4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 #define DEBUG_SUBSYSTEM S_LLITE
34 #include <linux/version.h>
35 #include <linux/user_namespace.h>
36 #ifdef HAVE_UIDGID_HEADER
37 # include <linux/uidgid.h>
39 #include <uapi/linux/lustre/lustre_param.h>
40 #include <lprocfs_status.h>
41 #include <obd_support.h>
43 #include "llite_internal.h"
44 #include "vvp_internal.h"
46 static struct kobject *llite_kobj;
47 static struct dentry *llite_root;
49 static void llite_kobj_release(struct kobject *kobj)
51 if (!IS_ERR_OR_NULL(llite_root)) {
52 debugfs_remove(llite_root);
59 static struct kobj_type llite_kobj_ktype = {
60 .release = llite_kobj_release,
61 .sysfs_ops = &lustre_sysfs_ops,
64 int llite_tunables_register(void)
68 llite_kobj = kzalloc(sizeof(*llite_kobj), GFP_KERNEL);
72 llite_kobj->kset = lustre_kset;
73 rc = kobject_init_and_add(llite_kobj, &llite_kobj_ktype,
74 &lustre_kset->kobj, "%s", "llite");
78 llite_root = debugfs_create_dir("llite", debugfs_lustre_root);
79 if (IS_ERR_OR_NULL(llite_root)) {
80 rc = llite_root ? PTR_ERR(llite_root) : -ENOMEM;
83 kobject_put(llite_kobj);
90 void llite_tunables_unregister(void)
92 kobject_put(llite_kobj);
96 /* <debugfs>/lustre/llite mount point registration */
97 static const struct file_operations ll_rw_extents_stats_fops;
98 static const struct file_operations ll_rw_extents_stats_pp_fops;
99 static const struct file_operations ll_rw_offset_stats_fops;
102 * ll_stats_pid_write() - Determine if stats collection should be enabled
103 * @buf: Buffer containing the data written
104 * @len: Number of bytes in the buffer
106 * Several proc files begin collecting stats when a value is written, and stop
107 * collecting when either '0' or 'disable' is written. This function checks the
108 * written value to see if collection should be enabled or disabled.
110 * Return: If '0' or 'disable' is provided, 0 is returned. If the text
111 * equivalent of a number is written, that number is returned. Otherwise,
112 * 1 is returned. Non-zero return values indicate collection should be enabled.
114 static s64 ll_stats_pid_write(const char __user *buf, size_t len)
116 unsigned long long value = 1;
120 rc = kstrtoull_from_user(buf, len, 0, &value);
121 if (rc < 0 && len < sizeof(kernbuf)) {
122 if (copy_from_user(kernbuf, buf, len))
126 if (kernbuf[len - 1] == '\n')
127 kernbuf[len - 1] = 0;
129 if (strncasecmp(kernbuf, "disable", 7) == 0)
136 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
139 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
141 struct obd_statfs osfs;
144 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
148 return sprintf(buf, "%u\n", osfs.os_bsize);
150 LUSTRE_RO_ATTR(blocksize);
152 static ssize_t stat_blocksize_show(struct kobject *kobj, struct attribute *attr,
155 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
158 return sprintf(buf, "%u\n", sbi->ll_stat_blksize);
161 static ssize_t stat_blocksize_store(struct kobject *kobj,
162 struct attribute *attr,
166 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
171 rc = kstrtouint(buffer, 10, &val);
175 if (val != 0 && (val < PAGE_SIZE || (val & (val - 1))) != 0)
178 sbi->ll_stat_blksize = val;
182 LUSTRE_RW_ATTR(stat_blocksize);
184 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
187 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
189 struct obd_statfs osfs;
194 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
198 blk_size = osfs.os_bsize >> 10;
199 result = osfs.os_blocks;
201 while (blk_size >>= 1)
204 return sprintf(buf, "%llu\n", result);
206 LUSTRE_RO_ATTR(kbytestotal);
208 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
211 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
213 struct obd_statfs osfs;
218 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
222 blk_size = osfs.os_bsize >> 10;
223 result = osfs.os_bfree;
225 while (blk_size >>= 1)
228 return sprintf(buf, "%llu\n", result);
230 LUSTRE_RO_ATTR(kbytesfree);
232 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
235 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
237 struct obd_statfs osfs;
242 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
246 blk_size = osfs.os_bsize >> 10;
247 result = osfs.os_bavail;
249 while (blk_size >>= 1)
252 return sprintf(buf, "%llu\n", result);
254 LUSTRE_RO_ATTR(kbytesavail);
256 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
259 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
261 struct obd_statfs osfs;
264 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
268 return sprintf(buf, "%llu\n", osfs.os_files);
270 LUSTRE_RO_ATTR(filestotal);
272 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
275 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
277 struct obd_statfs osfs;
280 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
284 return sprintf(buf, "%llu\n", osfs.os_ffree);
286 LUSTRE_RO_ATTR(filesfree);
288 static ssize_t client_type_show(struct kobject *kobj, struct attribute *attr,
291 return sprintf(buf, "local client\n");
293 LUSTRE_RO_ATTR(client_type);
295 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
298 return sprintf(buf, "lustre\n");
300 LUSTRE_RO_ATTR(fstype);
302 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
305 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
308 return sprintf(buf, "%s\n", sbi->ll_sb_uuid.uuid);
310 LUSTRE_RO_ATTR(uuid);
312 static int ll_site_stats_seq_show(struct seq_file *m, void *v)
314 struct super_block *sb = m->private;
317 * See description of statistical counters in struct cl_site, and
320 return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site), m);
323 LDEBUGFS_SEQ_FOPS_RO(ll_site_stats);
325 static int ll_max_readahead_mb_seq_show(struct seq_file *m, void *v)
