Whamcloud - gitweb
LU-11834 llite: fix temporary instance buffer size
[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         "selinux_policy",       /* 0x200 */
786         NULL
787 };
788
789 void obd_connect_seq_flags2str(struct seq_file *m, __u64 flags, __u64 flags2,
790                                const char *sep)
791 {
792         bool first = true;
793         __u64 mask;
794         int i;
795
796         for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
797                 if (flags & mask) {
798                         seq_printf(m, "%s%s",
799                                    first ? "" : sep, obd_connect_names[i]);
800                         first = false;
801                 }
802         }
803
804         if (flags & ~(mask - 1)) {
805                 seq_printf(m, "%sunknown_%#llx",
806                            first ? "" : sep, flags & ~(mask - 1));
807                 first = false;
808         }
809
810         if (!(flags & OBD_CONNECT_FLAGS2) || flags2 == 0)
811                 return;
812
813         for (i = 64, mask = 1; obd_connect_names[i] != NULL; i++, mask <<= 1) {
814                 if (flags2 & mask) {
815                         seq_printf(m, "%s%s",
816                                    first ? "" : sep, obd_connect_names[i]);
817                         first = false;
818                 }
819         }
820
821         if (flags2 & ~(mask - 1)) {
822                 seq_printf(m, "%sunknown2_%#llx",
823                            first ? "" : sep, flags2 & ~(mask - 1));
824                 first = false;
825         }
826 }
827 EXPORT_SYMBOL(obd_connect_seq_flags2str);
828
829 int obd_connect_flags2str(char *page, int count, __u64 flags, __u64 flags2,
830                           const char *sep)
831 {
832         __u64 mask;
833         int i, ret = 0;
834
835         for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
836                 if (flags & mask)
837                         ret += snprintf(page + ret, count - ret, "%s%s",
838                                         ret ? sep : "", obd_connect_names[i]);
839         }
840
841         if (flags & ~(mask - 1))
842                 ret += snprintf(page + ret, count - ret,
843                                 "%sunknown_%#llx",
844                                 ret ? sep : "", flags & ~(mask - 1));
845
846         if (!(flags & OBD_CONNECT_FLAGS2) || flags2 == 0)
847                 return ret;
848
849         for (i = 64, mask = 1; obd_connect_names[i] != NULL; i++, mask <<= 1) {
850                 if (flags2 & mask)
851                         ret += snprintf(page + ret, count - ret, "%s%s",
852                                         ret ? sep : "", obd_connect_names[i]);
853         }
854
855         if (flags2 & ~(mask - 1))
856                 ret += snprintf(page + ret, count - ret,
857                                 "%sunknown2_%#llx",
858                                 ret ? sep : "", flags2 & ~(mask - 1));
859
860         return ret;
861 }
862 EXPORT_SYMBOL(obd_connect_flags2str);
863
864 void
865 obd_connect_data_seqprint(struct seq_file *m, struct obd_connect_data *ocd)
866 {
867         __u64 flags;
868
869         LASSERT(ocd != NULL);
870         flags = ocd->ocd_connect_flags;
871
872         seq_printf(m, "    connect_data:\n"
873                    "       flags: %#llx\n"
874                    "       instance: %u\n",
875                    ocd->ocd_connect_flags,
876                    ocd->ocd_instance);
877         if (flags & OBD_CONNECT_VERSION)
878                 seq_printf(m, "       target_version: %u.%u.%u.%u\n",
879                            OBD_OCD_VERSION_MAJOR(ocd->ocd_version),
880                            OBD_OCD_VERSION_MINOR(ocd->ocd_version),
881                            OBD_OCD_VERSION_PATCH(ocd->ocd_version),
882                            OBD_OCD_VERSION_FIX(ocd->ocd_version));
883         if (flags & OBD_CONNECT_MDS)
884                 seq_printf(m, "       mdt_index: %d\n", ocd->ocd_group);
885         if (flags & OBD_CONNECT_GRANT)
886                 seq_printf(m, "       initial_grant: %d\n", ocd->ocd_grant);
887         if (flags & OBD_CONNECT_INDEX)
888                 seq_printf(m, "       target_index: %u\n", ocd->ocd_index);
889         if (flags & OBD_CONNECT_BRW_SIZE)
890                 seq_printf(m, "       max_brw_size: %d\n", ocd->ocd_brw_size);
891         if (flags & OBD_CONNECT_IBITS)
892                 seq_printf(m, "       ibits_known: %#llx\n",
893                            ocd->ocd_ibits_known);
894         if (flags & OBD_CONNECT_GRANT_PARAM)
895                 seq_printf(m, "       grant_block_size: %d\n"
896                            "       grant_inode_size: %d\n"
897                            "       grant_max_extent_size: %d\n"
898                            "       grant_extent_tax: %d\n",
899                            1 << ocd->ocd_grant_blkbits,
900                            1 << ocd->ocd_grant_inobits,
901                            ocd->ocd_grant_max_blks << ocd->ocd_grant_blkbits,
902                            ocd->ocd_grant_tax_kb << 10);
903         if (flags & OBD_CONNECT_TRANSNO)
904                 seq_printf(m, "       first_transno: %#llx\n",
905                            ocd->ocd_transno);
906         if (flags & OBD_CONNECT_CKSUM)
907                 seq_printf(m, "       cksum_types: %#x\n",
908                            ocd->ocd_cksum_types);
909         if (flags & OBD_CONNECT_MAX_EASIZE)
910                 seq_printf(m, "       max_easize: %d\n", ocd->ocd_max_easize);
911         if (flags & OBD_CONNECT_MAXBYTES)
912                 seq_printf(m, "       max_object_bytes: %llu\n",
913                            ocd->ocd_maxbytes);
914         if (flags & OBD_CONNECT_MULTIMODRPCS)
915                 seq_printf(m, "       max_mod_rpcs: %hu\n",
916                            ocd->ocd_maxmodrpcs);
917 }
918
919 int lprocfs_import_seq_show(struct seq_file *m, void *data)
920 {
921         char nidstr[LNET_NIDSTR_SIZE];
922         struct lprocfs_counter ret;
923         struct lprocfs_counter_header *header;
924         struct obd_device *obd = (struct obd_device *)data;
925         struct obd_import *imp;
926         struct obd_import_conn *conn;
927         struct obd_connect_data *ocd;
928         int j;
929         int k;
930         int rw = 0;
931
932         LASSERT(obd != NULL);
933         LPROCFS_CLIMP_CHECK(obd);
934         imp = obd->u.cli.cl_import;
935         ocd = &imp->imp_connect_data;
936
937         seq_printf(m, "import:\n"
938                    "    name: %s\n"
939                    "    target: %s\n"
940                    "    state: %s\n"
941                    "    connect_flags: [ ",
942                    obd->obd_name,
943                    obd2cli_tgt(obd),
944                    ptlrpc_import_state_name(imp->imp_state));
945         obd_connect_seq_flags2str(m, imp->imp_connect_data.ocd_connect_flags,
946                                   imp->imp_connect_data.ocd_connect_flags2,
947                                   ", ");
948         seq_printf(m, " ]\n");
949         obd_connect_data_seqprint(m, ocd);
950         seq_printf(m, "    import_flags: [ ");
951         obd_import_flags2str(imp, m);
952
953         seq_printf(m, " ]\n"
954                    "    connection:\n"
955                    "       failover_nids: [ ");
956         spin_lock(&imp->imp_lock);
957         j = 0;
958         list_for_each_entry(conn, &imp->imp_conn_list, oic_item) {
959                 libcfs_nid2str_r(conn->oic_conn->c_peer.nid,
960                                  nidstr, sizeof(nidstr));
961                 seq_printf(m, "%s%s", j ? ", " : "", nidstr);
962                 j++;
963         }
964         if (imp->imp_connection)
965                 libcfs_nid2str_r(imp->imp_connection->c_peer.nid,
966                                  nidstr, sizeof(nidstr));
967         else
968                 strncpy(nidstr, "<none>", sizeof(nidstr));
969         seq_printf(m, " ]\n"
970                    "       current_connection: %s\n"
971                    "       connection_attempts: %u\n"
972                    "       generation: %u\n"
973                    "       in-progress_invalidations: %u\n"
974                    "       idle: %lld sec\n",
975                    nidstr,
976                    imp->imp_conn_cnt,
977                    imp->imp_generation,
