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);
329 unsigned long ra_max_mb;
331 spin_lock(&sbi->ll_lock);
332 ra_max_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages);
333 spin_unlock(&sbi->ll_lock);
335 seq_printf(m, "%lu\n", ra_max_mb);
340 ll_max_readahead_mb_seq_write(struct file *file, const char __user *buffer,
341 size_t count, loff_t *off)
343 struct seq_file *m = file->private_data;
344 struct super_block *sb = m->private;
345 struct ll_sb_info *sbi = ll_s2sbi(sb);
346 s64 ra_max_mb, pages_number;
349 rc = lprocfs_str_with_units_to_s64(buffer, count, &ra_max_mb, 'M');
353 pages_number = round_up(ra_max_mb, 1024 * 1024) >> PAGE_SHIFT;
354 if (pages_number < 0 || pages_number > cfs_totalram_pages() / 2) {
356 CERROR("%s: can't set max_readahead_mb=%llu > %luMB\n",
357 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
358 PAGES_TO_MiB(cfs_totalram_pages()));
362 spin_lock(&sbi->ll_lock);
363 sbi->ll_ra_info.ra_max_pages = pages_number;
364 spin_unlock(&sbi->ll_lock);
369 LDEBUGFS_SEQ_FOPS(ll_max_readahead_mb);
371 static int ll_max_readahead_per_file_mb_seq_show(struct seq_file *m, void *v)
373 struct super_block *sb = m->private;
374 struct ll_sb_info *sbi = ll_s2sbi(sb);
375 unsigned long ra_max_file_mb;
377 spin_lock(&sbi->ll_lock);
378 ra_max_file_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file);
379 spin_unlock(&sbi->ll_lock);
381 seq_printf(m, "%lu\n", ra_max_file_mb);
386 ll_max_readahead_per_file_mb_seq_write(struct file *file,
387 const char __user *buffer,
388 size_t count, loff_t *off)
390 struct seq_file *m = file->private_data;
391 struct super_block *sb = m->private;
392 struct ll_sb_info *sbi = ll_s2sbi(sb);
393 s64 ra_max_file_mb, pages_number;
396 rc = lprocfs_str_with_units_to_s64(buffer, count, &ra_max_file_mb,
401 pages_number = round_up(ra_max_file_mb, 1024 * 1024) >> PAGE_SHIFT;
402 if (pages_number < 0 || pages_number > sbi->ll_ra_info.ra_max_pages) {
403 CERROR("%s: can't set max_readahead_per_file_mb=%llu > max_read_ahead_mb=%lu\n",
404 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
405 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
409 spin_lock(&sbi->ll_lock);
410 sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
411 spin_unlock(&sbi->ll_lock);
416 LDEBUGFS_SEQ_FOPS(ll_max_readahead_per_file_mb);
418 static int ll_max_read_ahead_whole_mb_seq_show(struct seq_file *m, void *v)
420 struct super_block *sb = m->private;
421 struct ll_sb_info *sbi = ll_s2sbi(sb);
422 unsigned long ra_max_whole_mb;
424 spin_lock(&sbi->ll_lock);
425 ra_max_whole_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_read_ahead_whole_pages);
426 spin_unlock(&sbi->ll_lock);
428 seq_printf(m, "%lu\n", ra_max_whole_mb);
433 ll_max_read_ahead_whole_mb_seq_write(struct file *file,
434 const char __user *buffer,
435 size_t count, loff_t *off)
437 struct seq_file *m = file->private_data;
438 struct super_block *sb = m->private;
439 struct ll_sb_info *sbi = ll_s2sbi(sb);
440 s64 ra_max_whole_mb, pages_number;
443 rc = lprocfs_str_with_units_to_s64(buffer, count, &ra_max_whole_mb,
448 pages_number = round_up(ra_max_whole_mb, 1024 * 1024) >> PAGE_SHIFT;
449 /* Cap this at the current max readahead window size, the readahead
450 * algorithm does this anyway so it's pointless to set it larger.
452 if (pages_number < 0 ||
453 pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
454 CERROR("%s: can't set max_read_ahead_whole_mb=%llu > max_read_ahead_per_file_mb=%lu\n",
455 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
456 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
460 spin_lock(&sbi->ll_lock);
461 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
462 spin_unlock(&sbi->ll_lock);
467 LDEBUGFS_SEQ_FOPS(ll_max_read_ahead_whole_mb);
469 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
471 struct super_block *sb = m->private;
472 struct ll_sb_info *sbi = ll_s2sbi(sb);
473 struct cl_client_cache *cache = sbi->ll_cache;
477 max_cached_mb = PAGES_TO_MiB(cache->ccc_lru_max);
478 unused_mb = PAGES_TO_MiB(atomic_long_read(&cache->ccc_lru_left));
479 seq_printf(m, "users: %d\n"
480 "max_cached_mb: %ld\n"
483 "reclaim_count: %u\n",
484 atomic_read(&cache->ccc_users),
486 max_cached_mb - unused_mb,
488 cache->ccc_lru_shrinkers);
492 static ssize_t ll_max_cached_mb_seq_write(struct file *file,
493 const char __user *buffer,
494 size_t count, loff_t *off)
496 struct seq_file *m = file->private_data;
497 struct super_block *sb = m->private;
498 struct ll_sb_info *sbi = ll_s2sbi(sb);
499 struct cl_client_cache *cache = sbi->ll_cache;
509 if (count >= sizeof(kernbuf))
512 if (copy_from_user(kernbuf, buffer, count))
516 buffer += lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count) -
518 rc = lprocfs_str_with_units_to_s64(buffer, count, &pages_number, 'M');
522 pages_number >>= PAGE_SHIFT;
524 if (pages_number < 0 || pages_number > cfs_totalram_pages()) {
525 CERROR("%s: can't set max cache more than %lu MB\n",
527 PAGES_TO_MiB(cfs_totalram_pages()));
530 /* Allow enough cache so clients can make well-formed RPCs */
531 pages_number = max_t(long, pages_number, PTLRPC_MAX_BRW_PAGES);
533 spin_lock(&sbi->ll_lock);
534 diff = pages_number - cache->ccc_lru_max;
535 spin_unlock(&sbi->ll_lock);
537 /* easy - add more LRU slots. */
539 atomic_long_add(diff, &cache->ccc_lru_left);
543 env = cl_env_get(&refcheck);
545 RETURN(PTR_ERR(env));
551 /* reduce LRU budget from free slots. */
555 ov = atomic_long_read(&cache->ccc_lru_left);
559 nv = ov > diff ? ov - diff : 0;
560 rc = atomic_long_cmpxchg(&cache->ccc_lru_left, ov, nv);
561 if (likely(ov == rc)) {
571 if (sbi->ll_dt_exp == NULL) { /* being initialized */
576 /* difficult - have to ask OSCs to drop LRU slots. */
578 rc = obd_set_info_async(env, sbi->ll_dt_exp,
579 sizeof(KEY_CACHE_LRU_SHRINK),
580 KEY_CACHE_LRU_SHRINK,
581 sizeof(tmp), &tmp, NULL);
585 cl_env_put(env, &refcheck);
589 spin_lock(&sbi->ll_lock);
590 cache->ccc_lru_max = pages_number;
591 spin_unlock(&sbi->ll_lock);
594 atomic_long_add(nrpages, &cache->ccc_lru_left);
599 LDEBUGFS_SEQ_FOPS(ll_max_cached_mb);
