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 #ifdef HAVE_UIDGID_HEADER
37 # include <linux/uidgid.h>
39 #include <uapi/linux/lustre/lustre_param.h>
40 #include <lprocfs_status.h>
41 #include <obd_support.h>
43 #include "llite_internal.h"
44 #include "vvp_internal.h"
46 static struct kobject *llite_kobj;
47 static struct dentry *llite_root;
49 int llite_tunables_register(void)
53 llite_kobj = class_setup_tunables("llite");
54 if (IS_ERR(llite_kobj))
55 return PTR_ERR(llite_kobj);
57 llite_root = debugfs_create_dir("llite", debugfs_lustre_root);
58 if (IS_ERR_OR_NULL(llite_root)) {
59 rc = llite_root ? PTR_ERR(llite_root) : -ENOMEM;
61 kobject_put(llite_kobj);
68 void llite_tunables_unregister(void)
71 kobject_put(llite_kobj);
75 if (!IS_ERR_OR_NULL(llite_root)) {
76 debugfs_remove(llite_root);
81 /* <debugfs>/lustre/llite mount point registration */
82 static const struct file_operations ll_rw_extents_stats_fops;
83 static const struct file_operations ll_rw_extents_stats_pp_fops;
84 static const struct file_operations ll_rw_offset_stats_fops;
87 * ll_stats_pid_write() - Determine if stats collection should be enabled
88 * @buf: Buffer containing the data written
89 * @len: Number of bytes in the buffer
91 * Several proc files begin collecting stats when a value is written, and stop
92 * collecting when either '0' or 'disable' is written. This function checks the
93 * written value to see if collection should be enabled or disabled.
95 * Return: If '0' or 'disable' is provided, 0 is returned. If the text
96 * equivalent of a number is written, that number is returned. Otherwise,
97 * 1 is returned. Non-zero return values indicate collection should be enabled.
99 static s64 ll_stats_pid_write(const char __user *buf, size_t len)
101 unsigned long long value = 1;
105 rc = kstrtoull_from_user(buf, len, 0, &value);
106 if (rc < 0 && len < sizeof(kernbuf)) {
107 if (copy_from_user(kernbuf, buf, len))
111 if (kernbuf[len - 1] == '\n')
112 kernbuf[len - 1] = 0;
114 if (strncasecmp(kernbuf, "disable", 7) == 0)
121 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
124 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
126 struct obd_statfs osfs;
129 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
133 return sprintf(buf, "%u\n", osfs.os_bsize);
135 LUSTRE_RO_ATTR(blocksize);
137 static ssize_t stat_blocksize_show(struct kobject *kobj, struct attribute *attr,
140 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
143 return sprintf(buf, "%u\n", sbi->ll_stat_blksize);
146 static ssize_t stat_blocksize_store(struct kobject *kobj,
147 struct attribute *attr,
151 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
156 rc = kstrtouint(buffer, 10, &val);
160 if (val != 0 && (val < PAGE_SIZE || (val & (val - 1))) != 0)
163 sbi->ll_stat_blksize = val;
167 LUSTRE_RW_ATTR(stat_blocksize);
169 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
172 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
174 struct obd_statfs osfs;
179 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
183 blk_size = osfs.os_bsize >> 10;
184 result = osfs.os_blocks;
186 while (blk_size >>= 1)
189 return sprintf(buf, "%llu\n", result);
191 LUSTRE_RO_ATTR(kbytestotal);
193 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
196 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
198 struct obd_statfs osfs;
203 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
207 blk_size = osfs.os_bsize >> 10;
208 result = osfs.os_bfree;
210 while (blk_size >>= 1)
213 return sprintf(buf, "%llu\n", result);
215 LUSTRE_RO_ATTR(kbytesfree);
217 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
220 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
222 struct obd_statfs osfs;
227 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
231 blk_size = osfs.os_bsize >> 10;
232 result = osfs.os_bavail;
234 while (blk_size >>= 1)
237 return sprintf(buf, "%llu\n", result);
239 LUSTRE_RO_ATTR(kbytesavail);
241 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
244 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
246 struct obd_statfs osfs;
249 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
253 return sprintf(buf, "%llu\n", osfs.os_files);
255 LUSTRE_RO_ATTR(filestotal);
257 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
260 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
262 struct obd_statfs osfs;
265 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
269 return sprintf(buf, "%llu\n", osfs.os_ffree);
271 LUSTRE_RO_ATTR(filesfree);
273 static ssize_t client_type_show(struct kobject *kobj, struct attribute *attr,
276 return sprintf(buf, "local client\n");
278 LUSTRE_RO_ATTR(client_type);
280 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
283 return sprintf(buf, "lustre\n");
285 LUSTRE_RO_ATTR(fstype);
287 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
290 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
293 return sprintf(buf, "%s\n", sbi->ll_sb_uuid.uuid);
295 LUSTRE_RO_ATTR(uuid);
297 static int ll_site_stats_seq_show(struct seq_file *m, void *v)
299 struct super_block *sb = m->private;
302 * See description of statistical counters in struct cl_site, and
305 return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site), m);
308 LDEBUGFS_SEQ_FOPS_RO(ll_site_stats);
310 static int ll_max_readahead_mb_seq_show(struct seq_file *m, void *v)
312 struct super_block *sb = m->private;
313 struct ll_sb_info *sbi = ll_s2sbi(sb);
317 spin_lock(&sbi->ll_lock);
318 pages_number = sbi->ll_ra_info.ra_max_pages;
319 spin_unlock(&sbi->ll_lock);
