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 num_exports_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", obd->obd_num_exports);
 154 }
 155 EXPORT_SYMBOL(num_exports_show);
 156
 157 ssize_t grant_check_threshold_show(struct kobject *kobj, struct attribute *attr,
 158                                    char *buf)
 159 {
 160         struct obd_device *obd = container_of(kobj, struct obd_device,
 161                                               obd_kset.kobj);
 162
 163         return scnprintf(buf, PAGE_SIZE, "%d\n",
 164                          obd->obd_grant_check_threshold);
 165 }
 166 EXPORT_SYMBOL(grant_check_threshold_show);
 167
 168 ssize_t grant_check_threshold_store(struct kobject *kobj,
 169                                     struct attribute *attr,
 170                                     const char *buffer, size_t count)
 171 {
 172         struct obd_device *obd = container_of(kobj, struct obd_device,
 173                                               obd_kset.kobj);
 174         int val;
 175         int rc;
 176
 177         rc = kstrtoint(buffer, 10, &val);
 178         if (rc)
 179                 return rc;
 180
 181         if (val < 0)
 182                 return -EINVAL;
 183         obd->obd_grant_check_threshold = val;
 184         return count;
 185 }
 186 EXPORT_SYMBOL(grant_check_threshold_store);
 187
 188 static int obd_export_flags2str(struct obd_export *exp, struct seq_file *m)
 189 {
 190         bool first = true;
 191
 192         flag2str(exp, failed);
 193         flag2str(exp, in_recovery);
 194         flag2str(exp, disconnected);
 195         flag2str(exp, connecting);
 196         flag2str(exp, no_recovery);
 197
 198         return 0;
 199 }
 200
 201 static int
 202 lprocfs_exp_print_export_seq(struct obd_export *exp, void *cb_data)
 203 {
 204         struct seq_file         *m = cb_data;
 205         struct obd_device       *obd;
 206         struct obd_connect_data *ocd;
 207
 208         LASSERT(exp != NULL);
 209         if (exp->exp_nid_stats == NULL)
 210                 goto out;
 211         obd = exp->exp_obd;
 212         ocd = &exp->exp_connect_data;
 213
 214         seq_printf(m, "%s:\n"
 215                    "    name: %s\n"
 216                    "    client: %s\n"
 217                    "    connect_flags: [ ",
 218                    obd_uuid2str(&exp->exp_client_uuid),
 219                    obd->obd_name,
 220                    obd_export_nid2str(exp));
 221         obd_connect_seq_flags2str(m, ocd->ocd_connect_flags,
 222                                   ocd->ocd_connect_flags2, ", ");
 223         seq_printf(m, " ]\n");
 224         obd_connect_data_seqprint(m, ocd);
 225         seq_printf(m, "    export_flags: [ ");
 226         obd_export_flags2str(exp, m);
 227         seq_printf(m, " ]\n");
 228
 229         if (obd->obd_type &&
 230             strcmp(obd->obd_type->typ_name, "obdfilter") == 0) {
 231                 struct filter_export_data *fed = &exp->exp_filter_data;
 232
 233                 seq_printf(m, "    grant:\n");
 234                 seq_printf(m, "       granted: %ld\n",
 235                         fed->fed_ted.ted_grant);
 236                 seq_printf(m, "       dirty: %ld\n",
 237                         fed->fed_ted.ted_dirty);
 238                 seq_printf(m, "       pending: %ld\n",
 239                         fed->fed_ted.ted_pending);
 240         }
 241
 242 out:
 243         return 0;
 244 }
 245
 246 /**
 247  * RPC connections are composed of an import and an export. Using the
 248  * lctl utility we can extract important information about the state.
 249  * The lprocfs_exp_export_seq_show routine displays the state information
 250  * for the export.
