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 static void llite_kobj_release(struct kobject *kobj)
51 if (!IS_ERR_OR_NULL(llite_root)) {
52 debugfs_remove(llite_root);
59 static struct kobj_type llite_kobj_ktype = {
60 .release = llite_kobj_release,
61 .sysfs_ops = &lustre_sysfs_ops,
64 int llite_tunables_register(void)
68 llite_kobj = kzalloc(sizeof(*llite_kobj), GFP_KERNEL);
72 llite_kobj->kset = lustre_kset;
73 rc = kobject_init_and_add(llite_kobj, &llite_kobj_ktype,
74 &lustre_kset->kobj, "%s", "llite");
78 llite_root = debugfs_create_dir("llite", debugfs_lustre_root);
79 if (IS_ERR_OR_NULL(llite_root)) {
80 rc = llite_root ? PTR_ERR(llite_root) : -ENOMEM;
83 kobject_put(llite_kobj);
90 void llite_tunables_unregister(void)
92 kobject_put(llite_kobj);
96 /* <debugfs>/lustre/llite mount point registration */
97 static const struct file_operations ll_rw_extents_stats_fops;
98 static const struct file_operations ll_rw_extents_stats_pp_fops;
99 static const struct file_operations ll_rw_offset_stats_fops;
102 * ll_stats_pid_write() - Determine if stats collection should be enabled
103 * @buf: Buffer containing the data written
104 * @len: Number of bytes in the buffer
106 * Several proc files begin collecting stats when a value is written, and stop
107 * collecting when either '0' or 'disable' is written. This function checks the
108 * written value to see if collection should be enabled or disabled.
110 * Return: If '0' or 'disable' is provided, 0 is returned. If the text
111 * equivalent of a number is written, that number is returned. Otherwise,
112 * 1 is returned. Non-zero return values indicate collection should be enabled.
114 static s64 ll_stats_pid_write(const char __user *buf, size_t len)
116 unsigned long long value = 1;
120 rc = kstrtoull_from_user(buf, len, 0, &value);
121 if (rc < 0 && len < sizeof(kernbuf)) {
122 if (copy_from_user(kernbuf, buf, len))
126 if (kernbuf[len - 1] == '\n')
127 kernbuf[len - 1] = 0;
129 if (strncasecmp(kernbuf, "disable", 7) == 0)
136 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
139 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
141 struct obd_statfs osfs;
144 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
148 return sprintf(buf, "%u\n", osfs.os_bsize);
150 LUSTRE_RO_ATTR(blocksize);
152 static ssize_t stat_blocksize_show(struct kobject *kobj, struct attribute *attr,
155 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
158 return sprintf(buf, "%u\n", sbi->ll_stat_blksize);
161 static ssize_t stat_blocksize_store(struct kobject *kobj,
162 struct attribute *attr,
166 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
171 rc = kstrtouint(buffer, 10, &val);
175 if (val != 0 && (val < PAGE_SIZE || (val & (val - 1))) != 0)
178 sbi->ll_stat_blksize = val;
182 LUSTRE_RW_ATTR(stat_blocksize);
184 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
187 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
189 struct obd_statfs osfs;
194 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
198 blk_size = osfs.os_bsize >> 10;
199 result = osfs.os_blocks;
201 while (blk_size >>= 1)
204 return sprintf(buf, "%llu\n", result);
206 LUSTRE_RO_ATTR(kbytestotal);
208 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
211 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
213 struct obd_statfs osfs;
218 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
222 blk_size = osfs.os_bsize >> 10;
223 result = osfs.os_bfree;
225 while (blk_size >>= 1)
228 return sprintf(buf, "%llu\n", result);
230 LUSTRE_RO_ATTR(kbytesfree);
232 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
235 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
237 struct obd_statfs osfs;
242 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
246 blk_size = osfs.os_bsize >> 10;
247 result = osfs.os_bavail;
249 while (blk_size >>= 1)
252 return sprintf(buf, "%llu\n", result);
254 LUSTRE_RO_ATTR(kbytesavail);
256 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
259 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
261 struct obd_statfs osfs;
264 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
268 return sprintf(buf, "%llu\n", osfs.os_files);
270 LUSTRE_RO_ATTR(filestotal);
272 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
275 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
277 struct obd_statfs osfs;
280 rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
284 return sprintf(buf, "%llu\n", osfs.os_ffree);
286 LUSTRE_RO_ATTR(filesfree);
288 static ssize_t client_type_show(struct kobject *kobj, struct attribute *attr,
291 return sprintf(buf, "local client\n");
293 LUSTRE_RO_ATTR(client_type);
295 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
298 return sprintf(buf, "lustre\n");
300 LUSTRE_RO_ATTR(fstype);
302 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
305 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
308 return sprintf(buf, "%s\n", sbi->ll_sb_uuid.uuid);
310 LUSTRE_RO_ATTR(uuid);
312 static int ll_site_stats_seq_show(struct seq_file *m, void *v)
314 struct super_block *sb = m->private;
317 * See description of statistical counters in struct cl_site, and
320 return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site), m);
323 LDEBUGFS_SEQ_FOPS_RO(ll_site_stats);
325 static int ll_max_readahead_mb_seq_show(struct seq_file *m, void *v)
327 struct super_block *sb = m->private;
328 struct ll_sb_info *sbi = ll_s2sbi(sb);
329 unsigned long ra_max_mb;
331 spin_lock(&sbi->ll_lock);
332 ra_max_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages);
333 spin_unlock(&sbi->ll_lock);
335 seq_printf(m, "%lu\n", ra_max_mb);
