4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 #define DEBUG_SUBSYSTEM S_LLITE
33 #include <linux/version.h>
34 #include <linux/user_namespace.h>
35 #include <linux/uidgid.h>
37 #include <uapi/linux/lustre/lustre_param.h>
38 #include <lprocfs_status.h>
39 #include <obd_support.h>
41 #include "llite_internal.h"
42 #include "lprocfs_status.h"
43 #include "vvp_internal.h"
45 static struct kobject *llite_kobj;
46 static struct dentry *llite_root;
48 static void llite_kobj_release(struct kobject *kobj)
50 if (!IS_ERR_OR_NULL(llite_root)) {
51 debugfs_remove(llite_root);
58 static struct kobj_type llite_kobj_ktype = {
59 .release = llite_kobj_release,
60 .sysfs_ops = &lustre_sysfs_ops,
63 int llite_tunables_register(void)
67 llite_kobj = kzalloc(sizeof(*llite_kobj), GFP_KERNEL);
71 llite_kobj->kset = lustre_kset;
72 rc = kobject_init_and_add(llite_kobj, &llite_kobj_ktype,
73 &lustre_kset->kobj, "%s", "llite");
77 llite_root = debugfs_create_dir("llite", debugfs_lustre_root);
81 kobject_put(llite_kobj);
87 void llite_tunables_unregister(void)
89 kobject_put(llite_kobj);
93 /* <debugfs>/lustre/llite mount point registration */
94 static const struct file_operations ll_rw_extents_stats_fops;
95 static const struct file_operations ll_rw_extents_stats_pp_fops;
96 static const struct file_operations ll_rw_offset_stats_fops;
99 * ll_stats_pid_write() - Determine if stats collection should be enabled
100 * @buf: Buffer containing the data written
101 * @len: Number of bytes in the buffer
103 * Several proc files begin collecting stats when a value is written, and stop
104 * collecting when either '0' or 'disable' is written. This function checks the
105 * written value to see if collection should be enabled or disabled.
107 * Return: If '0' or 'disable' is provided, 0 is returned. If the text
108 * equivalent of a number is written, that number is returned. Otherwise,
109 * 1 is returned. Non-zero return values indicate collection should be enabled.
111 static s64 ll_stats_pid_write(const char __user *buf, size_t len)
113 unsigned long long value = 1;
117 rc = kstrtoull_from_user(buf, len, 0, &value);
118 if (rc < 0 && len < sizeof(kernbuf)) {
119 if (copy_from_user(kernbuf, buf, len))
123 if (kernbuf[len - 1] == '\n')
124 kernbuf[len - 1] = 0;
126 if (strncasecmp(kernbuf, "disable", 7) == 0)
133 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
136 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
138 struct obd_statfs osfs;
141 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
145 return sprintf(buf, "%u\n", osfs.os_bsize);
147 LUSTRE_RO_ATTR(blocksize);
149 static ssize_t stat_blocksize_show(struct kobject *kobj, struct attribute *attr,
152 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
155 return sprintf(buf, "%u\n", sbi->ll_stat_blksize);
158 static ssize_t stat_blocksize_store(struct kobject *kobj,
159 struct attribute *attr,
163 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
168 rc = kstrtouint(buffer, 10, &val);
172 if (val != 0 && (val < PAGE_SIZE || (val & (val - 1))) != 0)
175 sbi->ll_stat_blksize = val;
179 LUSTRE_RW_ATTR(stat_blocksize);
181 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
184 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
186 struct obd_statfs osfs;
191 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
195 blk_size = osfs.os_bsize >> 10;
196 result = osfs.os_blocks;
198 while (blk_size >>= 1)
201 return sprintf(buf, "%llu\n", result);
203 LUSTRE_RO_ATTR(kbytestotal);
205 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
208 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
210 struct obd_statfs osfs;
215 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
219 blk_size = osfs.os_bsize >> 10;
220 result = osfs.os_bfree;
222 while (blk_size >>= 1)
225 return sprintf(buf, "%llu\n", result);
227 LUSTRE_RO_ATTR(kbytesfree);
229 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
232 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
234 struct obd_statfs osfs;
239 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
243 blk_size = osfs.os_bsize >> 10;
244 result = osfs.os_bavail;
246 while (blk_size >>= 1)
249 return sprintf(buf, "%llu\n", result);
251 LUSTRE_RO_ATTR(kbytesavail);
253 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
256 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
258 struct obd_statfs osfs;
261 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
265 return sprintf(buf, "%llu\n", osfs.os_files);
267 LUSTRE_RO_ATTR(filestotal);
269 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
272 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
274 struct obd_statfs osfs;
277 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
281 return sprintf(buf, "%llu\n", osfs.os_ffree);
283 LUSTRE_RO_ATTR(filesfree);
285 static ssize_t client_type_show(struct kobject *kobj, struct attribute *attr,
288 return sprintf(buf, "local client\n");
290 LUSTRE_RO_ATTR(client_type);
292 LUSTRE_RW_ATTR(foreign_symlink_enable);
294 LUSTRE_RW_ATTR(foreign_symlink_prefix);
296 LUSTRE_RW_ATTR(foreign_symlink_upcall);
298 LUSTRE_WO_ATTR(foreign_symlink_upcall_info);
300 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
303 return sprintf(buf, "lustre\n");
305 LUSTRE_RO_ATTR(fstype);
307 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
310 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
313 return sprintf(buf, "%s\n", sbi->ll_sb_uuid.uuid);
315 LUSTRE_RO_ATTR(uuid);
317 static int ll_site_stats_seq_show(struct seq_file *m, void *v)
319 struct super_block *sb = m->private;
322 * See description of statistical counters in struct cl_site, and
325 return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site), m);
328 LDEBUGFS_SEQ_FOPS_RO(ll_site_stats);
330 static ssize_t max_read_ahead_mb_show(struct kobject *kobj,
331 struct attribute *attr, char *buf)
333 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
336 return scnprintf(buf, PAGE_SIZE, "%lu\n",
337 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
340 static ssize_t max_read_ahead_mb_store(struct kobject *kobj,
341 struct attribute *attr,
342 const char *buffer, size_t count)
344 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
346 u64 ra_max_mb, pages_number;
349 rc = sysfs_memparse(buffer, count, &ra_max_mb, "MiB");
353 pages_number = round_up(ra_max_mb, 1024 * 1024) >> PAGE_SHIFT;
354 CDEBUG(D_INFO, "%s: set max_read_ahead_mb=%llu (%llu pages)\n",
355 sbi->ll_fsname, PAGES_TO_MiB(pages_number), pages_number);
356 if (pages_number > cfs_totalram_pages() / 2) {
358 CERROR("%s: cannot set max_read_ahead_mb=%llu > totalram/2=%luMB\n",
359 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
360 PAGES_TO_MiB(cfs_totalram_pages() / 2));
364 spin_lock(&sbi->ll_lock);
365 sbi->ll_ra_info.ra_max_pages = pages_number;
366 spin_unlock(&sbi->ll_lock);
370 LUSTRE_RW_ATTR(max_read_ahead_mb);
372 static ssize_t max_read_ahead_per_file_mb_show(struct kobject *kobj,
373 struct attribute *attr,
376 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
379 return scnprintf(buf, PAGE_SIZE, "%lu\n",
380 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
383 static ssize_t max_read_ahead_per_file_mb_store(struct kobject *kobj,
384 struct attribute *attr,
388 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
390 u64 ra_max_file_mb, pages_number;
393 rc = sysfs_memparse(buffer, count, &ra_max_file_mb, "MiB");
397 pages_number = round_up(ra_max_file_mb, 1024 * 1024) >> PAGE_SHIFT;
398 if (pages_number > sbi->ll_ra_info.ra_max_pages) {
399 CERROR("%s: cannot set max_read_ahead_per_file_mb=%llu > max_read_ahead_mb=%lu\n",
400 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
401 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
405 spin_lock(&sbi->ll_lock);
406 sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
407 spin_unlock(&sbi->ll_lock);
411 LUSTRE_RW_ATTR(max_read_ahead_per_file_mb);
413 static ssize_t max_read_ahead_whole_mb_show(struct kobject *kobj,
414 struct attribute *attr, char *buf)
416 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
419 return scnprintf(buf, PAGE_SIZE, "%lu\n",
420 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_read_ahead_whole_pages));
423 static ssize_t max_read_ahead_whole_mb_store(struct kobject *kobj,
424 struct attribute *attr,
425 const char *buffer, size_t count)
427 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
429 u64 ra_max_whole_mb, pages_number;
432 rc = sysfs_memparse(buffer, count, &ra_max_whole_mb, "MiB");
436 pages_number = round_up(ra_max_whole_mb, 1024 * 1024) >> PAGE_SHIFT;
437 /* Cap this at the current max readahead window size, the readahead
438 * algorithm does this anyway so it's pointless to set it larger.
