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);
81 kobject_put(llite_kobj);
87 void llite_tunables_unregister(void)
89 kobject_put(llite_kobj);
93 /* <debugfs>/lustre/llite mount point registration */
94 static const struct file_operations ll_rw_extents_stats_fops;
95 static const struct file_operations ll_rw_extents_stats_pp_fops;
96 static const struct file_operations ll_rw_offset_stats_fops;
99 * ll_stats_pid_write() - Determine if stats collection should be enabled
100 * @buf: Buffer containing the data written
101 * @len: Number of bytes in the buffer
103 * Several proc files begin collecting stats when a value is written, and stop
104 * collecting when either '0' or 'disable' is written. This function checks the
105 * written value to see if collection should be enabled or disabled.
107 * Return: If '0' or 'disable' is provided, 0 is returned. If the text
108 * equivalent of a number is written, that number is returned. Otherwise,
109 * 1 is returned. Non-zero return values indicate collection should be enabled.
111 static s64 ll_stats_pid_write(const char __user *buf, size_t len)
113 unsigned long long value = 1;
117 rc = kstrtoull_from_user(buf, len, 0, &value);
118 if (rc < 0 && len < sizeof(kernbuf)) {
119 if (copy_from_user(kernbuf, buf, len))
123 if (kernbuf[len - 1] == '\n')
124 kernbuf[len - 1] = 0;
126 if (strncasecmp(kernbuf, "disable", 7) == 0)
133 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
136 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
138 struct obd_statfs osfs;
141 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
145 return sprintf(buf, "%u\n", osfs.os_bsize);
147 LUSTRE_RO_ATTR(blocksize);
149 static ssize_t stat_blocksize_show(struct kobject *kobj, struct attribute *attr,
152 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
155 return sprintf(buf, "%u\n", sbi->ll_stat_blksize);
158 static ssize_t stat_blocksize_store(struct kobject *kobj,
159 struct attribute *attr,
163 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
168 rc = kstrtouint(buffer, 10, &val);
172 if (val != 0 && (val < PAGE_SIZE || (val & (val - 1))) != 0)
175 sbi->ll_stat_blksize = val;
179 LUSTRE_RW_ATTR(stat_blocksize);
181 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
184 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
186 struct obd_statfs osfs;
191 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
195 blk_size = osfs.os_bsize >> 10;
196 result = osfs.os_blocks;
198 while (blk_size >>= 1)
201 return sprintf(buf, "%llu\n", result);
203 LUSTRE_RO_ATTR(kbytestotal);
205 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
208 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
210 struct obd_statfs osfs;
215 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
219 blk_size = osfs.os_bsize >> 10;
220 result = osfs.os_bfree;
222 while (blk_size >>= 1)
225 return sprintf(buf, "%llu\n", result);
227 LUSTRE_RO_ATTR(kbytesfree);
229 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
232 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
234 struct obd_statfs osfs;
239 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
243 blk_size = osfs.os_bsize >> 10;
244 result = osfs.os_bavail;
246 while (blk_size >>= 1)
249 return sprintf(buf, "%llu\n", result);
251 LUSTRE_RO_ATTR(kbytesavail);
253 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
256 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
258 struct obd_statfs osfs;
261 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
265 return sprintf(buf, "%llu\n", osfs.os_files);
267 LUSTRE_RO_ATTR(filestotal);
269 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
272 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
274 struct obd_statfs osfs;
277 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
281 return sprintf(buf, "%llu\n", osfs.os_ffree);
283 LUSTRE_RO_ATTR(filesfree);
285 static ssize_t client_type_show(struct kobject *kobj, struct attribute *attr,
288 return sprintf(buf, "local client\n");
290 LUSTRE_RO_ATTR(client_type);
292 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
295 return sprintf(buf, "lustre\n");
297 LUSTRE_RO_ATTR(fstype);
299 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
302 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
305 return sprintf(buf, "%s\n", sbi->ll_sb_uuid.uuid);
307 LUSTRE_RO_ATTR(uuid);
309 static int ll_site_stats_seq_show(struct seq_file *m, void *v)
311 struct super_block *sb = m->private;
314 * See description of statistical counters in struct cl_site, and
317 return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site), m);
320 LDEBUGFS_SEQ_FOPS_RO(ll_site_stats);
322 static ssize_t max_read_ahead_mb_show(struct kobject *kobj,
323 struct attribute *attr, char *buf)
325 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
327 unsigned long ra_max_mb;
329 spin_lock(&sbi->ll_lock);
330 ra_max_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages);
331 spin_unlock(&sbi->ll_lock);
333 return snprintf(buf, PAGE_SIZE, "%lu\n", ra_max_mb);
336 static ssize_t max_read_ahead_mb_store(struct kobject *kobj,
337 struct attribute *attr,
338 const char *buffer, size_t count)
340 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
342 u64 ra_max_mb, pages_number;
345 rc = sysfs_memparse(buffer, count, &ra_max_mb, "MiB");
349 pages_number = round_up(ra_max_mb, 1024 * 1024) >> PAGE_SHIFT;
350 CDEBUG(D_INFO, "%s: set max_read_ahead_mb=%llu (%llu pages)\n",
351 sbi->ll_fsname, PAGES_TO_MiB(pages_number), pages_number);
352 if (pages_number > cfs_totalram_pages() / 2) {
354 CERROR("%s: cannot set max_read_ahead_mb=%llu > totalram/2=%luMB\n",
355 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
356 PAGES_TO_MiB(cfs_totalram_pages() / 2));
360 spin_lock(&sbi->ll_lock);
361 sbi->ll_ra_info.ra_max_pages = pages_number;
362 spin_unlock(&sbi->ll_lock);
366 LUSTRE_RW_ATTR(max_read_ahead_mb);
368 static ssize_t max_read_ahead_per_file_mb_show(struct kobject *kobj,
369 struct attribute *attr,
372 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
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 return snprintf(buf, PAGE_SIZE, "%lu\n", ra_max_file_mb);
383 static ssize_t max_read_ahead_per_file_mb_store(struct kobject *kobj,
384 struct attribute *attr,
388 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
390 u64 ra_max_file_mb, pages_number;
393 rc = sysfs_memparse(buffer, count, &ra_max_file_mb, "MiB");
397 pages_number = round_up(ra_max_file_mb, 1024 * 1024) >> PAGE_SHIFT;
398 if (pages_number > sbi->ll_ra_info.ra_max_pages) {
399 CERROR("%s: cannot set max_read_ahead_per_file_mb=%llu > max_read_ahead_mb=%lu\n",
400 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
401 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
405 spin_lock(&sbi->ll_lock);
406 sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
407 spin_unlock(&sbi->ll_lock);
411 LUSTRE_RW_ATTR(max_read_ahead_per_file_mb);
413 static ssize_t max_read_ahead_whole_mb_show(struct kobject *kobj,
414 struct attribute *attr, char *buf)
416 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
418 unsigned long ra_max_whole_mb;
420 spin_lock(&sbi->ll_lock);
421 ra_max_whole_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_read_ahead_whole_pages);
422 spin_unlock(&sbi->ll_lock);
424 return snprintf(buf, PAGE_SIZE, "%lu\n", ra_max_whole_mb);
427 static ssize_t max_read_ahead_whole_mb_store(struct kobject *kobj,
428 struct attribute *attr,
429 const char *buffer, size_t count)
431 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
433 u64 ra_max_whole_mb, pages_number;
436 rc = sysfs_memparse(buffer, count, &ra_max_whole_mb, "MiB");
440 pages_number = round_up(ra_max_whole_mb, 1024 * 1024) >> PAGE_SHIFT;
441 /* Cap this at the current max readahead window size, the readahead
442 * algorithm does this anyway so it's pointless to set it larger.
