Whamcloud - gitweb
663fdf14b7898b7a2bf53946849d51ec384c2071
[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 == NULL || name == NULL || fops == NULL)
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, const char *format, ...)
111 {
112         struct proc_dir_entry *entry;
113         char *dest;
114         va_list ap;
115
116         if (parent == NULL || format == NULL)
117                 return NULL;
118
119         OBD_ALLOC_WAIT(dest, MAX_STRING_SIZE + 1);
120         if (dest == NULL)
121                 return NULL;
122
123         va_start(ap, format);
124         vsnprintf(dest, MAX_STRING_SIZE, format, ap);
125         va_end(ap);
126
127         entry = proc_symlink(name, parent, dest);
128         if (entry == NULL)
129                 CERROR("LprocFS: Could not create symbolic link from "
130                        "%s to %s\n", name, dest);
131
132         OBD_FREE(dest, MAX_STRING_SIZE + 1);
133         return entry;
134 }
135 EXPORT_SYMBOL(lprocfs_add_symlink);
136
137 static const struct file_operations lprocfs_generic_fops = { };
138
139 int ldebugfs_add_vars(struct dentry *parent, struct lprocfs_vars *list,
140                       void *data)
141 {
142         if (IS_ERR_OR_NULL(parent) || IS_ERR_OR_NULL(list))
143                 return -EINVAL;
144
145         while (list->name) {
146                 struct dentry *entry;
147                 umode_t mode = 0;
148
149                 if (list->proc_mode != 0000) {
150                         mode = list->proc_mode;
151                 } else if (list->fops) {
152                         if (list->fops->read)
153                                 mode = 0444;
154                         if (list->fops->write)
155                                 mode |= 0200;
156                 }
157                 entry = debugfs_create_file(list->name, mode, parent,
158                                             list->data ? : data,
159                                             list->fops ? : &lprocfs_generic_fops);
160                 if (IS_ERR_OR_NULL(entry))
161                         return entry ? PTR_ERR(entry) : -ENOMEM;
162                 list++;
163         }
164         return 0;
165 }
166 EXPORT_SYMBOL_GPL(ldebugfs_add_vars);
167
168 /**
169  * Add /proc entries.
170  *
171  * \param root [in]  The parent proc entry on which new entry will be added.
172  * \param list [in]  Array of proc entries to be added.
173  * \param data [in]  The argument to be passed when entries read/write routines
174  *                   are called through /proc file.
175  *
176  * \retval 0   on success
177  *         < 0 on error
178  */
179 int
180 lprocfs_add_vars(struct proc_dir_entry *root, struct lprocfs_vars *list,
181                  void *data)
182 {
183         if (root == NULL || list == NULL)
184                 return -EINVAL;
185
186         while (list->name != NULL) {
187                 struct proc_dir_entry *proc;
188                 mode_t mode = 0;
189
190                 if (list->proc_mode != 0000) {
191                         mode = list->proc_mode;
192                 } else if (list->fops) {
193                         if (list->fops->read)
194                                 mode = 0444;
195                         if (list->fops->write)
196                                 mode |= 0200;
197                 }
198                 proc = proc_create_data(list->name, mode, root,
199                                         list->fops ?: &lprocfs_generic_fops,
200                                         list->data ?: data);
201                 if (proc == NULL)
202                         return -ENOMEM;
203                 list++;
204         }
205         return 0;
206 }
207 EXPORT_SYMBOL(lprocfs_add_vars);
208
209 void ldebugfs_remove(struct dentry **entryp)
210 {
211         debugfs_remove_recursive(*entryp);
212         *entryp = NULL;
213 }
214 EXPORT_SYMBOL_GPL(ldebugfs_remove);
215
216 #ifndef HAVE_REMOVE_PROC_SUBTREE
217 /* for b=10866, global variable */
218 DECLARE_RWSEM(_lprocfs_lock);
219 EXPORT_SYMBOL(_lprocfs_lock);
220
221 static void lprocfs_remove_nolock(struct proc_dir_entry **proot)
222 {
223         struct proc_dir_entry *root = *proot;
224         struct proc_dir_entry *temp = root;
225         struct proc_dir_entry *rm_entry;
226         struct proc_dir_entry *parent;
227
228         *proot = NULL;
229         if (root == NULL || IS_ERR(root))
230                 return;
231
232         parent = root->parent;
233         LASSERT(parent != NULL);
234
235         while (1) {
236                 while (temp->subdir != NULL)
237                         temp = temp->subdir;
238
239                 rm_entry = temp;
240                 temp = temp->parent;
241
242                 /* Memory corruption once caused this to fail, and
243                    without this LASSERT we would loop here forever. */
244                 LASSERTF(strlen(rm_entry->name) == rm_entry->namelen,
245                          "0x%p  %s/%s len %d\n", rm_entry, temp->name,
246                          rm_entry->name, (int)strlen(rm_entry->name));
247
248                 remove_proc_entry(rm_entry->name, temp);
249                 if (temp == parent)
250                         break;
251         }
252 }
253
254 int remove_proc_subtree(const char *name, struct proc_dir_entry *parent)
255 {
256         struct proc_dir_entry    *t = NULL;
257         struct proc_dir_entry   **p;
258         int                       len, busy = 0;
259
260         LASSERT(parent != NULL);
261         len = strlen(name);
262
263         down_write(&_lprocfs_lock);
264         /* lookup target name */
265         for (p = &parent->subdir; *p; p = &(*p)->next) {
266                 if ((*p)->namelen != len)
267                         continue;
268                 if (memcmp(name, (*p)->name, len))
269                         continue;
270                 t = *p;
271                 break;
272         }
273
274         if (t) {
275                 /* verify it's empty: do not count "num_refs" */
276                 for (p = &t->subdir; *p; p = &(*p)->next) {
277                         if ((*p)->namelen != strlen("num_refs")) {
278                                 busy = 1;
279                                 break;
280                         }
281                         if (memcmp("num_refs", (*p)->name,
282                                    strlen("num_refs"))) {
283                                 busy = 1;
284                                 break;
285                         }
286                 }
287         }
288
289         if (busy == 0)
290                 lprocfs_remove_nolock(&t);
291
292         up_write(&_lprocfs_lock);
293         return 0;
294 }
295 #endif /* !HAVE_REMOVE_PROC_SUBTREE */
296
297 #ifndef HAVE_PROC_REMOVE
298 void proc_remove(struct proc_dir_entry *de)
299 {
300 #ifndef HAVE_REMOVE_PROC_SUBTREE
301         down_write(&_lprocfs_lock); /* search vs remove race */
302         lprocfs_remove_nolock(&de);
303         up_write(&_lprocfs_lock);
304 #else
305         if (de)
306                 remove_proc_subtree(de->name, de->parent);
307 #endif
308 }
309 #endif
310
311 void lprocfs_remove(struct proc_dir_entry **rooth)
312 {
313         proc_remove(*rooth);
314         *rooth = NULL;
315 }
316 EXPORT_SYMBOL(lprocfs_remove);
317
318 void lprocfs_remove_proc_entry(const char *name, struct proc_dir_entry *parent)
319 {
320         LASSERT(parent != NULL);
321         remove_proc_entry(name, parent);
322 }
323 EXPORT_SYMBOL(lprocfs_remove_proc_entry);
324
325 struct dentry *ldebugfs_register(const char *name, struct dentry *parent,
326                                  struct lprocfs_vars *list, void *data)
327 {
328         struct dentry *entry;
329
330         entry = debugfs_create_dir(name, parent);
331         if (IS_ERR_OR_NULL(entry)) {
332                 entry = entry ?: ERR_PTR(-ENOMEM);
333                 goto out;
334         }
335
336         if (!IS_ERR_OR_NULL(list)) {
337                 int rc;
338
339                 rc = ldebugfs_add_vars(entry, list, data);
340                 if (rc) {
341                         debugfs_remove(entry);
342                         entry = ERR_PTR(rc);
343                 }
344         }
345 out:
346         return entry;
347 }
348 EXPORT_SYMBOL_GPL(ldebugfs_register);
349
350 struct proc_dir_entry *
351 lprocfs_register(const char *name, struct proc_dir_entry *parent,
352                  struct lprocfs_vars *list, void *data)
353 {
354         struct proc_dir_entry *newchild;
355
356         newchild = proc_mkdir(name, parent);
357         if (newchild == NULL)
358                 return ERR_PTR(-ENOMEM);
359
360         if (list != NULL) {
361                 int rc = lprocfs_add_vars(newchild, list, data);
362                 if (rc) {
363                         lprocfs_remove(&newchild);
364                         return ERR_PTR(rc);
365                 }
366         }
367         return newchild;
368 }
369 EXPORT_SYMBOL(lprocfs_register);
370
371 /* Generic callbacks */
372 int lprocfs_uuid_seq_show(struct seq_file *m, void *data)
373 {
374         struct obd_device *obd = data;
375
376         LASSERT(obd != NULL);
377         seq_printf(m, "%s\n", obd->obd_uuid.uuid);
378         return 0;
379 }
380 EXPORT_SYMBOL(lprocfs_uuid_seq_show);
381
382 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
383                          char *buf)
384 {
385         struct obd_device *obd = container_of(kobj, struct obd_device,
386                                               obd_kset.kobj);
387
388         return sprintf(buf, "%s\n", obd->obd_uuid.uuid);
389 }
390 LUSTRE_RO_ATTR(uuid);
391
392 int lprocfs_name_seq_show(struct seq_file *m, void *data)
393 {
394         struct obd_device *dev = data;
395
396         LASSERT(dev != NULL);
397         seq_printf(m, "%s\n", dev->obd_name);
398         return 0;
399 }
400 EXPORT_SYMBOL(lprocfs_name_seq_show);
401
402 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
403                               char *buf)
404 {
405         struct obd_device *obd = container_of(kobj, struct obd_device,
406                                               obd_kset.kobj);
407         struct obd_statfs osfs;
408         int rc;
409
410         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
411                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
412                         OBD_STATFS_NODELAY);
413         if (!rc)
414                 return sprintf(buf, "%u\n", osfs.os_bsize);
415
416         return rc;
417 }
418 LUSTRE_RO_ATTR(blocksize);
419
420 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
421                                 char *buf)
422 {
423         struct obd_device *obd = container_of(kobj, struct obd_device,
424                                               obd_kset.kobj);
425         struct obd_statfs osfs;
426         int rc;
427
428         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
429                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
430                         OBD_STATFS_NODELAY);
431         if (!rc) {
432                 u32 blk_size = osfs.os_bsize >> 10;
433                 u64 result = osfs.os_blocks;
434
435                 while (blk_size >>= 1)
436                         result <<= 1;
437
438                 return sprintf(buf, "%llu\n", result);
439         }
440
441         return rc;
442 }
443 LUSTRE_RO_ATTR(kbytestotal);
444
445 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
446                                char *buf)
447 {
448         struct obd_device *obd = container_of(kobj, struct obd_device,
449                                               obd_kset.kobj);
450         struct obd_statfs osfs;
451         int rc;
452
453         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
454                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
455                         OBD_STATFS_NODELAY);
456         if (!rc) {
457                 u32 blk_size = osfs.os_bsize >> 10;
458                 u64 result = osfs.os_bfree;
459
460                 while (blk_size >>= 1)
461                         result <<= 1;
462
463                 return sprintf(buf, "%llu\n", result);
464         }
465
466         return rc;
467 }
468 LUSTRE_RO_ATTR(kbytesfree);
469
