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
 120 int lprocfs_evict_client_open(struct inode *inode, struct file *f)
 121 {
 122         struct obd_device *obd = pde_data(file_inode(f));
 123
 124         atomic_inc(&obd->obd_evict_inprogress);
 125         return 0;
 126 }
 127
 128 int lprocfs_evict_client_release(struct inode *inode, struct file *f)
 129 {
 130         struct obd_device *obd = pde_data(file_inode(f));
 131
 132         atomic_dec(&obd->obd_evict_inprogress);
 133         wake_up(&obd->obd_evict_inprogress_waitq);
 134
 135         return 0;
 136 }
 137
 138 #define BUFLEN (UUID_MAX + 5)
 139
 140 ssize_t
 141 lprocfs_evict_client_seq_write(struct file *file, const char __user *buffer,
 142                                size_t count, loff_t *off)
 143 {
 144         struct seq_file *m = file->private_data;
 145         struct obd_device *obd = m->private;
 146         char *tmpbuf, *kbuf;
 147
 148         OBD_ALLOC(kbuf, BUFLEN);
 149         if (kbuf == NULL)
 150                 return -ENOMEM;
 151
 152         /*
 153          * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
 154          * bytes into kbuf, to ensure that the string is NUL-terminated.
 155          * UUID_MAX should include a trailing NUL already.
 156          */
 157         if (copy_from_user(kbuf, buffer,
 158                            min_t(unsigned long, BUFLEN - 1, count))) {
 159                 count = -EFAULT;
 160                 goto out;
 161         }
 162         tmpbuf = skip_spaces(kbuf);
 163         tmpbuf = strsep(&tmpbuf, " \t\n\f\v\r");
 164         class_incref(obd, __func__, current);
 165
 166         if (strncmp(tmpbuf, "nid:", 4) == 0)
 167                 obd_export_evict_by_nid(obd, tmpbuf + 4);
 168         else if (strncmp(tmpbuf, "uuid:", 5) == 0)
 169                 obd_export_evict_by_uuid(obd, tmpbuf + 5);
 170         else
 171                 obd_export_evict_by_uuid(obd, tmpbuf);
 172
 173         class_decref(obd, __func__, current);
 174
 175 out:
 176         OBD_FREE(kbuf, BUFLEN);
 177         return count;
 178 }
 179 EXPORT_SYMBOL(lprocfs_evict_client_seq_write);
 180
 181 #undef BUFLEN
 182
 183 ssize_t eviction_count_show(struct kobject *kobj, struct attribute *attr,
 184                          char *buf)
 185 {
 186         struct obd_device *obd = container_of(kobj, struct obd_device,
 187                                               obd_kset.kobj);
 188
 189         return scnprintf(buf, PAGE_SIZE, "%u\n",
 190                          atomic_read(&obd->obd_eviction_count));
 191 }
 192 EXPORT_SYMBOL(eviction_count_show);
 193
 194 ssize_t num_exports_show(struct kobject *kobj, struct attribute *attr,
 195                          char *buf)
 196 {
 197         struct obd_device *obd = container_of(kobj, struct obd_device,
 198                                               obd_kset.kobj);
 199
 200         return scnprintf(buf, PAGE_SIZE, "%u\n", obd->obd_num_exports);
 201 }
 202 EXPORT_SYMBOL(num_exports_show);
 203
 204 ssize_t grant_check_threshold_show(struct kobject *kobj, struct attribute *attr,
 205                                    char *buf)
 206 {
 207         struct obd_device *obd = container_of(kobj, struct obd_device,
 208                                               obd_kset.kobj);
 209
 210         return scnprintf(buf, PAGE_SIZE, "%d\n",
 211                          obd->obd_grant_check_threshold);
 212 }
 213 EXPORT_SYMBOL(grant_check_threshold_show);
 214
 215 ssize_t grant_check_threshold_store(struct kobject *kobj,
 216                                     struct attribute *attr,
 217                                     const char *buffer, size_t count)
 218 {
 219         struct obd_device *obd = container_of(kobj, struct obd_device,
 220                                               obd_kset.kobj);
 221         int val;
 222         int rc;
 223
 224         rc = kstrtoint(buffer, 10, &val);
 225         if (rc)
 226                 return rc;
 227
 228         if (val < 0)
 229                 return -EINVAL;
 230         obd->obd_grant_check_threshold = val;
 231         return count;
 232 }
 233 EXPORT_SYMBOL(grant_check_threshold_store);
 234
 235 static int obd_export_flags2str(struct obd_export *exp, struct seq_file *m)
 236 {
 237         bool first = true;
 238
 239         flag2str(exp, failed);
 240         flag2str(exp, in_recovery);
 241         flag2str(exp, disconnected);
 242         flag2str(exp, connecting);
 243         flag2str(exp, no_recovery);
 244
 245         return 0;
 246 }
 247
 248 static int
 249 lprocfs_exp_print_export_seq(struct obd_export *exp, void *cb_data)
 250 {
 251         struct seq_file         *m = cb_data;
 252         struct obd_device       *obd;
 253         struct obd_connect_data *ocd;
 254
 255         LASSERT(exp != NULL);
 256         if (exp->exp_nid_stats == NULL)
 257                 goto out;
 258         obd = exp->exp_obd;
 259         ocd = &exp->exp_connect_data;
 260
 261         seq_printf(m, "%s:\n"
 262                    "    name: %s\n"
 263                    "    client: %s\n"
 264                    "    connect_flags: [ ",
 265                    obd_uuid2str(&exp->exp_client_uuid),
 266                    obd->obd_name,
 267                    obd_export_nid2str(exp));
 268         obd_connect_seq_flags2str(m, ocd->ocd_connect_flags,
 269                                   ocd->ocd_connect_flags2, ", ");
 270         seq_printf(m, " ]\n");
 271         obd_connect_data_seqprint(m, ocd);
 272         seq_printf(m, "    export_flags: [ ");
 273         obd_export_flags2str(exp, m);
 274         seq_printf(m, " ]\n");
 275
 276         if (obd->obd_type &&
 277             strcmp(obd->obd_type->typ_name, "obdfilter") == 0) {
 278                 struct filter_export_data *fed = &exp->exp_filter_data;
 279
 280                 seq_printf(m, "    grant:\n");
 281                 seq_printf(m, "       granted: %ld\n",
 282                         fed->fed_ted.ted_grant);
 283                 seq_printf(m, "       dirty: %ld\n",
 284                         fed->fed_ted.ted_dirty);
 285                 seq_printf(m, "       pending: %ld\n",
 286                         fed->fed_ted.ted_pending);
 287         }
 288
 289 out:
 290         return 0;
 291 }
 292
 293 /**
 294  * RPC connections are composed of an import and an export. Using the
 295  * lctl utility we can extract important information about the state.
