4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 #define DEBUG_SUBSYSTEM S_LLITE
34 #include <linux/version.h>
35 #include <linux/user_namespace.h>
36 #include <linux/uidgid.h>
38 #include <uapi/linux/lustre/lustre_param.h>
39 #include <lprocfs_status.h>
40 #include <obd_support.h>
42 #include "llite_internal.h"
43 #include "vvp_internal.h"
45 static struct kobject *llite_kobj;
46 static struct dentry *llite_root;
48 static void llite_kobj_release(struct kobject *kobj)
50 if (!IS_ERR_OR_NULL(llite_root)) {
51 debugfs_remove(llite_root);
58 static struct kobj_type llite_kobj_ktype = {
59 .release = llite_kobj_release,
60 .sysfs_ops = &lustre_sysfs_ops,
63 int llite_tunables_register(void)
67 llite_kobj = kzalloc(sizeof(*llite_kobj), GFP_KERNEL);
71 llite_kobj->kset = lustre_kset;
72 rc = kobject_init_and_add(llite_kobj, &llite_kobj_ktype,
73 &lustre_kset->kobj, "%s", "llite");
77 llite_root = debugfs_create_dir("llite", debugfs_lustre_root);
81 kobject_put(llite_kobj);
87 void llite_tunables_unregister(void)
89 kobject_put(llite_kobj);
93 /* <debugfs>/lustre/llite mount point registration */
94 static const struct file_operations ll_rw_extents_stats_fops;
95 static const struct file_operations ll_rw_extents_stats_pp_fops;
96 static const struct file_operations ll_rw_offset_stats_fops;
99 * ll_stats_pid_write() - Determine if stats collection should be enabled
100 * @buf: Buffer containing the data written
101 * @len: Number of bytes in the buffer
103 * Several proc files begin collecting stats when a value is written, and stop
104 * collecting when either '0' or 'disable' is written. This function checks the
105 * written value to see if collection should be enabled or disabled.
107 * Return: If '0' or 'disable' is provided, 0 is returned. If the text
108 * equivalent of a number is written, that number is returned. Otherwise,
109 * 1 is returned. Non-zero return values indicate collection should be enabled.
111 static s64 ll_stats_pid_write(const char __user *buf, size_t len)
113 unsigned long long value = 1;
117 rc = kstrtoull_from_user(buf, len, 0, &value);
118 if (rc < 0 && len < sizeof(kernbuf)) {
119 if (copy_from_user(kernbuf, buf, len))
123 if (kernbuf[len - 1] == '\n')
124 kernbuf[len - 1] = 0;
126 if (strncasecmp(kernbuf, "disable", 7) == 0)
133 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
136 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
138 struct obd_statfs osfs;
141 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
145 return sprintf(buf, "%u\n", osfs.os_bsize);
147 LUSTRE_RO_ATTR(blocksize);
149 static ssize_t stat_blocksize_show(struct kobject *kobj, struct attribute *attr,
152 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
155 return sprintf(buf, "%u\n", sbi->ll_stat_blksize);
158 static ssize_t stat_blocksize_store(struct kobject *kobj,
159 struct attribute *attr,
163 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
168 rc = kstrtouint(buffer, 10, &val);
172 if (val != 0 && (val < PAGE_SIZE || (val & (val - 1))) != 0)
175 sbi->ll_stat_blksize = val;
179 LUSTRE_RW_ATTR(stat_blocksize);
181 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
184 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
186 struct obd_statfs osfs;
191 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
195 blk_size = osfs.os_bsize >> 10;
196 result = osfs.os_blocks;
198 while (blk_size >>= 1)
201 return sprintf(buf, "%llu\n", result);
203 LUSTRE_RO_ATTR(kbytestotal);
205 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
208 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
210 struct obd_statfs osfs;
215 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
219 blk_size = osfs.os_bsize >> 10;
220 result = osfs.os_bfree;
222 while (blk_size >>= 1)
225 return sprintf(buf, "%llu\n", result);
227 LUSTRE_RO_ATTR(kbytesfree);
229 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
232 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
234 struct obd_statfs osfs;
239 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
243 blk_size = osfs.os_bsize >> 10;
244 result = osfs.os_bavail;
246 while (blk_size >>= 1)
249 return sprintf(buf, "%llu\n", result);
251 LUSTRE_RO_ATTR(kbytesavail);
253 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
256 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
258 struct obd_statfs osfs;
261 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
265 return sprintf(buf, "%llu\n", osfs.os_files);
267 LUSTRE_RO_ATTR(filestotal);
269 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
272 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
274 struct obd_statfs osfs;
277 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
281 return sprintf(buf, "%llu\n", osfs.os_ffree);
283 LUSTRE_RO_ATTR(filesfree);
285 static ssize_t client_type_show(struct kobject *kobj, struct attribute *attr,
288 return sprintf(buf, "local client\n");
290 LUSTRE_RO_ATTR(client_type);
292 LUSTRE_RW_ATTR(foreign_symlink_enable);
294 LUSTRE_RW_ATTR(foreign_symlink_prefix);
296 LUSTRE_RW_ATTR(foreign_symlink_upcall);
298 LUSTRE_WO_ATTR(foreign_symlink_upcall_info);
300 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
303 return sprintf(buf, "lustre\n");
305 LUSTRE_RO_ATTR(fstype);
307 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
310 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
313 return sprintf(buf, "%s\n", sbi->ll_sb_uuid.uuid);
315 LUSTRE_RO_ATTR(uuid);
317 static int ll_site_stats_seq_show(struct seq_file *m, void *v)
319 struct super_block *sb = m->private;
322 * See description of statistical counters in struct cl_site, and
325 return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site), m);
328 LDEBUGFS_SEQ_FOPS_RO(ll_site_stats);
330 static ssize_t max_read_ahead_mb_show(struct kobject *kobj,
331 struct attribute *attr, char *buf)
