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 int ll_max_readahead_mb_seq_show(struct seq_file *m, void *v)
326 struct super_block *sb = m->private;
327 struct ll_sb_info *sbi = ll_s2sbi(sb);
328 unsigned long ra_max_mb;
330 spin_lock(&sbi->ll_lock);
331 ra_max_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages);
332 spin_unlock(&sbi->ll_lock);
334 seq_printf(m, "%lu\n", ra_max_mb);
339 ll_max_readahead_mb_seq_write(struct file *file, const char __user *buffer,
340 size_t count, loff_t *off)
342 struct seq_file *m = file->private_data;
343 struct super_block *sb = m->private;
344 struct ll_sb_info *sbi = ll_s2sbi(sb);
345 s64 ra_max_mb, pages_number;
348 rc = lprocfs_str_with_units_to_s64(buffer, count, &ra_max_mb, 'M');
352 pages_number = round_up(ra_max_mb, 1024 * 1024) >> PAGE_SHIFT;
353 if (pages_number < 0 || pages_number > cfs_totalram_pages() / 2) {
355 CERROR("%s: can't set max_readahead_mb=%llu > %luMB\n",
356 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
357 PAGES_TO_MiB(cfs_totalram_pages()));
361 spin_lock(&sbi->ll_lock);
362 sbi->ll_ra_info.ra_max_pages = pages_number;
363 spin_unlock(&sbi->ll_lock);
368 LDEBUGFS_SEQ_FOPS(ll_max_readahead_mb);
370 static int ll_max_readahead_per_file_mb_seq_show(struct seq_file *m, void *v)
372 struct super_block *sb = m->private;
373 struct ll_sb_info *sbi = ll_s2sbi(sb);
374 unsigned long ra_max_file_mb;
376 spin_lock(&sbi->ll_lock);
377 ra_max_file_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file);
378 spin_unlock(&sbi->ll_lock);
380 seq_printf(m, "%lu\n", ra_max_file_mb);
385 ll_max_readahead_per_file_mb_seq_write(struct file *file,
386 const char __user *buffer,
387 size_t count, loff_t *off)
389 struct seq_file *m = file->private_data;
390 struct super_block *sb = m->private;
391 struct ll_sb_info *sbi = ll_s2sbi(sb);
392 s64 ra_max_file_mb, pages_number;
395 rc = lprocfs_str_with_units_to_s64(buffer, count, &ra_max_file_mb,
400 pages_number = round_up(ra_max_file_mb, 1024 * 1024) >> PAGE_SHIFT;
401 if (pages_number < 0 || pages_number > sbi->ll_ra_info.ra_max_pages) {
402 CERROR("%s: can't set max_readahead_per_file_mb=%llu > max_read_ahead_mb=%lu\n",
403 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
404 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
408 spin_lock(&sbi->ll_lock);
409 sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
410 spin_unlock(&sbi->ll_lock);
415 LDEBUGFS_SEQ_FOPS(ll_max_readahead_per_file_mb);
417 static int ll_max_read_ahead_whole_mb_seq_show(struct seq_file *m, void *v)
419 struct super_block *sb = m->private;
420 struct ll_sb_info *sbi = ll_s2sbi(sb);
421 unsigned long ra_max_whole_mb;
423 spin_lock(&sbi->ll_lock);
424 ra_max_whole_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_read_ahead_whole_pages);
425 spin_unlock(&sbi->ll_lock);
427 seq_printf(m, "%lu\n", ra_max_whole_mb);
432 ll_max_read_ahead_whole_mb_seq_write(struct file *file,
433 const char __user *buffer,
434 size_t count, loff_t *off)
436 struct seq_file *m = file->private_data;
437 struct super_block *sb = m->private;
438 struct ll_sb_info *sbi = ll_s2sbi(sb);
439 s64 ra_max_whole_mb, pages_number;
442 rc = lprocfs_str_with_units_to_s64(buffer, count, &ra_max_whole_mb,
447 pages_number = round_up(ra_max_whole_mb, 1024 * 1024) >> PAGE_SHIFT;
448 /* Cap this at the current max readahead window size, the readahead
449 * algorithm does this anyway so it's pointless to set it larger.
