4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 #define DEBUG_SUBSYSTEM S_LLITE
34 #include <linux/version.h>
35 #include <linux/user_namespace.h>
36 #include <linux/uidgid.h>
38 #include <uapi/linux/lustre/lustre_param.h>
39 #include <lprocfs_status.h>
40 #include <obd_support.h>
42 #include "llite_internal.h"
43 #include "vvp_internal.h"
45 static struct kobject *llite_kobj;
46 static struct dentry *llite_root;
48 static void llite_kobj_release(struct kobject *kobj)
50 if (!IS_ERR_OR_NULL(llite_root)) {
51 debugfs_remove(llite_root);
58 static struct kobj_type llite_kobj_ktype = {
59 .release = llite_kobj_release,
60 .sysfs_ops = &lustre_sysfs_ops,
63 int llite_tunables_register(void)
67 llite_kobj = kzalloc(sizeof(*llite_kobj), GFP_KERNEL);
71 llite_kobj->kset = lustre_kset;
72 rc = kobject_init_and_add(llite_kobj, &llite_kobj_ktype,
73 &lustre_kset->kobj, "%s", "llite");
77 llite_root = debugfs_create_dir("llite", debugfs_lustre_root);
81 kobject_put(llite_kobj);
87 void llite_tunables_unregister(void)
89 kobject_put(llite_kobj);
93 /* <debugfs>/lustre/llite mount point registration */
94 static const struct file_operations ll_rw_extents_stats_fops;
95 static const struct file_operations ll_rw_extents_stats_pp_fops;
96 static const struct file_operations ll_rw_offset_stats_fops;
99 * ll_stats_pid_write() - Determine if stats collection should be enabled
100 * @buf: Buffer containing the data written
101 * @len: Number of bytes in the buffer
103 * Several proc files begin collecting stats when a value is written, and stop
104 * collecting when either '0' or 'disable' is written. This function checks the
105 * written value to see if collection should be enabled or disabled.
107 * Return: If '0' or 'disable' is provided, 0 is returned. If the text
108 * equivalent of a number is written, that number is returned. Otherwise,
109 * 1 is returned. Non-zero return values indicate collection should be enabled.
111 static s64 ll_stats_pid_write(const char __user *buf, size_t len)
113 unsigned long long value = 1;
117 rc = kstrtoull_from_user(buf, len, 0, &value);
118 if (rc < 0 && len < sizeof(kernbuf)) {
119 if (copy_from_user(kernbuf, buf, len))
123 if (kernbuf[len - 1] == '\n')
124 kernbuf[len - 1] = 0;
126 if (strncasecmp(kernbuf, "disable", 7) == 0)
133 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
136 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
138 struct obd_statfs osfs;
141 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
145 return sprintf(buf, "%u\n", osfs.os_bsize);
147 LUSTRE_RO_ATTR(blocksize);
149 static ssize_t stat_blocksize_show(struct kobject *kobj, struct attribute *attr,
152 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
155 return sprintf(buf, "%u\n", sbi->ll_stat_blksize);
158 static ssize_t stat_blocksize_store(struct kobject *kobj,
159 struct attribute *attr,
163 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
168 rc = kstrtouint(buffer, 10, &val);
172 if (val != 0 && (val < PAGE_SIZE || (val & (val - 1))) != 0)
175 sbi->ll_stat_blksize = val;
179 LUSTRE_RW_ATTR(stat_blocksize);
181 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
184 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
186 struct obd_statfs osfs;
191 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
195 blk_size = osfs.os_bsize >> 10;
196 result = osfs.os_blocks;
198 while (blk_size >>= 1)
201 return sprintf(buf, "%llu\n", result);
203 LUSTRE_RO_ATTR(kbytestotal);
205 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
208 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
210 struct obd_statfs osfs;
215 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
219 blk_size = osfs.os_bsize >> 10;
220 result = osfs.os_bfree;
222 while (blk_size >>= 1)
225 return sprintf(buf, "%llu\n", result);
227 LUSTRE_RO_ATTR(kbytesfree);
229 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
232 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
234 struct obd_statfs osfs;
239 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
243 blk_size = osfs.os_bsize >> 10;
244 result = osfs.os_bavail;
246 while (blk_size >>= 1)
249 return sprintf(buf, "%llu\n", result);
251 LUSTRE_RO_ATTR(kbytesavail);
253 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
256 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
258 struct obd_statfs osfs;
261 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
265 return sprintf(buf, "%llu\n", osfs.os_files);
267 LUSTRE_RO_ATTR(filestotal);
269 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
272 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
274 struct obd_statfs osfs;
277 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
281 return sprintf(buf, "%llu\n", osfs.os_ffree);
283 LUSTRE_RO_ATTR(filesfree);
285 static ssize_t client_type_show(struct kobject *kobj, struct attribute *attr,
288 return sprintf(buf, "local client\n");
290 LUSTRE_RO_ATTR(client_type);
292 LUSTRE_RW_ATTR(foreign_symlink_enable);
294 LUSTRE_RW_ATTR(foreign_symlink_prefix);
296 LUSTRE_RW_ATTR(foreign_symlink_upcall);
298 LUSTRE_WO_ATTR(foreign_symlink_upcall_info);
300 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
303 return sprintf(buf, "lustre\n");
305 LUSTRE_RO_ATTR(fstype);
307 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
310 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
313 return sprintf(buf, "%s\n", sbi->ll_sb_uuid.uuid);
315 LUSTRE_RO_ATTR(uuid);
317 static int ll_site_stats_seq_show(struct seq_file *m, void *v)
319 struct super_block *sb = m->private;
322 * See description of statistical counters in struct cl_site, and
325 return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site), m);
328 LDEBUGFS_SEQ_FOPS_RO(ll_site_stats);
