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 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
295 return sprintf(buf, "lustre\n");
297 LUSTRE_RO_ATTR(fstype);
299 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
302 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
305 return sprintf(buf, "%s\n", sbi->ll_sb_uuid.uuid);
307 LUSTRE_RO_ATTR(uuid);
309 static int ll_site_stats_seq_show(struct seq_file *m, void *v)
311 struct super_block *sb = m->private;
314 * See description of statistical counters in struct cl_site, and
317 return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site), m);
320 LDEBUGFS_SEQ_FOPS_RO(ll_site_stats);
322 static ssize_t max_read_ahead_mb_show(struct kobject *kobj,
323 struct attribute *attr, char *buf)
325 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
328 return scnprintf(buf, PAGE_SIZE, "%lu\n",
329 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
332 static ssize_t max_read_ahead_mb_store(struct kobject *kobj,
333 struct attribute *attr,
334 const char *buffer, size_t count)
336 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
338 u64 ra_max_mb, pages_number;
341 rc = sysfs_memparse(buffer, count, &ra_max_mb, "MiB");
345 pages_number = round_up(ra_max_mb, 1024 * 1024) >> PAGE_SHIFT;
346 CDEBUG(D_INFO, "%s: set max_read_ahead_mb=%llu (%llu pages)\n",
347 sbi->ll_fsname, PAGES_TO_MiB(pages_number), pages_number);
348 if (pages_number > cfs_totalram_pages() / 2) {
350 CERROR("%s: cannot set max_read_ahead_mb=%llu > totalram/2=%luMB\n",
351 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
352 PAGES_TO_MiB(cfs_totalram_pages() / 2));
356 spin_lock(&sbi->ll_lock);
357 sbi->ll_ra_info.ra_max_pages = pages_number;
358 spin_unlock(&sbi->ll_lock);
362 LUSTRE_RW_ATTR(max_read_ahead_mb);
364 static ssize_t max_read_ahead_per_file_mb_show(struct kobject *kobj,
365 struct attribute *attr,
368 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
371 return scnprintf(buf, PAGE_SIZE, "%lu\n",
372 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
375 static ssize_t max_read_ahead_per_file_mb_store(struct kobject *kobj,
376 struct attribute *attr,
380 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
382 u64 ra_max_file_mb, pages_number;
385 rc = sysfs_memparse(buffer, count, &ra_max_file_mb, "MiB");
389 pages_number = round_up(ra_max_file_mb, 1024 * 1024) >> PAGE_SHIFT;
390 if (pages_number > sbi->ll_ra_info.ra_max_pages) {
391 CERROR("%s: cannot set max_read_ahead_per_file_mb=%llu > max_read_ahead_mb=%lu\n",
392 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
393 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
397 spin_lock(&sbi->ll_lock);
398 sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
399 spin_unlock(&sbi->ll_lock);
403 LUSTRE_RW_ATTR(max_read_ahead_per_file_mb);
405 static ssize_t max_read_ahead_whole_mb_show(struct kobject *kobj,
406 struct attribute *attr, char *buf)
408 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
411 return scnprintf(buf, PAGE_SIZE, "%lu\n",
412 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_read_ahead_whole_pages));
415 static ssize_t max_read_ahead_whole_mb_store(struct kobject *kobj,
416 struct attribute *attr,
417 const char *buffer, size_t count)
419 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
421 u64 ra_max_whole_mb, pages_number;
424 rc = sysfs_memparse(buffer, count, &ra_max_whole_mb, "MiB");
428 pages_number = round_up(ra_max_whole_mb, 1024 * 1024) >> PAGE_SHIFT;
429 /* Cap this at the current max readahead window size, the readahead
430 * algorithm does this anyway so it's pointless to set it larger.
432 if (pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
433 CERROR("%s: cannot set max_read_ahead_whole_mb=%llu > max_read_ahead_per_file_mb=%lu\n",
434 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
435 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
440 spin_lock(&sbi->ll_lock);
441 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
442 spin_unlock(&sbi->ll_lock);
446 LUSTRE_RW_ATTR(max_read_ahead_whole_mb);
448 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
450 struct super_block *sb = m->private;
451 struct ll_sb_info *sbi = ll_s2sbi(sb);
452 struct cl_client_cache *cache = sbi->ll_cache;
456 mutex_lock(&cache->ccc_max_cache_mb_lock);
457 max_cached_mb = PAGES_TO_MiB(cache->ccc_lru_max);
458 unused_mb = PAGES_TO_MiB(atomic_long_read(&cache->ccc_lru_left));
459 mutex_unlock(&cache->ccc_max_cache_mb_lock);
460 seq_printf(m, "users: %d\n"
461 "max_cached_mb: %ld\n"
464 "reclaim_count: %u\n",
465 atomic_read(&cache->ccc_users),
467 max_cached_mb - unused_mb,
469 cache->ccc_lru_shrinkers);
473 static ssize_t ll_max_cached_mb_seq_write(struct file *file,
474 const char __user *buffer,
475 size_t count, loff_t *off)
477 struct seq_file *m = file->private_data;
478 struct super_block *sb = m->private;
479 struct ll_sb_info *sbi = ll_s2sbi(sb);
480 struct cl_client_cache *cache = sbi->ll_cache;
487 char kernbuf[128], *ptr;
490 if (count >= sizeof(kernbuf))
493 if (copy_from_user(kernbuf, buffer, count))
495 kernbuf[count] = '\0';
497 ptr = lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count);
498 rc = sysfs_memparse(ptr, count, &pages_number, "MiB");
502 pages_number >>= PAGE_SHIFT;
504 if (pages_number < 0 || pages_number > cfs_totalram_pages()) {
505 CERROR("%s: can't set max cache more than %lu MB\n",
507 PAGES_TO_MiB(cfs_totalram_pages()));
510 /* Allow enough cache so clients can make well-formed RPCs */
511 pages_number = max_t(long, pages_number, PTLRPC_MAX_BRW_PAGES);
513 mutex_lock(&cache->ccc_max_cache_mb_lock);
514 diff = pages_number - cache->ccc_lru_max;
516 /* easy - add more LRU slots. */
518 atomic_long_add(diff, &cache->ccc_lru_left);
522 env = cl_env_get(&refcheck);
524 GOTO(out_unlock, rc = PTR_ERR(env));
530 /* reduce LRU budget from free slots. */
532 long lru_left_old, lru_left_new, lru_left_ret;
534 lru_left_old = atomic_long_read(&cache->ccc_lru_left);
535 if (lru_left_old == 0)
538 lru_left_new = lru_left_old > diff ?