444 if (pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
445 CERROR("%s: cannot set max_read_ahead_whole_mb=%llu > max_read_ahead_per_file_mb=%lu\n",
446 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
447 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
452 spin_lock(&sbi->ll_lock);
453 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
454 spin_unlock(&sbi->ll_lock);
458 LUSTRE_RW_ATTR(max_read_ahead_whole_mb);
460 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
462 struct super_block *sb = m->private;
463 struct ll_sb_info *sbi = ll_s2sbi(sb);
464 struct cl_client_cache *cache = sbi->ll_cache;
468 mutex_lock(&cache->ccc_max_cache_mb_lock);
469 max_cached_mb = PAGES_TO_MiB(cache->ccc_lru_max);
470 unused_mb = PAGES_TO_MiB(atomic_long_read(&cache->ccc_lru_left));
471 mutex_unlock(&cache->ccc_max_cache_mb_lock);
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 mutex_lock(&cache->ccc_max_cache_mb_lock);
526 diff = pages_number - cache->ccc_lru_max;
528 /* easy - add more LRU slots. */
530 atomic_long_add(diff, &cache->ccc_lru_left);
534 env = cl_env_get(&refcheck);
536 GOTO(out_unlock, rc = PTR_ERR(env));
542 /* reduce LRU budget from free slots. */
544 long lru_left_old, lru_left_new, lru_left_ret;
546 lru_left_old = atomic_long_read(&cache->ccc_lru_left);
547 if (lru_left_old == 0)
550 lru_left_new = lru_left_old > diff ?
551 lru_left_old - diff : 0;
553 atomic_long_cmpxchg(&cache->ccc_lru_left,
556 if (likely(lru_left_old == lru_left_ret)) {
557 diff -= lru_left_old - lru_left_new;
558 nrpages += lru_left_old - lru_left_new;
566 if (sbi->ll_dt_exp == NULL) { /* being initialized */
571 /* Request extra free slots to avoid them all being used
572 * by other processes before this can continue shrinking.
574 tmp = diff + min_t(long, diff, MiB_TO_PAGES(1024));
575 /* difficult - have to ask OSCs to drop LRU slots. */
576 rc = obd_set_info_async(env, sbi->ll_dt_exp,
577 sizeof(KEY_CACHE_LRU_SHRINK),
578 KEY_CACHE_LRU_SHRINK,
579 sizeof(tmp), &tmp, NULL);
583 cl_env_put(env, &refcheck);
587 cache->ccc_lru_max = pages_number;
590 atomic_long_add(nrpages, &cache->ccc_lru_left);
593 mutex_unlock(&cache->ccc_max_cache_mb_lock);
596 LDEBUGFS_SEQ_FOPS(ll_max_cached_mb);
598 static ssize_t checksums_show(struct kobject *kobj, struct attribute *attr,
601 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
604 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
607 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
608 const char *buffer, size_t count)
610 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
620 rc = kstrtobool(buffer, &val);
624 sbi->ll_flags |= LL_SBI_CHECKSUM;
626 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
629 rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
630 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
632 CWARN("Failed to set OSC checksum flags: %d\n", rc);
636 LUSTRE_RW_ATTR(checksums);
638 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
640 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
641 enum stats_track_type type)
643 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
646 if (sbi->ll_stats_track_type == type)
647 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
648 else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
649 return sprintf(buf, "0 (all)\n");
651 return sprintf(buf, "untracked\n");
654 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
655 size_t count, enum stats_track_type type)
657 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
662 rc = kstrtoul(buffer, 10, &pid);
666 sbi->ll_stats_track_id = pid;
668 sbi->ll_stats_track_type = STATS_TRACK_ALL;
670 sbi->ll_stats_track_type = type;
671 lprocfs_clear_stats(sbi->ll_stats);
675 static ssize_t stats_track_pid_show(struct kobject *kobj,
676 struct attribute *attr,
679 return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
682 static ssize_t stats_track_pid_store(struct kobject *kobj,
683 struct attribute *attr,
687 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
689 LUSTRE_RW_ATTR(stats_track_pid);
691 static ssize_t stats_track_ppid_show(struct kobject *kobj,
692 struct attribute *attr,
695 return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
698 static ssize_t stats_track_ppid_store(struct kobject *kobj,
699 struct attribute *attr,
703 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
705 LUSTRE_RW_ATTR(stats_track_ppid);
707 static ssize_t stats_track_gid_show(struct kobject *kobj,
708 struct attribute *attr,