330 static ssize_t max_read_ahead_mb_show(struct kobject *kobj,
331 struct attribute *attr, char *buf)
333 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
336 return scnprintf(buf, PAGE_SIZE, "%lu\n",
337 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
340 static ssize_t max_read_ahead_mb_store(struct kobject *kobj,
341 struct attribute *attr,
342 const char *buffer, size_t count)
344 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
346 u64 ra_max_mb, pages_number;
349 rc = sysfs_memparse(buffer, count, &ra_max_mb, "MiB");
353 pages_number = round_up(ra_max_mb, 1024 * 1024) >> PAGE_SHIFT;
354 CDEBUG(D_INFO, "%s: set max_read_ahead_mb=%llu (%llu pages)\n",
355 sbi->ll_fsname, PAGES_TO_MiB(pages_number), pages_number);
356 if (pages_number > cfs_totalram_pages() / 2) {
358 CERROR("%s: cannot set max_read_ahead_mb=%llu > totalram/2=%luMB\n",
359 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
360 PAGES_TO_MiB(cfs_totalram_pages() / 2));
364 spin_lock(&sbi->ll_lock);
365 sbi->ll_ra_info.ra_max_pages = pages_number;
366 spin_unlock(&sbi->ll_lock);
370 LUSTRE_RW_ATTR(max_read_ahead_mb);
372 static ssize_t max_read_ahead_per_file_mb_show(struct kobject *kobj,
373 struct attribute *attr,
376 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
379 return scnprintf(buf, PAGE_SIZE, "%lu\n",
380 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
383 static ssize_t max_read_ahead_per_file_mb_store(struct kobject *kobj,
384 struct attribute *attr,
388 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
390 u64 ra_max_file_mb, pages_number;
393 rc = sysfs_memparse(buffer, count, &ra_max_file_mb, "MiB");
397 pages_number = round_up(ra_max_file_mb, 1024 * 1024) >> PAGE_SHIFT;
398 if (pages_number > sbi->ll_ra_info.ra_max_pages) {
399 CERROR("%s: cannot set max_read_ahead_per_file_mb=%llu > max_read_ahead_mb=%lu\n",
400 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
401 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
405 spin_lock(&sbi->ll_lock);
406 sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
407 spin_unlock(&sbi->ll_lock);
411 LUSTRE_RW_ATTR(max_read_ahead_per_file_mb);
413 static ssize_t max_read_ahead_whole_mb_show(struct kobject *kobj,
414 struct attribute *attr, char *buf)
416 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
419 return scnprintf(buf, PAGE_SIZE, "%lu\n",
420 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_read_ahead_whole_pages));
423 static ssize_t max_read_ahead_whole_mb_store(struct kobject *kobj,
424 struct attribute *attr,
425 const char *buffer, size_t count)
427 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
429 u64 ra_max_whole_mb, pages_number;
432 rc = sysfs_memparse(buffer, count, &ra_max_whole_mb, "MiB");
436 pages_number = round_up(ra_max_whole_mb, 1024 * 1024) >> PAGE_SHIFT;
437 /* Cap this at the current max readahead window size, the readahead
438 * algorithm does this anyway so it's pointless to set it larger.
440 if (pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
441 CERROR("%s: cannot set max_read_ahead_whole_mb=%llu > max_read_ahead_per_file_mb=%lu\n",
442 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
443 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
448 spin_lock(&sbi->ll_lock);
449 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
450 spin_unlock(&sbi->ll_lock);
454 LUSTRE_RW_ATTR(max_read_ahead_whole_mb);
456 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
458 struct super_block *sb = m->private;
459 struct ll_sb_info *sbi = ll_s2sbi(sb);
460 struct cl_client_cache *cache = sbi->ll_cache;
464 mutex_lock(&cache->ccc_max_cache_mb_lock);
465 max_cached_mb = PAGES_TO_MiB(cache->ccc_lru_max);
466 unused_mb = PAGES_TO_MiB(atomic_long_read(&cache->ccc_lru_left));
467 mutex_unlock(&cache->ccc_max_cache_mb_lock);
468 seq_printf(m, "users: %d\n"
469 "max_cached_mb: %ld\n"
472 "reclaim_count: %u\n",
473 atomic_read(&cache->ccc_users),
475 max_cached_mb - unused_mb,
477 cache->ccc_lru_shrinkers);
481 static ssize_t ll_max_cached_mb_seq_write(struct file *file,
482 const char __user *buffer,
483 size_t count, loff_t *off)
485 struct seq_file *m = file->private_data;
486 struct super_block *sb = m->private;
487 struct ll_sb_info *sbi = ll_s2sbi(sb);
488 struct cl_client_cache *cache = sbi->ll_cache;
495 char kernbuf[128], *ptr;
498 if (count >= sizeof(kernbuf))
501 if (copy_from_user(kernbuf, buffer, count))
503 kernbuf[count] = '\0';
505 ptr = lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count);
506 rc = sysfs_memparse(ptr, count, &pages_number, "MiB");
510 pages_number >>= PAGE_SHIFT;
512 if (pages_number < 0 || pages_number > cfs_totalram_pages()) {
513 CERROR("%s: can't set max cache more than %lu MB\n",
515 PAGES_TO_MiB(cfs_totalram_pages()));
518 /* Allow enough cache so clients can make well-formed RPCs */
519 pages_number = max_t(long, pages_number, PTLRPC_MAX_BRW_PAGES);
521 mutex_lock(&cache->ccc_max_cache_mb_lock);
522 diff = pages_number - cache->ccc_lru_max;
524 /* easy - add more LRU slots. */
526 atomic_long_add(diff, &cache->ccc_lru_left);
530 env = cl_env_get(&refcheck);
532 GOTO(out_unlock, rc = PTR_ERR(env));
538 /* reduce LRU budget from free slots. */
540 long lru_left_old, lru_left_new, lru_left_ret;
542 lru_left_old = atomic_long_read(&cache->ccc_lru_left);
543 if (lru_left_old == 0)
546 lru_left_new = lru_left_old > diff ?
547 lru_left_old - diff : 0;
549 atomic_long_cmpxchg(&cache->ccc_lru_left,
552 if (likely(lru_left_old == lru_left_ret)) {
553 diff -= lru_left_old - lru_left_new;
554 nrpages += lru_left_old - lru_left_new;
562 if (sbi->ll_dt_exp == NULL) { /* being initialized */
567 /* Request extra free slots to avoid them all being used
568 * by other processes before this can continue shrinking.