340 ll_max_readahead_mb_seq_write(struct file *file, const char __user *buffer,
341 size_t count, loff_t *off)
343 struct seq_file *m = file->private_data;
344 struct super_block *sb = m->private;
345 struct ll_sb_info *sbi = ll_s2sbi(sb);
346 s64 ra_max_mb, pages_number;
349 rc = lprocfs_str_with_units_to_s64(buffer, count, &ra_max_mb, 'M');
353 pages_number = round_up(ra_max_mb, 1024 * 1024) >> PAGE_SHIFT;
354 if (pages_number < 0 || pages_number > totalram_pages / 2) {
356 CERROR("%s: can't set max_readahead_mb=%llu > %luMB\n",
357 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
358 PAGES_TO_MiB(totalram_pages));
362 spin_lock(&sbi->ll_lock);
363 sbi->ll_ra_info.ra_max_pages = pages_number;
364 spin_unlock(&sbi->ll_lock);
369 LDEBUGFS_SEQ_FOPS(ll_max_readahead_mb);
371 static int ll_max_readahead_per_file_mb_seq_show(struct seq_file *m, void *v)
373 struct super_block *sb = m->private;
374 struct ll_sb_info *sbi = ll_s2sbi(sb);
375 unsigned long ra_max_file_mb;
377 spin_lock(&sbi->ll_lock);
378 ra_max_file_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file);
379 spin_unlock(&sbi->ll_lock);
381 seq_printf(m, "%lu\n", ra_max_file_mb);
386 ll_max_readahead_per_file_mb_seq_write(struct file *file,
387 const char __user *buffer,
388 size_t count, loff_t *off)
390 struct seq_file *m = file->private_data;
391 struct super_block *sb = m->private;
392 struct ll_sb_info *sbi = ll_s2sbi(sb);
393 s64 ra_max_file_mb, pages_number;
396 rc = lprocfs_str_with_units_to_s64(buffer, count, &ra_max_file_mb,
401 pages_number = round_up(ra_max_file_mb, 1024 * 1024) >> PAGE_SHIFT;
402 if (pages_number < 0 || pages_number > sbi->ll_ra_info.ra_max_pages) {
403 CERROR("%s: can't set max_readahead_per_file_mb=%llu > max_read_ahead_mb=%lu\n",
404 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
405 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
409 spin_lock(&sbi->ll_lock);
410 sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
411 spin_unlock(&sbi->ll_lock);
416 LDEBUGFS_SEQ_FOPS(ll_max_readahead_per_file_mb);
418 static int ll_max_read_ahead_whole_mb_seq_show(struct seq_file *m, void *v)
420 struct super_block *sb = m->private;
421 struct ll_sb_info *sbi = ll_s2sbi(sb);
422 unsigned long ra_max_whole_mb;
424 spin_lock(&sbi->ll_lock);
425 ra_max_whole_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_read_ahead_whole_pages);
426 spin_unlock(&sbi->ll_lock);
428 seq_printf(m, "%lu\n", ra_max_whole_mb);
433 ll_max_read_ahead_whole_mb_seq_write(struct file *file,
434 const char __user *buffer,
435 size_t count, loff_t *off)
437 struct seq_file *m = file->private_data;
438 struct super_block *sb = m->private;
439 struct ll_sb_info *sbi = ll_s2sbi(sb);
440 s64 ra_max_whole_mb, pages_number;
443 rc = lprocfs_str_with_units_to_s64(buffer, count, &ra_max_whole_mb,
448 pages_number = round_up(ra_max_whole_mb, 1024 * 1024) >> PAGE_SHIFT;
449 /* Cap this at the current max readahead window size, the readahead
450 * algorithm does this anyway so it's pointless to set it larger.
452 if (pages_number < 0 ||
453 pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
454 CERROR("%s: can't set max_read_ahead_whole_mb=%llu > max_read_ahead_per_file_mb=%lu\n",
455 sbi->ll_fsname, PAGES_TO_MiB(pages_number),
456 PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
460 spin_lock(&sbi->ll_lock);
461 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
462 spin_unlock(&sbi->ll_lock);
467 LDEBUGFS_SEQ_FOPS(ll_max_read_ahead_whole_mb);
469 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
471 struct super_block *sb = m->private;
472 struct ll_sb_info *sbi = ll_s2sbi(sb);
473 struct cl_client_cache *cache = sbi->ll_cache;
477 max_cached_mb = PAGES_TO_MiB(cache->ccc_lru_max);
478 unused_mb = PAGES_TO_MiB(atomic_long_read(&cache->ccc_lru_left));
479 seq_printf(m, "users: %d\n"
480 "max_cached_mb: %ld\n"
483 "reclaim_count: %u\n",
484 atomic_read(&cache->ccc_users),
486 max_cached_mb - unused_mb,
488 cache->ccc_lru_shrinkers);
492 static ssize_t ll_max_cached_mb_seq_write(struct file *file,
493 const char __user *buffer,
494 size_t count, loff_t *off)
496 struct seq_file *m = file->private_data;
497 struct super_block *sb = m->private;
498 struct ll_sb_info *sbi = ll_s2sbi(sb);
499 struct cl_client_cache *cache = sbi->ll_cache;
509 if (count >= sizeof(kernbuf))
512 if (copy_from_user(kernbuf, buffer, count))
516 buffer += lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count) -
518 rc = lprocfs_str_with_units_to_s64(buffer, count, &pages_number, 'M');
522 pages_number >>= PAGE_SHIFT;
524 if (pages_number < 0 || pages_number > totalram_pages) {
525 CERROR("%s: can't set max cache more than %lu MB\n",
527 PAGES_TO_MiB(totalram_pages));
530 /* Allow enough cache so clients can make well-formed RPCs */
531 pages_number = max_t(long, pages_number, PTLRPC_MAX_BRW_PAGES);
533 spin_lock(&sbi->ll_lock);
534 diff = pages_number - cache->ccc_lru_max;
535 spin_unlock(&sbi->ll_lock);
537 /* easy - add more LRU slots. */
539 atomic_long_add(diff, &cache->ccc_lru_left);
543 env = cl_env_get(&refcheck);
545 RETURN(PTR_ERR(env));
551 /* reduce LRU budget from free slots. */
555 ov = atomic_long_read(&cache->ccc_lru_left);
559 nv = ov > diff ? ov - diff : 0;
560 rc = atomic_long_cmpxchg(&cache->ccc_lru_left, ov, nv);
561 if (likely(ov == rc)) {
571 if (sbi->ll_dt_exp == NULL) { /* being initialized */
576 /* difficult - have to ask OSCs to drop LRU slots. */
578 rc = obd_set_info_async(env, sbi->ll_dt_exp,
579 sizeof(KEY_CACHE_LRU_SHRINK),
580 KEY_CACHE_LRU_SHRINK,
581 sizeof(tmp), &tmp, NULL);
585 cl_env_put(env, &refcheck);
