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 #include <linux/uidgid.h>
38 #include <uapi/linux/lustre/lustre_param.h>
39 #include <lprocfs_status.h>
40 #include <obd_support.h>
42 #include "llite_internal.h"
43 #include "vvp_internal.h"
45 static struct kobject *llite_kobj;
46 static struct dentry *llite_root;
48 static void llite_kobj_release(struct kobject *kobj)
50 if (!IS_ERR_OR_NULL(llite_root)) {
51 debugfs_remove(llite_root);
58 static struct kobj_type llite_kobj_ktype = {
59 .release = llite_kobj_release,
60 .sysfs_ops = &lustre_sysfs_ops,
63 int llite_tunables_register(void)
67 llite_kobj = kzalloc(sizeof(*llite_kobj), GFP_KERNEL);
71 llite_kobj->kset = lustre_kset;
72 rc = kobject_init_and_add(llite_kobj, &llite_kobj_ktype,
73 &lustre_kset->kobj, "%s", "llite");
77 llite_root = debugfs_create_dir("llite", debugfs_lustre_root);
78 if (IS_ERR_OR_NULL(llite_root)) {
79 rc = llite_root ? PTR_ERR(llite_root) : -ENOMEM;
82 kobject_put(llite_kobj);
89 void llite_tunables_unregister(void)
91 kobject_put(llite_kobj);
95 /* <debugfs>/lustre/llite mount point registration */
96 static const struct file_operations ll_rw_extents_stats_fops;
97 static const struct file_operations ll_rw_extents_stats_pp_fops;
98 static const struct file_operations ll_rw_offset_stats_fops;
101 * ll_stats_pid_write() - Determine if stats collection should be enabled
102 * @buf: Buffer containing the data written
103 * @len: Number of bytes in the buffer
105 * Several proc files begin collecting stats when a value is written, and stop
106 * collecting when either '0' or 'disable' is written. This function checks the
107 * written value to see if collection should be enabled or disabled.
109 * Return: If '0' or 'disable' is provided, 0 is returned. If the text
110 * equivalent of a number is written, that number is returned. Otherwise,
111 * 1 is returned. Non-zero return values indicate collection should be enabled.
113 static s64 ll_stats_pid_write(const char __user *buf, size_t len)
115 unsigned long long value = 1;
119 rc = kstrtoull_from_user(buf, len, 0, &value);
120 if (rc < 0 && len < sizeof(kernbuf)) {
121 if (copy_from_user(kernbuf, buf, len))
125 if (kernbuf[len - 1] == '\n')
126 kernbuf[len - 1] = 0;
128 if (strncasecmp(kernbuf, "disable", 7) == 0)
135 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
138 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
140 struct obd_statfs osfs;
143 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
147 return sprintf(buf, "%u\n", osfs.os_bsize);
149 LUSTRE_RO_ATTR(blocksize);
151 static ssize_t stat_blocksize_show(struct kobject *kobj, struct attribute *attr,
154 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
157 return sprintf(buf, "%u\n", sbi->ll_stat_blksize);
160 static ssize_t stat_blocksize_store(struct kobject *kobj,
161 struct attribute *attr,
165 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
170 rc = kstrtouint(buffer, 10, &val);
174 if (val != 0 && (val < PAGE_SIZE || (val & (val - 1))) != 0)
177 sbi->ll_stat_blksize = val;
181 LUSTRE_RW_ATTR(stat_blocksize);
183 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
186 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
188 struct obd_statfs osfs;
193 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
197 blk_size = osfs.os_bsize >> 10;
198 result = osfs.os_blocks;
200 while (blk_size >>= 1)
203 return sprintf(buf, "%llu\n", result);
205 LUSTRE_RO_ATTR(kbytestotal);
207 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
210 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
212 struct obd_statfs osfs;
217 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
221 blk_size = osfs.os_bsize >> 10;
222 result = osfs.os_bfree;
224 while (blk_size >>= 1)
227 return sprintf(buf, "%llu\n", result);
229 LUSTRE_RO_ATTR(kbytesfree);
231 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
234 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
236 struct obd_statfs osfs;
241 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
245 blk_size = osfs.os_bsize >> 10;
246 result = osfs.os_bavail;
248 while (blk_size >>= 1)
251 return sprintf(buf, "%llu\n", result);
253 LUSTRE_RO_ATTR(kbytesavail);
255 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
258 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
260 struct obd_statfs osfs;
263 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
267 return sprintf(buf, "%llu\n", osfs.os_files);
269 LUSTRE_RO_ATTR(filestotal);
271 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
274 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
276 struct obd_statfs osfs;
279 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
283 return sprintf(buf, "%llu\n", osfs.os_ffree);
285 LUSTRE_RO_ATTR(filesfree);
287 static ssize_t client_type_show(struct kobject *kobj, struct attribute *attr,
290 return sprintf(buf, "local client\n");
292 LUSTRE_RO_ATTR(client_type);
294 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
297 return sprintf(buf, "lustre\n");
299 LUSTRE_RO_ATTR(fstype);
301 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
304 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
307 return sprintf(buf, "%s\n", sbi->ll_sb_uuid.uuid);
309 LUSTRE_RO_ATTR(uuid);
311 static int ll_site_stats_seq_show(struct seq_file *m, void *v)
313 struct super_block *sb = m->private;
316 * See description of statistical counters in struct cl_site, and
319 return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site), m);
322 LDEBUGFS_SEQ_FOPS_RO(ll_site_stats);
324 static ssize_t max_read_ahead_mb_show(struct kobject *kobj,
325 struct attribute *attr, char *buf)
327 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
329 unsigned long ra_max_mb;
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);