321 mult = 1 << (20 - PAGE_SHIFT);
322 return lprocfs_seq_read_frac_helper(m, pages_number, mult);
326 ll_max_readahead_mb_seq_write(struct file *file, const char __user *buffer,
327 size_t count, loff_t *off)
329 struct seq_file *m = file->private_data;
330 struct super_block *sb = m->private;
331 struct ll_sb_info *sbi = ll_s2sbi(sb);
335 rc = lprocfs_str_with_units_to_s64(buffer, count, &pages_number, 'M');
339 pages_number >>= PAGE_SHIFT;
341 if (pages_number < 0 || pages_number > totalram_pages / 2) {
343 CERROR("%s: can't set max_readahead_mb=%lu > %luMB\n",
344 ll_get_fsname(sb, NULL, 0),
345 (unsigned long)pages_number >> (20 - PAGE_SHIFT),
346 totalram_pages >> (20 - PAGE_SHIFT + 1));
350 spin_lock(&sbi->ll_lock);
351 sbi->ll_ra_info.ra_max_pages = pages_number;
352 spin_unlock(&sbi->ll_lock);
357 LDEBUGFS_SEQ_FOPS(ll_max_readahead_mb);
359 static int ll_max_readahead_per_file_mb_seq_show(struct seq_file *m, void *v)
361 struct super_block *sb = m->private;
362 struct ll_sb_info *sbi = ll_s2sbi(sb);
366 spin_lock(&sbi->ll_lock);
367 pages_number = sbi->ll_ra_info.ra_max_pages_per_file;
368 spin_unlock(&sbi->ll_lock);
370 mult = 1 << (20 - PAGE_SHIFT);
371 return lprocfs_seq_read_frac_helper(m, pages_number, mult);
375 ll_max_readahead_per_file_mb_seq_write(struct file *file,
376 const char __user *buffer,
377 size_t count, loff_t *off)
379 struct seq_file *m = file->private_data;
380 struct super_block *sb = m->private;
381 struct ll_sb_info *sbi = ll_s2sbi(sb);
385 rc = lprocfs_str_with_units_to_s64(buffer, count, &pages_number, 'M');
389 pages_number >>= PAGE_SHIFT;
391 if (pages_number < 0 || pages_number > sbi->ll_ra_info.ra_max_pages) {
392 CERROR("%s: can't set max_readahead_per_file_mb=%lu > "
393 "max_read_ahead_mb=%lu\n", ll_get_fsname(sb, NULL, 0),
394 (unsigned long)pages_number >> (20 - PAGE_SHIFT),
395 sbi->ll_ra_info.ra_max_pages >> (20 - PAGE_SHIFT));
399 spin_lock(&sbi->ll_lock);
400 sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
401 spin_unlock(&sbi->ll_lock);
406 LDEBUGFS_SEQ_FOPS(ll_max_readahead_per_file_mb);
408 static int ll_max_read_ahead_whole_mb_seq_show(struct seq_file *m, void *v)
410 struct super_block *sb = m->private;
411 struct ll_sb_info *sbi = ll_s2sbi(sb);
415 spin_lock(&sbi->ll_lock);
416 pages_number = sbi->ll_ra_info.ra_max_read_ahead_whole_pages;
417 spin_unlock(&sbi->ll_lock);
419 mult = 1 << (20 - PAGE_SHIFT);
420 return lprocfs_seq_read_frac_helper(m, pages_number, mult);
424 ll_max_read_ahead_whole_mb_seq_write(struct file *file,
425 const char __user *buffer,
426 size_t count, loff_t *off)
428 struct seq_file *m = file->private_data;
429 struct super_block *sb = m->private;
430 struct ll_sb_info *sbi = ll_s2sbi(sb);
434 rc = lprocfs_str_with_units_to_s64(buffer, count, &pages_number, 'M');
438 pages_number >>= PAGE_SHIFT;
440 /* Cap this at the current max readahead window size, the readahead
441 * algorithm does this anyway so it's pointless to set it larger. */
442 if (pages_number < 0 ||
443 pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
444 int pages_shift = 20 - PAGE_SHIFT;
445 CERROR("%s: can't set max_read_ahead_whole_mb=%lu > "
446 "max_read_ahead_per_file_mb=%lu\n",
447 ll_get_fsname(sb, NULL, 0),
448 (unsigned long)pages_number >> pages_shift,
449 sbi->ll_ra_info.ra_max_pages_per_file >> pages_shift);
453 spin_lock(&sbi->ll_lock);
454 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
455 spin_unlock(&sbi->ll_lock);
460 LDEBUGFS_SEQ_FOPS(ll_max_read_ahead_whole_mb);
462 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
464 struct super_block *sb = m->private;
465 struct ll_sb_info *sbi = ll_s2sbi(sb);
466 struct cl_client_cache *cache = sbi->ll_cache;
467 int shift = 20 - PAGE_SHIFT;
471 max_cached_mb = cache->ccc_lru_max >> shift;
472 unused_mb = atomic_long_read(&cache->ccc_lru_left) >> shift;
473 seq_printf(m, "users: %d\n"
474 "max_cached_mb: %ld\n"
477 "reclaim_count: %u\n",
478 atomic_read(&cache->ccc_users),
480 max_cached_mb - unused_mb,
482 cache->ccc_lru_shrinkers);
486 static ssize_t ll_max_cached_mb_seq_write(struct file *file,
487 const char __user *buffer,
488 size_t count, loff_t *off)
490 struct seq_file *m = file->private_data;
491 struct super_block *sb = m->private;
492 struct ll_sb_info *sbi = ll_s2sbi(sb);
493 struct cl_client_cache *cache = sbi->ll_cache;
503 if (count >= sizeof(kernbuf))
506 if (copy_from_user(kernbuf, buffer, count))
510 buffer += lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count) -
512 rc = lprocfs_str_with_units_to_s64(buffer, count, &pages_number, 'M');
516 pages_number >>= PAGE_SHIFT;
518 if (pages_number < 0 || pages_number > totalram_pages) {
519 CERROR("%s: can't set max cache more than %lu MB\n",
520 ll_get_fsname(sb, NULL, 0),
521 totalram_pages >> (20 - PAGE_SHIFT));
524 /* Allow enough cache so clients can make well-formed RPCs */
525 pages_number = max_t(long, pages_number, PTLRPC_MAX_BRW_PAGES);
527 spin_lock(&sbi->ll_lock);
528 diff = pages_number - cache->ccc_lru_max;
529 spin_unlock(&sbi->ll_lock);
531 /* easy - add more LRU slots. */
533 atomic_long_add(diff, &cache->ccc_lru_left);
537 env = cl_env_get(&refcheck);
539 RETURN(PTR_ERR(env));
545 /* reduce LRU budget from free slots. */
549 ov = atomic_long_read(&cache->ccc_lru_left);
553 nv = ov > diff ? ov - diff : 0;
554 rc = atomic_long_cmpxchg(&cache->ccc_lru_left, ov, nv);
555 if (likely(ov == rc)) {
565 if (sbi->ll_dt_exp == NULL) { /* being initialized */
570 /* 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 spin_lock(&sbi->ll_lock);
584 cache->ccc_lru_max = pages_number;