978                    atomic_read(&imp->imp_inval_count),
979                    ktime_get_real_seconds() - imp->imp_last_reply_time);
980         spin_unlock(&imp->imp_lock);
981
982         if (!obd->obd_svc_stats)
983                 goto out_climp;
984
985         header = &obd->obd_svc_stats->ls_cnt_header[PTLRPC_REQWAIT_CNTR];
986         lprocfs_stats_collect(obd->obd_svc_stats, PTLRPC_REQWAIT_CNTR, &ret);
987         if (ret.lc_count != 0) {
988                 /* first argument to do_div MUST be __u64 */
989                 __u64 sum = ret.lc_sum;
990                 do_div(sum, ret.lc_count);
991                 ret.lc_sum = sum;
992         } else
993                 ret.lc_sum = 0;
994         seq_printf(m, "    rpcs:\n"
995                    "       inflight: %u\n"
996                    "       unregistering: %u\n"
997                    "       timeouts: %u\n"
998                    "       avg_waittime: %llu %s\n",
999                    atomic_read(&imp->imp_inflight),
1000                    atomic_read(&imp->imp_unregistering),
1001                    atomic_read(&imp->imp_timeouts),
1002                    ret.lc_sum, header->lc_units);
1003
1004         k = 0;
1005         for(j = 0; j < IMP_AT_MAX_PORTALS; j++) {
1006                 if (imp->imp_at.iat_portal[j] == 0)
1007                         break;
1008                 k = max_t(unsigned int, k,
1009                           at_get(&imp->imp_at.iat_service_estimate[j]));
1010         }
1011         seq_printf(m, "    service_estimates:\n"
1012                    "       services: %u sec\n"
1013                    "       network: %u sec\n",
1014                    k,
1015                    at_get(&imp->imp_at.iat_net_latency));
1016
1017         seq_printf(m, "    transactions:\n"
1018                    "       last_replay: %llu\n"
1019                    "       peer_committed: %llu\n"
1020                    "       last_checked: %llu\n",
1021                    imp->imp_last_replay_transno,
1022                    imp->imp_peer_committed_transno,
1023                    imp->imp_last_transno_checked);
1024
1025         /* avg data rates */
1026         for (rw = 0; rw <= 1; rw++) {
1027                 lprocfs_stats_collect(obd->obd_svc_stats,
1028                                       PTLRPC_LAST_CNTR + BRW_READ_BYTES + rw,
1029                                       &ret);
1030                 if (ret.lc_sum > 0 && ret.lc_count > 0) {
1031                         /* first argument to do_div MUST be __u64 */
1032                         __u64 sum = ret.lc_sum;
1033                         do_div(sum, ret.lc_count);
1034                         ret.lc_sum = sum;
1035                         seq_printf(m, "    %s_data_averages:\n"
1036                                    "       bytes_per_rpc: %llu\n",
1037                                    rw ? "write" : "read",
1038                                    ret.lc_sum);
1039                 }
1040                 k = (int)ret.lc_sum;
1041                 j = opcode_offset(OST_READ + rw) + EXTRA_MAX_OPCODES;
1042                 header = &obd->obd_svc_stats->ls_cnt_header[j];
1043                 lprocfs_stats_collect(obd->obd_svc_stats, j, &ret);
1044                 if (ret.lc_sum > 0 && ret.lc_count != 0) {
1045                         /* first argument to do_div MUST be __u64 */
1046                         __u64 sum = ret.lc_sum;
1047                         do_div(sum, ret.lc_count);
1048                         ret.lc_sum = sum;
1049                         seq_printf(m, "       %s_per_rpc: %llu\n",
1050                                    header->lc_units, ret.lc_sum);
1051                         j = (int)ret.lc_sum;
1052                         if (j > 0)
1053                                 seq_printf(m, "       MB_per_sec: %u.%.02u\n",
1054                                            k / j, (100 * k / j) % 100);
1055                 }
1056         }
1057
1058 out_climp:
1059         LPROCFS_CLIMP_EXIT(obd);
1060         return 0;
1061 }
1062 EXPORT_SYMBOL(lprocfs_import_seq_show);
1063
1064 int lprocfs_state_seq_show(struct seq_file *m, void *data)
1065 {
1066         struct obd_device *obd = (struct obd_device *)data;
1067         struct obd_import *imp;
1068         int j, k;
1069
1070         LASSERT(obd != NULL);
1071         LPROCFS_CLIMP_CHECK(obd);
1072         imp = obd->u.cli.cl_import;
1073
1074         seq_printf(m, "current_state: %s\n",
1075                    ptlrpc_import_state_name(imp->imp_state));
1076         seq_printf(m, "state_history:\n");
1077         k = imp->imp_state_hist_idx;
1078         for (j = 0; j < IMP_STATE_HIST_LEN; j++) {
1079                 struct import_state_hist *ish =
1080                         &imp->imp_state_hist[(k + j) % IMP_STATE_HIST_LEN];
1081                 if (ish->ish_state == 0)
1082                         continue;
1083                 seq_printf(m, " - [ %lld, %s ]\n", (s64)ish->ish_time,
1084                            ptlrpc_import_state_name(ish->ish_state));
1085         }
1086
1087         LPROCFS_CLIMP_EXIT(obd);
1088         return 0;
1089 }
1090 EXPORT_SYMBOL(lprocfs_state_seq_show);
1091
1092 int lprocfs_at_hist_helper(struct seq_file *m, struct adaptive_timeout *at)
1093 {
1094         int i;
1095         for (i = 0; i < AT_BINS; i++)
1096                 seq_printf(m, "%3u ", at->at_hist[i]);
1097         seq_printf(m, "\n");
1098         return 0;
1099 }
1100 EXPORT_SYMBOL(lprocfs_at_hist_helper);
1101
1102 /* See also ptlrpc_lprocfs_timeouts_show_seq */
1103 int lprocfs_timeouts_seq_show(struct seq_file *m, void *data)
1104 {
1105         struct obd_device *obd = (struct obd_device *)data;
1106         struct obd_import *imp;
1107         unsigned int cur, worst;
1108         time64_t now, worstt;
1109         int i;
1110
1111         LASSERT(obd != NULL);
1112         LPROCFS_CLIMP_CHECK(obd);
1113         imp = obd->u.cli.cl_import;
1114
1115         now = ktime_get_real_seconds();
1116
1117         /* Some network health info for kicks */
1118         seq_printf(m, "%-10s : %lld, %llds ago\n",
1119                    "last reply", (s64)imp->imp_last_reply_time,
1120                    (s64)(now - imp->imp_last_reply_time));
1121
1122         cur = at_get(&imp->imp_at.iat_net_latency);
1123         worst = imp->imp_at.iat_net_latency.at_worst_ever;
1124         worstt = imp->imp_at.iat_net_latency.at_worst_time;
1125         seq_printf(m, "%-10s : cur %3u  worst %3u (at %lld, %llds ago) ",
1126                    "network", cur, worst, (s64)worstt, (s64)(now - worstt));
1127         lprocfs_at_hist_helper(m, &imp->imp_at.iat_net_latency);
1128
1129         for(i = 0; i < IMP_AT_MAX_PORTALS; i++) {
1130                 if (imp->imp_at.iat_portal[i] == 0)
1131                         break;
1132                 cur = at_get(&imp->imp_at.iat_service_estimate[i]);
1133                 worst = imp->imp_at.iat_service_estimate[i].at_worst_ever;
1134                 worstt = imp->imp_at.iat_service_estimate[i].at_worst_time;
1135                 seq_printf(m, "portal %-2d  : cur %3u  worst %3u (at %lld, %llds ago) ",
1136                            imp->imp_at.iat_portal[i], cur, worst, (s64)worstt,
1137                            (s64)(now - worstt));
1138                 lprocfs_at_hist_helper(m, &imp->imp_at.iat_service_estimate[i]);
1139         }
1140
1141         LPROCFS_CLIMP_EXIT(obd);
1142         return 0;
1143 }
1144 EXPORT_SYMBOL(lprocfs_timeouts_seq_show);
1145
1146 int lprocfs_connect_flags_seq_show(struct seq_file *m, void *data)
1147 {
1148         struct obd_device *obd = data;
1149         __u64 flags;
1150         __u64 flags2;
1151
1152         LPROCFS_CLIMP_CHECK(obd);
1153         flags = obd->u.cli.cl_import->imp_connect_data.ocd_connect_flags;
1154         flags2 = obd->u.cli.cl_import->imp_connect_data.ocd_connect_flags2;
1155         seq_printf(m, "flags=%#llx\n", flags);
1156         seq_printf(m, "flags2=%#llx\n", flags2);
1157         obd_connect_seq_flags2str(m, flags, flags2, "\n");
1158         seq_printf(m, "\n");
1159         LPROCFS_CLIMP_EXIT(obd);
1160         return 0;
1161 }
1162 EXPORT_SYMBOL(lprocfs_connect_flags_seq_show);
1163
1164 static const struct attribute *obd_def_uuid_attrs[] = {
1165         &lustre_attr_uuid.attr,
1166         NULL,
1167 };
1168
1169 static const struct attribute *obd_def_attrs[] = {
1170         &lustre_attr_blocksize.attr,
