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/
31 #define DEBUG_SUBSYSTEM S_LLITE
33 #include <linux/version.h>
34 #include <linux/user_namespace.h>
35 #include <linux/uidgid.h>
37 #include <uapi/linux/lustre/lustre_param.h>
38 #include <lprocfs_status.h>
39 #include <obd_support.h>
41 #include "llite_internal.h"
42 #include "lprocfs_status.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 refcount_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;
502 char kernbuf[128], *ptr;
503 bool percent = false;
506 if (count >= sizeof(kernbuf))
509 if (copy_from_user(kernbuf, buffer, count))
512 if (count > 0 && kernbuf[count - 1] == '%') {
514 /* strip off the % */
515 kernbuf[count - 1] = '\0';
517 kernbuf[count] = '\0';
520 ptr = lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count);
522 rc = sysfs_memparse(ptr, count, &value, "B");
524 rc = sysfs_memparse(ptr, count, &value, "MiB");
529 pages_number = cfs_totalram_pages() * value / 100;
531 pages_number = value >> PAGE_SHIFT;
534 if (pages_number < 0 || pages_number > cfs_totalram_pages()) {
535 CERROR("%s: can't set max cache more than %lu MB\n",
537 PAGES_TO_MiB(cfs_totalram_pages()));
540 /* Allow enough cache so clients can make well-formed RPCs */
541 pages_number = max_t(long, pages_number, PTLRPC_MAX_BRW_PAGES);
543 mutex_lock(&cache->ccc_max_cache_mb_lock);
544 diff = pages_number - cache->ccc_lru_max;
546 /* easy - add more LRU slots. */
548 atomic_long_add(diff, &cache->ccc_lru_left);
552 env = cl_env_get(&refcheck);
554 GOTO(out_unlock, rc = PTR_ERR(env));
560 /* reduce LRU budget from free slots. */
562 long lru_left_old, lru_left_new, lru_left_ret;
564 lru_left_old = atomic_long_read(&cache->ccc_lru_left);
565 if (lru_left_old == 0)
568 lru_left_new = lru_left_old > diff ?
569 lru_left_old - diff : 0;
571 atomic_long_cmpxchg(&cache->ccc_lru_left,
574 if (likely(lru_left_old == lru_left_ret)) {
575 diff -= lru_left_old - lru_left_new;
576 nrpages += lru_left_old - lru_left_new;
584 if (sbi->ll_dt_exp == NULL) { /* being initialized */
589 /* Request extra free slots to avoid them all being used
590 * by other processes before this can continue shrinking.
592 tmp = diff + min_t(long, diff, MiB_TO_PAGES(1024));
593 /* difficult - have to ask OSCs to drop LRU slots. */
594 rc = obd_set_info_async(env, sbi->ll_dt_exp,
595 sizeof(KEY_CACHE_LRU_SHRINK),
596 KEY_CACHE_LRU_SHRINK,
597 sizeof(tmp), &tmp, NULL);
601 cl_env_put(env, &refcheck);
605 cache->ccc_lru_max = pages_number;
608 atomic_long_add(nrpages, &cache->ccc_lru_left);
611 mutex_unlock(&cache->ccc_max_cache_mb_lock);
614 LDEBUGFS_SEQ_FOPS(ll_max_cached_mb);
616 static ssize_t checksums_show(struct kobject *kobj, struct attribute *attr,
619 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
622 return scnprintf(buf, PAGE_SIZE, "%u\n",
623 test_bit(LL_SBI_CHECKSUM, sbi->ll_flags));
626 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
627 const char *buffer, size_t count)
629 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
639 rc = kstrtobool(buffer, &val);
643 set_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
645 clear_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
648 rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
649 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
651 CWARN("Failed to set OSC checksum flags: %d\n", rc);
655 LUSTRE_RW_ATTR(checksums);
657 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
659 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
660 enum stats_track_type type)
662 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
665 if (sbi->ll_stats_track_type == type)
666 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
667 else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
668 return sprintf(buf, "0 (all)\n");
670 return sprintf(buf, "untracked\n");
673 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
674 size_t count, enum stats_track_type type)
676 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
681 rc = kstrtoul(buffer, 10, &pid);
685 sbi->ll_stats_track_id = pid;
687 sbi->ll_stats_track_type = STATS_TRACK_ALL;
689 sbi->ll_stats_track_type = type;
690 lprocfs_stats_clear(sbi->ll_stats);
694 static ssize_t stats_track_pid_show(struct kobject *kobj,
695 struct attribute *attr,
698 return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
701 static ssize_t stats_track_pid_store(struct kobject *kobj,
702 struct attribute *attr,
706 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
708 LUSTRE_RW_ATTR(stats_track_pid);
710 static ssize_t stats_track_ppid_show(struct kobject *kobj,
711 struct attribute *attr,
714 return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
717 static ssize_t stats_track_ppid_store(struct kobject *kobj,
718 struct attribute *attr,
722 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
724 LUSTRE_RW_ATTR(stats_track_ppid);
726 static ssize_t stats_track_gid_show(struct kobject *kobj,
727 struct attribute *attr,
730 return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
733 static ssize_t stats_track_gid_store(struct kobject *kobj,
734 struct attribute *attr,
738 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
740 LUSTRE_RW_ATTR(stats_track_gid);
742 static ssize_t statahead_running_max_show(struct kobject *kobj,
743 struct attribute *attr,
746 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
749 return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_sa_running_max);
752 static ssize_t statahead_running_max_store(struct kobject *kobj,
753 struct attribute *attr,
757 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
762 rc = kstrtoul(buffer, 0, &val);
766 if (val <= LL_SA_RUNNING_MAX) {
767 sbi->ll_sa_running_max = val;
771 CERROR("Bad statahead_running_max value %lu. Valid values "
772 "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
776 LUSTRE_RW_ATTR(statahead_running_max);
778 static ssize_t statahead_batch_max_show(struct kobject *kobj,
779 struct attribute *attr,
782 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
785 return snprintf(buf, 16, "%u\n", sbi->ll_sa_batch_max);
788 static ssize_t statahead_batch_max_store(struct kobject *kobj,
789 struct attribute *attr,
793 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
798 rc = kstrtoul(buffer, 0, &val);
802 if (val > LL_SA_BATCH_MAX) {
803 CWARN("%s: statahead_batch_max value %lu limited to maximum %d\n",
804 sbi->ll_fsname, val, LL_SA_BATCH_MAX);
805 val = LL_SA_BATCH_MAX;
808 sbi->ll_sa_batch_max = val;
811 LUSTRE_RW_ATTR(statahead_batch_max);
813 static ssize_t statahead_max_show(struct kobject *kobj,
814 struct attribute *attr,
817 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
820 return sprintf(buf, "%u\n", sbi->ll_sa_max);
823 static ssize_t statahead_max_store(struct kobject *kobj,
824 struct attribute *attr,
828 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
833 rc = kstrtoul(buffer, 0, &val);
837 if (val > LL_SA_RPC_MAX) {
838 CWARN("%s: statahead_max value %lu limited to maximum %d\n",
839 sbi->ll_fsname, val, LL_SA_RPC_MAX);