585 spin_unlock(&sbi->ll_lock);
588 atomic_long_add(nrpages, &cache->ccc_lru_left);
593 LDEBUGFS_SEQ_FOPS(ll_max_cached_mb);
595 static ssize_t checksums_show(struct kobject *kobj, struct attribute *attr,
598 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
601 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
604 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
605 const char *buffer, size_t count)
607 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
617 rc = kstrtobool(buffer, &val);
621 sbi->ll_flags |= LL_SBI_CHECKSUM;
623 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
626 rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
627 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
629 CWARN("Failed to set OSC checksum flags: %d\n", rc);
633 LUSTRE_RW_ATTR(checksums);
635 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
637 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
638 enum stats_track_type type)
640 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
643 if (sbi->ll_stats_track_type == type)
644 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
645 else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
646 return sprintf(buf, "0 (all)\n");
648 return sprintf(buf, "untracked\n");
651 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
652 size_t count, enum stats_track_type type)
654 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
659 rc = kstrtoul(buffer, 10, &pid);
663 sbi->ll_stats_track_id = pid;
665 sbi->ll_stats_track_type = STATS_TRACK_ALL;
667 sbi->ll_stats_track_type = type;
668 lprocfs_clear_stats(sbi->ll_stats);
672 static ssize_t stats_track_pid_show(struct kobject *kobj,
673 struct attribute *attr,
676 return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
679 static ssize_t stats_track_pid_store(struct kobject *kobj,
680 struct attribute *attr,
684 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
686 LUSTRE_RW_ATTR(stats_track_pid);
688 static ssize_t stats_track_ppid_show(struct kobject *kobj,
689 struct attribute *attr,
692 return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
695 static ssize_t stats_track_ppid_store(struct kobject *kobj,
696 struct attribute *attr,
700 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
702 LUSTRE_RW_ATTR(stats_track_ppid);
704 static ssize_t stats_track_gid_show(struct kobject *kobj,
705 struct attribute *attr,
708 return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
711 static ssize_t stats_track_gid_store(struct kobject *kobj,
712 struct attribute *attr,
716 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
718 LUSTRE_RW_ATTR(stats_track_gid);
720 static ssize_t statahead_running_max_show(struct kobject *kobj,
721 struct attribute *attr,
724 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
727 return snprintf(buf, 16, "%u\n", sbi->ll_sa_running_max);
730 static ssize_t statahead_running_max_store(struct kobject *kobj,
731 struct attribute *attr,
735 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
740 rc = kstrtoul(buffer, 0, &val);
744 if (val <= LL_SA_RUNNING_MAX) {
745 sbi->ll_sa_running_max = val;
749 CERROR("Bad statahead_running_max value %lu. Valid values "
750 "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
754 LUSTRE_RW_ATTR(statahead_running_max);
756 static ssize_t statahead_max_show(struct kobject *kobj,
757 struct attribute *attr,
760 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
763 return sprintf(buf, "%u\n", sbi->ll_sa_max);
766 static ssize_t statahead_max_store(struct kobject *kobj,
767 struct attribute *attr,
771 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
776 rc = kstrtoul(buffer, 0, &val);
780 if (val <= LL_SA_RPC_MAX)
781 sbi->ll_sa_max = val;
783 CERROR("Bad statahead_max value %lu. Valid values are in the range [0, %d]\n",
788 LUSTRE_RW_ATTR(statahead_max);
790 static ssize_t statahead_agl_show(struct kobject *kobj,
791 struct attribute *attr,
794 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
797 return sprintf(buf, "%u\n", sbi->ll_flags & LL_SBI_AGL_ENABLED ? 1 : 0);
800 static ssize_t statahead_agl_store(struct kobject *kobj,
801 struct attribute *attr,
805 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
810 rc = kstrtobool(buffer, &val);
815 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
817 sbi->ll_flags &= ~LL_SBI_AGL_ENABLED;
821 LUSTRE_RW_ATTR(statahead_agl);
823 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
825 struct super_block *sb = m->private;
826 struct ll_sb_info *sbi = ll_s2sbi(sb);
828 seq_printf(m, "statahead total: %u\n"
829 "statahead wrong: %u\n"
831 atomic_read(&sbi->ll_sa_total),
832 atomic_read(&sbi->ll_sa_wrong),
833 atomic_read(&sbi->ll_agl_total));
837 LDEBUGFS_SEQ_FOPS_RO(ll_statahead_stats);
839 static ssize_t lazystatfs_show(struct kobject *kobj,
840 struct attribute *attr,
843 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
846 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
849 static ssize_t lazystatfs_store(struct kobject *kobj,
850 struct attribute *attr,
854 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
859 rc = kstrtobool(buffer, &val);
864 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
866 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
870 LUSTRE_RW_ATTR(lazystatfs);
872 static ssize_t max_easize_show(struct kobject *kobj,
873 struct attribute *attr,
876 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
881 rc = ll_get_max_mdsize(sbi, &ealen);
885 return sprintf(buf, "%u\n", ealen);
887 LUSTRE_RO_ATTR(max_easize);
890 * Get default_easize.