 251  *
 252  * \param[in] m         seq file
 253  * \param[in] data      unused
 254  *
 255  * \retval              0 on success
 256  *
 257  * The format of the export state information is like:
 258  * a793e354-49c0-aa11-8c4f-a4f2b1a1a92b:
 259  *     name: MGS
 260  *     client: 10.211.55.10@tcp
 261  *     connect_flags: [ version, barrier, adaptive_timeouts, ... ]
 262  *     connect_data:
 263  *        flags: 0x2000011005002020
 264  *        instance: 0
 265  *        target_version: 2.10.51.0
 266  *        export_flags: [ ... ]
 267  *
 268  */
 269 static int lprocfs_exp_export_seq_show(struct seq_file *m, void *data)
 270 {
 271         struct nid_stat *stats = m->private;
 272
 273         return obd_nid_export_for_each(stats->nid_obd, &stats->nid,
 274                                        lprocfs_exp_print_export_seq, m);
 275 }
 276 LPROC_SEQ_FOPS_RO(lprocfs_exp_export);
 277
 278 static void lprocfs_free_client_stats(struct nid_stat *client_stat)
 279 {
 280         CDEBUG(D_CONFIG, "stat %p - data %p/%p\n", client_stat,
 281                client_stat->nid_proc, client_stat->nid_stats);
 282
 283         LASSERTF(atomic_read(&client_stat->nid_exp_ref_count) == 0,
 284                  "nid %s:count %d\n", libcfs_nidstr(&client_stat->nid),
 285                  atomic_read(&client_stat->nid_exp_ref_count));
 286
 287         if (client_stat->nid_proc)
 288                 lprocfs_remove(&client_stat->nid_proc);
 289
 290         if (client_stat->nid_stats)
 291                 lprocfs_free_stats(&client_stat->nid_stats);
 292
 293         if (client_stat->nid_ldlm_stats)
 294                 lprocfs_free_stats(&client_stat->nid_ldlm_stats);
 295
 296         OBD_FREE_PTR(client_stat);
 297 }
 298
 299 void lprocfs_free_per_client_stats(struct obd_device *obd)
 300 {
 301         struct cfs_hash *hash = obd->obd_nid_stats_hash;
 302         struct nid_stat *stat;
 303         ENTRY;
 304
 305         /* we need extra list - because hash_exit called to early */
 306         /* not need locking because all clients is died */
 307         while (!list_empty(&obd->obd_nid_stats)) {
 308                 stat = list_entry(obd->obd_nid_stats.next,
 309                                   struct nid_stat, nid_list);
 310                 list_del_init(&stat->nid_list);
 311                 cfs_hash_del(hash, &stat->nid, &stat->nid_hash);
 312                 lprocfs_free_client_stats(stat);
 313         }
 314         EXIT;
 315 }
 316 EXPORT_SYMBOL(lprocfs_free_per_client_stats);
 317
 318 static int
 319 lprocfs_exp_print_nodemap_seq(struct obd_export *exp, void *cb_data)
 320 {
 321         struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap;
 322         struct seq_file *m = cb_data;
 323
 324         if (nodemap)
 325                 seq_printf(m, "%s\n", nodemap->nm_name);
 326         return 0;
 327 }
 328
 329 static int
 330 lprocfs_exp_nodemap_seq_show(struct seq_file *m, void *data)
 331 {
 332         struct nid_stat *stats = m->private;
 333
 334         return obd_nid_export_for_each(stats->nid_obd, &stats->nid,
 335                                        lprocfs_exp_print_nodemap_seq, m);
 336 }
 337 LPROC_SEQ_FOPS_RO(lprocfs_exp_nodemap);
 338
 339 static int
 340 lprocfs_exp_print_uuid_seq(struct obd_export *exp, void *cb_data)
 341 {
 342         struct seq_file *m = cb_data;
 343
 344         if (exp->exp_nid_stats)
 345                 seq_printf(m, "%s\n", obd_uuid2str(&exp->exp_client_uuid));
 346         return 0;
 347 }
 348
 349 static int lprocfs_exp_uuid_seq_show(struct seq_file *m, void *data)
 350 {
 351         struct nid_stat *stats = m->private;
 352
 353         return obd_nid_export_for_each(stats->nid_obd, &stats->nid,
 354                                        lprocfs_exp_print_uuid_seq, m);
 355 }
 356 LPROC_SEQ_FOPS_RO(lprocfs_exp_uuid);
 357
 358 #define HASH_NAME_LEN   16
 359
 360 static void ldebugfs_rhash_seq_show(const char *name, struct rhashtable *ht,
 361                                     struct seq_file *m)
 362 {
 363         unsigned int max_size = ht->p.max_size ? ht->p.max_size : UINT_MAX;
 364         struct bucket_table *tbl;
 365         int dist[8] = { 0, };
 366         int maxdep = 0;
 367         int i;
 368
 369         rcu_read_lock();
 370         tbl = rht_dereference(ht->tbl, ht);
 371         for (i = 0; i < tbl->size; i++) {
 372                 struct rhash_head *pos;
 373                 int count = 0;
 374
 375                 rht_for_each(pos, tbl, i)
 376                         count++;
 377
 378                 if (count)
 379                         maxdep = max(maxdep, count);
 380
 381                 dist[min(fls(count), 7)]++;
 382         }
 383
 384         seq_printf(m, "%-*s %5d %5d %10u %d.%03d 0.300 0.750 0x%03x %7d %7d %7d ",
 385                    HASH_NAME_LEN, name, tbl->size, ht->p.min_size, max_size,
 386                    atomic_read(&ht->nelems) / tbl->size,
 387                    atomic_read(&ht->nelems) * 1000 / tbl->size,
 388                    ht->p.automatic_shrinking, 0,
 389                    atomic_read(&ht->nelems), maxdep);
 390         rcu_read_unlock();
 391