451 if (pages_number < 0 ||
452 pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
453 CERROR("%s: can't set max_read_ahead_whole_mb=%llu > max_read_ahead_per_file_mb=%lu\n",
454 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
455 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
459 spin_lock(&sbi->ll_lock);
460 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
461 spin_unlock(&sbi->ll_lock);
466 LDEBUGFS_SEQ_FOPS(ll_max_read_ahead_whole_mb);
468 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
470 struct super_block *sb = m->private;
471 struct ll_sb_info *sbi = ll_s2sbi(sb);
472 struct cl_client_cache *cache = sbi->ll_cache;
476 max_cached_mb = PAGES_TO_MiB(cache->ccc_lru_max);
477 unused_mb = PAGES_TO_MiB(atomic_long_read(&cache->ccc_lru_left));
478 seq_printf(m, "users: %d\n"
479 "max_cached_mb: %ld\n"
482 "reclaim_count: %u\n",
483 atomic_read(&cache->ccc_users),
485 max_cached_mb - unused_mb,
487 cache->ccc_lru_shrinkers);
491 static ssize_t ll_max_cached_mb_seq_write(struct file *file,
492 const char __user *buffer,
493 size_t count, loff_t *off)
495 struct seq_file *m = file->private_data;
496 struct super_block *sb = m->private;
497 struct ll_sb_info *sbi = ll_s2sbi(sb);
498 struct cl_client_cache *cache = sbi->ll_cache;
508 if (count >= sizeof(kernbuf))
511 if (copy_from_user(kernbuf, buffer, count))
515 buffer += lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count) -
517 rc = lprocfs_str_with_units_to_s64(buffer, count, &pages_number, 'M');
521 pages_number >>= PAGE_SHIFT;
523 if (pages_number < 0 || pages_number > cfs_totalram_pages()) {
524 CERROR("%s: can't set max cache more than %lu MB\n",
526 PAGES_TO_MiB(cfs_totalram_pages()));
529 /* Allow enough cache so clients can make well-formed RPCs */
530 pages_number = max_t(long, pages_number, PTLRPC_MAX_BRW_PAGES);
532 spin_lock(&sbi->ll_lock);
533 diff = pages_number - cache->ccc_lru_max;
534 spin_unlock(&sbi->ll_lock);
536 /* easy - add more LRU slots. */
538 atomic_long_add(diff, &cache->ccc_lru_left);
542 env = cl_env_get(&refcheck);
544 RETURN(PTR_ERR(env));
550 /* reduce LRU budget from free slots. */
554 ov = atomic_long_read(&cache->ccc_lru_left);
558 nv = ov > diff ? ov - diff : 0;
559 rc = atomic_long_cmpxchg(&cache->ccc_lru_left, ov, nv);
560 if (likely(ov == rc)) {
570 if (sbi->ll_dt_exp == NULL) { /* being initialized */
575 /* difficult - have to ask OSCs to drop LRU slots. */
577 rc = obd_set_info_async(env, sbi->ll_dt_exp,
578 sizeof(KEY_CACHE_LRU_SHRINK),
579 KEY_CACHE_LRU_SHRINK,
580 sizeof(tmp), &tmp, NULL);
584 cl_env_put(env, &refcheck);
588 spin_lock(&sbi->ll_lock);
589 cache->ccc_lru_max = pages_number;
590 spin_unlock(&sbi->ll_lock);
593 atomic_long_add(nrpages, &cache->ccc_lru_left);
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,
1108 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1111 rc = kstrtouint(buffer, 10, &val);
1115 if (val < 1 || val > WQ_UNBOUND_MAX_ACTIVE) {
1116 CERROR("%s: cannot set max_read_ahead_async_active=%u %s than %u\n",
1117 sbi->ll_fsname, val,
1118 val < 1 ? "smaller" : "larger",
1119 val < 1 ? 1 : WQ_UNBOUND_MAX_ACTIVE);
1123 sbi->ll_ra_info.ra_async_max_active = val;
1124 workqueue_set_max_active(sbi->ll_ra_info.ll_readahead_wq, val);
1128 LUSTRE_RW_ATTR(max_read_ahead_async_active);
1130 static ssize_t read_ahead_async_file_threshold_mb_show(struct kobject *kobj,
1131 struct attribute *attr,
1134 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1137 return snprintf(buf, PAGE_SIZE, "%lu\n",
1138 PAGES_TO_MiB(sbi->ll_ra_info.ra_async_pages_per_file_threshold));
1142 read_ahead_async_file_threshold_mb_store(struct kobject *kobj,
1143 struct attribute *attr,
1144 const char *buffer, size_t count)
1146 unsigned long pages_number;
1147 unsigned long max_ra_per_file;
1148 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1152 rc = kstrtoul(buffer, 10, &pages_number);
1156 pages_number = MiB_TO_PAGES(pages_number);
1157 max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1158 if (pages_number < 0 || pages_number > max_ra_per_file) {
1159 CERROR("%s: can't set read_ahead_async_file_threshold_mb=%lu > "
1160 "max_read_readahead_per_file_mb=%lu\n", sbi->ll_fsname,
1161 PAGES_TO_MiB(pages_number),