440 if (pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
441 CERROR("%s: cannot set max_read_ahead_whole_mb=%llu > max_read_ahead_per_file_mb=%lu\n",
442 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
443 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
448 spin_lock(&sbi->ll_lock);
449 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
450 spin_unlock(&sbi->ll_lock);
454 LUSTRE_RW_ATTR(max_read_ahead_whole_mb);
456 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
458 struct super_block *sb = m->private;
459 struct ll_sb_info *sbi = ll_s2sbi(sb);
460 struct cl_client_cache *cache = sbi->ll_cache;
461 struct ll_ra_info *ra = &sbi->ll_ra_info;
465 mutex_lock(&cache->ccc_max_cache_mb_lock);
466 max_cached_mb = PAGES_TO_MiB(cache->ccc_lru_max);
467 unused_mb = PAGES_TO_MiB(atomic_long_read(&cache->ccc_lru_left));
468 mutex_unlock(&cache->ccc_max_cache_mb_lock);
470 seq_printf(m, "users: %d\n"
471 "max_cached_mb: %ld\n"
474 "reclaim_count: %u\n"
475 "max_read_ahead_mb: %lu\n"
476 "used_read_ahead_mb: %d\n",
477 refcount_read(&cache->ccc_users),
479 max_cached_mb - unused_mb,
481 cache->ccc_lru_shrinkers,
482 PAGES_TO_MiB(ra->ra_max_pages),
483 PAGES_TO_MiB(atomic_read(&ra->ra_cur_pages)));
487 static ssize_t ll_max_cached_mb_seq_write(struct file *file,
488 const char __user *buffer,
489 size_t count, loff_t *off)
491 struct seq_file *m = file->private_data;
492 struct super_block *sb = m->private;
493 struct ll_sb_info *sbi = ll_s2sbi(sb);
494 struct cl_client_cache *cache = sbi->ll_cache;
502 char kernbuf[128], *ptr;
503 bool percent = false;
506 if (count >= sizeof(kernbuf))
509 if (copy_from_user(kernbuf, buffer, count))
512 if (count > 0 && kernbuf[count - 1] == '%') {
514 /* strip off the % */
515 kernbuf[count - 1] = '\0';
517 kernbuf[count] = '\0';
520 ptr = lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count);
522 rc = sysfs_memparse(ptr, count, &value, "B");
524 rc = sysfs_memparse(ptr, count, &value, "MiB");
529 pages_number = cfs_totalram_pages() * value / 100;
531 pages_number = value >> PAGE_SHIFT;
533 if (pages_number < 0 || pages_number > cfs_totalram_pages()) {
534 CERROR("%s: can't set max cache more than %lu MB\n",
536 PAGES_TO_MiB(cfs_totalram_pages()));
539 /* Allow enough cache so clients can make well-formed RPCs */
540 pages_number = max_t(long, pages_number, PTLRPC_MAX_BRW_PAGES);
542 mutex_lock(&cache->ccc_max_cache_mb_lock);
543 diff = pages_number - cache->ccc_lru_max;
545 /* easy - add more LRU slots. */
547 atomic_long_add(diff, &cache->ccc_lru_left);
551 env = cl_env_get(&refcheck);
553 GOTO(out_unlock, rc = PTR_ERR(env));
559 /* reduce LRU budget from free slots. */
561 long lru_left_old, lru_left_new, lru_left_ret;
563 lru_left_old = atomic_long_read(&cache->ccc_lru_left);
564 if (lru_left_old == 0)
567 lru_left_new = lru_left_old > diff ?
568 lru_left_old - diff : 0;
570 atomic_long_cmpxchg(&cache->ccc_lru_left,
573 if (likely(lru_left_old == lru_left_ret)) {
574 diff -= lru_left_old - lru_left_new;
575 nrpages += lru_left_old - lru_left_new;
583 if (sbi->ll_dt_exp == NULL) { /* being initialized */
588 /* Request extra free slots to avoid them all being used
589 * by other processes before this can continue shrinking.
591 tmp = diff + min_t(long, diff, MiB_TO_PAGES(1024));
592 /* difficult - have to ask OSCs to drop LRU slots. */
593 rc = obd_set_info_async(env, sbi->ll_dt_exp,
594 sizeof(KEY_CACHE_LRU_SHRINK),
595 KEY_CACHE_LRU_SHRINK,
596 sizeof(tmp), &tmp, NULL);
600 cl_env_put(env, &refcheck);
604 cache->ccc_lru_max = pages_number;
607 atomic_long_add(nrpages, &cache->ccc_lru_left);
610 mutex_unlock(&cache->ccc_max_cache_mb_lock);
613 LDEBUGFS_SEQ_FOPS(ll_max_cached_mb);
615 static ssize_t checksums_show(struct kobject *kobj, struct attribute *attr,
618 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
621 return scnprintf(buf, PAGE_SIZE, "%u\n",
622 test_bit(LL_SBI_CHECKSUM, sbi->ll_flags));
625 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
626 const char *buffer, size_t count)
628 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
638 rc = kstrtobool(buffer, &val);
642 set_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
644 clear_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
647 rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
648 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
650 CWARN("Failed to set OSC checksum flags: %d\n", rc);
654 LUSTRE_RW_ATTR(checksums);
656 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
658 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
659 enum stats_track_type type)
661 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
664 if (sbi->ll_stats_track_type == type)
665 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
666 else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
667 return sprintf(buf, "0 (all)\n");
669 return sprintf(buf, "untracked\n");
672 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
673 size_t count, enum stats_track_type type)
675 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
680 rc = kstrtoul(buffer, 10, &pid);
684 sbi->ll_stats_track_id = pid;
686 sbi->ll_stats_track_type = STATS_TRACK_ALL;
688 sbi->ll_stats_track_type = type;
689 lprocfs_stats_clear(sbi->ll_stats);
693 static ssize_t stats_track_pid_show(struct kobject *kobj,
694 struct attribute *attr,
697 return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
700 static ssize_t stats_track_pid_store(struct kobject *kobj,
701 struct attribute *attr,
705 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
707 LUSTRE_RW_ATTR(stats_track_pid);
709 static ssize_t stats_track_ppid_show(struct kobject *kobj,
710 struct attribute *attr,
713 return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
716 static ssize_t stats_track_ppid_store(struct kobject *kobj,
717 struct attribute *attr,
721 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
723 LUSTRE_RW_ATTR(stats_track_ppid);
725 static ssize_t stats_track_gid_show(struct kobject *kobj,
726 struct attribute *attr,
729 return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
732 static ssize_t stats_track_gid_store(struct kobject *kobj,
733 struct attribute *attr,
737 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
739 LUSTRE_RW_ATTR(stats_track_gid);
741 static ssize_t enable_statahead_fname_show(struct kobject *kobj,
742 struct attribute *attr,
745 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
748 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_enable_statahead_fname);
751 static ssize_t enable_statahead_fname_store(struct kobject *kobj,
752 struct attribute *attr,
756 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
761 rc = kstrtobool(buffer, &val);
765 sbi->ll_enable_statahead_fname = val;
769 LUSTRE_RW_ATTR(enable_statahead_fname);
771 static ssize_t statahead_running_max_show(struct kobject *kobj,
772 struct attribute *attr,
775 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
778 return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_sa_running_max);
781 static ssize_t statahead_running_max_store(struct kobject *kobj,
782 struct attribute *attr,
786 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
791 rc = kstrtoul(buffer, 0, &val);
795 if (val <= LL_SA_RUNNING_MAX) {
796 sbi->ll_sa_running_max = val;
800 CERROR("Bad statahead_running_max value %lu. Valid values "
801 "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
805 LUSTRE_RW_ATTR(statahead_running_max);
807 static ssize_t statahead_batch_max_show(struct kobject *kobj,
808 struct attribute *attr,
811 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
814 return snprintf(buf, 16, "%u\n", sbi->ll_sa_batch_max);
817 static ssize_t statahead_batch_max_store(struct kobject *kobj,
818 struct attribute *attr,
822 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
827 rc = kstrtoul(buffer, 0, &val);
831 if (val > LL_SA_BATCH_MAX) {
832 CWARN("%s: statahead_batch_max value %lu limited to maximum %d\n",
833 sbi->ll_fsname, val, LL_SA_BATCH_MAX);
834 val = LL_SA_BATCH_MAX;
837 sbi->ll_sa_batch_max = val;
840 LUSTRE_RW_ATTR(statahead_batch_max);
842 static ssize_t statahead_max_show(struct kobject *kobj,
843 struct attribute *attr,
846 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
849 return sprintf(buf, "%u\n", sbi->ll_sa_max);
852 static ssize_t statahead_max_store(struct kobject *kobj,
853 struct attribute *attr,
857 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
862 rc = kstrtoul(buffer, 0, &val);
866 if (val > LL_SA_REQ_MAX) {
867 CWARN("%s: statahead_max value %lu limited to maximum %d\n",
868 sbi->ll_fsname, val, LL_SA_REQ_MAX);
872 sbi->ll_sa_max = val;
875 LUSTRE_RW_ATTR(statahead_max);
877 static ssize_t statahead_min_show(struct kobject *kobj,
878 struct attribute *attr,
881 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
884 return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_sa_min);
887 static ssize_t statahead_min_store(struct kobject *kobj,
888 struct attribute *attr,
892 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
897 rc = kstrtoul(buffer, 0, &val);
901 if (val < LL_SA_REQ_MIN)
902 CERROR("%s: bad statahead_min %lu < min %u\n",
903 sbi->ll_fsname, val, LL_SA_REQ_MIN);
904 else if (val > sbi->ll_sa_max)
905 CERROR("%s: bad statahead_min %lu > max %u\n",
906 sbi->ll_fsname, val, sbi->ll_sa_max);
908 sbi->ll_sa_min = val;
912 LUSTRE_RW_ATTR(statahead_min);
