4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 #define DEBUG_SUBSYSTEM S_LLITE
34 #include <linux/version.h>
35 #include <linux/user_namespace.h>
36 #ifdef HAVE_UIDGID_HEADER
37 # include <linux/uidgid.h>
39 #include <uapi/linux/lustre/lustre_param.h>
40 #include <lprocfs_status.h>
41 #include <obd_support.h>
43 #include "llite_internal.h"
44 #include "vvp_internal.h"
46 static struct kobject *llite_kobj;
47 static struct dentry *llite_root;
49 int llite_tunables_register(void)
53 llite_kobj = class_setup_tunables("llite");
54 if (IS_ERR(llite_kobj))
55 return PTR_ERR(llite_kobj);
57 llite_root = debugfs_create_dir("llite", debugfs_lustre_root);
58 if (IS_ERR_OR_NULL(llite_root)) {
59 rc = llite_root ? PTR_ERR(llite_root) : -ENOMEM;
61 kobject_put(llite_kobj);
68 void llite_tunables_unregister(void)
71 kobject_put(llite_kobj);
75 if (!IS_ERR_OR_NULL(llite_root)) {
76 debugfs_remove(llite_root);
81 /* <debugfs>/lustre/llite mount point registration */
82 static const struct file_operations ll_rw_extents_stats_fops;
83 static const struct file_operations ll_rw_extents_stats_pp_fops;
84 static const struct file_operations ll_rw_offset_stats_fops;
87 * ll_stats_pid_write() - Determine if stats collection should be enabled
88 * @buf: Buffer containing the data written
89 * @len: Number of bytes in the buffer
91 * Several proc files begin collecting stats when a value is written, and stop
92 * collecting when either '0' or 'disable' is written. This function checks the
93 * written value to see if collection should be enabled or disabled.
95 * Return: If '0' or 'disable' is provided, 0 is returned. If the text
96 * equivalent of a number is written, that number is returned. Otherwise,
97 * 1 is returned. Non-zero return values indicate collection should be enabled.
99 static s64 ll_stats_pid_write(const char __user *buf, size_t len)
101 unsigned long long value = 1;
105 rc = kstrtoull_from_user(buf, len, 0, &value);
106 if (rc < 0 && len < sizeof(kernbuf)) {
107 if (copy_from_user(kernbuf, buf, len))
111 if (kernbuf[len - 1] == '\n')
112 kernbuf[len - 1] = 0;
114 if (strncasecmp(kernbuf, "disable", 7) == 0)
121 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
124 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
126 struct obd_statfs osfs;
129 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
133 return sprintf(buf, "%u\n", osfs.os_bsize);
135 LUSTRE_RO_ATTR(blocksize);
137 static ssize_t stat_blocksize_show(struct kobject *kobj, struct attribute *attr,
140 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
143 return sprintf(buf, "%u\n", sbi->ll_stat_blksize);
146 static ssize_t stat_blocksize_store(struct kobject *kobj,
147 struct attribute *attr,
151 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
156 rc = kstrtouint(buffer, 10, &val);
160 if (val != 0 && (val < PAGE_SIZE || (val & (val - 1))) != 0)
163 sbi->ll_stat_blksize = val;
167 LUSTRE_RW_ATTR(stat_blocksize);
169 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
172 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
174 struct obd_statfs osfs;
179 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
183 blk_size = osfs.os_bsize >> 10;
184 result = osfs.os_blocks;
186 while (blk_size >>= 1)
189 return sprintf(buf, "%llu\n", result);
191 LUSTRE_RO_ATTR(kbytestotal);
193 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
196 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
198 struct obd_statfs osfs;
203 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
207 blk_size = osfs.os_bsize >> 10;
208 result = osfs.os_bfree;
210 while (blk_size >>= 1)
213 return sprintf(buf, "%llu\n", result);
215 LUSTRE_RO_ATTR(kbytesfree);
217 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
220 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
222 struct obd_statfs osfs;
227 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
231 blk_size = osfs.os_bsize >> 10;
232 result = osfs.os_bavail;
234 while (blk_size >>= 1)
237 return sprintf(buf, "%llu\n", result);
239 LUSTRE_RO_ATTR(kbytesavail);
241 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
244 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
246 struct obd_statfs osfs;
249 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
253 return sprintf(buf, "%llu\n", osfs.os_files);
255 LUSTRE_RO_ATTR(filestotal);
257 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
260 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
262 struct obd_statfs osfs;
265 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
269 return sprintf(buf, "%llu\n", osfs.os_ffree);
271 LUSTRE_RO_ATTR(filesfree);
273 static ssize_t client_type_show(struct kobject *kobj, struct attribute *attr,
276 return sprintf(buf, "local client\n");
278 LUSTRE_RO_ATTR(client_type);
280 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
283 return sprintf(buf, "lustre\n");
285 LUSTRE_RO_ATTR(fstype);
287 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
290 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
293 return sprintf(buf, "%s\n", sbi->ll_sb_uuid.uuid);
295 LUSTRE_RO_ATTR(uuid);
297 static int ll_site_stats_seq_show(struct seq_file *m, void *v)
299 struct super_block *sb = m->private;
302 * See description of statistical counters in struct cl_site, and
305 return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site), m);
308 LDEBUGFS_SEQ_FOPS_RO(ll_site_stats);
310 static int ll_max_readahead_mb_seq_show(struct seq_file *m, void *v)
312 struct super_block *sb = m->private;
313 struct ll_sb_info *sbi = ll_s2sbi(sb);
314 unsigned long ra_max_mb;
316 spin_lock(&sbi->ll_lock);
317 ra_max_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages);
318 spin_unlock(&sbi->ll_lock);
320 seq_printf(m, "%lu\n", ra_max_mb);
325 ll_max_readahead_mb_seq_write(struct file *file, const char __user *buffer,
326 size_t count, loff_t *off)
