Whamcloud - gitweb
LU-1757 osc: clarify short_io_bytes is maximum value
[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                    "       idle: %lld sec\n",
1001                    nidstr,
1002                    imp->imp_conn_cnt,
1003                    imp->imp_generation,
1004                    atomic_read(&imp->imp_inval_count),
1005                    ktime_get_real_seconds() - imp->imp_last_reply_time);
1006         spin_unlock(&imp->imp_lock);
1007
1008         if (obd->obd_svc_stats == NULL)
1009                 goto out_climp;
1010
1011         header = &obd->obd_svc_stats->ls_cnt_header[PTLRPC_REQWAIT_CNTR];
1012         lprocfs_stats_collect(obd->obd_svc_stats, PTLRPC_REQWAIT_CNTR, &ret);
1013         if (ret.lc_count != 0) {
1014                 /* first argument to do_div MUST be __u64 */
1015                 __u64 sum = ret.lc_sum;
1016                 do_div(sum, ret.lc_count);
1017                 ret.lc_sum = sum;
1018         } else
1019                 ret.lc_sum = 0;
1020         seq_printf(m, "    rpcs:\n"
1021                    "       inflight: %u\n"
1022                    "       unregistering: %u\n"
1023                    "       timeouts: %u\n"
1024                    "       avg_waittime: %llu %s\n",
1025                    atomic_read(&imp->imp_inflight),
1026                    atomic_read(&imp->imp_unregistering),
1027                    atomic_read(&imp->imp_timeouts),
1028                    ret.lc_sum, header->lc_units);
1029
1030         k = 0;
1031         for(j = 0; j < IMP_AT_MAX_PORTALS; j++) {
1032                 if (imp->imp_at.iat_portal[j] == 0)
1033                         break;
1034                 k = max_t(unsigned int, k,
1035                           at_get(&imp->imp_at.iat_service_estimate[j]));
1036         }
1037         seq_printf(m, "    service_estimates:\n"
1038                    "       services: %u sec\n"
1039                    "       network: %u sec\n",
1040                    k,
1041                    at_get(&imp->imp_at.iat_net_latency));
1042
1043         seq_printf(m, "    transactions:\n"
1044                    "       last_replay: %llu\n"
1045                    "       peer_committed: %llu\n"
1046                    "       last_checked: %llu\n",
1047                    imp->imp_last_replay_transno,
1048                    imp->imp_peer_committed_transno,
1049                    imp->imp_last_transno_checked);
1050
1051         /* avg data rates */
1052         for (rw = 0; rw <= 1; rw++) {
1053                 lprocfs_stats_collect(obd->obd_svc_stats,
1054                                       PTLRPC_LAST_CNTR + BRW_READ_BYTES + rw,
1055                                       &ret);
1056                 if (ret.lc_sum > 0 && ret.lc_count > 0) {
1057                         /* first argument to do_div MUST be __u64 */
1058                         __u64 sum = ret.lc_sum;
1059                         do_div(sum, ret.lc_count);
1060                         ret.lc_sum = sum;
1061                         seq_printf(m, "    %s_data_averages:\n"
1062                                    "       bytes_per_rpc: %llu\n",
1063                                    rw ? "write" : "read",
1064                                    ret.lc_sum);
1065                 }
1066                 k = (int)ret.lc_sum;
1067                 j = opcode_offset(OST_READ + rw) + EXTRA_MAX_OPCODES;
1068                 header = &obd->obd_svc_stats->ls_cnt_header[j];
1069                 lprocfs_stats_collect(obd->obd_svc_stats, j, &ret);
1070                 if (ret.lc_sum > 0 && ret.lc_count != 0) {
1071                         /* first argument to do_div MUST be __u64 */
1072                         __u64 sum = ret.lc_sum;
1073                         do_div(sum, ret.lc_count);
1074                         ret.lc_sum = sum;
1075                         seq_printf(m, "       %s_per_rpc: %llu\n",
1076                                    header->lc_units, ret.lc_sum);
1077                         j = (int)ret.lc_sum;
1078                         if (j > 0)
1079                                 seq_printf(m, "       MB_per_sec: %u.%.02u\n",
1080                                            k / j, (100 * k / j) % 100);
1081                 }
1082         }
1083
1084 out_climp:
1085         LPROCFS_CLIMP_EXIT(obd);
1086         return 0;
1087 }
1088 EXPORT_SYMBOL(lprocfs_import_seq_show);
1089
1090 int lprocfs_state_seq_show(struct seq_file *m, void *data)
1091 {
1092         struct obd_device *obd = (struct obd_device *)data;
1093         struct obd_import *imp;
1094         int j, k;
1095
1096         LASSERT(obd != NULL);
1097         LPROCFS_CLIMP_CHECK(obd);
1098         imp = obd->u.cli.cl_import;
1099
1100         seq_printf(m, "current_state: %s\n",
1101                    ptlrpc_import_state_name(imp->imp_state));
1102         seq_printf(m, "state_history:\n");
1103         k = imp->imp_state_hist_idx;
1104         for (j = 0; j < IMP_STATE_HIST_LEN; j++) {
1105                 struct import_state_hist *ish =
1106                         &imp->imp_state_hist[(k + j) % IMP_STATE_HIST_LEN];
1107                 if (ish->ish_state == 0)
1108                         continue;
1109                 seq_printf(m, " - [ %lld, %s ]\n", (s64)ish->ish_time,
1110                            ptlrpc_import_state_name(ish->ish_state));
1111         }
1112
1113         LPROCFS_CLIMP_EXIT(obd);
1114         return 0;
1115 }
1116 EXPORT_SYMBOL(lprocfs_state_seq_show);
1117
1118 int lprocfs_at_hist_helper(struct seq_file *m, struct adaptive_timeout *at)
1119 {
1120         int i;
1121         for (i = 0; i < AT_BINS; i++)
1122                 seq_printf(m, "%3u ", at->at_hist[i]);
1123         seq_printf(m, "\n");
1124         return 0;
1125 }
1126 EXPORT_SYMBOL(lprocfs_at_hist_helper);
1127
1128 /* See also ptlrpc_lprocfs_timeouts_show_seq */
1129 int lprocfs_timeouts_seq_show(struct seq_file *m, void *data)
1130 {
1131         struct obd_device *obd = (struct obd_device *)data;
1132         struct obd_import *imp;
1133         unsigned int cur, worst;
1134         time64_t now, worstt;
1135         int i;
1136
1137         LASSERT(obd != NULL);
1138         LPROCFS_CLIMP_CHECK(obd);
1139         imp = obd->u.cli.cl_import;
1140
1141         now = ktime_get_real_seconds();
1142
1143         /* Some network health info for kicks */
1144         seq_printf(m, "%-10s : %lld, %llds ago\n",
1145                    "last reply", (s64)imp->imp_last_reply_time,
1146                    (s64)(now - imp->imp_last_reply_time));
1147
1148         cur = at_get(&imp->imp_at.iat_net_latency);
1149         worst = imp->imp_at.iat_net_latency.at_worst_ever;
1150         worstt = imp->imp_at.iat_net_latency.at_worst_time;
1151         seq_printf(m, "%-10s : cur %3u  worst %3u (at %lld, %llds ago) ",
1152                    "network", cur, worst, (s64)worstt, (s64)(now - worstt));
1153         lprocfs_at_hist_helper(m, &imp->imp_at.iat_net_latency);
1154
1155         for(i = 0; i < IMP_AT_MAX_PORTALS; i++) {
1156                 if (imp->imp_at.iat_portal[i] == 0)
1157                         break;
1158                 cur = at_get(&imp->imp_at.iat_service_estimate[i]);