 296  * The lprocfs_exp_export_seq_show routine displays the state information
 297  * for the export.
 298  *
 299  * \param[in] m         seq file
 300  * \param[in] data      unused
 301  *
 302  * \retval              0 on success
 303  *
 304  * The format of the export state information is like:
 305  * a793e354-49c0-aa11-8c4f-a4f2b1a1a92b:
 306  *     name: MGS
 307  *     client: 10.211.55.10@tcp
 308  *     connect_flags: [ version, barrier, adaptive_timeouts, ... ]
 309  *     connect_data:
 310  *        flags: 0x2000011005002020
 311  *        instance: 0
 312  *        target_version: 2.10.51.0
 313  *        export_flags: [ ... ]
 314  *
 315  */
 316 static int lprocfs_exp_export_seq_show(struct seq_file *m, void *data)
 317 {
 318         struct nid_stat *stats = m->private;
 319
 320         return obd_nid_export_for_each(stats->nid_obd, &stats->nid,
 321                                        lprocfs_exp_print_export_seq, m);
 322 }
 323 LPROC_SEQ_FOPS_RO(lprocfs_exp_export);
 324
 325 static void lprocfs_free_client_stats(struct nid_stat *client_stat)
 326 {
 327         CDEBUG(D_CONFIG, "stat %p - data %p/%p\n", client_stat,
 328                client_stat->nid_proc, client_stat->nid_stats);
 329
 330         LASSERTF(atomic_read(&client_stat->nid_exp_ref_count) == 0,
 331                  "nid %s:count %d\n", libcfs_nidstr(&client_stat->nid),
 332                  atomic_read(&client_stat->nid_exp_ref_count));
 333
 334         if (client_stat->nid_proc)
 335                 lprocfs_remove(&client_stat->nid_proc);
 336
 337         if (client_stat->nid_stats)
 338                 lprocfs_stats_free(&client_stat->nid_stats);
 339
 340         if (client_stat->nid_ldlm_stats)
 341                 lprocfs_stats_free(&client_stat->nid_ldlm_stats);
 342
 343         OBD_FREE_PTR(client_stat);
 344 }
 345
 346 void lprocfs_free_per_client_stats(struct obd_device *obd)
 347 {
 348         struct cfs_hash *hash = obd->obd_nid_stats_hash;
 349         struct nid_stat *stat;
 350         ENTRY;
 351
 352         /* we need extra list - because hash_exit called to early */
 353         /* not need locking because all clients is died */
 354         while (!list_empty(&obd->obd_nid_stats)) {
 355                 stat = list_entry(obd->obd_nid_stats.next,
 356                                   struct nid_stat, nid_list);
 357                 list_del_init(&stat->nid_list);
 358                 cfs_hash_del(hash, &stat->nid, &stat->nid_hash);
 359                 lprocfs_free_client_stats(stat);
 360         }
 361         EXIT;
 362 }
 363 EXPORT_SYMBOL(lprocfs_free_per_client_stats);
 364
 365 static int
 366 lprocfs_exp_print_nodemap_seq(struct obd_export *exp, void *cb_data)
 367 {
 368         struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap;
 369         struct seq_file *m = cb_data;
 370
 371         if (nodemap)
 372                 seq_printf(m, "%s\n", nodemap->nm_name);
 373         return 0;
 374 }
 375
 376 static int
 377 lprocfs_exp_nodemap_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_nodemap_seq, m);
 383 }
 384 LPROC_SEQ_FOPS_RO(lprocfs_exp_nodemap);
 385
 386 static int
 387 lprocfs_exp_print_uuid_seq(struct obd_export *exp, void *cb_data)
 388 {
 389         struct seq_file *m = cb_data;
 390
 391         if (exp->exp_nid_stats)
 392                 seq_printf(m, "%s\n", obd_uuid2str(&exp->exp_client_uuid));
 393         return 0;
 394 }
 395
 396 static int lprocfs_exp_uuid_seq_show(struct seq_file *m, void *data)
 397 {
 398         struct nid_stat *stats = m->private;
 399
 400         return obd_nid_export_for_each(stats->nid_obd, &stats->nid,
 401                                        lprocfs_exp_print_uuid_seq, m);
 402 }
 403 LPROC_SEQ_FOPS_RO(lprocfs_exp_uuid);
 404
 405 #define HASH_NAME_LEN   16
 406
 407 static void ldebugfs_rhash_seq_show(const char *name, struct rhashtable *ht,
 408                                     struct seq_file *m)
 409 {
 410         unsigned int max_size = ht->p.max_size ? ht->p.max_size : UINT_MAX;
 411         struct bucket_table *tbl;
 412         int dist[8] = { 0, };
 413         int maxdep = 0;
 414         int i;
 415
 416         rcu_read_lock();
 417         tbl = rht_dereference(ht->tbl, ht);
 418         for (i = 0; i < tbl->size; i++) {
 419                 struct rhash_head *pos;
 420                 int count = 0;
 421
 422                 rht_for_each(pos, tbl, i)
 423                         count++;
 424
 425                 if (count)
 426                         maxdep = max(maxdep, count);
 427
 428                 dist[min(fls(count), 7)]++;
 429         }
 430
 431         seq_printf(m, "%-*s %5d %5d %10u %d.%03d 0.300 0.750 0x%03x %7d %7d %7d ",
 432                    HASH_NAME_LEN, name, tbl->size, ht->p.min_size, max_size,
 433                    atomic_read(&ht->nelems) / tbl->size,
 434                    atomic_read(&ht->nelems) * 1000 / tbl->size,