601 static ssize_t checksums_show(struct kobject *kobj, struct attribute *attr,
604 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
607 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
610 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
611 const char *buffer, size_t count)
613 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
623 rc = kstrtobool(buffer, &val);
627 sbi->ll_flags |= LL_SBI_CHECKSUM;
629 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
632 rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
633 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
635 CWARN("Failed to set OSC checksum flags: %d\n", rc);
639 LUSTRE_RW_ATTR(checksums);
641 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
643 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
644 enum stats_track_type type)
646 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
649 if (sbi->ll_stats_track_type == type)
650 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
651 else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
652 return sprintf(buf, "0 (all)\n");
654 return sprintf(buf, "untracked\n");
657 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
658 size_t count, enum stats_track_type type)
660 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
665 rc = kstrtoul(buffer, 10, &pid);
669 sbi->ll_stats_track_id = pid;
671 sbi->ll_stats_track_type = STATS_TRACK_ALL;
673 sbi->ll_stats_track_type = type;
674 lprocfs_clear_stats(sbi->ll_stats);
678 static ssize_t stats_track_pid_show(struct kobject *kobj,
679 struct attribute *attr,
682 return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
685 static ssize_t stats_track_pid_store(struct kobject *kobj,
686 struct attribute *attr,
690 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
692 LUSTRE_RW_ATTR(stats_track_pid);
694 static ssize_t stats_track_ppid_show(struct kobject *kobj,
695 struct attribute *attr,
698 return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
701 static ssize_t stats_track_ppid_store(struct kobject *kobj,
702 struct attribute *attr,
706 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
708 LUSTRE_RW_ATTR(stats_track_ppid);
710 static ssize_t stats_track_gid_show(struct kobject *kobj,
711 struct attribute *attr,
714 return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
717 static ssize_t stats_track_gid_store(struct kobject *kobj,
718 struct attribute *attr,
722 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
724 LUSTRE_RW_ATTR(stats_track_gid);
726 static ssize_t statahead_running_max_show(struct kobject *kobj,
727 struct attribute *attr,
730 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
733 return snprintf(buf, 16, "%u\n", sbi->ll_sa_running_max);
736 static ssize_t statahead_running_max_store(struct kobject *kobj,
737 struct attribute *attr,
741 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
746 rc = kstrtoul(buffer, 0, &val);
750 if (val <= LL_SA_RUNNING_MAX) {
751 sbi->ll_sa_running_max = val;
755 CERROR("Bad statahead_running_max value %lu. Valid values "
756 "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
760 LUSTRE_RW_ATTR(statahead_running_max);
762 static ssize_t statahead_max_show(struct kobject *kobj,
763 struct attribute *attr,
766 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
769 return sprintf(buf, "%u\n", sbi->ll_sa_max);
772 static ssize_t statahead_max_store(struct kobject *kobj,
773 struct attribute *attr,
777 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
782 rc = kstrtoul(buffer, 0, &val);
786 if (val <= LL_SA_RPC_MAX)
787 sbi->ll_sa_max = val;
789 CERROR("Bad statahead_max value %lu. Valid values are in the range [0, %d]\n",
794 LUSTRE_RW_ATTR(statahead_max);
796 static ssize_t statahead_agl_show(struct kobject *kobj,
797 struct attribute *attr,
800 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
803 return sprintf(buf, "%u\n", sbi->ll_flags & LL_SBI_AGL_ENABLED ? 1 : 0);
806 static ssize_t statahead_agl_store(struct kobject *kobj,
807 struct attribute *attr,
811 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
816 rc = kstrtobool(buffer, &val);
821 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
823 sbi->ll_flags &= ~LL_SBI_AGL_ENABLED;
827 LUSTRE_RW_ATTR(statahead_agl);
829 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
831 struct super_block *sb = m->private;
832 struct ll_sb_info *sbi = ll_s2sbi(sb);
834 seq_printf(m, "statahead total: %u\n"
835 "statahead wrong: %u\n"
837 atomic_read(&sbi->ll_sa_total),
838 atomic_read(&sbi->ll_sa_wrong),
839 atomic_read(&sbi->ll_agl_total));
843 LDEBUGFS_SEQ_FOPS_RO(ll_statahead_stats);
845 static ssize_t lazystatfs_show(struct kobject *kobj,
846 struct attribute *attr,
849 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
852 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
855 static ssize_t lazystatfs_store(struct kobject *kobj,
856 struct attribute *attr,
860 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
865 rc = kstrtobool(buffer, &val);
870 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
872 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
876 LUSTRE_RW_ATTR(lazystatfs);
878 static ssize_t statfs_max_age_show(struct kobject *kobj, struct attribute *attr,
881 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
884 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_statfs_max_age);
887 static ssize_t statfs_max_age_store(struct kobject *kobj,
888 struct attribute *attr, const char *buffer,
891 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
896 rc = kstrtouint(buffer, 10, &val);
899 if (val > OBD_STATFS_CACHE_MAX_AGE)
902 sbi->ll_statfs_max_age = val;
906 LUSTRE_RW_ATTR(statfs_max_age);
908 static ssize_t max_easize_show(struct kobject *kobj,
909 struct attribute *attr,
912 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
917 rc = ll_get_max_mdsize(sbi, &ealen);
921 /* Limit xattr size returned to userspace based on kernel maximum */
922 return snprintf(buf, PAGE_SIZE, "%u\n",
923 ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
925 LUSTRE_RO_ATTR(max_easize);
928 * Get default_easize.