892 * \see client_obd::cl_default_mds_easize
894 * \param[in] m seq_file handle
895 * \param[in] v unused for single entry
897 * \retval 0 on success
898 * \retval negative negated errno on failure
900 static ssize_t default_easize_show(struct kobject *kobj,
901 struct attribute *attr,
904 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
909 rc = ll_get_default_mdsize(sbi, &ealen);
913 return sprintf(buf, "%u\n", ealen);
917 * Set default_easize.
919 * Range checking on the passed value is handled by
920 * ll_set_default_mdsize().
922 * \see client_obd::cl_default_mds_easize
924 * \param[in] file proc file
925 * \param[in] buffer string passed from user space
926 * \param[in] count \a buffer length
927 * \param[in] off unused for single entry
929 * \retval positive \a count on success
930 * \retval negative negated errno on failure
932 static ssize_t default_easize_store(struct kobject *kobj,
933 struct attribute *attr,
937 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
945 rc = kstrtouint(buffer, 10, &val);
949 rc = ll_set_default_mdsize(sbi, val);
955 LUSTRE_RW_ATTR(default_easize);
957 static int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
959 const char *str[] = LL_SBI_FLAGS;
960 struct super_block *sb = m->private;
961 int flags = ll_s2sbi(sb)->ll_flags;
965 if (ARRAY_SIZE(str) <= i) {
966 CERROR("%s: Revise array LL_SBI_FLAGS to match sbi "
967 "flags please.\n", ll_get_fsname(sb, NULL, 0));
972 seq_printf(m, "%s ", str[i]);
976 seq_printf(m, "\b\n");
980 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
982 static ssize_t xattr_cache_show(struct kobject *kobj,
983 struct attribute *attr,
986 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
989 return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
992 static ssize_t xattr_cache_store(struct kobject *kobj,
993 struct attribute *attr,
997 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1002 rc = kstrtobool(buffer, &val);
1006 if (val && !(sbi->ll_flags & LL_SBI_XATTR_CACHE))
1009 sbi->ll_xattr_cache_enabled = val;
1010 sbi->ll_xattr_cache_set = 1;
1014 LUSTRE_RW_ATTR(xattr_cache);
1016 static ssize_t tiny_write_show(struct kobject *kobj,
1017 struct attribute *attr,
1020 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1023 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_TINY_WRITE));
1026 static ssize_t tiny_write_store(struct kobject *kobj,
1027 struct attribute *attr,
1031 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1036 rc = kstrtobool(buffer, &val);
1040 spin_lock(&sbi->ll_lock);
1042 sbi->ll_flags |= LL_SBI_TINY_WRITE;
1044 sbi->ll_flags &= ~LL_SBI_TINY_WRITE;
1045 spin_unlock(&sbi->ll_lock);
1049 LUSTRE_RW_ATTR(tiny_write);
1051 static ssize_t fast_read_show(struct kobject *kobj,
1052 struct attribute *attr,
1055 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1058 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_FAST_READ));
1061 static ssize_t fast_read_store(struct kobject *kobj,
1062 struct attribute *attr,
1066 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1071 rc = kstrtobool(buffer, &val);
1075 spin_lock(&sbi->ll_lock);
1077 sbi->ll_flags |= LL_SBI_FAST_READ;
1079 sbi->ll_flags &= ~LL_SBI_FAST_READ;
1080 spin_unlock(&sbi->ll_lock);
1084 LUSTRE_RW_ATTR(fast_read);
1086 static ssize_t pio_show(struct kobject *kobj,
1087 struct attribute *attr,
1090 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1093 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_PIO));
1096 static ssize_t pio_store(struct kobject *kobj,
1097 struct attribute *attr,
1101 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1106 rc = kstrtobool(buffer, &val);
1110 spin_lock(&sbi->ll_lock);
1112 sbi->ll_flags |= LL_SBI_PIO;
1114 sbi->ll_flags &= ~LL_SBI_PIO;
1115 spin_unlock(&sbi->ll_lock);
1119 LUSTRE_RW_ATTR(pio);
1121 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1123 struct super_block *sb = m->private;
1124 struct ll_sb_info *sbi = ll_s2sbi(sb);
1125 struct cl_client_cache *cache = sbi->ll_cache;
1129 pages = atomic_long_read(&cache->ccc_unstable_nr);
1130 mb = (pages * PAGE_SIZE) >> 20;
1132 seq_printf(m, "unstable_check: %8d\n"
1133 "unstable_pages: %12ld\n"
1134 "unstable_mb: %8d\n",
1135 cache->ccc_unstable_check, pages, mb);
1139 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1140 const char __user *buffer,
1141 size_t count, loff_t *unused)
1143 struct seq_file *seq = file->private_data;
1144 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1151 if (count >= sizeof(kernbuf))
1154 if (copy_from_user(kernbuf, buffer, count))
1158 buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1160 rc = kstrtobool_from_user(buffer, count, &val);
1164 /* borrow lru lock to set the value */
1165 spin_lock(&sbi->ll_cache->ccc_lru_lock);
1166 sbi->ll_cache->ccc_unstable_check = val;
1167 spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1172 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1174 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1176 struct super_block *sb = m->private;
1177 struct ll_sb_info *sbi = ll_s2sbi(sb);
1178 struct root_squash_info *squash = &sbi->ll_squash;