 435                    ht->p.automatic_shrinking, 0,
 436                    atomic_read(&ht->nelems), maxdep);
 437         rcu_read_unlock();
 438
 439         for (i = 0; i < 8; i++)
 440                 seq_printf(m, "%d%c",  dist[i], (i == 7) ? '\n' : '/');
 441 }
 442
 443 static int
 444 lprocfs_exp_print_hash_seq(struct obd_export *exp, void *cb_data)
 445
 446 {
 447         struct obd_device *obd = exp->exp_obd;
 448         struct seq_file *m = cb_data;
 449
 450         if (exp->exp_lock_hash != NULL) {
 451                 seq_printf(m, "%-*s   cur   min        max theta t-min t-max flags rehash   count distribution\n",
 452                            HASH_NAME_LEN, "name");
 453                 ldebugfs_rhash_seq_show("NID_HASH", &obd->obd_nid_hash.ht, m);
 454         }
 455         return 0;
 456 }
 457
 458 static int lprocfs_exp_hash_seq_show(struct seq_file *m, void *data)
 459 {
 460         struct nid_stat *stats = m->private;
 461
 462         return obd_nid_export_for_each(stats->nid_obd, &stats->nid,
 463                                        lprocfs_exp_print_hash_seq, m);
 464 }
 465 LPROC_SEQ_FOPS_RO(lprocfs_exp_hash);
 466
 467 int lprocfs_exp_print_replydata_seq(struct obd_export *exp, void *cb_data)
 468
 469 {
 470         struct seq_file *m = cb_data;
 471         struct tg_export_data *ted = &exp->exp_target_data;
 472
 473         seq_printf(m, "reply_cnt: %d\n"
 474                    "reply_max: %d\n"
 475                    "reply_released_by_xid: %d\n"
 476                    "reply_released_by_tag: %d\n\n",
 477                    ted->ted_reply_cnt,
 478                    ted->ted_reply_max,
 479                    ted->ted_release_xid,
 480                    ted->ted_release_tag);
 481         return 0;
 482 }
 483
 484 int lprocfs_exp_replydata_seq_show(struct seq_file *m, void *data)
 485 {
 486         struct nid_stat *stats = m->private;
 487
 488         return obd_nid_export_for_each(stats->nid_obd, &stats->nid,
 489                                        lprocfs_exp_print_replydata_seq, m);
 490 }
 491 LPROC_SEQ_FOPS_RO(lprocfs_exp_replydata);
 492
 493 int lprocfs_exp_print_fmd_count_seq(struct obd_export *exp, void *cb_data)
 494 {
 495         struct seq_file *m = cb_data;
 496         struct tg_export_data *ted = &exp->exp_target_data;
 497
 498         seq_printf(m, "%d\n", ted->ted_fmd_count);
 499
 500         return 0;
 501 }
 502
 503 int lprocfs_exp_fmd_count_seq_show(struct seq_file *m, void *data)
 504 {
 505         struct nid_stat *stats = m->private;
 506
 507         return obd_nid_export_for_each(stats->nid_obd, &stats->nid,
 508                                        lprocfs_exp_print_fmd_count_seq, m);
 509 }
 510 LPROC_SEQ_FOPS_RO(lprocfs_exp_fmd_count);
 511
 512 int lprocfs_nid_stats_clear_seq_show(struct seq_file *m, void *data)
 513 {
 514         seq_puts(m, "Write into this file to clear all nid stats and stale nid entries\n");
 515         return 0;
 516 }
 517 EXPORT_SYMBOL(lprocfs_nid_stats_clear_seq_show);
 518
 519 static int lprocfs_nid_stats_clear_write_cb(void *obj, void *data)
 520 {
 521         struct nid_stat *stat = obj;
 522         ENTRY;
 523
 524         CDEBUG(D_INFO, "refcnt %d\n", atomic_read(&stat->nid_exp_ref_count));
 525         if (atomic_read(&stat->nid_exp_ref_count) == 1) {
 526                 /* object has only hash references. */
 527                 spin_lock(&stat->nid_obd->obd_nid_lock);
 528                 list_move(&stat->nid_list, data);
 529                 spin_unlock(&stat->nid_obd->obd_nid_lock);
 530                 RETURN(1);
 531         }
 532         /* we has reference to object - only clear data*/
 533         if (stat->nid_stats)
 534                 lprocfs_stats_clear(stat->nid_stats);
 535
 536         RETURN(0);
 537 }
 538
 539 ssize_t
 540 lprocfs_nid_stats_clear_seq_write(struct file *file, const char __user *buffer,
 541                                         size_t count, loff_t *off)
 542 {
 543         struct seq_file *m = file->private_data;
 544         struct obd_device *obd = m->private;
 545         struct nid_stat *client_stat;
 546         LIST_HEAD(free_list);
 547
 548         cfs_hash_cond_del(obd->obd_nid_stats_hash,
 549                           lprocfs_nid_stats_clear_write_cb, &free_list);
 550
 551         while (!list_empty(&free_list)) {
 552                 client_stat = list_entry(free_list.next, struct nid_stat,
 553                                          nid_list);
 554                 list_del_init(&client_stat->nid_list);
 555                 lprocfs_free_client_stats(client_stat);
 556         }
 557         return count;
 558 }
 559 EXPORT_SYMBOL(lprocfs_nid_stats_clear_seq_write);
 560
 561 int lprocfs_exp_setup(struct obd_export *exp, struct lnet_nid *nid)
 562 {
 563         struct nid_stat *new_stat, *old_stat;
 564         struct obd_device *obd = NULL;