 392         for (i = 0; i < 8; i++)
 393                 seq_printf(m, "%d%c",  dist[i], (i == 7) ? '\n' : '/');
 394 }
 395
 396 static int
 397 lprocfs_exp_print_hash_seq(struct obd_export *exp, void *cb_data)
 398
 399 {
 400         struct obd_device *obd = exp->exp_obd;
 401         struct seq_file *m = cb_data;
 402
 403         if (exp->exp_lock_hash != NULL) {
 404                 seq_printf(m, "%-*s   cur   min        max theta t-min t-max flags rehash   count distribution\n",
 405                            HASH_NAME_LEN, "name");
 406                 ldebugfs_rhash_seq_show("NID_HASH", &obd->obd_nid_hash.ht, m);
 407         }
 408         return 0;
 409 }
 410
 411 static int lprocfs_exp_hash_seq_show(struct seq_file *m, void *data)
 412 {
 413         struct nid_stat *stats = m->private;
 414
 415         return obd_nid_export_for_each(stats->nid_obd, &stats->nid,
 416                                        lprocfs_exp_print_hash_seq, m);
 417 }
 418 LPROC_SEQ_FOPS_RO(lprocfs_exp_hash);
 419
 420 int lprocfs_exp_print_replydata_seq(struct obd_export *exp, void *cb_data)
 421
 422 {
 423         struct seq_file *m = cb_data;
 424         struct tg_export_data *ted = &exp->exp_target_data;
 425
 426         seq_printf(m, "reply_cnt: %d\n"
 427                    "reply_max: %d\n"
 428                    "reply_released_by_xid: %d\n"
 429                    "reply_released_by_tag: %d\n\n",
 430                    ted->ted_reply_cnt,
 431                    ted->ted_reply_max,
 432                    ted->ted_release_xid,
 433                    ted->ted_release_tag);
 434         return 0;
 435 }
 436
 437 int lprocfs_exp_replydata_seq_show(struct seq_file *m, void *data)
 438 {
 439         struct nid_stat *stats = m->private;
 440
 441         return obd_nid_export_for_each(stats->nid_obd, &stats->nid,
 442                                        lprocfs_exp_print_replydata_seq, m);
 443 }
 444 LPROC_SEQ_FOPS_RO(lprocfs_exp_replydata);
 445
 446 int lprocfs_exp_print_fmd_count_seq(struct obd_export *exp, void *cb_data)
 447 {
 448         struct seq_file *m = cb_data;
 449         struct tg_export_data *ted = &exp->exp_target_data;
 450
 451         seq_printf(m, "%d\n", ted->ted_fmd_count);
 452
 453         return 0;
 454 }
 455
 456 int lprocfs_exp_fmd_count_seq_show(struct seq_file *m, void *data)
 457 {
 458         struct nid_stat *stats = m->private;
 459
 460         return obd_nid_export_for_each(stats->nid_obd, &stats->nid,
 461                                        lprocfs_exp_print_fmd_count_seq, m);
 462 }
 463 LPROC_SEQ_FOPS_RO(lprocfs_exp_fmd_count);
 464
 465 int lprocfs_nid_stats_clear_seq_show(struct seq_file *m, void *data)
 466 {
 467         seq_puts(m, "Write into this file to clear all nid stats and stale nid entries\n");
 468         return 0;
 469 }
 470 EXPORT_SYMBOL(lprocfs_nid_stats_clear_seq_show);
 471
 472 static int lprocfs_nid_stats_clear_write_cb(void *obj, void *data)
 473 {
 474         struct nid_stat *stat = obj;
 475         ENTRY;
 476
 477         CDEBUG(D_INFO, "refcnt %d\n", atomic_read(&stat->nid_exp_ref_count));
 478         if (atomic_read(&stat->nid_exp_ref_count) == 1) {
 479                 /* object has only hash references. */
 480                 spin_lock(&stat->nid_obd->obd_nid_lock);
 481                 list_move(&stat->nid_list, data);
 482                 spin_unlock(&stat->nid_obd->obd_nid_lock);
 483                 RETURN(1);
 484         }
 485         /* we has reference to object - only clear data*/
 486         if (stat->nid_stats)
 487                 lprocfs_clear_stats(stat->nid_stats);
 488
 489         RETURN(0);
 490 }
 491
 492 ssize_t
 493 lprocfs_nid_stats_clear_seq_write(struct file *file, const char __user *buffer,
 494                                         size_t count, loff_t *off)
 495 {
 496         struct seq_file *m = file->private_data;
 497         struct obd_device *obd = m->private;
 498         struct nid_stat *client_stat;
 499         LIST_HEAD(free_list);
 500
 501         cfs_hash_cond_del(obd->obd_nid_stats_hash,
 502                           lprocfs_nid_stats_clear_write_cb, &free_list);
 503
 504         while (!list_empty(&free_list)) {
 505                 client_stat = list_entry(free_list.next, struct nid_stat,
 506                                          nid_list);
 507                 list_del_init(&client_stat->nid_list);
 508                 lprocfs_free_client_stats(client_stat);
 509         }
 510         return count;
 511 }
 512 EXPORT_SYMBOL(lprocfs_nid_stats_clear_seq_write);
 513
 514 int lprocfs_exp_setup(struct obd_export *exp, lnet_nid_t *nid)
 515 {
 516         struct nid_stat *new_stat, *old_stat;
 517         struct obd_device *obd = NULL;
 518         struct proc_dir_entry *entry;
 519         char nidstr[LNET_NIDSTR_SIZE];
 520         int rc = 0;
 521         ENTRY;
 522
 523         if (!exp || !exp->exp_obd || !exp->exp_obd->obd_proc_exports_entry ||
 524             !exp->exp_obd->obd_nid_stats_hash)
 525                 RETURN(-EINVAL);
 526
 527         /* not test against zero because eric say:
 528          * You may only test nid against another nid, or LNET_NID_ANY.
