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 int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1088 struct super_block *sb = m->private;
1089 struct ll_sb_info *sbi = ll_s2sbi(sb);
1090 struct cl_client_cache *cache = sbi->ll_cache;
1094 pages = atomic_long_read(&cache->ccc_unstable_nr);
1095 mb = (pages * PAGE_SIZE) >> 20;
1097 seq_printf(m, "unstable_check: %8d\n"
1098 "unstable_pages: %12ld\n"
1099 "unstable_mb: %8d\n",
1100 cache->ccc_unstable_check, pages, mb);
1104 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1105 const char __user *buffer,
1106 size_t count, loff_t *unused)
1108 struct seq_file *seq = file->private_data;
1109 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1116 if (count >= sizeof(kernbuf))
1119 if (copy_from_user(kernbuf, buffer, count))
1123 buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1125 rc = kstrtobool_from_user(buffer, count, &val);
1129 /* borrow lru lock to set the value */
1130 spin_lock(&sbi->ll_cache->ccc_lru_lock);
1131 sbi->ll_cache->ccc_unstable_check = val;
1132 spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1137 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1139 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1141 struct super_block *sb = m->private;
1142 struct ll_sb_info *sbi = ll_s2sbi(sb);
1143 struct root_squash_info *squash = &sbi->ll_squash;
1145 seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1149 static ssize_t ll_root_squash_seq_write(struct file *file,
1150 const char __user *buffer,
1151 size_t count, loff_t *off)
1153 struct seq_file *m = file->private_data;
1154 struct super_block *sb = m->private;
1155 struct ll_sb_info *sbi = ll_s2sbi(sb);
1156 struct root_squash_info *squash = &sbi->ll_squash;
1158 return lprocfs_wr_root_squash(buffer, count, squash,
1159 ll_get_fsname(sb, NULL, 0));
1162 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1164 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1166 struct super_block *sb = m->private;
1167 struct ll_sb_info *sbi = ll_s2sbi(sb);
1168 struct root_squash_info *squash = &sbi->ll_squash;
1171 down_read(&squash->rsi_sem);
1172 if (!list_empty(&squash->rsi_nosquash_nids)) {
1173 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1174 &squash->rsi_nosquash_nids);
1178 seq_puts(m, "NONE\n");
1180 up_read(&squash->rsi_sem);
1185 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1186 const char __user *buffer,
1187 size_t count, loff_t *off)
1189 struct seq_file *m = file->private_data;
1190 struct super_block *sb = m->private;
1191 struct ll_sb_info *sbi = ll_s2sbi(sb);
1192 struct root_squash_info *squash = &sbi->ll_squash;
1195 rc = lprocfs_wr_nosquash_nids(buffer, count, squash,
1196 ll_get_fsname(sb, NULL, 0));
1200 ll_compute_rootsquash_state(sbi);
1205 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1207 struct lprocfs_vars lprocfs_llite_obd_vars[] = {
1209 .fops = &ll_site_stats_fops },
1210 { .name = "max_read_ahead_mb",
1211 .fops = &ll_max_readahead_mb_fops },
1212 { .name = "max_read_ahead_per_file_mb",
1213 .fops = &ll_max_readahead_per_file_mb_fops },
1214 { .name = "max_read_ahead_whole_mb",
1215 .fops = &ll_max_read_ahead_whole_mb_fops },
1216 { .name = "max_cached_mb",
1217 .fops = &ll_max_cached_mb_fops },
1218 { .name = "statahead_stats",
1219 .fops = &ll_statahead_stats_fops },
1220 { .name = "unstable_stats",
1221 .fops = &ll_unstable_stats_fops },
1222 { .name = "sbi_flags",
1223 .fops = &ll_sbi_flags_fops },
1224 { .name = "root_squash",
1225 .fops = &ll_root_squash_fops },
1226 { .name = "nosquash_nids",
1227 .fops = &ll_nosquash_nids_fops },
1231 #define MAX_STRING_SIZE 128