1180 seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1184 static ssize_t ll_root_squash_seq_write(struct file *file,
1185 const char __user *buffer,
1186 size_t count, loff_t *off)
1188 struct seq_file *m = file->private_data;
1189 struct super_block *sb = m->private;
1190 struct ll_sb_info *sbi = ll_s2sbi(sb);
1191 struct root_squash_info *squash = &sbi->ll_squash;
1193 return lprocfs_wr_root_squash(buffer, count, squash,
1194 ll_get_fsname(sb, NULL, 0));
1197 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1199 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1201 struct super_block *sb = m->private;
1202 struct ll_sb_info *sbi = ll_s2sbi(sb);
1203 struct root_squash_info *squash = &sbi->ll_squash;
1206 down_read(&squash->rsi_sem);
1207 if (!list_empty(&squash->rsi_nosquash_nids)) {
1208 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1209 &squash->rsi_nosquash_nids);
1213 seq_puts(m, "NONE\n");
1215 up_read(&squash->rsi_sem);
1220 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1221 const char __user *buffer,
1222 size_t count, loff_t *off)
1224 struct seq_file *m = file->private_data;
1225 struct super_block *sb = m->private;
1226 struct ll_sb_info *sbi = ll_s2sbi(sb);
1227 struct root_squash_info *squash = &sbi->ll_squash;
1230 rc = lprocfs_wr_nosquash_nids(buffer, count, squash,
1231 ll_get_fsname(sb, NULL, 0));
1235 ll_compute_rootsquash_state(sbi);
1240 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1242 struct lprocfs_vars lprocfs_llite_obd_vars[] = {
1244 .fops = &ll_site_stats_fops },
1245 { .name = "max_read_ahead_mb",
1246 .fops = &ll_max_readahead_mb_fops },
1247 { .name = "max_read_ahead_per_file_mb",
1248 .fops = &ll_max_readahead_per_file_mb_fops },
1249 { .name = "max_read_ahead_whole_mb",
1250 .fops = &ll_max_read_ahead_whole_mb_fops },
1251 { .name = "max_cached_mb",
1252 .fops = &ll_max_cached_mb_fops },
1253 { .name = "statahead_stats",
1254 .fops = &ll_statahead_stats_fops },
1255 { .name = "unstable_stats",
1256 .fops = &ll_unstable_stats_fops },
1257 { .name = "sbi_flags",
1258 .fops = &ll_sbi_flags_fops },
1259 { .name = "root_squash",
1260 .fops = &ll_root_squash_fops },
1261 { .name = "nosquash_nids",
1262 .fops = &ll_nosquash_nids_fops },
1266 #define MAX_STRING_SIZE 128
1268 static struct attribute *llite_attrs[] = {
1269 &lustre_attr_blocksize.attr,
1270 &lustre_attr_stat_blocksize.attr,
1271 &lustre_attr_kbytestotal.attr,
1272 &lustre_attr_kbytesfree.attr,
1273 &lustre_attr_kbytesavail.attr,
1274 &lustre_attr_filestotal.attr,
1275 &lustre_attr_filesfree.attr,
1276 &lustre_attr_client_type.attr,
1277 &lustre_attr_fstype.attr,
1278 &lustre_attr_uuid.attr,
1279 &lustre_attr_checksums.attr,
1280 &lustre_attr_checksum_pages.attr,
1281 &lustre_attr_stats_track_pid.attr,
1282 &lustre_attr_stats_track_ppid.attr,
1283 &lustre_attr_stats_track_gid.attr,
1284 &lustre_attr_statahead_running_max.attr,
1285 &lustre_attr_statahead_max.attr,
1286 &lustre_attr_statahead_agl.attr,
1287 &lustre_attr_lazystatfs.attr,
1288 &lustre_attr_max_easize.attr,
1289 &lustre_attr_default_easize.attr,
1290 &lustre_attr_xattr_cache.attr,
1291 &lustre_attr_fast_read.attr,
1292 &lustre_attr_pio.attr,
1293 &lustre_attr_tiny_write.attr,
1297 static void llite_kobj_release(struct kobject *kobj)
1299 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1301 complete(&sbi->ll_kobj_unregister);
1304 static struct kobj_type llite_ktype = {
1305 .default_attrs = llite_attrs,
1306 .sysfs_ops = &lustre_sysfs_ops,
1307 .release = llite_kobj_release,
1310 static const struct llite_file_opcode {
1314 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1315 /* file operation */
1316 { LPROC_LL_DIRTY_HITS, LPROCFS_TYPE_REGS, "dirty_pages_hits" },
1317 { LPROC_LL_DIRTY_MISSES, LPROCFS_TYPE_REGS, "dirty_pages_misses" },
1318 { LPROC_LL_READ_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
1320 { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
1322 { LPROC_LL_BRW_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
1324 { LPROC_LL_BRW_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
1326 { LPROC_LL_IOCTL, LPROCFS_TYPE_REGS, "ioctl" },
1327 { LPROC_LL_OPEN, LPROCFS_TYPE_REGS, "open" },
1328 { LPROC_LL_RELEASE, LPROCFS_TYPE_REGS, "close" },
1329 { LPROC_LL_MAP, LPROCFS_TYPE_REGS, "mmap" },
1330 { LPROC_LL_FAULT, LPROCFS_TYPE_REGS, "page_fault" },
1331 { LPROC_LL_MKWRITE, LPROCFS_TYPE_REGS, "page_mkwrite" },
1332 { LPROC_LL_LLSEEK, LPROCFS_TYPE_REGS, "seek" },
1333 { LPROC_LL_FSYNC, LPROCFS_TYPE_REGS, "fsync" },
1334 { LPROC_LL_READDIR, LPROCFS_TYPE_REGS, "readdir" },
1335 /* inode operation */
1336 { LPROC_LL_SETATTR, LPROCFS_TYPE_REGS, "setattr" },
1337 { LPROC_LL_TRUNC, LPROCFS_TYPE_REGS, "truncate" },
1338 { LPROC_LL_FLOCK, LPROCFS_TYPE_REGS, "flock" },
1339 { LPROC_LL_GETATTR, LPROCFS_TYPE_REGS, "getattr" },
1340 /* dir inode operation */
1341 { LPROC_LL_CREATE, LPROCFS_TYPE_REGS, "create" },
1342 { LPROC_LL_LINK, LPROCFS_TYPE_REGS, "link" },
1343 { LPROC_LL_UNLINK, LPROCFS_TYPE_REGS, "unlink" },
1344 { LPROC_LL_SYMLINK, LPROCFS_TYPE_REGS, "symlink" },
1345 { LPROC_LL_MKDIR, LPROCFS_TYPE_REGS, "mkdir" },
1346 { LPROC_LL_RMDIR, LPROCFS_TYPE_REGS, "rmdir" },