333 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
336 return scnprintf(buf, PAGE_SIZE, "%lu\n",
337 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
340 static ssize_t max_read_ahead_mb_store(struct kobject *kobj,
341 struct attribute *attr,
342 const char *buffer, size_t count)
344 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
346 u64 ra_max_mb, pages_number;
349 rc = sysfs_memparse(buffer, count, &ra_max_mb, "MiB");
353 pages_number = round_up(ra_max_mb, 1024 * 1024) >> PAGE_SHIFT;
354 CDEBUG(D_INFO, "%s: set max_read_ahead_mb=%llu (%llu pages)\n",
355 sbi->ll_fsname, PAGES_TO_MiB(pages_number), pages_number);
356 if (pages_number > cfs_totalram_pages() / 2) {
358 CERROR("%s: cannot set max_read_ahead_mb=%llu > totalram/2=%luMB\n",
359 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
360 PAGES_TO_MiB(cfs_totalram_pages() / 2));
364 spin_lock(&sbi->ll_lock);
365 sbi->ll_ra_info.ra_max_pages = pages_number;
366 spin_unlock(&sbi->ll_lock);
370 LUSTRE_RW_ATTR(max_read_ahead_mb);
372 static ssize_t max_read_ahead_per_file_mb_show(struct kobject *kobj,
373 struct attribute *attr,
376 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
379 return scnprintf(buf, PAGE_SIZE, "%lu\n",
380 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
383 static ssize_t max_read_ahead_per_file_mb_store(struct kobject *kobj,
384 struct attribute *attr,
388 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
390 u64 ra_max_file_mb, pages_number;
393 rc = sysfs_memparse(buffer, count, &ra_max_file_mb, "MiB");
397 pages_number = round_up(ra_max_file_mb, 1024 * 1024) >> PAGE_SHIFT;
398 if (pages_number > sbi->ll_ra_info.ra_max_pages) {
399 CERROR("%s: cannot set max_read_ahead_per_file_mb=%llu > max_read_ahead_mb=%lu\n",
400 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
401 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
405 spin_lock(&sbi->ll_lock);
406 sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
407 spin_unlock(&sbi->ll_lock);
411 LUSTRE_RW_ATTR(max_read_ahead_per_file_mb);
413 static ssize_t max_read_ahead_whole_mb_show(struct kobject *kobj,
414 struct attribute *attr, char *buf)
416 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
419 return scnprintf(buf, PAGE_SIZE, "%lu\n",
420 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_read_ahead_whole_pages));
423 static ssize_t max_read_ahead_whole_mb_store(struct kobject *kobj,
424 struct attribute *attr,
425 const char *buffer, size_t count)
427 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
429 u64 ra_max_whole_mb, pages_number;
432 rc = sysfs_memparse(buffer, count, &ra_max_whole_mb, "MiB");
436 pages_number = round_up(ra_max_whole_mb, 1024 * 1024) >> PAGE_SHIFT;
437 /* Cap this at the current max readahead window size, the readahead
438 * algorithm does this anyway so it's pointless to set it larger.
440 if (pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
441 CERROR("%s: cannot set max_read_ahead_whole_mb=%llu > max_read_ahead_per_file_mb=%lu\n",
442 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
443 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
448 spin_lock(&sbi->ll_lock);
449 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
450 spin_unlock(&sbi->ll_lock);
454 LUSTRE_RW_ATTR(max_read_ahead_whole_mb);
456 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
458 struct super_block *sb = m->private;
459 struct ll_sb_info *sbi = ll_s2sbi(sb);
460 struct cl_client_cache *cache = sbi->ll_cache;
461 struct ll_ra_info *ra = &sbi->ll_ra_info;
465 mutex_lock(&cache->ccc_max_cache_mb_lock);
466 max_cached_mb = PAGES_TO_MiB(cache->ccc_lru_max);
467 unused_mb = PAGES_TO_MiB(atomic_long_read(&cache->ccc_lru_left));
468 mutex_unlock(&cache->ccc_max_cache_mb_lock);
470 seq_printf(m, "users: %d\n"
471 "max_cached_mb: %ld\n"
474 "reclaim_count: %u\n"
475 "max_read_ahead_mb: %lu\n"
476 "used_read_ahead_mb: %d\n",
477 atomic_read(&cache->ccc_users),
479 max_cached_mb - unused_mb,
481 cache->ccc_lru_shrinkers,
482 PAGES_TO_MiB(ra->ra_max_pages),
483 PAGES_TO_MiB(atomic_read(&ra->ra_cur_pages)));
487 static ssize_t ll_max_cached_mb_seq_write(struct file *file,
488 const char __user *buffer,
489 size_t count, loff_t *off)
491 struct seq_file *m = file->private_data;
492 struct super_block *sb = m->private;
493 struct ll_sb_info *sbi = ll_s2sbi(sb);
494 struct cl_client_cache *cache = sbi->ll_cache;
501 char kernbuf[128], *ptr;
504 if (count >= sizeof(kernbuf))
507 if (copy_from_user(kernbuf, buffer, count))
509 kernbuf[count] = '\0';
511 ptr = lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count);
512 rc = sysfs_memparse(ptr, count, &pages_number, "MiB");
516 pages_number >>= PAGE_SHIFT;
518 if (pages_number < 0 || pages_number > cfs_totalram_pages()) {
519 CERROR("%s: can't set max cache more than %lu MB\n",
521 PAGES_TO_MiB(cfs_totalram_pages()));
524 /* Allow enough cache so clients can make well-formed RPCs */
525 pages_number = max_t(long, pages_number, PTLRPC_MAX_BRW_PAGES);
527 mutex_lock(&cache->ccc_max_cache_mb_lock);
528 diff = pages_number - cache->ccc_lru_max;
530 /* easy - add more LRU slots. */
532 atomic_long_add(diff, &cache->ccc_lru_left);
536 env = cl_env_get(&refcheck);
538 GOTO(out_unlock, rc = PTR_ERR(env));
544 /* reduce LRU budget from free slots. */
546 long lru_left_old, lru_left_new, lru_left_ret;
548 lru_left_old = atomic_long_read(&cache->ccc_lru_left);
549 if (lru_left_old == 0)
552 lru_left_new = lru_left_old > diff ?
553 lru_left_old - diff : 0;
555 atomic_long_cmpxchg(&cache->ccc_lru_left,
558 if (likely(lru_left_old == lru_left_ret)) {
559 diff -= lru_left_old - lru_left_new;
560 nrpages += lru_left_old - lru_left_new;
568 if (sbi->ll_dt_exp == NULL) { /* being initialized */
573 /* Request extra free slots to avoid them all being used
574 * by other processes before this can continue shrinking.
