Whamcloud - gitweb
LU-10092 llite: Add persistent cache on client
[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.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) 2012, 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 #define DEBUG_SUBSYSTEM S_LLITE
33
34 #include <linux/version.h>
35 #include <linux/user_namespace.h>
36 #ifdef HAVE_UIDGID_HEADER
37 # include <linux/uidgid.h>
38 #endif
39 #include <uapi/linux/lustre/lustre_param.h>
40 #include <lprocfs_status.h>
41 #include <obd_support.h>
42
43 #include "llite_internal.h"
44 #include "vvp_internal.h"
45
46 static struct kobject *llite_kobj;
47 static struct dentry *llite_root;
48
49 static void llite_kobj_release(struct kobject *kobj)
50 {
51         if (!IS_ERR_OR_NULL(llite_root)) {
52                 debugfs_remove(llite_root);
53                 llite_root = NULL;
54         }
55
56         kfree(kobj);
57 }
58
59 static struct kobj_type llite_kobj_ktype = {
60         .release        = llite_kobj_release,
61         .sysfs_ops      = &lustre_sysfs_ops,
62 };
63
64 int llite_tunables_register(void)
65 {
66         int rc;
67
68         llite_kobj = kzalloc(sizeof(*llite_kobj), GFP_KERNEL);
69         if (!llite_kobj)
70                 return -ENOMEM;
71
72         llite_kobj->kset = lustre_kset;
73         rc = kobject_init_and_add(llite_kobj, &llite_kobj_ktype,
74                                   &lustre_kset->kobj, "%s", "llite");
75         if (rc)
76                 goto free_kobj;
77
78         llite_root = debugfs_create_dir("llite", debugfs_lustre_root);
79         if (IS_ERR_OR_NULL(llite_root)) {
80                 rc = llite_root ? PTR_ERR(llite_root) : -ENOMEM;
81                 llite_root = NULL;
82 free_kobj:
83                 kobject_put(llite_kobj);
84                 llite_kobj = NULL;
85         }
86
87         return rc;
88 }
89
90 void llite_tunables_unregister(void)
91 {
92         kobject_put(llite_kobj);
93         llite_kobj = NULL;
94 }
95
96 /* <debugfs>/lustre/llite mount point registration */
97 static const struct file_operations ll_rw_extents_stats_fops;
98 static const struct file_operations ll_rw_extents_stats_pp_fops;
99 static const struct file_operations ll_rw_offset_stats_fops;
100
101 /**
102  * ll_stats_pid_write() - Determine if stats collection should be enabled
103  * @buf: Buffer containing the data written
104  * @len: Number of bytes in the buffer
105  *
106  * Several proc files begin collecting stats when a value is written, and stop
107  * collecting when either '0' or 'disable' is written. This function checks the
108  * written value to see if collection should be enabled or disabled.
109  *
110  * Return: If '0' or 'disable' is provided, 0 is returned. If the text
111  * equivalent of a number is written, that number is returned. Otherwise,
112  * 1 is returned. Non-zero return values indicate collection should be enabled.
113  */
114 static s64 ll_stats_pid_write(const char __user *buf, size_t len)
115 {
116         unsigned long long value = 1;
117         char kernbuf[16];
118         int rc;
119
120         rc = kstrtoull_from_user(buf, len, 0, &value);
121         if (rc < 0 && len < sizeof(kernbuf)) {
122                 if (copy_from_user(kernbuf, buf, len))
123                         return -EFAULT;
124                 kernbuf[len] = 0;
125
126                 if (kernbuf[len - 1] == '\n')
127                         kernbuf[len - 1] = 0;
128
129                 if (strncasecmp(kernbuf, "disable", 7) == 0)
130                         value = 0;
131         }
132
133         return value;
134 }
135
136 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
137                               char *buf)
138 {
139         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
140                                               ll_kset.kobj);
141         struct obd_statfs osfs;
142         int rc;
143
144         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
145         if (rc)
146                 return rc;
147
148         return sprintf(buf, "%u\n", osfs.os_bsize);
149 }
150 LUSTRE_RO_ATTR(blocksize);
151
152 static ssize_t stat_blocksize_show(struct kobject *kobj, struct attribute *attr,
153                                    char *buf)
154 {
155         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
156                                               ll_kset.kobj);
157
158         return sprintf(buf, "%u\n", sbi->ll_stat_blksize);
159 }
160
161 static ssize_t stat_blocksize_store(struct kobject *kobj,
162                                     struct attribute *attr,
163                                     const char *buffer,
164                                     size_t count)
165 {
166         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
167                                               ll_kset.kobj);
168         unsigned int val;
169         int rc;
170
171         rc = kstrtouint(buffer, 10, &val);
172         if (rc)
173                 return rc;
174
175         if (val != 0 && (val < PAGE_SIZE || (val & (val - 1))) != 0)
176                 return -ERANGE;
177
178         sbi->ll_stat_blksize = val;
179
180         return count;
181 }
182 LUSTRE_RW_ATTR(stat_blocksize);
183
184 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
185                                 char *buf)
186 {
187         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
188                                               ll_kset.kobj);
189         struct obd_statfs osfs;
190         u32 blk_size;
191         u64 result;
192         int rc;
193
194         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
195         if (rc)
196                 return rc;
197
198         blk_size = osfs.os_bsize >> 10;
199         result = osfs.os_blocks;
200
201         while (blk_size >>= 1)
202                 result <<= 1;
203
204         return sprintf(buf, "%llu\n", result);
205 }
206 LUSTRE_RO_ATTR(kbytestotal);
207
208 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
209                                char *buf)
210 {
211         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
212                                               ll_kset.kobj);
213         struct obd_statfs osfs;
214         u32 blk_size;
215         u64 result;
216         int rc;
217
218         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
219         if (rc)
220                 return rc;
221
222         blk_size = osfs.os_bsize >> 10;
223         result = osfs.os_bfree;
224
225         while (blk_size >>= 1)
226                 result <<= 1;
227
228         return sprintf(buf, "%llu\n", result);
229 }
230 LUSTRE_RO_ATTR(kbytesfree);
231
232 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
233                                 char *buf)
234 {
235         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
236                                               ll_kset.kobj);
237         struct obd_statfs osfs;
238         u32 blk_size;
239         u64 result;
240         int rc;
241
242         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
243         if (rc)
244                 return rc;
245
246         blk_size = osfs.os_bsize >> 10;
247         result = osfs.os_bavail;
248
249         while (blk_size >>= 1)
250                 result <<= 1;
251
252         return sprintf(buf, "%llu\n", result);
253 }
254 LUSTRE_RO_ATTR(kbytesavail);
255
256 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
257                                char *buf)
258 {
259         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
260                                               ll_kset.kobj);
261         struct obd_statfs osfs;
262         int rc;
263
264         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
265         if (rc)
266                 return rc;
267
268         return sprintf(buf, "%llu\n", osfs.os_files);
269 }
270 LUSTRE_RO_ATTR(filestotal);
271
272 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
273                               char *buf)
274 {
275         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
276                                               ll_kset.kobj);
277         struct obd_statfs osfs;
278         int rc;
279
280         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
281         if (rc)
282                 return rc;
283
284         return sprintf(buf, "%llu\n", osfs.os_ffree);
285 }
286 LUSTRE_RO_ATTR(filesfree);
287
288 static ssize_t client_type_show(struct kobject *kobj, struct attribute *attr,
289                                 char *buf)
290 {
291         return sprintf(buf, "local client\n");
292 }
293 LUSTRE_RO_ATTR(client_type);
294
295 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
296                            char *buf)
297 {
298         return sprintf(buf, "lustre\n");
299 }
300 LUSTRE_RO_ATTR(fstype);
301
302 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
303                          char *buf)
304 {
305         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
306                                               ll_kset.kobj);
307
308         return sprintf(buf, "%s\n", sbi->ll_sb_uuid.uuid);
309 }
310 LUSTRE_RO_ATTR(uuid);
311
312 static int ll_site_stats_seq_show(struct seq_file *m, void *v)
313 {
314         struct super_block *sb = m->private;
315
316         /*
317          * See description of statistical counters in struct cl_site, and
318          * struct lu_site.