843 sbi->ll_sa_max = val;
846 LUSTRE_RW_ATTR(statahead_max);
848 static ssize_t statahead_agl_show(struct kobject *kobj,
849 struct attribute *attr,
852 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
855 return scnprintf(buf, PAGE_SIZE, "%u\n",
856 test_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags));
859 static ssize_t statahead_agl_store(struct kobject *kobj,
860 struct attribute *attr,
864 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
869 rc = kstrtobool(buffer, &val);
874 set_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags);
876 clear_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags);
880 LUSTRE_RW_ATTR(statahead_agl);
882 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
884 struct super_block *sb = m->private;
885 struct ll_sb_info *sbi = ll_s2sbi(sb);
887 seq_printf(m, "statahead total: %u\n"
888 "statahead wrong: %u\n"
892 atomic_read(&sbi->ll_sa_total),
893 atomic_read(&sbi->ll_sa_wrong),
894 atomic_read(&sbi->ll_agl_total),
895 atomic_read(&sbi->ll_sa_hit_total),
896 atomic_read(&sbi->ll_sa_miss_total));
900 static ssize_t ll_statahead_stats_seq_write(struct file *file,
901 const char __user *buffer,
902 size_t count, loff_t *off)
904 struct seq_file *m = file->private_data;
905 struct super_block *sb = m->private;
906 struct ll_sb_info *sbi = ll_s2sbi(sb);
908 atomic_set(&sbi->ll_sa_total, 0);
909 atomic_set(&sbi->ll_sa_wrong, 0);
910 atomic_set(&sbi->ll_agl_total, 0);
911 atomic_set(&sbi->ll_sa_hit_total, 0);
912 atomic_set(&sbi->ll_sa_miss_total, 0);
916 LDEBUGFS_SEQ_FOPS(ll_statahead_stats);
918 static ssize_t lazystatfs_show(struct kobject *kobj,
919 struct attribute *attr,
922 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
925 return scnprintf(buf, PAGE_SIZE, "%u\n",
926 test_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags));
929 static ssize_t lazystatfs_store(struct kobject *kobj,
930 struct attribute *attr,
934 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
939 rc = kstrtobool(buffer, &val);
944 set_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags);
946 clear_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags);
950 LUSTRE_RW_ATTR(lazystatfs);
952 static ssize_t statfs_max_age_show(struct kobject *kobj, struct attribute *attr,
955 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
958 return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_statfs_max_age);
961 static ssize_t statfs_max_age_store(struct kobject *kobj,
962 struct attribute *attr, const char *buffer,
965 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
970 rc = kstrtouint(buffer, 10, &val);
973 if (val > OBD_STATFS_CACHE_MAX_AGE)
976 sbi->ll_statfs_max_age = val;
980 LUSTRE_RW_ATTR(statfs_max_age);
982 static ssize_t max_easize_show(struct kobject *kobj,
983 struct attribute *attr,
986 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
991 rc = ll_get_max_mdsize(sbi, &ealen);
995 /* Limit xattr size returned to userspace based on kernel maximum */
996 return scnprintf(buf, PAGE_SIZE, "%u\n",
997 ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
999 LUSTRE_RO_ATTR(max_easize);
1002 * Get default_easize.
1004 * \see client_obd::cl_default_mds_easize
1006 * \param[in] m seq_file handle
1007 * \param[in] v unused for single entry
1009 * \retval 0 on success
1010 * \retval negative negated errno on failure
1012 static ssize_t default_easize_show(struct kobject *kobj,
1013 struct attribute *attr,
1016 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1021 rc = ll_get_default_mdsize(sbi, &ealen);
1025 /* Limit xattr size returned to userspace based on kernel maximum */
1026 return scnprintf(buf, PAGE_SIZE, "%u\n",
1027 ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
1031 * Set default_easize.
1033 * Range checking on the passed value is handled by
1034 * ll_set_default_mdsize().
1036 * \see client_obd::cl_default_mds_easize
1038 * \param[in] file proc file
1039 * \param[in] buffer string passed from user space
1040 * \param[in] count \a buffer length
1041 * \param[in] off unused for single entry
1043 * \retval positive \a count on success
1044 * \retval negative negated errno on failure
1046 static ssize_t default_easize_store(struct kobject *kobj,
1047 struct attribute *attr,
1051 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1059 rc = kstrtouint(buffer, 10, &val);
1063 rc = ll_set_default_mdsize(sbi, val);
1069 LUSTRE_RW_ATTR(default_easize);
1071 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
1073 static ssize_t xattr_cache_show(struct kobject *kobj,
1074 struct attribute *attr,
1077 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1080 return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
1083 static ssize_t xattr_cache_store(struct kobject *kobj,
1084 struct attribute *attr,
1088 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1093 rc = kstrtobool(buffer, &val);
1097 if (val && !test_bit(LL_SBI_XATTR_CACHE, sbi->ll_flags))
1100 sbi->ll_xattr_cache_enabled = val;
1101 sbi->ll_xattr_cache_set = 1;
1105 LUSTRE_RW_ATTR(xattr_cache);
1107 static ssize_t tiny_write_show(struct kobject *kobj,
1108 struct attribute *attr,
1111 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1114 return scnprintf(buf, PAGE_SIZE, "%u\n",
1115 test_bit(LL_SBI_TINY_WRITE, sbi->ll_flags));
1118 static ssize_t tiny_write_store(struct kobject *kobj,
1119 struct attribute *attr,
1123 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1128 rc = kstrtobool(buffer, &val);
1132 spin_lock(&sbi->ll_lock);
1134 set_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
1136 clear_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
1137 spin_unlock(&sbi->ll_lock);
1141 LUSTRE_RW_ATTR(tiny_write);
1143 static ssize_t parallel_dio_show(struct kobject *kobj,
1144 struct attribute *attr,
1147 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1150 return snprintf(buf, PAGE_SIZE, "%u\n",
1151 test_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags));
1154 static ssize_t parallel_dio_store(struct kobject *kobj,
1155 struct attribute *attr,
1159 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1164 rc = kstrtobool(buffer, &val);
1168 spin_lock(&sbi->ll_lock);
1170 set_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
1172 clear_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
1173 spin_unlock(&sbi->ll_lock);
1177 LUSTRE_RW_ATTR(parallel_dio);
1179 static ssize_t max_read_ahead_async_active_show(struct kobject *kobj,
1180 struct attribute *attr,
1183 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1186 return scnprintf(buf, PAGE_SIZE, "%u\n",
1187 sbi->ll_ra_info.ra_async_max_active);
1190 static ssize_t max_read_ahead_async_active_store(struct kobject *kobj,
1191 struct attribute *attr,
1195 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1200 rc = kstrtouint(buffer, 10, &val);
1205 * It doesn't make any sense to make it exceed what
1206 * workqueue could acutally support. This can easily
1207 * over subscripe the cores but Lustre internally
1208 * throttles to avoid those impacts.