576 tmp = diff + min_t(long, diff, MiB_TO_PAGES(1024));
577 /* 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 cache->ccc_lru_max = pages_number;
592 atomic_long_add(nrpages, &cache->ccc_lru_left);
595 mutex_unlock(&cache->ccc_max_cache_mb_lock);
598 LDEBUGFS_SEQ_FOPS(ll_max_cached_mb);
600 static ssize_t checksums_show(struct kobject *kobj, struct attribute *attr,
603 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
606 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
609 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
610 const char *buffer, size_t count)
612 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
622 rc = kstrtobool(buffer, &val);
626 sbi->ll_flags |= LL_SBI_CHECKSUM;
628 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
631 rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
632 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
634 CWARN("Failed to set OSC checksum flags: %d\n", rc);
638 LUSTRE_RW_ATTR(checksums);
640 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
642 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
643 enum stats_track_type type)
645 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
648 if (sbi->ll_stats_track_type == type)
649 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
650 else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
651 return sprintf(buf, "0 (all)\n");
653 return sprintf(buf, "untracked\n");
656 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
657 size_t count, enum stats_track_type type)
659 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
664 rc = kstrtoul(buffer, 10, &pid);
668 sbi->ll_stats_track_id = pid;
670 sbi->ll_stats_track_type = STATS_TRACK_ALL;
672 sbi->ll_stats_track_type = type;
673 lprocfs_clear_stats(sbi->ll_stats);
677 static ssize_t stats_track_pid_show(struct kobject *kobj,
678 struct attribute *attr,
681 return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
684 static ssize_t stats_track_pid_store(struct kobject *kobj,
685 struct attribute *attr,
689 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
691 LUSTRE_RW_ATTR(stats_track_pid);
693 static ssize_t stats_track_ppid_show(struct kobject *kobj,
694 struct attribute *attr,
697 return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
700 static ssize_t stats_track_ppid_store(struct kobject *kobj,
701 struct attribute *attr,
705 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
707 LUSTRE_RW_ATTR(stats_track_ppid);
709 static ssize_t stats_track_gid_show(struct kobject *kobj,
710 struct attribute *attr,
713 return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
716 static ssize_t stats_track_gid_store(struct kobject *kobj,
717 struct attribute *attr,
721 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
723 LUSTRE_RW_ATTR(stats_track_gid);
725 static ssize_t statahead_running_max_show(struct kobject *kobj,
726 struct attribute *attr,
729 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
732 return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_sa_running_max);
735 static ssize_t statahead_running_max_store(struct kobject *kobj,
736 struct attribute *attr,
740 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
745 rc = kstrtoul(buffer, 0, &val);
749 if (val <= LL_SA_RUNNING_MAX) {
750 sbi->ll_sa_running_max = val;
754 CERROR("Bad statahead_running_max value %lu. Valid values "
755 "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
759 LUSTRE_RW_ATTR(statahead_running_max);
761 static ssize_t statahead_max_show(struct kobject *kobj,
762 struct attribute *attr,
765 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
768 return sprintf(buf, "%u\n", sbi->ll_sa_max);
771 static ssize_t statahead_max_store(struct kobject *kobj,
772 struct attribute *attr,
776 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
781 rc = kstrtoul(buffer, 0, &val);
785 if (val <= LL_SA_RPC_MAX)
786 sbi->ll_sa_max = val;
788 CERROR("Bad statahead_max value %lu. Valid values are in the range [0, %d]\n",
793 LUSTRE_RW_ATTR(statahead_max);
795 static ssize_t statahead_agl_show(struct kobject *kobj,
796 struct attribute *attr,
799 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
802 return sprintf(buf, "%u\n", sbi->ll_flags & LL_SBI_AGL_ENABLED ? 1 : 0);
805 static ssize_t statahead_agl_store(struct kobject *kobj,
806 struct attribute *attr,
810 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
815 rc = kstrtobool(buffer, &val);
820 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
822 sbi->ll_flags &= ~LL_SBI_AGL_ENABLED;
826 LUSTRE_RW_ATTR(statahead_agl);
828 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
830 struct super_block *sb = m->private;
831 struct ll_sb_info *sbi = ll_s2sbi(sb);
833 seq_printf(m, "statahead total: %u\n"
834 "statahead wrong: %u\n"
836 atomic_read(&sbi->ll_sa_total),
837 atomic_read(&sbi->ll_sa_wrong),
838 atomic_read(&sbi->ll_agl_total));
842 LDEBUGFS_SEQ_FOPS_RO(ll_statahead_stats);
844 static ssize_t lazystatfs_show(struct kobject *kobj,
845 struct attribute *attr,
848 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
851 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
854 static ssize_t lazystatfs_store(struct kobject *kobj,
855 struct attribute *attr,
859 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
864 rc = kstrtobool(buffer, &val);
869 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
871 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
875 LUSTRE_RW_ATTR(lazystatfs);
877 static ssize_t statfs_max_age_show(struct kobject *kobj, struct attribute *attr,
880 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
883 return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_statfs_max_age);
886 static ssize_t statfs_max_age_store(struct kobject *kobj,
887 struct attribute *attr, const char *buffer,
890 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
895 rc = kstrtouint(buffer, 10, &val);
898 if (val > OBD_STATFS_CACHE_MAX_AGE)
901 sbi->ll_statfs_max_age = val;
905 LUSTRE_RW_ATTR(statfs_max_age);
907 static ssize_t max_easize_show(struct kobject *kobj,
908 struct attribute *attr,
911 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
916 rc = ll_get_max_mdsize(sbi, &ealen);
920 /* Limit xattr size returned to userspace based on kernel maximum */
921 return scnprintf(buf, PAGE_SIZE, "%u\n",
922 ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
924 LUSTRE_RO_ATTR(max_easize);
927 * Get default_easize.
929 * \see client_obd::cl_default_mds_easize
931 * \param[in] m seq_file handle
932 * \param[in] v unused for single entry
934 * \retval 0 on success
935 * \retval negative negated errno on failure
937 static ssize_t default_easize_show(struct kobject *kobj,
938 struct attribute *attr,
941 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
946 rc = ll_get_default_mdsize(sbi, &ealen);
950 /* Limit xattr size returned to userspace based on kernel maximum */
951 return scnprintf(buf, PAGE_SIZE, "%u\n",
952 ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
956 * Set default_easize.
958 * Range checking on the passed value is handled by
959 * ll_set_default_mdsize().