1159                 worst = imp->imp_at.iat_service_estimate[i].at_worst_ever;
1160                 worstt = imp->imp_at.iat_service_estimate[i].at_worst_time;
1161                 seq_printf(m, "portal %-2d  : cur %3u  worst %3u (at %lld, %llds ago) ",
1162                            imp->imp_at.iat_portal[i], cur, worst, (s64)worstt,
1163                            (s64)(now - worstt));
1164                 lprocfs_at_hist_helper(m, &imp->imp_at.iat_service_estimate[i]);
1165         }
1166
1167         LPROCFS_CLIMP_EXIT(obd);
1168         return 0;
1169 }
1170 EXPORT_SYMBOL(lprocfs_timeouts_seq_show);
1171
1172 int lprocfs_connect_flags_seq_show(struct seq_file *m, void *data)
1173 {
1174         struct obd_device *obd = data;
1175         __u64 flags;
1176         __u64 flags2;
1177
1178         LPROCFS_CLIMP_CHECK(obd);
1179         flags = obd->u.cli.cl_import->imp_connect_data.ocd_connect_flags;
1180         flags2 = obd->u.cli.cl_import->imp_connect_data.ocd_connect_flags2;
1181         seq_printf(m, "flags=%#llx\n", flags);
1182         seq_printf(m, "flags2=%#llx\n", flags2);
1183         obd_connect_seq_flags2str(m, flags, flags2, "\n");
1184         seq_printf(m, "\n");
1185         LPROCFS_CLIMP_EXIT(obd);
1186         return 0;
1187 }
1188 EXPORT_SYMBOL(lprocfs_connect_flags_seq_show);
1189
1190 static const struct attribute *obd_def_uuid_attrs[] = {
1191         &lustre_attr_uuid.attr,
1192         NULL,
1193 };
1194
1195 static const struct attribute *obd_def_attrs[] = {
1196         &lustre_attr_blocksize.attr,
1197         &lustre_attr_kbytestotal.attr,
1198         &lustre_attr_kbytesfree.attr,
1199         &lustre_attr_kbytesavail.attr,
1200         &lustre_attr_filestotal.attr,
1201         &lustre_attr_filesfree.attr,
1202         &lustre_attr_uuid.attr,
1203         NULL,
1204 };
1205
1206 static void obd_sysfs_release(struct kobject *kobj)
1207 {
1208         struct obd_device *obd = container_of(kobj, struct obd_device,
1209                                               obd_kset.kobj);
1210
1211         complete(&obd->obd_kobj_unregister);
1212 }
1213
1214 int lprocfs_obd_setup(struct obd_device *obd, bool uuid_only)
1215 {
1216         struct lprocfs_vars *debugfs_vars = NULL;
1217         int rc;
1218
1219         if (!obd || obd->obd_magic != OBD_DEVICE_MAGIC)
1220                 return -ENODEV;
1221
1222         rc = kobject_set_name(&obd->obd_kset.kobj, "%s", obd->obd_name);
1223         if (rc)
1224                 return rc;
1225
1226         obd->obd_ktype.sysfs_ops = &lustre_sysfs_ops;
1227         obd->obd_ktype.release = obd_sysfs_release;
1228
1229         obd->obd_kset.kobj.parent = obd->obd_type->typ_kobj;
1230         obd->obd_kset.kobj.ktype = &obd->obd_ktype;
1231         init_completion(&obd->obd_kobj_unregister);
1232         rc = kset_register(&obd->obd_kset);
1233         if (rc)
1234                 return rc;
1235
1236         if (uuid_only)
1237                 obd->obd_attrs = obd_def_uuid_attrs;
1238         else
1239                 obd->obd_attrs = obd_def_attrs;
1240
1241         rc = sysfs_create_files(&obd->obd_kset.kobj, obd->obd_attrs);
1242         if (rc) {
1243                 kset_unregister(&obd->obd_kset);
1244                 return rc;
1245         }
1246
1247         if (!obd->obd_type->typ_procroot)
1248                 debugfs_vars = obd->obd_vars;
1249         obd->obd_debugfs_entry = ldebugfs_register(obd->obd_name,
1250                                                    obd->obd_type->typ_debugfs_entry,
1251                                                    debugfs_vars, obd);
1252         if (IS_ERR_OR_NULL(obd->obd_debugfs_entry)) {
1253                 rc = obd->obd_debugfs_entry ? PTR_ERR(obd->obd_debugfs_entry)
1254                                             : -ENOMEM;
1255                 CERROR("error %d setting up debugfs for %s\n",
1256                        rc, obd->obd_name);
1257                 obd->obd_debugfs_entry = NULL;
1258
1259                 sysfs_remove_files(&obd->obd_kset.kobj, obd->obd_attrs);
1260                 obd->obd_attrs = NULL;
1261                 kset_unregister(&obd->obd_kset);
1262                 return rc;
1263         }
1264
1265         if (obd->obd_proc_entry || !obd->obd_type->typ_procroot)
1266                 GOTO(already_registered, rc);
1267
1268         obd->obd_proc_entry = lprocfs_register(obd->obd_name,
1269                                                obd->obd_type->typ_procroot,
1270                                                obd->obd_vars, obd);
1271         if (IS_ERR(obd->obd_proc_entry)) {
1272                 rc = PTR_ERR(obd->obd_proc_entry);
1273                 CERROR("error %d setting up lprocfs for %s\n",rc,obd->obd_name);
1274                 obd->obd_proc_entry = NULL;
1275
1276                 ldebugfs_remove(&obd->obd_debugfs_entry);
1277                 sysfs_remove_files(&obd->obd_kset.kobj, obd->obd_attrs);
1278                 obd->obd_attrs = NULL;
1279                 kset_unregister(&obd->obd_kset);
1280                 return rc;
1281         }
1282 already_registered:
1283         return rc;
1284 }
1285 EXPORT_SYMBOL(lprocfs_obd_setup);
1286
1287 int lprocfs_obd_cleanup(struct obd_device *obd)
1288 {
1289         if (!obd)
1290                 return -EINVAL;
1291
1292         if (obd->obd_proc_exports_entry) {
1293                 /* Should be no exports left */
1294                 lprocfs_remove(&obd->obd_proc_exports_entry);
1295                 obd->obd_proc_exports_entry = NULL;
1296         }
1297
1298         if (obd->obd_proc_entry) {
1299                 lprocfs_remove(&obd->obd_proc_entry);
1300                 obd->obd_proc_entry = NULL;
1301         }
1302
1303         if (!IS_ERR_OR_NULL(obd->obd_debugfs_entry))
1304                 ldebugfs_remove(&obd->obd_debugfs_entry);
1305
1306         /* obd device never allocated a kset */
1307         if (!obd->obd_kset.kobj.state_initialized)
1308                 return 0;
1309
1310         if (obd->obd_attrs) {
1311                 sysfs_remove_files(&obd->obd_kset.kobj, obd->obd_attrs);
1312                 obd->obd_attrs = NULL;
1313         }
1314
1315         kset_unregister(&obd->obd_kset);
1316         wait_for_completion(&obd->obd_kobj_unregister);
1317         return 0;
1318 }
1319 EXPORT_SYMBOL(lprocfs_obd_cleanup);
1320
1321 int lprocfs_stats_alloc_one(struct lprocfs_stats *stats, unsigned int cpuid)
1322 {
1323         struct lprocfs_counter  *cntr;
1324         unsigned int            percpusize;
1325         int                     rc = -ENOMEM;
1326         unsigned long           flags = 0;
1327         int                     i;
1328
1329         LASSERT(stats->ls_percpu[cpuid] == NULL);
1330         LASSERT((stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) == 0);
1331
1332         percpusize = lprocfs_stats_counter_size(stats);
1333         LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[cpuid], percpusize);
1334         if (stats->ls_percpu[cpuid] != NULL) {
1335                 rc = 0;
1336                 if (unlikely(stats->ls_biggest_alloc_num <= cpuid)) {
