Whamcloud - gitweb
LU-2139 osc: Track and limit "unstable" pages
[fs/lustre-release.git] / lustre / llite / lproc_llite.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2012, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  */
36 #define DEBUG_SUBSYSTEM S_LLITE
37
38 #include <linux/version.h>
39 #include <lustre_lite.h>
40 #include <lprocfs_status.h>
41 #include <linux/seq_file.h>
42 #include <obd_support.h>
43
44 #include "llite_internal.h"
45
46 struct proc_dir_entry *proc_lustre_fs_root;
47
48 #ifdef LPROCFS
49 /* /proc/lustre/llite mount point registration */
50 extern struct file_operations vvp_dump_pgcache_file_ops;
51 struct file_operations ll_rw_extents_stats_fops;
52 struct file_operations ll_rw_extents_stats_pp_fops;
53 struct file_operations ll_rw_offset_stats_fops;
54
55 static int ll_rd_blksize(char *page, char **start, off_t off, int count,
56                          int *eof, void *data)
57 {
58         struct super_block *sb = (struct super_block *)data;
59         struct obd_statfs osfs;
60         int rc;
61
62         LASSERT(sb != NULL);
63         rc = ll_statfs_internal(sb, &osfs,
64                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
65                                 OBD_STATFS_NODELAY);
66         if (!rc) {
67               *eof = 1;
68               rc = snprintf(page, count, "%u\n", osfs.os_bsize);
69         }
70
71         return rc;
72 }
73
74 static int ll_rd_kbytestotal(char *page, char **start, off_t off, int count,
75                              int *eof, void *data)
76 {
77         struct super_block *sb = (struct super_block *)data;
78         struct obd_statfs osfs;
79         int rc;
80
81         LASSERT(sb != NULL);
82         rc = ll_statfs_internal(sb, &osfs,
83                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
84                                 OBD_STATFS_NODELAY);
85         if (!rc) {
86                 __u32 blk_size = osfs.os_bsize >> 10;
87                 __u64 result = osfs.os_blocks;
88
89                 while (blk_size >>= 1)
90                         result <<= 1;
91
92                 *eof = 1;
93                 rc = snprintf(page, count, LPU64"\n", result);
94         }
95         return rc;
96
97 }
98
99 static int ll_rd_kbytesfree(char *page, char **start, off_t off, int count,
100                             int *eof, void *data)
101 {
102         struct super_block *sb = (struct super_block *)data;
103         struct obd_statfs osfs;
104         int rc;
105
106         LASSERT(sb != NULL);
107         rc = ll_statfs_internal(sb, &osfs,
108                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
109                                 OBD_STATFS_NODELAY);
110         if (!rc) {
111                 __u32 blk_size = osfs.os_bsize >> 10;
112                 __u64 result = osfs.os_bfree;
113
114                 while (blk_size >>= 1)
115                         result <<= 1;
116
117                 *eof = 1;
118                 rc = snprintf(page, count, LPU64"\n", result);
119         }
120         return rc;
121 }
122
123 static int ll_rd_kbytesavail(char *page, char **start, off_t off, int count,
124                              int *eof, void *data)
125 {
126         struct super_block *sb = (struct super_block *)data;
127         struct obd_statfs osfs;
128         int rc;
129
130         LASSERT(sb != NULL);
131         rc = ll_statfs_internal(sb, &osfs,
132                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
133                                 OBD_STATFS_NODELAY);
134         if (!rc) {
135                 __u32 blk_size = osfs.os_bsize >> 10;
136                 __u64 result = osfs.os_bavail;
137
138                 while (blk_size >>= 1)
139                         result <<= 1;
140
141                 *eof = 1;
142                 rc = snprintf(page, count, LPU64"\n", result);
143         }
144         return rc;
145 }
146
147 static int ll_rd_filestotal(char *page, char **start, off_t off, int count,
148                             int *eof, void *data)
149 {
150         struct super_block *sb = (struct super_block *)data;
151         struct obd_statfs osfs;
152         int rc;
153
154         LASSERT(sb != NULL);
155         rc = ll_statfs_internal(sb, &osfs,
156                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
157                                 OBD_STATFS_NODELAY);
158         if (!rc) {
159                  *eof = 1;
160                  rc = snprintf(page, count, LPU64"\n", osfs.os_files);
161         }
162         return rc;
163 }
164
165 static int ll_rd_filesfree(char *page, char **start, off_t off, int count,
166                            int *eof, void *data)
167 {
168         struct super_block *sb = (struct super_block *)data;
169         struct obd_statfs osfs;
170         int rc;
171
172         LASSERT(sb != NULL);
173         rc = ll_statfs_internal(sb, &osfs,
174                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
175                                 OBD_STATFS_NODELAY);
176         if (!rc) {
177                  *eof = 1;
178                  rc = snprintf(page, count, LPU64"\n", osfs.os_ffree);
179         }
180         return rc;
181
182 }
183
184 static int ll_rd_client_type(char *page, char **start, off_t off, int count,
185                             int *eof, void *data)
186 {
187         struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)data);
188         int rc;
189
190         LASSERT(sbi != NULL);
191
192         *eof = 1;
193         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
194                 rc = snprintf(page, count, "remote client\n");
195         else
196                 rc = snprintf(page, count, "local client\n");
197
198         return rc;
199 }
200
201 static int ll_rd_fstype(char *page, char **start, off_t off, int count,
202                         int *eof, void *data)
203 {
204         struct super_block *sb = (struct super_block*)data;
205
206         LASSERT(sb != NULL);
207         *eof = 1;
208         return snprintf(page, count, "%s\n", sb->s_type->name);
209 }
210
211 static int ll_rd_sb_uuid(char *page, char **start, off_t off, int count,
212                          int *eof, void *data)
213 {
214         struct super_block *sb = (struct super_block *)data;
215
216         LASSERT(sb != NULL);
217         *eof = 1;
218         return snprintf(page, count, "%s\n", ll_s2sbi(sb)->ll_sb_uuid.uuid);
219 }
220
221 static int ll_rd_site_stats(char *page, char **start, off_t off,
222                             int count, int *eof, void *data)
223 {
224         struct super_block *sb = data;
225
226         /*
227          * See description of statistical counters in struct cl_site, and
228          * struct lu_site.