930 * \see client_obd::cl_default_mds_easize
932 * \param[in] m seq_file handle
933 * \param[in] v unused for single entry
935 * \retval 0 on success
936 * \retval negative negated errno on failure
938 static ssize_t default_easize_show(struct kobject *kobj,
939 struct attribute *attr,
942 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
947 rc = ll_get_default_mdsize(sbi, &ealen);
951 /* Limit xattr size returned to userspace based on kernel maximum */
952 return snprintf(buf, PAGE_SIZE, "%u\n",
953 ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
957 * Set default_easize.
959 * Range checking on the passed value is handled by
960 * ll_set_default_mdsize().
962 * \see client_obd::cl_default_mds_easize
964 * \param[in] file proc file
965 * \param[in] buffer string passed from user space
966 * \param[in] count \a buffer length
967 * \param[in] off unused for single entry
969 * \retval positive \a count on success
970 * \retval negative negated errno on failure
972 static ssize_t default_easize_store(struct kobject *kobj,
973 struct attribute *attr,
977 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
985 rc = kstrtouint(buffer, 10, &val);
989 rc = ll_set_default_mdsize(sbi, val);
995 LUSTRE_RW_ATTR(default_easize);
997 static int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
999 const char *str[] = LL_SBI_FLAGS;
1000 struct super_block *sb = m->private;
1001 int flags = ll_s2sbi(sb)->ll_flags;
1004 while (flags != 0) {
1005 if (ARRAY_SIZE(str) <= i) {
1006 CERROR("%s: Revise array LL_SBI_FLAGS to match sbi "
1007 "flags please.\n", ll_s2sbi(sb)->ll_fsname);
1012 seq_printf(m, "%s ", str[i]);
1016 seq_printf(m, "\b\n");
1020 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
1022 static ssize_t xattr_cache_show(struct kobject *kobj,
1023 struct attribute *attr,
1026 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1029 return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
1032 static ssize_t xattr_cache_store(struct kobject *kobj,
1033 struct attribute *attr,
1037 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1042 rc = kstrtobool(buffer, &val);
1046 if (val && !(sbi->ll_flags & LL_SBI_XATTR_CACHE))
1049 sbi->ll_xattr_cache_enabled = val;
1050 sbi->ll_xattr_cache_set = 1;
1054 LUSTRE_RW_ATTR(xattr_cache);
1056 static ssize_t tiny_write_show(struct kobject *kobj,
1057 struct attribute *attr,
1060 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1063 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_TINY_WRITE));
1066 static ssize_t tiny_write_store(struct kobject *kobj,
1067 struct attribute *attr,
1071 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1076 rc = kstrtobool(buffer, &val);
1080 spin_lock(&sbi->ll_lock);
1082 sbi->ll_flags |= LL_SBI_TINY_WRITE;
1084 sbi->ll_flags &= ~LL_SBI_TINY_WRITE;
1085 spin_unlock(&sbi->ll_lock);
1089 LUSTRE_RW_ATTR(tiny_write);
1091 static ssize_t max_read_ahead_async_active_show(struct kobject *kobj,
1092 struct attribute *attr,
1095 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1098 return snprintf(buf, PAGE_SIZE, "%u\n",
1099 sbi->ll_ra_info.ra_async_max_active);
1102 static ssize_t max_read_ahead_async_active_store(struct kobject *kobj,
1103 struct attribute *attr,
1109 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1112 rc = kstrtouint(buffer, 10, &val);
1116 if (val < 1 || val > WQ_UNBOUND_MAX_ACTIVE) {
1117 CERROR("%s: cannot set max_read_ahead_async_active=%u %s than %u\n",
1118 sbi->ll_fsname, val,
1119 val < 1 ? "smaller" : "larger",
1120 val < 1 ? 1 : WQ_UNBOUND_MAX_ACTIVE);
1124 sbi->ll_ra_info.ra_async_max_active = val;
1125 workqueue_set_max_active(sbi->ll_ra_info.ll_readahead_wq, val);
1129 LUSTRE_RW_ATTR(max_read_ahead_async_active);
1131 static ssize_t read_ahead_async_file_threshold_mb_show(struct kobject *kobj,
1132 struct attribute *attr,
1135 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1138 return snprintf(buf, PAGE_SIZE, "%lu\n",
1139 PAGES_TO_MiB(sbi->ll_ra_info.ra_async_pages_per_file_threshold));
1143 read_ahead_async_file_threshold_mb_store(struct kobject *kobj,
1144 struct attribute *attr,
1145 const char *buffer, size_t count)
1147 unsigned long pages_number;
1148 unsigned long max_ra_per_file;
1149 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1153 rc = kstrtoul(buffer, 10, &pages_number);
1157 pages_number = MiB_TO_PAGES(pages_number);
1158 max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1159 if (pages_number < 0 || pages_number > max_ra_per_file) {
1160 CERROR("%s: can't set read_ahead_async_file_threshold_mb=%lu > "
1161 "max_read_readahead_per_file_mb=%lu\n", sbi->ll_fsname,
1162 PAGES_TO_MiB(pages_number),
1163 PAGES_TO_MiB(max_ra_per_file));
1166 sbi->ll_ra_info.ra_async_pages_per_file_threshold = pages_number;
1170 LUSTRE_RW_ATTR(read_ahead_async_file_threshold_mb);
1172 static ssize_t fast_read_show(struct kobject *kobj,
1173 struct attribute *attr,
1176 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1179 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_FAST_READ));