1233 static struct attribute *llite_attrs[] = {
1234 &lustre_attr_blocksize.attr,
1235 &lustre_attr_stat_blocksize.attr,
1236 &lustre_attr_kbytestotal.attr,
1237 &lustre_attr_kbytesfree.attr,
1238 &lustre_attr_kbytesavail.attr,
1239 &lustre_attr_filestotal.attr,
1240 &lustre_attr_filesfree.attr,
1241 &lustre_attr_client_type.attr,
1242 &lustre_attr_fstype.attr,
1243 &lustre_attr_uuid.attr,
1244 &lustre_attr_checksums.attr,
1245 &lustre_attr_checksum_pages.attr,
1246 &lustre_attr_stats_track_pid.attr,
1247 &lustre_attr_stats_track_ppid.attr,
1248 &lustre_attr_stats_track_gid.attr,
1249 &lustre_attr_statahead_running_max.attr,
1250 &lustre_attr_statahead_max.attr,
1251 &lustre_attr_statahead_agl.attr,
1252 &lustre_attr_lazystatfs.attr,
1253 &lustre_attr_max_easize.attr,
1254 &lustre_attr_default_easize.attr,
1255 &lustre_attr_xattr_cache.attr,
1256 &lustre_attr_fast_read.attr,
1257 &lustre_attr_tiny_write.attr,
1261 static void llite_kobj_release(struct kobject *kobj)
1263 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1265 complete(&sbi->ll_kobj_unregister);
1268 static struct kobj_type llite_ktype = {
1269 .default_attrs = llite_attrs,
1270 .sysfs_ops = &lustre_sysfs_ops,
1271 .release = llite_kobj_release,
1274 static const struct llite_file_opcode {
1278 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1279 /* file operation */
1280 { LPROC_LL_DIRTY_HITS, LPROCFS_TYPE_REGS, "dirty_pages_hits" },
1281 { LPROC_LL_DIRTY_MISSES, LPROCFS_TYPE_REGS, "dirty_pages_misses" },
1282 { LPROC_LL_READ_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
1284 { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
1286 { LPROC_LL_BRW_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
1288 { LPROC_LL_BRW_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
1290 { LPROC_LL_IOCTL, LPROCFS_TYPE_REGS, "ioctl" },
1291 { LPROC_LL_OPEN, LPROCFS_TYPE_REGS, "open" },
1292 { LPROC_LL_RELEASE, LPROCFS_TYPE_REGS, "close" },
1293 { LPROC_LL_MAP, LPROCFS_TYPE_REGS, "mmap" },
1294 { LPROC_LL_FAULT, LPROCFS_TYPE_REGS, "page_fault" },
1295 { LPROC_LL_MKWRITE, LPROCFS_TYPE_REGS, "page_mkwrite" },
1296 { LPROC_LL_LLSEEK, LPROCFS_TYPE_REGS, "seek" },
1297 { LPROC_LL_FSYNC, LPROCFS_TYPE_REGS, "fsync" },
1298 { LPROC_LL_READDIR, LPROCFS_TYPE_REGS, "readdir" },
1299 /* inode operation */
1300 { LPROC_LL_SETATTR, LPROCFS_TYPE_REGS, "setattr" },
1301 { LPROC_LL_TRUNC, LPROCFS_TYPE_REGS, "truncate" },
1302 { LPROC_LL_FLOCK, LPROCFS_TYPE_REGS, "flock" },
1303 { LPROC_LL_GETATTR, LPROCFS_TYPE_REGS, "getattr" },
1304 /* dir inode operation */
1305 { LPROC_LL_CREATE, LPROCFS_TYPE_REGS, "create" },
1306 { LPROC_LL_LINK, LPROCFS_TYPE_REGS, "link" },
1307 { LPROC_LL_UNLINK, LPROCFS_TYPE_REGS, "unlink" },
1308 { LPROC_LL_SYMLINK, LPROCFS_TYPE_REGS, "symlink" },
1309 { LPROC_LL_MKDIR, LPROCFS_TYPE_REGS, "mkdir" },
1310 { LPROC_LL_RMDIR, LPROCFS_TYPE_REGS, "rmdir" },
1311 { LPROC_LL_MKNOD, LPROCFS_TYPE_REGS, "mknod" },
1312 { LPROC_LL_RENAME, LPROCFS_TYPE_REGS, "rename" },
1313 /* special inode operation */
1314 { LPROC_LL_STAFS, LPROCFS_TYPE_REGS, "statfs" },
1315 { LPROC_LL_ALLOC_INODE, LPROCFS_TYPE_REGS, "alloc_inode" },
1316 { LPROC_LL_SETXATTR, LPROCFS_TYPE_REGS, "setxattr" },
1317 { LPROC_LL_GETXATTR, LPROCFS_TYPE_REGS, "getxattr" },
1318 { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REGS, "getxattr_hits" },
1319 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_REGS, "listxattr" },
1320 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_REGS, "removexattr" },
1321 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_REGS, "inode_permission" },
1324 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