711 return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
714 static ssize_t stats_track_gid_store(struct kobject *kobj,
715 struct attribute *attr,
719 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
721 LUSTRE_RW_ATTR(stats_track_gid);
723 static ssize_t statahead_running_max_show(struct kobject *kobj,
724 struct attribute *attr,
727 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
730 return snprintf(buf, 16, "%u\n", sbi->ll_sa_running_max);
733 static ssize_t statahead_running_max_store(struct kobject *kobj,
734 struct attribute *attr,
738 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
743 rc = kstrtoul(buffer, 0, &val);
747 if (val <= LL_SA_RUNNING_MAX) {
748 sbi->ll_sa_running_max = val;
752 CERROR("Bad statahead_running_max value %lu. Valid values "
753 "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
757 LUSTRE_RW_ATTR(statahead_running_max);
759 static ssize_t statahead_max_show(struct kobject *kobj,
760 struct attribute *attr,
763 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
766 return sprintf(buf, "%u\n", sbi->ll_sa_max);
769 static ssize_t statahead_max_store(struct kobject *kobj,
770 struct attribute *attr,
774 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
779 rc = kstrtoul(buffer, 0, &val);
783 if (val <= LL_SA_RPC_MAX)
784 sbi->ll_sa_max = val;
786 CERROR("Bad statahead_max value %lu. Valid values are in the range [0, %d]\n",
791 LUSTRE_RW_ATTR(statahead_max);
793 static ssize_t statahead_agl_show(struct kobject *kobj,
794 struct attribute *attr,
797 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
800 return sprintf(buf, "%u\n", sbi->ll_flags & LL_SBI_AGL_ENABLED ? 1 : 0);
803 static ssize_t statahead_agl_store(struct kobject *kobj,
804 struct attribute *attr,
808 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
813 rc = kstrtobool(buffer, &val);
818 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
820 sbi->ll_flags &= ~LL_SBI_AGL_ENABLED;
824 LUSTRE_RW_ATTR(statahead_agl);
826 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
828 struct super_block *sb = m->private;
829 struct ll_sb_info *sbi = ll_s2sbi(sb);
831 seq_printf(m, "statahead total: %u\n"
832 "statahead wrong: %u\n"
834 atomic_read(&sbi->ll_sa_total),
835 atomic_read(&sbi->ll_sa_wrong),
836 atomic_read(&sbi->ll_agl_total));
840 LDEBUGFS_SEQ_FOPS_RO(ll_statahead_stats);
842 static ssize_t lazystatfs_show(struct kobject *kobj,
843 struct attribute *attr,
846 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
849 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
852 static ssize_t lazystatfs_store(struct kobject *kobj,
853 struct attribute *attr,
857 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
862 rc = kstrtobool(buffer, &val);
867 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
869 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
873 LUSTRE_RW_ATTR(lazystatfs);
875 static ssize_t statfs_max_age_show(struct kobject *kobj, struct attribute *attr,
878 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
881 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_statfs_max_age);
884 static ssize_t statfs_max_age_store(struct kobject *kobj,
885 struct attribute *attr, const char *buffer,
888 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
893 rc = kstrtouint(buffer, 10, &val);
896 if (val > OBD_STATFS_CACHE_MAX_AGE)
899 sbi->ll_statfs_max_age = val;
903 LUSTRE_RW_ATTR(statfs_max_age);
905 static ssize_t max_easize_show(struct kobject *kobj,
906 struct attribute *attr,
909 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
914 rc = ll_get_max_mdsize(sbi, &ealen);
918 /* Limit xattr size returned to userspace based on kernel maximum */
919 return snprintf(buf, PAGE_SIZE, "%u\n",
920 ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
922 LUSTRE_RO_ATTR(max_easize);
925 * Get default_easize.
927 * \see client_obd::cl_default_mds_easize
929 * \param[in] m seq_file handle
930 * \param[in] v unused for single entry
932 * \retval 0 on success
933 * \retval negative negated errno on failure
935 static ssize_t default_easize_show(struct kobject *kobj,
936 struct attribute *attr,
939 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
944 rc = ll_get_default_mdsize(sbi, &ealen);
948 /* Limit xattr size returned to userspace based on kernel maximum */
949 return snprintf(buf, PAGE_SIZE, "%u\n",
950 ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
954 * Set default_easize.
956 * Range checking on the passed value is handled by
957 * ll_set_default_mdsize().