961 * \see client_obd::cl_default_mds_easize
963 * \param[in] file proc file
964 * \param[in] buffer string passed from user space
965 * \param[in] count \a buffer length
966 * \param[in] off unused for single entry
968 * \retval positive \a count on success
969 * \retval negative negated errno on failure
971 static ssize_t default_easize_store(struct kobject *kobj,
972 struct attribute *attr,
976 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
984 rc = kstrtouint(buffer, 10, &val);
988 rc = ll_set_default_mdsize(sbi, val);
994 LUSTRE_RW_ATTR(default_easize);
996 static int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
998 const char *const str[] = LL_SBI_FLAGS;
999 struct super_block *sb = m->private;
1000 int flags = ll_s2sbi(sb)->ll_flags;
1003 while (flags != 0) {
1004 if (ARRAY_SIZE(str) <= i) {
1005 CERROR("%s: Revise array LL_SBI_FLAGS to match sbi "
1006 "flags please.\n", ll_s2sbi(sb)->ll_fsname);
1011 seq_printf(m, "%s ", str[i]);
1015 seq_printf(m, "\b\n");
1019 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
1021 static ssize_t xattr_cache_show(struct kobject *kobj,
1022 struct attribute *attr,
1025 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1028 return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
1031 static ssize_t xattr_cache_store(struct kobject *kobj,
1032 struct attribute *attr,
1036 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1041 rc = kstrtobool(buffer, &val);
1045 if (val && !(sbi->ll_flags & LL_SBI_XATTR_CACHE))
1048 sbi->ll_xattr_cache_enabled = val;
1049 sbi->ll_xattr_cache_set = 1;
1053 LUSTRE_RW_ATTR(xattr_cache);
1055 static ssize_t tiny_write_show(struct kobject *kobj,
1056 struct attribute *attr,
1059 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1062 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_TINY_WRITE));
1065 static ssize_t tiny_write_store(struct kobject *kobj,
1066 struct attribute *attr,
1070 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1075 rc = kstrtobool(buffer, &val);
1079 spin_lock(&sbi->ll_lock);
1081 sbi->ll_flags |= LL_SBI_TINY_WRITE;
1083 sbi->ll_flags &= ~LL_SBI_TINY_WRITE;
1084 spin_unlock(&sbi->ll_lock);
1088 LUSTRE_RW_ATTR(tiny_write);
1090 static ssize_t max_read_ahead_async_active_show(struct kobject *kobj,
1091 struct attribute *attr,
1094 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1097 return scnprintf(buf, PAGE_SIZE, "%u\n",
1098 sbi->ll_ra_info.ra_async_max_active);
1101 static ssize_t max_read_ahead_async_active_store(struct kobject *kobj,
1102 struct attribute *attr,
1106 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1111 rc = kstrtouint(buffer, 10, &val);
1116 * It doesn't make any sense to make it exceed what
1117 * workqueue could acutally support. This can easily
1118 * over subscripe the cores but Lustre internally
1119 * throttles to avoid those impacts.
1121 if (val > WQ_UNBOUND_MAX_ACTIVE) {
1122 CERROR("%s: cannot set max_read_ahead_async_active=%u larger than %u\n",
1123 sbi->ll_fsname, val, WQ_UNBOUND_MAX_ACTIVE);
1127 spin_lock(&sbi->ll_lock);
1128 sbi->ll_ra_info.ra_async_max_active = val;
1129 spin_unlock(&sbi->ll_lock);
1133 LUSTRE_RW_ATTR(max_read_ahead_async_active);
1135 static ssize_t read_ahead_async_file_threshold_mb_show(struct kobject *kobj,
1136 struct attribute *attr,
1139 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1142 return scnprintf(buf, PAGE_SIZE, "%lu\n", PAGES_TO_MiB(
1143 sbi->ll_ra_info.ra_async_pages_per_file_threshold));
1147 read_ahead_async_file_threshold_mb_store(struct kobject *kobj,
1148 struct attribute *attr,
1149 const char *buffer, size_t count)
1151 unsigned long pages_number;
1152 unsigned long max_ra_per_file;
1153 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1157 rc = kstrtoul(buffer, 10, &pages_number);
1161 pages_number = MiB_TO_PAGES(pages_number);
1162 max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1163 if (pages_number < 0 || pages_number > max_ra_per_file) {
1164 CERROR("%s: can't set read_ahead_async_file_threshold_mb=%lu > "
1165 "max_read_readahead_per_file_mb=%lu\n", sbi->ll_fsname,
1166 PAGES_TO_MiB(pages_number),
1167 PAGES_TO_MiB(max_ra_per_file));
1170 sbi->ll_ra_info.ra_async_pages_per_file_threshold = pages_number;
1174 LUSTRE_RW_ATTR(read_ahead_async_file_threshold_mb);
1176 static ssize_t read_ahead_range_kb_show(struct kobject *kobj,
1177 struct attribute *attr,char *buf)
1179 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1182 return snprintf(buf, PAGE_SIZE, "%lu\n",
1183 sbi->ll_ra_info.ra_range_pages << (PAGE_SHIFT - 10));
1187 read_ahead_range_kb_store(struct kobject *kobj,
1188 struct attribute *attr,
1189 const char *buffer, size_t count)
1191 unsigned long pages_number;
1192 unsigned long max_ra_per_file;
1194 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1198 rc = sysfs_memparse(buffer, count, &val, "KiB");
1202 pages_number = val >> PAGE_SHIFT;
1203 /* Disable mmap range read */
1204 if (pages_number == 0)
1207 max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1208 if (pages_number > max_ra_per_file ||
1209 pages_number < RA_MIN_MMAP_RANGE_PAGES)
1213 spin_lock(&sbi->ll_lock);
1214 sbi->ll_ra_info.ra_range_pages = pages_number;
1215 spin_unlock(&sbi->ll_lock);
1219 LUSTRE_RW_ATTR(read_ahead_range_kb);
1221 static ssize_t fast_read_show(struct kobject *kobj,
1222 struct attribute *attr,
1225 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1228 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_FAST_READ));
1231 static ssize_t fast_read_store(struct kobject *kobj,
1232 struct attribute *attr,
1236 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1241 rc = kstrtobool(buffer, &val);
1245 spin_lock(&sbi->ll_lock);
1247 sbi->ll_flags |= LL_SBI_FAST_READ;
1249 sbi->ll_flags &= ~LL_SBI_FAST_READ;
1250 spin_unlock(&sbi->ll_lock);
1254 LUSTRE_RW_ATTR(fast_read);
1256 static ssize_t file_heat_show(struct kobject *kobj,
1257 struct attribute *attr,
1260 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1263 return scnprintf(buf, PAGE_SIZE, "%u\n",
1264 !!(sbi->ll_flags & LL_SBI_FILE_HEAT));