327 struct super_block *sb = m->private;
328 struct ll_sb_info *sbi = ll_s2sbi(sb);
332 spin_lock(&sbi->ll_lock);
333 pages_number = sbi->ll_ra_info.ra_max_pages;
334 spin_unlock(&sbi->ll_lock);
336 mult = 1 << (20 - PAGE_SHIFT);
337 return lprocfs_seq_read_frac_helper(m, pages_number, mult);
341 ll_max_readahead_mb_seq_write(struct file *file, const char __user *buffer,
342 size_t count, loff_t *off)
344 struct seq_file *m = file->private_data;
345 struct super_block *sb = m->private;
346 struct ll_sb_info *sbi = ll_s2sbi(sb);
350 rc = lprocfs_str_with_units_to_s64(buffer, count, &pages_number, 'M');
354 pages_number >>= PAGE_SHIFT;
356 if (pages_number < 0 || pages_number > totalram_pages / 2) {
358 CERROR("%s: can't set max_readahead_mb=%lu > %luMB\n",
359 ll_get_fsname(sb, NULL, 0),
360 (unsigned long)pages_number >> (20 - PAGE_SHIFT),
361 totalram_pages >> (20 - PAGE_SHIFT + 1));
365 spin_lock(&sbi->ll_lock);
366 sbi->ll_ra_info.ra_max_pages = pages_number;
367 spin_unlock(&sbi->ll_lock);
372 LDEBUGFS_SEQ_FOPS(ll_max_readahead_mb);
374 static int ll_max_readahead_per_file_mb_seq_show(struct seq_file *m, void *v)
376 struct super_block *sb = m->private;
377 struct ll_sb_info *sbi = ll_s2sbi(sb);
381 spin_lock(&sbi->ll_lock);
382 pages_number = sbi->ll_ra_info.ra_max_pages_per_file;
383 spin_unlock(&sbi->ll_lock);
385 mult = 1 << (20 - PAGE_SHIFT);
386 return lprocfs_seq_read_frac_helper(m, pages_number, mult);
390 ll_max_readahead_per_file_mb_seq_write(struct file *file,
391 const char __user *buffer,
392 size_t count, loff_t *off)
394 struct seq_file *m = file->private_data;
395 struct super_block *sb = m->private;
396 struct ll_sb_info *sbi = ll_s2sbi(sb);
400 rc = lprocfs_str_with_units_to_s64(buffer, count, &pages_number, 'M');
404 pages_number >>= PAGE_SHIFT;
406 if (pages_number < 0 || pages_number > sbi->ll_ra_info.ra_max_pages) {
407 CERROR("%s: can't set max_readahead_per_file_mb=%lu > "
408 "max_read_ahead_mb=%lu\n", ll_get_fsname(sb, NULL, 0),
409 (unsigned long)pages_number >> (20 - PAGE_SHIFT),
410 sbi->ll_ra_info.ra_max_pages >> (20 - PAGE_SHIFT));
414 spin_lock(&sbi->ll_lock);
415 sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
416 spin_unlock(&sbi->ll_lock);
421 LDEBUGFS_SEQ_FOPS(ll_max_readahead_per_file_mb);
423 static int ll_max_read_ahead_whole_mb_seq_show(struct seq_file *m, void *v)
425 struct super_block *sb = m->private;
426 struct ll_sb_info *sbi = ll_s2sbi(sb);
430 spin_lock(&sbi->ll_lock);
431 pages_number = sbi->ll_ra_info.ra_max_read_ahead_whole_pages;
432 spin_unlock(&sbi->ll_lock);
434 mult = 1 << (20 - PAGE_SHIFT);
435 return lprocfs_seq_read_frac_helper(m, pages_number, mult);
439 ll_max_read_ahead_whole_mb_seq_write(struct file *file,
440 const char __user *buffer,
441 size_t count, loff_t *off)
443 struct seq_file *m = file->private_data;
444 struct super_block *sb = m->private;
445 struct ll_sb_info *sbi = ll_s2sbi(sb);
449 rc = lprocfs_str_with_units_to_s64(buffer, count, &pages_number, 'M');
453 pages_number >>= PAGE_SHIFT;
455 /* Cap this at the current max readahead window size, the readahead
456 * algorithm does this anyway so it's pointless to set it larger. */
457 if (pages_number < 0 ||
458 pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
459 int pages_shift = 20 - PAGE_SHIFT;
460 CERROR("%s: can't set max_read_ahead_whole_mb=%lu > "
461 "max_read_ahead_per_file_mb=%lu\n",
462 ll_get_fsname(sb, NULL, 0),
463 (unsigned long)pages_number >> pages_shift,
464 sbi->ll_ra_info.ra_max_pages_per_file >> pages_shift);
468 spin_lock(&sbi->ll_lock);
469 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
470 spin_unlock(&sbi->ll_lock);
475 LDEBUGFS_SEQ_FOPS(ll_max_read_ahead_whole_mb);
477 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
479 struct super_block *sb = m->private;
480 struct ll_sb_info *sbi = ll_s2sbi(sb);
481 struct cl_client_cache *cache = sbi->ll_cache;
482 int shift = 20 - PAGE_SHIFT;
486 max_cached_mb = cache->ccc_lru_max >> shift;
487 unused_mb = atomic_long_read(&cache->ccc_lru_left) >> shift;
488 seq_printf(m, "users: %d\n"
489 "max_cached_mb: %ld\n"
492 "reclaim_count: %u\n",
493 atomic_read(&cache->ccc_users),
495 max_cached_mb - unused_mb,
497 cache->ccc_lru_shrinkers);
501 static ssize_t ll_max_cached_mb_seq_write(struct file *file,
502 const char __user *buffer,
503 size_t count, loff_t *off)
505 struct seq_file *m = file->private_data;
506 struct super_block *sb = m->private;
507 struct ll_sb_info *sbi = ll_s2sbi(sb);
508 struct cl_client_cache *cache = sbi->ll_cache;
518 if (count >= sizeof(kernbuf))
521 if (copy_from_user(kernbuf, buffer, count))
525 buffer += lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count) -
527 rc = lprocfs_str_with_units_to_s64(buffer, count, &pages_number, 'M');
531 pages_number >>= PAGE_SHIFT;
533 if (pages_number < 0 || pages_number > totalram_pages) {
534 CERROR("%s: can't set max cache more than %lu MB\n",
535 ll_get_fsname(sb, NULL, 0),
536 totalram_pages >> (20 - PAGE_SHIFT));
539 /* Allow enough cache so clients can make well-formed RPCs */
540 pages_number = max_t(long, pages_number, PTLRPC_MAX_BRW_PAGES);
542 spin_lock(&sbi->ll_lock);
543 diff = pages_number - cache->ccc_lru_max;
544 spin_unlock(&sbi->ll_lock);
546 /* easy - add more LRU slots. */
548 atomic_long_add(diff, &cache->ccc_lru_left);
552 env = cl_env_get(&refcheck);
554 RETURN(PTR_ERR(env));
560 /* reduce LRU budget from free slots. */
564 ov = atomic_long_read(&cache->ccc_lru_left);
568 nv = ov > diff ? ov - diff : 0;
569 rc = atomic_long_cmpxchg(&cache->ccc_lru_left, ov, nv);
570 if (likely(ov == rc)) {
580 if (sbi->ll_dt_exp == NULL) { /* being initialized */
585 /* difficult - have to ask OSCs to drop LRU slots. */
587 rc = obd_set_info_async(env, sbi->ll_dt_exp,