470 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
471                                 char *buf)
472 {
473         struct obd_device *obd = container_of(kobj, struct obd_device,
474                                               obd_kset.kobj);
475         struct obd_statfs osfs;
476         int rc;
477
478         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
479                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
480                         OBD_STATFS_NODELAY);
481         if (!rc) {
482                 u32 blk_size = osfs.os_bsize >> 10;
483                 u64 result = osfs.os_bavail;
484
485                 while (blk_size >>= 1)
486                         result <<= 1;
487
488                 return sprintf(buf, "%llu\n", result);
489         }
490
491         return rc;
492 }
493 LUSTRE_RO_ATTR(kbytesavail);
494
495 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
496                                char *buf)
497 {
498         struct obd_device *obd = container_of(kobj, struct obd_device,
499                                               obd_kset.kobj);
500         struct obd_statfs osfs;
501         int rc;
502
503         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
504                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
505                         OBD_STATFS_NODELAY);
506         if (!rc)
507                 return sprintf(buf, "%llu\n", osfs.os_files);
508
509         return rc;
510 }
511 LUSTRE_RO_ATTR(filestotal);
512
513 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
514                               char *buf)
515 {
516         struct obd_device *obd = container_of(kobj, struct obd_device,
517                                               obd_kset.kobj);
518         struct obd_statfs osfs;
519         int rc;
520
521         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
522                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
523                         OBD_STATFS_NODELAY);
524         if (!rc)
525                 return sprintf(buf, "%llu\n", osfs.os_ffree);
526
527         return rc;
528 }
529 LUSTRE_RO_ATTR(filesfree);
530
531 ssize_t conn_uuid_show(struct kobject *kobj, struct attribute *attr, char *buf)
532 {
533         struct obd_device *obd = container_of(kobj, struct obd_device,
534                                               obd_kset.kobj);
535         struct ptlrpc_connection *conn;
536         ssize_t count;
537
538         LPROCFS_CLIMP_CHECK(obd);
539         conn = obd->u.cli.cl_import->imp_connection;
540         if (conn && obd->u.cli.cl_import)
541                 count = sprintf(buf, "%s\n", conn->c_remote_uuid.uuid);
542         else
543                 count = sprintf(buf, "%s\n", "<none>");
544
545         LPROCFS_CLIMP_EXIT(obd);
546         return count;
547 }
548 EXPORT_SYMBOL(conn_uuid_show);
549
550 int lprocfs_server_uuid_seq_show(struct seq_file *m, void *data)
551 {
552         struct obd_device *obd = data;
553         struct obd_import *imp;
554         char *imp_state_name = NULL;
555         int rc = 0;
556
557         LASSERT(obd != NULL);
558         LPROCFS_CLIMP_CHECK(obd);
559         imp = obd->u.cli.cl_import;
560         imp_state_name = ptlrpc_import_state_name(imp->imp_state);
561         seq_printf(m, "%s\t%s%s\n", obd2cli_tgt(obd), imp_state_name,
562                    imp->imp_deactive ? "\tDEACTIVATED" : "");
563
564         LPROCFS_CLIMP_EXIT(obd);
565         return rc;
566 }
567 EXPORT_SYMBOL(lprocfs_server_uuid_seq_show);
568
569 int lprocfs_conn_uuid_seq_show(struct seq_file *m, void *data)
570 {
571         struct obd_device *obd = data;
572         struct ptlrpc_connection *conn;
573         int rc = 0;
574
575         LASSERT(obd != NULL);
576
577         LPROCFS_CLIMP_CHECK(obd);
578         conn = obd->u.cli.cl_import->imp_connection;
579         if (conn && obd->u.cli.cl_import)
580                 seq_printf(m, "%s\n", conn->c_remote_uuid.uuid);
581         else
582                 seq_printf(m, "%s\n", "<none>");
583
584         LPROCFS_CLIMP_EXIT(obd);
585         return rc;
586 }
587 EXPORT_SYMBOL(lprocfs_conn_uuid_seq_show);
588
589 /** add up per-cpu counters */
590
591 /**
592  * Lock statistics structure for access, possibly only on this CPU.
593  *
594  * The statistics struct may be allocated with per-CPU structures for
595  * efficient concurrent update (usually only on server-wide stats), or
596  * as a single global struct (e.g. for per-client or per-job statistics),
597  * so the required locking depends on the type of structure allocated.
598  *
599  * For per-CPU statistics, pin the thread to the current cpuid so that
600  * will only access the statistics for that CPU.  If the stats structure
601  * for the current CPU has not been allocated (or previously freed),
602  * allocate it now.  The per-CPU statistics do not need locking since
603  * the thread is pinned to the CPU during update.
604  *
605  * For global statistics, lock the stats structure to prevent concurrent update.
606  *
607  * \param[in] stats     statistics structure to lock
608  * \param[in] opc       type of operation:
609  *                      LPROCFS_GET_SMP_ID: "lock" and return current CPU index
610  *                              for incrementing statistics for that CPU
611  *                      LPROCFS_GET_NUM_CPU: "lock" and return number of used
612  *                              CPU indices to iterate over all indices
613  * \param[out] flags    CPU interrupt saved state for IRQ-safe locking
614  *
615  * \retval cpuid of current thread or number of allocated structs
616  * \retval negative on error (only for opc LPROCFS_GET_SMP_ID + per-CPU stats)
617  */
618 int lprocfs_stats_lock(struct lprocfs_stats *stats,
619                        enum lprocfs_stats_lock_ops opc,
620                        unsigned long *flags)
621 {
622         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) {
623                 if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
624                         spin_lock_irqsave(&stats->ls_lock, *flags);
625                 else
626                         spin_lock(&stats->ls_lock);
627                 return opc == LPROCFS_GET_NUM_CPU ? 1 : 0;
628         }
629
630         switch (opc) {
631         case LPROCFS_GET_SMP_ID: {
632                 unsigned int cpuid = get_cpu();
633
634                 if (unlikely(!stats->ls_percpu[cpuid])) {
635                         int rc = lprocfs_stats_alloc_one(stats, cpuid);
636
637                         if (rc < 0) {
638                                 put_cpu();
639                                 return rc;
640                         }
641                 }
642                 return cpuid;
643         }
644         case LPROCFS_GET_NUM_CPU:
645                 return stats->ls_biggest_alloc_num;
646         default:
647                 LBUG();
648         }
649 }
650
651 /**
652  * Unlock statistics structure after access.
653  *
654  * Unlock the lock acquired via lprocfs_stats_lock() for global statistics,
655  * or unpin this thread from the current cpuid for per-CPU statistics.
656  *
657  * This function must be called using the same arguments as used when calling
658  * lprocfs_stats_lock() so that the correct operation can be performed.
659  *
660  * \param[in] stats     statistics structure to unlock
661  * \param[in] opc       type of operation (current cpuid or number of structs)
662  * \param[in] flags     CPU interrupt saved state for IRQ-safe locking
663  */
664 void lprocfs_stats_unlock(struct lprocfs_stats *stats,
665                           enum lprocfs_stats_lock_ops opc,
666                           unsigned long *flags)
667 {
668         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) {
669                 if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
670                         spin_unlock_irqrestore(&stats->ls_lock, *flags);
671                 else
672                         spin_unlock(&stats->ls_lock);
673         } else if (opc == LPROCFS_GET_SMP_ID) {
674                 put_cpu();
675         }
676 }
677
678 /** add up per-cpu counters */
679 void lprocfs_stats_collect(struct lprocfs_stats *stats, int idx,
680                            struct lprocfs_counter *cnt)
681 {
682         unsigned int                    num_entry;
683         struct lprocfs_counter          *percpu_cntr;
684         int                             i;
685         unsigned long                   flags = 0;
686
687         memset(cnt, 0, sizeof(*cnt));
688
689         if (stats == NULL) {
690                 /* set count to 1 to avoid divide-by-zero errs in callers */
691                 cnt->lc_count = 1;
692                 return;
693         }
694
695         cnt->lc_min = LC_MIN_INIT;
696
697         num_entry = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
698
699         for (i = 0; i < num_entry; i++) {
700                 if (stats->ls_percpu[i] == NULL)
701                         continue;
702                 percpu_cntr = lprocfs_stats_counter_get(stats, i, idx);
703
704                 cnt->lc_count += percpu_cntr->lc_count;
705                 cnt->lc_sum += percpu_cntr->lc_sum;
706                 if (percpu_cntr->lc_min < cnt->lc_min)
707                         cnt->lc_min = percpu_cntr->lc_min;
708                 if (percpu_cntr->lc_max > cnt->lc_max)
709                         cnt->lc_max = percpu_cntr->lc_max;
710                 cnt->lc_sumsquare += percpu_cntr->lc_sumsquare;
711         }
712
713         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
714 }
715
716 static void obd_import_flags2str(struct obd_import *imp, struct seq_file *m)
717 {
718         bool first = true;
719
720         if (imp->imp_obd->obd_no_recov) {
721                 seq_printf(m, "no_recov");
722                 first = false;
723         }
724
725         flag2str(imp, invalid);
726         flag2str(imp, deactive);
727         flag2str(imp, replayable);
728         flag2str(imp, delayed_recovery);
729         flag2str(imp, vbr_failed);
730         flag2str(imp, pingable);
731         flag2str(imp, resend_replay);
732         flag2str(imp, no_pinger_recover);
733         flag2str(imp, connect_tried);
734 }
735
736 static const char *obd_connect_names[] = {
737         /* flags names  */
738         "read_only",
739         "lov_index",
740         "connect_from_mds",
741         "write_grant",
742         "server_lock",
743         "version",
744         "request_portal",
745         "acl",
746         "xattr",
747         "create_on_write",
748         "truncate_lock",
749         "initial_transno",
750         "inode_bit_locks",
751         "barrier",
752         "getattr_by_fid",
753         "no_oh_for_devices",
754         "remote_client",
755         "remote_client_by_force",
756         "max_byte_per_rpc",
757         "64bit_qdata",
758         "mds_capability",
759         "oss_capability",
760         "early_lock_cancel",
761         "som",
762         "adaptive_timeouts",
763         "lru_resize",
764         "mds_mds_connection",
765         "real_conn",
766         "change_qunit_size",
767         "alt_checksum_algorithm",
768         "fid_is_enabled",
769         "version_recovery",
770         "pools",
771         "grant_shrink",
772         "skip_orphan",
773         "large_ea",
774         "full20",
775         "layout_lock",
776         "64bithash",
777         "object_max_bytes",
778         "imp_recov",
779         "jobstats",
780         "umask",
781         "einprogress",
782         "grant_param",
783         "flock_owner",
784         "lvb_type",
785         "nanoseconds_times",
786         "lightweight_conn",
787         "short_io",
788         "pingless",
789         "flock_deadlock",
790         "disp_stripe",
791         "open_by_fid",
792         "lfsck",
793         "unknown",
794         "unlink_close",
795         "multi_mod_rpcs",
796         "dir_stripe",
797         "subtree",
798         "lockahead",
799         "bulk_mbits",
800         "compact_obdo",
801         "second_flags",
802         /* flags2 names */
803         "file_secctx",  /* 0x01 */
804         "lockaheadv2",  /* 0x02 */
805         "dir_migrate",  /* 0x04 */
806         "unknown",      /* 0x08 */
807         "unknown",      /* 0x10 */