335 return snprintf(buf, PAGE_SIZE, "%lu\n", ra_max_mb);
338 static ssize_t max_read_ahead_mb_store(struct kobject *kobj,
339 struct attribute *attr,
340 const char *buffer, size_t count)
342 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
344 u64 ra_max_mb, pages_number;
347 rc = sysfs_memparse(buffer, count, &ra_max_mb, "MiB");
351 pages_number = round_up(ra_max_mb, 1024 * 1024) >> PAGE_SHIFT;
352 CDEBUG(D_INFO, "%s: set max_read_ahead_mb=%llu (%llu pages)\n",
353 sbi->ll_fsname, PAGES_TO_MiB(pages_number), pages_number);
354 if (pages_number > cfs_totalram_pages() / 2) {
356 CERROR("%s: cannot set max_read_ahead_mb=%llu > totalram/2=%luMB\n",
357 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
358 PAGES_TO_MiB(cfs_totalram_pages() / 2));
362 spin_lock(&sbi->ll_lock);
363 sbi->ll_ra_info.ra_max_pages = pages_number;
364 spin_unlock(&sbi->ll_lock);
368 LUSTRE_RW_ATTR(max_read_ahead_mb);
370 static ssize_t max_read_ahead_per_file_mb_show(struct kobject *kobj,
371 struct attribute *attr,
374 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
376 unsigned long ra_max_file_mb;
378 spin_lock(&sbi->ll_lock);
379 ra_max_file_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file);
380 spin_unlock(&sbi->ll_lock);
382 return snprintf(buf, PAGE_SIZE, "%lu\n", ra_max_file_mb);
385 static ssize_t max_read_ahead_per_file_mb_store(struct kobject *kobj,
386 struct attribute *attr,
390 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
392 u64 ra_max_file_mb, pages_number;
395 rc = sysfs_memparse(buffer, count, &ra_max_file_mb, "MiB");
399 pages_number = round_up(ra_max_file_mb, 1024 * 1024) >> PAGE_SHIFT;
400 if (pages_number > sbi->ll_ra_info.ra_max_pages) {
401 CERROR("%s: cannot set max_read_ahead_per_file_mb=%llu > max_read_ahead_mb=%lu\n",
402 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
403 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
407 spin_lock(&sbi->ll_lock);
408 sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
409 spin_unlock(&sbi->ll_lock);
413 LUSTRE_RW_ATTR(max_read_ahead_per_file_mb);
415 static ssize_t max_read_ahead_whole_mb_show(struct kobject *kobj,
416 struct attribute *attr, char *buf)
418 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
420 unsigned long ra_max_whole_mb;
422 spin_lock(&sbi->ll_lock);
423 ra_max_whole_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_read_ahead_whole_pages);
424 spin_unlock(&sbi->ll_lock);
426 return snprintf(buf, PAGE_SIZE, "%lu\n", ra_max_whole_mb);
429 static ssize_t max_read_ahead_whole_mb_store(struct kobject *kobj,
430 struct attribute *attr,
431 const char *buffer, size_t count)
433 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
435 u64 ra_max_whole_mb, pages_number;
438 rc = sysfs_memparse(buffer, count, &ra_max_whole_mb, "MiB");
442 pages_number = round_up(ra_max_whole_mb, 1024 * 1024) >> PAGE_SHIFT;
443 /* Cap this at the current max readahead window size, the readahead
444 * algorithm does this anyway so it's pointless to set it larger.
446 if (pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
447 CERROR("%s: cannot set max_read_ahead_whole_mb=%llu > max_read_ahead_per_file_mb=%lu\n",
448 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
449 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
454 spin_lock(&sbi->ll_lock);
455 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
456 spin_unlock(&sbi->ll_lock);
460 LUSTRE_RW_ATTR(max_read_ahead_whole_mb);
462 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
464 struct super_block *sb = m->private;
465 struct ll_sb_info *sbi = ll_s2sbi(sb);
466 struct cl_client_cache *cache = sbi->ll_cache;
470 mutex_lock(&cache->ccc_max_cache_mb_lock);
471 max_cached_mb = PAGES_TO_MiB(cache->ccc_lru_max);
472 unused_mb = PAGES_TO_MiB(atomic_long_read(&cache->ccc_lru_left));
473 mutex_unlock(&cache->ccc_max_cache_mb_lock);
474 seq_printf(m, "users: %d\n"
475 "max_cached_mb: %ld\n"
478 "reclaim_count: %u\n",
479 atomic_read(&cache->ccc_users),
481 max_cached_mb - unused_mb,
483 cache->ccc_lru_shrinkers);
487 static ssize_t ll_max_cached_mb_seq_write(struct file *file,
488 const char __user *buffer,
489 size_t count, loff_t *off)
491 struct seq_file *m = file->private_data;
492 struct super_block *sb = m->private;
493 struct ll_sb_info *sbi = ll_s2sbi(sb);
494 struct cl_client_cache *cache = sbi->ll_cache;
501 char kernbuf[128], *ptr;
504 if (count >= sizeof(kernbuf))
507 if (copy_from_user(kernbuf, buffer, count))
509 kernbuf[count] = '\0';
511 ptr = lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count);
512 rc = sysfs_memparse(ptr, count, &pages_number, "MiB");
516 pages_number >>= PAGE_SHIFT;
518 if (pages_number < 0 || pages_number > cfs_totalram_pages()) {
519 CERROR("%s: can't set max cache more than %lu MB\n",
521 PAGES_TO_MiB(cfs_totalram_pages()));
524 /* Allow enough cache so clients can make well-formed RPCs */
525 pages_number = max_t(long, pages_number, PTLRPC_MAX_BRW_PAGES);
527 mutex_lock(&cache->ccc_max_cache_mb_lock);
528 diff = pages_number - cache->ccc_lru_max;
530 /* easy - add more LRU slots. */
532 atomic_long_add(diff, &cache->ccc_lru_left);
536 env = cl_env_get(&refcheck);
538 GOTO(out_unlock, rc = PTR_ERR(env));
544 /* reduce LRU budget from free slots. */
546 long lru_left_old, lru_left_new, lru_left_ret;
548 lru_left_old = atomic_long_read(&cache->ccc_lru_left);
549 if (lru_left_old == 0)
552 lru_left_new = lru_left_old > diff ?
553 lru_left_old - diff : 0;
555 atomic_long_cmpxchg(&cache->ccc_lru_left,
558 if (likely(lru_left_old == lru_left_ret)) {
559 diff -= lru_left_old - lru_left_new;
560 nrpages += lru_left_old - lru_left_new;
568 if (sbi->ll_dt_exp == NULL) { /* being initialized */
573 /* Request extra free slots to avoid them all being used
574 * by other processes before this can continue shrinking.