 565         struct proc_dir_entry *entry;
 566         char nidstr[LNET_NIDSTR_SIZE];
 567         int rc = 0;
 568         ENTRY;
 569
 570         if (!exp || !exp->exp_obd || !exp->exp_obd->obd_proc_exports_entry ||
 571             !exp->exp_obd->obd_nid_stats_hash)
 572                 RETURN(-EINVAL);
 573
 574         /* not test against zero because eric say:
 575          * You may only test nid against another nid, or LNET_NID_ANY.
 576          * Anything else is nonsense.*/
 577         if (nid == NULL || LNET_NID_IS_ANY(nid))
 578                 RETURN(-EALREADY);
 579
 580         libcfs_nidstr_r(nid, nidstr, sizeof(nidstr));
 581
 582         spin_lock(&exp->exp_lock);
 583         if (exp->exp_nid_stats != NULL) {
 584                 spin_unlock(&exp->exp_lock);
 585                 RETURN(-EALREADY);
 586         }
 587         spin_unlock(&exp->exp_lock);
 588
 589         obd = exp->exp_obd;
 590
 591         CDEBUG(D_CONFIG, "using hash %p\n", obd->obd_nid_stats_hash);
 592
 593         OBD_ALLOC_PTR(new_stat);
 594         if (new_stat == NULL)
 595                 RETURN(-ENOMEM);
 596
 597         new_stat->nid = *nid;
 598         new_stat->nid_obd = exp->exp_obd;
 599         /* we need set default refcount to 1 to balance obd_disconnect */
 600         atomic_set(&new_stat->nid_exp_ref_count, 1);
 601
 602         old_stat = cfs_hash_findadd_unique(obd->obd_nid_stats_hash,
 603                                            &new_stat->nid,
 604                                            &new_stat->nid_hash);
 605         CDEBUG(D_INFO, "Found stats %p for nid %s - ref %d\n",
 606                old_stat, nidstr, atomic_read(&old_stat->nid_exp_ref_count));
 607
 608         /* Return -EALREADY here so that we know that the /proc
 609          * entry already has been created */
 610         if (old_stat != new_stat) {
 611                 spin_lock(&exp->exp_lock);
 612                 if (exp->exp_nid_stats) {
 613                         LASSERT(exp->exp_nid_stats == old_stat);
 614                         nidstat_putref(exp->exp_nid_stats);
 615                 }
 616                 exp->exp_nid_stats = old_stat;
 617                 spin_unlock(&exp->exp_lock);
 618                 GOTO(destroy_new, rc = -EALREADY);
 619         }
 620         /* not found - create */
 621         new_stat->nid_proc = lprocfs_register(nidstr,
 622                                               obd->obd_proc_exports_entry,
 623                                               NULL, NULL);
 624
 625         if (IS_ERR(new_stat->nid_proc)) {
 626                 rc = PTR_ERR(new_stat->nid_proc);
 627                 new_stat->nid_proc = NULL;
 628                 CERROR("%s: cannot create proc entry for export %s: rc = %d\n",
 629                        obd->obd_name, nidstr, rc);
 630                 GOTO(destroy_new_ns, rc);
 631         }
 632
 633         entry = lprocfs_add_simple(new_stat->nid_proc, "nodemap", new_stat,
 634                                    &lprocfs_exp_nodemap_fops);
 635         if (IS_ERR(entry)) {
 636                 rc = PTR_ERR(entry);
 637                 CWARN("%s: error adding the nodemap 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, "uuid", new_stat,
 643                                    &lprocfs_exp_uuid_fops);
 644         if (IS_ERR(entry)) {
 645                 rc = PTR_ERR(entry);
 646                 CWARN("%s: error adding the NID stats file: rc = %d\n",
 647                       obd->obd_name, rc);
 648                 GOTO(destroy_new_ns, rc);
 649         }
 650
 651         entry = lprocfs_add_simple(new_stat->nid_proc, "hash", new_stat,
 652                                    &lprocfs_exp_hash_fops);
 653         if (IS_ERR(entry)) {
 654                 rc = PTR_ERR(entry);
 655                 CWARN("%s: error adding the hash file: rc = %d\n",
 656                       obd->obd_name, rc);
 657                 GOTO(destroy_new_ns, rc);
 658         }
 659
 660         entry = lprocfs_add_simple(new_stat->nid_proc, "export",
 661                                    new_stat, &lprocfs_exp_export_fops);
 662         if (IS_ERR(entry)) {
 663                 rc = PTR_ERR(entry);
 664                 CWARN("%s: error adding the export file: rc = %d\n",
 665                       obd->obd_name, rc);
 666                 GOTO(destroy_new_ns, rc);
 667         }
 668
 669         entry = lprocfs_add_simple(new_stat->nid_proc, "reply_data", new_stat,
 670                                    &lprocfs_exp_replydata_fops);
 671         if (IS_ERR(entry)) {
 672                 rc = PTR_ERR(entry);
 673                 CWARN("%s: error adding the reply_data file: rc = %d\n",
 674                       obd->obd_name, rc);
 675                 GOTO(destroy_new_ns, rc);
 676         }
 677
 678         entry = lprocfs_add_simple(new_stat->nid_proc, "fmd_count", new_stat,
 679                                    &lprocfs_exp_fmd_count_fops);