589 spin_lock(&sbi->ll_lock);
590 cache->ccc_lru_max = pages_number;
591 spin_unlock(&sbi->ll_lock);
594 atomic_long_add(nrpages, &cache->ccc_lru_left);
599 LDEBUGFS_SEQ_FOPS(ll_max_cached_mb);
601 static ssize_t checksums_show(struct kobject *kobj, struct attribute *attr,
604 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
607 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
610 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
611 const char *buffer, size_t count)
613 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
623 rc = kstrtobool(buffer, &val);
627 sbi->ll_flags |= LL_SBI_CHECKSUM;
629 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
632 rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
633 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
635 CWARN("Failed to set OSC checksum flags: %d\n", rc);
639 LUSTRE_RW_ATTR(checksums);
641 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
643 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
644 enum stats_track_type type)
646 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
649 if (sbi->ll_stats_track_type == type)
650 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
651 else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
652 return sprintf(buf, "0 (all)\n");
654 return sprintf(buf, "untracked\n");
657 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
658 size_t count, enum stats_track_type type)
660 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
665 rc = kstrtoul(buffer, 10, &pid);
669 sbi->ll_stats_track_id = pid;
671 sbi->ll_stats_track_type = STATS_TRACK_ALL;
673 sbi->ll_stats_track_type = type;
674 lprocfs_clear_stats(sbi->ll_stats);
678 static ssize_t stats_track_pid_show(struct kobject *kobj,
679 struct attribute *attr,
682 return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
685 static ssize_t stats_track_pid_store(struct kobject *kobj,
686 struct attribute *attr,
690 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
692 LUSTRE_RW_ATTR(stats_track_pid);
694 static ssize_t stats_track_ppid_show(struct kobject *kobj,
695 struct attribute *attr,
698 return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
701 static ssize_t stats_track_ppid_store(struct kobject *kobj,
702 struct attribute *attr,
706 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
708 LUSTRE_RW_ATTR(stats_track_ppid);
710 static ssize_t stats_track_gid_show(struct kobject *kobj,
711 struct attribute *attr,
714 return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
717 static ssize_t stats_track_gid_store(struct kobject *kobj,
718 struct attribute *attr,
722 return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
724 LUSTRE_RW_ATTR(stats_track_gid);
726 static ssize_t statahead_running_max_show(struct kobject *kobj,
727 struct attribute *attr,
730 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
733 return snprintf(buf, 16, "%u\n", sbi->ll_sa_running_max);
736 static ssize_t statahead_running_max_store(struct kobject *kobj,
737 struct attribute *attr,
741 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
746 rc = kstrtoul(buffer, 0, &val);
750 if (val <= LL_SA_RUNNING_MAX) {
751 sbi->ll_sa_running_max = val;
755 CERROR("Bad statahead_running_max value %lu. Valid values "
756 "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
760 LUSTRE_RW_ATTR(statahead_running_max);
762 static ssize_t statahead_max_show(struct kobject *kobj,
763 struct attribute *attr,
766 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
769 return sprintf(buf, "%u\n", sbi->ll_sa_max);
772 static ssize_t statahead_max_store(struct kobject *kobj,
773 struct attribute *attr,
777 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
782 rc = kstrtoul(buffer, 0, &val);
786 if (val <= LL_SA_RPC_MAX)
787 sbi->ll_sa_max = val;
789 CERROR("Bad statahead_max value %lu. Valid values are in the range [0, %d]\n",
794 LUSTRE_RW_ATTR(statahead_max);
796 static ssize_t statahead_agl_show(struct kobject *kobj,
797 struct attribute *attr,
800 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
803 return sprintf(buf, "%u\n", sbi->ll_flags & LL_SBI_AGL_ENABLED ? 1 : 0);
806 static ssize_t statahead_agl_store(struct kobject *kobj,
807 struct attribute *attr,
811 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
816 rc = kstrtobool(buffer, &val);
821 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
823 sbi->ll_flags &= ~LL_SBI_AGL_ENABLED;
827 LUSTRE_RW_ATTR(statahead_agl);
829 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
831 struct super_block *sb = m->private;
832 struct ll_sb_info *sbi = ll_s2sbi(sb);
834 seq_printf(m, "statahead total: %u\n"
835 "statahead wrong: %u\n"
837 atomic_read(&sbi->ll_sa_total),
838 atomic_read(&sbi->ll_sa_wrong),
839 atomic_read(&sbi->ll_agl_total));
843 LDEBUGFS_SEQ_FOPS_RO(ll_statahead_stats);
845 static ssize_t lazystatfs_show(struct kobject *kobj,
846 struct attribute *attr,
849 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
852 return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
855 static ssize_t lazystatfs_store(struct kobject *kobj,
856 struct attribute *attr,
860 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
865 rc = kstrtobool(buffer, &val);
870 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
872 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
876 LUSTRE_RW_ATTR(lazystatfs);
878 static ssize_t max_easize_show(struct kobject *kobj,
879 struct attribute *attr,
882 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
887 rc = ll_get_max_mdsize(sbi, &ealen);
891 return sprintf(buf, "%u\n", ealen);
893 LUSTRE_RO_ATTR(max_easize);
896 * Get default_easize.
