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 static int obd_export_flags2str(struct obd_export *exp, struct seq_file *m)
 158 {
 159         bool first = true;
 160
 161         flag2str(exp, failed);
 162         flag2str(exp, in_recovery);
 163         flag2str(exp, disconnected);
 164         flag2str(exp, connecting);
 165         flag2str(exp, no_recovery);
 166
 167         return 0;
 168 }
 169
 170 static int
 171 lprocfs_exp_print_export_seq(struct obd_export *exp, void *cb_data)
 172 {
 173         struct seq_file         *m = cb_data;
 174         struct obd_device       *obd;
 175         struct obd_connect_data *ocd;
 176
 177         LASSERT(exp != NULL);
 178         if (exp->exp_nid_stats == NULL)
 179                 goto out;
 180         obd = exp->exp_obd;
 181         ocd = &exp->exp_connect_data;
 182
 183         seq_printf(m, "%s:\n"
 184                    "    name: %s\n"
 185                    "    client: %s\n"
 186                    "    connect_flags: [ ",
 187                    obd_uuid2str(&exp->exp_client_uuid),
 188                    obd->obd_name,
 189                    obd_export_nid2str(exp));
 190         obd_connect_seq_flags2str(m, ocd->ocd_connect_flags,
 191                                   ocd->ocd_connect_flags2, ", ");
 192         seq_printf(m, " ]\n");
 193         obd_connect_data_seqprint(m, ocd);
 194         seq_printf(m, "    export_flags: [ ");
 195         obd_export_flags2str(exp, m);
 196         seq_printf(m, " ]\n");
 197
 198         if (obd->obd_type &&
 199             strcmp(obd->obd_type->typ_name, "obdfilter") == 0) {
 200                 struct filter_export_data *fed = &exp->exp_filter_data;
 201
 202                 seq_printf(m, "    grant:\n");
 203                 seq_printf(m, "       granted: %ld\n",
 204                         fed->fed_ted.ted_grant);
 205                 seq_printf(m, "       dirty: %ld\n",
 206                         fed->fed_ted.ted_dirty);
 207                 seq_printf(m, "       pending: %ld\n",
 208                         fed->fed_ted.ted_pending);
 209         }
 210
 211 out:
 212         return 0;
 213 }
 214
 215 /**
 216  * RPC connections are composed of an import and an export. Using the
 217  * lctl utility we can extract important information about the state.
 218  * The lprocfs_exp_export_seq_show routine displays the state information
 219  * for the export.
 220  *
 221  * \param[in] m         seq file
 222  * \param[in] data      unused
 223  *
 224  * \retval              0 on success
 225  *
 226  * The format of the export state information is like:
 227  * a793e354-49c0-aa11-8c4f-a4f2b1a1a92b:
 228  *     name: MGS
 229  *     client: 10.211.55.10@tcp
 230  *     connect_flags: [ version, barrier, adaptive_timeouts, ... ]
 231  *     connect_data:
 232  *        flags: 0x2000011005002020
 233  *        instance: 0
 234  *        target_version: 2.10.51.0
 235  *        export_flags: [ ... ]
 236  *
 237  */
 238 static int lprocfs_exp_export_seq_show(struct seq_file *m, void *data)
 239 {
 240         struct nid_stat *stats = m->private;
 241
 242         return obd_nid_export_for_each(stats->nid_obd, stats->nid,
 243                                        lprocfs_exp_print_export_seq, m);
 244 }
 245 LPROC_SEQ_FOPS_RO(lprocfs_exp_export);
 246
 247 static void lprocfs_free_client_stats(struct nid_stat *client_stat)
 248 {
 249         CDEBUG(D_CONFIG, "stat %p - data %p/%p\n", client_stat,
 250                client_stat->nid_proc, client_stat->nid_stats);
 251
 252         LASSERTF(atomic_read(&client_stat->nid_exp_ref_count) == 0,
 253                  "nid %s:count %d\n", libcfs_nid2str(client_stat->nid),
 254                  atomic_read(&client_stat->nid_exp_ref_count));
 255
 256         if (client_stat->nid_proc)
 257                 lprocfs_remove(&client_stat->nid_proc);
 258
 259         if (client_stat->nid_stats)
 260                 lprocfs_free_stats(&client_stat->nid_stats);
 261
 262         if (client_stat->nid_ldlm_stats)
 263                 lprocfs_free_stats(&client_stat->nid_ldlm_stats);
 264
 265         OBD_FREE_PTR(client_stat);
 266 }
 267
 268 void lprocfs_free_per_client_stats(struct obd_device *obd)
 269 {