588 sizeof(KEY_CACHE_LRU_SHRINK),
589 KEY_CACHE_LRU_SHRINK,
590 sizeof(tmp), &tmp, NULL);
594 cl_env_put(env, &refcheck);
598 spin_lock(&sbi->ll_lock);
599 cache->ccc_lru_max = pages_number;
600 spin_unlock(&sbi->ll_lock);
603 atomic_long_add(nrpages, &cache->ccc_lru_left);
608 LDEBUGFS_SEQ_FOPS(ll_max_cached_mb);
610 static ssize_t checksums_show(struct kobject *kobj, struct attribute *attr,
613 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
616 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
619 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
620 const char *buffer, size_t count)
622 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
632 rc = kstrtobool(buffer, &val);
636 sbi->ll_flags |= LL_SBI_CHECKSUM;
638 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
641 rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
642 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
644 CWARN("Failed to set OSC checksum flags: %d\n", rc);
648 LUSTRE_RW_ATTR(checksums);
650 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
652 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
653 enum stats_track_type type)
655 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
658 if (sbi->ll_stats_track_type == type)
659 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
660 else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
661 return sprintf(buf, "0 (all)\n");
663 return sprintf(buf, "untracked\n");
666 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
667 size_t count, enum stats_track_type type)
669 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
674 rc = kstrtoul(buffer, 10, &pid);
678 sbi->ll_stats_track_id = pid;
680 sbi->ll_stats_track_type = STATS_TRACK_ALL;
682 sbi->ll_stats_track_type = type;
683 lprocfs_clear_stats(sbi->ll_stats);
687 static ssize_t stats_track_pid_show(struct kobject *kobj,
688 struct attribute *attr,
691 return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
694 static ssize_t stats_track_pid_store(struct kobject *kobj,
695 struct attribute *attr,
699 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
701 LUSTRE_RW_ATTR(stats_track_pid);
703 static ssize_t stats_track_ppid_show(struct kobject *kobj,
704 struct attribute *attr,
707 return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
710 static ssize_t stats_track_ppid_store(struct kobject *kobj,
711 struct attribute *attr,
715 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
717 LUSTRE_RW_ATTR(stats_track_ppid);
719 static ssize_t stats_track_gid_show(struct kobject *kobj,
720 struct attribute *attr,
723 return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
726 static ssize_t stats_track_gid_store(struct kobject *kobj,
727 struct attribute *attr,
731 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
733 LUSTRE_RW_ATTR(stats_track_gid);
735 static ssize_t statahead_running_max_show(struct kobject *kobj,
736 struct attribute *attr,
739 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
742 return snprintf(buf, 16, "%u\n", sbi->ll_sa_running_max);
745 static ssize_t statahead_running_max_store(struct kobject *kobj,
746 struct attribute *attr,
750 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
755 rc = kstrtoul(buffer, 0, &val);
759 if (val <= LL_SA_RUNNING_MAX) {
760 sbi->ll_sa_running_max = val;
764 CERROR("Bad statahead_running_max value %lu. Valid values "
765 "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
769 LUSTRE_RW_ATTR(statahead_running_max);
771 static ssize_t statahead_max_show(struct kobject *kobj,
772 struct attribute *attr,
775 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
778 return sprintf(buf, "%u\n", sbi->ll_sa_max);
781 static ssize_t statahead_max_store(struct kobject *kobj,
782 struct attribute *attr,
786 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
791 rc = kstrtoul(buffer, 0, &val);
795 if (val <= LL_SA_RPC_MAX)
796 sbi->ll_sa_max = val;
798 CERROR("Bad statahead_max value %lu. Valid values are in the range [0, %d]\n",
803 LUSTRE_RW_ATTR(statahead_max);
805 static ssize_t statahead_agl_show(struct kobject *kobj,
806 struct attribute *attr,
809 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
812 return sprintf(buf, "%u\n", sbi->ll_flags & LL_SBI_AGL_ENABLED ? 1 : 0);
815 static ssize_t statahead_agl_store(struct kobject *kobj,
816 struct attribute *attr,
820 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
825 rc = kstrtobool(buffer, &val);
830 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
832 sbi->ll_flags &= ~LL_SBI_AGL_ENABLED;
836 LUSTRE_RW_ATTR(statahead_agl);
838 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
840 struct super_block *sb = m->private;
841 struct ll_sb_info *sbi = ll_s2sbi(sb);
843 seq_printf(m, "statahead total: %u\n"
844 "statahead wrong: %u\n"
846 atomic_read(&sbi->ll_sa_total),
847 atomic_read(&sbi->ll_sa_wrong),
848 atomic_read(&sbi->ll_agl_total));
852 LDEBUGFS_SEQ_FOPS_RO(ll_statahead_stats);
854 static ssize_t lazystatfs_show(struct kobject *kobj,
855 struct attribute *attr,
858 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
861 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
864 static ssize_t lazystatfs_store(struct kobject *kobj,
865 struct attribute *attr,
869 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
874 rc = kstrtobool(buffer, &val);
879 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
881 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
885 LUSTRE_RW_ATTR(lazystatfs);
887 static ssize_t max_easize_show(struct kobject *kobj,
888 struct attribute *attr,
891 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
896 rc = ll_get_max_mdsize(sbi, &ealen);
900 return sprintf(buf, "%u\n", ealen);
902 LUSTRE_RO_ATTR(max_easize);
905 * Get default_easize.