 680         if (IS_ERR(entry)) {
 681                 rc = PTR_ERR(entry);
 682                 CWARN("%s: error adding the fmd_count file: rc = %d\n",
 683                       obd->obd_name, rc);
 684                 GOTO(destroy_new_ns, rc);
 685         }
 686
 687         spin_lock(&exp->exp_lock);
 688         exp->exp_nid_stats = new_stat;
 689         spin_unlock(&exp->exp_lock);
 690
 691         /* protect competitive add to list, not need locking on destroy */
 692         spin_lock(&obd->obd_nid_lock);
 693         list_add(&new_stat->nid_list, &obd->obd_nid_stats);
 694         spin_unlock(&obd->obd_nid_lock);
 695
 696         RETURN(0);
 697
 698 destroy_new_ns:
 699         if (new_stat->nid_proc != NULL)
 700                 lprocfs_remove(&new_stat->nid_proc);
 701         cfs_hash_del(obd->obd_nid_stats_hash, &new_stat->nid,
 702                      &new_stat->nid_hash);
 703
 704 destroy_new:
 705         nidstat_putref(new_stat);
 706         OBD_FREE_PTR(new_stat);
 707         RETURN(rc);
 708 }
 709 EXPORT_SYMBOL(lprocfs_exp_setup);
 710
 711 int lprocfs_exp_cleanup(struct obd_export *exp)
 712 {
 713         struct nid_stat *stat = exp->exp_nid_stats;
 714
 715         if (!stat || !exp->exp_obd)
 716                 RETURN(0);
 717
 718         nidstat_putref(exp->exp_nid_stats);
 719         exp->exp_nid_stats = NULL;
 720
 721         return 0;
 722 }
 723
 724 int lprocfs_alloc_obd_stats(struct obd_device *obd, unsigned int num_stats)
 725 {
 726         struct lprocfs_stats *stats;
 727         int rc;
 728
 729         LASSERT(obd->obd_stats == NULL);
 730         LASSERT(obd->obd_proc_entry != NULL);
 731
 732         stats = lprocfs_stats_alloc(num_stats, 0);
 733         if (stats == NULL)
 734                 return -ENOMEM;
 735
 736         rc = lprocfs_stats_register(obd->obd_proc_entry, "stats", stats);
 737         if (rc < 0)
 738                 lprocfs_stats_free(&stats);
 739         else
 740                 obd->obd_stats = stats;
 741
 742         return rc;
 743 }
 744 EXPORT_SYMBOL(lprocfs_alloc_obd_stats);
 745
 746 void lprocfs_free_obd_stats(struct obd_device *obd)
 747 {
 748         if (obd->obd_stats)
 749                 lprocfs_stats_free(&obd->obd_stats);
 750 }
 751 EXPORT_SYMBOL(lprocfs_free_obd_stats);
 752
 753 static void display_brw_stats(struct seq_file *seq, const char *name,
 754                               const char *units, struct obd_hist_pcpu *read,
 755                               struct obd_hist_pcpu *write, bool scale)
 756 {
 757         unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
 758         unsigned int i;
 759
 760         seq_printf(seq, "\n%26s read      |     write\n", " ");
 761         seq_printf(seq, "%-22s %-5s %% cum %% |  %-11s %% cum %%\n",
 762                    name, units, units);
 763
 764         read_tot = lprocfs_oh_sum_pcpu(read);
 765         write_tot = lprocfs_oh_sum_pcpu(write);
 766
 767         if (!read_tot && !write_tot)
 768                 return;
 769
 770         for (i = 0; i < OBD_HIST_MAX; i++) {
 771                 r = lprocfs_oh_counter_pcpu(read, i);
 772                 w = lprocfs_oh_counter_pcpu(write, i);
 773                 read_cum += r;
 774                 write_cum += w;
 775                 if (read_cum == 0 && write_cum == 0)
 776                         continue;
 777
 778                 if (!scale)
 779                         seq_printf(seq, "%u", i);
 780                 else if (i < 10)
 781                         seq_printf(seq, "%lu", BIT(i));
 782                 else if (i < 20)
 783                         seq_printf(seq, "%luK", BIT(i - 10));
 784                 else
 785                         seq_printf(seq, "%luM", BIT(i - 20));
 786
 787                 seq_printf(seq, ":\t\t%10lu %3u %3u   | %4lu %3u %3u\n",
 788                            r, pct(r, read_tot), pct(read_cum, read_tot),
 789                            w, pct(w, write_tot), pct(write_cum, write_tot));
 790
 791                 if (read_cum == read_tot && write_cum == write_tot)
 792                         break;
 793         }
 794 }
 795
 796 static const struct brw_stats_props brw_props[] = {
 797         { .bsp_name     = "pages per bulk r/w",
 798           .bsp_units    = "rpcs",
 799           .bsp_scale    = true                          },
 800         { .bsp_name     = "discontiguous pages",
 801           .bsp_units    = "rpcs",
 802           .bsp_scale    = false                         },
 803         { .bsp_name     = "discontiguous blocks",
 804           .bsp_units    = "rpcs",
 805           .bsp_scale    = false                         },
 806         { .bsp_name     = "disk fragmented I/Os",
 807           .bsp_units    = "ios",
 808           .bsp_scale    = false                         },
 809         { .bsp_name     = "disk I/Os in flight",
 810           .bsp_units    = "ios",