 270         struct cfs_hash *hash = obd->obd_nid_stats_hash;
 271         struct nid_stat *stat;
 272         ENTRY;
 273
 274         /* we need extra list - because hash_exit called to early */
 275         /* not need locking because all clients is died */
 276         while (!list_empty(&obd->obd_nid_stats)) {
 277                 stat = list_entry(obd->obd_nid_stats.next,
 278                                   struct nid_stat, nid_list);
 279                 list_del_init(&stat->nid_list);
 280                 cfs_hash_del(hash, &stat->nid, &stat->nid_hash);
 281                 lprocfs_free_client_stats(stat);
 282         }
 283         EXIT;
 284 }
 285 EXPORT_SYMBOL(lprocfs_free_per_client_stats);
 286
 287 static int
 288 lprocfs_exp_print_nodemap_seq(struct obd_export *exp, void *cb_data)
 289 {
 290         struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap;
 291         struct seq_file *m = cb_data;
 292
 293         if (nodemap)
 294                 seq_printf(m, "%s\n", nodemap->nm_name);
 295         return 0;
 296 }
 297
 298 static int
 299 lprocfs_exp_nodemap_seq_show(struct seq_file *m, void *data)
 300 {
 301         struct nid_stat *stats = m->private;
 302
 303         return obd_nid_export_for_each(stats->nid_obd, stats->nid,
 304                                        lprocfs_exp_print_nodemap_seq, m);
 305 }
 306 LPROC_SEQ_FOPS_RO(lprocfs_exp_nodemap);
 307
 308 static int
 309 lprocfs_exp_print_uuid_seq(struct obd_export *exp, void *cb_data)
 310 {
 311         struct seq_file *m = cb_data;
 312
 313         if (exp->exp_nid_stats)
 314                 seq_printf(m, "%s\n", obd_uuid2str(&exp->exp_client_uuid));
 315         return 0;
 316 }
 317
 318 static int lprocfs_exp_uuid_seq_show(struct seq_file *m, void *data)
 319 {
 320         struct nid_stat *stats = m->private;
 321
 322         return obd_nid_export_for_each(stats->nid_obd, stats->nid,
 323                                        lprocfs_exp_print_uuid_seq, m);
 324 }
 325 LPROC_SEQ_FOPS_RO(lprocfs_exp_uuid);
 326
 327 #define HASH_NAME_LEN   16
 328
 329 static void ldebugfs_rhash_seq_show(const char *name, struct rhashtable *ht,
 330                                     struct seq_file *m)
 331 {
 332         unsigned int max_size = ht->p.max_size ? ht->p.max_size : UINT_MAX;
 333         struct bucket_table *tbl;
 334         int dist[8] = { 0, };
 335         int maxdep = 0;
 336         int i;
 337
 338         rcu_read_lock();
 339         tbl = rht_dereference(ht->tbl, ht);
 340         for (i = 0; i < tbl->size; i++) {
 341                 struct rhash_head *pos;
 342                 int count = 0;
 343
 344                 rht_for_each(pos, tbl, i)
 345                         count++;
 346
 347                 if (count)
 348                         maxdep = max(maxdep, count);
 349
 350                 dist[min(fls(count), 7)]++;
 351         }
 352
 353         seq_printf(m, "%-*s %5d %5d %10u %d.%03d 0.300 0.750 0x%03x %7d %7d %7d ",
 354                    HASH_NAME_LEN, name, tbl->size, ht->p.min_size, max_size,
 355                    atomic_read(&ht->nelems) / tbl->size,
 356                    atomic_read(&ht->nelems) * 1000 / tbl->size,
 357                    ht->p.automatic_shrinking, 0,
 358                    atomic_read(&ht->nelems), maxdep);
 359         rcu_read_unlock();
 360
 361         for (i = 0; i < 8; i++)
 362                 seq_printf(m, "%d%c",  dist[i], (i == 7) ? '\n' : '/');
 363 }
 364
 365 static int
 366 lprocfs_exp_print_hash_seq(struct obd_export *exp, void *cb_data)
 367
 368 {
 369         struct obd_device *obd = exp->exp_obd;
 370         struct seq_file *m = cb_data;
 371
 372         if (exp->exp_lock_hash != NULL) {
 373                 seq_printf(m, "%-*s   cur   min        max theta t-min t-max flags rehash   count distribution\n",
 374                            HASH_NAME_LEN, "name");
 375                 ldebugfs_rhash_seq_show("NID_HASH", &obd->obd_nid_hash.ht, m);
 376         }
 377         return 0;
 378 }
 379
 380 static int lprocfs_exp_hash_seq_show(struct seq_file *m, void *data)
 381 {
 382         struct nid_stat *stats = m->private;
 383
 384         return obd_nid_export_for_each(stats->nid_obd, stats->nid,
 385                                        lprocfs_exp_print_hash_seq, m);