319          */
320         return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site), m);
321 }
322
323 LDEBUGFS_SEQ_FOPS_RO(ll_site_stats);
324
325 static int ll_max_readahead_mb_seq_show(struct seq_file *m, void *v)
326 {
327         struct super_block *sb = m->private;
328         struct ll_sb_info *sbi = ll_s2sbi(sb);
329         unsigned long ra_max_mb;
330
331         spin_lock(&sbi->ll_lock);
332         ra_max_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages);
333         spin_unlock(&sbi->ll_lock);
334
335         seq_printf(m, "%lu\n", ra_max_mb);
336         return 0;
337 }
338
339 static ssize_t
340 ll_max_readahead_mb_seq_write(struct file *file, const char __user *buffer,
341                               size_t count, loff_t *off)
342 {
343         struct seq_file *m = file->private_data;
344         struct super_block *sb = m->private;
345         struct ll_sb_info *sbi = ll_s2sbi(sb);
346         s64 ra_max_mb, pages_number;
347         int rc;
348
349         rc = lprocfs_str_with_units_to_s64(buffer, count, &ra_max_mb, 'M');
350         if (rc)
351                 return rc;
352
353         pages_number = round_up(ra_max_mb, 1024 * 1024) >> PAGE_SHIFT;
354         if (pages_number < 0 || pages_number > totalram_pages / 2) {
355                 /* 1/2 of RAM */
356                 CERROR("%s: can't set max_readahead_mb=%llu > %luMB\n",
357                        sbi->ll_fsname, PAGES_TO_MiB(pages_number),
358                        PAGES_TO_MiB(totalram_pages));
359                 return -ERANGE;
360         }
361
362         spin_lock(&sbi->ll_lock);
363         sbi->ll_ra_info.ra_max_pages = pages_number;
364         spin_unlock(&sbi->ll_lock);
365
366         return count;
367 }
368
369 LDEBUGFS_SEQ_FOPS(ll_max_readahead_mb);
370
371 static int ll_max_readahead_per_file_mb_seq_show(struct seq_file *m, void *v)
372 {
373         struct super_block *sb = m->private;
374         struct ll_sb_info *sbi = ll_s2sbi(sb);
375         unsigned long ra_max_file_mb;
376
377         spin_lock(&sbi->ll_lock);
378         ra_max_file_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file);
379         spin_unlock(&sbi->ll_lock);
380
381         seq_printf(m, "%lu\n", ra_max_file_mb);
382         return 0;
383 }
384
385 static ssize_t
386 ll_max_readahead_per_file_mb_seq_write(struct file *file,
387                                        const char __user *buffer,
388                                        size_t count, loff_t *off)
389 {
390         struct seq_file *m = file->private_data;
391         struct super_block *sb = m->private;
392         struct ll_sb_info *sbi = ll_s2sbi(sb);
393         s64 ra_max_file_mb, pages_number;
394         int rc;
395
396         rc = lprocfs_str_with_units_to_s64(buffer, count, &ra_max_file_mb,
397                                            'M');
398         if (rc)
399                 return rc;
400
401         pages_number = round_up(ra_max_file_mb, 1024 * 1024) >> PAGE_SHIFT;
402         if (pages_number < 0 || pages_number > sbi->ll_ra_info.ra_max_pages) {
403                 CERROR("%s: can't set max_readahead_per_file_mb=%llu > max_read_ahead_mb=%lu\n",
404                        sbi->ll_fsname, PAGES_TO_MiB(pages_number),
405                        PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
406                 return -ERANGE;
407         }
408
409         spin_lock(&sbi->ll_lock);
410         sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
411         spin_unlock(&sbi->ll_lock);
412
413         return count;
414 }
415
416 LDEBUGFS_SEQ_FOPS(ll_max_readahead_per_file_mb);
417
418 static int ll_max_read_ahead_whole_mb_seq_show(struct seq_file *m, void *v)
419 {
420         struct super_block *sb = m->private;
421         struct ll_sb_info *sbi = ll_s2sbi(sb);
422         unsigned long ra_max_whole_mb;
423
424         spin_lock(&sbi->ll_lock);
425         ra_max_whole_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_read_ahead_whole_pages);
426         spin_unlock(&sbi->ll_lock);
427
428         seq_printf(m, "%lu\n", ra_max_whole_mb);
429         return 0;
430 }
431
432 static ssize_t
433 ll_max_read_ahead_whole_mb_seq_write(struct file *file,
434                                      const char __user *buffer,
435                                      size_t count, loff_t *off)
436 {
437         struct seq_file *m = file->private_data;
438         struct super_block *sb = m->private;
439         struct ll_sb_info *sbi = ll_s2sbi(sb);
440         s64 ra_max_whole_mb, pages_number;
441         int rc;
442
443         rc = lprocfs_str_with_units_to_s64(buffer, count, &ra_max_whole_mb,
444                                            'M');
445         if (rc)
446                 return rc;
447
448         pages_number = round_up(ra_max_whole_mb, 1024 * 1024) >> PAGE_SHIFT;
449         /* Cap this at the current max readahead window size, the readahead
450          * algorithm does this anyway so it's pointless to set it larger.
451          */
452         if (pages_number < 0 ||
453             pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
454                 CERROR("%s: can't set max_read_ahead_whole_mb=%llu > max_read_ahead_per_file_mb=%lu\n",
455                        sbi->ll_fsname, PAGES_TO_MiB(pages_number),
456                        PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
457                 return -ERANGE;
458         }
459
460         spin_lock(&sbi->ll_lock);
461         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
462         spin_unlock(&sbi->ll_lock);
463
464         return count;
465 }
466
467 LDEBUGFS_SEQ_FOPS(ll_max_read_ahead_whole_mb);
468
469 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
470 {
471         struct super_block     *sb    = m->private;
472         struct ll_sb_info      *sbi   = ll_s2sbi(sb);
473         struct cl_client_cache *cache = sbi->ll_cache;
474         long max_cached_mb;
475         long unused_mb;
476
477         max_cached_mb = PAGES_TO_MiB(cache->ccc_lru_max);
478         unused_mb = PAGES_TO_MiB(atomic_long_read(&cache->ccc_lru_left));
479         seq_printf(m, "users: %d\n"
480                       "max_cached_mb: %ld\n"
481                       "used_mb: %ld\n"
482                       "unused_mb: %ld\n"
483                       "reclaim_count: %u\n",
484                    atomic_read(&cache->ccc_users),
485                    max_cached_mb,
486                    max_cached_mb - unused_mb,
487                    unused_mb,
488                    cache->ccc_lru_shrinkers);
489         return 0;
490 }
491
492 static ssize_t ll_max_cached_mb_seq_write(struct file *file,
493                                           const char __user *buffer,
494                                           size_t count, loff_t *off)
495 {
496         struct seq_file *m = file->private_data;
497         struct super_block *sb = m->private;
498         struct ll_sb_info *sbi = ll_s2sbi(sb);
499         struct cl_client_cache *cache = sbi->ll_cache;
500         struct lu_env *env;
501         long diff = 0;
502         long nrpages = 0;
503         __u16 refcheck;
504         __s64 pages_number;
505         int rc;
506         char kernbuf[128];
507
508         ENTRY;
509         if (count >= sizeof(kernbuf))
510                 RETURN(-EINVAL);
511
512         if (copy_from_user(kernbuf, buffer, count))
513                 RETURN(-EFAULT);
514         kernbuf[count] = 0;
515
516         buffer += lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count) -
517                   kernbuf;
518         rc = lprocfs_str_with_units_to_s64(buffer, count, &pages_number, 'M');
519         if (rc)
520                 RETURN(rc);
521
522         pages_number >>= PAGE_SHIFT;
523
524         if (pages_number < 0 || pages_number > totalram_pages) {
525                 CERROR("%s: can't set max cache more than %lu MB\n",
526                        sbi->ll_fsname,
527                        PAGES_TO_MiB(totalram_pages));
528                 RETURN(-ERANGE);
529         }
530         /* Allow enough cache so clients can make well-formed RPCs */
531         pages_number = max_t(long, pages_number, PTLRPC_MAX_BRW_PAGES);
532
533         spin_lock(&sbi->ll_lock);
534         diff = pages_number - cache->ccc_lru_max;
535         spin_unlock(&sbi->ll_lock);
536
537         /* easy - add more LRU slots. */
538         if (diff >= 0) {
539                 atomic_long_add(diff, &cache->ccc_lru_left);
540                 GOTO(out, rc = 0);
541         }
542
543         env = cl_env_get(&refcheck);
544         if (IS_ERR(env))
545                 RETURN(PTR_ERR(env));
546
547         diff = -diff;
548         while (diff > 0) {
549                 long tmp;
550
551                 /* reduce LRU budget from free slots. */
552                 do {
553                         long ov, nv;
554
555                         ov = atomic_long_read(&cache->ccc_lru_left);
556                         if (ov == 0)
557                                 break;
558
559                         nv = ov > diff ? ov - diff : 0;
560                         rc = atomic_long_cmpxchg(&cache->ccc_lru_left, ov, nv);
561                         if (likely(ov == rc)) {
562                                 diff -= ov - nv;
563                                 nrpages += ov - nv;
564                                 break;
565                         }
566                 } while (1);
567
568                 if (diff <= 0)
569                         break;
570
571                 if (sbi->ll_dt_exp == NULL) { /* being initialized */
572                         rc = -ENODEV;
573                         break;
574                 }
575
576                 /* difficult - have to ask OSCs to drop LRU slots. */
577                 tmp = diff << 1;
578                 rc = obd_set_info_async(env, sbi->ll_dt_exp,
579                                 sizeof(KEY_CACHE_LRU_SHRINK),
580                                 KEY_CACHE_LRU_SHRINK,
581                                 sizeof(tmp), &tmp, NULL);
582                 if (rc < 0)
583                         break;
584         }
585         cl_env_put(env, &refcheck);
586
587 out:
588         if (rc >= 0) {
589                 spin_lock(&sbi->ll_lock);
590                 cache->ccc_lru_max = pages_number;
591                 spin_unlock(&sbi->ll_lock);
592                 rc = count;
593         } else {
594                 atomic_long_add(nrpages, &cache->ccc_lru_left);
595         }
596         return rc;
597 }
598
599 LDEBUGFS_SEQ_FOPS(ll_max_cached_mb);
600
