These changes are the raw update to linux-4.4.6-rt14. Kernel sources
[kvmfornfv.git] / kernel / drivers / staging / lustre / lustre / ptlrpc / lproc_ptlrpc.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.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2012, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  */
36 #define DEBUG_SUBSYSTEM S_CLASS
37
38 #include "../include/obd_support.h"
39 #include "../include/obd.h"
40 #include "../include/lprocfs_status.h"
41 #include "../include/lustre/lustre_idl.h"
42 #include "../include/lustre_net.h"
43 #include "../include/obd_class.h"
44 #include "ptlrpc_internal.h"
45
46 static struct ll_rpc_opcode {
47         __u32       opcode;
48         const char *opname;
49 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
50         { OST_REPLY,    "ost_reply" },
51         { OST_GETATTR,      "ost_getattr" },
52         { OST_SETATTR,      "ost_setattr" },
53         { OST_READ,      "ost_read" },
54         { OST_WRITE,    "ost_write" },
55         { OST_CREATE,       "ost_create" },
56         { OST_DESTROY,      "ost_destroy" },
57         { OST_GET_INFO,     "ost_get_info" },
58         { OST_CONNECT,      "ost_connect" },
59         { OST_DISCONNECT,   "ost_disconnect" },
60         { OST_PUNCH,    "ost_punch" },
61         { OST_OPEN,      "ost_open" },
62         { OST_CLOSE,    "ost_close" },
63         { OST_STATFS,       "ost_statfs" },
64         { 14,           NULL },    /* formerly OST_SAN_READ */
65         { 15,           NULL },    /* formerly OST_SAN_WRITE */
66         { OST_SYNC,      "ost_sync" },
67         { OST_SET_INFO,     "ost_set_info" },
68         { OST_QUOTACHECK,   "ost_quotacheck" },
69         { OST_QUOTACTL,     "ost_quotactl" },
70         { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
71         { MDS_GETATTR,      "mds_getattr" },
72         { MDS_GETATTR_NAME, "mds_getattr_lock" },
73         { MDS_CLOSE,    "mds_close" },
74         { MDS_REINT,    "mds_reint" },
75         { MDS_READPAGE,     "mds_readpage" },
76         { MDS_CONNECT,      "mds_connect" },
77         { MDS_DISCONNECT,   "mds_disconnect" },
78         { MDS_GETSTATUS,    "mds_getstatus" },
79         { MDS_STATFS,       "mds_statfs" },
80         { MDS_PIN,        "mds_pin" },
81         { MDS_UNPIN,    "mds_unpin" },
82         { MDS_SYNC,      "mds_sync" },
83         { MDS_DONE_WRITING, "mds_done_writing" },
84         { MDS_SET_INFO,     "mds_set_info" },
85         { MDS_QUOTACHECK,   "mds_quotacheck" },
86         { MDS_QUOTACTL,     "mds_quotactl" },
87         { MDS_GETXATTR,     "mds_getxattr" },
88         { MDS_SETXATTR,     "mds_setxattr" },
89         { MDS_WRITEPAGE,    "mds_writepage" },
90         { MDS_IS_SUBDIR,    "mds_is_subdir" },
91         { MDS_GET_INFO,     "mds_get_info" },
92         { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
93         { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
94         { MDS_HSM_ACTION,   "mds_hsm_action" },
95         { MDS_HSM_PROGRESS, "mds_hsm_progress" },
96         { MDS_HSM_REQUEST,  "mds_hsm_request" },
97         { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
98         { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
99         { MDS_SWAP_LAYOUTS,     "mds_swap_layouts" },
100         { LDLM_ENQUEUE,     "ldlm_enqueue" },
101         { LDLM_CONVERT,     "ldlm_convert" },
102         { LDLM_CANCEL,      "ldlm_cancel" },
103         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
104         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
105         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
106         { LDLM_SET_INFO,    "ldlm_set_info" },
107         { MGS_CONNECT,      "mgs_connect" },
108         { MGS_DISCONNECT,   "mgs_disconnect" },
109         { MGS_EXCEPTION,    "mgs_exception" },
110         { MGS_TARGET_REG,   "mgs_target_reg" },
111         { MGS_TARGET_DEL,   "mgs_target_del" },
112         { MGS_SET_INFO,     "mgs_set_info" },
113         { MGS_CONFIG_READ,  "mgs_config_read" },
114         { OBD_PING,      "obd_ping" },
115         { OBD_LOG_CANCEL,       "llog_cancel" },
116         { OBD_QC_CALLBACK,  "obd_quota_callback" },
117         { OBD_IDX_READ,     "dt_index_read" },
118         { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_open" },
119         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
120         { LLOG_ORIGIN_HANDLE_READ_HEADER, "llog_origin_handle_read_header" },
121         { LLOG_ORIGIN_HANDLE_WRITE_REC,  "llog_origin_handle_write_rec" },
122         { LLOG_ORIGIN_HANDLE_CLOSE,      "llog_origin_handle_close" },
123         { LLOG_ORIGIN_CONNECT,     "llog_origin_connect" },
124         { LLOG_CATINFO,           "llog_catinfo" },
125         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
126         { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
127         { QUOTA_DQACQ,      "quota_acquire" },
128         { QUOTA_DQREL,      "quota_release" },
129         { SEQ_QUERY,    "seq_query" },
130         { SEC_CTX_INIT,     "sec_ctx_init" },
131         { SEC_CTX_INIT_CONT, "sec_ctx_init_cont" },
132         { SEC_CTX_FINI,     "sec_ctx_fini" },
133         { FLD_QUERY,    "fld_query" },
134         { UPDATE_OBJ,       "update_obj" },
135 };
136
137 static struct ll_eopcode {
138         __u32       opcode;
139         const char *opname;
140 } ll_eopcode_table[EXTRA_LAST_OPC] = {
141         { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
142         { LDLM_PLAIN_ENQUEUE,   "ldlm_plain_enqueue" },
143         { LDLM_EXTENT_ENQUEUE,  "ldlm_extent_enqueue" },
144         { LDLM_FLOCK_ENQUEUE,   "ldlm_flock_enqueue" },
145         { LDLM_IBITS_ENQUEUE,   "ldlm_ibits_enqueue" },
146         { MDS_REINT_SETATTR,    "mds_reint_setattr" },
147         { MDS_REINT_CREATE,     "mds_reint_create" },
148         { MDS_REINT_LINK,       "mds_reint_link" },
149         { MDS_REINT_UNLINK,     "mds_reint_unlink" },