 386 }
 387 LPROC_SEQ_FOPS_RO(lprocfs_exp_hash);
 388
 389 int lprocfs_exp_print_replydata_seq(struct obd_export *exp, void *cb_data)
 390
 391 {
 392         struct seq_file *m = cb_data;
 393         struct tg_export_data *ted = &exp->exp_target_data;
 394
 395         seq_printf(m, "reply_cnt: %d\n"
 396                    "reply_max: %d\n"
 397                    "reply_released_by_xid: %d\n"
 398                    "reply_released_by_tag: %d\n\n",
 399                    ted->ted_reply_cnt,
 400                    ted->ted_reply_max,
 401                    ted->ted_release_xid,
 402                    ted->ted_release_tag);
 403         return 0;
 404 }
 405
 406 int lprocfs_exp_replydata_seq_show(struct seq_file *m, void *data)
 407 {
 408         struct nid_stat *stats = m->private;
 409
 410         return obd_nid_export_for_each(stats->nid_obd, stats->nid,
 411                                        lprocfs_exp_print_replydata_seq, m);
 412 }
 413 LPROC_SEQ_FOPS_RO(lprocfs_exp_replydata);
 414
 415 int lprocfs_exp_print_fmd_count_seq(struct obd_export *exp, void *cb_data)
 416 {
 417         struct seq_file *m = cb_data;
 418         struct tg_export_data *ted = &exp->exp_target_data;
 419
 420         seq_printf(m, "%d\n", ted->ted_fmd_count);
 421
 422         return 0;
 423 }
 424
 425 int lprocfs_exp_fmd_count_seq_show(struct seq_file *m, void *data)
 426 {
 427         struct nid_stat *stats = m->private;
 428
 429         return obd_nid_export_for_each(stats->nid_obd, stats->nid,
 430                                        lprocfs_exp_print_fmd_count_seq, m);
 431 }
 432 LPROC_SEQ_FOPS_RO(lprocfs_exp_fmd_count);
 433
 434 int lprocfs_nid_stats_clear_seq_show(struct seq_file *m, void *data)
 435 {
 436         seq_puts(m, "Write into this file to clear all nid stats and stale nid entries\n");
 437         return 0;
 438 }
 439 EXPORT_SYMBOL(lprocfs_nid_stats_clear_seq_show);
 440
 441 static int lprocfs_nid_stats_clear_write_cb(void *obj, void *data)
 442 {
 443         struct nid_stat *stat = obj;
 444         ENTRY;
 445
 446         CDEBUG(D_INFO, "refcnt %d\n", atomic_read(&stat->nid_exp_ref_count));
 447         if (atomic_read(&stat->nid_exp_ref_count) == 1) {
 448                 /* object has only hash references. */
 449                 spin_lock(&stat->nid_obd->obd_nid_lock);
 450                 list_move(&stat->nid_list, data);
 451                 spin_unlock(&stat->nid_obd->obd_nid_lock);
 452                 RETURN(1);
 453         }
 454         /* we has reference to object - only clear data*/
 455         if (stat->nid_stats)
 456                 lprocfs_clear_stats(stat->nid_stats);
 457
 458         RETURN(0);
 459 }
 460
 461 ssize_t
 462 lprocfs_nid_stats_clear_seq_write(struct file *file, const char __user *buffer,
 463                                         size_t count, loff_t *off)
 464 {
 465         struct seq_file *m = file->private_data;
 466         struct obd_device *obd = m->private;
 467         struct nid_stat *client_stat;
 468         LIST_HEAD(free_list);
 469
 470         cfs_hash_cond_del(obd->obd_nid_stats_hash,
 471                           lprocfs_nid_stats_clear_write_cb, &free_list);
 472
 473         while (!list_empty(&free_list)) {
 474                 client_stat = list_entry(free_list.next, struct nid_stat,
 475                                          nid_list);
 476                 list_del_init(&client_stat->nid_list);
 477                 lprocfs_free_client_stats(client_stat);
 478         }
 479         return count;
 480 }
 481 EXPORT_SYMBOL(lprocfs_nid_stats_clear_seq_write);
 482
 483 int lprocfs_exp_setup(struct obd_export *exp, lnet_nid_t *nid)
 484 {
 485         struct nid_stat *new_stat, *old_stat;
 486         struct obd_device *obd = NULL;
 487         struct proc_dir_entry *entry;
 488         char nidstr[LNET_NIDSTR_SIZE];
 489         int rc = 0;
 490         ENTRY;
 491
 492         if (!exp || !exp->exp_obd || !exp->exp_obd->obd_proc_exports_entry ||
 493             !exp->exp_obd->obd_nid_stats_hash)
 494                 RETURN(-EINVAL);
 495
 496         /* not test against zero because eric say:
 497          * You may only test nid against another nid, or LNET_NID_ANY.