229          */
230         return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site),
231                                    page, count);
232 }
233
234 static int ll_rd_max_readahead_mb(char *page, char **start, off_t off,
235                                    int count, int *eof, void *data)
236 {
237         struct super_block *sb = data;
238         struct ll_sb_info *sbi = ll_s2sbi(sb);
239         long pages_number;
240         int mult;
241
242         spin_lock(&sbi->ll_lock);
243         pages_number = sbi->ll_ra_info.ra_max_pages;
244         spin_unlock(&sbi->ll_lock);
245
246         mult = 1 << (20 - PAGE_CACHE_SHIFT);
247         return lprocfs_read_frac_helper(page, count, pages_number, mult);
248 }
249
250 static int ll_wr_max_readahead_mb(struct file *file, const char *buffer,
251                                   unsigned long count, void *data)
252 {
253         struct super_block *sb = data;
254         struct ll_sb_info *sbi = ll_s2sbi(sb);
255         int mult, rc, pages_number;
256
257         mult = 1 << (20 - CFS_PAGE_SHIFT);
258         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
259         if (rc)
260                 return rc;
261
262         if (pages_number < 0 || pages_number > cfs_num_physpages / 2) {
263                 CERROR("can't set file readahead more than %lu MB\n",
264                        cfs_num_physpages >> (20 - CFS_PAGE_SHIFT + 1)); /*1/2 of RAM*/
265                 return -ERANGE;
266         }
267
268         spin_lock(&sbi->ll_lock);
269         sbi->ll_ra_info.ra_max_pages = pages_number;
270         spin_unlock(&sbi->ll_lock);
271
272         return count;
273 }
274
275 static int ll_rd_max_readahead_per_file_mb(char *page, char **start, off_t off,
276                                            int count, int *eof, void *data)
277 {
278         struct super_block *sb = data;
279         struct ll_sb_info *sbi = ll_s2sbi(sb);
280         long pages_number;
281         int mult;
282
283         spin_lock(&sbi->ll_lock);
284         pages_number = sbi->ll_ra_info.ra_max_pages_per_file;
285         spin_unlock(&sbi->ll_lock);
286
287         mult = 1 << (20 - CFS_PAGE_SHIFT);
288         return lprocfs_read_frac_helper(page, count, pages_number, mult);
289 }
290
291 static int ll_wr_max_readahead_per_file_mb(struct file *file, const char *buffer,
292                                           unsigned long count, void *data)
293 {
294         struct super_block *sb = data;
295         struct ll_sb_info *sbi = ll_s2sbi(sb);
296         int mult, rc, pages_number;
297
298         mult = 1 << (20 - CFS_PAGE_SHIFT);
299         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
300         if (rc)
301                 return rc;
302
303         if (pages_number < 0 ||
304                 pages_number > sbi->ll_ra_info.ra_max_pages) {
305                 CERROR("can't set file readahead more than"
306                        "max_read_ahead_mb %lu MB\n",
307                        sbi->ll_ra_info.ra_max_pages);
308                 return -ERANGE;
309         }
310
311         spin_lock(&sbi->ll_lock);
312         sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
313         spin_unlock(&sbi->ll_lock);
314
315         return count;
316 }
317
318 static int ll_rd_max_read_ahead_whole_mb(char *page, char **start, off_t off,
319                                          int count, int *eof, void *data)
320 {
321         struct super_block *sb = data;
322         struct ll_sb_info *sbi = ll_s2sbi(sb);
323         long pages_number;
324         int mult;
325
326         spin_lock(&sbi->ll_lock);
327         pages_number = sbi->ll_ra_info.ra_max_read_ahead_whole_pages;
328         spin_unlock(&sbi->ll_lock);
329
330         mult = 1 << (20 - CFS_PAGE_SHIFT);
331         return lprocfs_read_frac_helper(page, count, pages_number, mult);
332 }
333
334 static int ll_wr_max_read_ahead_whole_mb(struct file *file, const char *buffer,
335                                          unsigned long count, void *data)
336 {
337         struct super_block *sb = data;
338         struct ll_sb_info *sbi = ll_s2sbi(sb);
339         int mult, rc, pages_number;
340
341         mult = 1 << (20 - CFS_PAGE_SHIFT);
342         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
343         if (rc)
344                 return rc;
345
346         /* Cap this at the current max readahead window size, the readahead
347          * algorithm does this anyway so it's pointless to set it larger. */
348         if (pages_number < 0 ||
349             pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
350                 CERROR("can't set max_read_ahead_whole_mb more than "
351                        "max_read_ahead_per_file_mb: %lu\n",
352                         sbi->ll_ra_info.ra_max_pages_per_file >> (20 - CFS_PAGE_SHIFT));
353                 return -ERANGE;
354         }
355
356         spin_lock(&sbi->ll_lock);
357         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
358         spin_unlock(&sbi->ll_lock);
359
360         return count;
361 }
362
363 static int ll_rd_max_cached_mb(char *page, char **start, off_t off,
364                                int count, int *eof, void *data)
365 {
366         struct super_block     *sb    = data;
367         struct ll_sb_info      *sbi   = ll_s2sbi(sb);
368         struct cl_client_cache *cache = &sbi->ll_cache;
369         int shift = 20 - CFS_PAGE_SHIFT;
370         int max_cached_mb;
371         int unused_mb;
372
373         *eof = 1;
374         max_cached_mb = cache->ccc_lru_max >> shift;
375         unused_mb = cfs_atomic_read(&cache->ccc_lru_left) >> shift;
376         return snprintf(page, count,
377                         "users: %d\n"
378                         "max_cached_mb: %d\n"
379                         "used_mb: %d\n"
380                         "unused_mb: %d\n"
381                         "reclaim_count: %u\n",
382                         cfs_atomic_read(&cache->ccc_users),
383                         max_cached_mb,
384                         max_cached_mb - unused_mb,
385                         unused_mb,
386                         cache->ccc_lru_shrinkers);
387 }
388
389 static int ll_wr_max_cached_mb(struct file *file, const char *buffer,
390                                unsigned long count, void *data)
391 {
392         struct super_block *sb = data;
393         struct ll_sb_info *sbi = ll_s2sbi(sb);
394         struct cl_client_cache *cache = &sbi->ll_cache;
395         int mult, rc, pages_number;
396         int diff = 0;
397         int nrpages = 0;
398         ENTRY;
399
400         mult = 1 << (20 - CFS_PAGE_SHIFT);
401         buffer = lprocfs_find_named_value(buffer, "max_cached_mb:", &count);
402         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
403         if (rc)
404                 RETURN(rc);
405
406         if (pages_number < 0 || pages_number > cfs_num_physpages) {
407                 CERROR("%s: can't set max cache more than %lu MB\n",
408                        ll_get_fsname(sb, NULL, 0),