570 tmp = diff + min_t(long, diff, MiB_TO_PAGES(1024));
571 /* difficult - have to ask OSCs to drop LRU slots. */
572 rc = obd_set_info_async(env, sbi->ll_dt_exp,
573 sizeof(KEY_CACHE_LRU_SHRINK),
574 KEY_CACHE_LRU_SHRINK,
575 sizeof(tmp), &tmp, NULL);
579 cl_env_put(env, &refcheck);
583 cache->ccc_lru_max = pages_number;
586 atomic_long_add(nrpages, &cache->ccc_lru_left);
589 mutex_unlock(&cache->ccc_max_cache_mb_lock);
592 LDEBUGFS_SEQ_FOPS(ll_max_cached_mb);
594 static ssize_t checksums_show(struct kobject *kobj, struct attribute *attr,
597 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
600 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
603 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
604 const char *buffer, size_t count)
606 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
616 rc = kstrtobool(buffer, &val);
620 sbi->ll_flags |= LL_SBI_CHECKSUM;
622 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
625 rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
626 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
628 CWARN("Failed to set OSC checksum flags: %d\n", rc);
632 LUSTRE_RW_ATTR(checksums);
634 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
636 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
637 enum stats_track_type type)
639 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
642 if (sbi->ll_stats_track_type == type)
643 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
644 else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
645 return sprintf(buf, "0 (all)\n");
647 return sprintf(buf, "untracked\n");
650 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
651 size_t count, enum stats_track_type type)
653 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
658 rc = kstrtoul(buffer, 10, &pid);
662 sbi->ll_stats_track_id = pid;
664 sbi->ll_stats_track_type = STATS_TRACK_ALL;
666 sbi->ll_stats_track_type = type;
667 lprocfs_clear_stats(sbi->ll_stats);
671 static ssize_t stats_track_pid_show(struct kobject *kobj,
672 struct attribute *attr,
675 return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
678 static ssize_t stats_track_pid_store(struct kobject *kobj,
679 struct attribute *attr,
683 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
685 LUSTRE_RW_ATTR(stats_track_pid);
687 static ssize_t stats_track_ppid_show(struct kobject *kobj,
688 struct attribute *attr,
691 return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
694 static ssize_t stats_track_ppid_store(struct kobject *kobj,
695 struct attribute *attr,
699 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
701 LUSTRE_RW_ATTR(stats_track_ppid);
703 static ssize_t stats_track_gid_show(struct kobject *kobj,
704 struct attribute *attr,
707 return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
710 static ssize_t stats_track_gid_store(struct kobject *kobj,
711 struct attribute *attr,
715 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
717 LUSTRE_RW_ATTR(stats_track_gid);
719 static ssize_t statahead_running_max_show(struct kobject *kobj,
720 struct attribute *attr,
723 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
726 return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_sa_running_max);
729 static ssize_t statahead_running_max_store(struct kobject *kobj,
730 struct attribute *attr,
734 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
739 rc = kstrtoul(buffer, 0, &val);
743 if (val <= LL_SA_RUNNING_MAX) {
744 sbi->ll_sa_running_max = val;
748 CERROR("Bad statahead_running_max value %lu. Valid values "
749 "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
753 LUSTRE_RW_ATTR(statahead_running_max);
755 static ssize_t statahead_max_show(struct kobject *kobj,
756 struct attribute *attr,
759 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
762 return sprintf(buf, "%u\n", sbi->ll_sa_max);
765 static ssize_t statahead_max_store(struct kobject *kobj,
766 struct attribute *attr,
770 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
775 rc = kstrtoul(buffer, 0, &val);
779 if (val <= LL_SA_RPC_MAX)
780 sbi->ll_sa_max = val;
782 CERROR("Bad statahead_max value %lu. Valid values are in the range [0, %d]\n",
787 LUSTRE_RW_ATTR(statahead_max);
789 static ssize_t statahead_agl_show(struct kobject *kobj,
790 struct attribute *attr,
793 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
796 return sprintf(buf, "%u\n", sbi->ll_flags & LL_SBI_AGL_ENABLED ? 1 : 0);
799 static ssize_t statahead_agl_store(struct kobject *kobj,
800 struct attribute *attr,
804 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
809 rc = kstrtobool(buffer, &val);
814 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
816 sbi->ll_flags &= ~LL_SBI_AGL_ENABLED;
820 LUSTRE_RW_ATTR(statahead_agl);
822 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
824 struct super_block *sb = m->private;
825 struct ll_sb_info *sbi = ll_s2sbi(sb);
827 seq_printf(m, "statahead total: %u\n"
828 "statahead wrong: %u\n"
830 atomic_read(&sbi->ll_sa_total),
831 atomic_read(&sbi->ll_sa_wrong),
832 atomic_read(&sbi->ll_agl_total));
836 LDEBUGFS_SEQ_FOPS_RO(ll_statahead_stats);
838 static ssize_t lazystatfs_show(struct kobject *kobj,
839 struct attribute *attr,
842 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
845 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
848 static ssize_t lazystatfs_store(struct kobject *kobj,
849 struct attribute *attr,
853 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
858 rc = kstrtobool(buffer, &val);
863 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
865 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
869 LUSTRE_RW_ATTR(lazystatfs);
871 static ssize_t statfs_max_age_show(struct kobject *kobj, struct attribute *attr,
874 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
877 return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_statfs_max_age);
880 static ssize_t statfs_max_age_store(struct kobject *kobj,
881 struct attribute *attr, const char *buffer,
884 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
889 rc = kstrtouint(buffer, 10, &val);
892 if (val > OBD_STATFS_CACHE_MAX_AGE)
895 sbi->ll_statfs_max_age = val;
899 LUSTRE_RW_ATTR(statfs_max_age);
901 static ssize_t max_easize_show(struct kobject *kobj,
902 struct attribute *attr,
905 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
910 rc = ll_get_max_mdsize(sbi, &ealen);
914 /* Limit xattr size returned to userspace based on kernel maximum */
915 return scnprintf(buf, PAGE_SIZE, "%u\n",
916 ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
918 LUSTRE_RO_ATTR(max_easize);
921 * Get default_easize.
923 * \see client_obd::cl_default_mds_easize
925 * \param[in] m seq_file handle
926 * \param[in] v unused for single entry
928 * \retval 0 on success
929 * \retval negative negated errno on failure
931 static ssize_t default_easize_show(struct kobject *kobj,
932 struct attribute *attr,
935 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
940 rc = ll_get_default_mdsize(sbi, &ealen);
944 /* Limit xattr size returned to userspace based on kernel maximum */
945 return scnprintf(buf, PAGE_SIZE, "%u\n",
946 ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
950 * Set default_easize.
952 * Range checking on the passed value is handled by
953 * ll_set_default_mdsize().