1337                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1338                                 spin_lock_irqsave(&stats->ls_lock, flags);
1339                         else
1340                                 spin_lock(&stats->ls_lock);
1341                         if (stats->ls_biggest_alloc_num <= cpuid)
1342                                 stats->ls_biggest_alloc_num = cpuid + 1;
1343                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) {
1344                                 spin_unlock_irqrestore(&stats->ls_lock, flags);
1345                         } else {
1346                                 spin_unlock(&stats->ls_lock);
1347                         }
1348                 }
1349                 /* initialize the ls_percpu[cpuid] non-zero counter */
1350                 for (i = 0; i < stats->ls_num; ++i) {
1351                         cntr = lprocfs_stats_counter_get(stats, cpuid, i);
1352                         cntr->lc_min = LC_MIN_INIT;
1353                 }
1354         }
1355         return rc;
1356 }
1357
1358 struct lprocfs_stats *lprocfs_alloc_stats(unsigned int num,
1359                                           enum lprocfs_stats_flags flags)
1360 {
1361         struct lprocfs_stats    *stats;
1362         unsigned int            num_entry;
1363         unsigned int            percpusize = 0;
1364         int                     i;
1365
1366         if (num == 0)
1367                 return NULL;
1368
1369         if (lprocfs_no_percpu_stats != 0)
1370                 flags |= LPROCFS_STATS_FLAG_NOPERCPU;
1371
1372         if (flags & LPROCFS_STATS_FLAG_NOPERCPU)
1373                 num_entry = 1;
1374         else
1375                 num_entry = num_possible_cpus();
1376
1377         /* alloc percpu pointers for all possible cpu slots */
1378         LIBCFS_ALLOC(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1379         if (stats == NULL)
1380                 return NULL;
1381
1382         stats->ls_num = num;
1383         stats->ls_flags = flags;
1384         spin_lock_init(&stats->ls_lock);
1385
1386         /* alloc num of counter headers */
1387         LIBCFS_ALLOC(stats->ls_cnt_header,
1388                      stats->ls_num * sizeof(struct lprocfs_counter_header));
1389         if (stats->ls_cnt_header == NULL)
1390                 goto fail;
1391
1392         if ((flags & LPROCFS_STATS_FLAG_NOPERCPU) != 0) {
1393                 /* contains only one set counters */
1394                 percpusize = lprocfs_stats_counter_size(stats);
1395                 LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[0], percpusize);
1396                 if (stats->ls_percpu[0] == NULL)
1397                         goto fail;
1398                 stats->ls_biggest_alloc_num = 1;
1399         } else if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0) {
1400                 /* alloc all percpu data, currently only obd_memory use this */
1401                 for (i = 0; i < num_entry; ++i)
1402                         if (lprocfs_stats_alloc_one(stats, i) < 0)
1403                                 goto fail;
1404         }
1405
1406         return stats;
1407
1408 fail:
1409         lprocfs_free_stats(&stats);
1410         return NULL;
1411 }
1412 EXPORT_SYMBOL(lprocfs_alloc_stats);
1413
1414 void lprocfs_free_stats(struct lprocfs_stats **statsh)
1415 {
1416         struct lprocfs_stats *stats = *statsh;
1417         unsigned int num_entry;
1418         unsigned int percpusize;
1419         unsigned int i;
1420
1421         if (stats == NULL || stats->ls_num == 0)
1422                 return;
1423         *statsh = NULL;
1424
1425         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU)
1426                 num_entry = 1;
1427         else
1428                 num_entry = num_possible_cpus();
1429
1430         percpusize = lprocfs_stats_counter_size(stats);
1431         for (i = 0; i < num_entry; i++)
1432                 if (stats->ls_percpu[i] != NULL)
1433                         LIBCFS_FREE(stats->ls_percpu[i], percpusize);
1434         if (stats->ls_cnt_header != NULL)
1435                 LIBCFS_FREE(stats->ls_cnt_header, stats->ls_num *
1436                                         sizeof(struct lprocfs_counter_header));
1437         LIBCFS_FREE(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1438 }
1439 EXPORT_SYMBOL(lprocfs_free_stats);
1440
1441 u64 lprocfs_stats_collector(struct lprocfs_stats *stats, int idx,
1442                             enum lprocfs_fields_flags field)
1443 {
1444         unsigned long flags = 0;
1445         unsigned int num_cpu;
1446         unsigned int i;
1447         u64 ret = 0;
1448
1449         LASSERT(stats);
1450
1451         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1452         for (i = 0; i < num_cpu; i++) {
1453                 struct lprocfs_counter *cntr;
1454
1455                 if (!stats->ls_percpu[i])
1456                         continue;
1457
1458                 cntr = lprocfs_stats_counter_get(stats, i, idx);
1459                 ret += lprocfs_read_helper(cntr, &stats->ls_cnt_header[idx],
1460                                            stats->ls_flags, field);
1461         }
1462         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1463         return ret;
1464 }
1465 EXPORT_SYMBOL(lprocfs_stats_collector);
1466
1467 void lprocfs_clear_stats(struct lprocfs_stats *stats)
1468 {
1469         struct lprocfs_counter          *percpu_cntr;
1470         int                             i;
1471         int                             j;
1472         unsigned int                    num_entry;
1473         unsigned long                   flags = 0;
1474
1475         num_entry = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1476
1477         for (i = 0; i < num_entry; i++) {
1478                 if (stats->ls_percpu[i] == NULL)
1479                         continue;
1480                 for (j = 0; j < stats->ls_num; j++) {
1481                         percpu_cntr = lprocfs_stats_counter_get(stats, i, j);
1482                         percpu_cntr->lc_count           = 0;
1483                         percpu_cntr->lc_min             = LC_MIN_INIT;
1484                         percpu_cntr->lc_max             = 0;
1485                         percpu_cntr->lc_sumsquare       = 0;
1486                         percpu_cntr->lc_sum             = 0;
1487                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1488                                 percpu_cntr->lc_sum_irq = 0;
1489                 }
1490         }
1491
1492         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1493 }
1494 EXPORT_SYMBOL(lprocfs_clear_stats);
1495
1496 static ssize_t lprocfs_stats_seq_write(struct file *file,
1497                                        const char __user *buf,
1498                                        size_t len, loff_t *off)
1499 {
1500         struct seq_file *seq = file->private_data;
1501         struct lprocfs_stats *stats = seq->private;
1502
1503         lprocfs_clear_stats(stats);
1504
1505         return len;
1506 }
1507
1508 static void *lprocfs_stats_seq_start(struct seq_file *p, loff_t *pos)
1509 {
1510         struct lprocfs_stats *stats = p->private;
1511