1328 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1329 lprocfs_counter_add(sbi->ll_stats, op, count);
1330 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1331 sbi->ll_stats_track_id == current->pid)
1332 lprocfs_counter_add(sbi->ll_stats, op, count);
1333 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1334 sbi->ll_stats_track_id == current->parent->pid)
1335 lprocfs_counter_add(sbi->ll_stats, op, count);
1336 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1337 sbi->ll_stats_track_id ==
1338 from_kgid(&init_user_ns, current_gid()))
1339 lprocfs_counter_add(sbi->ll_stats, op, count);
1341 EXPORT_SYMBOL(ll_stats_ops_tally);
1343 static const char *ra_stat_string[] = {
1344 [RA_STAT_HIT] = "hits",
1345 [RA_STAT_MISS] = "misses",
1346 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1347 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1348 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1349 [RA_STAT_FAILED_MATCH] = "failed lock match",
1350 [RA_STAT_DISCARDED] = "read but discarded",
1351 [RA_STAT_ZERO_LEN] = "zero length file",
1352 [RA_STAT_ZERO_WINDOW] = "zero size window",
1353 [RA_STAT_EOF] = "read-ahead to EOF",
1354 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1355 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1356 [RA_STAT_FAILED_REACH_END] = "failed to reach end"
1359 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1361 struct lustre_sb_info *lsi = s2lsi(sb);
1362 struct ll_sb_info *sbi = ll_s2sbi(sb);
1368 if (IS_ERR_OR_NULL(llite_root))
1371 sbi->ll_debugfs_entry = ldebugfs_register(name, llite_root,
1372 lprocfs_llite_obd_vars, sb);
1373 if (IS_ERR_OR_NULL(sbi->ll_debugfs_entry)) {
1374 err = sbi->ll_debugfs_entry ? PTR_ERR(sbi->ll_debugfs_entry) :
1376 sbi->ll_debugfs_entry = NULL;
1380 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "dump_page_cache",0444,
1381 &vvp_dump_pgcache_file_ops, sbi);
1383 CWARN("Error adding the dump_page_cache file\n");
1385 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "extents_stats", 0644,
1386 &ll_rw_extents_stats_fops, sbi);
1388 CWARN("Error adding the extent_stats file\n");
1390 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry,
1391 "extents_stats_per_process", 0644,
1392 &ll_rw_extents_stats_pp_fops, sbi);
1394 CWARN("Error adding the extents_stats_per_process file\n");
1396 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "offset_stats", 0644,
1397 &ll_rw_offset_stats_fops, sbi);
1399 CWARN("Error adding the offset_stats file\n");
1401 /* File operations stats */
1402 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1403 LPROCFS_STATS_FLAG_NONE);
1404 if (sbi->ll_stats == NULL)
1405 GOTO(out_debugfs, err = -ENOMEM);
1407 /* do counter init */
1408 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1409 u32 type = llite_opcode_table[id].type;
1412 if (type & LPROCFS_TYPE_REGS)
1414 else if (type & LPROCFS_TYPE_BYTES)
1416 else if (type & LPROCFS_TYPE_PAGES)
1418 lprocfs_counter_init(sbi->ll_stats,
1419 llite_opcode_table[id].opcode,
1420 (type & LPROCFS_CNTR_AVGMINMAX),
1421 llite_opcode_table[id].opname, ptr);
1424 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "stats",
1427 GOTO(out_stats, err);
1429 sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1430 LPROCFS_STATS_FLAG_NONE);
1431 if (sbi->ll_ra_stats == NULL)
1432 GOTO(out_stats, err = -ENOMEM);
1434 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1435 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1436 ra_stat_string[id], "pages");
1438 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "read_ahead_stats",
1441 GOTO(out_ra_stats, err);
1444 /* Yes we also register sysfs mount kset here as well */