955 * \see client_obd::cl_default_mds_easize
957 * \param[in] file proc file
958 * \param[in] buffer string passed from user space
959 * \param[in] count \a buffer length
960 * \param[in] off unused for single entry
962 * \retval positive \a count on success
963 * \retval negative negated errno on failure
965 static ssize_t default_easize_store(struct kobject *kobj,
966 struct attribute *attr,
970 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
978 rc = kstrtouint(buffer, 10, &val);
982 rc = ll_set_default_mdsize(sbi, val);
988 LUSTRE_RW_ATTR(default_easize);
990 static int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
992 const char *const str[] = LL_SBI_FLAGS;
993 struct super_block *sb = m->private;
994 int flags = ll_s2sbi(sb)->ll_flags;
998 if (ARRAY_SIZE(str) <= i) {
999 CERROR("%s: Revise array LL_SBI_FLAGS to match sbi "
1000 "flags please.\n", ll_s2sbi(sb)->ll_fsname);
1005 seq_printf(m, "%s ", str[i]);
1009 seq_printf(m, "\b\n");
1013 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
1015 static ssize_t xattr_cache_show(struct kobject *kobj,
1016 struct attribute *attr,
1019 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1022 return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
1025 static ssize_t xattr_cache_store(struct kobject *kobj,
1026 struct attribute *attr,
1030 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1035 rc = kstrtobool(buffer, &val);
1039 if (val && !(sbi->ll_flags & LL_SBI_XATTR_CACHE))
1042 sbi->ll_xattr_cache_enabled = val;
1043 sbi->ll_xattr_cache_set = 1;
1047 LUSTRE_RW_ATTR(xattr_cache);
1049 static ssize_t tiny_write_show(struct kobject *kobj,
1050 struct attribute *attr,
1053 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1056 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_TINY_WRITE));
1059 static ssize_t tiny_write_store(struct kobject *kobj,
1060 struct attribute *attr,
1064 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1069 rc = kstrtobool(buffer, &val);
1073 spin_lock(&sbi->ll_lock);
1075 sbi->ll_flags |= LL_SBI_TINY_WRITE;
1077 sbi->ll_flags &= ~LL_SBI_TINY_WRITE;
1078 spin_unlock(&sbi->ll_lock);
1082 LUSTRE_RW_ATTR(tiny_write);
1084 static ssize_t max_read_ahead_async_active_show(struct kobject *kobj,
1085 struct attribute *attr,
1088 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1091 return scnprintf(buf, PAGE_SIZE, "%u\n",
1092 sbi->ll_ra_info.ra_async_max_active);
1095 static ssize_t max_read_ahead_async_active_store(struct kobject *kobj,
1096 struct attribute *attr,
1100 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1105 rc = kstrtouint(buffer, 10, &val);
1110 * It doesn't make any sense to make it exceed what
1111 * workqueue could acutally support. This can easily
1112 * over subscripe the cores but Lustre internally
1113 * throttles to avoid those impacts.
1115 if (val > WQ_UNBOUND_MAX_ACTIVE) {
1116 CERROR("%s: cannot set max_read_ahead_async_active=%u larger than %u\n",
1117 sbi->ll_fsname, val, WQ_UNBOUND_MAX_ACTIVE);
1121 spin_lock(&sbi->ll_lock);
1122 sbi->ll_ra_info.ra_async_max_active = val;
1123 spin_unlock(&sbi->ll_lock);
1127 LUSTRE_RW_ATTR(max_read_ahead_async_active);
1129 static ssize_t read_ahead_async_file_threshold_mb_show(struct kobject *kobj,
1130 struct attribute *attr,
1133 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1136 return scnprintf(buf, PAGE_SIZE, "%lu\n", PAGES_TO_MiB(
1137 sbi->ll_ra_info.ra_async_pages_per_file_threshold));
1141 read_ahead_async_file_threshold_mb_store(struct kobject *kobj,
1142 struct attribute *attr,
1143 const char *buffer, size_t count)
1145 unsigned long pages_number;
1146 unsigned long max_ra_per_file;
1147 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1151 rc = kstrtoul(buffer, 10, &pages_number);
1155 pages_number = MiB_TO_PAGES(pages_number);
1156 max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1157 if (pages_number < 0 || pages_number > max_ra_per_file) {
1158 CERROR("%s: can't set read_ahead_async_file_threshold_mb=%lu > "
1159 "max_read_readahead_per_file_mb=%lu\n", sbi->ll_fsname,
1160 PAGES_TO_MiB(pages_number),
1161 PAGES_TO_MiB(max_ra_per_file));
1164 sbi->ll_ra_info.ra_async_pages_per_file_threshold = pages_number;
1168 LUSTRE_RW_ATTR(read_ahead_async_file_threshold_mb);
1170 static ssize_t read_ahead_range_kb_show(struct kobject *kobj,
1171 struct attribute *attr,char *buf)
1173 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1176 return snprintf(buf, PAGE_SIZE, "%lu\n",
1177 sbi->ll_ra_info.ra_range_pages << (PAGE_SHIFT - 10));
1181 read_ahead_range_kb_store(struct kobject *kobj,
1182 struct attribute *attr,
1183 const char *buffer, size_t count)
1185 unsigned long pages_number;
1186 unsigned long max_ra_per_file;
1188 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1192 rc = sysfs_memparse(buffer, count, &val, "KiB");
1196 pages_number = val >> PAGE_SHIFT;
1197 /* Disable mmap range read */
1198 if (pages_number == 0)
1201 max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1202 if (pages_number > max_ra_per_file ||
1203 pages_number < RA_MIN_MMAP_RANGE_PAGES)
1207 spin_lock(&sbi->ll_lock);
1208 sbi->ll_ra_info.ra_range_pages = pages_number;
1209 spin_unlock(&sbi->ll_lock);
1213 LUSTRE_RW_ATTR(read_ahead_range_kb);
1215 static ssize_t fast_read_show(struct kobject *kobj,
1216 struct attribute *attr,
1219 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1222 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_FAST_READ));
1225 static ssize_t fast_read_store(struct kobject *kobj,
1226 struct attribute *attr,
1230 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1235 rc = kstrtobool(buffer, &val);
1239 spin_lock(&sbi->ll_lock);
1241 sbi->ll_flags |= LL_SBI_FAST_READ;
1243 sbi->ll_flags &= ~LL_SBI_FAST_READ;
1244 spin_unlock(&sbi->ll_lock);
1248 LUSTRE_RW_ATTR(fast_read);
1250 static ssize_t file_heat_show(struct kobject *kobj,
1251 struct attribute *attr,
1254 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1257 return scnprintf(buf, PAGE_SIZE, "%u\n",
1258 !!(sbi->ll_flags & LL_SBI_FILE_HEAT));