601 static ssize_t checksums_show(struct kobject *kobj, struct attribute *attr,
602                               char *buf)
603 {
604         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
605                                               ll_kset.kobj);
606
607         return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
608 }
609
610 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
611                                const char *buffer, size_t count)
612 {
613         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
614                                               ll_kset.kobj);
615         bool val;
616         int tmp;
617         int rc;
618
619         if (!sbi->ll_dt_exp)
620                 /* Not set up yet */
621                 return -EAGAIN;
622
623         rc = kstrtobool(buffer, &val);
624         if (rc)
625                 return rc;
626         if (val)
627                 sbi->ll_flags |= LL_SBI_CHECKSUM;
628         else
629                 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
630         tmp = val;
631
632         rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
633                                 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
634         if (rc)
635                 CWARN("Failed to set OSC checksum flags: %d\n", rc);
636
637         return count;
638 }
639 LUSTRE_RW_ATTR(checksums);
640
641 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
642
643 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
644                               enum stats_track_type type)
645 {
646         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
647                                               ll_kset.kobj);
648
649         if (sbi->ll_stats_track_type == type)
650                 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
651         else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
652                 return sprintf(buf, "0 (all)\n");
653
654         return sprintf(buf, "untracked\n");
655 }
656
657 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
658                               size_t count, enum stats_track_type type)
659 {
660         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
661                                               ll_kset.kobj);
662         unsigned long pid;
663         int rc;
664
665         rc = kstrtoul(buffer, 10, &pid);
666         if (rc)
667                 return rc;
668
669         sbi->ll_stats_track_id = pid;
670         if (pid == 0)
671                 sbi->ll_stats_track_type = STATS_TRACK_ALL;
672         else
673                 sbi->ll_stats_track_type = type;
674         lprocfs_clear_stats(sbi->ll_stats);
675         return count;
676 }
677
678 static ssize_t stats_track_pid_show(struct kobject *kobj,
679                                     struct attribute *attr,
680                                     char *buf)
681 {
682         return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
683 }
684
685 static ssize_t stats_track_pid_store(struct kobject *kobj,
686                                      struct attribute *attr,
687                                      const char *buffer,
688                                      size_t count)
689 {
690         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
691 }
692 LUSTRE_RW_ATTR(stats_track_pid);
693
694 static ssize_t stats_track_ppid_show(struct kobject *kobj,
695                                      struct attribute *attr,
696                                      char *buf)
697 {
698         return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
699 }
700
701 static ssize_t stats_track_ppid_store(struct kobject *kobj,
702                                       struct attribute *attr,
703                                       const char *buffer,
704                                       size_t count)
705 {
706         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
707 }
708 LUSTRE_RW_ATTR(stats_track_ppid);
709
710 static ssize_t stats_track_gid_show(struct kobject *kobj,
711                                     struct attribute *attr,
712                                     char *buf)
713 {
714         return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
715 }
716
717 static ssize_t stats_track_gid_store(struct kobject *kobj,
718                                      struct attribute *attr,
719                                      const char *buffer,
720                                      size_t count)
721 {
722         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
723 }
724 LUSTRE_RW_ATTR(stats_track_gid);
725
726 static ssize_t statahead_running_max_show(struct kobject *kobj,
727                                           struct attribute *attr,
728                                           char *buf)
729 {
730         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
731                                               ll_kset.kobj);
732
733         return snprintf(buf, 16, "%u\n", sbi->ll_sa_running_max);
734 }
735
736 static ssize_t statahead_running_max_store(struct kobject *kobj,
737                                            struct attribute *attr,
738                                            const char *buffer,
739                                            size_t count)
740 {
741         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
742                                               ll_kset.kobj);
743         unsigned long val;
744         int rc;
745
746         rc = kstrtoul(buffer, 0, &val);
747         if (rc)
748                 return rc;
749
750         if (val <= LL_SA_RUNNING_MAX) {
751                 sbi->ll_sa_running_max = val;
752                 return count;
753         }
754
755         CERROR("Bad statahead_running_max value %lu. Valid values "
756                "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
757
758         return -ERANGE;
759 }
760 LUSTRE_RW_ATTR(statahead_running_max);
761
762 static ssize_t statahead_max_show(struct kobject *kobj,
763                                   struct attribute *attr,
764                                   char *buf)
765 {
766         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
767                                               ll_kset.kobj);
768
769         return sprintf(buf, "%u\n", sbi->ll_sa_max);
770 }
771
772 static ssize_t statahead_max_store(struct kobject *kobj,
773                                    struct attribute *attr,
774                                    const char *buffer,
775                                    size_t count)
776 {
777         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
778                                               ll_kset.kobj);
779         unsigned long val;
780         int rc;
781
782         rc = kstrtoul(buffer, 0, &val);
783         if (rc)
784                 return rc;
785
786         if (val <= LL_SA_RPC_MAX)
787                 sbi->ll_sa_max = val;
788         else
789                 CERROR("Bad statahead_max value %lu. Valid values are in the range [0, %d]\n",
790                        val, LL_SA_RPC_MAX);
791
792         return count;
793 }
794 LUSTRE_RW_ATTR(statahead_max);
795
796 static ssize_t statahead_agl_show(struct kobject *kobj,
797                                   struct attribute *attr,
798                                   char *buf)
799 {
800         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
801                                               ll_kset.kobj);
802
803         return sprintf(buf, "%u\n", sbi->ll_flags & LL_SBI_AGL_ENABLED ? 1 : 0);
804 }
805
806 static ssize_t statahead_agl_store(struct kobject *kobj,
807                                    struct attribute *attr,
808                                    const char *buffer,
809                                    size_t count)
810 {
811         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
812                                               ll_kset.kobj);
813         bool val;
814         int rc;
815
816         rc = kstrtobool(buffer, &val);
817         if (rc)
818                 return rc;
819
820         if (val)
821                 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
822         else
823                 sbi->ll_flags &= ~LL_SBI_AGL_ENABLED;
824
825         return count;
826 }
827 LUSTRE_RW_ATTR(statahead_agl);
828
829 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
830 {
831         struct super_block *sb = m->private;
832         struct ll_sb_info *sbi = ll_s2sbi(sb);
833
834         seq_printf(m, "statahead total: %u\n"
835                       "statahead wrong: %u\n"
836                       "agl total: %u\n",
837                    atomic_read(&sbi->ll_sa_total),
838                    atomic_read(&sbi->ll_sa_wrong),
839                    atomic_read(&sbi->ll_agl_total));
840         return 0;
841 }
842
843 LDEBUGFS_SEQ_FOPS_RO(ll_statahead_stats);
844
845 static ssize_t lazystatfs_show(struct kobject *kobj,
846                                struct attribute *attr,
847                                char *buf)
848 {
849         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
850                                               ll_kset.kobj);
851
852         return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
853 }
854
855 static ssize_t lazystatfs_store(struct kobject *kobj,
856                                 struct attribute *attr,
857                                 const char *buffer,
858                                 size_t count)
859 {
860         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
861                                               ll_kset.kobj);
862         bool val;
863         int rc;
864
865         rc = kstrtobool(buffer, &val);
866         if (rc)
867                 return rc;
868
869         if (val)
870                 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
871         else
872                 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
873
874         return count;
875 }
876 LUSTRE_RW_ATTR(lazystatfs);
877
878 static ssize_t max_easize_show(struct kobject *kobj,
879                                struct attribute *attr,
880                                char *buf)
881 {
882         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
883                                               ll_kset.kobj);
884         unsigned int ealen;
885         int rc;
886
887         rc = ll_get_max_mdsize(sbi, &ealen);
888         if (rc)
889                 return rc;
890
891         return sprintf(buf, "%u\n", ealen);
892 }
893 LUSTRE_RO_ATTR(max_easize);
894
895 /**
896  * Get default_easize.