409                        cfs_num_physpages >> (20 - CFS_PAGE_SHIFT));
410                 RETURN(-ERANGE);
411         }
412
413         if (sbi->ll_dt_exp == NULL)
414                 RETURN(-ENODEV);
415
416         spin_lock(&sbi->ll_lock);
417         diff = pages_number - cache->ccc_lru_max;
418         spin_unlock(&sbi->ll_lock);
419
420         /* easy - add more LRU slots. */
421         if (diff >= 0) {
422                 cfs_atomic_add(diff, &cache->ccc_lru_left);
423                 GOTO(out, rc = 0);
424         }
425
426         diff = -diff;
427         while (diff > 0) {
428                 int tmp;
429
430                 /* reduce LRU budget from free slots. */
431                 do {
432                         int ov, nv;
433
434                         ov = cfs_atomic_read(&cache->ccc_lru_left);
435                         if (ov == 0)
436                                 break;
437
438                         nv = ov > diff ? ov - diff : 0;
439                         rc = cfs_atomic_cmpxchg(&cache->ccc_lru_left, ov, nv);
440                         if (likely(ov == rc)) {
441                                 diff -= ov - nv;
442                                 nrpages += ov - nv;
443                                 break;
444                         }
445                 } while (1);
446
447                 if (diff <= 0)
448                         break;
449
450                 /* difficult - have to ask OSCs to drop LRU slots. */
451                 tmp = diff << 1;
452                 rc = obd_set_info_async(NULL, sbi->ll_dt_exp,
453                                 sizeof(KEY_CACHE_LRU_SHRINK),
454                                 KEY_CACHE_LRU_SHRINK,
455                                 sizeof(tmp), &tmp, NULL);
456                 if (rc < 0)
457                         break;
458         }
459
460 out:
461         if (rc >= 0) {
462                 spin_lock(&sbi->ll_lock);
463                 cache->ccc_lru_max = pages_number;
464                 spin_unlock(&sbi->ll_lock);
465                 rc = count;
466         } else {
467                 cfs_atomic_add(nrpages, &cache->ccc_lru_left);
468         }
469         return rc;
470 }
471
472 static int ll_rd_checksum(char *page, char **start, off_t off,
473                           int count, int *eof, void *data)
474 {
475         struct super_block *sb = data;
476         struct ll_sb_info *sbi = ll_s2sbi(sb);
477
478         return snprintf(page, count, "%u\n",
479                         (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
480 }
481
482 static int ll_wr_checksum(struct file *file, const char *buffer,
483                           unsigned long count, void *data)
484 {
485         struct super_block *sb = data;
486         struct ll_sb_info *sbi = ll_s2sbi(sb);
487         int val, rc;
488
489         if (!sbi->ll_dt_exp)
490                 /* Not set up yet */
491                 return -EAGAIN;
492
493         rc = lprocfs_write_helper(buffer, count, &val);
494         if (rc)
495                 return rc;
496         if (val)
497                 sbi->ll_flags |= LL_SBI_CHECKSUM;
498         else
499                 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
500
501         rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
502                                 KEY_CHECKSUM, sizeof(val), &val, NULL);
503         if (rc)
504                 CWARN("Failed to set OSC checksum flags: %d\n", rc);
505
506         return count;
507 }
508
509 static int ll_rd_max_rw_chunk(char *page, char **start, off_t off,
510                           int count, int *eof, void *data)
511 {
512         struct super_block *sb = data;
513
514         return snprintf(page, count, "%lu\n", ll_s2sbi(sb)->ll_max_rw_chunk);
515 }
516
517 static int ll_wr_max_rw_chunk(struct file *file, const char *buffer,
518                           unsigned long count, void *data)
519 {
520         struct super_block *sb = data;
521         int rc, val;
522
523         rc = lprocfs_write_helper(buffer, count, &val);
524         if (rc)
525                 return rc;
526         ll_s2sbi(sb)->ll_max_rw_chunk = val;
527         return count;
528 }
529
530 static int ll_rd_track_id(char *page, int count, void *data,
531                           enum stats_track_type type)
532 {
533         struct super_block *sb = data;
534
535         if (ll_s2sbi(sb)->ll_stats_track_type == type) {
536                 return snprintf(page, count, "%d\n",
537                                 ll_s2sbi(sb)->ll_stats_track_id);
538
539         } else if (ll_s2sbi(sb)->ll_stats_track_type == STATS_TRACK_ALL) {
540                 return snprintf(page, count, "0 (all)\n");
541         } else {
542                 return snprintf(page, count, "untracked\n");
543         }
544 }
545
546 static int ll_wr_track_id(const char *buffer, unsigned long count, void *data,
547                           enum stats_track_type type)
548 {
549         struct super_block *sb = data;
550         int rc, pid;
551
552         rc = lprocfs_write_helper(buffer, count, &pid);
553         if (rc)
554                 return rc;
555         ll_s2sbi(sb)->ll_stats_track_id = pid;
556         if (pid == 0)
557                 ll_s2sbi(sb)->ll_stats_track_type = STATS_TRACK_ALL;
558         else
559                 ll_s2sbi(sb)->ll_stats_track_type = type;
560         lprocfs_clear_stats(ll_s2sbi(sb)->ll_stats);
561         return count;
562 }
563
564 static int ll_rd_track_pid(char *page, char **start, off_t off,
565                           int count, int *eof, void *data)
566 {
567         return (ll_rd_track_id(page, count, data, STATS_TRACK_PID));
568 }
569
570 static int ll_wr_track_pid(struct file *file, const char *buffer,
571                           unsigned long count, void *data)
572 {
573         return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PID));
574 }
575
576 static int ll_rd_track_ppid(char *page, char **start, off_t off,
577                           int count, int *eof, void *data)
578 {
579         return (ll_rd_track_id(page, count, data, STATS_TRACK_PPID));
580 }
581
582 static int ll_wr_track_ppid(struct file *file, const char *buffer,
583                           unsigned long count, void *data)
584 {
585         return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PPID));
586 }
587
588 static int ll_rd_track_gid(char *page, char **start, off_t off,
589                           int count, int *eof, void *data)
590 {
591         return (ll_rd_track_id(page, count, data, STATS_TRACK_GID));
592 }
593
594 static int ll_wr_track_gid(struct file *file, const char *buffer,
595                           unsigned long count, void *data)
596 {
597         return (ll_wr_track_id(buffer, count, data, STATS_TRACK_GID));
598 }
599
600 static int ll_rd_statahead_max(char *page, char **start, off_t off,
601                                int count, int *eof, void *data)
602 {
603         struct super_block *sb = data;
604         struct ll_sb_info *sbi = ll_s2sbi(sb);
605
606         return snprintf(page, count, "%u\n", sbi->ll_sa_max);