 498          * Anything else is nonsense.*/
 499         if (nid == NULL || *nid == LNET_NID_ANY)
 500                 RETURN(-EALREADY);
 501
 502         libcfs_nid2str_r(*nid, nidstr, sizeof(nidstr));
 503
 504         spin_lock(&exp->exp_lock);
 505         if (exp->exp_nid_stats != NULL) {
 506                 spin_unlock(&exp->exp_lock);
 507                 RETURN(-EALREADY);
 508         }
 509         spin_unlock(&exp->exp_lock);
 510
 511         obd = exp->exp_obd;
 512
 513         CDEBUG(D_CONFIG, "using hash %p\n", obd->obd_nid_stats_hash);
 514
 515         OBD_ALLOC_PTR(new_stat);
 516         if (new_stat == NULL)
 517                 RETURN(-ENOMEM);
 518
 519         new_stat->nid     = *nid;
 520         new_stat->nid_obd = exp->exp_obd;
 521         /* we need set default refcount to 1 to balance obd_disconnect */
 522         atomic_set(&new_stat->nid_exp_ref_count, 1);
 523
 524         old_stat = cfs_hash_findadd_unique(obd->obd_nid_stats_hash,
 525                                            nid, &new_stat->nid_hash);
 526         CDEBUG(D_INFO, "Found stats %p for nid %s - ref %d\n",
 527                old_stat, nidstr, atomic_read(&old_stat->nid_exp_ref_count));
 528
 529         /* Return -EALREADY here so that we know that the /proc
 530          * entry already has been created */
 531         if (old_stat != new_stat) {
 532                 spin_lock(&exp->exp_lock);
 533                 if (exp->exp_nid_stats) {
 534                         LASSERT(exp->exp_nid_stats == old_stat);
 535                         nidstat_putref(exp->exp_nid_stats);
 536                 }
 537                 exp->exp_nid_stats = old_stat;
 538                 spin_unlock(&exp->exp_lock);
 539                 GOTO(destroy_new, rc = -EALREADY);
 540         }
 541         /* not found - create */
 542         new_stat->nid_proc = lprocfs_register(nidstr,
 543                                               obd->obd_proc_exports_entry,
 544                                               NULL, NULL);
 545
 546         if (IS_ERR(new_stat->nid_proc)) {
 547                 rc = PTR_ERR(new_stat->nid_proc);
 548                 new_stat->nid_proc = NULL;
 549                 CERROR("%s: cannot create proc entry for export %s: rc = %d\n",
 550                        obd->obd_name, nidstr, rc);
 551                 GOTO(destroy_new_ns, rc);
 552         }
 553
 554         entry = lprocfs_add_simple(new_stat->nid_proc, "nodemap", new_stat,
 555                                    &lprocfs_exp_nodemap_fops);
 556         if (IS_ERR(entry)) {
 557                 rc = PTR_ERR(entry);
 558                 CWARN("%s: error adding the nodemap file: rc = %d\n",
 559                       obd->obd_name, rc);
 560                 GOTO(destroy_new_ns, rc);
 561         }
 562
 563         entry = lprocfs_add_simple(new_stat->nid_proc, "uuid", new_stat,
 564                                    &lprocfs_exp_uuid_fops);
 565         if (IS_ERR(entry)) {
 566                 rc = PTR_ERR(entry);
 567                 CWARN("%s: error adding the NID stats file: rc = %d\n",
 568                       obd->obd_name, rc);
 569                 GOTO(destroy_new_ns, rc);
 570         }
 571
 572         entry = lprocfs_add_simple(new_stat->nid_proc, "hash", new_stat,
 573                                    &lprocfs_exp_hash_fops);
 574         if (IS_ERR(entry)) {
 575                 rc = PTR_ERR(entry);
 576                 CWARN("%s: error adding the hash file: rc = %d\n",
 577                       obd->obd_name, rc);
 578                 GOTO(destroy_new_ns, rc);
 579         }
 580
 581         entry = lprocfs_add_simple(new_stat->nid_proc, "export",
 582                                    new_stat, &lprocfs_exp_export_fops);
 583         if (IS_ERR(entry)) {
 584                 rc = PTR_ERR(entry);
 585                 CWARN("%s: error adding the export file: rc = %d\n",
 586                       obd->obd_name, rc);
 587                 GOTO(destroy_new_ns, rc);
 588         }
 589
 590         entry = lprocfs_add_simple(new_stat->nid_proc, "reply_data", new_stat,
 591                                    &lprocfs_exp_replydata_fops);
 592         if (IS_ERR(entry)) {
 593                 rc = PTR_ERR(entry);
 594                 CWARN("%s: error adding the reply_data file: rc = %d\n",
 595                       obd->obd_name, rc);
 596                 GOTO(destroy_new_ns, rc);
 597         }
 598
 599         entry = lprocfs_add_simple(new_stat->nid_proc, "fmd_count", new_stat,
 600                                    &lprocfs_exp_fmd_count_fops);
 601         if (IS_ERR(entry)) {
 602                 rc = PTR_ERR(entry);
 603                 CWARN("%s: error adding the fmd_count file: rc = %d\n",
