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 max_cached_mb = PAGES_TO_MiB(cache->ccc_lru_max);
471 unused_mb = PAGES_TO_MiB(atomic_long_read(&cache->ccc_lru_left));
472 seq_printf(m, "users: %d\n"
473 "max_cached_mb: %ld\n"
476 "reclaim_count: %u\n",
477 atomic_read(&cache->ccc_users),
479 max_cached_mb - unused_mb,
481 cache->ccc_lru_shrinkers);
485 static ssize_t ll_max_cached_mb_seq_write(struct file *file,
486 const char __user *buffer,
487 size_t count, loff_t *off)
489 struct seq_file *m = file->private_data;
490 struct super_block *sb = m->private;
491 struct ll_sb_info *sbi = ll_s2sbi(sb);
492 struct cl_client_cache *cache = sbi->ll_cache;
499 char kernbuf[128], *ptr;
502 if (count >= sizeof(kernbuf))
505 if (copy_from_user(kernbuf, buffer, count))
507 kernbuf[count] = '\0';
509 ptr = lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count);
510 rc = sysfs_memparse(ptr, count, &pages_number, "MiB");
514 pages_number >>= PAGE_SHIFT;
516 if (pages_number < 0 || pages_number > cfs_totalram_pages()) {
517 CERROR("%s: can't set max cache more than %lu MB\n",
519 PAGES_TO_MiB(cfs_totalram_pages()));
522 /* Allow enough cache so clients can make well-formed RPCs */
523 pages_number = max_t(long, pages_number, PTLRPC_MAX_BRW_PAGES);
525 spin_lock(&sbi->ll_lock);
526 diff = pages_number - cache->ccc_lru_max;
527 spin_unlock(&sbi->ll_lock);
529 /* easy - add more LRU slots. */
531 atomic_long_add(diff, &cache->ccc_lru_left);
535 env = cl_env_get(&refcheck);
537 RETURN(PTR_ERR(env));
543 /* reduce LRU budget from free slots. */
547 ov = atomic_long_read(&cache->ccc_lru_left);
551 nv = ov > diff ? ov - diff : 0;
552 retv = atomic_long_cmpxchg(&cache->ccc_lru_left,
554 if (likely(ov == retv)) {
564 if (sbi->ll_dt_exp == NULL) { /* being initialized */
569 /* difficult - have to ask OSCs to drop LRU slots. */
571 rc = obd_set_info_async(env, sbi->ll_dt_exp,
572 sizeof(KEY_CACHE_LRU_SHRINK),
573 KEY_CACHE_LRU_SHRINK,
574 sizeof(tmp), &tmp, NULL);
578 cl_env_put(env, &refcheck);
582 spin_lock(&sbi->ll_lock);
583 cache->ccc_lru_max = pages_number;
584 spin_unlock(&sbi->ll_lock);
587 atomic_long_add(nrpages, &cache->ccc_lru_left);
591 LDEBUGFS_SEQ_FOPS(ll_max_cached_mb);
593 static ssize_t checksums_show(struct kobject *kobj, struct attribute *attr,
596 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
599 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
602 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
603 const char *buffer, size_t count)
605 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
615 rc = kstrtobool(buffer, &val);
619 sbi->ll_flags |= LL_SBI_CHECKSUM;
621 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
624 rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
625 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
627 CWARN("Failed to set OSC checksum flags: %d\n", rc);
631 LUSTRE_RW_ATTR(checksums);
633 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
635 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
636 enum stats_track_type type)
638 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
641 if (sbi->ll_stats_track_type == type)
642 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
643 else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
644 return sprintf(buf, "0 (all)\n");
646 return sprintf(buf, "untracked\n");
649 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
650 size_t count, enum stats_track_type type)
652 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
657 rc = kstrtoul(buffer, 10, &pid);
661 sbi->ll_stats_track_id = pid;
663 sbi->ll_stats_track_type = STATS_TRACK_ALL;
665 sbi->ll_stats_track_type = type;
666 lprocfs_clear_stats(sbi->ll_stats);
670 static ssize_t stats_track_pid_show(struct kobject *kobj,
671 struct attribute *attr,
674 return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
677 static ssize_t stats_track_pid_store(struct kobject *kobj,
678 struct attribute *attr,
682 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
684 LUSTRE_RW_ATTR(stats_track_pid);
686 static ssize_t stats_track_ppid_show(struct kobject *kobj,
687 struct attribute *attr,
690 return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
693 static ssize_t stats_track_ppid_store(struct kobject *kobj,
694 struct attribute *attr,
698 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
700 LUSTRE_RW_ATTR(stats_track_ppid);
702 static ssize_t stats_track_gid_show(struct kobject *kobj,
703 struct attribute *attr,