607 }
608
609 static int ll_wr_statahead_max(struct file *file, const char *buffer,
610                                unsigned long count, void *data)
611 {
612         struct super_block *sb = data;
613         struct ll_sb_info *sbi = ll_s2sbi(sb);
614         int val, rc;
615
616         rc = lprocfs_write_helper(buffer, count, &val);
617         if (rc)
618                 return rc;
619
620         if (val >= 0 && val <= LL_SA_RPC_MAX)
621                 sbi->ll_sa_max = val;
622         else
623                 CERROR("Bad statahead_max value %d. Valid values are in the "
624                        "range [0, %d]\n", val, LL_SA_RPC_MAX);
625
626         return count;
627 }
628
629 static int ll_rd_statahead_agl(char *page, char **start, off_t off,
630                                int count, int *eof, void *data)
631 {
632         struct super_block *sb = data;
633         struct ll_sb_info *sbi = ll_s2sbi(sb);
634
635         return snprintf(page, count, "%u\n",
636                         sbi->ll_flags & LL_SBI_AGL_ENABLED ? 1 : 0);
637 }
638
639 static int ll_wr_statahead_agl(struct file *file, const char *buffer,
640                                unsigned long count, void *data)
641 {
642         struct super_block *sb = data;
643         struct ll_sb_info *sbi = ll_s2sbi(sb);
644         int val, rc;
645
646         rc = lprocfs_write_helper(buffer, count, &val);
647         if (rc)
648                 return rc;
649
650         if (val)
651                 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
652         else
653                 sbi->ll_flags &= ~LL_SBI_AGL_ENABLED;
654
655         return count;
656 }
657
658 static int ll_rd_statahead_stats(char *page, char **start, off_t off,
659                                  int count, int *eof, void *data)
660 {
661         struct super_block *sb = data;
662         struct ll_sb_info *sbi = ll_s2sbi(sb);
663
664         return snprintf(page, count,
665                         "statahead total: %u\n"
666                         "statahead wrong: %u\n"
667                         "agl total: %u\n",
668                         atomic_read(&sbi->ll_sa_total),
669                         atomic_read(&sbi->ll_sa_wrong),
670                         atomic_read(&sbi->ll_agl_total));
671 }
672
673 static int ll_rd_lazystatfs(char *page, char **start, off_t off,
674                             int count, int *eof, void *data)
675 {
676         struct super_block *sb = data;
677         struct ll_sb_info *sbi = ll_s2sbi(sb);
678
679         return snprintf(page, count, "%u\n",
680                         (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
681 }
682
683 static int ll_wr_lazystatfs(struct file *file, const char *buffer,
684                             unsigned long count, void *data)
685 {
686         struct super_block *sb = data;
687         struct ll_sb_info *sbi = ll_s2sbi(sb);
688         int val, rc;
689
690         rc = lprocfs_write_helper(buffer, count, &val);
691         if (rc)
692                 return rc;
693
694         if (val)
695                 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
696         else
697                 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
698
699         return count;
700 }
701
702 static int ll_rd_maxea_size(char *page, char **start, off_t off,
703                             int count, int *eof, void *data)
704 {
705         struct super_block *sb = data;
706         struct ll_sb_info *sbi = ll_s2sbi(sb);
707         unsigned int ealen;
708         int rc;
709
710         rc = ll_get_max_mdsize(sbi, &ealen);
711         if (rc)
712                 return rc;
713
714         return snprintf(page, count, "%u\n", ealen);
715 }
716
717 static int ll_rd_sbi_flags(char *page, char **start, off_t off,
718                                 int count, int *eof, void *data)
719 {
720         const char *str[] = LL_SBI_FLAGS;
721         struct super_block *sb = data;
722         int flags = ll_s2sbi(sb)->ll_flags;
723         int i = 0;
724         int rc = 0;
725
726         while (flags != 0) {
727                 if (ARRAY_SIZE(str) <= i) {
728                         CERROR("%s: Revise array LL_SBI_FLAGS to match sbi "
729                                 "flags please.\n", ll_get_fsname(sb, NULL, 0));
730                         return -EINVAL;
731                 }
732
733                 if (flags & 0x1)
734                         rc += snprintf(page + rc, count - rc, "%s ", str[i]);
735                 flags >>= 1;
736                 ++i;
737         }
738         if (rc > 0)
739                 rc += snprintf(page + rc, count - rc, "\b\n");
740         return rc;
741 }
742
743 static int ll_rd_unstable_stats(char *page, char **start, off_t off,
744                               int count, int *eof, void *data)
745 {
746         struct super_block      *sb    = data;
747         struct ll_sb_info       *sbi   = ll_s2sbi(sb);
748         struct cl_client_cache  *cache = &sbi->ll_cache;
749         int pages, mb, rc;
750
751         pages = cfs_atomic_read(&cache->ccc_unstable_nr);
752         mb    = (pages * CFS_PAGE_SIZE) >> 20;
753
754         rc = snprintf(page, count, "unstable_pages: %8d\n"
755                                    "unstable_mb:    %8d\n", pages, mb);
756
757         return rc;
758 }
759
760 static struct lprocfs_vars lprocfs_llite_obd_vars[] = {
761         { "uuid",         ll_rd_sb_uuid,          0, 0 },
762         //{ "mntpt_path",   ll_rd_path,             0, 0 },
763         { "fstype",       ll_rd_fstype,           0, 0 },
764         { "site",         ll_rd_site_stats,       0, 0 },
765         { "blocksize",    ll_rd_blksize,          0, 0 },
766         { "kbytestotal",  ll_rd_kbytestotal,      0, 0 },
767         { "kbytesfree",   ll_rd_kbytesfree,       0, 0 },
768         { "kbytesavail",  ll_rd_kbytesavail,      0, 0 },
769         { "filestotal",   ll_rd_filestotal,       0, 0 },
770         { "filesfree",    ll_rd_filesfree,        0, 0 },
771         { "client_type",  ll_rd_client_type,      0, 0 },
772         //{ "filegroups",   lprocfs_rd_filegroups,  0, 0 },
773         { "max_read_ahead_mb", ll_rd_max_readahead_mb,
774                                ll_wr_max_readahead_mb, 0 },
775         { "max_read_ahead_per_file_mb", ll_rd_max_readahead_per_file_mb,
776                                         ll_wr_max_readahead_per_file_mb, 0 },
777         { "max_read_ahead_whole_mb", ll_rd_max_read_ahead_whole_mb,
778                                      ll_wr_max_read_ahead_whole_mb, 0 },
779         { "max_cached_mb",    ll_rd_max_cached_mb, ll_wr_max_cached_mb, 0 },
780         { "checksum_pages",   ll_rd_checksum, ll_wr_checksum, 0 },
781         { "max_rw_chunk",     ll_rd_max_rw_chunk, ll_wr_max_rw_chunk, 0 },
782         { "stats_track_pid",  ll_rd_track_pid, ll_wr_track_pid, 0 },
783         { "stats_track_ppid", ll_rd_track_ppid, ll_wr_track_ppid, 0 },
784         { "stats_track_gid",  ll_rd_track_gid, ll_wr_track_gid, 0 },
785         { "statahead_max",    ll_rd_statahead_max, ll_wr_statahead_max, 0 },