1347 { LPROC_LL_MKNOD, LPROCFS_TYPE_REGS, "mknod" },
1348 { LPROC_LL_RENAME, LPROCFS_TYPE_REGS, "rename" },
1349 /* special inode operation */
1350 { LPROC_LL_STAFS, LPROCFS_TYPE_REGS, "statfs" },
1351 { LPROC_LL_ALLOC_INODE, LPROCFS_TYPE_REGS, "alloc_inode" },
1352 { LPROC_LL_SETXATTR, LPROCFS_TYPE_REGS, "setxattr" },
1353 { LPROC_LL_GETXATTR, LPROCFS_TYPE_REGS, "getxattr" },
1354 { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REGS, "getxattr_hits" },
1355 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_REGS, "listxattr" },
1356 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_REGS, "removexattr" },
1357 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_REGS, "inode_permission" },
1360 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
1364 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1365 lprocfs_counter_add(sbi->ll_stats, op, count);
1366 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1367 sbi->ll_stats_track_id == current->pid)
1368 lprocfs_counter_add(sbi->ll_stats, op, count);
1369 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1370 sbi->ll_stats_track_id == current->parent->pid)
1371 lprocfs_counter_add(sbi->ll_stats, op, count);
1372 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1373 sbi->ll_stats_track_id ==
1374 from_kgid(&init_user_ns, current_gid()))
1375 lprocfs_counter_add(sbi->ll_stats, op, count);
1377 EXPORT_SYMBOL(ll_stats_ops_tally);
1379 static const char *ra_stat_string[] = {
1380 [RA_STAT_HIT] = "hits",
1381 [RA_STAT_MISS] = "misses",
1382 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1383 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1384 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1385 [RA_STAT_FAILED_MATCH] = "failed lock match",
1386 [RA_STAT_DISCARDED] = "read but discarded",
1387 [RA_STAT_ZERO_LEN] = "zero length file",
1388 [RA_STAT_ZERO_WINDOW] = "zero size window",
1389 [RA_STAT_EOF] = "read-ahead to EOF",
1390 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1391 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1392 [RA_STAT_FAILED_REACH_END] = "failed to reach end"
1395 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1397 struct lustre_sb_info *lsi = s2lsi(sb);
1398 struct ll_sb_info *sbi = ll_s2sbi(sb);
1404 if (IS_ERR_OR_NULL(llite_root))
1407 sbi->ll_debugfs_entry = ldebugfs_register(name, llite_root,
1408 lprocfs_llite_obd_vars, sb);
1409 if (IS_ERR_OR_NULL(sbi->ll_debugfs_entry)) {
1410 err = sbi->ll_debugfs_entry ? PTR_ERR(sbi->ll_debugfs_entry) :
1412 sbi->ll_debugfs_entry = NULL;
1416 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "dump_page_cache",0444,
1417 &vvp_dump_pgcache_file_ops, sbi);
1419 CWARN("Error adding the dump_page_cache file\n");
1421 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "extents_stats", 0644,
1422 &ll_rw_extents_stats_fops, sbi);
1424 CWARN("Error adding the extent_stats file\n");
1426 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry,
1427 "extents_stats_per_process", 0644,
1428 &ll_rw_extents_stats_pp_fops, sbi);
1430 CWARN("Error adding the extents_stats_per_process file\n");
1432 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "offset_stats", 0644,
1433 &ll_rw_offset_stats_fops, sbi);
1435 CWARN("Error adding the offset_stats file\n");
1437 /* File operations stats */
1438 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1439 LPROCFS_STATS_FLAG_NONE);
1440 if (sbi->ll_stats == NULL)
1441 GOTO(out_debugfs, err = -ENOMEM);
1443 /* do counter init */
1444 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1445 u32 type = llite_opcode_table[id].type;
1448 if (type & LPROCFS_TYPE_REGS)
1450 else if (type & LPROCFS_TYPE_BYTES)
1452 else if (type & LPROCFS_TYPE_PAGES)
1454 lprocfs_counter_init(sbi->ll_stats,
1455 llite_opcode_table[id].opcode,
1456 (type & LPROCFS_CNTR_AVGMINMAX),
1457 llite_opcode_table[id].opname, ptr);
1460 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "stats",
1463 GOTO(out_stats, err);
1465 sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1466 LPROCFS_STATS_FLAG_NONE);
1467 if (sbi->ll_ra_stats == NULL)
1468 GOTO(out_stats, err = -ENOMEM);
1470 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1471 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1472 ra_stat_string[id], "pages");
1474 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "read_ahead_stats",
1477 GOTO(out_ra_stats, err);
1480 /* Yes we also register sysfs mount kset here as well */
1481 sbi->ll_kset.kobj.parent = llite_kobj;
1482 sbi->ll_kset.kobj.ktype = &llite_ktype;
1483 init_completion(&sbi->ll_kobj_unregister);
1484 err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1486 GOTO(out_ra_stats, err);
1488 err = kset_register(&sbi->ll_kset);
1490 GOTO(out_ra_stats, err);
1492 lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1496 lprocfs_free_stats(&sbi->ll_ra_stats);
1498 lprocfs_free_stats(&sbi->ll_stats);
1500 ldebugfs_remove(&sbi->ll_debugfs_entry);
1505 void ll_debugfs_unregister_super(struct super_block *sb)
1507 struct lustre_sb_info *lsi = s2lsi(sb);
1508 struct ll_sb_info *sbi = ll_s2sbi(sb);
1510 if (!IS_ERR_OR_NULL(sbi->ll_debugfs_entry))