706 return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
709 static ssize_t stats_track_gid_store(struct kobject *kobj,
710 struct attribute *attr,
714 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
716 LUSTRE_RW_ATTR(stats_track_gid);
718 static ssize_t statahead_running_max_show(struct kobject *kobj,
719 struct attribute *attr,
722 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
725 return snprintf(buf, 16, "%u\n", sbi->ll_sa_running_max);
728 static ssize_t statahead_running_max_store(struct kobject *kobj,
729 struct attribute *attr,
733 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
738 rc = kstrtoul(buffer, 0, &val);
742 if (val <= LL_SA_RUNNING_MAX) {
743 sbi->ll_sa_running_max = val;
747 CERROR("Bad statahead_running_max value %lu. Valid values "
748 "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
752 LUSTRE_RW_ATTR(statahead_running_max);
754 static ssize_t statahead_max_show(struct kobject *kobj,
755 struct attribute *attr,
758 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
761 return sprintf(buf, "%u\n", sbi->ll_sa_max);
764 static ssize_t statahead_max_store(struct kobject *kobj,
765 struct attribute *attr,
769 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
774 rc = kstrtoul(buffer, 0, &val);
778 if (val <= LL_SA_RPC_MAX)
779 sbi->ll_sa_max = val;
781 CERROR("Bad statahead_max value %lu. Valid values are in the range [0, %d]\n",
786 LUSTRE_RW_ATTR(statahead_max);
788 static ssize_t statahead_agl_show(struct kobject *kobj,
789 struct attribute *attr,
792 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
795 return sprintf(buf, "%u\n", sbi->ll_flags & LL_SBI_AGL_ENABLED ? 1 : 0);
798 static ssize_t statahead_agl_store(struct kobject *kobj,
799 struct attribute *attr,
803 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
808 rc = kstrtobool(buffer, &val);
813 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
815 sbi->ll_flags &= ~LL_SBI_AGL_ENABLED;
819 LUSTRE_RW_ATTR(statahead_agl);
821 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
823 struct super_block *sb = m->private;
824 struct ll_sb_info *sbi = ll_s2sbi(sb);
826 seq_printf(m, "statahead total: %u\n"
827 "statahead wrong: %u\n"
829 atomic_read(&sbi->ll_sa_total),
830 atomic_read(&sbi->ll_sa_wrong),
831 atomic_read(&sbi->ll_agl_total));
835 LDEBUGFS_SEQ_FOPS_RO(ll_statahead_stats);
837 static ssize_t lazystatfs_show(struct kobject *kobj,
838 struct attribute *attr,
841 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
844 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
847 static ssize_t lazystatfs_store(struct kobject *kobj,
848 struct attribute *attr,
852 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
857 rc = kstrtobool(buffer, &val);
862 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
864 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
868 LUSTRE_RW_ATTR(lazystatfs);
870 static ssize_t statfs_max_age_show(struct kobject *kobj, struct attribute *attr,
873 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
876 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_statfs_max_age);
879 static ssize_t statfs_max_age_store(struct kobject *kobj,
880 struct attribute *attr, const char *buffer,
883 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
888 rc = kstrtouint(buffer, 10, &val);
891 if (val > OBD_STATFS_CACHE_MAX_AGE)
894 sbi->ll_statfs_max_age = val;
898 LUSTRE_RW_ATTR(statfs_max_age);
900 static ssize_t max_easize_show(struct kobject *kobj,
901 struct attribute *attr,
904 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
909 rc = ll_get_max_mdsize(sbi, &ealen);
913 /* Limit xattr size returned to userspace based on kernel maximum */
914 return snprintf(buf, PAGE_SIZE, "%u\n",
915 ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
917 LUSTRE_RO_ATTR(max_easize);
920 * Get default_easize.
922 * \see client_obd::cl_default_mds_easize
924 * \param[in] m seq_file handle
925 * \param[in] v unused for single entry
927 * \retval 0 on success
928 * \retval negative negated errno on failure
930 static ssize_t default_easize_show(struct kobject *kobj,
931 struct attribute *attr,
934 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
939 rc = ll_get_default_mdsize(sbi, &ealen);
943 /* Limit xattr size returned to userspace based on kernel maximum */
944 return snprintf(buf, PAGE_SIZE, "%u\n",
945 ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
949 * Set default_easize.
951 * Range checking on the passed value is handled by
952 * ll_set_default_mdsize().