1162 PAGES_TO_MiB(max_ra_per_file));
1165 sbi->ll_ra_info.ra_async_pages_per_file_threshold = pages_number;
1169 LUSTRE_RW_ATTR(read_ahead_async_file_threshold_mb);
1171 static ssize_t fast_read_show(struct kobject *kobj,
1172 struct attribute *attr,
1175 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1178 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_FAST_READ));
1181 static ssize_t fast_read_store(struct kobject *kobj,
1182 struct attribute *attr,
1186 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1191 rc = kstrtobool(buffer, &val);
1195 spin_lock(&sbi->ll_lock);
1197 sbi->ll_flags |= LL_SBI_FAST_READ;
1199 sbi->ll_flags &= ~LL_SBI_FAST_READ;
1200 spin_unlock(&sbi->ll_lock);
1204 LUSTRE_RW_ATTR(fast_read);
1206 static ssize_t file_heat_show(struct kobject *kobj,
1207 struct attribute *attr,
1210 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1213 return snprintf(buf, PAGE_SIZE, "%u\n",
1214 !!(sbi->ll_flags & LL_SBI_FILE_HEAT));
1217 static ssize_t file_heat_store(struct kobject *kobj,
1218 struct attribute *attr,
1222 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1227 rc = kstrtobool(buffer, &val);
1231 spin_lock(&sbi->ll_lock);
1233 sbi->ll_flags |= LL_SBI_FILE_HEAT;
1235 sbi->ll_flags &= ~LL_SBI_FILE_HEAT;
1236 spin_unlock(&sbi->ll_lock);
1240 LUSTRE_RW_ATTR(file_heat);
1242 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1243 struct attribute *attr,
1246 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1249 return snprintf(buf, PAGE_SIZE, "%u\n",
1250 (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1253 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1254 struct attribute *attr,
1258 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1263 rc = kstrtoul(buffer, 10, &val);
1267 if (val < 0 || val > 100)
1270 sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1274 LUSTRE_RW_ATTR(heat_decay_percentage);
1276 static ssize_t heat_period_second_show(struct kobject *kobj,
1277 struct attribute *attr,
1280 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1283 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1286 static ssize_t heat_period_second_store(struct kobject *kobj,
1287 struct attribute *attr,
1291 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1296 rc = kstrtoul(buffer, 10, &val);
1303 sbi->ll_heat_period_second = val;
1307 LUSTRE_RW_ATTR(heat_period_second);
1309 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1311 struct super_block *sb = m->private;
1312 struct ll_sb_info *sbi = ll_s2sbi(sb);
1313 struct cl_client_cache *cache = sbi->ll_cache;
1317 pages = atomic_long_read(&cache->ccc_unstable_nr);
1318 mb = (pages * PAGE_SIZE) >> 20;
1320 seq_printf(m, "unstable_check: %8d\n"
1321 "unstable_pages: %12ld\n"
1322 "unstable_mb: %8d\n",
1323 cache->ccc_unstable_check, pages, mb);
1327 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1328 const char __user *buffer,
1329 size_t count, loff_t *unused)
1331 struct seq_file *seq = file->private_data;
1332 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1339 if (count >= sizeof(kernbuf))
1342 if (copy_from_user(kernbuf, buffer, count))
1346 buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1348 rc = kstrtobool_from_user(buffer, count, &val);
1352 /* borrow lru lock to set the value */
1353 spin_lock(&sbi->ll_cache->ccc_lru_lock);
1354 sbi->ll_cache->ccc_unstable_check = val;
1355 spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1360 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1362 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1364 struct super_block *sb = m->private;
1365 struct ll_sb_info *sbi = ll_s2sbi(sb);
1366 struct root_squash_info *squash = &sbi->ll_squash;
1368 seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1372 static ssize_t ll_root_squash_seq_write(struct file *file,
1373 const char __user *buffer,
1374 size_t count, loff_t *off)
1376 struct seq_file *m = file->private_data;
1377 struct super_block *sb = m->private;
1378 struct ll_sb_info *sbi = ll_s2sbi(sb);
1379 struct root_squash_info *squash = &sbi->ll_squash;
1381 return lprocfs_wr_root_squash(buffer, count, squash, sbi->ll_fsname);
1384 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1386 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1388 struct super_block *sb = m->private;