959 * \see client_obd::cl_default_mds_easize
961 * \param[in] file proc file
962 * \param[in] buffer string passed from user space
963 * \param[in] count \a buffer length
964 * \param[in] off unused for single entry
966 * \retval positive \a count on success
967 * \retval negative negated errno on failure
969 static ssize_t default_easize_store(struct kobject *kobj,
970 struct attribute *attr,
974 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
982 rc = kstrtouint(buffer, 10, &val);
986 rc = ll_set_default_mdsize(sbi, val);
992 LUSTRE_RW_ATTR(default_easize);
994 static int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
996 const char *str[] = LL_SBI_FLAGS;
997 struct super_block *sb = m->private;
998 int flags = ll_s2sbi(sb)->ll_flags;
1001 while (flags != 0) {
1002 if (ARRAY_SIZE(str) <= i) {
1003 CERROR("%s: Revise array LL_SBI_FLAGS to match sbi "
1004 "flags please.\n", ll_s2sbi(sb)->ll_fsname);
1009 seq_printf(m, "%s ", str[i]);
1013 seq_printf(m, "\b\n");
1017 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
1019 static ssize_t xattr_cache_show(struct kobject *kobj,
1020 struct attribute *attr,
1023 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1026 return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
1029 static ssize_t xattr_cache_store(struct kobject *kobj,
1030 struct attribute *attr,
1034 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1039 rc = kstrtobool(buffer, &val);
1043 if (val && !(sbi->ll_flags & LL_SBI_XATTR_CACHE))
1046 sbi->ll_xattr_cache_enabled = val;
1047 sbi->ll_xattr_cache_set = 1;
1051 LUSTRE_RW_ATTR(xattr_cache);
1053 static ssize_t tiny_write_show(struct kobject *kobj,
1054 struct attribute *attr,
1057 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1060 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_TINY_WRITE));
1063 static ssize_t tiny_write_store(struct kobject *kobj,
1064 struct attribute *attr,
1068 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1073 rc = kstrtobool(buffer, &val);
1077 spin_lock(&sbi->ll_lock);
1079 sbi->ll_flags |= LL_SBI_TINY_WRITE;
1081 sbi->ll_flags &= ~LL_SBI_TINY_WRITE;
1082 spin_unlock(&sbi->ll_lock);
1086 LUSTRE_RW_ATTR(tiny_write);
1088 static ssize_t max_read_ahead_async_active_show(struct kobject *kobj,
1089 struct attribute *attr,
1092 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1095 return snprintf(buf, PAGE_SIZE, "%u\n",
1096 sbi->ll_ra_info.ra_async_max_active);
1099 static ssize_t max_read_ahead_async_active_store(struct kobject *kobj,
1100 struct attribute *attr,
1104 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1109 rc = kstrtouint(buffer, 10, &val);
1114 * It doesn't make any sense to make it exceed what
1115 * workqueue could acutally support. This can easily
1116 * over subscripe the cores but Lustre internally
1117 * throttles to avoid those impacts.
1119 if (val > WQ_UNBOUND_MAX_ACTIVE) {
1120 CERROR("%s: cannot set max_read_ahead_async_active=%u larger than %u\n",
1121 sbi->ll_fsname, val, WQ_UNBOUND_MAX_ACTIVE);
1125 spin_lock(&sbi->ll_lock);
1126 sbi->ll_ra_info.ra_async_max_active = val;
1127 spin_unlock(&sbi->ll_lock);
1131 LUSTRE_RW_ATTR(max_read_ahead_async_active);
1133 static ssize_t read_ahead_async_file_threshold_mb_show(struct kobject *kobj,
1134 struct attribute *attr,
1137 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1140 return snprintf(buf, PAGE_SIZE, "%lu\n",
1141 PAGES_TO_MiB(sbi->ll_ra_info.ra_async_pages_per_file_threshold));
1145 read_ahead_async_file_threshold_mb_store(struct kobject *kobj,
1146 struct attribute *attr,
1147 const char *buffer, size_t count)
1149 unsigned long pages_number;
1150 unsigned long max_ra_per_file;
1151 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1155 rc = kstrtoul(buffer, 10, &pages_number);
1159 pages_number = MiB_TO_PAGES(pages_number);
1160 max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1161 if (pages_number < 0 || pages_number > max_ra_per_file) {
1162 CERROR("%s: can't set read_ahead_async_file_threshold_mb=%lu > "
1163 "max_read_readahead_per_file_mb=%lu\n", sbi->ll_fsname,
1164 PAGES_TO_MiB(pages_number),
1165 PAGES_TO_MiB(max_ra_per_file));
1168 sbi->ll_ra_info.ra_async_pages_per_file_threshold = pages_number;
1172 LUSTRE_RW_ATTR(read_ahead_async_file_threshold_mb);
1174 static ssize_t fast_read_show(struct kobject *kobj,
1175 struct attribute *attr,
1178 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1181 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_FAST_READ));
1184 static ssize_t fast_read_store(struct kobject *kobj,
1185 struct attribute *attr,
1189 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1194 rc = kstrtobool(buffer, &val);
1198 spin_lock(&sbi->ll_lock);
1200 sbi->ll_flags |= LL_SBI_FAST_READ;
1202 sbi->ll_flags &= ~LL_SBI_FAST_READ;
1203 spin_unlock(&sbi->ll_lock);
1207 LUSTRE_RW_ATTR(fast_read);
1209 static ssize_t file_heat_show(struct kobject *kobj,
1210 struct attribute *attr,
1213 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1216 return snprintf(buf, PAGE_SIZE, "%u\n",
1217 !!(sbi->ll_flags & LL_SBI_FILE_HEAT));
1220 static ssize_t file_heat_store(struct kobject *kobj,
1221 struct attribute *attr,
1225 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1230 rc = kstrtobool(buffer, &val);
1234 spin_lock(&sbi->ll_lock);
1236 sbi->ll_flags |= LL_SBI_FILE_HEAT;
1238 sbi->ll_flags &= ~LL_SBI_FILE_HEAT;
1239 spin_unlock(&sbi->ll_lock);
1243 LUSTRE_RW_ATTR(file_heat);
1245 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1246 struct attribute *attr,
1249 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1252 return snprintf(buf, PAGE_SIZE, "%u\n",
1253 (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1256 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1257 struct attribute *attr,