1511 ldebugfs_remove(&sbi->ll_debugfs_entry);
1514 sysfs_remove_link(&sbi->ll_kset.kobj,
1515 sbi->ll_dt_obd->obd_type->typ_name);
1518 sysfs_remove_link(&sbi->ll_kset.kobj,
1519 sbi->ll_md_obd->obd_type->typ_name);
1521 kobject_put(lsi->lsi_kobj);
1523 kset_unregister(&sbi->ll_kset);
1524 wait_for_completion(&sbi->ll_kobj_unregister);
1526 lprocfs_free_stats(&sbi->ll_ra_stats);
1527 lprocfs_free_stats(&sbi->ll_stats);
1529 #undef MAX_STRING_SIZE
1531 #define pct(a,b) (b ? a * 100 / b : 0)
1533 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1534 struct seq_file *seq, int which)
1536 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1537 unsigned long start, end, r, w;
1538 char *unitp = "KMGTPEZY";
1540 struct per_process_info *pp_info = &io_extents->pp_extents[which];
1546 for(i = 0; i < LL_HIST_MAX; i++) {
1547 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1548 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1551 for(i = 0; i < LL_HIST_MAX; i++) {
1552 r = pp_info->pp_r_hist.oh_buckets[i];
1553 w = pp_info->pp_w_hist.oh_buckets[i];
1556 end = BIT(i + LL_HIST_START - units);
1557 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4lu %4lu | "
1558 "%14lu %4lu %4lu\n", start, *unitp, end, *unitp,
1559 (i == LL_HIST_MAX - 1) ? '+' : ' ',
1560 r, pct(r, read_tot), pct(read_cum, read_tot),
1561 w, pct(w, write_tot), pct(write_cum, write_tot));
1563 if (start == BIT(10)) {
1568 if (read_cum == read_tot && write_cum == write_tot)
1573 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1575 struct timespec64 now;
1576 struct ll_sb_info *sbi = seq->private;
1577 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1580 ktime_get_real_ts64(&now);
1582 if (!sbi->ll_rw_stats_on) {
1583 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1586 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1587 (s64)now.tv_sec, now.tv_nsec);
1588 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1589 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1590 "extents", "calls", "%", "cum%",
1591 "calls", "%", "cum%");
1592 spin_lock(&sbi->ll_pp_extent_lock);
1593 for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1594 if (io_extents->pp_extents[k].pid != 0) {
1595 seq_printf(seq, "\nPID: %d\n",
1596 io_extents->pp_extents[k].pid);
1597 ll_display_extents_info(io_extents, seq, k);
1600 spin_unlock(&sbi->ll_pp_extent_lock);
1604 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1605 const char __user *buf,
1609 struct seq_file *seq = file->private_data;
1610 struct ll_sb_info *sbi = seq->private;
1611 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1618 value = ll_stats_pid_write(buf, len);
1621 sbi->ll_rw_stats_on = 0;
1623 sbi->ll_rw_stats_on = 1;
1625 spin_lock(&sbi->ll_pp_extent_lock);
1626 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1627 io_extents->pp_extents[i].pid = 0;
1628 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1629 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1631 spin_unlock(&sbi->ll_pp_extent_lock);
1635 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1637 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1639 struct timespec64 now;
1640 struct ll_sb_info *sbi = seq->private;
1641 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1643 ktime_get_real_ts64(&now);
1645 if (!sbi->ll_rw_stats_on) {
1646 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1649 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1650 (s64)now.tv_sec, now.tv_nsec);
1652 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1653 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1654 "extents", "calls", "%", "cum%",
1655 "calls", "%", "cum%");
1656 spin_lock(&sbi->ll_lock);
1657 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1658 spin_unlock(&sbi->ll_lock);
1663 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1664 const char __user *buf,
1665 size_t len, loff_t *off)
1667 struct seq_file *seq = file->private_data;
1668 struct ll_sb_info *sbi = seq->private;
1669 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1676 value = ll_stats_pid_write(buf, len);
1679 sbi->ll_rw_stats_on = 0;
1681 sbi->ll_rw_stats_on = 1;
1683 spin_lock(&sbi->ll_pp_extent_lock);
1684 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1685 io_extents->pp_extents[i].pid = 0;
1686 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1687 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1689 spin_unlock(&sbi->ll_pp_extent_lock);
1694 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1696 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1697 struct ll_file_data *file, loff_t pos,
1698 size_t count, int rw)
1701 struct ll_rw_process_info *process;
1702 struct ll_rw_process_info *offset;
1703 int *off_count = &sbi->ll_rw_offset_entry_count;
1704 int *process_count = &sbi->ll_offset_process_count;
1705 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1707 if(!sbi->ll_rw_stats_on)
1709 process = sbi->ll_rw_process_info;
1710 offset = sbi->ll_rw_offset_info;
1712 spin_lock(&sbi->ll_pp_extent_lock);
1713 /* Extent statistics */
1714 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1715 if(io_extents->pp_extents[i].pid == pid) {