914 static ssize_t statahead_timeout_show(struct kobject *kobj,
915 struct attribute *attr,
918 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
921 return scnprintf(buf, PAGE_SIZE, "%lu\n", sbi->ll_sa_timeout);
924 static ssize_t statahead_timeout_store(struct kobject *kobj,
925 struct attribute *attr,
929 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
934 rc = kstrtoul(buffer, 0, &val);
938 sbi->ll_sa_timeout = val;
941 LUSTRE_RW_ATTR(statahead_timeout);
943 static ssize_t statahead_agl_show(struct kobject *kobj,
944 struct attribute *attr,
947 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
950 return scnprintf(buf, PAGE_SIZE, "%u\n",
951 test_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags));
954 static ssize_t statahead_agl_store(struct kobject *kobj,
955 struct attribute *attr,
959 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
964 rc = kstrtobool(buffer, &val);
969 set_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags);
971 clear_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags);
975 LUSTRE_RW_ATTR(statahead_agl);
977 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
979 struct super_block *sb = m->private;
980 struct ll_sb_info *sbi = ll_s2sbi(sb);
982 seq_printf(m, "statahead total: %u\n"
983 "statahead wrong: %u\n"
989 atomic_read(&sbi->ll_sa_total),
990 atomic_read(&sbi->ll_sa_wrong),
991 atomic_read(&sbi->ll_agl_total),
992 atomic_read(&sbi->ll_sa_list_total),
993 atomic_read(&sbi->ll_sa_fname_total),
994 atomic_read(&sbi->ll_sa_hit_total),
995 atomic_read(&sbi->ll_sa_miss_total));
999 static ssize_t ll_statahead_stats_seq_write(struct file *file,
1000 const char __user *buffer,
1001 size_t count, loff_t *off)
1003 struct seq_file *m = file->private_data;
1004 struct super_block *sb = m->private;
1005 struct ll_sb_info *sbi = ll_s2sbi(sb);
1007 atomic_set(&sbi->ll_sa_total, 0);
1008 atomic_set(&sbi->ll_sa_wrong, 0);
1009 atomic_set(&sbi->ll_agl_total, 0);
1010 atomic_set(&sbi->ll_sa_list_total, 0);
1011 atomic_set(&sbi->ll_sa_fname_total, 0);
1012 atomic_set(&sbi->ll_sa_hit_total, 0);
1013 atomic_set(&sbi->ll_sa_miss_total, 0);
1017 LDEBUGFS_SEQ_FOPS(ll_statahead_stats);
1019 static ssize_t lazystatfs_show(struct kobject *kobj,
1020 struct attribute *attr,
1023 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1026 return scnprintf(buf, PAGE_SIZE, "%u\n",
1027 test_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags));
1030 static ssize_t lazystatfs_store(struct kobject *kobj,
1031 struct attribute *attr,
1035 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1040 rc = kstrtobool(buffer, &val);
1045 set_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags);
1047 clear_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags);
1051 LUSTRE_RW_ATTR(lazystatfs);
1053 static ssize_t statfs_max_age_show(struct kobject *kobj, struct attribute *attr,
1056 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1059 return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_statfs_max_age);
1062 static ssize_t statfs_max_age_store(struct kobject *kobj,
1063 struct attribute *attr, const char *buffer,
1066 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1071 rc = kstrtouint(buffer, 10, &val);
1074 if (val > OBD_STATFS_CACHE_MAX_AGE)
1077 sbi->ll_statfs_max_age = val;
1081 LUSTRE_RW_ATTR(statfs_max_age);
1083 static ssize_t statfs_project_show(struct kobject *kobj,
1084 struct attribute *attr,
1087 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1090 return scnprintf(buf, PAGE_SIZE, "%u\n",
1091 test_bit(LL_SBI_STATFS_PROJECT, sbi->ll_flags));
1094 static ssize_t statfs_project_store(struct kobject *kobj,
1095 struct attribute *attr,
1099 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1104 rc = kstrtobool(buffer, &val);
1109 set_bit(LL_SBI_STATFS_PROJECT, sbi->ll_flags);
1111 clear_bit(LL_SBI_STATFS_PROJECT, sbi->ll_flags);
1115 LUSTRE_RW_ATTR(statfs_project);
1117 static ssize_t max_easize_show(struct kobject *kobj,
1118 struct attribute *attr,
1121 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1126 rc = ll_get_max_mdsize(sbi, &ealen);
1130 /* Limit xattr size returned to userspace based on kernel maximum */
1131 return scnprintf(buf, PAGE_SIZE, "%u\n",
1132 ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
1134 LUSTRE_RO_ATTR(max_easize);
1137 * Get default_easize.
1139 * \see client_obd::cl_default_mds_easize
1141 * \param[in] m seq_file handle
1142 * \param[in] v unused for single entry
1144 * \retval 0 on success
1145 * \retval negative negated errno on failure
1147 static ssize_t default_easize_show(struct kobject *kobj,
1148 struct attribute *attr,
1151 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1156 rc = ll_get_default_mdsize(sbi, &ealen);
1160 /* Limit xattr size returned to userspace based on kernel maximum */
1161 return scnprintf(buf, PAGE_SIZE, "%u\n",
1162 ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
1166 * Set default_easize.
1168 * Range checking on the passed value is handled by
1169 * ll_set_default_mdsize().
1171 * \see client_obd::cl_default_mds_easize
1173 * \param[in] file proc file
1174 * \param[in] buffer string passed from user space
1175 * \param[in] count \a buffer length
1176 * \param[in] off unused for single entry
1178 * \retval positive \a count on success
1179 * \retval negative negated errno on failure
1181 static ssize_t default_easize_store(struct kobject *kobj,
1182 struct attribute *attr,
1186 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1194 rc = kstrtouint(buffer, 10, &val);
1198 rc = ll_set_default_mdsize(sbi, val);
1204 LUSTRE_RW_ATTR(default_easize);
1206 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
1208 static ssize_t xattr_cache_show(struct kobject *kobj,
1209 struct attribute *attr,
1212 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1215 return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
1218 static ssize_t xattr_cache_store(struct kobject *kobj,
1219 struct attribute *attr,
1223 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1228 rc = kstrtobool(buffer, &val);
1232 if (val && !test_bit(LL_SBI_XATTR_CACHE, sbi->ll_flags))
1235 sbi->ll_xattr_cache_enabled = val;
1236 sbi->ll_xattr_cache_set = 1;
1240 LUSTRE_RW_ATTR(xattr_cache);
1242 static ssize_t tiny_write_show(struct kobject *kobj,
1243 struct attribute *attr,
1246 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1249 return scnprintf(buf, PAGE_SIZE, "%u\n",
1250 test_bit(LL_SBI_TINY_WRITE, sbi->ll_flags));
1253 static ssize_t tiny_write_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 set_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
1271 clear_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
1272 spin_unlock(&sbi->ll_lock);
1276 LUSTRE_RW_ATTR(tiny_write);
1278 static ssize_t unaligned_dio_show(struct kobject *kobj,
1279 struct attribute *attr,
1282 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1285 return scnprintf(buf, PAGE_SIZE, "%u\n",
1286 test_bit(LL_SBI_UNALIGNED_DIO, sbi->ll_flags));
1289 static ssize_t unaligned_dio_store(struct kobject *kobj,
1290 struct attribute *attr,
1294 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1299 rc = kstrtobool(buffer, &val);
1303 spin_lock(&sbi->ll_lock);
1305 set_bit(LL_SBI_UNALIGNED_DIO, sbi->ll_flags);
1307 clear_bit(LL_SBI_UNALIGNED_DIO, sbi->ll_flags);
1308 spin_unlock(&sbi->ll_lock);
1312 LUSTRE_RW_ATTR(unaligned_dio);
1314 static ssize_t parallel_dio_show(struct kobject *kobj,
1315 struct attribute *attr,
1318 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1321 return snprintf(buf, PAGE_SIZE, "%u\n",
1322 test_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags));
1325 static ssize_t parallel_dio_store(struct kobject *kobj,
1326 struct attribute *attr,
1330 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1335 rc = kstrtobool(buffer, &val);
1339 spin_lock(&sbi->ll_lock);
1341 set_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
1343 clear_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
1344 spin_unlock(&sbi->ll_lock);
1348 LUSTRE_RW_ATTR(parallel_dio);
1350 static ssize_t hybrid_io_show(struct kobject *kobj, struct attribute *attr,
1353 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1356 return snprintf(buf, PAGE_SIZE, "%u\n",
1357 test_bit(LL_SBI_HYBRID_IO, sbi->ll_flags));
1360 static ssize_t hybrid_io_store(struct kobject *kobj, struct attribute *attr,
1361 const char *buffer, size_t count)
1363 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1368 rc = kstrtobool(buffer, &val);
1372 spin_lock(&sbi->ll_lock);
1374 set_bit(LL_SBI_HYBRID_IO, sbi->ll_flags);
1376 clear_bit(LL_SBI_HYBRID_IO, sbi->ll_flags);
1377 spin_unlock(&sbi->ll_lock);
1381 LUSTRE_RW_ATTR(hybrid_io);
1383 static ssize_t max_read_ahead_async_active_show(struct kobject *kobj,
1384 struct attribute *attr,
1387 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1390 return scnprintf(buf, PAGE_SIZE, "%u\n",
1391 sbi->ll_ra_info.ra_async_max_active);
1394 static ssize_t max_read_ahead_async_active_store(struct kobject *kobj,
1395 struct attribute *attr,
1399 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1404 rc = kstrtouint(buffer, 10, &val);
1409 * It doesn't make any sense to make it exceed what
1410 * workqueue could acutally support. This can easily
1411 * over subscripe the cores but Lustre internally
1412 * throttles to avoid those impacts.