1261 static ssize_t file_heat_store(struct kobject *kobj,
1262 struct attribute *attr,
1266 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1271 rc = kstrtobool(buffer, &val);
1275 spin_lock(&sbi->ll_lock);
1277 sbi->ll_flags |= LL_SBI_FILE_HEAT;
1279 sbi->ll_flags &= ~LL_SBI_FILE_HEAT;
1280 spin_unlock(&sbi->ll_lock);
1284 LUSTRE_RW_ATTR(file_heat);
1286 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1287 struct attribute *attr,
1290 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1293 return scnprintf(buf, PAGE_SIZE, "%u\n",
1294 (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1297 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1298 struct attribute *attr,
1302 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1307 rc = kstrtoul(buffer, 10, &val);
1311 if (val < 0 || val > 100)
1314 sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1318 LUSTRE_RW_ATTR(heat_decay_percentage);
1320 static ssize_t heat_period_second_show(struct kobject *kobj,
1321 struct attribute *attr,
1324 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1327 return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1330 static ssize_t heat_period_second_store(struct kobject *kobj,
1331 struct attribute *attr,
1335 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1340 rc = kstrtoul(buffer, 10, &val);
1347 sbi->ll_heat_period_second = val;
1351 LUSTRE_RW_ATTR(heat_period_second);
1353 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1355 struct super_block *sb = m->private;
1356 struct ll_sb_info *sbi = ll_s2sbi(sb);
1357 struct cl_client_cache *cache = sbi->ll_cache;
1361 pages = atomic_long_read(&cache->ccc_unstable_nr);
1362 mb = (pages * PAGE_SIZE) >> 20;
1364 seq_printf(m, "unstable_check: %8d\n"
1365 "unstable_pages: %12ld\n"
1366 "unstable_mb: %8d\n",
1367 cache->ccc_unstable_check, pages, mb);
1371 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1372 const char __user *buffer,
1373 size_t count, loff_t *unused)
1375 struct seq_file *seq = file->private_data;
1376 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1383 if (count >= sizeof(kernbuf))
1386 if (copy_from_user(kernbuf, buffer, count))
1390 buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1392 rc = kstrtobool_from_user(buffer, count, &val);
1396 /* borrow lru lock to set the value */
1397 spin_lock(&sbi->ll_cache->ccc_lru_lock);
1398 sbi->ll_cache->ccc_unstable_check = val;
1399 spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1404 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1406 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1408 struct super_block *sb = m->private;
1409 struct ll_sb_info *sbi = ll_s2sbi(sb);
1410 struct root_squash_info *squash = &sbi->ll_squash;
1412 seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1416 static ssize_t ll_root_squash_seq_write(struct file *file,
1417 const char __user *buffer,
1418 size_t count, loff_t *off)
1420 struct seq_file *m = file->private_data;
1421 struct super_block *sb = m->private;
1422 struct ll_sb_info *sbi = ll_s2sbi(sb);
1423 struct root_squash_info *squash = &sbi->ll_squash;
1425 return lprocfs_wr_root_squash(buffer, count, squash, sbi->ll_fsname);
1428 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1430 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1432 struct super_block *sb = m->private;
1433 struct ll_sb_info *sbi = ll_s2sbi(sb);
1434 struct root_squash_info *squash = &sbi->ll_squash;
1437 spin_lock(&squash->rsi_lock);
1438 if (!list_empty(&squash->rsi_nosquash_nids)) {
1439 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1440 &squash->rsi_nosquash_nids);
1444 seq_puts(m, "NONE\n");
1446 spin_unlock(&squash->rsi_lock);
1451 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1452 const char __user *buffer,
1453 size_t count, loff_t *off)
1455 struct seq_file *m = file->private_data;
1456 struct super_block *sb = m->private;
1457 struct ll_sb_info *sbi = ll_s2sbi(sb);
1458 struct root_squash_info *squash = &sbi->ll_squash;
1461 rc = lprocfs_wr_nosquash_nids(buffer, count, squash, sbi->ll_fsname);
1465 ll_compute_rootsquash_state(sbi);
1470 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1472 static int ll_pcc_seq_show(struct seq_file *m, void *v)
1474 struct super_block *sb = m->private;
1475 struct ll_sb_info *sbi = ll_s2sbi(sb);
1477 return pcc_super_dump(&sbi->ll_pcc_super, m);
1480 static ssize_t ll_pcc_seq_write(struct file *file, const char __user *buffer,
1481 size_t count, loff_t *off)
1483 struct seq_file *m = file->private_data;
1484 struct super_block *sb = m->private;
1485 struct ll_sb_info *sbi = ll_s2sbi(sb);
1489 if (count >= LPROCFS_WR_PCC_MAX_CMD)
1492 if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_PCC))
1495 OBD_ALLOC(kernbuf, count + 1);
1496 if (kernbuf == NULL)
1499 if (copy_from_user(kernbuf, buffer, count))
1500 GOTO(out_free_kernbuff, rc = -EFAULT);
1502 rc = pcc_cmd_handle(kernbuf, count, &sbi->ll_pcc_super);
1504 OBD_FREE(kernbuf, count + 1);
1505 return rc ? rc : count;
1507 LDEBUGFS_SEQ_FOPS(ll_pcc);
1509 struct ldebugfs_vars lprocfs_llite_obd_vars[] = {
1511 .fops = &ll_site_stats_fops },
1512 { .name = "max_cached_mb",
1513 .fops = &ll_max_cached_mb_fops },
1514 { .name = "statahead_stats",
1515 .fops = &ll_statahead_stats_fops },
1516 { .name = "unstable_stats",
1517 .fops = &ll_unstable_stats_fops },
1518 { .name = "sbi_flags",
1519 .fops = &ll_sbi_flags_fops },
1520 { .name = "root_squash",
1521 .fops = &ll_root_squash_fops },
1522 { .name = "nosquash_nids",
1523 .fops = &ll_nosquash_nids_fops },
1525 .fops = &ll_pcc_fops, },
1529 #define MAX_STRING_SIZE 128
1531 static struct attribute *llite_attrs[] = {
1532 &lustre_attr_blocksize.attr,
1533 &lustre_attr_stat_blocksize.attr,
1534 &lustre_attr_kbytestotal.attr,
1535 &lustre_attr_kbytesfree.attr,
1536 &lustre_attr_kbytesavail.attr,
1537 &lustre_attr_filestotal.attr,
1538 &lustre_attr_filesfree.attr,
1539 &lustre_attr_client_type.attr,
1540 &lustre_attr_foreign_symlink_enable.attr,
1541 &lustre_attr_foreign_symlink_prefix.attr,