1512         return (*pos < stats->ls_num) ? pos : NULL;
1513 }
1514
1515 static void lprocfs_stats_seq_stop(struct seq_file *p, void *v)
1516 {
1517 }
1518
1519 static void *lprocfs_stats_seq_next(struct seq_file *p, void *v, loff_t *pos)
1520 {
1521         (*pos)++;
1522
1523         return lprocfs_stats_seq_start(p, pos);
1524 }
1525
1526 /* seq file export of one lprocfs counter */
1527 static int lprocfs_stats_seq_show(struct seq_file *p, void *v)
1528 {
1529         struct lprocfs_stats            *stats  = p->private;
1530         struct lprocfs_counter_header   *hdr;
1531         struct lprocfs_counter           ctr;
1532         int                              idx    = *(loff_t *)v;
1533
1534         if (idx == 0) {
1535                 struct timespec64 now;
1536
1537                 ktime_get_real_ts64(&now);
1538                 seq_printf(p, "%-25s %llu.%09lu secs.nsecs\n",
1539                            "snapshot_time", (s64)now.tv_sec, now.tv_nsec);
1540         }
1541
1542         hdr = &stats->ls_cnt_header[idx];
1543         lprocfs_stats_collect(stats, idx, &ctr);
1544
1545         if (ctr.lc_count == 0)
1546                 return 0;
1547
1548         seq_printf(p, "%-25s %lld samples [%s]", hdr->lc_name,
1549                    ctr.lc_count, hdr->lc_units);
1550
1551         if ((hdr->lc_config & LPROCFS_CNTR_AVGMINMAX) && ctr.lc_count > 0) {
1552                 seq_printf(p, " %lld %lld %lld",
1553                            ctr.lc_min, ctr.lc_max, ctr.lc_sum);
1554                 if (hdr->lc_config & LPROCFS_CNTR_STDDEV)
1555                         seq_printf(p, " %llu", ctr.lc_sumsquare);
1556         }
1557         seq_putc(p, '\n');
1558         return 0;
1559 }
1560
1561 static const struct seq_operations lprocfs_stats_seq_sops = {
1562         .start  = lprocfs_stats_seq_start,
1563         .stop   = lprocfs_stats_seq_stop,
1564         .next   = lprocfs_stats_seq_next,
1565         .show   = lprocfs_stats_seq_show,
1566 };
1567
1568 static int lprocfs_stats_seq_open(struct inode *inode, struct file *file)
1569 {
1570         struct seq_file *seq;
1571         int rc;
1572
1573         rc = LPROCFS_ENTRY_CHECK(inode);
1574         if (rc < 0)
1575                 return rc;
1576
1577         rc = seq_open(file, &lprocfs_stats_seq_sops);
1578         if (rc)
1579                 return rc;
1580         seq = file->private_data;
1581         seq->private = inode->i_private ? inode->i_private : PDE_DATA(inode);
1582         return 0;
1583 }
1584
1585 static const struct file_operations lprocfs_stats_seq_fops = {
1586         .owner   = THIS_MODULE,
1587         .open    = lprocfs_stats_seq_open,
1588         .read    = seq_read,
1589         .write   = lprocfs_stats_seq_write,
1590         .llseek  = seq_lseek,
1591         .release = lprocfs_seq_release,
1592 };
1593
1594 int ldebugfs_register_stats(struct dentry *parent, const char *name,
1595                             struct lprocfs_stats *stats)
1596 {
1597         struct dentry *entry;
1598
1599         LASSERT(!IS_ERR_OR_NULL(parent));
1600
1601         entry = debugfs_create_file(name, 0644, parent, stats,
1602                                     &lprocfs_stats_seq_fops);
1603         if (IS_ERR_OR_NULL(entry))
1604                 return entry ? PTR_ERR(entry) : -ENOMEM;
1605
1606         return 0;
1607 }
1608 EXPORT_SYMBOL_GPL(ldebugfs_register_stats);
1609
1610 int lprocfs_register_stats(struct proc_dir_entry *root, const char *name,
1611                            struct lprocfs_stats *stats)
1612 {
1613         struct proc_dir_entry *entry;
1614         LASSERT(root != NULL);
1615
1616         entry = proc_create_data(name, 0644, root,
1617                                  &lprocfs_stats_seq_fops, stats);
1618         if (entry == NULL)
1619                 return -ENOMEM;
1620         return 0;
1621 }
1622 EXPORT_SYMBOL(lprocfs_register_stats);
1623
1624 void lprocfs_counter_init(struct lprocfs_stats *stats, int index,
1625                           unsigned conf, const char *name, const char *units)
1626 {
1627         struct lprocfs_counter_header   *header;
1628         struct lprocfs_counter          *percpu_cntr;
1629         unsigned long                   flags = 0;
1630         unsigned int                    i;
1631         unsigned int                    num_cpu;
1632
1633         LASSERT(stats != NULL);
1634
1635         header = &stats->ls_cnt_header[index];
1636         LASSERTF(header != NULL, "Failed to allocate stats header:[%d]%s/%s\n",
1637                  index, name, units);
1638
1639         header->lc_config = conf;
1640         header->lc_name   = name;
1641         header->lc_units  = units;
1642
1643         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1644         for (i = 0; i < num_cpu; ++i) {
1645                 if (stats->ls_percpu[i] == NULL)
1646                         continue;
1647                 percpu_cntr = lprocfs_stats_counter_get(stats, i, index);
1648                 percpu_cntr->lc_count           = 0;
1649                 percpu_cntr->lc_min             = LC_MIN_INIT;
1650                 percpu_cntr->lc_max             = 0;
1651                 percpu_cntr->lc_sumsquare       = 0;
1652                 percpu_cntr->lc_sum             = 0;
1653                 if ((stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1654                         percpu_cntr->lc_sum_irq = 0;
1655         }
1656         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1657 }
1658 EXPORT_SYMBOL(lprocfs_counter_init);
1659
1660 static const char * const mps_stats[] = {
1661         [LPROC_MD_CLOSE]                = "close",
1662         [LPROC_MD_CREATE]               = "create",
1663         [LPROC_MD_ENQUEUE]              = "enqueue",
1664         [LPROC_MD_GETATTR]              = "getattr",
1665         [LPROC_MD_INTENT_LOCK]          = "intent_lock",
1666         [LPROC_MD_LINK]                 = "link",
1667         [LPROC_MD_RENAME]               = "rename",
1668         [LPROC_MD_SETATTR]              = "setattr",
1669         [LPROC_MD_FSYNC]                = "fsync",
1670         [LPROC_MD_READ_PAGE]            = "read_page",
1671         [LPROC_MD_UNLINK]               = "unlink",
1672         [LPROC_MD_SETXATTR]             = "setxattr",
1673         [LPROC_MD_GETXATTR]             = "getxattr",
1674         [LPROC_MD_INTENT_GETATTR_ASYNC] = "intent_getattr_async",
1675         [LPROC_MD_REVALIDATE_LOCK]      = "revalidate_lock",
1676 };
1677
1678 int lprocfs_alloc_md_stats(struct obd_device *obd,
1679                            unsigned int num_private_stats)
1680 {
1681         struct lprocfs_stats *stats;
1682         unsigned int num_stats;
1683         int rc, i;
1684
1685         /* TODO Ensure that this function is only used where
1686          * appropriate by adding an assertion to the effect that
1687          * obd->obd_type->typ_md_ops is not NULL. We can't do this now
1688          * because mdt_procfs_init() uses this function to allocate
1689          * the stats backing /proc/fs/lustre/mdt/.../md_stats but the
1690          * mdt layer does not use the md_ops interface. This is
1691          * confusing and a waste of memory. See LU-2484.