786         { "statahead_agl",    ll_rd_statahead_agl, ll_wr_statahead_agl, 0 },
787         { "statahead_stats",  ll_rd_statahead_stats, 0, 0 },
788         { "lazystatfs",       ll_rd_lazystatfs, ll_wr_lazystatfs, 0 },
789         { "max_easize",       ll_rd_maxea_size, 0, 0 },
790         { "sbi_flags",        ll_rd_sbi_flags, 0, 0 },
791         { "unstable_stats",   ll_rd_unstable_stats, 0, 0},
792         { 0 }
793 };
794
795 #define MAX_STRING_SIZE 128
796
797 struct llite_file_opcode {
798         __u32       opcode;
799         __u32       type;
800         const char *opname;
801 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
802         /* file operation */
803         { LPROC_LL_DIRTY_HITS,     LPROCFS_TYPE_REGS, "dirty_pages_hits" },
804         { LPROC_LL_DIRTY_MISSES,   LPROCFS_TYPE_REGS, "dirty_pages_misses" },
805         { LPROC_LL_READ_BYTES,     LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
806                                    "read_bytes" },
807         { LPROC_LL_WRITE_BYTES,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
808                                    "write_bytes" },
809         { LPROC_LL_BRW_READ,       LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
810                                    "brw_read" },
811         { LPROC_LL_BRW_WRITE,      LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
812                                    "brw_write" },
813         { LPROC_LL_OSC_READ,       LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
814                                    "osc_read" },
815         { LPROC_LL_OSC_WRITE,      LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
816                                    "osc_write" },
817         { LPROC_LL_IOCTL,          LPROCFS_TYPE_REGS, "ioctl" },
818         { LPROC_LL_OPEN,           LPROCFS_TYPE_REGS, "open" },
819         { LPROC_LL_RELEASE,        LPROCFS_TYPE_REGS, "close" },
820         { LPROC_LL_MAP,            LPROCFS_TYPE_REGS, "mmap" },
821         { LPROC_LL_LLSEEK,         LPROCFS_TYPE_REGS, "seek" },
822         { LPROC_LL_FSYNC,          LPROCFS_TYPE_REGS, "fsync" },
823         { LPROC_LL_READDIR,        LPROCFS_TYPE_REGS, "readdir" },
824         /* inode operation */
825         { LPROC_LL_SETATTR,        LPROCFS_TYPE_REGS, "setattr" },
826         { LPROC_LL_TRUNC,          LPROCFS_TYPE_REGS, "truncate" },
827         { LPROC_LL_FLOCK,          LPROCFS_TYPE_REGS, "flock" },
828         { LPROC_LL_GETATTR,        LPROCFS_TYPE_REGS, "getattr" },
829         /* dir inode operation */
830         { LPROC_LL_CREATE,         LPROCFS_TYPE_REGS, "create" },
831         { LPROC_LL_LINK,           LPROCFS_TYPE_REGS, "link" },
832         { LPROC_LL_UNLINK,         LPROCFS_TYPE_REGS, "unlink" },
833         { LPROC_LL_SYMLINK,        LPROCFS_TYPE_REGS, "symlink" },
834         { LPROC_LL_MKDIR,          LPROCFS_TYPE_REGS, "mkdir" },
835         { LPROC_LL_RMDIR,          LPROCFS_TYPE_REGS, "rmdir" },
836         { LPROC_LL_MKNOD,          LPROCFS_TYPE_REGS, "mknod" },
837         { LPROC_LL_RENAME,         LPROCFS_TYPE_REGS, "rename" },
838         /* special inode operation */
839         { LPROC_LL_STAFS,          LPROCFS_TYPE_REGS, "statfs" },
840         { LPROC_LL_ALLOC_INODE,    LPROCFS_TYPE_REGS, "alloc_inode" },
841         { LPROC_LL_SETXATTR,       LPROCFS_TYPE_REGS, "setxattr" },
842         { LPROC_LL_GETXATTR,       LPROCFS_TYPE_REGS, "getxattr" },
843         { LPROC_LL_LISTXATTR,      LPROCFS_TYPE_REGS, "listxattr" },
844         { LPROC_LL_REMOVEXATTR,    LPROCFS_TYPE_REGS, "removexattr" },
845         { LPROC_LL_INODE_PERM,     LPROCFS_TYPE_REGS, "inode_permission" },
846 };
847
848 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
849 {
850         if (!sbi->ll_stats)
851                 return;
852         if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
853                 lprocfs_counter_add(sbi->ll_stats, op, count);
854         else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
855                  sbi->ll_stats_track_id == current->pid)
856                 lprocfs_counter_add(sbi->ll_stats, op, count);
857         else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
858                  sbi->ll_stats_track_id == current->parent->pid)
859                 lprocfs_counter_add(sbi->ll_stats, op, count);
860         else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
861                  sbi->ll_stats_track_id == cfs_curproc_gid())
862                 lprocfs_counter_add(sbi->ll_stats, op, count);
863 }
864 EXPORT_SYMBOL(ll_stats_ops_tally);
865
866 static const char *ra_stat_string[] = {
867         [RA_STAT_HIT] = "hits",
868         [RA_STAT_MISS] = "misses",
869         [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
870         [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
871         [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
872         [RA_STAT_FAILED_MATCH] = "failed lock match",
873         [RA_STAT_DISCARDED] = "read but discarded",
874         [RA_STAT_ZERO_LEN] = "zero length file",
875         [RA_STAT_ZERO_WINDOW] = "zero size window",
876         [RA_STAT_EOF] = "read-ahead to EOF",
877         [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
878         [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
879 };
880
881
882 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
883                                 struct super_block *sb, char *osc, char *mdc)
884 {
885         struct lprocfs_vars lvars[2];
886         struct lustre_sb_info *lsi = s2lsi(sb);
887         struct ll_sb_info *sbi = ll_s2sbi(sb);
888         struct obd_device *obd;
889         char name[MAX_STRING_SIZE + 1], *ptr;
890         int err, id, len, rc;
891         ENTRY;
892
893         memset(lvars, 0, sizeof(lvars));
894
895         name[MAX_STRING_SIZE] = '\0';
896         lvars[0].name = name;
897
898         LASSERT(sbi != NULL);
899         LASSERT(mdc != NULL);
900         LASSERT(osc != NULL);
901
902         /* Get fsname */
903         len = strlen(lsi->lsi_lmd->lmd_profile);
904         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
905         if (ptr && (strcmp(ptr, "-client") == 0))
906                 len -= 7;
907
908         /* Mount info */
909         snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
910                  lsi->lsi_lmd->lmd_profile, sb);
911
912         sbi->ll_proc_root = lprocfs_register(name, parent, NULL, NULL);
913         if (IS_ERR(sbi->ll_proc_root)) {
914                 err = PTR_ERR(sbi->ll_proc_root);
915                 sbi->ll_proc_root = NULL;
916                 RETURN(err);
917         }
918
919         rc = lprocfs_seq_create(sbi->ll_proc_root, "dump_page_cache", 0444,
920                                 &vvp_dump_pgcache_file_ops, sbi);
921         if (rc)
922                 CWARN("Error adding the dump_page_cache file\n");
923
924         rc = lprocfs_seq_create(sbi->ll_proc_root, "extents_stats", 0644,
925                                 &ll_rw_extents_stats_fops, sbi);