1542 &lustre_attr_foreign_symlink_upcall.attr,
1543 &lustre_attr_foreign_symlink_upcall_info.attr,
1544 &lustre_attr_fstype.attr,
1545 &lustre_attr_uuid.attr,
1546 &lustre_attr_checksums.attr,
1547 &lustre_attr_checksum_pages.attr,
1548 &lustre_attr_max_read_ahead_mb.attr,
1549 &lustre_attr_max_read_ahead_per_file_mb.attr,
1550 &lustre_attr_max_read_ahead_whole_mb.attr,
1551 &lustre_attr_max_read_ahead_async_active.attr,
1552 &lustre_attr_read_ahead_async_file_threshold_mb.attr,
1553 &lustre_attr_read_ahead_range_kb.attr,
1554 &lustre_attr_stats_track_pid.attr,
1555 &lustre_attr_stats_track_ppid.attr,
1556 &lustre_attr_stats_track_gid.attr,
1557 &lustre_attr_statahead_running_max.attr,
1558 &lustre_attr_statahead_max.attr,
1559 &lustre_attr_statahead_agl.attr,
1560 &lustre_attr_lazystatfs.attr,
1561 &lustre_attr_statfs_max_age.attr,
1562 &lustre_attr_max_easize.attr,
1563 &lustre_attr_default_easize.attr,
1564 &lustre_attr_xattr_cache.attr,
1565 &lustre_attr_fast_read.attr,
1566 &lustre_attr_tiny_write.attr,
1567 &lustre_attr_file_heat.attr,
1568 &lustre_attr_heat_decay_percentage.attr,
1569 &lustre_attr_heat_period_second.attr,
1573 static void sbi_kobj_release(struct kobject *kobj)
1575 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1577 complete(&sbi->ll_kobj_unregister);
1580 static struct kobj_type sbi_ktype = {
1581 .default_attrs = llite_attrs,
1582 .sysfs_ops = &lustre_sysfs_ops,
1583 .release = sbi_kobj_release,
1586 static const struct llite_file_opcode {
1590 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1591 /* file operation */
1592 { LPROC_LL_READ_BYTES, LPROCFS_TYPE_BYTES_FULL, "read_bytes" },
1593 { LPROC_LL_WRITE_BYTES, LPROCFS_TYPE_BYTES_FULL, "write_bytes" },
1594 { LPROC_LL_READ, LPROCFS_TYPE_LATENCY, "read" },
1595 { LPROC_LL_WRITE, LPROCFS_TYPE_LATENCY, "write" },
1596 { LPROC_LL_IOCTL, LPROCFS_TYPE_REQS, "ioctl" },
1597 { LPROC_LL_OPEN, LPROCFS_TYPE_LATENCY, "open" },
1598 { LPROC_LL_RELEASE, LPROCFS_TYPE_LATENCY, "close" },
1599 { LPROC_LL_MMAP, LPROCFS_TYPE_LATENCY, "mmap" },
1600 { LPROC_LL_FAULT, LPROCFS_TYPE_LATENCY, "page_fault" },
1601 { LPROC_LL_MKWRITE, LPROCFS_TYPE_LATENCY, "page_mkwrite" },
1602 { LPROC_LL_LLSEEK, LPROCFS_TYPE_LATENCY, "seek" },
1603 { LPROC_LL_FSYNC, LPROCFS_TYPE_LATENCY, "fsync" },
1604 { LPROC_LL_READDIR, LPROCFS_TYPE_LATENCY, "readdir" },
1605 /* inode operation */
1606 { LPROC_LL_SETATTR, LPROCFS_TYPE_LATENCY, "setattr" },
1607 { LPROC_LL_TRUNC, LPROCFS_TYPE_LATENCY, "truncate" },
1608 { LPROC_LL_FLOCK, LPROCFS_TYPE_LATENCY, "flock" },
1609 { LPROC_LL_GETATTR, LPROCFS_TYPE_LATENCY, "getattr" },
1610 { LPROC_LL_FALLOCATE, LPROCFS_TYPE_LATENCY, "fallocate"},
1611 /* dir inode operation */
1612 { LPROC_LL_CREATE, LPROCFS_TYPE_LATENCY, "create" },
1613 { LPROC_LL_LINK, LPROCFS_TYPE_LATENCY, "link" },
1614 { LPROC_LL_UNLINK, LPROCFS_TYPE_LATENCY, "unlink" },
1615 { LPROC_LL_SYMLINK, LPROCFS_TYPE_LATENCY, "symlink" },
1616 { LPROC_LL_MKDIR, LPROCFS_TYPE_LATENCY, "mkdir" },
1617 { LPROC_LL_RMDIR, LPROCFS_TYPE_LATENCY, "rmdir" },
1618 { LPROC_LL_MKNOD, LPROCFS_TYPE_LATENCY, "mknod" },
1619 { LPROC_LL_RENAME, LPROCFS_TYPE_LATENCY, "rename" },
1620 /* special inode operation */
1621 { LPROC_LL_STATFS, LPROCFS_TYPE_LATENCY, "statfs" },
1622 { LPROC_LL_SETXATTR, LPROCFS_TYPE_LATENCY, "setxattr" },
1623 { LPROC_LL_GETXATTR, LPROCFS_TYPE_LATENCY, "getxattr" },
1624 { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REQS, "getxattr_hits" },
1625 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_LATENCY, "listxattr" },
1626 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_LATENCY, "removexattr" },
1627 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_LATENCY, "inode_permission" },
1630 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count)
1635 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1636 lprocfs_counter_add(sbi->ll_stats, op, count);
1637 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1638 sbi->ll_stats_track_id == current->pid)
1639 lprocfs_counter_add(sbi->ll_stats, op, count);
1640 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1641 sbi->ll_stats_track_id == current->parent->pid)
1642 lprocfs_counter_add(sbi->ll_stats, op, count);
1643 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1644 sbi->ll_stats_track_id ==
1645 from_kgid(&init_user_ns, current_gid()))
1646 lprocfs_counter_add(sbi->ll_stats, op, count);
1648 EXPORT_SYMBOL(ll_stats_ops_tally);
1650 static const char *const ra_stat_string[] = {
1651 [RA_STAT_HIT] = "hits",
1652 [RA_STAT_MISS] = "misses",
1653 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1654 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1655 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1656 [RA_STAT_FAILED_MATCH] = "failed lock match",
1657 [RA_STAT_DISCARDED] = "read but discarded",
1658 [RA_STAT_ZERO_LEN] = "zero length file",
1659 [RA_STAT_ZERO_WINDOW] = "zero size window",
1660 [RA_STAT_EOF] = "read-ahead to EOF",
1661 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1662 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1663 [RA_STAT_FAILED_REACH_END] = "failed to reach end",
1664 [RA_STAT_ASYNC] = "async readahead",
1665 [RA_STAT_FAILED_FAST_READ] = "failed to fast read",
1666 [RA_STAT_MMAP_RANGE_READ] = "mmap range read",
1669 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1671 struct lustre_sb_info *lsi = s2lsi(sb);
1672 struct ll_sb_info *sbi = ll_s2sbi(sb);
1678 if (IS_ERR_OR_NULL(llite_root))
1681 sbi->ll_debugfs_entry = debugfs_create_dir(name, llite_root);
1682 ldebugfs_add_vars(sbi->ll_debugfs_entry, lprocfs_llite_obd_vars, sb);
1684 debugfs_create_file("dump_page_cache", 0444, sbi->ll_debugfs_entry, sbi,
1685 &vvp_dump_pgcache_file_ops);
1687 debugfs_create_file("extents_stats", 0644, sbi->ll_debugfs_entry, sbi,
1688 &ll_rw_extents_stats_fops);
1690 debugfs_create_file("extents_stats_per_process", 0644,
1691 sbi->ll_debugfs_entry, sbi,
1692 &ll_rw_extents_stats_pp_fops);
1694 debugfs_create_file("offset_stats", 0644, sbi->ll_debugfs_entry, sbi,
1695 &ll_rw_offset_stats_fops);
