Whamcloud - gitweb
Branch b1_6
[fs/lustre-release.git] / lustre / obdfilter / lproc_obdfilter.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  *  Copyright (C) 2002, 2003 Cluster File Systems, Inc.
5  *
6  *   This file is part of the Lustre file system, http://www.lustre.org
7  *   Lustre is a trademark of Cluster File Systems, Inc.
8  *
9  *   You may have signed or agreed to another license before downloading
10  *   this software.  If so, you are bound by the terms and conditions
11  *   of that agreement, and the following does not apply to you.  See the
12  *   LICENSE file included with this distribution for more information.
13  *
14  *   If you did not agree to a different license, then this copy of Lustre
15  *   is open source software; you can redistribute it and/or modify it
16  *   under the terms of version 2 of the GNU General Public License as
17  *   published by the Free Software Foundation.
18  *
19  *   In either case, Lustre is distributed in the hope that it will be
20  *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
21  *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
22  *   license text for more details.
23  *
24  */
25 #define DEBUG_SUBSYSTEM S_CLASS
26
27 #include <linux/version.h>
28 #include <lprocfs_status.h>
29 #include <obd.h>
30 #include <linux/seq_file.h>
31 #include <linux/version.h>
32
33 #include "filter_internal.h"
34
35 #ifdef LPROCFS
36 static int lprocfs_filter_rd_groups(char *page, char **start, off_t off,
37                                     int count, int *eof, void *data)
38 {
39         *eof = 1;
40         return snprintf(page, count, "%u\n", FILTER_GROUPS);
41 }
42
43 static int lprocfs_filter_rd_tot_dirty(char *page, char **start, off_t off,
44                                        int count, int *eof, void *data)
45 {
46         struct obd_device *obd = (struct obd_device *)data;
47
48         LASSERT(obd != NULL);
49         *eof = 1;
50         return snprintf(page, count, LPU64"\n", obd->u.filter.fo_tot_dirty);
51 }
52
53 static int lprocfs_filter_rd_tot_granted(char *page, char **start, off_t off,
54                                          int count, int *eof, void *data)
55 {
56         struct obd_device *obd = (struct obd_device *)data;
57
58         LASSERT(obd != NULL);
59         *eof = 1;
60         return snprintf(page, count, LPU64"\n", obd->u.filter.fo_tot_granted);
61 }
62
63 static int lprocfs_filter_rd_tot_pending(char *page, char **start, off_t off,
64                                          int count, int *eof, void *data)
65 {
66         struct obd_device *obd = (struct obd_device *)data;
67
68         LASSERT(obd != NULL);
69         *eof = 1;
70         return snprintf(page, count, LPU64"\n", obd->u.filter.fo_tot_pending);
71 }
72
73 static int lprocfs_filter_rd_mntdev(char *page, char **start, off_t off,
74                                     int count, int *eof, void *data)
75 {
76         struct obd_device *obd = (struct obd_device *)data;
77
78         LASSERT(obd != NULL);
79         LASSERT(obd->u.filter.fo_vfsmnt->mnt_devname);
80         *eof = 1;
81         return snprintf(page, count, "%s\n",
82                         obd->u.filter.fo_vfsmnt->mnt_devname);
83 }
84
85 static int lprocfs_filter_rd_last_id(char *page, char **start, off_t off,
86                                      int count, int *eof, void *data)
87 {
88         struct obd_device *obd = data;
89
90         if (obd == NULL)
91                 return 0;
92
93         return snprintf(page, count, LPU64"\n",
94                         filter_last_id(&obd->u.filter, 0));
95 }
96
97 int lprocfs_filter_rd_readcache(char *page, char **start, off_t off, int count,
98                                 int *eof, void *data)
99 {
100         struct obd_device *obd = data;
101         int rc;
102
103         rc = snprintf(page, count, LPU64"\n",
104                       obd->u.filter.fo_readcache_max_filesize);
105         return rc;
106 }
107
108 int lprocfs_filter_wr_readcache(struct file *file, const char *buffer,
109                                 unsigned long count, void *data)
110 {
111         struct obd_device *obd = data;
112         __u64 val;
113         int rc;
114
115         rc = lprocfs_write_u64_helper(buffer, count, &val);
116         if (rc)
117                 return rc;
118
119         obd->u.filter.fo_readcache_max_filesize = val;
120         return count;
121 }
122
123 int lprocfs_filter_rd_fmd_max_num(char *page, char **start, off_t off,