1210 if (val > WQ_UNBOUND_MAX_ACTIVE) {
1211 CERROR("%s: cannot set max_read_ahead_async_active=%u larger than %u\n",
1212 sbi->ll_fsname, val, WQ_UNBOUND_MAX_ACTIVE);
1216 spin_lock(&sbi->ll_lock);
1217 sbi->ll_ra_info.ra_async_max_active = val;
1218 spin_unlock(&sbi->ll_lock);
1222 LUSTRE_RW_ATTR(max_read_ahead_async_active);
1224 static ssize_t read_ahead_async_file_threshold_mb_show(struct kobject *kobj,
1225 struct attribute *attr,
1228 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1231 return scnprintf(buf, PAGE_SIZE, "%lu\n", PAGES_TO_MiB(
1232 sbi->ll_ra_info.ra_async_pages_per_file_threshold));
1236 read_ahead_async_file_threshold_mb_store(struct kobject *kobj,
1237 struct attribute *attr,
1238 const char *buffer, size_t count)
1240 unsigned long pages_number;
1241 unsigned long max_ra_per_file;
1242 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1246 rc = kstrtoul(buffer, 10, &pages_number);
1250 pages_number = MiB_TO_PAGES(pages_number);
1251 max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1252 if (pages_number < 0 || pages_number > max_ra_per_file) {
1253 CERROR("%s: can't set read_ahead_async_file_threshold_mb=%lu > "
1254 "max_read_readahead_per_file_mb=%lu\n", sbi->ll_fsname,
1255 PAGES_TO_MiB(pages_number),
1256 PAGES_TO_MiB(max_ra_per_file));
1259 sbi->ll_ra_info.ra_async_pages_per_file_threshold = pages_number;
1263 LUSTRE_RW_ATTR(read_ahead_async_file_threshold_mb);
1265 static ssize_t read_ahead_range_kb_show(struct kobject *kobj,
1266 struct attribute *attr,char *buf)
1268 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1271 return snprintf(buf, PAGE_SIZE, "%lu\n",
1272 sbi->ll_ra_info.ra_range_pages << (PAGE_SHIFT - 10));
1276 read_ahead_range_kb_store(struct kobject *kobj,
1277 struct attribute *attr,
1278 const char *buffer, size_t count)
1280 unsigned long pages_number;
1281 unsigned long max_ra_per_file;
1283 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1287 rc = sysfs_memparse(buffer, count, &val, "KiB");
1291 pages_number = val >> PAGE_SHIFT;
1292 /* Disable mmap range read */
1293 if (pages_number == 0)
1296 max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1297 if (pages_number > max_ra_per_file ||
1298 pages_number < RA_MIN_MMAP_RANGE_PAGES)
1302 spin_lock(&sbi->ll_lock);
1303 sbi->ll_ra_info.ra_range_pages = pages_number;
1304 spin_unlock(&sbi->ll_lock);
1308 LUSTRE_RW_ATTR(read_ahead_range_kb);
1310 static ssize_t fast_read_show(struct kobject *kobj,
1311 struct attribute *attr,
1314 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1317 return scnprintf(buf, PAGE_SIZE, "%u\n",
1318 test_bit(LL_SBI_FAST_READ, sbi->ll_flags));
1321 static ssize_t fast_read_store(struct kobject *kobj,
1322 struct attribute *attr,
1326 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1331 rc = kstrtobool(buffer, &val);
1335 spin_lock(&sbi->ll_lock);
1337 set_bit(LL_SBI_FAST_READ, sbi->ll_flags);
1339 clear_bit(LL_SBI_FAST_READ, sbi->ll_flags);
1340 spin_unlock(&sbi->ll_lock);
1344 LUSTRE_RW_ATTR(fast_read);
1346 static ssize_t file_heat_show(struct kobject *kobj,
1347 struct attribute *attr,
1350 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1353 return scnprintf(buf, PAGE_SIZE, "%u\n",
1354 test_bit(LL_SBI_FILE_HEAT, sbi->ll_flags));
1357 static ssize_t file_heat_store(struct kobject *kobj,
1358 struct attribute *attr,
1362 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1367 rc = kstrtobool(buffer, &val);
1371 spin_lock(&sbi->ll_lock);
1373 set_bit(LL_SBI_FILE_HEAT, sbi->ll_flags);
1375 clear_bit(LL_SBI_FILE_HEAT, sbi->ll_flags);
1376 spin_unlock(&sbi->ll_lock);
1380 LUSTRE_RW_ATTR(file_heat);
1382 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1383 struct attribute *attr,
1386 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1389 return scnprintf(buf, PAGE_SIZE, "%u\n",
1390 (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1393 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1394 struct attribute *attr,
1398 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1403 rc = kstrtoul(buffer, 10, &val);
1407 if (val < 0 || val > 100)
1410 sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1414 LUSTRE_RW_ATTR(heat_decay_percentage);
1416 static ssize_t heat_period_second_show(struct kobject *kobj,
1417 struct attribute *attr,
1420 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1423 return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1426 static ssize_t heat_period_second_store(struct kobject *kobj,
1427 struct attribute *attr,
1431 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1436 rc = kstrtoul(buffer, 10, &val);
1443 sbi->ll_heat_period_second = val;
1447 LUSTRE_RW_ATTR(heat_period_second);
1449 static ssize_t opencache_threshold_count_show(struct kobject *kobj,
1450 struct attribute *attr,
1453 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1456 if (sbi->ll_oc_thrsh_count)
1457 return snprintf(buf, PAGE_SIZE, "%u\n",
1458 sbi->ll_oc_thrsh_count);
1460 return snprintf(buf, PAGE_SIZE, "off\n");
1463 static ssize_t opencache_threshold_count_store(struct kobject *kobj,
1464 struct attribute *attr,
1468 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1473 rc = kstrtouint(buffer, 10, &val);
1476 /* also accept "off" to disable and "on" to always cache */
1477 rc = kstrtobool(buffer, &enable);
1482 sbi->ll_oc_thrsh_count = val;
1486 LUSTRE_RW_ATTR(opencache_threshold_count);
1488 static ssize_t opencache_threshold_ms_show(struct kobject *kobj,
1489 struct attribute *attr,
1492 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1495 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_oc_thrsh_ms);
1498 static ssize_t opencache_threshold_ms_store(struct kobject *kobj,
1499 struct attribute *attr,
1503 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1508 rc = kstrtouint(buffer, 10, &val);
1512 sbi->ll_oc_thrsh_ms = val;