576 tmp = diff + min_t(long, diff, MiB_TO_PAGES(1024));
577 /* difficult - have to ask OSCs to drop LRU slots. */
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);
585 cl_env_put(env, &refcheck);
589 cache->ccc_lru_max = pages_number;
592 atomic_long_add(nrpages, &cache->ccc_lru_left);
595 mutex_unlock(&cache->ccc_max_cache_mb_lock);
598 LDEBUGFS_SEQ_FOPS(ll_max_cached_mb);
600 static ssize_t checksums_show(struct kobject *kobj, struct attribute *attr,
603 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
606 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
609 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
610 const char *buffer, size_t count)
612 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
622 rc = kstrtobool(buffer, &val);
626 sbi->ll_flags |= LL_SBI_CHECKSUM;
628 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
631 rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
632 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
634 CWARN("Failed to set OSC checksum flags: %d\n", rc);
638 LUSTRE_RW_ATTR(checksums);
640 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
642 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
643 enum stats_track_type type)
645 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
648 if (sbi->ll_stats_track_type == type)
649 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
650 else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
651 return sprintf(buf, "0 (all)\n");
653 return sprintf(buf, "untracked\n");
656 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
657 size_t count, enum stats_track_type type)
659 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
664 rc = kstrtoul(buffer, 10, &pid);
668 sbi->ll_stats_track_id = pid;
670 sbi->ll_stats_track_type = STATS_TRACK_ALL;
672 sbi->ll_stats_track_type = type;
673 lprocfs_clear_stats(sbi->ll_stats);
677 static ssize_t stats_track_pid_show(struct kobject *kobj,
678 struct attribute *attr,
681 return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
684 static ssize_t stats_track_pid_store(struct kobject *kobj,
685 struct attribute *attr,
689 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
691 LUSTRE_RW_ATTR(stats_track_pid);
693 static ssize_t stats_track_ppid_show(struct kobject *kobj,
694 struct attribute *attr,
697 return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
700 static ssize_t stats_track_ppid_store(struct kobject *kobj,
701 struct attribute *attr,
705 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
707 LUSTRE_RW_ATTR(stats_track_ppid);
709 static ssize_t stats_track_gid_show(struct kobject *kobj,
710 struct attribute *attr,
713 return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
716 static ssize_t stats_track_gid_store(struct kobject *kobj,
717 struct attribute *attr,
721 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
723 LUSTRE_RW_ATTR(stats_track_gid);
725 static ssize_t statahead_running_max_show(struct kobject *kobj,
726 struct attribute *attr,
729 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
732 return snprintf(buf, 16, "%u\n", sbi->ll_sa_running_max);
735 static ssize_t statahead_running_max_store(struct kobject *kobj,
736 struct attribute *attr,
740 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
745 rc = kstrtoul(buffer, 0, &val);
749 if (val <= LL_SA_RUNNING_MAX) {
750 sbi->ll_sa_running_max = val;
754 CERROR("Bad statahead_running_max value %lu. Valid values "
755 "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
759 LUSTRE_RW_ATTR(statahead_running_max);
761 static ssize_t statahead_max_show(struct kobject *kobj,
762 struct attribute *attr,
765 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
768 return sprintf(buf, "%u\n", sbi->ll_sa_max);
771 static ssize_t statahead_max_store(struct kobject *kobj,
772 struct attribute *attr,
776 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
781 rc = kstrtoul(buffer, 0, &val);
785 if (val <= LL_SA_RPC_MAX)
786 sbi->ll_sa_max = val;
788 CERROR("Bad statahead_max value %lu. Valid values are in the range [0, %d]\n",
793 LUSTRE_RW_ATTR(statahead_max);
795 static ssize_t statahead_agl_show(struct kobject *kobj,
796 struct attribute *attr,
799 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
802 return sprintf(buf, "%u\n", sbi->ll_flags & LL_SBI_AGL_ENABLED ? 1 : 0);
805 static ssize_t statahead_agl_store(struct kobject *kobj,
806 struct attribute *attr,
810 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
815 rc = kstrtobool(buffer, &val);
820 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
822 sbi->ll_flags &= ~LL_SBI_AGL_ENABLED;
826 LUSTRE_RW_ATTR(statahead_agl);
828 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
830 struct super_block *sb = m->private;
831 struct ll_sb_info *sbi = ll_s2sbi(sb);
833 seq_printf(m, "statahead total: %u\n"
834 "statahead wrong: %u\n"
836 atomic_read(&sbi->ll_sa_total),
837 atomic_read(&sbi->ll_sa_wrong),
838 atomic_read(&sbi->ll_agl_total));
842 LDEBUGFS_SEQ_FOPS_RO(ll_statahead_stats);
844 static ssize_t lazystatfs_show(struct kobject *kobj,
845 struct attribute *attr,
848 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
851 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
854 static ssize_t lazystatfs_store(struct kobject *kobj,
855 struct attribute *attr,
859 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
864 rc = kstrtobool(buffer, &val);
869 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
871 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
875 LUSTRE_RW_ATTR(lazystatfs);
877 static ssize_t statfs_max_age_show(struct kobject *kobj, struct attribute *attr,
880 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
883 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_statfs_max_age);
886 static ssize_t statfs_max_age_store(struct kobject *kobj,
887 struct attribute *attr, const char *buffer,
890 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
895 rc = kstrtouint(buffer, 10, &val);
898 if (val > OBD_STATFS_CACHE_MAX_AGE)
901 sbi->ll_statfs_max_age = val;
905 LUSTRE_RW_ATTR(statfs_max_age);
907 static ssize_t max_easize_show(struct kobject *kobj,
908 struct attribute *attr,
911 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
916 rc = ll_get_max_mdsize(sbi, &ealen);
920 /* Limit xattr size returned to userspace based on kernel maximum */
921 return snprintf(buf, PAGE_SIZE, "%u\n",
922 ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
924 LUSTRE_RO_ATTR(max_easize);
927 * Get default_easize.
929 * \see client_obd::cl_default_mds_easize
931 * \param[in] m seq_file handle
932 * \param[in] v unused for single entry
934 * \retval 0 on success
935 * \retval negative negated errno on failure
937 static ssize_t default_easize_show(struct kobject *kobj,
938 struct attribute *attr,
941 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
946 rc = ll_get_default_mdsize(sbi, &ealen);
950 /* Limit xattr size returned to userspace based on kernel maximum */
951 return snprintf(buf, PAGE_SIZE, "%u\n",
952 ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
956 * Set default_easize.
958 * Range checking on the passed value is handled by
959 * ll_set_default_mdsize().