124                                   int count, int *eof, void *data)
125 {
126         struct obd_device *obd = data;
127         int rc;
128
129         rc = snprintf(page, count, "%u\n", obd->u.filter.fo_fmd_max_num);
130         return rc;
131 }
132
133 int lprocfs_filter_wr_fmd_max_num(struct file *file, const char *buffer,
134                                   unsigned long count, void *data)
135 {
136         struct obd_device *obd = data;
137         int val;
138         int rc;
139
140         rc = lprocfs_write_helper(buffer, count, &val);
141         if (rc)
142                 return rc;
143
144         if (val > 65536 || val < 1)
145                 return -EINVAL;
146
147         obd->u.filter.fo_fmd_max_num = val;
148         return count;
149 }
150
151 int lprocfs_filter_rd_fmd_max_age(char *page, char **start, off_t off,
152                                   int count, int *eof, void *data)
153 {
154         struct obd_device *obd = data;
155         int rc;
156
157         rc = snprintf(page, count, "%u\n", obd->u.filter.fo_fmd_max_age / HZ);
158         return rc;
159 }
160
161 int lprocfs_filter_wr_fmd_max_age(struct file *file, const char *buffer,
162                                   unsigned long count, void *data)
163 {
164         struct obd_device *obd = data;
165         int val;
166         int rc;
167
168         rc = lprocfs_write_helper(buffer, count, &val);
169         if (rc)
170                 return rc;
171
172         if (val > 65536 || val < 1)
173                 return -EINVAL;
174
175         obd->u.filter.fo_fmd_max_age = val * HZ;
176         return count;
177 }
178
179 static struct lprocfs_vars lprocfs_filter_obd_vars[] = {
180         { "uuid",         lprocfs_rd_uuid,          0, 0 },
181         { "blocksize",    lprocfs_rd_blksize,       0, 0 },
182         { "kbytestotal",  lprocfs_rd_kbytestotal,   0, 0 },
183         { "kbytesfree",   lprocfs_rd_kbytesfree,    0, 0 },
184         { "kbytesavail",  lprocfs_rd_kbytesavail,   0, 0 },
185         { "filestotal",   lprocfs_rd_filestotal,    0, 0 },
186         { "filesfree",    lprocfs_rd_filesfree,     0, 0 },
187         { "filegroups",   lprocfs_filter_rd_groups, 0, 0 },
188         { "fstype",       lprocfs_rd_fstype,        0, 0 },
189         { "mntdev",       lprocfs_filter_rd_mntdev, 0, 0 },
190         { "last_id",      lprocfs_filter_rd_last_id,0, 0 },
191         { "tot_dirty",    lprocfs_filter_rd_tot_dirty,   0, 0 },
192         { "tot_pending",  lprocfs_filter_rd_tot_pending, 0, 0 },
193         { "tot_granted",  lprocfs_filter_rd_tot_granted, 0, 0 },
194         { "recovery_status", lprocfs_obd_rd_recovery_status, 0, 0 },
195 #ifdef CRAY_XT3
196         { "recovery_maxtime", lprocfs_obd_rd_recovery_maxtime,
197                               lprocfs_obd_wr_recovery_maxtime, 0},
198 #endif
199         { "evict_client", 0, lprocfs_wr_evict_client, 0,
200                                 &lprocfs_evict_client_fops},
201         { "num_exports",  lprocfs_rd_num_exports,   0, 0 },
202         { "readcache_max_filesize",
203                           lprocfs_filter_rd_readcache,
204                           lprocfs_filter_wr_readcache, 0 },
205 #ifdef HAVE_QUOTA_SUPPORT
206         { "quota_bunit_sz", lprocfs_quota_rd_bunit,
207                             lprocfs_quota_wr_bunit, 0},
208         { "quota_btune_sz", lprocfs_quota_rd_btune,
209                             lprocfs_quota_wr_btune, 0},
210         { "quota_iunit_sz", lprocfs_quota_rd_iunit,
211                             lprocfs_quota_wr_iunit, 0},
212         { "quota_itune_sz", lprocfs_quota_rd_itune,
213                             lprocfs_quota_wr_itune, 0},
214         { "quota_type",     lprocfs_quota_rd_type,
215                             lprocfs_quota_wr_type, 0},
216         { "quota_switch_seconds",  lprocfs_quota_rd_switch_seconds,
217                             lprocfs_quota_wr_switch_seconds, 0 },
218
219 #endif
220         { "client_cache_count", lprocfs_filter_rd_fmd_max_num,
221                           lprocfs_filter_wr_fmd_max_num, 0 },
222         { "client_cache_seconds", lprocfs_filter_rd_fmd_max_age,
223                           lprocfs_filter_wr_fmd_max_age, 0 },
224         { 0 }
225 };
226
227 static struct lprocfs_vars lprocfs_filter_module_vars[] = {
228         { "num_refs",     lprocfs_rd_numrefs,       0, 0 },
229         { 0 }
230 };
231