539 lru_left_old - diff : 0;
541 atomic_long_cmpxchg(&cache->ccc_lru_left,
544 if (likely(lru_left_old == lru_left_ret)) {
545 diff -= lru_left_old - lru_left_new;
546 nrpages += lru_left_old - lru_left_new;
554 if (sbi->ll_dt_exp == NULL) { /* being initialized */
559 /* Request extra free slots to avoid them all being used
560 * by other processes before this can continue shrinking.
562 tmp = diff + min_t(long, diff, MiB_TO_PAGES(1024));
563 /* difficult - have to ask OSCs to drop LRU slots. */
564 rc = obd_set_info_async(env, sbi->ll_dt_exp,
565 sizeof(KEY_CACHE_LRU_SHRINK),
566 KEY_CACHE_LRU_SHRINK,
567 sizeof(tmp), &tmp, NULL);
571 cl_env_put(env, &refcheck);
575 cache->ccc_lru_max = pages_number;
578 atomic_long_add(nrpages, &cache->ccc_lru_left);
581 mutex_unlock(&cache->ccc_max_cache_mb_lock);
584 LDEBUGFS_SEQ_FOPS(ll_max_cached_mb);
586 static ssize_t checksums_show(struct kobject *kobj, struct attribute *attr,
589 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
592 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
595 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
596 const char *buffer, size_t count)
598 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
608 rc = kstrtobool(buffer, &val);
612 sbi->ll_flags |= LL_SBI_CHECKSUM;
614 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
617 rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
618 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
620 CWARN("Failed to set OSC checksum flags: %d\n", rc);
624 LUSTRE_RW_ATTR(checksums);
626 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
628 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
629 enum stats_track_type type)
631 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
634 if (sbi->ll_stats_track_type == type)
635 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
636 else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
637 return sprintf(buf, "0 (all)\n");
639 return sprintf(buf, "untracked\n");
642 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
643 size_t count, enum stats_track_type type)
645 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
650 rc = kstrtoul(buffer, 10, &pid);
654 sbi->ll_stats_track_id = pid;
656 sbi->ll_stats_track_type = STATS_TRACK_ALL;
658 sbi->ll_stats_track_type = type;
659 lprocfs_clear_stats(sbi->ll_stats);
663 static ssize_t stats_track_pid_show(struct kobject *kobj,
664 struct attribute *attr,
667 return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
670 static ssize_t stats_track_pid_store(struct kobject *kobj,
671 struct attribute *attr,
675 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
677 LUSTRE_RW_ATTR(stats_track_pid);
679 static ssize_t stats_track_ppid_show(struct kobject *kobj,
680 struct attribute *attr,
683 return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
686 static ssize_t stats_track_ppid_store(struct kobject *kobj,
687 struct attribute *attr,
691 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
693 LUSTRE_RW_ATTR(stats_track_ppid);
695 static ssize_t stats_track_gid_show(struct kobject *kobj,
696 struct attribute *attr,
699 return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
702 static ssize_t stats_track_gid_store(struct kobject *kobj,
703 struct attribute *attr,
707 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
709 LUSTRE_RW_ATTR(stats_track_gid);
711 static ssize_t statahead_running_max_show(struct kobject *kobj,
712 struct attribute *attr,
715 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
718 return snprintf(buf, 16, "%u\n", sbi->ll_sa_running_max);
721 static ssize_t statahead_running_max_store(struct kobject *kobj,
722 struct attribute *attr,
726 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
731 rc = kstrtoul(buffer, 0, &val);
735 if (val <= LL_SA_RUNNING_MAX) {
736 sbi->ll_sa_running_max = val;
740 CERROR("Bad statahead_running_max value %lu. Valid values "
741 "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
745 LUSTRE_RW_ATTR(statahead_running_max);
747 static ssize_t statahead_max_show(struct kobject *kobj,
748 struct attribute *attr,
751 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
754 return sprintf(buf, "%u\n", sbi->ll_sa_max);
757 static ssize_t statahead_max_store(struct kobject *kobj,
758 struct attribute *attr,
762 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
767 rc = kstrtoul(buffer, 0, &val);
771 if (val <= LL_SA_RPC_MAX)
772 sbi->ll_sa_max = val;
774 CERROR("Bad statahead_max value %lu. Valid values are in the range [0, %d]\n",
779 LUSTRE_RW_ATTR(statahead_max);
781 static ssize_t statahead_agl_show(struct kobject *kobj,
782 struct attribute *attr,
785 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
788 return sprintf(buf, "%u\n", sbi->ll_flags & LL_SBI_AGL_ENABLED ? 1 : 0);
791 static ssize_t statahead_agl_store(struct kobject *kobj,
792 struct attribute *attr,
796 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
801 rc = kstrtobool(buffer, &val);
806 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
808 sbi->ll_flags &= ~LL_SBI_AGL_ENABLED;
812 LUSTRE_RW_ATTR(statahead_agl);
814 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
816 struct super_block *sb = m->private;
817 struct ll_sb_info *sbi = ll_s2sbi(sb);
819 seq_printf(m, "statahead total: %u\n"
820 "statahead wrong: %u\n"
822 atomic_read(&sbi->ll_sa_total),
823 atomic_read(&sbi->ll_sa_wrong),
824 atomic_read(&sbi->ll_agl_total));
828 LDEBUGFS_SEQ_FOPS_RO(ll_statahead_stats);
830 static ssize_t lazystatfs_show(struct kobject *kobj,
831 struct attribute *attr,
834 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
837 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
840 static ssize_t lazystatfs_store(struct kobject *kobj,
841 struct attribute *attr,
845 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
850 rc = kstrtobool(buffer, &val);
855 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
857 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
861 LUSTRE_RW_ATTR(lazystatfs);
863 static ssize_t statfs_max_age_show(struct kobject *kobj, struct attribute *attr,
866 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
869 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_statfs_max_age);
872 static ssize_t statfs_max_age_store(struct kobject *kobj,
873 struct attribute *attr, const char *buffer,
876 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
881 rc = kstrtouint(buffer, 10, &val);
884 if (val > OBD_STATFS_CACHE_MAX_AGE)
887 sbi->ll_statfs_max_age = val;
891 LUSTRE_RW_ATTR(statfs_max_age);
893 static ssize_t max_easize_show(struct kobject *kobj,
894 struct attribute *attr,
897 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
902 rc = ll_get_max_mdsize(sbi, &ealen);
906 /* Limit xattr size returned to userspace based on kernel maximum */
907 return snprintf(buf, PAGE_SIZE, "%u\n",
908 ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
910 LUSTRE_RO_ATTR(max_easize);
913 * Get default_easize.