898 * \see client_obd::cl_default_mds_easize
900 * \param[in] m seq_file handle
901 * \param[in] v unused for single entry
903 * \retval 0 on success
904 * \retval negative negated errno on failure
906 static ssize_t default_easize_show(struct kobject *kobj,
907 struct attribute *attr,
910 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
915 rc = ll_get_default_mdsize(sbi, &ealen);
919 return sprintf(buf, "%u\n", ealen);
923 * Set default_easize.
925 * Range checking on the passed value is handled by
926 * ll_set_default_mdsize().
928 * \see client_obd::cl_default_mds_easize
930 * \param[in] file proc file
931 * \param[in] buffer string passed from user space
932 * \param[in] count \a buffer length
933 * \param[in] off unused for single entry
935 * \retval positive \a count on success
936 * \retval negative negated errno on failure
938 static ssize_t default_easize_store(struct kobject *kobj,
939 struct attribute *attr,
943 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
951 rc = kstrtouint(buffer, 10, &val);
955 rc = ll_set_default_mdsize(sbi, val);
961 LUSTRE_RW_ATTR(default_easize);
963 static int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
965 const char *str[] = LL_SBI_FLAGS;
966 struct super_block *sb = m->private;
967 int flags = ll_s2sbi(sb)->ll_flags;
971 if (ARRAY_SIZE(str) <= i) {
972 CERROR("%s: Revise array LL_SBI_FLAGS to match sbi "
973 "flags please.\n", ll_s2sbi(sb)->ll_fsname);
978 seq_printf(m, "%s ", str[i]);
982 seq_printf(m, "\b\n");
986 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
988 static ssize_t xattr_cache_show(struct kobject *kobj,
989 struct attribute *attr,
992 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
995 return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
998 static ssize_t xattr_cache_store(struct kobject *kobj,
999 struct attribute *attr,
1003 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1008 rc = kstrtobool(buffer, &val);
1012 if (val && !(sbi->ll_flags & LL_SBI_XATTR_CACHE))
1015 sbi->ll_xattr_cache_enabled = val;
1016 sbi->ll_xattr_cache_set = 1;
1020 LUSTRE_RW_ATTR(xattr_cache);
1022 static ssize_t tiny_write_show(struct kobject *kobj,
1023 struct attribute *attr,
1026 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1029 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_TINY_WRITE));
1032 static ssize_t tiny_write_store(struct kobject *kobj,
1033 struct attribute *attr,
1037 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1042 rc = kstrtobool(buffer, &val);
1046 spin_lock(&sbi->ll_lock);
1048 sbi->ll_flags |= LL_SBI_TINY_WRITE;
1050 sbi->ll_flags &= ~LL_SBI_TINY_WRITE;
1051 spin_unlock(&sbi->ll_lock);
1055 LUSTRE_RW_ATTR(tiny_write);
1057 static ssize_t fast_read_show(struct kobject *kobj,
1058 struct attribute *attr,
1061 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1064 return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_FAST_READ));
1067 static ssize_t fast_read_store(struct kobject *kobj,
1068 struct attribute *attr,
1072 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1077 rc = kstrtobool(buffer, &val);
1081 spin_lock(&sbi->ll_lock);
1083 sbi->ll_flags |= LL_SBI_FAST_READ;
1085 sbi->ll_flags &= ~LL_SBI_FAST_READ;
1086 spin_unlock(&sbi->ll_lock);
1090 LUSTRE_RW_ATTR(fast_read);
1092 static ssize_t file_heat_show(struct kobject *kobj,
1093 struct attribute *attr,
1096 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1099 return snprintf(buf, PAGE_SIZE, "%u\n",
1100 !!(sbi->ll_flags & LL_SBI_FILE_HEAT));
1103 static ssize_t file_heat_store(struct kobject *kobj,
1104 struct attribute *attr,
1108 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1113 rc = kstrtobool(buffer, &val);
1117 spin_lock(&sbi->ll_lock);
1119 sbi->ll_flags |= LL_SBI_FILE_HEAT;
1121 sbi->ll_flags &= ~LL_SBI_FILE_HEAT;
1122 spin_unlock(&sbi->ll_lock);
1126 LUSTRE_RW_ATTR(file_heat);
1128 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1129 struct attribute *attr,
1132 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1135 return snprintf(buf, PAGE_SIZE, "%u\n",
1136 (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1139 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1140 struct attribute *attr,
1144 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1149 rc = kstrtoul(buffer, 10, &val);
1153 if (val < 0 || val > 100)
1156 sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1160 LUSTRE_RW_ATTR(heat_decay_percentage);
1162 static ssize_t heat_period_second_show(struct kobject *kobj,
1163 struct attribute *attr,
1166 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1169 return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1172 static ssize_t heat_period_second_store(struct kobject *kobj,
1173 struct attribute *attr,
1177 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1182 rc = kstrtoul(buffer, 10, &val);
1189 sbi->ll_heat_period_second = val;
1193 LUSTRE_RW_ATTR(heat_period_second);
1195 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1197 struct super_block *sb = m->private;
1198 struct ll_sb_info *sbi = ll_s2sbi(sb);
1199 struct cl_client_cache *cache = sbi->ll_cache;
1203 pages = atomic_long_read(&cache->ccc_unstable_nr);
1204 mb = (pages * PAGE_SIZE) >> 20;
1206 seq_printf(m, "unstable_check: %8d\n"
1207 "unstable_pages: %12ld\n"
1208 "unstable_mb: %8d\n",
1209 cache->ccc_unstable_check, pages, mb);
1213 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1214 const char __user *buffer,
1215 size_t count, loff_t *unused)
1217 struct seq_file *seq = file->private_data;