808         "flr",          /* 0x20 */
809         "wbc",          /* 0x40 */
810         "lock_convert",  /* 0x80 */
811         "archive_id_array",     /* 0x100 */
812         NULL
813 };
814
815 void obd_connect_seq_flags2str(struct seq_file *m, __u64 flags, __u64 flags2,
816                                const char *sep)
817 {
818         bool first = true;
819         __u64 mask;
820         int i;
821
822         for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
823                 if (flags & mask) {
824                         seq_printf(m, "%s%s",
825                                    first ? "" : sep, obd_connect_names[i]);
826                         first = false;
827                 }
828         }
829
830         if (flags & ~(mask - 1)) {
831                 seq_printf(m, "%sunknown_%#llx",
832                            first ? "" : sep, flags & ~(mask - 1));
833                 first = false;
834         }
835
836         if (!(flags & OBD_CONNECT_FLAGS2) || flags2 == 0)
837                 return;
838
839         for (i = 64, mask = 1; obd_connect_names[i] != NULL; i++, mask <<= 1) {
840                 if (flags2 & mask) {
841                         seq_printf(m, "%s%s",
842                                    first ? "" : sep, obd_connect_names[i]);
843                         first = false;
844                 }
845         }
846
847         if (flags2 & ~(mask - 1)) {
848                 seq_printf(m, "%sunknown2_%#llx",
849                            first ? "" : sep, flags2 & ~(mask - 1));
850                 first = false;
851         }
852 }
853 EXPORT_SYMBOL(obd_connect_seq_flags2str);
854
855 int obd_connect_flags2str(char *page, int count, __u64 flags, __u64 flags2,
856                           const char *sep)
857 {
858         __u64 mask;
859         int i, ret = 0;
860
861         for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
862                 if (flags & mask)
863                         ret += snprintf(page + ret, count - ret, "%s%s",
864                                         ret ? sep : "", obd_connect_names[i]);
865         }
866
867         if (flags & ~(mask - 1))
868                 ret += snprintf(page + ret, count - ret,
869                                 "%sunknown_%#llx",
870                                 ret ? sep : "", flags & ~(mask - 1));
871
872         if (!(flags & OBD_CONNECT_FLAGS2) || flags2 == 0)
873                 return ret;
874
875         for (i = 64, mask = 1; obd_connect_names[i] != NULL; i++, mask <<= 1) {
876                 if (flags2 & mask)
877                         ret += snprintf(page + ret, count - ret, "%s%s",
878                                         ret ? sep : "", obd_connect_names[i]);
879         }
880
881         if (flags2 & ~(mask - 1))
882                 ret += snprintf(page + ret, count - ret,
883                                 "%sunknown2_%#llx",
884                                 ret ? sep : "", flags2 & ~(mask - 1));
885
886         return ret;
887 }
888 EXPORT_SYMBOL(obd_connect_flags2str);
889
890 void
891 obd_connect_data_seqprint(struct seq_file *m, struct obd_connect_data *ocd)
892 {
893         __u64 flags;
894
895         LASSERT(ocd != NULL);
896         flags = ocd->ocd_connect_flags;
897
898         seq_printf(m, "    connect_data:\n"
899                    "       flags: %#llx\n"
900                    "       instance: %u\n",
901                    ocd->ocd_connect_flags,
902                    ocd->ocd_instance);
903         if (flags & OBD_CONNECT_VERSION)
904                 seq_printf(m, "       target_version: %u.%u.%u.%u\n",
905                            OBD_OCD_VERSION_MAJOR(ocd->ocd_version),
906                            OBD_OCD_VERSION_MINOR(ocd->ocd_version),
907                            OBD_OCD_VERSION_PATCH(ocd->ocd_version),
908                            OBD_OCD_VERSION_FIX(ocd->ocd_version));
909         if (flags & OBD_CONNECT_MDS)
910                 seq_printf(m, "       mdt_index: %d\n", ocd->ocd_group);
911         if (flags & OBD_CONNECT_GRANT)
912                 seq_printf(m, "       initial_grant: %d\n", ocd->ocd_grant);
913         if (flags & OBD_CONNECT_INDEX)
914                 seq_printf(m, "       target_index: %u\n", ocd->ocd_index);
915         if (flags & OBD_CONNECT_BRW_SIZE)
916                 seq_printf(m, "       max_brw_size: %d\n", ocd->ocd_brw_size);
917         if (flags & OBD_CONNECT_IBITS)
918                 seq_printf(m, "       ibits_known: %#llx\n",
919                            ocd->ocd_ibits_known);
920         if (flags & OBD_CONNECT_GRANT_PARAM)
921                 seq_printf(m, "       grant_block_size: %d\n"
922                            "       grant_inode_size: %d\n"
923                            "       grant_max_extent_size: %d\n"
924                            "       grant_extent_tax: %d\n",
925                            1 << ocd->ocd_grant_blkbits,
926                            1 << ocd->ocd_grant_inobits,
927                            ocd->ocd_grant_max_blks << ocd->ocd_grant_blkbits,
928                            ocd->ocd_grant_tax_kb << 10);
929         if (flags & OBD_CONNECT_TRANSNO)
930                 seq_printf(m, "       first_transno: %#llx\n",
931                            ocd->ocd_transno);
932         if (flags & OBD_CONNECT_CKSUM)
933                 seq_printf(m, "       cksum_types: %#x\n",
934                            ocd->ocd_cksum_types);
935         if (flags & OBD_CONNECT_MAX_EASIZE)
936                 seq_printf(m, "       max_easize: %d\n", ocd->ocd_max_easize);
937         if (flags & OBD_CONNECT_MAXBYTES)
938                 seq_printf(m, "       max_object_bytes: %llu\n",
939                            ocd->ocd_maxbytes);
940         if (flags & OBD_CONNECT_MULTIMODRPCS)
941                 seq_printf(m, "       max_mod_rpcs: %hu\n",
942                            ocd->ocd_maxmodrpcs);
943 }
944
945 int lprocfs_import_seq_show(struct seq_file *m, void *data)
946 {
947         char                            nidstr[LNET_NIDSTR_SIZE];
948         struct lprocfs_counter          ret;
949         struct lprocfs_counter_header   *header;
950         struct obd_device               *obd    = (struct obd_device *)data;
951         struct obd_import               *imp;
952         struct obd_import_conn          *conn;
953         struct obd_connect_data         *ocd;
954         int                             j;
955         int                             k;
956         int                             rw      = 0;
957
958         LASSERT(obd != NULL);
959         LPROCFS_CLIMP_CHECK(obd);
960         imp = obd->u.cli.cl_import;
961         ocd = &imp->imp_connect_data;
962
963         seq_printf(m, "import:\n"
964                    "    name: %s\n"
965                    "    target: %s\n"
966                    "    state: %s\n"
967                    "    connect_flags: [ ",
968                    obd->obd_name,
969                    obd2cli_tgt(obd),
970                    ptlrpc_import_state_name(imp->imp_state));
971         obd_connect_seq_flags2str(m, imp->imp_connect_data.ocd_connect_flags,
972                                   imp->imp_connect_data.ocd_connect_flags2,
973                                   ", ");
974         seq_printf(m, " ]\n");
975         obd_connect_data_seqprint(m, ocd);
976         seq_printf(m, "    import_flags: [ ");
977         obd_import_flags2str(imp, m);
978
979         seq_printf(m, " ]\n"
980                    "    connection:\n"
981                    "       failover_nids: [ ");
982         spin_lock(&imp->imp_lock);
983         j = 0;
984         list_for_each_entry(conn, &imp->imp_conn_list, oic_item) {
985                 libcfs_nid2str_r(conn->oic_conn->c_peer.nid,
986                                  nidstr, sizeof(nidstr));
987                 seq_printf(m, "%s%s", j ? ", " : "", nidstr);
988                 j++;
989         }
990         if (imp->imp_connection != NULL)
991                 libcfs_nid2str_r(imp->imp_connection->c_peer.nid,
992                                  nidstr, sizeof(nidstr));
993         else
994                 strncpy(nidstr, "<none>", sizeof(nidstr));
995         seq_printf(m, " ]\n"
996                    "       current_connection: %s\n"
997                    "       connection_attempts: %u\n"
998                    "       generation: %u\n"
999                    "       in-progress_invalidations: %u\n",
1000                    nidstr,
1001                    imp->imp_conn_cnt,
1002                    imp->imp_generation,
1003                    atomic_read(&imp->imp_inval_count));
1004         spin_unlock(&imp->imp_lock);
1005
1006         if (obd->obd_svc_stats == NULL)
1007                 goto out_climp;
1008
1009         header = &obd->obd_svc_stats->ls_cnt_header[PTLRPC_REQWAIT_CNTR];
1010         lprocfs_stats_collect(obd->obd_svc_stats, PTLRPC_REQWAIT_CNTR, &ret);
1011         if (ret.lc_count != 0) {
1012                 /* first argument to do_div MUST be __u64 */
1013                 __u64 sum = ret.lc_sum;
1014                 do_div(sum, ret.lc_count);
1015                 ret.lc_sum = sum;
1016         } else
1017                 ret.lc_sum = 0;
1018         seq_printf(m, "    rpcs:\n"
1019                    "       inflight: %u\n"
1020                    "       unregistering: %u\n"
1021                    "       timeouts: %u\n"
1022                    "       avg_waittime: %llu %s\n",
1023                    atomic_read(&imp->imp_inflight),
1024                    atomic_read(&imp->imp_unregistering),
1025                    atomic_read(&imp->imp_timeouts),
1026                    ret.lc_sum, header->lc_units);
1027
1028         k = 0;
1029         for(j = 0; j < IMP_AT_MAX_PORTALS; j++) {
1030                 if (imp->imp_at.iat_portal[j] == 0)
1031                         break;
1032                 k = max_t(unsigned int, k,
1033                           at_get(&imp->imp_at.iat_service_estimate[j]));
1034         }
1035         seq_printf(m, "    service_estimates:\n"
1036                    "       services: %u sec\n"
1037                    "       network: %u sec\n",
1038                    k,
1039                    at_get(&imp->imp_at.iat_net_latency));
1040
1041         seq_printf(m, "    transactions:\n"
1042                    "       last_replay: %llu\n"
1043                    "       peer_committed: %llu\n"
1044                    "       last_checked: %llu\n",
1045                    imp->imp_last_replay_transno,
1046                    imp->imp_peer_committed_transno,
1047                    imp->imp_last_transno_checked);
1048
1049         /* avg data rates */
1050         for (rw = 0; rw <= 1; rw++) {
1051                 lprocfs_stats_collect(obd->obd_svc_stats,
1052                                       PTLRPC_LAST_CNTR + BRW_READ_BYTES + rw,
1053                                       &ret);
1054                 if (ret.lc_sum > 0 && ret.lc_count > 0) {
1055                         /* first argument to do_div MUST be __u64 */
1056                         __u64 sum = ret.lc_sum;
1057                         do_div(sum, ret.lc_count);
1058                         ret.lc_sum = sum;
1059                         seq_printf(m, "    %s_data_averages:\n"
1060                                    "       bytes_per_rpc: %llu\n",
1061                                    rw ? "write" : "read",
1062                                    ret.lc_sum);
1063                 }
1064                 k = (int)ret.lc_sum;
1065                 j = opcode_offset(OST_READ + rw) + EXTRA_MAX_OPCODES;
1066                 header = &obd->obd_svc_stats->ls_cnt_header[j];
1067                 lprocfs_stats_collect(obd->obd_svc_stats, j, &ret);
1068                 if (ret.lc_sum > 0 && ret.lc_count != 0) {
1069                         /* first argument to do_div MUST be __u64 */
1070                         __u64 sum = ret.lc_sum;
1071                         do_div(sum, ret.lc_count);
1072                         ret.lc_sum = sum;
1073                         seq_printf(m, "       %s_per_rpc: %llu\n",
1074                                    header->lc_units, ret.lc_sum);
1075                         j = (int)ret.lc_sum;
1076                         if (j > 0)
1077                                 seq_printf(m, "       MB_per_sec: %u.%.02u\n",