961 * \see client_obd::cl_default_mds_easize
963 * \param[in] file proc file
964 * \param[in] buffer string passed from user space
965 * \param[in] count \a buffer length
966 * \param[in] off unused for single entry
968 * \retval positive \a count on success
969 * \retval negative negated errno on failure
971 static ssize_t default_easize_store(struct kobject *kobj,
972 struct attribute *attr,
976 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
984 rc = kstrtouint(buffer, 10, &val);
988 rc = ll_set_default_mdsize(sbi, val);
994 LUSTRE_RW_ATTR(default_easize);
996 static int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
998 const char *str[] = LL_SBI_FLAGS;
999 struct super_block *sb = m->private;
1000 int flags = ll_s2sbi(sb)->ll_flags;
1003 while (flags != 0) {
1004 if (ARRAY_SIZE(str) <= i) {
1005 CERROR("%s: Revise array LL_SBI_FLAGS to match sbi "
1006 "flags please.\n", ll_s2sbi(sb)->ll_fsname);
1011 seq_printf(m, "%s ", str[i]);
1015 seq_printf(m, "\b\n");
1019 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
1021 static ssize_t xattr_cache_show(struct kobject *kobj,
1022 struct attribute *attr,
1025 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1028 return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
1031 static ssize_t xattr_cache_store(struct kobject *kobj,
1032 struct attribute *attr,
1036 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1041 rc = kstrtobool(buffer, &val);
1045 if (val && !(sbi->ll_flags & LL_SBI_XATTR_CACHE))
1048 sbi->ll_xattr_cache_enabled = val;
1049 sbi->ll_xattr_cache_set = 1;
1053 LUSTRE_RW_ATTR(xattr_cache);
1055 static ssize_t tiny_write_show(struct kobject *kobj,
1056 struct attribute *attr,
1059 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1062 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_TINY_WRITE));
1065 static ssize_t tiny_write_store(struct kobject *kobj,
1066 struct attribute *attr,
1070 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1075 rc = kstrtobool(buffer, &val);
1079 spin_lock(&sbi->ll_lock);
1081 sbi->ll_flags |= LL_SBI_TINY_WRITE;
1083 sbi->ll_flags &= ~LL_SBI_TINY_WRITE;
1084 spin_unlock(&sbi->ll_lock);
1088 LUSTRE_RW_ATTR(tiny_write);
1090 static ssize_t max_read_ahead_async_active_show(struct kobject *kobj,
1091 struct attribute *attr,
1094 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1097 return snprintf(buf, PAGE_SIZE, "%u\n",
1098 sbi->ll_ra_info.ra_async_max_active);
1101 static ssize_t max_read_ahead_async_active_store(struct kobject *kobj,
1102 struct attribute *attr,
1106 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1111 rc = kstrtouint(buffer, 10, &val);
1116 * It doesn't make any sense to make it exceed what
1117 * workqueue could acutally support. This can easily
1118 * over subscripe the cores but Lustre internally
1119 * throttles to avoid those impacts.
1121 if (val > WQ_UNBOUND_MAX_ACTIVE) {
1122 CERROR("%s: cannot set max_read_ahead_async_active=%u larger than %u\n",
1123 sbi->ll_fsname, val, WQ_UNBOUND_MAX_ACTIVE);
1127 spin_lock(&sbi->ll_lock);
1128 sbi->ll_ra_info.ra_async_max_active = val;
1129 spin_unlock(&sbi->ll_lock);
1133 LUSTRE_RW_ATTR(max_read_ahead_async_active);
1135 static ssize_t read_ahead_async_file_threshold_mb_show(struct kobject *kobj,
1136 struct attribute *attr,
1139 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1142 return snprintf(buf, PAGE_SIZE, "%lu\n",
1143 PAGES_TO_MiB(sbi->ll_ra_info.ra_async_pages_per_file_threshold));
1147 read_ahead_async_file_threshold_mb_store(struct kobject *kobj,
1148 struct attribute *attr,
1149 const char *buffer, size_t count)
1151 unsigned long pages_number;
1152 unsigned long max_ra_per_file;
1153 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1157 rc = kstrtoul(buffer, 10, &pages_number);
1161 pages_number = MiB_TO_PAGES(pages_number);
1162 max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1163 if (pages_number < 0 || pages_number > max_ra_per_file) {
1164 CERROR("%s: can't set read_ahead_async_file_threshold_mb=%lu > "
1165 "max_read_readahead_per_file_mb=%lu\n", sbi->ll_fsname,
1166 PAGES_TO_MiB(pages_number),
1167 PAGES_TO_MiB(max_ra_per_file));
1170 sbi->ll_ra_info.ra_async_pages_per_file_threshold = pages_number;
1174 LUSTRE_RW_ATTR(read_ahead_async_file_threshold_mb);
1176 static ssize_t fast_read_show(struct kobject *kobj,
1177 struct attribute *attr,
1180 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1183 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_FAST_READ));
1186 static ssize_t fast_read_store(struct kobject *kobj,
1187 struct attribute *attr,
1191 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1196 rc = kstrtobool(buffer, &val);
1200 spin_lock(&sbi->ll_lock);
1202 sbi->ll_flags |= LL_SBI_FAST_READ;
1204 sbi->ll_flags &= ~LL_SBI_FAST_READ;
1205 spin_unlock(&sbi->ll_lock);
1209 LUSTRE_RW_ATTR(fast_read);
1211 static ssize_t file_heat_show(struct kobject *kobj,
1212 struct attribute *attr,
1215 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1218 return snprintf(buf, PAGE_SIZE, "%u\n",
1219 !!(sbi->ll_flags & LL_SBI_FILE_HEAT));
1222 static ssize_t file_heat_store(struct kobject *kobj,
1223 struct attribute *attr,
1227 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1232 rc = kstrtobool(buffer, &val);
1236 spin_lock(&sbi->ll_lock);
1238 sbi->ll_flags |= LL_SBI_FILE_HEAT;
1240 sbi->ll_flags &= ~LL_SBI_FILE_HEAT;
1241 spin_unlock(&sbi->ll_lock);
1245 LUSTRE_RW_ATTR(file_heat);
1247 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1248 struct attribute *attr,
1251 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1254 return snprintf(buf, PAGE_SIZE, "%u\n",
1255 (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1258 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1259 struct attribute *attr,