1692          */
1693         LASSERT(obd->obd_proc_entry != NULL);
1694         LASSERT(obd->obd_md_stats == NULL);
1695
1696         num_stats = ARRAY_SIZE(mps_stats) + num_private_stats;
1697         stats = lprocfs_alloc_stats(num_stats, 0);
1698         if (!stats)
1699                 return -ENOMEM;
1700
1701         for (i = 0; i < ARRAY_SIZE(mps_stats); i++) {
1702                 lprocfs_counter_init(stats, i, 0, mps_stats[i], "reqs");
1703                 if (!stats->ls_cnt_header[i].lc_name) {
1704                         CERROR("Missing md_stat initializer md_op operation at offset %d. Aborting.\n",
1705                                i);
1706                         LBUG();
1707                 }
1708         }
1709
1710         rc = lprocfs_register_stats(obd->obd_proc_entry, "md_stats", stats);
1711         if (rc < 0) {
1712                 lprocfs_free_stats(&stats);
1713         } else {
1714                 obd->obd_md_stats = stats;
1715         }
1716
1717         return rc;
1718 }
1719 EXPORT_SYMBOL(lprocfs_alloc_md_stats);
1720
1721 void lprocfs_free_md_stats(struct obd_device *obd)
1722 {
1723         struct lprocfs_stats *stats = obd->obd_md_stats;
1724
1725         if (stats != NULL) {
1726                 obd->obd_md_stats = NULL;
1727                 lprocfs_free_stats(&stats);
1728         }
1729 }
1730 EXPORT_SYMBOL(lprocfs_free_md_stats);
1731
1732 void lprocfs_init_ldlm_stats(struct lprocfs_stats *ldlm_stats)
1733 {
1734         lprocfs_counter_init(ldlm_stats,
1735                              LDLM_ENQUEUE - LDLM_FIRST_OPC,
1736                              0, "ldlm_enqueue", "reqs");
1737         lprocfs_counter_init(ldlm_stats,
1738                              LDLM_CONVERT - LDLM_FIRST_OPC,
1739                              0, "ldlm_convert", "reqs");
1740         lprocfs_counter_init(ldlm_stats,
1741                              LDLM_CANCEL - LDLM_FIRST_OPC,
1742                              0, "ldlm_cancel", "reqs");
1743         lprocfs_counter_init(ldlm_stats,
1744                              LDLM_BL_CALLBACK - LDLM_FIRST_OPC,
1745                              0, "ldlm_bl_callback", "reqs");
1746         lprocfs_counter_init(ldlm_stats,
1747                              LDLM_CP_CALLBACK - LDLM_FIRST_OPC,
1748                              0, "ldlm_cp_callback", "reqs");
1749         lprocfs_counter_init(ldlm_stats,
1750                              LDLM_GL_CALLBACK - LDLM_FIRST_OPC,
1751                              0, "ldlm_gl_callback", "reqs");
1752 }
1753 EXPORT_SYMBOL(lprocfs_init_ldlm_stats);
1754
1755 __s64 lprocfs_read_helper(struct lprocfs_counter *lc,
1756                           struct lprocfs_counter_header *header,
1757                           enum lprocfs_stats_flags flags,
1758                           enum lprocfs_fields_flags field)
1759 {
1760         __s64 ret = 0;
1761
1762         if (lc == NULL || header == NULL)
1763                 RETURN(0);
1764
1765         switch (field) {
1766                 case LPROCFS_FIELDS_FLAGS_CONFIG:
1767                         ret = header->lc_config;
1768                         break;
1769                 case LPROCFS_FIELDS_FLAGS_SUM:
1770                         ret = lc->lc_sum;
1771                         if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1772                                 ret += lc->lc_sum_irq;
1773                         break;
1774                 case LPROCFS_FIELDS_FLAGS_MIN:
1775                         ret = lc->lc_min;
1776                         break;
1777                 case LPROCFS_FIELDS_FLAGS_MAX:
1778                         ret = lc->lc_max;
1779                         break;
1780                 case LPROCFS_FIELDS_FLAGS_AVG:
1781                         ret = (lc->lc_max - lc->lc_min) / 2;
1782                         break;
1783                 case LPROCFS_FIELDS_FLAGS_SUMSQUARE:
1784                         ret = lc->lc_sumsquare;
1785                         break;
1786                 case LPROCFS_FIELDS_FLAGS_COUNT:
1787                         ret = lc->lc_count;
1788                         break;
1789                 default:
1790                         break;
1791         };
1792         RETURN(ret);
1793 }
1794 EXPORT_SYMBOL(lprocfs_read_helper);
1795
1796 int lprocfs_read_frac_helper(char *buffer, unsigned long count, long val,
1797                              int mult)
1798 {
1799         long decimal_val, frac_val;
1800         int prtn;
1801
1802         if (count < 10)
1803                 return -EINVAL;
1804
1805         decimal_val = val / mult;
1806         prtn = snprintf(buffer, count, "%ld", decimal_val);
1807         frac_val = val % mult;
1808
1809         if (prtn < (count - 4) && frac_val > 0) {
1810                 long temp_frac;
1811                 int i, temp_mult = 1, frac_bits = 0;
1812
1813                 temp_frac = frac_val * 10;
1814                 buffer[prtn++] = '.';
1815                 while (frac_bits < 2 && (temp_frac / mult) < 1 ) {
1816                         /* only reserved 2 bits fraction */
1817                         buffer[prtn++] ='0';
1818                         temp_frac *= 10;
1819                         frac_bits++;
1820                 }
1821                 /*
1822                  * Need to think these cases :
1823                  *      1. #echo x.00 > /proc/xxx       output result : x
1824                  *      2. #echo x.0x > /proc/xxx       output result : x.0x
1825                  *      3. #echo x.x0 > /proc/xxx       output result : x.x
1826                  *      4. #echo x.xx > /proc/xxx       output result : x.xx
1827                  *      Only reserved 2 bits fraction.
1828                  */
1829                 for (i = 0; i < (5 - prtn); i++)
1830                         temp_mult *= 10;
1831
1832                 frac_bits = min((int)count - prtn, 3 - frac_bits);
1833                 prtn += snprintf(buffer + prtn, frac_bits, "%ld",
1834                                  frac_val * temp_mult / mult);
1835
1836                 prtn--;
1837                 while(buffer[prtn] < '1' || buffer[prtn] > '9') {
1838                         prtn--;
1839                         if (buffer[prtn] == '.') {
1840                                 prtn--;
1841                                 break;
1842                         }
1843                 }
1844                 prtn++;
1845         }
1846         buffer[prtn++] ='\n';
1847         return prtn;
1848 }
1849 EXPORT_SYMBOL(lprocfs_read_frac_helper);
1850
1851 int lprocfs_seq_read_frac_helper(struct seq_file *m, long val, int mult)
1852 {
1853         long decimal_val, frac_val;
1854
1855         decimal_val = val / mult;
1856         seq_printf(m, "%ld", decimal_val);
1857         frac_val = val % mult;
1858
1859         if (frac_val > 0) {
1860                 frac_val *= 100;
1861                 frac_val /= mult;
1862         }
1863         if (frac_val > 0) {
1864                 /* Three cases: x0, xx, 0x */
1865                 if ((frac_val % 10) != 0)
1866                         seq_printf(m, ".%ld", frac_val);
1867                 else
1868                         seq_printf(m, ".%ld", frac_val / 10);
1869         }
1870
1871         seq_printf(m, "\n");
1872         return 0;
1873 }
1874 EXPORT_SYMBOL(lprocfs_seq_read_frac_helper);
1875
1876 /* Obtains the conversion factor for the unit specified */
1877 static int get_mult(char unit, __u64 *mult)
1878 {
1879         __u64 units = 1;
1880
1881         switch (unit) {
1882         /* peta, tera, giga, mega, and kilo */
1883         case 'p':
1884         case 'P':
1885                 units <<= 10;
1886         case 't':
1887         case 'T':
1888                 units <<= 10;
1889         case 'g':
1890         case 'G':
1891                 units <<= 10;
1892         case 'm':
1893         case 'M':
1894                 units <<= 10;
1895         case 'k':
1896         case 'K':
1897                 units <<= 10;
1898                 break;
1899         /* some tests expect % to be accepted */
1900         case '%':
1901                 units = 1;
1902                 break;
1903         default:
1904                 return -EINVAL;
1905         }
1906
1907         *mult = units;
1908
1909         return 0;
1910 }
1911
1912 /*
1913  * Ensures the numeric string is valid. The function provides the final
1914  * multiplier in the case a unit exists at the end of the string. It also
1915  * locates the start of the whole and fractional parts (if any). This
1916  * function modifies the string so kstrtoull can be used to parse both
1917  * the whole and fraction portions. This function also figures out
1918  * the base of the number.