1723 sbi->ll_extent_process_count =
1724 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1725 cur = sbi->ll_extent_process_count;
1726 io_extents->pp_extents[cur].pid = pid;
1727 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1728 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1731 for(i = 0; (count >= BIT(LL_HIST_START << i)) &&
1732 (i < (LL_HIST_MAX - 1)); i++);
1734 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1735 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1737 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1738 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1740 spin_unlock(&sbi->ll_pp_extent_lock);
1742 spin_lock(&sbi->ll_process_lock);
1743 /* Offset statistics */
1744 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1745 if (process[i].rw_pid == pid) {
1746 if (process[i].rw_last_file != file) {
1747 process[i].rw_range_start = pos;
1748 process[i].rw_last_file_pos = pos + count;
1749 process[i].rw_smallest_extent = count;
1750 process[i].rw_largest_extent = count;
1751 process[i].rw_offset = 0;
1752 process[i].rw_last_file = file;
1753 spin_unlock(&sbi->ll_process_lock);
1756 if (process[i].rw_last_file_pos != pos) {
1758 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1759 offset[*off_count].rw_op = process[i].rw_op;
1760 offset[*off_count].rw_pid = pid;
1761 offset[*off_count].rw_range_start =
1762 process[i].rw_range_start;
1763 offset[*off_count].rw_range_end =
1764 process[i].rw_last_file_pos;
1765 offset[*off_count].rw_smallest_extent =
1766 process[i].rw_smallest_extent;
1767 offset[*off_count].rw_largest_extent =
1768 process[i].rw_largest_extent;
1769 offset[*off_count].rw_offset =
1770 process[i].rw_offset;
1771 process[i].rw_op = rw;
1772 process[i].rw_range_start = pos;
1773 process[i].rw_smallest_extent = count;
1774 process[i].rw_largest_extent = count;
1775 process[i].rw_offset = pos -
1776 process[i].rw_last_file_pos;
1778 if(process[i].rw_smallest_extent > count)
1779 process[i].rw_smallest_extent = count;
1780 if(process[i].rw_largest_extent < count)
1781 process[i].rw_largest_extent = count;
1782 process[i].rw_last_file_pos = pos + count;
1783 spin_unlock(&sbi->ll_process_lock);
1787 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1788 process[*process_count].rw_pid = pid;
1789 process[*process_count].rw_op = rw;
1790 process[*process_count].rw_range_start = pos;
1791 process[*process_count].rw_last_file_pos = pos + count;
1792 process[*process_count].rw_smallest_extent = count;
1793 process[*process_count].rw_largest_extent = count;
1794 process[*process_count].rw_offset = 0;
1795 process[*process_count].rw_last_file = file;
1796 spin_unlock(&sbi->ll_process_lock);
1799 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1801 struct timespec64 now;
1802 struct ll_sb_info *sbi = seq->private;
1803 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1804 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1807 ktime_get_real_ts64(&now);
1809 if (!sbi->ll_rw_stats_on) {
1810 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1813 spin_lock(&sbi->ll_process_lock);
1815 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1816 (s64)now.tv_sec, now.tv_nsec);
1817 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1818 "R/W", "PID", "RANGE START", "RANGE END",
1819 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1821 /* We stored the discontiguous offsets here; print them first */
1822 for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1823 if (offset[i].rw_pid != 0)
1825 "%3c %10d %14llu %14llu %17lu %17lu %14llu\n",
1826 offset[i].rw_op == READ ? 'R' : 'W',
1828 offset[i].rw_range_start,
1829 offset[i].rw_range_end,
1830 (unsigned long)offset[i].rw_smallest_extent,
1831 (unsigned long)offset[i].rw_largest_extent,
1832 offset[i].rw_offset);
1835 /* Then print the current offsets for each process */
1836 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1837 if (process[i].rw_pid != 0)
1839 "%3c %10d %14llu %14llu %17lu %17lu %14llu\n",
1840 process[i].rw_op == READ ? 'R' : 'W',
1842 process[i].rw_range_start,
1843 process[i].rw_last_file_pos,
1844 (unsigned long)process[i].rw_smallest_extent,
1845 (unsigned long)process[i].rw_largest_extent,
1846 process[i].rw_offset);
1848 spin_unlock(&sbi->ll_process_lock);
1853 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
1854 const char __user *buf,
1855 size_t len, loff_t *off)
1857 struct seq_file *seq = file->private_data;
1858 struct ll_sb_info *sbi = seq->private;
1859 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1860 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1866 value = ll_stats_pid_write(buf, len);
1869 sbi->ll_rw_stats_on = 0;
1871 sbi->ll_rw_stats_on = 1;
1873 spin_lock(&sbi->ll_process_lock);
1874 sbi->ll_offset_process_count = 0;
1875 sbi->ll_rw_offset_entry_count = 0;
1876 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1877 LL_PROCESS_HIST_MAX);
1878 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1879 LL_OFFSET_HIST_MAX);
1880 spin_unlock(&sbi->ll_process_lock);
1885 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);