328 struct seq_file *m = file->private_data;
329 struct super_block *sb = m->private;
330 struct ll_sb_info *sbi = ll_s2sbi(sb);
331 s64 ra_max_mb, pages_number;
334 rc = lprocfs_str_with_units_to_s64(buffer, count, &ra_max_mb, 'M');
338 pages_number = round_up(ra_max_mb, 1024 * 1024) >> PAGE_SHIFT;
339 if (pages_number < 0 || pages_number > cfs_totalram_pages() / 2) {
341 CERROR("%s: can't set max_readahead_mb=%llu > %luMB\n",
342 ll_get_fsname(sb, NULL, 0), PAGES_TO_MiB(pages_number),
343 PAGES_TO_MiB(cfs_totalram_pages()));
347 spin_lock(&sbi->ll_lock);
348 sbi->ll_ra_info.ra_max_pages = pages_number;
349 spin_unlock(&sbi->ll_lock);
354 LDEBUGFS_SEQ_FOPS(ll_max_readahead_mb);
356 static int ll_max_readahead_per_file_mb_seq_show(struct seq_file *m, void *v)
358 struct super_block *sb = m->private;
359 struct ll_sb_info *sbi = ll_s2sbi(sb);
360 unsigned long ra_max_file_mb;
362 spin_lock(&sbi->ll_lock);
363 ra_max_file_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file);
364 spin_unlock(&sbi->ll_lock);
366 seq_printf(m, "%lu\n", ra_max_file_mb);
371 ll_max_readahead_per_file_mb_seq_write(struct file *file,
372 const char __user *buffer,
373 size_t count, loff_t *off)
375 struct seq_file *m = file->private_data;
376 struct super_block *sb = m->private;
377 struct ll_sb_info *sbi = ll_s2sbi(sb);
378 s64 ra_max_file_mb, pages_number;
381 rc = lprocfs_str_with_units_to_s64(buffer, count, &ra_max_file_mb,
386 pages_number = round_up(ra_max_file_mb, 1024 * 1024) >> PAGE_SHIFT;
387 if (pages_number < 0 || pages_number > sbi->ll_ra_info.ra_max_pages) {
388 CERROR("%s: can't set max_readahead_per_file_mb=%llu > max_read_ahead_mb=%lu\n",
389 ll_get_fsname(sb, NULL, 0), PAGES_TO_MiB(pages_number),
390 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
394 spin_lock(&sbi->ll_lock);
395 sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
396 spin_unlock(&sbi->ll_lock);
401 LDEBUGFS_SEQ_FOPS(ll_max_readahead_per_file_mb);
403 static int ll_max_read_ahead_whole_mb_seq_show(struct seq_file *m, void *v)
405 struct super_block *sb = m->private;
406 struct ll_sb_info *sbi = ll_s2sbi(sb);
407 unsigned long ra_max_whole_mb;
409 spin_lock(&sbi->ll_lock);
410 ra_max_whole_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_read_ahead_whole_pages);
411 spin_unlock(&sbi->ll_lock);
413 seq_printf(m, "%lu\n", ra_max_whole_mb);
418 ll_max_read_ahead_whole_mb_seq_write(struct file *file,
419 const char __user *buffer,
420 size_t count, loff_t *off)
422 struct seq_file *m = file->private_data;
423 struct super_block *sb = m->private;
424 struct ll_sb_info *sbi = ll_s2sbi(sb);
425 s64 ra_max_whole_mb, pages_number;
428 rc = lprocfs_str_with_units_to_s64(buffer, count, &ra_max_whole_mb,
433 pages_number = round_up(ra_max_whole_mb, 1024 * 1024) >> PAGE_SHIFT;
434 /* Cap this at the current max readahead window size, the readahead
435 * algorithm does this anyway so it's pointless to set it larger.
437 if (pages_number < 0 ||
438 pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
439 CERROR("%s: can't set max_read_ahead_whole_mb=%llu > max_read_ahead_per_file_mb=%lu\n",
440 ll_get_fsname(sb, NULL, 0), PAGES_TO_MiB(pages_number),
441 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
445 spin_lock(&sbi->ll_lock);
446 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
447 spin_unlock(&sbi->ll_lock);
452 LDEBUGFS_SEQ_FOPS(ll_max_read_ahead_whole_mb);
454 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
456 struct super_block *sb = m->private;
457 struct ll_sb_info *sbi = ll_s2sbi(sb);
458 struct cl_client_cache *cache = sbi->ll_cache;
462 max_cached_mb = PAGES_TO_MiB(cache->ccc_lru_max);
463 unused_mb = PAGES_TO_MiB(atomic_long_read(&cache->ccc_lru_left));
464 seq_printf(m, "users: %d\n"
465 "max_cached_mb: %ld\n"
468 "reclaim_count: %u\n",
469 atomic_read(&cache->ccc_users),
471 max_cached_mb - unused_mb,
473 cache->ccc_lru_shrinkers);
477 static ssize_t ll_max_cached_mb_seq_write(struct file *file,
478 const char __user *buffer,
479 size_t count, loff_t *off)
481 struct seq_file *m = file->private_data;
482 struct super_block *sb = m->private;
483 struct ll_sb_info *sbi = ll_s2sbi(sb);
484 struct cl_client_cache *cache = sbi->ll_cache;
494 if (count >= sizeof(kernbuf))
497 if (copy_from_user(kernbuf, buffer, count))
501 buffer += lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count) -
503 rc = lprocfs_str_with_units_to_s64(buffer, count, &pages_number, 'M');
507 pages_number >>= PAGE_SHIFT;
509 if (pages_number < 0 || pages_number > cfs_totalram_pages()) {
510 CERROR("%s: can't set max cache more than %lu MB\n",
511 ll_get_fsname(sb, NULL, 0),
512 PAGES_TO_MiB(cfs_totalram_pages()));
515 /* Allow enough cache so clients can make well-formed RPCs */
516 pages_number = max_t(long, pages_number, PTLRPC_MAX_BRW_PAGES);
518 spin_lock(&sbi->ll_lock);
519 diff = pages_number - cache->ccc_lru_max;
520 spin_unlock(&sbi->ll_lock);
522 /* easy - add more LRU slots. */
524 atomic_long_add(diff, &cache->ccc_lru_left);
528 env = cl_env_get(&refcheck);
530 RETURN(PTR_ERR(env));
536 /* reduce LRU budget from free slots. */
540 ov = atomic_long_read(&cache->ccc_lru_left);
544 nv = ov > diff ? ov - diff : 0;
545 rc = atomic_long_cmpxchg(&cache->ccc_lru_left, ov, nv);
546 if (likely(ov == rc)) {
556 if (sbi->ll_dt_exp == NULL) { /* being initialized */
561 /* difficult - have to ask OSCs to drop LRU slots. */
563 rc = obd_set_info_async(env, sbi->ll_dt_exp,
564 sizeof(KEY_CACHE_LRU_SHRINK),
565 KEY_CACHE_LRU_SHRINK,
566 sizeof(tmp), &tmp, NULL);
570 cl_env_put(env, &refcheck);
574 spin_lock(&sbi->ll_lock);
575 cache->ccc_lru_max = pages_number;
576 spin_unlock(&sbi->ll_lock);
579 atomic_long_add(nrpages, &cache->ccc_lru_left);
584 LDEBUGFS_SEQ_FOPS(ll_max_cached_mb);
586 static ssize_t checksums_show(struct kobject *kobj, struct attribute *attr,