1445 sbi->ll_kset.kobj.parent = llite_kobj;
1446 sbi->ll_kset.kobj.ktype = &llite_ktype;
1447 init_completion(&sbi->ll_kobj_unregister);
1448 err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1450 GOTO(out_ra_stats, err);
1452 err = kset_register(&sbi->ll_kset);
1454 GOTO(out_ra_stats, err);
1456 lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1460 lprocfs_free_stats(&sbi->ll_ra_stats);
1462 lprocfs_free_stats(&sbi->ll_stats);
1464 ldebugfs_remove(&sbi->ll_debugfs_entry);
1469 void ll_debugfs_unregister_super(struct super_block *sb)
1471 struct lustre_sb_info *lsi = s2lsi(sb);
1472 struct ll_sb_info *sbi = ll_s2sbi(sb);
1474 if (!IS_ERR_OR_NULL(sbi->ll_debugfs_entry))
1475 ldebugfs_remove(&sbi->ll_debugfs_entry);
1478 sysfs_remove_link(&sbi->ll_kset.kobj,
1479 sbi->ll_dt_obd->obd_type->typ_name);
1482 sysfs_remove_link(&sbi->ll_kset.kobj,
1483 sbi->ll_md_obd->obd_type->typ_name);
1485 kobject_put(lsi->lsi_kobj);
1487 kset_unregister(&sbi->ll_kset);
1488 wait_for_completion(&sbi->ll_kobj_unregister);
1490 lprocfs_free_stats(&sbi->ll_ra_stats);
1491 lprocfs_free_stats(&sbi->ll_stats);
1493 #undef MAX_STRING_SIZE
1495 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1496 struct seq_file *seq, int which)
1498 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1499 unsigned long start, end, r, w;
1500 char *unitp = "KMGTPEZY";
1502 struct per_process_info *pp_info = &io_extents->pp_extents[which];
1508 for(i = 0; i < LL_HIST_MAX; i++) {
1509 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1510 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1513 for(i = 0; i < LL_HIST_MAX; i++) {
1514 r = pp_info->pp_r_hist.oh_buckets[i];
1515 w = pp_info->pp_w_hist.oh_buckets[i];
1518 end = BIT(i + LL_HIST_START - units);
1519 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u | "
1520 "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1521 (i == LL_HIST_MAX - 1) ? '+' : ' ',
1522 r, pct(r, read_tot), pct(read_cum, read_tot),
1523 w, pct(w, write_tot), pct(write_cum, write_tot));
1525 if (start == BIT(10)) {
1530 if (read_cum == read_tot && write_cum == write_tot)
1535 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1537 struct timespec64 now;
1538 struct ll_sb_info *sbi = seq->private;
1539 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1542 ktime_get_real_ts64(&now);
1544 if (!sbi->ll_rw_stats_on) {
1545 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1548 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1549 (s64)now.tv_sec, now.tv_nsec);
1550 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1551 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1552 "extents", "calls", "%", "cum%",
1553 "calls", "%", "cum%");
1554 spin_lock(&sbi->ll_pp_extent_lock);
1555 for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1556 if (io_extents->pp_extents[k].pid != 0) {
1557 seq_printf(seq, "\nPID: %d\n",
1558 io_extents->pp_extents[k].pid);
1559 ll_display_extents_info(io_extents, seq, k);
1562 spin_unlock(&sbi->ll_pp_extent_lock);
1566 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1567 const char __user *buf,
1571 struct seq_file *seq = file->private_data;
1572 struct ll_sb_info *sbi = seq->private;
1573 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1580 value = ll_stats_pid_write(buf, len);
1583 sbi->ll_rw_stats_on = 0;
1585 sbi->ll_rw_stats_on = 1;
1587 spin_lock(&sbi->ll_pp_extent_lock);
1588 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1589 io_extents->pp_extents[i].pid = 0;
1590 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1591 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1593 spin_unlock(&sbi->ll_pp_extent_lock);