 604                       obd->obd_name, rc);
 605                 GOTO(destroy_new_ns, rc);
 606         }
 607
 608         spin_lock(&exp->exp_lock);
 609         exp->exp_nid_stats = new_stat;
 610         spin_unlock(&exp->exp_lock);
 611
 612         /* protect competitive add to list, not need locking on destroy */
 613         spin_lock(&obd->obd_nid_lock);
 614         list_add(&new_stat->nid_list, &obd->obd_nid_stats);
 615         spin_unlock(&obd->obd_nid_lock);
 616
 617         RETURN(0);
 618
 619 destroy_new_ns:
 620         if (new_stat->nid_proc != NULL)
 621                 lprocfs_remove(&new_stat->nid_proc);
 622         cfs_hash_del(obd->obd_nid_stats_hash, nid, &new_stat->nid_hash);
 623
 624 destroy_new:
 625         nidstat_putref(new_stat);
 626         OBD_FREE_PTR(new_stat);
 627         RETURN(rc);
 628 }
 629 EXPORT_SYMBOL(lprocfs_exp_setup);
 630
 631 int lprocfs_exp_cleanup(struct obd_export *exp)
 632 {
 633         struct nid_stat *stat = exp->exp_nid_stats;
 634
 635         if (!stat || !exp->exp_obd)
 636                 RETURN(0);
 637
 638         nidstat_putref(exp->exp_nid_stats);
 639         exp->exp_nid_stats = NULL;
 640
 641         return 0;
 642 }
 643
 644 int lprocfs_alloc_obd_stats(struct obd_device *obd, unsigned int num_stats)
 645 {
 646         struct lprocfs_stats *stats;
 647         int rc;
 648
 649         LASSERT(obd->obd_stats == NULL);
 650         LASSERT(obd->obd_proc_entry != NULL);
 651
 652         stats = lprocfs_alloc_stats(num_stats, 0);
 653         if (stats == NULL)
 654                 return -ENOMEM;
 655
 656         rc = lprocfs_register_stats(obd->obd_proc_entry, "stats", stats);
 657         if (rc < 0)
 658                 lprocfs_free_stats(&stats);
 659         else
 660                 obd->obd_stats = stats;
 661
 662         return rc;
 663 }
 664 EXPORT_SYMBOL(lprocfs_alloc_obd_stats);
 665
 666 void lprocfs_free_obd_stats(struct obd_device *obd)
 667 {
 668         if (obd->obd_stats)
 669                 lprocfs_free_stats(&obd->obd_stats);
 670 }
 671 EXPORT_SYMBOL(lprocfs_free_obd_stats);
 672
 673 static void display_brw_stats(struct seq_file *seq, const char *name,
 674                               const char *units, struct obd_histogram *read,
 675                               struct obd_histogram *write, bool scale)
 676 {
 677         unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
 678         unsigned int i;
 679
 680         seq_printf(seq, "\n%26s read      |     write\n", " ");
 681         seq_printf(seq, "%-22s %-5s %% cum %% |  %-11s %% cum %%\n",
 682                    name, units, units);
 683
 684         read_tot = lprocfs_oh_sum(read);
 685         write_tot = lprocfs_oh_sum(write);
 686
 687         if (!read_tot && !write_tot)
 688                 return;
 689
 690         for (i = 0; i < OBD_HIST_MAX; i++) {
 691                 r = read->oh_buckets[i];
 692                 w = write->oh_buckets[i];
 693                 read_cum += r;
 694                 write_cum += w;
 695                 if (read_cum == 0 && write_cum == 0)
 696                         continue;
 697
 698                 if (!scale)
 699                         seq_printf(seq, "%u", i);
 700                 else if (i < 10)
 701                         seq_printf(seq, "%lu", BIT(i));
 702                 else if (i < 20)
 703                         seq_printf(seq, "%luK", BIT(i - 10));
 704                 else
 705                         seq_printf(seq, "%luM", BIT(i - 20));
 706
 707                 seq_printf(seq, ":\t\t%10lu %3u %3u   | %4lu %3u %3u\n",
 708                            r, pct(r, read_tot), pct(read_cum, read_tot),
 709                            w, pct(w, write_tot), pct(write_cum, write_tot));
 710
 711                 if (read_cum == read_tot && write_cum == write_tot)
 712                         break;
 713         }
 714 }
 715
 716 static const struct brw_stats_props brw_props[] = {
 717         { .bsp_name     = "pages per bulk r/w",
 718           .bsp_units    = "rpcs",
 719           .bsp_scale    = true                          },
 720         { .bsp_name     = "discontiguous pages",
 721           .bsp_units    = "rpcs",
 722           .bsp_scale    = false                         },
 723         { .bsp_name     = "discontiguous blocks",
 724           .bsp_units    = "rpcs",
 725           .bsp_scale    = false                         },
 726         { .bsp_name     = "disk fragmented I/Os",
 727           .bsp_units    = "ios",
 728           .bsp_scale    = false                         },
 729         { .bsp_name     = "disk I/Os in flight",
 730           .bsp_units    = "ios",
 731           .bsp_scale    = false                         },
 732         { .bsp_name     = "I/O time (1/1000s)",