907 * \see client_obd::cl_default_mds_easize
909 * \param[in] m seq_file handle
910 * \param[in] v unused for single entry
912 * \retval 0 on success
913 * \retval negative negated errno on failure
915 static ssize_t default_easize_show(struct kobject *kobj,
916 struct attribute *attr,
919 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
924 rc = ll_get_default_mdsize(sbi, &ealen);
928 return sprintf(buf, "%u\n", ealen);
932 * Set default_easize.
934 * Range checking on the passed value is handled by
935 * ll_set_default_mdsize().
937 * \see client_obd::cl_default_mds_easize
939 * \param[in] file proc file
940 * \param[in] buffer string passed from user space
941 * \param[in] count \a buffer length
942 * \param[in] off unused for single entry
944 * \retval positive \a count on success
945 * \retval negative negated errno on failure
947 static ssize_t default_easize_store(struct kobject *kobj,
948 struct attribute *attr,
952 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
960 rc = kstrtouint(buffer, 10, &val);
964 rc = ll_set_default_mdsize(sbi, val);
970 LUSTRE_RW_ATTR(default_easize);
972 static int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
974 const char *str[] = LL_SBI_FLAGS;
975 struct super_block *sb = m->private;
976 int flags = ll_s2sbi(sb)->ll_flags;
980 if (ARRAY_SIZE(str) <= i) {
981 CERROR("%s: Revise array LL_SBI_FLAGS to match sbi "
982 "flags please.\n", ll_get_fsname(sb, NULL, 0));
987 seq_printf(m, "%s ", str[i]);
991 seq_printf(m, "\b\n");
995 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
997 static ssize_t xattr_cache_show(struct kobject *kobj,
998 struct attribute *attr,
1001 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1004 return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
1007 static ssize_t xattr_cache_store(struct kobject *kobj,
1008 struct attribute *attr,
1012 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1017 rc = kstrtobool(buffer, &val);
1021 if (val && !(sbi->ll_flags & LL_SBI_XATTR_CACHE))
1024 sbi->ll_xattr_cache_enabled = val;
1025 sbi->ll_xattr_cache_set = 1;
1029 LUSTRE_RW_ATTR(xattr_cache);
1031 static ssize_t tiny_write_show(struct kobject *kobj,
1032 struct attribute *attr,
1035 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1038 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_TINY_WRITE));
1041 static ssize_t tiny_write_store(struct kobject *kobj,
1042 struct attribute *attr,
1046 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1051 rc = kstrtobool(buffer, &val);
1055 spin_lock(&sbi->ll_lock);
1057 sbi->ll_flags |= LL_SBI_TINY_WRITE;
1059 sbi->ll_flags &= ~LL_SBI_TINY_WRITE;
1060 spin_unlock(&sbi->ll_lock);
1064 LUSTRE_RW_ATTR(tiny_write);
1066 static ssize_t fast_read_show(struct kobject *kobj,
1067 struct attribute *attr,
1070 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1073 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_FAST_READ));
1076 static ssize_t fast_read_store(struct kobject *kobj,
1077 struct attribute *attr,
1081 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1086 rc = kstrtobool(buffer, &val);
1090 spin_lock(&sbi->ll_lock);
1092 sbi->ll_flags |= LL_SBI_FAST_READ;
1094 sbi->ll_flags &= ~LL_SBI_FAST_READ;
1095 spin_unlock(&sbi->ll_lock);
1099 LUSTRE_RW_ATTR(fast_read);
1101 static ssize_t file_heat_show(struct kobject *kobj,
1102 struct attribute *attr,
1105 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1108 return snprintf(buf, PAGE_SIZE, "%u\n",
1109 !!(sbi->ll_flags & LL_SBI_FILE_HEAT));
1112 static ssize_t file_heat_store(struct kobject *kobj,
1113 struct attribute *attr,
1117 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1122 rc = kstrtobool(buffer, &val);
1126 spin_lock(&sbi->ll_lock);
1128 sbi->ll_flags |= LL_SBI_FILE_HEAT;
1130 sbi->ll_flags &= ~LL_SBI_FILE_HEAT;
1131 spin_unlock(&sbi->ll_lock);
1135 LUSTRE_RW_ATTR(file_heat);
1137 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1138 struct attribute *attr,
1141 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1144 return snprintf(buf, PAGE_SIZE, "%u\n",
1145 (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1148 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1149 struct attribute *attr,
1153 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1158 rc = kstrtoul(buffer, 10, &val);
1162 if (val < 0 || val > 100)
1165 sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1169 LUSTRE_RW_ATTR(heat_decay_percentage);
1171 static ssize_t heat_period_second_show(struct kobject *kobj,
1172 struct attribute *attr,
1175 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1178 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1181 static ssize_t heat_period_second_store(struct kobject *kobj,
1182 struct attribute *attr,
1186 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1191 rc = kstrtoul(buffer, 10, &val);
1198 sbi->ll_heat_period_second = val;
1202 LUSTRE_RW_ATTR(heat_period_second);
1204 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1206 struct super_block *sb = m->private;
1207 struct ll_sb_info *sbi = ll_s2sbi(sb);
1208 struct cl_client_cache *cache = sbi->ll_cache;
1212 pages = atomic_long_read(&cache->ccc_unstable_nr);
1213 mb = (pages * PAGE_SIZE) >> 20;
1215 seq_printf(m, "unstable_check: %8d\n"
1216 "unstable_pages: %12ld\n"
1217 "unstable_mb: %8d\n",
1218 cache->ccc_unstable_check, pages, mb);