954 * \see client_obd::cl_default_mds_easize
956 * \param[in] file proc file
957 * \param[in] buffer string passed from user space
958 * \param[in] count \a buffer length
959 * \param[in] off unused for single entry
961 * \retval positive \a count on success
962 * \retval negative negated errno on failure
964 static ssize_t default_easize_store(struct kobject *kobj,
965 struct attribute *attr,
969 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
977 rc = kstrtouint(buffer, 10, &val);
981 rc = ll_set_default_mdsize(sbi, val);
987 LUSTRE_RW_ATTR(default_easize);
989 static int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
991 const char *str[] = LL_SBI_FLAGS;
992 struct super_block *sb = m->private;
993 int flags = ll_s2sbi(sb)->ll_flags;
997 if (ARRAY_SIZE(str) <= i) {
998 CERROR("%s: Revise array LL_SBI_FLAGS to match sbi "
999 "flags please.\n", ll_s2sbi(sb)->ll_fsname);
1004 seq_printf(m, "%s ", str[i]);
1008 seq_printf(m, "\b\n");
1012 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
1014 static ssize_t xattr_cache_show(struct kobject *kobj,
1015 struct attribute *attr,
1018 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1021 return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
1024 static ssize_t xattr_cache_store(struct kobject *kobj,
1025 struct attribute *attr,
1029 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1034 rc = kstrtobool(buffer, &val);
1038 if (val && !(sbi->ll_flags & LL_SBI_XATTR_CACHE))
1041 sbi->ll_xattr_cache_enabled = val;
1042 sbi->ll_xattr_cache_set = 1;
1046 LUSTRE_RW_ATTR(xattr_cache);
1048 static ssize_t tiny_write_show(struct kobject *kobj,
1049 struct attribute *attr,
1052 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1055 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_TINY_WRITE));
1058 static ssize_t tiny_write_store(struct kobject *kobj,
1059 struct attribute *attr,
1063 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1068 rc = kstrtobool(buffer, &val);
1072 spin_lock(&sbi->ll_lock);
1074 sbi->ll_flags |= LL_SBI_TINY_WRITE;
1076 sbi->ll_flags &= ~LL_SBI_TINY_WRITE;
1077 spin_unlock(&sbi->ll_lock);
1081 LUSTRE_RW_ATTR(tiny_write);
1083 static ssize_t max_read_ahead_async_active_show(struct kobject *kobj,
1084 struct attribute *attr,
1087 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1090 return snprintf(buf, PAGE_SIZE, "%u\n",
1091 sbi->ll_ra_info.ra_async_max_active);
1094 static ssize_t max_read_ahead_async_active_store(struct kobject *kobj,
1095 struct attribute *attr,
1099 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1104 rc = kstrtouint(buffer, 10, &val);
1108 if (val < 1 || val > WQ_UNBOUND_MAX_ACTIVE) {
1109 CERROR("%s: cannot set max_read_ahead_async_active=%u %s than %u\n",
1110 sbi->ll_fsname, val,
1111 val < 1 ? "smaller" : "larger",
1112 val < 1 ? 1 : WQ_UNBOUND_MAX_ACTIVE);
1116 spin_lock(&sbi->ll_lock);
1117 sbi->ll_ra_info.ra_async_max_active = val;
1118 spin_unlock(&sbi->ll_lock);
1119 workqueue_set_max_active(sbi->ll_ra_info.ll_readahead_wq, val);
1123 LUSTRE_RW_ATTR(max_read_ahead_async_active);
1125 static ssize_t read_ahead_async_file_threshold_mb_show(struct kobject *kobj,
1126 struct attribute *attr,
1129 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1132 return snprintf(buf, PAGE_SIZE, "%lu\n",
1133 PAGES_TO_MiB(sbi->ll_ra_info.ra_async_pages_per_file_threshold));
1137 read_ahead_async_file_threshold_mb_store(struct kobject *kobj,
1138 struct attribute *attr,
1139 const char *buffer, size_t count)
1141 unsigned long pages_number;
1142 unsigned long max_ra_per_file;
1143 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1147 rc = kstrtoul(buffer, 10, &pages_number);
1151 pages_number = MiB_TO_PAGES(pages_number);
1152 max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1153 if (pages_number < 0 || pages_number > max_ra_per_file) {
1154 CERROR("%s: can't set read_ahead_async_file_threshold_mb=%lu > "
1155 "max_read_readahead_per_file_mb=%lu\n", sbi->ll_fsname,
1156 PAGES_TO_MiB(pages_number),
1157 PAGES_TO_MiB(max_ra_per_file));
1160 sbi->ll_ra_info.ra_async_pages_per_file_threshold = pages_number;
1164 LUSTRE_RW_ATTR(read_ahead_async_file_threshold_mb);
1166 static ssize_t fast_read_show(struct kobject *kobj,
1167 struct attribute *attr,
1170 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1173 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_FAST_READ));
1176 static ssize_t fast_read_store(struct kobject *kobj,
1177 struct attribute *attr,
1181 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1186 rc = kstrtobool(buffer, &val);
1190 spin_lock(&sbi->ll_lock);
1192 sbi->ll_flags |= LL_SBI_FAST_READ;
1194 sbi->ll_flags &= ~LL_SBI_FAST_READ;
1195 spin_unlock(&sbi->ll_lock);
1199 LUSTRE_RW_ATTR(fast_read);
1201 static ssize_t file_heat_show(struct kobject *kobj,
1202 struct attribute *attr,
1205 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1208 return snprintf(buf, PAGE_SIZE, "%u\n",