1171         &lustre_attr_kbytestotal.attr,
1172         &lustre_attr_kbytesfree.attr,
1173         &lustre_attr_kbytesavail.attr,
1174         &lustre_attr_filestotal.attr,
1175         &lustre_attr_filesfree.attr,
1176         &lustre_attr_uuid.attr,
1177         NULL,
1178 };
1179
1180 static void obd_sysfs_release(struct kobject *kobj)
1181 {
1182         struct obd_device *obd = container_of(kobj, struct obd_device,
1183                                               obd_kset.kobj);
1184
1185         complete(&obd->obd_kobj_unregister);
1186 }
1187
1188 int lprocfs_obd_setup(struct obd_device *obd, bool uuid_only)
1189 {
1190         struct lprocfs_vars *debugfs_vars = NULL;
1191         int rc;
1192
1193         if (!obd || obd->obd_magic != OBD_DEVICE_MAGIC)
1194                 return -ENODEV;
1195
1196         rc = kobject_set_name(&obd->obd_kset.kobj, "%s", obd->obd_name);
1197         if (rc)
1198                 return rc;
1199
1200         obd->obd_ktype.sysfs_ops = &lustre_sysfs_ops;
1201         obd->obd_ktype.release = obd_sysfs_release;
1202
1203         obd->obd_kset.kobj.parent = obd->obd_type->typ_kobj;
1204         obd->obd_kset.kobj.ktype = &obd->obd_ktype;
1205         init_completion(&obd->obd_kobj_unregister);
1206         rc = kset_register(&obd->obd_kset);
1207         if (rc)
1208                 return rc;
1209
1210         if (uuid_only)
1211                 obd->obd_attrs = obd_def_uuid_attrs;
1212         else
1213                 obd->obd_attrs = obd_def_attrs;
1214
1215         rc = sysfs_create_files(&obd->obd_kset.kobj, obd->obd_attrs);
1216         if (rc) {
1217                 kset_unregister(&obd->obd_kset);
1218                 return rc;
1219         }
1220
1221         if (!obd->obd_type->typ_procroot)
1222                 debugfs_vars = obd->obd_vars;
1223         obd->obd_debugfs_entry = ldebugfs_register(obd->obd_name,
1224                                                    obd->obd_type->typ_debugfs_entry,
1225                                                    debugfs_vars, obd);
1226         if (IS_ERR_OR_NULL(obd->obd_debugfs_entry)) {
1227                 rc = obd->obd_debugfs_entry ? PTR_ERR(obd->obd_debugfs_entry)
1228                                             : -ENOMEM;
1229                 CERROR("error %d setting up debugfs for %s\n",
1230                        rc, obd->obd_name);
1231                 obd->obd_debugfs_entry = NULL;
1232
1233                 sysfs_remove_files(&obd->obd_kset.kobj, obd->obd_attrs);
1234                 obd->obd_attrs = NULL;
1235                 kset_unregister(&obd->obd_kset);
1236                 return rc;
1237         }
1238
1239         if (obd->obd_proc_entry || !obd->obd_type->typ_procroot)
1240                 GOTO(already_registered, rc);
1241
1242         obd->obd_proc_entry = lprocfs_register(obd->obd_name,
1243                                                obd->obd_type->typ_procroot,
1244                                                obd->obd_vars, obd);
1245         if (IS_ERR(obd->obd_proc_entry)) {
1246                 rc = PTR_ERR(obd->obd_proc_entry);
1247                 CERROR("error %d setting up lprocfs for %s\n",rc,obd->obd_name);
1248                 obd->obd_proc_entry = NULL;
1249
1250                 ldebugfs_remove(&obd->obd_debugfs_entry);
1251                 sysfs_remove_files(&obd->obd_kset.kobj, obd->obd_attrs);
1252                 obd->obd_attrs = NULL;
1253                 kset_unregister(&obd->obd_kset);
1254                 return rc;
1255         }
1256 already_registered:
1257         return rc;
1258 }
1259 EXPORT_SYMBOL(lprocfs_obd_setup);
1260
1261 int lprocfs_obd_cleanup(struct obd_device *obd)
1262 {
1263         if (!obd)
1264                 return -EINVAL;
1265
1266         if (obd->obd_proc_exports_entry) {
1267                 /* Should be no exports left */
1268                 lprocfs_remove(&obd->obd_proc_exports_entry);
1269                 obd->obd_proc_exports_entry = NULL;
1270         }
1271
1272         if (obd->obd_proc_entry) {
1273                 lprocfs_remove(&obd->obd_proc_entry);
1274                 obd->obd_proc_entry = NULL;
1275         }
1276
1277         if (!IS_ERR_OR_NULL(obd->obd_debugfs_entry))
1278                 ldebugfs_remove(&obd->obd_debugfs_entry);
1279
1280         /* obd device never allocated a kset */
1281         if (!obd->obd_kset.kobj.state_initialized)
1282                 return 0;
1283
1284         if (obd->obd_attrs) {
1285                 sysfs_remove_files(&obd->obd_kset.kobj, obd->obd_attrs);
1286                 obd->obd_attrs = NULL;
1287         }
1288
1289         kset_unregister(&obd->obd_kset);
1290         wait_for_completion(&obd->obd_kobj_unregister);
1291         return 0;
1292 }
1293 EXPORT_SYMBOL(lprocfs_obd_cleanup);
1294
1295 int lprocfs_stats_alloc_one(struct lprocfs_stats *stats, unsigned int cpuid)
1296 {
1297         struct lprocfs_counter *cntr;
1298         unsigned int percpusize;
1299         int rc = -ENOMEM;
1300         unsigned long flags = 0;
1301         int i;
1302
1303         LASSERT(stats->ls_percpu[cpuid] == NULL);
1304         LASSERT((stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) == 0);
1305
1306         percpusize = lprocfs_stats_counter_size(stats);
1307         LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[cpuid], percpusize);
1308         if (stats->ls_percpu[cpuid]) {
1309                 rc = 0;
1310                 if (unlikely(stats->ls_biggest_alloc_num <= cpuid)) {
1311                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1312                                 spin_lock_irqsave(&stats->ls_lock, flags);
1313                         else
1314                                 spin_lock(&stats->ls_lock);
1315                         if (stats->ls_biggest_alloc_num <= cpuid)
1316                                 stats->ls_biggest_alloc_num = cpuid + 1;
1317                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) {
1318                                 spin_unlock_irqrestore(&stats->ls_lock, flags);
1319                         } else {
1320                                 spin_unlock(&stats->ls_lock);
1321                         }
1322                 }
1323                 /* initialize the ls_percpu[cpuid] non-zero counter */
1324                 for (i = 0; i < stats->ls_num; ++i) {
1325                         cntr = lprocfs_stats_counter_get(stats, cpuid, i);
1326                         cntr->lc_min = LC_MIN_INIT;
1327                 }
1328         }
1329         return rc;
1330 }
1331
1332 struct lprocfs_stats *lprocfs_alloc_stats(unsigned int num,
1333                                           enum lprocfs_stats_flags flags)
1334 {
1335         struct lprocfs_stats *stats;
1336         unsigned int num_entry;
1337         unsigned int percpusize = 0;
1338         int i;
1339
1340         if (num == 0)
1341                 return NULL;
1342
1343         if (lprocfs_no_percpu_stats != 0)
1344                 flags |= LPROCFS_STATS_FLAG_NOPERCPU;
1345
1346         if (flags & LPROCFS_STATS_FLAG_NOPERCPU)
1347                 num_entry = 1;
1348         else
1349                 num_entry = num_possible_cpus();
1350
1351         /* alloc percpu pointers for all possible cpu slots */
1352         LIBCFS_ALLOC(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1353         if (!stats)
1354                 return NULL;
1355
1356         stats->ls_num = num;
1357         stats->ls_flags = flags;
1358         spin_lock_init(&stats->ls_lock);
1359
1360         /* alloc num of counter headers */
1361         LIBCFS_ALLOC(stats->ls_cnt_header,
1362                      stats->ls_num * sizeof(struct lprocfs_counter_header));
1363         if (!stats->ls_cnt_header)
1364                 goto fail;
1365
1366         if ((flags & LPROCFS_STATS_FLAG_NOPERCPU) != 0) {
1367                 /* contains only one set counters */
1368                 percpusize = lprocfs_stats_counter_size(stats);
1369                 LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[0], percpusize);
1370                 if (!stats->ls_percpu[0])
1371                         goto fail;
1372                 stats->ls_biggest_alloc_num = 1;
1373         } else if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0) {
1374                 /* alloc all percpu data, currently only obd_memory use this */
1375                 for (i = 0; i < num_entry; ++i)