1182 static ssize_t fast_read_store(struct kobject *kobj,
1183 struct attribute *attr,
1187 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1192 rc = kstrtobool(buffer, &val);
1196 spin_lock(&sbi->ll_lock);
1198 sbi->ll_flags |= LL_SBI_FAST_READ;
1200 sbi->ll_flags &= ~LL_SBI_FAST_READ;
1201 spin_unlock(&sbi->ll_lock);
1205 LUSTRE_RW_ATTR(fast_read);
1207 static ssize_t file_heat_show(struct kobject *kobj,
1208 struct attribute *attr,
1211 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1214 return snprintf(buf, PAGE_SIZE, "%u\n",
1215 !!(sbi->ll_flags & LL_SBI_FILE_HEAT));
1218 static ssize_t file_heat_store(struct kobject *kobj,
1219 struct attribute *attr,
1223 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1228 rc = kstrtobool(buffer, &val);
1232 spin_lock(&sbi->ll_lock);
1234 sbi->ll_flags |= LL_SBI_FILE_HEAT;
1236 sbi->ll_flags &= ~LL_SBI_FILE_HEAT;
1237 spin_unlock(&sbi->ll_lock);
1241 LUSTRE_RW_ATTR(file_heat);
1243 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1244 struct attribute *attr,
1247 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1250 return snprintf(buf, PAGE_SIZE, "%u\n",
1251 (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1254 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1255 struct attribute *attr,
1259 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1264 rc = kstrtoul(buffer, 10, &val);
1268 if (val < 0 || val > 100)
1271 sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1275 LUSTRE_RW_ATTR(heat_decay_percentage);
1277 static ssize_t heat_period_second_show(struct kobject *kobj,
1278 struct attribute *attr,
1281 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1284 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1287 static ssize_t heat_period_second_store(struct kobject *kobj,
1288 struct attribute *attr,
1292 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1297 rc = kstrtoul(buffer, 10, &val);
1304 sbi->ll_heat_period_second = val;
1308 LUSTRE_RW_ATTR(heat_period_second);
1310 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1312 struct super_block *sb = m->private;
1313 struct ll_sb_info *sbi = ll_s2sbi(sb);
1314 struct cl_client_cache *cache = sbi->ll_cache;
1318 pages = atomic_long_read(&cache->ccc_unstable_nr);
1319 mb = (pages * PAGE_SIZE) >> 20;
1321 seq_printf(m, "unstable_check: %8d\n"
1322 "unstable_pages: %12ld\n"
1323 "unstable_mb: %8d\n",
1324 cache->ccc_unstable_check, pages, mb);
1328 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1329 const char __user *buffer,
1330 size_t count, loff_t *unused)
1332 struct seq_file *seq = file->private_data;
1333 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1340 if (count >= sizeof(kernbuf))
1343 if (copy_from_user(kernbuf, buffer, count))
1347 buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1349 rc = kstrtobool_from_user(buffer, count, &val);
1353 /* borrow lru lock to set the value */
1354 spin_lock(&sbi->ll_cache->ccc_lru_lock);
1355 sbi->ll_cache->ccc_unstable_check = val;
1356 spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1361 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1363 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1365 struct super_block *sb = m->private;
1366 struct ll_sb_info *sbi = ll_s2sbi(sb);
1367 struct root_squash_info *squash = &sbi->ll_squash;
1369 seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1373 static ssize_t ll_root_squash_seq_write(struct file *file,
1374 const char __user *buffer,
1375 size_t count, loff_t *off)
1377 struct seq_file *m = file->private_data;
1378 struct super_block *sb = m->private;
1379 struct ll_sb_info *sbi = ll_s2sbi(sb);
1380 struct root_squash_info *squash = &sbi->ll_squash;
1382 return lprocfs_wr_root_squash(buffer, count, squash, sbi->ll_fsname);
1385 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1387 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1389 struct super_block *sb = m->private;
1390 struct ll_sb_info *sbi = ll_s2sbi(sb);
1391 struct root_squash_info *squash = &sbi->ll_squash;
1394 spin_lock(&squash->rsi_lock);
1395 if (!list_empty(&squash->rsi_nosquash_nids)) {
1396 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1397 &squash->rsi_nosquash_nids);
1401 seq_puts(m, "NONE\n");
1403 spin_unlock(&squash->rsi_lock);
1408 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1409 const char __user *buffer,
1410 size_t count, loff_t *off)
1412 struct seq_file *m = file->private_data;
1413 struct super_block *sb = m->private;
1414 struct ll_sb_info *sbi = ll_s2sbi(sb);
1415 struct root_squash_info *squash = &sbi->ll_squash;
1418 rc = lprocfs_wr_nosquash_nids(buffer, count, squash, sbi->ll_fsname);
1422 ll_compute_rootsquash_state(sbi);
1427 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1429 static int ll_pcc_seq_show(struct seq_file *m, void *v)
1431 struct super_block *sb = m->private;
1432 struct ll_sb_info *sbi = ll_s2sbi(sb);
1434 return pcc_super_dump(&sbi->ll_pcc_super, m);
1437 static ssize_t ll_pcc_seq_write(struct file *file, const char __user *buffer,