 811           .bsp_scale    = false                         },
 812         { .bsp_name     = "I/O time (1/1000s)",
 813           .bsp_units    = "ios",
 814           .bsp_scale    = true                          },
 815         { .bsp_name     = "disk I/O size",
 816           .bsp_units    = "ios",
 817           .bsp_scale    = true                          },
 818         { .bsp_name     = "block maps msec",
 819           .bsp_units    = "maps",
 820           .bsp_scale    = true,                         },
 821 };
 822
 823 static int brw_stats_seq_show(struct seq_file *seq, void *v)
 824 {
 825         struct brw_stats *brw_stats = seq->private;
 826         int i;
 827
 828         /* this sampling races with updates */
 829         lprocfs_stats_header(seq, ktime_get_real(), brw_stats->bs_init, 25,
 830                              ":", true, "");
 831
 832         for (i = 0; i < ARRAY_SIZE(brw_stats->bs_props); i++) {
 833                 if (!brw_stats->bs_props[i].bsp_name)
 834                         continue;
 835
 836                 display_brw_stats(seq, brw_stats->bs_props[i].bsp_name,
 837                                   brw_stats->bs_props[i].bsp_units,
 838                                   &brw_stats->bs_hist[i * 2],
 839                                   &brw_stats->bs_hist[i * 2 + 1],
 840                                   brw_stats->bs_props[i].bsp_scale);
 841         }
 842
 843         return 0;
 844 }
 845
 846 static ssize_t brw_stats_seq_write(struct file *file,
 847                                    const char __user *buf,
 848                                    size_t len, loff_t *off)
 849 {
 850         struct seq_file *seq = file->private_data;
 851         struct brw_stats *brw_stats = seq->private;
 852         int i;
 853
 854         for (i = 0; i < BRW_RW_STATS_NUM; i++)
 855                 lprocfs_oh_clear_pcpu(&brw_stats->bs_hist[i]);
 856         brw_stats->bs_init = ktime_get_real();
 857
 858         return len;
 859 }
 860
 861 LDEBUGFS_SEQ_FOPS(brw_stats);
 862
 863 int lprocfs_init_brw_stats(struct brw_stats *brw_stats)
 864 {
 865         int i, result;
 866
 867         for (i = 0; i < BRW_RW_STATS_NUM; i++) {
 868                 result = lprocfs_oh_alloc_pcpu(&brw_stats->bs_hist[i]);
 869                 if (result)
 870                         break;
 871         }
 872
 873         return result;
 874 }
 875 EXPORT_SYMBOL(lprocfs_init_brw_stats);
 876
 877 void lprocfs_fini_brw_stats(struct brw_stats *brw_stats)
 878 {
 879         int i;
 880
 881         for (i = 0; i < BRW_RW_STATS_NUM; i++)
 882                 lprocfs_oh_release_pcpu(&brw_stats->bs_hist[i]);
 883 }
 884 EXPORT_SYMBOL(lprocfs_fini_brw_stats);
 885
 886 void ldebugfs_register_osd_stats(struct dentry *parent,
 887                                  struct brw_stats *brw_stats,
 888                                  struct lprocfs_stats *stats)
 889 {
 890         int i;
 891
 892         LASSERT(brw_stats);
 893         brw_stats->bs_init = ktime_get_real();
 894         for (i = 0; i < BRW_RW_STATS_NUM; i++) {
 895                 struct brw_stats_props *props = brw_stats->bs_props;
 896
 897                 if (i % 2) {
 898                         props[i / 2].bsp_name = brw_props[i / 2].bsp_name;
 899                         props[i / 2].bsp_units = brw_props[i / 2].bsp_units;
 900                         props[i / 2].bsp_scale = brw_props[i / 2].bsp_scale;
 901                 }
 902         }
 903
 904         if (!parent)
 905                 return;
 906
 907         debugfs_create_file("brw_stats", 0644, parent, brw_stats,
 908                             &brw_stats_fops);
 909
 910         if (stats)
 911                 debugfs_create_file("stats", 0644, parent, stats,
 912                                     &ldebugfs_stats_seq_fops);
 913 }
 914 EXPORT_SYMBOL(ldebugfs_register_osd_stats);
 915
 916 int lprocfs_hash_seq_show(struct seq_file *m, void *data)
 917 {
 918         struct obd_device *obd = m->private;
 919
 920         if (obd == NULL)
 921                 return 0;
 922
 923         /* header for rhashtable state */
 924         seq_printf(m, "%-*s   cur   min        max theta t-min t-max flags  rehash   count  maxdep distribution\n",
 925                    HASH_NAME_LEN, "name");
 926         ldebugfs_rhash_seq_show("UUID_HASH", &obd->obd_uuid_hash, m);
 927         ldebugfs_rhash_seq_show("NID_HASH", &obd->obd_nid_hash.ht, m);
 928
 929         cfs_hash_debug_header(m);
 930         cfs_hash_debug_str(obd->obd_nid_stats_hash, m);
 931         return 0;
 932 }
 933 EXPORT_SYMBOL(lprocfs_hash_seq_show);
 934
 935 int lprocfs_recovery_status_seq_show(struct seq_file *m, void *data)
 936 {
 937         struct obd_device *obd = m->private;
 938         struct target_distribute_txn_data *tdtd;
 939
 940         LASSERT(obd != NULL);
 941
 942         seq_printf(m, "status: ");