1218 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1225 if (count >= sizeof(kernbuf))
1228 if (copy_from_user(kernbuf, buffer, count))
1232 buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1234 rc = kstrtobool_from_user(buffer, count, &val);
1238 /* borrow lru lock to set the value */
1239 spin_lock(&sbi->ll_cache->ccc_lru_lock);
1240 sbi->ll_cache->ccc_unstable_check = val;
1241 spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1246 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1248 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1250 struct super_block *sb = m->private;
1251 struct ll_sb_info *sbi = ll_s2sbi(sb);
1252 struct root_squash_info *squash = &sbi->ll_squash;
1254 seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1258 static ssize_t ll_root_squash_seq_write(struct file *file,
1259 const char __user *buffer,
1260 size_t count, loff_t *off)
1262 struct seq_file *m = file->private_data;
1263 struct super_block *sb = m->private;
1264 struct ll_sb_info *sbi = ll_s2sbi(sb);
1265 struct root_squash_info *squash = &sbi->ll_squash;
1267 return lprocfs_wr_root_squash(buffer, count, squash, sbi->ll_fsname);
1270 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1272 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1274 struct super_block *sb = m->private;
1275 struct ll_sb_info *sbi = ll_s2sbi(sb);
1276 struct root_squash_info *squash = &sbi->ll_squash;
1279 down_read(&squash->rsi_sem);
1280 if (!list_empty(&squash->rsi_nosquash_nids)) {
1281 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1282 &squash->rsi_nosquash_nids);
1286 seq_puts(m, "NONE\n");
1288 up_read(&squash->rsi_sem);
1293 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1294 const char __user *buffer,
1295 size_t count, loff_t *off)
1297 struct seq_file *m = file->private_data;
1298 struct super_block *sb = m->private;
1299 struct ll_sb_info *sbi = ll_s2sbi(sb);
1300 struct root_squash_info *squash = &sbi->ll_squash;
1303 rc = lprocfs_wr_nosquash_nids(buffer, count, squash, sbi->ll_fsname);
1307 ll_compute_rootsquash_state(sbi);
1312 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1314 static int ll_pcc_seq_show(struct seq_file *m, void *v)
1316 struct super_block *sb = m->private;
1317 struct ll_sb_info *sbi = ll_s2sbi(sb);
1319 return pcc_super_dump(&sbi->ll_pcc_super, m);
1322 static ssize_t ll_pcc_seq_write(struct file *file, const char __user *buffer,
1323 size_t count, loff_t *off)
1325 struct seq_file *m = file->private_data;
1326 struct super_block *sb = m->private;
1327 struct ll_sb_info *sbi = ll_s2sbi(sb);
1331 if (count >= LPROCFS_WR_PCC_MAX_CMD)
1334 if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_PCC))
1337 OBD_ALLOC(kernbuf, count + 1);
1338 if (kernbuf == NULL)
1341 if (copy_from_user(kernbuf, buffer, count))
1342 GOTO(out_free_kernbuff, rc = -EFAULT);
1344 rc = pcc_cmd_handle(kernbuf, count, &sbi->ll_pcc_super);
1346 OBD_FREE(kernbuf, count + 1);
1347 return rc ? rc : count;
1349 LPROC_SEQ_FOPS(ll_pcc);
1351 struct lprocfs_vars lprocfs_llite_obd_vars[] = {
1353 .fops = &ll_site_stats_fops },
1354 { .name = "max_read_ahead_mb",
1355 .fops = &ll_max_readahead_mb_fops },
1356 { .name = "max_read_ahead_per_file_mb",
1357 .fops = &ll_max_readahead_per_file_mb_fops },
1358 { .name = "max_read_ahead_whole_mb",
1359 .fops = &ll_max_read_ahead_whole_mb_fops },
1360 { .name = "max_cached_mb",
1361 .fops = &ll_max_cached_mb_fops },
1362 { .name = "statahead_stats",
1363 .fops = &ll_statahead_stats_fops },
1364 { .name = "unstable_stats",
1365 .fops = &ll_unstable_stats_fops },
1366 { .name = "sbi_flags",
1367 .fops = &ll_sbi_flags_fops },
1368 { .name = "root_squash",
1369 .fops = &ll_root_squash_fops },
1370 { .name = "nosquash_nids",
1371 .fops = &ll_nosquash_nids_fops },
1373 .fops = &ll_pcc_fops, },
1377 #define MAX_STRING_SIZE 128
1379 static struct attribute *llite_attrs[] = {
1380 &lustre_attr_blocksize.attr,
1381 &lustre_attr_stat_blocksize.attr,
1382 &lustre_attr_kbytestotal.attr,
1383 &lustre_attr_kbytesfree.attr,
1384 &lustre_attr_kbytesavail.attr,
1385 &lustre_attr_filestotal.attr,
1386 &lustre_attr_filesfree.attr,
1387 &lustre_attr_client_type.attr,
1388 &lustre_attr_fstype.attr,
1389 &lustre_attr_uuid.attr,
1390 &lustre_attr_checksums.attr,
1391 &lustre_attr_checksum_pages.attr,
1392 &lustre_attr_stats_track_pid.attr,
1393 &lustre_attr_stats_track_ppid.attr,
1394 &lustre_attr_stats_track_gid.attr,
1395 &lustre_attr_statahead_running_max.attr,
1396 &lustre_attr_statahead_max.attr,
1397 &lustre_attr_statahead_agl.attr,
1398 &lustre_attr_lazystatfs.attr,
1399 &lustre_attr_max_easize.attr,
1400 &lustre_attr_default_easize.attr,
1401 &lustre_attr_xattr_cache.attr,
1402 &lustre_attr_fast_read.attr,
1403 &lustre_attr_tiny_write.attr,
1404 &lustre_attr_file_heat.attr,
1405 &lustre_attr_heat_decay_percentage.attr,
1406 &lustre_attr_heat_period_second.attr,
1410 static void sbi_kobj_release(struct kobject *kobj)
1412 struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1414 complete(&sbi->ll_kobj_unregister);
1417 static struct kobj_type sbi_ktype = {
1418 .default_attrs = llite_attrs,
1419 .sysfs_ops = &lustre_sysfs_ops,
1420 .release = sbi_kobj_release,
1423 static const struct llite_file_opcode {
1427 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1428 /* file operation */
1429 { LPROC_LL_DIRTY_HITS, LPROCFS_TYPE_REGS, "dirty_pages_hits" },
1430 { LPROC_LL_DIRTY_MISSES, LPROCFS_TYPE_REGS, "dirty_pages_misses" },
1431 { LPROC_LL_READ_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
1433 { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