926         if (rc)
927                 CWARN("Error adding the extent_stats file\n");
928
929         rc = lprocfs_seq_create(sbi->ll_proc_root, "extents_stats_per_process",
930                                 0644, &ll_rw_extents_stats_pp_fops, sbi);
931         if (rc)
932                 CWARN("Error adding the extents_stats_per_process file\n");
933
934         rc = lprocfs_seq_create(sbi->ll_proc_root, "offset_stats", 0644,
935                                 &ll_rw_offset_stats_fops, sbi);
936         if (rc)
937                 CWARN("Error adding the offset_stats file\n");
938
939         /* File operations stats */
940         sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
941                                             LPROCFS_STATS_FLAG_NONE);
942         if (sbi->ll_stats == NULL)
943                 GOTO(out, err = -ENOMEM);
944         /* do counter init */
945         for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
946                 __u32 type = llite_opcode_table[id].type;
947                 void *ptr = NULL;
948                 if (type & LPROCFS_TYPE_REGS)
949                         ptr = "regs";
950                 else if (type & LPROCFS_TYPE_BYTES)
951                         ptr = "bytes";
952                 else if (type & LPROCFS_TYPE_PAGES)
953                         ptr = "pages";
954                 lprocfs_counter_init(sbi->ll_stats,
955                                      llite_opcode_table[id].opcode,
956                                      (type & LPROCFS_CNTR_AVGMINMAX),
957                                      llite_opcode_table[id].opname, ptr);
958         }
959         err = lprocfs_register_stats(sbi->ll_proc_root, "stats", sbi->ll_stats);
960         if (err)
961                 GOTO(out, err);
962
963         sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
964                                                LPROCFS_STATS_FLAG_NONE);
965         if (sbi->ll_ra_stats == NULL)
966                 GOTO(out, err = -ENOMEM);
967
968         for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
969                 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
970                                      ra_stat_string[id], "pages");
971         err = lprocfs_register_stats(sbi->ll_proc_root, "read_ahead_stats",
972                                      sbi->ll_ra_stats);
973         if (err)
974                 GOTO(out, err);
975
976
977         err = lprocfs_add_vars(sbi->ll_proc_root, lprocfs_llite_obd_vars, sb);
978         if (err)
979                 GOTO(out, err);
980
981         /* MDC info */
982         obd = class_name2obd(mdc);
983
984         LASSERT(obd != NULL);
985         LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
986         LASSERT(obd->obd_type->typ_name != NULL);
987
988         snprintf(name, MAX_STRING_SIZE, "%s/common_name",
989                  obd->obd_type->typ_name);
990         lvars[0].read_fptr = lprocfs_rd_name;
991         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
992         if (err)
993                 GOTO(out, err);
994
995         snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
996         lvars[0].read_fptr = lprocfs_rd_uuid;
997         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
998         if (err)
999                 GOTO(out, err);
1000
1001         /* OSC */
1002         obd = class_name2obd(osc);
1003
1004         LASSERT(obd != NULL);
1005         LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
1006         LASSERT(obd->obd_type->typ_name != NULL);
1007
1008         snprintf(name, MAX_STRING_SIZE, "%s/common_name",
1009                  obd->obd_type->typ_name);
1010         lvars[0].read_fptr = lprocfs_rd_name;
1011         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
1012         if (err)
1013                 GOTO(out, err);
1014
1015         snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
1016         lvars[0].read_fptr = lprocfs_rd_uuid;
1017         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
1018 out:
1019         if (err) {
1020                 lprocfs_remove(&sbi->ll_proc_root);
1021                 lprocfs_free_stats(&sbi->ll_ra_stats);
1022                 lprocfs_free_stats(&sbi->ll_stats);
1023         }
1024         RETURN(err);
1025 }
1026
1027 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi)
1028 {
1029         if (sbi->ll_proc_root) {
1030                 lprocfs_remove(&sbi->ll_proc_root);
1031                 lprocfs_free_stats(&sbi->ll_ra_stats);
1032                 lprocfs_free_stats(&sbi->ll_stats);
1033         }
1034 }
1035 #undef MAX_STRING_SIZE
1036
1037 #define pct(a,b) (b ? a * 100 / b : 0)
1038
1039 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1040                                    struct seq_file *seq, int which)
1041 {
1042         unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1043         unsigned long start, end, r, w;
1044         char *unitp = "KMGTPEZY";
1045         int i, units = 10;
1046         struct per_process_info *pp_info = &io_extents->pp_extents[which];
1047
1048         read_cum = 0;
1049         write_cum = 0;
1050         start = 0;
1051
1052         for(i = 0; i < LL_HIST_MAX; i++) {
1053                 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1054                 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1055         }
1056
1057         for(i = 0; i < LL_HIST_MAX; i++) {
1058                 r = pp_info->pp_r_hist.oh_buckets[i];
1059                 w = pp_info->pp_w_hist.oh_buckets[i];
1060                 read_cum += r;
1061                 write_cum += w;
1062                 end = 1 << (i + LL_HIST_START - units);
1063                 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4lu %4lu  | "
1064                            "%14lu %4lu %4lu\n", start, *unitp, end, *unitp,
1065                            (i == LL_HIST_MAX - 1) ? '+' : ' ',
1066                            r, pct(r, read_tot), pct(read_cum, read_tot),
1067                            w, pct(w, write_tot), pct(write_cum, write_tot));
1068                 start = end;
1069                 if (start == 1<<10) {
1070                         start = 1;
1071                         units += 10;
1072                         unitp++;
1073                 }
1074                 if (read_cum == read_tot && write_cum == write_tot)
1075                         break;
1076         }
1077 }
1078
1079 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1080 {
1081         struct timeval now;
1082         struct ll_sb_info *sbi = seq->private;
1083         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1084         int k;
1085
1086         cfs_gettimeofday(&now);
1087
1088         if (!sbi->ll_rw_stats_on) {
1089                 seq_printf(seq, "disabled\n"
1090                                 "write anything in this file to activate, "
1091                                 "then 0 or \"[D/d]isabled\" to deactivate\n");
1092                 return 0;
1093         }
