1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 #define DEBUG_SUBSYSTEM S_LLITE
38 #include <linux/version.h>
39 #include <lustre_lite.h>
40 #include <lprocfs_status.h>
41 #include <linux/seq_file.h>
42 #include <obd_support.h>
44 #include "llite_internal.h"
46 struct proc_dir_entry *proc_lustre_fs_root;
49 /* /proc/lustre/llite mount point registration */
50 extern struct file_operations vvp_dump_pgcache_file_ops;
51 struct file_operations ll_rw_extents_stats_fops;
52 struct file_operations ll_rw_extents_stats_pp_fops;
53 struct file_operations ll_rw_offset_stats_fops;
55 static int ll_rd_blksize(char *page, char **start, off_t off, int count,
58 struct super_block *sb = (struct super_block *)data;
59 struct obd_statfs osfs;
63 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
67 rc = snprintf(page, count, "%u\n", osfs.os_bsize);
73 static int ll_rd_kbytestotal(char *page, char **start, off_t off, int count,
76 struct super_block *sb = (struct super_block *)data;
77 struct obd_statfs osfs;
81 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
84 __u32 blk_size = osfs.os_bsize >> 10;
85 __u64 result = osfs.os_blocks;
87 while (blk_size >>= 1)
91 rc = snprintf(page, count, LPU64"\n", result);
97 static int ll_rd_kbytesfree(char *page, char **start, off_t off, int count,
100 struct super_block *sb = (struct super_block *)data;
101 struct obd_statfs osfs;
105 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
108 __u32 blk_size = osfs.os_bsize >> 10;
109 __u64 result = osfs.os_bfree;
111 while (blk_size >>= 1)
115 rc = snprintf(page, count, LPU64"\n", result);
120 static int ll_rd_kbytesavail(char *page, char **start, off_t off, int count,
121 int *eof, void *data)
123 struct super_block *sb = (struct super_block *)data;
124 struct obd_statfs osfs;
128 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
131 __u32 blk_size = osfs.os_bsize >> 10;
132 __u64 result = osfs.os_bavail;
134 while (blk_size >>= 1)
138 rc = snprintf(page, count, LPU64"\n", result);
143 static int ll_rd_filestotal(char *page, char **start, off_t off, int count,
144 int *eof, void *data)
146 struct super_block *sb = (struct super_block *)data;
147 struct obd_statfs osfs;
151 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
155 rc = snprintf(page, count, LPU64"\n", osfs.os_files);
160 static int ll_rd_filesfree(char *page, char **start, off_t off, int count,
161 int *eof, void *data)
163 struct super_block *sb = (struct super_block *)data;
164 struct obd_statfs osfs;
168 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
172 rc = snprintf(page, count, LPU64"\n", osfs.os_ffree);
178 static int ll_rd_client_type(char *page, char **start, off_t off, int count,
179 int *eof, void *data)
181 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)data);
184 LASSERT(sbi != NULL);
187 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
188 rc = snprintf(page, count, "remote client\n");
190 rc = snprintf(page, count, "local client\n");
195 static int ll_rd_fstype(char *page, char **start, off_t off, int count,
196 int *eof, void *data)
198 struct super_block *sb = (struct super_block*)data;
202 return snprintf(page, count, "%s\n", sb->s_type->name);
205 static int ll_rd_sb_uuid(char *page, char **start, off_t off, int count,
206 int *eof, void *data)
208 struct super_block *sb = (struct super_block *)data;
212 return snprintf(page, count, "%s\n", ll_s2sbi(sb)->ll_sb_uuid.uuid);
215 static int ll_rd_site_stats(char *page, char **start, off_t off,
216 int count, int *eof, void *data)
218 struct super_block *sb = data;
221 * See description of statistical counters in struct cl_site, and
224 return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site),