232 void filter_tally(struct obd_export *exp, struct page **pages, int nr_pages,
233                   unsigned long *blocks, int blocks_per_page, int wr)
234 {
235         struct filter_obd *filter = &exp->exp_obd->u.filter;
236         struct filter_export_data *fed = &exp->exp_filter_data;
237         struct page *last_page = NULL;
238         unsigned long *last_block = NULL;
239         unsigned long discont_pages = 0;
240         unsigned long discont_blocks = 0;
241         int i;
242
243         if (nr_pages == 0)
244                 return;
245
246         lprocfs_oh_tally_log2(&filter->fo_filter_stats.hist[BRW_R_PAGES + wr],
247                               nr_pages);
248         lprocfs_oh_tally_log2(&fed->fed_brw_stats.hist[BRW_R_PAGES + wr],
249                               nr_pages);
250         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_brw_stats)
251                 lprocfs_oh_tally_log2(&exp->exp_nid_stats->nid_brw_stats->hist[BRW_W_PAGES + wr],
252                                       nr_pages);
253
254         while (nr_pages-- > 0) {
255                 if (last_page && (*pages)->index != (last_page->index + 1))
256                         discont_pages++;
257                 last_page = *pages;
258                 pages++;
259                 for (i = 0; i < blocks_per_page; i++) {
260                         if (last_block && *blocks != (*last_block + 1))
261                                 discont_blocks++;
262                         last_block = blocks++;
263                 }
264         }
265
266         lprocfs_oh_tally(&filter->fo_filter_stats.hist[BRW_R_DISCONT_PAGES +wr],
267                          discont_pages);
268         lprocfs_oh_tally(&fed->fed_brw_stats.hist[BRW_R_DISCONT_PAGES + wr],
269                          discont_pages);
270         lprocfs_oh_tally(&filter->fo_filter_stats.hist[BRW_R_DISCONT_BLOCKS+wr],
271                          discont_blocks);
272         lprocfs_oh_tally(&fed->fed_brw_stats.hist[BRW_R_DISCONT_BLOCKS + wr],
273                          discont_blocks);
274
275         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_brw_stats) {
276                 lprocfs_oh_tally_log2(&exp->exp_nid_stats->nid_brw_stats->hist[BRW_W_DISCONT_PAGES + wr],
277                                       discont_pages);
278                 lprocfs_oh_tally_log2(&exp->exp_nid_stats->nid_brw_stats->hist[BRW_W_DISCONT_BLOCKS + wr],
279                                       discont_blocks);
280         }
281 }
282
283 #define pct(a,b) (b ? a * 100 / b : 0)
284
285 static void display_brw_stats(struct seq_file *seq, char *name, char *units,
286         struct obd_histogram *read, struct obd_histogram *write, int log2)
287 {
288         unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
289         int i;
290
291         seq_printf(seq, "\n%26s read      |     write\n", " ");
292         seq_printf(seq, "%-22s %-5s %% cum %% |  %-5s %% cum %%\n", 
293                    name, units, units);
294
295         read_tot = lprocfs_oh_sum(read);
296         write_tot = lprocfs_oh_sum(write);
297         for (i = 0; i < OBD_HIST_MAX; i++) {
298                 r = read->oh_buckets[i];
299                 w = write->oh_buckets[i];
300                 read_cum += r;
301                 write_cum += w;
302                 if (read_cum == 0 && write_cum == 0)
303                         continue;
304
305                 if (!log2) 
306                         seq_printf(seq, "%u", i);
307                 else if (i < 10)
308                         seq_printf(seq, "%u", 1<<i);
309                 else if (i < 20)
310                         seq_printf(seq, "%uK", 1<<(i-10));
311                 else
312                         seq_printf(seq, "%uM", 1<<(i-20));
313
314                 seq_printf(seq, ":\t\t%10lu %3lu %3lu   | %4lu %3lu %3lu\n",
315                            r, pct(r, read_tot), pct(read_cum, read_tot), 
316                            w, pct(w, write_tot), pct(write_cum, write_tot));
317
318                 if (read_cum == read_tot && write_cum == write_tot)
319                         break;
320         }
321 }
322
323 static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
324 {
325         struct timeval now;
326
327         /* this sampling races with updates */
328         do_gettimeofday(&now);
329         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
330                    now.tv_sec, now.tv_usec);
331