897  *
898  * \see client_obd::cl_default_mds_easize
899  *
900  * \param[in] m         seq_file handle
901  * \param[in] v         unused for single entry
902  *
903  * \retval 0            on success
904  * \retval negative     negated errno on failure
905  */
906 static ssize_t default_easize_show(struct kobject *kobj,
907                                    struct attribute *attr,
908                                    char *buf)
909 {
910         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
911                                               ll_kset.kobj);
912         unsigned int ealen;
913         int rc;
914
915         rc = ll_get_default_mdsize(sbi, &ealen);
916         if (rc)
917                 return rc;
918
919         return sprintf(buf, "%u\n", ealen);
920 }
921
922 /**
923  * Set default_easize.
924  *
925  * Range checking on the passed value is handled by
926  * ll_set_default_mdsize().
927  *
928  * \see client_obd::cl_default_mds_easize
929  *
930  * \param[in] file      proc file
931  * \param[in] buffer    string passed from user space
932  * \param[in] count     \a buffer length
933  * \param[in] off       unused for single entry
934  *
935  * \retval positive     \a count on success
936  * \retval negative     negated errno on failure
937  */
938 static ssize_t default_easize_store(struct kobject *kobj,
939                                     struct attribute *attr,
940                                     const char *buffer,
941                                     size_t count)
942 {
943         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
944                                               ll_kset.kobj);
945         unsigned int val;
946         int rc;
947
948         if (count == 0)
949                 return 0;
950
951         rc = kstrtouint(buffer, 10, &val);
952         if (rc)
953                 return rc;
954
955         rc = ll_set_default_mdsize(sbi, val);
956         if (rc)
957                 return rc;
958
959         return count;
960 }
961 LUSTRE_RW_ATTR(default_easize);
962
963 static int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
964 {
965         const char *str[] = LL_SBI_FLAGS;
966         struct super_block *sb = m->private;
967         int flags = ll_s2sbi(sb)->ll_flags;
968         int i = 0;
969
970         while (flags != 0) {
971                 if (ARRAY_SIZE(str) <= i) {
972                         CERROR("%s: Revise array LL_SBI_FLAGS to match sbi "
973                                 "flags please.\n", ll_s2sbi(sb)->ll_fsname);
974                         return -EINVAL;
975                 }
976
977                 if (flags & 0x1)
978                         seq_printf(m, "%s ", str[i]);
979                 flags >>= 1;
980                 ++i;
981         }
982         seq_printf(m, "\b\n");
983         return 0;
984 }
985
986 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
987
988 static ssize_t xattr_cache_show(struct kobject *kobj,
989                                 struct attribute *attr,
990                                 char *buf)
991 {
992         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
993                                               ll_kset.kobj);
994
995         return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
996 }
997
998 static ssize_t xattr_cache_store(struct kobject *kobj,
999                                  struct attribute *attr,
1000                                  const char *buffer,
1001                                  size_t count)
1002 {
1003         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1004                                               ll_kset.kobj);
1005         bool val;
1006         int rc;
1007
1008         rc = kstrtobool(buffer, &val);
1009         if (rc)
1010                 return rc;
1011
1012         if (val && !(sbi->ll_flags & LL_SBI_XATTR_CACHE))
1013                 return -ENOTSUPP;
1014
1015         sbi->ll_xattr_cache_enabled = val;
1016         sbi->ll_xattr_cache_set = 1;
1017
1018         return count;
1019 }
1020 LUSTRE_RW_ATTR(xattr_cache);
1021
1022 static ssize_t tiny_write_show(struct kobject *kobj,
1023                                struct attribute *attr,
1024                                char *buf)
1025 {
1026         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1027                                               ll_kset.kobj);
1028
1029         return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_TINY_WRITE));
1030 }
1031
1032 static ssize_t tiny_write_store(struct kobject *kobj,
1033                                 struct attribute *attr,
1034                                 const char *buffer,
1035                                 size_t count)
1036 {
1037         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1038                                               ll_kset.kobj);
1039         bool val;
1040         int rc;
1041
1042         rc = kstrtobool(buffer, &val);
1043         if (rc)
1044                 return rc;
1045
1046         spin_lock(&sbi->ll_lock);
1047         if (val)
1048                 sbi->ll_flags |= LL_SBI_TINY_WRITE;
1049         else
1050                 sbi->ll_flags &= ~LL_SBI_TINY_WRITE;
1051         spin_unlock(&sbi->ll_lock);
1052
1053         return count;
1054 }
1055 LUSTRE_RW_ATTR(tiny_write);
1056
1057 static ssize_t fast_read_show(struct kobject *kobj,
1058                               struct attribute *attr,
1059                               char *buf)
1060 {
1061         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1062                                               ll_kset.kobj);
1063
1064         return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_FAST_READ));
1065 }
1066
1067 static ssize_t fast_read_store(struct kobject *kobj,
1068                                struct attribute *attr,
1069                                const char *buffer,
1070                                size_t count)
1071 {
1072         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1073                                               ll_kset.kobj);
1074         bool val;
1075         int rc;
1076
1077         rc = kstrtobool(buffer, &val);
1078         if (rc)
1079                 return rc;
1080
1081         spin_lock(&sbi->ll_lock);
1082         if (val)
1083                 sbi->ll_flags |= LL_SBI_FAST_READ;
1084         else
1085                 sbi->ll_flags &= ~LL_SBI_FAST_READ;
1086         spin_unlock(&sbi->ll_lock);
1087
1088         return count;
1089 }
1090 LUSTRE_RW_ATTR(fast_read);
1091
1092 static ssize_t file_heat_show(struct kobject *kobj,
1093                               struct attribute *attr,
1094                               char *buf)
1095 {
1096         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1097                                               ll_kset.kobj);
1098
1099         return snprintf(buf, PAGE_SIZE, "%u\n",
1100                         !!(sbi->ll_flags & LL_SBI_FILE_HEAT));
1101 }
1102
1103 static ssize_t file_heat_store(struct kobject *kobj,
1104                                struct attribute *attr,
1105                                const char *buffer,
1106                                size_t count)
1107 {
1108         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1109                                               ll_kset.kobj);
1110         bool val;
1111         int rc;
1112
1113         rc = kstrtobool(buffer, &val);
1114         if (rc)
1115                 return rc;
1116
1117         spin_lock(&sbi->ll_lock);
1118         if (val)
1119                 sbi->ll_flags |= LL_SBI_FILE_HEAT;
1120         else
1121                 sbi->ll_flags &= ~LL_SBI_FILE_HEAT;
1122         spin_unlock(&sbi->ll_lock);
1123
1124         return count;
1125 }
1126 LUSTRE_RW_ATTR(file_heat);
1127
1128 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1129                                           struct attribute *attr,
1130                                           char *buf)
1131 {
1132         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1133                                               ll_kset.kobj);
1134
1135         return snprintf(buf, PAGE_SIZE, "%u\n",
1136                        (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1137 }
1138
1139 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1140                                            struct attribute *attr,
1141                                            const char *buffer,
1142                                            size_t count)
1143 {
1144         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1145                                               ll_kset.kobj);
1146         unsigned long val;
1147         int rc;
1148
1149         rc = kstrtoul(buffer, 10, &val);
1150         if (rc)
1151                 return rc;
1152
1153         if (val < 0 || val > 100)
1154                 return -ERANGE;
1155
1156         sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1157
1158         return count;
1159 }
1160 LUSTRE_RW_ATTR(heat_decay_percentage);
1161
1162 static ssize_t heat_period_second_show(struct kobject *kobj,