150         { MDS_REINT_RENAME,     "mds_reint_rename" },
151         { MDS_REINT_OPEN,       "mds_reint_open" },
152         { MDS_REINT_SETXATTR,   "mds_reint_setxattr" },
153         { BRW_READ_BYTES,       "read_bytes" },
154         { BRW_WRITE_BYTES,      "write_bytes" },
155 };
156
157 const char *ll_opcode2str(__u32 opcode)
158 {
159         /* When one of the assertions below fail, chances are that:
160          *     1) A new opcode was added in include/lustre/lustre_idl.h,
161          *      but is missing from the table above.
162          * or  2) The opcode space was renumbered or rearranged,
163          *      and the opcode_offset() function in
164          *      ptlrpc_internal.h needs to be modified.
165          */
166         __u32 offset = opcode_offset(opcode);
167
168         LASSERTF(offset < LUSTRE_MAX_OPCODES,
169                  "offset %u >= LUSTRE_MAX_OPCODES %u\n",
170                  offset, LUSTRE_MAX_OPCODES);
171         LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
172                  "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
173                  offset, ll_rpc_opcode_table[offset].opcode, opcode);
174         return ll_rpc_opcode_table[offset].opname;
175 }
176
177 static const char *ll_eopcode2str(__u32 opcode)
178 {
179         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
180         return ll_eopcode_table[opcode].opname;
181 }
182
183 static void
184 ptlrpc_ldebugfs_register(struct dentry *root, char *dir,
185                          char *name,
186                          struct dentry **debugfs_root_ret,
187                          struct lprocfs_stats **stats_ret)
188 {
189         struct dentry *svc_debugfs_entry;
190         struct lprocfs_stats *svc_stats;
191         int i, rc;
192         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
193                                           LPROCFS_CNTR_STDDEV;
194
195         LASSERT(*debugfs_root_ret == NULL);
196         LASSERT(*stats_ret == NULL);
197
198         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,
199                                         0);
200         if (svc_stats == NULL)
201                 return;
202
203         if (dir != NULL) {
204                 svc_debugfs_entry = ldebugfs_register(dir, root, NULL, NULL);
205                 if (IS_ERR(svc_debugfs_entry)) {
206                         lprocfs_free_stats(&svc_stats);
207                         return;
208                 }
209         } else {
210                 svc_debugfs_entry = root;
211         }
212
213         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
214                              svc_counter_config, "req_waittime", "usec");
215         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
216                              svc_counter_config, "req_qdepth", "reqs");
217         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
218                              svc_counter_config, "req_active", "reqs");
219         lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
220                              svc_counter_config, "req_timeout", "sec");
221         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
222                              svc_counter_config, "reqbuf_avail", "bufs");
223         for (i = 0; i < EXTRA_LAST_OPC; i++) {
224                 char *units;
225
226                 switch (i) {
227                 case BRW_WRITE_BYTES:
228                 case BRW_READ_BYTES:
229                         units = "bytes";
230                         break;
231                 default:
232                         units = "reqs";
233                         break;
234                 }
235                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
236                                      svc_counter_config,
237                                      ll_eopcode2str(i), units);
238         }
239         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
240                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
241
242                 lprocfs_counter_init(svc_stats,
243                                      EXTRA_MAX_OPCODES + i, svc_counter_config,
244                                      ll_opcode2str(opcode), "usec");
245         }
246
247         rc = ldebugfs_register_stats(svc_debugfs_entry, name, svc_stats);
248         if (rc < 0) {
249                 if (dir != NULL)
250                         ldebugfs_remove(&svc_debugfs_entry);
251                 lprocfs_free_stats(&svc_stats);
252         } else {
253                 if (dir != NULL)
254                         *debugfs_root_ret = svc_debugfs_entry;
255                 *stats_ret = svc_stats;
256         }
257 }
258
259 static int
260 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
261 {
262         struct ptlrpc_service *svc = m->private;
263         struct ptlrpc_service_part *svcpt;
264         int total = 0;
265         int i;
266
267         ptlrpc_service_for_each_part(svcpt, i, svc)
268                 total += svcpt->scp_hist_nrqbds;
269
270         seq_printf(m, "%d\n", total);
271         return 0;
272 }
273
274 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
275
276 static int
277 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
278 {
279         struct ptlrpc_service *svc = m->private;
280         struct ptlrpc_service_part *svcpt;
281         int total = 0;
282         int i;
283
284         ptlrpc_service_for_each_part(svcpt, i, svc)
285                 total += svc->srv_hist_nrqbds_cpt_max;
286
287         seq_printf(m, "%d\n", total);
288         return 0;
289 }
290
291 static ssize_t
292 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
293                                          const char __user *buffer,
294                                          size_t count, loff_t *off)
295 {
296         struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
297         int bufpages;