1209 !!(sbi->ll_flags & LL_SBI_FILE_HEAT));
1212 static ssize_t file_heat_store(struct kobject *kobj,
1213 struct attribute *attr,
1217 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1222 rc = kstrtobool(buffer, &val);
1226 spin_lock(&sbi->ll_lock);
1228 sbi->ll_flags |= LL_SBI_FILE_HEAT;
1230 sbi->ll_flags &= ~LL_SBI_FILE_HEAT;
1231 spin_unlock(&sbi->ll_lock);
1235 LUSTRE_RW_ATTR(file_heat);
1237 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1238 struct attribute *attr,
1241 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1244 return snprintf(buf, PAGE_SIZE, "%u\n",
1245 (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1248 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1249 struct attribute *attr,
1253 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1258 rc = kstrtoul(buffer, 10, &val);
1262 if (val < 0 || val > 100)
1265 sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1269 LUSTRE_RW_ATTR(heat_decay_percentage);
1271 static ssize_t heat_period_second_show(struct kobject *kobj,
1272 struct attribute *attr,
1275 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1278 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1281 static ssize_t heat_period_second_store(struct kobject *kobj,
1282 struct attribute *attr,
1286 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1291 rc = kstrtoul(buffer, 10, &val);
1298 sbi->ll_heat_period_second = val;
1302 LUSTRE_RW_ATTR(heat_period_second);
1304 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1306 struct super_block *sb = m->private;
1307 struct ll_sb_info *sbi = ll_s2sbi(sb);
1308 struct cl_client_cache *cache = sbi->ll_cache;
1312 pages = atomic_long_read(&cache->ccc_unstable_nr);
1313 mb = (pages * PAGE_SIZE) >> 20;
1315 seq_printf(m, "unstable_check: %8d\n"
1316 "unstable_pages: %12ld\n"
1317 "unstable_mb: %8d\n",
1318 cache->ccc_unstable_check, pages, mb);
1322 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1323 const char __user *buffer,
1324 size_t count, loff_t *unused)
1326 struct seq_file *seq = file->private_data;
1327 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1334 if (count >= sizeof(kernbuf))
1337 if (copy_from_user(kernbuf, buffer, count))
1341 buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1343 rc = kstrtobool_from_user(buffer, count, &val);
1347 /* borrow lru lock to set the value */
1348 spin_lock(&sbi->ll_cache->ccc_lru_lock);
1349 sbi->ll_cache->ccc_unstable_check = val;
1350 spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1355 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1357 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1359 struct super_block *sb = m->private;
1360 struct ll_sb_info *sbi = ll_s2sbi(sb);
1361 struct root_squash_info *squash = &sbi->ll_squash;
1363 seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1367 static ssize_t ll_root_squash_seq_write(struct file *file,
1368 const char __user *buffer,
1369 size_t count, loff_t *off)
1371 struct seq_file *m = file->private_data;
1372 struct super_block *sb = m->private;
1373 struct ll_sb_info *sbi = ll_s2sbi(sb);
1374 struct root_squash_info *squash = &sbi->ll_squash;
1376 return lprocfs_wr_root_squash(buffer, count, squash, sbi->ll_fsname);
1379 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1381 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1383 struct super_block *sb = m->private;
1384 struct ll_sb_info *sbi = ll_s2sbi(sb);
1385 struct root_squash_info *squash = &sbi->ll_squash;
1388 spin_lock(&squash->rsi_lock);
1389 if (!list_empty(&squash->rsi_nosquash_nids)) {
1390 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1391 &squash->rsi_nosquash_nids);
1395 seq_puts(m, "NONE\n");
1397 spin_unlock(&squash->rsi_lock);
1402 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1403 const char __user *buffer,
1404 size_t count, loff_t *off)
1406 struct seq_file *m = file->private_data;
1407 struct super_block *sb = m->private;
1408 struct ll_sb_info *sbi = ll_s2sbi(sb);
1409 struct root_squash_info *squash = &sbi->ll_squash;
1412 rc = lprocfs_wr_nosquash_nids(buffer, count, squash, sbi->ll_fsname);
1416 ll_compute_rootsquash_state(sbi);
1421 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1423 static int ll_pcc_seq_show(struct seq_file *m, void *v)
1425 struct super_block *sb = m->private;
1426 struct ll_sb_info *sbi = ll_s2sbi(sb);
1428 return pcc_super_dump(&sbi->ll_pcc_super, m);
1431 static ssize_t ll_pcc_seq_write(struct file *file, const char __user *buffer,
1432 size_t count, loff_t *off)
1434 struct seq_file *m = file->private_data;
1435 struct super_block *sb = m->private;
1436 struct ll_sb_info *sbi = ll_s2sbi(sb);
1440 if (count >= LPROCFS_WR_PCC_MAX_CMD)
1443 if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_PCC))
1446 OBD_ALLOC(kernbuf, count + 1);
1447 if (kernbuf == NULL)
1450 if (copy_from_user(kernbuf, buffer, count))