1919  */
1920 static int preprocess_numeric_str(char *buffer, __u64 *mult, __u64 def_mult,
1921                                   bool allow_units, char **whole, char **frac,
1922                                   unsigned int *base)
1923 {
1924         bool hit_decimal = false;
1925         bool hit_unit = false;
1926         int rc = 0;
1927         char *start;
1928         *mult = def_mult;
1929         *whole = NULL;
1930         *frac = NULL;
1931         *base = 10;
1932
1933         /* a hex string if it starts with "0x" */
1934         if (buffer[0] == '0' && tolower(buffer[1]) == 'x') {
1935                 *base = 16;
1936                 buffer += 2;
1937         }
1938
1939         start = buffer;
1940
1941         while (*buffer) {
1942                 /* allow for a single new line before the null terminator */
1943                 if (*buffer == '\n') {
1944                         *buffer = '\0';
1945                         buffer++;
1946
1947                         if (*buffer)
1948                                 return -EINVAL;
1949
1950                         break;
1951                 }
1952
1953                 /* any chars after our unit indicates a malformed string */
1954                 if (hit_unit)
1955                         return -EINVAL;
1956
1957                 /* ensure we only hit one decimal */
1958                 if (*buffer == '.') {
1959                         if (hit_decimal)
1960                                 return -EINVAL;
1961
1962                         /* if past start, there's a whole part */
1963                         if (start != buffer)
1964                                 *whole = start;
1965
1966                         *buffer = '\0';
1967                         start = buffer + 1;
1968                         hit_decimal = true;
1969                 } else if (!isdigit(*buffer) &&
1970                            !(*base == 16 && isxdigit(*buffer))) {
1971                         if (allow_units) {
1972                                 /* if we allow units, attempt to get mult */
1973                                 hit_unit = true;
1974                                 rc = get_mult(*buffer, mult);
1975                                 if (rc)
1976                                         return rc;
1977
1978                                 /* string stops here, but keep processing */
1979                                 *buffer = '\0';
1980                         } else {
1981                                 /* bad string */
1982                                 return -EINVAL;
1983                         }
1984                 }
1985
1986                 buffer++;
1987         }
1988
1989         if (hit_decimal) {
1990                 /* hit a decimal, make sure there's a fractional part */
1991                 if (!*start)
1992                         return -EINVAL;
1993
1994                 *frac = start;
1995         } else {
1996                 /* didn't hit a decimal, but may have a whole part */
1997                 if (start != buffer && *start)
1998                         *whole = start;
1999         }
2000
2001         /* malformed string if we didn't get anything */
2002         if (!*frac && !*whole)
2003                 return -EINVAL;
2004
2005         return 0;
2006 }
2007
2008 /*
2009  * Parses a numeric string which can contain a whole and fraction portion
2010  * into a __u64. Accepts a multiplier to apply to the value parsed. Also
2011  * allows the string to have a unit at the end. The function handles
2012  * wrapping of the final unsigned value.
2013  */
2014 static int str_to_u64_parse(char *buffer, unsigned long count,
2015                             __u64 *val, __u64 def_mult, bool allow_units)
2016 {
2017         __u64 whole = 0;
2018         __u64 frac = 0;
2019         unsigned int frac_d = 1;
2020         __u64 wrap_indicator = ULLONG_MAX;
2021         int rc = 0;
2022         __u64 mult;
2023         char *strwhole;
2024         char *strfrac;
2025         unsigned int base = 10;
2026
2027         rc = preprocess_numeric_str(buffer, &mult, def_mult, allow_units,
2028                                     &strwhole, &strfrac, &base);
2029
2030         if (rc)
2031                 return rc;
2032
2033         if (mult == 0) {
2034                 *val = 0;
2035                 return 0;
2036         }
2037
2038         /* the multiplier limits how large the value can be */
2039         wrap_indicator = div64_u64(wrap_indicator, mult);
2040
2041         if (strwhole) {
2042                 rc = kstrtoull(strwhole, base, &whole);
2043                 if (rc)
2044                         return rc;
2045
2046                 if (whole > wrap_indicator)
2047                         return -ERANGE;
2048
2049                 whole *= mult;
2050         }
2051
2052         if (strfrac) {
2053                 if (strlen(strfrac) > 10)
2054                         strfrac[10] = '\0';
2055
2056                 rc = kstrtoull(strfrac, base, &frac);
2057                 if (rc)
2058                         return rc;
2059
2060                 /* determine power of fractional portion */
2061                 while (*strfrac) {
2062                         frac_d *= base;
2063                         strfrac++;
2064                 }
2065
2066                 /* fractional portion is too large to perform calculation */
2067                 if (frac > wrap_indicator)
2068                         return -ERANGE;
2069
2070                 frac *= mult;
2071                 do_div(frac, frac_d);
2072         }
2073
2074         /* check that the sum of whole and fraction fits in u64 */
2075         if (whole > (ULLONG_MAX - frac))
2076                 return -ERANGE;
2077
2078         *val = whole + frac;
2079
2080         return 0;
2081 }
2082
2083 /*
2084  * This function parses numeric/hex strings into __s64. It accepts a multiplier
2085  * which will apply to the value parsed. It also can allow the string to
2086  * have a unit as the last character. The function handles overflow/underflow
2087  * of the signed integer.
2088  */
2089 static int str_to_s64_internal(const char __user *buffer, unsigned long count,
2090                                __s64 *val, __u64 def_mult, bool allow_units)
2091 {
2092         char kernbuf[22];
2093         __u64 tmp;
2094         unsigned int offset = 0;
2095         int signed sign = 1;
2096         __u64 max = LLONG_MAX;
2097         int rc = 0;
2098
2099         if (count > (sizeof(kernbuf) - 1))
2100                 return -EINVAL;
2101
2102         if (copy_from_user(kernbuf, buffer, count))
2103                 return -EFAULT;
2104
2105         kernbuf[count] = '\0';
2106
2107         /* keep track of our sign */
2108         if (*kernbuf == '-') {
2109                 sign = -1;
2110                 offset++;
2111                 /* equivalent to max = -LLONG_MIN, avoids overflow */
2112                 max++;
2113         }
2114
2115         rc = str_to_u64_parse(kernbuf + offset, count - offset,
2116                               &tmp, def_mult, allow_units);
2117         if (rc)
2118                 return rc;
2119
2120         /* check for overflow/underflow */
2121         if (max < tmp)
2122                 return -ERANGE;
2123
2124         *val = (__s64)tmp * sign;
2125
2126         return 0;
2127 }
2128
2129 /**
2130  * Convert a user string into a signed 64 bit number. This function produces
2131  * an error when the value parsed from the string times multiplier underflows or
2132  * overflows. This function only accepts strings that contains digits, an
2133  * optional decimal, and a char representing a unit at the end. If a unit is
2134  * specified in the string, the multiplier provided by the caller is ignored.
2135  * This function can also accept hexadecimal strings which are prefixed with
2136  * "0x".
2137  *
2138  * \param[in] buffer    string consisting of numbers, a decimal, and a unit
2139  * \param[in] count     buffer length
2140  * \param[in] val       if successful, the value represented by the string
2141  * \param[in] defunit   default unit if string doesn't contain one
2142  *
2143  * \retval              0 on success
2144  * \retval              negative number on error
2145  */
2146 int lprocfs_str_with_units_to_s64(const char __user *buffer,
2147                                   unsigned long count, __s64 *val, char defunit)
2148 {
2149         __u64 mult = 1;
2150         int rc;
2151
2152         if (defunit != '1') {
2153                 rc = get_mult(defunit, &mult);
2154                 if (rc)
2155                         return rc;
2156         }
2157
2158         return str_to_s64_internal(buffer, count, val, mult, true);
2159 }
2160 EXPORT_SYMBOL(lprocfs_str_with_units_to_s64);
2161
2162 char *lprocfs_strnstr(const char *s1, const char *s2, size_t len)
2163 {
2164         size_t l2;
2165
2166         l2 = strlen(s2);
2167         if (!l2)
2168                 return (char *)s1;
2169         while (len >= l2) {
2170                 len--;
2171                 if (!memcmp(s1, s2, l2))
2172                         return (char *)s1;
2173                 s1++;
2174         }
2175         return NULL;
2176 }
2177 EXPORT_SYMBOL(lprocfs_strnstr);
2178
2179 /**
2180  * Find the string \a name in the input \a buffer, and return a pointer to the
2181  * value immediately following \a name, reducing \a count appropriately.