1261 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1266 rc = kstrtoul(buffer, 10, &val);
1270 if (val < 0 || val > 100)
1273 sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1277 LUSTRE_RW_ATTR(heat_decay_percentage);
1279 static ssize_t heat_period_second_show(struct kobject *kobj,
1280 struct attribute *attr,
1283 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1286 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1289 static ssize_t heat_period_second_store(struct kobject *kobj,
1290 struct attribute *attr,
1294 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1299 rc = kstrtoul(buffer, 10, &val);
1306 sbi->ll_heat_period_second = val;
1310 LUSTRE_RW_ATTR(heat_period_second);
1312 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1314 struct super_block *sb = m->private;
1315 struct ll_sb_info *sbi = ll_s2sbi(sb);
1316 struct cl_client_cache *cache = sbi->ll_cache;
1320 pages = atomic_long_read(&cache->ccc_unstable_nr);
1321 mb = (pages * PAGE_SIZE) >> 20;
1323 seq_printf(m, "unstable_check: %8d\n"
1324 "unstable_pages: %12ld\n"
1325 "unstable_mb: %8d\n",
1326 cache->ccc_unstable_check, pages, mb);
1330 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1331 const char __user *buffer,
1332 size_t count, loff_t *unused)
1334 struct seq_file *seq = file->private_data;
1335 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1342 if (count >= sizeof(kernbuf))
1345 if (copy_from_user(kernbuf, buffer, count))
1349 buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1351 rc = kstrtobool_from_user(buffer, count, &val);
1355 /* borrow lru lock to set the value */
1356 spin_lock(&sbi->ll_cache->ccc_lru_lock);
1357 sbi->ll_cache->ccc_unstable_check = val;
1358 spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1363 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1365 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1367 struct super_block *sb = m->private;
1368 struct ll_sb_info *sbi = ll_s2sbi(sb);
1369 struct root_squash_info *squash = &sbi->ll_squash;
1371 seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1375 static ssize_t ll_root_squash_seq_write(struct file *file,
1376 const char __user *buffer,
1377 size_t count, loff_t *off)
1379 struct seq_file *m = file->private_data;
1380 struct super_block *sb = m->private;
1381 struct ll_sb_info *sbi = ll_s2sbi(sb);
1382 struct root_squash_info *squash = &sbi->ll_squash;
1384 return lprocfs_wr_root_squash(buffer, count, squash, sbi->ll_fsname);
1387 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1389 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1391 struct super_block *sb = m->private;
1392 struct ll_sb_info *sbi = ll_s2sbi(sb);
1393 struct root_squash_info *squash = &sbi->ll_squash;
1396 spin_lock(&squash->rsi_lock);
1397 if (!list_empty(&squash->rsi_nosquash_nids)) {
1398 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1399 &squash->rsi_nosquash_nids);
1403 seq_puts(m, "NONE\n");
1405 spin_unlock(&squash->rsi_lock);
1410 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1411 const char __user *buffer,
1412 size_t count, loff_t *off)
1414 struct seq_file *m = file->private_data;
1415 struct super_block *sb = m->private;
1416 struct ll_sb_info *sbi = ll_s2sbi(sb);
1417 struct root_squash_info *squash = &sbi->ll_squash;
1420 rc = lprocfs_wr_nosquash_nids(buffer, count, squash, sbi->ll_fsname);
1424 ll_compute_rootsquash_state(sbi);
1429 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1431 static int ll_pcc_seq_show(struct seq_file *m, void *v)
1433 struct super_block *sb = m->private;
1434 struct ll_sb_info *sbi = ll_s2sbi(sb);
1436 return pcc_super_dump(&sbi->ll_pcc_super, m);
1439 static ssize_t ll_pcc_seq_write(struct file *file, const char __user *buffer,
1440 size_t count, loff_t *off)
1442 struct seq_file *m = file->private_data;
1443 struct super_block *sb = m->private;
1444 struct ll_sb_info *sbi = ll_s2sbi(sb);
1448 if (count >= LPROCFS_WR_PCC_MAX_CMD)
1451 if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_PCC))
1454 OBD_ALLOC(kernbuf, count + 1);
1455 if (kernbuf == NULL)
1458 if (copy_from_user(kernbuf, buffer, count))
1459 GOTO(out_free_kernbuff, rc = -EFAULT);
1461 rc = pcc_cmd_handle(kernbuf, count, &sbi->ll_pcc_super);
1463 OBD_FREE(kernbuf, count + 1);
1464 return rc ? rc : count;
1466 LPROC_SEQ_FOPS(ll_pcc);
1468 struct lprocfs_vars lprocfs_llite_obd_vars[] = {
1470 .fops = &ll_site_stats_fops },
1471 { .name = "max_cached_mb",
1472 .fops = &ll_max_cached_mb_fops },
1473 { .name = "statahead_stats",
1474 .fops = &ll_statahead_stats_fops },
1475 { .name = "unstable_stats",
1476 .fops = &ll_unstable_stats_fops },
1477 { .name = "sbi_flags",
1478 .fops = &ll_sbi_flags_fops },
1479 { .name = "root_squash",
1480 .fops = &ll_root_squash_fops },
1481 { .name = "nosquash_nids",
1482 .fops = &ll_nosquash_nids_fops },
1484 .fops = &ll_pcc_fops, },
1488 #define MAX_STRING_SIZE 128
1490 static struct attribute *llite_attrs[] = {
1491 &lustre_attr_blocksize.attr,
1492 &lustre_attr_stat_blocksize.attr,
1493 &lustre_attr_kbytestotal.attr,
1494 &lustre_attr_kbytesfree.attr,
1495 &lustre_attr_kbytesavail.attr,
1496 &lustre_attr_filestotal.attr,
1497 &lustre_attr_filesfree.attr,
1498 &lustre_attr_client_type.attr,
1499 &lustre_attr_fstype.attr,
1500 &lustre_attr_uuid.attr,
1501 &lustre_attr_checksums.attr,
1502 &lustre_attr_checksum_pages.attr,
1503 &lustre_attr_max_read_ahead_mb.attr,
1504 &lustre_attr_max_read_ahead_per_file_mb.attr,
1505 &lustre_attr_max_read_ahead_whole_mb.attr,
1506 &lustre_attr_max_read_ahead_async_active.attr,
1507 &lustre_attr_read_ahead_async_file_threshold_mb.attr,
1508 &lustre_attr_stats_track_pid.attr,
1509 &lustre_attr_stats_track_ppid.attr,
1510 &lustre_attr_stats_track_gid.attr,