228 static int ll_rd_max_readahead_mb(char *page, char **start, off_t off,
229 int count, int *eof, void *data)
231 struct super_block *sb = data;
232 struct ll_sb_info *sbi = ll_s2sbi(sb);
236 spin_lock(&sbi->ll_lock);
237 pages_number = sbi->ll_ra_info.ra_max_pages;
238 spin_unlock(&sbi->ll_lock);
240 mult = 1 << (20 - PAGE_CACHE_SHIFT);
241 return lprocfs_read_frac_helper(page, count, pages_number, mult);
244 static int ll_wr_max_readahead_mb(struct file *file, const char *buffer,
245 unsigned long count, void *data)
247 struct super_block *sb = data;
248 struct ll_sb_info *sbi = ll_s2sbi(sb);
249 int mult, rc, pages_number;
251 mult = 1 << (20 - CFS_PAGE_SHIFT);
252 rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
256 if (pages_number < 0 || pages_number > num_physpages / 2) {
257 CERROR("can't set file readahead more than %lu MB\n",
258 num_physpages >> (20 - CFS_PAGE_SHIFT + 1)); /*1/2 of RAM*/
262 spin_lock(&sbi->ll_lock);
263 sbi->ll_ra_info.ra_max_pages = pages_number;
264 spin_unlock(&sbi->ll_lock);
269 static int ll_rd_max_read_ahead_whole_mb(char *page, char **start, off_t off,
270 int count, int *eof, void *data)
272 struct super_block *sb = data;
273 struct ll_sb_info *sbi = ll_s2sbi(sb);
277 spin_lock(&sbi->ll_lock);
278 pages_number = sbi->ll_ra_info.ra_max_read_ahead_whole_pages;
279 spin_unlock(&sbi->ll_lock);
281 mult = 1 << (20 - CFS_PAGE_SHIFT);
282 return lprocfs_read_frac_helper(page, count, pages_number, mult);
285 static int ll_wr_max_read_ahead_whole_mb(struct file *file, const char *buffer,
286 unsigned long count, void *data)
288 struct super_block *sb = data;
289 struct ll_sb_info *sbi = ll_s2sbi(sb);
290 int mult, rc, pages_number;
292 mult = 1 << (20 - CFS_PAGE_SHIFT);
293 rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
297 /* Cap this at the current max readahead window size, the readahead
298 * algorithm does this anyway so it's pointless to set it larger. */
299 if (pages_number < 0 || pages_number > sbi->ll_ra_info.ra_max_pages) {
300 CERROR("can't set max_read_ahead_whole_mb more than "
301 "max_read_ahead_mb: %lu\n",
302 sbi->ll_ra_info.ra_max_pages >> (20 - CFS_PAGE_SHIFT));
306 spin_lock(&sbi->ll_lock);
307 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
308 spin_unlock(&sbi->ll_lock);
313 static int ll_rd_max_cached_mb(char *page, char **start, off_t off,
314 int count, int *eof, void *data)
316 struct super_block *sb = data;
317 struct ll_sb_info *sbi = ll_s2sbi(sb);
321 spin_lock(&sbi->ll_lock);
322 pages_number = sbi->ll_async_page_max;
323 spin_unlock(&sbi->ll_lock);
325 mult = 1 << (20 - CFS_PAGE_SHIFT);
326 return lprocfs_read_frac_helper(page, count, pages_number, mult);;
329 static int ll_wr_max_cached_mb(struct file *file, const char *buffer,
330 unsigned long count, void *data)
332 struct super_block *sb = data;
333 struct ll_sb_info *sbi = ll_s2sbi(sb);
334 int mult, rc, pages_number;
336 mult = 1 << (20 - CFS_PAGE_SHIFT);
337 rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
341 if (pages_number < 0 || pages_number > num_physpages) {
342 CERROR("can't set max cache more than %lu MB\n",
343 num_physpages >> (20 - CFS_PAGE_SHIFT));
347 spin_lock(&sbi->ll_lock);
348 sbi->ll_async_page_max = pages_number ;
349 spin_unlock(&sbi->ll_lock);
352 /* Not set up yet, don't call llap_shrink_cache */
358 static int ll_rd_checksum(char *page, char **start, off_t off,
359 int count, int *eof, void *data)
361 struct super_block *sb = data;
362 struct ll_sb_info *sbi = ll_s2sbi(sb);
364 return snprintf(page, count, "%u\n",
365 (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
368 static int ll_wr_checksum(struct file *file, const char *buffer,
369 unsigned long count, void *data)