1414 if (val > WQ_UNBOUND_MAX_ACTIVE) {
1415 CERROR("%s: cannot set max_read_ahead_async_active=%u larger than %u\n",
1416 sbi->ll_fsname, val, WQ_UNBOUND_MAX_ACTIVE);
1420 spin_lock(&sbi->ll_lock);
1421 sbi->ll_ra_info.ra_async_max_active = val;
1422 spin_unlock(&sbi->ll_lock);
1426 LUSTRE_RW_ATTR(max_read_ahead_async_active);
1428 static ssize_t read_ahead_async_file_threshold_mb_show(struct kobject *kobj,
1429 struct attribute *attr,
1432 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1435 return scnprintf(buf, PAGE_SIZE, "%lu\n", PAGES_TO_MiB(
1436 sbi->ll_ra_info.ra_async_pages_per_file_threshold));
1440 read_ahead_async_file_threshold_mb_store(struct kobject *kobj,
1441 struct attribute *attr,
1442 const char *buffer, size_t count)
1444 unsigned long pages_number;
1445 unsigned long max_ra_per_file;
1446 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1450 rc = kstrtoul(buffer, 10, &pages_number);
1454 pages_number = MiB_TO_PAGES(pages_number);
1455 max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1456 if (pages_number < 0 || pages_number > max_ra_per_file) {
1457 CERROR("%s: can't set read_ahead_async_file_threshold_mb=%lu > "
1458 "max_read_readahead_per_file_mb=%lu\n", sbi->ll_fsname,
1459 PAGES_TO_MiB(pages_number),
1460 PAGES_TO_MiB(max_ra_per_file));
1463 sbi->ll_ra_info.ra_async_pages_per_file_threshold = pages_number;
1467 LUSTRE_RW_ATTR(read_ahead_async_file_threshold_mb);
1469 static ssize_t read_ahead_range_kb_show(struct kobject *kobj,
1470 struct attribute *attr, char *buf)
1472 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1475 return snprintf(buf, PAGE_SIZE, "%lu\n",
1476 sbi->ll_ra_info.ra_range_pages << (PAGE_SHIFT - 10));
1480 read_ahead_range_kb_store(struct kobject *kobj,
1481 struct attribute *attr,
1482 const char *buffer, size_t count)
1484 unsigned long pages_number;
1485 unsigned long max_ra_per_file;
1487 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1491 rc = sysfs_memparse(buffer, count, &val, "KiB");
1495 pages_number = val >> PAGE_SHIFT;
1496 /* Disable mmap range read */
1497 if (pages_number == 0)
1500 max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1501 if (pages_number > max_ra_per_file ||
1502 pages_number < RA_MIN_MMAP_RANGE_PAGES)
1506 spin_lock(&sbi->ll_lock);
1507 sbi->ll_ra_info.ra_range_pages = pages_number;
1508 spin_unlock(&sbi->ll_lock);
1512 LUSTRE_RW_ATTR(read_ahead_range_kb);
1514 static ssize_t fast_read_show(struct kobject *kobj,
1515 struct attribute *attr,
1518 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1521 return scnprintf(buf, PAGE_SIZE, "%u\n",
1522 test_bit(LL_SBI_FAST_READ, sbi->ll_flags));
1525 static ssize_t fast_read_store(struct kobject *kobj,
1526 struct attribute *attr,
1530 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1535 rc = kstrtobool(buffer, &val);
1539 spin_lock(&sbi->ll_lock);
1541 set_bit(LL_SBI_FAST_READ, sbi->ll_flags);
1543 clear_bit(LL_SBI_FAST_READ, sbi->ll_flags);
1544 spin_unlock(&sbi->ll_lock);
1548 LUSTRE_RW_ATTR(fast_read);
1550 static ssize_t file_heat_show(struct kobject *kobj,
1551 struct attribute *attr,
1554 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1557 return scnprintf(buf, PAGE_SIZE, "%u\n",
1558 test_bit(LL_SBI_FILE_HEAT, sbi->ll_flags));
1561 static ssize_t file_heat_store(struct kobject *kobj,
1562 struct attribute *attr,
1566 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1571 rc = kstrtobool(buffer, &val);
1575 spin_lock(&sbi->ll_lock);
1577 set_bit(LL_SBI_FILE_HEAT, sbi->ll_flags);
1579 clear_bit(LL_SBI_FILE_HEAT, sbi->ll_flags);
1580 spin_unlock(&sbi->ll_lock);
1584 LUSTRE_RW_ATTR(file_heat);
1586 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1587 struct attribute *attr,
1590 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1593 return scnprintf(buf, PAGE_SIZE, "%u\n",
1594 (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1597 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1598 struct attribute *attr,
1602 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1607 rc = kstrtoul(buffer, 10, &val);
1611 if (val < 0 || val > 100)
1614 sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1618 LUSTRE_RW_ATTR(heat_decay_percentage);
1620 static ssize_t heat_period_second_show(struct kobject *kobj,
1621 struct attribute *attr,
1624 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1627 return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1630 static ssize_t heat_period_second_store(struct kobject *kobj,
1631 struct attribute *attr,
1635 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1640 rc = kstrtoul(buffer, 10, &val);
1647 sbi->ll_heat_period_second = val;
1651 LUSTRE_RW_ATTR(heat_period_second);
1653 static ssize_t opencache_threshold_count_show(struct kobject *kobj,
1654 struct attribute *attr,
1657 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1660 if (sbi->ll_oc_thrsh_count)
1661 return snprintf(buf, PAGE_SIZE, "%u\n",
1662 sbi->ll_oc_thrsh_count);
1664 return snprintf(buf, PAGE_SIZE, "off\n");
1667 static ssize_t opencache_threshold_count_store(struct kobject *kobj,
1668 struct attribute *attr,
1672 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1677 rc = kstrtouint(buffer, 10, &val);
1680 /* also accept "off" to disable and "on" to always cache */
1681 rc = kstrtobool(buffer, &enable);
1686 sbi->ll_oc_thrsh_count = val;
1690 LUSTRE_RW_ATTR(opencache_threshold_count);
1692 static ssize_t opencache_threshold_ms_show(struct kobject *kobj,
1693 struct attribute *attr,
1696 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1699 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_oc_thrsh_ms);
1702 static ssize_t opencache_threshold_ms_store(struct kobject *kobj,
1703 struct attribute *attr,
1707 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1712 rc = kstrtouint(buffer, 10, &val);
1716 sbi->ll_oc_thrsh_ms = val;
1720 LUSTRE_RW_ATTR(opencache_threshold_ms);
1722 static ssize_t opencache_max_ms_show(struct kobject *kobj,
1723 struct attribute *attr,
1726 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1729 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_oc_max_ms);
1732 static ssize_t opencache_max_ms_store(struct kobject *kobj,
1733 struct attribute *attr,
1737 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1742 rc = kstrtouint(buffer, 10, &val);
1746 sbi->ll_oc_max_ms = val;
1750 LUSTRE_RW_ATTR(opencache_max_ms);
1752 static ssize_t inode_cache_show(struct kobject *kobj,
1753 struct attribute *attr,
1756 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1759 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_inode_cache_enabled);
1762 static ssize_t inode_cache_store(struct kobject *kobj,
1763 struct attribute *attr,
1767 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1772 rc = kstrtobool(buffer, &val);
1776 sbi->ll_inode_cache_enabled = val;
1780 LUSTRE_RW_ATTR(inode_cache);
1782 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1784 struct super_block *sb = m->private;
1785 struct ll_sb_info *sbi = ll_s2sbi(sb);
1786 struct cl_client_cache *cache = sbi->ll_cache;