1267 static ssize_t file_heat_store(struct kobject *kobj,
1268 struct attribute *attr,
1272 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1277 rc = kstrtobool(buffer, &val);
1281 spin_lock(&sbi->ll_lock);
1283 sbi->ll_flags |= LL_SBI_FILE_HEAT;
1285 sbi->ll_flags &= ~LL_SBI_FILE_HEAT;
1286 spin_unlock(&sbi->ll_lock);
1290 LUSTRE_RW_ATTR(file_heat);
1292 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1293 struct attribute *attr,
1296 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1299 return scnprintf(buf, PAGE_SIZE, "%u\n",
1300 (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1303 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1304 struct attribute *attr,
1308 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1313 rc = kstrtoul(buffer, 10, &val);
1317 if (val < 0 || val > 100)
1320 sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1324 LUSTRE_RW_ATTR(heat_decay_percentage);
1326 static ssize_t heat_period_second_show(struct kobject *kobj,
1327 struct attribute *attr,
1330 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1333 return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1336 static ssize_t heat_period_second_store(struct kobject *kobj,
1337 struct attribute *attr,
1341 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1346 rc = kstrtoul(buffer, 10, &val);
1353 sbi->ll_heat_period_second = val;
1357 LUSTRE_RW_ATTR(heat_period_second);
1359 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1361 struct super_block *sb = m->private;
1362 struct ll_sb_info *sbi = ll_s2sbi(sb);
1363 struct cl_client_cache *cache = sbi->ll_cache;
1367 pages = atomic_long_read(&cache->ccc_unstable_nr);
1368 mb = (pages * PAGE_SIZE) >> 20;
1370 seq_printf(m, "unstable_check: %8d\n"
1371 "unstable_pages: %12ld\n"
1372 "unstable_mb: %8d\n",
1373 cache->ccc_unstable_check, pages, mb);
1377 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1378 const char __user *buffer,
1379 size_t count, loff_t *unused)
1381 struct seq_file *seq = file->private_data;
1382 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1389 if (count >= sizeof(kernbuf))
1392 if (copy_from_user(kernbuf, buffer, count))
1396 buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1398 rc = kstrtobool_from_user(buffer, count, &val);
1402 /* borrow lru lock to set the value */
1403 spin_lock(&sbi->ll_cache->ccc_lru_lock);
1404 sbi->ll_cache->ccc_unstable_check = val;
1405 spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1410 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1412 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1414 struct super_block *sb = m->private;
1415 struct ll_sb_info *sbi = ll_s2sbi(sb);
1416 struct root_squash_info *squash = &sbi->ll_squash;
1418 seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1422 static ssize_t ll_root_squash_seq_write(struct file *file,
1423 const char __user *buffer,
1424 size_t count, loff_t *off)
1426 struct seq_file *m = file->private_data;
1427 struct super_block *sb = m->private;
1428 struct ll_sb_info *sbi = ll_s2sbi(sb);
1429 struct root_squash_info *squash = &sbi->ll_squash;
1431 return lprocfs_wr_root_squash(buffer, count, squash, sbi->ll_fsname);
1434 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1436 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1438 struct super_block *sb = m->private;
1439 struct ll_sb_info *sbi = ll_s2sbi(sb);
1440 struct root_squash_info *squash = &sbi->ll_squash;
1443 spin_lock(&squash->rsi_lock);
1444 if (!list_empty(&squash->rsi_nosquash_nids)) {
1445 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1446 &squash->rsi_nosquash_nids);
1450 seq_puts(m, "NONE\n");
1452 spin_unlock(&squash->rsi_lock);
1457 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1458 const char __user *buffer,
1459 size_t count, loff_t *off)
1461 struct seq_file *m = file->private_data;
1462 struct super_block *sb = m->private;
1463 struct ll_sb_info *sbi = ll_s2sbi(sb);
1464 struct root_squash_info *squash = &sbi->ll_squash;
1467 rc = lprocfs_wr_nosquash_nids(buffer, count, squash, sbi->ll_fsname);
1471 ll_compute_rootsquash_state(sbi);
1476 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1478 static int ll_pcc_seq_show(struct seq_file *m, void *v)
1480 struct super_block *sb = m->private;
1481 struct ll_sb_info *sbi = ll_s2sbi(sb);
1483 return pcc_super_dump(&sbi->ll_pcc_super, m);
1486 static ssize_t ll_pcc_seq_write(struct file *file, const char __user *buffer,
1487 size_t count, loff_t *off)
1489 struct seq_file *m = file->private_data;
1490 struct super_block *sb = m->private;
1491 struct ll_sb_info *sbi = ll_s2sbi(sb);
1495 if (count >= LPROCFS_WR_PCC_MAX_CMD)
1498 if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_PCC))
1501 OBD_ALLOC(kernbuf, count + 1);
1502 if (kernbuf == NULL)
1505 if (copy_from_user(kernbuf, buffer, count))
1506 GOTO(out_free_kernbuff, rc = -EFAULT);
1508 rc = pcc_cmd_handle(kernbuf, count, &sbi->ll_pcc_super);
1510 OBD_FREE(kernbuf, count + 1);
1511 return rc ? rc : count;
1513 LDEBUGFS_SEQ_FOPS(ll_pcc);
1515 struct ldebugfs_vars lprocfs_llite_obd_vars[] = {
1517 .fops = &ll_site_stats_fops },
1518 { .name = "max_cached_mb",
1519 .fops = &ll_max_cached_mb_fops },
1520 { .name = "statahead_stats",
1521 .fops = &ll_statahead_stats_fops },
1522 { .name = "unstable_stats",
1523 .fops = &ll_unstable_stats_fops },
1524 { .name = "sbi_flags",
1525 .fops = &ll_sbi_flags_fops },
1526 { .name = "root_squash",
1527 .fops = &ll_root_squash_fops },
1528 { .name = "nosquash_nids",
1529 .fops = &ll_nosquash_nids_fops },
1531 .fops = &ll_pcc_fops, },
1535 #define MAX_STRING_SIZE 128
1537 static struct attribute *llite_attrs[] = {
1538 &lustre_attr_blocksize.attr,
1539 &lustre_attr_stat_blocksize.attr,
1540 &lustre_attr_kbytestotal.attr,
1541 &lustre_attr_kbytesfree.attr,
1542 &lustre_attr_kbytesavail.attr,
1543 &lustre_attr_filestotal.attr,
1544 &lustre_attr_filesfree.attr,
1545 &lustre_attr_client_type.attr,
1546 &lustre_attr_foreign_symlink_enable.attr,
1547 &lustre_attr_foreign_symlink_prefix.attr,