 529          * Anything else is nonsense.*/
 530         if (nid == NULL || *nid == LNET_NID_ANY)
 531                 RETURN(-EALREADY);
 532
 533         libcfs_nid2str_r(*nid, nidstr, sizeof(nidstr));
 534
 535         spin_lock(&exp->exp_lock);
 536         if (exp->exp_nid_stats != NULL) {
 537                 spin_unlock(&exp->exp_lock);
 538                 RETURN(-EALREADY);
 539         }
 540         spin_unlock(&exp->exp_lock);
 541
 542         obd = exp->exp_obd;
 543
 544         CDEBUG(D_CONFIG, "using hash %p\n", obd->obd_nid_stats_hash);
 545
 546         OBD_ALLOC_PTR(new_stat);
 547         if (new_stat == NULL)
 548                 RETURN(-ENOMEM);
 549
 550         lnet_nid4_to_nid(*nid, &new_stat->nid);
 551         new_stat->nid_obd = exp->exp_obd;
 552         /* we need set default refcount to 1 to balance obd_disconnect */
 553         atomic_set(&new_stat->nid_exp_ref_count, 1);
 554
 555         old_stat = cfs_hash_findadd_unique(obd->obd_nid_stats_hash,
 556                                            &new_stat->nid,
 557                                            &new_stat->nid_hash);
 558         CDEBUG(D_INFO, "Found stats %p for nid %s - ref %d\n",
 559                old_stat, nidstr, atomic_read(&old_stat->nid_exp_ref_count));
 560
 561         /* Return -EALREADY here so that we know that the /proc
 562          * entry already has been created */
 563         if (old_stat != new_stat) {
 564                 spin_lock(&exp->exp_lock);
 565                 if (exp->exp_nid_stats) {
 566                         LASSERT(exp->exp_nid_stats == old_stat);
 567                         nidstat_putref(exp->exp_nid_stats);
 568                 }
 569                 exp->exp_nid_stats = old_stat;
 570                 spin_unlock(&exp->exp_lock);
 571                 GOTO(destroy_new, rc = -EALREADY);
 572         }
 573         /* not found - create */
 574         new_stat->nid_proc = lprocfs_register(nidstr,
 575                                               obd->obd_proc_exports_entry,
 576                                               NULL, NULL);
 577
 578         if (IS_ERR(new_stat->nid_proc)) {
 579                 rc = PTR_ERR(new_stat->nid_proc);
 580                 new_stat->nid_proc = NULL;
 581                 CERROR("%s: cannot create proc entry for export %s: rc = %d\n",
 582                        obd->obd_name, nidstr, rc);
 583                 GOTO(destroy_new_ns, rc);
 584         }
 585
 586         entry = lprocfs_add_simple(new_stat->nid_proc, "nodemap", new_stat,
 587                                    &lprocfs_exp_nodemap_fops);
 588         if (IS_ERR(entry)) {
 589                 rc = PTR_ERR(entry);
 590                 CWARN("%s: error adding the nodemap file: rc = %d\n",
 591                       obd->obd_name, rc);
 592                 GOTO(destroy_new_ns, rc);
 593         }
 594
 595         entry = lprocfs_add_simple(new_stat->nid_proc, "uuid", new_stat,
 596                                    &lprocfs_exp_uuid_fops);
 597         if (IS_ERR(entry)) {
 598                 rc = PTR_ERR(entry);
 599                 CWARN("%s: error adding the NID stats file: rc = %d\n",
 600                       obd->obd_name, rc);
 601                 GOTO(destroy_new_ns, rc);
 602         }
 603
 604         entry = lprocfs_add_simple(new_stat->nid_proc, "hash", new_stat,
 605                                    &lprocfs_exp_hash_fops);
 606         if (IS_ERR(entry)) {
 607                 rc = PTR_ERR(entry);
 608                 CWARN("%s: error adding the hash file: rc = %d\n",
 609                       obd->obd_name, rc);
 610                 GOTO(destroy_new_ns, rc);
 611         }
 612
 613         entry = lprocfs_add_simple(new_stat->nid_proc, "export",
 614                                    new_stat, &lprocfs_exp_export_fops);
 615         if (IS_ERR(entry)) {
 616                 rc = PTR_ERR(entry);
 617                 CWARN("%s: error adding the export file: rc = %d\n",
 618                       obd->obd_name, rc);
 619                 GOTO(destroy_new_ns, rc);
 620         }
 621
 622         entry = lprocfs_add_simple(new_stat->nid_proc, "reply_data", new_stat,
 623                                    &lprocfs_exp_replydata_fops);
 624         if (IS_ERR(entry)) {
 625                 rc = PTR_ERR(entry);
 626                 CWARN("%s: error adding the reply_data file: rc = %d\n",
 627                       obd->obd_name, rc);
 628                 GOTO(destroy_new_ns, rc);
 629         }
 630
 631         entry = lprocfs_add_simple(new_stat->nid_proc, "fmd_count", new_stat,
 632                                    &lprocfs_exp_fmd_count_fops);
 633         if (IS_ERR(entry)) {
 634                 rc = PTR_ERR(entry);
 635                 CWARN("%s: error adding the fmd_count file: rc = %d\n",
 636                       obd->obd_name, rc);
 637                 GOTO(destroy_new_ns, rc);
 638         }
 639
 640         spin_lock(&exp->exp_lock);
 641         exp->exp_nid_stats = new_stat;