1163                                        struct attribute *attr,
1164                                        char *buf)
1165 {
1166         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1167                                               ll_kset.kobj);
1168
1169         return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1170 }
1171
1172 static ssize_t heat_period_second_store(struct kobject *kobj,
1173                                         struct attribute *attr,
1174                                         const char *buffer,
1175                                         size_t count)
1176 {
1177         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1178                                               ll_kset.kobj);
1179         unsigned long val;
1180         int rc;
1181
1182         rc = kstrtoul(buffer, 10, &val);
1183         if (rc)
1184                 return rc;
1185
1186         if (val <= 0)
1187                 return -ERANGE;
1188
1189         sbi->ll_heat_period_second = val;
1190
1191         return count;
1192 }
1193 LUSTRE_RW_ATTR(heat_period_second);
1194
1195 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1196 {
1197         struct super_block      *sb    = m->private;
1198         struct ll_sb_info       *sbi   = ll_s2sbi(sb);
1199         struct cl_client_cache  *cache = sbi->ll_cache;
1200         long pages;
1201         int mb;
1202
1203         pages = atomic_long_read(&cache->ccc_unstable_nr);
1204         mb    = (pages * PAGE_SIZE) >> 20;
1205
1206         seq_printf(m, "unstable_check:     %8d\n"
1207                       "unstable_pages: %12ld\n"
1208                       "unstable_mb:        %8d\n",
1209                    cache->ccc_unstable_check, pages, mb);
1210         return 0;
1211 }
1212
1213 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1214                                            const char __user *buffer,
1215                                            size_t count, loff_t *unused)
1216 {
1217         struct seq_file *seq = file->private_data;
1218         struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1219         char kernbuf[128];
1220         bool val;
1221         int rc;
1222
1223         if (count == 0)
1224                 return 0;
1225         if (count >= sizeof(kernbuf))
1226                 return -EINVAL;
1227
1228         if (copy_from_user(kernbuf, buffer, count))
1229                 return -EFAULT;
1230         kernbuf[count] = 0;
1231
1232         buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1233                   kernbuf;
1234         rc = kstrtobool_from_user(buffer, count, &val);
1235         if (rc < 0)
1236                 return rc;
1237
1238         /* borrow lru lock to set the value */
1239         spin_lock(&sbi->ll_cache->ccc_lru_lock);
1240         sbi->ll_cache->ccc_unstable_check = val;
1241         spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1242
1243         return count;
1244 }
1245
1246 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1247
1248 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1249 {
1250         struct super_block *sb = m->private;
1251         struct ll_sb_info *sbi = ll_s2sbi(sb);
1252         struct root_squash_info *squash = &sbi->ll_squash;
1253
1254         seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1255         return 0;
1256 }
1257
1258 static ssize_t ll_root_squash_seq_write(struct file *file,
1259                                         const char __user *buffer,
1260                                         size_t count, loff_t *off)
1261 {
1262         struct seq_file *m = file->private_data;
1263         struct super_block *sb = m->private;
1264         struct ll_sb_info *sbi = ll_s2sbi(sb);
1265         struct root_squash_info *squash = &sbi->ll_squash;
1266
1267         return lprocfs_wr_root_squash(buffer, count, squash, sbi->ll_fsname);
1268 }
1269
1270 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1271
1272 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1273 {
1274         struct super_block *sb = m->private;
1275         struct ll_sb_info *sbi = ll_s2sbi(sb);
1276         struct root_squash_info *squash = &sbi->ll_squash;
1277         int len;
1278
1279         down_read(&squash->rsi_sem);
1280         if (!list_empty(&squash->rsi_nosquash_nids)) {
1281                 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1282                                         &squash->rsi_nosquash_nids);
1283                 m->count += len;
1284                 seq_putc(m, '\n');
1285         } else {
1286                 seq_puts(m, "NONE\n");
1287         }
1288         up_read(&squash->rsi_sem);
1289
1290         return 0;
1291 }
1292
1293 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1294                                           const char __user *buffer,
1295                                           size_t count, loff_t *off)
1296 {
1297         struct seq_file *m = file->private_data;
1298         struct super_block *sb = m->private;
1299         struct ll_sb_info *sbi = ll_s2sbi(sb);
1300         struct root_squash_info *squash = &sbi->ll_squash;
1301         int rc;
1302
1303         rc = lprocfs_wr_nosquash_nids(buffer, count, squash, sbi->ll_fsname);
1304         if (rc < 0)
1305                 return rc;
1306
1307         ll_compute_rootsquash_state(sbi);
1308
1309         return rc;
1310 }
1311
1312 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1313
1314 static int ll_pcc_seq_show(struct seq_file *m, void *v)
1315 {
1316         struct super_block *sb = m->private;
1317         struct ll_sb_info *sbi = ll_s2sbi(sb);
1318
1319         return pcc_super_dump(&sbi->ll_pcc_super, m);
1320 }
1321
1322 static ssize_t ll_pcc_seq_write(struct file *file, const char __user *buffer,
1323                                 size_t count, loff_t *off)
1324 {
1325         struct seq_file *m = file->private_data;
1326         struct super_block *sb = m->private;
1327         struct ll_sb_info *sbi = ll_s2sbi(sb);
1328         int rc;
1329         char *kernbuf;
1330
1331         if (count >= LPROCFS_WR_PCC_MAX_CMD)
1332                 return -EINVAL;
1333
1334         if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_PCC))
1335                 return -EOPNOTSUPP;
1336
1337         OBD_ALLOC(kernbuf, count + 1);
1338         if (kernbuf == NULL)
1339                 return -ENOMEM;
1340
1341         if (copy_from_user(kernbuf, buffer, count))
1342                 GOTO(out_free_kernbuff, rc = -EFAULT);
1343
1344         rc = pcc_cmd_handle(kernbuf, count, &sbi->ll_pcc_super);
1345 out_free_kernbuff:
1346         OBD_FREE(kernbuf, count + 1);
1347         return rc ? rc : count;
1348 }
1349 LPROC_SEQ_FOPS(ll_pcc);
1350
1351 struct lprocfs_vars lprocfs_llite_obd_vars[] = {
1352         { .name =       "site",
1353           .fops =       &ll_site_stats_fops                     },
1354         { .name =       "max_read_ahead_mb",
1355           .fops =       &ll_max_readahead_mb_fops               },
1356         { .name =       "max_read_ahead_per_file_mb",
1357           .fops =       &ll_max_readahead_per_file_mb_fops      },
1358         { .name =       "max_read_ahead_whole_mb",
1359           .fops =       &ll_max_read_ahead_whole_mb_fops        },
1360         { .name =       "max_cached_mb",
1361           .fops =       &ll_max_cached_mb_fops                  },
1362         { .name =       "statahead_stats",
1363           .fops =       &ll_statahead_stats_fops                },
1364         { .name =       "unstable_stats",
1365           .fops =       &ll_unstable_stats_fops                 },
1366         { .name =       "sbi_flags",
1367           .fops =       &ll_sbi_flags_fops                      },
1368         { .name =       "root_squash",
1369           .fops =       &ll_root_squash_fops                    },
1370         { .name =       "nosquash_nids",
1371           .fops =       &ll_nosquash_nids_fops                  },
1372         { .name =       "pcc",
1373           .fops =       &ll_pcc_fops,                           },
1374         { NULL }
1375 };
1376
1377 #define MAX_STRING_SIZE 128
1378
1379 static struct attribute *llite_attrs[] = {
1380         &lustre_attr_blocksize.attr,
1381         &lustre_attr_stat_blocksize.attr,
1382         &lustre_attr_kbytestotal.attr,
1383         &lustre_attr_kbytesfree.attr,
1384         &lustre_attr_kbytesavail.attr,
1385         &lustre_attr_filestotal.attr,
1386         &lustre_attr_filesfree.attr,
1387         &lustre_attr_client_type.attr,
1388         &lustre_attr_fstype.attr,
1389         &lustre_attr_uuid.attr,
1390         &lustre_attr_checksums.attr,
1391         &lustre_attr_checksum_pages.attr,
1392         &lustre_attr_stats_track_pid.attr,
1393         &lustre_attr_stats_track_ppid.attr,
1394         &lustre_attr_stats_track_gid.attr,
1395         &lustre_attr_statahead_running_max.attr,
1396         &lustre_attr_statahead_max.attr,
1397         &lustre_attr_statahead_agl.attr,
1398         &lustre_attr_lazystatfs.attr,
1399         &lustre_attr_max_easize.attr,