1094         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1095                    now.tv_sec, now.tv_usec);
1096         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1097         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1098                    "extents", "calls", "%", "cum%",
1099                    "calls", "%", "cum%");
1100         spin_lock(&sbi->ll_pp_extent_lock);
1101         for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1102                 if (io_extents->pp_extents[k].pid != 0) {
1103                         seq_printf(seq, "\nPID: %d\n",
1104                                    io_extents->pp_extents[k].pid);
1105                         ll_display_extents_info(io_extents, seq, k);
1106                 }
1107         }
1108         spin_unlock(&sbi->ll_pp_extent_lock);
1109         return 0;
1110 }
1111
1112 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1113                                                 const char *buf, size_t len,
1114                                                 loff_t *off)
1115 {
1116         struct seq_file *seq = file->private_data;
1117         struct ll_sb_info *sbi = seq->private;
1118         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1119         int i;
1120         int value = 1, rc = 0;
1121
1122         rc = lprocfs_write_helper(buf, len, &value);
1123         if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1124                        strcmp(buf, "Disabled") == 0))
1125                 value = 0;
1126
1127         if (value == 0)
1128                 sbi->ll_rw_stats_on = 0;
1129         else
1130                 sbi->ll_rw_stats_on = 1;
1131
1132         spin_lock(&sbi->ll_pp_extent_lock);
1133         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1134                 io_extents->pp_extents[i].pid = 0;
1135                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1136                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1137         }
1138         spin_unlock(&sbi->ll_pp_extent_lock);
1139         return len;
1140 }
1141
1142 LPROC_SEQ_FOPS(ll_rw_extents_stats_pp);
1143
1144 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1145 {
1146         struct timeval now;
1147         struct ll_sb_info *sbi = seq->private;
1148         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1149
1150         cfs_gettimeofday(&now);
1151
1152         if (!sbi->ll_rw_stats_on) {
1153                 seq_printf(seq, "disabled\n"
1154                                 "write anything in this file to activate, "
1155                                 "then 0 or \"[D/d]isabled\" to deactivate\n");
1156                 return 0;
1157         }
1158         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1159                    now.tv_sec, now.tv_usec);
1160
1161         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1162         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1163                    "extents", "calls", "%", "cum%",
1164                    "calls", "%", "cum%");
1165         spin_lock(&sbi->ll_lock);
1166         ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1167         spin_unlock(&sbi->ll_lock);
1168
1169         return 0;
1170 }
1171
1172 static ssize_t ll_rw_extents_stats_seq_write(struct file *file, const char *buf,
1173                                         size_t len, loff_t *off)
1174 {
1175         struct seq_file *seq = file->private_data;
1176         struct ll_sb_info *sbi = seq->private;
1177         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1178         int i;
1179         int value = 1, rc = 0;
1180
1181         rc = lprocfs_write_helper(buf, len, &value);
1182         if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1183                        strcmp(buf, "Disabled") == 0))
1184                 value = 0;
1185
1186         if (value == 0)
1187                 sbi->ll_rw_stats_on = 0;
1188         else
1189                 sbi->ll_rw_stats_on = 1;
1190         spin_lock(&sbi->ll_pp_extent_lock);
1191         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1192                 io_extents->pp_extents[i].pid = 0;
1193                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1194                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1195         }
1196         spin_unlock(&sbi->ll_pp_extent_lock);
1197
1198         return len;
1199 }
1200
1201 LPROC_SEQ_FOPS(ll_rw_extents_stats);
1202
1203 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1204                        struct ll_file_data *file, loff_t pos,
1205                        size_t count, int rw)
1206 {
1207         int i, cur = -1;
1208         struct ll_rw_process_info *process;
1209         struct ll_rw_process_info *offset;
1210         int *off_count = &sbi->ll_rw_offset_entry_count;
1211         int *process_count = &sbi->ll_offset_process_count;
1212         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1213
1214         if(!sbi->ll_rw_stats_on)
1215                 return;
1216         process = sbi->ll_rw_process_info;
1217         offset = sbi->ll_rw_offset_info;
1218
1219         spin_lock(&sbi->ll_pp_extent_lock);
1220         /* Extent statistics */
1221         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1222                 if(io_extents->pp_extents[i].pid == pid) {
1223                         cur = i;
1224                         break;
1225                 }
1226         }
1227
1228         if (cur == -1) {
1229                 /* new process */
1230                 sbi->ll_extent_process_count =
1231                         (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1232                 cur = sbi->ll_extent_process_count;
1233                 io_extents->pp_extents[cur].pid = pid;
1234                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1235                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1236         }
1237
1238         for(i = 0; (count >= (1 << LL_HIST_START << i)) &&
1239              (i < (LL_HIST_MAX - 1)); i++);
1240         if (rw == 0) {
1241                 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1242                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1243         } else {
1244                 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1245                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1246         }
1247         spin_unlock(&sbi->ll_pp_extent_lock);
1248
1249         spin_lock(&sbi->ll_process_lock);
1250         /* Offset statistics */
1251         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1252                 if (process[i].rw_pid == pid) {
1253                         if (process[i].rw_last_file != file) {
1254                                 process[i].rw_range_start = pos;
1255                                 process[i].rw_last_file_pos = pos + count;
1256                                 process[i].rw_smallest_extent = count;
1257                                 process[i].rw_largest_extent = count;
1258                                 process[i].rw_offset = 0;