1078                                            k / j, (100 * k / j) % 100);
1079                 }
1080         }
1081
1082 out_climp:
1083         LPROCFS_CLIMP_EXIT(obd);
1084         return 0;
1085 }
1086 EXPORT_SYMBOL(lprocfs_import_seq_show);
1087
1088 int lprocfs_state_seq_show(struct seq_file *m, void *data)
1089 {
1090         struct obd_device *obd = (struct obd_device *)data;
1091         struct obd_import *imp;
1092         int j, k;
1093
1094         LASSERT(obd != NULL);
1095         LPROCFS_CLIMP_CHECK(obd);
1096         imp = obd->u.cli.cl_import;
1097
1098         seq_printf(m, "current_state: %s\n",
1099                    ptlrpc_import_state_name(imp->imp_state));
1100         seq_printf(m, "state_history:\n");
1101         k = imp->imp_state_hist_idx;
1102         for (j = 0; j < IMP_STATE_HIST_LEN; j++) {
1103                 struct import_state_hist *ish =
1104                         &imp->imp_state_hist[(k + j) % IMP_STATE_HIST_LEN];
1105                 if (ish->ish_state == 0)
1106                         continue;
1107                 seq_printf(m, " - [ %lld, %s ]\n", (s64)ish->ish_time,
1108                            ptlrpc_import_state_name(ish->ish_state));
1109         }
1110
1111         LPROCFS_CLIMP_EXIT(obd);
1112         return 0;
1113 }
1114 EXPORT_SYMBOL(lprocfs_state_seq_show);
1115
1116 int lprocfs_at_hist_helper(struct seq_file *m, struct adaptive_timeout *at)
1117 {
1118         int i;
1119         for (i = 0; i < AT_BINS; i++)
1120                 seq_printf(m, "%3u ", at->at_hist[i]);
1121         seq_printf(m, "\n");
1122         return 0;
1123 }
1124 EXPORT_SYMBOL(lprocfs_at_hist_helper);
1125
1126 /* See also ptlrpc_lprocfs_timeouts_show_seq */
1127 int lprocfs_timeouts_seq_show(struct seq_file *m, void *data)
1128 {
1129         struct obd_device *obd = (struct obd_device *)data;
1130         struct obd_import *imp;
1131         unsigned int cur, worst;
1132         time64_t now, worstt;
1133         int i;
1134
1135         LASSERT(obd != NULL);
1136         LPROCFS_CLIMP_CHECK(obd);
1137         imp = obd->u.cli.cl_import;
1138
1139         now = ktime_get_real_seconds();
1140
1141         /* Some network health info for kicks */
1142         seq_printf(m, "%-10s : %lld, %llds ago\n",
1143                    "last reply", (s64)imp->imp_last_reply_time,
1144                    (s64)(now - imp->imp_last_reply_time));
1145
1146         cur = at_get(&imp->imp_at.iat_net_latency);
1147         worst = imp->imp_at.iat_net_latency.at_worst_ever;
1148         worstt = imp->imp_at.iat_net_latency.at_worst_time;
1149         seq_printf(m, "%-10s : cur %3u  worst %3u (at %lld, %llds ago) ",
1150                    "network", cur, worst, (s64)worstt, (s64)(now - worstt));
1151         lprocfs_at_hist_helper(m, &imp->imp_at.iat_net_latency);
1152
1153         for(i = 0; i < IMP_AT_MAX_PORTALS; i++) {
1154                 if (imp->imp_at.iat_portal[i] == 0)
1155                         break;
1156                 cur = at_get(&imp->imp_at.iat_service_estimate[i]);
1157                 worst = imp->imp_at.iat_service_estimate[i].at_worst_ever;
1158                 worstt = imp->imp_at.iat_service_estimate[i].at_worst_time;
1159                 seq_printf(m, "portal %-2d  : cur %3u  worst %3u (at %lld, %llds ago) ",
1160                            imp->imp_at.iat_portal[i], cur, worst, (s64)worstt,
1161                            (s64)(now - worstt));
1162                 lprocfs_at_hist_helper(m, &imp->imp_at.iat_service_estimate[i]);
1163         }
1164
1165         LPROCFS_CLIMP_EXIT(obd);
1166         return 0;
1167 }
1168 EXPORT_SYMBOL(lprocfs_timeouts_seq_show);
1169
1170 int lprocfs_connect_flags_seq_show(struct seq_file *m, void *data)
1171 {
1172         struct obd_device *obd = data;
1173         __u64 flags;
1174         __u64 flags2;
1175
1176         LPROCFS_CLIMP_CHECK(obd);
1177         flags = obd->u.cli.cl_import->imp_connect_data.ocd_connect_flags;
1178         flags2 = obd->u.cli.cl_import->imp_connect_data.ocd_connect_flags2;
1179         seq_printf(m, "flags=%#llx\n", flags);
1180         seq_printf(m, "flags2=%#llx\n", flags2);
1181         obd_connect_seq_flags2str(m, flags, flags2, "\n");
1182         seq_printf(m, "\n");
1183         LPROCFS_CLIMP_EXIT(obd);
1184         return 0;
1185 }
1186 EXPORT_SYMBOL(lprocfs_connect_flags_seq_show);
1187
1188 static const struct attribute *obd_def_uuid_attrs[] = {
1189         &lustre_attr_uuid.attr,
1190         NULL,
1191 };
1192
1193 static const struct attribute *obd_def_attrs[] = {
1194         &lustre_attr_blocksize.attr,
1195         &lustre_attr_kbytestotal.attr,
1196         &lustre_attr_kbytesfree.attr,
1197         &lustre_attr_kbytesavail.attr,
1198         &lustre_attr_filestotal.attr,
1199         &lustre_attr_filesfree.attr,
1200         &lustre_attr_uuid.attr,
1201         NULL,
1202 };
1203
1204 static void obd_sysfs_release(struct kobject *kobj)
1205 {
1206         struct obd_device *obd = container_of(kobj, struct obd_device,
1207                                               obd_kset.kobj);
1208
1209         complete(&obd->obd_kobj_unregister);
1210 }
1211
1212 int lprocfs_obd_setup(struct obd_device *obd, bool uuid_only)
1213 {
1214         struct lprocfs_vars *debugfs_vars = NULL;
1215         int rc;
1216
1217         if (!obd || obd->obd_magic != OBD_DEVICE_MAGIC)
1218                 return -ENODEV;
1219
1220         rc = kobject_set_name(&obd->obd_kset.kobj, "%s", obd->obd_name);
1221         if (rc)
1222                 return rc;
1223
1224         obd->obd_ktype.sysfs_ops = &lustre_sysfs_ops;
1225         obd->obd_ktype.release = obd_sysfs_release;
1226
1227         obd->obd_kset.kobj.parent = obd->obd_type->typ_kobj;
1228         obd->obd_kset.kobj.ktype = &obd->obd_ktype;
1229         init_completion(&obd->obd_kobj_unregister);
1230         rc = kset_register(&obd->obd_kset);
1231         if (rc)
1232                 return rc;
1233
1234         if (uuid_only)
1235                 obd->obd_attrs = obd_def_uuid_attrs;
1236         else
1237                 obd->obd_attrs = obd_def_attrs;
1238
1239         rc = sysfs_create_files(&obd->obd_kset.kobj, obd->obd_attrs);
1240         if (rc) {
1241                 kset_unregister(&obd->obd_kset);
1242                 return rc;
1243         }
1244
1245         if (!obd->obd_type->typ_procroot)
1246                 debugfs_vars = obd->obd_vars;
1247         obd->obd_debugfs_entry = ldebugfs_register(obd->obd_name,
1248                                                    obd->obd_type->typ_debugfs_entry,
1249                                                    debugfs_vars, obd);
1250         if (IS_ERR_OR_NULL(obd->obd_debugfs_entry)) {
1251                 rc = obd->obd_debugfs_entry ? PTR_ERR(obd->obd_debugfs_entry)
1252                                             : -ENOMEM;
1253                 CERROR("error %d setting up debugfs for %s\n",
1254                        rc, obd->obd_name);
1255                 obd->obd_debugfs_entry = NULL;
1256
1257                 sysfs_remove_files(&obd->obd_kset.kobj, obd->obd_attrs);
1258                 obd->obd_attrs = NULL;
1259                 kset_unregister(&obd->obd_kset);
1260                 return rc;
1261         }
1262
1263         if (obd->obd_proc_entry || !obd->obd_type->typ_procroot)
1264                 GOTO(already_registered, rc);
1265
1266         obd->obd_proc_entry = lprocfs_register(obd->obd_name,
1267                                                obd->obd_type->typ_procroot,
1268                                                obd->obd_vars, obd);
1269         if (IS_ERR(obd->obd_proc_entry)) {
1270                 rc = PTR_ERR(obd->obd_proc_entry);
1271                 CERROR("error %d setting up lprocfs for %s\n",rc,obd->obd_name);
1272                 obd->obd_proc_entry = NULL;
1273
1274                 ldebugfs_remove(&obd->obd_debugfs_entry);
1275                 sysfs_remove_files(&obd->obd_kset.kobj, obd->obd_attrs);
1276                 obd->obd_attrs = NULL;
1277                 kset_unregister(&obd->obd_kset);
1278                 return rc;
1279         }
1280 already_registered:
1281         return rc;
1282 }
1283 EXPORT_SYMBOL(lprocfs_obd_setup);
1284
1285 int lprocfs_obd_cleanup(struct obd_device *obd)
1286 {
1287         if (!obd)
1288                 return -EINVAL;
1289
1290         if (obd->obd_proc_exports_entry) {
1291                 /* Should be no exports left */
1292                 lprocfs_remove(&obd->obd_proc_exports_entry);
1293                 obd->obd_proc_exports_entry = NULL;
1294         }
1295
1296         if (obd->obd_proc_entry) {
1297                 lprocfs_remove(&obd->obd_proc_entry);
1298                 obd->obd_proc_entry = NULL;
1299         }
1300
1301         if (!IS_ERR_OR_NULL(obd->obd_debugfs_entry))
1302                 ldebugfs_remove(&obd->obd_debugfs_entry);
1303
1304         /* obd device never allocated a kset */
1305         if (!obd->obd_kset.kobj.state_initialized)
1306                 return 0;
1307
1308         if (obd->obd_attrs) {
1309                 sysfs_remove_files(&obd->obd_kset.kobj, obd->obd_attrs);
1310                 obd->obd_attrs = NULL;
1311         }
1312
1313         kset_unregister(&obd->obd_kset);
1314         wait_for_completion(&obd->obd_kobj_unregister);
1315         return 0;
1316 }
1317 EXPORT_SYMBOL(lprocfs_obd_cleanup);
1318
1319 int lprocfs_stats_alloc_one(struct lprocfs_stats *stats, unsigned int cpuid)
1320 {
1321         struct lprocfs_counter  *cntr;
1322         unsigned int            percpusize;
1323         int                     rc = -ENOMEM;
1324         unsigned long           flags = 0;
1325         int                     i;
1326
1327         LASSERT(stats->ls_percpu[cpuid] == NULL);
1328         LASSERT((stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) == 0);
1329
1330         percpusize = lprocfs_stats_counter_size(stats);
1331         LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[cpuid], percpusize);
1332         if (stats->ls_percpu[cpuid] != NULL) {
1333                 rc = 0;
1334                 if (unlikely(stats->ls_biggest_alloc_num <= cpuid)) {
1335                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1336                                 spin_lock_irqsave(&stats->ls_lock, flags);
1337                         else
1338                                 spin_lock(&stats->ls_lock);
1339                         if (stats->ls_biggest_alloc_num <= cpuid)
1340                                 stats->ls_biggest_alloc_num = cpuid + 1;
1341                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) {
1342                                 spin_unlock_irqrestore(&stats->ls_lock, flags);
1343                         } else {
1344                                 spin_unlock(&stats->ls_lock);
1345                         }
1346                 }
1347                 /* initialize the ls_percpu[cpuid] non-zero counter */
1348                 for (i = 0; i < stats->ls_num; ++i) {
1349                         cntr = lprocfs_stats_counter_get(stats, cpuid, i);
1350                         cntr->lc_min = LC_MIN_INIT;
1351                 }
1352         }
1353         return rc;
1354 }
1355
1356 struct lprocfs_stats *lprocfs_alloc_stats(unsigned int num,
1357                                           enum lprocfs_stats_flags flags)
1358 {
1359         struct lprocfs_stats    *stats;
1360         unsigned int            num_entry;
1361         unsigned int            percpusize = 0;
1362         int                     i;
1363
1364         if (num == 0)
1365                 return NULL;
1366
1367         if (lprocfs_no_percpu_stats != 0)
1368                 flags |= LPROCFS_STATS_FLAG_NOPERCPU;
1369
1370         if (flags & LPROCFS_STATS_FLAG_NOPERCPU)
1371                 num_entry = 1;
1372         else
1373                 num_entry = num_possible_cpus();
1374
1375         /* alloc percpu pointers for all possible cpu slots */
1376         LIBCFS_ALLOC(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1377         if (stats == NULL)
1378                 return NULL;
1379
1380         stats->ls_num = num;
1381         stats->ls_flags = flags;
1382         spin_lock_init(&stats->ls_lock);
1383
1384         /* alloc num of counter headers */