371 struct super_block *sb = data;
372 struct ll_sb_info *sbi = ll_s2sbi(sb);
379 rc = lprocfs_write_helper(buffer, count, &val);
383 sbi->ll_flags |= LL_SBI_CHECKSUM;
385 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
387 rc = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
388 KEY_CHECKSUM, sizeof(val), &val, NULL);
390 CWARN("Failed to set OSC checksum flags: %d\n", rc);
395 static int ll_rd_max_rw_chunk(char *page, char **start, off_t off,
396 int count, int *eof, void *data)
398 struct super_block *sb = data;
400 return snprintf(page, count, "%lu\n", ll_s2sbi(sb)->ll_max_rw_chunk);
403 static int ll_wr_max_rw_chunk(struct file *file, const char *buffer,
404 unsigned long count, void *data)
406 struct super_block *sb = data;
409 rc = lprocfs_write_helper(buffer, count, &val);
412 ll_s2sbi(sb)->ll_max_rw_chunk = val;
416 static int ll_rd_track_id(char *page, int count, void *data,
417 enum stats_track_type type)
419 struct super_block *sb = data;
421 if (ll_s2sbi(sb)->ll_stats_track_type == type) {
422 return snprintf(page, count, "%d\n",
423 ll_s2sbi(sb)->ll_stats_track_id);
425 } else if (ll_s2sbi(sb)->ll_stats_track_type == STATS_TRACK_ALL) {
426 return snprintf(page, count, "0 (all)\n");
428 return snprintf(page, count, "untracked\n");
432 static int ll_wr_track_id(const char *buffer, unsigned long count, void *data,
433 enum stats_track_type type)
435 struct super_block *sb = data;
438 rc = lprocfs_write_helper(buffer, count, &pid);
441 ll_s2sbi(sb)->ll_stats_track_id = pid;
443 ll_s2sbi(sb)->ll_stats_track_type = STATS_TRACK_ALL;
445 ll_s2sbi(sb)->ll_stats_track_type = type;
446 lprocfs_clear_stats(ll_s2sbi(sb)->ll_stats);
450 static int ll_rd_track_pid(char *page, char **start, off_t off,
451 int count, int *eof, void *data)
453 return (ll_rd_track_id(page, count, data, STATS_TRACK_PID));
456 static int ll_wr_track_pid(struct file *file, const char *buffer,
457 unsigned long count, void *data)
459 return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PID));
462 static int ll_rd_track_ppid(char *page, char **start, off_t off,
463 int count, int *eof, void *data)
465 return (ll_rd_track_id(page, count, data, STATS_TRACK_PPID));
468 static int ll_wr_track_ppid(struct file *file, const char *buffer,
469 unsigned long count, void *data)
471 return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PPID));
474 static int ll_rd_track_gid(char *page, char **start, off_t off,
475 int count, int *eof, void *data)
477 return (ll_rd_track_id(page, count, data, STATS_TRACK_GID));
480 static int ll_wr_track_gid(struct file *file, const char *buffer,
481 unsigned long count, void *data)
483 return (ll_wr_track_id(buffer, count, data, STATS_TRACK_GID));
486 static int ll_rd_statahead_max(char *page, char **start, off_t off,
487 int count, int *eof, void *data)
489 struct super_block *sb = data;
490 struct ll_sb_info *sbi = ll_s2sbi(sb);
492 return snprintf(page, count, "%u\n", sbi->ll_sa_max);
495 static int ll_wr_statahead_max(struct file *file, const char *buffer,
496 unsigned long count, void *data)
498 struct super_block *sb = data;
499 struct ll_sb_info *sbi = ll_s2sbi(sb);
502 rc = lprocfs_write_helper(buffer, count, &val);
506 if (val >= 0 && val <= LL_SA_RPC_MAX)
507 sbi->ll_sa_max = val;
509 CERROR("Bad statahead_max value %d. Valid values are in the "
510 "range [0, %d]\n", val, LL_SA_RPC_MAX);
515 static int ll_rd_statahead_stats(char *page, char **start, off_t off,
516 int count, int *eof, void *data)
518 struct super_block *sb = data;
519 struct ll_sb_info *sbi = ll_s2sbi(sb);
521 return snprintf(page, count,
522 "statahead wrong: %u\n"
523 "statahead total: %u\n"
527 "miss count: %llu\n",
536 static struct lprocfs_vars lprocfs_llite_obd_vars[] = {