1438 size_t count, loff_t *off)
1440 struct seq_file *m = file->private_data;
1441 struct super_block *sb = m->private;
1442 struct ll_sb_info *sbi = ll_s2sbi(sb);
1446 if (count >= LPROCFS_WR_PCC_MAX_CMD)
1449 if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_PCC))
1452 OBD_ALLOC(kernbuf, count + 1);
1453 if (kernbuf == NULL)
1456 if (copy_from_user(kernbuf, buffer, count))
1457 GOTO(out_free_kernbuff, rc = -EFAULT);
1459 rc = pcc_cmd_handle(kernbuf, count, &sbi->ll_pcc_super);
1461 OBD_FREE(kernbuf, count + 1);
1462 return rc ? rc : count;
1464 LPROC_SEQ_FOPS(ll_pcc);
1466 struct lprocfs_vars lprocfs_llite_obd_vars[] = {
1468 .fops = &ll_site_stats_fops },
1469 { .name = "max_read_ahead_mb",
1470 .fops = &ll_max_readahead_mb_fops },
1471 { .name = "max_read_ahead_per_file_mb",
1472 .fops = &ll_max_readahead_per_file_mb_fops },
1473 { .name = "max_read_ahead_whole_mb",
1474 .fops = &ll_max_read_ahead_whole_mb_fops },
1475 { .name = "max_cached_mb",
1476 .fops = &ll_max_cached_mb_fops },
1477 { .name = "statahead_stats",
1478 .fops = &ll_statahead_stats_fops },
1479 { .name = "unstable_stats",
1480 .fops = &ll_unstable_stats_fops },
1481 { .name = "sbi_flags",
1482 .fops = &ll_sbi_flags_fops },
1483 { .name = "root_squash",
1484 .fops = &ll_root_squash_fops },
1485 { .name = "nosquash_nids",
1486 .fops = &ll_nosquash_nids_fops },
1488 .fops = &ll_pcc_fops, },
1492 #define MAX_STRING_SIZE 128
1494 static struct attribute *llite_attrs[] = {
1495 &lustre_attr_blocksize.attr,
1496 &lustre_attr_stat_blocksize.attr,
1497 &lustre_attr_kbytestotal.attr,
1498 &lustre_attr_kbytesfree.attr,
1499 &lustre_attr_kbytesavail.attr,
1500 &lustre_attr_filestotal.attr,
1501 &lustre_attr_filesfree.attr,
1502 &lustre_attr_client_type.attr,
1503 &lustre_attr_fstype.attr,
1504 &lustre_attr_uuid.attr,
1505 &lustre_attr_checksums.attr,
1506 &lustre_attr_checksum_pages.attr,
1507 &lustre_attr_stats_track_pid.attr,
1508 &lustre_attr_stats_track_ppid.attr,
1509 &lustre_attr_stats_track_gid.attr,
1510 &lustre_attr_statahead_running_max.attr,
1511 &lustre_attr_statahead_max.attr,
1512 &lustre_attr_statahead_agl.attr,
1513 &lustre_attr_lazystatfs.attr,
1514 &lustre_attr_statfs_max_age.attr,
1515 &lustre_attr_max_easize.attr,
1516 &lustre_attr_default_easize.attr,
1517 &lustre_attr_xattr_cache.attr,
1518 &lustre_attr_fast_read.attr,
1519 &lustre_attr_tiny_write.attr,
1520 &lustre_attr_file_heat.attr,
1521 &lustre_attr_heat_decay_percentage.attr,
1522 &lustre_attr_heat_period_second.attr,
1523 &lustre_attr_max_read_ahead_async_active.attr,
1524 &lustre_attr_read_ahead_async_file_threshold_mb.attr,
1528 static void sbi_kobj_release(struct kobject *kobj)
1530 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1532 complete(&sbi->ll_kobj_unregister);
1535 static struct kobj_type sbi_ktype = {
1536 .default_attrs = llite_attrs,
1537 .sysfs_ops = &lustre_sysfs_ops,
1538 .release = sbi_kobj_release,
1541 #define LPROCFS_TYPE_LATENCY \
1542 (LPROCFS_TYPE_USEC | LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV)
1543 static const struct llite_file_opcode {
1547 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1548 /* file operation */
1549 { LPROC_LL_READ_BYTES, LPROCFS_CNTR_AVGMINMAX | LPROCFS_TYPE_BYTES,
1551 { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX | LPROCFS_TYPE_BYTES,
1553 { LPROC_LL_READ, LPROCFS_TYPE_LATENCY, "read" },
1554 { LPROC_LL_WRITE, LPROCFS_TYPE_LATENCY, "write" },
1555 { LPROC_LL_IOCTL, LPROCFS_TYPE_REQS, "ioctl" },
1556 { LPROC_LL_OPEN, LPROCFS_TYPE_LATENCY, "open" },
1557 { LPROC_LL_RELEASE, LPROCFS_TYPE_LATENCY, "close" },
1558 { LPROC_LL_MMAP, LPROCFS_TYPE_LATENCY, "mmap" },
1559 { LPROC_LL_FAULT, LPROCFS_TYPE_LATENCY, "page_fault" },
1560 { LPROC_LL_MKWRITE, LPROCFS_TYPE_LATENCY, "page_mkwrite" },
1561 { LPROC_LL_LLSEEK, LPROCFS_TYPE_LATENCY, "seek" },
1562 { LPROC_LL_FSYNC, LPROCFS_TYPE_LATENCY, "fsync" },
1563 { LPROC_LL_READDIR, LPROCFS_TYPE_LATENCY, "readdir" },
1564 /* inode operation */
1565 { LPROC_LL_SETATTR, LPROCFS_TYPE_LATENCY, "setattr" },
1566 { LPROC_LL_TRUNC, LPROCFS_TYPE_LATENCY, "truncate" },
1567 { LPROC_LL_FLOCK, LPROCFS_TYPE_LATENCY, "flock" },
1568 { LPROC_LL_GETATTR, LPROCFS_TYPE_LATENCY, "getattr" },
1569 /* dir inode operation */
1570 { LPROC_LL_CREATE, LPROCFS_TYPE_LATENCY, "create" },
1571 { LPROC_LL_LINK, LPROCFS_TYPE_LATENCY, "link" },
1572 { LPROC_LL_UNLINK, LPROCFS_TYPE_LATENCY, "unlink" },
1573 { LPROC_LL_SYMLINK, LPROCFS_TYPE_LATENCY, "symlink" },
1574 { LPROC_LL_MKDIR, LPROCFS_TYPE_LATENCY, "mkdir" },
1575 { LPROC_LL_RMDIR, LPROCFS_TYPE_LATENCY, "rmdir" },
1576 { LPROC_LL_MKNOD, LPROCFS_TYPE_LATENCY, "mknod" },
1577 { LPROC_LL_RENAME, LPROCFS_TYPE_LATENCY, "rename" },
1578 /* special inode operation */
1579 { LPROC_LL_STATFS, LPROCFS_TYPE_LATENCY, "statfs" },
1580 { LPROC_LL_SETXATTR, LPROCFS_TYPE_LATENCY, "setxattr" },
1581 { LPROC_LL_GETXATTR, LPROCFS_TYPE_LATENCY, "getxattr" },
1582 { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REQS, "getxattr_hits" },