1790 pages = atomic_long_read(&cache->ccc_unstable_nr);
1791 mb = (pages * PAGE_SIZE) >> 20;
1793 seq_printf(m, "unstable_check: %8d\n"
1794 "unstable_pages: %12ld\n"
1795 "unstable_mb: %8d\n",
1796 cache->ccc_unstable_check, pages, mb);
1800 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1801 const char __user *buffer,
1802 size_t count, loff_t *unused)
1804 struct seq_file *seq = file->private_data;
1805 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1812 if (count >= sizeof(kernbuf))
1815 if (copy_from_user(kernbuf, buffer, count))
1819 buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1821 rc = kstrtobool_from_user(buffer, count, &val);
1825 /* borrow lru lock to set the value */
1826 spin_lock(&sbi->ll_cache->ccc_lru_lock);
1827 sbi->ll_cache->ccc_unstable_check = val;
1828 spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1833 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1835 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1837 struct super_block *sb = m->private;
1838 struct ll_sb_info *sbi = ll_s2sbi(sb);
1839 struct root_squash_info *squash = &sbi->ll_squash;
1841 seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1845 static ssize_t ll_root_squash_seq_write(struct file *file,
1846 const char __user *buffer,
1847 size_t count, loff_t *off)
1849 struct seq_file *m = file->private_data;
1850 struct super_block *sb = m->private;
1851 struct ll_sb_info *sbi = ll_s2sbi(sb);
1852 struct root_squash_info *squash = &sbi->ll_squash;
1854 return lprocfs_wr_root_squash(buffer, count, squash, sbi->ll_fsname);
1857 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1859 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1861 struct super_block *sb = m->private;
1862 struct ll_sb_info *sbi = ll_s2sbi(sb);
1863 struct root_squash_info *squash = &sbi->ll_squash;
1866 spin_lock(&squash->rsi_lock);
1867 if (!list_empty(&squash->rsi_nosquash_nids)) {
1868 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1869 &squash->rsi_nosquash_nids);
1873 seq_puts(m, "NONE\n");
1875 spin_unlock(&squash->rsi_lock);
1880 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1881 const char __user *buffer,
1882 size_t count, loff_t *off)
1884 struct seq_file *m = file->private_data;
1885 struct super_block *sb = m->private;
1886 struct ll_sb_info *sbi = ll_s2sbi(sb);
1887 struct root_squash_info *squash = &sbi->ll_squash;
1890 rc = lprocfs_wr_nosquash_nids(buffer, count, squash, sbi->ll_fsname);
1894 ll_compute_rootsquash_state(sbi);
1899 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1901 #if defined(CONFIG_LL_ENCRYPTION)
1902 static ssize_t enable_filename_encryption_show(struct kobject *kobj,
1903 struct attribute *attr,
1906 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1908 struct lustre_sb_info *lsi = sbi->lsi;
1910 return snprintf(buffer, PAGE_SIZE, "%u\n",
1911 lsi->lsi_flags & LSI_FILENAME_ENC ? 1 : 0);
1914 static ssize_t enable_filename_encryption_store(struct kobject *kobj,
1915 struct attribute *attr,
1919 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1921 struct lustre_sb_info *lsi = sbi->lsi;
1925 rc = kstrtobool(buffer, &val);
1930 if (!ll_sbi_has_name_encrypt(sbi)) {
1931 /* server does not support name encryption,
1932 * so force it to NULL on client
1934 CDEBUG(D_SEC, "%s: server does not support name encryption\n",
1936 lsi->lsi_flags &= ~LSI_FILENAME_ENC;
1940 lsi->lsi_flags |= LSI_FILENAME_ENC;
1942 lsi->lsi_flags &= ~LSI_FILENAME_ENC;
1948 LUSTRE_RW_ATTR(enable_filename_encryption);
1949 #endif /* CONFIG_LL_ENCRYPTION */
1951 #if defined(CONFIG_LL_ENCRYPTION) || defined(HAVE_LUSTRE_CRYPTO)
1952 static ssize_t filename_enc_use_old_base64_show(struct kobject *kobj,
1953 struct attribute *attr,
1956 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1958 struct lustre_sb_info *lsi = sbi->lsi;
1960 return snprintf(buffer, PAGE_SIZE, "%u\n",
1961 lsi->lsi_flags & LSI_FILENAME_ENC_B64_OLD_CLI ? 1 : 0);
1964 static ssize_t filename_enc_use_old_base64_store(struct kobject *kobj,
1965 struct attribute *attr,
1969 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1971 struct lustre_sb_info *lsi = sbi->lsi;
1975 rc = kstrtobool(buffer, &val);
1980 if (!ll_sbi_has_name_encrypt(sbi)) {
1981 /* server does not support name encryption,
1982 * so force it to NULL on client
1985 "%s: server does not support name encryption\n",
1987 lsi->lsi_flags &= ~LSI_FILENAME_ENC_B64_OLD_CLI;
1991 lsi->lsi_flags |= LSI_FILENAME_ENC_B64_OLD_CLI;
1993 lsi->lsi_flags &= ~LSI_FILENAME_ENC_B64_OLD_CLI;
1999 LUSTRE_RW_ATTR(filename_enc_use_old_base64);
2000 #endif /* CONFIG_LL_ENCRYPTION || HAVE_LUSTRE_CRYPTO */
2002 static int ll_pcc_seq_show(struct seq_file *m, void *v)
2004 struct super_block *sb = m->private;
2005 struct ll_sb_info *sbi = ll_s2sbi(sb);
2007 return pcc_super_dump(&sbi->ll_pcc_super, m);
2010 static ssize_t ll_pcc_seq_write(struct file *file, const char __user *buffer,
2011 size_t count, loff_t *off)
2013 struct seq_file *m = file->private_data;
2014 struct super_block *sb = m->private;
2015 struct ll_sb_info *sbi = ll_s2sbi(sb);
2019 if (count >= LPROCFS_WR_PCC_MAX_CMD)
2022 if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_PCC))
2025 OBD_ALLOC(kernbuf, count + 1);
2026 if (kernbuf == NULL)
2029 if (copy_from_user(kernbuf, buffer, count))
2030 GOTO(out_free_kernbuff, rc = -EFAULT);
2032 rc = pcc_cmd_handle(kernbuf, count, &sbi->ll_pcc_super);
2034 OBD_FREE(kernbuf, count + 1);
2035 return rc ? rc : count;
2037 LDEBUGFS_SEQ_FOPS(ll_pcc);
2039 struct ldebugfs_vars lprocfs_llite_obd_vars[] = {
2041 .fops = &ll_site_stats_fops },
2042 { .name = "max_cached_mb",
2043 .fops = &ll_max_cached_mb_fops },
2044 { .name = "statahead_stats",
2045 .fops = &ll_statahead_stats_fops },
2046 { .name = "unstable_stats",
2047 .fops = &ll_unstable_stats_fops },
2048 { .name = "sbi_flags",
2049 .fops = &ll_sbi_flags_fops },
2050 { .name = "root_squash",
2051 .fops = &ll_root_squash_fops },
2052 { .name = "nosquash_nids",
2053 .fops = &ll_nosquash_nids_fops },
2055 .fops = &ll_pcc_fops, },
2059 #define MAX_STRING_SIZE 128
2061 static struct attribute *llite_attrs[] = {
2062 &lustre_attr_blocksize.attr,
2063 &lustre_attr_stat_blocksize.attr,
2064 &lustre_attr_kbytestotal.attr,
2065 &lustre_attr_kbytesfree.attr,
2066 &lustre_attr_kbytesavail.attr,
2067 &lustre_attr_filestotal.attr,
2068 &lustre_attr_filesfree.attr,
2069 &lustre_attr_client_type.attr,
2070 &lustre_attr_foreign_symlink_enable.attr,
2071 &lustre_attr_foreign_symlink_prefix.attr,
2072 &lustre_attr_foreign_symlink_upcall.attr,
2073 &lustre_attr_foreign_symlink_upcall_info.attr,
2074 &lustre_attr_fstype.attr,
2075 &lustre_attr_uuid.attr,
2076 &lustre_attr_checksums.attr,
2077 &lustre_attr_checksum_pages.attr,
2078 &lustre_attr_max_read_ahead_mb.attr,
2079 &lustre_attr_max_read_ahead_per_file_mb.attr,
2080 &lustre_attr_max_read_ahead_whole_mb.attr,
2081 &lustre_attr_max_read_ahead_async_active.attr,
2082 &lustre_attr_read_ahead_async_file_threshold_mb.attr,
2083 &lustre_attr_read_ahead_range_kb.attr,
2084 &lustre_attr_stats_track_pid.attr,
2085 &lustre_attr_stats_track_ppid.attr,