1548 &lustre_attr_foreign_symlink_upcall.attr,
1549 &lustre_attr_foreign_symlink_upcall_info.attr,
1550 &lustre_attr_fstype.attr,
1551 &lustre_attr_uuid.attr,
1552 &lustre_attr_checksums.attr,
1553 &lustre_attr_checksum_pages.attr,
1554 &lustre_attr_max_read_ahead_mb.attr,
1555 &lustre_attr_max_read_ahead_per_file_mb.attr,
1556 &lustre_attr_max_read_ahead_whole_mb.attr,
1557 &lustre_attr_max_read_ahead_async_active.attr,
1558 &lustre_attr_read_ahead_async_file_threshold_mb.attr,
1559 &lustre_attr_read_ahead_range_kb.attr,
1560 &lustre_attr_stats_track_pid.attr,
1561 &lustre_attr_stats_track_ppid.attr,
1562 &lustre_attr_stats_track_gid.attr,
1563 &lustre_attr_statahead_running_max.attr,
1564 &lustre_attr_statahead_max.attr,
1565 &lustre_attr_statahead_agl.attr,
1566 &lustre_attr_lazystatfs.attr,
1567 &lustre_attr_statfs_max_age.attr,
1568 &lustre_attr_max_easize.attr,
1569 &lustre_attr_default_easize.attr,
1570 &lustre_attr_xattr_cache.attr,
1571 &lustre_attr_fast_read.attr,
1572 &lustre_attr_tiny_write.attr,
1573 &lustre_attr_file_heat.attr,
1574 &lustre_attr_heat_decay_percentage.attr,
1575 &lustre_attr_heat_period_second.attr,
1579 static void sbi_kobj_release(struct kobject *kobj)
1581 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1583 complete(&sbi->ll_kobj_unregister);
1586 static struct kobj_type sbi_ktype = {
1587 .default_attrs = llite_attrs,
1588 .sysfs_ops = &lustre_sysfs_ops,
1589 .release = sbi_kobj_release,
1592 static const struct llite_file_opcode {
1596 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1597 /* file operation */
1598 { LPROC_LL_READ_BYTES, LPROCFS_TYPE_BYTES_FULL, "read_bytes" },
1599 { LPROC_LL_WRITE_BYTES, LPROCFS_TYPE_BYTES_FULL, "write_bytes" },
1600 { LPROC_LL_READ, LPROCFS_TYPE_LATENCY, "read" },
1601 { LPROC_LL_WRITE, LPROCFS_TYPE_LATENCY, "write" },
1602 { LPROC_LL_IOCTL, LPROCFS_TYPE_REQS, "ioctl" },
1603 { LPROC_LL_OPEN, LPROCFS_TYPE_LATENCY, "open" },
1604 { LPROC_LL_RELEASE, LPROCFS_TYPE_LATENCY, "close" },
1605 { LPROC_LL_MMAP, LPROCFS_TYPE_LATENCY, "mmap" },
1606 { LPROC_LL_FAULT, LPROCFS_TYPE_LATENCY, "page_fault" },
1607 { LPROC_LL_MKWRITE, LPROCFS_TYPE_LATENCY, "page_mkwrite" },
1608 { LPROC_LL_LLSEEK, LPROCFS_TYPE_LATENCY, "seek" },
1609 { LPROC_LL_FSYNC, LPROCFS_TYPE_LATENCY, "fsync" },
1610 { LPROC_LL_READDIR, LPROCFS_TYPE_LATENCY, "readdir" },
1611 /* inode operation */
1612 { LPROC_LL_SETATTR, LPROCFS_TYPE_LATENCY, "setattr" },
1613 { LPROC_LL_TRUNC, LPROCFS_TYPE_LATENCY, "truncate" },
1614 { LPROC_LL_FLOCK, LPROCFS_TYPE_LATENCY, "flock" },
1615 { LPROC_LL_GETATTR, LPROCFS_TYPE_LATENCY, "getattr" },
1616 { LPROC_LL_FALLOCATE, LPROCFS_TYPE_LATENCY, "fallocate"},
1617 /* dir inode operation */
1618 { LPROC_LL_CREATE, LPROCFS_TYPE_LATENCY, "create" },
1619 { LPROC_LL_LINK, LPROCFS_TYPE_LATENCY, "link" },
1620 { LPROC_LL_UNLINK, LPROCFS_TYPE_LATENCY, "unlink" },
1621 { LPROC_LL_SYMLINK, LPROCFS_TYPE_LATENCY, "symlink" },
1622 { LPROC_LL_MKDIR, LPROCFS_TYPE_LATENCY, "mkdir" },
1623 { LPROC_LL_RMDIR, LPROCFS_TYPE_LATENCY, "rmdir" },
1624 { LPROC_LL_MKNOD, LPROCFS_TYPE_LATENCY, "mknod" },
1625 { LPROC_LL_RENAME, LPROCFS_TYPE_LATENCY, "rename" },
1626 /* special inode operation */
1627 { LPROC_LL_STATFS, LPROCFS_TYPE_LATENCY, "statfs" },
1628 { LPROC_LL_SETXATTR, LPROCFS_TYPE_LATENCY, "setxattr" },
1629 { LPROC_LL_GETXATTR, LPROCFS_TYPE_LATENCY, "getxattr" },
1630 { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REQS, "getxattr_hits" },
1631 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_LATENCY, "listxattr" },
1632 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_LATENCY, "removexattr" },
1633 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_LATENCY, "inode_permission" },
1636 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count)
1641 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1642 lprocfs_counter_add(sbi->ll_stats, op, count);
1643 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1644 sbi->ll_stats_track_id == current->pid)
1645 lprocfs_counter_add(sbi->ll_stats, op, count);
1646 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1647 sbi->ll_stats_track_id == current->parent->pid)
1648 lprocfs_counter_add(sbi->ll_stats, op, count);
1649 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1650 sbi->ll_stats_track_id ==
1651 from_kgid(&init_user_ns, current_gid()))
1652 lprocfs_counter_add(sbi->ll_stats, op, count);
1654 EXPORT_SYMBOL(ll_stats_ops_tally);
1656 static const char *const ra_stat_string[] = {
1657 [RA_STAT_HIT] = "hits",
1658 [RA_STAT_MISS] = "misses",
1659 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1660 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1661 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1662 [RA_STAT_FAILED_MATCH] = "failed lock match",
1663 [RA_STAT_DISCARDED] = "read but discarded",
1664 [RA_STAT_ZERO_LEN] = "zero length file",
1665 [RA_STAT_ZERO_WINDOW] = "zero size window",
1666 [RA_STAT_EOF] = "read-ahead to EOF",
1667 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1668 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1669 [RA_STAT_FAILED_REACH_END] = "failed to reach end",
1670 [RA_STAT_ASYNC] = "async readahead",
1671 [RA_STAT_FAILED_FAST_READ] = "failed to fast read",
1672 [RA_STAT_MMAP_RANGE_READ] = "mmap range read",
1675 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1677 struct lustre_sb_info *lsi = s2lsi(sb);
1678 struct ll_sb_info *sbi = ll_s2sbi(sb);
1684 if (IS_ERR_OR_NULL(llite_root))
1687 sbi->ll_debugfs_entry = debugfs_create_dir(name, llite_root);
1688 ldebugfs_add_vars(sbi->ll_debugfs_entry, lprocfs_llite_obd_vars, sb);
1690 debugfs_create_file("dump_page_cache", 0444, sbi->ll_debugfs_entry, sbi,
1691 &vvp_dump_pgcache_file_ops);
1693 debugfs_create_file("extents_stats", 0644, sbi->ll_debugfs_entry, sbi,
1694 &ll_rw_extents_stats_fops);
1696 debugfs_create_file("extents_stats_per_process", 0644,
1697 sbi->ll_debugfs_entry, sbi,
1698 &ll_rw_extents_stats_pp_fops);
1700 debugfs_create_file("offset_stats", 0644, sbi->ll_debugfs_entry, sbi,
1701 &ll_rw_offset_stats_fops);