915 * \see client_obd::cl_default_mds_easize
917 * \param[in] m seq_file handle
918 * \param[in] v unused for single entry
920 * \retval 0 on success
921 * \retval negative negated errno on failure
923 static ssize_t default_easize_show(struct kobject *kobj,
924 struct attribute *attr,
927 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
932 rc = ll_get_default_mdsize(sbi, &ealen);
936 /* Limit xattr size returned to userspace based on kernel maximum */
937 return snprintf(buf, PAGE_SIZE, "%u\n",
938 ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
942 * Set default_easize.
944 * Range checking on the passed value is handled by
945 * ll_set_default_mdsize().
947 * \see client_obd::cl_default_mds_easize
949 * \param[in] file proc file
950 * \param[in] buffer string passed from user space
951 * \param[in] count \a buffer length
952 * \param[in] off unused for single entry
954 * \retval positive \a count on success
955 * \retval negative negated errno on failure
957 static ssize_t default_easize_store(struct kobject *kobj,
958 struct attribute *attr,
962 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
970 rc = kstrtouint(buffer, 10, &val);
974 rc = ll_set_default_mdsize(sbi, val);
980 LUSTRE_RW_ATTR(default_easize);
982 static int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
984 const char *str[] = LL_SBI_FLAGS;
985 struct super_block *sb = m->private;
986 int flags = ll_s2sbi(sb)->ll_flags;
990 if (ARRAY_SIZE(str) <= i) {
991 CERROR("%s: Revise array LL_SBI_FLAGS to match sbi "
992 "flags please.\n", ll_s2sbi(sb)->ll_fsname);
997 seq_printf(m, "%s ", str[i]);
1001 seq_printf(m, "\b\n");
1005 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
1007 static ssize_t xattr_cache_show(struct kobject *kobj,
1008 struct attribute *attr,
1011 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1014 return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
1017 static ssize_t xattr_cache_store(struct kobject *kobj,
1018 struct attribute *attr,
1022 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1027 rc = kstrtobool(buffer, &val);
1031 if (val && !(sbi->ll_flags & LL_SBI_XATTR_CACHE))
1034 sbi->ll_xattr_cache_enabled = val;
1035 sbi->ll_xattr_cache_set = 1;
1039 LUSTRE_RW_ATTR(xattr_cache);
1041 static ssize_t tiny_write_show(struct kobject *kobj,
1042 struct attribute *attr,
1045 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1048 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_TINY_WRITE));
1051 static ssize_t tiny_write_store(struct kobject *kobj,
1052 struct attribute *attr,
1056 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1061 rc = kstrtobool(buffer, &val);
1065 spin_lock(&sbi->ll_lock);
1067 sbi->ll_flags |= LL_SBI_TINY_WRITE;
1069 sbi->ll_flags &= ~LL_SBI_TINY_WRITE;
1070 spin_unlock(&sbi->ll_lock);
1074 LUSTRE_RW_ATTR(tiny_write);
1076 static ssize_t max_read_ahead_async_active_show(struct kobject *kobj,
1077 struct attribute *attr,
1080 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1083 return snprintf(buf, PAGE_SIZE, "%u\n",
1084 sbi->ll_ra_info.ra_async_max_active);
1087 static ssize_t max_read_ahead_async_active_store(struct kobject *kobj,
1088 struct attribute *attr,
1092 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1097 rc = kstrtouint(buffer, 10, &val);
1102 * It doesn't make any sense to make it exceed what
1103 * workqueue could acutally support. This can easily
1104 * over subscripe the cores but Lustre internally
1105 * throttles to avoid those impacts.