298         int val;
299         int rc;
300
301         rc = lprocfs_write_helper(buffer, count, &val);
302         if (rc < 0)
303                 return rc;
304
305         if (val < 0)
306                 return -ERANGE;
307
308         /* This sanity check is more of an insanity check; we can still
309          * hose a kernel by allowing the request history to grow too
310          * far. */
311         bufpages = (svc->srv_buf_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
312         if (val > totalram_pages / (2 * bufpages))
313                 return -ERANGE;
314
315         spin_lock(&svc->srv_lock);
316
317         if (val == 0)
318                 svc->srv_hist_nrqbds_cpt_max = 0;
319         else
320                 svc->srv_hist_nrqbds_cpt_max = max(1, (val / svc->srv_ncpts));
321
322         spin_unlock(&svc->srv_lock);
323
324         return count;
325 }
326
327 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
328
329 static ssize_t threads_min_show(struct kobject *kobj, struct attribute *attr,
330                                 char *buf)
331 {
332         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
333                                                   srv_kobj);
334
335         return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_init * svc->srv_ncpts);
336 }
337
338 static ssize_t threads_min_store(struct kobject *kobj, struct attribute *attr,
339                                  const char *buffer, size_t count)
340 {
341         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
342                                                   srv_kobj);
343         unsigned long val;
344         int rc = kstrtoul(buffer, 10, &val);
345
346         if (rc < 0)
347                 return rc;
348
349         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
350                 return -ERANGE;
351
352         spin_lock(&svc->srv_lock);
353         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
354                 spin_unlock(&svc->srv_lock);
355                 return -ERANGE;
356         }
357
358         svc->srv_nthrs_cpt_init = val / svc->srv_ncpts;
359
360         spin_unlock(&svc->srv_lock);
361
362         return count;
363 }
364 LUSTRE_RW_ATTR(threads_min);
365
366 static ssize_t threads_started_show(struct kobject *kobj,
367                                     struct attribute *attr,
368                                     char *buf)
369 {
370         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
371                                                   srv_kobj);
372         struct ptlrpc_service_part *svcpt;
373         int total = 0;
374         int i;
375
376         ptlrpc_service_for_each_part(svcpt, i, svc)
377                 total += svcpt->scp_nthrs_running;
378
379         return sprintf(buf, "%d\n", total);
380 }
381 LUSTRE_RO_ATTR(threads_started);
382
383 static ssize_t threads_max_show(struct kobject *kobj, struct attribute *attr,
384                                 char *buf)
385 {
386         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
387                                                   srv_kobj);
388
389         return sprintf(buf, "%d\n", svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
390 }
391
392 static ssize_t threads_max_store(struct kobject *kobj, struct attribute *attr,
393                                  const char *buffer, size_t count)
394 {
395         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
396                                                   srv_kobj);
397         unsigned long val;
398         int rc = kstrtoul(buffer, 10, &val);
399
400         if (rc < 0)
401                 return rc;
402
403         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
404                 return -ERANGE;
405
406         spin_lock(&svc->srv_lock);
407         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
408                 spin_unlock(&svc->srv_lock);
409                 return -ERANGE;
410         }
411
412         svc->srv_nthrs_cpt_limit = val / svc->srv_ncpts;
413
414         spin_unlock(&svc->srv_lock);
415
416         return count;
417 }
418 LUSTRE_RW_ATTR(threads_max);
419
420 /**
421  * \addtogoup nrs
422  * @{
423  */
424
425 /**
426  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
427  *
428  * \param[in] state The policy state
429  */
430 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
431 {
432         switch (state) {
433         default:
434                 LBUG();
435         case NRS_POL_STATE_INVALID:
436                 return "invalid";
437         case NRS_POL_STATE_STOPPED:
438                 return "stopped";
439         case NRS_POL_STATE_STOPPING:
440                 return "stopping";
441         case NRS_POL_STATE_STARTING:
442                 return "starting";
443         case NRS_POL_STATE_STARTED:
444                 return "started";
445         }
446 }
447
448 /**
449  * Obtains status information for \a policy.
450  *
451  * Information is copied in \a info.
452  *
453  * \param[in] policy The policy
454  * \param[out] info  Holds returned status information
455  */
456 static void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
457                                 struct ptlrpc_nrs_pol_info *info)
458 {
459         LASSERT(policy != NULL);
460         LASSERT(info != NULL);
461         assert_spin_locked(&policy->pol_nrs->nrs_lock);
462
463         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
464
465         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
466         info->pi_state       = policy->pol_state;
467         /**
468          * XXX: These are accessed without holding
469          * ptlrpc_service_part::scp_req_lock.
470          */
471         info->pi_req_queued  = policy->pol_req_queued;
472         info->pi_req_started = policy->pol_req_started;
473 }
474
475 /**
476  * Reads and prints policy status information for all policies of a PTLRPC
477  * service.
