -
1004                          ktime_get_seconds()));
1005         seq_printf(m, "connected_clients: %d/%d\n",
1006                    atomic_read(&obd->obd_connected_clients),
1007                    atomic_read(&obd->obd_max_recoverable_clients));
1008         /* Number of clients that have completed recovery */
1009         seq_printf(m, "req_replay_clients: %d\n",
1010                    atomic_read(&obd->obd_req_replay_clients));
1011         seq_printf(m, "lock_repay_clients: %d\n",
1012                    atomic_read(&obd->obd_lock_replay_clients));
1013         seq_printf(m, "completed_clients: %d\n",
1014                    atomic_read(&obd->obd_connected_clients) -
1015                    atomic_read(&obd->obd_lock_replay_clients));
1016         seq_printf(m, "evicted_clients: %d\n", obd->obd_stale_clients);
1017         seq_printf(m, "replayed_requests: %d\n", obd->obd_replayed_requests);
1018         seq_printf(m, "queued_requests: %d\n",
1019                    obd->obd_requests_queued_for_recovery);
1020         seq_printf(m, "next_transno: %lld\n",
1021                    obd->obd_next_recovery_transno);
1022 out:
1023         return 0;
1024 }
1025 EXPORT_SYMBOL(lprocfs_recovery_status_seq_show);
1026
1027 ssize_t ir_factor_show(struct kobject *kobj, struct attribute *attr,
1028                        char *buf)
1029 {
1030         struct obd_device *obd = container_of(kobj, struct obd_device,
1031                                               obd_kset.kobj);
1032
1033         return scnprintf(buf, PAGE_SIZE, "%d\n", obd->obd_recovery_ir_factor);
1034 }
1035 EXPORT_SYMBOL(ir_factor_show);
1036
1037 ssize_t ir_factor_store(struct kobject *kobj, struct attribute *attr,
1038                         const char *buffer, size_t count)
1039 {
1040         struct obd_device *obd = container_of(kobj, struct obd_device,
1041                                               obd_kset.kobj);
1042         int val;
1043         int rc;
1044
1045         rc = kstrtoint(buffer, 10, &val);
1046         if (rc)
1047                 return rc;
1048
1049         if (val < OBD_IR_FACTOR_MIN || val > OBD_IR_FACTOR_MAX)
1050                 return -EINVAL;
1051
1052         obd->obd_recovery_ir_factor = val;
1053         return count;
1054 }
1055 EXPORT_SYMBOL(ir_factor_store);
1056
1057 int lprocfs_checksum_dump_seq_show(struct seq_file *m, void *data)
1058 {
1059         struct obd_device *obd = m->private;
1060
1061         LASSERT(obd != NULL);
1062         seq_printf(m, "%d\n", obd->obd_checksum_dump);
1063         return 0;
1064 }
1065 EXPORT_SYMBOL(lprocfs_checksum_dump_seq_show);
1066
1067 ssize_t
1068 lprocfs_checksum_dump_seq_write(struct file *file, const char __user *buffer,
1069                             size_t count, loff_t *off)
1070 {
1071         struct seq_file *m = file->private_data;
1072         struct obd_device *obd = m->private;
1073         bool val;
1074         int rc;
1075
1076         LASSERT(obd != NULL);
1077         rc = kstrtobool_from_user(buffer, count, &val);
1078         if (rc)
1079                 return rc;
1080
1081         obd->obd_checksum_dump = val;
1082         return count;
1083 }
1084 EXPORT_SYMBOL(lprocfs_checksum_dump_seq_write);
1085
1086 ssize_t recovery_time_soft_show(struct kobject *kobj, struct attribute *attr,
1087                                 char *buf)
1088 {
1089         struct obd_device *obd = container_of(kobj, struct obd_device,
1090                                               obd_kset.kobj);
1091
1092         return scnprintf(buf, PAGE_SIZE, "%d\n", obd->obd_recovery_timeout);
1093 }
1094 EXPORT_SYMBOL(recovery_time_soft_show);
1095
1096 ssize_t recovery_time_soft_store(struct kobject *kobj,
1097                                  struct attribute *attr,
1098                                  const char *buffer, size_t count)
1099 {
1100         struct obd_device *obd = container_of(kobj, struct obd_device,
1101                                               obd_kset.kobj);
1102         unsigned int val;
1103         int rc;
1104
1105         rc = kstrtouint(buffer, 0, &val);
1106         if (rc)
1107                 return rc;
1108
1109         obd->obd_recovery_timeout = val;
1110         return count;
1111 }
1112 EXPORT_SYMBOL(recovery_time_soft_store);
1113
1114 ssize_t recovery_time_hard_show(struct kobject *kobj, struct attribute *attr,
1115                                 char *buf)
1116 {
1117         struct obd_device *obd = container_of(kobj, struct obd_device,
1118                                               obd_kset.kobj);
1119
1120         return scnprintf(buf, PAGE_SIZE, "%d\n", obd->obd_recovery_time_hard);
1121 }
1122 EXPORT_SYMBOL(recovery_time_hard_show);
1123
1124 ssize_t recovery_time_hard_store(struct kobject *kobj,
1125                                  struct attribute *attr,
1126                                  const char *buffer, size_t count)
1127 {
1128         struct obd_device *obd = container_of(kobj, struct obd_device,
1129                                               obd_kset.kobj);
1130         unsigned int val;
1131         int rc;
1132
1133         rc = kstrtouint(buffer, 0, &val);
1134         if (rc)
1135                 return rc;
1136
1137         obd->obd_recovery_time_hard = val;
1138         return count;
1139 }
1140 EXPORT_SYMBOL(recovery_time_hard_store);
1141
1142 ssize_t instance_show(struct kobject *kobj, struct attribute *attr,
1143                       char *buf)
1144 {
1145         struct obd_device *obd = container_of(kobj, struct obd_device,
1146                                               obd_kset.kobj);
1147         struct obd_device_target *target = obd2obt(obd);
1148
1149         LASSERT(target->obt_magic == OBT_MAGIC);
1150         return scnprintf(buf, PAGE_SIZE, "%u\n", obd2obt(obd)->obt_instance);
1151 }
1152 EXPORT_SYMBOL(instance_show);
1153
1154 #endif /* CONFIG_PROC_FS*/