537 { "uuid", ll_rd_sb_uuid, 0, 0 },
538 //{ "mntpt_path", ll_rd_path, 0, 0 },
539 { "fstype", ll_rd_fstype, 0, 0 },
540 { "site", ll_rd_site_stats, 0, 0 },
541 { "blocksize", ll_rd_blksize, 0, 0 },
542 { "kbytestotal", ll_rd_kbytestotal, 0, 0 },
543 { "kbytesfree", ll_rd_kbytesfree, 0, 0 },
544 { "kbytesavail", ll_rd_kbytesavail, 0, 0 },
545 { "filestotal", ll_rd_filestotal, 0, 0 },
546 { "filesfree", ll_rd_filesfree, 0, 0 },
547 { "client_type", ll_rd_client_type, 0, 0 },
548 //{ "filegroups", lprocfs_rd_filegroups, 0, 0 },
549 { "max_read_ahead_mb", ll_rd_max_readahead_mb,
550 ll_wr_max_readahead_mb, 0 },
551 { "max_read_ahead_whole_mb", ll_rd_max_read_ahead_whole_mb,
552 ll_wr_max_read_ahead_whole_mb, 0 },
553 { "max_cached_mb", ll_rd_max_cached_mb, ll_wr_max_cached_mb, 0 },
554 { "checksum_pages", ll_rd_checksum, ll_wr_checksum, 0 },
555 { "max_rw_chunk", ll_rd_max_rw_chunk, ll_wr_max_rw_chunk, 0 },
556 { "stats_track_pid", ll_rd_track_pid, ll_wr_track_pid, 0 },
557 { "stats_track_ppid", ll_rd_track_ppid, ll_wr_track_ppid, 0 },
558 { "stats_track_gid", ll_rd_track_gid, ll_wr_track_gid, 0 },
559 { "statahead_max", ll_rd_statahead_max, ll_wr_statahead_max, 0 },
560 { "statahead_stats", ll_rd_statahead_stats, 0, 0 },
564 #define MAX_STRING_SIZE 128
566 struct llite_file_opcode {
570 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
572 { LPROC_LL_DIRTY_HITS, LPROCFS_TYPE_REGS, "dirty_pages_hits" },
573 { LPROC_LL_DIRTY_MISSES, LPROCFS_TYPE_REGS, "dirty_pages_misses" },
574 { LPROC_LL_WB_WRITEPAGE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
575 "writeback_from_writepage" },
576 { LPROC_LL_WB_PRESSURE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
577 "writeback_from_pressure" },
578 { LPROC_LL_WB_OK, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
579 "writeback_ok_pages" },
580 { LPROC_LL_WB_FAIL, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
581 "writeback_failed_pages" },
582 { LPROC_LL_READ_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
584 { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
586 { LPROC_LL_BRW_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
588 { LPROC_LL_BRW_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
591 { LPROC_LL_IOCTL, LPROCFS_TYPE_REGS, "ioctl" },
592 { LPROC_LL_OPEN, LPROCFS_TYPE_REGS, "open" },
593 { LPROC_LL_RELEASE, LPROCFS_TYPE_REGS, "close" },
594 { LPROC_LL_MAP, LPROCFS_TYPE_REGS, "mmap" },
595 { LPROC_LL_LLSEEK, LPROCFS_TYPE_REGS, "seek" },
596 { LPROC_LL_FSYNC, LPROCFS_TYPE_REGS, "fsync" },
597 /* inode operation */
598 { LPROC_LL_SETATTR, LPROCFS_TYPE_REGS, "setattr" },
599 { LPROC_LL_TRUNC, LPROCFS_TYPE_REGS, "truncate" },
600 { LPROC_LL_LOCKLESS_TRUNC, LPROCFS_TYPE_REGS, "lockless_truncate"},
601 { LPROC_LL_FLOCK, LPROCFS_TYPE_REGS, "flock" },
602 { LPROC_LL_GETATTR, LPROCFS_TYPE_REGS, "getattr" },
603 /* special inode operation */
604 { LPROC_LL_STAFS, LPROCFS_TYPE_REGS, "statfs" },
605 { LPROC_LL_ALLOC_INODE, LPROCFS_TYPE_REGS, "alloc_inode" },
606 { LPROC_LL_SETXATTR, LPROCFS_TYPE_REGS, "setxattr" },
607 { LPROC_LL_GETXATTR, LPROCFS_TYPE_REGS, "getxattr" },
608 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_REGS, "listxattr" },
609 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_REGS, "removexattr" },
610 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_REGS, "inode_permission" },
611 { LPROC_LL_DIRECT_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
613 { LPROC_LL_DIRECT_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
615 { LPROC_LL_LOCKLESS_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
616 "lockless_read_bytes" },
617 { LPROC_LL_LOCKLESS_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