1107 if (val > WQ_UNBOUND_MAX_ACTIVE) {
1108 CERROR("%s: cannot set max_read_ahead_async_active=%u larger than %u\n",
1109 sbi->ll_fsname, val, WQ_UNBOUND_MAX_ACTIVE);
1113 spin_lock(&sbi->ll_lock);
1114 sbi->ll_ra_info.ra_async_max_active = val;
1115 spin_unlock(&sbi->ll_lock);
1119 LUSTRE_RW_ATTR(max_read_ahead_async_active);
1121 static ssize_t read_ahead_async_file_threshold_mb_show(struct kobject *kobj,
1122 struct attribute *attr,
1125 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1128 return snprintf(buf, PAGE_SIZE, "%lu\n",
1129 PAGES_TO_MiB(sbi->ll_ra_info.ra_async_pages_per_file_threshold));
1133 read_ahead_async_file_threshold_mb_store(struct kobject *kobj,
1134 struct attribute *attr,
1135 const char *buffer, size_t count)
1137 unsigned long pages_number;
1138 unsigned long max_ra_per_file;
1139 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1143 rc = kstrtoul(buffer, 10, &pages_number);
1147 pages_number = MiB_TO_PAGES(pages_number);
1148 max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1149 if (pages_number < 0 || pages_number > max_ra_per_file) {
1150 CERROR("%s: can't set read_ahead_async_file_threshold_mb=%lu > "
1151 "max_read_readahead_per_file_mb=%lu\n", sbi->ll_fsname,
1152 PAGES_TO_MiB(pages_number),
1153 PAGES_TO_MiB(max_ra_per_file));
1156 sbi->ll_ra_info.ra_async_pages_per_file_threshold = pages_number;
1160 LUSTRE_RW_ATTR(read_ahead_async_file_threshold_mb);
1162 static ssize_t read_ahead_range_kb_show(struct kobject *kobj,
1163 struct attribute *attr,char *buf)
1165 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1168 return snprintf(buf, PAGE_SIZE, "%lu\n",
1169 sbi->ll_ra_info.ra_range_pages << (PAGE_SHIFT - 10));
1173 read_ahead_range_kb_store(struct kobject *kobj,
1174 struct attribute *attr,
1175 const char *buffer, size_t count)
1177 unsigned long pages_number;
1178 unsigned long max_ra_per_file;
1180 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1184 rc = sysfs_memparse(buffer, count, &val, "KiB");
1188 pages_number = val >> PAGE_SHIFT;
1189 /* Disable mmap range read */
1190 if (pages_number == 0)
1193 max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1194 if (pages_number > max_ra_per_file ||
1195 pages_number < RA_MIN_MMAP_RANGE_PAGES)
1199 spin_lock(&sbi->ll_lock);
1200 sbi->ll_ra_info.ra_range_pages = pages_number;
1201 spin_unlock(&sbi->ll_lock);
1205 LUSTRE_RW_ATTR(read_ahead_range_kb);
1207 static ssize_t fast_read_show(struct kobject *kobj,
1208 struct attribute *attr,
1211 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1214 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_FAST_READ));
1217 static ssize_t fast_read_store(struct kobject *kobj,
1218 struct attribute *attr,
1222 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1227 rc = kstrtobool(buffer, &val);
1231 spin_lock(&sbi->ll_lock);
1233 sbi->ll_flags |= LL_SBI_FAST_READ;
1235 sbi->ll_flags &= ~LL_SBI_FAST_READ;
1236 spin_unlock(&sbi->ll_lock);
1240 LUSTRE_RW_ATTR(fast_read);
1242 static ssize_t file_heat_show(struct kobject *kobj,
1243 struct attribute *attr,
1246 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1249 return snprintf(buf, PAGE_SIZE, "%u\n",
1250 !!(sbi->ll_flags & LL_SBI_FILE_HEAT));
1253 static ssize_t file_heat_store(struct kobject *kobj,
1254 struct attribute *attr,
1258 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1263 rc = kstrtobool(buffer, &val);
1267 spin_lock(&sbi->ll_lock);
1269 sbi->ll_flags |= LL_SBI_FILE_HEAT;
1271 sbi->ll_flags &= ~LL_SBI_FILE_HEAT;
1272 spin_unlock(&sbi->ll_lock);
1276 LUSTRE_RW_ATTR(file_heat);
1278 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1279 struct attribute *attr,
1282 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1285 return snprintf(buf, PAGE_SIZE, "%u\n",
1286 (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1289 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1290 struct attribute *attr,
1294 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1299 rc = kstrtoul(buffer, 10, &val);
1303 if (val < 0 || val > 100)
1306 sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1310 LUSTRE_RW_ATTR(heat_decay_percentage);
1312 static ssize_t heat_period_second_show(struct kobject *kobj,
1313 struct attribute *attr,
1316 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1319 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1322 static ssize_t heat_period_second_store(struct kobject *kobj,
1323 struct attribute *attr,
1327 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1332 rc = kstrtoul(buffer, 10, &val);
1339 sbi->ll_heat_period_second = val;
1343 LUSTRE_RW_ATTR(heat_period_second);
1345 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1347 struct super_block *sb = m->private;
1348 struct ll_sb_info *sbi = ll_s2sbi(sb);
1349 struct cl_client_cache *cache = sbi->ll_cache;
1353 pages = atomic_long_read(&cache->ccc_unstable_nr);
1354 mb = (pages * PAGE_SIZE) >> 20;
1356 seq_printf(m, "unstable_check: %8d\n"
1357 "unstable_pages: %12ld\n"
1358 "unstable_mb: %8d\n",
1359 cache->ccc_unstable_check, pages, mb);
1363 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1364 const char __user *buffer,
1365 size_t count, loff_t *unused)
1367 struct seq_file *seq = file->private_data;
1368 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1375 if (count >= sizeof(kernbuf))
1378 if (copy_from_user(kernbuf, buffer, count))
1382 buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1384 rc = kstrtobool_from_user(buffer, count, &val);