 733           .bsp_units    = "ios",
 734           .bsp_scale    = true                          },
 735         { .bsp_name     = "disk I/O size",
 736           .bsp_units    = "ios",
 737           .bsp_scale    = true                          },
 738 };
 739
 740 static int brw_stats_seq_show(struct seq_file *seq, void *v)
 741 {
 742         struct brw_stats *brw_stats = seq->private;
 743         int i;
 744
 745         /* this sampling races with updates */
 746         lprocfs_stats_header(seq, ktime_get(), brw_stats->bs_init, 25, ":", 1);
 747
 748         for (i = 0; i < ARRAY_SIZE(brw_stats->bs_props); i++) {
 749                 if (!brw_stats->bs_props[i].bsp_name)
 750                         continue;
 751
 752                 display_brw_stats(seq, brw_stats->bs_props[i].bsp_name,
 753                                   brw_stats->bs_props[i].bsp_units,
 754                                   &brw_stats->bs_hist[i * 2],
 755                                   &brw_stats->bs_hist[i * 2 + 1],
 756                                   brw_stats->bs_props[i].bsp_scale);
 757         }
 758
 759         return 0;
 760 }
 761
 762 static ssize_t brw_stats_seq_write(struct file *file,
 763                                    const char __user *buf,
 764                                    size_t len, loff_t *off)
 765 {
 766         struct seq_file *seq = file->private_data;
 767         struct brw_stats *brw_stats = seq->private;
 768         int i;
 769
 770         for (i = 0; i < BRW_RW_STATS_NUM; i++)
 771                 lprocfs_oh_clear(&brw_stats->bs_hist[i]);
 772
 773         return len;
 774 }
 775
 776 LDEBUGFS_SEQ_FOPS(brw_stats);
 777
 778 void ldebugfs_register_osd_stats(struct dentry *parent,
 779                                  struct brw_stats *brw_stats,
 780                                  struct lprocfs_stats *stats)
 781 {
 782         int i;
 783
 784         LASSERT(brw_stats);
 785         brw_stats->bs_init = ktime_get();
 786         for (i = 0; i < BRW_RW_STATS_NUM; i++) {
 787                 struct brw_stats_props *props = brw_stats->bs_props;
 788
 789                 spin_lock_init(&brw_stats->bs_hist[i].oh_lock);
 790                 if (i % 2) {
 791                         props[i / 2].bsp_name = brw_props[i / 2].bsp_name;
 792                         props[i / 2].bsp_units = brw_props[i / 2].bsp_units;
 793                         props[i / 2].bsp_scale = brw_props[i / 2].bsp_scale;
 794                 }
 795         }
 796
 797         if (!parent)
 798                 return;
 799
 800         debugfs_create_file("brw_stats", 0644, parent, brw_stats,
 801                             &brw_stats_fops);
 802
 803         if (stats)
 804                 debugfs_create_file("stats", 0644, parent, stats,
 805                                     &ldebugfs_stats_seq_fops);
 806 }
 807 EXPORT_SYMBOL(ldebugfs_register_osd_stats);
 808
 809 int lprocfs_hash_seq_show(struct seq_file *m, void *data)
 810 {
 811         struct obd_device *obd = m->private;
 812
 813         if (obd == NULL)
 814                 return 0;
 815
 816         /* header for rhashtable state */
 817         seq_printf(m, "%-*s   cur   min        max theta t-min t-max flags  rehash   count  maxdep distribution\n",
 818                    HASH_NAME_LEN, "name");
 819         ldebugfs_rhash_seq_show("UUID_HASH", &obd->obd_uuid_hash, m);
 820         ldebugfs_rhash_seq_show("NID_HASH", &obd->obd_nid_hash.ht, m);
 821
 822         cfs_hash_debug_header(m);
 823         cfs_hash_debug_str(obd->obd_nid_stats_hash, m);
 824         return 0;
 825 }
 826 EXPORT_SYMBOL(lprocfs_hash_seq_show);
 827
 828 int lprocfs_recovery_status_seq_show(struct seq_file *m, void *data)
 829 {
 830         struct obd_device *obd = m->private;
 831         struct target_distribute_txn_data *tdtd;
 832
 833         LASSERT(obd != NULL);
 834
 835         seq_printf(m, "status: ");
 836         if (atomic_read(&obd->obd_max_recoverable_clients) == 0) {
 837                 seq_printf(m, "INACTIVE\n");
 838                 goto out;
 839         }
 840
 841         /* There is gap between client data read from storage and setting
 842          * obd_recovering so check obd_recovery_end as well to make sure
 843          * recovery is really finished
 844          */
 845         if (obd->obd_recovery_end > 0 && !obd->obd_recovering) {
 846                 seq_printf(m, "COMPLETE\n");
 847                 seq_printf(m, "recovery_start: %lld\n",
 848                            (s64)ktime_get_real_seconds() -
 849                            (ktime_get_seconds() - obd->obd_recovery_start));
 850                 seq_printf(m, "recovery_duration: %lld\n",
 851                            obd->obd_recovery_end ?