1400         &lustre_attr_default_easize.attr,
1401         &lustre_attr_xattr_cache.attr,
1402         &lustre_attr_fast_read.attr,
1403         &lustre_attr_tiny_write.attr,
1404         &lustre_attr_file_heat.attr,
1405         &lustre_attr_heat_decay_percentage.attr,
1406         &lustre_attr_heat_period_second.attr,
1407         NULL,
1408 };
1409
1410 static void sbi_kobj_release(struct kobject *kobj)
1411 {
1412         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1413                                               ll_kset.kobj);
1414         complete(&sbi->ll_kobj_unregister);
1415 }
1416
1417 static struct kobj_type sbi_ktype = {
1418         .default_attrs  = llite_attrs,
1419         .sysfs_ops      = &lustre_sysfs_ops,
1420         .release        = sbi_kobj_release,
1421 };
1422
1423 static const struct llite_file_opcode {
1424         __u32       opcode;
1425         __u32       type;
1426         const char *opname;
1427 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1428         /* file operation */
1429         { LPROC_LL_DIRTY_HITS,     LPROCFS_TYPE_REGS, "dirty_pages_hits" },
1430         { LPROC_LL_DIRTY_MISSES,   LPROCFS_TYPE_REGS, "dirty_pages_misses" },
1431         { LPROC_LL_READ_BYTES,     LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
1432                                    "read_bytes" },
1433         { LPROC_LL_WRITE_BYTES,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
1434                                    "write_bytes" },
1435         { LPROC_LL_BRW_READ,       LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
1436                                    "brw_read" },
1437         { LPROC_LL_BRW_WRITE,      LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
1438                                    "brw_write" },
1439         { LPROC_LL_IOCTL,          LPROCFS_TYPE_REGS, "ioctl" },
1440         { LPROC_LL_OPEN,           LPROCFS_TYPE_REGS, "open" },
1441         { LPROC_LL_RELEASE,        LPROCFS_TYPE_REGS, "close" },
1442         { LPROC_LL_MAP,            LPROCFS_TYPE_REGS, "mmap" },
1443         { LPROC_LL_FAULT,          LPROCFS_TYPE_REGS, "page_fault" },
1444         { LPROC_LL_MKWRITE,        LPROCFS_TYPE_REGS, "page_mkwrite" },
1445         { LPROC_LL_LLSEEK,         LPROCFS_TYPE_REGS, "seek" },
1446         { LPROC_LL_FSYNC,          LPROCFS_TYPE_REGS, "fsync" },
1447         { LPROC_LL_READDIR,        LPROCFS_TYPE_REGS, "readdir" },
1448         /* inode operation */
1449         { LPROC_LL_SETATTR,        LPROCFS_TYPE_REGS, "setattr" },
1450         { LPROC_LL_TRUNC,          LPROCFS_TYPE_REGS, "truncate" },
1451         { LPROC_LL_FLOCK,          LPROCFS_TYPE_REGS, "flock" },
1452         { LPROC_LL_GETATTR,        LPROCFS_TYPE_REGS, "getattr" },
1453         /* dir inode operation */
1454         { LPROC_LL_CREATE,         LPROCFS_TYPE_REGS, "create" },
1455         { LPROC_LL_LINK,           LPROCFS_TYPE_REGS, "link" },
1456         { LPROC_LL_UNLINK,         LPROCFS_TYPE_REGS, "unlink" },
1457         { LPROC_LL_SYMLINK,        LPROCFS_TYPE_REGS, "symlink" },
1458         { LPROC_LL_MKDIR,          LPROCFS_TYPE_REGS, "mkdir" },
1459         { LPROC_LL_RMDIR,          LPROCFS_TYPE_REGS, "rmdir" },
1460         { LPROC_LL_MKNOD,          LPROCFS_TYPE_REGS, "mknod" },
1461         { LPROC_LL_RENAME,         LPROCFS_TYPE_REGS, "rename" },
1462         /* special inode operation */
1463         { LPROC_LL_STAFS,          LPROCFS_TYPE_REGS, "statfs" },
1464         { LPROC_LL_ALLOC_INODE,    LPROCFS_TYPE_REGS, "alloc_inode" },
1465         { LPROC_LL_SETXATTR,       LPROCFS_TYPE_REGS, "setxattr" },
1466         { LPROC_LL_GETXATTR,       LPROCFS_TYPE_REGS, "getxattr" },
1467         { LPROC_LL_GETXATTR_HITS,  LPROCFS_TYPE_REGS, "getxattr_hits" },
1468         { LPROC_LL_LISTXATTR,      LPROCFS_TYPE_REGS, "listxattr" },
1469         { LPROC_LL_REMOVEXATTR,    LPROCFS_TYPE_REGS, "removexattr" },
1470         { LPROC_LL_INODE_PERM,     LPROCFS_TYPE_REGS, "inode_permission" },
1471 };
1472
1473 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
1474 {
1475         if (!sbi->ll_stats)
1476                 return;
1477         if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1478                 lprocfs_counter_add(sbi->ll_stats, op, count);
1479         else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1480                  sbi->ll_stats_track_id == current->pid)
1481                 lprocfs_counter_add(sbi->ll_stats, op, count);
1482         else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1483                  sbi->ll_stats_track_id == current->parent->pid)
1484                 lprocfs_counter_add(sbi->ll_stats, op, count);
1485         else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1486                  sbi->ll_stats_track_id ==
1487                         from_kgid(&init_user_ns, current_gid()))
1488                 lprocfs_counter_add(sbi->ll_stats, op, count);
1489 }
1490 EXPORT_SYMBOL(ll_stats_ops_tally);
1491
1492 static const char *ra_stat_string[] = {
1493         [RA_STAT_HIT] = "hits",
1494         [RA_STAT_MISS] = "misses",
1495         [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1496         [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1497         [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1498         [RA_STAT_FAILED_MATCH] = "failed lock match",
1499         [RA_STAT_DISCARDED] = "read but discarded",
1500         [RA_STAT_ZERO_LEN] = "zero length file",
1501         [RA_STAT_ZERO_WINDOW] = "zero size window",
1502         [RA_STAT_EOF] = "read-ahead to EOF",
1503         [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1504         [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1505         [RA_STAT_FAILED_REACH_END] = "failed to reach end"
1506 };
1507
1508 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1509 {
1510         struct lustre_sb_info *lsi = s2lsi(sb);
1511         struct ll_sb_info *sbi = ll_s2sbi(sb);
1512         int err, id, rc;
1513
1514         ENTRY;
1515         LASSERT(sbi);
1516
1517         if (IS_ERR_OR_NULL(llite_root))
1518                 goto out_ll_kset;
1519
1520         sbi->ll_debugfs_entry = ldebugfs_register(name, llite_root,
1521                                                   lprocfs_llite_obd_vars, sb);
1522         if (IS_ERR_OR_NULL(sbi->ll_debugfs_entry)) {
1523                 err = sbi->ll_debugfs_entry ? PTR_ERR(sbi->ll_debugfs_entry) :
1524                                               -ENOMEM;
1525                 sbi->ll_debugfs_entry = NULL;
1526                 RETURN(err);
1527         }
1528
1529         rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "dump_page_cache",0444,
1530                                  &vvp_dump_pgcache_file_ops, sbi);
1531         if (rc)
1532                 CWARN("Error adding the dump_page_cache file\n");
1533
1534         rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "extents_stats", 0644,
1535                                  &ll_rw_extents_stats_fops, sbi);
1536         if (rc)
1537                 CWARN("Error adding the extent_stats file\n");
1538
1539         rc = ldebugfs_seq_create(sbi->ll_debugfs_entry,
1540                                  "extents_stats_per_process", 0644,
1541                                  &ll_rw_extents_stats_pp_fops, sbi);
1542         if (rc)
1543                 CWARN("Error adding the extents_stats_per_process file\n");
1544
1545         rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "offset_stats", 0644,
1546                                  &ll_rw_offset_stats_fops, sbi);
1547         if (rc)
1548                 CWARN("Error adding the offset_stats file\n");
1549
1550         /* File operations stats */
1551         sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1552                                             LPROCFS_STATS_FLAG_NONE);
1553         if (sbi->ll_stats == NULL)
1554                 GOTO(out_debugfs, err = -ENOMEM);
1555
1556         /* do counter init */
1557         for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1558                 u32 type = llite_opcode_table[id].type;
1559                 void *ptr = NULL;
1560
1561                 if (type & LPROCFS_TYPE_REGS)
1562                         ptr = "regs";
1563                 else if (type & LPROCFS_TYPE_BYTES)
1564                         ptr = "bytes";
1565                 else if (type & LPROCFS_TYPE_PAGES)
1566                         ptr = "pages";
1567                 lprocfs_counter_init(sbi->ll_stats,
1568                                      llite_opcode_table[id].opcode,
1569                                      (type & LPROCFS_CNTR_AVGMINMAX),
1570                                      llite_opcode_table[id].opname, ptr);
1571         }
1572
1573         err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "stats",
1574                                       sbi->ll_stats);
1575         if (err)
1576                 GOTO(out_stats, err);
1577
1578         sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1579                                                LPROCFS_STATS_FLAG_NONE);
1580         if (sbi->ll_ra_stats == NULL)