1451 GOTO(out_free_kernbuff, rc = -EFAULT);
1453 rc = pcc_cmd_handle(kernbuf, count, &sbi->ll_pcc_super);
1455 OBD_FREE(kernbuf, count + 1);
1456 return rc ? rc : count;
1458 LPROC_SEQ_FOPS(ll_pcc);
1460 struct lprocfs_vars lprocfs_llite_obd_vars[] = {
1462 .fops = &ll_site_stats_fops },
1463 { .name = "max_cached_mb",
1464 .fops = &ll_max_cached_mb_fops },
1465 { .name = "statahead_stats",
1466 .fops = &ll_statahead_stats_fops },
1467 { .name = "unstable_stats",
1468 .fops = &ll_unstable_stats_fops },
1469 { .name = "sbi_flags",
1470 .fops = &ll_sbi_flags_fops },
1471 { .name = "root_squash",
1472 .fops = &ll_root_squash_fops },
1473 { .name = "nosquash_nids",
1474 .fops = &ll_nosquash_nids_fops },
1476 .fops = &ll_pcc_fops, },
1480 #define MAX_STRING_SIZE 128
1482 static struct attribute *llite_attrs[] = {
1483 &lustre_attr_blocksize.attr,
1484 &lustre_attr_stat_blocksize.attr,
1485 &lustre_attr_kbytestotal.attr,
1486 &lustre_attr_kbytesfree.attr,
1487 &lustre_attr_kbytesavail.attr,
1488 &lustre_attr_filestotal.attr,
1489 &lustre_attr_filesfree.attr,
1490 &lustre_attr_client_type.attr,
1491 &lustre_attr_fstype.attr,
1492 &lustre_attr_uuid.attr,
1493 &lustre_attr_checksums.attr,
1494 &lustre_attr_checksum_pages.attr,
1495 &lustre_attr_max_read_ahead_mb.attr,
1496 &lustre_attr_max_read_ahead_per_file_mb.attr,
1497 &lustre_attr_max_read_ahead_whole_mb.attr,
1498 &lustre_attr_max_read_ahead_async_active.attr,
1499 &lustre_attr_read_ahead_async_file_threshold_mb.attr,
1500 &lustre_attr_stats_track_pid.attr,
1501 &lustre_attr_stats_track_ppid.attr,
1502 &lustre_attr_stats_track_gid.attr,
1503 &lustre_attr_statahead_running_max.attr,
1504 &lustre_attr_statahead_max.attr,
1505 &lustre_attr_statahead_agl.attr,
1506 &lustre_attr_lazystatfs.attr,
1507 &lustre_attr_statfs_max_age.attr,
1508 &lustre_attr_max_easize.attr,
1509 &lustre_attr_default_easize.attr,
1510 &lustre_attr_xattr_cache.attr,
1511 &lustre_attr_fast_read.attr,
1512 &lustre_attr_tiny_write.attr,
1513 &lustre_attr_file_heat.attr,
1514 &lustre_attr_heat_decay_percentage.attr,
1515 &lustre_attr_heat_period_second.attr,
1519 static void sbi_kobj_release(struct kobject *kobj)
1521 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1523 complete(&sbi->ll_kobj_unregister);
1526 static struct kobj_type sbi_ktype = {
1527 .default_attrs = llite_attrs,
1528 .sysfs_ops = &lustre_sysfs_ops,
1529 .release = sbi_kobj_release,
1532 #define LPROCFS_TYPE_LATENCY \
1533 (LPROCFS_TYPE_USEC | LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV)
1534 static const struct llite_file_opcode {
1538 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1539 /* file operation */
1540 { LPROC_LL_READ_BYTES, LPROCFS_CNTR_AVGMINMAX | LPROCFS_TYPE_BYTES,
1542 { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX | LPROCFS_TYPE_BYTES,
1544 { LPROC_LL_READ, LPROCFS_TYPE_LATENCY, "read" },
1545 { LPROC_LL_WRITE, LPROCFS_TYPE_LATENCY, "write" },
1546 { LPROC_LL_IOCTL, LPROCFS_TYPE_REQS, "ioctl" },
1547 { LPROC_LL_OPEN, LPROCFS_TYPE_LATENCY, "open" },
1548 { LPROC_LL_RELEASE, LPROCFS_TYPE_LATENCY, "close" },
1549 { LPROC_LL_MMAP, LPROCFS_TYPE_LATENCY, "mmap" },
1550 { LPROC_LL_FAULT, LPROCFS_TYPE_LATENCY, "page_fault" },
1551 { LPROC_LL_MKWRITE, LPROCFS_TYPE_LATENCY, "page_mkwrite" },
1552 { LPROC_LL_LLSEEK, LPROCFS_TYPE_LATENCY, "seek" },
1553 { LPROC_LL_FSYNC, LPROCFS_TYPE_LATENCY, "fsync" },
1554 { LPROC_LL_READDIR, LPROCFS_TYPE_LATENCY, "readdir" },
1555 /* inode operation */
1556 { LPROC_LL_SETATTR, LPROCFS_TYPE_LATENCY, "setattr" },
1557 { LPROC_LL_TRUNC, LPROCFS_TYPE_LATENCY, "truncate" },
1558 { LPROC_LL_FLOCK, LPROCFS_TYPE_LATENCY, "flock" },
1559 { LPROC_LL_GETATTR, LPROCFS_TYPE_LATENCY, "getattr" },
1560 /* dir inode operation */
1561 { LPROC_LL_CREATE, LPROCFS_TYPE_LATENCY, "create" },
1562 { LPROC_LL_LINK, LPROCFS_TYPE_LATENCY, "link" },
1563 { LPROC_LL_UNLINK, LPROCFS_TYPE_LATENCY, "unlink" },
1564 { LPROC_LL_SYMLINK, LPROCFS_TYPE_LATENCY, "symlink" },
1565 { LPROC_LL_MKDIR, LPROCFS_TYPE_LATENCY, "mkdir" },
1566 { LPROC_LL_RMDIR, LPROCFS_TYPE_LATENCY, "rmdir" },
1567 { LPROC_LL_MKNOD, LPROCFS_TYPE_LATENCY, "mknod" },
1568 { LPROC_LL_RENAME, LPROCFS_TYPE_LATENCY, "rename" },
1569 /* special inode operation */
1570 { LPROC_LL_STATFS, LPROCFS_TYPE_LATENCY, "statfs" },
1571 { LPROC_LL_SETXATTR, LPROCFS_TYPE_LATENCY, "setxattr" },
1572 { LPROC_LL_GETXATTR, LPROCFS_TYPE_LATENCY, "getxattr" },
1573 { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REQS, "getxattr_hits" },
1574 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_LATENCY, "listxattr" },
1575 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_LATENCY, "removexattr" },
1576 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_LATENCY, "inode_permission" },
1579 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count)
1584 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1585 lprocfs_counter_add(sbi->ll_stats, op, count);