1222 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1223 const char __user *buffer,
1224 size_t count, loff_t *unused)
1226 struct seq_file *seq = file->private_data;
1227 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1234 if (count >= sizeof(kernbuf))
1237 if (copy_from_user(kernbuf, buffer, count))
1241 buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1243 rc = kstrtobool_from_user(buffer, count, &val);
1247 /* borrow lru lock to set the value */
1248 spin_lock(&sbi->ll_cache->ccc_lru_lock);
1249 sbi->ll_cache->ccc_unstable_check = val;
1250 spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1255 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1257 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1259 struct super_block *sb = m->private;
1260 struct ll_sb_info *sbi = ll_s2sbi(sb);
1261 struct root_squash_info *squash = &sbi->ll_squash;
1263 seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1267 static ssize_t ll_root_squash_seq_write(struct file *file,
1268 const char __user *buffer,
1269 size_t count, loff_t *off)
1271 struct seq_file *m = file->private_data;
1272 struct super_block *sb = m->private;
1273 struct ll_sb_info *sbi = ll_s2sbi(sb);
1274 struct root_squash_info *squash = &sbi->ll_squash;
1276 return lprocfs_wr_root_squash(buffer, count, squash,
1277 ll_get_fsname(sb, NULL, 0));
1280 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1282 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1284 struct super_block *sb = m->private;
1285 struct ll_sb_info *sbi = ll_s2sbi(sb);
1286 struct root_squash_info *squash = &sbi->ll_squash;
1289 down_read(&squash->rsi_sem);
1290 if (!list_empty(&squash->rsi_nosquash_nids)) {
1291 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1292 &squash->rsi_nosquash_nids);
1296 seq_puts(m, "NONE\n");
1298 up_read(&squash->rsi_sem);
1303 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1304 const char __user *buffer,
1305 size_t count, loff_t *off)
1307 struct seq_file *m = file->private_data;
1308 struct super_block *sb = m->private;
1309 struct ll_sb_info *sbi = ll_s2sbi(sb);
1310 struct root_squash_info *squash = &sbi->ll_squash;
1313 rc = lprocfs_wr_nosquash_nids(buffer, count, squash,
1314 ll_get_fsname(sb, NULL, 0));
1318 ll_compute_rootsquash_state(sbi);
1323 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1325 struct lprocfs_vars lprocfs_llite_obd_vars[] = {
1327 .fops = &ll_site_stats_fops },
1328 { .name = "max_read_ahead_mb",
1329 .fops = &ll_max_readahead_mb_fops },
1330 { .name = "max_read_ahead_per_file_mb",
1331 .fops = &ll_max_readahead_per_file_mb_fops },
1332 { .name = "max_read_ahead_whole_mb",
1333 .fops = &ll_max_read_ahead_whole_mb_fops },
1334 { .name = "max_cached_mb",
1335 .fops = &ll_max_cached_mb_fops },
1336 { .name = "statahead_stats",
1337 .fops = &ll_statahead_stats_fops },
1338 { .name = "unstable_stats",
1339 .fops = &ll_unstable_stats_fops },
1340 { .name = "sbi_flags",
1341 .fops = &ll_sbi_flags_fops },
1342 { .name = "root_squash",
1343 .fops = &ll_root_squash_fops },
1344 { .name = "nosquash_nids",
1345 .fops = &ll_nosquash_nids_fops },
1349 #define MAX_STRING_SIZE 128
1351 static struct attribute *llite_attrs[] = {
1352 &lustre_attr_blocksize.attr,
1353 &lustre_attr_stat_blocksize.attr,
1354 &lustre_attr_kbytestotal.attr,
1355 &lustre_attr_kbytesfree.attr,
1356 &lustre_attr_kbytesavail.attr,
1357 &lustre_attr_filestotal.attr,
1358 &lustre_attr_filesfree.attr,
1359 &lustre_attr_client_type.attr,
1360 &lustre_attr_fstype.attr,
1361 &lustre_attr_uuid.attr,
1362 &lustre_attr_checksums.attr,
1363 &lustre_attr_checksum_pages.attr,
1364 &lustre_attr_stats_track_pid.attr,
1365 &lustre_attr_stats_track_ppid.attr,
1366 &lustre_attr_stats_track_gid.attr,
1367 &lustre_attr_statahead_running_max.attr,
1368 &lustre_attr_statahead_max.attr,
1369 &lustre_attr_statahead_agl.attr,
1370 &lustre_attr_lazystatfs.attr,
1371 &lustre_attr_max_easize.attr,
1372 &lustre_attr_default_easize.attr,
1373 &lustre_attr_xattr_cache.attr,
1374 &lustre_attr_fast_read.attr,
1375 &lustre_attr_tiny_write.attr,
1376 &lustre_attr_file_heat.attr,
1377 &lustre_attr_heat_decay_percentage.attr,
1378 &lustre_attr_heat_period_second.attr,
1382 static void sbi_kobj_release(struct kobject *kobj)
1384 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1386 complete(&sbi->ll_kobj_unregister);
1389 static struct kobj_type sbi_ktype = {
1390 .default_attrs = llite_attrs,
1391 .sysfs_ops = &lustre_sysfs_ops,
1392 .release = sbi_kobj_release,
1395 static const struct llite_file_opcode {
1399 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1400 /* file operation */
1401 { LPROC_LL_DIRTY_HITS, LPROCFS_TYPE_REGS, "dirty_pages_hits" },
1402 { LPROC_LL_DIRTY_MISSES, LPROCFS_TYPE_REGS, "dirty_pages_misses" },
1403 { LPROC_LL_READ_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
1405 { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
1407 { LPROC_LL_BRW_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
1409 { LPROC_LL_BRW_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
1411 { LPROC_LL_IOCTL, LPROCFS_TYPE_REGS, "ioctl" },
1412 { LPROC_LL_OPEN, LPROCFS_TYPE_REGS, "open" },
1413 { LPROC_LL_RELEASE, LPROCFS_TYPE_REGS, "close" },
1414 { LPROC_LL_MAP, LPROCFS_TYPE_REGS, "mmap" },
1415 { LPROC_LL_FAULT, LPROCFS_TYPE_REGS, "page_fault" },
1416 { LPROC_LL_MKWRITE, LPROCFS_TYPE_REGS, "page_mkwrite" },