2182  * If \a name is not found the original \a buffer is returned.
2183  */
2184 char *lprocfs_find_named_value(const char *buffer, const char *name,
2185                                 size_t *count)
2186 {
2187         char *val;
2188         size_t buflen = *count;
2189
2190         /* there is no strnstr() in rhel5 and ubuntu kernels */
2191         val = lprocfs_strnstr(buffer, name, buflen);
2192         if (val == NULL)
2193                 return (char *)buffer;
2194
2195         val += strlen(name);                             /* skip prefix */
2196         while (val < buffer + buflen && isspace(*val)) /* skip separator */
2197                 val++;
2198
2199         *count = 0;
2200         while (val < buffer + buflen && isalnum(*val)) {
2201                 ++*count;
2202                 ++val;
2203         }
2204
2205         return val - *count;
2206 }
2207 EXPORT_SYMBOL(lprocfs_find_named_value);
2208
2209 int ldebugfs_seq_create(struct dentry *parent, const char *name, umode_t mode,
2210                         const struct file_operations *seq_fops, void *data)
2211 {
2212         struct dentry *entry;
2213
2214         /* Disallow secretly (un)writable entries. */
2215         LASSERT((!seq_fops->write) == (!(mode & 0222)));
2216
2217         entry = debugfs_create_file(name, mode, parent, data, seq_fops);
2218         if (IS_ERR_OR_NULL(entry))
2219                 return entry ? PTR_ERR(entry) : -ENOMEM;
2220
2221         return 0;
2222 }
2223 EXPORT_SYMBOL_GPL(ldebugfs_seq_create);
2224
2225 int lprocfs_seq_create(struct proc_dir_entry *parent,
2226                        const char *name,
2227                        mode_t mode,
2228                        const struct file_operations *seq_fops,
2229                        void *data)
2230 {
2231         struct proc_dir_entry *entry;
2232         ENTRY;
2233
2234         /* Disallow secretly (un)writable entries. */
2235         LASSERT((seq_fops->write == NULL) == ((mode & 0222) == 0));
2236
2237         entry = proc_create_data(name, mode, parent, seq_fops, data);
2238
2239         if (entry == NULL)
2240                 RETURN(-ENOMEM);
2241
2242         RETURN(0);
2243 }
2244 EXPORT_SYMBOL(lprocfs_seq_create);
2245
2246 int lprocfs_obd_seq_create(struct obd_device *dev,
2247                            const char *name,
2248                            mode_t mode,
2249                            const struct file_operations *seq_fops,
2250                            void *data)
2251 {
2252         return (lprocfs_seq_create(dev->obd_proc_entry, name,
2253                                    mode, seq_fops, data));
2254 }
2255 EXPORT_SYMBOL(lprocfs_obd_seq_create);
2256
2257 void lprocfs_oh_tally(struct obd_histogram *oh, unsigned int value)
2258 {
2259         if (value >= OBD_HIST_MAX)
2260                 value = OBD_HIST_MAX - 1;
2261
2262         spin_lock(&oh->oh_lock);
2263         oh->oh_buckets[value]++;
2264         spin_unlock(&oh->oh_lock);
2265 }
2266 EXPORT_SYMBOL(lprocfs_oh_tally);
2267
2268 void lprocfs_oh_tally_log2(struct obd_histogram *oh, unsigned int value)
2269 {
2270         unsigned int val = 0;
2271
2272         if (likely(value != 0))
2273                 val = min(fls(value - 1), OBD_HIST_MAX);
2274
2275         lprocfs_oh_tally(oh, val);
2276 }
2277 EXPORT_SYMBOL(lprocfs_oh_tally_log2);
2278
2279 unsigned long lprocfs_oh_sum(struct obd_histogram *oh)
2280 {
2281         unsigned long ret = 0;
2282         int i;
2283
2284         for (i = 0; i < OBD_HIST_MAX; i++)
2285                 ret +=  oh->oh_buckets[i];
2286         return ret;
2287 }
2288 EXPORT_SYMBOL(lprocfs_oh_sum);
2289
2290 void lprocfs_oh_clear(struct obd_histogram *oh)
2291 {
2292         spin_lock(&oh->oh_lock);
2293         memset(oh->oh_buckets, 0, sizeof(oh->oh_buckets));
2294         spin_unlock(&oh->oh_lock);
2295 }
2296 EXPORT_SYMBOL(lprocfs_oh_clear);
2297
2298 ssize_t lustre_attr_show(struct kobject *kobj,
2299                          struct attribute *attr, char *buf)
2300 {
2301         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
2302
2303         return a->show ? a->show(kobj, attr, buf) : 0;
2304 }
2305 EXPORT_SYMBOL_GPL(lustre_attr_show);
2306
2307 ssize_t lustre_attr_store(struct kobject *kobj, struct attribute *attr,
2308                           const char *buf, size_t len)
2309 {
2310         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
2311
2312         return a->store ? a->store(kobj, attr, buf, len) : len;
2313 }
2314 EXPORT_SYMBOL_GPL(lustre_attr_store);
2315
2316 const struct sysfs_ops lustre_sysfs_ops = {
2317         .show  = lustre_attr_show,
2318         .store = lustre_attr_store,
2319 };
2320 EXPORT_SYMBOL_GPL(lustre_sysfs_ops);
2321
2322 int lprocfs_obd_max_pages_per_rpc_seq_show(struct seq_file *m, void *data)
2323 {
2324         struct obd_device *dev = data;
2325         struct client_obd *cli = &dev->u.cli;
2326
2327         spin_lock(&cli->cl_loi_list_lock);
2328         seq_printf(m, "%d\n", cli->cl_max_pages_per_rpc);
2329         spin_unlock(&cli->cl_loi_list_lock);
2330         return 0;
2331 }
2332 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_show);
2333
2334 ssize_t lprocfs_obd_max_pages_per_rpc_seq_write(struct file *file,
2335                                                 const char __user *buffer,
2336                                                 size_t count, loff_t *off)
2337 {
2338         struct obd_device *dev =
2339                 ((struct seq_file *)file->private_data)->private;
2340         struct client_obd *cli = &dev->u.cli;
2341         struct obd_connect_data *ocd = &cli->cl_import->imp_connect_data;
2342         int chunk_mask, rc;
2343         s64 val;
2344
2345         rc = lprocfs_str_with_units_to_s64(buffer, count, &val, '1');
2346         if (rc)
2347                 return rc;
2348         if (val < 0)
2349                 return -ERANGE;
2350
2351         /* if the max_pages is specified in bytes, convert to pages */
2352         if (val >= ONE_MB_BRW_SIZE)
2353                 val >>= PAGE_SHIFT;
2354
2355         LPROCFS_CLIMP_CHECK(dev);
2356
2357         chunk_mask = ~((1 << (cli->cl_chunkbits - PAGE_SHIFT)) - 1);
2358         /* max_pages_per_rpc must be chunk aligned */
2359         val = (val + ~chunk_mask) & chunk_mask;
2360         if (val == 0 || (ocd->ocd_brw_size != 0 &&
2361                          val > ocd->ocd_brw_size >> PAGE_SHIFT)) {
2362                 LPROCFS_CLIMP_EXIT(dev);
2363                 return -ERANGE;
2364         }
2365         spin_lock(&cli->cl_loi_list_lock);
2366         cli->cl_max_pages_per_rpc = val;
2367         client_adjust_max_dirty(cli);
2368         spin_unlock(&cli->cl_loi_list_lock);
2369
2370         LPROCFS_CLIMP_EXIT(dev);
2371         return count;