1586 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1587 sbi->ll_stats_track_id == current->pid)
1588 lprocfs_counter_add(sbi->ll_stats, op, count);
1589 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1590 sbi->ll_stats_track_id == current->parent->pid)
1591 lprocfs_counter_add(sbi->ll_stats, op, count);
1592 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1593 sbi->ll_stats_track_id ==
1594 from_kgid(&init_user_ns, current_gid()))
1595 lprocfs_counter_add(sbi->ll_stats, op, count);
1597 EXPORT_SYMBOL(ll_stats_ops_tally);
1599 static const char *ra_stat_string[] = {
1600 [RA_STAT_HIT] = "hits",
1601 [RA_STAT_MISS] = "misses",
1602 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1603 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1604 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1605 [RA_STAT_FAILED_MATCH] = "failed lock match",
1606 [RA_STAT_DISCARDED] = "read but discarded",
1607 [RA_STAT_ZERO_LEN] = "zero length file",
1608 [RA_STAT_ZERO_WINDOW] = "zero size window",
1609 [RA_STAT_EOF] = "read-ahead to EOF",
1610 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1611 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1612 [RA_STAT_FAILED_REACH_END] = "failed to reach end",
1613 [RA_STAT_ASYNC] = "async readahead",
1614 [RA_STAT_FAILED_FAST_READ] = "failed to fast read",
1617 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1619 struct lustre_sb_info *lsi = s2lsi(sb);
1620 struct ll_sb_info *sbi = ll_s2sbi(sb);
1626 if (IS_ERR_OR_NULL(llite_root))
1629 sbi->ll_debugfs_entry = ldebugfs_register(name, llite_root,
1630 lprocfs_llite_obd_vars, sb);
1631 if (IS_ERR_OR_NULL(sbi->ll_debugfs_entry)) {
1632 err = sbi->ll_debugfs_entry ? PTR_ERR(sbi->ll_debugfs_entry) :
1634 sbi->ll_debugfs_entry = NULL;
1638 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "dump_page_cache",0444,
1639 &vvp_dump_pgcache_file_ops, sbi);
1641 CWARN("Error adding the dump_page_cache file\n");
1643 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "extents_stats", 0644,
1644 &ll_rw_extents_stats_fops, sbi);
1646 CWARN("Error adding the extent_stats file\n");
1648 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry,
1649 "extents_stats_per_process", 0644,
1650 &ll_rw_extents_stats_pp_fops, sbi);
1652 CWARN("Error adding the extents_stats_per_process file\n");
1654 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "offset_stats", 0644,
1655 &ll_rw_offset_stats_fops, sbi);
1657 CWARN("Error adding the offset_stats file\n");
1659 /* File operations stats */
1660 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1661 LPROCFS_STATS_FLAG_NONE);
1662 if (sbi->ll_stats == NULL)
1663 GOTO(out_debugfs, err = -ENOMEM);
1665 /* do counter init */
1666 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1667 u32 type = llite_opcode_table[id].type;
1670 if (type & LPROCFS_TYPE_REQS)
1672 else if (type & LPROCFS_TYPE_BYTES)
1674 else if (type & LPROCFS_TYPE_PAGES)
1676 else if (type & LPROCFS_TYPE_USEC)
1678 lprocfs_counter_init(sbi->ll_stats,
1679 llite_opcode_table[id].opcode,
1680 (type & LPROCFS_CNTR_AVGMINMAX),
1681 llite_opcode_table[id].opname, ptr);
1684 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "stats",
1687 GOTO(out_stats, err);
1689 sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1690 LPROCFS_STATS_FLAG_NONE);
1691 if (sbi->ll_ra_stats == NULL)
1692 GOTO(out_stats, err = -ENOMEM);
1694 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1695 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1696 ra_stat_string[id], "pages");
1698 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "read_ahead_stats",
1701 GOTO(out_ra_stats, err);
1704 /* Yes we also register sysfs mount kset here as well */
1705 sbi->ll_kset.kobj.parent = llite_kobj;
1706 sbi->ll_kset.kobj.ktype = &sbi_ktype;
1707 init_completion(&sbi->ll_kobj_unregister);
1708 err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1710 GOTO(out_ra_stats, err);
1712 err = kset_register(&sbi->ll_kset);
1714 GOTO(out_ra_stats, err);
1716 lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1720 lprocfs_free_stats(&sbi->ll_ra_stats);
1722 lprocfs_free_stats(&sbi->ll_stats);
1724 debugfs_remove_recursive(sbi->ll_debugfs_entry);
1729 void ll_debugfs_unregister_super(struct super_block *sb)
1731 struct lustre_sb_info *lsi = s2lsi(sb);
1732 struct ll_sb_info *sbi = ll_s2sbi(sb);
1734 debugfs_remove_recursive(sbi->ll_debugfs_entry);
1737 sysfs_remove_link(&sbi->ll_kset.kobj,
1738 sbi->ll_dt_obd->obd_type->typ_name);
1741 sysfs_remove_link(&sbi->ll_kset.kobj,
1742 sbi->ll_md_obd->obd_type->typ_name);
1744 kobject_put(lsi->lsi_kobj);
1746 kset_unregister(&sbi->ll_kset);
1747 wait_for_completion(&sbi->ll_kobj_unregister);
1749 lprocfs_free_stats(&sbi->ll_ra_stats);
1750 lprocfs_free_stats(&sbi->ll_stats);
1752 #undef MAX_STRING_SIZE
