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 int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1103 struct super_block *sb = m->private;
1104 struct ll_sb_info *sbi = ll_s2sbi(sb);
1105 struct cl_client_cache *cache = sbi->ll_cache;
1109 pages = atomic_long_read(&cache->ccc_unstable_nr);
1110 mb = (pages * PAGE_SIZE) >> 20;
1112 seq_printf(m, "unstable_check: %8d\n"
1113 "unstable_pages: %12ld\n"
1114 "unstable_mb: %8d\n",
1115 cache->ccc_unstable_check, pages, mb);
1119 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1120 const char __user *buffer,
1121 size_t count, loff_t *unused)
1123 struct seq_file *seq = file->private_data;
1124 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1131 if (count >= sizeof(kernbuf))
1134 if (copy_from_user(kernbuf, buffer, count))
1138 buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1140 rc = kstrtobool_from_user(buffer, count, &val);
1144 /* borrow lru lock to set the value */
1145 spin_lock(&sbi->ll_cache->ccc_lru_lock);
1146 sbi->ll_cache->ccc_unstable_check = val;
1147 spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1152 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1154 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1156 struct super_block *sb = m->private;
1157 struct ll_sb_info *sbi = ll_s2sbi(sb);
1158 struct root_squash_info *squash = &sbi->ll_squash;
1160 seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1164 static ssize_t ll_root_squash_seq_write(struct file *file,
1165 const char __user *buffer,
1166 size_t count, loff_t *off)
1168 struct seq_file *m = file->private_data;
1169 struct super_block *sb = m->private;
1170 struct ll_sb_info *sbi = ll_s2sbi(sb);
1171 struct root_squash_info *squash = &sbi->ll_squash;
1173 return lprocfs_wr_root_squash(buffer, count, squash,
1174 ll_get_fsname(sb, NULL, 0));
1177 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1179 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1181 struct super_block *sb = m->private;
1182 struct ll_sb_info *sbi = ll_s2sbi(sb);
1183 struct root_squash_info *squash = &sbi->ll_squash;
1186 down_read(&squash->rsi_sem);
1187 if (!list_empty(&squash->rsi_nosquash_nids)) {
1188 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1189 &squash->rsi_nosquash_nids);
1193 seq_puts(m, "NONE\n");
1195 up_read(&squash->rsi_sem);
1200 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1201 const char __user *buffer,
1202 size_t count, loff_t *off)
1204 struct seq_file *m = file->private_data;
1205 struct super_block *sb = m->private;
1206 struct ll_sb_info *sbi = ll_s2sbi(sb);
1207 struct root_squash_info *squash = &sbi->ll_squash;
1210 rc = lprocfs_wr_nosquash_nids(buffer, count, squash,
1211 ll_get_fsname(sb, NULL, 0));
1215 ll_compute_rootsquash_state(sbi);
1220 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1222 struct lprocfs_vars lprocfs_llite_obd_vars[] = {
1224 .fops = &ll_site_stats_fops },
1225 { .name = "max_read_ahead_mb",
1226 .fops = &ll_max_readahead_mb_fops },
1227 { .name = "max_read_ahead_per_file_mb",
1228 .fops = &ll_max_readahead_per_file_mb_fops },
1229 { .name = "max_read_ahead_whole_mb",
1230 .fops = &ll_max_read_ahead_whole_mb_fops },
1231 { .name = "max_cached_mb",
1232 .fops = &ll_max_cached_mb_fops },
1233 { .name = "statahead_stats",
1234 .fops = &ll_statahead_stats_fops },
1235 { .name = "unstable_stats",
1236 .fops = &ll_unstable_stats_fops },
1237 { .name = "sbi_flags",
1238 .fops = &ll_sbi_flags_fops },
1239 { .name = "root_squash",
1240 .fops = &ll_root_squash_fops },
1241 { .name = "nosquash_nids",
1242 .fops = &ll_nosquash_nids_fops },
1246 #define MAX_STRING_SIZE 128