1516 LUSTRE_RW_ATTR(opencache_threshold_ms);
1518 static ssize_t opencache_max_ms_show(struct kobject *kobj,
1519 struct attribute *attr,
1522 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1525 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_oc_max_ms);
1528 static ssize_t opencache_max_ms_store(struct kobject *kobj,
1529 struct attribute *attr,
1533 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1538 rc = kstrtouint(buffer, 10, &val);
1542 sbi->ll_oc_max_ms = val;
1546 LUSTRE_RW_ATTR(opencache_max_ms);
1548 static ssize_t inode_cache_show(struct kobject *kobj,
1549 struct attribute *attr,
1552 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1555 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_inode_cache_enabled);
1558 static ssize_t inode_cache_store(struct kobject *kobj,
1559 struct attribute *attr,
1563 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1568 rc = kstrtobool(buffer, &val);
1572 sbi->ll_inode_cache_enabled = val;
1576 LUSTRE_RW_ATTR(inode_cache);
1578 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1580 struct super_block *sb = m->private;
1581 struct ll_sb_info *sbi = ll_s2sbi(sb);
1582 struct cl_client_cache *cache = sbi->ll_cache;
1586 pages = atomic_long_read(&cache->ccc_unstable_nr);
1587 mb = (pages * PAGE_SIZE) >> 20;
1589 seq_printf(m, "unstable_check: %8d\n"
1590 "unstable_pages: %12ld\n"
1591 "unstable_mb: %8d\n",
1592 cache->ccc_unstable_check, pages, mb);
1596 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1597 const char __user *buffer,
1598 size_t count, loff_t *unused)
1600 struct seq_file *seq = file->private_data;
1601 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1608 if (count >= sizeof(kernbuf))
1611 if (copy_from_user(kernbuf, buffer, count))
1615 buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1617 rc = kstrtobool_from_user(buffer, count, &val);
1621 /* borrow lru lock to set the value */
1622 spin_lock(&sbi->ll_cache->ccc_lru_lock);
1623 sbi->ll_cache->ccc_unstable_check = val;
1624 spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1629 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1631 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1633 struct super_block *sb = m->private;
1634 struct ll_sb_info *sbi = ll_s2sbi(sb);
1635 struct root_squash_info *squash = &sbi->ll_squash;
1637 seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1641 static ssize_t ll_root_squash_seq_write(struct file *file,
1642 const char __user *buffer,
1643 size_t count, loff_t *off)
1645 struct seq_file *m = file->private_data;
1646 struct super_block *sb = m->private;
1647 struct ll_sb_info *sbi = ll_s2sbi(sb);
1648 struct root_squash_info *squash = &sbi->ll_squash;
1650 return lprocfs_wr_root_squash(buffer, count, squash, sbi->ll_fsname);
1653 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1655 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1657 struct super_block *sb = m->private;
1658 struct ll_sb_info *sbi = ll_s2sbi(sb);
1659 struct root_squash_info *squash = &sbi->ll_squash;
1662 spin_lock(&squash->rsi_lock);
1663 if (!list_empty(&squash->rsi_nosquash_nids)) {
1664 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1665 &squash->rsi_nosquash_nids);
1669 seq_puts(m, "NONE\n");
1671 spin_unlock(&squash->rsi_lock);
1676 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1677 const char __user *buffer,
1678 size_t count, loff_t *off)
1680 struct seq_file *m = file->private_data;
1681 struct super_block *sb = m->private;
1682 struct ll_sb_info *sbi = ll_s2sbi(sb);
1683 struct root_squash_info *squash = &sbi->ll_squash;
1686 rc = lprocfs_wr_nosquash_nids(buffer, count, squash, sbi->ll_fsname);
1690 ll_compute_rootsquash_state(sbi);
1695 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1697 #if defined(CONFIG_LL_ENCRYPTION)
1698 static ssize_t enable_filename_encryption_show(struct kobject *kobj,
1699 struct attribute *attr,
1702 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1704 struct lustre_sb_info *lsi = sbi->lsi;
1706 return snprintf(buffer, PAGE_SIZE, "%u\n",
1707 lsi->lsi_flags & LSI_FILENAME_ENC ? 1 : 0);
1710 static ssize_t enable_filename_encryption_store(struct kobject *kobj,
1711 struct attribute *attr,
1715 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1717 struct lustre_sb_info *lsi = sbi->lsi;
1721 rc = kstrtobool(buffer, &val);
1726 if (!ll_sbi_has_name_encrypt(sbi)) {
1727 /* server does not support name encryption,
1728 * so force it to NULL on client
1730 CDEBUG(D_SEC, "%s: server does not support name encryption\n",
1732 lsi->lsi_flags &= ~LSI_FILENAME_ENC;
1736 lsi->lsi_flags |= LSI_FILENAME_ENC;
1738 lsi->lsi_flags &= ~LSI_FILENAME_ENC;
1744 LUSTRE_RW_ATTR(enable_filename_encryption);
1745 #endif /* CONFIG_LL_ENCRYPTION */
1747 #if defined(CONFIG_LL_ENCRYPTION) || defined(HAVE_LUSTRE_CRYPTO)
1748 static ssize_t filename_enc_use_old_base64_show(struct kobject *kobj,
1749 struct attribute *attr,
1752 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1754 struct lustre_sb_info *lsi = sbi->lsi;
1756 return snprintf(buffer, PAGE_SIZE, "%u\n",
1757 lsi->lsi_flags & LSI_FILENAME_ENC_B64_OLD_CLI ? 1 : 0);
1760 static ssize_t filename_enc_use_old_base64_store(struct kobject *kobj,
1761 struct attribute *attr,
1765 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1767 struct lustre_sb_info *lsi = sbi->lsi;
1771 rc = kstrtobool(buffer, &val);
1776 if (!ll_sbi_has_name_encrypt(sbi)) {
1777 /* server does not support name encryption,
1778 * so force it to NULL on client
1781 "%s: server does not support name encryption\n",
1783 lsi->lsi_flags &= ~LSI_FILENAME_ENC_B64_OLD_CLI;
1787 lsi->lsi_flags |= LSI_FILENAME_ENC_B64_OLD_CLI;
1789 lsi->lsi_flags &= ~LSI_FILENAME_ENC_B64_OLD_CLI;
1795 LUSTRE_RW_ATTR(filename_enc_use_old_base64);
1796 #endif /* CONFIG_LL_ENCRYPTION || HAVE_LUSTRE_CRYPTO */
1798 static int ll_pcc_seq_show(struct seq_file *m, void *v)