589 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
592 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
595 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
596 const char *buffer, size_t count)
598 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
608 rc = kstrtobool(buffer, &val);
612 sbi->ll_flags |= LL_SBI_CHECKSUM;
614 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
617 rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
618 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
620 CWARN("Failed to set OSC checksum flags: %d\n", rc);
624 LUSTRE_RW_ATTR(checksums);
626 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
628 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
629 enum stats_track_type type)
631 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
634 if (sbi->ll_stats_track_type == type)
635 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
636 else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
637 return sprintf(buf, "0 (all)\n");
639 return sprintf(buf, "untracked\n");
642 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
643 size_t count, enum stats_track_type type)
645 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
650 rc = kstrtoul(buffer, 10, &pid);
654 sbi->ll_stats_track_id = pid;
656 sbi->ll_stats_track_type = STATS_TRACK_ALL;
658 sbi->ll_stats_track_type = type;
659 lprocfs_clear_stats(sbi->ll_stats);
663 static ssize_t stats_track_pid_show(struct kobject *kobj,
664 struct attribute *attr,
667 return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
670 static ssize_t stats_track_pid_store(struct kobject *kobj,
671 struct attribute *attr,
675 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
677 LUSTRE_RW_ATTR(stats_track_pid);
679 static ssize_t stats_track_ppid_show(struct kobject *kobj,
680 struct attribute *attr,
683 return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
686 static ssize_t stats_track_ppid_store(struct kobject *kobj,
687 struct attribute *attr,
691 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
693 LUSTRE_RW_ATTR(stats_track_ppid);
695 static ssize_t stats_track_gid_show(struct kobject *kobj,
696 struct attribute *attr,
699 return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
702 static ssize_t stats_track_gid_store(struct kobject *kobj,
703 struct attribute *attr,
707 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
709 LUSTRE_RW_ATTR(stats_track_gid);
711 static ssize_t statahead_running_max_show(struct kobject *kobj,
712 struct attribute *attr,
715 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
718 return snprintf(buf, 16, "%u\n", sbi->ll_sa_running_max);
721 static ssize_t statahead_running_max_store(struct kobject *kobj,
722 struct attribute *attr,
726 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
731 rc = kstrtoul(buffer, 0, &val);
735 if (val <= LL_SA_RUNNING_MAX) {
736 sbi->ll_sa_running_max = val;
740 CERROR("Bad statahead_running_max value %lu. Valid values "
741 "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
745 LUSTRE_RW_ATTR(statahead_running_max);
747 static ssize_t statahead_max_show(struct kobject *kobj,
748 struct attribute *attr,
751 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
754 return sprintf(buf, "%u\n", sbi->ll_sa_max);
757 static ssize_t statahead_max_store(struct kobject *kobj,
758 struct attribute *attr,
762 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
767 rc = kstrtoul(buffer, 0, &val);
771 if (val <= LL_SA_RPC_MAX)
772 sbi->ll_sa_max = val;
774 CERROR("Bad statahead_max value %lu. Valid values are in the range [0, %d]\n",
779 LUSTRE_RW_ATTR(statahead_max);
781 static ssize_t statahead_agl_show(struct kobject *kobj,
782 struct attribute *attr,
785 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
788 return sprintf(buf, "%u\n", sbi->ll_flags & LL_SBI_AGL_ENABLED ? 1 : 0);
791 static ssize_t statahead_agl_store(struct kobject *kobj,
792 struct attribute *attr,
796 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
801 rc = kstrtobool(buffer, &val);
806 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
808 sbi->ll_flags &= ~LL_SBI_AGL_ENABLED;
812 LUSTRE_RW_ATTR(statahead_agl);
814 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
816 struct super_block *sb = m->private;
817 struct ll_sb_info *sbi = ll_s2sbi(sb);
819 seq_printf(m, "statahead total: %u\n"
820 "statahead wrong: %u\n"
822 atomic_read(&sbi->ll_sa_total),
823 atomic_read(&sbi->ll_sa_wrong),
824 atomic_read(&sbi->ll_agl_total));
828 LDEBUGFS_SEQ_FOPS_RO(ll_statahead_stats);
830 static ssize_t lazystatfs_show(struct kobject *kobj,
831 struct attribute *attr,
834 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
837 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
840 static ssize_t lazystatfs_store(struct kobject *kobj,
841 struct attribute *attr,
845 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
850 rc = kstrtobool(buffer, &val);
855 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
857 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
861 LUSTRE_RW_ATTR(lazystatfs);
863 static ssize_t statfs_max_age_show(struct kobject *kobj, struct attribute *attr,
866 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
869 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_statfs_max_age);
872 static ssize_t statfs_max_age_store(struct kobject *kobj,
873 struct attribute *attr, const char *buffer,
876 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
881 rc = kstrtouint(buffer, 10, &val);
884 if (val > OBD_STATFS_CACHE_MAX_AGE)
887 sbi->ll_statfs_max_age = val;
891 LUSTRE_RW_ATTR(statfs_max_age);
893 static ssize_t max_easize_show(struct kobject *kobj,
894 struct attribute *attr,
897 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
902 rc = ll_get_max_mdsize(sbi, &ealen);
906 return sprintf(buf, "%u\n", ealen);
908 LUSTRE_RO_ATTR(max_easize);
911 * Get default_easize.