1259                                 process[i].rw_last_file = file;
1260                                 spin_unlock(&sbi->ll_process_lock);
1261                                 return;
1262                         }
1263                         if (process[i].rw_last_file_pos != pos) {
1264                                 *off_count =
1265                                     (*off_count + 1) % LL_OFFSET_HIST_MAX;
1266                                 offset[*off_count].rw_op = process[i].rw_op;
1267                                 offset[*off_count].rw_pid = pid;
1268                                 offset[*off_count].rw_range_start =
1269                                         process[i].rw_range_start;
1270                                 offset[*off_count].rw_range_end =
1271                                         process[i].rw_last_file_pos;
1272                                 offset[*off_count].rw_smallest_extent =
1273                                         process[i].rw_smallest_extent;
1274                                 offset[*off_count].rw_largest_extent =
1275                                         process[i].rw_largest_extent;
1276                                 offset[*off_count].rw_offset =
1277                                         process[i].rw_offset;
1278                                 process[i].rw_op = rw;
1279                                 process[i].rw_range_start = pos;
1280                                 process[i].rw_smallest_extent = count;
1281                                 process[i].rw_largest_extent = count;
1282                                 process[i].rw_offset = pos -
1283                                         process[i].rw_last_file_pos;
1284                         }
1285                         if(process[i].rw_smallest_extent > count)
1286                                 process[i].rw_smallest_extent = count;
1287                         if(process[i].rw_largest_extent < count)
1288                                 process[i].rw_largest_extent = count;
1289                         process[i].rw_last_file_pos = pos + count;
1290                         spin_unlock(&sbi->ll_process_lock);
1291                         return;
1292                 }
1293         }
1294         *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1295         process[*process_count].rw_pid = pid;
1296         process[*process_count].rw_op = rw;
1297         process[*process_count].rw_range_start = pos;
1298         process[*process_count].rw_last_file_pos = pos + count;
1299         process[*process_count].rw_smallest_extent = count;
1300         process[*process_count].rw_largest_extent = count;
1301         process[*process_count].rw_offset = 0;
1302         process[*process_count].rw_last_file = file;
1303         spin_unlock(&sbi->ll_process_lock);
1304 }
1305
1306 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1307 {
1308         struct timeval now;
1309         struct ll_sb_info *sbi = seq->private;
1310         struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1311         struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1312         int i;
1313
1314         cfs_gettimeofday(&now);
1315
1316         if (!sbi->ll_rw_stats_on) {
1317                 seq_printf(seq, "disabled\n"
1318                                 "write anything in this file to activate, "
1319                                 "then 0 or \"[D/d]isabled\" to deactivate\n");
1320                 return 0;
1321         }
1322         spin_lock(&sbi->ll_process_lock);
1323
1324         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1325                    now.tv_sec, now.tv_usec);
1326         seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1327                    "R/W", "PID", "RANGE START", "RANGE END",
1328                    "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1329         /* We stored the discontiguous offsets here; print them first */
1330         for(i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1331                 if (offset[i].rw_pid != 0)
1332                         seq_printf(seq,"%3c %10d %14Lu %14Lu %17lu %17lu %14Lu",
1333                                    offset[i].rw_op ? 'W' : 'R',
1334                                    offset[i].rw_pid,
1335                                    offset[i].rw_range_start,
1336                                    offset[i].rw_range_end,
1337                                    (unsigned long)offset[i].rw_smallest_extent,
1338                                    (unsigned long)offset[i].rw_largest_extent,
1339                                    offset[i].rw_offset);
1340         }
1341         /* Then print the current offsets for each process */
1342         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1343                 if (process[i].rw_pid != 0)
1344                         seq_printf(seq,"%3c %10d %14Lu %14Lu %17lu %17lu %14Lu",
1345                                    process[i].rw_op ? 'W' : 'R',
1346                                    process[i].rw_pid,
1347                                    process[i].rw_range_start,
1348                                    process[i].rw_last_file_pos,
1349                                    (unsigned long)process[i].rw_smallest_extent,
1350                                    (unsigned long)process[i].rw_largest_extent,
1351                                    process[i].rw_offset);
1352         }
1353         spin_unlock(&sbi->ll_process_lock);
1354
1355         return 0;
1356 }
1357
1358 static ssize_t ll_rw_offset_stats_seq_write(struct file *file, const char *buf,
1359                                        size_t len, loff_t *off)
1360 {
1361         struct seq_file *seq = file->private_data;
1362         struct ll_sb_info *sbi = seq->private;
1363         struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1364         struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1365         int value = 1, rc = 0;
1366
1367         rc = lprocfs_write_helper(buf, len, &value);
1368
1369         if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1370                            strcmp(buf, "Disabled") == 0))
1371                 value = 0;
1372
1373         if (value == 0)
1374                 sbi->ll_rw_stats_on = 0;
1375         else
1376                 sbi->ll_rw_stats_on = 1;
1377
1378         spin_lock(&sbi->ll_process_lock);
1379         sbi->ll_offset_process_count = 0;
1380         sbi->ll_rw_offset_entry_count = 0;
1381         memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1382                LL_PROCESS_HIST_MAX);
1383         memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1384                LL_OFFSET_HIST_MAX);
1385         spin_unlock(&sbi->ll_process_lock);
1386
1387         return len;
1388 }
1389
1390 LPROC_SEQ_FOPS(ll_rw_offset_stats);
1391
1392 void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars)
1393 {
1394     lvars->module_vars  = NULL;
1395     lvars->obd_vars     = lprocfs_llite_obd_vars;
1396 }
1397 #endif /* LPROCFS */