1800 struct super_block *sb = m->private;
1801 struct ll_sb_info *sbi = ll_s2sbi(sb);
1803 return pcc_super_dump(&sbi->ll_pcc_super, m);
1806 static ssize_t ll_pcc_seq_write(struct file *file, const char __user *buffer,
1807 size_t count, loff_t *off)
1809 struct seq_file *m = file->private_data;
1810 struct super_block *sb = m->private;
1811 struct ll_sb_info *sbi = ll_s2sbi(sb);
1815 if (count >= LPROCFS_WR_PCC_MAX_CMD)
1818 if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_PCC))
1821 OBD_ALLOC(kernbuf, count + 1);
1822 if (kernbuf == NULL)
1825 if (copy_from_user(kernbuf, buffer, count))
1826 GOTO(out_free_kernbuff, rc = -EFAULT);
1828 rc = pcc_cmd_handle(kernbuf, count, &sbi->ll_pcc_super);
1830 OBD_FREE(kernbuf, count + 1);
1831 return rc ? rc : count;
1833 LDEBUGFS_SEQ_FOPS(ll_pcc);
1835 struct ldebugfs_vars lprocfs_llite_obd_vars[] = {
1837 .fops = &ll_site_stats_fops },
1838 { .name = "max_cached_mb",
1839 .fops = &ll_max_cached_mb_fops },
1840 { .name = "statahead_stats",
1841 .fops = &ll_statahead_stats_fops },
1842 { .name = "unstable_stats",
1843 .fops = &ll_unstable_stats_fops },
1844 { .name = "sbi_flags",
1845 .fops = &ll_sbi_flags_fops },
1846 { .name = "root_squash",
1847 .fops = &ll_root_squash_fops },
1848 { .name = "nosquash_nids",
1849 .fops = &ll_nosquash_nids_fops },
1851 .fops = &ll_pcc_fops, },
1855 #define MAX_STRING_SIZE 128
1857 static struct attribute *llite_attrs[] = {
1858 &lustre_attr_blocksize.attr,
1859 &lustre_attr_stat_blocksize.attr,
1860 &lustre_attr_kbytestotal.attr,
1861 &lustre_attr_kbytesfree.attr,
1862 &lustre_attr_kbytesavail.attr,
1863 &lustre_attr_filestotal.attr,
1864 &lustre_attr_filesfree.attr,
1865 &lustre_attr_client_type.attr,
1866 &lustre_attr_foreign_symlink_enable.attr,
1867 &lustre_attr_foreign_symlink_prefix.attr,
1868 &lustre_attr_foreign_symlink_upcall.attr,
1869 &lustre_attr_foreign_symlink_upcall_info.attr,
1870 &lustre_attr_fstype.attr,
1871 &lustre_attr_uuid.attr,
1872 &lustre_attr_checksums.attr,
1873 &lustre_attr_checksum_pages.attr,
1874 &lustre_attr_max_read_ahead_mb.attr,
1875 &lustre_attr_max_read_ahead_per_file_mb.attr,
1876 &lustre_attr_max_read_ahead_whole_mb.attr,
1877 &lustre_attr_max_read_ahead_async_active.attr,
1878 &lustre_attr_read_ahead_async_file_threshold_mb.attr,
1879 &lustre_attr_read_ahead_range_kb.attr,
1880 &lustre_attr_stats_track_pid.attr,
1881 &lustre_attr_stats_track_ppid.attr,
1882 &lustre_attr_stats_track_gid.attr,
1883 &lustre_attr_statahead_running_max.attr,
1884 &lustre_attr_statahead_batch_max.attr,
1885 &lustre_attr_statahead_max.attr,
1886 &lustre_attr_statahead_agl.attr,
1887 &lustre_attr_lazystatfs.attr,
1888 &lustre_attr_statfs_max_age.attr,
1889 &lustre_attr_max_easize.attr,
1890 &lustre_attr_default_easize.attr,
1891 &lustre_attr_xattr_cache.attr,
1892 &lustre_attr_fast_read.attr,
1893 &lustre_attr_tiny_write.attr,
1894 &lustre_attr_parallel_dio.attr,
1895 &lustre_attr_file_heat.attr,
1896 &lustre_attr_heat_decay_percentage.attr,
1897 &lustre_attr_heat_period_second.attr,
1898 &lustre_attr_opencache_threshold_count.attr,
1899 &lustre_attr_opencache_threshold_ms.attr,
1900 &lustre_attr_opencache_max_ms.attr,
1901 &lustre_attr_inode_cache.attr,
1902 #ifdef CONFIG_LL_ENCRYPTION
1903 &lustre_attr_enable_filename_encryption.attr,
1905 #if defined(CONFIG_LL_ENCRYPTION) || defined(HAVE_LUSTRE_CRYPTO)
1906 &lustre_attr_filename_enc_use_old_base64.attr,
1911 KOBJ_ATTRIBUTE_GROUPS(llite); /* creates llite_groups */
1913 static void sbi_kobj_release(struct kobject *kobj)
1915 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1917 complete(&sbi->ll_kobj_unregister);
1920 static struct kobj_type sbi_ktype = {
1921 .default_groups = KOBJ_ATTR_GROUPS(llite),
1922 .sysfs_ops = &lustre_sysfs_ops,
1923 .release = sbi_kobj_release,
1926 static const struct llite_file_opcode {
1928 enum lprocfs_counter_config lfo_config;
1929 const char *lfo_opname;
1930 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1931 /* file operation */
1932 { LPROC_LL_READ_BYTES, LPROCFS_TYPE_BYTES_FULL, "read_bytes" },
1933 { LPROC_LL_WRITE_BYTES, LPROCFS_TYPE_BYTES_FULL, "write_bytes" },
1934 { LPROC_LL_READ, LPROCFS_TYPE_LATENCY, "read" },
1935 { LPROC_LL_WRITE, LPROCFS_TYPE_LATENCY, "write" },
1936 { LPROC_LL_IOCTL, LPROCFS_TYPE_REQS, "ioctl" },
1937 { LPROC_LL_OPEN, LPROCFS_TYPE_LATENCY, "open" },
1938 { LPROC_LL_RELEASE, LPROCFS_TYPE_LATENCY, "close" },
1939 { LPROC_LL_MMAP, LPROCFS_TYPE_LATENCY, "mmap" },
1940 { LPROC_LL_FAULT, LPROCFS_TYPE_LATENCY, "page_fault" },
1941 { LPROC_LL_MKWRITE, LPROCFS_TYPE_LATENCY, "page_mkwrite" },
1942 { LPROC_LL_LLSEEK, LPROCFS_TYPE_LATENCY, "seek" },
1943 { LPROC_LL_FSYNC, LPROCFS_TYPE_LATENCY, "fsync" },
1944 { LPROC_LL_READDIR, LPROCFS_TYPE_LATENCY, "readdir" },
1945 { LPROC_LL_INODE_OCOUNT, LPROCFS_TYPE_REQS | LPROCFS_CNTR_AVGMINMAX |
1946 LPROCFS_CNTR_STDDEV, "opencount" },
1947 { LPROC_LL_INODE_OPCLTM,LPROCFS_TYPE_LATENCY, "openclosetime" },
1948 /* inode operation */
1949 { LPROC_LL_SETATTR, LPROCFS_TYPE_LATENCY, "setattr" },
1950 { LPROC_LL_TRUNC, LPROCFS_TYPE_LATENCY, "truncate" },
1951 { LPROC_LL_FLOCK, LPROCFS_TYPE_LATENCY, "flock" },
1952 { LPROC_LL_GETATTR, LPROCFS_TYPE_LATENCY, "getattr" },
1953 { LPROC_LL_FALLOCATE, LPROCFS_TYPE_LATENCY, "fallocate"},
1954 /* dir inode operation */
1955 { LPROC_LL_CREATE, LPROCFS_TYPE_LATENCY, "create" },
1956 { LPROC_LL_LINK, LPROCFS_TYPE_LATENCY, "link" },
1957 { LPROC_LL_UNLINK, LPROCFS_TYPE_LATENCY, "unlink" },
1958 { LPROC_LL_SYMLINK, LPROCFS_TYPE_LATENCY, "symlink" },
1959 { LPROC_LL_MKDIR, LPROCFS_TYPE_LATENCY, "mkdir" },
1960 { LPROC_LL_RMDIR, LPROCFS_TYPE_LATENCY, "rmdir" },
1961 { LPROC_LL_MKNOD, LPROCFS_TYPE_LATENCY, "mknod" },
1962 { LPROC_LL_RENAME, LPROCFS_TYPE_LATENCY, "rename" },
1963 /* special inode operation */
1964 { LPROC_LL_STATFS, LPROCFS_TYPE_LATENCY, "statfs" },