1417 { LPROC_LL_LLSEEK, LPROCFS_TYPE_REGS, "seek" },
1418 { LPROC_LL_FSYNC, LPROCFS_TYPE_REGS, "fsync" },
1419 { LPROC_LL_READDIR, LPROCFS_TYPE_REGS, "readdir" },
1420 /* inode operation */
1421 { LPROC_LL_SETATTR, LPROCFS_TYPE_REGS, "setattr" },
1422 { LPROC_LL_TRUNC, LPROCFS_TYPE_REGS, "truncate" },
1423 { LPROC_LL_FLOCK, LPROCFS_TYPE_REGS, "flock" },
1424 { LPROC_LL_GETATTR, LPROCFS_TYPE_REGS, "getattr" },
1425 /* dir inode operation */
1426 { LPROC_LL_CREATE, LPROCFS_TYPE_REGS, "create" },
1427 { LPROC_LL_LINK, LPROCFS_TYPE_REGS, "link" },
1428 { LPROC_LL_UNLINK, LPROCFS_TYPE_REGS, "unlink" },
1429 { LPROC_LL_SYMLINK, LPROCFS_TYPE_REGS, "symlink" },
1430 { LPROC_LL_MKDIR, LPROCFS_TYPE_REGS, "mkdir" },
1431 { LPROC_LL_RMDIR, LPROCFS_TYPE_REGS, "rmdir" },
1432 { LPROC_LL_MKNOD, LPROCFS_TYPE_REGS, "mknod" },
1433 { LPROC_LL_RENAME, LPROCFS_TYPE_REGS, "rename" },
1434 /* special inode operation */
1435 { LPROC_LL_STAFS, LPROCFS_TYPE_REGS, "statfs" },
1436 { LPROC_LL_ALLOC_INODE, LPROCFS_TYPE_REGS, "alloc_inode" },
1437 { LPROC_LL_SETXATTR, LPROCFS_TYPE_REGS, "setxattr" },
1438 { LPROC_LL_GETXATTR, LPROCFS_TYPE_REGS, "getxattr" },
1439 { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REGS, "getxattr_hits" },
1440 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_REGS, "listxattr" },
1441 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_REGS, "removexattr" },
1442 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_REGS, "inode_permission" },
1445 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
1449 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1450 lprocfs_counter_add(sbi->ll_stats, op, count);
1451 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1452 sbi->ll_stats_track_id == current->pid)
1453 lprocfs_counter_add(sbi->ll_stats, op, count);
1454 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1455 sbi->ll_stats_track_id == current->parent->pid)
1456 lprocfs_counter_add(sbi->ll_stats, op, count);
1457 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1458 sbi->ll_stats_track_id ==
1459 from_kgid(&init_user_ns, current_gid()))
1460 lprocfs_counter_add(sbi->ll_stats, op, count);
1462 EXPORT_SYMBOL(ll_stats_ops_tally);
1464 static const char *ra_stat_string[] = {
1465 [RA_STAT_HIT] = "hits",
1466 [RA_STAT_MISS] = "misses",
1467 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1468 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1469 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1470 [RA_STAT_FAILED_MATCH] = "failed lock match",
1471 [RA_STAT_DISCARDED] = "read but discarded",
1472 [RA_STAT_ZERO_LEN] = "zero length file",
1473 [RA_STAT_ZERO_WINDOW] = "zero size window",
1474 [RA_STAT_EOF] = "read-ahead to EOF",
1475 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1476 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1477 [RA_STAT_FAILED_REACH_END] = "failed to reach end"
1480 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1482 struct lustre_sb_info *lsi = s2lsi(sb);
1483 struct ll_sb_info *sbi = ll_s2sbi(sb);
1489 if (IS_ERR_OR_NULL(llite_root))
1492 sbi->ll_debugfs_entry = ldebugfs_register(name, llite_root,
1493 lprocfs_llite_obd_vars, sb);
1494 if (IS_ERR_OR_NULL(sbi->ll_debugfs_entry)) {
1495 err = sbi->ll_debugfs_entry ? PTR_ERR(sbi->ll_debugfs_entry) :
1497 sbi->ll_debugfs_entry = NULL;
1501 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "dump_page_cache",0444,
1502 &vvp_dump_pgcache_file_ops, sbi);
1504 CWARN("Error adding the dump_page_cache file\n");
1506 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "extents_stats", 0644,
1507 &ll_rw_extents_stats_fops, sbi);
1509 CWARN("Error adding the extent_stats file\n");
1511 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry,
1512 "extents_stats_per_process", 0644,
1513 &ll_rw_extents_stats_pp_fops, sbi);
1515 CWARN("Error adding the extents_stats_per_process file\n");
1517 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "offset_stats", 0644,
1518 &ll_rw_offset_stats_fops, sbi);
1520 CWARN("Error adding the offset_stats file\n");
1522 /* File operations stats */
1523 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1524 LPROCFS_STATS_FLAG_NONE);
1525 if (sbi->ll_stats == NULL)
1526 GOTO(out_debugfs, err = -ENOMEM);
1528 /* do counter init */
1529 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1530 u32 type = llite_opcode_table[id].type;
1533 if (type & LPROCFS_TYPE_REGS)
1535 else if (type & LPROCFS_TYPE_BYTES)
1537 else if (type & LPROCFS_TYPE_PAGES)
1539 lprocfs_counter_init(sbi->ll_stats,
1540 llite_opcode_table[id].opcode,
1541 (type & LPROCFS_CNTR_AVGMINMAX),
1542 llite_opcode_table[id].opname, ptr);
1545 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "stats",
1548 GOTO(out_stats, err);
1550 sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1551 LPROCFS_STATS_FLAG_NONE);
1552 if (sbi->ll_ra_stats == NULL)
1553 GOTO(out_stats, err = -ENOMEM);
1555 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1556 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1557 ra_stat_string[id], "pages");
1559 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "read_ahead_stats",
1562 GOTO(out_ra_stats, err);
1565 /* Yes we also register sysfs mount kset here as well */
1566 sbi->ll_kset.kobj.parent = llite_kobj;
1567 sbi->ll_kset.kobj.ktype = &sbi_ktype;
1568 init_completion(&sbi->ll_kobj_unregister);