1376                         if (lprocfs_stats_alloc_one(stats, i) < 0)
1377                                 goto fail;
1378         }
1379
1380         return stats;
1381
1382 fail:
1383         lprocfs_free_stats(&stats);
1384         return NULL;
1385 }
1386 EXPORT_SYMBOL(lprocfs_alloc_stats);
1387
1388 void lprocfs_free_stats(struct lprocfs_stats **statsh)
1389 {
1390         struct lprocfs_stats *stats = *statsh;
1391         unsigned int num_entry;
1392         unsigned int percpusize;
1393         unsigned int i;
1394
1395         if (!stats || stats->ls_num == 0)
1396                 return;
1397         *statsh = NULL;
1398
1399         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU)
1400                 num_entry = 1;
1401         else
1402                 num_entry = num_possible_cpus();
1403
1404         percpusize = lprocfs_stats_counter_size(stats);
1405         for (i = 0; i < num_entry; i++)
1406                 if (stats->ls_percpu[i])
1407                         LIBCFS_FREE(stats->ls_percpu[i], percpusize);
1408         if (stats->ls_cnt_header)
1409                 LIBCFS_FREE(stats->ls_cnt_header, stats->ls_num *
1410                                         sizeof(struct lprocfs_counter_header));
1411         LIBCFS_FREE(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1412 }
1413 EXPORT_SYMBOL(lprocfs_free_stats);
1414
1415 u64 lprocfs_stats_collector(struct lprocfs_stats *stats, int idx,
1416                             enum lprocfs_fields_flags field)
1417 {
1418         unsigned long flags = 0;
1419         unsigned int num_cpu;
1420         unsigned int i;
1421         u64 ret = 0;
1422
1423         LASSERT(stats);
1424
1425         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1426         for (i = 0; i < num_cpu; i++) {
1427                 struct lprocfs_counter *cntr;
1428
1429                 if (!stats->ls_percpu[i])
1430                         continue;
1431
1432                 cntr = lprocfs_stats_counter_get(stats, i, idx);
1433                 ret += lprocfs_read_helper(cntr, &stats->ls_cnt_header[idx],
1434                                            stats->ls_flags, field);
1435         }
1436         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1437         return ret;
1438 }
1439 EXPORT_SYMBOL(lprocfs_stats_collector);
1440
1441 void lprocfs_clear_stats(struct lprocfs_stats *stats)
1442 {
1443         struct lprocfs_counter *percpu_cntr;
1444         int i;
1445         int j;
1446         unsigned int num_entry;
1447         unsigned long flags = 0;
1448
1449         num_entry = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1450
1451         for (i = 0; i < num_entry; i++) {
1452                 if (!stats->ls_percpu[i])
1453                         continue;
1454                 for (j = 0; j < stats->ls_num; j++) {
1455                         percpu_cntr = lprocfs_stats_counter_get(stats, i, j);
1456                         percpu_cntr->lc_count           = 0;
1457                         percpu_cntr->lc_min             = LC_MIN_INIT;
1458                         percpu_cntr->lc_max             = 0;
1459                         percpu_cntr->lc_sumsquare       = 0;
1460                         percpu_cntr->lc_sum             = 0;
1461                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1462                                 percpu_cntr->lc_sum_irq = 0;
1463                 }
1464         }
1465
1466         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1467 }
1468 EXPORT_SYMBOL(lprocfs_clear_stats);
1469
1470 static ssize_t lprocfs_stats_seq_write(struct file *file,
1471                                        const char __user *buf,
1472                                        size_t len, loff_t *off)
1473 {
1474         struct seq_file *seq = file->private_data;
1475         struct lprocfs_stats *stats = seq->private;
1476
1477         lprocfs_clear_stats(stats);
1478
1479         return len;
1480 }
1481
1482 static void *lprocfs_stats_seq_start(struct seq_file *p, loff_t *pos)
1483 {
1484         struct lprocfs_stats *stats = p->private;
1485
1486         return (*pos < stats->ls_num) ? pos : NULL;
1487 }
1488
1489 static void lprocfs_stats_seq_stop(struct seq_file *p, void *v)
1490 {
1491 }
1492
1493 static void *lprocfs_stats_seq_next(struct seq_file *p, void *v, loff_t *pos)
1494 {
1495         (*pos)++;
1496
1497         return lprocfs_stats_seq_start(p, pos);
1498 }
1499
1500 /* seq file export of one lprocfs counter */
1501 static int lprocfs_stats_seq_show(struct seq_file *p, void *v)
1502 {
1503         struct lprocfs_stats *stats = p->private;
1504         struct lprocfs_counter_header *hdr;
1505         struct lprocfs_counter ctr;
1506         int idx = *(loff_t *)v;
1507
1508         if (idx == 0) {
1509                 struct timespec64 now;
1510
1511                 ktime_get_real_ts64(&now);
1512                 seq_printf(p, "%-25s %llu.%09lu secs.nsecs\n",
1513                            "snapshot_time", (s64)now.tv_sec, now.tv_nsec);
1514         }
1515
1516         hdr = &stats->ls_cnt_header[idx];
1517         lprocfs_stats_collect(stats, idx, &ctr);
1518
1519         if (ctr.lc_count == 0)
1520                 return 0;
1521
1522         seq_printf(p, "%-25s %lld samples [%s]", hdr->lc_name,
1523                    ctr.lc_count, hdr->lc_units);
1524
1525         if ((hdr->lc_config & LPROCFS_CNTR_AVGMINMAX) && ctr.lc_count > 0) {
1526                 seq_printf(p, " %lld %lld %lld",
1527                            ctr.lc_min, ctr.lc_max, ctr.lc_sum);
1528                 if (hdr->lc_config & LPROCFS_CNTR_STDDEV)
1529                         seq_printf(p, " %llu", ctr.lc_sumsquare);
1530         }
1531         seq_putc(p, '\n');
1532         return 0;
1533 }
1534
1535 static const struct seq_operations lprocfs_stats_seq_sops = {
1536         .start  = lprocfs_stats_seq_start,
1537         .stop   = lprocfs_stats_seq_stop,
1538         .next   = lprocfs_stats_seq_next,
1539         .show   = lprocfs_stats_seq_show,
1540 };
1541
1542 static int lprocfs_stats_seq_open(struct inode *inode, struct file *file)
1543 {
1544         struct seq_file *seq;
1545         int rc;
1546
1547         rc = LPROCFS_ENTRY_CHECK(inode);
1548         if (rc < 0)
1549                 return rc;
1550
1551         rc = seq_open(file, &lprocfs_stats_seq_sops);
1552         if (rc)
1553                 return rc;
1554         seq = file->private_data;
1555         seq->private = inode->i_private ? inode->i_private : PDE_DATA(inode);
1556         return 0;
1557 }
1558
1559 static const struct file_operations lprocfs_stats_seq_fops = {
1560         .owner   = THIS_MODULE,
1561         .open    = lprocfs_stats_seq_open,
1562         .read    = seq_read,
1563         .write   = lprocfs_stats_seq_write,
1564         .llseek  = seq_lseek,
1565         .release = lprocfs_seq_release,
1566 };
1567
1568 int ldebugfs_register_stats(struct dentry *parent, const char *name,
1569                             struct lprocfs_stats *stats)
1570 {
1571         struct dentry *entry;
1572
1573         LASSERT(!IS_ERR_OR_NULL(parent));
1574
1575         entry = debugfs_create_file(name, 0644, parent, stats,
1576                                     &lprocfs_stats_seq_fops);
1577         if (IS_ERR_OR_NULL(entry))
1578                 return entry ? PTR_ERR(entry) : -ENOMEM;
1579
1580         return 0;
1581 }
1582 EXPORT_SYMBOL_GPL(ldebugfs_register_stats);
1583
1584 int lprocfs_register_stats(struct proc_dir_entry *root, const char *name,
1585                            struct lprocfs_stats *stats)
1586 {
1587         struct proc_dir_entry *entry;
1588         LASSERT(root != NULL);
1589
1590         entry = proc_create_data(name, 0644, root,
1591                                  &lprocfs_stats_seq_fops, stats);
1592         if (!entry)
1593                 return -ENOMEM;
1594         return 0;
1595 }
1596 EXPORT_SYMBOL(lprocfs_register_stats);
1597
1598 void lprocfs_counter_init(struct lprocfs_stats *stats, int index,
1599                           unsigned conf, const char *name, const char *units)
1600 {
1601         struct lprocfs_counter_header *header;
1602         struct lprocfs_counter *percpu_cntr;
1603         unsigned long flags = 0;
1604         unsigned int i;
1605         unsigned int num_cpu;
1606
1607         LASSERT(stats != NULL);
1608
1609         header = &stats->ls_cnt_header[index];