1581                 GOTO(out_stats, err = -ENOMEM);
1582
1583         for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1584                 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1585                                      ra_stat_string[id], "pages");
1586
1587         err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "read_ahead_stats",
1588                                       sbi->ll_ra_stats);
1589         if (err)
1590                 GOTO(out_ra_stats, err);
1591
1592 out_ll_kset:
1593         /* Yes we also register sysfs mount kset here as well */
1594         sbi->ll_kset.kobj.parent = llite_kobj;
1595         sbi->ll_kset.kobj.ktype = &sbi_ktype;
1596         init_completion(&sbi->ll_kobj_unregister);
1597         err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1598         if (err)
1599                 GOTO(out_ra_stats, err);
1600
1601         err = kset_register(&sbi->ll_kset);
1602         if (err)
1603                 GOTO(out_ra_stats, err);
1604
1605         lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1606
1607         RETURN(0);
1608 out_ra_stats:
1609         lprocfs_free_stats(&sbi->ll_ra_stats);
1610 out_stats:
1611         lprocfs_free_stats(&sbi->ll_stats);
1612 out_debugfs:
1613         ldebugfs_remove(&sbi->ll_debugfs_entry);
1614
1615         RETURN(err);
1616 }
1617
1618 void ll_debugfs_unregister_super(struct super_block *sb)
1619 {
1620         struct lustre_sb_info *lsi = s2lsi(sb);
1621         struct ll_sb_info *sbi = ll_s2sbi(sb);
1622
1623         if (!IS_ERR_OR_NULL(sbi->ll_debugfs_entry))
1624                 ldebugfs_remove(&sbi->ll_debugfs_entry);
1625
1626         if (sbi->ll_dt_obd)
1627                 sysfs_remove_link(&sbi->ll_kset.kobj,
1628                                   sbi->ll_dt_obd->obd_type->typ_name);
1629
1630         if (sbi->ll_md_obd)
1631                 sysfs_remove_link(&sbi->ll_kset.kobj,
1632                                   sbi->ll_md_obd->obd_type->typ_name);
1633
1634         kobject_put(lsi->lsi_kobj);
1635
1636         kset_unregister(&sbi->ll_kset);
1637         wait_for_completion(&sbi->ll_kobj_unregister);
1638
1639         lprocfs_free_stats(&sbi->ll_ra_stats);
1640         lprocfs_free_stats(&sbi->ll_stats);
1641 }
1642 #undef MAX_STRING_SIZE
1643
1644 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1645                                    struct seq_file *seq, int which)
1646 {
1647         unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1648         unsigned long start, end, r, w;
1649         char *unitp = "KMGTPEZY";
1650         int i, units = 10;
1651         struct per_process_info *pp_info = &io_extents->pp_extents[which];
1652
1653         read_cum = 0;
1654         write_cum = 0;
1655         start = 0;
1656
1657         for(i = 0; i < LL_HIST_MAX; i++) {
1658                 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1659                 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1660         }
1661
1662         for(i = 0; i < LL_HIST_MAX; i++) {
1663                 r = pp_info->pp_r_hist.oh_buckets[i];
1664                 w = pp_info->pp_w_hist.oh_buckets[i];
1665                 read_cum += r;
1666                 write_cum += w;
1667                 end = BIT(i + LL_HIST_START - units);
1668                 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u  | "
1669                            "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1670                            (i == LL_HIST_MAX - 1) ? '+' : ' ',
1671                            r, pct(r, read_tot), pct(read_cum, read_tot),
1672                            w, pct(w, write_tot), pct(write_cum, write_tot));
1673                 start = end;
1674                 if (start == BIT(10)) {
1675                         start = 1;
1676                         units += 10;
1677                         unitp++;
1678                 }
1679                 if (read_cum == read_tot && write_cum == write_tot)
1680                         break;
1681         }
1682 }
1683
1684 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1685 {
1686         struct timespec64 now;
1687         struct ll_sb_info *sbi = seq->private;
1688         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1689         int k;
1690
1691         ktime_get_real_ts64(&now);
1692
1693         if (!sbi->ll_rw_stats_on) {
1694                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1695                 return 0;
1696         }
1697         seq_printf(seq, "snapshot_time:         %llu.%09lu (secs.nsecs)\n",
1698                    (s64)now.tv_sec, now.tv_nsec);
1699         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1700         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1701                    "extents", "calls", "%", "cum%",
1702                    "calls", "%", "cum%");
1703         spin_lock(&sbi->ll_pp_extent_lock);
1704         for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1705                 if (io_extents->pp_extents[k].pid != 0) {
1706                         seq_printf(seq, "\nPID: %d\n",
1707                                    io_extents->pp_extents[k].pid);
1708                         ll_display_extents_info(io_extents, seq, k);
1709                 }
1710         }
1711         spin_unlock(&sbi->ll_pp_extent_lock);
1712         return 0;
1713 }
1714
1715 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1716                                                 const char __user *buf,
1717                                                 size_t len,
1718                                                 loff_t *off)
1719 {
1720         struct seq_file *seq = file->private_data;
1721         struct ll_sb_info *sbi = seq->private;
1722         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1723         int i;
1724         __s64 value;
1725
1726         if (len == 0)
1727                 return -EINVAL;
1728
1729         value = ll_stats_pid_write(buf, len);
1730
1731         if (value == 0)
1732                 sbi->ll_rw_stats_on = 0;
1733         else
1734                 sbi->ll_rw_stats_on = 1;
1735
1736         spin_lock(&sbi->ll_pp_extent_lock);
1737         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1738                 io_extents->pp_extents[i].pid = 0;
1739                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1740                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1741         }
1742         spin_unlock(&sbi->ll_pp_extent_lock);
1743         return len;
1744 }
1745
1746 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1747
1748 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1749 {
1750         struct timespec64 now;
1751         struct ll_sb_info *sbi = seq->private;
1752         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1753
1754         ktime_get_real_ts64(&now);
1755
1756         if (!sbi->ll_rw_stats_on) {
1757                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1758                 return 0;
1759         }
1760         seq_printf(seq, "snapshot_time:         %llu.%09lu (secs.nsecs)\n",
1761                    (s64)now.tv_sec, now.tv_nsec);
1762
1763         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1764         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1765                    "extents", "calls", "%", "cum%",
1766                    "calls", "%", "cum%");
1767         spin_lock(&sbi->ll_lock);
1768         ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1769         spin_unlock(&sbi->ll_lock);
1770
1771         return 0;
1772 }
1773
1774 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1775                                              const char __user *buf,
1776                                              size_t len, loff_t *off)
1777 {
1778         struct seq_file *seq = file->private_data;
1779         struct ll_sb_info *sbi = seq->private;
1780         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1781         int i;
1782         __s64 value;
1783
1784         if (len == 0)
1785                 return -EINVAL;
1786
1787         value = ll_stats_pid_write(buf, len);
1788
1789         if (value == 0)
1790                 sbi->ll_rw_stats_on = 0;
1791         else
1792                 sbi->ll_rw_stats_on = 1;
1793
1794         spin_lock(&sbi->ll_pp_extent_lock);
1795         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1796                 io_extents->pp_extents[i].pid = 0;
1797                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1798                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1799         }
1800         spin_unlock(&sbi->ll_pp_extent_lock);
1801
1802         return len;
1803 }
1804
1805 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1806
1807 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1808                        struct ll_file_data *file, loff_t pos,
1809                        size_t count, int rw)
1810 {
1811         int i, cur = -1;
1812         struct ll_rw_process_info *process;
1813         struct ll_rw_process_info *offset;