1388 /* borrow lru lock to set the value */
1389 spin_lock(&sbi->ll_cache->ccc_lru_lock);
1390 sbi->ll_cache->ccc_unstable_check = val;
1391 spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1396 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1398 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1400 struct super_block *sb = m->private;
1401 struct ll_sb_info *sbi = ll_s2sbi(sb);
1402 struct root_squash_info *squash = &sbi->ll_squash;
1404 seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1408 static ssize_t ll_root_squash_seq_write(struct file *file,
1409 const char __user *buffer,
1410 size_t count, loff_t *off)
1412 struct seq_file *m = file->private_data;
1413 struct super_block *sb = m->private;
1414 struct ll_sb_info *sbi = ll_s2sbi(sb);
1415 struct root_squash_info *squash = &sbi->ll_squash;
1417 return lprocfs_wr_root_squash(buffer, count, squash, sbi->ll_fsname);
1420 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1422 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1424 struct super_block *sb = m->private;
1425 struct ll_sb_info *sbi = ll_s2sbi(sb);
1426 struct root_squash_info *squash = &sbi->ll_squash;
1429 spin_lock(&squash->rsi_lock);
1430 if (!list_empty(&squash->rsi_nosquash_nids)) {
1431 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1432 &squash->rsi_nosquash_nids);
1436 seq_puts(m, "NONE\n");
1438 spin_unlock(&squash->rsi_lock);
1443 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1444 const char __user *buffer,
1445 size_t count, loff_t *off)
1447 struct seq_file *m = file->private_data;
1448 struct super_block *sb = m->private;
1449 struct ll_sb_info *sbi = ll_s2sbi(sb);
1450 struct root_squash_info *squash = &sbi->ll_squash;
1453 rc = lprocfs_wr_nosquash_nids(buffer, count, squash, sbi->ll_fsname);
1457 ll_compute_rootsquash_state(sbi);
1462 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1464 static int ll_pcc_seq_show(struct seq_file *m, void *v)
1466 struct super_block *sb = m->private;
1467 struct ll_sb_info *sbi = ll_s2sbi(sb);
1469 return pcc_super_dump(&sbi->ll_pcc_super, m);
1472 static ssize_t ll_pcc_seq_write(struct file *file, const char __user *buffer,
1473 size_t count, loff_t *off)
1475 struct seq_file *m = file->private_data;
1476 struct super_block *sb = m->private;
1477 struct ll_sb_info *sbi = ll_s2sbi(sb);
1481 if (count >= LPROCFS_WR_PCC_MAX_CMD)
1484 if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_PCC))
1487 OBD_ALLOC(kernbuf, count + 1);
1488 if (kernbuf == NULL)
1491 if (copy_from_user(kernbuf, buffer, count))
1492 GOTO(out_free_kernbuff, rc = -EFAULT);
1494 rc = pcc_cmd_handle(kernbuf, count, &sbi->ll_pcc_super);
1496 OBD_FREE(kernbuf, count + 1);
1497 return rc ? rc : count;
1499 LDEBUGFS_SEQ_FOPS(ll_pcc);
1501 struct ldebugfs_vars lprocfs_llite_obd_vars[] = {
1503 .fops = &ll_site_stats_fops },
1504 { .name = "max_cached_mb",
1505 .fops = &ll_max_cached_mb_fops },
1506 { .name = "statahead_stats",
1507 .fops = &ll_statahead_stats_fops },
1508 { .name = "unstable_stats",
1509 .fops = &ll_unstable_stats_fops },
1510 { .name = "sbi_flags",
1511 .fops = &ll_sbi_flags_fops },
1512 { .name = "root_squash",
1513 .fops = &ll_root_squash_fops },
1514 { .name = "nosquash_nids",
1515 .fops = &ll_nosquash_nids_fops },
1517 .fops = &ll_pcc_fops, },
1521 #define MAX_STRING_SIZE 128
1523 static struct attribute *llite_attrs[] = {
1524 &lustre_attr_blocksize.attr,
1525 &lustre_attr_stat_blocksize.attr,
1526 &lustre_attr_kbytestotal.attr,
1527 &lustre_attr_kbytesfree.attr,
1528 &lustre_attr_kbytesavail.attr,
1529 &lustre_attr_filestotal.attr,
1530 &lustre_attr_filesfree.attr,
1531 &lustre_attr_client_type.attr,
1532 &lustre_attr_fstype.attr,
1533 &lustre_attr_uuid.attr,
1534 &lustre_attr_checksums.attr,
1535 &lustre_attr_checksum_pages.attr,
1536 &lustre_attr_max_read_ahead_mb.attr,
1537 &lustre_attr_max_read_ahead_per_file_mb.attr,
1538 &lustre_attr_max_read_ahead_whole_mb.attr,
1539 &lustre_attr_max_read_ahead_async_active.attr,
1540 &lustre_attr_read_ahead_async_file_threshold_mb.attr,
1541 &lustre_attr_read_ahead_range_kb.attr,
1542 &lustre_attr_stats_track_pid.attr,
1543 &lustre_attr_stats_track_ppid.attr,
1544 &lustre_attr_stats_track_gid.attr,
1545 &lustre_attr_statahead_running_max.attr,
1546 &lustre_attr_statahead_max.attr,
1547 &lustre_attr_statahead_agl.attr,
1548 &lustre_attr_lazystatfs.attr,
1549 &lustre_attr_statfs_max_age.attr,
1550 &lustre_attr_max_easize.attr,
1551 &lustre_attr_default_easize.attr,
1552 &lustre_attr_xattr_cache.attr,
1553 &lustre_attr_fast_read.attr,
1554 &lustre_attr_tiny_write.attr,
1555 &lustre_attr_file_heat.attr,
1556 &lustre_attr_heat_decay_percentage.attr,
1557 &lustre_attr_heat_period_second.attr,
1561 static void sbi_kobj_release(struct kobject *kobj)
1563 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1565 complete(&sbi->ll_kobj_unregister);
1568 static struct kobj_type sbi_ktype = {
1569 .default_attrs = llite_attrs,
1570 .sysfs_ops = &lustre_sysfs_ops,
1571 .release = sbi_kobj_release,
1574 static const struct llite_file_opcode {
1578 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1579 /* file operation */
1580 { LPROC_LL_READ_BYTES, LPROCFS_TYPE_BYTES_FULL, "read_bytes" },
1581 { LPROC_LL_WRITE_BYTES, LPROCFS_TYPE_BYTES_FULL, "write_bytes" },
1582 { LPROC_LL_READ, LPROCFS_TYPE_LATENCY, "read" },
1583 { LPROC_LL_WRITE, LPROCFS_TYPE_LATENCY, "write" },