478  */
479 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
480 {
481         struct ptlrpc_service *svc = m->private;
482         struct ptlrpc_service_part *svcpt;
483         struct ptlrpc_nrs *nrs;
484         struct ptlrpc_nrs_policy *policy;
485         struct ptlrpc_nrs_pol_info *infos;
486         struct ptlrpc_nrs_pol_info tmp;
487         unsigned num_pols;
488         unsigned pol_idx = 0;
489         bool hp = false;
490         int i;
491         int rc = 0;
492
493         /**
494          * Serialize NRS core lprocfs operations with policy registration/
495          * unregistration.
496          */
497         mutex_lock(&nrs_core.nrs_mutex);
498
499         /**
500          * Use the first service partition's regular NRS head in order to obtain
501          * the number of policies registered with NRS heads of this service. All
502          * service partitions will have the same number of policies.
503          */
504         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
505
506         spin_lock(&nrs->nrs_lock);
507         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
508         spin_unlock(&nrs->nrs_lock);
509
510         infos = kcalloc(num_pols, sizeof(*infos), GFP_NOFS);
511         if (infos == NULL) {
512                 rc = -ENOMEM;
513                 goto unlock;
514         }
515 again:
516
517         ptlrpc_service_for_each_part(svcpt, i, svc) {
518                 nrs = nrs_svcpt2nrs(svcpt, hp);
519                 spin_lock(&nrs->nrs_lock);
520
521                 pol_idx = 0;
522
523                 list_for_each_entry(policy, &nrs->nrs_policy_list,
524                                         pol_list) {
525                         LASSERT(pol_idx < num_pols);
526
527                         nrs_policy_get_info_locked(policy, &tmp);
528                         /**
529                          * Copy values when handling the first service
530                          * partition.
531                          */
532                         if (i == 0) {
533                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
534                                        NRS_POL_NAME_MAX);
535                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
536                                        sizeof(tmp.pi_state));
537                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
538                                 /**
539                                  * For the rest of the service partitions
540                                  * sanity-check the values we get.
541                                  */
542                         } else {
543                                 LASSERT(strncmp(infos[pol_idx].pi_name,
544                                                 tmp.pi_name,
545                                                 NRS_POL_NAME_MAX) == 0);
546                                 /**
547                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
548                                  * because it may be different between
549                                  * instances of the same policy in different
550                                  * service partitions.
551                                  */
552                                 LASSERT(infos[pol_idx].pi_fallback ==
553                                         tmp.pi_fallback);
554                         }
555
556                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
557                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
558
559                         pol_idx++;
560                 }
561                 spin_unlock(&nrs->nrs_lock);
562         }
563
564         /**
565          * Policy status information output is in YAML format.
566          * For example:
567          *
568          *      regular_requests:
569          *        - name: fifo
570          *          state: started
571          *          fallback: yes
572          *          queued: 0
573          *          active: 0
574          *
575          *        - name: crrn
576          *          state: started
577          *          fallback: no
578          *          queued: 2015
579          *          active: 384
580          *
581          *      high_priority_requests:
582          *        - name: fifo
583          *          state: started
584          *          fallback: yes
585          *          queued: 0
586          *          active: 2
587          *
588          *        - name: crrn
589          *          state: stopped
590          *          fallback: no
591          *          queued: 0
592          *          active: 0
593          */
594         seq_printf(m, "%s\n",
595                       !hp ?  "\nregular_requests:" : "high_priority_requests:");
596
597         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
598                 seq_printf(m,  "  - name: %s\n"
599                                "    state: %s\n"
600                                "    fallback: %s\n"
601                                "    queued: %-20d\n"
602                                "    active: %-20d\n\n",
603                                infos[pol_idx].pi_name,
604                                nrs_state2str(infos[pol_idx].pi_state),
605                                infos[pol_idx].pi_fallback ? "yes" : "no",
606                                (int)infos[pol_idx].pi_req_queued,
607                                (int)infos[pol_idx].pi_req_started);
608         }
609
610         if (!hp && nrs_svc_has_hp(svc)) {
611                 memset(infos, 0, num_pols * sizeof(*infos));
612
613                 /**
614                  * Redo the processing for the service's HP NRS heads' policies.
615                  */
616                 hp = true;
617                 goto again;
618         }
619
620         kfree(infos);
621 unlock:
622         mutex_unlock(&nrs_core.nrs_mutex);
623
624         return rc;
625 }
626
627 /**
628  * The longest valid command string is the maximum policy name size, plus the
629  * length of the " reg" substring
630  */
631 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1)
632
633 /**
634  * Starts and stops a given policy on a PTLRPC service.
635  *
636  * Commands consist of the policy name, followed by an optional [reg|hp] token;
637  * if the optional token is omitted, the operation is performed on both the
638  * regular and high-priority (if the service has one) NRS head.