1263 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1268 rc = kstrtoul(buffer, 10, &val);
1272 if (val < 0 || val > 100)
1275 sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1279 LUSTRE_RW_ATTR(heat_decay_percentage);
1281 static ssize_t heat_period_second_show(struct kobject *kobj,
1282 struct attribute *attr,
1285 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1288 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1291 static ssize_t heat_period_second_store(struct kobject *kobj,
1292 struct attribute *attr,
1296 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1301 rc = kstrtoul(buffer, 10, &val);
1308 sbi->ll_heat_period_second = val;
1312 LUSTRE_RW_ATTR(heat_period_second);
1314 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1316 struct super_block *sb = m->private;
1317 struct ll_sb_info *sbi = ll_s2sbi(sb);
1318 struct cl_client_cache *cache = sbi->ll_cache;
1322 pages = atomic_long_read(&cache->ccc_unstable_nr);
1323 mb = (pages * PAGE_SIZE) >> 20;
1325 seq_printf(m, "unstable_check: %8d\n"
1326 "unstable_pages: %12ld\n"
1327 "unstable_mb: %8d\n",
1328 cache->ccc_unstable_check, pages, mb);
1332 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1333 const char __user *buffer,
1334 size_t count, loff_t *unused)
1336 struct seq_file *seq = file->private_data;
1337 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1344 if (count >= sizeof(kernbuf))
1347 if (copy_from_user(kernbuf, buffer, count))
1351 buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1353 rc = kstrtobool_from_user(buffer, count, &val);
1357 /* borrow lru lock to set the value */
1358 spin_lock(&sbi->ll_cache->ccc_lru_lock);
1359 sbi->ll_cache->ccc_unstable_check = val;
1360 spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1365 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1367 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1369 struct super_block *sb = m->private;
1370 struct ll_sb_info *sbi = ll_s2sbi(sb);
1371 struct root_squash_info *squash = &sbi->ll_squash;
1373 seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1377 static ssize_t ll_root_squash_seq_write(struct file *file,
1378 const char __user *buffer,
1379 size_t count, loff_t *off)
1381 struct seq_file *m = file->private_data;
1382 struct super_block *sb = m->private;
1383 struct ll_sb_info *sbi = ll_s2sbi(sb);
1384 struct root_squash_info *squash = &sbi->ll_squash;
1386 return lprocfs_wr_root_squash(buffer, count, squash, sbi->ll_fsname);
1389 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1391 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1393 struct super_block *sb = m->private;
1394 struct ll_sb_info *sbi = ll_s2sbi(sb);
1395 struct root_squash_info *squash = &sbi->ll_squash;
1398 spin_lock(&squash->rsi_lock);
1399 if (!list_empty(&squash->rsi_nosquash_nids)) {
1400 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1401 &squash->rsi_nosquash_nids);
1405 seq_puts(m, "NONE\n");
1407 spin_unlock(&squash->rsi_lock);
1412 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1413 const char __user *buffer,
1414 size_t count, loff_t *off)
1416 struct seq_file *m = file->private_data;
1417 struct super_block *sb = m->private;
1418 struct ll_sb_info *sbi = ll_s2sbi(sb);
1419 struct root_squash_info *squash = &sbi->ll_squash;
1422 rc = lprocfs_wr_nosquash_nids(buffer, count, squash, sbi->ll_fsname);
1426 ll_compute_rootsquash_state(sbi);
1431 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1433 static int ll_pcc_seq_show(struct seq_file *m, void *v)
1435 struct super_block *sb = m->private;
1436 struct ll_sb_info *sbi = ll_s2sbi(sb);
1438 return pcc_super_dump(&sbi->ll_pcc_super, m);
1441 static ssize_t ll_pcc_seq_write(struct file *file, const char __user *buffer,
1442 size_t count, loff_t *off)
1444 struct seq_file *m = file->private_data;
1445 struct super_block *sb = m->private;
1446 struct ll_sb_info *sbi = ll_s2sbi(sb);
1450 if (count >= LPROCFS_WR_PCC_MAX_CMD)
1453 if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_PCC))
1456 OBD_ALLOC(kernbuf, count + 1);
1457 if (kernbuf == NULL)
1460 if (copy_from_user(kernbuf, buffer, count))
1461 GOTO(out_free_kernbuff, rc = -EFAULT);
1463 rc = pcc_cmd_handle(kernbuf, count, &sbi->ll_pcc_super);
1465 OBD_FREE(kernbuf, count + 1);
1466 return rc ? rc : count;
1468 LPROC_SEQ_FOPS(ll_pcc);
1470 struct lprocfs_vars lprocfs_llite_obd_vars[] = {
1472 .fops = &ll_site_stats_fops },
1473 { .name = "max_cached_mb",
1474 .fops = &ll_max_cached_mb_fops },
1475 { .name = "statahead_stats",
1476 .fops = &ll_statahead_stats_fops },
1477 { .name = "unstable_stats",
1478 .fops = &ll_unstable_stats_fops },
1479 { .name = "sbi_flags",
1480 .fops = &ll_sbi_flags_fops },
1481 { .name = "root_squash",
1482 .fops = &ll_root_squash_fops },
1483 { .name = "nosquash_nids",
1484 .fops = &ll_nosquash_nids_fops },
1486 .fops = &ll_pcc_fops, },
1490 #define MAX_STRING_SIZE 128
1492 static struct attribute *llite_attrs[] = {
1493 &lustre_attr_blocksize.attr,
1494 &lustre_attr_stat_blocksize.attr,
1495 &lustre_attr_kbytestotal.attr,
1496 &lustre_attr_kbytesfree.attr,
1497 &lustre_attr_kbytesavail.attr,
1498 &lustre_attr_filestotal.attr,
1499 &lustre_attr_filesfree.attr,
1500 &lustre_attr_client_type.attr,
1501 &lustre_attr_fstype.attr,
1502 &lustre_attr_uuid.attr,
1503 &lustre_attr_checksums.attr,
1504 &lustre_attr_checksum_pages.attr,
1505 &lustre_attr_max_read_ahead_mb.attr,
1506 &lustre_attr_max_read_ahead_per_file_mb.attr,
1507 &lustre_attr_max_read_ahead_whole_mb.attr,
1508 &lustre_attr_max_read_ahead_async_active.attr,
1509 &lustre_attr_read_ahead_async_file_threshold_mb.attr,
1510 &lustre_attr_stats_track_pid.attr,
1511 &lustre_attr_stats_track_ppid.attr,
1512 &lustre_attr_stats_track_gid.attr,
1513 &lustre_attr_statahead_running_max.attr,
1514 &lustre_attr_statahead_max.attr,
1515 &lustre_attr_statahead_agl.attr,