1965 { LPROC_LL_SETXATTR, LPROCFS_TYPE_LATENCY, "setxattr" },
1966 { LPROC_LL_GETXATTR, LPROCFS_TYPE_LATENCY, "getxattr" },
1967 { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REQS, "getxattr_hits" },
1968 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_LATENCY, "listxattr" },
1969 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_LATENCY, "removexattr" },
1970 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_LATENCY, "inode_permission" },
1973 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count)
1978 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1979 lprocfs_counter_add(sbi->ll_stats, op, count);
1980 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1981 sbi->ll_stats_track_id == current->pid)
1982 lprocfs_counter_add(sbi->ll_stats, op, count);
1983 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1984 sbi->ll_stats_track_id == current->real_parent->pid)
1985 lprocfs_counter_add(sbi->ll_stats, op, count);
1986 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1987 sbi->ll_stats_track_id ==
1988 from_kgid(&init_user_ns, current_gid()))
1989 lprocfs_counter_add(sbi->ll_stats, op, count);
1991 EXPORT_SYMBOL(ll_stats_ops_tally);
1993 static const char *const ra_stat_string[] = {
1994 [RA_STAT_HIT] = "hits",
1995 [RA_STAT_MISS] = "misses",
1996 [RA_STAT_DISTANT_READPAGE] = "readpage_not_consecutive",
1997 [RA_STAT_MISS_IN_WINDOW] = "miss_inside_window",
1998 [RA_STAT_FAILED_GRAB_PAGE] = "failed_grab_cache_page",
1999 [RA_STAT_FAILED_MATCH] = "failed_lock_match",
2000 [RA_STAT_DISCARDED] = "read_but_discarded",
2001 [RA_STAT_ZERO_LEN] = "zero_length_file",
2002 [RA_STAT_ZERO_WINDOW] = "zero_size_window",
2003 [RA_STAT_EOF] = "readahead_to_eof",
2004 [RA_STAT_MAX_IN_FLIGHT] = "hit_max_readahead_issue",
2005 [RA_STAT_WRONG_GRAB_PAGE] = "wrong_page_from_grab_cache_page",
2006 [RA_STAT_FAILED_REACH_END] = "failed_to_reach_end",
2007 [RA_STAT_ASYNC] = "async_readahead",
2008 [RA_STAT_FAILED_FAST_READ] = "failed_to_fast_read",
2009 [RA_STAT_MMAP_RANGE_READ] = "mmap_range_read",
2010 [RA_STAT_READAHEAD_PAGES] = "readahead_pages"
2013 int ll_debugfs_register_super(struct super_block *sb, const char *name)
2015 struct lustre_sb_info *lsi = s2lsi(sb);
2016 struct ll_sb_info *sbi = ll_s2sbi(sb);
2022 if (IS_ERR_OR_NULL(llite_root))
2025 sbi->ll_debugfs_entry = debugfs_create_dir(name, llite_root);
2026 ldebugfs_add_vars(sbi->ll_debugfs_entry, lprocfs_llite_obd_vars, sb);
2028 debugfs_create_file("dump_page_cache", 0444, sbi->ll_debugfs_entry, sbi,
2029 &vvp_dump_pgcache_file_ops);
2031 debugfs_create_file("extents_stats", 0644, sbi->ll_debugfs_entry, sbi,
2032 &ll_rw_extents_stats_fops);
2034 debugfs_create_file("extents_stats_per_process", 0644,
2035 sbi->ll_debugfs_entry, sbi,
2036 &ll_rw_extents_stats_pp_fops);
2038 debugfs_create_file("offset_stats", 0644, sbi->ll_debugfs_entry, sbi,
2039 &ll_rw_offset_stats_fops);
2041 /* File operations stats */
2042 sbi->ll_stats = lprocfs_stats_alloc(LPROC_LL_FILE_OPCODES,
2043 LPROCFS_STATS_FLAG_NONE);
2044 if (sbi->ll_stats == NULL)
2045 GOTO(out_debugfs, err = -ENOMEM);
2047 /* do counter init */
2048 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++)
2049 lprocfs_counter_init(sbi->ll_stats,
2050 llite_opcode_table[id].lfo_opcode,
2051 llite_opcode_table[id].lfo_config,
2052 llite_opcode_table[id].lfo_opname);
2054 debugfs_create_file("stats", 0644, sbi->ll_debugfs_entry,
2055 sbi->ll_stats, &ldebugfs_stats_seq_fops);
2057 sbi->ll_ra_stats = lprocfs_stats_alloc(ARRAY_SIZE(ra_stat_string),
2058 LPROCFS_STATS_FLAG_NONE);
2059 if (sbi->ll_ra_stats == NULL)
2060 GOTO(out_stats, err = -ENOMEM);
2062 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++) {
2063 if (id == RA_STAT_READAHEAD_PAGES)
2064 lprocfs_counter_init(sbi->ll_ra_stats, id,
2065 LPROCFS_TYPE_PAGES |
2066 LPROCFS_CNTR_AVGMINMAX,
2067 ra_stat_string[id]);
2069 lprocfs_counter_init(sbi->ll_ra_stats, id,
2071 ra_stat_string[id]);
2074 debugfs_create_file("read_ahead_stats", 0644, sbi->ll_debugfs_entry,
2075 sbi->ll_ra_stats, &ldebugfs_stats_seq_fops);
2078 /* Yes we also register sysfs mount kset here as well */
2079 sbi->ll_kset.kobj.parent = llite_kobj;
2080 sbi->ll_kset.kobj.ktype = &sbi_ktype;
2081 init_completion(&sbi->ll_kobj_unregister);
2082 err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
2084 GOTO(out_ra_stats, err);
2086 err = kset_register(&sbi->ll_kset);
2088 GOTO(out_ra_stats, err);
2090 lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
2094 lprocfs_stats_free(&sbi->ll_ra_stats);
2096 lprocfs_stats_free(&sbi->ll_stats);
2098 debugfs_remove_recursive(sbi->ll_debugfs_entry);
2103 void ll_debugfs_unregister_super(struct super_block *sb)
2105 struct lustre_sb_info *lsi = s2lsi(sb);
2106 struct ll_sb_info *sbi = ll_s2sbi(sb);
2108 debugfs_remove_recursive(sbi->ll_debugfs_entry);
2111 sysfs_remove_link(&sbi->ll_kset.kobj,
2112 sbi->ll_dt_obd->obd_type->typ_name);
2115 sysfs_remove_link(&sbi->ll_kset.kobj,
2116 sbi->ll_md_obd->obd_type->typ_name);
2118 kobject_put(lsi->lsi_kobj);
2120 kset_unregister(&sbi->ll_kset);
2121 wait_for_completion(&sbi->ll_kobj_unregister);
2123 lprocfs_stats_free(&sbi->ll_ra_stats);
2124 lprocfs_stats_free(&sbi->ll_stats);
2126 #undef MAX_STRING_SIZE
2128 static void ll_display_extents_info(struct ll_rw_extents_info *rw_extents,
2129 struct seq_file *seq, int which)
2131 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
2132 unsigned long start, end, r, w;
2133 char *unitp = "KMGTPEZY";
2135 struct per_process_info *pp_info;
2137 pp_info = &rw_extents->pp_extents[which];
2142 for (i = 0; i < LL_HIST_MAX; i++) {
2143 read_tot += pp_info->pp_r_hist.oh_buckets[i];
2144 write_tot += pp_info->pp_w_hist.oh_buckets[i];
2147 for (i = 0; i < LL_HIST_MAX; i++) {
2148 r = pp_info->pp_r_hist.oh_buckets[i];
2149 w = pp_info->pp_w_hist.oh_buckets[i];