 852                            obd->obd_recovery_end - obd->obd_recovery_start :
 853                            ktime_get_seconds() - obd->obd_recovery_start);
 854                 /* Number of clients that have completed recovery */
 855                 seq_printf(m, "completed_clients: %d/%d\n",
 856                            atomic_read(&obd->obd_max_recoverable_clients) -
 857                            obd->obd_stale_clients,
 858                            atomic_read(&obd->obd_max_recoverable_clients));
 859                 seq_printf(m, "replayed_requests: %d\n",
 860                            obd->obd_replayed_requests);
 861                 seq_printf(m, "last_transno: %lld\n",
 862                            obd->obd_next_recovery_transno - 1);
 863                 seq_printf(m, "VBR: %s\n", obd->obd_version_recov ?
 864                            "ENABLED" : "DISABLED");
 865                 seq_printf(m, "IR: %s\n", obd->obd_no_ir ?
 866                            "DISABLED" : "ENABLED");
 867                 goto out;
 868         }
 869
 870         tdtd = obd->u.obt.obt_lut->lut_tdtd;
 871         if (tdtd && tdtd->tdtd_show_update_logs_retrievers) {
 872                 char *buf;
 873                 int size = 0;
 874                 int count = 0;
 875
 876                 buf = tdtd->tdtd_show_update_logs_retrievers(
 877                         tdtd->tdtd_show_retrievers_cbdata,
 878                         &size, &count);
 879                 if (count > 0) {
 880                         seq_printf(m, "WAITING\n");
 881                         seq_printf(m, "non-ready MDTs: %s\n",
 882                                    buf ? buf : "unknown (not enough RAM)");
 883                         seq_printf(m, "recovery_start: %lld\n",
 884                                    (s64)ktime_get_real_seconds() -
 885                                    (ktime_get_seconds() -
 886                                     obd->obd_recovery_start));
 887                         seq_printf(m, "time_waited: %lld\n",
 888                                    (s64)(ktime_get_seconds() -
 889                                          obd->obd_recovery_start));
 890                 }
 891
 892                 if (buf != NULL)
 893                         OBD_FREE(buf, size);
 894
 895                 if (likely(count > 0))
 896                         goto out;
 897         }
 898
 899         /* recovery won't start until the clients connect */
 900         if (obd->obd_recovery_start == 0) {
 901                 seq_printf(m, "WAITING_FOR_CLIENTS\n");
 902                 goto out;
 903         }
 904
 905         seq_printf(m, "RECOVERING\n");
 906         seq_printf(m, "recovery_start: %lld\n", (s64)ktime_get_real_seconds() -
 907                    (ktime_get_seconds() - obd->obd_recovery_start));
 908         seq_printf(m, "time_remaining: %lld\n",
 909                    ktime_get_seconds() >=
 910                    obd->obd_recovery_start +
 911                    obd->obd_recovery_timeout ? 0 :
 912                    (s64)(obd->obd_recovery_start +
 913                          obd->obd_recovery_timeout -
 914                          ktime_get_seconds()));
 915         seq_printf(m, "connected_clients: %d/%d\n",
 916                    atomic_read(&obd->obd_connected_clients),
 917                    atomic_read(&obd->obd_max_recoverable_clients));
 918         /* Number of clients that have completed recovery */
 919         seq_printf(m, "req_replay_clients: %d\n",
 920                    atomic_read(&obd->obd_req_replay_clients));
 921         seq_printf(m, "lock_repay_clients: %d\n",
 922                    atomic_read(&obd->obd_lock_replay_clients));
 923         seq_printf(m, "completed_clients: %d\n",
 924                    atomic_read(&obd->obd_connected_clients) -
 925                    atomic_read(&obd->obd_lock_replay_clients));
 926         seq_printf(m, "evicted_clients: %d\n", obd->obd_stale_clients);
 927         seq_printf(m, "replayed_requests: %d\n", obd->obd_replayed_requests);
 928         seq_printf(m, "queued_requests: %d\n",
 929                    obd->obd_requests_queued_for_recovery);
 930         seq_printf(m, "next_transno: %lld\n",
 931                    obd->obd_next_recovery_transno);
 932 out:
 933         return 0;
 934 }
 935 EXPORT_SYMBOL(lprocfs_recovery_status_seq_show);
 936
 937 ssize_t ir_factor_show(struct kobject *kobj, struct attribute *attr,
 938                        char *buf)
 939 {