1584 { LPROC_LL_IOCTL, LPROCFS_TYPE_REQS, "ioctl" },
1585 { LPROC_LL_OPEN, LPROCFS_TYPE_LATENCY, "open" },
1586 { LPROC_LL_RELEASE, LPROCFS_TYPE_LATENCY, "close" },
1587 { LPROC_LL_MMAP, LPROCFS_TYPE_LATENCY, "mmap" },
1588 { LPROC_LL_FAULT, LPROCFS_TYPE_LATENCY, "page_fault" },
1589 { LPROC_LL_MKWRITE, LPROCFS_TYPE_LATENCY, "page_mkwrite" },
1590 { LPROC_LL_LLSEEK, LPROCFS_TYPE_LATENCY, "seek" },
1591 { LPROC_LL_FSYNC, LPROCFS_TYPE_LATENCY, "fsync" },
1592 { LPROC_LL_READDIR, LPROCFS_TYPE_LATENCY, "readdir" },
1593 /* inode operation */
1594 { LPROC_LL_SETATTR, LPROCFS_TYPE_LATENCY, "setattr" },
1595 { LPROC_LL_TRUNC, LPROCFS_TYPE_LATENCY, "truncate" },
1596 { LPROC_LL_FLOCK, LPROCFS_TYPE_LATENCY, "flock" },
1597 { LPROC_LL_GETATTR, LPROCFS_TYPE_LATENCY, "getattr" },
1598 { LPROC_LL_FALLOCATE, LPROCFS_TYPE_LATENCY, "fallocate"},
1599 /* dir inode operation */
1600 { LPROC_LL_CREATE, LPROCFS_TYPE_LATENCY, "create" },
1601 { LPROC_LL_LINK, LPROCFS_TYPE_LATENCY, "link" },
1602 { LPROC_LL_UNLINK, LPROCFS_TYPE_LATENCY, "unlink" },
1603 { LPROC_LL_SYMLINK, LPROCFS_TYPE_LATENCY, "symlink" },
1604 { LPROC_LL_MKDIR, LPROCFS_TYPE_LATENCY, "mkdir" },
1605 { LPROC_LL_RMDIR, LPROCFS_TYPE_LATENCY, "rmdir" },
1606 { LPROC_LL_MKNOD, LPROCFS_TYPE_LATENCY, "mknod" },
1607 { LPROC_LL_RENAME, LPROCFS_TYPE_LATENCY, "rename" },
1608 /* special inode operation */
1609 { LPROC_LL_STATFS, LPROCFS_TYPE_LATENCY, "statfs" },
1610 { LPROC_LL_SETXATTR, LPROCFS_TYPE_LATENCY, "setxattr" },
1611 { LPROC_LL_GETXATTR, LPROCFS_TYPE_LATENCY, "getxattr" },
1612 { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REQS, "getxattr_hits" },
1613 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_LATENCY, "listxattr" },
1614 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_LATENCY, "removexattr" },
1615 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_LATENCY, "inode_permission" },
1618 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count)
1623 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1624 lprocfs_counter_add(sbi->ll_stats, op, count);
1625 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1626 sbi->ll_stats_track_id == current->pid)
1627 lprocfs_counter_add(sbi->ll_stats, op, count);
1628 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1629 sbi->ll_stats_track_id == current->parent->pid)
1630 lprocfs_counter_add(sbi->ll_stats, op, count);
1631 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1632 sbi->ll_stats_track_id ==
1633 from_kgid(&init_user_ns, current_gid()))
1634 lprocfs_counter_add(sbi->ll_stats, op, count);
1636 EXPORT_SYMBOL(ll_stats_ops_tally);
1638 static const char *ra_stat_string[] = {
1639 [RA_STAT_HIT] = "hits",
1640 [RA_STAT_MISS] = "misses",
1641 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1642 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1643 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1644 [RA_STAT_FAILED_MATCH] = "failed lock match",
1645 [RA_STAT_DISCARDED] = "read but discarded",
1646 [RA_STAT_ZERO_LEN] = "zero length file",
1647 [RA_STAT_ZERO_WINDOW] = "zero size window",
1648 [RA_STAT_EOF] = "read-ahead to EOF",
1649 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1650 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1651 [RA_STAT_FAILED_REACH_END] = "failed to reach end",
1652 [RA_STAT_ASYNC] = "async readahead",
1653 [RA_STAT_FAILED_FAST_READ] = "failed to fast read",
1654 [RA_STAT_MMAP_RANGE_READ] = "mmap range read",
1657 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1659 struct lustre_sb_info *lsi = s2lsi(sb);
1660 struct ll_sb_info *sbi = ll_s2sbi(sb);
1666 if (IS_ERR_OR_NULL(llite_root))
1669 sbi->ll_debugfs_entry = debugfs_create_dir(name, llite_root);
1670 ldebugfs_add_vars(sbi->ll_debugfs_entry, lprocfs_llite_obd_vars, sb);
1672 debugfs_create_file("dump_page_cache", 0444, sbi->ll_debugfs_entry, sbi,
1673 &vvp_dump_pgcache_file_ops);
1675 debugfs_create_file("extents_stats", 0644, sbi->ll_debugfs_entry, sbi,
1676 &ll_rw_extents_stats_fops);
1678 debugfs_create_file("extents_stats_per_process", 0644,
1679 sbi->ll_debugfs_entry, sbi,
1680 &ll_rw_extents_stats_pp_fops);
1682 debugfs_create_file("offset_stats", 0644, sbi->ll_debugfs_entry, sbi,
1683 &ll_rw_offset_stats_fops);
1685 /* File operations stats */
1686 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1687 LPROCFS_STATS_FLAG_NONE);
1688 if (sbi->ll_stats == NULL)
1689 GOTO(out_debugfs, err = -ENOMEM);
1691 /* do counter init */
1692 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1693 u32 type = llite_opcode_table[id].type;
1694 void *ptr = "unknown";
1696 if (type & LPROCFS_TYPE_REQS)
1698 else if (type & LPROCFS_TYPE_BYTES)
1700 else if (type & LPROCFS_TYPE_USEC)
1702 lprocfs_counter_init(sbi->ll_stats,
1703 llite_opcode_table[id].opcode, type,
1704 llite_opcode_table[id].opname, ptr);
1707 debugfs_create_file("stats", 0644, sbi->ll_debugfs_entry,
1708 sbi->ll_stats, &ldebugfs_stats_seq_fops);
1710 sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1711 LPROCFS_STATS_FLAG_NONE);
1712 if (sbi->ll_ra_stats == NULL)
1713 GOTO(out_stats, err = -ENOMEM);
1715 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1716 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1717 ra_stat_string[id], "pages");
1719 debugfs_create_file("read_ahead_stats", 0644, sbi->ll_debugfs_entry,
1720 sbi->ll_ra_stats, &ldebugfs_stats_seq_fops);
1723 /* Yes we also register sysfs mount kset here as well */
1724 sbi->ll_kset.kobj.parent = llite_kobj;