2152 end = 1 << (i + LL_HIST_START - units);
2153 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u | "
2154 "%14lu %4u %4u\n", start, *unitp, end, *unitp,
2155 (i == LL_HIST_MAX - 1) ? '+' : ' ',
2156 r, pct(r, read_tot), pct(read_cum, read_tot),
2157 w, pct(w, write_tot), pct(write_cum, write_tot));
2159 if (start == (1 << 10)) {
2164 if (read_cum == read_tot && write_cum == write_tot)
2169 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
2171 struct ll_sb_info *sbi = seq->private;
2172 struct ll_rw_extents_info *rw_extents = sbi->ll_rw_extents_info;
2175 if (!sbi->ll_rw_stats_on || !rw_extents) {
2176 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2180 spin_lock(&sbi->ll_pp_extent_lock);
2181 lprocfs_stats_header(seq, ktime_get_real(), rw_extents->pp_init, 25,
2183 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
2184 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
2185 "extents", "calls", "%", "cum%", "calls", "%", "cum%");
2187 for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
2188 if (rw_extents->pp_extents[k].pid != 0) {
2189 seq_printf(seq, "\nPID: %d\n",
2190 rw_extents->pp_extents[k].pid);
2191 ll_display_extents_info(rw_extents, seq, k);
2194 spin_unlock(&sbi->ll_pp_extent_lock);
2198 static int alloc_rw_stats_info(struct ll_sb_info *sbi)
2200 struct ll_rw_extents_info *rw_extents;
2201 struct ll_rw_process_info *offset;
2202 struct ll_rw_process_info *process;
2205 OBD_ALLOC(rw_extents, sizeof(*rw_extents));
2209 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
2210 spin_lock_init(&rw_extents->pp_extents[i].pp_r_hist.oh_lock);
2211 spin_lock_init(&rw_extents->pp_extents[i].pp_w_hist.oh_lock);
2213 rw_extents->pp_init = ktime_get_real();
2215 spin_lock(&sbi->ll_pp_extent_lock);
2216 if (!sbi->ll_rw_extents_info)
2217 sbi->ll_rw_extents_info = rw_extents;
2218 spin_unlock(&sbi->ll_pp_extent_lock);
2219 /* another writer allocated the struct before we got the lock */
2220 if (sbi->ll_rw_extents_info != rw_extents)
2221 OBD_FREE(rw_extents, sizeof(*rw_extents));
2223 OBD_ALLOC(process, sizeof(*process) * LL_PROCESS_HIST_MAX);
2225 GOTO(out, rc = -ENOMEM);
2226 OBD_ALLOC(offset, sizeof(*offset) * LL_OFFSET_HIST_MAX);
2228 GOTO(out_free, rc = -ENOMEM);
2230 spin_lock(&sbi->ll_process_lock);
2231 if (!sbi->ll_rw_process_info)
2232 sbi->ll_rw_process_info = process;
2233 if (!sbi->ll_rw_offset_info)
2234 sbi->ll_rw_offset_info = offset;
2235 spin_unlock(&sbi->ll_process_lock);
2236 sbi->ll_process_stats_init = ktime_get_real();
2238 /* another writer allocated the structs before we got the lock */
2239 if (sbi->ll_rw_offset_info != offset)
2240 OBD_FREE(offset, sizeof(*offset) * LL_OFFSET_HIST_MAX);
2241 if (sbi->ll_rw_process_info != process) {
2243 OBD_FREE(process, sizeof(*process) * LL_PROCESS_HIST_MAX);
2250 void ll_free_rw_stats_info(struct ll_sb_info *sbi)
2252 if (sbi->ll_rw_extents_info) {
2253 OBD_FREE(sbi->ll_rw_extents_info,
2254 sizeof(*sbi->ll_rw_extents_info));
2255 sbi->ll_rw_extents_info = NULL;
2257 if (sbi->ll_rw_offset_info) {
2258 OBD_FREE(sbi->ll_rw_offset_info,
2259 sizeof(*sbi->ll_rw_offset_info) * LL_OFFSET_HIST_MAX);
2260 sbi->ll_rw_offset_info = NULL;
2262 if (sbi->ll_rw_process_info) {
2263 OBD_FREE(sbi->ll_rw_process_info,
2264 sizeof(*sbi->ll_rw_process_info) * LL_PROCESS_HIST_MAX);
2265 sbi->ll_rw_process_info = NULL;
2269 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
2270 const char __user *buf,
2271 size_t len, loff_t *off)
2273 struct seq_file *seq = file->private_data;
2274 struct ll_sb_info *sbi = seq->private;
2275 struct ll_rw_extents_info *rw_extents;
2282 value = ll_stats_pid_write(buf, len);
2285 sbi->ll_rw_stats_on = 0;
2287 if (!sbi->ll_rw_extents_info) {
2288 int rc = alloc_rw_stats_info(sbi);
2293 sbi->ll_rw_stats_on = 1;
2297 spin_lock(&sbi->ll_pp_extent_lock);
2298 rw_extents = sbi->ll_rw_extents_info;
2300 rw_extents->pp_init = ktime_get_real();
2301 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2302 rw_extents->pp_extents[i].pid = 0;
2303 lprocfs_oh_clear(&rw_extents->pp_extents[i].pp_r_hist);
2304 lprocfs_oh_clear(&rw_extents->pp_extents[i].pp_w_hist);
2307 spin_unlock(&sbi->ll_pp_extent_lock);
2312 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
2314 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
2316 struct ll_sb_info *sbi = seq->private;
2317 struct ll_rw_extents_info *rw_extents = sbi->ll_rw_extents_info;
2319 if (!sbi->ll_rw_stats_on || !rw_extents) {
2320 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2324 spin_lock(&sbi->ll_lock);
2325 lprocfs_stats_header(seq, ktime_get_real(), rw_extents->pp_init, 25,
2328 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
2329 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
2330 "extents", "calls", "%", "cum%",
2331 "calls", "%", "cum%");
2333 ll_display_extents_info(rw_extents, seq, LL_PROCESS_HIST_MAX);
2334 spin_unlock(&sbi->ll_lock);
2339 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
2340 const char __user *buf,
2341 size_t len, loff_t *off)
2343 struct seq_file *seq = file->private_data;
2344 struct ll_sb_info *sbi = seq->private;
2345 struct ll_rw_extents_info *rw_extents;
2352 value = ll_stats_pid_write(buf, len);
2355 sbi->ll_rw_stats_on = 0;
2357 if (!sbi->ll_rw_extents_info) {
2358 int rc = alloc_rw_stats_info(sbi);
2363 sbi->ll_rw_stats_on = 1;
2366 spin_lock(&sbi->ll_pp_extent_lock);
2367 rw_extents = sbi->ll_rw_extents_info;
2369 rw_extents->pp_init = ktime_get_real();
2370 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
2371 rw_extents->pp_extents[i].pid = 0;
2372 lprocfs_oh_clear(&rw_extents->pp_extents[i].pp_r_hist);
2373 lprocfs_oh_clear(&rw_extents->pp_extents[i].pp_w_hist);
2376 spin_unlock(&sbi->ll_pp_extent_lock);
2381 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
2383 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