639  */
640 static ssize_t ptlrpc_lprocfs_nrs_seq_write(struct file *file,
641                                             const char __user *buffer,
642                                             size_t count, loff_t *off)
643 {
644         struct ptlrpc_service *svc = ((struct seq_file *)file->private_data)->private;
645         enum ptlrpc_nrs_queue_type queue = PTLRPC_NRS_QUEUE_BOTH;
646         char *cmd;
647         char *cmd_copy = NULL;
648         char *token;
649         int rc = 0;
650
651         if (count >= LPROCFS_NRS_WR_MAX_CMD)
652                 return -EINVAL;
653
654         cmd = kzalloc(LPROCFS_NRS_WR_MAX_CMD, GFP_NOFS);
655         if (!cmd)
656                 return -ENOMEM;
657         /**
658          * strsep() modifies its argument, so keep a copy
659          */
660         cmd_copy = cmd;
661
662         if (copy_from_user(cmd, buffer, count)) {
663                 rc = -EFAULT;
664                 goto out;
665         }
666
667         cmd[count] = '\0';
668
669         token = strsep(&cmd, " ");
670
671         if (strlen(token) > NRS_POL_NAME_MAX - 1) {
672                 rc = -EINVAL;
673                 goto out;
674         }
675
676         /**
677          * No [reg|hp] token has been specified
678          */
679         if (cmd == NULL)
680                 goto default_queue;
681
682         /**
683          * The second token is either NULL, or an optional [reg|hp] string
684          */
685         if (strcmp(cmd, "reg") == 0)
686                 queue = PTLRPC_NRS_QUEUE_REG;
687         else if (strcmp(cmd, "hp") == 0)
688                 queue = PTLRPC_NRS_QUEUE_HP;
689         else {
690                 rc = -EINVAL;
691                 goto out;
692         }
693
694 default_queue:
695
696         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc)) {
697                 rc = -ENODEV;
698                 goto out;
699         } else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
700                 queue = PTLRPC_NRS_QUEUE_REG;
701
702         /**
703          * Serialize NRS core lprocfs operations with policy registration/
704          * unregistration.
705          */
706         mutex_lock(&nrs_core.nrs_mutex);
707
708         rc = ptlrpc_nrs_policy_control(svc, queue, token, PTLRPC_NRS_CTL_START,
709                                        false, NULL);
710
711         mutex_unlock(&nrs_core.nrs_mutex);
712 out:
713         kfree(cmd_copy);
714
715         return rc < 0 ? rc : count;
716 }
717
718 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
719
720 /** @} nrs */
721
722 struct ptlrpc_srh_iterator {
723         int                     srhi_idx;
724         __u64                   srhi_seq;
725         struct ptlrpc_request   *srhi_req;
726 };
727
728 static int
729 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
730                                     struct ptlrpc_srh_iterator *srhi,
731                                     __u64 seq)
732 {
733         struct list_head *e;
734         struct ptlrpc_request *req;
735
736         if (srhi->srhi_req != NULL &&
737             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
738             srhi->srhi_seq <= seq) {
739                 /* If srhi_req was set previously, hasn't been culled and
740                  * we're searching for a seq on or after it (i.e. more
741                  * recent), search from it onwards.
742                  * Since the service history is LRU (i.e. culled reqs will
743                  * be near the head), we shouldn't have to do long
744                  * re-scans */
745                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
746                          "%s:%d: seek seq %llu, request seq %llu\n",
747                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
748                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
749                 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
750                          "%s:%d: seek offset %llu, request seq %llu, last culled %llu\n",
751                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
752                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
753                 e = &srhi->srhi_req->rq_history_list;
754         } else {
755                 /* search from start */
756                 e = svcpt->scp_hist_reqs.next;
757         }
758
759         while (e != &svcpt->scp_hist_reqs) {
760                 req = list_entry(e, struct ptlrpc_request, rq_history_list);
761
762                 if (req->rq_history_seq >= seq) {
763                         srhi->srhi_seq = req->rq_history_seq;
764                         srhi->srhi_req = req;
765                         return 0;
766                 }
767                 e = e->next;
768         }
769
770         return -ENOENT;
771 }
772
773 /*
774  * ptlrpc history sequence is used as "position" of seq_file, in some case,
775  * seq_read() will increase "position" to indicate reading the next
776  * element, however, low bits of history sequence are reserved for CPT id
777  * (check the details from comments before ptlrpc_req_add_history), which
778  * means seq_read() might change CPT id of history sequence and never
779  * finish reading of requests on a CPT. To make it work, we have to shift
780  * CPT id to high bits and timestamp to low bits, so seq_read() will only
781  * increase timestamp which can correctly indicate the next position.