1703 /* File operations stats */
1704 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1705 LPROCFS_STATS_FLAG_NONE);
1706 if (sbi->ll_stats == NULL)
1707 GOTO(out_debugfs, err = -ENOMEM);
1709 /* do counter init */
1710 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1711 u32 type = llite_opcode_table[id].type;
1712 void *ptr = "unknown";
1714 if (type & LPROCFS_TYPE_REQS)
1716 else if (type & LPROCFS_TYPE_BYTES)
1718 else if (type & LPROCFS_TYPE_USEC)
1720 lprocfs_counter_init(sbi->ll_stats,
1721 llite_opcode_table[id].opcode, type,
1722 llite_opcode_table[id].opname, ptr);
1725 debugfs_create_file("stats", 0644, sbi->ll_debugfs_entry,
1726 sbi->ll_stats, &ldebugfs_stats_seq_fops);
1728 sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1729 LPROCFS_STATS_FLAG_NONE);
1730 if (sbi->ll_ra_stats == NULL)
1731 GOTO(out_stats, err = -ENOMEM);
1733 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1734 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1735 ra_stat_string[id], "pages");
1737 debugfs_create_file("read_ahead_stats", 0644, sbi->ll_debugfs_entry,
1738 sbi->ll_ra_stats, &ldebugfs_stats_seq_fops);
1741 /* Yes we also register sysfs mount kset here as well */
1742 sbi->ll_kset.kobj.parent = llite_kobj;
1743 sbi->ll_kset.kobj.ktype = &sbi_ktype;
1744 init_completion(&sbi->ll_kobj_unregister);
1745 err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1747 GOTO(out_ra_stats, err);
1749 err = kset_register(&sbi->ll_kset);
1751 GOTO(out_ra_stats, err);
1753 lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1757 lprocfs_free_stats(&sbi->ll_ra_stats);
1759 lprocfs_free_stats(&sbi->ll_stats);
1761 debugfs_remove_recursive(sbi->ll_debugfs_entry);
1766 void ll_debugfs_unregister_super(struct super_block *sb)
1768 struct lustre_sb_info *lsi = s2lsi(sb);
1769 struct ll_sb_info *sbi = ll_s2sbi(sb);
1771 debugfs_remove_recursive(sbi->ll_debugfs_entry);
1774 sysfs_remove_link(&sbi->ll_kset.kobj,
1775 sbi->ll_dt_obd->obd_type->typ_name);
1778 sysfs_remove_link(&sbi->ll_kset.kobj,
1779 sbi->ll_md_obd->obd_type->typ_name);
1781 kobject_put(lsi->lsi_kobj);
1783 kset_unregister(&sbi->ll_kset);
1784 wait_for_completion(&sbi->ll_kobj_unregister);
1786 lprocfs_free_stats(&sbi->ll_ra_stats);
1787 lprocfs_free_stats(&sbi->ll_stats);
1789 #undef MAX_STRING_SIZE
1791 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1792 struct seq_file *seq, int which)
1794 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1795 unsigned long start, end, r, w;
1796 char *unitp = "KMGTPEZY";
1798 struct per_process_info *pp_info = &io_extents->pp_extents[which];
1804 for(i = 0; i < LL_HIST_MAX; i++) {
1805 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1806 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1809 for(i = 0; i < LL_HIST_MAX; i++) {
1810 r = pp_info->pp_r_hist.oh_buckets[i];
1811 w = pp_info->pp_w_hist.oh_buckets[i];
1814 end = 1 << (i + LL_HIST_START - units);
1815 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u | "
1816 "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1817 (i == LL_HIST_MAX - 1) ? '+' : ' ',
1818 r, pct(r, read_tot), pct(read_cum, read_tot),
1819 w, pct(w, write_tot), pct(write_cum, write_tot));
1821 if (start == (1 << 10)) {
1826 if (read_cum == read_tot && write_cum == write_tot)
1831 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1833 struct timespec64 now;
1834 struct ll_sb_info *sbi = seq->private;
1835 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1838 ktime_get_real_ts64(&now);
1840 if (!sbi->ll_rw_stats_on) {
1841 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1844 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1845 (s64)now.tv_sec, now.tv_nsec);
1846 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1847 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1848 "extents", "calls", "%", "cum%",
1849 "calls", "%", "cum%");
1850 spin_lock(&sbi->ll_pp_extent_lock);
1851 for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1852 if (io_extents->pp_extents[k].pid != 0) {
1853 seq_printf(seq, "\nPID: %d\n",
1854 io_extents->pp_extents[k].pid);
1855 ll_display_extents_info(io_extents, seq, k);
1858 spin_unlock(&sbi->ll_pp_extent_lock);
1862 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1863 const char __user *buf,
1867 struct seq_file *seq = file->private_data;
1868 struct ll_sb_info *sbi = seq->private;
1869 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1876 value = ll_stats_pid_write(buf, len);
1879 sbi->ll_rw_stats_on = 0;
1881 sbi->ll_rw_stats_on = 1;
1883 spin_lock(&sbi->ll_pp_extent_lock);
1884 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1885 io_extents->pp_extents[i].pid = 0;
1886 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1887 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1889 spin_unlock(&sbi->ll_pp_extent_lock);
1893 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1895 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1897 struct timespec64 now;
1898 struct ll_sb_info *sbi = seq->private;
1899 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1901 ktime_get_real_ts64(&now);
1903 if (!sbi->ll_rw_stats_on) {
1904 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1907 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1908 (s64)now.tv_sec, now.tv_nsec);
1910 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1911 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1912 "extents", "calls", "%", "cum%",
1913 "calls", "%", "cum%");
1914 spin_lock(&sbi->ll_lock);
1915 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1916 spin_unlock(&sbi->ll_lock);
1921 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1922 const char __user *buf,
1923 size_t len, loff_t *off)
1925 struct seq_file *seq = file->private_data;
1926 struct ll_sb_info *sbi = seq->private;
1927 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1934 value = ll_stats_pid_write(buf, len);
1937 sbi->ll_rw_stats_on = 0;
1939 sbi->ll_rw_stats_on = 1;
1941 spin_lock(&sbi->ll_pp_extent_lock);