1248 static struct attribute *llite_attrs[] = {
1249 &lustre_attr_blocksize.attr,
1250 &lustre_attr_stat_blocksize.attr,
1251 &lustre_attr_kbytestotal.attr,
1252 &lustre_attr_kbytesfree.attr,
1253 &lustre_attr_kbytesavail.attr,
1254 &lustre_attr_filestotal.attr,
1255 &lustre_attr_filesfree.attr,
1256 &lustre_attr_client_type.attr,
1257 &lustre_attr_fstype.attr,
1258 &lustre_attr_uuid.attr,
1259 &lustre_attr_checksums.attr,
1260 &lustre_attr_checksum_pages.attr,
1261 &lustre_attr_stats_track_pid.attr,
1262 &lustre_attr_stats_track_ppid.attr,
1263 &lustre_attr_stats_track_gid.attr,
1264 &lustre_attr_statahead_running_max.attr,
1265 &lustre_attr_statahead_max.attr,
1266 &lustre_attr_statahead_agl.attr,
1267 &lustre_attr_lazystatfs.attr,
1268 &lustre_attr_max_easize.attr,
1269 &lustre_attr_default_easize.attr,
1270 &lustre_attr_xattr_cache.attr,
1271 &lustre_attr_fast_read.attr,
1272 &lustre_attr_tiny_write.attr,
1276 static void sbi_kobj_release(struct kobject *kobj)
1278 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1280 complete(&sbi->ll_kobj_unregister);
1283 static struct kobj_type sbi_ktype = {
1284 .default_attrs = llite_attrs,
1285 .sysfs_ops = &lustre_sysfs_ops,
1286 .release = sbi_kobj_release,
1289 static const struct llite_file_opcode {
1293 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1294 /* file operation */
1295 { LPROC_LL_DIRTY_HITS, LPROCFS_TYPE_REGS, "dirty_pages_hits" },
1296 { LPROC_LL_DIRTY_MISSES, LPROCFS_TYPE_REGS, "dirty_pages_misses" },
1297 { LPROC_LL_READ_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
1299 { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
1301 { LPROC_LL_BRW_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
1303 { LPROC_LL_BRW_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
1305 { LPROC_LL_IOCTL, LPROCFS_TYPE_REGS, "ioctl" },
1306 { LPROC_LL_OPEN, LPROCFS_TYPE_REGS, "open" },
1307 { LPROC_LL_RELEASE, LPROCFS_TYPE_REGS, "close" },
1308 { LPROC_LL_MAP, LPROCFS_TYPE_REGS, "mmap" },
1309 { LPROC_LL_FAULT, LPROCFS_TYPE_REGS, "page_fault" },
1310 { LPROC_LL_MKWRITE, LPROCFS_TYPE_REGS, "page_mkwrite" },
1311 { LPROC_LL_LLSEEK, LPROCFS_TYPE_REGS, "seek" },
1312 { LPROC_LL_FSYNC, LPROCFS_TYPE_REGS, "fsync" },
1313 { LPROC_LL_READDIR, LPROCFS_TYPE_REGS, "readdir" },
1314 /* inode operation */
1315 { LPROC_LL_SETATTR, LPROCFS_TYPE_REGS, "setattr" },
1316 { LPROC_LL_TRUNC, LPROCFS_TYPE_REGS, "truncate" },
1317 { LPROC_LL_FLOCK, LPROCFS_TYPE_REGS, "flock" },
1318 { LPROC_LL_GETATTR, LPROCFS_TYPE_REGS, "getattr" },
1319 /* dir inode operation */
1320 { LPROC_LL_CREATE, LPROCFS_TYPE_REGS, "create" },
1321 { LPROC_LL_LINK, LPROCFS_TYPE_REGS, "link" },
1322 { LPROC_LL_UNLINK, LPROCFS_TYPE_REGS, "unlink" },
1323 { LPROC_LL_SYMLINK, LPROCFS_TYPE_REGS, "symlink" },
1324 { LPROC_LL_MKDIR, LPROCFS_TYPE_REGS, "mkdir" },
1325 { LPROC_LL_RMDIR, LPROCFS_TYPE_REGS, "rmdir" },
1326 { LPROC_LL_MKNOD, LPROCFS_TYPE_REGS, "mknod" },
1327 { LPROC_LL_RENAME, LPROCFS_TYPE_REGS, "rename" },
1328 /* special inode operation */
1329 { LPROC_LL_STAFS, LPROCFS_TYPE_REGS, "statfs" },
1330 { LPROC_LL_ALLOC_INODE, LPROCFS_TYPE_REGS, "alloc_inode" },
1331 { LPROC_LL_SETXATTR, LPROCFS_TYPE_REGS, "setxattr" },
1332 { LPROC_LL_GETXATTR, LPROCFS_TYPE_REGS, "getxattr" },
1333 { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REGS, "getxattr_hits" },
1334 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_REGS, "listxattr" },
1335 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_REGS, "removexattr" },
1336 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_REGS, "inode_permission" },
1339 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