1389 struct ll_sb_info *sbi = ll_s2sbi(sb);
1390 struct root_squash_info *squash = &sbi->ll_squash;
1393 spin_lock(&squash->rsi_lock);
1394 if (!list_empty(&squash->rsi_nosquash_nids)) {
1395 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1396 &squash->rsi_nosquash_nids);
1400 seq_puts(m, "NONE\n");
1402 spin_unlock(&squash->rsi_lock);
1407 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1408 const char __user *buffer,
1409 size_t count, loff_t *off)
1411 struct seq_file *m = file->private_data;
1412 struct super_block *sb = m->private;
1413 struct ll_sb_info *sbi = ll_s2sbi(sb);
1414 struct root_squash_info *squash = &sbi->ll_squash;
1417 rc = lprocfs_wr_nosquash_nids(buffer, count, squash, sbi->ll_fsname);
1421 ll_compute_rootsquash_state(sbi);
1426 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1428 static int ll_pcc_seq_show(struct seq_file *m, void *v)
1430 struct super_block *sb = m->private;
1431 struct ll_sb_info *sbi = ll_s2sbi(sb);
1433 return pcc_super_dump(&sbi->ll_pcc_super, m);
1436 static ssize_t ll_pcc_seq_write(struct file *file, const char __user *buffer,
1437 size_t count, loff_t *off)
1439 struct seq_file *m = file->private_data;
1440 struct super_block *sb = m->private;
1441 struct ll_sb_info *sbi = ll_s2sbi(sb);
1445 if (count >= LPROCFS_WR_PCC_MAX_CMD)
1448 if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_PCC))
1451 OBD_ALLOC(kernbuf, count + 1);
1452 if (kernbuf == NULL)
1455 if (copy_from_user(kernbuf, buffer, count))
1456 GOTO(out_free_kernbuff, rc = -EFAULT);
1458 rc = pcc_cmd_handle(kernbuf, count, &sbi->ll_pcc_super);
1460 OBD_FREE(kernbuf, count + 1);
1461 return rc ? rc : count;
1463 LPROC_SEQ_FOPS(ll_pcc);
1465 struct lprocfs_vars lprocfs_llite_obd_vars[] = {
1467 .fops = &ll_site_stats_fops },
1468 { .name = "max_read_ahead_mb",
1469 .fops = &ll_max_readahead_mb_fops },
1470 { .name = "max_read_ahead_per_file_mb",
1471 .fops = &ll_max_readahead_per_file_mb_fops },
1472 { .name = "max_read_ahead_whole_mb",
1473 .fops = &ll_max_read_ahead_whole_mb_fops },
1474 { .name = "max_cached_mb",
1475 .fops = &ll_max_cached_mb_fops },
1476 { .name = "statahead_stats",
1477 .fops = &ll_statahead_stats_fops },
1478 { .name = "unstable_stats",
1479 .fops = &ll_unstable_stats_fops },
1480 { .name = "sbi_flags",
1481 .fops = &ll_sbi_flags_fops },
1482 { .name = "root_squash",
1483 .fops = &ll_root_squash_fops },
1484 { .name = "nosquash_nids",
1485 .fops = &ll_nosquash_nids_fops },
1487 .fops = &ll_pcc_fops, },
1491 #define MAX_STRING_SIZE 128
1493 static struct attribute *llite_attrs[] = {
1494 &lustre_attr_blocksize.attr,
1495 &lustre_attr_stat_blocksize.attr,
1496 &lustre_attr_kbytestotal.attr,
1497 &lustre_attr_kbytesfree.attr,
1498 &lustre_attr_kbytesavail.attr,
1499 &lustre_attr_filestotal.attr,
1500 &lustre_attr_filesfree.attr,
1501 &lustre_attr_client_type.attr,
1502 &lustre_attr_fstype.attr,
1503 &lustre_attr_uuid.attr,
1504 &lustre_attr_checksums.attr,
1505 &lustre_attr_checksum_pages.attr,
1506 &lustre_attr_stats_track_pid.attr,
1507 &lustre_attr_stats_track_ppid.attr,
1508 &lustre_attr_stats_track_gid.attr,
1509 &lustre_attr_statahead_running_max.attr,
1510 &lustre_attr_statahead_max.attr,
1511 &lustre_attr_statahead_agl.attr,
1512 &lustre_attr_lazystatfs.attr,
1513 &lustre_attr_statfs_max_age.attr,
1514 &lustre_attr_max_easize.attr,
1515 &lustre_attr_default_easize.attr,
1516 &lustre_attr_xattr_cache.attr,
1517 &lustre_attr_fast_read.attr,
1518 &lustre_attr_tiny_write.attr,
1519 &lustre_attr_file_heat.attr,
1520 &lustre_attr_heat_decay_percentage.attr,
1521 &lustre_attr_heat_period_second.attr,
1522 &lustre_attr_max_read_ahead_async_active.attr,
1523 &lustre_attr_read_ahead_async_file_threshold_mb.attr,
1527 static void sbi_kobj_release(struct kobject *kobj)
1529 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1531 complete(&sbi->ll_kobj_unregister);
1534 static struct kobj_type sbi_ktype = {
1535 .default_attrs = llite_attrs,
1536 .sysfs_ops = &lustre_sysfs_ops,
1537 .release = sbi_kobj_release,
1540 #define LPROCFS_TYPE_LATENCY \
1541 (LPROCFS_TYPE_USEC | LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV)
1542 static const struct llite_file_opcode {