1435 { LPROC_LL_BRW_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
1437 { LPROC_LL_BRW_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
1439 { LPROC_LL_IOCTL, LPROCFS_TYPE_REGS, "ioctl" },
1440 { LPROC_LL_OPEN, LPROCFS_TYPE_REGS, "open" },
1441 { LPROC_LL_RELEASE, LPROCFS_TYPE_REGS, "close" },
1442 { LPROC_LL_MAP, LPROCFS_TYPE_REGS, "mmap" },
1443 { LPROC_LL_FAULT, LPROCFS_TYPE_REGS, "page_fault" },
1444 { LPROC_LL_MKWRITE, LPROCFS_TYPE_REGS, "page_mkwrite" },
1445 { LPROC_LL_LLSEEK, LPROCFS_TYPE_REGS, "seek" },
1446 { LPROC_LL_FSYNC, LPROCFS_TYPE_REGS, "fsync" },
1447 { LPROC_LL_READDIR, LPROCFS_TYPE_REGS, "readdir" },
1448 /* inode operation */
1449 { LPROC_LL_SETATTR, LPROCFS_TYPE_REGS, "setattr" },
1450 { LPROC_LL_TRUNC, LPROCFS_TYPE_REGS, "truncate" },
1451 { LPROC_LL_FLOCK, LPROCFS_TYPE_REGS, "flock" },
1452 { LPROC_LL_GETATTR, LPROCFS_TYPE_REGS, "getattr" },
1453 /* dir inode operation */
1454 { LPROC_LL_CREATE, LPROCFS_TYPE_REGS, "create" },
1455 { LPROC_LL_LINK, LPROCFS_TYPE_REGS, "link" },
1456 { LPROC_LL_UNLINK, LPROCFS_TYPE_REGS, "unlink" },
1457 { LPROC_LL_SYMLINK, LPROCFS_TYPE_REGS, "symlink" },
1458 { LPROC_LL_MKDIR, LPROCFS_TYPE_REGS, "mkdir" },
1459 { LPROC_LL_RMDIR, LPROCFS_TYPE_REGS, "rmdir" },
1460 { LPROC_LL_MKNOD, LPROCFS_TYPE_REGS, "mknod" },
1461 { LPROC_LL_RENAME, LPROCFS_TYPE_REGS, "rename" },
1462 /* special inode operation */
1463 { LPROC_LL_STAFS, LPROCFS_TYPE_REGS, "statfs" },
1464 { LPROC_LL_ALLOC_INODE, LPROCFS_TYPE_REGS, "alloc_inode" },
1465 { LPROC_LL_SETXATTR, LPROCFS_TYPE_REGS, "setxattr" },
1466 { LPROC_LL_GETXATTR, LPROCFS_TYPE_REGS, "getxattr" },
1467 { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REGS, "getxattr_hits" },
1468 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_REGS, "listxattr" },
1469 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_REGS, "removexattr" },
1470 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_REGS, "inode_permission" },
1473 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
1477 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1478 lprocfs_counter_add(sbi->ll_stats, op, count);
1479 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1480 sbi->ll_stats_track_id == current->pid)
1481 lprocfs_counter_add(sbi->ll_stats, op, count);
1482 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1483 sbi->ll_stats_track_id == current->parent->pid)
1484 lprocfs_counter_add(sbi->ll_stats, op, count);
1485 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1486 sbi->ll_stats_track_id ==
1487 from_kgid(&init_user_ns, current_gid()))
1488 lprocfs_counter_add(sbi->ll_stats, op, count);
1490 EXPORT_SYMBOL(ll_stats_ops_tally);
1492 static const char *ra_stat_string[] = {
1493 [RA_STAT_HIT] = "hits",
1494 [RA_STAT_MISS] = "misses",
1495 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1496 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1497 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1498 [RA_STAT_FAILED_MATCH] = "failed lock match",
1499 [RA_STAT_DISCARDED] = "read but discarded",
1500 [RA_STAT_ZERO_LEN] = "zero length file",
1501 [RA_STAT_ZERO_WINDOW] = "zero size window",
1502 [RA_STAT_EOF] = "read-ahead to EOF",
1503 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1504 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1505 [RA_STAT_FAILED_REACH_END] = "failed to reach end"
1508 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1510 struct lustre_sb_info *lsi = s2lsi(sb);
1511 struct ll_sb_info *sbi = ll_s2sbi(sb);
1517 if (IS_ERR_OR_NULL(llite_root))
1520 sbi->ll_debugfs_entry = ldebugfs_register(name, llite_root,
1521 lprocfs_llite_obd_vars, sb);
1522 if (IS_ERR_OR_NULL(sbi->ll_debugfs_entry)) {
1523 err = sbi->ll_debugfs_entry ? PTR_ERR(sbi->ll_debugfs_entry) :
1525 sbi->ll_debugfs_entry = NULL;
1529 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "dump_page_cache",0444,
1530 &vvp_dump_pgcache_file_ops, sbi);
1532 CWARN("Error adding the dump_page_cache file\n");
1534 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "extents_stats", 0644,
1535 &ll_rw_extents_stats_fops, sbi);
1537 CWARN("Error adding the extent_stats file\n");
1539 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry,
1540 "extents_stats_per_process", 0644,
1541 &ll_rw_extents_stats_pp_fops, sbi);
1543 CWARN("Error adding the extents_stats_per_process file\n");
1545 rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "offset_stats", 0644,
1546 &ll_rw_offset_stats_fops, sbi);
1548 CWARN("Error adding the offset_stats file\n");
1550 /* File operations stats */
1551 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1552 LPROCFS_STATS_FLAG_NONE);
1553 if (sbi->ll_stats == NULL)
1554 GOTO(out_debugfs, err = -ENOMEM);
1556 /* do counter init */
1557 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1558 u32 type = llite_opcode_table[id].type;
1561 if (type & LPROCFS_TYPE_REGS)
1563 else if (type & LPROCFS_TYPE_BYTES)
1565 else if (type & LPROCFS_TYPE_PAGES)
1567 lprocfs_counter_init(sbi->ll_stats,
1568 llite_opcode_table[id].opcode,
1569 (type & LPROCFS_CNTR_AVGMINMAX),
1570 llite_opcode_table[id].opname, ptr);
1573 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "stats",
1576 GOTO(out_stats, err);
1578 sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1579 LPROCFS_STATS_FLAG_NONE);
1580 if (sbi->ll_ra_stats == NULL)
1581 GOTO(out_stats, err = -ENOMEM);
1583 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1584 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1585 ra_stat_string[id], "pages");