1725 sbi->ll_kset.kobj.ktype = &sbi_ktype;
1726 init_completion(&sbi->ll_kobj_unregister);
1727 err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1729 GOTO(out_ra_stats, err);
1731 err = kset_register(&sbi->ll_kset);
1733 GOTO(out_ra_stats, err);
1735 lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1739 lprocfs_free_stats(&sbi->ll_ra_stats);
1741 lprocfs_free_stats(&sbi->ll_stats);
1743 debugfs_remove_recursive(sbi->ll_debugfs_entry);
1748 void ll_debugfs_unregister_super(struct super_block *sb)
1750 struct lustre_sb_info *lsi = s2lsi(sb);
1751 struct ll_sb_info *sbi = ll_s2sbi(sb);
1753 debugfs_remove_recursive(sbi->ll_debugfs_entry);
1756 sysfs_remove_link(&sbi->ll_kset.kobj,
1757 sbi->ll_dt_obd->obd_type->typ_name);
1760 sysfs_remove_link(&sbi->ll_kset.kobj,
1761 sbi->ll_md_obd->obd_type->typ_name);
1763 kobject_put(lsi->lsi_kobj);
1765 kset_unregister(&sbi->ll_kset);
1766 wait_for_completion(&sbi->ll_kobj_unregister);
1768 lprocfs_free_stats(&sbi->ll_ra_stats);
1769 lprocfs_free_stats(&sbi->ll_stats);
1771 #undef MAX_STRING_SIZE
1773 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1774 struct seq_file *seq, int which)
1776 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1777 unsigned long start, end, r, w;
1778 char *unitp = "KMGTPEZY";
1780 struct per_process_info *pp_info = &io_extents->pp_extents[which];
1786 for(i = 0; i < LL_HIST_MAX; i++) {
1787 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1788 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1791 for(i = 0; i < LL_HIST_MAX; i++) {
1792 r = pp_info->pp_r_hist.oh_buckets[i];
1793 w = pp_info->pp_w_hist.oh_buckets[i];
1796 end = 1 << (i + LL_HIST_START - units);
1797 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u | "
1798 "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1799 (i == LL_HIST_MAX - 1) ? '+' : ' ',
1800 r, pct(r, read_tot), pct(read_cum, read_tot),
1801 w, pct(w, write_tot), pct(write_cum, write_tot));
1803 if (start == (1 << 10)) {
1808 if (read_cum == read_tot && write_cum == write_tot)
1813 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1815 struct timespec64 now;
1816 struct ll_sb_info *sbi = seq->private;
1817 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1820 ktime_get_real_ts64(&now);
1822 if (!sbi->ll_rw_stats_on) {
1823 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1826 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1827 (s64)now.tv_sec, now.tv_nsec);
1828 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1829 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1830 "extents", "calls", "%", "cum%",
1831 "calls", "%", "cum%");
1832 spin_lock(&sbi->ll_pp_extent_lock);
1833 for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1834 if (io_extents->pp_extents[k].pid != 0) {
1835 seq_printf(seq, "\nPID: %d\n",
1836 io_extents->pp_extents[k].pid);
1837 ll_display_extents_info(io_extents, seq, k);
1840 spin_unlock(&sbi->ll_pp_extent_lock);
1844 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1845 const char __user *buf,
1849 struct seq_file *seq = file->private_data;
1850 struct ll_sb_info *sbi = seq->private;
1851 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1858 value = ll_stats_pid_write(buf, len);
1861 sbi->ll_rw_stats_on = 0;
1863 sbi->ll_rw_stats_on = 1;
1865 spin_lock(&sbi->ll_pp_extent_lock);
1866 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1867 io_extents->pp_extents[i].pid = 0;
1868 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1869 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1871 spin_unlock(&sbi->ll_pp_extent_lock);
1875 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1877 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1879 struct timespec64 now;
1880 struct ll_sb_info *sbi = seq->private;
1881 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1883 ktime_get_real_ts64(&now);
1885 if (!sbi->ll_rw_stats_on) {
1886 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1889 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1890 (s64)now.tv_sec, now.tv_nsec);
1892 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1893 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1894 "extents", "calls", "%", "cum%",
1895 "calls", "%", "cum%");
1896 spin_lock(&sbi->ll_lock);
1897 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1898 spin_unlock(&sbi->ll_lock);
1903 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1904 const char __user *buf,
1905 size_t len, loff_t *off)
1907 struct seq_file *seq = file->private_data;
1908 struct ll_sb_info *sbi = seq->private;
1909 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1916 value = ll_stats_pid_write(buf, len);
1919 sbi->ll_rw_stats_on = 0;
1921 sbi->ll_rw_stats_on = 1;
1923 spin_lock(&sbi->ll_pp_extent_lock);
1924 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1925 io_extents->pp_extents[i].pid = 0;
1926 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1927 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1929 spin_unlock(&sbi->ll_pp_extent_lock);
1934 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1936 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1937 struct ll_file_data *file, loff_t pos,
1938 size_t count, int rw)
1941 struct ll_rw_process_info *process;
1942 struct ll_rw_process_info *offset;
1943 int *off_count = &sbi->ll_rw_offset_entry_count;
1944 int *process_count = &sbi->ll_offset_process_count;