1385         LIBCFS_ALLOC(stats->ls_cnt_header,
1386                      stats->ls_num * sizeof(struct lprocfs_counter_header));
1387         if (stats->ls_cnt_header == NULL)
1388                 goto fail;
1389
1390         if ((flags & LPROCFS_STATS_FLAG_NOPERCPU) != 0) {
1391                 /* contains only one set counters */
1392                 percpusize = lprocfs_stats_counter_size(stats);
1393                 LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[0], percpusize);
1394                 if (stats->ls_percpu[0] == NULL)
1395                         goto fail;
1396                 stats->ls_biggest_alloc_num = 1;
1397         } else if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0) {
1398                 /* alloc all percpu data, currently only obd_memory use this */
1399                 for (i = 0; i < num_entry; ++i)
1400                         if (lprocfs_stats_alloc_one(stats, i) < 0)
1401                                 goto fail;
1402         }
1403
1404         return stats;
1405
1406 fail:
1407         lprocfs_free_stats(&stats);
1408         return NULL;
1409 }
1410 EXPORT_SYMBOL(lprocfs_alloc_stats);
1411
1412 void lprocfs_free_stats(struct lprocfs_stats **statsh)
1413 {
1414         struct lprocfs_stats *stats = *statsh;
1415         unsigned int num_entry;
1416         unsigned int percpusize;
1417         unsigned int i;
1418
1419         if (stats == NULL || stats->ls_num == 0)
1420                 return;
1421         *statsh = NULL;
1422
1423         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU)
1424                 num_entry = 1;
1425         else
1426                 num_entry = num_possible_cpus();
1427
1428         percpusize = lprocfs_stats_counter_size(stats);
1429         for (i = 0; i < num_entry; i++)
1430                 if (stats->ls_percpu[i] != NULL)
1431                         LIBCFS_FREE(stats->ls_percpu[i], percpusize);
1432         if (stats->ls_cnt_header != NULL)
1433                 LIBCFS_FREE(stats->ls_cnt_header, stats->ls_num *
1434                                         sizeof(struct lprocfs_counter_header));
1435         LIBCFS_FREE(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1436 }
1437 EXPORT_SYMBOL(lprocfs_free_stats);
1438
1439 u64 lprocfs_stats_collector(struct lprocfs_stats *stats, int idx,
1440                             enum lprocfs_fields_flags field)
1441 {
1442         unsigned long flags = 0;
1443         unsigned int num_cpu;
1444         unsigned int i;
1445         u64 ret = 0;
1446
1447         LASSERT(stats);
1448
1449         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1450         for (i = 0; i < num_cpu; i++) {
1451                 struct lprocfs_counter *cntr;
1452
1453                 if (!stats->ls_percpu[i])
1454                         continue;
1455
1456                 cntr = lprocfs_stats_counter_get(stats, i, idx);
1457                 ret += lprocfs_read_helper(cntr, &stats->ls_cnt_header[idx],
1458                                            stats->ls_flags, field);
1459         }
1460         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1461         return ret;
1462 }
1463 EXPORT_SYMBOL(lprocfs_stats_collector);
1464
1465 void lprocfs_clear_stats(struct lprocfs_stats *stats)
1466 {
1467         struct lprocfs_counter          *percpu_cntr;
1468         int                             i;
1469         int                             j;
1470         unsigned int                    num_entry;
1471         unsigned long                   flags = 0;
1472
1473         num_entry = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1474
1475         for (i = 0; i < num_entry; i++) {
1476                 if (stats->ls_percpu[i] == NULL)
1477                         continue;
1478                 for (j = 0; j < stats->ls_num; j++) {
1479                         percpu_cntr = lprocfs_stats_counter_get(stats, i, j);
1480                         percpu_cntr->lc_count           = 0;
1481                         percpu_cntr->lc_min             = LC_MIN_INIT;
1482                         percpu_cntr->lc_max             = 0;
1483                         percpu_cntr->lc_sumsquare       = 0;
1484                         percpu_cntr->lc_sum             = 0;
1485                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1486                                 percpu_cntr->lc_sum_irq = 0;
1487                 }
1488         }
1489
1490         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1491 }
1492 EXPORT_SYMBOL(lprocfs_clear_stats);
1493
1494 static ssize_t lprocfs_stats_seq_write(struct file *file,
1495                                        const char __user *buf,
1496                                        size_t len, loff_t *off)
1497 {
1498         struct seq_file *seq = file->private_data;
1499         struct lprocfs_stats *stats = seq->private;
1500
1501         lprocfs_clear_stats(stats);
1502
1503         return len;
1504 }
1505
1506 static void *lprocfs_stats_seq_start(struct seq_file *p, loff_t *pos)
1507 {
1508         struct lprocfs_stats *stats = p->private;
1509
1510         return (*pos < stats->ls_num) ? pos : NULL;
1511 }
1512
1513 static void lprocfs_stats_seq_stop(struct seq_file *p, void *v)
1514 {
1515 }
1516
1517 static void *lprocfs_stats_seq_next(struct seq_file *p, void *v, loff_t *pos)
1518 {
1519         (*pos)++;
1520
1521         return lprocfs_stats_seq_start(p, pos);
1522 }
1523
1524 /* seq file export of one lprocfs counter */
1525 static int lprocfs_stats_seq_show(struct seq_file *p, void *v)
1526 {
1527         struct lprocfs_stats            *stats  = p->private;
1528         struct lprocfs_counter_header   *hdr;
1529         struct lprocfs_counter           ctr;
1530         int                              idx    = *(loff_t *)v;
1531
1532         if (idx == 0) {
1533                 struct timespec64 now;
1534
1535                 ktime_get_real_ts64(&now);
1536                 seq_printf(p, "%-25s %llu.%09lu secs.nsecs\n",
1537                            "snapshot_time", (s64)now.tv_sec, now.tv_nsec);
1538         }
1539
1540         hdr = &stats->ls_cnt_header[idx];
1541         lprocfs_stats_collect(stats, idx, &ctr);
1542
1543         if (ctr.lc_count == 0)
1544                 return 0;
1545
1546         seq_printf(p, "%-25s %lld samples [%s]", hdr->lc_name,
1547                    ctr.lc_count, hdr->lc_units);
1548
1549         if ((hdr->lc_config & LPROCFS_CNTR_AVGMINMAX) && ctr.lc_count > 0) {
1550                 seq_printf(p, " %lld %lld %lld",
1551                            ctr.lc_min, ctr.lc_max, ctr.lc_sum);
1552                 if (hdr->lc_config & LPROCFS_CNTR_STDDEV)
1553                         seq_printf(p, " %llu", ctr.lc_sumsquare);
1554         }
1555         seq_putc(p, '\n');
1556         return 0;
1557 }
1558
1559 static const struct seq_operations lprocfs_stats_seq_sops = {
1560         .start  = lprocfs_stats_seq_start,
1561         .stop   = lprocfs_stats_seq_stop,
1562         .next   = lprocfs_stats_seq_next,
1563         .show   = lprocfs_stats_seq_show,
1564 };
1565
1566 static int lprocfs_stats_seq_open(struct inode *inode, struct file *file)
1567 {
1568         struct seq_file *seq;
1569         int rc;
1570
1571         rc = LPROCFS_ENTRY_CHECK(inode);
1572         if (rc < 0)
1573                 return rc;
1574
1575         rc = seq_open(file, &lprocfs_stats_seq_sops);
1576         if (rc)
1577                 return rc;
1578         seq = file->private_data;
1579         seq->private = inode->i_private ? inode->i_private : PDE_DATA(inode);
1580         return 0;
1581 }
1582
1583 static const struct file_operations lprocfs_stats_seq_fops = {
1584         .owner   = THIS_MODULE,
1585         .open    = lprocfs_stats_seq_open,
1586         .read    = seq_read,
1587         .write   = lprocfs_stats_seq_write,
1588         .llseek  = seq_lseek,
1589         .release = lprocfs_seq_release,
1590 };
1591
1592 int ldebugfs_register_stats(struct dentry *parent, const char *name,
1593                             struct lprocfs_stats *stats)
1594 {
1595         struct dentry *entry;
1596
1597         LASSERT(!IS_ERR_OR_NULL(parent));
1598
1599         entry = debugfs_create_file(name, 0644, parent, stats,
1600                                     &lprocfs_stats_seq_fops);
1601         if (IS_ERR_OR_NULL(entry))
1602                 return entry ? PTR_ERR(entry) : -ENOMEM;
1603
1604         return 0;
1605 }
1606 EXPORT_SYMBOL_GPL(ldebugfs_register_stats);
1607
1608 int lprocfs_register_stats(struct proc_dir_entry *root, const char *name,
1609                            struct lprocfs_stats *stats)
1610 {
1611         struct proc_dir_entry *entry;
1612         LASSERT(root != NULL);
1613
1614         entry = proc_create_data(name, 0644, root,
1615                                  &lprocfs_stats_seq_fops, stats);
1616         if (entry == NULL)
1617                 return -ENOMEM;
1618         return 0;
1619 }
1620 EXPORT_SYMBOL(lprocfs_register_stats);
1621
1622 void lprocfs_counter_init(struct lprocfs_stats *stats, int index,
1623                           unsigned conf, const char *name, const char *units)
1624 {
1625         struct lprocfs_counter_header   *header;
1626         struct lprocfs_counter          *percpu_cntr;
1627         unsigned long                   flags = 0;
1628         unsigned int                    i;
1629         unsigned int                    num_cpu;
1630
1631         LASSERT(stats != NULL);
1632
1633         header = &stats->ls_cnt_header[index];
1634         LASSERTF(header != NULL, "Failed to allocate stats header:[%d]%s/%s\n",
1635                  index, name, units);
1636
1637         header->lc_config = conf;
1638         header->lc_name   = name;
1639         header->lc_units  = units;
1640
1641         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1642         for (i = 0; i < num_cpu; ++i) {
1643                 if (stats->ls_percpu[i] == NULL)
1644                         continue;
1645                 percpu_cntr = lprocfs_stats_counter_get(stats, i, index);
1646                 percpu_cntr->lc_count           = 0;
1647                 percpu_cntr->lc_min             = LC_MIN_INIT;
1648                 percpu_cntr->lc_max             = 0;
1649                 percpu_cntr->lc_sumsquare       = 0;
1650                 percpu_cntr->lc_sum             = 0;
1651                 if ((stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1652                         percpu_cntr->lc_sum_irq = 0;
1653         }
1654         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1655 }
1656 EXPORT_SYMBOL(lprocfs_counter_init);
1657
1658 static const char * const mps_stats[] = {
1659         [LPROC_MD_CLOSE]                = "close",
1660         [LPROC_MD_CREATE]               = "create",
1661         [LPROC_MD_ENQUEUE]              = "enqueue",
1662         [LPROC_MD_GETATTR]              = "getattr",
1663         [LPROC_MD_INTENT_LOCK]          = "intent_lock",
1664         [LPROC_MD_LINK]                 = "link",
1665         [LPROC_MD_RENAME]               = "rename",
1666         [LPROC_MD_SETATTR]              = "setattr",
1667         [LPROC_MD_FSYNC]                = "fsync",
1668         [LPROC_MD_READ_PAGE]            = "read_page",
1669         [LPROC_MD_UNLINK]               = "unlink",
1670         [LPROC_MD_SETXATTR]             = "setxattr",
1671         [LPROC_MD_GETXATTR]             = "getxattr",
1672         [LPROC_MD_INTENT_GETATTR_ASYNC] = "intent_getattr_async",
1673         [LPROC_MD_REVALIDATE_LOCK]      = "revalidate_lock",
1674 };
1675
1676 int lprocfs_alloc_md_stats(struct obd_device *obd,
1677                            unsigned int num_private_stats)
1678 {
1679         struct lprocfs_stats *stats;
1680         unsigned int num_stats;
1681         int rc, i;
1682
1683         /* TODO Ensure that this function is only used where
1684          * appropriate by adding an assertion to the effect that
1685          * obd->obd_type->typ_md_ops is not NULL. We can't do this now
1686          * because mdt_procfs_init() uses this function to allocate
1687          * the stats backing /proc/fs/lustre/mdt/.../md_stats but the
1688          * mdt layer does not use the md_ops interface. This is
1689          * confusing and a waste of memory. See LU-2484.