782  */
783
784 /* convert seq_file pos to cpt */
785 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
786         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
787          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
788
789 /* make up seq_file pos from cpt */
790 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
791         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
792          (cpt) << (64 - (svc)->srv_cpt_bits))
793
794 /* convert sequence to position */
795 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
796         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
797          ((seq) >> (svc)->srv_cpt_bits) |               \
798          ((seq) << (64 - (svc)->srv_cpt_bits)))
799
800 /* convert position to sequence */
801 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
802         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
803          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
804          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
805
806 static void *
807 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
808 {
809         struct ptlrpc_service           *svc = s->private;
810         struct ptlrpc_service_part      *svcpt;
811         struct ptlrpc_srh_iterator      *srhi;
812         unsigned int                    cpt;
813         int                             rc;
814         int                             i;
815
816         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
817                 CWARN("Failed to read request history because size of loff_t %d can't match size of u64\n",
818                       (int)sizeof(loff_t));
819                 return NULL;
820         }
821
822         srhi = kzalloc(sizeof(*srhi), GFP_NOFS);
823         if (!srhi)
824                 return NULL;
825
826         srhi->srhi_seq = 0;
827         srhi->srhi_req = NULL;
828
829         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
830
831         ptlrpc_service_for_each_part(svcpt, i, svc) {
832                 if (i < cpt) /* skip */
833                         continue;
834                 if (i > cpt) /* make up the lowest position for this CPT */
835                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
836
837                 spin_lock(&svcpt->scp_lock);
838                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
839                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
840                 spin_unlock(&svcpt->scp_lock);
841                 if (rc == 0) {
842                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
843                         srhi->srhi_idx = i;
844                         return srhi;
845                 }
846         }
847
848         kfree(srhi);
849         return NULL;
850 }
851
852 static void
853 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
854 {
855         struct ptlrpc_srh_iterator *srhi = iter;
856
857         kfree(srhi);
858 }
859
860 static void *
861 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
862                                     void *iter, loff_t *pos)
863 {
864         struct ptlrpc_service *svc = s->private;
865         struct ptlrpc_srh_iterator *srhi = iter;
866         struct ptlrpc_service_part *svcpt;
867         __u64 seq;
868         int rc;
869         int i;
870
871         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
872                 svcpt = svc->srv_parts[i];
873
874                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
875                         srhi->srhi_req = NULL;
876                         seq = srhi->srhi_seq = 0;
877                 } else { /* the next sequence */
878                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
879                 }
880
881                 spin_lock(&svcpt->scp_lock);
882                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
883                 spin_unlock(&svcpt->scp_lock);
884                 if (rc == 0) {
885                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
886                         srhi->srhi_idx = i;
887                         return srhi;
888                 }
889         }
890
891         kfree(srhi);
892         return NULL;
893 }
894
895 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
896 {
897         struct ptlrpc_service *svc = s->private;
898         struct ptlrpc_srh_iterator *srhi = iter;
899         struct ptlrpc_service_part *svcpt;
900         struct ptlrpc_request *req;
901         int rc;
902
903         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
904
905         svcpt = svc->srv_parts[srhi->srhi_idx];
906
907         spin_lock(&svcpt->scp_lock);
908
909         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
910
911         if (rc == 0) {
912                 char nidstr[LNET_NIDSTR_SIZE];
913
914                 req = srhi->srhi_req;
915
916                 libcfs_nid2str_r(req->rq_self, nidstr, sizeof(nidstr));
917                 /* Print common req fields.
918                  * CAVEAT EMPTOR: we're racing with the service handler
919                  * here.  The request could contain any old crap, so you
920                  * must be just as careful as the service's request
921                  * parser. Currently I only print stuff here I know is OK
922                  * to look at coz it was set up in request_in_callback()!!! */
923                 seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:%lld:%lds(%+lds) ",
924                            req->rq_history_seq, nidstr,
925                            libcfs_id2str(req->rq_peer), req->rq_xid,
926                            req->rq_reqlen, ptlrpc_rqphase2str(req),
927                            (s64)req->rq_arrival_time.tv_sec,
928                            (long)(req->rq_sent - req->rq_arrival_time.tv_sec),
929                            (long)(req->rq_sent - req->rq_deadline));
930                 if (svc->srv_ops.so_req_printer == NULL)
931                         seq_putc(s, '\n');
932                 else
933                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
934         }
935
936         spin_unlock(&svcpt->scp_lock);
937         return rc;
938 }
939
940 static int
941 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
942 {
943         static struct seq_operations sops = {
944                 .start = ptlrpc_lprocfs_svc_req_history_start,
945                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
946                 .next  = ptlrpc_lprocfs_svc_req_history_next,
947                 .show  = ptlrpc_lprocfs_svc_req_history_show,
948         };
949         struct seq_file *seqf;
950         int rc;
951
952         rc = seq_open(file, &sops);
953         if (rc)
954                 return rc;
955
956         seqf = file->private_data;
957         seqf->private = inode->i_private;
958         return 0;
959 }
960
961 /* See also lprocfs_rd_timeouts */
962 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
963 {
964         struct ptlrpc_service *svc = m->private;
965         struct ptlrpc_service_part *svcpt;
966         struct dhms ts;
967         time64_t worstt;
968         unsigned int cur;
969         unsigned int worst;
970         int i;
971
972         if (AT_OFF) {
973                 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
974                                obd_timeout);
975                 return 0;
976         }
977
978         ptlrpc_service_for_each_part(svcpt, i, svc) {
979                 cur     = at_get(&svcpt->scp_at_estimate);
980                 worst   = svcpt->scp_at_estimate.at_worst_ever;
981                 worstt  = svcpt->scp_at_estimate.at_worst_time;
982                 s2dhms(&ts, ktime_get_real_seconds() - worstt);
983
984                 seq_printf(m, "%10s : cur %3u  worst %3u (at %lld, "
985                               DHMS_FMT" ago) ", "service",
986                               cur, worst, (s64)worstt, DHMS_VARS(&ts));
987
988                 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
989         }
990
991         return 0;
992 }
993