618 "lockless_write_bytes" },
622 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
626 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
627 lprocfs_counter_add(sbi->ll_stats, op, count);
628 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
629 sbi->ll_stats_track_id == current->pid)
630 lprocfs_counter_add(sbi->ll_stats, op, count);
631 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
632 sbi->ll_stats_track_id == current->p_pptr->pid)
633 lprocfs_counter_add(sbi->ll_stats, op, count);
634 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
635 sbi->ll_stats_track_id == current->gid)
636 lprocfs_counter_add(sbi->ll_stats, op, count);
638 EXPORT_SYMBOL(ll_stats_ops_tally);
640 static const char *ra_stat_string[] = {
641 [RA_STAT_HIT] = "hits",
642 [RA_STAT_MISS] = "misses",
643 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
644 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
645 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
646 [RA_STAT_FAILED_MATCH] = "failed lock match",
647 [RA_STAT_DISCARDED] = "read but discarded",
648 [RA_STAT_ZERO_LEN] = "zero length file",
649 [RA_STAT_ZERO_WINDOW] = "zero size window",
650 [RA_STAT_EOF] = "read-ahead to EOF",
651 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
652 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
656 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
657 struct super_block *sb, char *osc, char *mdc)
659 struct lprocfs_vars lvars[2];
660 struct lustre_sb_info *lsi = s2lsi(sb);
661 struct ll_sb_info *sbi = ll_s2sbi(sb);
662 struct obd_device *obd;
663 char name[MAX_STRING_SIZE + 1], *ptr;
664 int err, id, len, rc;
667 memset(lvars, 0, sizeof(lvars));
669 name[MAX_STRING_SIZE] = '\0';
670 lvars[0].name = name;
672 LASSERT(sbi != NULL);
673 LASSERT(mdc != NULL);
674 LASSERT(osc != NULL);
677 len = strlen(lsi->lsi_lmd->lmd_profile);
678 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
679 if (ptr && (strcmp(ptr, "-client") == 0))
683 snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
684 lsi->lsi_lmd->lmd_profile, sb);
686 sbi->ll_proc_root = lprocfs_register(name, parent, NULL, NULL);
687 if (IS_ERR(sbi->ll_proc_root)) {
688 err = PTR_ERR(sbi->ll_proc_root);
689 sbi->ll_proc_root = NULL;
693 rc = lprocfs_seq_create(sbi->ll_proc_root, "dump_page_cache", 0444,
694 &vvp_dump_pgcache_file_ops, sbi);
696 CWARN("Error adding the dump_page_cache file\n");
698 rc = lprocfs_seq_create(sbi->ll_proc_root, "extents_stats", 0644,
699 &ll_rw_extents_stats_fops, sbi);
701 CWARN("Error adding the extent_stats file\n");
703 rc = lprocfs_seq_create(sbi->ll_proc_root, "extents_stats_per_process",
704 0644, &ll_rw_extents_stats_pp_fops, sbi);
706 CWARN("Error adding the extents_stats_per_process file\n");
708 rc = lprocfs_seq_create(sbi->ll_proc_root, "offset_stats", 0644,
709 &ll_rw_offset_stats_fops, sbi);
711 CWARN("Error adding the offset_stats file\n");
713 /* File operations stats */
714 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
715 LPROCFS_STATS_FLAG_PERCPU);
716 if (sbi->ll_stats == NULL)
717 GOTO(out, err = -ENOMEM);
718 /* do counter init */
719 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
720 __u32 type = llite_opcode_table[id].type;
722 if (type & LPROCFS_TYPE_REGS)
724 else if (type & LPROCFS_TYPE_BYTES)
726 else if (type & LPROCFS_TYPE_PAGES)
728 lprocfs_counter_init(sbi->ll_stats,
729 llite_opcode_table[id].opcode,
730 (type & LPROCFS_CNTR_AVGMINMAX),
731 llite_opcode_table[id].opname, ptr);
733 err = lprocfs_register_stats(sbi->ll_proc_root, "stats", sbi->ll_stats);
737 sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
738 LPROCFS_STATS_FLAG_PERCPU);
739 if (sbi->ll_ra_stats == NULL)
740 GOTO(out, err = -ENOMEM);