2384 struct ll_file_data *file, loff_t pos,
2385 size_t count, int rw)
2388 struct ll_rw_process_info *process;
2389 struct ll_rw_process_info *offset;
2390 int *off_count = &sbi->ll_rw_offset_entry_count;
2391 int *process_count = &sbi->ll_offset_process_count;
2392 struct ll_rw_extents_info *rw_extents;
2394 if (!sbi->ll_rw_stats_on)
2397 spin_lock(&sbi->ll_pp_extent_lock);
2398 rw_extents = sbi->ll_rw_extents_info;
2400 spin_unlock(&sbi->ll_pp_extent_lock);
2404 /* Extent statistics */
2405 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2406 if (rw_extents->pp_extents[i].pid == pid) {
2414 sbi->ll_extent_process_count =
2415 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
2416 cur = sbi->ll_extent_process_count;
2417 rw_extents->pp_extents[cur].pid = pid;
2418 lprocfs_oh_clear(&rw_extents->pp_extents[cur].pp_r_hist);
2419 lprocfs_oh_clear(&rw_extents->pp_extents[cur].pp_w_hist);
2422 for (i = 0; (count >= 1 << (LL_HIST_START + i)) &&
2423 (i < (LL_HIST_MAX - 1)); i++);
2425 rw_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
2426 rw_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
2428 rw_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
2429 rw_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
2431 spin_unlock(&sbi->ll_pp_extent_lock);
2433 spin_lock(&sbi->ll_process_lock);
2434 process = sbi->ll_rw_process_info;
2435 offset = sbi->ll_rw_offset_info;
2436 if (!process || !offset)
2439 /* Offset statistics */
2440 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2441 if (process[i].rw_pid == pid) {
2442 if (process[i].rw_last_file != file) {
2443 process[i].rw_range_start = pos;
2444 process[i].rw_last_file_pos = pos + count;
2445 process[i].rw_smallest_extent = count;
2446 process[i].rw_largest_extent = count;
2447 process[i].rw_offset = 0;
2448 process[i].rw_last_file = file;
2451 if (process[i].rw_last_file_pos != pos) {
2453 (*off_count + 1) % LL_OFFSET_HIST_MAX;
2454 offset[*off_count].rw_op = process[i].rw_op;
2455 offset[*off_count].rw_pid = pid;
2456 offset[*off_count].rw_range_start =
2457 process[i].rw_range_start;
2458 offset[*off_count].rw_range_end =
2459 process[i].rw_last_file_pos;
2460 offset[*off_count].rw_smallest_extent =
2461 process[i].rw_smallest_extent;
2462 offset[*off_count].rw_largest_extent =
2463 process[i].rw_largest_extent;
2464 offset[*off_count].rw_offset =
2465 process[i].rw_offset;
2466 process[i].rw_op = rw;
2467 process[i].rw_range_start = pos;
2468 process[i].rw_smallest_extent = count;
2469 process[i].rw_largest_extent = count;
2470 process[i].rw_offset = pos -
2471 process[i].rw_last_file_pos;
2473 if (process[i].rw_smallest_extent > count)
2474 process[i].rw_smallest_extent = count;
2475 if (process[i].rw_largest_extent < count)
2476 process[i].rw_largest_extent = count;
2477 process[i].rw_last_file_pos = pos + count;
2481 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
2482 process[*process_count].rw_pid = pid;
2483 process[*process_count].rw_op = rw;
2484 process[*process_count].rw_range_start = pos;
2485 process[*process_count].rw_last_file_pos = pos + count;
2486 process[*process_count].rw_smallest_extent = count;
2487 process[*process_count].rw_largest_extent = count;
2488 process[*process_count].rw_offset = 0;
2489 process[*process_count].rw_last_file = file;
2492 spin_unlock(&sbi->ll_process_lock);
2495 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
2497 struct ll_sb_info *sbi = seq->private;
2498 struct ll_rw_process_info *offset;
2499 struct ll_rw_process_info *process;
2502 if (!sbi->ll_rw_stats_on) {
2503 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2507 spin_lock(&sbi->ll_process_lock);
2508 lprocfs_stats_header(seq, ktime_get_real(), sbi->ll_process_stats_init,
2510 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
2511 "R/W", "PID", "RANGE START", "RANGE END",
2512 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
2514 /* We stored the discontiguous offsets here; print them first */
2515 offset = sbi->ll_rw_offset_info;
2516 for (i = 0; offset && i < LL_OFFSET_HIST_MAX; i++) {
2517 if (offset[i].rw_pid != 0)
2519 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2520 offset[i].rw_op == READ ? 'R' : 'W',
2522 offset[i].rw_range_start,
2523 offset[i].rw_range_end,
2524 (unsigned long)offset[i].rw_smallest_extent,
2525 (unsigned long)offset[i].rw_largest_extent,
2526 offset[i].rw_offset);
2529 /* Then print the current offsets for each process */
2530 process = sbi->ll_rw_process_info;
2531 for (i = 0; process && i < LL_PROCESS_HIST_MAX; i++) {
2532 if (process[i].rw_pid != 0)
2534 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2535 process[i].rw_op == READ ? 'R' : 'W',
2537 process[i].rw_range_start,
2538 process[i].rw_last_file_pos,
2539 (unsigned long)process[i].rw_smallest_extent,
2540 (unsigned long)process[i].rw_largest_extent,
2541 process[i].rw_offset);
2543 spin_unlock(&sbi->ll_process_lock);
2548 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
2549 const char __user *buf,
2550 size_t len, loff_t *off)
2552 struct seq_file *seq = file->private_data;
2553 struct ll_sb_info *sbi = seq->private;
2559 value = ll_stats_pid_write(buf, len);
2562 sbi->ll_rw_stats_on = 0;
2564 if (!sbi->ll_rw_process_info || !sbi->ll_rw_offset_info) {
2565 int rc = alloc_rw_stats_info(sbi);
2570 sbi->ll_rw_stats_on = 1;
2573 spin_lock(&sbi->ll_process_lock);
2574 sbi->ll_offset_process_count = 0;
2575 sbi->ll_rw_offset_entry_count = 0;
2576 sbi->ll_process_stats_init = ktime_get_real();
2577 if (sbi->ll_rw_process_info)
2578 memset(sbi->ll_rw_process_info, 0,
2579 sizeof(struct ll_rw_process_info) * LL_PROCESS_HIST_MAX);
2580 if (sbi->ll_rw_offset_info)
2581 memset(sbi->ll_rw_offset_info, 0,
2582 sizeof(struct ll_rw_process_info) * LL_OFFSET_HIST_MAX);
2583 spin_unlock(&sbi->ll_process_lock);
2588 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);