1343 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1344 lprocfs_counter_add(sbi->ll_stats, op, count);
1345 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1346 sbi->ll_stats_track_id == current->pid)
1347 lprocfs_counter_add(sbi->ll_stats, op, count);
1348 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1349 sbi->ll_stats_track_id == current->parent->pid)
1350 lprocfs_counter_add(sbi->ll_stats, op, count);
1351 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1352 sbi->ll_stats_track_id ==
1353 from_kgid(&init_user_ns, current_gid()))
1354 lprocfs_counter_add(sbi->ll_stats, op, count);
1356 EXPORT_SYMBOL(ll_stats_ops_tally);
1358 static const char *ra_stat_string[] = {
1359 [RA_STAT_HIT] = "hits",
1360 [RA_STAT_MISS] = "misses",
1361 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1362 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1363 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1364 [RA_STAT_FAILED_MATCH] = "failed lock match",
1365 [RA_STAT_DISCARDED] = "read but discarded",
1366 [RA_STAT_ZERO_LEN] = "zero length file",
1367 [RA_STAT_ZERO_WINDOW] = "zero size window",
1368 [RA_STAT_EOF] = "read-ahead to EOF",
1369 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1370 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1371 [RA_STAT_FAILED_REACH_END] = "failed to reach end"
1374 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1376 struct lustre_sb_info *lsi = s2lsi(sb);
1377 struct ll_sb_info *sbi = ll_s2sbi(sb);
1383 if (IS_ERR_OR_NULL(llite_root))
1386 sbi->ll_debugfs_entry = ldebugfs_register(name, llite_root,
1387 lprocfs_llite_obd_vars, sb);
1388 if (IS_ERR_OR_NULL(sbi->ll_debugfs_entry)) {
1389 err = sbi->ll_debugfs_entry ? PTR_ERR(sbi->ll_debugfs_entry) :
1391 sbi->ll_debugfs_entry = NULL;
1395 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "dump_page_cache",0444,
1396 &vvp_dump_pgcache_file_ops, sbi);
1398 CWARN("Error adding the dump_page_cache file\n");
1400 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "extents_stats", 0644,
1401 &ll_rw_extents_stats_fops, sbi);
1403 CWARN("Error adding the extent_stats file\n");
1405 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry,
1406 "extents_stats_per_process", 0644,
1407 &ll_rw_extents_stats_pp_fops, sbi);
1409 CWARN("Error adding the extents_stats_per_process file\n");
1411 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "offset_stats", 0644,
1412 &ll_rw_offset_stats_fops, sbi);
1414 CWARN("Error adding the offset_stats file\n");
1416 /* File operations stats */
1417 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1418 LPROCFS_STATS_FLAG_NONE);
1419 if (sbi->ll_stats == NULL)
1420 GOTO(out_debugfs, err = -ENOMEM);
1422 /* do counter init */
1423 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1424 u32 type = llite_opcode_table[id].type;
1427 if (type & LPROCFS_TYPE_REGS)
1429 else if (type & LPROCFS_TYPE_BYTES)
1431 else if (type & LPROCFS_TYPE_PAGES)
1433 lprocfs_counter_init(sbi->ll_stats,
1434 llite_opcode_table[id].opcode,
1435 (type & LPROCFS_CNTR_AVGMINMAX),
1436 llite_opcode_table[id].opname, ptr);
1439 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "stats",
1442 GOTO(out_stats, err);
1444 sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1445 LPROCFS_STATS_FLAG_NONE);
1446 if (sbi->ll_ra_stats == NULL)
1447 GOTO(out_stats, err = -ENOMEM);
1449 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1450 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1451 ra_stat_string[id], "pages");
1453 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "read_ahead_stats",
1456 GOTO(out_ra_stats, err);
1459 /* Yes we also register sysfs mount kset here as well */