332         display_brw_stats(seq, "pages per bulk r/w", "rpcs",
333                           &brw_stats->hist[BRW_R_PAGES],
334                           &brw_stats->hist[BRW_W_PAGES], 1);
335
336         display_brw_stats(seq, "discontiguous pages", "rpcs",
337                           &brw_stats->hist[BRW_R_DISCONT_PAGES],
338                           &brw_stats->hist[BRW_W_DISCONT_PAGES], 0);
339
340         display_brw_stats(seq, "discontiguous blocks", "rpcs",
341                           &brw_stats->hist[BRW_R_DISCONT_BLOCKS],
342                           &brw_stats->hist[BRW_W_DISCONT_BLOCKS], 0);
343
344         display_brw_stats(seq, "disk fragmented I/Os", "ios",
345                           &brw_stats->hist[BRW_R_DIO_FRAGS],
346                           &brw_stats->hist[BRW_W_DIO_FRAGS], 0);
347
348 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
349         display_brw_stats(seq, "disk I/Os in flight", "ios",
350                           &brw_stats->hist[BRW_R_RPC_HIST],
351                           &brw_stats->hist[BRW_W_RPC_HIST], 0);
352
353         {
354                 char title[24];
355                 sprintf(title, "I/O time (1/%ds)", HZ);
356                 display_brw_stats(seq, title, "ios",
357                                   &brw_stats->hist[BRW_R_IO_TIME],
358                                   &brw_stats->hist[BRW_W_IO_TIME], 1);
359         }
360
361         display_brw_stats(seq, "disk I/O size", "ios",
362                           &brw_stats->hist[BRW_R_DISK_IOSIZE],
363                           &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
364 #endif
365 }
366
367 #undef pct
368
369 static int filter_brw_stats_seq_show(struct seq_file *seq, void *v)
370 {
371         struct obd_device *dev = seq->private;
372         struct filter_obd *filter = &dev->u.filter;
373
374         brw_stats_show(seq, &filter->fo_filter_stats);
375
376         return 0;
377 }
378
379 static ssize_t filter_brw_stats_seq_write(struct file *file, const char *buf,
380                                        size_t len, loff_t *off)
381 {
382         struct seq_file *seq = file->private_data;
383         struct obd_device *dev = seq->private;
384         struct filter_obd *filter = &dev->u.filter;
385         int i;
386
387         for (i = 0; i < BRW_LAST; i++)
388                 lprocfs_oh_clear(&filter->fo_filter_stats.hist[i]);
389
390         return len;
391 }
392
393 LPROC_SEQ_FOPS(filter_brw_stats);
394
395 int lproc_filter_attach_seqstat(struct obd_device *dev)
396 {
397         return lprocfs_obd_seq_create(dev, "brw_stats", 0444,
398                                       &filter_brw_stats_fops, dev);
399 }
400
401 static int filter_per_export_stats_seq_show(struct seq_file *seq, void *v)
402 {
403         struct filter_export_data *fed = seq->private;
404
405         brw_stats_show(seq, &fed->fed_brw_stats);
406
407         return 0;
408 }
409
410 static ssize_t filter_per_export_stats_seq_write(struct file *file,
411                                        const char *buf, size_t len, loff_t *off)
412 {
413         struct seq_file *seq = file->private_data;
414         struct filter_export_data *fed = seq->private;
415         int i;
416
417         for (i = 0; i < BRW_LAST; i++)
418                 lprocfs_oh_clear(&fed->fed_brw_stats.hist[i]);
419
420         return len;
421 }
422
423 LPROC_SEQ_FOPS(filter_per_export_stats);
424
425 void lprocfs_filter_init_vars(struct lprocfs_static_vars *lvars)
426 {
427     lvars->module_vars  = lprocfs_filter_module_vars;
428     lvars->obd_vars     = lprocfs_filter_obd_vars;
429 }
430
431 static int filter_per_nid_stats_seq_show(struct seq_file *seq, void *v)
432 {
433         nid_stat_t *tmp = seq->private;
434
435         if (tmp->nid_brw_stats)
436                 brw_stats_show(seq, tmp->nid_brw_stats);
437
438         return 0;
439 }
440
441 static ssize_t filter_per_nid_stats_seq_write(struct file *file,
442                                               const char *buf, size_t len,
443                                               loff_t *off)
444 {
445         struct seq_file *seq = file->private_data;
446         nid_stat_t *tmp = seq->private;
447         int i;
448
449         if (tmp->nid_brw_stats)
450                 for (i = 0; i < BRW_LAST; i++)
451                         lprocfs_oh_clear(&tmp->nid_brw_stats->hist[i]);
452
453         return len;
454 }
455
456 LPROC_SEQ_FOPS(filter_per_nid_stats);
457
458 #endif /* LPROCFS */