Whamcloud - gitweb
2ea931e2f8a2348e9521e72c2246d5f4fe0c0345
[fs/lustre-release.git] / lustre / obdclass / lprocfs_status.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) 2011, 2017, Intel Corporation.
27  */
28 /*
29  * This file is part of Lustre, http://www.lustre.org/
30  * Lustre is a trademark of Sun Microsystems, Inc.
31  *
32  * lustre/obdclass/lprocfs_status.c
33  *
34  * Author: Hariharan Thantry <thantry@users.sourceforge.net>
35  */
36
37 #define DEBUG_SUBSYSTEM S_CLASS
38
39 #include <obd_class.h>
40 #include <lprocfs_status.h>
41
42 #ifdef CONFIG_PROC_FS
43
44 static int lprocfs_no_percpu_stats = 0;
45 module_param(lprocfs_no_percpu_stats, int, 0644);
46 MODULE_PARM_DESC(lprocfs_no_percpu_stats, "Do not alloc percpu data for lprocfs stats");
47
48 #define MAX_STRING_SIZE 128
49
50 int lprocfs_single_release(struct inode *inode, struct file *file)
51 {
52         return single_release(inode, file);
53 }
54 EXPORT_SYMBOL(lprocfs_single_release);
55
56 int lprocfs_seq_release(struct inode *inode, struct file *file)
57 {
58         return seq_release(inode, file);
59 }
60 EXPORT_SYMBOL(lprocfs_seq_release);
61
62 struct dentry *ldebugfs_add_simple(struct dentry *root,
63                                    char *name, void *data,
64                                    const struct file_operations *fops)
65 {
66         struct dentry *entry;
67         umode_t mode = 0;
68
69         if (!root || !name || !fops)
70                 return ERR_PTR(-EINVAL);
71
72         if (fops->read)
73                 mode = 0444;
74         if (fops->write)
75                 mode |= 0200;
76         entry = debugfs_create_file(name, mode, root, data, fops);
77         if (IS_ERR_OR_NULL(entry)) {
78                 CERROR("LprocFS: No memory to create <debugfs> entry %s", name);
79                 return entry ?: ERR_PTR(-ENOMEM);
80         }
81         return entry;
82 }
83 EXPORT_SYMBOL(ldebugfs_add_simple);
84
85 struct proc_dir_entry *
86 lprocfs_add_simple(struct proc_dir_entry *root, char *name,
87                    void *data, const struct file_operations *fops)
88 {
89         struct proc_dir_entry *proc;
90         mode_t mode = 0;
91
92         if (!root || !name || !fops)
93                 return ERR_PTR(-EINVAL);
94
95         if (fops->read)
96                 mode = 0444;
97         if (fops->write)
98                 mode |= 0200;
99         proc = proc_create_data(name, mode, root, fops, data);
100         if (!proc) {
101                 CERROR("LprocFS: No memory to create /proc entry %s\n",
102                        name);
103                 return ERR_PTR(-ENOMEM);
104         }
105         return proc;
106 }
107 EXPORT_SYMBOL(lprocfs_add_simple);
108
109 struct proc_dir_entry *lprocfs_add_symlink(const char *name,
110                                            struct proc_dir_entry *parent,
111                                            const char *format, ...)
112 {
113         struct proc_dir_entry *entry;
114         char *dest;
115         va_list ap;
116
117         if (!parent || !format)
118                 return NULL;
119
120         OBD_ALLOC_WAIT(dest, MAX_STRING_SIZE + 1);
121         if (!dest)
122                 return NULL;
123
124         va_start(ap, format);
125         vsnprintf(dest, MAX_STRING_SIZE, format, ap);
126         va_end(ap);
127
128         entry = proc_symlink(name, parent, dest);
129         if (!entry)
130                 CERROR("LprocFS: Could not create symbolic link from "
131                        "%s to %s\n", name, dest);
132
133         OBD_FREE(dest, MAX_STRING_SIZE + 1);
134         return entry;
135 }
136 EXPORT_SYMBOL(lprocfs_add_symlink);
137
138 static const struct file_operations lprocfs_generic_fops = { };
139
140 int ldebugfs_add_vars(struct dentry *parent, struct lprocfs_vars *list,
141                       void *data)
142 {
143         if (IS_ERR_OR_NULL(parent) || IS_ERR_OR_NULL(list))
144                 return -EINVAL;
145
146         while (list->name) {
147                 struct dentry *entry;
148                 umode_t mode = 0;
149
150                 if (list->proc_mode != 0000) {
151                         mode = list->proc_mode;
152                 } else if (list->fops) {
153                         if (list->fops->read)
154                                 mode = 0444;
155                         if (list->fops->write)
156                                 mode |= 0200;
157                 }
158                 entry = debugfs_create_file(list->name, mode, parent,
159                                             list->data ? : data,
160                                             list->fops ? : &lprocfs_generic_fops);
161                 if (IS_ERR_OR_NULL(entry))
162                         return entry ? PTR_ERR(entry) : -ENOMEM;
163                 list++;
164         }
165         return 0;
166 }
167 EXPORT_SYMBOL_GPL(ldebugfs_add_vars);
168
169 /**
170  * Add /proc entries.
171  *
172  * \param root [in]  The parent proc entry on which new entry will be added.
173  * \param list [in]  Array of proc entries to be added.
174  * \param data [in]  The argument to be passed when entries read/write routines
175  *                   are called through /proc file.
176  *
177  * \retval 0   on success
178  *         < 0 on error
179  */
180 int
181 lprocfs_add_vars(struct proc_dir_entry *root, struct lprocfs_vars *list,
182                  void *data)
183 {
184         if (!root || !list)
185                 return -EINVAL;
186
187         while (list->name) {
188                 struct proc_dir_entry *proc;
189                 mode_t mode = 0;
190
191                 if (list->proc_mode != 0000) {
192                         mode = list->proc_mode;
193                 } else if (list->fops) {
194                         if (list->fops->read)
195                                 mode = 0444;
196                         if (list->fops->write)
197                                 mode |= 0200;
198                 }
199                 proc = proc_create_data(list->name, mode, root,
200                                         list->fops ?: &lprocfs_generic_fops,
201                                         list->data ?: data);
202                 if (!proc)
203                         return -ENOMEM;
204                 list++;
205         }
206         return 0;
207 }
208 EXPORT_SYMBOL(lprocfs_add_vars);
209
210 void ldebugfs_remove(struct dentry **entryp)
211 {
212         debugfs_remove_recursive(*entryp);
213         *entryp = NULL;
214 }
215 EXPORT_SYMBOL_GPL(ldebugfs_remove);
216
217 #ifndef HAVE_REMOVE_PROC_SUBTREE
218 /* for b=10866, global variable */
219 DECLARE_RWSEM(_lprocfs_lock);
220 EXPORT_SYMBOL(_lprocfs_lock);
221
222 static void lprocfs_remove_nolock(struct proc_dir_entry **proot)
223 {
224         struct proc_dir_entry *root = *proot;
225         struct proc_dir_entry *temp = root;
226         struct proc_dir_entry *rm_entry;
227         struct proc_dir_entry *parent;
228
229         *proot = NULL;
230         if (!root || IS_ERR(root))
231                 return;
232
233         parent = root->parent;
234         LASSERT(parent != NULL);
235
236         while (1) {
237                 while (temp->subdir)
238                         temp = temp->subdir;
239
240                 rm_entry = temp;
241                 temp = temp->parent;
242
243                 /*
244                  * Memory corruption once caused this to fail, and
245                  * without this LASSERT we would loop here forever.
246                  */
247                 LASSERTF(strlen(rm_entry->name) == rm_entry->namelen,
248                          "0x%p  %s/%s len %d\n", rm_entry, temp->name,
249                          rm_entry->name, (int)strlen(rm_entry->name));
250
251                 remove_proc_entry(rm_entry->name, temp);
252                 if (temp == parent)
253                         break;
254         }
255 }
256
257 int remove_proc_subtree(const char *name, struct proc_dir_entry *parent)
258 {
259         struct proc_dir_entry *t = NULL;
260         struct proc_dir_entry **p;
261         int len, busy = 0;
262
263         LASSERT(parent != NULL);
264         len = strlen(name);
265
266         down_write(&_lprocfs_lock);
267         /* lookup target name */
268         for (p = &parent->subdir; *p; p = &(*p)->next) {
269                 if ((*p)->namelen != len)
270                         continue;
271                 if (memcmp(name, (*p)->name, len))
272                         continue;
273                 t = *p;
274                 break;
275         }
276
277         if (t) {
278                 /* verify it's empty: do not count "num_refs" */
279                 for (p = &t->subdir; *p; p = &(*p)->next) {
280                         if ((*p)->namelen != strlen("num_refs")) {
281                                 busy = 1;
282                                 break;
283                         }
284                         if (memcmp("num_refs", (*p)->name,
285                                    strlen("num_refs"))) {
286                                 busy = 1;
287                                 break;
288                         }
289                 }
290         }
291
292         if (busy == 0)
293                 lprocfs_remove_nolock(&t);
294
295         up_write(&_lprocfs_lock);
296         return 0;
297 }
298 #endif /* !HAVE_REMOVE_PROC_SUBTREE */
299
300 #ifndef HAVE_PROC_REMOVE
301 void proc_remove(struct proc_dir_entry *de)
302 {
303 #ifndef HAVE_REMOVE_PROC_SUBTREE
304         down_write(&_lprocfs_lock); /* search vs remove race */
305         lprocfs_remove_nolock(&de);
306         up_write(&_lprocfs_lock);
307 #else
308         if (de)
309                 remove_proc_subtree(de->name, de->parent);
310 #endif
311 }
312 #endif
313
314 void lprocfs_remove(struct proc_dir_entry **rooth)
315 {
316         proc_remove(*rooth);
317         *rooth = NULL;
318 }
319 EXPORT_SYMBOL(lprocfs_remove);
320
321 void lprocfs_remove_proc_entry(const char *name, struct proc_dir_entry *parent)
322 {
323         LASSERT(parent != NULL);
324         remove_proc_entry(name, parent);
325 }
326 EXPORT_SYMBOL(lprocfs_remove_proc_entry);
327
328 struct dentry *ldebugfs_register(const char *name, struct dentry *parent,
329                                  struct lprocfs_vars *list, void *data)
330 {
331         struct dentry *entry;
332
333         entry = debugfs_create_dir(name, parent);
334         if (IS_ERR_OR_NULL(entry)) {
335                 entry = entry ?: ERR_PTR(-ENOMEM);
336                 goto out;
337         }
338
339         if (!IS_ERR_OR_NULL(list)) {
340                 int rc;
341
342                 rc = ldebugfs_add_vars(entry, list, data);
343                 if (rc) {
344                         debugfs_remove(entry);
345                         entry = ERR_PTR(rc);
346                 }
347         }
348 out:
349         return entry;
350 }
351 EXPORT_SYMBOL_GPL(ldebugfs_register);
352
353 struct proc_dir_entry *
354 lprocfs_register(const char *name, struct proc_dir_entry *parent,
355                  struct lprocfs_vars *list, void *data)
356 {
357         struct proc_dir_entry *newchild;
358
359         newchild = proc_mkdir(name, parent);
360         if (!newchild)
361                 return ERR_PTR(-ENOMEM);
362
363         if (list) {
364                 int rc = lprocfs_add_vars(newchild, list, data);
365                 if (rc) {
366                         lprocfs_remove(&newchild);
367                         return ERR_PTR(rc);
368                 }
369         }
370         return newchild;
371 }
372 EXPORT_SYMBOL(lprocfs_register);
373
374 /* Generic callbacks */
375 int lprocfs_uuid_seq_show(struct seq_file *m, void *data)
376 {
377         struct obd_device *obd = data;
378
379         LASSERT(obd != NULL);
380         seq_printf(m, "%s\n", obd->obd_uuid.uuid);
381         return 0;
382 }
383 EXPORT_SYMBOL(lprocfs_uuid_seq_show);
384
385 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
386                          char *buf)
387 {
388         struct obd_device *obd = container_of(kobj, struct obd_device,
389                                               obd_kset.kobj);
390
391         return sprintf(buf, "%s\n", obd->obd_uuid.uuid);
392 }
393 LUSTRE_RO_ATTR(uuid);
394
395 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
396                               char *buf)
397 {
398         struct obd_device *obd = container_of(kobj, struct obd_device,
399                                               obd_kset.kobj);
400         struct obd_statfs osfs;
401         int rc;