1754 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1755 struct seq_file *seq, int which)
1757 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1758 unsigned long start, end, r, w;
1759 char *unitp = "KMGTPEZY";
1761 struct per_process_info *pp_info = &io_extents->pp_extents[which];
1767 for(i = 0; i < LL_HIST_MAX; i++) {
1768 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1769 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1772 for(i = 0; i < LL_HIST_MAX; i++) {
1773 r = pp_info->pp_r_hist.oh_buckets[i];
1774 w = pp_info->pp_w_hist.oh_buckets[i];
1777 end = BIT(i + LL_HIST_START - units);
1778 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u | "
1779 "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1780 (i == LL_HIST_MAX - 1) ? '+' : ' ',
1781 r, pct(r, read_tot), pct(read_cum, read_tot),
1782 w, pct(w, write_tot), pct(write_cum, write_tot));
1784 if (start == BIT(10)) {
1789 if (read_cum == read_tot && write_cum == write_tot)
1794 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1796 struct timespec64 now;
1797 struct ll_sb_info *sbi = seq->private;
1798 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1801 ktime_get_real_ts64(&now);
1803 if (!sbi->ll_rw_stats_on) {
1804 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1807 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1808 (s64)now.tv_sec, now.tv_nsec);
1809 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1810 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1811 "extents", "calls", "%", "cum%",
1812 "calls", "%", "cum%");
1813 spin_lock(&sbi->ll_pp_extent_lock);
1814 for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1815 if (io_extents->pp_extents[k].pid != 0) {
1816 seq_printf(seq, "\nPID: %d\n",
1817 io_extents->pp_extents[k].pid);
1818 ll_display_extents_info(io_extents, seq, k);
1821 spin_unlock(&sbi->ll_pp_extent_lock);
1825 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1826 const char __user *buf,
1830 struct seq_file *seq = file->private_data;
1831 struct ll_sb_info *sbi = seq->private;
1832 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1839 value = ll_stats_pid_write(buf, len);
1842 sbi->ll_rw_stats_on = 0;
1844 sbi->ll_rw_stats_on = 1;
1846 spin_lock(&sbi->ll_pp_extent_lock);
1847 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1848 io_extents->pp_extents[i].pid = 0;
1849 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1850 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1852 spin_unlock(&sbi->ll_pp_extent_lock);
1856 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1858 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1860 struct timespec64 now;
1861 struct ll_sb_info *sbi = seq->private;
1862 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1864 ktime_get_real_ts64(&now);
1866 if (!sbi->ll_rw_stats_on) {
1867 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1870 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1871 (s64)now.tv_sec, now.tv_nsec);
1873 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1874 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1875 "extents", "calls", "%", "cum%",
1876 "calls", "%", "cum%");
1877 spin_lock(&sbi->ll_lock);
1878 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1879 spin_unlock(&sbi->ll_lock);
1884 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1885 const char __user *buf,
1886 size_t len, loff_t *off)
1888 struct seq_file *seq = file->private_data;
1889 struct ll_sb_info *sbi = seq->private;
1890 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1897 value = ll_stats_pid_write(buf, len);
1900 sbi->ll_rw_stats_on = 0;
1902 sbi->ll_rw_stats_on = 1;
1904 spin_lock(&sbi->ll_pp_extent_lock);
1905 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1906 io_extents->pp_extents[i].pid = 0;
1907 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1908 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1910 spin_unlock(&sbi->ll_pp_extent_lock);
1915 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1917 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1918 struct ll_file_data *file, loff_t pos,
1919 size_t count, int rw)
1922 struct ll_rw_process_info *process;
1923 struct ll_rw_process_info *offset;
1924 int *off_count = &sbi->ll_rw_offset_entry_count;
1925 int *process_count = &sbi->ll_offset_process_count;
1926 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1928 if(!sbi->ll_rw_stats_on)
1930 process = sbi->ll_rw_process_info;
1931 offset = sbi->ll_rw_offset_info;
1933 spin_lock(&sbi->ll_pp_extent_lock);
1934 /* Extent statistics */
1935 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1936 if(io_extents->pp_extents[i].pid == pid) {
1944 sbi->ll_extent_process_count =
1945 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1946 cur = sbi->ll_extent_process_count;
1947 io_extents->pp_extents[cur].pid = pid;