 642         spin_unlock(&exp->exp_lock);
 643
 644         /* protect competitive add to list, not need locking on destroy */
 645         spin_lock(&obd->obd_nid_lock);
 646         list_add(&new_stat->nid_list, &obd->obd_nid_stats);
 647         spin_unlock(&obd->obd_nid_lock);
 648
 649         RETURN(0);
 650
 651 destroy_new_ns:
 652         if (new_stat->nid_proc != NULL)
 653                 lprocfs_remove(&new_stat->nid_proc);
 654         cfs_hash_del(obd->obd_nid_stats_hash, &new_stat->nid,
 655                      &new_stat->nid_hash);
 656
 657 destroy_new:
 658         nidstat_putref(new_stat);
 659         OBD_FREE_PTR(new_stat);
 660         RETURN(rc);
 661 }
 662 EXPORT_SYMBOL(lprocfs_exp_setup);
 663
 664 int lprocfs_exp_cleanup(struct obd_export *exp)
 665 {
 666         struct nid_stat *stat = exp->exp_nid_stats;
 667
 668         if (!stat || !exp->exp_obd)
 669                 RETURN(0);
 670
 671         nidstat_putref(exp->exp_nid_stats);
 672         exp->exp_nid_stats = NULL;
 673
 674         return 0;
 675 }
 676
 677 int lprocfs_alloc_obd_stats(struct obd_device *obd, unsigned int num_stats)
 678 {
 679         struct lprocfs_stats *stats;
 680         int rc;
 681
 682         LASSERT(obd->obd_stats == NULL);
 683         LASSERT(obd->obd_proc_entry != NULL);
 684
 685         stats = lprocfs_alloc_stats(num_stats, 0);
 686         if (stats == NULL)
 687                 return -ENOMEM;
 688
 689         rc = lprocfs_register_stats(obd->obd_proc_entry, "stats", stats);
 690         if (rc < 0)
 691                 lprocfs_free_stats(&stats);
 692         else
 693                 obd->obd_stats = stats;
 694
 695         return rc;
 696 }
 697 EXPORT_SYMBOL(lprocfs_alloc_obd_stats);
 698
 699 void lprocfs_free_obd_stats(struct obd_device *obd)
 700 {
 701         if (obd->obd_stats)
 702                 lprocfs_free_stats(&obd->obd_stats);
 703 }
 704 EXPORT_SYMBOL(lprocfs_free_obd_stats);
 705
 706 static void display_brw_stats(struct seq_file *seq, const char *name,
 707                               const char *units, struct obd_hist_pcpu *read,
 708                               struct obd_hist_pcpu *write, bool scale)
 709 {
 710         unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
 711         unsigned int i;
 712
 713         seq_printf(seq, "\n%26s read      |     write\n", " ");
 714         seq_printf(seq, "%-22s %-5s %% cum %% |  %-11s %% cum %%\n",
 715                    name, units, units);
 716
 717         read_tot = lprocfs_oh_sum_pcpu(read);
 718         write_tot = lprocfs_oh_sum_pcpu(write);
 719
 720         if (!read_tot && !write_tot)
 721                 return;
 722
 723         for (i = 0; i < OBD_HIST_MAX; i++) {
 724                 r = lprocfs_oh_counter_pcpu(read, i);
 725                 w = lprocfs_oh_counter_pcpu(write, i);
 726                 read_cum += r;
 727                 write_cum += w;
 728                 if (read_cum == 0 && write_cum == 0)
 729                         continue;
 730
 731                 if (!scale)
 732                         seq_printf(seq, "%u", i);
 733                 else if (i < 10)
 734                         seq_printf(seq, "%lu", BIT(i));
 735                 else if (i < 20)
 736                         seq_printf(seq, "%luK", BIT(i - 10));
 737                 else
 738                         seq_printf(seq, "%luM", BIT(i - 20));
 739
 740                 seq_printf(seq, ":\t\t%10lu %3u %3u   | %4lu %3u %3u\n",
 741                            r, pct(r, read_tot), pct(read_cum, read_tot),
 742                            w, pct(w, write_tot), pct(write_cum, write_tot));
 743
 744                 if (read_cum == read_tot && write_cum == write_tot)
 745                         break;
 746         }
 747 }
 748
 749 static const struct brw_stats_props brw_props[] = {
 750         { .bsp_name     = "pages per bulk r/w",
 751           .bsp_units    = "rpcs",
 752           .bsp_scale    = true                          },
 753         { .bsp_name     = "discontiguous pages",
 754           .bsp_units    = "rpcs",
 755           .bsp_scale    = false                         },
 756         { .bsp_name     = "discontiguous blocks",
 757           .bsp_units    = "rpcs",
 758           .bsp_scale    = false                         },
 759         { .bsp_name     = "disk fragmented I/Os",
 760           .bsp_units    = "ios",
 761           .bsp_scale    = false                         },
 762         { .bsp_name     = "disk I/Os in flight",
 763           .bsp_units    = "ios",
 764           .bsp_scale    = false                         },
 765         { .bsp_name     = "I/O time (1/1000s)",
 766           .bsp_units    = "ios",
 767           .bsp_scale    = true                          },
 768         { .bsp_name     = "disk I/O size",
 769           .bsp_units    = "ios",
 770           .bsp_scale    = true                          },
 771 };
 772
 773 static int brw_stats_seq_show(struct seq_file *seq, void *v)
 774 {