402
403         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
404                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
405                         OBD_STATFS_NODELAY);
406         if (!rc)
407                 return sprintf(buf, "%u\n", osfs.os_bsize);
408
409         return rc;
410 }
411 LUSTRE_RO_ATTR(blocksize);
412
413 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
414                                 char *buf)
415 {
416         struct obd_device *obd = container_of(kobj, struct obd_device,
417                                               obd_kset.kobj);
418         struct obd_statfs osfs;
419         int rc;
420
421         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
422                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
423                         OBD_STATFS_NODELAY);
424         if (!rc) {
425                 u32 blk_size = osfs.os_bsize >> 10;
426                 u64 result = osfs.os_blocks;
427
428                 while (blk_size >>= 1)
429                         result <<= 1;
430
431                 return sprintf(buf, "%llu\n", result);
432         }
433
434         return rc;
435 }
436 LUSTRE_RO_ATTR(kbytestotal);
437
438 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
439                                char *buf)
440 {
441         struct obd_device *obd = container_of(kobj, struct obd_device,
442                                               obd_kset.kobj);
443         struct obd_statfs osfs;
444         int rc;
445
446         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
447                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
448                         OBD_STATFS_NODELAY);
449         if (!rc) {
450                 u32 blk_size = osfs.os_bsize >> 10;
451                 u64 result = osfs.os_bfree;
452
453                 while (blk_size >>= 1)
454                         result <<= 1;
455
456                 return sprintf(buf, "%llu\n", result);
457         }
458
459         return rc;
460 }
461 LUSTRE_RO_ATTR(kbytesfree);
462
463 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
464                                 char *buf)
465 {
466         struct obd_device *obd = container_of(kobj, struct obd_device,
467                                               obd_kset.kobj);
468         struct obd_statfs osfs;
469         int rc;
470
471         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
472                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
473                         OBD_STATFS_NODELAY);
474         if (!rc) {
475                 u32 blk_size = osfs.os_bsize >> 10;
476                 u64 result = osfs.os_bavail;
477
478                 while (blk_size >>= 1)
479                         result <<= 1;
480
481                 return sprintf(buf, "%llu\n", result);
482         }
483
484         return rc;
485 }
486 LUSTRE_RO_ATTR(kbytesavail);
487
488 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
489                                char *buf)
490 {
491         struct obd_device *obd = container_of(kobj, struct obd_device,
492                                               obd_kset.kobj);
493         struct obd_statfs osfs;
494         int rc;
495
496         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
497                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
498                         OBD_STATFS_NODELAY);
499         if (!rc)
500                 return sprintf(buf, "%llu\n", osfs.os_files);
501
502         return rc;
503 }
504 LUSTRE_RO_ATTR(filestotal);
505
506 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
507                               char *buf)
508 {
509         struct obd_device *obd = container_of(kobj, struct obd_device,
510                                               obd_kset.kobj);
511         struct obd_statfs osfs;
512         int rc;
513
514         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
515                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
516                         OBD_STATFS_NODELAY);
517         if (!rc)
518                 return sprintf(buf, "%llu\n", osfs.os_ffree);
519
520         return rc;
521 }
522 LUSTRE_RO_ATTR(filesfree);
523
524 ssize_t conn_uuid_show(struct kobject *kobj, struct attribute *attr, char *buf)
525 {
526         struct obd_device *obd = container_of(kobj, struct obd_device,
527                                               obd_kset.kobj);
528         struct ptlrpc_connection *conn;
529         ssize_t count;
530
531         LPROCFS_CLIMP_CHECK(obd);
532         conn = obd->u.cli.cl_import->imp_connection;
533         if (conn && obd->u.cli.cl_import)
534                 count = sprintf(buf, "%s\n", conn->c_remote_uuid.uuid);
535         else
536                 count = sprintf(buf, "%s\n", "<none>");
537
538         LPROCFS_CLIMP_EXIT(obd);
539         return count;
540 }
541 EXPORT_SYMBOL(conn_uuid_show);
542
543 int lprocfs_server_uuid_seq_show(struct seq_file *m, void *data)
544 {
545         struct obd_device *obd = data;
546         struct obd_import *imp;
547         char *imp_state_name = NULL;
548         int rc = 0;
549
550         LASSERT(obd != NULL);
551         LPROCFS_CLIMP_CHECK(obd);
552         imp = obd->u.cli.cl_import;
553         imp_state_name = ptlrpc_import_state_name(imp->imp_state);
554         seq_printf(m, "%s\t%s%s\n", obd2cli_tgt(obd), imp_state_name,
555                    imp->imp_deactive ? "\tDEACTIVATED" : "");
556
557         LPROCFS_CLIMP_EXIT(obd);
558         return rc;
559 }
560 EXPORT_SYMBOL(lprocfs_server_uuid_seq_show);
561
562 /** add up per-cpu counters */
563
564 /**
565  * Lock statistics structure for access, possibly only on this CPU.
566  *
567  * The statistics struct may be allocated with per-CPU structures for
568  * efficient concurrent update (usually only on server-wide stats), or
569  * as a single global struct (e.g. for per-client or per-job statistics),
570  * so the required locking depends on the type of structure allocated.
571  *
572  * For per-CPU statistics, pin the thread to the current cpuid so that
573  * will only access the statistics for that CPU.  If the stats structure
574  * for the current CPU has not been allocated (or previously freed),
575  * allocate it now.  The per-CPU statistics do not need locking since
576  * the thread is pinned to the CPU during update.
577  *
578  * For global statistics, lock the stats structure to prevent concurrent update.
579  *
580  * \param[in] stats     statistics structure to lock
581  * \param[in] opc       type of operation:
582  *                      LPROCFS_GET_SMP_ID: "lock" and return current CPU index
583  *                              for incrementing statistics for that CPU
584  *                      LPROCFS_GET_NUM_CPU: "lock" and return number of used
585  *                              CPU indices to iterate over all indices
586  * \param[out] flags    CPU interrupt saved state for IRQ-safe locking
587  *
588  * \retval cpuid of current thread or number of allocated structs
589  * \retval negative on error (only for opc LPROCFS_GET_SMP_ID + per-CPU stats)
590  */
591 int lprocfs_stats_lock(struct lprocfs_stats *stats,
592                        enum lprocfs_stats_lock_ops opc,
593                        unsigned long *flags)
594 {
595         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) {
596                 if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
597                         spin_lock_irqsave(&stats->ls_lock, *flags);
598                 else
599                         spin_lock(&stats->ls_lock);
600                 return opc == LPROCFS_GET_NUM_CPU ? 1 : 0;
601         }
602
603         switch (opc) {
604         case LPROCFS_GET_SMP_ID: {
605                 unsigned int cpuid = get_cpu();
606
607                 if (unlikely(!stats->ls_percpu[cpuid])) {
608                         int rc = lprocfs_stats_alloc_one(stats, cpuid);
609
610                         if (rc < 0) {
611                                 put_cpu();
612                                 return rc;
613                         }
614                 }
615                 return cpuid;
616         }
617         case LPROCFS_GET_NUM_CPU:
618                 return stats->ls_biggest_alloc_num;
619         default:
620                 LBUG();
621         }
622 }
623
624 /**
625  * Unlock statistics structure after access.
626  *
627  * Unlock the lock acquired via lprocfs_stats_lock() for global statistics,
628  * or unpin this thread from the current cpuid for per-CPU statistics.
629  *
630  * This function must be called using the same arguments as used when calling
631  * lprocfs_stats_lock() so that the correct operation can be performed.
632  *
633  * \param[in] stats     statistics structure to unlock
634  * \param[in] opc       type of operation (current cpuid or number of structs)
635  * \param[in] flags     CPU interrupt saved state for IRQ-safe locking
636  */
637 void lprocfs_stats_unlock(struct lprocfs_stats *stats,
638                           enum lprocfs_stats_lock_ops opc,
639                           unsigned long *flags)
640 {
641         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) {
642                 if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
643                         spin_unlock_irqrestore(&stats->ls_lock, *flags);
644                 else
645                         spin_unlock(&stats->ls_lock);
646         } else if (opc == LPROCFS_GET_SMP_ID) {
647                 put_cpu();
648         }
649 }
650
651 /** add up per-cpu counters */
652 void lprocfs_stats_collect(struct lprocfs_stats *stats, int idx,
653                            struct lprocfs_counter *cnt)
654 {
655         unsigned int num_entry;
656         struct lprocfs_counter *percpu_cntr;
657         int i;
658         unsigned long flags = 0;
659
660         memset(cnt, 0, sizeof(*cnt));
661
662         if (!stats) {
663                 /* set count to 1 to avoid divide-by-zero errs in callers */
664                 cnt->lc_count = 1;
665                 return;
666         }
667
668         cnt->lc_min = LC_MIN_INIT;
669
670         num_entry = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
671
672         for (i = 0; i < num_entry; i++) {
673                 if (!stats->ls_percpu[i])
674                         continue;
675                 percpu_cntr = lprocfs_stats_counter_get(stats, i, idx);
676
677                 cnt->lc_count += percpu_cntr->lc_count;
678                 cnt->lc_sum += percpu_cntr->lc_sum;
679                 if (percpu_cntr->lc_min < cnt->lc_min)
680                         cnt->lc_min = percpu_cntr->lc_min;
681                 if (percpu_cntr->lc_max > cnt->lc_max)
682                         cnt->lc_max = percpu_cntr->lc_max;
683                 cnt->lc_sumsquare += percpu_cntr->lc_sumsquare;
684         }
685
686         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
687 }
688
689 static void obd_import_flags2str(struct obd_import *imp, struct seq_file *m)
690 {
691         bool first = true;
692
693         if (imp->imp_obd->obd_no_recov) {
694                 seq_printf(m, "no_recov");
695                 first = false;
696         }
697
698         flag2str(imp, invalid);
699         flag2str(imp, deactive);
700         flag2str(imp, replayable);
701         flag2str(imp, delayed_recovery);
702         flag2str(imp, vbr_failed);
703         flag2str(imp, pingable);
704         flag2str(imp, resend_replay);
705         flag2str(imp, no_pinger_recover);
706         flag2str(imp, connect_tried);
707 }
708
709 static const char *obd_connect_names[] = {
710         /* flags names  */
711         "read_only",
712         "lov_index",
713         "connect_from_mds",
714         "write_grant",
715         "server_lock",
716         "version",
717         "request_portal",
718         "acl",
719         "xattr",
720         "create_on_write",
721         "truncate_lock",
722         "initial_transno",
723         "inode_bit_locks",
724         "barrier",
725         "getattr_by_fid",
726         "no_oh_for_devices",
727         "remote_client",
728         "remote_client_by_force",
729         "max_byte_per_rpc",
730         "64bit_qdata",
731         "mds_capability",
732         "oss_capability",
733         "early_lock_cancel",
734         "som",
735         "adaptive_timeouts",
736         "lru_resize",
737         "mds_mds_connection",
738         "real_conn",
739         "change_qunit_size",
740         "alt_checksum_algorithm",
741         "fid_is_enabled",
742         "version_recovery",
