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