1942 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1943 io_extents->pp_extents[i].pid = 0;
1944 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1945 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1947 spin_unlock(&sbi->ll_pp_extent_lock);
1952 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1954 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1955 struct ll_file_data *file, loff_t pos,
1956 size_t count, int rw)
1959 struct ll_rw_process_info *process;
1960 struct ll_rw_process_info *offset;
1961 int *off_count = &sbi->ll_rw_offset_entry_count;
1962 int *process_count = &sbi->ll_offset_process_count;
1963 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1965 if(!sbi->ll_rw_stats_on)
1967 process = sbi->ll_rw_process_info;
1968 offset = sbi->ll_rw_offset_info;
1970 spin_lock(&sbi->ll_pp_extent_lock);
1971 /* Extent statistics */
1972 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1973 if(io_extents->pp_extents[i].pid == pid) {
1981 sbi->ll_extent_process_count =
1982 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1983 cur = sbi->ll_extent_process_count;
1984 io_extents->pp_extents[cur].pid = pid;
1985 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1986 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1989 for (i = 0; (count >= 1 << (LL_HIST_START + i)) &&
1990 (i < (LL_HIST_MAX - 1)); i++);
1992 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1993 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1995 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1996 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1998 spin_unlock(&sbi->ll_pp_extent_lock);
2000 spin_lock(&sbi->ll_process_lock);
2001 /* Offset statistics */
2002 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2003 if (process[i].rw_pid == pid) {
2004 if (process[i].rw_last_file != file) {
2005 process[i].rw_range_start = pos;
2006 process[i].rw_last_file_pos = pos + count;
2007 process[i].rw_smallest_extent = count;
2008 process[i].rw_largest_extent = count;
2009 process[i].rw_offset = 0;
2010 process[i].rw_last_file = file;
2011 spin_unlock(&sbi->ll_process_lock);
2014 if (process[i].rw_last_file_pos != pos) {
2016 (*off_count + 1) % LL_OFFSET_HIST_MAX;
2017 offset[*off_count].rw_op = process[i].rw_op;
2018 offset[*off_count].rw_pid = pid;
2019 offset[*off_count].rw_range_start =
2020 process[i].rw_range_start;
2021 offset[*off_count].rw_range_end =
2022 process[i].rw_last_file_pos;
2023 offset[*off_count].rw_smallest_extent =
2024 process[i].rw_smallest_extent;
2025 offset[*off_count].rw_largest_extent =
2026 process[i].rw_largest_extent;
2027 offset[*off_count].rw_offset =
2028 process[i].rw_offset;
2029 process[i].rw_op = rw;
2030 process[i].rw_range_start = pos;
2031 process[i].rw_smallest_extent = count;
2032 process[i].rw_largest_extent = count;
2033 process[i].rw_offset = pos -
2034 process[i].rw_last_file_pos;
2036 if(process[i].rw_smallest_extent > count)
2037 process[i].rw_smallest_extent = count;
2038 if(process[i].rw_largest_extent < count)
2039 process[i].rw_largest_extent = count;
2040 process[i].rw_last_file_pos = pos + count;
2041 spin_unlock(&sbi->ll_process_lock);
2045 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
2046 process[*process_count].rw_pid = pid;
2047 process[*process_count].rw_op = rw;
2048 process[*process_count].rw_range_start = pos;
2049 process[*process_count].rw_last_file_pos = pos + count;
2050 process[*process_count].rw_smallest_extent = count;
2051 process[*process_count].rw_largest_extent = count;
2052 process[*process_count].rw_offset = 0;
2053 process[*process_count].rw_last_file = file;
2054 spin_unlock(&sbi->ll_process_lock);
2057 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
2059 struct timespec64 now;
2060 struct ll_sb_info *sbi = seq->private;
2061 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
2062 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
2065 ktime_get_real_ts64(&now);
2067 if (!sbi->ll_rw_stats_on) {
2068 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2071 spin_lock(&sbi->ll_process_lock);
2073 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
2074 (s64)now.tv_sec, now.tv_nsec);
2075 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
2076 "R/W", "PID", "RANGE START", "RANGE END",
2077 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
2079 /* We stored the discontiguous offsets here; print them first */
2080 for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
2081 if (offset[i].rw_pid != 0)
2083 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2084 offset[i].rw_op == READ ? 'R' : 'W',
2086 offset[i].rw_range_start,
2087 offset[i].rw_range_end,
2088 (unsigned long)offset[i].rw_smallest_extent,
2089 (unsigned long)offset[i].rw_largest_extent,
2090 offset[i].rw_offset);
2093 /* Then print the current offsets for each process */
2094 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2095 if (process[i].rw_pid != 0)
2097 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2098 process[i].rw_op == READ ? 'R' : 'W',
2100 process[i].rw_range_start,
2101 process[i].rw_last_file_pos,
2102 (unsigned long)process[i].rw_smallest_extent,
2103 (unsigned long)process[i].rw_largest_extent,
2104 process[i].rw_offset);
2106 spin_unlock(&sbi->ll_process_lock);
2111 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
2112 const char __user *buf,
2113 size_t len, loff_t *off)
2115 struct seq_file *seq = file->private_data;
2116 struct ll_sb_info *sbi = seq->private;
2117 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
2118 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
2124 value = ll_stats_pid_write(buf, len);
2127 sbi->ll_rw_stats_on = 0;
2129 sbi->ll_rw_stats_on = 1;
2131 spin_lock(&sbi->ll_process_lock);
2132 sbi->ll_offset_process_count = 0;
2133 sbi->ll_rw_offset_entry_count = 0;
2134 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
2135 LL_PROCESS_HIST_MAX);
2136 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
2137 LL_OFFSET_HIST_MAX);
2138 spin_unlock(&sbi->ll_process_lock);
2143 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);