1516 &lustre_attr_lazystatfs.attr,
1517 &lustre_attr_statfs_max_age.attr,
1518 &lustre_attr_max_easize.attr,
1519 &lustre_attr_default_easize.attr,
1520 &lustre_attr_xattr_cache.attr,
1521 &lustre_attr_fast_read.attr,
1522 &lustre_attr_tiny_write.attr,
1523 &lustre_attr_file_heat.attr,
1524 &lustre_attr_heat_decay_percentage.attr,
1525 &lustre_attr_heat_period_second.attr,
1529 static void sbi_kobj_release(struct kobject *kobj)
1531 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1533 complete(&sbi->ll_kobj_unregister);
1536 static struct kobj_type sbi_ktype = {
1537 .default_attrs = llite_attrs,
1538 .sysfs_ops = &lustre_sysfs_ops,
1539 .release = sbi_kobj_release,
1542 #define LPROCFS_TYPE_LATENCY \
1543 (LPROCFS_TYPE_USEC | LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV)
1544 static const struct llite_file_opcode {
1548 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1549 /* file operation */
1550 { LPROC_LL_READ_BYTES, LPROCFS_CNTR_AVGMINMAX | LPROCFS_TYPE_BYTES,
1552 { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX | LPROCFS_TYPE_BYTES,
1554 { LPROC_LL_READ, LPROCFS_TYPE_LATENCY, "read" },
1555 { LPROC_LL_WRITE, LPROCFS_TYPE_LATENCY, "write" },
1556 { LPROC_LL_IOCTL, LPROCFS_TYPE_REQS, "ioctl" },
1557 { LPROC_LL_OPEN, LPROCFS_TYPE_LATENCY, "open" },
1558 { LPROC_LL_RELEASE, LPROCFS_TYPE_LATENCY, "close" },
1559 { LPROC_LL_MMAP, LPROCFS_TYPE_LATENCY, "mmap" },
1560 { LPROC_LL_FAULT, LPROCFS_TYPE_LATENCY, "page_fault" },
1561 { LPROC_LL_MKWRITE, LPROCFS_TYPE_LATENCY, "page_mkwrite" },
1562 { LPROC_LL_LLSEEK, LPROCFS_TYPE_LATENCY, "seek" },
1563 { LPROC_LL_FSYNC, LPROCFS_TYPE_LATENCY, "fsync" },
1564 { LPROC_LL_READDIR, LPROCFS_TYPE_LATENCY, "readdir" },
1565 /* inode operation */
1566 { LPROC_LL_SETATTR, LPROCFS_TYPE_LATENCY, "setattr" },
1567 { LPROC_LL_TRUNC, LPROCFS_TYPE_LATENCY, "truncate" },
1568 { LPROC_LL_FLOCK, LPROCFS_TYPE_LATENCY, "flock" },
1569 { LPROC_LL_GETATTR, LPROCFS_TYPE_LATENCY, "getattr" },
1570 /* dir inode operation */
1571 { LPROC_LL_CREATE, LPROCFS_TYPE_LATENCY, "create" },
1572 { LPROC_LL_LINK, LPROCFS_TYPE_LATENCY, "link" },
1573 { LPROC_LL_UNLINK, LPROCFS_TYPE_LATENCY, "unlink" },
1574 { LPROC_LL_SYMLINK, LPROCFS_TYPE_LATENCY, "symlink" },
1575 { LPROC_LL_MKDIR, LPROCFS_TYPE_LATENCY, "mkdir" },
1576 { LPROC_LL_RMDIR, LPROCFS_TYPE_LATENCY, "rmdir" },
1577 { LPROC_LL_MKNOD, LPROCFS_TYPE_LATENCY, "mknod" },
1578 { LPROC_LL_RENAME, LPROCFS_TYPE_LATENCY, "rename" },
1579 /* special inode operation */
1580 { LPROC_LL_STATFS, LPROCFS_TYPE_LATENCY, "statfs" },
1581 { LPROC_LL_SETXATTR, LPROCFS_TYPE_LATENCY, "setxattr" },
1582 { LPROC_LL_GETXATTR, LPROCFS_TYPE_LATENCY, "getxattr" },
1583 { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REQS, "getxattr_hits" },
1584 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_LATENCY, "listxattr" },
1585 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_LATENCY, "removexattr" },
1586 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_LATENCY, "inode_permission" },
1589 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count)
1594 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1595 lprocfs_counter_add(sbi->ll_stats, op, count);
1596 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1597 sbi->ll_stats_track_id == current->pid)
1598 lprocfs_counter_add(sbi->ll_stats, op, count);
1599 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1600 sbi->ll_stats_track_id == current->parent->pid)
1601 lprocfs_counter_add(sbi->ll_stats, op, count);
1602 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1603 sbi->ll_stats_track_id ==
1604 from_kgid(&init_user_ns, current_gid()))
1605 lprocfs_counter_add(sbi->ll_stats, op, count);
1607 EXPORT_SYMBOL(ll_stats_ops_tally);
1609 static const char *ra_stat_string[] = {
1610 [RA_STAT_HIT] = "hits",
1611 [RA_STAT_MISS] = "misses",
1612 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1613 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1614 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1615 [RA_STAT_FAILED_MATCH] = "failed lock match",
1616 [RA_STAT_DISCARDED] = "read but discarded",
1617 [RA_STAT_ZERO_LEN] = "zero length file",
1618 [RA_STAT_ZERO_WINDOW] = "zero size window",
1619 [RA_STAT_EOF] = "read-ahead to EOF",
1620 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1621 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1622 [RA_STAT_FAILED_REACH_END] = "failed to reach end",
1623 [RA_STAT_ASYNC] = "async readahead",
1624 [RA_STAT_FAILED_FAST_READ] = "failed to fast read",
1627 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1629 struct lustre_sb_info *lsi = s2lsi(sb);
1630 struct ll_sb_info *sbi = ll_s2sbi(sb);
1636 if (IS_ERR_OR_NULL(llite_root))
1639 sbi->ll_debugfs_entry = debugfs_create_dir(name, llite_root);
1640 ldebugfs_add_vars(sbi->ll_debugfs_entry, lprocfs_llite_obd_vars, sb);
1642 debugfs_create_file("dump_page_cache", 0444, sbi->ll_debugfs_entry, sbi,
1643 &vvp_dump_pgcache_file_ops);
1645 debugfs_create_file("extents_stats", 0644, sbi->ll_debugfs_entry, sbi,
1646 &ll_rw_extents_stats_fops);
1648 debugfs_create_file("extents_stats_per_process", 0644,
1649 sbi->ll_debugfs_entry, sbi,
1650 &ll_rw_extents_stats_pp_fops);
1652 debugfs_create_file("offset_stats", 0644, sbi->ll_debugfs_entry, sbi,
1653 &ll_rw_offset_stats_fops);
1655 /* File operations stats */
1656 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1657 LPROCFS_STATS_FLAG_NONE);
1658 if (sbi->ll_stats == NULL)
1659 GOTO(out_debugfs, err = -ENOMEM);
1661 /* do counter init */
1662 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1663 u32 type = llite_opcode_table[id].type;
1666 if (type & LPROCFS_TYPE_REQS)
1668 else if (type & LPROCFS_TYPE_BYTES)
1670 else if (type & LPROCFS_TYPE_PAGES)
1672 else if (type & LPROCFS_TYPE_USEC)