1546 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1547 /* file operation */
1548 { LPROC_LL_READ_BYTES, LPROCFS_CNTR_AVGMINMAX | LPROCFS_TYPE_BYTES,
1550 { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX | LPROCFS_TYPE_BYTES,
1552 { LPROC_LL_READ, LPROCFS_TYPE_LATENCY, "read" },
1553 { LPROC_LL_WRITE, LPROCFS_TYPE_LATENCY, "write" },
1554 { LPROC_LL_IOCTL, LPROCFS_TYPE_REQS, "ioctl" },
1555 { LPROC_LL_OPEN, LPROCFS_TYPE_LATENCY, "open" },
1556 { LPROC_LL_RELEASE, LPROCFS_TYPE_LATENCY, "close" },
1557 { LPROC_LL_MMAP, LPROCFS_TYPE_LATENCY, "mmap" },
1558 { LPROC_LL_FAULT, LPROCFS_TYPE_LATENCY, "page_fault" },
1559 { LPROC_LL_MKWRITE, LPROCFS_TYPE_LATENCY, "page_mkwrite" },
1560 { LPROC_LL_LLSEEK, LPROCFS_TYPE_LATENCY, "seek" },
1561 { LPROC_LL_FSYNC, LPROCFS_TYPE_LATENCY, "fsync" },
1562 { LPROC_LL_READDIR, LPROCFS_TYPE_LATENCY, "readdir" },
1563 /* inode operation */
1564 { LPROC_LL_SETATTR, LPROCFS_TYPE_LATENCY, "setattr" },
1565 { LPROC_LL_TRUNC, LPROCFS_TYPE_LATENCY, "truncate" },
1566 { LPROC_LL_FLOCK, LPROCFS_TYPE_LATENCY, "flock" },
1567 { LPROC_LL_GETATTR, LPROCFS_TYPE_LATENCY, "getattr" },
1568 /* dir inode operation */
1569 { LPROC_LL_CREATE, LPROCFS_TYPE_LATENCY, "create" },
1570 { LPROC_LL_LINK, LPROCFS_TYPE_LATENCY, "link" },
1571 { LPROC_LL_UNLINK, LPROCFS_TYPE_LATENCY, "unlink" },
1572 { LPROC_LL_SYMLINK, LPROCFS_TYPE_LATENCY, "symlink" },
1573 { LPROC_LL_MKDIR, LPROCFS_TYPE_LATENCY, "mkdir" },
1574 { LPROC_LL_RMDIR, LPROCFS_TYPE_LATENCY, "rmdir" },
1575 { LPROC_LL_MKNOD, LPROCFS_TYPE_LATENCY, "mknod" },
1576 { LPROC_LL_RENAME, LPROCFS_TYPE_LATENCY, "rename" },
1577 /* special inode operation */
1578 { LPROC_LL_STATFS, LPROCFS_TYPE_LATENCY, "statfs" },
1579 { LPROC_LL_SETXATTR, LPROCFS_TYPE_LATENCY, "setxattr" },
1580 { LPROC_LL_GETXATTR, LPROCFS_TYPE_LATENCY, "getxattr" },
1581 { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REQS, "getxattr_hits" },
1582 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_LATENCY, "listxattr" },
1583 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_LATENCY, "removexattr" },
1584 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_LATENCY, "inode_permission" },
1587 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count)
1592 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1593 lprocfs_counter_add(sbi->ll_stats, op, count);
1594 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1595 sbi->ll_stats_track_id == current->pid)
1596 lprocfs_counter_add(sbi->ll_stats, op, count);
1597 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1598 sbi->ll_stats_track_id == current->parent->pid)
1599 lprocfs_counter_add(sbi->ll_stats, op, count);
1600 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1601 sbi->ll_stats_track_id ==
1602 from_kgid(&init_user_ns, current_gid()))
1603 lprocfs_counter_add(sbi->ll_stats, op, count);
1605 EXPORT_SYMBOL(ll_stats_ops_tally);
1607 static const char *ra_stat_string[] = {
1608 [RA_STAT_HIT] = "hits",
1609 [RA_STAT_MISS] = "misses",
1610 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1611 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1612 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1613 [RA_STAT_FAILED_MATCH] = "failed lock match",
1614 [RA_STAT_DISCARDED] = "read but discarded",
1615 [RA_STAT_ZERO_LEN] = "zero length file",
1616 [RA_STAT_ZERO_WINDOW] = "zero size window",
1617 [RA_STAT_EOF] = "read-ahead to EOF",
1618 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1619 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1620 [RA_STAT_FAILED_REACH_END] = "failed to reach end",
1621 [RA_STAT_ASYNC] = "async readahead",
1622 [RA_STAT_FAILED_FAST_READ] = "failed to fast read",
1625 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1627 struct lustre_sb_info *lsi = s2lsi(sb);
1628 struct ll_sb_info *sbi = ll_s2sbi(sb);
1634 if (IS_ERR_OR_NULL(llite_root))
1637 sbi->ll_debugfs_entry = ldebugfs_register(name, llite_root,
1638 lprocfs_llite_obd_vars, sb);
1639 if (IS_ERR_OR_NULL(sbi->ll_debugfs_entry)) {
1640 err = sbi->ll_debugfs_entry ? PTR_ERR(sbi->ll_debugfs_entry) :
1642 sbi->ll_debugfs_entry = NULL;
1646 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "dump_page_cache",0444,
1647 &vvp_dump_pgcache_file_ops, sbi);