1690          */
1691         LASSERT(obd->obd_proc_entry != NULL);
1692         LASSERT(obd->obd_md_stats == NULL);
1693
1694         num_stats = ARRAY_SIZE(mps_stats) + num_private_stats;
1695         stats = lprocfs_alloc_stats(num_stats, 0);
1696         if (!stats)
1697                 return -ENOMEM;
1698
1699         for (i = 0; i < ARRAY_SIZE(mps_stats); i++) {
1700                 lprocfs_counter_init(stats, i, 0, mps_stats[i], "reqs");
1701                 if (!stats->ls_cnt_header[i].lc_name) {
1702                         CERROR("Missing md_stat initializer md_op operation at offset %d. Aborting.\n",
1703                                i);
1704                         LBUG();
1705                 }
1706         }
1707
1708         rc = lprocfs_register_stats(obd->obd_proc_entry, "md_stats", stats);
1709         if (rc < 0) {
1710                 lprocfs_free_stats(&stats);
1711         } else {
1712                 obd->obd_md_stats = stats;
1713         }
1714
1715         return rc;
1716 }
1717 EXPORT_SYMBOL(lprocfs_alloc_md_stats);
1718
1719 void lprocfs_free_md_stats(struct obd_device *obd)
1720 {
1721         struct lprocfs_stats *stats = obd->obd_md_stats;
1722
1723         if (stats != NULL) {
1724                 obd->obd_md_stats = NULL;
1725                 lprocfs_free_stats(&stats);
1726         }
1727 }
1728 EXPORT_SYMBOL(lprocfs_free_md_stats);
1729
1730 void lprocfs_init_ldlm_stats(struct lprocfs_stats *ldlm_stats)
1731 {
1732         lprocfs_counter_init(ldlm_stats,
1733                              LDLM_ENQUEUE - LDLM_FIRST_OPC,
1734                              0, "ldlm_enqueue", "reqs");
1735         lprocfs_counter_init(ldlm_stats,
1736                              LDLM_CONVERT - LDLM_FIRST_OPC,
1737                              0, "ldlm_convert", "reqs");
1738         lprocfs_counter_init(ldlm_stats,
1739                              LDLM_CANCEL - LDLM_FIRST_OPC,
1740                              0, "ldlm_cancel", "reqs");
1741         lprocfs_counter_init(ldlm_stats,
1742                              LDLM_BL_CALLBACK - LDLM_FIRST_OPC,
1743                              0, "ldlm_bl_callback", "reqs");
1744         lprocfs_counter_init(ldlm_stats,
1745                              LDLM_CP_CALLBACK - LDLM_FIRST_OPC,
1746                              0, "ldlm_cp_callback", "reqs");
1747         lprocfs_counter_init(ldlm_stats,
1748                              LDLM_GL_CALLBACK - LDLM_FIRST_OPC,
1749                              0, "ldlm_gl_callback", "reqs");
1750 }
1751 EXPORT_SYMBOL(lprocfs_init_ldlm_stats);
1752
1753 __s64 lprocfs_read_helper(struct lprocfs_counter *lc,
1754                           struct lprocfs_counter_header *header,
1755                           enum lprocfs_stats_flags flags,
1756                           enum lprocfs_fields_flags field)
1757 {
1758         __s64 ret = 0;
1759
1760         if (lc == NULL || header == NULL)
1761                 RETURN(0);
1762
1763         switch (field) {
1764                 case LPROCFS_FIELDS_FLAGS_CONFIG:
1765                         ret = header->lc_config;
1766                         break;
1767                 case LPROCFS_FIELDS_FLAGS_SUM:
1768                         ret = lc->lc_sum;
1769                         if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1770                                 ret += lc->lc_sum_irq;
1771                         break;
1772                 case LPROCFS_FIELDS_FLAGS_MIN:
1773                         ret = lc->lc_min;
1774                         break;
1775                 case LPROCFS_FIELDS_FLAGS_MAX:
1776                         ret = lc->lc_max;
1777                         break;
1778                 case LPROCFS_FIELDS_FLAGS_AVG:
1779                         ret = (lc->lc_max - lc->lc_min) / 2;
1780                         break;
1781                 case LPROCFS_FIELDS_FLAGS_SUMSQUARE:
1782                         ret = lc->lc_sumsquare;
1783                         break;
1784                 case LPROCFS_FIELDS_FLAGS_COUNT:
1785                         ret = lc->lc_count;
1786                         break;
1787                 default:
1788                         break;
1789         };
1790         RETURN(ret);
1791 }
1792 EXPORT_SYMBOL(lprocfs_read_helper);
1793
1794 int lprocfs_read_frac_helper(char *buffer, unsigned long count, long val,
1795                              int mult)
1796 {
1797         long decimal_val, frac_val;
1798         int prtn;
1799
1800         if (count < 10)
1801                 return -EINVAL;
1802
1803         decimal_val = val / mult;
1804         prtn = snprintf(buffer, count, "%ld", decimal_val);
1805         frac_val = val % mult;
1806
1807         if (prtn < (count - 4) && frac_val > 0) {
1808                 long temp_frac;
1809                 int i, temp_mult = 1, frac_bits = 0;
1810
1811                 temp_frac = frac_val * 10;
1812                 buffer[prtn++] = '.';
1813                 while (frac_bits < 2 && (temp_frac / mult) < 1 ) {
1814                         /* only reserved 2 bits fraction */
1815                         buffer[prtn++] ='0';
1816                         temp_frac *= 10;
1817                         frac_bits++;
1818                 }
1819                 /*
1820                  * Need to think these cases :
1821                  *      1. #echo x.00 > /proc/xxx       output result : x
1822                  *      2. #echo x.0x > /proc/xxx       output result : x.0x
1823                  *      3. #echo x.x0 > /proc/xxx       output result : x.x
1824                  *      4. #echo x.xx > /proc/xxx       output result : x.xx
1825                  *      Only reserved 2 bits fraction.
1826                  */
1827                 for (i = 0; i < (5 - prtn); i++)
1828                         temp_mult *= 10;
1829
1830                 frac_bits = min((int)count - prtn, 3 - frac_bits);
1831                 prtn += snprintf(buffer + prtn, frac_bits, "%ld",
1832                                  frac_val * temp_mult / mult);
1833
1834                 prtn--;
1835                 while(buffer[prtn] < '1' || buffer[prtn] > '9') {
1836                         prtn--;
1837                         if (buffer[prtn] == '.') {
1838                                 prtn--;
1839                                 break;
1840                         }
1841                 }
1842                 prtn++;
1843         }
1844         buffer[prtn++] ='\n';
1845         return prtn;
1846 }
1847 EXPORT_SYMBOL(lprocfs_read_frac_helper);
1848
1849 int lprocfs_seq_read_frac_helper(struct seq_file *m, long val, int mult)
1850 {
1851         long decimal_val, frac_val;
1852
1853         decimal_val = val / mult;
1854         seq_printf(m, "%ld", decimal_val);
1855         frac_val = val % mult;
1856
1857         if (frac_val > 0) {
1858                 frac_val *= 100;
1859                 frac_val /= mult;
1860         }
1861         if (frac_val > 0) {
1862                 /* Three cases: x0, xx, 0x */
1863                 if ((frac_val % 10) != 0)
1864                         seq_printf(m, ".%ld", frac_val);
1865                 else
1866                         seq_printf(m, ".%ld", frac_val / 10);
1867         }
1868
1869         seq_printf(m, "\n");
1870         return 0;
1871 }
1872 EXPORT_SYMBOL(lprocfs_seq_read_frac_helper);
1873
1874 /* Obtains the conversion factor for the unit specified */
1875 static int get_mult(char unit, __u64 *mult)
1876 {
1877         __u64 units = 1;
1878
1879         switch (unit) {
1880         /* peta, tera, giga, mega, and kilo */
1881         case 'p':
1882         case 'P':
1883                 units <<= 10;
1884         case 't':
1885         case 'T':
1886                 units <<= 10;
1887         case 'g':
1888         case 'G':
1889                 units <<= 10;
1890         case 'm':
1891         case 'M':
1892                 units <<= 10;
1893         case 'k':
1894         case 'K':
1895                 units <<= 10;
1896                 break;
1897         /* some tests expect % to be accepted */
1898         case '%':
1899                 units = 1;
1900                 break;
1901         default:
1902                 return -EINVAL;
1903         }
1904
1905         *mult = units;
1906
1907         return 0;
1908 }
1909
1910 /*
1911  * Ensures the numeric string is valid. The function provides the final
1912  * multiplier in the case a unit exists at the end of the string. It also
1913  * locates the start of the whole and fractional parts (if any). This
1914  * function modifies the string so kstrtoull can be used to parse both
1915  * the whole and fraction portions. This function also figures out
1916  * the base of the number.