1511 &lustre_attr_statahead_running_max.attr,
1512 &lustre_attr_statahead_max.attr,
1513 &lustre_attr_statahead_agl.attr,
1514 &lustre_attr_lazystatfs.attr,
1515 &lustre_attr_statfs_max_age.attr,
1516 &lustre_attr_max_easize.attr,
1517 &lustre_attr_default_easize.attr,
1518 &lustre_attr_xattr_cache.attr,
1519 &lustre_attr_fast_read.attr,
1520 &lustre_attr_tiny_write.attr,
1521 &lustre_attr_file_heat.attr,
1522 &lustre_attr_heat_decay_percentage.attr,
1523 &lustre_attr_heat_period_second.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 static const struct llite_file_opcode {
1544 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1545 /* file operation */
1546 { LPROC_LL_READ_BYTES, LPROCFS_TYPE_BYTES_FULL, "read_bytes" },
1547 { LPROC_LL_WRITE_BYTES, LPROCFS_TYPE_BYTES_FULL, "write_bytes" },
1548 { LPROC_LL_READ, LPROCFS_TYPE_LATENCY, "read" },
1549 { LPROC_LL_WRITE, LPROCFS_TYPE_LATENCY, "write" },
1550 { LPROC_LL_IOCTL, LPROCFS_TYPE_REQS, "ioctl" },
1551 { LPROC_LL_OPEN, LPROCFS_TYPE_LATENCY, "open" },
1552 { LPROC_LL_RELEASE, LPROCFS_TYPE_LATENCY, "close" },
1553 { LPROC_LL_MMAP, LPROCFS_TYPE_LATENCY, "mmap" },
1554 { LPROC_LL_FAULT, LPROCFS_TYPE_LATENCY, "page_fault" },
1555 { LPROC_LL_MKWRITE, LPROCFS_TYPE_LATENCY, "page_mkwrite" },
1556 { LPROC_LL_LLSEEK, LPROCFS_TYPE_LATENCY, "seek" },
1557 { LPROC_LL_FSYNC, LPROCFS_TYPE_LATENCY, "fsync" },
1558 { LPROC_LL_READDIR, LPROCFS_TYPE_LATENCY, "readdir" },
1559 /* inode operation */
1560 { LPROC_LL_SETATTR, LPROCFS_TYPE_LATENCY, "setattr" },
1561 { LPROC_LL_TRUNC, LPROCFS_TYPE_LATENCY, "truncate" },
1562 { LPROC_LL_FLOCK, LPROCFS_TYPE_LATENCY, "flock" },
1563 { LPROC_LL_GETATTR, LPROCFS_TYPE_LATENCY, "getattr" },
1564 { LPROC_LL_FALLOCATE, LPROCFS_TYPE_LATENCY, "fallocate"},
1565 /* dir inode operation */
1566 { LPROC_LL_CREATE, LPROCFS_TYPE_LATENCY, "create" },
1567 { LPROC_LL_LINK, LPROCFS_TYPE_LATENCY, "link" },
1568 { LPROC_LL_UNLINK, LPROCFS_TYPE_LATENCY, "unlink" },
1569 { LPROC_LL_SYMLINK, LPROCFS_TYPE_LATENCY, "symlink" },
1570 { LPROC_LL_MKDIR, LPROCFS_TYPE_LATENCY, "mkdir" },
1571 { LPROC_LL_RMDIR, LPROCFS_TYPE_LATENCY, "rmdir" },
1572 { LPROC_LL_MKNOD, LPROCFS_TYPE_LATENCY, "mknod" },
1573 { LPROC_LL_RENAME, LPROCFS_TYPE_LATENCY, "rename" },
1574 /* special inode operation */
1575 { LPROC_LL_STATFS, LPROCFS_TYPE_LATENCY, "statfs" },
1576 { LPROC_LL_SETXATTR, LPROCFS_TYPE_LATENCY, "setxattr" },
1577 { LPROC_LL_GETXATTR, LPROCFS_TYPE_LATENCY, "getxattr" },
1578 { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REQS, "getxattr_hits" },
1579 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_LATENCY, "listxattr" },
1580 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_LATENCY, "removexattr" },
1581 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_LATENCY, "inode_permission" },
1584 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count)
1589 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1590 lprocfs_counter_add(sbi->ll_stats, op, count);
1591 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1592 sbi->ll_stats_track_id == current->pid)
1593 lprocfs_counter_add(sbi->ll_stats, op, count);
1594 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1595 sbi->ll_stats_track_id == current->parent->pid)
1596 lprocfs_counter_add(sbi->ll_stats, op, count);
1597 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1598 sbi->ll_stats_track_id ==
1599 from_kgid(&init_user_ns, current_gid()))
1600 lprocfs_counter_add(sbi->ll_stats, op, count);
1602 EXPORT_SYMBOL(ll_stats_ops_tally);
1604 static const char *ra_stat_string[] = {
1605 [RA_STAT_HIT] = "hits",
1606 [RA_STAT_MISS] = "misses",
1607 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1608 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1609 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1610 [RA_STAT_FAILED_MATCH] = "failed lock match",
1611 [RA_STAT_DISCARDED] = "read but discarded",
1612 [RA_STAT_ZERO_LEN] = "zero length file",
1613 [RA_STAT_ZERO_WINDOW] = "zero size window",
1614 [RA_STAT_EOF] = "read-ahead to EOF",
1615 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1616 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1617 [RA_STAT_FAILED_REACH_END] = "failed to reach end",
1618 [RA_STAT_ASYNC] = "async readahead",
1619 [RA_STAT_FAILED_FAST_READ] = "failed to fast read",
1622 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1624 struct lustre_sb_info *lsi = s2lsi(sb);
1625 struct ll_sb_info *sbi = ll_s2sbi(sb);
1631 if (IS_ERR_OR_NULL(llite_root))
1634 sbi->ll_debugfs_entry = debugfs_create_dir(name, llite_root);
1635 ldebugfs_add_vars(sbi->ll_debugfs_entry, lprocfs_llite_obd_vars, sb);
1637 debugfs_create_file("dump_page_cache", 0444, sbi->ll_debugfs_entry, sbi,
1638 &vvp_dump_pgcache_file_ops);
1640 debugfs_create_file("extents_stats", 0644, sbi->ll_debugfs_entry, sbi,
1641 &ll_rw_extents_stats_fops);
1643 debugfs_create_file("extents_stats_per_process", 0644,
1644 sbi->ll_debugfs_entry, sbi,
1645 &ll_rw_extents_stats_pp_fops);
1647 debugfs_create_file("offset_stats", 0644, sbi->ll_debugfs_entry, sbi,
1648 &ll_rw_offset_stats_fops);
1650 /* File operations stats */
1651 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1652 LPROCFS_STATS_FLAG_NONE);
1653 if (sbi->ll_stats == NULL)
1654 GOTO(out_debugfs, err = -ENOMEM);
1656 /* do counter init */
1657 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1658 u32 type = llite_opcode_table[id].type;
1661 if (type & LPROCFS_TYPE_REQS)
1663 else if (type & LPROCFS_TYPE_BYTES)
1665 else if (type & LPROCFS_TYPE_PAGES)
1667 else if (type & LPROCFS_TYPE_USEC)
1669 lprocfs_counter_init(sbi->ll_stats,
1670 llite_opcode_table[id].opcode,
1671 (type & LPROCFS_CNTR_AVGMINMAX),
1672 llite_opcode_table[id].opname, ptr);