1697 /* File operations stats */
1698 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1699 LPROCFS_STATS_FLAG_NONE);
1700 if (sbi->ll_stats == NULL)
1701 GOTO(out_debugfs, err = -ENOMEM);
1703 /* do counter init */
1704 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1705 u32 type = llite_opcode_table[id].type;
1706 void *ptr = "unknown";
1708 if (type & LPROCFS_TYPE_REQS)
1710 else if (type & LPROCFS_TYPE_BYTES)
1712 else if (type & LPROCFS_TYPE_USEC)
1714 lprocfs_counter_init(sbi->ll_stats,
1715 llite_opcode_table[id].opcode, type,
1716 llite_opcode_table[id].opname, ptr);
1719 debugfs_create_file("stats", 0644, sbi->ll_debugfs_entry,
1720 sbi->ll_stats, &ldebugfs_stats_seq_fops);
1722 sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1723 LPROCFS_STATS_FLAG_NONE);
1724 if (sbi->ll_ra_stats == NULL)
1725 GOTO(out_stats, err = -ENOMEM);
1727 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1728 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1729 ra_stat_string[id], "pages");
1731 debugfs_create_file("read_ahead_stats", 0644, sbi->ll_debugfs_entry,
1732 sbi->ll_ra_stats, &ldebugfs_stats_seq_fops);
1735 /* Yes we also register sysfs mount kset here as well */
1736 sbi->ll_kset.kobj.parent = llite_kobj;
1737 sbi->ll_kset.kobj.ktype = &sbi_ktype;
1738 init_completion(&sbi->ll_kobj_unregister);
1739 err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1741 GOTO(out_ra_stats, err);
1743 err = kset_register(&sbi->ll_kset);
1745 GOTO(out_ra_stats, err);
1747 lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1751 lprocfs_free_stats(&sbi->ll_ra_stats);
1753 lprocfs_free_stats(&sbi->ll_stats);
1755 debugfs_remove_recursive(sbi->ll_debugfs_entry);
1760 void ll_debugfs_unregister_super(struct super_block *sb)
1762 struct lustre_sb_info *lsi = s2lsi(sb);
1763 struct ll_sb_info *sbi = ll_s2sbi(sb);
1765 debugfs_remove_recursive(sbi->ll_debugfs_entry);
1768 sysfs_remove_link(&sbi->ll_kset.kobj,
1769 sbi->ll_dt_obd->obd_type->typ_name);
1772 sysfs_remove_link(&sbi->ll_kset.kobj,
1773 sbi->ll_md_obd->obd_type->typ_name);
1775 kobject_put(lsi->lsi_kobj);
1777 kset_unregister(&sbi->ll_kset);
1778 wait_for_completion(&sbi->ll_kobj_unregister);
1780 lprocfs_free_stats(&sbi->ll_ra_stats);
1781 lprocfs_free_stats(&sbi->ll_stats);
1783 #undef MAX_STRING_SIZE
1785 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1786 struct seq_file *seq, int which)
1788 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1789 unsigned long start, end, r, w;
1790 char *unitp = "KMGTPEZY";
1792 struct per_process_info *pp_info = &io_extents->pp_extents[which];
1798 for(i = 0; i < LL_HIST_MAX; i++) {
1799 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1800 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1803 for(i = 0; i < LL_HIST_MAX; i++) {
1804 r = pp_info->pp_r_hist.oh_buckets[i];
1805 w = pp_info->pp_w_hist.oh_buckets[i];
1808 end = 1 << (i + LL_HIST_START - units);
1809 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u | "
1810 "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1811 (i == LL_HIST_MAX - 1) ? '+' : ' ',
1812 r, pct(r, read_tot), pct(read_cum, read_tot),
1813 w, pct(w, write_tot), pct(write_cum, write_tot));
1815 if (start == (1 << 10)) {
1820 if (read_cum == read_tot && write_cum == write_tot)
1825 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1827 struct timespec64 now;
1828 struct ll_sb_info *sbi = seq->private;
1829 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1832 ktime_get_real_ts64(&now);
1834 if (!sbi->ll_rw_stats_on) {
1835 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1838 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1839 (s64)now.tv_sec, now.tv_nsec);
1840 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1841 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1842 "extents", "calls", "%", "cum%",
1843 "calls", "%", "cum%");
1844 spin_lock(&sbi->ll_pp_extent_lock);
1845 for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1846 if (io_extents->pp_extents[k].pid != 0) {
1847 seq_printf(seq, "\nPID: %d\n",
1848 io_extents->pp_extents[k].pid);
1849 ll_display_extents_info(io_extents, seq, k);
1852 spin_unlock(&sbi->ll_pp_extent_lock);
1856 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1857 const char __user *buf,
1861 struct seq_file *seq = file->private_data;
1862 struct ll_sb_info *sbi = seq->private;
1863 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1870 value = ll_stats_pid_write(buf, len);
1873 sbi->ll_rw_stats_on = 0;
1875 sbi->ll_rw_stats_on = 1;
1877 spin_lock(&sbi->ll_pp_extent_lock);
1878 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1879 io_extents->pp_extents[i].pid = 0;
1880 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1881 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1883 spin_unlock(&sbi->ll_pp_extent_lock);
1887 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1889 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1891 struct timespec64 now;
1892 struct ll_sb_info *sbi = seq->private;
1893 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1895 ktime_get_real_ts64(&now);
1897 if (!sbi->ll_rw_stats_on) {
1898 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1901 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1902 (s64)now.tv_sec, now.tv_nsec);
1904 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1905 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1906 "extents", "calls", "%", "cum%",
1907 "calls", "%", "cum%");
1908 spin_lock(&sbi->ll_lock);
1909 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1910 spin_unlock(&sbi->ll_lock);
1915 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1916 const char __user *buf,
1917 size_t len, loff_t *off)
1919 struct seq_file *seq = file->private_data;
1920 struct ll_sb_info *sbi = seq->private;
1921 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1928 value = ll_stats_pid_write(buf, len);
1931 sbi->ll_rw_stats_on = 0;
1933 sbi->ll_rw_stats_on = 1;
1935 spin_lock(&sbi->ll_pp_extent_lock);
1936 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1937 io_extents->pp_extents[i].pid = 0;