1610         LASSERTF(header != NULL, "Failed to allocate stats header:[%d]%s/%s\n",
1611                  index, name, units);
1612
1613         header->lc_config = conf;
1614         header->lc_name   = name;
1615         header->lc_units  = units;
1616
1617         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1618         for (i = 0; i < num_cpu; ++i) {
1619                 if (!stats->ls_percpu[i])
1620                         continue;
1621                 percpu_cntr = lprocfs_stats_counter_get(stats, i, index);
1622                 percpu_cntr->lc_count           = 0;
1623                 percpu_cntr->lc_min             = LC_MIN_INIT;
1624                 percpu_cntr->lc_max             = 0;
1625                 percpu_cntr->lc_sumsquare       = 0;
1626                 percpu_cntr->lc_sum             = 0;
1627                 if ((stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1628                         percpu_cntr->lc_sum_irq = 0;
1629         }
1630         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1631 }
1632 EXPORT_SYMBOL(lprocfs_counter_init);
1633
1634 static const char * const mps_stats[] = {
1635         [LPROC_MD_CLOSE]                = "close",
1636         [LPROC_MD_CREATE]               = "create",
1637         [LPROC_MD_ENQUEUE]              = "enqueue",
1638         [LPROC_MD_GETATTR]              = "getattr",
1639         [LPROC_MD_INTENT_LOCK]          = "intent_lock",
1640         [LPROC_MD_LINK]                 = "link",
1641         [LPROC_MD_RENAME]               = "rename",
1642         [LPROC_MD_SETATTR]              = "setattr",
1643         [LPROC_MD_FSYNC]                = "fsync",
1644         [LPROC_MD_READ_PAGE]            = "read_page",
1645         [LPROC_MD_UNLINK]               = "unlink",
1646         [LPROC_MD_SETXATTR]             = "setxattr",
1647         [LPROC_MD_GETXATTR]             = "getxattr",
1648         [LPROC_MD_INTENT_GETATTR_ASYNC] = "intent_getattr_async",
1649         [LPROC_MD_REVALIDATE_LOCK]      = "revalidate_lock",
1650 };
1651
1652 int lprocfs_alloc_md_stats(struct obd_device *obd,
1653                            unsigned int num_private_stats)
1654 {
1655         struct lprocfs_stats *stats;
1656         unsigned int num_stats;
1657         int rc, i;
1658
1659         /*
1660          * TODO Ensure that this function is only used where
1661          * appropriate by adding an assertion to the effect that
1662          * obd->obd_type->typ_md_ops is not NULL. We can't do this now
1663          * because mdt_procfs_init() uses this function to allocate
1664          * the stats backing /proc/fs/lustre/mdt/.../md_stats but the
1665          * mdt layer does not use the md_ops interface. This is
1666          * confusing and a waste of memory. See LU-2484.
1667          */
1668         LASSERT(obd->obd_proc_entry != NULL);
1669         LASSERT(obd->obd_md_stats == NULL);
1670
1671         num_stats = ARRAY_SIZE(mps_stats) + num_private_stats;
1672         stats = lprocfs_alloc_stats(num_stats, 0);
1673         if (!stats)
1674                 return -ENOMEM;
1675
1676         for (i = 0; i < ARRAY_SIZE(mps_stats); i++) {
1677                 lprocfs_counter_init(stats, i, 0, mps_stats[i], "reqs");
1678                 if (!stats->ls_cnt_header[i].lc_name) {
1679                         CERROR("Missing md_stat initializer md_op operation at offset %d. Aborting.\n",
1680                                i);
1681                         LBUG();
1682                 }
1683         }
1684
1685         rc = lprocfs_register_stats(obd->obd_proc_entry, "md_stats", stats);
1686         if (rc < 0) {
1687                 lprocfs_free_stats(&stats);
1688         } else {
1689                 obd->obd_md_stats = stats;
1690         }
1691
1692         return rc;
1693 }
1694 EXPORT_SYMBOL(lprocfs_alloc_md_stats);
1695
1696 void lprocfs_free_md_stats(struct obd_device *obd)
1697 {
1698         struct lprocfs_stats *stats = obd->obd_md_stats;
1699
1700         if (stats) {
1701                 obd->obd_md_stats = NULL;
1702                 lprocfs_free_stats(&stats);
1703         }
1704 }
1705 EXPORT_SYMBOL(lprocfs_free_md_stats);
1706
1707 void lprocfs_init_ldlm_stats(struct lprocfs_stats *ldlm_stats)
1708 {
1709         lprocfs_counter_init(ldlm_stats,
1710                              LDLM_ENQUEUE - LDLM_FIRST_OPC,
1711                              0, "ldlm_enqueue", "reqs");
1712         lprocfs_counter_init(ldlm_stats,
1713                              LDLM_CONVERT - LDLM_FIRST_OPC,
1714                              0, "ldlm_convert", "reqs");
1715         lprocfs_counter_init(ldlm_stats,
1716                              LDLM_CANCEL - LDLM_FIRST_OPC,
1717                              0, "ldlm_cancel", "reqs");
1718         lprocfs_counter_init(ldlm_stats,
1719                              LDLM_BL_CALLBACK - LDLM_FIRST_OPC,
1720                              0, "ldlm_bl_callback", "reqs");
1721         lprocfs_counter_init(ldlm_stats,
1722                              LDLM_CP_CALLBACK - LDLM_FIRST_OPC,
1723                              0, "ldlm_cp_callback", "reqs");
1724         lprocfs_counter_init(ldlm_stats,
1725                              LDLM_GL_CALLBACK - LDLM_FIRST_OPC,
1726                              0, "ldlm_gl_callback", "reqs");
1727 }
1728 EXPORT_SYMBOL(lprocfs_init_ldlm_stats);
1729
1730 __s64 lprocfs_read_helper(struct lprocfs_counter *lc,
1731                           struct lprocfs_counter_header *header,
1732                           enum lprocfs_stats_flags flags,
1733                           enum lprocfs_fields_flags field)
1734 {
1735         __s64 ret = 0;
1736
1737         if (!lc || !header)
1738                 RETURN(0);
1739
1740         switch (field) {
1741                 case LPROCFS_FIELDS_FLAGS_CONFIG:
1742                         ret = header->lc_config;
1743                         break;
1744                 case LPROCFS_FIELDS_FLAGS_SUM:
1745                         ret = lc->lc_sum;
1746                         if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1747                                 ret += lc->lc_sum_irq;
1748                         break;
1749                 case LPROCFS_FIELDS_FLAGS_MIN:
1750                         ret = lc->lc_min;
1751                         break;
1752                 case LPROCFS_FIELDS_FLAGS_MAX:
1753                         ret = lc->lc_max;
1754                         break;
1755                 case LPROCFS_FIELDS_FLAGS_AVG:
1756                         ret = (lc->lc_max - lc->lc_min) / 2;
1757                         break;
1758                 case LPROCFS_FIELDS_FLAGS_SUMSQUARE:
1759                         ret = lc->lc_sumsquare;
1760                         break;
1761                 case LPROCFS_FIELDS_FLAGS_COUNT:
1762                         ret = lc->lc_count;
1763                         break;
1764                 default:
1765                         break;
1766         };
1767         RETURN(ret);
1768 }
1769 EXPORT_SYMBOL(lprocfs_read_helper);
1770
1771 int lprocfs_read_frac_helper(char *buffer, unsigned long count, long val,
1772                              int mult)
1773 {
1774         long decimal_val, frac_val;
1775         int prtn;
1776
1777         if (count < 10)
1778                 return -EINVAL;
1779
1780         decimal_val = val / mult;
1781         prtn = snprintf(buffer, count, "%ld", decimal_val);
1782         frac_val = val % mult;
1783
1784         if (prtn < (count - 4) && frac_val > 0) {
1785                 long temp_frac;
1786                 int i, temp_mult = 1, frac_bits = 0;
1787
1788                 temp_frac = frac_val * 10;
1789                 buffer[prtn++] = '.';
1790                 while (frac_bits < 2 && (temp_frac / mult) < 1) {
1791                         /* only reserved 2 bits fraction */
1792                         buffer[prtn++] = '0';
1793                         temp_frac *= 10;
1794                         frac_bits++;
1795                 }
1796                 /*
1797                  * Need to think these cases :
1798                  *      1. #echo x.00 > /proc/xxx       output result : x
1799                  *      2. #echo x.0x > /proc/xxx       output result : x.0x
1800                  *      3. #echo x.x0 > /proc/xxx       output result : x.x
1801                  *      4. #echo x.xx > /proc/xxx       output result : x.xx
1802                  *      Only reserved 2 bits fraction.