1917  */
1918 static int preprocess_numeric_str(char *buffer, __u64 *mult, __u64 def_mult,
1919                                   bool allow_units, char **whole, char **frac,
1920                                   unsigned int *base)
1921 {
1922         bool hit_decimal = false;
1923         bool hit_unit = false;
1924         int rc = 0;
1925         char *start;
1926         *mult = def_mult;
1927         *whole = NULL;
1928         *frac = NULL;
1929         *base = 10;
1930
1931         /* a hex string if it starts with "0x" */
1932         if (buffer[0] == '0' && tolower(buffer[1]) == 'x') {
1933                 *base = 16;
1934                 buffer += 2;
1935         }
1936
1937         start = buffer;
1938
1939         while (*buffer) {
1940                 /* allow for a single new line before the null terminator */
1941                 if (*buffer == '\n') {
1942                         *buffer = '\0';
1943                         buffer++;
1944
1945                         if (*buffer)
1946                                 return -EINVAL;
1947
1948                         break;
1949                 }
1950
1951                 /* any chars after our unit indicates a malformed string */
1952                 if (hit_unit)
1953                         return -EINVAL;
1954
1955                 /* ensure we only hit one decimal */
1956                 if (*buffer == '.') {
1957                         if (hit_decimal)
1958                                 return -EINVAL;
1959
1960                         /* if past start, there's a whole part */
1961                         if (start != buffer)
1962                                 *whole = start;
1963
1964                         *buffer = '\0';
1965                         start = buffer + 1;
1966                         hit_decimal = true;
1967                 } else if (!isdigit(*buffer) &&
1968                            !(*base == 16 && isxdigit(*buffer))) {
1969                         if (allow_units) {
1970                                 /* if we allow units, attempt to get mult */
1971                                 hit_unit = true;
1972                                 rc = get_mult(*buffer, mult);
1973                                 if (rc)
1974                                         return rc;
1975
1976                                 /* string stops here, but keep processing */
1977                                 *buffer = '\0';
1978                         } else {
1979                                 /* bad string */
1980                                 return -EINVAL;
1981                         }
1982                 }
1983
1984                 buffer++;
1985         }
1986
1987         if (hit_decimal) {
1988                 /* hit a decimal, make sure there's a fractional part */
1989                 if (!*start)
1990                         return -EINVAL;
1991
1992                 *frac = start;
1993         } else {
1994                 /* didn't hit a decimal, but may have a whole part */
1995                 if (start != buffer && *start)
1996                         *whole = start;
1997         }
1998
1999         /* malformed string if we didn't get anything */
2000         if (!*frac && !*whole)
2001                 return -EINVAL;
2002
2003         return 0;
2004 }
2005
2006 /*
2007  * Parses a numeric string which can contain a whole and fraction portion
2008  * into a __u64. Accepts a multiplier to apply to the value parsed. Also
2009  * allows the string to have a unit at the end. The function handles
2010  * wrapping of the final unsigned value.
2011  */
2012 static int str_to_u64_parse(char *buffer, unsigned long count,
2013                             __u64 *val, __u64 def_mult, bool allow_units)
2014 {
2015         __u64 whole = 0;
2016         __u64 frac = 0;
2017         unsigned int frac_d = 1;
2018         __u64 wrap_indicator = ULLONG_MAX;
2019         int rc = 0;
2020         __u64 mult;
2021         char *strwhole;
2022         char *strfrac;
2023         unsigned int base = 10;
2024
2025         rc = preprocess_numeric_str(buffer, &mult, def_mult, allow_units,
2026                                     &strwhole, &strfrac, &base);
2027
2028         if (rc)
2029                 return rc;
2030
2031         if (mult == 0) {
2032                 *val = 0;
2033                 return 0;
2034         }
2035
2036         /* the multiplier limits how large the value can be */
2037         wrap_indicator = div64_u64(wrap_indicator, mult);
2038
2039         if (strwhole) {
2040                 rc = kstrtoull(strwhole, base, &whole);
2041                 if (rc)
2042                         return rc;
2043
2044                 if (whole > wrap_indicator)
2045                         return -ERANGE;
2046
2047                 whole *= mult;
2048         }
2049
2050         if (strfrac) {
2051                 if (strlen(strfrac) > 10)
2052                         strfrac[10] = '\0';
2053
2054                 rc = kstrtoull(strfrac, base, &frac);
2055                 if (rc)
2056                         return rc;
2057
2058                 /* determine power of fractional portion */
2059                 while (*strfrac) {
2060                         frac_d *= base;
2061                         strfrac++;
2062                 }
2063
2064                 /* fractional portion is too large to perform calculation */
2065                 if (frac > wrap_indicator)
2066                         return -ERANGE;
2067
2068                 frac *= mult;
2069                 do_div(frac, frac_d);
2070         }
2071
2072         /* check that the sum of whole and fraction fits in u64 */
2073         if (whole > (ULLONG_MAX - frac))
2074                 return -ERANGE;
2075
2076         *val = whole + frac;
2077
2078         return 0;
2079 }
2080
2081 /*
2082  * This function parses numeric/hex strings into __s64. It accepts a multiplier
2083  * which will apply to the value parsed. It also can allow the string to
2084  * have a unit as the last character. The function handles overflow/underflow
2085  * of the signed integer.
2086  */
2087 static int str_to_s64_internal(const char __user *buffer, unsigned long count,
2088                                __s64 *val, __u64 def_mult, bool allow_units)
2089 {
2090         char kernbuf[22];
2091         __u64 tmp;
2092         unsigned int offset = 0;
2093         int signed sign = 1;
2094         __u64 max = LLONG_MAX;
2095         int rc = 0;
2096
2097         if (count > (sizeof(kernbuf) - 1))
2098                 return -EINVAL;
2099
2100         if (copy_from_user(kernbuf, buffer, count))
2101                 return -EFAULT;
2102
2103         kernbuf[count] = '\0';
2104
2105         /* keep track of our sign */
2106         if (*kernbuf == '-') {
2107                 sign = -1;
2108                 offset++;
2109                 /* equivalent to max = -LLONG_MIN, avoids overflow */
2110                 max++;
2111         }
2112
2113         rc = str_to_u64_parse(kernbuf + offset, count - offset,
2114                               &tmp, def_mult, allow_units);
2115         if (rc)
2116                 return rc;
2117
2118         /* check for overflow/underflow */
2119         if (max < tmp)
2120                 return -ERANGE;
2121
2122         *val = (__s64)tmp * sign;
2123
2124         return 0;
2125 }
2126
2127 /**
2128  * Convert a user string into a signed 64 bit number. This function produces
2129  * an error when the value parsed from the string times multiplier underflows or
2130  * overflows. This function only accepts strings that contains digits, an
2131  * optional decimal, and a char representing a unit at the end. If a unit is
2132  * specified in the string, the multiplier provided by the caller is ignored.
2133  * This function can also accept hexadecimal strings which are prefixed with
2134  * "0x".
2135  *
2136  * \param[in] buffer    string consisting of numbers, a decimal, and a unit
2137  * \param[in] count     buffer length
2138  * \param[in] val       if successful, the value represented by the string
2139  * \param[in] defunit   default unit if string doesn't contain one
2140  *
2141  * \retval              0 on success
2142  * \retval              negative number on error
2143  */
2144 int lprocfs_str_with_units_to_s64(const char __user *buffer,
2145                                   unsigned long count, __s64 *val, char defunit)
2146 {
2147         __u64 mult = 1;
2148         int rc;
2149
2150         if (defunit != '1') {
2151                 rc = get_mult(defunit, &mult);
2152                 if (rc)
2153                         return rc;
2154         }
2155
2156         return str_to_s64_internal(buffer, count, val, mult, true);
2157 }
2158 EXPORT_SYMBOL(lprocfs_str_with_units_to_s64);
2159
2160 char *lprocfs_strnstr(const char *s1, const char *s2, size_t len)
2161 {
2162         size_t l2;
2163
2164         l2 = strlen(s2);
2165         if (!l2)
2166                 return (char *)s1;
2167         while (len >= l2) {
2168                 len--;
2169                 if (!memcmp(s1, s2, l2))
2170                         return (char *)s1;
2171                 s1++;
2172         }
2173         return NULL;
2174 }
2175 EXPORT_SYMBOL(lprocfs_strnstr);
2176
2177 /**
2178  * Find the string \a name in the input \a buffer, and return a pointer to the
2179  * value immediately following \a name, reducing \a count appropriately.
2180  * If \a name is not found the original \a buffer is returned.