913 * \see client_obd::cl_default_mds_easize
915 * \param[in] m seq_file handle
916 * \param[in] v unused for single entry
918 * \retval 0 on success
919 * \retval negative negated errno on failure
921 static ssize_t default_easize_show(struct kobject *kobj,
922 struct attribute *attr,
925 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
930 rc = ll_get_default_mdsize(sbi, &ealen);
934 return sprintf(buf, "%u\n", ealen);
938 * Set default_easize.
940 * Range checking on the passed value is handled by
941 * ll_set_default_mdsize().
943 * \see client_obd::cl_default_mds_easize
945 * \param[in] file proc file
946 * \param[in] buffer string passed from user space
947 * \param[in] count \a buffer length
948 * \param[in] off unused for single entry
950 * \retval positive \a count on success
951 * \retval negative negated errno on failure
953 static ssize_t default_easize_store(struct kobject *kobj,
954 struct attribute *attr,
958 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
966 rc = kstrtouint(buffer, 10, &val);
970 rc = ll_set_default_mdsize(sbi, val);
976 LUSTRE_RW_ATTR(default_easize);
978 static int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
980 const char *str[] = LL_SBI_FLAGS;
981 struct super_block *sb = m->private;
982 int flags = ll_s2sbi(sb)->ll_flags;
986 if (ARRAY_SIZE(str) <= i) {
987 CERROR("%s: Revise array LL_SBI_FLAGS to match sbi "
988 "flags please.\n", ll_get_fsname(sb, NULL, 0));
993 seq_printf(m, "%s ", str[i]);
997 seq_printf(m, "\b\n");
1001 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
1003 static ssize_t xattr_cache_show(struct kobject *kobj,
1004 struct attribute *attr,
1007 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1010 return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
1013 static ssize_t xattr_cache_store(struct kobject *kobj,
1014 struct attribute *attr,
1018 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1023 rc = kstrtobool(buffer, &val);
1027 if (val && !(sbi->ll_flags & LL_SBI_XATTR_CACHE))
1030 sbi->ll_xattr_cache_enabled = val;
1031 sbi->ll_xattr_cache_set = 1;
1035 LUSTRE_RW_ATTR(xattr_cache);
1037 static ssize_t tiny_write_show(struct kobject *kobj,
1038 struct attribute *attr,
1041 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1044 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_TINY_WRITE));
1047 static ssize_t tiny_write_store(struct kobject *kobj,
1048 struct attribute *attr,
1052 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1057 rc = kstrtobool(buffer, &val);
1061 spin_lock(&sbi->ll_lock);
1063 sbi->ll_flags |= LL_SBI_TINY_WRITE;
1065 sbi->ll_flags &= ~LL_SBI_TINY_WRITE;
1066 spin_unlock(&sbi->ll_lock);
1070 LUSTRE_RW_ATTR(tiny_write);
1072 static ssize_t fast_read_show(struct kobject *kobj,
1073 struct attribute *attr,
1076 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1079 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_FAST_READ));
1082 static ssize_t fast_read_store(struct kobject *kobj,
1083 struct attribute *attr,
1087 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1092 rc = kstrtobool(buffer, &val);
1096 spin_lock(&sbi->ll_lock);
1098 sbi->ll_flags |= LL_SBI_FAST_READ;
1100 sbi->ll_flags &= ~LL_SBI_FAST_READ;
1101 spin_unlock(&sbi->ll_lock);
1105 LUSTRE_RW_ATTR(fast_read);
1107 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1109 struct super_block *sb = m->private;
1110 struct ll_sb_info *sbi = ll_s2sbi(sb);
1111 struct cl_client_cache *cache = sbi->ll_cache;
1115 pages = atomic_long_read(&cache->ccc_unstable_nr);
1116 mb = (pages * PAGE_SIZE) >> 20;
1118 seq_printf(m, "unstable_check: %8d\n"
1119 "unstable_pages: %12ld\n"
1120 "unstable_mb: %8d\n",
1121 cache->ccc_unstable_check, pages, mb);
1125 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1126 const char __user *buffer,
1127 size_t count, loff_t *unused)
1129 struct seq_file *seq = file->private_data;
1130 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1137 if (count >= sizeof(kernbuf))
1140 if (copy_from_user(kernbuf, buffer, count))
1144 buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1146 rc = kstrtobool_from_user(buffer, count, &val);
1150 /* borrow lru lock to set the value */
1151 spin_lock(&sbi->ll_cache->ccc_lru_lock);
1152 sbi->ll_cache->ccc_unstable_check = val;
1153 spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1158 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1160 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1162 struct super_block *sb = m->private;
1163 struct ll_sb_info *sbi = ll_s2sbi(sb);
1164 struct root_squash_info *squash = &sbi->ll_squash;
1166 seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1170 static ssize_t ll_root_squash_seq_write(struct file *file,
1171 const char __user *buffer,
1172 size_t count, loff_t *off)
1174 struct seq_file *m = file->private_data;
1175 struct super_block *sb = m->private;
1176 struct ll_sb_info *sbi = ll_s2sbi(sb);
1177 struct root_squash_info *squash = &sbi->ll_squash;
1179 return lprocfs_wr_root_squash(buffer, count, squash,