 775         struct brw_stats *brw_stats = seq->private;
 776         int i;
 777
 778         /* this sampling races with updates */
 779         lprocfs_stats_header(seq, ktime_get(), brw_stats->bs_init, 25, ":", 1);
 780
 781         for (i = 0; i < ARRAY_SIZE(brw_stats->bs_props); i++) {
 782                 if (!brw_stats->bs_props[i].bsp_name)
 783                         continue;
 784
 785                 display_brw_stats(seq, brw_stats->bs_props[i].bsp_name,
 786                                   brw_stats->bs_props[i].bsp_units,
 787                                   &brw_stats->bs_hist[i * 2],
 788                                   &brw_stats->bs_hist[i * 2 + 1],
 789                                   brw_stats->bs_props[i].bsp_scale);
 790         }
 791
 792         return 0;
 793 }
 794
 795 static ssize_t brw_stats_seq_write(struct file *file,
 796                                    const char __user *buf,
 797                                    size_t len, loff_t *off)
 798 {
 799         struct seq_file *seq = file->private_data;
 800         struct brw_stats *brw_stats = seq->private;
 801         int i;
 802
 803         for (i = 0; i < BRW_RW_STATS_NUM; i++)
 804                 lprocfs_oh_clear_pcpu(&brw_stats->bs_hist[i]);
 805
 806         return len;
 807 }
 808
 809 LDEBUGFS_SEQ_FOPS(brw_stats);
 810
 811 int lprocfs_init_brw_stats(struct brw_stats *brw_stats)
 812 {
 813         int i, result;
 814
 815         for (i = 0; i < BRW_RW_STATS_NUM; i++) {
 816                 result = lprocfs_oh_alloc_pcpu(&brw_stats->bs_hist[i]);
 817                 if (result)
 818                         break;
 819         }
 820
 821         return result;
 822 }
 823 EXPORT_SYMBOL(lprocfs_init_brw_stats);
 824
 825 void lprocfs_fini_brw_stats(struct brw_stats *brw_stats)
 826 {
 827         int i;
 828
 829         for (i = 0; i < BRW_RW_STATS_NUM; i++)
 830                 lprocfs_oh_release_pcpu(&brw_stats->bs_hist[i]);
 831 }
 832 EXPORT_SYMBOL(lprocfs_fini_brw_stats);
 833
 834 void ldebugfs_register_osd_stats(struct dentry *parent,
 835                                  struct brw_stats *brw_stats,
 836                                  struct lprocfs_stats *stats)
 837 {
 838         int i;
 839
 840         LASSERT(brw_stats);
 841         brw_stats->bs_init = ktime_get();
 842         for (i = 0; i < BRW_RW_STATS_NUM; i++) {
 843                 struct brw_stats_props *props = brw_stats->bs_props;
 844
 845                 if (i % 2) {
 846                         props[i / 2].bsp_name = brw_props[i / 2].bsp_name;
 847                         props[i / 2].bsp_units = brw_props[i / 2].bsp_units;
 848                         props[i / 2].bsp_scale = brw_props[i / 2].bsp_scale;
 849                 }
 850         }
 851
 852         if (!parent)
 853                 return;
 854
 855         debugfs_create_file("brw_stats", 0644, parent, brw_stats,
 856                             &brw_stats_fops);
 857
 858         if (stats)
 859                 debugfs_create_file("stats", 0644, parent, stats,
 860                                     &ldebugfs_stats_seq_fops);
 861 }
 862 EXPORT_SYMBOL(ldebugfs_register_osd_stats);
 863
 864 int lprocfs_hash_seq_show(struct seq_file *m, void *data)
 865 {
 866         struct obd_device *obd = m->private;
 867
 868         if (obd == NULL)
 869                 return 0;
 870
 871         /* header for rhashtable state */
 872         seq_printf(m, "%-*s   cur   min        max theta t-min t-max flags  rehash   count  maxdep distribution\n",
 873                    HASH_NAME_LEN, "name");
 874         ldebugfs_rhash_seq_show("UUID_HASH", &obd->obd_uuid_hash, m);
 875         ldebugfs_rhash_seq_show("NID_HASH", &obd->obd_nid_hash.ht, m);
 876
 877         cfs_hash_debug_header(m);
 878         cfs_hash_debug_str(obd->obd_nid_stats_hash, m);
 879         return 0;
 880 }
 881 EXPORT_SYMBOL(lprocfs_hash_seq_show);
 882
 883 int lprocfs_recovery_status_seq_show(struct seq_file *m, void *data)
 884 {
 885         struct obd_device *obd = m->private;
 886         struct target_distribute_txn_data *tdtd;
 887
 888         LASSERT(obd != NULL);
 889
 890         seq_printf(m, "status: ");
 891         if (atomic_read(&obd->obd_max_recoverable_clients) == 0) {
 892                 seq_printf(m, "INACTIVE\n");
 893                 goto out;
 894         }
 895
 896         /* There is gap between client data read from storage and setting
 897          * obd_recovering so check obd_recovery_end as well to make sure
 898          * recovery is really finished
 899          */
 900         if (obd->obd_recovery_end > 0 && !obd->obd_recovering) {
 901                 seq_printf(m, "COMPLETE\n");
 902                 seq_printf(m, "recovery_start: %lld\n",
 903                            (s64)ktime_get_real_seconds() -
 904                            (ktime_get_seconds() - obd->obd_recovery_start));
 905                 seq_printf(m, "recovery_duration: %lld\n",
 906                            obd->obd_recovery_end ?