1649 CWARN("Error adding the dump_page_cache file\n");
1651 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "extents_stats", 0644,
1652 &ll_rw_extents_stats_fops, sbi);
1654 CWARN("Error adding the extent_stats file\n");
1656 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry,
1657 "extents_stats_per_process", 0644,
1658 &ll_rw_extents_stats_pp_fops, sbi);
1660 CWARN("Error adding the extents_stats_per_process file\n");
1662 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "offset_stats", 0644,
1663 &ll_rw_offset_stats_fops, sbi);
1665 CWARN("Error adding the offset_stats file\n");
1667 /* File operations stats */
1668 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1669 LPROCFS_STATS_FLAG_NONE);
1670 if (sbi->ll_stats == NULL)
1671 GOTO(out_debugfs, err = -ENOMEM);
1673 /* do counter init */
1674 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1675 u32 type = llite_opcode_table[id].type;
1678 if (type & LPROCFS_TYPE_REQS)
1680 else if (type & LPROCFS_TYPE_BYTES)
1682 else if (type & LPROCFS_TYPE_PAGES)
1684 else if (type & LPROCFS_TYPE_USEC)
1686 lprocfs_counter_init(sbi->ll_stats,
1687 llite_opcode_table[id].opcode,
1688 (type & LPROCFS_CNTR_AVGMINMAX),
1689 llite_opcode_table[id].opname, ptr);
1692 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "stats",
1695 GOTO(out_stats, err);
1697 sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1698 LPROCFS_STATS_FLAG_NONE);
1699 if (sbi->ll_ra_stats == NULL)
1700 GOTO(out_stats, err = -ENOMEM);
1702 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1703 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1704 ra_stat_string[id], "pages");
1706 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "read_ahead_stats",
1709 GOTO(out_ra_stats, err);
1712 /* Yes we also register sysfs mount kset here as well */
1713 sbi->ll_kset.kobj.parent = llite_kobj;
1714 sbi->ll_kset.kobj.ktype = &sbi_ktype;
1715 init_completion(&sbi->ll_kobj_unregister);
1716 err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1718 GOTO(out_ra_stats, err);
1720 err = kset_register(&sbi->ll_kset);
1722 GOTO(out_ra_stats, err);
1724 lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1728 lprocfs_free_stats(&sbi->ll_ra_stats);
1730 lprocfs_free_stats(&sbi->ll_stats);
1732 debugfs_remove_recursive(sbi->ll_debugfs_entry);
1737 void ll_debugfs_unregister_super(struct super_block *sb)
1739 struct lustre_sb_info *lsi = s2lsi(sb);
1740 struct ll_sb_info *sbi = ll_s2sbi(sb);
1742 debugfs_remove_recursive(sbi->ll_debugfs_entry);
1745 sysfs_remove_link(&sbi->ll_kset.kobj,
1746 sbi->ll_dt_obd->obd_type->typ_name);
1749 sysfs_remove_link(&sbi->ll_kset.kobj,
1750 sbi->ll_md_obd->obd_type->typ_name);
1752 kobject_put(lsi->lsi_kobj);
1754 kset_unregister(&sbi->ll_kset);
1755 wait_for_completion(&sbi->ll_kobj_unregister);
1757 lprocfs_free_stats(&sbi->ll_ra_stats);
1758 lprocfs_free_stats(&sbi->ll_stats);
1760 #undef MAX_STRING_SIZE
1762 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1763 struct seq_file *seq, int which)
1765 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1766 unsigned long start, end, r, w;
1767 char *unitp = "KMGTPEZY";
1769 struct per_process_info *pp_info = &io_extents->pp_extents[which];
1775 for(i = 0; i < LL_HIST_MAX; i++) {
1776 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1777 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1780 for(i = 0; i < LL_HIST_MAX; i++) {
1781 r = pp_info->pp_r_hist.oh_buckets[i];
1782 w = pp_info->pp_w_hist.oh_buckets[i];
1785 end = BIT(i + LL_HIST_START - units);
1786 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u | "
1787 "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1788 (i == LL_HIST_MAX - 1) ? '+' : ' ',
1789 r, pct(r, read_tot), pct(read_cum, read_tot),
1790 w, pct(w, write_tot), pct(write_cum, write_tot));
1792 if (start == BIT(10)) {
1797 if (read_cum == read_tot && write_cum == write_tot)
1802 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1804 struct timespec64 now;
1805 struct ll_sb_info *sbi = seq->private;
1806 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1809 ktime_get_real_ts64(&now);
1811 if (!sbi->ll_rw_stats_on) {
1812 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1815 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1816 (s64)now.tv_sec, now.tv_nsec);