1180 ll_get_fsname(sb, NULL, 0));
1183 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1185 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1187 struct super_block *sb = m->private;
1188 struct ll_sb_info *sbi = ll_s2sbi(sb);
1189 struct root_squash_info *squash = &sbi->ll_squash;
1192 down_read(&squash->rsi_sem);
1193 if (!list_empty(&squash->rsi_nosquash_nids)) {
1194 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1195 &squash->rsi_nosquash_nids);
1199 seq_puts(m, "NONE\n");
1201 up_read(&squash->rsi_sem);
1206 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1207 const char __user *buffer,
1208 size_t count, loff_t *off)
1210 struct seq_file *m = file->private_data;
1211 struct super_block *sb = m->private;
1212 struct ll_sb_info *sbi = ll_s2sbi(sb);
1213 struct root_squash_info *squash = &sbi->ll_squash;
1216 rc = lprocfs_wr_nosquash_nids(buffer, count, squash,
1217 ll_get_fsname(sb, NULL, 0));
1221 ll_compute_rootsquash_state(sbi);
1226 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1228 struct lprocfs_vars lprocfs_llite_obd_vars[] = {
1230 .fops = &ll_site_stats_fops },
1231 { .name = "max_read_ahead_mb",
1232 .fops = &ll_max_readahead_mb_fops },
1233 { .name = "max_read_ahead_per_file_mb",
1234 .fops = &ll_max_readahead_per_file_mb_fops },
1235 { .name = "max_read_ahead_whole_mb",
1236 .fops = &ll_max_read_ahead_whole_mb_fops },
1237 { .name = "max_cached_mb",
1238 .fops = &ll_max_cached_mb_fops },
1239 { .name = "statahead_stats",
1240 .fops = &ll_statahead_stats_fops },
1241 { .name = "unstable_stats",
1242 .fops = &ll_unstable_stats_fops },
1243 { .name = "sbi_flags",
1244 .fops = &ll_sbi_flags_fops },
1245 { .name = "root_squash",
1246 .fops = &ll_root_squash_fops },
1247 { .name = "nosquash_nids",
1248 .fops = &ll_nosquash_nids_fops },
1252 #define MAX_STRING_SIZE 128
1254 static struct attribute *llite_attrs[] = {
1255 &lustre_attr_blocksize.attr,
1256 &lustre_attr_stat_blocksize.attr,
1257 &lustre_attr_kbytestotal.attr,
1258 &lustre_attr_kbytesfree.attr,
1259 &lustre_attr_kbytesavail.attr,
1260 &lustre_attr_filestotal.attr,
1261 &lustre_attr_filesfree.attr,
1262 &lustre_attr_client_type.attr,
1263 &lustre_attr_fstype.attr,
1264 &lustre_attr_uuid.attr,
1265 &lustre_attr_checksums.attr,
1266 &lustre_attr_checksum_pages.attr,
1267 &lustre_attr_stats_track_pid.attr,
1268 &lustre_attr_stats_track_ppid.attr,
1269 &lustre_attr_stats_track_gid.attr,
1270 &lustre_attr_statahead_running_max.attr,
1271 &lustre_attr_statahead_max.attr,
1272 &lustre_attr_statahead_agl.attr,
1273 &lustre_attr_lazystatfs.attr,
1274 &lustre_attr_statfs_max_age.attr,
1275 &lustre_attr_max_easize.attr,
1276 &lustre_attr_default_easize.attr,
1277 &lustre_attr_xattr_cache.attr,
1278 &lustre_attr_fast_read.attr,
1279 &lustre_attr_tiny_write.attr,
1283 static void llite_kobj_release(struct kobject *kobj)
1285 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1287 complete(&sbi->ll_kobj_unregister);
1290 static struct kobj_type llite_ktype = {
1291 .default_attrs = llite_attrs,
1292 .sysfs_ops = &lustre_sysfs_ops,
1293 .release = llite_kobj_release,
1296 static const struct llite_file_opcode {
1300 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1301 /* file operation */
1302 { LPROC_LL_DIRTY_HITS, LPROCFS_TYPE_REGS, "dirty_pages_hits" },
1303 { LPROC_LL_DIRTY_MISSES, LPROCFS_TYPE_REGS, "dirty_pages_misses" },
1304 { LPROC_LL_READ_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
1306 { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
1308 { LPROC_LL_BRW_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
1310 { LPROC_LL_BRW_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
1312 { LPROC_LL_IOCTL, LPROCFS_TYPE_REGS, "ioctl" },
1313 { LPROC_LL_OPEN, LPROCFS_TYPE_REGS, "open" },
1314 { LPROC_LL_RELEASE, LPROCFS_TYPE_REGS, "close" },
1315 { LPROC_LL_MAP, LPROCFS_TYPE_REGS, "mmap" },
1316 { LPROC_LL_FAULT, LPROCFS_TYPE_REGS, "page_fault" },
1317 { LPROC_LL_MKWRITE, LPROCFS_TYPE_REGS, "page_mkwrite" },
1318 { LPROC_LL_LLSEEK, LPROCFS_TYPE_REGS, "seek" },
1319 { LPROC_LL_FSYNC, LPROCFS_TYPE_REGS, "fsync" },
1320 { LPROC_LL_READDIR, LPROCFS_TYPE_REGS, "readdir" },
1321 /* inode operation */
1322 { LPROC_LL_SETATTR, LPROCFS_TYPE_REGS, "setattr" },
1323 { LPROC_LL_TRUNC, LPROCFS_TYPE_REGS, "truncate" },
1324 { LPROC_LL_FLOCK, LPROCFS_TYPE_REGS, "flock" },
1325 { LPROC_LL_GETATTR, LPROCFS_TYPE_REGS, "getattr" },
1326 /* dir inode operation */
1327 { LPROC_LL_CREATE, LPROCFS_TYPE_REGS, "create" },
1328 { LPROC_LL_LINK, LPROCFS_TYPE_REGS, "link" },
1329 { LPROC_LL_UNLINK, LPROCFS_TYPE_REGS, "unlink" },
1330 { LPROC_LL_SYMLINK, LPROCFS_TYPE_REGS, "symlink" },
1331 { LPROC_LL_MKDIR, LPROCFS_TYPE_REGS, "mkdir" },
1332 { LPROC_LL_RMDIR, LPROCFS_TYPE_REGS, "rmdir" },
1333 { LPROC_LL_MKNOD, LPROCFS_TYPE_REGS, "mknod" },
1334 { LPROC_LL_RENAME, LPROCFS_TYPE_REGS, "rename" },
1335 /* special inode operation */
1336 { LPROC_LL_STAFS, LPROCFS_TYPE_REGS, "statfs" },
1337 { LPROC_LL_ALLOC_INODE, LPROCFS_TYPE_REGS, "alloc_inode" },