 943         if (atomic_read(&obd->obd_max_recoverable_clients) == 0) {
 944                 seq_printf(m, "INACTIVE\n");
 945                 goto out;
 946         }
 947
 948         /* There is gap between client data read from storage and setting
 949          * obd_recovering so check obd_recovery_end as well to make sure
 950          * recovery is really finished
 951          */
 952         if (obd->obd_recovery_end > 0 && !obd->obd_recovering) {
 953                 seq_printf(m, "COMPLETE\n");
 954                 seq_printf(m, "recovery_start: %lld\n",
 955                            (s64)ktime_get_real_seconds() -
 956                            (ktime_get_seconds() - obd->obd_recovery_start));
 957                 seq_printf(m, "recovery_duration: %lld\n",
 958                            obd->obd_recovery_end ?
 959                            obd->obd_recovery_end - obd->obd_recovery_start :
 960                            ktime_get_seconds() - obd->obd_recovery_start);
 961                 /* Number of clients that have completed recovery */
 962                 seq_printf(m, "completed_clients: %d/%d\n",
 963                            atomic_read(&obd->obd_max_recoverable_clients) -
 964                            obd->obd_stale_clients,
 965                            atomic_read(&obd->obd_max_recoverable_clients));
 966                 seq_printf(m, "replayed_requests: %d\n",
 967                            obd->obd_replayed_requests);
 968                 seq_printf(m, "last_transno: %lld\n",
 969                            obd->obd_next_recovery_transno - 1);
 970                 seq_printf(m, "VBR: %s\n", obd->obd_version_recov ?
 971                            "ENABLED" : "DISABLED");
 972                 seq_printf(m, "IR: %s\n", obd->obd_no_ir ?
 973                            "DISABLED" : "ENABLED");
 974                 goto out;
 975         }
 976
 977         tdtd = obd2obt(obd)->obt_lut->lut_tdtd;
 978         if (tdtd && tdtd->tdtd_show_update_logs_retrievers) {
 979                 char *buf;
 980                 int size = 0;
 981                 int count = 0;
 982
 983                 buf = tdtd->tdtd_show_update_logs_retrievers(
 984                         tdtd->tdtd_show_retrievers_cbdata,
 985                         &size, &count);
 986                 if (count > 0) {
 987                         seq_printf(m, "WAITING\n");
 988                         seq_printf(m, "non-ready MDTs: %s\n",
 989                                    buf ? buf : "unknown (not enough RAM)");
 990                         seq_printf(m, "recovery_start: %lld\n",
 991                                    (s64)ktime_get_real_seconds() -
 992                                    (ktime_get_seconds() -
 993                                     obd->obd_recovery_start));
 994                         seq_printf(m, "time_waited: %lld\n",
 995                                    (s64)(ktime_get_seconds() -
 996                                          obd->obd_recovery_start));
 997                 }
 998
 999                 if (buf != NULL)
1000                         OBD_FREE(buf, size);
1001
1002                 if (likely(count > 0))
1003                         goto out;
1004         }
1005
1006         /* recovery won't start until the clients connect */
1007         if (obd->obd_recovery_start == 0) {
1008                 seq_printf(m, "WAITING_FOR_CLIENTS\n");
1009                 goto out;
1010         }
1011
1012         seq_printf(m, "RECOVERING\n");
1013         seq_printf(m, "recovery_start: %lld\n", (s64)ktime_get_real_seconds() -
1014                    (ktime_get_seconds() - obd->obd_recovery_start));
1015         seq_printf(m, "time_remaining: %lld\n",
1016                    ktime_get_seconds() >=
1017                    obd->obd_recovery_start +
1018                    obd->obd_recovery_timeout ? 0 :
1019                    (s64)(obd->obd_recovery_start +
1020                          obd->obd_recovery_timeout -
1021                          ktime_get_seconds()));
1022         seq_printf(m, "connected_clients: %d/%d\n",
1023                    atomic_read(&obd->obd_connected_clients),
1024                    atomic_read(&obd->obd_max_recoverable_clients));
1025         /* Number of clients that have completed recovery */
1026         seq_printf(m, "req_replay_clients: %d\n",
1027                    atomic_read(&obd->obd_req_replay_clients));
1028         seq_printf(m, "lock_repay_clients: %d\n",
1029                    atomic_read(&obd->obd_lock_replay_clients));
1030         seq_printf(m, "completed_clients: %d\n",
1031                    atomic_read(&obd->obd_connected_clients) -
1032                    atomic_read(&obd->obd_lock_replay_clients));
1033         seq_printf(m, "evicted_clients: %d\n", obd->obd_stale_clients);
1034         seq_printf(m, "replayed_requests: %d\n", obd->obd_replayed_requests);
1035         seq_printf(m, "queued_requests: %d\n",