2086 &lustre_attr_stats_track_gid.attr,
2087 &lustre_attr_enable_statahead_fname.attr,
2088 &lustre_attr_statahead_running_max.attr,
2089 &lustre_attr_statahead_batch_max.attr,
2090 &lustre_attr_statahead_max.attr,
2091 &lustre_attr_statahead_min.attr,
2092 &lustre_attr_statahead_timeout.attr,
2093 &lustre_attr_statahead_agl.attr,
2094 &lustre_attr_lazystatfs.attr,
2095 &lustre_attr_statfs_max_age.attr,
2096 &lustre_attr_statfs_project.attr,
2097 &lustre_attr_max_easize.attr,
2098 &lustre_attr_default_easize.attr,
2099 &lustre_attr_xattr_cache.attr,
2100 &lustre_attr_fast_read.attr,
2101 &lustre_attr_tiny_write.attr,
2102 &lustre_attr_parallel_dio.attr,
2103 &lustre_attr_unaligned_dio.attr,
2104 &lustre_attr_hybrid_io.attr,
2105 &lustre_attr_file_heat.attr,
2106 &lustre_attr_heat_decay_percentage.attr,
2107 &lustre_attr_heat_period_second.attr,
2108 &lustre_attr_opencache_threshold_count.attr,
2109 &lustre_attr_opencache_threshold_ms.attr,
2110 &lustre_attr_opencache_max_ms.attr,
2111 &lustre_attr_inode_cache.attr,
2112 #ifdef CONFIG_LL_ENCRYPTION
2113 &lustre_attr_enable_filename_encryption.attr,
2115 #if defined(CONFIG_LL_ENCRYPTION) || defined(HAVE_LUSTRE_CRYPTO)
2116 &lustre_attr_filename_enc_use_old_base64.attr,
2121 KOBJ_ATTRIBUTE_GROUPS(llite); /* creates llite_groups */
2123 static void sbi_kobj_release(struct kobject *kobj)
2125 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
2127 complete(&sbi->ll_kobj_unregister);
2130 static struct kobj_type sbi_ktype = {
2131 .default_groups = KOBJ_ATTR_GROUPS(llite),
2132 .sysfs_ops = &lustre_sysfs_ops,
2133 .release = sbi_kobj_release,
2136 static const struct llite_file_opcode {
2138 enum lprocfs_counter_config lfo_config;
2139 const char *lfo_opname;
2140 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
2141 /* file operation */
2142 { LPROC_LL_READ_BYTES, LPROCFS_TYPE_BYTES_FULL, "read_bytes" },
2143 { LPROC_LL_WRITE_BYTES, LPROCFS_TYPE_BYTES_FULL, "write_bytes" },
2144 { LPROC_LL_READ, LPROCFS_TYPE_LATENCY, "read" },
2145 { LPROC_LL_WRITE, LPROCFS_TYPE_LATENCY, "write" },
2146 { LPROC_LL_IOCTL, LPROCFS_TYPE_REQS, "ioctl" },
2147 { LPROC_LL_OPEN, LPROCFS_TYPE_LATENCY, "open" },
2148 { LPROC_LL_RELEASE, LPROCFS_TYPE_LATENCY, "close" },
2149 { LPROC_LL_MMAP, LPROCFS_TYPE_LATENCY, "mmap" },
2150 { LPROC_LL_FAULT, LPROCFS_TYPE_LATENCY, "page_fault" },
2151 { LPROC_LL_MKWRITE, LPROCFS_TYPE_LATENCY, "page_mkwrite" },
2152 { LPROC_LL_LLSEEK, LPROCFS_TYPE_LATENCY, "seek" },
2153 { LPROC_LL_FSYNC, LPROCFS_TYPE_LATENCY, "fsync" },
2154 { LPROC_LL_READDIR, LPROCFS_TYPE_LATENCY, "readdir" },
2155 { LPROC_LL_INODE_OCOUNT, LPROCFS_TYPE_REQS | LPROCFS_CNTR_AVGMINMAX |
2156 LPROCFS_CNTR_STDDEV, "opencount" },
2157 { LPROC_LL_INODE_OPCLTM, LPROCFS_TYPE_LATENCY, "openclosetime" },
2158 /* inode operation */
2159 { LPROC_LL_SETATTR, LPROCFS_TYPE_LATENCY, "setattr" },
2160 { LPROC_LL_TRUNC, LPROCFS_TYPE_LATENCY, "truncate" },
2161 { LPROC_LL_FLOCK, LPROCFS_TYPE_LATENCY, "flock" },
2162 { LPROC_LL_GETATTR, LPROCFS_TYPE_LATENCY, "getattr" },
2163 { LPROC_LL_FALLOCATE, LPROCFS_TYPE_LATENCY, "fallocate"},
2164 /* dir inode operation */
2165 { LPROC_LL_CREATE, LPROCFS_TYPE_LATENCY, "create" },
2166 { LPROC_LL_LINK, LPROCFS_TYPE_LATENCY, "link" },
2167 { LPROC_LL_UNLINK, LPROCFS_TYPE_LATENCY, "unlink" },
2168 { LPROC_LL_SYMLINK, LPROCFS_TYPE_LATENCY, "symlink" },
2169 { LPROC_LL_MKDIR, LPROCFS_TYPE_LATENCY, "mkdir" },
2170 { LPROC_LL_RMDIR, LPROCFS_TYPE_LATENCY, "rmdir" },
2171 { LPROC_LL_MKNOD, LPROCFS_TYPE_LATENCY, "mknod" },
2172 { LPROC_LL_RENAME, LPROCFS_TYPE_LATENCY, "rename" },
2173 /* special inode operation */
2174 { LPROC_LL_STATFS, LPROCFS_TYPE_LATENCY, "statfs" },
2175 { LPROC_LL_SETXATTR, LPROCFS_TYPE_LATENCY, "setxattr" },
2176 { LPROC_LL_GETXATTR, LPROCFS_TYPE_LATENCY, "getxattr" },
2177 { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REQS, "getxattr_hits" },
2178 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_LATENCY, "listxattr" },
2179 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_LATENCY, "removexattr" },
2180 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_LATENCY, "inode_permission" },
2183 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count)
2188 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
2189 lprocfs_counter_add(sbi->ll_stats, op, count);
2190 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
2191 sbi->ll_stats_track_id == current->pid)
2192 lprocfs_counter_add(sbi->ll_stats, op, count);
2193 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
2194 sbi->ll_stats_track_id == current->real_parent->pid)
2195 lprocfs_counter_add(sbi->ll_stats, op, count);
2196 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
2197 sbi->ll_stats_track_id ==
2198 from_kgid(&init_user_ns, current_gid()))
2199 lprocfs_counter_add(sbi->ll_stats, op, count);
2201 EXPORT_SYMBOL(ll_stats_ops_tally);
2203 static const char *const ra_stat_string[] = {
2204 [RA_STAT_HIT] = "hits",
2205 [RA_STAT_MISS] = "misses",
2206 [RA_STAT_DISTANT_READPAGE] = "readpage_not_consecutive",
2207 [RA_STAT_MISS_IN_WINDOW] = "miss_inside_window",
2208 [RA_STAT_FAILED_GRAB_PAGE] = "failed_grab_cache_page",
2209 [RA_STAT_FAILED_MATCH] = "failed_lock_match",
2210 [RA_STAT_DISCARDED] = "read_but_discarded",
2211 [RA_STAT_ZERO_LEN] = "zero_length_file",
2212 [RA_STAT_ZERO_WINDOW] = "zero_size_window",
2213 [RA_STAT_EOF] = "readahead_to_eof",
2214 [RA_STAT_MAX_IN_FLIGHT] = "hit_max_readahead_issue",
2215 [RA_STAT_WRONG_GRAB_PAGE] = "wrong_page_from_grab_cache_page",
2216 [RA_STAT_FAILED_REACH_END] = "failed_to_reach_end",
2217 [RA_STAT_ASYNC] = "async_readahead",
2218 [RA_STAT_FAILED_FAST_READ] = "failed_to_fast_read",
2219 [RA_STAT_MMAP_RANGE_READ] = "mmap_range_read",
2220 [RA_STAT_READAHEAD_PAGES] = "readahead_pages"
2223 int ll_debugfs_register_super(struct super_block *sb, const char *name)
2225 struct lustre_sb_info *lsi = s2lsi(sb);
2226 struct ll_sb_info *sbi = ll_s2sbi(sb);
2232 if (IS_ERR_OR_NULL(llite_root))
2235 sbi->ll_debugfs_entry = debugfs_create_dir(name, llite_root);
2236 ldebugfs_add_vars(sbi->ll_debugfs_entry, lprocfs_llite_obd_vars, sb);
2238 debugfs_create_file("dump_page_cache", 0444, sbi->ll_debugfs_entry, sbi,
2239 &vvp_dump_pgcache_file_ops);
2241 debugfs_create_file("extents_stats", 0644, sbi->ll_debugfs_entry, sbi,
2242 &ll_rw_extents_stats_fops);
2244 debugfs_create_file("extents_stats_per_process", 0644,
2245 sbi->ll_debugfs_entry, sbi,
2246 &ll_rw_extents_stats_pp_fops);
2248 debugfs_create_file("offset_stats", 0644, sbi->ll_debugfs_entry, sbi,
2249 &ll_rw_offset_stats_fops);
2251 /* File operations stats */
2252 sbi->ll_stats = lprocfs_stats_alloc(LPROC_LL_FILE_OPCODES,
2253 LPROCFS_STATS_FLAG_NONE);
2254 if (sbi->ll_stats == NULL)
2255 GOTO(out_debugfs, err = -ENOMEM);
2257 /* do counter init */
2258 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++)
2259 lprocfs_counter_init(sbi->ll_stats,
2260 llite_opcode_table[id].lfo_opcode,
2261 llite_opcode_table[id].lfo_config,
2262 llite_opcode_table[id].lfo_opname);
2264 debugfs_create_file("stats", 0644, sbi->ll_debugfs_entry,