1948 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1949 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1952 for (i = 0; (count >= BIT(LL_HIST_START + i)) &&
1953 (i < (LL_HIST_MAX - 1)); i++);
1955 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1956 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1958 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1959 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1961 spin_unlock(&sbi->ll_pp_extent_lock);
1963 spin_lock(&sbi->ll_process_lock);
1964 /* Offset statistics */
1965 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1966 if (process[i].rw_pid == pid) {
1967 if (process[i].rw_last_file != file) {
1968 process[i].rw_range_start = pos;
1969 process[i].rw_last_file_pos = pos + count;
1970 process[i].rw_smallest_extent = count;
1971 process[i].rw_largest_extent = count;
1972 process[i].rw_offset = 0;
1973 process[i].rw_last_file = file;
1974 spin_unlock(&sbi->ll_process_lock);
1977 if (process[i].rw_last_file_pos != pos) {
1979 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1980 offset[*off_count].rw_op = process[i].rw_op;
1981 offset[*off_count].rw_pid = pid;
1982 offset[*off_count].rw_range_start =
1983 process[i].rw_range_start;
1984 offset[*off_count].rw_range_end =
1985 process[i].rw_last_file_pos;
1986 offset[*off_count].rw_smallest_extent =
1987 process[i].rw_smallest_extent;
1988 offset[*off_count].rw_largest_extent =
1989 process[i].rw_largest_extent;
1990 offset[*off_count].rw_offset =
1991 process[i].rw_offset;
1992 process[i].rw_op = rw;
1993 process[i].rw_range_start = pos;
1994 process[i].rw_smallest_extent = count;
1995 process[i].rw_largest_extent = count;
1996 process[i].rw_offset = pos -
1997 process[i].rw_last_file_pos;
1999 if(process[i].rw_smallest_extent > count)
2000 process[i].rw_smallest_extent = count;
2001 if(process[i].rw_largest_extent < count)
2002 process[i].rw_largest_extent = count;
2003 process[i].rw_last_file_pos = pos + count;
2004 spin_unlock(&sbi->ll_process_lock);
2008 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
2009 process[*process_count].rw_pid = pid;
2010 process[*process_count].rw_op = rw;
2011 process[*process_count].rw_range_start = pos;
2012 process[*process_count].rw_last_file_pos = pos + count;
2013 process[*process_count].rw_smallest_extent = count;
2014 process[*process_count].rw_largest_extent = count;
2015 process[*process_count].rw_offset = 0;
2016 process[*process_count].rw_last_file = file;
2017 spin_unlock(&sbi->ll_process_lock);
2020 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
2022 struct timespec64 now;
2023 struct ll_sb_info *sbi = seq->private;
2024 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
2025 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
2028 ktime_get_real_ts64(&now);
2030 if (!sbi->ll_rw_stats_on) {
2031 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2034 spin_lock(&sbi->ll_process_lock);
2036 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
2037 (s64)now.tv_sec, now.tv_nsec);
2038 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
2039 "R/W", "PID", "RANGE START", "RANGE END",
2040 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
2042 /* We stored the discontiguous offsets here; print them first */
2043 for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
2044 if (offset[i].rw_pid != 0)
2046 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2047 offset[i].rw_op == READ ? 'R' : 'W',
2049 offset[i].rw_range_start,
2050 offset[i].rw_range_end,
2051 (unsigned long)offset[i].rw_smallest_extent,
2052 (unsigned long)offset[i].rw_largest_extent,
2053 offset[i].rw_offset);
2056 /* Then print the current offsets for each process */
2057 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2058 if (process[i].rw_pid != 0)
2060 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2061 process[i].rw_op == READ ? 'R' : 'W',
2063 process[i].rw_range_start,
2064 process[i].rw_last_file_pos,
2065 (unsigned long)process[i].rw_smallest_extent,
2066 (unsigned long)process[i].rw_largest_extent,
2067 process[i].rw_offset);
2069 spin_unlock(&sbi->ll_process_lock);
2074 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
2075 const char __user *buf,
2076 size_t len, loff_t *off)
2078 struct seq_file *seq = file->private_data;
2079 struct ll_sb_info *sbi = seq->private;
2080 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
2081 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
2087 value = ll_stats_pid_write(buf, len);
2090 sbi->ll_rw_stats_on = 0;
2092 sbi->ll_rw_stats_on = 1;
2094 spin_lock(&sbi->ll_process_lock);
2095 sbi->ll_offset_process_count = 0;
2096 sbi->ll_rw_offset_entry_count = 0;
2097 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
2098 LL_PROCESS_HIST_MAX);
2099 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
2100 LL_OFFSET_HIST_MAX);
2101 spin_unlock(&sbi->ll_process_lock);
2106 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);