1674 lprocfs_counter_init(sbi->ll_stats,
1675 llite_opcode_table[id].opcode,
1676 (type & LPROCFS_CNTR_AVGMINMAX),
1677 llite_opcode_table[id].opname, ptr);
1680 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "stats",
1683 GOTO(out_stats, err);
1685 sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1686 LPROCFS_STATS_FLAG_NONE);
1687 if (sbi->ll_ra_stats == NULL)
1688 GOTO(out_stats, err = -ENOMEM);
1690 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1691 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1692 ra_stat_string[id], "pages");
1694 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "read_ahead_stats",
1697 GOTO(out_ra_stats, err);
1700 /* Yes we also register sysfs mount kset here as well */
1701 sbi->ll_kset.kobj.parent = llite_kobj;
1702 sbi->ll_kset.kobj.ktype = &sbi_ktype;
1703 init_completion(&sbi->ll_kobj_unregister);
1704 err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1706 GOTO(out_ra_stats, err);
1708 err = kset_register(&sbi->ll_kset);
1710 GOTO(out_ra_stats, err);
1712 lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1716 lprocfs_free_stats(&sbi->ll_ra_stats);
1718 lprocfs_free_stats(&sbi->ll_stats);
1720 debugfs_remove_recursive(sbi->ll_debugfs_entry);
1725 void ll_debugfs_unregister_super(struct super_block *sb)
1727 struct lustre_sb_info *lsi = s2lsi(sb);
1728 struct ll_sb_info *sbi = ll_s2sbi(sb);
1730 debugfs_remove_recursive(sbi->ll_debugfs_entry);
1733 sysfs_remove_link(&sbi->ll_kset.kobj,
1734 sbi->ll_dt_obd->obd_type->typ_name);
1737 sysfs_remove_link(&sbi->ll_kset.kobj,
1738 sbi->ll_md_obd->obd_type->typ_name);
1740 kobject_put(lsi->lsi_kobj);
1742 kset_unregister(&sbi->ll_kset);
1743 wait_for_completion(&sbi->ll_kobj_unregister);
1745 lprocfs_free_stats(&sbi->ll_ra_stats);
1746 lprocfs_free_stats(&sbi->ll_stats);
1748 #undef MAX_STRING_SIZE
1750 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1751 struct seq_file *seq, int which)
1753 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1754 unsigned long start, end, r, w;
1755 char *unitp = "KMGTPEZY";
1757 struct per_process_info *pp_info = &io_extents->pp_extents[which];
1763 for(i = 0; i < LL_HIST_MAX; i++) {
1764 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1765 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1768 for(i = 0; i < LL_HIST_MAX; i++) {
1769 r = pp_info->pp_r_hist.oh_buckets[i];
1770 w = pp_info->pp_w_hist.oh_buckets[i];
1773 end = BIT(i + LL_HIST_START - units);
1774 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u | "
1775 "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1776 (i == LL_HIST_MAX - 1) ? '+' : ' ',
1777 r, pct(r, read_tot), pct(read_cum, read_tot),
1778 w, pct(w, write_tot), pct(write_cum, write_tot));
1780 if (start == BIT(10)) {
1785 if (read_cum == read_tot && write_cum == write_tot)
1790 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1792 struct timespec64 now;
1793 struct ll_sb_info *sbi = seq->private;
1794 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1797 ktime_get_real_ts64(&now);
1799 if (!sbi->ll_rw_stats_on) {
1800 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1803 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1804 (s64)now.tv_sec, now.tv_nsec);
1805 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1806 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1807 "extents", "calls", "%", "cum%",
1808 "calls", "%", "cum%");
1809 spin_lock(&sbi->ll_pp_extent_lock);
1810 for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1811 if (io_extents->pp_extents[k].pid != 0) {
1812 seq_printf(seq, "\nPID: %d\n",
1813 io_extents->pp_extents[k].pid);
1814 ll_display_extents_info(io_extents, seq, k);
1817 spin_unlock(&sbi->ll_pp_extent_lock);
1821 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1822 const char __user *buf,
1826 struct seq_file *seq = file->private_data;
1827 struct ll_sb_info *sbi = seq->private;
1828 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1835 value = ll_stats_pid_write(buf, len);
1838 sbi->ll_rw_stats_on = 0;
1840 sbi->ll_rw_stats_on = 1;
1842 spin_lock(&sbi->ll_pp_extent_lock);
1843 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1844 io_extents->pp_extents[i].pid = 0;
1845 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1846 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1848 spin_unlock(&sbi->ll_pp_extent_lock);
1852 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1854 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1856 struct timespec64 now;
1857 struct ll_sb_info *sbi = seq->private;
1858 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1860 ktime_get_real_ts64(&now);
1862 if (!sbi->ll_rw_stats_on) {
1863 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1866 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1867 (s64)now.tv_sec, now.tv_nsec);
1869 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1870 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1871 "extents", "calls", "%", "cum%",
1872 "calls", "%", "cum%");
1873 spin_lock(&sbi->ll_lock);
1874 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1875 spin_unlock(&sbi->ll_lock);
1880 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1881 const char __user *buf,
1882 size_t len, loff_t *off)
1884 struct seq_file *seq = file->private_data;
1885 struct ll_sb_info *sbi = seq->private;
1886 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1893 value = ll_stats_pid_write(buf, len);
1896 sbi->ll_rw_stats_on = 0;
1898 sbi->ll_rw_stats_on = 1;
1900 spin_lock(&sbi->ll_pp_extent_lock);
1901 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1902 io_extents->pp_extents[i].pid = 0;
1903 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1904 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1906 spin_unlock(&sbi->ll_pp_extent_lock);