743         "pools",
744         "grant_shrink",
745         "skip_orphan",
746         "large_ea",
747         "full20",
748         "layout_lock",
749         "64bithash",
750         "object_max_bytes",
751         "imp_recov",
752         "jobstats",
753         "umask",
754         "einprogress",
755         "grant_param",
756         "flock_owner",
757         "lvb_type",
758         "nanoseconds_times",
759         "lightweight_conn",
760         "short_io",
761         "pingless",
762         "flock_deadlock",
763         "disp_stripe",
764         "open_by_fid",
765         "lfsck",
766         "unknown",
767         "unlink_close",
768         "multi_mod_rpcs",
769         "dir_stripe",
770         "subtree",
771         "lockahead",
772         "bulk_mbits",
773         "compact_obdo",
774         "second_flags",
775         /* flags2 names */
776         "file_secctx",  /* 0x01 */
777         "lockaheadv2",  /* 0x02 */
778         "dir_migrate",  /* 0x04 */
779         "unknown",      /* 0x08 */
780         "unknown",      /* 0x10 */
781         "flr",          /* 0x20 */
782         "wbc",          /* 0x40 */
783         "lock_convert",  /* 0x80 */
784         "archive_id_array",     /* 0x100 */
785         "unknown",              /* 0x200 */
786         "selinux_policy",       /* 0x400 */
787         "lsom",                 /* 0x800 */
788         "pcc",                  /* 0x1000 */
789         "plain_layout",         /* 0x2000 */
790         NULL
791 };
792
793 void obd_connect_seq_flags2str(struct seq_file *m, __u64 flags, __u64 flags2,
794                                const char *sep)
795 {
796         bool first = true;
797         __u64 mask;
798         int i;
799
800         for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
801                 if (flags & mask) {
802                         seq_printf(m, "%s%s",
803                                    first ? "" : sep, obd_connect_names[i]);
804                         first = false;
805                 }
806         }
807
808         if (flags & ~(mask - 1)) {
809                 seq_printf(m, "%sunknown_%#llx",
810                            first ? "" : sep, flags & ~(mask - 1));
811                 first = false;
812         }
813
814         if (!(flags & OBD_CONNECT_FLAGS2) || flags2 == 0)
815                 return;
816
817         for (i = 64, mask = 1; obd_connect_names[i] != NULL; i++, mask <<= 1) {
818                 if (flags2 & mask) {
819                         seq_printf(m, "%s%s",
820                                    first ? "" : sep, obd_connect_names[i]);
821                         first = false;
822                 }
823         }
824
825         if (flags2 & ~(mask - 1)) {
826                 seq_printf(m, "%sunknown2_%#llx",
827                            first ? "" : sep, flags2 & ~(mask - 1));
828                 first = false;
829         }
830 }
831 EXPORT_SYMBOL(obd_connect_seq_flags2str);
832
833 int obd_connect_flags2str(char *page, int count, __u64 flags, __u64 flags2,
834                           const char *sep)
835 {
836         __u64 mask;
837         int i, ret = 0;
838
839         for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
840                 if (flags & mask)
841                         ret += snprintf(page + ret, count - ret, "%s%s",
842                                         ret ? sep : "", obd_connect_names[i]);
843         }
844
845         if (flags & ~(mask - 1))
846                 ret += snprintf(page + ret, count - ret,
847                                 "%sunknown_%#llx",
848                                 ret ? sep : "", flags & ~(mask - 1));
849
850         if (!(flags & OBD_CONNECT_FLAGS2) || flags2 == 0)
851                 return ret;
852
853         for (i = 64, mask = 1; obd_connect_names[i] != NULL; i++, mask <<= 1) {
854                 if (flags2 & mask)
855                         ret += snprintf(page + ret, count - ret, "%s%s",
856                                         ret ? sep : "", obd_connect_names[i]);
857         }
858
859         if (flags2 & ~(mask - 1))
860                 ret += snprintf(page + ret, count - ret,
861                                 "%sunknown2_%#llx",
862                                 ret ? sep : "", flags2 & ~(mask - 1));
863
864         return ret;
865 }
866 EXPORT_SYMBOL(obd_connect_flags2str);
867
868 void
869 obd_connect_data_seqprint(struct seq_file *m, struct obd_connect_data *ocd)
870 {
871         __u64 flags;
872
873         LASSERT(ocd != NULL);
874         flags = ocd->ocd_connect_flags;
875
876         seq_printf(m, "    connect_data:\n"
877                    "       flags: %#llx\n"
878                    "       instance: %u\n",
879                    ocd->ocd_connect_flags,
880                    ocd->ocd_instance);
881         if (flags & OBD_CONNECT_VERSION)
882                 seq_printf(m, "       target_version: %u.%u.%u.%u\n",
883                            OBD_OCD_VERSION_MAJOR(ocd->ocd_version),
884                            OBD_OCD_VERSION_MINOR(ocd->ocd_version),
885                            OBD_OCD_VERSION_PATCH(ocd->ocd_version),
886                            OBD_OCD_VERSION_FIX(ocd->ocd_version));
887         if (flags & OBD_CONNECT_MDS)
888                 seq_printf(m, "       mdt_index: %d\n", ocd->ocd_group);
889         if (flags & OBD_CONNECT_GRANT)
890                 seq_printf(m, "       initial_grant: %d\n", ocd->ocd_grant);
891         if (flags & OBD_CONNECT_INDEX)
892                 seq_printf(m, "       target_index: %u\n", ocd->ocd_index);
893         if (flags & OBD_CONNECT_BRW_SIZE)
894                 seq_printf(m, "       max_brw_size: %d\n", ocd->ocd_brw_size);
895         if (flags & OBD_CONNECT_IBITS)
896                 seq_printf(m, "       ibits_known: %#llx\n",
897                            ocd->ocd_ibits_known);
898         if (flags & OBD_CONNECT_GRANT_PARAM)
899                 seq_printf(m, "       grant_block_size: %d\n"
900                            "       grant_inode_size: %d\n"
901                            "       grant_max_extent_size: %d\n"
902                            "       grant_extent_tax: %d\n",
903                            1 << ocd->ocd_grant_blkbits,
904                            1 << ocd->ocd_grant_inobits,
905                            ocd->ocd_grant_max_blks << ocd->ocd_grant_blkbits,
906                            ocd->ocd_grant_tax_kb << 10);
907         if (flags & OBD_CONNECT_TRANSNO)
908                 seq_printf(m, "       first_transno: %#llx\n",
909                            ocd->ocd_transno);
910         if (flags & OBD_CONNECT_CKSUM)
911                 seq_printf(m, "       cksum_types: %#x\n",
912                            ocd->ocd_cksum_types);
913         if (flags & OBD_CONNECT_MAX_EASIZE)
914                 seq_printf(m, "       max_easize: %d\n", ocd->ocd_max_easize);
915         if (flags & OBD_CONNECT_MAXBYTES)
916                 seq_printf(m, "       max_object_bytes: %llu\n",
917                            ocd->ocd_maxbytes);
918         if (flags & OBD_CONNECT_MULTIMODRPCS)
919                 seq_printf(m, "       max_mod_rpcs: %hu\n",
920                            ocd->ocd_maxmodrpcs);
921 }
922
923 int lprocfs_import_seq_show(struct seq_file *m, void *data)
924 {
925         char nidstr[LNET_NIDSTR_SIZE];
926         struct lprocfs_counter ret;
927         struct lprocfs_counter_header *header;
928         struct obd_device *obd = (struct obd_device *)data;
929         struct obd_import *imp;
930         struct obd_import_conn *conn;
931         struct obd_connect_data *ocd;
932         int j;
933         int k;
934         int rw = 0;
935
936         LASSERT(obd != NULL);
937         LPROCFS_CLIMP_CHECK(obd);
938         imp = obd->u.cli.cl_import;
939         ocd = &imp->imp_connect_data;
940
941         seq_printf(m, "import:\n"
942                    "    name: %s\n"
943                    "    target: %s\n"
944                    "    state: %s\n"
945                    "    connect_flags: [ ",
946                    obd->obd_name,
947                    obd2cli_tgt(obd),
948                    ptlrpc_import_state_name(imp->imp_state));
949         obd_connect_seq_flags2str(m, imp->imp_connect_data.ocd_connect_flags,
950                                   imp->imp_connect_data.ocd_connect_flags2,
951                                   ", ");
952         seq_printf(m, " ]\n");
953         obd_connect_data_seqprint(m, ocd);
954         seq_printf(m, "    import_flags: [ ");
955         obd_import_flags2str(imp, m);
956
957         seq_printf(m, " ]\n"
958                    "    connection:\n"
959                    "       failover_nids: [ ");
960         spin_lock(&imp->imp_lock);
961         j = 0;
962         list_for_each_entry(conn, &imp->imp_conn_list, oic_item) {
963                 libcfs_nid2str_r(conn->oic_conn->c_peer.nid,
964                                  nidstr, sizeof(nidstr));
965                 seq_printf(m, "%s%s", j ? ", " : "", nidstr);
966                 j++;
967         }
968         if (imp->imp_connection)
969                 libcfs_nid2str_r(imp->imp_connection->c_peer.nid,
970                                  nidstr, sizeof(nidstr));
971         else
972                 strncpy(nidstr, "<none>", sizeof(nidstr));
973         seq_printf(m, " ]\n"
974                    "       current_connection: %s\n"
975                    "       connection_attempts: %u\n"
976                    "       generation: %u\n"
977                    "       in-progress_invalidations: %u\n"
978                    "       idle: %lld sec\n",
979                    nidstr,
980                    imp->imp_conn_cnt,
981                    imp->imp_generation,
982                    atomic_read(&imp->imp_inval_count),
983                    ktime_get_real_seconds() - imp->imp_last_reply_time);
984         spin_unlock(&imp->imp_lock);
985
986         if (!obd->obd_svc_stats)
987                 goto out_climp;
988
989         header = &obd->obd_svc_stats->ls_cnt_header[PTLRPC_REQWAIT_CNTR];
990         lprocfs_stats_collect(obd->obd_svc_stats, PTLRPC_REQWAIT_CNTR, &ret);
991         if (ret.lc_count != 0) {
992                 /* first argument to do_div MUST be __u64 */
993                 __u64 sum = ret.lc_sum;
994                 do_div(sum, ret.lc_count);
995                 ret.lc_sum = sum;
996         } else
997                 ret.lc_sum = 0;
998         seq_printf(m, "    rpcs:\n"
999                    "       inflight: %u\n"
1000                    "       unregistering: %u\n"
1001                    "       timeouts: %u\n"
1002                    "       avg_waittime: %llu %s\n",
1003                    atomic_read(&imp->imp_inflight),
1004                    atomic_read(&imp->imp_unregistering),
1005                    atomic_read(&imp->imp_timeouts),
1006                    ret.lc_sum, header->lc_units);
1007
1008         k = 0;
1009         for(j = 0; j < IMP_AT_MAX_PORTALS; j++) {
1010                 if (imp->imp_at.iat_portal[j] == 0)
1011                         break;
1012                 k = max_t(unsigned int, k,
1013                           at_get(&imp->imp_at.iat_service_estimate[j]));
1014         }
1015         seq_printf(m, "    service_estimates:\n"
1016                    "       services: %u sec\n"
1017                    "       network: %u sec\n",
1018                    k,
1019                    at_get(&imp->imp_at.iat_net_latency));
1020
1021         seq_printf(m, "    transactions:\n"
1022                    "       last_replay: %llu\n"
1023                    "       peer_committed: %llu\n"
1024                    "       last_checked: %llu\n",
1025                    imp->imp_last_replay_transno,
1026                    imp->imp_peer_committed_transno,
1027                    imp->imp_last_transno_checked);
1028
1029         /* avg data rates */
1030         for (rw = 0; rw <= 1; rw++) {
1031                 lprocfs_stats_collect(obd->obd_svc_stats,
1032                                       PTLRPC_LAST_CNTR + BRW_READ_BYTES + rw,
1033                                       &ret);
1034                 if (ret.lc_sum > 0 && ret.lc_count > 0) {
1035                         /* first argument to do_div MUST be __u64 */
1036                         __u64 sum = ret.lc_sum;
1037                         do_div(sum, ret.lc_count);
1038                         ret.lc_sum = sum;
1039                         seq_printf(m, "    %s_data_averages:\n"
1040                                    "       bytes_per_rpc: %llu\n",