2372 }
2373 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_write);
2374
2375 ssize_t short_io_bytes_show(struct kobject *kobj, struct attribute *attr,
2376                             char *buf)
2377 {
2378         struct obd_device *dev = container_of(kobj, struct obd_device,
2379                                               obd_kset.kobj);
2380         struct client_obd *cli = &dev->u.cli;
2381         int rc;
2382
2383         spin_lock(&cli->cl_loi_list_lock);
2384         rc = sprintf(buf, "%d\n", cli->cl_max_short_io_bytes);
2385         spin_unlock(&cli->cl_loi_list_lock);
2386         return rc;
2387 }
2388 EXPORT_SYMBOL(short_io_bytes_show);
2389
2390 /* Used to catch people who think they're specifying pages. */
2391 #define MIN_SHORT_IO_BYTES 64
2392
2393 ssize_t short_io_bytes_store(struct kobject *kobj, struct attribute *attr,
2394                              const char *buffer, size_t count)
2395 {
2396         struct obd_device *dev = container_of(kobj, struct obd_device,
2397                                               obd_kset.kobj);
2398         struct client_obd *cli = &dev->u.cli;
2399         u32 val;
2400         int rc;
2401
2402         LPROCFS_CLIMP_CHECK(dev);
2403
2404         rc = kstrtouint(buffer, 0, &val);
2405         if (rc)
2406                 GOTO(out, rc);
2407
2408         if (val && (val < MIN_SHORT_IO_BYTES || val > OBD_MAX_SHORT_IO_BYTES))
2409                 GOTO(out, rc = -ERANGE);
2410
2411         rc = count;
2412
2413         spin_lock(&cli->cl_loi_list_lock);
2414         if (val > (cli->cl_max_pages_per_rpc << PAGE_SHIFT))
2415                 rc = -ERANGE;
2416         else
2417                 cli->cl_max_short_io_bytes = val;
2418         spin_unlock(&cli->cl_loi_list_lock);
2419
2420 out:
2421         LPROCFS_CLIMP_EXIT(dev);
2422         return rc;
2423 }
2424 EXPORT_SYMBOL(short_io_bytes_store);
2425
2426 int lprocfs_wr_root_squash(const char __user *buffer, unsigned long count,
2427                            struct root_squash_info *squash, char *name)
2428 {
2429         int rc;
2430         char kernbuf[64], *tmp, *errmsg;
2431         unsigned long uid, gid;
2432         ENTRY;
2433
2434         if (count >= sizeof(kernbuf)) {
2435                 errmsg = "string too long";
2436                 GOTO(failed_noprint, rc = -EINVAL);
2437         }
2438         if (copy_from_user(kernbuf, buffer, count)) {
2439                 errmsg = "bad address";
2440                 GOTO(failed_noprint, rc = -EFAULT);
2441         }
2442         kernbuf[count] = '\0';
2443
2444         /* look for uid gid separator */
2445         tmp = strchr(kernbuf, ':');
2446         if (tmp == NULL) {
2447                 errmsg = "needs uid:gid format";
2448                 GOTO(failed, rc = -EINVAL);
2449         }
2450         *tmp = '\0';
2451         tmp++;
2452
2453         /* parse uid */
2454         if (kstrtoul(kernbuf, 0, &uid) != 0) {
2455                 errmsg = "bad uid";
2456                 GOTO(failed, rc = -EINVAL);
2457         }
2458
2459         /* parse gid */
2460         if (kstrtoul(tmp, 0, &gid) != 0) {
2461                 errmsg = "bad gid";
2462                 GOTO(failed, rc = -EINVAL);
2463         }
2464
2465         squash->rsi_uid = uid;
2466         squash->rsi_gid = gid;
2467
2468         LCONSOLE_INFO("%s: root_squash is set to %u:%u\n",
2469                       name, squash->rsi_uid, squash->rsi_gid);
2470         RETURN(count);
2471
2472 failed:
2473         if (tmp != NULL) {
2474                 tmp--;
2475                 *tmp = ':';
2476         }
2477         CWARN("%s: failed to set root_squash to \"%s\", %s, rc = %d\n",
2478               name, kernbuf, errmsg, rc);
2479         RETURN(rc);
2480 failed_noprint:
2481         CWARN("%s: failed to set root_squash due to %s, rc = %d\n",
2482               name, errmsg, rc);
2483         RETURN(rc);
2484 }
2485 EXPORT_SYMBOL(lprocfs_wr_root_squash);
2486
2487
2488 int lprocfs_wr_nosquash_nids(const char __user *buffer, unsigned long count,
2489                              struct root_squash_info *squash, char *name)
2490 {
2491         int rc;
2492         char *kernbuf = NULL;
2493         char *errmsg;
2494         struct list_head tmp;
2495         int len = count;
2496         ENTRY;
2497
2498         if (count > 4096) {
2499                 errmsg = "string too long";
2500                 GOTO(failed, rc = -EINVAL);
2501         }
2502
2503         OBD_ALLOC(kernbuf, count + 1);
2504         if (kernbuf == NULL) {
2505                 errmsg = "no memory";
2506                 GOTO(failed, rc = -ENOMEM);
2507         }
2508         if (copy_from_user(kernbuf, buffer, count)) {
2509                 errmsg = "bad address";
2510                 GOTO(failed, rc = -EFAULT);
2511         }
2512         kernbuf[count] = '\0';
2513
2514         if (count > 0 && kernbuf[count - 1] == '\n')
2515                 len = count - 1;
2516
2517         if ((len == 4 && strncmp(kernbuf, "NONE", len) == 0) ||
2518             (len == 5 && strncmp(kernbuf, "clear", len) == 0)) {
2519                 /* empty string is special case */
2520                 down_write(&squash->rsi_sem);
2521                 if (!list_empty(&squash->rsi_nosquash_nids))
2522                         cfs_free_nidlist(&squash->rsi_nosquash_nids);
2523                 up_write(&squash->rsi_sem);
2524                 LCONSOLE_INFO("%s: nosquash_nids is cleared\n", name);
2525                 OBD_FREE(kernbuf, count + 1);
2526                 RETURN(count);
2527         }
2528
2529         INIT_LIST_HEAD(&tmp);
2530         if (cfs_parse_nidlist(kernbuf, count, &tmp) <= 0) {
2531                 errmsg = "can't parse";
2532                 GOTO(failed, rc = -EINVAL);
2533         }
2534         LCONSOLE_INFO("%s: nosquash_nids set to %s\n",
2535                       name, kernbuf);
2536         OBD_FREE(kernbuf, count + 1);
2537         kernbuf = NULL;
2538
2539         down_write(&squash->rsi_sem);
2540         if (!list_empty(&squash->rsi_nosquash_nids))
2541                 cfs_free_nidlist(&squash->rsi_nosquash_nids);
2542         list_splice(&tmp, &squash->rsi_nosquash_nids);
2543         up_write(&squash->rsi_sem);
2544
2545         RETURN(count);
2546
2547 failed:
2548         if (kernbuf) {
2549                 CWARN("%s: failed to set nosquash_nids to \"%s\", %s rc = %d\n",
2550                       name, kernbuf, errmsg, rc);
2551                 OBD_FREE(kernbuf, count + 1);
2552         } else {
2553                 CWARN("%s: failed to set nosquash_nids due to %s rc = %d\n",
2554                       name, errmsg, rc);
2555         }
2556         RETURN(rc);
2557 }
2558 EXPORT_SYMBOL(lprocfs_wr_nosquash_nids);
2559
2560 #endif /* CONFIG_PROC_FS*/