1675 debugfs_create_file("stats", 0644, sbi->ll_debugfs_entry,
1676 sbi->ll_stats, &lprocfs_stats_seq_fops);
1678 sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1679 LPROCFS_STATS_FLAG_NONE);
1680 if (sbi->ll_ra_stats == NULL)
1681 GOTO(out_stats, err = -ENOMEM);
1683 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1684 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1685 ra_stat_string[id], "pages");
1687 debugfs_create_file("read_ahead_stats", 0644, sbi->ll_debugfs_entry,
1688 sbi->ll_ra_stats, &lprocfs_stats_seq_fops);
1691 /* Yes we also register sysfs mount kset here as well */
1692 sbi->ll_kset.kobj.parent = llite_kobj;
1693 sbi->ll_kset.kobj.ktype = &sbi_ktype;
1694 init_completion(&sbi->ll_kobj_unregister);
1695 err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1697 GOTO(out_ra_stats, err);
1699 err = kset_register(&sbi->ll_kset);
1701 GOTO(out_ra_stats, err);
1703 lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1707 lprocfs_free_stats(&sbi->ll_ra_stats);
1709 lprocfs_free_stats(&sbi->ll_stats);
1711 debugfs_remove_recursive(sbi->ll_debugfs_entry);
1716 void ll_debugfs_unregister_super(struct super_block *sb)
1718 struct lustre_sb_info *lsi = s2lsi(sb);
1719 struct ll_sb_info *sbi = ll_s2sbi(sb);
1721 debugfs_remove_recursive(sbi->ll_debugfs_entry);
1724 sysfs_remove_link(&sbi->ll_kset.kobj,
1725 sbi->ll_dt_obd->obd_type->typ_name);
1728 sysfs_remove_link(&sbi->ll_kset.kobj,
1729 sbi->ll_md_obd->obd_type->typ_name);
1731 kobject_put(lsi->lsi_kobj);
1733 kset_unregister(&sbi->ll_kset);
1734 wait_for_completion(&sbi->ll_kobj_unregister);
1736 lprocfs_free_stats(&sbi->ll_ra_stats);
1737 lprocfs_free_stats(&sbi->ll_stats);
1739 #undef MAX_STRING_SIZE
1741 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1742 struct seq_file *seq, int which)
1744 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1745 unsigned long start, end, r, w;
1746 char *unitp = "KMGTPEZY";
1748 struct per_process_info *pp_info = &io_extents->pp_extents[which];
1754 for(i = 0; i < LL_HIST_MAX; i++) {
1755 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1756 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1759 for(i = 0; i < LL_HIST_MAX; i++) {
1760 r = pp_info->pp_r_hist.oh_buckets[i];
1761 w = pp_info->pp_w_hist.oh_buckets[i];
1764 end = 1 << (i + LL_HIST_START - units);
1765 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u | "
1766 "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1767 (i == LL_HIST_MAX - 1) ? '+' : ' ',
1768 r, pct(r, read_tot), pct(read_cum, read_tot),
1769 w, pct(w, write_tot), pct(write_cum, write_tot));
1771 if (start == (1 << 10)) {
1776 if (read_cum == read_tot && write_cum == write_tot)
1781 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1783 struct timespec64 now;
1784 struct ll_sb_info *sbi = seq->private;
1785 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1788 ktime_get_real_ts64(&now);
1790 if (!sbi->ll_rw_stats_on) {
1791 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1794 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1795 (s64)now.tv_sec, now.tv_nsec);
1796 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1797 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1798 "extents", "calls", "%", "cum%",
1799 "calls", "%", "cum%");
1800 spin_lock(&sbi->ll_pp_extent_lock);
1801 for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1802 if (io_extents->pp_extents[k].pid != 0) {
1803 seq_printf(seq, "\nPID: %d\n",
1804 io_extents->pp_extents[k].pid);
1805 ll_display_extents_info(io_extents, seq, k);
1808 spin_unlock(&sbi->ll_pp_extent_lock);
1812 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1813 const char __user *buf,
1817 struct seq_file *seq = file->private_data;
1818 struct ll_sb_info *sbi = seq->private;
1819 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1826 value = ll_stats_pid_write(buf, len);
1829 sbi->ll_rw_stats_on = 0;
1831 sbi->ll_rw_stats_on = 1;
1833 spin_lock(&sbi->ll_pp_extent_lock);
1834 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1835 io_extents->pp_extents[i].pid = 0;
1836 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1837 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1839 spin_unlock(&sbi->ll_pp_extent_lock);
1843 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1845 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1847 struct timespec64 now;
1848 struct ll_sb_info *sbi = seq->private;
1849 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1851 ktime_get_real_ts64(&now);
1853 if (!sbi->ll_rw_stats_on) {
1854 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1857 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1858 (s64)now.tv_sec, now.tv_nsec);
1860 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1861 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1862 "extents", "calls", "%", "cum%",
1863 "calls", "%", "cum%");
1864 spin_lock(&sbi->ll_lock);
1865 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1866 spin_unlock(&sbi->ll_lock);
1871 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1872 const char __user *buf,
1873 size_t len, loff_t *off)
1875 struct seq_file *seq = file->private_data;
1876 struct ll_sb_info *sbi = seq->private;
1877 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1884 value = ll_stats_pid_write(buf, len);
1887 sbi->ll_rw_stats_on = 0;
1889 sbi->ll_rw_stats_on = 1;
1891 spin_lock(&sbi->ll_pp_extent_lock);
1892 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1893 io_extents->pp_extents[i].pid = 0;
1894 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1895 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1897 spin_unlock(&sbi->ll_pp_extent_lock);
1902 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1904 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1905 struct ll_file_data *file, loff_t pos,