 907                            obd->obd_recovery_end - obd->obd_recovery_start :
 908                            ktime_get_seconds() - obd->obd_recovery_start);
 909                 /* Number of clients that have completed recovery */
 910                 seq_printf(m, "completed_clients: %d/%d\n",
 911                            atomic_read(&obd->obd_max_recoverable_clients) -
 912                            obd->obd_stale_clients,
 913                            atomic_read(&obd->obd_max_recoverable_clients));
 914                 seq_printf(m, "replayed_requests: %d\n",
 915                            obd->obd_replayed_requests);
 916                 seq_printf(m, "last_transno: %lld\n",
 917                            obd->obd_next_recovery_transno - 1);
 918                 seq_printf(m, "VBR: %s\n", obd->obd_version_recov ?
 919                            "ENABLED" : "DISABLED");
 920                 seq_printf(m, "IR: %s\n", obd->obd_no_ir ?
 921                            "DISABLED" : "ENABLED");
 922                 goto out;
 923         }
 924
 925         tdtd = obd->u.obt.obt_lut->lut_tdtd;
 926         if (tdtd && tdtd->tdtd_show_update_logs_retrievers) {
 927                 char *buf;
 928                 int size = 0;
 929                 int count = 0;
 930
 931                 buf = tdtd->tdtd_show_update_logs_retrievers(
 932                         tdtd->tdtd_show_retrievers_cbdata,
 933                         &size, &count);
 934                 if (count > 0) {
 935                         seq_printf(m, "WAITING\n");
 936                         seq_printf(m, "non-ready MDTs: %s\n",
 937                                    buf ? buf : "unknown (not enough RAM)");
 938                         seq_printf(m, "recovery_start: %lld\n",
 939                                    (s64)ktime_get_real_seconds() -
 940                                    (ktime_get_seconds() -
 941                                     obd->obd_recovery_start));
 942                         seq_printf(m, "time_waited: %lld\n",
 943                                    (s64)(ktime_get_seconds() -
 944                                          obd->obd_recovery_start));
 945                 }
 946
 947                 if (buf != NULL)
 948                         OBD_FREE(buf, size);
 949
 950                 if (likely(count > 0))
 951                         goto out;
 952         }
 953
 954         /* recovery won't start until the clients connect */
 955         if (obd->obd_recovery_start == 0) {
 956                 seq_printf(m, "WAITING_FOR_CLIENTS\n");
 957                 goto out;
 958         }
 959
 960         seq_printf(m, "RECOVERING\n");
 961         seq_printf(m, "recovery_start: %lld\n", (s64)ktime_get_real_seconds() -
 962                    (ktime_get_seconds() - obd->obd_recovery_start));
 963         seq_printf(m, "time_remaining: %lld\n",
 964                    ktime_get_seconds() >=
 965                    obd->obd_recovery_start +
 966                    obd->obd_recovery_timeout ? 0 :
 967                    (s64)(obd->obd_recovery_start +
 968                          obd->obd_recovery_timeout -
 969                          ktime_get_seconds()));
 970         seq_printf(m, "connected_clients: %d/%d\n",
 971                    atomic_read(&obd->obd_connected_clients),
 972                    atomic_read(&obd->obd_max_recoverable_clients));
 973         /* Number of clients that have completed recovery */
 974         seq_printf(m, "req_replay_clients: %d\n",
 975                    atomic_read(&obd->obd_req_replay_clients));
 976         seq_printf(m, "lock_repay_clients: %d\n",
 977                    atomic_read(&obd->obd_lock_replay_clients));
 978         seq_printf(m, "completed_clients: %d\n",
 979                    atomic_read(&obd->obd_connected_clients) -
 980                    atomic_read(&obd->obd_lock_replay_clients));
 981         seq_printf(m, "evicted_clients: %d\n", obd->obd_stale_clients);
 982         seq_printf(m, "replayed_requests: %d\n", obd->obd_replayed_requests);
 983         seq_printf(m, "queued_requests: %d\n",
 984                    obd->obd_requests_queued_for_recovery);
 985         seq_printf(m, "next_transno: %lld\n",
 986                    obd->obd_next_recovery_transno);
 987 out:
 988         return 0;
 989 }
 990 EXPORT_SYMBOL(lprocfs_recovery_status_seq_show);
 991
 992 ssize_t ir_factor_show(struct kobject *kobj, struct attribute *attr,
 993                        char *buf)
 994 {