1460 sbi->ll_kset.kobj.parent = llite_kobj;
1461 sbi->ll_kset.kobj.ktype = &sbi_ktype;
1462 init_completion(&sbi->ll_kobj_unregister);
1463 err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1465 GOTO(out_ra_stats, err);
1467 err = kset_register(&sbi->ll_kset);
1469 GOTO(out_ra_stats, err);
1471 lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1475 lprocfs_free_stats(&sbi->ll_ra_stats);
1477 lprocfs_free_stats(&sbi->ll_stats);
1479 ldebugfs_remove(&sbi->ll_debugfs_entry);
1484 void ll_debugfs_unregister_super(struct super_block *sb)
1486 struct lustre_sb_info *lsi = s2lsi(sb);
1487 struct ll_sb_info *sbi = ll_s2sbi(sb);
1489 if (!IS_ERR_OR_NULL(sbi->ll_debugfs_entry))
1490 ldebugfs_remove(&sbi->ll_debugfs_entry);
1493 sysfs_remove_link(&sbi->ll_kset.kobj,
1494 sbi->ll_dt_obd->obd_type->typ_name);
1497 sysfs_remove_link(&sbi->ll_kset.kobj,
1498 sbi->ll_md_obd->obd_type->typ_name);
1500 kobject_put(lsi->lsi_kobj);
1502 kset_unregister(&sbi->ll_kset);
1503 wait_for_completion(&sbi->ll_kobj_unregister);
1505 lprocfs_free_stats(&sbi->ll_ra_stats);
1506 lprocfs_free_stats(&sbi->ll_stats);
1508 #undef MAX_STRING_SIZE
1510 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1511 struct seq_file *seq, int which)
1513 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1514 unsigned long start, end, r, w;
1515 char *unitp = "KMGTPEZY";
1517 struct per_process_info *pp_info = &io_extents->pp_extents[which];
1523 for(i = 0; i < LL_HIST_MAX; i++) {
1524 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1525 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1528 for(i = 0; i < LL_HIST_MAX; i++) {
1529 r = pp_info->pp_r_hist.oh_buckets[i];
1530 w = pp_info->pp_w_hist.oh_buckets[i];
1533 end = BIT(i + LL_HIST_START - units);
1534 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u | "
1535 "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1536 (i == LL_HIST_MAX - 1) ? '+' : ' ',
1537 r, pct(r, read_tot), pct(read_cum, read_tot),
1538 w, pct(w, write_tot), pct(write_cum, write_tot));
1540 if (start == BIT(10)) {
1545 if (read_cum == read_tot && write_cum == write_tot)
1550 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1552 struct timespec64 now;
1553 struct ll_sb_info *sbi = seq->private;
1554 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1557 ktime_get_real_ts64(&now);
1559 if (!sbi->ll_rw_stats_on) {
1560 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1563 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1564 (s64)now.tv_sec, now.tv_nsec);
1565 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1566 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1567 "extents", "calls", "%", "cum%",
1568 "calls", "%", "cum%");
1569 spin_lock(&sbi->ll_pp_extent_lock);
1570 for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1571 if (io_extents->pp_extents[k].pid != 0) {
1572 seq_printf(seq, "\nPID: %d\n",
1573 io_extents->pp_extents[k].pid);
1574 ll_display_extents_info(io_extents, seq, k);
1577 spin_unlock(&sbi->ll_pp_extent_lock);
1581 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1582 const char __user *buf,
1586 struct seq_file *seq = file->private_data;
1587 struct ll_sb_info *sbi = seq->private;
1588 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1595 value = ll_stats_pid_write(buf, len);
1598 sbi->ll_rw_stats_on = 0;
1600 sbi->ll_rw_stats_on = 1;
1602 spin_lock(&sbi->ll_pp_extent_lock);
1603 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1604 io_extents->pp_extents[i].pid = 0;
1605 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1606 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1608 spin_unlock(&sbi->ll_pp_extent_lock);