1945 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1947 if(!sbi->ll_rw_stats_on)
1949 process = sbi->ll_rw_process_info;
1950 offset = sbi->ll_rw_offset_info;
1952 spin_lock(&sbi->ll_pp_extent_lock);
1953 /* Extent statistics */
1954 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1955 if(io_extents->pp_extents[i].pid == pid) {
1963 sbi->ll_extent_process_count =
1964 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1965 cur = sbi->ll_extent_process_count;
1966 io_extents->pp_extents[cur].pid = pid;
1967 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1968 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1971 for (i = 0; (count >= 1 << (LL_HIST_START + i)) &&
1972 (i < (LL_HIST_MAX - 1)); i++);
1974 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1975 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1977 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1978 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1980 spin_unlock(&sbi->ll_pp_extent_lock);
1982 spin_lock(&sbi->ll_process_lock);
1983 /* Offset statistics */
1984 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1985 if (process[i].rw_pid == pid) {
1986 if (process[i].rw_last_file != file) {
1987 process[i].rw_range_start = pos;
1988 process[i].rw_last_file_pos = pos + count;
1989 process[i].rw_smallest_extent = count;
1990 process[i].rw_largest_extent = count;
1991 process[i].rw_offset = 0;
1992 process[i].rw_last_file = file;
1993 spin_unlock(&sbi->ll_process_lock);
1996 if (process[i].rw_last_file_pos != pos) {
1998 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1999 offset[*off_count].rw_op = process[i].rw_op;
2000 offset[*off_count].rw_pid = pid;
2001 offset[*off_count].rw_range_start =
2002 process[i].rw_range_start;
2003 offset[*off_count].rw_range_end =
2004 process[i].rw_last_file_pos;
2005 offset[*off_count].rw_smallest_extent =
2006 process[i].rw_smallest_extent;
2007 offset[*off_count].rw_largest_extent =
2008 process[i].rw_largest_extent;
2009 offset[*off_count].rw_offset =
2010 process[i].rw_offset;
2011 process[i].rw_op = rw;
2012 process[i].rw_range_start = pos;
2013 process[i].rw_smallest_extent = count;
2014 process[i].rw_largest_extent = count;
2015 process[i].rw_offset = pos -
2016 process[i].rw_last_file_pos;
2018 if(process[i].rw_smallest_extent > count)
2019 process[i].rw_smallest_extent = count;
2020 if(process[i].rw_largest_extent < count)
2021 process[i].rw_largest_extent = count;
2022 process[i].rw_last_file_pos = pos + count;
2023 spin_unlock(&sbi->ll_process_lock);
2027 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
2028 process[*process_count].rw_pid = pid;
2029 process[*process_count].rw_op = rw;
2030 process[*process_count].rw_range_start = pos;
2031 process[*process_count].rw_last_file_pos = pos + count;
2032 process[*process_count].rw_smallest_extent = count;
2033 process[*process_count].rw_largest_extent = count;
2034 process[*process_count].rw_offset = 0;
2035 process[*process_count].rw_last_file = file;
2036 spin_unlock(&sbi->ll_process_lock);
2039 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
2041 struct timespec64 now;
2042 struct ll_sb_info *sbi = seq->private;
2043 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
2044 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
2047 ktime_get_real_ts64(&now);
2049 if (!sbi->ll_rw_stats_on) {
2050 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2053 spin_lock(&sbi->ll_process_lock);
2055 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
2056 (s64)now.tv_sec, now.tv_nsec);
2057 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
2058 "R/W", "PID", "RANGE START", "RANGE END",
2059 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
2061 /* We stored the discontiguous offsets here; print them first */
2062 for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
2063 if (offset[i].rw_pid != 0)
2065 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2066 offset[i].rw_op == READ ? 'R' : 'W',
2068 offset[i].rw_range_start,
2069 offset[i].rw_range_end,
2070 (unsigned long)offset[i].rw_smallest_extent,
2071 (unsigned long)offset[i].rw_largest_extent,
2072 offset[i].rw_offset);
2075 /* Then print the current offsets for each process */
2076 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2077 if (process[i].rw_pid != 0)
2079 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2080 process[i].rw_op == READ ? 'R' : 'W',
2082 process[i].rw_range_start,
2083 process[i].rw_last_file_pos,
2084 (unsigned long)process[i].rw_smallest_extent,
2085 (unsigned long)process[i].rw_largest_extent,
2086 process[i].rw_offset);
2088 spin_unlock(&sbi->ll_process_lock);
2093 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
2094 const char __user *buf,
2095 size_t len, loff_t *off)
2097 struct seq_file *seq = file->private_data;
2098 struct ll_sb_info *sbi = seq->private;
2099 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
2100 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
2106 value = ll_stats_pid_write(buf, len);
2109 sbi->ll_rw_stats_on = 0;
2111 sbi->ll_rw_stats_on = 1;
2113 spin_lock(&sbi->ll_process_lock);
2114 sbi->ll_offset_process_count = 0;
2115 sbi->ll_rw_offset_entry_count = 0;
2116 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
2117 LL_PROCESS_HIST_MAX);
2118 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
2119 LL_OFFSET_HIST_MAX);
2120 spin_unlock(&sbi->ll_process_lock);
2125 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);