 940         struct obd_device *obd = container_of(kobj, struct obd_device,
 941                                               obd_kset.kobj);
 942
 943         return scnprintf(buf, PAGE_SIZE, "%d\n", obd->obd_recovery_ir_factor);
 944 }
 945 EXPORT_SYMBOL(ir_factor_show);
 946
 947 ssize_t ir_factor_store(struct kobject *kobj, struct attribute *attr,
 948                         const char *buffer, size_t count)
 949 {
 950         struct obd_device *obd = container_of(kobj, struct obd_device,
 951                                               obd_kset.kobj);
 952         int val;
 953         int rc;
 954
 955         rc = kstrtoint(buffer, 10, &val);
 956         if (rc)
 957                 return rc;
 958
 959         if (val < OBD_IR_FACTOR_MIN || val > OBD_IR_FACTOR_MAX)
 960                 return -EINVAL;
 961
 962         obd->obd_recovery_ir_factor = val;
 963         return count;
 964 }
 965 EXPORT_SYMBOL(ir_factor_store);
 966
 967 int lprocfs_checksum_dump_seq_show(struct seq_file *m, void *data)
 968 {
 969         struct obd_device *obd = m->private;
 970
 971         LASSERT(obd != NULL);
 972         seq_printf(m, "%d\n", obd->obd_checksum_dump);
 973         return 0;
 974 }
 975 EXPORT_SYMBOL(lprocfs_checksum_dump_seq_show);
 976
 977 ssize_t
 978 lprocfs_checksum_dump_seq_write(struct file *file, const char __user *buffer,
 979                             size_t count, loff_t *off)
 980 {
 981         struct seq_file *m = file->private_data;
 982         struct obd_device *obd = m->private;
 983         bool val;
 984         int rc;
 985
 986         LASSERT(obd != NULL);
 987         rc = kstrtobool_from_user(buffer, count, &val);
 988         if (rc)
 989                 return rc;
 990
 991         obd->obd_checksum_dump = val;
 992         return count;
 993 }
 994 EXPORT_SYMBOL(lprocfs_checksum_dump_seq_write);
 995
 996 ssize_t recovery_time_soft_show(struct kobject *kobj, struct attribute *attr,
 997                                 char *buf)
 998 {
 999         struct obd_device *obd = container_of(kobj, struct obd_device,
1000                                               obd_kset.kobj);
1001
1002         return scnprintf(buf, PAGE_SIZE, "%d\n", obd->obd_recovery_timeout);
1003 }
1004 EXPORT_SYMBOL(recovery_time_soft_show);
1005
1006 ssize_t recovery_time_soft_store(struct kobject *kobj,
1007                                  struct attribute *attr,
1008                                  const char *buffer, size_t count)
1009 {
1010         struct obd_device *obd = container_of(kobj, struct obd_device,
1011                                               obd_kset.kobj);
1012         unsigned int val;
1013         int rc;
1014
1015         rc = kstrtouint(buffer, 0, &val);
1016         if (rc)
1017                 return rc;
1018
1019         obd->obd_recovery_timeout = val;
1020         return count;
1021 }
1022 EXPORT_SYMBOL(recovery_time_soft_store);
1023
1024 ssize_t recovery_time_hard_show(struct kobject *kobj, struct attribute *attr,
1025                                 char *buf)
1026 {
1027         struct obd_device *obd = container_of(kobj, struct obd_device,
1028                                               obd_kset.kobj);
1029
1030         return scnprintf(buf, PAGE_SIZE, "%d\n", obd->obd_recovery_time_hard);
1031 }
1032 EXPORT_SYMBOL(recovery_time_hard_show);
1033
1034 ssize_t recovery_time_hard_store(struct kobject *kobj,
1035                                  struct attribute *attr,
1036                                  const char *buffer, size_t count)
1037 {
1038         struct obd_device *obd = container_of(kobj, struct obd_device,
1039                                               obd_kset.kobj);
1040         unsigned int val;
1041         int rc;
1042
1043         rc = kstrtouint(buffer, 0, &val);
1044         if (rc)
1045                 return rc;
1046
1047         obd->obd_recovery_time_hard = val;
1048         return count;
1049 }
1050 EXPORT_SYMBOL(recovery_time_hard_store);
1051
1052 ssize_t instance_show(struct kobject *kobj, struct attribute *attr,
1053                       char *buf)
1054 {
1055         struct obd_device *obd = container_of(kobj, struct obd_device,
1056                                               obd_kset.kobj);
1057         struct obd_device_target *target = &obd->u.obt;
1058
1059         LASSERT(target->obt_magic == OBT_MAGIC);
1060         return scnprintf(buf, PAGE_SIZE, "%u\n", obd->u.obt.obt_instance);
1061 }
1062 EXPORT_SYMBOL(instance_show);
1063
1064 #endif /* CONFIG_PROC_FS*/