2265 sbi->ll_stats, &ldebugfs_stats_seq_fops);
2267 sbi->ll_ra_stats = lprocfs_stats_alloc(ARRAY_SIZE(ra_stat_string),
2268 LPROCFS_STATS_FLAG_NONE);
2269 if (sbi->ll_ra_stats == NULL)
2270 GOTO(out_stats, err = -ENOMEM);
2272 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++) {
2273 if (id == RA_STAT_READAHEAD_PAGES)
2274 lprocfs_counter_init(sbi->ll_ra_stats, id,
2275 LPROCFS_TYPE_PAGES |
2276 LPROCFS_CNTR_AVGMINMAX,
2277 ra_stat_string[id]);
2279 lprocfs_counter_init(sbi->ll_ra_stats, id,
2281 ra_stat_string[id]);
2284 debugfs_create_file("read_ahead_stats", 0644, sbi->ll_debugfs_entry,
2285 sbi->ll_ra_stats, &ldebugfs_stats_seq_fops);
2288 /* Yes we also register sysfs mount kset here as well */
2289 sbi->ll_kset.kobj.parent = llite_kobj;
2290 sbi->ll_kset.kobj.ktype = &sbi_ktype;
2291 init_completion(&sbi->ll_kobj_unregister);
2292 err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
2294 GOTO(out_ra_stats, err);
2296 err = kset_register(&sbi->ll_kset);
2298 GOTO(out_ra_stats, err);
2300 lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
2304 lprocfs_stats_free(&sbi->ll_ra_stats);
2306 lprocfs_stats_free(&sbi->ll_stats);
2308 debugfs_remove_recursive(sbi->ll_debugfs_entry);
2313 void ll_debugfs_unregister_super(struct super_block *sb)
2315 struct lustre_sb_info *lsi = s2lsi(sb);
2316 struct ll_sb_info *sbi = ll_s2sbi(sb);
2318 debugfs_remove_recursive(sbi->ll_debugfs_entry);
2321 sysfs_remove_link(&sbi->ll_kset.kobj,
2322 sbi->ll_dt_obd->obd_type->typ_name);
2325 sysfs_remove_link(&sbi->ll_kset.kobj,
2326 sbi->ll_md_obd->obd_type->typ_name);
2328 kobject_put(lsi->lsi_kobj);
2330 kset_unregister(&sbi->ll_kset);
2331 wait_for_completion(&sbi->ll_kobj_unregister);
2333 lprocfs_stats_free(&sbi->ll_ra_stats);
2334 lprocfs_stats_free(&sbi->ll_stats);
2336 #undef MAX_STRING_SIZE
2338 static void ll_display_extents_info(struct ll_rw_extents_info *rw_extents,
2339 struct seq_file *seq, int which)
2341 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
2342 unsigned long start, end, r, w;
2343 char *unitp = "KMGTPEZY";
2345 struct per_process_info *pp_info;
2347 pp_info = &rw_extents->pp_extents[which];
2352 for (i = 0; i < LL_HIST_MAX; i++) {
2353 read_tot += pp_info->pp_r_hist.oh_buckets[i];
2354 write_tot += pp_info->pp_w_hist.oh_buckets[i];
2357 for (i = 0; i < LL_HIST_MAX; i++) {
2358 r = pp_info->pp_r_hist.oh_buckets[i];
2359 w = pp_info->pp_w_hist.oh_buckets[i];
2362 end = 1 << (i + LL_HIST_START - units);
2363 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u | "
2364 "%14lu %4u %4u\n", start, *unitp, end, *unitp,
2365 (i == LL_HIST_MAX - 1) ? '+' : ' ',
2366 r, pct(r, read_tot), pct(read_cum, read_tot),
2367 w, pct(w, write_tot), pct(write_cum, write_tot));
2369 if (start == (1 << 10)) {
2374 if (read_cum == read_tot && write_cum == write_tot)
2379 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
2381 struct ll_sb_info *sbi = seq->private;
2382 struct ll_rw_extents_info *rw_extents = sbi->ll_rw_extents_info;
2385 if (!sbi->ll_rw_stats_on || !rw_extents) {
2386 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2390 spin_lock(&sbi->ll_pp_extent_lock);
2391 lprocfs_stats_header(seq, ktime_get_real(), rw_extents->pp_init, 25,
2393 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
2394 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
2395 "extents", "calls", "%", "cum%", "calls", "%", "cum%");
2397 for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
2398 if (rw_extents->pp_extents[k].pid != 0) {
2399 seq_printf(seq, "\nPID: %d\n",
2400 rw_extents->pp_extents[k].pid);
2401 ll_display_extents_info(rw_extents, seq, k);
2404 spin_unlock(&sbi->ll_pp_extent_lock);
2408 static int alloc_rw_stats_info(struct ll_sb_info *sbi)
2410 struct ll_rw_extents_info *rw_extents;
2411 struct ll_rw_process_info *offset;
2412 struct ll_rw_process_info *process;
2415 OBD_ALLOC(rw_extents, sizeof(*rw_extents));
2419 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
2420 spin_lock_init(&rw_extents->pp_extents[i].pp_r_hist.oh_lock);
2421 spin_lock_init(&rw_extents->pp_extents[i].pp_w_hist.oh_lock);
2423 rw_extents->pp_init = ktime_get_real();
2425 spin_lock(&sbi->ll_pp_extent_lock);
2426 if (!sbi->ll_rw_extents_info)
2427 sbi->ll_rw_extents_info = rw_extents;
2428 spin_unlock(&sbi->ll_pp_extent_lock);
2429 /* another writer allocated the struct before we got the lock */
2430 if (sbi->ll_rw_extents_info != rw_extents)
2431 OBD_FREE(rw_extents, sizeof(*rw_extents));
2433 OBD_ALLOC(process, sizeof(*process) * LL_PROCESS_HIST_MAX);
2435 GOTO(out, rc = -ENOMEM);
2436 OBD_ALLOC(offset, sizeof(*offset) * LL_OFFSET_HIST_MAX);
2438 GOTO(out_free, rc = -ENOMEM);
2440 spin_lock(&sbi->ll_process_lock);
2441 if (!sbi->ll_rw_process_info)
2442 sbi->ll_rw_process_info = process;
2443 if (!sbi->ll_rw_offset_info)
2444 sbi->ll_rw_offset_info = offset;
2445 spin_unlock(&sbi->ll_process_lock);
2446 sbi->ll_process_stats_init = ktime_get_real();
2448 /* another writer allocated the structs before we got the lock */
2449 if (sbi->ll_rw_offset_info != offset)
2450 OBD_FREE(offset, sizeof(*offset) * LL_OFFSET_HIST_MAX);
2451 if (sbi->ll_rw_process_info != process) {
2453 OBD_FREE(process, sizeof(*process) * LL_PROCESS_HIST_MAX);
2460 void ll_free_rw_stats_info(struct ll_sb_info *sbi)
2462 if (sbi->ll_rw_extents_info) {
2463 OBD_FREE(sbi->ll_rw_extents_info,
2464 sizeof(*sbi->ll_rw_extents_info));
2465 sbi->ll_rw_extents_info = NULL;
2467 if (sbi->ll_rw_offset_info) {
2468 OBD_FREE(sbi->ll_rw_offset_info,
2469 sizeof(*sbi->ll_rw_offset_info) * LL_OFFSET_HIST_MAX);
2470 sbi->ll_rw_offset_info = NULL;
2472 if (sbi->ll_rw_process_info) {
2473 OBD_FREE(sbi->ll_rw_process_info,
2474 sizeof(*sbi->ll_rw_process_info) * LL_PROCESS_HIST_MAX);
2475 sbi->ll_rw_process_info = NULL;
2479 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
2480 const char __user *buf,
2481 size_t len, loff_t *off)
2483 struct seq_file *seq = file->private_data;
2484 struct ll_sb_info *sbi = seq->private;
2485 struct ll_rw_extents_info *rw_extents;
2492 value = ll_stats_pid_write(buf, len);
2495 sbi->ll_rw_stats_on = 0;
2497 if (!sbi->ll_rw_extents_info) {
2498 int rc = alloc_rw_stats_info(sbi);
2503 sbi->ll_rw_stats_on = 1;
2507 spin_lock(&sbi->ll_pp_extent_lock);
2508 rw_extents = sbi->ll_rw_extents_info;
2510 rw_extents->pp_init = ktime_get_real();
2511 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2512 rw_extents->pp_extents[i].pid = 0;
2513 lprocfs_oh_clear(&rw_extents->pp_extents[i].pp_r_hist);
2514 lprocfs_oh_clear(&rw_extents->pp_extents[i].pp_w_hist);
2517 spin_unlock(&sbi->ll_pp_extent_lock);
2522 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
2524 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
2526 struct ll_sb_info *sbi = seq->private;
2527 struct ll_rw_extents_info *rw_extents = sbi->ll_rw_extents_info;
2529 if (!sbi->ll_rw_stats_on || !rw_extents) {
2530 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2534 spin_lock(&sbi->ll_lock);
2535 lprocfs_stats_header(seq, ktime_get_real(), rw_extents->pp_init, 25,
2538 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