994 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
995
996 static ssize_t high_priority_ratio_show(struct kobject *kobj,
997                                         struct attribute *attr,
998                                         char *buf)
999 {
1000         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1001                                                   srv_kobj);
1002         return sprintf(buf, "%d\n", svc->srv_hpreq_ratio);
1003 }
1004
1005 static ssize_t high_priority_ratio_store(struct kobject *kobj,
1006                                          struct attribute *attr,
1007                                          const char *buffer,
1008                                          size_t count)
1009 {
1010         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1011                                                   srv_kobj);
1012         int rc;
1013         int val;
1014
1015         rc = kstrtoint(buffer, 10, &val);
1016         if (rc < 0)
1017                 return rc;
1018
1019         if (val < 0)
1020                 return -ERANGE;
1021
1022         spin_lock(&svc->srv_lock);
1023         svc->srv_hpreq_ratio = val;
1024         spin_unlock(&svc->srv_lock);
1025
1026         return count;
1027 }
1028 LUSTRE_RW_ATTR(high_priority_ratio);
1029
1030 static struct attribute *ptlrpc_svc_attrs[] = {
1031         &lustre_attr_threads_min.attr,
1032         &lustre_attr_threads_started.attr,
1033         &lustre_attr_threads_max.attr,
1034         &lustre_attr_high_priority_ratio.attr,
1035         NULL,
1036 };
1037
1038 static void ptlrpc_sysfs_svc_release(struct kobject *kobj)
1039 {
1040         struct ptlrpc_service *svc = container_of(kobj, struct ptlrpc_service,
1041                                                   srv_kobj);
1042
1043         complete(&svc->srv_kobj_unregister);
1044 }
1045
1046 static struct kobj_type ptlrpc_svc_ktype = {
1047         .default_attrs  = ptlrpc_svc_attrs,
1048         .sysfs_ops      = &lustre_sysfs_ops,
1049         .release        = ptlrpc_sysfs_svc_release,
1050 };
1051
1052 void ptlrpc_sysfs_unregister_service(struct ptlrpc_service *svc)
1053 {
1054         /* Let's see if we had a chance at initialization first */
1055         if (svc->srv_kobj.kset) {
1056                 kobject_put(&svc->srv_kobj);
1057                 wait_for_completion(&svc->srv_kobj_unregister);
1058         }
1059 }
1060
1061 int ptlrpc_sysfs_register_service(struct kset *parent,
1062                                   struct ptlrpc_service *svc)
1063 {
1064         int rc;
1065
1066         svc->srv_kobj.kset = parent;
1067         init_completion(&svc->srv_kobj_unregister);
1068         rc = kobject_init_and_add(&svc->srv_kobj, &ptlrpc_svc_ktype, NULL,
1069                                   "%s", svc->srv_name);
1070
1071         return rc;
1072 }
1073
1074 void ptlrpc_ldebugfs_register_service(struct dentry *entry,
1075                                       struct ptlrpc_service *svc)
1076 {
1077         struct lprocfs_vars lproc_vars[] = {
1078                 {.name       = "req_buffer_history_len",
1079                  .fops       = &ptlrpc_lprocfs_req_history_len_fops,
1080                  .data       = svc},
1081                 {.name       = "req_buffer_history_max",
1082                  .fops       = &ptlrpc_lprocfs_req_history_max_fops,
1083                  .data       = svc},
1084                 {.name       = "timeouts",
1085                  .fops       = &ptlrpc_lprocfs_timeouts_fops,
1086                  .data       = svc},
1087                 {.name       = "nrs_policies",
1088                  .fops       = &ptlrpc_lprocfs_nrs_fops,
1089                  .data       = svc},
1090                 {NULL}
1091         };
1092         static const struct file_operations req_history_fops = {
1093                 .owner       = THIS_MODULE,
1094                 .open   = ptlrpc_lprocfs_svc_req_history_open,
1095                 .read   = seq_read,
1096                 .llseek      = seq_lseek,
1097                 .release     = lprocfs_seq_release,
1098         };
1099
1100         int rc;
1101
1102         ptlrpc_ldebugfs_register(entry, svc->srv_name,
1103                                  "stats", &svc->srv_debugfs_entry,
1104                                  &svc->srv_stats);
1105
1106         if (svc->srv_debugfs_entry == NULL)
1107                 return;
1108
1109         ldebugfs_add_vars(svc->srv_debugfs_entry, lproc_vars, NULL);
1110
1111         rc = ldebugfs_seq_create(svc->srv_debugfs_entry, "req_history",
1112                                  0400, &req_history_fops, svc);
1113         if (rc)
1114                 CWARN("Error adding the req_history file\n");
1115 }
1116
1117 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1118 {
1119         ptlrpc_ldebugfs_register(obddev->obd_debugfs_entry, NULL, "stats",
1120                                  &obddev->obd_svc_debugfs_entry,
1121                                  &obddev->obd_svc_stats);
1122 }
1123 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1124
1125 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1126 {
1127         struct lprocfs_stats *svc_stats;
1128         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1129         int opc = opcode_offset(op);
1130
1131         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1132         if (svc_stats == NULL || opc <= 0)
1133                 return;
1134         LASSERT(opc < LUSTRE_MAX_OPCODES);
1135         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1136                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1137 }
1138
1139 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1140 {
1141         struct lprocfs_stats *svc_stats;
1142         int idx;
1143
1144         if (!req->rq_import)
1145                 return;
1146         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1147         if (!svc_stats)
1148                 return;
1149         idx = lustre_msg_get_opc(req->rq_reqmsg);
1150         switch (idx) {
1151         case OST_READ:
1152                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1153                 break;
1154         case OST_WRITE:
1155                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1156                 break;
1157         default:
1158                 LASSERTF(0, "unsupported opcode %u\n", idx);
1159                 break;
1160         }
1161
1162         lprocfs_counter_add(svc_stats, idx, bytes);
1163 }
1164
1165 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1166
1167 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1168 {
1169         if (svc->srv_debugfs_entry != NULL)
1170                 ldebugfs_remove(&svc->srv_debugfs_entry);
1171
1172         if (svc->srv_stats)
1173                 lprocfs_free_stats(&svc->srv_stats);
1174 }
1175
1176 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1177 {
1178         if (!IS_ERR_OR_NULL(obd->obd_svc_debugfs_entry))
1179                 ldebugfs_remove(&obd->obd_svc_debugfs_entry);
1180
1181         if (obd->obd_svc_stats)
1182                 lprocfs_free_stats(&obd->obd_svc_stats);
1183 }
1184 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1185
1186 #undef BUFLEN
1187
1188 int lprocfs_wr_ping(struct file *file, const char __user *buffer,
1189                     size_t count, loff_t *off)
1190 {
1191         struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1192         struct ptlrpc_request *req;
1193         int rc;
1194
1195         rc = lprocfs_climp_check(obd);
1196         if (rc)
1197                 return rc;
1198
1199         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1200         LPROCFS_CLIMP_EXIT(obd);
1201         if (req == NULL)
1202                 return -ENOMEM;
1203
1204         req->rq_send_state = LUSTRE_IMP_FULL;
1205
1206         rc = ptlrpc_queue_wait(req);
1207
1208         ptlrpc_req_finished(req);
1209         if (rc >= 0)
1210                 return count;
1211         return rc;
1212 }
1213 EXPORT_SYMBOL(lprocfs_wr_ping);
1214
1215 /* Write the connection UUID to this file to attempt to connect to that node.