1597 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1599 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1601 struct timespec64 now;
1602 struct ll_sb_info *sbi = seq->private;
1603 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1605 ktime_get_real_ts64(&now);
1607 if (!sbi->ll_rw_stats_on) {
1608 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1611 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1612 (s64)now.tv_sec, now.tv_nsec);
1614 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1615 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1616 "extents", "calls", "%", "cum%",
1617 "calls", "%", "cum%");
1618 spin_lock(&sbi->ll_lock);
1619 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1620 spin_unlock(&sbi->ll_lock);
1625 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1626 const char __user *buf,
1627 size_t len, loff_t *off)
1629 struct seq_file *seq = file->private_data;
1630 struct ll_sb_info *sbi = seq->private;
1631 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1638 value = ll_stats_pid_write(buf, len);
1641 sbi->ll_rw_stats_on = 0;
1643 sbi->ll_rw_stats_on = 1;
1645 spin_lock(&sbi->ll_pp_extent_lock);
1646 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1647 io_extents->pp_extents[i].pid = 0;
1648 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1649 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1651 spin_unlock(&sbi->ll_pp_extent_lock);
1656 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1658 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1659 struct ll_file_data *file, loff_t pos,
1660 size_t count, int rw)
1663 struct ll_rw_process_info *process;
1664 struct ll_rw_process_info *offset;
1665 int *off_count = &sbi->ll_rw_offset_entry_count;
1666 int *process_count = &sbi->ll_offset_process_count;
1667 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1669 if(!sbi->ll_rw_stats_on)
1671 process = sbi->ll_rw_process_info;
1672 offset = sbi->ll_rw_offset_info;
1674 spin_lock(&sbi->ll_pp_extent_lock);
1675 /* Extent statistics */
1676 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1677 if(io_extents->pp_extents[i].pid == pid) {
1685 sbi->ll_extent_process_count =
1686 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1687 cur = sbi->ll_extent_process_count;
1688 io_extents->pp_extents[cur].pid = pid;
1689 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1690 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1693 for(i = 0; (count >= BIT(LL_HIST_START << i)) &&
1694 (i < (LL_HIST_MAX - 1)); i++);
1696 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1697 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1699 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1700 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1702 spin_unlock(&sbi->ll_pp_extent_lock);
1704 spin_lock(&sbi->ll_process_lock);
1705 /* Offset statistics */
1706 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1707 if (process[i].rw_pid == pid) {
1708 if (process[i].rw_last_file != file) {
1709 process[i].rw_range_start = pos;
1710 process[i].rw_last_file_pos = pos + count;
1711 process[i].rw_smallest_extent = count;
1712 process[i].rw_largest_extent = count;
1713 process[i].rw_offset = 0;
1714 process[i].rw_last_file = file;
1715 spin_unlock(&sbi->ll_process_lock);
1718 if (process[i].rw_last_file_pos != pos) {
1720 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1721 offset[*off_count].rw_op = process[i].rw_op;
1722 offset[*off_count].rw_pid = pid;
1723 offset[*off_count].rw_range_start =
1724 process[i].rw_range_start;
1725 offset[*off_count].rw_range_end =