742 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
743 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
744 ra_stat_string[id], "pages");
745 err = lprocfs_register_stats(sbi->ll_proc_root, "read_ahead_stats",
751 err = lprocfs_add_vars(sbi->ll_proc_root, lprocfs_llite_obd_vars, sb);
756 obd = class_name2obd(mdc);
758 LASSERT(obd != NULL);
759 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
760 LASSERT(obd->obd_type->typ_name != NULL);
762 snprintf(name, MAX_STRING_SIZE, "%s/common_name",
763 obd->obd_type->typ_name);
764 lvars[0].read_fptr = lprocfs_rd_name;
765 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
769 snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
770 lvars[0].read_fptr = lprocfs_rd_uuid;
771 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
776 obd = class_name2obd(osc);
778 LASSERT(obd != NULL);
779 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
780 LASSERT(obd->obd_type->typ_name != NULL);
782 snprintf(name, MAX_STRING_SIZE, "%s/common_name",
783 obd->obd_type->typ_name);
784 lvars[0].read_fptr = lprocfs_rd_name;
785 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
789 snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
790 lvars[0].read_fptr = lprocfs_rd_uuid;
791 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
794 lprocfs_remove(&sbi->ll_proc_root);
795 lprocfs_free_stats(&sbi->ll_ra_stats);
796 lprocfs_free_stats(&sbi->ll_stats);
801 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi)
803 if (sbi->ll_proc_root) {
804 lprocfs_remove(&sbi->ll_proc_root);
805 lprocfs_free_stats(&sbi->ll_ra_stats);
806 lprocfs_free_stats(&sbi->ll_stats);
809 #undef MAX_STRING_SIZE
811 #define pct(a,b) (b ? a * 100 / b : 0)
813 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
814 struct seq_file *seq, int which)
816 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
817 unsigned long start, end, r, w;
818 char *unitp = "KMGTPEZY";
820 struct per_process_info *pp_info = &io_extents->pp_extents[which];
826 for(i = 0; i < LL_HIST_MAX; i++) {
827 read_tot += pp_info->pp_r_hist.oh_buckets[i];
828 write_tot += pp_info->pp_w_hist.oh_buckets[i];
831 for(i = 0; i < LL_HIST_MAX; i++) {
832 r = pp_info->pp_r_hist.oh_buckets[i];
833 w = pp_info->pp_w_hist.oh_buckets[i];
836 end = 1 << (i + LL_HIST_START - units);
837 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4lu %4lu | "
838 "%14lu %4lu %4lu\n", start, *unitp, end, *unitp,
839 (i == LL_HIST_MAX - 1) ? '+' : ' ',
840 r, pct(r, read_tot), pct(read_cum, read_tot),
841 w, pct(w, write_tot), pct(write_cum, write_tot));
843 if (start == 1<<10) {
848 if (read_cum == read_tot && write_cum == write_tot)
853 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
856 struct ll_sb_info *sbi = seq->private;
857 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
860 do_gettimeofday(&now);
862 if (!sbi->ll_rw_stats_on) {
863 seq_printf(seq, "disabled\n"
864 "write anything in this file to activate, "
865 "then 0 or \"[D/d]isabled\" to deactivate\n");
868 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
869 now.tv_sec, now.tv_usec);
870 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
871 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
872 "extents", "calls", "%", "cum%",
873 "calls", "%", "cum%");
874 spin_lock(&sbi->ll_pp_extent_lock);
875 for(k = 0; k < LL_PROCESS_HIST_MAX; k++) {
876 if(io_extents->pp_extents[k].pid != 0) {
877 seq_printf(seq, "\nPID: %d\n",
878 io_extents->pp_extents[k].pid);
879 ll_display_extents_info(io_extents, seq, k);
882 spin_unlock(&sbi->ll_pp_extent_lock);
886 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
887 const char *buf, size_t len,
890 struct seq_file *seq = file->private_data;