1817 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1818 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1819 "extents", "calls", "%", "cum%",
1820 "calls", "%", "cum%");
1821 spin_lock(&sbi->ll_pp_extent_lock);
1822 for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1823 if (io_extents->pp_extents[k].pid != 0) {
1824 seq_printf(seq, "\nPID: %d\n",
1825 io_extents->pp_extents[k].pid);
1826 ll_display_extents_info(io_extents, seq, k);
1829 spin_unlock(&sbi->ll_pp_extent_lock);
1833 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1834 const char __user *buf,
1838 struct seq_file *seq = file->private_data;
1839 struct ll_sb_info *sbi = seq->private;
1840 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1847 value = ll_stats_pid_write(buf, len);
1850 sbi->ll_rw_stats_on = 0;
1852 sbi->ll_rw_stats_on = 1;
1854 spin_lock(&sbi->ll_pp_extent_lock);
1855 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1856 io_extents->pp_extents[i].pid = 0;
1857 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1858 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1860 spin_unlock(&sbi->ll_pp_extent_lock);
1864 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1866 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1868 struct timespec64 now;
1869 struct ll_sb_info *sbi = seq->private;
1870 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1872 ktime_get_real_ts64(&now);
1874 if (!sbi->ll_rw_stats_on) {
1875 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1878 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1879 (s64)now.tv_sec, now.tv_nsec);
1881 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1882 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1883 "extents", "calls", "%", "cum%",
1884 "calls", "%", "cum%");
1885 spin_lock(&sbi->ll_lock);
1886 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1887 spin_unlock(&sbi->ll_lock);
1892 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1893 const char __user *buf,
1894 size_t len, loff_t *off)
1896 struct seq_file *seq = file->private_data;
1897 struct ll_sb_info *sbi = seq->private;
1898 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1905 value = ll_stats_pid_write(buf, len);
1908 sbi->ll_rw_stats_on = 0;
1910 sbi->ll_rw_stats_on = 1;
1912 spin_lock(&sbi->ll_pp_extent_lock);
1913 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1914 io_extents->pp_extents[i].pid = 0;
1915 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1916 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1918 spin_unlock(&sbi->ll_pp_extent_lock);
1923 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1925 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1926 struct ll_file_data *file, loff_t pos,
1927 size_t count, int rw)
1930 struct ll_rw_process_info *process;
1931 struct ll_rw_process_info *offset;
1932 int *off_count = &sbi->ll_rw_offset_entry_count;
1933 int *process_count = &sbi->ll_offset_process_count;
1934 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1936 if(!sbi->ll_rw_stats_on)
1938 process = sbi->ll_rw_process_info;
1939 offset = sbi->ll_rw_offset_info;
1941 spin_lock(&sbi->ll_pp_extent_lock);
1942 /* Extent statistics */
1943 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1944 if(io_extents->pp_extents[i].pid == pid) {
1952 sbi->ll_extent_process_count =
1953 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1954 cur = sbi->ll_extent_process_count;
1955 io_extents->pp_extents[cur].pid = pid;
1956 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1957 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1960 for (i = 0; (count >= BIT(LL_HIST_START + i)) &&
1961 (i < (LL_HIST_MAX - 1)); i++);
1963 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1964 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1966 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1967 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1969 spin_unlock(&sbi->ll_pp_extent_lock);
1971 spin_lock(&sbi->ll_process_lock);
1972 /* Offset statistics */
1973 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1974 if (process[i].rw_pid == pid) {
1975 if (process[i].rw_last_file != file) {
1976 process[i].rw_range_start = pos;