1041                                    rw ? "write" : "read",
1042                                    ret.lc_sum);
1043                 }
1044                 k = (int)ret.lc_sum;
1045                 j = opcode_offset(OST_READ + rw) + EXTRA_MAX_OPCODES;
1046                 header = &obd->obd_svc_stats->ls_cnt_header[j];
1047                 lprocfs_stats_collect(obd->obd_svc_stats, j, &ret);
1048                 if (ret.lc_sum > 0 && ret.lc_count != 0) {
1049                         /* first argument to do_div MUST be __u64 */
1050                         __u64 sum = ret.lc_sum;
1051                         do_div(sum, ret.lc_count);
1052                         ret.lc_sum = sum;
1053                         seq_printf(m, "       %s_per_rpc: %llu\n",
1054                                    header->lc_units, ret.lc_sum);
1055                         j = (int)ret.lc_sum;
1056                         if (j > 0)
1057                                 seq_printf(m, "       MB_per_sec: %u.%.02u\n",
1058                                            k / j, (100 * k / j) % 100);
1059                 }
1060         }
1061
1062 out_climp:
1063         LPROCFS_CLIMP_EXIT(obd);
1064         return 0;
1065 }
1066 EXPORT_SYMBOL(lprocfs_import_seq_show);
1067
1068 int lprocfs_state_seq_show(struct seq_file *m, void *data)
1069 {
1070         struct obd_device *obd = (struct obd_device *)data;
1071         struct obd_import *imp;
1072         int j, k;
1073
1074         LASSERT(obd != NULL);
1075         LPROCFS_CLIMP_CHECK(obd);
1076         imp = obd->u.cli.cl_import;
1077
1078         seq_printf(m, "current_state: %s\n",
1079                    ptlrpc_import_state_name(imp->imp_state));
1080         seq_printf(m, "state_history:\n");
1081         k = imp->imp_state_hist_idx;
1082         for (j = 0; j < IMP_STATE_HIST_LEN; j++) {
1083                 struct import_state_hist *ish =
1084                         &imp->imp_state_hist[(k + j) % IMP_STATE_HIST_LEN];
1085                 if (ish->ish_state == 0)
1086                         continue;
1087                 seq_printf(m, " - [ %lld, %s ]\n", (s64)ish->ish_time,
1088                            ptlrpc_import_state_name(ish->ish_state));
1089         }
1090
1091         LPROCFS_CLIMP_EXIT(obd);
1092         return 0;
1093 }
1094 EXPORT_SYMBOL(lprocfs_state_seq_show);
1095
1096 int lprocfs_at_hist_helper(struct seq_file *m, struct adaptive_timeout *at)
1097 {
1098         int i;
1099         for (i = 0; i < AT_BINS; i++)
1100                 seq_printf(m, "%3u ", at->at_hist[i]);
1101         seq_printf(m, "\n");
1102         return 0;
1103 }
1104 EXPORT_SYMBOL(lprocfs_at_hist_helper);
1105
1106 /* See also ptlrpc_lprocfs_timeouts_show_seq */
1107 int lprocfs_timeouts_seq_show(struct seq_file *m, void *data)
1108 {
1109         struct obd_device *obd = (struct obd_device *)data;
1110         struct obd_import *imp;
1111         unsigned int cur, worst;
1112         time64_t now, worstt;
1113         int i;
1114
1115         LASSERT(obd != NULL);
1116         LPROCFS_CLIMP_CHECK(obd);
1117         imp = obd->u.cli.cl_import;
1118
1119         now = ktime_get_real_seconds();
1120
1121         /* Some network health info for kicks */
1122         seq_printf(m, "%-10s : %lld, %llds ago\n",
1123                    "last reply", (s64)imp->imp_last_reply_time,
1124                    (s64)(now - imp->imp_last_reply_time));
1125
1126         cur = at_get(&imp->imp_at.iat_net_latency);
1127         worst = imp->imp_at.iat_net_latency.at_worst_ever;
1128         worstt = imp->imp_at.iat_net_latency.at_worst_time;
1129         seq_printf(m, "%-10s : cur %3u  worst %3u (at %lld, %llds ago) ",
1130                    "network", cur, worst, (s64)worstt, (s64)(now - worstt));
1131         lprocfs_at_hist_helper(m, &imp->imp_at.iat_net_latency);
1132
1133         for(i = 0; i < IMP_AT_MAX_PORTALS; i++) {
1134                 if (imp->imp_at.iat_portal[i] == 0)
1135                         break;
1136                 cur = at_get(&imp->imp_at.iat_service_estimate[i]);
1137                 worst = imp->imp_at.iat_service_estimate[i].at_worst_ever;
1138                 worstt = imp->imp_at.iat_service_estimate[i].at_worst_time;
1139                 seq_printf(m, "portal %-2d  : cur %3u  worst %3u (at %lld, %llds ago) ",
1140                            imp->imp_at.iat_portal[i], cur, worst, (s64)worstt,
1141                            (s64)(now - worstt));
1142                 lprocfs_at_hist_helper(m, &imp->imp_at.iat_service_estimate[i]);
1143         }
1144
1145         LPROCFS_CLIMP_EXIT(obd);
1146         return 0;
1147 }
1148 EXPORT_SYMBOL(lprocfs_timeouts_seq_show);
1149
1150 int lprocfs_connect_flags_seq_show(struct seq_file *m, void *data)
1151 {
1152         struct obd_device *obd = data;
1153         __u64 flags;
1154         __u64 flags2;
1155
1156         LPROCFS_CLIMP_CHECK(obd);
1157         flags = obd->u.cli.cl_import->imp_connect_data.ocd_connect_flags;
1158         flags2 = obd->u.cli.cl_import->imp_connect_data.ocd_connect_flags2;
1159         seq_printf(m, "flags=%#llx\n", flags);
1160         seq_printf(m, "flags2=%#llx\n", flags2);
1161         obd_connect_seq_flags2str(m, flags, flags2, "\n");
1162         seq_printf(m, "\n");
1163         LPROCFS_CLIMP_EXIT(obd);
1164         return 0;
1165 }
1166 EXPORT_SYMBOL(lprocfs_connect_flags_seq_show);
1167
1168 static const struct attribute *obd_def_uuid_attrs[] = {
1169         &lustre_attr_uuid.attr,
1170         NULL,
1171 };
1172
1173 static const struct attribute *obd_def_attrs[] = {
1174         &lustre_attr_blocksize.attr,
1175         &lustre_attr_kbytestotal.attr,
1176         &lustre_attr_kbytesfree.attr,
1177         &lustre_attr_kbytesavail.attr,
1178         &lustre_attr_filestotal.attr,
1179         &lustre_attr_filesfree.attr,
1180         &lustre_attr_uuid.attr,
1181         NULL,
1182 };
1183
1184 static void obd_sysfs_release(struct kobject *kobj)
1185 {
1186         struct obd_device *obd = container_of(kobj, struct obd_device,
1187                                               obd_kset.kobj);
1188
1189         complete(&obd->obd_kobj_unregister);
1190 }
1191
1192 int lprocfs_obd_setup(struct obd_device *obd, bool uuid_only)
1193 {
1194         struct lprocfs_vars *debugfs_vars = NULL;
1195         int rc;
1196
1197         if (!obd || obd->obd_magic != OBD_DEVICE_MAGIC)
1198                 return -ENODEV;
1199
1200         rc = kobject_set_name(&obd->obd_kset.kobj, "%s", obd->obd_name);
1201         if (rc)
1202                 return rc;
1203
1204         obd->obd_ktype.sysfs_ops = &lustre_sysfs_ops;
1205         obd->obd_ktype.release = obd_sysfs_release;
1206
1207         obd->obd_kset.kobj.parent = obd->obd_type->typ_kobj;
1208         obd->obd_kset.kobj.ktype = &obd->obd_ktype;
1209         init_completion(&obd->obd_kobj_unregister);
1210         rc = kset_register(&obd->obd_kset);
1211         if (rc)
1212                 return rc;
1213
1214         if (uuid_only)
1215                 obd->obd_attrs = obd_def_uuid_attrs;
1216         else
1217                 obd->obd_attrs = obd_def_attrs;
1218
1219         rc = sysfs_create_files(&obd->obd_kset.kobj, obd->obd_attrs);
1220         if (rc) {
1221                 kset_unregister(&obd->obd_kset);
1222                 return rc;
1223         }
1224
1225         if (!obd->obd_type->typ_procroot)
1226                 debugfs_vars = obd->obd_vars;
1227         obd->obd_debugfs_entry = ldebugfs_register(obd->obd_name,
1228                                                    obd->obd_type->typ_debugfs_entry,
1229                                                    debugfs_vars, obd);
1230         if (IS_ERR_OR_NULL(obd->obd_debugfs_entry)) {
1231                 rc = obd->obd_debugfs_entry ? PTR_ERR(obd->obd_debugfs_entry)
1232                                             : -ENOMEM;
1233                 CERROR("error %d setting up debugfs for %s\n",
1234                        rc, obd->obd_name);
1235                 obd->obd_debugfs_entry = NULL;
1236
1237                 sysfs_remove_files(&obd->obd_kset.kobj, obd->obd_attrs);
1238                 obd->obd_attrs = NULL;
1239                 kset_unregister(&obd->obd_kset);
1240                 return rc;
1241         }
1242
1243         if (obd->obd_proc_entry || !obd->obd_type->typ_procroot)
1244                 GOTO(already_registered, rc);
1245
1246         obd->obd_proc_entry = lprocfs_register(obd->obd_name,
1247                                                obd->obd_type->typ_procroot,
1248                                                obd->obd_vars, obd);
1249         if (IS_ERR(obd->obd_proc_entry)) {
1250                 rc = PTR_ERR(obd->obd_proc_entry);
1251                 CERROR("error %d setting up lprocfs for %s\n",rc,obd->obd_name);
1252                 obd->obd_proc_entry = NULL;
1253
1254                 ldebugfs_remove(&obd->obd_debugfs_entry);
1255                 sysfs_remove_files(&obd->obd_kset.kobj, obd->obd_attrs);
1256                 obd->obd_attrs = NULL;
1257                 kset_unregister(&obd->obd_kset);
1258                 return rc;
1259         }
1260 already_registered:
1261         return rc;
1262 }
1263 EXPORT_SYMBOL(lprocfs_obd_setup);
1264
1265 int lprocfs_obd_cleanup(struct obd_device *obd)
1266 {
1267         if (!obd)
1268                 return -EINVAL;
1269
1270         if (obd->obd_proc_exports_entry) {
1271                 /* Should be no exports left */
1272                 lprocfs_remove(&obd->obd_proc_exports_entry);
1273                 obd->obd_proc_exports_entry = NULL;
1274         }
1275
1276         if (obd->obd_proc_entry) {
1277                 lprocfs_remove(&obd->obd_proc_entry);
1278                 obd->obd_proc_entry = NULL;
1279         }
1280
1281         if (!IS_ERR_OR_NULL(obd->obd_debugfs_entry))
1282                 ldebugfs_remove(&obd->obd_debugfs_entry);
1283
1284         /* obd device never allocated a kset */
1285         if (!obd->obd_kset.kobj.state_initialized)
1286                 return 0;
1287
1288         if (obd->obd_attrs) {
1289                 sysfs_remove_files(&obd->obd_kset.kobj, obd->obd_attrs);
1290                 obd->obd_attrs = NULL;
1291         }
1292
1293         kset_unregister(&obd->obd_kset);
1294         wait_for_completion(&obd->obd_kobj_unregister);
1295         return 0;
1296 }
1297 EXPORT_SYMBOL(lprocfs_obd_cleanup);
1298
1299 int lprocfs_stats_alloc_one(struct lprocfs_stats *stats, unsigned int cpuid)
1300 {
1301         struct lprocfs_counter *cntr;
1302         unsigned int percpusize;
1303         int rc = -ENOMEM;
1304         unsigned long flags = 0;
1305         int i;
1306
1307         LASSERT(stats->ls_percpu[cpuid] == NULL);
1308         LASSERT((stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) == 0);
1309
1310         percpusize = lprocfs_stats_counter_size(stats);
1311         LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[cpuid], percpusize);
1312         if (stats->ls_percpu[cpuid]) {
1313                 rc = 0;
1314                 if (unlikely(stats->ls_biggest_alloc_num <= cpuid)) {
1315                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1316                                 spin_lock_irqsave(&stats->ls_lock, flags);
1317                         else
1318                                 spin_lock(&stats->ls_lock);
1319                         if (stats->ls_biggest_alloc_num <= cpuid)
1320                                 stats->ls_biggest_alloc_num = cpuid + 1;
1321                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) {
1322                                 spin_unlock_irqrestore(&stats->ls_lock, flags);
1323                         } else {
1324                                 spin_unlock(&stats->ls_lock);
1325                         }
1326                 }
1327                 /* initialize the ls_percpu[cpuid] non-zero counter */
1328                 for (i = 0; i < stats->ls_num; ++i) {
1329                         cntr = lprocfs_stats_counter_get(stats, cpuid, i);
1330                         cntr->lc_min = LC_MIN_INIT;
1331                 }
1332         }
1333         return rc;
1334 }
1335
1336 struct lprocfs_stats *lprocfs_alloc_stats(unsigned int num,
1337                                           enum lprocfs_stats_flags flags)
1338 {
1339         struct lprocfs_stats *stats;
1340         unsigned int num_entry;
1341         unsigned int percpusize = 0;
1342         int i;
1343
1344         if (num == 0)
1345                 return NULL;
1346
1347         if (lprocfs_no_percpu_stats != 0)
1348                 flags |= LPROCFS_STATS_FLAG_NOPERCPU;