1036                    obd->obd_requests_queued_for_recovery);
1037         seq_printf(m, "next_transno: %lld\n",
1038                    obd->obd_next_recovery_transno);
1039 out:
1040         return 0;
1041 }
1042 EXPORT_SYMBOL(lprocfs_recovery_status_seq_show);
1043
1044 ssize_t ir_factor_show(struct kobject *kobj, struct attribute *attr,
1045                        char *buf)
1046 {
1047         struct obd_device *obd = container_of(kobj, struct obd_device,
1048                                               obd_kset.kobj);
1049
1050         return scnprintf(buf, PAGE_SIZE, "%d\n", obd->obd_recovery_ir_factor);
1051 }
1052 EXPORT_SYMBOL(ir_factor_show);
1053
1054 ssize_t ir_factor_store(struct kobject *kobj, struct attribute *attr,
1055                         const char *buffer, size_t count)
1056 {
1057         struct obd_device *obd = container_of(kobj, struct obd_device,
1058                                               obd_kset.kobj);
1059         int val;
1060         int rc;
1061
1062         rc = kstrtoint(buffer, 10, &val);
1063         if (rc)
1064                 return rc;
1065
1066         if (val < OBD_IR_FACTOR_MIN || val > OBD_IR_FACTOR_MAX)
1067                 return -EINVAL;
1068
1069         obd->obd_recovery_ir_factor = val;
1070         return count;
1071 }
1072 EXPORT_SYMBOL(ir_factor_store);
1073
1074 int lprocfs_checksum_dump_seq_show(struct seq_file *m, void *data)
1075 {
1076         struct obd_device *obd = m->private;
1077
1078         LASSERT(obd != NULL);
1079         seq_printf(m, "%d\n", obd->obd_checksum_dump);
1080         return 0;
1081 }
1082 EXPORT_SYMBOL(lprocfs_checksum_dump_seq_show);
1083
1084 ssize_t
1085 lprocfs_checksum_dump_seq_write(struct file *file, const char __user *buffer,
1086                             size_t count, loff_t *off)
1087 {
1088         struct seq_file *m = file->private_data;
1089         struct obd_device *obd = m->private;
1090         bool val;
1091         int rc;
1092
1093         LASSERT(obd != NULL);
1094         rc = kstrtobool_from_user(buffer, count, &val);
1095         if (rc)
1096                 return rc;
1097
1098         obd->obd_checksum_dump = val;
1099         return count;
1100 }
1101 EXPORT_SYMBOL(lprocfs_checksum_dump_seq_write);
1102
1103 ssize_t recovery_time_soft_show(struct kobject *kobj, struct attribute *attr,
1104                                 char *buf)
1105 {
1106         struct obd_device *obd = container_of(kobj, struct obd_device,
1107                                               obd_kset.kobj);
1108
1109         return scnprintf(buf, PAGE_SIZE, "%d\n", obd->obd_recovery_timeout);
1110 }
1111 EXPORT_SYMBOL(recovery_time_soft_show);
1112
1113 ssize_t recovery_time_soft_store(struct kobject *kobj,
1114                                  struct attribute *attr,
1115                                  const char *buffer, size_t count)
1116 {
1117         struct obd_device *obd = container_of(kobj, struct obd_device,
1118                                               obd_kset.kobj);
1119         unsigned int val;
1120         int rc;
1121
1122         rc = kstrtouint(buffer, 0, &val);
1123         if (rc)
1124                 return rc;
1125
1126         obd->obd_recovery_timeout = val;
1127         return count;
1128 }
1129 EXPORT_SYMBOL(recovery_time_soft_store);
1130
1131 ssize_t recovery_time_hard_show(struct kobject *kobj, struct attribute *attr,
1132                                 char *buf)
1133 {
1134         struct obd_device *obd = container_of(kobj, struct obd_device,
1135                                               obd_kset.kobj);
1136
1137         return scnprintf(buf, PAGE_SIZE, "%d\n", obd->obd_recovery_time_hard);
1138 }
1139 EXPORT_SYMBOL(recovery_time_hard_show);
1140
1141 ssize_t recovery_time_hard_store(struct kobject *kobj,
1142                                  struct attribute *attr,
1143                                  const char *buffer, size_t count)
1144 {
1145         struct obd_device *obd = container_of(kobj, struct obd_device,
1146                                               obd_kset.kobj);
1147         unsigned int val;
1148         int rc;
1149
1150         rc = kstrtouint(buffer, 0, &val);
1151         if (rc)
1152                 return rc;
1153
1154         obd->obd_recovery_time_hard = val;
1155         return count;
1156 }
1157 EXPORT_SYMBOL(recovery_time_hard_store);
1158
1159 ssize_t instance_show(struct kobject *kobj, struct attribute *attr,
1160                       char *buf)
1161 {
1162         struct obd_device *obd = container_of(kobj, struct obd_device,
1163                                               obd_kset.kobj);
1164         struct obd_device_target *target = obd2obt(obd);
1165
1166         LASSERT(target->obt_magic == OBT_MAGIC);
1167         return scnprintf(buf, PAGE_SIZE, "%u\n", obd2obt(obd)->obt_instance);
1168 }
1169 EXPORT_SYMBOL(instance_show);
1170
1171 #endif /* CONFIG_PROC_FS*/