1216  * The connection UUID is a node's primary NID. For example,
1217  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1218  */
1219 int lprocfs_wr_import(struct file *file, const char __user *buffer,
1220                       size_t count, loff_t *off)
1221 {
1222         struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1223         struct obd_import *imp = obd->u.cli.cl_import;
1224         char *kbuf = NULL;
1225         char *uuid;
1226         char *ptr;
1227         int do_reconn = 1;
1228         const char prefix[] = "connection=";
1229         const int prefix_len = sizeof(prefix) - 1;
1230
1231         if (count > PAGE_CACHE_SIZE - 1 || count <= prefix_len)
1232                 return -EINVAL;
1233
1234         kbuf = kzalloc(count + 1, GFP_NOFS);
1235         if (!kbuf)
1236                 return -ENOMEM;
1237
1238         if (copy_from_user(kbuf, buffer, count)) {
1239                 count = -EFAULT;
1240                 goto out;
1241         }
1242
1243         kbuf[count] = 0;
1244
1245         /* only support connection=uuid::instance now */
1246         if (strncmp(prefix, kbuf, prefix_len) != 0) {
1247                 count = -EINVAL;
1248                 goto out;
1249         }
1250
1251         uuid = kbuf + prefix_len;
1252         ptr = strstr(uuid, "::");
1253         if (ptr) {
1254                 __u32 inst;
1255                 char *endptr;
1256
1257                 *ptr = 0;
1258                 do_reconn = 0;
1259                 ptr += strlen("::");
1260                 inst = simple_strtoul(ptr, &endptr, 10);
1261                 if (*endptr) {
1262                         CERROR("config: wrong instance # %s\n", ptr);
1263                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1264                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted target(%u/%u), reconnecting...\n",
1265                                imp->imp_obd->obd_name,
1266                                imp->imp_connect_data.ocd_instance, inst);
1267                         do_reconn = 1;
1268                 } else {
1269                         CDEBUG(D_INFO, "IR: %s has already been connecting to new target(%u)\n",
1270                                imp->imp_obd->obd_name, inst);
1271                 }
1272         }
1273
1274         if (do_reconn)
1275                 ptlrpc_recover_import(imp, uuid, 1);
1276
1277 out:
1278         kfree(kbuf);
1279         return count;
1280 }
1281 EXPORT_SYMBOL(lprocfs_wr_import);
1282
1283 int lprocfs_rd_pinger_recov(struct seq_file *m, void *n)
1284 {
1285         struct obd_device *obd = m->private;
1286         struct obd_import *imp = obd->u.cli.cl_import;
1287         int rc;
1288
1289         rc = lprocfs_climp_check(obd);
1290         if (rc)
1291                 return rc;
1292
1293         seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1294         LPROCFS_CLIMP_EXIT(obd);
1295
1296         return 0;
1297 }
1298 EXPORT_SYMBOL(lprocfs_rd_pinger_recov);
1299
1300 int lprocfs_wr_pinger_recov(struct file *file, const char __user *buffer,
1301                       size_t count, loff_t *off)
1302 {
1303         struct obd_device *obd = ((struct seq_file *)file->private_data)->private;
1304         struct client_obd *cli = &obd->u.cli;
1305         struct obd_import *imp = cli->cl_import;
1306         int rc, val;
1307
1308         rc = lprocfs_write_helper(buffer, count, &val);
1309         if (rc < 0)
1310                 return rc;
1311
1312         if (val != 0 && val != 1)
1313                 return -ERANGE;
1314
1315         rc = lprocfs_climp_check(obd);
1316         if (rc)
1317                 return rc;
1318
1319         spin_lock(&imp->imp_lock);
1320         imp->imp_no_pinger_recover = !val;
1321         spin_unlock(&imp->imp_lock);
1322         LPROCFS_CLIMP_EXIT(obd);
1323
1324         return count;
1325
1326 }
1327 EXPORT_SYMBOL(lprocfs_wr_pinger_recov);