1349
1350         if (flags & LPROCFS_STATS_FLAG_NOPERCPU)
1351                 num_entry = 1;
1352         else
1353                 num_entry = num_possible_cpus();
1354
1355         /* alloc percpu pointers for all possible cpu slots */
1356         LIBCFS_ALLOC(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1357         if (!stats)
1358                 return NULL;
1359
1360         stats->ls_num = num;
1361         stats->ls_flags = flags;
1362         spin_lock_init(&stats->ls_lock);
1363
1364         /* alloc num of counter headers */
1365         LIBCFS_ALLOC(stats->ls_cnt_header,
1366                      stats->ls_num * sizeof(struct lprocfs_counter_header));
1367         if (!stats->ls_cnt_header)
1368                 goto fail;
1369
1370         if ((flags & LPROCFS_STATS_FLAG_NOPERCPU) != 0) {
1371                 /* contains only one set counters */
1372                 percpusize = lprocfs_stats_counter_size(stats);
1373                 LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[0], percpusize);
1374                 if (!stats->ls_percpu[0])
1375                         goto fail;
1376                 stats->ls_biggest_alloc_num = 1;
1377         } else if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0) {
1378                 /* alloc all percpu data, currently only obd_memory use this */
1379                 for (i = 0; i < num_entry; ++i)
1380                         if (lprocfs_stats_alloc_one(stats, i) < 0)
1381                                 goto fail;
1382         }
1383
1384         return stats;
1385
1386 fail:
1387         lprocfs_free_stats(&stats);
1388         return NULL;
1389 }
1390 EXPORT_SYMBOL(lprocfs_alloc_stats);
1391
1392 void lprocfs_free_stats(struct lprocfs_stats **statsh)
1393 {
1394         struct lprocfs_stats *stats = *statsh;
1395         unsigned int num_entry;
1396         unsigned int percpusize;
1397         unsigned int i;
1398
1399         if (!stats || stats->ls_num == 0)
1400                 return;
1401         *statsh = NULL;
1402
1403         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU)
1404                 num_entry = 1;
1405         else
1406                 num_entry = num_possible_cpus();
1407
1408         percpusize = lprocfs_stats_counter_size(stats);
1409         for (i = 0; i < num_entry; i++)
1410                 if (stats->ls_percpu[i])
1411                         LIBCFS_FREE(stats->ls_percpu[i], percpusize);
1412         if (stats->ls_cnt_header)
1413                 LIBCFS_FREE(stats->ls_cnt_header, stats->ls_num *
1414                                         sizeof(struct lprocfs_counter_header));
1415         LIBCFS_FREE(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1416 }
1417 EXPORT_SYMBOL(lprocfs_free_stats);
1418
1419 u64 lprocfs_stats_collector(struct lprocfs_stats *stats, int idx,
1420                             enum lprocfs_fields_flags field)
1421 {
1422         unsigned long flags = 0;
1423         unsigned int num_cpu;
1424         unsigned int i;
1425         u64 ret = 0;
1426
1427         LASSERT(stats);
1428
1429         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1430         for (i = 0; i < num_cpu; i++) {
1431                 struct lprocfs_counter *cntr;
1432
1433                 if (!stats->ls_percpu[i])
1434                         continue;
1435
1436                 cntr = lprocfs_stats_counter_get(stats, i, idx);
1437                 ret += lprocfs_read_helper(cntr, &stats->ls_cnt_header[idx],
1438                                            stats->ls_flags, field);
1439         }
1440         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1441         return ret;
1442 }
1443 EXPORT_SYMBOL(lprocfs_stats_collector);
1444
1445 void lprocfs_clear_stats(struct lprocfs_stats *stats)
1446 {
1447         struct lprocfs_counter *percpu_cntr;
1448         int i;
1449         int j;
1450         unsigned int num_entry;
1451         unsigned long flags = 0;
1452
1453         num_entry = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1454
1455         for (i = 0; i < num_entry; i++) {
1456                 if (!stats->ls_percpu[i])
1457                         continue;
1458                 for (j = 0; j < stats->ls_num; j++) {
1459                         percpu_cntr = lprocfs_stats_counter_get(stats, i, j);
1460                         percpu_cntr->lc_count           = 0;
1461                         percpu_cntr->lc_min             = LC_MIN_INIT;
1462                         percpu_cntr->lc_max             = 0;
1463                         percpu_cntr->lc_sumsquare       = 0;
1464                         percpu_cntr->lc_sum             = 0;
1465                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1466                                 percpu_cntr->lc_sum_irq = 0;
1467                 }
1468         }
1469
1470         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1471 }
1472 EXPORT_SYMBOL(lprocfs_clear_stats);
1473
1474 static ssize_t lprocfs_stats_seq_write(struct file *file,
1475                                        const char __user *buf,
1476                                        size_t len, loff_t *off)
1477 {
1478         struct seq_file *seq = file->private_data;
1479         struct lprocfs_stats *stats = seq->private;
1480
1481         lprocfs_clear_stats(stats);
1482
1483         return len;
1484 }
1485
1486 static void *lprocfs_stats_seq_start(struct seq_file *p, loff_t *pos)
1487 {
1488         struct lprocfs_stats *stats = p->private;
1489
1490         return (*pos < stats->ls_num) ? pos : NULL;
1491 }
1492
1493 static void lprocfs_stats_seq_stop(struct seq_file *p, void *v)
1494 {
1495 }
1496
1497 static void *lprocfs_stats_seq_next(struct seq_file *p, void *v, loff_t *pos)
1498 {
1499         (*pos)++;
1500
1501         return lprocfs_stats_seq_start(p, pos);
1502 }
1503
1504 /* seq file export of one lprocfs counter */
1505 static int lprocfs_stats_seq_show(struct seq_file *p, void *v)
1506 {
1507         struct lprocfs_stats *stats = p->private;
1508         struct lprocfs_counter_header *hdr;
1509         struct lprocfs_counter ctr;
1510         int idx = *(loff_t *)v;
1511
1512         if (idx == 0) {
1513                 struct timespec64 now;
1514
1515                 ktime_get_real_ts64(&now);
1516                 seq_printf(p, "%-25s %llu.%09lu secs.nsecs\n",
1517                            "snapshot_time", (s64)now.tv_sec, now.tv_nsec);
1518         }
1519
1520         hdr = &stats->ls_cnt_header[idx];
1521         lprocfs_stats_collect(stats, idx, &ctr);
1522
1523         if (ctr.lc_count == 0)
1524                 return 0;
1525
1526         seq_printf(p, "%-25s %lld samples [%s]", hdr->lc_name,
1527                    ctr.lc_count, hdr->lc_units);
1528
1529         if ((hdr->lc_config & LPROCFS_CNTR_AVGMINMAX) && ctr.lc_count > 0) {
1530                 seq_printf(p, " %lld %lld %lld",
1531                            ctr.lc_min, ctr.lc_max, ctr.lc_sum);
1532                 if (hdr->lc_config & LPROCFS_CNTR_STDDEV)
1533                         seq_printf(p, " %llu", ctr.lc_sumsquare);
1534         }
1535         seq_putc(p, '\n');
1536         return 0;
1537 }
1538
1539 static const struct seq_operations lprocfs_stats_seq_sops = {
1540         .start  = lprocfs_stats_seq_start,
1541         .stop   = lprocfs_stats_seq_stop,
1542         .next   = lprocfs_stats_seq_next,
1543         .show   = lprocfs_stats_seq_show,
1544 };
1545
1546 static int lprocfs_stats_seq_open(struct inode *inode, struct file *file)
1547 {
1548         struct seq_file *seq;
1549         int rc;
1550
1551         rc = LPROCFS_ENTRY_CHECK(inode);
1552         if (rc < 0)
1553                 return rc;
1554
1555         rc = seq_open(file, &lprocfs_stats_seq_sops);
1556         if (rc)
1557                 return rc;
1558         seq = file->private_data;
1559         seq->private = inode->i_private ? inode->i_private : PDE_DATA(inode);
1560         return 0;
1561 }
1562
1563 static const struct file_operations lprocfs_stats_seq_fops = {
1564         .owner   = THIS_MODULE,
1565         .open    = lprocfs_stats_seq_open,
1566         .read    = seq_read,
1567         .write   = lprocfs_stats_seq_write,
1568         .llseek  = seq_lseek,
1569         .release = lprocfs_seq_release,
1570 };
1571
1572 int ldebugfs_register_stats(struct dentry *parent, const char *name,
1573                             struct lprocfs_stats *stats)
1574 {
1575         struct dentry *entry;
1576
1577         LASSERT(!IS_ERR_OR_NULL(parent));
1578
1579         entry = debugfs_create_file(name, 0644, parent, stats,
1580                                     &lprocfs_stats_seq_fops);
1581         if (IS_ERR_OR_NULL(entry))
1582                 return entry ? PTR_ERR(entry) : -ENOMEM;
1583
1584         return 0;
1585 }
1586 EXPORT_SYMBOL_GPL(ldebugfs_register_stats);
1587
1588 int lprocfs_register_stats(struct proc_dir_entry *root, const char *name,
1589                            struct lprocfs_stats *stats)
1590 {
1591         struct proc_dir_entry *entry;
1592         LASSERT(root != NULL);
1593
1594         entry = proc_create_data(name, 0644, root,
1595                                  &lprocfs_stats_seq_fops, stats);
1596         if (!entry)
1597                 return -ENOMEM;
1598         return 0;
1599 }
1600 EXPORT_SYMBOL(lprocfs_register_stats);
1601
1602 void lprocfs_counter_init(struct lprocfs_stats *stats, int index,
1603                           unsigned conf, const char *name, const char *units)
1604 {
1605         struct lprocfs_counter_header *header;
1606         struct lprocfs_counter *percpu_cntr;
1607         unsigned long flags = 0;
1608         unsigned int i;
1609         unsigned int num_cpu;
1610
1611         LASSERT(stats != NULL);
1612
1613         header = &stats->ls_cnt_header[index];
1614         LASSERTF(header != NULL, "Failed to allocate stats header:[%d]%s/%s\n",
1615                  index, name, units);
1616
1617         header->lc_config = conf;
1618         header->lc_name   = name;
1619         header->lc_units  = units;
1620
1621         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1622         for (i = 0; i < num_cpu; ++i) {
1623                 if (!stats->ls_percpu[i])
1624                         continue;
1625                 percpu_cntr = lprocfs_stats_counter_get(stats, i, index);
1626                 percpu_cntr->lc_count           = 0;
1627                 percpu_cntr->lc_min             = LC_MIN_INIT;
1628                 percpu_cntr->lc_max             = 0;
1629                 percpu_cntr->lc_sumsquare       = 0;
1630                 percpu_cntr->lc_sum             = 0;
1631                 if ((stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1632                         percpu_cntr->lc_sum_irq = 0;
1633         }
1634         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1635 }
1636 EXPORT_SYMBOL(lprocfs_counter_init);
1637
1638 static const char * const mps_stats[] = {
1639         [LPROC_MD_CLOSE]                = "close",
1640         [LPROC_MD_CREATE]               = "create",
1641         [LPROC_MD_ENQUEUE]              = "enqueue",
1642         [LPROC_MD_GETATTR]              = "getattr",
1643         [LPROC_MD_INTENT_LOCK]          = "intent_lock",
1644         [LPROC_MD_LINK]                 = "link",
1645         [LPROC_MD_RENAME]               = "rename",
1646         [LPROC_MD_SETATTR]              = "setattr",
1647         [LPROC_MD_FSYNC]                = "fsync",
1648         [LPROC_MD_READ_PAGE]            = "read_page",
1649         [LPROC_MD_UNLINK]               = "unlink",
1650         [LPROC_MD_SETXATTR]             = "setxattr",
1651         [LPROC_MD_GETXATTR]             = "getxattr",
1652         [LPROC_MD_INTENT_GETATTR_ASYNC] = "intent_getattr_async",
1653         [LPROC_MD_REVALIDATE_LOCK]      = "revalidate_lock",
1654 };
1655
1656 int lprocfs_alloc_md_stats(struct obd_device *obd,
1657                            unsigned int num_private_stats)
1658 {
1659         struct lprocfs_stats *stats;
1660         unsigned int num_stats;
1661         int rc, i;
1662
1663         /*
1664          * TODO Ensure that this function is only used where
1665          * appropriate by adding an assertion to the effect that
1666          * obd->obd_type->typ_md_ops is not NULL. We can't do this now
1667          * because mdt_procfs_init() uses this function to allocate
1668          * the stats backing /proc/fs/lustre/mdt/.../md_stats but the
1669          * mdt layer does not use the md_ops interface. This is
1670          * confusing and a waste of memory. See LU-2484.