2539 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
2540 "extents", "calls", "%", "cum%",
2541 "calls", "%", "cum%");
2543 ll_display_extents_info(rw_extents, seq, LL_PROCESS_HIST_MAX);
2544 spin_unlock(&sbi->ll_lock);
2549 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
2550 const char __user *buf,
2551 size_t len, loff_t *off)
2553 struct seq_file *seq = file->private_data;
2554 struct ll_sb_info *sbi = seq->private;
2555 struct ll_rw_extents_info *rw_extents;
2562 value = ll_stats_pid_write(buf, len);
2565 sbi->ll_rw_stats_on = 0;
2567 if (!sbi->ll_rw_extents_info) {
2568 int rc = alloc_rw_stats_info(sbi);
2573 sbi->ll_rw_stats_on = 1;
2576 spin_lock(&sbi->ll_pp_extent_lock);
2577 rw_extents = sbi->ll_rw_extents_info;
2579 rw_extents->pp_init = ktime_get_real();
2580 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
2581 rw_extents->pp_extents[i].pid = 0;
2582 lprocfs_oh_clear(&rw_extents->pp_extents[i].pp_r_hist);
2583 lprocfs_oh_clear(&rw_extents->pp_extents[i].pp_w_hist);
2586 spin_unlock(&sbi->ll_pp_extent_lock);
2591 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
2593 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
2594 struct ll_file_data *file, loff_t pos,
2595 size_t count, int rw)
2598 struct ll_rw_process_info *process;
2599 struct ll_rw_process_info *offset;
2600 int *off_count = &sbi->ll_rw_offset_entry_count;
2601 int *process_count = &sbi->ll_offset_process_count;
2602 struct ll_rw_extents_info *rw_extents;
2604 if (!sbi->ll_rw_stats_on)
2607 spin_lock(&sbi->ll_pp_extent_lock);
2608 rw_extents = sbi->ll_rw_extents_info;
2610 spin_unlock(&sbi->ll_pp_extent_lock);
2614 /* Extent statistics */
2615 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2616 if (rw_extents->pp_extents[i].pid == pid) {
2624 sbi->ll_extent_process_count =
2625 (sbi->ll_extent_process_count + 1) %
2626 LL_PROCESS_HIST_MAX;
2627 cur = sbi->ll_extent_process_count;
2628 rw_extents->pp_extents[cur].pid = pid;
2629 lprocfs_oh_clear(&rw_extents->pp_extents[cur].pp_r_hist);
2630 lprocfs_oh_clear(&rw_extents->pp_extents[cur].pp_w_hist);
2633 for (i = 0; (count >= 1 << (LL_HIST_START + i)) &&
2634 (i < (LL_HIST_MAX - 1)); i++);
2636 rw_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
2637 rw_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
2639 rw_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
2640 rw_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
2642 spin_unlock(&sbi->ll_pp_extent_lock);
2644 spin_lock(&sbi->ll_process_lock);
2645 process = sbi->ll_rw_process_info;
2646 offset = sbi->ll_rw_offset_info;
2647 if (!process || !offset)
2650 /* Offset statistics */
2651 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2652 if (process[i].rw_pid == pid) {
2653 if (process[i].rw_last_file != file) {
2654 process[i].rw_range_start = pos;
2655 process[i].rw_last_file_pos = pos + count;
2656 process[i].rw_smallest_extent = count;
2657 process[i].rw_largest_extent = count;
2658 process[i].rw_offset = 0;
2659 process[i].rw_last_file = file;
2662 if (process[i].rw_last_file_pos != pos) {
2664 (*off_count + 1) % LL_OFFSET_HIST_MAX;
2665 offset[*off_count].rw_op = process[i].rw_op;
2666 offset[*off_count].rw_pid = pid;
2667 offset[*off_count].rw_range_start =
2668 process[i].rw_range_start;
2669 offset[*off_count].rw_range_end =
2670 process[i].rw_last_file_pos;
2671 offset[*off_count].rw_smallest_extent =
2672 process[i].rw_smallest_extent;
2673 offset[*off_count].rw_largest_extent =
2674 process[i].rw_largest_extent;
2675 offset[*off_count].rw_offset =
2676 process[i].rw_offset;
2677 process[i].rw_op = rw;
2678 process[i].rw_range_start = pos;
2679 process[i].rw_smallest_extent = count;
2680 process[i].rw_largest_extent = count;
2681 process[i].rw_offset = pos -
2682 process[i].rw_last_file_pos;
2684 if (process[i].rw_smallest_extent > count)
2685 process[i].rw_smallest_extent = count;
2686 if (process[i].rw_largest_extent < count)
2687 process[i].rw_largest_extent = count;
2688 process[i].rw_last_file_pos = pos + count;
2692 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
2693 process[*process_count].rw_pid = pid;
2694 process[*process_count].rw_op = rw;
2695 process[*process_count].rw_range_start = pos;
2696 process[*process_count].rw_last_file_pos = pos + count;
2697 process[*process_count].rw_smallest_extent = count;
2698 process[*process_count].rw_largest_extent = count;
2699 process[*process_count].rw_offset = 0;
2700 process[*process_count].rw_last_file = file;
2703 spin_unlock(&sbi->ll_process_lock);
2706 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
2708 struct ll_sb_info *sbi = seq->private;
2709 struct ll_rw_process_info *offset;
2710 struct ll_rw_process_info *process;
2713 if (!sbi->ll_rw_stats_on) {
2715 "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2719 spin_lock(&sbi->ll_process_lock);
2720 lprocfs_stats_header(seq, ktime_get_real(), sbi->ll_process_stats_init,
2722 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
2723 "R/W", "PID", "RANGE START", "RANGE END",
2724 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
2726 /* We stored the discontiguous offsets here; print them first */
2727 offset = sbi->ll_rw_offset_info;
2728 for (i = 0; offset && i < LL_OFFSET_HIST_MAX; i++) {
2729 if (offset[i].rw_pid != 0)
2731 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2732 offset[i].rw_op == READ ? 'R' : 'W',
2734 offset[i].rw_range_start,
2735 offset[i].rw_range_end,
2736 (unsigned long)offset[i].rw_smallest_extent,
2737 (unsigned long)offset[i].rw_largest_extent,
2738 offset[i].rw_offset);
2741 /* Then print the current offsets for each process */
2742 process = sbi->ll_rw_process_info;
2743 for (i = 0; process && i < LL_PROCESS_HIST_MAX; i++) {
2744 if (process[i].rw_pid != 0)
2746 "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2747 process[i].rw_op == READ ? 'R' : 'W',
2749 process[i].rw_range_start,
2750 process[i].rw_last_file_pos,
2751 (unsigned long)process[i].rw_smallest_extent,
2752 (unsigned long)process[i].rw_largest_extent,
2753 process[i].rw_offset);
2755 spin_unlock(&sbi->ll_process_lock);
2760 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
2761 const char __user *buf,
2762 size_t len, loff_t *off)
2764 struct seq_file *seq = file->private_data;
2765 struct ll_sb_info *sbi = seq->private;
2771 value = ll_stats_pid_write(buf, len);
2774 sbi->ll_rw_stats_on = 0;
2776 if (!sbi->ll_rw_process_info || !sbi->ll_rw_offset_info) {
2777 int rc = alloc_rw_stats_info(sbi);
2782 sbi->ll_rw_stats_on = 1;
2785 spin_lock(&sbi->ll_process_lock);
2786 sbi->ll_offset_process_count = 0;
2787 sbi->ll_rw_offset_entry_count = 0;
2788 sbi->ll_process_stats_init = ktime_get_real();
2789 if (sbi->ll_rw_process_info)
2790 memset(sbi->ll_rw_process_info, 0,
2791 sizeof(struct ll_rw_process_info) * LL_PROCESS_HIST_MAX);
2792 if (sbi->ll_rw_offset_info)
2793 memset(sbi->ll_rw_offset_info, 0,
2794 sizeof(struct ll_rw_process_info) * LL_OFFSET_HIST_MAX);
2795 spin_unlock(&sbi->ll_process_lock);
2800 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);