1612 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1614 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1616 struct timespec64 now;
1617 struct ll_sb_info *sbi = seq->private;
1618 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1620 ktime_get_real_ts64(&now);
1622 if (!sbi->ll_rw_stats_on) {
1623 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1626 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1627 (s64)now.tv_sec, now.tv_nsec);
1629 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1630 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1631 "extents", "calls", "%", "cum%",
1632 "calls", "%", "cum%");
1633 spin_lock(&sbi->ll_lock);
1634 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1635 spin_unlock(&sbi->ll_lock);
1640 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1641 const char __user *buf,
1642 size_t len, loff_t *off)
1644 struct seq_file *seq = file->private_data;
1645 struct ll_sb_info *sbi = seq->private;
1646 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1653 value = ll_stats_pid_write(buf, len);
1656 sbi->ll_rw_stats_on = 0;
1658 sbi->ll_rw_stats_on = 1;
1660 spin_lock(&sbi->ll_pp_extent_lock);
1661 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1662 io_extents->pp_extents[i].pid = 0;
1663 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1664 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1666 spin_unlock(&sbi->ll_pp_extent_lock);
1671 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1673 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1674 struct ll_file_data *file, loff_t pos,
1675 size_t count, int rw)
1678 struct ll_rw_process_info *process;
1679 struct ll_rw_process_info *offset;
1680 int *off_count = &sbi->ll_rw_offset_entry_count;
1681 int *process_count = &sbi->ll_offset_process_count;
1682 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1684 if(!sbi->ll_rw_stats_on)
1686 process = sbi->ll_rw_process_info;
1687 offset = sbi->ll_rw_offset_info;
1689 spin_lock(&sbi->ll_pp_extent_lock);
1690 /* Extent statistics */
1691 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1692 if(io_extents->pp_extents[i].pid == pid) {
1700 sbi->ll_extent_process_count =
1701 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1702 cur = sbi->ll_extent_process_count;
1703 io_extents->pp_extents[cur].pid = pid;
1704 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1705 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1708 for(i = 0; (count >= BIT(LL_HIST_START << i)) &&
1709 (i < (LL_HIST_MAX - 1)); i++);
1711 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1712 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1714 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1715 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1717 spin_unlock(&sbi->ll_pp_extent_lock);
1719 spin_lock(&sbi->ll_process_lock);
1720 /* Offset statistics */
1721 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1722 if (process[i].rw_pid == pid) {
1723 if (process[i].rw_last_file != file) {
1724 process[i].rw_range_start = pos;
1725 process[i].rw_last_file_pos = pos + count;
1726 process[i].rw_smallest_extent = count;
1727 process[i].rw_largest_extent = count;
1728 process[i].rw_offset = 0;
1729 process[i].rw_last_file = file;
1730 spin_unlock(&sbi->ll_process_lock);
1733 if (process[i].rw_last_file_pos != pos) {
1735 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1736 offset[*off_count].rw_op = process[i].rw_op;
1737 offset[*off_count].rw_pid = pid;
1738 offset[*off_count].rw_range_start =
1739 process[i].rw_range_start;
1740 offset[*off_count].rw_range_end =