1338 { LPROC_LL_SETXATTR, LPROCFS_TYPE_REGS, "setxattr" },
1339 { LPROC_LL_GETXATTR, LPROCFS_TYPE_REGS, "getxattr" },
1340 { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REGS, "getxattr_hits" },
1341 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_REGS, "listxattr" },
1342 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_REGS, "removexattr" },
1343 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_REGS, "inode_permission" },
1346 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
1350 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1351 lprocfs_counter_add(sbi->ll_stats, op, count);
1352 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1353 sbi->ll_stats_track_id == current->pid)
1354 lprocfs_counter_add(sbi->ll_stats, op, count);
1355 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1356 sbi->ll_stats_track_id == current->parent->pid)
1357 lprocfs_counter_add(sbi->ll_stats, op, count);
1358 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1359 sbi->ll_stats_track_id ==
1360 from_kgid(&init_user_ns, current_gid()))
1361 lprocfs_counter_add(sbi->ll_stats, op, count);
1363 EXPORT_SYMBOL(ll_stats_ops_tally);
1365 static const char *ra_stat_string[] = {
1366 [RA_STAT_HIT] = "hits",
1367 [RA_STAT_MISS] = "misses",
1368 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1369 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1370 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1371 [RA_STAT_FAILED_MATCH] = "failed lock match",
1372 [RA_STAT_DISCARDED] = "read but discarded",
1373 [RA_STAT_ZERO_LEN] = "zero length file",
1374 [RA_STAT_ZERO_WINDOW] = "zero size window",
1375 [RA_STAT_EOF] = "read-ahead to EOF",
1376 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1377 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1378 [RA_STAT_FAILED_REACH_END] = "failed to reach end"
1381 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1383 struct lustre_sb_info *lsi = s2lsi(sb);
1384 struct ll_sb_info *sbi = ll_s2sbi(sb);
1390 if (IS_ERR_OR_NULL(llite_root))
1393 sbi->ll_debugfs_entry = ldebugfs_register(name, llite_root,
1394 lprocfs_llite_obd_vars, sb);
1395 if (IS_ERR_OR_NULL(sbi->ll_debugfs_entry)) {
1396 err = sbi->ll_debugfs_entry ? PTR_ERR(sbi->ll_debugfs_entry) :
1398 sbi->ll_debugfs_entry = NULL;
1402 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "dump_page_cache",0444,
1403 &vvp_dump_pgcache_file_ops, sbi);
1405 CWARN("Error adding the dump_page_cache file\n");
1407 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "extents_stats", 0644,
1408 &ll_rw_extents_stats_fops, sbi);
1410 CWARN("Error adding the extent_stats file\n");
1412 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry,
1413 "extents_stats_per_process", 0644,
1414 &ll_rw_extents_stats_pp_fops, sbi);
1416 CWARN("Error adding the extents_stats_per_process file\n");
1418 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "offset_stats", 0644,
1419 &ll_rw_offset_stats_fops, sbi);
1421 CWARN("Error adding the offset_stats file\n");
1423 /* File operations stats */
1424 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1425 LPROCFS_STATS_FLAG_NONE);
1426 if (sbi->ll_stats == NULL)
1427 GOTO(out_debugfs, err = -ENOMEM);
1429 /* do counter init */
1430 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1431 u32 type = llite_opcode_table[id].type;
1434 if (type & LPROCFS_TYPE_REGS)
1436 else if (type & LPROCFS_TYPE_BYTES)
1438 else if (type & LPROCFS_TYPE_PAGES)
1440 lprocfs_counter_init(sbi->ll_stats,
1441 llite_opcode_table[id].opcode,
1442 (type & LPROCFS_CNTR_AVGMINMAX),
1443 llite_opcode_table[id].opname, ptr);
1446 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "stats",
1449 GOTO(out_stats, err);
1451 sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1452 LPROCFS_STATS_FLAG_NONE);
1453 if (sbi->ll_ra_stats == NULL)
1454 GOTO(out_stats, err = -ENOMEM);
1456 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1457 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1458 ra_stat_string[id], "pages");
1460 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "read_ahead_stats",
1463 GOTO(out_ra_stats, err);
1466 /* Yes we also register sysfs mount kset here as well */
1467 sbi->ll_kset.kobj.parent = llite_kobj;
1468 sbi->ll_kset.kobj.ktype = &llite_ktype;
1469 init_completion(&sbi->ll_kobj_unregister);
1470 err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1472 GOTO(out_ra_stats, err);
1474 err = kset_register(&sbi->ll_kset);
1476 GOTO(out_ra_stats, err);
1478 lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1482 lprocfs_free_stats(&sbi->ll_ra_stats);
1484 lprocfs_free_stats(&sbi->ll_stats);
1486 ldebugfs_remove(&sbi->ll_debugfs_entry);
1491 void ll_debugfs_unregister_super(struct super_block *sb)
1493 struct lustre_sb_info *lsi = s2lsi(sb);
1494 struct ll_sb_info *sbi = ll_s2sbi(sb);
1496 if (!IS_ERR_OR_NULL(sbi->ll_debugfs_entry))
1497 ldebugfs_remove(&sbi->ll_debugfs_entry);
1500 sysfs_remove_link(&sbi->ll_kset.kobj,
1501 sbi->ll_dt_obd->obd_type->typ_name);
1504 sysfs_remove_link(&sbi->ll_kset.kobj,