1977 process[i].rw_last_file_pos = pos + count;
1978 process[i].rw_smallest_extent = count;
1979 process[i].rw_largest_extent = count;
1980 process[i].rw_offset = 0;
1981 process[i].rw_last_file = file;
1982 spin_unlock(&sbi->ll_process_lock);
1985 if (process[i].rw_last_file_pos != pos) {
1987 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1988 offset[*off_count].rw_op = process[i].rw_op;
1989 offset[*off_count].rw_pid = pid;
1990 offset[*off_count].rw_range_start =
1991 process[i].rw_range_start;
1992 offset[*off_count].rw_range_end =
1993 process[i].rw_last_file_pos;
1994 offset[*off_count].rw_smallest_extent =
1995 process[i].rw_smallest_extent;
1996 offset[*off_count].rw_largest_extent =
1997 process[i].rw_largest_extent;
1998 offset[*off_count].rw_offset =
1999 process[i].rw_offset;
2000 process[i].rw_op = rw;
2001 process[i].rw_range_start = pos;
2002 process[i].rw_smallest_extent = count;
2003 process[i].rw_largest_extent = count;
2004 process[i].rw_offset = pos -
2005 process[i].rw_last_file_pos;
2007 if(process[i].rw_smallest_extent > count)
2008 process[i].rw_smallest_extent = count;
2009 if(process[i].rw_largest_extent < count)
2010 process[i].rw_largest_extent = count;
2011 process[i].rw_last_file_pos = pos + count;
2012 spin_unlock(&sbi->ll_process_lock);
2016 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
2017 process[*process_count].rw_pid = pid;
2018 process[*process_count].rw_op = rw;
2019 process[*process_count].rw_range_start = pos;
2020 process[*process_count].rw_last_file_pos = pos + count;
2021 process[*process_count].rw_smallest_extent = count;
2022 process[*process_count].rw_largest_extent = count;
2023 process[*process_count].rw_offset = 0;
2024 process[*process_count].rw_last_file = file;
2025 spin_unlock(&sbi->ll_process_lock);
2028 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
2030 struct timespec64 now;
2031 struct ll_sb_info *sbi = seq->private;
2032 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
2033 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
2036 ktime_get_real_ts64(&now);
2038 if (!sbi->ll_rw_stats_on) {
2039 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2042 spin_lock(&sbi->ll_process_lock);
2044 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
2045 (s64)now.tv_sec, now.tv_nsec);
2046 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
2047 "R/W", "PID", "RANGE START", "RANGE END",
2048 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
2050 /* We stored the discontiguous offsets here; print them first */
2051 for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
2052 if (offset[i].rw_pid != 0)
2054 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2055 offset[i].rw_op == READ ? 'R' : 'W',
2057 offset[i].rw_range_start,
2058 offset[i].rw_range_end,
2059 (unsigned long)offset[i].rw_smallest_extent,
2060 (unsigned long)offset[i].rw_largest_extent,
2061 offset[i].rw_offset);
2064 /* Then print the current offsets for each process */
2065 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2066 if (process[i].rw_pid != 0)
2068 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2069 process[i].rw_op == READ ? 'R' : 'W',
2071 process[i].rw_range_start,
2072 process[i].rw_last_file_pos,
2073 (unsigned long)process[i].rw_smallest_extent,
2074 (unsigned long)process[i].rw_largest_extent,
2075 process[i].rw_offset);
2077 spin_unlock(&sbi->ll_process_lock);
2082 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
2083 const char __user *buf,
2084 size_t len, loff_t *off)
2086 struct seq_file *seq = file->private_data;
2087 struct ll_sb_info *sbi = seq->private;
2088 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
2089 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
2095 value = ll_stats_pid_write(buf, len);
2098 sbi->ll_rw_stats_on = 0;
2100 sbi->ll_rw_stats_on = 1;
2102 spin_lock(&sbi->ll_process_lock);
2103 sbi->ll_offset_process_count = 0;
2104 sbi->ll_rw_offset_entry_count = 0;
2105 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
2106 LL_PROCESS_HIST_MAX);
2107 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
2108 LL_OFFSET_HIST_MAX);
2109 spin_unlock(&sbi->ll_process_lock);
2114 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);