1583 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_LATENCY, "listxattr" },
1584 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_LATENCY, "removexattr" },
1585 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_LATENCY, "inode_permission" },
1588 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count)
1593 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1594 lprocfs_counter_add(sbi->ll_stats, op, count);
1595 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1596 sbi->ll_stats_track_id == current->pid)
1597 lprocfs_counter_add(sbi->ll_stats, op, count);
1598 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1599 sbi->ll_stats_track_id == current->parent->pid)
1600 lprocfs_counter_add(sbi->ll_stats, op, count);
1601 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1602 sbi->ll_stats_track_id ==
1603 from_kgid(&init_user_ns, current_gid()))
1604 lprocfs_counter_add(sbi->ll_stats, op, count);
1606 EXPORT_SYMBOL(ll_stats_ops_tally);
1608 static const char *ra_stat_string[] = {
1609 [RA_STAT_HIT] = "hits",
1610 [RA_STAT_MISS] = "misses",
1611 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1612 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1613 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1614 [RA_STAT_FAILED_MATCH] = "failed lock match",
1615 [RA_STAT_DISCARDED] = "read but discarded",
1616 [RA_STAT_ZERO_LEN] = "zero length file",
1617 [RA_STAT_ZERO_WINDOW] = "zero size window",
1618 [RA_STAT_EOF] = "read-ahead to EOF",
1619 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1620 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1621 [RA_STAT_FAILED_REACH_END] = "failed to reach end",
1622 [RA_STAT_ASYNC] = "async readahead",
1623 [RA_STAT_FAILED_FAST_READ] = "failed to fast read",
1626 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1628 struct lustre_sb_info *lsi = s2lsi(sb);
1629 struct ll_sb_info *sbi = ll_s2sbi(sb);
1635 if (IS_ERR_OR_NULL(llite_root))
1638 sbi->ll_debugfs_entry = ldebugfs_register(name, llite_root,
1639 lprocfs_llite_obd_vars, sb);
1640 if (IS_ERR_OR_NULL(sbi->ll_debugfs_entry)) {
1641 err = sbi->ll_debugfs_entry ? PTR_ERR(sbi->ll_debugfs_entry) :
1643 sbi->ll_debugfs_entry = NULL;
1647 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "dump_page_cache",0444,
1648 &vvp_dump_pgcache_file_ops, sbi);
1650 CWARN("Error adding the dump_page_cache file\n");
1652 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "extents_stats", 0644,
1653 &ll_rw_extents_stats_fops, sbi);
1655 CWARN("Error adding the extent_stats file\n");
1657 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry,
1658 "extents_stats_per_process", 0644,
1659 &ll_rw_extents_stats_pp_fops, sbi);
1661 CWARN("Error adding the extents_stats_per_process file\n");
1663 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "offset_stats", 0644,
1664 &ll_rw_offset_stats_fops, sbi);
1666 CWARN("Error adding the offset_stats file\n");
1668 /* File operations stats */
1669 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1670 LPROCFS_STATS_FLAG_NONE);
1671 if (sbi->ll_stats == NULL)
1672 GOTO(out_debugfs, err = -ENOMEM);
1674 /* do counter init */
1675 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1676 u32 type = llite_opcode_table[id].type;
1679 if (type & LPROCFS_TYPE_REQS)
1681 else if (type & LPROCFS_TYPE_BYTES)
1683 else if (type & LPROCFS_TYPE_PAGES)
1685 else if (type & LPROCFS_TYPE_USEC)
1687 lprocfs_counter_init(sbi->ll_stats,
1688 llite_opcode_table[id].opcode,
1689 (type & LPROCFS_CNTR_AVGMINMAX),
1690 llite_opcode_table[id].opname, ptr);
1693 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "stats",
1696 GOTO(out_stats, err);
1698 sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1699 LPROCFS_STATS_FLAG_NONE);
1700 if (sbi->ll_ra_stats == NULL)
1701 GOTO(out_stats, err = -ENOMEM);
1703 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1704 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1705 ra_stat_string[id], "pages");
1707 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "read_ahead_stats",
1710 GOTO(out_ra_stats, err);
1713 /* Yes we also register sysfs mount kset here as well */
1714 sbi->ll_kset.kobj.parent = llite_kobj;
1715 sbi->ll_kset.kobj.ktype = &sbi_ktype;
1716 init_completion(&sbi->ll_kobj_unregister);
1717 err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1719 GOTO(out_ra_stats, err);
1721 err = kset_register(&sbi->ll_kset);
1723 GOTO(out_ra_stats, err);
1725 lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1729 lprocfs_free_stats(&sbi->ll_ra_stats);
1731 lprocfs_free_stats(&sbi->ll_stats);
1733 debugfs_remove_recursive(sbi->ll_debugfs_entry);
1738 void ll_debugfs_unregister_super(struct super_block *sb)
1740 struct lustre_sb_info *lsi = s2lsi(sb);
1741 struct ll_sb_info *sbi = ll_s2sbi(sb);
1743 debugfs_remove_recursive(sbi->ll_debugfs_entry);
1746 sysfs_remove_link(&sbi->ll_kset.kobj,
1747 sbi->ll_dt_obd->obd_type->typ_name);
1750 sysfs_remove_link(&sbi->ll_kset.kobj,
1751 sbi->ll_md_obd->obd_type->typ_name);