1671          */
1672         LASSERT(obd->obd_proc_entry != NULL);
1673         LASSERT(obd->obd_md_stats == NULL);
1674
1675         num_stats = ARRAY_SIZE(mps_stats) + num_private_stats;
1676         stats = lprocfs_alloc_stats(num_stats, 0);
1677         if (!stats)
1678                 return -ENOMEM;
1679
1680         for (i = 0; i < ARRAY_SIZE(mps_stats); i++) {
1681                 lprocfs_counter_init(stats, i, 0, mps_stats[i], "reqs");
1682                 if (!stats->ls_cnt_header[i].lc_name) {
1683                         CERROR("Missing md_stat initializer md_op operation at offset %d. Aborting.\n",
1684                                i);
1685                         LBUG();
1686                 }
1687         }
1688
1689         rc = lprocfs_register_stats(obd->obd_proc_entry, "md_stats", stats);
1690         if (rc < 0) {
1691                 lprocfs_free_stats(&stats);
1692         } else {
1693                 obd->obd_md_stats = stats;
1694         }
1695
1696         return rc;
1697 }
1698 EXPORT_SYMBOL(lprocfs_alloc_md_stats);
1699
1700 void lprocfs_free_md_stats(struct obd_device *obd)
1701 {
1702         struct lprocfs_stats *stats = obd->obd_md_stats;
1703
1704         if (stats) {
1705                 obd->obd_md_stats = NULL;
1706                 lprocfs_free_stats(&stats);
1707         }
1708 }
1709 EXPORT_SYMBOL(lprocfs_free_md_stats);
1710
1711 void lprocfs_init_ldlm_stats(struct lprocfs_stats *ldlm_stats)
1712 {
1713         lprocfs_counter_init(ldlm_stats,
1714                              LDLM_ENQUEUE - LDLM_FIRST_OPC,
1715                              0, "ldlm_enqueue", "reqs");
1716         lprocfs_counter_init(ldlm_stats,
1717                              LDLM_CONVERT - LDLM_FIRST_OPC,
1718                              0, "ldlm_convert", "reqs");
1719         lprocfs_counter_init(ldlm_stats,
1720                              LDLM_CANCEL - LDLM_FIRST_OPC,
1721                              0, "ldlm_cancel", "reqs");
1722         lprocfs_counter_init(ldlm_stats,
1723                              LDLM_BL_CALLBACK - LDLM_FIRST_OPC,
1724                              0, "ldlm_bl_callback", "reqs");
1725         lprocfs_counter_init(ldlm_stats,
1726                              LDLM_CP_CALLBACK - LDLM_FIRST_OPC,
1727                              0, "ldlm_cp_callback", "reqs");
1728         lprocfs_counter_init(ldlm_stats,
1729                              LDLM_GL_CALLBACK - LDLM_FIRST_OPC,
1730                              0, "ldlm_gl_callback", "reqs");
1731 }
1732 EXPORT_SYMBOL(lprocfs_init_ldlm_stats);
1733
1734 __s64 lprocfs_read_helper(struct lprocfs_counter *lc,
1735                           struct lprocfs_counter_header *header,
1736                           enum lprocfs_stats_flags flags,
1737                           enum lprocfs_fields_flags field)
1738 {
1739         __s64 ret = 0;
1740
1741         if (!lc || !header)
1742                 RETURN(0);
1743
1744         switch (field) {
1745                 case LPROCFS_FIELDS_FLAGS_CONFIG:
1746                         ret = header->lc_config;
1747                         break;
1748                 case LPROCFS_FIELDS_FLAGS_SUM:
1749                         ret = lc->lc_sum;
1750                         if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1751                                 ret += lc->lc_sum_irq;
1752                         break;
1753                 case LPROCFS_FIELDS_FLAGS_MIN:
1754                         ret = lc->lc_min;
1755                         break;
1756                 case LPROCFS_FIELDS_FLAGS_MAX:
1757                         ret = lc->lc_max;
1758                         break;
1759                 case LPROCFS_FIELDS_FLAGS_AVG:
1760                         ret = (lc->lc_max - lc->lc_min) / 2;
1761                         break;
1762                 case LPROCFS_FIELDS_FLAGS_SUMSQUARE:
1763                         ret = lc->lc_sumsquare;
1764                         break;
1765                 case LPROCFS_FIELDS_FLAGS_COUNT:
1766                         ret = lc->lc_count;
1767                         break;
1768                 default:
1769                         break;
1770         };
1771         RETURN(ret);
1772 }
1773 EXPORT_SYMBOL(lprocfs_read_helper);
1774
1775 int lprocfs_read_frac_helper(char *buffer, unsigned long count, long val,
1776                              int mult)
1777 {
1778         long decimal_val, frac_val;
1779         int prtn;
1780
1781         if (count < 10)
1782                 return -EINVAL;
1783
1784         decimal_val = val / mult;
1785         prtn = snprintf(buffer, count, "%ld", decimal_val);
1786         frac_val = val % mult;
1787
1788         if (prtn < (count - 4) && frac_val > 0) {
1789                 long temp_frac;
1790                 int i, temp_mult = 1, frac_bits = 0;
1791
1792                 temp_frac = frac_val * 10;
1793                 buffer[prtn++] = '.';
1794                 while (frac_bits < 2 && (temp_frac / mult) < 1) {
1795                         /* only reserved 2 bits fraction */
1796                         buffer[prtn++] = '0';
1797                         temp_frac *= 10;
1798                         frac_bits++;
1799                 }
1800                 /*
1801                  * Need to think these cases :
1802                  *      1. #echo x.00 > /proc/xxx       output result : x
1803                  *      2. #echo x.0x > /proc/xxx       output result : x.0x
1804                  *      3. #echo x.x0 > /proc/xxx       output result : x.x
1805                  *      4. #echo x.xx > /proc/xxx       output result : x.xx
1806                  *      Only reserved 2 bits fraction.
1807                  */
1808                 for (i = 0; i < (5 - prtn); i++)
1809                         temp_mult *= 10;
1810
1811                 frac_bits = min((int)count - prtn, 3 - frac_bits);
1812                 prtn += snprintf(buffer + prtn, frac_bits, "%ld",
1813                                  frac_val * temp_mult / mult);
1814
1815                 prtn--;
1816                 while (buffer[prtn] < '1' || buffer[prtn] > '9') {
1817                         prtn--;
1818                         if (buffer[prtn] == '.') {
1819                                 prtn--;
1820                                 break;
1821                         }
1822                 }
1823                 prtn++;
1824         }
1825         buffer[prtn++] = '\n';
1826         return prtn;
1827 }
1828 EXPORT_SYMBOL(lprocfs_read_frac_helper);
1829
1830 int lprocfs_seq_read_frac_helper(struct seq_file *m, long val, int mult)
1831 {
1832         long decimal_val, frac_val;
1833
1834         decimal_val = val / mult;
1835         seq_printf(m, "%ld", decimal_val);
1836         frac_val = val % mult;
1837
1838         if (frac_val > 0) {
1839                 frac_val *= 100;
1840                 frac_val /= mult;
1841         }
1842         if (frac_val > 0) {
1843                 /* Three cases: x0, xx, 0x */
1844                 if ((frac_val % 10) != 0)
1845                         seq_printf(m, ".%ld", frac_val);
1846                 else
1847                         seq_printf(m, ".%ld", frac_val / 10);
1848         }
1849
1850         seq_printf(m, "\n");
1851         return 0;
1852 }
1853 EXPORT_SYMBOL(lprocfs_seq_read_frac_helper);
1854
1855 /* Obtains the conversion factor for the unit specified */
1856 static int get_mult(char unit, __u64 *mult)
1857 {
1858         __u64 units = 1;
1859
1860         switch (unit) {
1861         /* peta, tera, giga, mega, and kilo */
1862         case 'p':
1863         case 'P':
1864                 units <<= 10;
1865         case 't':
1866         case 'T':
1867                 units <<= 10;
1868         case 'g':
1869         case 'G':
1870                 units <<= 10;
1871         case 'm':
1872         case 'M':
1873                 units <<= 10;
1874         case 'k':
1875         case 'K':
1876                 units <<= 10;
1877                 break;
1878         /* some tests expect % to be accepted */
1879         case '%':
1880                 units = 1;
1881                 break;
1882         default:
1883                 return -EINVAL;
1884         }
1885
1886         *mult = units;
1887
1888         return 0;
1889 }
1890
1891 /*
1892  * Ensures the numeric string is valid. The function provides the final
1893  * multiplier in the case a unit exists at the end of the string. It also
1894  * locates the start of the whole and fractional parts (if any). This
1895  * function modifies the string so kstrtoull can be used to parse both
1896  * the whole and fraction portions. This function also figures out
1897  * the base of the number.