1569 err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1571 GOTO(out_ra_stats, err);
1573 err = kset_register(&sbi->ll_kset);
1575 GOTO(out_ra_stats, err);
1577 lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1581 lprocfs_free_stats(&sbi->ll_ra_stats);
1583 lprocfs_free_stats(&sbi->ll_stats);
1585 ldebugfs_remove(&sbi->ll_debugfs_entry);
1590 void ll_debugfs_unregister_super(struct super_block *sb)
1592 struct lustre_sb_info *lsi = s2lsi(sb);
1593 struct ll_sb_info *sbi = ll_s2sbi(sb);
1595 if (!IS_ERR_OR_NULL(sbi->ll_debugfs_entry))
1596 ldebugfs_remove(&sbi->ll_debugfs_entry);
1599 sysfs_remove_link(&sbi->ll_kset.kobj,
1600 sbi->ll_dt_obd->obd_type->typ_name);
1603 sysfs_remove_link(&sbi->ll_kset.kobj,
1604 sbi->ll_md_obd->obd_type->typ_name);
1606 kobject_put(lsi->lsi_kobj);
1608 kset_unregister(&sbi->ll_kset);
1609 wait_for_completion(&sbi->ll_kobj_unregister);
1611 lprocfs_free_stats(&sbi->ll_ra_stats);
1612 lprocfs_free_stats(&sbi->ll_stats);
1614 #undef MAX_STRING_SIZE
1616 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1617 struct seq_file *seq, int which)
1619 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1620 unsigned long start, end, r, w;
1621 char *unitp = "KMGTPEZY";
1623 struct per_process_info *pp_info = &io_extents->pp_extents[which];
1629 for(i = 0; i < LL_HIST_MAX; i++) {
1630 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1631 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1634 for(i = 0; i < LL_HIST_MAX; i++) {
1635 r = pp_info->pp_r_hist.oh_buckets[i];
1636 w = pp_info->pp_w_hist.oh_buckets[i];
1639 end = BIT(i + LL_HIST_START - units);
1640 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u | "
1641 "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1642 (i == LL_HIST_MAX - 1) ? '+' : ' ',
1643 r, pct(r, read_tot), pct(read_cum, read_tot),
1644 w, pct(w, write_tot), pct(write_cum, write_tot));
1646 if (start == BIT(10)) {
1651 if (read_cum == read_tot && write_cum == write_tot)
1656 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1658 struct timespec64 now;
1659 struct ll_sb_info *sbi = seq->private;
1660 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1663 ktime_get_real_ts64(&now);
1665 if (!sbi->ll_rw_stats_on) {
1666 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1669 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1670 (s64)now.tv_sec, now.tv_nsec);
1671 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1672 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1673 "extents", "calls", "%", "cum%",
1674 "calls", "%", "cum%");
1675 spin_lock(&sbi->ll_pp_extent_lock);
1676 for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1677 if (io_extents->pp_extents[k].pid != 0) {
1678 seq_printf(seq, "\nPID: %d\n",
1679 io_extents->pp_extents[k].pid);
1680 ll_display_extents_info(io_extents, seq, k);
1683 spin_unlock(&sbi->ll_pp_extent_lock);
1687 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1688 const char __user *buf,
1692 struct seq_file *seq = file->private_data;
1693 struct ll_sb_info *sbi = seq->private;
1694 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1701 value = ll_stats_pid_write(buf, len);
1704 sbi->ll_rw_stats_on = 0;
1706 sbi->ll_rw_stats_on = 1;
1708 spin_lock(&sbi->ll_pp_extent_lock);
1709 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1710 io_extents->pp_extents[i].pid = 0;
1711 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1712 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1714 spin_unlock(&sbi->ll_pp_extent_lock);
1718 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1720 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1722 struct timespec64 now;
1723 struct ll_sb_info *sbi = seq->private;
1724 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1726 ktime_get_real_ts64(&now);
1728 if (!sbi->ll_rw_stats_on) {
1729 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1732 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1733 (s64)now.tv_sec, now.tv_nsec);
1735 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1736 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1737 "extents", "calls", "%", "cum%",
1738 "calls", "%", "cum%");
1739 spin_lock(&sbi->ll_lock);
1740 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1741 spin_unlock(&sbi->ll_lock);
1746 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1747 const char __user *buf,
1748 size_t len, loff_t *off)
1750 struct seq_file *seq = file->private_data;
1751 struct ll_sb_info *sbi = seq->private;
1752 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1759 value = ll_stats_pid_write(buf, len);
1762 sbi->ll_rw_stats_on = 0;
1764 sbi->ll_rw_stats_on = 1;
1766 spin_lock(&sbi->ll_pp_extent_lock);
1767 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1768 io_extents->pp_extents[i].pid = 0;
1769 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1770 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1772 spin_unlock(&sbi->ll_pp_extent_lock);
1777 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1779 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1780 struct ll_file_data *file, loff_t pos,
1781 size_t count, int rw)
1784 struct ll_rw_process_info *process;
1785 struct ll_rw_process_info *offset;
1786 int *off_count = &sbi->ll_rw_offset_entry_count;
1787 int *process_count = &sbi->ll_offset_process_count;
1788 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1790 if(!sbi->ll_rw_stats_on)