1587 err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "read_ahead_stats",
1590 GOTO(out_ra_stats, err);
1593 /* Yes we also register sysfs mount kset here as well */
1594 sbi->ll_kset.kobj.parent = llite_kobj;
1595 sbi->ll_kset.kobj.ktype = &sbi_ktype;
1596 init_completion(&sbi->ll_kobj_unregister);
1597 err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1599 GOTO(out_ra_stats, err);
1601 err = kset_register(&sbi->ll_kset);
1603 GOTO(out_ra_stats, err);
1605 lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1609 lprocfs_free_stats(&sbi->ll_ra_stats);
1611 lprocfs_free_stats(&sbi->ll_stats);
1613 ldebugfs_remove(&sbi->ll_debugfs_entry);
1618 void ll_debugfs_unregister_super(struct super_block *sb)
1620 struct lustre_sb_info *lsi = s2lsi(sb);
1621 struct ll_sb_info *sbi = ll_s2sbi(sb);
1623 if (!IS_ERR_OR_NULL(sbi->ll_debugfs_entry))
1624 ldebugfs_remove(&sbi->ll_debugfs_entry);
1627 sysfs_remove_link(&sbi->ll_kset.kobj,
1628 sbi->ll_dt_obd->obd_type->typ_name);
1631 sysfs_remove_link(&sbi->ll_kset.kobj,
1632 sbi->ll_md_obd->obd_type->typ_name);
1634 kobject_put(lsi->lsi_kobj);
1636 kset_unregister(&sbi->ll_kset);
1637 wait_for_completion(&sbi->ll_kobj_unregister);
1639 lprocfs_free_stats(&sbi->ll_ra_stats);
1640 lprocfs_free_stats(&sbi->ll_stats);
1642 #undef MAX_STRING_SIZE
1644 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1645 struct seq_file *seq, int which)
1647 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1648 unsigned long start, end, r, w;
1649 char *unitp = "KMGTPEZY";
1651 struct per_process_info *pp_info = &io_extents->pp_extents[which];
1657 for(i = 0; i < LL_HIST_MAX; i++) {
1658 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1659 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1662 for(i = 0; i < LL_HIST_MAX; i++) {
1663 r = pp_info->pp_r_hist.oh_buckets[i];
1664 w = pp_info->pp_w_hist.oh_buckets[i];
1667 end = BIT(i + LL_HIST_START - units);
1668 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u | "
1669 "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1670 (i == LL_HIST_MAX - 1) ? '+' : ' ',
1671 r, pct(r, read_tot), pct(read_cum, read_tot),
1672 w, pct(w, write_tot), pct(write_cum, write_tot));
1674 if (start == BIT(10)) {
1679 if (read_cum == read_tot && write_cum == write_tot)
1684 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1686 struct timespec64 now;
1687 struct ll_sb_info *sbi = seq->private;
1688 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1691 ktime_get_real_ts64(&now);
1693 if (!sbi->ll_rw_stats_on) {
1694 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1697 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1698 (s64)now.tv_sec, now.tv_nsec);
1699 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1700 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1701 "extents", "calls", "%", "cum%",
1702 "calls", "%", "cum%");
1703 spin_lock(&sbi->ll_pp_extent_lock);
1704 for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1705 if (io_extents->pp_extents[k].pid != 0) {
1706 seq_printf(seq, "\nPID: %d\n",
1707 io_extents->pp_extents[k].pid);
1708 ll_display_extents_info(io_extents, seq, k);
1711 spin_unlock(&sbi->ll_pp_extent_lock);
1715 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1716 const char __user *buf,
1720 struct seq_file *seq = file->private_data;
1721 struct ll_sb_info *sbi = seq->private;
1722 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1729 value = ll_stats_pid_write(buf, len);
1732 sbi->ll_rw_stats_on = 0;
1734 sbi->ll_rw_stats_on = 1;
1736 spin_lock(&sbi->ll_pp_extent_lock);
1737 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1738 io_extents->pp_extents[i].pid = 0;
1739 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1740 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1742 spin_unlock(&sbi->ll_pp_extent_lock);
1746 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1748 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1750 struct timespec64 now;
1751 struct ll_sb_info *sbi = seq->private;
1752 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1754 ktime_get_real_ts64(&now);
1756 if (!sbi->ll_rw_stats_on) {
1757 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1760 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1761 (s64)now.tv_sec, now.tv_nsec);
1763 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1764 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1765 "extents", "calls", "%", "cum%",
1766 "calls", "%", "cum%");
1767 spin_lock(&sbi->ll_lock);
1768 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1769 spin_unlock(&sbi->ll_lock);
1774 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1775 const char __user *buf,
1776 size_t len, loff_t *off)
1778 struct seq_file *seq = file->private_data;
1779 struct ll_sb_info *sbi = seq->private;
1780 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1787 value = ll_stats_pid_write(buf, len);
1790 sbi->ll_rw_stats_on = 0;
1792 sbi->ll_rw_stats_on = 1;
1794 spin_lock(&sbi->ll_pp_extent_lock);
1795 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1796 io_extents->pp_extents[i].pid = 0;
1797 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1798 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1800 spin_unlock(&sbi->ll_pp_extent_lock);
1805 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1807 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1808 struct ll_file_data *file, loff_t pos,
1809 size_t count, int rw)
1812 struct ll_rw_process_info *process;
1813 struct ll_rw_process_info *offset;