1898  */
1899 static int preprocess_numeric_str(char *buffer, __u64 *mult, __u64 def_mult,
1900                                   bool allow_units, char **whole, char **frac,
1901                                   unsigned int *base)
1902 {
1903         bool hit_decimal = false;
1904         bool hit_unit = false;
1905         int rc = 0;
1906         char *start;
1907         *mult = def_mult;
1908         *whole = NULL;
1909         *frac = NULL;
1910         *base = 10;
1911
1912         /* a hex string if it starts with "0x" */
1913         if (buffer[0] == '0' && tolower(buffer[1]) == 'x') {
1914                 *base = 16;
1915                 buffer += 2;
1916         }
1917
1918         start = buffer;
1919
1920         while (*buffer) {
1921                 /* allow for a single new line before the null terminator */
1922                 if (*buffer == '\n') {
1923                         *buffer = '\0';
1924                         buffer++;
1925
1926                         if (*buffer)
1927                                 return -EINVAL;
1928
1929                         break;
1930                 }
1931
1932                 /* any chars after our unit indicates a malformed string */
1933                 if (hit_unit)
1934                         return -EINVAL;
1935
1936                 /* ensure we only hit one decimal */
1937                 if (*buffer == '.') {
1938                         if (hit_decimal)
1939                                 return -EINVAL;
1940
1941                         /* if past start, there's a whole part */
1942                         if (start != buffer)
1943                                 *whole = start;
1944
1945                         *buffer = '\0';
1946                         start = buffer + 1;
1947                         hit_decimal = true;
1948                 } else if (!isdigit(*buffer) &&
1949                            !(*base == 16 && isxdigit(*buffer))) {
1950                         if (allow_units) {
1951                                 /* if we allow units, attempt to get mult */
1952                                 hit_unit = true;
1953                                 rc = get_mult(*buffer, mult);
1954                                 if (rc)
1955                                         return rc;
1956
1957                                 /* string stops here, but keep processing */
1958                                 *buffer = '\0';
1959                         } else {
1960                                 /* bad string */
1961                                 return -EINVAL;
1962                         }
1963                 }
1964
1965                 buffer++;
1966         }
1967
1968         if (hit_decimal) {
1969                 /* hit a decimal, make sure there's a fractional part */
1970                 if (!*start)
1971                         return -EINVAL;
1972
1973                 *frac = start;
1974         } else {
1975                 /* didn't hit a decimal, but may have a whole part */
1976                 if (start != buffer && *start)
1977                         *whole = start;
1978         }
1979
1980         /* malformed string if we didn't get anything */
1981         if (!*frac && !*whole)
1982                 return -EINVAL;
1983
1984         return 0;
1985 }
1986
1987 /*
1988  * Parses a numeric string which can contain a whole and fraction portion
1989  * into a __u64. Accepts a multiplier to apply to the value parsed. Also
1990  * allows the string to have a unit at the end. The function handles
1991  * wrapping of the final unsigned value.
1992  */
1993 static int str_to_u64_parse(char *buffer, unsigned long count,
1994                             __u64 *val, __u64 def_mult, bool allow_units)
1995 {
1996         __u64 whole = 0;
1997         __u64 frac = 0;
1998         unsigned int frac_d = 1;
1999         __u64 wrap_indicator = ULLONG_MAX;
2000         int rc = 0;
2001         __u64 mult;
2002         char *strwhole;
2003         char *strfrac;
2004         unsigned int base = 10;
2005
2006         rc = preprocess_numeric_str(buffer, &mult, def_mult, allow_units,
2007                                     &strwhole, &strfrac, &base);
2008
2009         if (rc)
2010                 return rc;
2011
2012         if (mult == 0) {
2013                 *val = 0;
2014                 return 0;
2015         }
2016
2017         /* the multiplier limits how large the value can be */
2018         wrap_indicator = div64_u64(wrap_indicator, mult);
2019
2020         if (strwhole) {
2021                 rc = kstrtoull(strwhole, base, &whole);
2022                 if (rc)
2023                         return rc;
2024
2025                 if (whole > wrap_indicator)
2026                         return -ERANGE;
2027
2028                 whole *= mult;
2029         }
2030
2031         if (strfrac) {
2032                 if (strlen(strfrac) > 10)
2033                         strfrac[10] = '\0';
2034
2035                 rc = kstrtoull(strfrac, base, &frac);
2036                 if (rc)
2037                         return rc;
2038
2039                 /* determine power of fractional portion */
2040                 while (*strfrac) {
2041                         frac_d *= base;
2042                         strfrac++;
2043                 }
2044
2045                 /* fractional portion is too large to perform calculation */
2046                 if (frac > wrap_indicator)
2047                         return -ERANGE;
2048
2049                 frac *= mult;
2050                 do_div(frac, frac_d);
2051         }
2052
2053         /* check that the sum of whole and fraction fits in u64 */
2054         if (whole > (ULLONG_MAX - frac))
2055                 return -ERANGE;
2056
2057         *val = whole + frac;
2058
2059         return 0;
2060 }
2061
2062 /*
2063  * This function parses numeric/hex strings into __s64. It accepts a multiplier
2064  * which will apply to the value parsed. It also can allow the string to
2065  * have a unit as the last character. The function handles overflow/underflow
2066  * of the signed integer.
2067  */
2068 static int str_to_s64_internal(const char __user *buffer, unsigned long count,
2069                                __s64 *val, __u64 def_mult, bool allow_units)
2070 {
2071         char kernbuf[22];
2072         __u64 tmp;
2073         unsigned int offset = 0;
2074         int signed sign = 1;
2075         __u64 max = LLONG_MAX;
2076         int rc = 0;
2077
2078         if (count > (sizeof(kernbuf) - 1))
2079                 return -EINVAL;
2080
2081         if (copy_from_user(kernbuf, buffer, count))
2082                 return -EFAULT;
2083
2084         kernbuf[count] = '\0';
2085
2086         /* keep track of our sign */
2087         if (*kernbuf == '-') {
2088                 sign = -1;
2089                 offset++;
2090                 /* equivalent to max = -LLONG_MIN, avoids overflow */
2091                 max++;
2092         }
2093
2094         rc = str_to_u64_parse(kernbuf + offset, count - offset,
2095                               &tmp, def_mult, allow_units);
2096         if (rc)
2097                 return rc;
2098
2099         /* check for overflow/underflow */
2100         if (max < tmp)
2101                 return -ERANGE;
2102
2103         *val = (__s64)tmp * sign;
2104
2105         return 0;
2106 }
2107
2108 /**
2109  * Convert a user string into a signed 64 bit number. This function produces
2110  * an error when the value parsed from the string times multiplier underflows or
2111  * overflows. This function only accepts strings that contains digits, an
2112  * optional decimal, and a char representing a unit at the end. If a unit is
2113  * specified in the string, the multiplier provided by the caller is ignored.
2114  * This function can also accept hexadecimal strings which are prefixed with
2115  * "0x".
2116  *
2117  * \param[in] buffer    string consisting of numbers, a decimal, and a unit
2118  * \param[in] count     buffer length
2119  * \param[in] val       if successful, the value represented by the string
2120  * \param[in] defunit   default unit if string doesn't contain one
2121  *
2122  * \retval              0 on success
2123  * \retval              negative number on error
2124  */
2125 int lprocfs_str_with_units_to_s64(const char __user *buffer,
2126                                   unsigned long count, __s64 *val, char defunit)
2127 {
2128         __u64 mult = 1;
2129         int rc;
2130
2131         if (defunit != '1') {
2132                 rc = get_mult(defunit, &mult);
2133                 if (rc)
2134                         return rc;
2135         }
2136
2137         return str_to_s64_internal(buffer, count, val, mult, true);
2138 }
2139 EXPORT_SYMBOL(lprocfs_str_with_units_to_s64);
2140
2141 char *lprocfs_strnstr(const char *s1, const char *s2, size_t len)
2142 {
2143         size_t l2;
2144
2145         l2 = strlen(s2);
2146         if (!l2)
2147                 return (char *)s1;
2148         while (len >= l2) {
2149                 len--;
2150                 if (!memcmp(s1, s2, l2))
2151                         return (char *)s1;
2152                 s1++;
2153         }
2154         return NULL;
2155 }
2156 EXPORT_SYMBOL(lprocfs_strnstr);
2157
2158 /**
2159  * Find the string \a name in the input \a buffer, and return a pointer to the
2160  * value immediately following \a name, reducing \a count appropriately.
2161  * If \a name is not found the original \a buffer is returned.