1803                  */
1804                 for (i = 0; i < (5 - prtn); i++)
1805                         temp_mult *= 10;
1806
1807                 frac_bits = min((int)count - prtn, 3 - frac_bits);
1808                 prtn += snprintf(buffer + prtn, frac_bits, "%ld",
1809                                  frac_val * temp_mult / mult);
1810
1811                 prtn--;
1812                 while (buffer[prtn] < '1' || buffer[prtn] > '9') {
1813                         prtn--;
1814                         if (buffer[prtn] == '.') {
1815                                 prtn--;
1816                                 break;
1817                         }
1818                 }
1819                 prtn++;
1820         }
1821         buffer[prtn++] = '\n';
1822         return prtn;
1823 }
1824 EXPORT_SYMBOL(lprocfs_read_frac_helper);
1825
1826 int lprocfs_seq_read_frac_helper(struct seq_file *m, long val, int mult)
1827 {
1828         long decimal_val, frac_val;
1829
1830         decimal_val = val / mult;
1831         seq_printf(m, "%ld", decimal_val);
1832         frac_val = val % mult;
1833
1834         if (frac_val > 0) {
1835                 frac_val *= 100;
1836                 frac_val /= mult;
1837         }
1838         if (frac_val > 0) {
1839                 /* Three cases: x0, xx, 0x */
1840                 if ((frac_val % 10) != 0)
1841                         seq_printf(m, ".%ld", frac_val);
1842                 else
1843                         seq_printf(m, ".%ld", frac_val / 10);
1844         }
1845
1846         seq_printf(m, "\n");
1847         return 0;
1848 }
1849 EXPORT_SYMBOL(lprocfs_seq_read_frac_helper);
1850
1851 /* Obtains the conversion factor for the unit specified */
1852 static int get_mult(char unit, __u64 *mult)
1853 {
1854         __u64 units = 1;
1855
1856         switch (unit) {
1857         /* peta, tera, giga, mega, and kilo */
1858         case 'p':
1859         case 'P':
1860                 units <<= 10;
1861         case 't':
1862         case 'T':
1863                 units <<= 10;
1864         case 'g':
1865         case 'G':
1866                 units <<= 10;
1867         case 'm':
1868         case 'M':
1869                 units <<= 10;
1870         case 'k':
1871         case 'K':
1872                 units <<= 10;
1873                 break;
1874         /* some tests expect % to be accepted */
1875         case '%':
1876                 units = 1;
1877                 break;
1878         default:
1879                 return -EINVAL;
1880         }
1881
1882         *mult = units;
1883
1884         return 0;
1885 }
1886
1887 /*
1888  * Ensures the numeric string is valid. The function provides the final
1889  * multiplier in the case a unit exists at the end of the string. It also
1890  * locates the start of the whole and fractional parts (if any). This
1891  * function modifies the string so kstrtoull can be used to parse both
1892  * the whole and fraction portions. This function also figures out
1893  * the base of the number.
1894  */
1895 static int preprocess_numeric_str(char *buffer, __u64 *mult, __u64 def_mult,
1896                                   bool allow_units, char **whole, char **frac,
1897                                   unsigned int *base)
1898 {
1899         bool hit_decimal = false;
1900         bool hit_unit = false;
1901         int rc = 0;
1902         char *start;
1903         *mult = def_mult;
1904         *whole = NULL;
1905         *frac = NULL;
1906         *base = 10;
1907
1908         /* a hex string if it starts with "0x" */
1909         if (buffer[0] == '0' && tolower(buffer[1]) == 'x') {
1910                 *base = 16;
1911                 buffer += 2;
1912         }
1913
1914         start = buffer;
1915
1916         while (*buffer) {
1917                 /* allow for a single new line before the null terminator */
1918                 if (*buffer == '\n') {
1919                         *buffer = '\0';
1920                         buffer++;
1921
1922                         if (*buffer)
1923                                 return -EINVAL;
1924
1925                         break;
1926                 }
1927
1928                 /* any chars after our unit indicates a malformed string */
1929                 if (hit_unit)
1930                         return -EINVAL;
1931
1932                 /* ensure we only hit one decimal */
1933                 if (*buffer == '.') {
1934                         if (hit_decimal)
1935                                 return -EINVAL;
1936
1937                         /* if past start, there's a whole part */
1938                         if (start != buffer)
1939                                 *whole = start;
1940
1941                         *buffer = '\0';
1942                         start = buffer + 1;
1943                         hit_decimal = true;
1944                 } else if (!isdigit(*buffer) &&
1945                            !(*base == 16 && isxdigit(*buffer))) {
1946                         if (allow_units) {
1947                                 /* if we allow units, attempt to get mult */
1948                                 hit_unit = true;
1949                                 rc = get_mult(*buffer, mult);
1950                                 if (rc)
1951                                         return rc;
1952
1953                                 /* string stops here, but keep processing */
1954                                 *buffer = '\0';
1955                         } else {
1956                                 /* bad string */
1957                                 return -EINVAL;
1958                         }
1959                 }
1960
1961                 buffer++;
1962         }
1963
1964         if (hit_decimal) {
1965                 /* hit a decimal, make sure there's a fractional part */
1966                 if (!*start)
1967                         return -EINVAL;
1968
1969                 *frac = start;
1970         } else {
1971                 /* didn't hit a decimal, but may have a whole part */
1972                 if (start != buffer && *start)
1973                         *whole = start;
1974         }
1975
1976         /* malformed string if we didn't get anything */
1977         if (!*frac && !*whole)
1978                 return -EINVAL;
1979
1980         return 0;
1981 }
1982
1983 /*
1984  * Parses a numeric string which can contain a whole and fraction portion
1985  * into a __u64. Accepts a multiplier to apply to the value parsed. Also
1986  * allows the string to have a unit at the end. The function handles
1987  * wrapping of the final unsigned value.
1988  */
1989 static int str_to_u64_parse(char *buffer, unsigned long count,
1990                             __u64 *val, __u64 def_mult, bool allow_units)
1991 {
1992         __u64 whole = 0;
1993         __u64 frac = 0;
1994         unsigned int frac_d = 1;
1995         __u64 wrap_indicator = ULLONG_MAX;
1996         int rc = 0;
1997         __u64 mult;
1998         char *strwhole;
1999         char *strfrac;
2000         unsigned int base = 10;
2001
2002         rc = preprocess_numeric_str(buffer, &mult, def_mult, allow_units,
2003                                     &strwhole, &strfrac, &base);
2004
2005         if (rc)
2006                 return rc;
2007
2008         if (mult == 0) {
2009                 *val = 0;
2010                 return 0;
2011         }
2012
2013         /* the multiplier limits how large the value can be */
2014         wrap_indicator = div64_u64(wrap_indicator, mult);
2015
2016         if (strwhole) {
2017                 rc = kstrtoull(strwhole, base, &whole);
2018                 if (rc)
2019                         return rc;
2020
2021                 if (whole > wrap_indicator)
2022                         return -ERANGE;
2023
2024                 whole *= mult;
2025         }
2026
2027         if (strfrac) {
2028                 if (strlen(strfrac) > 10)
2029                         strfrac[10] = '\0';
2030
2031                 rc = kstrtoull(strfrac, base, &frac);
2032                 if (rc)
2033                         return rc;
2034
2035                 /* determine power of fractional portion */
2036                 while (*strfrac) {
2037                         frac_d *= base;
2038                         strfrac++;
2039                 }
2040
2041                 /* fractional portion is too large to perform calculation */
2042                 if (frac > wrap_indicator)
2043                         return -ERANGE;
2044
2045                 frac *= mult;
2046                 do_div(frac, frac_d);
2047         }
2048
2049         /* check that the sum of whole and fraction fits in u64 */
2050         if (whole > (ULLONG_MAX - frac))
2051                 return -ERANGE;
2052
2053         *val = whole + frac;
2054
2055         return 0;
2056 }
2057
2058 /*
2059  * This function parses numeric/hex strings into __s64. It accepts a multiplier
2060  * which will apply to the value parsed. It also can allow the string to
2061  * have a unit as the last character. The function handles overflow/underflow
2062  * of the signed integer.