1505 sbi->ll_md_obd->obd_type->typ_name);
1507 kobject_put(lsi->lsi_kobj);
1509 kset_unregister(&sbi->ll_kset);
1510 wait_for_completion(&sbi->ll_kobj_unregister);
1512 lprocfs_free_stats(&sbi->ll_ra_stats);
1513 lprocfs_free_stats(&sbi->ll_stats);
1515 #undef MAX_STRING_SIZE
1517 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1518 struct seq_file *seq, int which)
1520 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1521 unsigned long start, end, r, w;
1522 char *unitp = "KMGTPEZY";
1524 struct per_process_info *pp_info = &io_extents->pp_extents[which];
1530 for(i = 0; i < LL_HIST_MAX; i++) {
1531 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1532 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1535 for(i = 0; i < LL_HIST_MAX; i++) {
1536 r = pp_info->pp_r_hist.oh_buckets[i];
1537 w = pp_info->pp_w_hist.oh_buckets[i];
1540 end = BIT(i + LL_HIST_START - units);
1541 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u | "
1542 "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1543 (i == LL_HIST_MAX - 1) ? '+' : ' ',
1544 r, pct(r, read_tot), pct(read_cum, read_tot),
1545 w, pct(w, write_tot), pct(write_cum, write_tot));
1547 if (start == BIT(10)) {
1552 if (read_cum == read_tot && write_cum == write_tot)
1557 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1559 struct timespec64 now;
1560 struct ll_sb_info *sbi = seq->private;
1561 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1564 ktime_get_real_ts64(&now);
1566 if (!sbi->ll_rw_stats_on) {
1567 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1570 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1571 (s64)now.tv_sec, now.tv_nsec);
1572 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1573 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1574 "extents", "calls", "%", "cum%",
1575 "calls", "%", "cum%");
1576 spin_lock(&sbi->ll_pp_extent_lock);
1577 for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1578 if (io_extents->pp_extents[k].pid != 0) {
1579 seq_printf(seq, "\nPID: %d\n",
1580 io_extents->pp_extents[k].pid);
1581 ll_display_extents_info(io_extents, seq, k);
1584 spin_unlock(&sbi->ll_pp_extent_lock);
1588 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1589 const char __user *buf,
1593 struct seq_file *seq = file->private_data;
1594 struct ll_sb_info *sbi = seq->private;
1595 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1602 value = ll_stats_pid_write(buf, len);
1605 sbi->ll_rw_stats_on = 0;
1607 sbi->ll_rw_stats_on = 1;
1609 spin_lock(&sbi->ll_pp_extent_lock);
1610 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1611 io_extents->pp_extents[i].pid = 0;
1612 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1613 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1615 spin_unlock(&sbi->ll_pp_extent_lock);
1619 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1621 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1623 struct timespec64 now;
1624 struct ll_sb_info *sbi = seq->private;
1625 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1627 ktime_get_real_ts64(&now);
1629 if (!sbi->ll_rw_stats_on) {
1630 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1633 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1634 (s64)now.tv_sec, now.tv_nsec);
1636 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1637 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1638 "extents", "calls", "%", "cum%",
1639 "calls", "%", "cum%");
1640 spin_lock(&sbi->ll_lock);
1641 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1642 spin_unlock(&sbi->ll_lock);
1647 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1648 const char __user *buf,
1649 size_t len, loff_t *off)
1651 struct seq_file *seq = file->private_data;
1652 struct ll_sb_info *sbi = seq->private;
1653 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1660 value = ll_stats_pid_write(buf, len);
1663 sbi->ll_rw_stats_on = 0;
1665 sbi->ll_rw_stats_on = 1;
1667 spin_lock(&sbi->ll_pp_extent_lock);
1668 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1669 io_extents->pp_extents[i].pid = 0;
1670 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1671 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1673 spin_unlock(&sbi->ll_pp_extent_lock);
1678 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1680 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1681 struct ll_file_data *file, loff_t pos,
1682 size_t count, int rw)
1685 struct ll_rw_process_info *process;
1686 struct ll_rw_process_info *offset;
1687 int *off_count = &sbi->ll_rw_offset_entry_count;
1688 int *process_count = &sbi->ll_offset_process_count;
1689 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1691 if(!sbi->ll_rw_stats_on)
1693 process = sbi->ll_rw_process_info;
1694 offset = sbi->ll_rw_offset_info;
1696 spin_lock(&sbi->ll_pp_extent_lock);
1697 /* Extent statistics */
1698 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1699 if(io_extents->pp_extents[i].pid == pid) {
1707 sbi->ll_extent_process_count =
1708 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1709 cur = sbi->ll_extent_process_count;