1726 process[i].rw_last_file_pos;
1727 offset[*off_count].rw_smallest_extent =
1728 process[i].rw_smallest_extent;
1729 offset[*off_count].rw_largest_extent =
1730 process[i].rw_largest_extent;
1731 offset[*off_count].rw_offset =
1732 process[i].rw_offset;
1733 process[i].rw_op = rw;
1734 process[i].rw_range_start = pos;
1735 process[i].rw_smallest_extent = count;
1736 process[i].rw_largest_extent = count;
1737 process[i].rw_offset = pos -
1738 process[i].rw_last_file_pos;
1740 if(process[i].rw_smallest_extent > count)
1741 process[i].rw_smallest_extent = count;
1742 if(process[i].rw_largest_extent < count)
1743 process[i].rw_largest_extent = count;
1744 process[i].rw_last_file_pos = pos + count;
1745 spin_unlock(&sbi->ll_process_lock);
1749 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1750 process[*process_count].rw_pid = pid;
1751 process[*process_count].rw_op = rw;
1752 process[*process_count].rw_range_start = pos;
1753 process[*process_count].rw_last_file_pos = pos + count;
1754 process[*process_count].rw_smallest_extent = count;
1755 process[*process_count].rw_largest_extent = count;
1756 process[*process_count].rw_offset = 0;
1757 process[*process_count].rw_last_file = file;
1758 spin_unlock(&sbi->ll_process_lock);
1761 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1763 struct timespec64 now;
1764 struct ll_sb_info *sbi = seq->private;
1765 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1766 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1769 ktime_get_real_ts64(&now);
1771 if (!sbi->ll_rw_stats_on) {
1772 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1775 spin_lock(&sbi->ll_process_lock);
1777 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1778 (s64)now.tv_sec, now.tv_nsec);
1779 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1780 "R/W", "PID", "RANGE START", "RANGE END",
1781 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1783 /* We stored the discontiguous offsets here; print them first */
1784 for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1785 if (offset[i].rw_pid != 0)
1787 "%3c %10d %14llu %14llu %17lu %17lu %14llu\n",
1788 offset[i].rw_op == READ ? 'R' : 'W',
1790 offset[i].rw_range_start,
1791 offset[i].rw_range_end,
1792 (unsigned long)offset[i].rw_smallest_extent,
1793 (unsigned long)offset[i].rw_largest_extent,
1794 offset[i].rw_offset);
1797 /* Then print the current offsets for each process */
1798 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1799 if (process[i].rw_pid != 0)
1801 "%3c %10d %14llu %14llu %17lu %17lu %14llu\n",
1802 process[i].rw_op == READ ? 'R' : 'W',
1804 process[i].rw_range_start,
1805 process[i].rw_last_file_pos,
1806 (unsigned long)process[i].rw_smallest_extent,
1807 (unsigned long)process[i].rw_largest_extent,
1808 process[i].rw_offset);
1810 spin_unlock(&sbi->ll_process_lock);
1815 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
1816 const char __user *buf,
1817 size_t len, loff_t *off)
1819 struct seq_file *seq = file->private_data;
1820 struct ll_sb_info *sbi = seq->private;
1821 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1822 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1828 value = ll_stats_pid_write(buf, len);
1831 sbi->ll_rw_stats_on = 0;
1833 sbi->ll_rw_stats_on = 1;
1835 spin_lock(&sbi->ll_process_lock);
1836 sbi->ll_offset_process_count = 0;
1837 sbi->ll_rw_offset_entry_count = 0;
1838 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1839 LL_PROCESS_HIST_MAX);
1840 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1841 LL_OFFSET_HIST_MAX);
1842 spin_unlock(&sbi->ll_process_lock);
1847 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);