2181  */
2182 char *lprocfs_find_named_value(const char *buffer, const char *name,
2183                                 size_t *count)
2184 {
2185         char *val;
2186         size_t buflen = *count;
2187
2188         /* there is no strnstr() in rhel5 and ubuntu kernels */
2189         val = lprocfs_strnstr(buffer, name, buflen);
2190         if (val == NULL)
2191                 return (char *)buffer;
2192
2193         val += strlen(name);                             /* skip prefix */
2194         while (val < buffer + buflen && isspace(*val)) /* skip separator */
2195                 val++;
2196
2197         *count = 0;
2198         while (val < buffer + buflen && isalnum(*val)) {
2199                 ++*count;
2200                 ++val;
2201         }
2202
2203         return val - *count;
2204 }
2205 EXPORT_SYMBOL(lprocfs_find_named_value);
2206
2207 int ldebugfs_seq_create(struct dentry *parent, const char *name, umode_t mode,
2208                         const struct file_operations *seq_fops, void *data)
2209 {
2210         struct dentry *entry;
2211
2212         /* Disallow secretly (un)writable entries. */
2213         LASSERT((!seq_fops->write) == (!(mode & 0222)));
2214
2215         entry = debugfs_create_file(name, mode, parent, data, seq_fops);
2216         if (IS_ERR_OR_NULL(entry))
2217                 return entry ? PTR_ERR(entry) : -ENOMEM;
2218
2219         return 0;
2220 }
2221 EXPORT_SYMBOL_GPL(ldebugfs_seq_create);
2222
2223 int lprocfs_seq_create(struct proc_dir_entry *parent,
2224                        const char *name,
2225                        mode_t mode,
2226                        const struct file_operations *seq_fops,
2227                        void *data)
2228 {
2229         struct proc_dir_entry *entry;
2230         ENTRY;
2231
2232         /* Disallow secretly (un)writable entries. */
2233         LASSERT((seq_fops->write == NULL) == ((mode & 0222) == 0));
2234
2235         entry = proc_create_data(name, mode, parent, seq_fops, data);
2236
2237         if (entry == NULL)
2238                 RETURN(-ENOMEM);
2239
2240         RETURN(0);
2241 }
2242 EXPORT_SYMBOL(lprocfs_seq_create);
2243
2244 int lprocfs_obd_seq_create(struct obd_device *dev,
2245                            const char *name,
2246                            mode_t mode,
2247                            const struct file_operations *seq_fops,
2248                            void *data)
2249 {
2250         return (lprocfs_seq_create(dev->obd_proc_entry, name,
2251                                    mode, seq_fops, data));
2252 }
2253 EXPORT_SYMBOL(lprocfs_obd_seq_create);
2254
2255 void lprocfs_oh_tally(struct obd_histogram *oh, unsigned int value)
2256 {
2257         if (value >= OBD_HIST_MAX)
2258                 value = OBD_HIST_MAX - 1;
2259
2260         spin_lock(&oh->oh_lock);
2261         oh->oh_buckets[value]++;
2262         spin_unlock(&oh->oh_lock);
2263 }
2264 EXPORT_SYMBOL(lprocfs_oh_tally);
2265
2266 void lprocfs_oh_tally_log2(struct obd_histogram *oh, unsigned int value)
2267 {
2268         unsigned int val = 0;
2269
2270         if (likely(value != 0))
2271                 val = min(fls(value - 1), OBD_HIST_MAX);
2272
2273         lprocfs_oh_tally(oh, val);
2274 }
2275 EXPORT_SYMBOL(lprocfs_oh_tally_log2);
2276
2277 unsigned long lprocfs_oh_sum(struct obd_histogram *oh)
2278 {
2279         unsigned long ret = 0;
2280         int i;
2281
2282         for (i = 0; i < OBD_HIST_MAX; i++)
2283                 ret +=  oh->oh_buckets[i];
2284         return ret;
2285 }
2286 EXPORT_SYMBOL(lprocfs_oh_sum);
2287
2288 void lprocfs_oh_clear(struct obd_histogram *oh)
2289 {
2290         spin_lock(&oh->oh_lock);
2291         memset(oh->oh_buckets, 0, sizeof(oh->oh_buckets));
2292         spin_unlock(&oh->oh_lock);
2293 }
2294 EXPORT_SYMBOL(lprocfs_oh_clear);
2295
2296 ssize_t lustre_attr_show(struct kobject *kobj,
2297                          struct attribute *attr, char *buf)
2298 {
2299         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
2300
2301         return a->show ? a->show(kobj, attr, buf) : 0;
2302 }
2303 EXPORT_SYMBOL_GPL(lustre_attr_show);
2304
2305 ssize_t lustre_attr_store(struct kobject *kobj, struct attribute *attr,
2306                           const char *buf, size_t len)
2307 {
2308         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
2309
2310         return a->store ? a->store(kobj, attr, buf, len) : len;
2311 }
2312 EXPORT_SYMBOL_GPL(lustre_attr_store);
2313
2314 const struct sysfs_ops lustre_sysfs_ops = {
2315         .show  = lustre_attr_show,
2316         .store = lustre_attr_store,
2317 };
2318 EXPORT_SYMBOL_GPL(lustre_sysfs_ops);
2319
2320 int lprocfs_obd_max_pages_per_rpc_seq_show(struct seq_file *m, void *data)
2321 {
2322         struct obd_device *dev = data;
2323         struct client_obd *cli = &dev->u.cli;
2324
2325         spin_lock(&cli->cl_loi_list_lock);
2326         seq_printf(m, "%d\n", cli->cl_max_pages_per_rpc);
2327         spin_unlock(&cli->cl_loi_list_lock);
2328         return 0;
2329 }
2330 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_show);
2331
2332 ssize_t lprocfs_obd_max_pages_per_rpc_seq_write(struct file *file,
2333                                                 const char __user *buffer,
2334                                                 size_t count, loff_t *off)
2335 {
2336         struct obd_device *dev =
2337                 ((struct seq_file *)file->private_data)->private;
2338         struct client_obd *cli = &dev->u.cli;
2339         struct obd_connect_data *ocd = &cli->cl_import->imp_connect_data;
2340         int chunk_mask, rc;
2341         s64 val;
2342
2343         rc = lprocfs_str_with_units_to_s64(buffer, count, &val, '1');
2344         if (rc)
2345                 return rc;
2346         if (val < 0)
2347                 return -ERANGE;
2348
2349         /* if the max_pages is specified in bytes, convert to pages */
2350         if (val >= ONE_MB_BRW_SIZE)
2351                 val >>= PAGE_SHIFT;
2352
2353         LPROCFS_CLIMP_CHECK(dev);
2354
2355         chunk_mask = ~((1 << (cli->cl_chunkbits - PAGE_SHIFT)) - 1);
2356         /* max_pages_per_rpc must be chunk aligned */
2357         val = (val + ~chunk_mask) & chunk_mask;
2358         if (val == 0 || (ocd->ocd_brw_size != 0 &&
2359                          val > ocd->ocd_brw_size >> PAGE_SHIFT)) {
2360                 LPROCFS_CLIMP_EXIT(dev);
2361                 return -ERANGE;
2362         }
2363         spin_lock(&cli->cl_loi_list_lock);
2364         cli->cl_max_pages_per_rpc = val;
2365         client_adjust_max_dirty(cli);
2366         spin_unlock(&cli->cl_loi_list_lock);
2367
2368         LPROCFS_CLIMP_EXIT(dev);
2369         return count;
2370 }
2371 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_write);
2372
2373 ssize_t short_io_bytes_show(struct kobject *kobj, struct attribute *attr,
2374                             char *buf)
2375 {
2376         struct obd_device *dev = container_of(kobj, struct obd_device,
2377                                               obd_kset.kobj);
2378         struct client_obd *cli = &dev->u.cli;
2379         int rc;
2380
2381         spin_lock(&cli->cl_loi_list_lock);
2382         rc = sprintf(buf, "%d\n", cli->cl_short_io_bytes);
2383         spin_unlock(&cli->cl_loi_list_lock);
2384         return rc;
2385 }
2386 EXPORT_SYMBOL(short_io_bytes_show);
2387
2388 /* Used to catch people who think they're specifying pages. */
2389 #define MIN_SHORT_IO_BYTES 64
2390
2391 ssize_t short_io_bytes_store(struct kobject *kobj, struct attribute *attr,
2392                              const char *buffer, size_t count)
2393 {
2394         struct obd_device *dev = container_of(kobj, struct obd_device,
2395                                               obd_kset.kobj);
2396         struct client_obd *cli = &dev->u.cli;
2397         u32 val;
2398         int rc;
2399
2400         LPROCFS_CLIMP_CHECK(dev);
2401
2402         rc = kstrtouint(buffer, 0, &val);
2403         if (rc)
2404                 GOTO(out, rc);
2405
2406         if (val < MIN_SHORT_IO_BYTES || val > OBD_MAX_SHORT_IO_BYTES)
2407                 GOTO(out, rc = -ERANGE);
2408
2409         rc = count;
2410
2411         spin_lock(&cli->cl_loi_list_lock);
2412         if (val > (cli->cl_max_pages_per_rpc << PAGE_SHIFT))
2413                 rc = -ERANGE;
2414         else
2415                 cli->cl_short_io_bytes = val;
2416         spin_unlock(&cli->cl_loi_list_lock);
2417
2418 out:
2419         LPROCFS_CLIMP_EXIT(dev);
2420         return rc;
2421 }
2422 EXPORT_SYMBOL(short_io_bytes_store);
2423
2424 int lprocfs_wr_root_squash(const char __user *buffer, unsigned long count,
2425                            struct root_squash_info *squash, char *name)
2426 {
2427         int rc;
2428         char kernbuf[64], *tmp, *errmsg;
2429         unsigned long uid, gid;
2430         ENTRY;
2431
2432         if (count >= sizeof(kernbuf)) {
2433                 errmsg = "string too long";
2434                 GOTO(failed_noprint, rc = -EINVAL);
2435         }
2436         if (copy_from_user(kernbuf, buffer, count)) {
2437                 errmsg = "bad address";
2438                 GOTO(failed_noprint, rc = -EFAULT);
2439         }
2440         kernbuf[count] = '\0';
2441
2442         /* look for uid gid separator */
2443         tmp = strchr(kernbuf, ':');
2444         if (tmp == NULL) {
2445                 errmsg = "needs uid:gid format";
2446                 GOTO(failed, rc = -EINVAL);
2447         }
2448         *tmp = '\0';
2449         tmp++;
2450
2451         /* parse uid */
2452         if (kstrtoul(kernbuf, 0, &uid) != 0) {
2453                 errmsg = "bad uid";
2454                 GOTO(failed, rc = -EINVAL);
2455         }
2456
2457         /* parse gid */
2458         if (kstrtoul(tmp, 0, &gid) != 0) {
2459                 errmsg = "bad gid";
2460                 GOTO(failed, rc = -EINVAL);
2461         }
2462
2463         squash->rsi_uid = uid;
2464         squash->rsi_gid = gid;
2465
2466         LCONSOLE_INFO("%s: root_squash is set to %u:%u\n",
2467                       name, squash->rsi_uid, squash->rsi_gid);
2468         RETURN(count);
2469
2470 failed:
2471         if (tmp != NULL) {
2472                 tmp--;
2473                 *tmp = ':';
2474         }
2475         CWARN("%s: failed to set root_squash to \"%s\", %s, rc = %d\n",
2476               name, kernbuf, errmsg, rc);
2477         RETURN(rc);
2478 failed_noprint:
2479         CWARN("%s: failed to set root_squash due to %s, rc = %d\n",
2480               name, errmsg, rc);
2481         RETURN(rc);
2482 }
2483 EXPORT_SYMBOL(lprocfs_wr_root_squash);
2484
2485
2486 int lprocfs_wr_nosquash_nids(const char __user *buffer, unsigned long count,
2487                              struct root_squash_info *squash, char *name)
2488 {
2489         int rc;
2490         char *kernbuf = NULL;
2491         char *errmsg;
2492         struct list_head tmp;
2493         int len = count;
2494         ENTRY;
2495
2496         if (count > 4096) {
2497                 errmsg = "string too long";
2498                 GOTO(failed, rc = -EINVAL);
2499         }
2500
2501         OBD_ALLOC(kernbuf, count + 1);
2502         if (kernbuf == NULL) {
2503                 errmsg = "no memory";
2504                 GOTO(failed, rc = -ENOMEM);
2505         }
2506         if (copy_from_user(kernbuf, buffer, count)) {
2507                 errmsg = "bad address";
2508                 GOTO(failed, rc = -EFAULT);
2509         }
2510         kernbuf[count] = '\0';
2511
2512         if (count > 0 && kernbuf[count - 1] == '\n')
2513                 len = count - 1;
2514
2515         if ((len == 4 && strncmp(kernbuf, "NONE", len) == 0) ||
2516             (len == 5 && strncmp(kernbuf, "clear", len) == 0)) {
2517                 /* empty string is special case */
2518                 down_write(&squash->rsi_sem);
2519                 if (!list_empty(&squash->rsi_nosquash_nids))
2520                         cfs_free_nidlist(&squash->rsi_nosquash_nids);
2521                 up_write(&squash->rsi_sem);
2522                 LCONSOLE_INFO("%s: nosquash_nids is cleared\n", name);
2523                 OBD_FREE(kernbuf, count + 1);
2524                 RETURN(count);
2525         }
2526
2527         INIT_LIST_HEAD(&tmp);
2528         if (cfs_parse_nidlist(kernbuf, count, &tmp) <= 0) {
2529                 errmsg = "can't parse";
2530                 GOTO(failed, rc = -EINVAL);
2531         }
2532         LCONSOLE_INFO("%s: nosquash_nids set to %s\n",
2533                       name, kernbuf);
2534         OBD_FREE(kernbuf, count + 1);
2535         kernbuf = NULL;
2536
2537         down_write(&squash->rsi_sem);
2538         if (!list_empty(&squash->rsi_nosquash_nids))
2539                 cfs_free_nidlist(&squash->rsi_nosquash_nids);
2540         list_splice(&tmp, &squash->rsi_nosquash_nids);
2541         up_write(&squash->rsi_sem);
2542
2543         RETURN(count);
2544
2545 failed:
2546         if (kernbuf) {
2547                 CWARN("%s: failed to set nosquash_nids to \"%s\", %s rc = %d\n",
2548                       name, kernbuf, errmsg, rc);
2549                 OBD_FREE(kernbuf, count + 1);
2550         } else {
2551                 CWARN("%s: failed to set nosquash_nids due to %s rc = %d\n",
2552                       name, errmsg, rc);
2553         }
2554         RETURN(rc);
2555 }
2556 EXPORT_SYMBOL(lprocfs_wr_nosquash_nids);
2557
2558 #endif /* CONFIG_PROC_FS*/