2063  */
2064 static int str_to_s64_internal(const char __user *buffer, unsigned long count,
2065                                __s64 *val, __u64 def_mult, bool allow_units)
2066 {
2067         char kernbuf[22];
2068         __u64 tmp;
2069         unsigned int offset = 0;
2070         int signed sign = 1;
2071         __u64 max = LLONG_MAX;
2072         int rc = 0;
2073
2074         if (count > (sizeof(kernbuf) - 1))
2075                 return -EINVAL;
2076
2077         if (copy_from_user(kernbuf, buffer, count))
2078                 return -EFAULT;
2079
2080         kernbuf[count] = '\0';
2081
2082         /* keep track of our sign */
2083         if (*kernbuf == '-') {
2084                 sign = -1;
2085                 offset++;
2086                 /* equivalent to max = -LLONG_MIN, avoids overflow */
2087                 max++;
2088         }
2089
2090         rc = str_to_u64_parse(kernbuf + offset, count - offset,
2091                               &tmp, def_mult, allow_units);
2092         if (rc)
2093                 return rc;
2094
2095         /* check for overflow/underflow */
2096         if (max < tmp)
2097                 return -ERANGE;
2098
2099         *val = (__s64)tmp * sign;
2100
2101         return 0;
2102 }
2103
2104 /**
2105  * Convert a user string into a signed 64 bit number. This function produces
2106  * an error when the value parsed from the string times multiplier underflows or
2107  * overflows. This function only accepts strings that contains digits, an
2108  * optional decimal, and a char representing a unit at the end. If a unit is
2109  * specified in the string, the multiplier provided by the caller is ignored.
2110  * This function can also accept hexadecimal strings which are prefixed with
2111  * "0x".
2112  *
2113  * \param[in] buffer    string consisting of numbers, a decimal, and a unit
2114  * \param[in] count     buffer length
2115  * \param[in] val       if successful, the value represented by the string
2116  * \param[in] defunit   default unit if string doesn't contain one
2117  *
2118  * \retval              0 on success
2119  * \retval              negative number on error
2120  */
2121 int lprocfs_str_with_units_to_s64(const char __user *buffer,
2122                                   unsigned long count, __s64 *val, char defunit)
2123 {
2124         __u64 mult = 1;
2125         int rc;
2126
2127         if (defunit != '1') {
2128                 rc = get_mult(defunit, &mult);
2129                 if (rc)
2130                         return rc;
2131         }
2132
2133         return str_to_s64_internal(buffer, count, val, mult, true);
2134 }
2135 EXPORT_SYMBOL(lprocfs_str_with_units_to_s64);
2136
2137 char *lprocfs_strnstr(const char *s1, const char *s2, size_t len)
2138 {
2139         size_t l2;
2140
2141         l2 = strlen(s2);
2142         if (!l2)
2143                 return (char *)s1;
2144         while (len >= l2) {
2145                 len--;
2146                 if (!memcmp(s1, s2, l2))
2147                         return (char *)s1;
2148                 s1++;
2149         }
2150         return NULL;
2151 }
2152 EXPORT_SYMBOL(lprocfs_strnstr);
2153
2154 /**
2155  * Find the string \a name in the input \a buffer, and return a pointer to the
2156  * value immediately following \a name, reducing \a count appropriately.
2157  * If \a name is not found the original \a buffer is returned.
2158  */
2159 char *lprocfs_find_named_value(const char *buffer, const char *name,
2160                                 size_t *count)
2161 {
2162         char *val;
2163         size_t buflen = *count;
2164
2165         /* there is no strnstr() in rhel5 and ubuntu kernels */
2166         val = lprocfs_strnstr(buffer, name, buflen);
2167         if (!val)
2168                 return (char *)buffer;
2169
2170         val += strlen(name);                             /* skip prefix */
2171         while (val < buffer + buflen && isspace(*val)) /* skip separator */
2172                 val++;
2173
2174         *count = 0;
2175         while (val < buffer + buflen && isalnum(*val)) {
2176                 ++*count;
2177                 ++val;
2178         }
2179
2180         return val - *count;
2181 }
2182 EXPORT_SYMBOL(lprocfs_find_named_value);
2183
2184 int ldebugfs_seq_create(struct dentry *parent, const char *name, umode_t mode,
2185                         const struct file_operations *seq_fops, void *data)
2186 {
2187         struct dentry *entry;
2188
2189         /* Disallow secretly (un)writable entries. */
2190         LASSERT((!seq_fops->write) == (!(mode & 0222)));
2191
2192         entry = debugfs_create_file(name, mode, parent, data, seq_fops);
2193         if (IS_ERR_OR_NULL(entry))
2194                 return entry ? PTR_ERR(entry) : -ENOMEM;
2195
2196         return 0;
2197 }
2198 EXPORT_SYMBOL_GPL(ldebugfs_seq_create);
2199
2200 int lprocfs_seq_create(struct proc_dir_entry *parent,
2201                        const char *name,
2202                        mode_t mode,
2203                        const struct file_operations *seq_fops,
2204                        void *data)
2205 {
2206         struct proc_dir_entry *entry;
2207         ENTRY;
2208
2209         /* Disallow secretly (un)writable entries. */
2210         LASSERT((seq_fops->write == NULL) == ((mode & 0222) == 0));
2211
2212         entry = proc_create_data(name, mode, parent, seq_fops, data);
2213
2214         if (!entry)
2215                 RETURN(-ENOMEM);
2216
2217         RETURN(0);
2218 }
2219 EXPORT_SYMBOL(lprocfs_seq_create);
2220
2221 int lprocfs_obd_seq_create(struct obd_device *dev,
2222                            const char *name,
2223                            mode_t mode,
2224                            const struct file_operations *seq_fops,
2225                            void *data)
2226 {
2227         return (lprocfs_seq_create(dev->obd_proc_entry, name,
2228                                    mode, seq_fops, data));
2229 }
2230 EXPORT_SYMBOL(lprocfs_obd_seq_create);
2231
2232 void lprocfs_oh_tally(struct obd_histogram *oh, unsigned int value)
2233 {
2234         if (value >= OBD_HIST_MAX)
2235                 value = OBD_HIST_MAX - 1;
2236
2237         spin_lock(&oh->oh_lock);
2238         oh->oh_buckets[value]++;
2239         spin_unlock(&oh->oh_lock);
2240 }
2241 EXPORT_SYMBOL(lprocfs_oh_tally);
2242
2243 void lprocfs_oh_tally_log2(struct obd_histogram *oh, unsigned int value)
2244 {
2245         unsigned int val = 0;
2246
2247         if (likely(value != 0))
2248                 val = min(fls(value - 1), OBD_HIST_MAX);
2249
2250         lprocfs_oh_tally(oh, val);
2251 }
2252 EXPORT_SYMBOL(lprocfs_oh_tally_log2);
2253
2254 unsigned long lprocfs_oh_sum(struct obd_histogram *oh)
2255 {
2256         unsigned long ret = 0;
2257         int i;
2258
2259         for (i = 0; i < OBD_HIST_MAX; i++)
2260                 ret +=  oh->oh_buckets[i];
2261         return ret;
2262 }
2263 EXPORT_SYMBOL(lprocfs_oh_sum);
2264
2265 void lprocfs_oh_clear(struct obd_histogram *oh)
2266 {
2267         spin_lock(&oh->oh_lock);
2268         memset(oh->oh_buckets, 0, sizeof(oh->oh_buckets));
2269         spin_unlock(&oh->oh_lock);
2270 }
2271 EXPORT_SYMBOL(lprocfs_oh_clear);
2272
2273 ssize_t lustre_attr_show(struct kobject *kobj,
2274                          struct attribute *attr, char *buf)
2275 {
2276         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
2277
2278         return a->show ? a->show(kobj, attr, buf) : 0;
2279 }
2280 EXPORT_SYMBOL_GPL(lustre_attr_show);
2281
2282 ssize_t lustre_attr_store(struct kobject *kobj, struct attribute *attr,
2283                           const char *buf, size_t len)
2284 {
2285         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
2286
2287         return a->store ? a->store(kobj, attr, buf, len) : len;
2288 }
2289 EXPORT_SYMBOL_GPL(lustre_attr_store);
2290
2291 const struct sysfs_ops lustre_sysfs_ops = {
2292         .show  = lustre_attr_show,
2293         .store = lustre_attr_store,
2294 };
2295 EXPORT_SYMBOL_GPL(lustre_sysfs_ops);
2296
2297 int lprocfs_obd_max_pages_per_rpc_seq_show(struct seq_file *m, void *data)
2298 {
2299         struct obd_device *dev = data;
2300         struct client_obd *cli = &dev->u.cli;
2301
2302         spin_lock(&cli->cl_loi_list_lock);
2303         seq_printf(m, "%d\n", cli->cl_max_pages_per_rpc);
2304         spin_unlock(&cli->cl_loi_list_lock);
2305         return 0;
2306 }