1938 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1939 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1941 spin_unlock(&sbi->ll_pp_extent_lock);
1946 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1948 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1949 struct ll_file_data *file, loff_t pos,
1950 size_t count, int rw)
1953 struct ll_rw_process_info *process;
1954 struct ll_rw_process_info *offset;
1955 int *off_count = &sbi->ll_rw_offset_entry_count;
1956 int *process_count = &sbi->ll_offset_process_count;
1957 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1959 if(!sbi->ll_rw_stats_on)
1961 process = sbi->ll_rw_process_info;
1962 offset = sbi->ll_rw_offset_info;
1964 spin_lock(&sbi->ll_pp_extent_lock);
1965 /* Extent statistics */
1966 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1967 if(io_extents->pp_extents[i].pid == pid) {
1975 sbi->ll_extent_process_count =
1976 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1977 cur = sbi->ll_extent_process_count;
1978 io_extents->pp_extents[cur].pid = pid;
1979 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1980 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1983 for (i = 0; (count >= 1 << (LL_HIST_START + i)) &&
1984 (i < (LL_HIST_MAX - 1)); i++);
1986 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1987 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1989 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1990 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1992 spin_unlock(&sbi->ll_pp_extent_lock);
1994 spin_lock(&sbi->ll_process_lock);
1995 /* Offset statistics */
1996 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1997 if (process[i].rw_pid == pid) {
1998 if (process[i].rw_last_file != file) {
1999 process[i].rw_range_start = pos;
2000 process[i].rw_last_file_pos = pos + count;
2001 process[i].rw_smallest_extent = count;
2002 process[i].rw_largest_extent = count;
2003 process[i].rw_offset = 0;
2004 process[i].rw_last_file = file;
2005 spin_unlock(&sbi->ll_process_lock);
2008 if (process[i].rw_last_file_pos != pos) {
2010 (*off_count + 1) % LL_OFFSET_HIST_MAX;
2011 offset[*off_count].rw_op = process[i].rw_op;
2012 offset[*off_count].rw_pid = pid;
2013 offset[*off_count].rw_range_start =
2014 process[i].rw_range_start;
2015 offset[*off_count].rw_range_end =
2016 process[i].rw_last_file_pos;
2017 offset[*off_count].rw_smallest_extent =
2018 process[i].rw_smallest_extent;
2019 offset[*off_count].rw_largest_extent =
2020 process[i].rw_largest_extent;
2021 offset[*off_count].rw_offset =
2022 process[i].rw_offset;
2023 process[i].rw_op = rw;
2024 process[i].rw_range_start = pos;
2025 process[i].rw_smallest_extent = count;
2026 process[i].rw_largest_extent = count;
2027 process[i].rw_offset = pos -
2028 process[i].rw_last_file_pos;
2030 if(process[i].rw_smallest_extent > count)
2031 process[i].rw_smallest_extent = count;
2032 if(process[i].rw_largest_extent < count)
2033 process[i].rw_largest_extent = count;
2034 process[i].rw_last_file_pos = pos + count;
2035 spin_unlock(&sbi->ll_process_lock);
2039 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
2040 process[*process_count].rw_pid = pid;
2041 process[*process_count].rw_op = rw;
2042 process[*process_count].rw_range_start = pos;
2043 process[*process_count].rw_last_file_pos = pos + count;
2044 process[*process_count].rw_smallest_extent = count;
2045 process[*process_count].rw_largest_extent = count;
2046 process[*process_count].rw_offset = 0;
2047 process[*process_count].rw_last_file = file;
2048 spin_unlock(&sbi->ll_process_lock);
2051 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
2053 struct timespec64 now;
2054 struct ll_sb_info *sbi = seq->private;
2055 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
2056 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
2059 ktime_get_real_ts64(&now);
2061 if (!sbi->ll_rw_stats_on) {
2062 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2065 spin_lock(&sbi->ll_process_lock);
2067 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
2068 (s64)now.tv_sec, now.tv_nsec);
2069 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
2070 "R/W", "PID", "RANGE START", "RANGE END",
2071 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
2073 /* We stored the discontiguous offsets here; print them first */
2074 for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
2075 if (offset[i].rw_pid != 0)
2077 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2078 offset[i].rw_op == READ ? 'R' : 'W',
2080 offset[i].rw_range_start,
2081 offset[i].rw_range_end,
2082 (unsigned long)offset[i].rw_smallest_extent,
2083 (unsigned long)offset[i].rw_largest_extent,
2084 offset[i].rw_offset);
2087 /* Then print the current offsets for each process */
2088 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2089 if (process[i].rw_pid != 0)
2091 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2092 process[i].rw_op == READ ? 'R' : 'W',
2094 process[i].rw_range_start,
2095 process[i].rw_last_file_pos,
2096 (unsigned long)process[i].rw_smallest_extent,
2097 (unsigned long)process[i].rw_largest_extent,
2098 process[i].rw_offset);
2100 spin_unlock(&sbi->ll_process_lock);
2105 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
2106 const char __user *buf,
2107 size_t len, loff_t *off)
2109 struct seq_file *seq = file->private_data;
2110 struct ll_sb_info *sbi = seq->private;
2111 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
2112 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
2118 value = ll_stats_pid_write(buf, len);
2121 sbi->ll_rw_stats_on = 0;
2123 sbi->ll_rw_stats_on = 1;
2125 spin_lock(&sbi->ll_process_lock);
2126 sbi->ll_offset_process_count = 0;
2127 sbi->ll_rw_offset_entry_count = 0;
2128 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
2129 LL_PROCESS_HIST_MAX);
2130 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
2131 LL_OFFSET_HIST_MAX);
2132 spin_unlock(&sbi->ll_process_lock);
2137 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);