1814         int *off_count = &sbi->ll_rw_offset_entry_count;
1815         int *process_count = &sbi->ll_offset_process_count;
1816         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1817
1818         if(!sbi->ll_rw_stats_on)
1819                 return;
1820         process = sbi->ll_rw_process_info;
1821         offset = sbi->ll_rw_offset_info;
1822
1823         spin_lock(&sbi->ll_pp_extent_lock);
1824         /* Extent statistics */
1825         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1826                 if(io_extents->pp_extents[i].pid == pid) {
1827                         cur = i;
1828                         break;
1829                 }
1830         }
1831
1832         if (cur == -1) {
1833                 /* new process */
1834                 sbi->ll_extent_process_count =
1835                         (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1836                 cur = sbi->ll_extent_process_count;
1837                 io_extents->pp_extents[cur].pid = pid;
1838                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1839                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1840         }
1841
1842         for(i = 0; (count >= BIT(LL_HIST_START << i)) &&
1843              (i < (LL_HIST_MAX - 1)); i++);
1844         if (rw == 0) {
1845                 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1846                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1847         } else {
1848                 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1849                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1850         }
1851         spin_unlock(&sbi->ll_pp_extent_lock);
1852
1853         spin_lock(&sbi->ll_process_lock);
1854         /* Offset statistics */
1855         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1856                 if (process[i].rw_pid == pid) {
1857                         if (process[i].rw_last_file != file) {
1858                                 process[i].rw_range_start = pos;
1859                                 process[i].rw_last_file_pos = pos + count;
1860                                 process[i].rw_smallest_extent = count;
1861                                 process[i].rw_largest_extent = count;
1862                                 process[i].rw_offset = 0;
1863                                 process[i].rw_last_file = file;
1864                                 spin_unlock(&sbi->ll_process_lock);
1865                                 return;
1866                         }
1867                         if (process[i].rw_last_file_pos != pos) {
1868                                 *off_count =
1869                                     (*off_count + 1) % LL_OFFSET_HIST_MAX;
1870                                 offset[*off_count].rw_op = process[i].rw_op;
1871                                 offset[*off_count].rw_pid = pid;
1872                                 offset[*off_count].rw_range_start =
1873                                         process[i].rw_range_start;
1874                                 offset[*off_count].rw_range_end =
1875                                         process[i].rw_last_file_pos;
1876                                 offset[*off_count].rw_smallest_extent =
1877                                         process[i].rw_smallest_extent;
1878                                 offset[*off_count].rw_largest_extent =
1879                                         process[i].rw_largest_extent;
1880                                 offset[*off_count].rw_offset =
1881                                         process[i].rw_offset;
1882                                 process[i].rw_op = rw;
1883                                 process[i].rw_range_start = pos;
1884                                 process[i].rw_smallest_extent = count;
1885                                 process[i].rw_largest_extent = count;
1886                                 process[i].rw_offset = pos -
1887                                         process[i].rw_last_file_pos;
1888                         }
1889                         if(process[i].rw_smallest_extent > count)
1890                                 process[i].rw_smallest_extent = count;
1891                         if(process[i].rw_largest_extent < count)
1892                                 process[i].rw_largest_extent = count;
1893                         process[i].rw_last_file_pos = pos + count;
1894                         spin_unlock(&sbi->ll_process_lock);
1895                         return;
1896                 }
1897         }
1898         *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1899         process[*process_count].rw_pid = pid;
1900         process[*process_count].rw_op = rw;
1901         process[*process_count].rw_range_start = pos;
1902         process[*process_count].rw_last_file_pos = pos + count;
1903         process[*process_count].rw_smallest_extent = count;
1904         process[*process_count].rw_largest_extent = count;
1905         process[*process_count].rw_offset = 0;
1906         process[*process_count].rw_last_file = file;
1907         spin_unlock(&sbi->ll_process_lock);
1908 }
1909
1910 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1911 {
1912         struct timespec64 now;
1913         struct ll_sb_info *sbi = seq->private;
1914         struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1915         struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1916         int i;
1917
1918         ktime_get_real_ts64(&now);
1919
1920         if (!sbi->ll_rw_stats_on) {
1921                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1922                 return 0;
1923         }
1924         spin_lock(&sbi->ll_process_lock);
1925
1926         seq_printf(seq, "snapshot_time:         %llu.%09lu (secs.nsecs)\n",
1927                    (s64)now.tv_sec, now.tv_nsec);
1928         seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1929                    "R/W", "PID", "RANGE START", "RANGE END",
1930                    "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1931
1932         /* We stored the discontiguous offsets here; print them first */
1933         for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1934                 if (offset[i].rw_pid != 0)
1935                         seq_printf(seq,
1936                                   "%3c %10d %14llu %14llu %17lu %17lu %14llu\n",
1937                                    offset[i].rw_op == READ ? 'R' : 'W',
1938                                    offset[i].rw_pid,
1939                                    offset[i].rw_range_start,
1940                                    offset[i].rw_range_end,
1941                                    (unsigned long)offset[i].rw_smallest_extent,
1942                                    (unsigned long)offset[i].rw_largest_extent,
1943                                    offset[i].rw_offset);
1944         }
1945
1946         /* Then print the current offsets for each process */
1947         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1948                 if (process[i].rw_pid != 0)
1949                         seq_printf(seq,
1950                                   "%3c %10d %14llu %14llu %17lu %17lu %14llu\n",
1951                                    process[i].rw_op == READ ? 'R' : 'W',
1952                                    process[i].rw_pid,
1953                                    process[i].rw_range_start,
1954                                    process[i].rw_last_file_pos,
1955                                    (unsigned long)process[i].rw_smallest_extent,
1956                                    (unsigned long)process[i].rw_largest_extent,
1957                                    process[i].rw_offset);
1958         }
1959         spin_unlock(&sbi->ll_process_lock);
1960
1961         return 0;
1962 }
1963
1964 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
1965                                             const char __user *buf,
1966                                             size_t len, loff_t *off)
1967 {
1968         struct seq_file *seq = file->private_data;
1969         struct ll_sb_info *sbi = seq->private;
1970         struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1971         struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1972         __s64 value;
1973
1974         if (len == 0)
1975                 return -EINVAL;
1976
1977         value = ll_stats_pid_write(buf, len);
1978
1979         if (value == 0)
1980                 sbi->ll_rw_stats_on = 0;
1981         else
1982                 sbi->ll_rw_stats_on = 1;
1983
1984         spin_lock(&sbi->ll_process_lock);
1985         sbi->ll_offset_process_count = 0;
1986         sbi->ll_rw_offset_entry_count = 0;
1987         memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1988                LL_PROCESS_HIST_MAX);
1989         memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1990                LL_OFFSET_HIST_MAX);
1991         spin_unlock(&sbi->ll_process_lock);
1992
1993         return len;
1994 }
1995
1996 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);