2307 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_show);
2308
2309 ssize_t lprocfs_obd_max_pages_per_rpc_seq_write(struct file *file,
2310                                                 const char __user *buffer,
2311                                                 size_t count, loff_t *off)
2312 {
2313         struct obd_device *dev =
2314                 ((struct seq_file *)file->private_data)->private;
2315         struct client_obd *cli = &dev->u.cli;
2316         struct obd_connect_data *ocd = &cli->cl_import->imp_connect_data;
2317         int chunk_mask, rc;
2318         s64 val;
2319
2320         rc = lprocfs_str_with_units_to_s64(buffer, count, &val, '1');
2321         if (rc)
2322                 return rc;
2323         if (val < 0)
2324                 return -ERANGE;
2325
2326         /* if the max_pages is specified in bytes, convert to pages */
2327         if (val >= ONE_MB_BRW_SIZE)
2328                 val >>= PAGE_SHIFT;
2329
2330         LPROCFS_CLIMP_CHECK(dev);
2331
2332         chunk_mask = ~((1 << (cli->cl_chunkbits - PAGE_SHIFT)) - 1);
2333         /* max_pages_per_rpc must be chunk aligned */
2334         val = (val + ~chunk_mask) & chunk_mask;
2335         if (val == 0 || (ocd->ocd_brw_size != 0 &&
2336                          val > ocd->ocd_brw_size >> PAGE_SHIFT)) {
2337                 LPROCFS_CLIMP_EXIT(dev);
2338                 return -ERANGE;
2339         }
2340         spin_lock(&cli->cl_loi_list_lock);
2341         cli->cl_max_pages_per_rpc = val;
2342         client_adjust_max_dirty(cli);
2343         spin_unlock(&cli->cl_loi_list_lock);
2344
2345         LPROCFS_CLIMP_EXIT(dev);
2346         return count;
2347 }
2348 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_write);
2349
2350 ssize_t short_io_bytes_show(struct kobject *kobj, struct attribute *attr,
2351                             char *buf)
2352 {
2353         struct obd_device *dev = container_of(kobj, struct obd_device,
2354                                               obd_kset.kobj);
2355         struct client_obd *cli = &dev->u.cli;
2356         int rc;
2357
2358         spin_lock(&cli->cl_loi_list_lock);
2359         rc = sprintf(buf, "%d\n", cli->cl_max_short_io_bytes);
2360         spin_unlock(&cli->cl_loi_list_lock);
2361         return rc;
2362 }
2363 EXPORT_SYMBOL(short_io_bytes_show);
2364
2365 /* Used to catch people who think they're specifying pages. */
2366 #define MIN_SHORT_IO_BYTES 64
2367
2368 ssize_t short_io_bytes_store(struct kobject *kobj, struct attribute *attr,
2369                              const char *buffer, size_t count)
2370 {
2371         struct obd_device *dev = container_of(kobj, struct obd_device,
2372                                               obd_kset.kobj);
2373         struct client_obd *cli = &dev->u.cli;
2374         u32 val;
2375         int rc;
2376
2377         LPROCFS_CLIMP_CHECK(dev);
2378
2379         rc = kstrtouint(buffer, 0, &val);
2380         if (rc)
2381                 GOTO(out, rc);
2382
2383         if (val && (val < MIN_SHORT_IO_BYTES || val > OBD_MAX_SHORT_IO_BYTES))
2384                 GOTO(out, rc = -ERANGE);
2385
2386         rc = count;
2387
2388         spin_lock(&cli->cl_loi_list_lock);
2389         if (val > (cli->cl_max_pages_per_rpc << PAGE_SHIFT))
2390                 rc = -ERANGE;
2391         else
2392                 cli->cl_max_short_io_bytes = val;
2393         spin_unlock(&cli->cl_loi_list_lock);
2394
2395 out:
2396         LPROCFS_CLIMP_EXIT(dev);
2397         return rc;
2398 }
2399 EXPORT_SYMBOL(short_io_bytes_store);
2400
2401 int lprocfs_wr_root_squash(const char __user *buffer, unsigned long count,
2402                            struct root_squash_info *squash, char *name)
2403 {
2404         int rc;
2405         char kernbuf[64], *tmp, *errmsg;
2406         unsigned long uid, gid;
2407         ENTRY;
2408
2409         if (count >= sizeof(kernbuf)) {
2410                 errmsg = "string too long";
2411                 GOTO(failed_noprint, rc = -EINVAL);
2412         }
2413         if (copy_from_user(kernbuf, buffer, count)) {
2414                 errmsg = "bad address";
2415                 GOTO(failed_noprint, rc = -EFAULT);
2416         }
2417         kernbuf[count] = '\0';
2418
2419         /* look for uid gid separator */
2420         tmp = strchr(kernbuf, ':');
2421         if (!tmp) {
2422                 errmsg = "needs uid:gid format";
2423                 GOTO(failed, rc = -EINVAL);
2424         }
2425         *tmp = '\0';
2426         tmp++;
2427
2428         /* parse uid */
2429         if (kstrtoul(kernbuf, 0, &uid) != 0) {
2430                 errmsg = "bad uid";
2431                 GOTO(failed, rc = -EINVAL);
2432         }
2433
2434         /* parse gid */
2435         if (kstrtoul(tmp, 0, &gid) != 0) {
2436                 errmsg = "bad gid";
2437                 GOTO(failed, rc = -EINVAL);
2438         }
2439
2440         squash->rsi_uid = uid;
2441         squash->rsi_gid = gid;
2442
2443         LCONSOLE_INFO("%s: root_squash is set to %u:%u\n",
2444                       name, squash->rsi_uid, squash->rsi_gid);
2445         RETURN(count);
2446
2447 failed:
2448         if (tmp) {
2449                 tmp--;
2450                 *tmp = ':';
2451         }
2452         CWARN("%s: failed to set root_squash to \"%s\", %s, rc = %d\n",
2453               name, kernbuf, errmsg, rc);
2454         RETURN(rc);
2455 failed_noprint:
2456         CWARN("%s: failed to set root_squash due to %s, rc = %d\n",
2457               name, errmsg, rc);
2458         RETURN(rc);
2459 }
2460 EXPORT_SYMBOL(lprocfs_wr_root_squash);
2461
2462
2463 int lprocfs_wr_nosquash_nids(const char __user *buffer, unsigned long count,
2464                              struct root_squash_info *squash, char *name)
2465 {
2466         int rc;
2467         char *kernbuf = NULL;
2468         char *errmsg;
2469         struct list_head tmp;
2470         int len = count;
2471         ENTRY;
2472
2473         if (count > 4096) {
2474                 errmsg = "string too long";
2475                 GOTO(failed, rc = -EINVAL);
2476         }
2477
2478         OBD_ALLOC(kernbuf, count + 1);
2479         if (!kernbuf) {
2480                 errmsg = "no memory";
2481                 GOTO(failed, rc = -ENOMEM);
2482         }
2483         if (copy_from_user(kernbuf, buffer, count)) {
2484                 errmsg = "bad address";
2485                 GOTO(failed, rc = -EFAULT);
2486         }
2487         kernbuf[count] = '\0';
2488
2489         if (count > 0 && kernbuf[count - 1] == '\n')
2490                 len = count - 1;
2491
2492         if ((len == 4 && strncmp(kernbuf, "NONE", len) == 0) ||
2493             (len == 5 && strncmp(kernbuf, "clear", len) == 0)) {
2494                 /* empty string is special case */
2495                 down_write(&squash->rsi_sem);
2496                 if (!list_empty(&squash->rsi_nosquash_nids))
2497                         cfs_free_nidlist(&squash->rsi_nosquash_nids);
2498                 up_write(&squash->rsi_sem);
2499                 LCONSOLE_INFO("%s: nosquash_nids is cleared\n", name);
2500                 OBD_FREE(kernbuf, count + 1);
2501                 RETURN(count);
2502         }
2503
2504         INIT_LIST_HEAD(&tmp);
2505         if (cfs_parse_nidlist(kernbuf, count, &tmp) <= 0) {
2506                 errmsg = "can't parse";
2507                 GOTO(failed, rc = -EINVAL);
2508         }
2509         LCONSOLE_INFO("%s: nosquash_nids set to %s\n",
2510                       name, kernbuf);
2511         OBD_FREE(kernbuf, count + 1);
2512         kernbuf = NULL;
2513
2514         down_write(&squash->rsi_sem);
2515         if (!list_empty(&squash->rsi_nosquash_nids))
2516                 cfs_free_nidlist(&squash->rsi_nosquash_nids);
2517         list_splice(&tmp, &squash->rsi_nosquash_nids);
2518         up_write(&squash->rsi_sem);
2519
2520         RETURN(count);
2521
2522 failed:
2523         if (kernbuf) {
2524                 CWARN("%s: failed to set nosquash_nids to \"%s\", %s rc = %d\n",
2525                       name, kernbuf, errmsg, rc);
2526                 OBD_FREE(kernbuf, count + 1);
2527         } else {
2528                 CWARN("%s: failed to set nosquash_nids due to %s rc = %d\n",
2529                       name, errmsg, rc);
2530         }
2531         RETURN(rc);
2532 }
2533 EXPORT_SYMBOL(lprocfs_wr_nosquash_nids);
2534
2535 #endif /* CONFIG_PROC_FS*/