1911 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1913 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1914 struct ll_file_data *file, loff_t pos,
1915 size_t count, int rw)
1918 struct ll_rw_process_info *process;
1919 struct ll_rw_process_info *offset;
1920 int *off_count = &sbi->ll_rw_offset_entry_count;
1921 int *process_count = &sbi->ll_offset_process_count;
1922 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1924 if(!sbi->ll_rw_stats_on)
1926 process = sbi->ll_rw_process_info;
1927 offset = sbi->ll_rw_offset_info;
1929 spin_lock(&sbi->ll_pp_extent_lock);
1930 /* Extent statistics */
1931 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1932 if(io_extents->pp_extents[i].pid == pid) {
1940 sbi->ll_extent_process_count =
1941 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1942 cur = sbi->ll_extent_process_count;
1943 io_extents->pp_extents[cur].pid = pid;
1944 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1945 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1948 for (i = 0; (count >= BIT(LL_HIST_START + i)) &&
1949 (i < (LL_HIST_MAX - 1)); i++);
1951 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1952 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1954 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1955 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1957 spin_unlock(&sbi->ll_pp_extent_lock);
1959 spin_lock(&sbi->ll_process_lock);
1960 /* Offset statistics */
1961 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1962 if (process[i].rw_pid == pid) {
1963 if (process[i].rw_last_file != file) {
1964 process[i].rw_range_start = pos;
1965 process[i].rw_last_file_pos = pos + count;
1966 process[i].rw_smallest_extent = count;
1967 process[i].rw_largest_extent = count;
1968 process[i].rw_offset = 0;
1969 process[i].rw_last_file = file;
1970 spin_unlock(&sbi->ll_process_lock);
1973 if (process[i].rw_last_file_pos != pos) {
1975 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1976 offset[*off_count].rw_op = process[i].rw_op;
1977 offset[*off_count].rw_pid = pid;
1978 offset[*off_count].rw_range_start =
1979 process[i].rw_range_start;
1980 offset[*off_count].rw_range_end =
1981 process[i].rw_last_file_pos;
1982 offset[*off_count].rw_smallest_extent =
1983 process[i].rw_smallest_extent;
1984 offset[*off_count].rw_largest_extent =
1985 process[i].rw_largest_extent;
1986 offset[*off_count].rw_offset =
1987 process[i].rw_offset;
1988 process[i].rw_op = rw;
1989 process[i].rw_range_start = pos;
1990 process[i].rw_smallest_extent = count;
1991 process[i].rw_largest_extent = count;
1992 process[i].rw_offset = pos -
1993 process[i].rw_last_file_pos;
1995 if(process[i].rw_smallest_extent > count)
1996 process[i].rw_smallest_extent = count;
1997 if(process[i].rw_largest_extent < count)
1998 process[i].rw_largest_extent = count;
1999 process[i].rw_last_file_pos = pos + count;
2000 spin_unlock(&sbi->ll_process_lock);
2004 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
2005 process[*process_count].rw_pid = pid;
2006 process[*process_count].rw_op = rw;
2007 process[*process_count].rw_range_start = pos;
2008 process[*process_count].rw_last_file_pos = pos + count;
2009 process[*process_count].rw_smallest_extent = count;
2010 process[*process_count].rw_largest_extent = count;
2011 process[*process_count].rw_offset = 0;
2012 process[*process_count].rw_last_file = file;
2013 spin_unlock(&sbi->ll_process_lock);
2016 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
2018 struct timespec64 now;
2019 struct ll_sb_info *sbi = seq->private;
2020 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
2021 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
2024 ktime_get_real_ts64(&now);
2026 if (!sbi->ll_rw_stats_on) {
2027 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2030 spin_lock(&sbi->ll_process_lock);
2032 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
2033 (s64)now.tv_sec, now.tv_nsec);
2034 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
2035 "R/W", "PID", "RANGE START", "RANGE END",
2036 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
2038 /* We stored the discontiguous offsets here; print them first */
2039 for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
2040 if (offset[i].rw_pid != 0)
2042 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2043 offset[i].rw_op == READ ? 'R' : 'W',
2045 offset[i].rw_range_start,
2046 offset[i].rw_range_end,
2047 (unsigned long)offset[i].rw_smallest_extent,
2048 (unsigned long)offset[i].rw_largest_extent,
2049 offset[i].rw_offset);
2052 /* Then print the current offsets for each process */
2053 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2054 if (process[i].rw_pid != 0)
2056 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2057 process[i].rw_op == READ ? 'R' : 'W',
2059 process[i].rw_range_start,
2060 process[i].rw_last_file_pos,
2061 (unsigned long)process[i].rw_smallest_extent,
2062 (unsigned long)process[i].rw_largest_extent,
2063 process[i].rw_offset);
2065 spin_unlock(&sbi->ll_process_lock);
2070 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
2071 const char __user *buf,
2072 size_t len, loff_t *off)
2074 struct seq_file *seq = file->private_data;
2075 struct ll_sb_info *sbi = seq->private;
2076 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
2077 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
2083 value = ll_stats_pid_write(buf, len);
2086 sbi->ll_rw_stats_on = 0;
2088 sbi->ll_rw_stats_on = 1;
2090 spin_lock(&sbi->ll_process_lock);
2091 sbi->ll_offset_process_count = 0;
2092 sbi->ll_rw_offset_entry_count = 0;
2093 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
2094 LL_PROCESS_HIST_MAX);
2095 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
2096 LL_OFFSET_HIST_MAX);
2097 spin_unlock(&sbi->ll_process_lock);
2102 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);