1753 kobject_put(lsi->lsi_kobj);
1755 kset_unregister(&sbi->ll_kset);
1756 wait_for_completion(&sbi->ll_kobj_unregister);
1758 lprocfs_free_stats(&sbi->ll_ra_stats);
1759 lprocfs_free_stats(&sbi->ll_stats);
1761 #undef MAX_STRING_SIZE
1763 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1764 struct seq_file *seq, int which)
1766 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1767 unsigned long start, end, r, w;
1768 char *unitp = "KMGTPEZY";
1770 struct per_process_info *pp_info = &io_extents->pp_extents[which];
1776 for(i = 0; i < LL_HIST_MAX; i++) {
1777 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1778 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1781 for(i = 0; i < LL_HIST_MAX; i++) {
1782 r = pp_info->pp_r_hist.oh_buckets[i];
1783 w = pp_info->pp_w_hist.oh_buckets[i];
1786 end = BIT(i + LL_HIST_START - units);
1787 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u | "
1788 "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1789 (i == LL_HIST_MAX - 1) ? '+' : ' ',
1790 r, pct(r, read_tot), pct(read_cum, read_tot),
1791 w, pct(w, write_tot), pct(write_cum, write_tot));
1793 if (start == BIT(10)) {
1798 if (read_cum == read_tot && write_cum == write_tot)
1803 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1805 struct timespec64 now;
1806 struct ll_sb_info *sbi = seq->private;
1807 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1810 ktime_get_real_ts64(&now);
1812 if (!sbi->ll_rw_stats_on) {
1813 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1816 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1817 (s64)now.tv_sec, now.tv_nsec);
1818 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1819 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1820 "extents", "calls", "%", "cum%",
1821 "calls", "%", "cum%");
1822 spin_lock(&sbi->ll_pp_extent_lock);
1823 for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1824 if (io_extents->pp_extents[k].pid != 0) {
1825 seq_printf(seq, "\nPID: %d\n",
1826 io_extents->pp_extents[k].pid);
1827 ll_display_extents_info(io_extents, seq, k);
1830 spin_unlock(&sbi->ll_pp_extent_lock);
1834 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1835 const char __user *buf,
1839 struct seq_file *seq = file->private_data;
1840 struct ll_sb_info *sbi = seq->private;
1841 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1848 value = ll_stats_pid_write(buf, len);
1851 sbi->ll_rw_stats_on = 0;
1853 sbi->ll_rw_stats_on = 1;
1855 spin_lock(&sbi->ll_pp_extent_lock);
1856 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1857 io_extents->pp_extents[i].pid = 0;
1858 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1859 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1861 spin_unlock(&sbi->ll_pp_extent_lock);
1865 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1867 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1869 struct timespec64 now;
1870 struct ll_sb_info *sbi = seq->private;
1871 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1873 ktime_get_real_ts64(&now);
1875 if (!sbi->ll_rw_stats_on) {
1876 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1879 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1880 (s64)now.tv_sec, now.tv_nsec);
1882 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1883 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1884 "extents", "calls", "%", "cum%",
1885 "calls", "%", "cum%");
1886 spin_lock(&sbi->ll_lock);
1887 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1888 spin_unlock(&sbi->ll_lock);
1893 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1894 const char __user *buf,
1895 size_t len, loff_t *off)
1897 struct seq_file *seq = file->private_data;
1898 struct ll_sb_info *sbi = seq->private;
1899 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1906 value = ll_stats_pid_write(buf, len);
1909 sbi->ll_rw_stats_on = 0;
1911 sbi->ll_rw_stats_on = 1;
1913 spin_lock(&sbi->ll_pp_extent_lock);
1914 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1915 io_extents->pp_extents[i].pid = 0;
1916 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1917 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1919 spin_unlock(&sbi->ll_pp_extent_lock);
1924 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1926 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1927 struct ll_file_data *file, loff_t pos,
1928 size_t count, int rw)
1931 struct ll_rw_process_info *process;
1932 struct ll_rw_process_info *offset;
1933 int *off_count = &sbi->ll_rw_offset_entry_count;
1934 int *process_count = &sbi->ll_offset_process_count;
1935 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1937 if(!sbi->ll_rw_stats_on)
1939 process = sbi->ll_rw_process_info;
1940 offset = sbi->ll_rw_offset_info;
1942 spin_lock(&sbi->ll_pp_extent_lock);
1943 /* Extent statistics */
1944 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1945 if(io_extents->pp_extents[i].pid == pid) {
1953 sbi->ll_extent_process_count =
1954 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1955 cur = sbi->ll_extent_process_count;