2162  */
2163 char *lprocfs_find_named_value(const char *buffer, const char *name,
2164                                 size_t *count)
2165 {
2166         char *val;
2167         size_t buflen = *count;
2168
2169         /* there is no strnstr() in rhel5 and ubuntu kernels */
2170         val = lprocfs_strnstr(buffer, name, buflen);
2171         if (!val)
2172                 return (char *)buffer;
2173
2174         val += strlen(name);                             /* skip prefix */
2175         while (val < buffer + buflen && isspace(*val)) /* skip separator */
2176                 val++;
2177
2178         *count = 0;
2179         while (val < buffer + buflen && isalnum(*val)) {
2180                 ++*count;
2181                 ++val;
2182         }
2183
2184         return val - *count;
2185 }
2186 EXPORT_SYMBOL(lprocfs_find_named_value);
2187
2188 int ldebugfs_seq_create(struct dentry *parent, const char *name, umode_t mode,
2189                         const struct file_operations *seq_fops, void *data)
2190 {
2191         struct dentry *entry;
2192
2193         /* Disallow secretly (un)writable entries. */
2194         LASSERT((!seq_fops->write) == (!(mode & 0222)));
2195
2196         entry = debugfs_create_file(name, mode, parent, data, seq_fops);
2197         if (IS_ERR_OR_NULL(entry))
2198                 return entry ? PTR_ERR(entry) : -ENOMEM;
2199
2200         return 0;
2201 }
2202 EXPORT_SYMBOL_GPL(ldebugfs_seq_create);
2203
2204 int lprocfs_seq_create(struct proc_dir_entry *parent,
2205                        const char *name,
2206                        mode_t mode,
2207                        const struct file_operations *seq_fops,
2208                        void *data)
2209 {
2210         struct proc_dir_entry *entry;
2211         ENTRY;
2212
2213         /* Disallow secretly (un)writable entries. */
2214         LASSERT((seq_fops->write == NULL) == ((mode & 0222) == 0));
2215
2216         entry = proc_create_data(name, mode, parent, seq_fops, data);
2217
2218         if (!entry)
2219                 RETURN(-ENOMEM);
2220
2221         RETURN(0);
2222 }
2223 EXPORT_SYMBOL(lprocfs_seq_create);
2224
2225 int lprocfs_obd_seq_create(struct obd_device *dev,
2226                            const char *name,
2227                            mode_t mode,
2228                            const struct file_operations *seq_fops,
2229                            void *data)
2230 {
2231         return (lprocfs_seq_create(dev->obd_proc_entry, name,
2232                                    mode, seq_fops, data));
2233 }
2234 EXPORT_SYMBOL(lprocfs_obd_seq_create);
2235
2236 void lprocfs_oh_tally(struct obd_histogram *oh, unsigned int value)
2237 {
2238         if (value >= OBD_HIST_MAX)
2239                 value = OBD_HIST_MAX - 1;
2240
2241         spin_lock(&oh->oh_lock);
2242         oh->oh_buckets[value]++;
2243         spin_unlock(&oh->oh_lock);
2244 }
2245 EXPORT_SYMBOL(lprocfs_oh_tally);
2246
2247 void lprocfs_oh_tally_log2(struct obd_histogram *oh, unsigned int value)
2248 {
2249         unsigned int val = 0;
2250
2251         if (likely(value != 0))
2252                 val = min(fls(value - 1), OBD_HIST_MAX);
2253
2254         lprocfs_oh_tally(oh, val);
2255 }
2256 EXPORT_SYMBOL(lprocfs_oh_tally_log2);
2257
2258 unsigned long lprocfs_oh_sum(struct obd_histogram *oh)
2259 {
2260         unsigned long ret = 0;
2261         int i;
2262
2263         for (i = 0; i < OBD_HIST_MAX; i++)
2264                 ret +=  oh->oh_buckets[i];
2265         return ret;
2266 }
2267 EXPORT_SYMBOL(lprocfs_oh_sum);
2268
2269 void lprocfs_oh_clear(struct obd_histogram *oh)
2270 {
2271         spin_lock(&oh->oh_lock);
2272         memset(oh->oh_buckets, 0, sizeof(oh->oh_buckets));
2273         spin_unlock(&oh->oh_lock);
2274 }
2275 EXPORT_SYMBOL(lprocfs_oh_clear);
2276
2277 ssize_t lustre_attr_show(struct kobject *kobj,
2278                          struct attribute *attr, char *buf)
2279 {
2280         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
2281
2282         return a->show ? a->show(kobj, attr, buf) : 0;
2283 }
2284 EXPORT_SYMBOL_GPL(lustre_attr_show);
2285
2286 ssize_t lustre_attr_store(struct kobject *kobj, struct attribute *attr,
2287                           const char *buf, size_t len)
2288 {
2289         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
2290
2291         return a->store ? a->store(kobj, attr, buf, len) : len;
2292 }
2293 EXPORT_SYMBOL_GPL(lustre_attr_store);
2294
2295 const struct sysfs_ops lustre_sysfs_ops = {
2296         .show  = lustre_attr_show,
2297         .store = lustre_attr_store,
2298 };
2299 EXPORT_SYMBOL_GPL(lustre_sysfs_ops);
2300
2301 int lprocfs_obd_max_pages_per_rpc_seq_show(struct seq_file *m, void *data)
2302 {
2303         struct obd_device *dev = data;
2304         struct client_obd *cli = &dev->u.cli;
2305
2306         spin_lock(&cli->cl_loi_list_lock);
2307         seq_printf(m, "%d\n", cli->cl_max_pages_per_rpc);
2308         spin_unlock(&cli->cl_loi_list_lock);
2309         return 0;
2310 }
2311 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_show);
2312
2313 ssize_t lprocfs_obd_max_pages_per_rpc_seq_write(struct file *file,
2314                                                 const char __user *buffer,
2315                                                 size_t count, loff_t *off)
2316 {
2317         struct obd_device *dev =
2318                 ((struct seq_file *)file->private_data)->private;
2319         struct client_obd *cli = &dev->u.cli;
2320         struct obd_connect_data *ocd = &cli->cl_import->imp_connect_data;
2321         int chunk_mask, rc;
2322         s64 val;
2323
2324         rc = lprocfs_str_with_units_to_s64(buffer, count, &val, '1');
2325         if (rc)
2326                 return rc;
2327         if (val < 0)
2328                 return -ERANGE;
2329
2330         /* if the max_pages is specified in bytes, convert to pages */
2331         if (val >= ONE_MB_BRW_SIZE)
2332                 val >>= PAGE_SHIFT;
2333
2334         LPROCFS_CLIMP_CHECK(dev);
2335
2336         chunk_mask = ~((1 << (cli->cl_chunkbits - PAGE_SHIFT)) - 1);
2337         /* max_pages_per_rpc must be chunk aligned */
2338         val = (val + ~chunk_mask) & chunk_mask;
2339         if (val == 0 || (ocd->ocd_brw_size != 0 &&
2340                          val > ocd->ocd_brw_size >> PAGE_SHIFT)) {
2341                 LPROCFS_CLIMP_EXIT(dev);
2342                 return -ERANGE;
2343         }
2344         spin_lock(&cli->cl_loi_list_lock);
2345         cli->cl_max_pages_per_rpc = val;
2346         client_adjust_max_dirty(cli);
2347         spin_unlock(&cli->cl_loi_list_lock);
2348
2349         LPROCFS_CLIMP_EXIT(dev);
2350         return count;
2351 }
2352 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_write);
2353
2354 ssize_t short_io_bytes_show(struct kobject *kobj, struct attribute *attr,
2355                             char *buf)
2356 {
2357         struct obd_device *dev = container_of(kobj, struct obd_device,
2358                                               obd_kset.kobj);
2359         struct client_obd *cli = &dev->u.cli;
2360         int rc;
2361
2362         spin_lock(&cli->cl_loi_list_lock);
2363         rc = sprintf(buf, "%d\n", cli->cl_max_short_io_bytes);
2364         spin_unlock(&cli->cl_loi_list_lock);
2365         return rc;
2366 }
2367 EXPORT_SYMBOL(short_io_bytes_show);
2368
2369 /* Used to catch people who think they're specifying pages. */
2370 #define MIN_SHORT_IO_BYTES 64U
2371
2372 ssize_t short_io_bytes_store(struct kobject *kobj, struct attribute *attr,
2373                              const char *buffer, size_t count)
2374 {
2375         struct obd_device *dev = container_of(kobj, struct obd_device,
2376                                               obd_kset.kobj);
2377         struct client_obd *cli = &dev->u.cli;
2378         u32 val;
2379         int rc;
2380
2381         LPROCFS_CLIMP_CHECK(dev);
2382
2383         rc = kstrtouint(buffer, 0, &val);
2384         if (rc)
2385                 GOTO(out, rc);
2386
2387         if (val && (val < MIN_SHORT_IO_BYTES || val > OBD_MAX_SHORT_IO_BYTES))
2388                 GOTO(out, rc = -ERANGE);
2389
2390         rc = count;
2391
2392         spin_lock(&cli->cl_loi_list_lock);
2393         if (val > (cli->cl_max_pages_per_rpc << PAGE_SHIFT))
2394                 rc = -ERANGE;
2395         else
2396                 cli->cl_max_short_io_bytes = val;
2397         spin_unlock(&cli->cl_loi_list_lock);
2398
2399 out:
2400         LPROCFS_CLIMP_EXIT(dev);
2401         return rc;
2402 }
2403 EXPORT_SYMBOL(short_io_bytes_store);
2404
2405 int lprocfs_wr_root_squash(const char __user *buffer, unsigned long count,
2406                            struct root_squash_info *squash, char *name)
2407 {
2408         int rc;
2409         char kernbuf[64], *tmp, *errmsg;
2410         unsigned long uid, gid;
2411         ENTRY;
2412
2413         if (count >= sizeof(kernbuf)) {
2414                 errmsg = "string too long";
2415                 GOTO(failed_noprint, rc = -EINVAL);
2416         }
2417         if (copy_from_user(kernbuf, buffer, count)) {
2418                 errmsg = "bad address";
2419                 GOTO(failed_noprint, rc = -EFAULT);
2420         }
2421         kernbuf[count] = '\0';
2422
2423         /* look for uid gid separator */
2424         tmp = strchr(kernbuf, ':');
2425         if (!tmp) {
2426                 errmsg = "needs uid:gid format";
2427                 GOTO(failed, rc = -EINVAL);
2428         }
2429         *tmp = '\0';
2430         tmp++;
2431
2432         /* parse uid */
2433         if (kstrtoul(kernbuf, 0, &uid) != 0) {
2434                 errmsg = "bad uid";
2435                 GOTO(failed, rc = -EINVAL);
2436         }
2437
2438         /* parse gid */
2439         if (kstrtoul(tmp, 0, &gid) != 0) {
2440                 errmsg = "bad gid";
2441                 GOTO(failed, rc = -EINVAL);
2442         }
2443
2444         squash->rsi_uid = uid;
2445         squash->rsi_gid = gid;
2446
2447         LCONSOLE_INFO("%s: root_squash is set to %u:%u\n",
2448                       name, squash->rsi_uid, squash->rsi_gid);
2449         RETURN(count);
2450
2451 failed:
2452         if (tmp) {
2453                 tmp--;
2454                 *tmp = ':';
2455         }
2456         CWARN("%s: failed to set root_squash to \"%s\", %s, rc = %d\n",
2457               name, kernbuf, errmsg, rc);
2458         RETURN(rc);
2459 failed_noprint:
2460         CWARN("%s: failed to set root_squash due to %s, rc = %d\n",
2461               name, errmsg, rc);
2462         RETURN(rc);
2463 }
2464 EXPORT_SYMBOL(lprocfs_wr_root_squash);
2465
2466
2467 int lprocfs_wr_nosquash_nids(const char __user *buffer, unsigned long count,
2468                              struct root_squash_info *squash, char *name)
2469 {
2470         int rc;
2471         char *kernbuf = NULL;
2472         char *errmsg;
2473         struct list_head tmp;
2474         int len = count;
2475         ENTRY;
2476
2477         if (count > 4096) {
2478                 errmsg = "string too long";
2479                 GOTO(failed, rc = -EINVAL);
2480         }
2481
2482         OBD_ALLOC(kernbuf, count + 1);
2483         if (!kernbuf) {
2484                 errmsg = "no memory";
2485                 GOTO(failed, rc = -ENOMEM);
2486         }
2487         if (copy_from_user(kernbuf, buffer, count)) {
2488                 errmsg = "bad address";
2489                 GOTO(failed, rc = -EFAULT);
2490         }
2491         kernbuf[count] = '\0';
2492
2493         if (count > 0 && kernbuf[count - 1] == '\n')
2494                 len = count - 1;
2495
2496         if ((len == 4 && strncmp(kernbuf, "NONE", len) == 0) ||
2497             (len == 5 && strncmp(kernbuf, "clear", len) == 0)) {
2498                 /* empty string is special case */
2499                 down_write(&squash->rsi_sem);
2500                 if (!list_empty(&squash->rsi_nosquash_nids))
2501                         cfs_free_nidlist(&squash->rsi_nosquash_nids);
2502                 up_write(&squash->rsi_sem);
2503                 LCONSOLE_INFO("%s: nosquash_nids is cleared\n", name);
2504                 OBD_FREE(kernbuf, count + 1);
2505                 RETURN(count);
2506         }
2507
2508         INIT_LIST_HEAD(&tmp);
2509         if (cfs_parse_nidlist(kernbuf, count, &tmp) <= 0) {
2510                 errmsg = "can't parse";
2511                 GOTO(failed, rc = -EINVAL);
2512         }
2513         LCONSOLE_INFO("%s: nosquash_nids set to %s\n",
2514                       name, kernbuf);
2515         OBD_FREE(kernbuf, count + 1);
2516         kernbuf = NULL;
2517
2518         down_write(&squash->rsi_sem);
2519         if (!list_empty(&squash->rsi_nosquash_nids))
2520                 cfs_free_nidlist(&squash->rsi_nosquash_nids);
2521         list_splice(&tmp, &squash->rsi_nosquash_nids);
2522         up_write(&squash->rsi_sem);
2523
2524         RETURN(count);
2525
2526 failed:
2527         if (kernbuf) {
2528                 CWARN("%s: failed to set nosquash_nids to \"%s\", %s rc = %d\n",
2529                       name, kernbuf, errmsg, rc);
2530                 OBD_FREE(kernbuf, count + 1);
2531         } else {
2532                 CWARN("%s: failed to set nosquash_nids due to %s rc = %d\n",
2533                       name, errmsg, rc);
2534         }
2535         RETURN(rc);
2536 }
2537 EXPORT_SYMBOL(lprocfs_wr_nosquash_nids);
2538
2539 #endif /* CONFIG_PROC_FS*/