 995         struct obd_device *obd = container_of(kobj, struct obd_device,
 996                                               obd_kset.kobj);
 997
 998         return scnprintf(buf, PAGE_SIZE, "%d\n", obd->obd_recovery_ir_factor);
 999 }
1000 EXPORT_SYMBOL(ir_factor_show);
1001
1002 ssize_t ir_factor_store(struct kobject *kobj, struct attribute *attr,
1003                         const char *buffer, size_t count)
1004 {
1005         struct obd_device *obd = container_of(kobj, struct obd_device,
1006                                               obd_kset.kobj);
1007         int val;
1008         int rc;
1009
1010         rc = kstrtoint(buffer, 10, &val);
1011         if (rc)
1012                 return rc;
1013
1014         if (val < OBD_IR_FACTOR_MIN || val > OBD_IR_FACTOR_MAX)
1015                 return -EINVAL;
1016
1017         obd->obd_recovery_ir_factor = val;
1018         return count;
1019 }
1020 EXPORT_SYMBOL(ir_factor_store);
1021
1022 int lprocfs_checksum_dump_seq_show(struct seq_file *m, void *data)
1023 {
1024         struct obd_device *obd = m->private;
1025
1026         LASSERT(obd != NULL);
1027         seq_printf(m, "%d\n", obd->obd_checksum_dump);
1028         return 0;
1029 }
1030 EXPORT_SYMBOL(lprocfs_checksum_dump_seq_show);
1031
1032 ssize_t
1033 lprocfs_checksum_dump_seq_write(struct file *file, const char __user *buffer,
1034                             size_t count, loff_t *off)
1035 {
1036         struct seq_file *m = file->private_data;
1037         struct obd_device *obd = m->private;
1038         bool val;
1039         int rc;
1040
1041         LASSERT(obd != NULL);
1042         rc = kstrtobool_from_user(buffer, count, &val);
1043         if (rc)
1044                 return rc;
1045
1046         obd->obd_checksum_dump = val;
1047         return count;
1048 }
1049 EXPORT_SYMBOL(lprocfs_checksum_dump_seq_write);
1050
1051 ssize_t recovery_time_soft_show(struct kobject *kobj, struct attribute *attr,
1052                                 char *buf)
1053 {
1054         struct obd_device *obd = container_of(kobj, struct obd_device,
1055                                               obd_kset.kobj);
1056
1057         return scnprintf(buf, PAGE_SIZE, "%d\n", obd->obd_recovery_timeout);
1058 }
1059 EXPORT_SYMBOL(recovery_time_soft_show);
1060
1061 ssize_t recovery_time_soft_store(struct kobject *kobj,
1062                                  struct attribute *attr,
1063                                  const char *buffer, size_t count)
1064 {
1065         struct obd_device *obd = container_of(kobj, struct obd_device,
1066                                               obd_kset.kobj);
1067         unsigned int val;
1068         int rc;
1069
1070         rc = kstrtouint(buffer, 0, &val);
1071         if (rc)
1072                 return rc;
1073
1074         obd->obd_recovery_timeout = val;
1075         return count;
1076 }
1077 EXPORT_SYMBOL(recovery_time_soft_store);
1078
1079 ssize_t recovery_time_hard_show(struct kobject *kobj, struct attribute *attr,
1080                                 char *buf)
1081 {
1082         struct obd_device *obd = container_of(kobj, struct obd_device,
1083                                               obd_kset.kobj);
1084
1085         return scnprintf(buf, PAGE_SIZE, "%d\n", obd->obd_recovery_time_hard);
1086 }
1087 EXPORT_SYMBOL(recovery_time_hard_show);
1088
1089 ssize_t recovery_time_hard_store(struct kobject *kobj,
1090                                  struct attribute *attr,
1091                                  const char *buffer, size_t count)
1092 {
1093         struct obd_device *obd = container_of(kobj, struct obd_device,
1094                                               obd_kset.kobj);
1095         unsigned int val;
1096         int rc;
1097
1098         rc = kstrtouint(buffer, 0, &val);
1099         if (rc)
1100                 return rc;
1101
1102         obd->obd_recovery_time_hard = val;
1103         return count;
1104 }
1105 EXPORT_SYMBOL(recovery_time_hard_store);
1106
1107 ssize_t instance_show(struct kobject *kobj, struct attribute *attr,
1108                       char *buf)
1109 {
1110         struct obd_device *obd = container_of(kobj, struct obd_device,
1111                                               obd_kset.kobj);
1112         struct obd_device_target *target = &obd->u.obt;
1113
1114         LASSERT(target->obt_magic == OBT_MAGIC);
1115         return scnprintf(buf, PAGE_SIZE, "%u\n", obd->u.obt.obt_instance);
1116 }
1117 EXPORT_SYMBOL(instance_show);
1118
1119 #endif /* CONFIG_PROC_FS*/