1814 int *off_count = &sbi->ll_rw_offset_entry_count;
1815 int *process_count = &sbi->ll_offset_process_count;
1816 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1818 if(!sbi->ll_rw_stats_on)
1820 process = sbi->ll_rw_process_info;
1821 offset = sbi->ll_rw_offset_info;
1823 spin_lock(&sbi->ll_pp_extent_lock);
1824 /* Extent statistics */
1825 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1826 if(io_extents->pp_extents[i].pid == pid) {
1834 sbi->ll_extent_process_count =
1835 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1836 cur = sbi->ll_extent_process_count;
1837 io_extents->pp_extents[cur].pid = pid;
1838 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1839 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1842 for(i = 0; (count >= BIT(LL_HIST_START << i)) &&
1843 (i < (LL_HIST_MAX - 1)); i++);
1845 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1846 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1848 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1849 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1851 spin_unlock(&sbi->ll_pp_extent_lock);
1853 spin_lock(&sbi->ll_process_lock);
1854 /* Offset statistics */
1855 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1856 if (process[i].rw_pid == pid) {
1857 if (process[i].rw_last_file != file) {
1858 process[i].rw_range_start = pos;
1859 process[i].rw_last_file_pos = pos + count;
1860 process[i].rw_smallest_extent = count;
1861 process[i].rw_largest_extent = count;
1862 process[i].rw_offset = 0;
1863 process[i].rw_last_file = file;
1864 spin_unlock(&sbi->ll_process_lock);
1867 if (process[i].rw_last_file_pos != pos) {
1869 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1870 offset[*off_count].rw_op = process[i].rw_op;
1871 offset[*off_count].rw_pid = pid;
1872 offset[*off_count].rw_range_start =
1873 process[i].rw_range_start;
1874 offset[*off_count].rw_range_end =
1875 process[i].rw_last_file_pos;
1876 offset[*off_count].rw_smallest_extent =
1877 process[i].rw_smallest_extent;
1878 offset[*off_count].rw_largest_extent =
1879 process[i].rw_largest_extent;
1880 offset[*off_count].rw_offset =
1881 process[i].rw_offset;
1882 process[i].rw_op = rw;
1883 process[i].rw_range_start = pos;
1884 process[i].rw_smallest_extent = count;
1885 process[i].rw_largest_extent = count;
1886 process[i].rw_offset = pos -
1887 process[i].rw_last_file_pos;
1889 if(process[i].rw_smallest_extent > count)
1890 process[i].rw_smallest_extent = count;
1891 if(process[i].rw_largest_extent < count)
1892 process[i].rw_largest_extent = count;
1893 process[i].rw_last_file_pos = pos + count;
1894 spin_unlock(&sbi->ll_process_lock);
1898 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1899 process[*process_count].rw_pid = pid;
1900 process[*process_count].rw_op = rw;
1901 process[*process_count].rw_range_start = pos;
1902 process[*process_count].rw_last_file_pos = pos + count;
1903 process[*process_count].rw_smallest_extent = count;
1904 process[*process_count].rw_largest_extent = count;
1905 process[*process_count].rw_offset = 0;
1906 process[*process_count].rw_last_file = file;
1907 spin_unlock(&sbi->ll_process_lock);
1910 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1912 struct timespec64 now;
1913 struct ll_sb_info *sbi = seq->private;
1914 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1915 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1918 ktime_get_real_ts64(&now);
1920 if (!sbi->ll_rw_stats_on) {
1921 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1924 spin_lock(&sbi->ll_process_lock);
1926 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
1927 (s64)now.tv_sec, now.tv_nsec);
1928 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1929 "R/W", "PID", "RANGE START", "RANGE END",
1930 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1932 /* We stored the discontiguous offsets here; print them first */
1933 for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1934 if (offset[i].rw_pid != 0)
1936 "%3c %10d %14llu %14llu %17lu %17lu %14llu\n",
1937 offset[i].rw_op == READ ? 'R' : 'W',
1939 offset[i].rw_range_start,
1940 offset[i].rw_range_end,
1941 (unsigned long)offset[i].rw_smallest_extent,
1942 (unsigned long)offset[i].rw_largest_extent,
1943 offset[i].rw_offset);
1946 /* Then print the current offsets for each process */
1947 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1948 if (process[i].rw_pid != 0)
1950 "%3c %10d %14llu %14llu %17lu %17lu %14llu\n",
1951 process[i].rw_op == READ ? 'R' : 'W',
1953 process[i].rw_range_start,
1954 process[i].rw_last_file_pos,
1955 (unsigned long)process[i].rw_smallest_extent,
1956 (unsigned long)process[i].rw_largest_extent,
1957 process[i].rw_offset);
1959 spin_unlock(&sbi->ll_process_lock);
1964 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
1965 const char __user *buf,
1966 size_t len, loff_t *off)
1968 struct seq_file *seq = file->private_data;
1969 struct ll_sb_info *sbi = seq->private;
1970 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1971 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1977 value = ll_stats_pid_write(buf, len);
1980 sbi->ll_rw_stats_on = 0;
1982 sbi->ll_rw_stats_on = 1;
1984 spin_lock(&sbi->ll_process_lock);
1985 sbi->ll_offset_process_count = 0;
1986 sbi->ll_rw_offset_entry_count = 0;
1987 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1988 LL_PROCESS_HIST_MAX);
1989 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1990 LL_OFFSET_HIST_MAX);
1991 spin_unlock(&sbi->ll_process_lock);
1996 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);