1956 io_extents->pp_extents[cur].pid = pid;
1957 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1958 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1961 for (i = 0; (count >= BIT(LL_HIST_START + i)) &&
1962 (i < (LL_HIST_MAX - 1)); i++);
1964 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1965 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1967 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1968 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1970 spin_unlock(&sbi->ll_pp_extent_lock);
1972 spin_lock(&sbi->ll_process_lock);
1973 /* Offset statistics */
1974 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1975 if (process[i].rw_pid == pid) {
1976 if (process[i].rw_last_file != file) {
1977 process[i].rw_range_start = pos;
1978 process[i].rw_last_file_pos = pos + count;
1979 process[i].rw_smallest_extent = count;
1980 process[i].rw_largest_extent = count;
1981 process[i].rw_offset = 0;
1982 process[i].rw_last_file = file;
1983 spin_unlock(&sbi->ll_process_lock);
1986 if (process[i].rw_last_file_pos != pos) {
1988 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1989 offset[*off_count].rw_op = process[i].rw_op;
1990 offset[*off_count].rw_pid = pid;
1991 offset[*off_count].rw_range_start =
1992 process[i].rw_range_start;
1993 offset[*off_count].rw_range_end =
1994 process[i].rw_last_file_pos;
1995 offset[*off_count].rw_smallest_extent =
1996 process[i].rw_smallest_extent;
1997 offset[*off_count].rw_largest_extent =
1998 process[i].rw_largest_extent;
1999 offset[*off_count].rw_offset =
2000 process[i].rw_offset;
2001 process[i].rw_op = rw;
2002 process[i].rw_range_start = pos;
2003 process[i].rw_smallest_extent = count;
2004 process[i].rw_largest_extent = count;
2005 process[i].rw_offset = pos -
2006 process[i].rw_last_file_pos;
2008 if(process[i].rw_smallest_extent > count)
2009 process[i].rw_smallest_extent = count;
2010 if(process[i].rw_largest_extent < count)
2011 process[i].rw_largest_extent = count;
2012 process[i].rw_last_file_pos = pos + count;
2013 spin_unlock(&sbi->ll_process_lock);
2017 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
2018 process[*process_count].rw_pid = pid;
2019 process[*process_count].rw_op = rw;
2020 process[*process_count].rw_range_start = pos;
2021 process[*process_count].rw_last_file_pos = pos + count;
2022 process[*process_count].rw_smallest_extent = count;
2023 process[*process_count].rw_largest_extent = count;
2024 process[*process_count].rw_offset = 0;
2025 process[*process_count].rw_last_file = file;
2026 spin_unlock(&sbi->ll_process_lock);
2029 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
2031 struct timespec64 now;
2032 struct ll_sb_info *sbi = seq->private;
2033 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
2034 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
2037 ktime_get_real_ts64(&now);
2039 if (!sbi->ll_rw_stats_on) {
2040 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2043 spin_lock(&sbi->ll_process_lock);
2045 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
2046 (s64)now.tv_sec, now.tv_nsec);
2047 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
2048 "R/W", "PID", "RANGE START", "RANGE END",
2049 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
2051 /* We stored the discontiguous offsets here; print them first */
2052 for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
2053 if (offset[i].rw_pid != 0)
2055 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2056 offset[i].rw_op == READ ? 'R' : 'W',
2058 offset[i].rw_range_start,
2059 offset[i].rw_range_end,
2060 (unsigned long)offset[i].rw_smallest_extent,
2061 (unsigned long)offset[i].rw_largest_extent,
2062 offset[i].rw_offset);
2065 /* Then print the current offsets for each process */
2066 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2067 if (process[i].rw_pid != 0)
2069 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2070 process[i].rw_op == READ ? 'R' : 'W',
2072 process[i].rw_range_start,
2073 process[i].rw_last_file_pos,
2074 (unsigned long)process[i].rw_smallest_extent,
2075 (unsigned long)process[i].rw_largest_extent,
2076 process[i].rw_offset);
2078 spin_unlock(&sbi->ll_process_lock);
2083 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
2084 const char __user *buf,
2085 size_t len, loff_t *off)
2087 struct seq_file *seq = file->private_data;
2088 struct ll_sb_info *sbi = seq->private;
2089 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
2090 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
2096 value = ll_stats_pid_write(buf, len);
2099 sbi->ll_rw_stats_on = 0;
2101 sbi->ll_rw_stats_on = 1;
2103 spin_lock(&sbi->ll_process_lock);
2104 sbi->ll_offset_process_count = 0;
2105 sbi->ll_rw_offset_entry_count = 0;
2106 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
2107 LL_PROCESS_HIST_MAX);
2108 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
2109 LL_OFFSET_HIST_MAX);
2110 spin_unlock(&sbi->ll_process_lock);
2115 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);