1710 io_extents->pp_extents[cur].pid = pid;
1711 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1712 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1715 for (i = 0; (count >= BIT(LL_HIST_START + i)) &&
1716 (i < (LL_HIST_MAX - 1)); i++);
1718 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1719 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1721 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1722 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1724 spin_unlock(&sbi->ll_pp_extent_lock);
1726 spin_lock(&sbi->ll_process_lock);
1727 /* Offset statistics */
1728 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1729 if (process[i].rw_pid == pid) {
1730 if (process[i].rw_last_file != file) {
1731 process[i].rw_range_start = pos;
1732 process[i].rw_last_file_pos = pos + count;
1733 process[i].rw_smallest_extent = count;
1734 process[i].rw_largest_extent = count;
1735 process[i].rw_offset = 0;
1736 process[i].rw_last_file = file;
1737 spin_unlock(&sbi->ll_process_lock);
1740 if (process[i].rw_last_file_pos != pos) {
1742 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1743 offset[*off_count].rw_op = process[i].rw_op;
1744 offset[*off_count].rw_pid = pid;
1745 offset[*off_count].rw_range_start =
1746 process[i].rw_range_start;
1747 offset[*off_count].rw_range_end =
1748 process[i].rw_last_file_pos;
1749 offset[*off_count].rw_smallest_extent =
1750 process[i].rw_smallest_extent;
1751 offset[*off_count].rw_largest_extent =
1752 process[i].rw_largest_extent;
1753 offset[*off_count].rw_offset =
1754 process[i].rw_offset;
1755 process[i].rw_op = rw;
1756 process[i].rw_range_start = pos;
1757 process[i].rw_smallest_extent = count;
1758 process[i].rw_largest_extent = count;
1759 process[i].rw_offset = pos -
1760 process[i].rw_last_file_pos;
1762 if(process[i].rw_smallest_extent > count)
1763 process[i].rw_smallest_extent = count;
1764 if(process[i].rw_largest_extent < count)
1765 process[i].rw_largest_extent = count;
1766 process[i].rw_last_file_pos = pos + count;
1767 spin_unlock(&sbi->ll_process_lock);
1771 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1772 process[*process_count].rw_pid = pid;
1773 process[*process_count].rw_op = rw;
1774 process[*process_count].rw_range_start = pos;
1775 process[*process_count].rw_last_file_pos = pos + count;
1776 process[*process_count].rw_smallest_extent = count;
1777 process[*process_count].rw_largest_extent = count;
1778 process[*process_count].rw_offset = 0;
1779 process[*process_count].rw_last_file = file;
1780 spin_unlock(&sbi->ll_process_lock);
1783 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1785 struct timespec64 now;
1786 struct ll_sb_info *sbi = seq->private;
1787 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1788 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1791 ktime_get_real_ts64(&now);
1793 if (!sbi->ll_rw_stats_on) {
1794 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1797 spin_lock(&sbi->ll_process_lock);
1799 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1800 (s64)now.tv_sec, now.tv_nsec);
1801 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1802 "R/W", "PID", "RANGE START", "RANGE END",
1803 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1805 /* We stored the discontiguous offsets here; print them first */
1806 for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1807 if (offset[i].rw_pid != 0)
1809 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
1810 offset[i].rw_op == READ ? 'R' : 'W',
1812 offset[i].rw_range_start,
1813 offset[i].rw_range_end,
1814 (unsigned long)offset[i].rw_smallest_extent,
1815 (unsigned long)offset[i].rw_largest_extent,
1816 offset[i].rw_offset);
1819 /* Then print the current offsets for each process */
1820 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1821 if (process[i].rw_pid != 0)
1823 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
1824 process[i].rw_op == READ ? 'R' : 'W',
1826 process[i].rw_range_start,
1827 process[i].rw_last_file_pos,
1828 (unsigned long)process[i].rw_smallest_extent,
1829 (unsigned long)process[i].rw_largest_extent,
1830 process[i].rw_offset);
1832 spin_unlock(&sbi->ll_process_lock);
1837 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
1838 const char __user *buf,
1839 size_t len, loff_t *off)
1841 struct seq_file *seq = file->private_data;
1842 struct ll_sb_info *sbi = seq->private;
1843 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1844 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1850 value = ll_stats_pid_write(buf, len);
1853 sbi->ll_rw_stats_on = 0;
1855 sbi->ll_rw_stats_on = 1;
1857 spin_lock(&sbi->ll_process_lock);
1858 sbi->ll_offset_process_count = 0;
1859 sbi->ll_rw_offset_entry_count = 0;
1860 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1861 LL_PROCESS_HIST_MAX);
1862 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1863 LL_OFFSET_HIST_MAX);
1864 spin_unlock(&sbi->ll_process_lock);
1869 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);