1906 size_t count, int rw)
1909 struct ll_rw_process_info *process;
1910 struct ll_rw_process_info *offset;
1911 int *off_count = &sbi->ll_rw_offset_entry_count;
1912 int *process_count = &sbi->ll_offset_process_count;
1913 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1915 if(!sbi->ll_rw_stats_on)
1917 process = sbi->ll_rw_process_info;
1918 offset = sbi->ll_rw_offset_info;
1920 spin_lock(&sbi->ll_pp_extent_lock);
1921 /* Extent statistics */
1922 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1923 if(io_extents->pp_extents[i].pid == pid) {
1931 sbi->ll_extent_process_count =
1932 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1933 cur = sbi->ll_extent_process_count;
1934 io_extents->pp_extents[cur].pid = pid;
1935 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1936 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1939 for (i = 0; (count >= 1 << (LL_HIST_START + i)) &&
1940 (i < (LL_HIST_MAX - 1)); i++);
1942 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1943 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1945 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1946 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1948 spin_unlock(&sbi->ll_pp_extent_lock);
1950 spin_lock(&sbi->ll_process_lock);
1951 /* Offset statistics */
1952 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1953 if (process[i].rw_pid == pid) {
1954 if (process[i].rw_last_file != file) {
1955 process[i].rw_range_start = pos;
1956 process[i].rw_last_file_pos = pos + count;
1957 process[i].rw_smallest_extent = count;
1958 process[i].rw_largest_extent = count;
1959 process[i].rw_offset = 0;
1960 process[i].rw_last_file = file;
1961 spin_unlock(&sbi->ll_process_lock);
1964 if (process[i].rw_last_file_pos != pos) {
1966 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1967 offset[*off_count].rw_op = process[i].rw_op;
1968 offset[*off_count].rw_pid = pid;
1969 offset[*off_count].rw_range_start =
1970 process[i].rw_range_start;
1971 offset[*off_count].rw_range_end =
1972 process[i].rw_last_file_pos;
1973 offset[*off_count].rw_smallest_extent =
1974 process[i].rw_smallest_extent;
1975 offset[*off_count].rw_largest_extent =
1976 process[i].rw_largest_extent;
1977 offset[*off_count].rw_offset =
1978 process[i].rw_offset;
1979 process[i].rw_op = rw;
1980 process[i].rw_range_start = pos;
1981 process[i].rw_smallest_extent = count;
1982 process[i].rw_largest_extent = count;
1983 process[i].rw_offset = pos -
1984 process[i].rw_last_file_pos;
1986 if(process[i].rw_smallest_extent > count)
1987 process[i].rw_smallest_extent = count;
1988 if(process[i].rw_largest_extent < count)
1989 process[i].rw_largest_extent = count;
1990 process[i].rw_last_file_pos = pos + count;
1991 spin_unlock(&sbi->ll_process_lock);
1995 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1996 process[*process_count].rw_pid = pid;
1997 process[*process_count].rw_op = rw;
1998 process[*process_count].rw_range_start = pos;
1999 process[*process_count].rw_last_file_pos = pos + count;
2000 process[*process_count].rw_smallest_extent = count;
2001 process[*process_count].rw_largest_extent = count;
2002 process[*process_count].rw_offset = 0;
2003 process[*process_count].rw_last_file = file;
2004 spin_unlock(&sbi->ll_process_lock);
2007 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
2009 struct timespec64 now;
2010 struct ll_sb_info *sbi = seq->private;
2011 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
2012 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
2015 ktime_get_real_ts64(&now);
2017 if (!sbi->ll_rw_stats_on) {
2018 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2021 spin_lock(&sbi->ll_process_lock);
2023 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
2024 (s64)now.tv_sec, now.tv_nsec);
2025 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
2026 "R/W", "PID", "RANGE START", "RANGE END",
2027 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
2029 /* We stored the discontiguous offsets here; print them first */
2030 for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
2031 if (offset[i].rw_pid != 0)
2033 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2034 offset[i].rw_op == READ ? 'R' : 'W',
2036 offset[i].rw_range_start,
2037 offset[i].rw_range_end,
2038 (unsigned long)offset[i].rw_smallest_extent,
2039 (unsigned long)offset[i].rw_largest_extent,
2040 offset[i].rw_offset);
2043 /* Then print the current offsets for each process */
2044 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2045 if (process[i].rw_pid != 0)
2047 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2048 process[i].rw_op == READ ? 'R' : 'W',
2050 process[i].rw_range_start,
2051 process[i].rw_last_file_pos,
2052 (unsigned long)process[i].rw_smallest_extent,
2053 (unsigned long)process[i].rw_largest_extent,
2054 process[i].rw_offset);
2056 spin_unlock(&sbi->ll_process_lock);
2061 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
2062 const char __user *buf,
2063 size_t len, loff_t *off)
2065 struct seq_file *seq = file->private_data;
2066 struct ll_sb_info *sbi = seq->private;
2067 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
2068 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
2074 value = ll_stats_pid_write(buf, len);
2077 sbi->ll_rw_stats_on = 0;
2079 sbi->ll_rw_stats_on = 1;
2081 spin_lock(&sbi->ll_process_lock);
2082 sbi->ll_offset_process_count = 0;
2083 sbi->ll_rw_offset_entry_count = 0;
2084 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
2085 LL_PROCESS_HIST_MAX);
2086 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
2087 LL_OFFSET_HIST_MAX);
2088 spin_unlock(&sbi->ll_process_lock);
2093 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);