1792 process = sbi->ll_rw_process_info;
1793 offset = sbi->ll_rw_offset_info;
1795 spin_lock(&sbi->ll_pp_extent_lock);
1796 /* Extent statistics */
1797 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1798 if(io_extents->pp_extents[i].pid == pid) {
1806 sbi->ll_extent_process_count =
1807 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1808 cur = sbi->ll_extent_process_count;
1809 io_extents->pp_extents[cur].pid = pid;
1810 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1811 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1814 for(i = 0; (count >= BIT(LL_HIST_START << i)) &&
1815 (i < (LL_HIST_MAX - 1)); i++);
1817 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1818 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1820 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1821 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1823 spin_unlock(&sbi->ll_pp_extent_lock);
1825 spin_lock(&sbi->ll_process_lock);
1826 /* Offset statistics */
1827 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1828 if (process[i].rw_pid == pid) {
1829 if (process[i].rw_last_file != file) {
1830 process[i].rw_range_start = pos;
1831 process[i].rw_last_file_pos = pos + count;
1832 process[i].rw_smallest_extent = count;
1833 process[i].rw_largest_extent = count;
1834 process[i].rw_offset = 0;
1835 process[i].rw_last_file = file;
1836 spin_unlock(&sbi->ll_process_lock);
1839 if (process[i].rw_last_file_pos != pos) {
1841 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1842 offset[*off_count].rw_op = process[i].rw_op;
1843 offset[*off_count].rw_pid = pid;
1844 offset[*off_count].rw_range_start =
1845 process[i].rw_range_start;
1846 offset[*off_count].rw_range_end =
1847 process[i].rw_last_file_pos;
1848 offset[*off_count].rw_smallest_extent =
1849 process[i].rw_smallest_extent;
1850 offset[*off_count].rw_largest_extent =
1851 process[i].rw_largest_extent;
1852 offset[*off_count].rw_offset =
1853 process[i].rw_offset;
1854 process[i].rw_op = rw;
1855 process[i].rw_range_start = pos;
1856 process[i].rw_smallest_extent = count;
1857 process[i].rw_largest_extent = count;
1858 process[i].rw_offset = pos -
1859 process[i].rw_last_file_pos;
1861 if(process[i].rw_smallest_extent > count)
1862 process[i].rw_smallest_extent = count;
1863 if(process[i].rw_largest_extent < count)
1864 process[i].rw_largest_extent = count;
1865 process[i].rw_last_file_pos = pos + count;
1866 spin_unlock(&sbi->ll_process_lock);
1870 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1871 process[*process_count].rw_pid = pid;
1872 process[*process_count].rw_op = rw;
1873 process[*process_count].rw_range_start = pos;
1874 process[*process_count].rw_last_file_pos = pos + count;
1875 process[*process_count].rw_smallest_extent = count;
1876 process[*process_count].rw_largest_extent = count;
1877 process[*process_count].rw_offset = 0;
1878 process[*process_count].rw_last_file = file;
1879 spin_unlock(&sbi->ll_process_lock);
1882 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1884 struct timespec64 now;
1885 struct ll_sb_info *sbi = seq->private;
1886 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1887 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1890 ktime_get_real_ts64(&now);
1892 if (!sbi->ll_rw_stats_on) {
1893 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1896 spin_lock(&sbi->ll_process_lock);
1898 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1899 (s64)now.tv_sec, now.tv_nsec);
1900 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1901 "R/W", "PID", "RANGE START", "RANGE END",
1902 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1904 /* We stored the discontiguous offsets here; print them first */
1905 for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1906 if (offset[i].rw_pid != 0)
1908 "%3c %10d %14llu %14llu %17lu %17lu %14llu\n",
1909 offset[i].rw_op == READ ? 'R' : 'W',
1911 offset[i].rw_range_start,
1912 offset[i].rw_range_end,
1913 (unsigned long)offset[i].rw_smallest_extent,
1914 (unsigned long)offset[i].rw_largest_extent,
1915 offset[i].rw_offset);
1918 /* Then print the current offsets for each process */
1919 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1920 if (process[i].rw_pid != 0)
1922 "%3c %10d %14llu %14llu %17lu %17lu %14llu\n",
1923 process[i].rw_op == READ ? 'R' : 'W',
1925 process[i].rw_range_start,
1926 process[i].rw_last_file_pos,
1927 (unsigned long)process[i].rw_smallest_extent,
1928 (unsigned long)process[i].rw_largest_extent,
1929 process[i].rw_offset);
1931 spin_unlock(&sbi->ll_process_lock);
1936 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
1937 const char __user *buf,
1938 size_t len, loff_t *off)
1940 struct seq_file *seq = file->private_data;
1941 struct ll_sb_info *sbi = seq->private;
1942 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1943 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1949 value = ll_stats_pid_write(buf, len);
1952 sbi->ll_rw_stats_on = 0;
1954 sbi->ll_rw_stats_on = 1;
1956 spin_lock(&sbi->ll_process_lock);
1957 sbi->ll_offset_process_count = 0;
1958 sbi->ll_rw_offset_entry_count = 0;
1959 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1960 LL_PROCESS_HIST_MAX);
1961 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1962 LL_OFFSET_HIST_MAX);
1963 spin_unlock(&sbi->ll_process_lock);
1968 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);