1741 process[i].rw_last_file_pos;
1742 offset[*off_count].rw_smallest_extent =
1743 process[i].rw_smallest_extent;
1744 offset[*off_count].rw_largest_extent =
1745 process[i].rw_largest_extent;
1746 offset[*off_count].rw_offset =
1747 process[i].rw_offset;
1748 process[i].rw_op = rw;
1749 process[i].rw_range_start = pos;
1750 process[i].rw_smallest_extent = count;
1751 process[i].rw_largest_extent = count;
1752 process[i].rw_offset = pos -
1753 process[i].rw_last_file_pos;
1755 if(process[i].rw_smallest_extent > count)
1756 process[i].rw_smallest_extent = count;
1757 if(process[i].rw_largest_extent < count)
1758 process[i].rw_largest_extent = count;
1759 process[i].rw_last_file_pos = pos + count;
1760 spin_unlock(&sbi->ll_process_lock);
1764 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1765 process[*process_count].rw_pid = pid;
1766 process[*process_count].rw_op = rw;
1767 process[*process_count].rw_range_start = pos;
1768 process[*process_count].rw_last_file_pos = pos + count;
1769 process[*process_count].rw_smallest_extent = count;
1770 process[*process_count].rw_largest_extent = count;
1771 process[*process_count].rw_offset = 0;
1772 process[*process_count].rw_last_file = file;
1773 spin_unlock(&sbi->ll_process_lock);
1776 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1778 struct timespec64 now;
1779 struct ll_sb_info *sbi = seq->private;
1780 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1781 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1784 ktime_get_real_ts64(&now);
1786 if (!sbi->ll_rw_stats_on) {
1787 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1790 spin_lock(&sbi->ll_process_lock);
1792 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1793 (s64)now.tv_sec, now.tv_nsec);
1794 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1795 "R/W", "PID", "RANGE START", "RANGE END",
1796 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1798 /* We stored the discontiguous offsets here; print them first */
1799 for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1800 if (offset[i].rw_pid != 0)
1802 "%3c %10d %14llu %14llu %17lu %17lu %14llu\n",
1803 offset[i].rw_op == READ ? 'R' : 'W',
1805 offset[i].rw_range_start,
1806 offset[i].rw_range_end,
1807 (unsigned long)offset[i].rw_smallest_extent,
1808 (unsigned long)offset[i].rw_largest_extent,
1809 offset[i].rw_offset);
1812 /* Then print the current offsets for each process */
1813 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1814 if (process[i].rw_pid != 0)
1816 "%3c %10d %14llu %14llu %17lu %17lu %14llu\n",
1817 process[i].rw_op == READ ? 'R' : 'W',
1819 process[i].rw_range_start,
1820 process[i].rw_last_file_pos,
1821 (unsigned long)process[i].rw_smallest_extent,
1822 (unsigned long)process[i].rw_largest_extent,
1823 process[i].rw_offset);
1825 spin_unlock(&sbi->ll_process_lock);
1830 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
1831 const char __user *buf,
1832 size_t len, loff_t *off)
1834 struct seq_file *seq = file->private_data;
1835 struct ll_sb_info *sbi = seq->private;
1836 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1837 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1843 value = ll_stats_pid_write(buf, len);
1846 sbi->ll_rw_stats_on = 0;
1848 sbi->ll_rw_stats_on = 1;
1850 spin_lock(&sbi->ll_process_lock);
1851 sbi->ll_offset_process_count = 0;
1852 sbi->ll_rw_offset_entry_count = 0;
1853 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1854 LL_PROCESS_HIST_MAX);
1855 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1856 LL_OFFSET_HIST_MAX);
1857 spin_unlock(&sbi->ll_process_lock);
1862 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);