891 struct ll_sb_info *sbi = seq->private;
892 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
894 int value = 1, rc = 0;
896 rc = lprocfs_write_helper(buf, len, &value);
897 if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
898 strcmp(buf, "Disabled") == 0))
902 sbi->ll_rw_stats_on = 0;
904 sbi->ll_rw_stats_on = 1;
906 spin_lock(&sbi->ll_pp_extent_lock);
907 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
908 io_extents->pp_extents[i].pid = 0;
909 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
910 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
912 spin_unlock(&sbi->ll_pp_extent_lock);
916 LPROC_SEQ_FOPS(ll_rw_extents_stats_pp);
918 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
921 struct ll_sb_info *sbi = seq->private;
922 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
924 do_gettimeofday(&now);
926 if (!sbi->ll_rw_stats_on) {
927 seq_printf(seq, "disabled\n"
928 "write anything in this file to activate, "
929 "then 0 or \"[D/d]isabled\" to deactivate\n");
932 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
933 now.tv_sec, now.tv_usec);
935 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
936 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
937 "extents", "calls", "%", "cum%",
938 "calls", "%", "cum%");
939 spin_lock(&sbi->ll_lock);
940 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
941 spin_unlock(&sbi->ll_lock);
946 static ssize_t ll_rw_extents_stats_seq_write(struct file *file, const char *buf,
947 size_t len, loff_t *off)
949 struct seq_file *seq = file->private_data;
950 struct ll_sb_info *sbi = seq->private;
951 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
953 int value = 1, rc = 0;
955 rc = lprocfs_write_helper(buf, len, &value);
956 if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
957 strcmp(buf, "Disabled") == 0))
961 sbi->ll_rw_stats_on = 0;
963 sbi->ll_rw_stats_on = 1;
964 spin_lock(&sbi->ll_pp_extent_lock);
965 for(i = 0; i <= LL_PROCESS_HIST_MAX; i++)
967 io_extents->pp_extents[i].pid = 0;
968 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
969 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
971 spin_unlock(&sbi->ll_pp_extent_lock);
976 LPROC_SEQ_FOPS(ll_rw_extents_stats);
978 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
979 struct ll_file_data *file, loff_t pos,
980 size_t count, int rw)
983 struct ll_rw_process_info *process;
984 struct ll_rw_process_info *offset;
985 int *off_count = &sbi->ll_rw_offset_entry_count;
986 int *process_count = &sbi->ll_offset_process_count;
987 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
989 if(!sbi->ll_rw_stats_on)
991 process = sbi->ll_rw_process_info;
992 offset = sbi->ll_rw_offset_info;
994 spin_lock(&sbi->ll_pp_extent_lock);
995 /* Extent statistics */
996 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
997 if(io_extents->pp_extents[i].pid == pid) {
1005 sbi->ll_extent_process_count =
1006 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1007 cur = sbi->ll_extent_process_count;
1008 io_extents->pp_extents[cur].pid = pid;
1009 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1010 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1013 for(i = 0; (count >= (1 << LL_HIST_START << i)) &&
1014 (i < (LL_HIST_MAX - 1)); i++);
1016 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1017 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1019 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1020 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1022 spin_unlock(&sbi->ll_pp_extent_lock);
1024 spin_lock(&sbi->ll_process_lock);
1025 /* Offset statistics */
1026 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1027 if (process[i].rw_pid == pid) {
1028 if (process[i].rw_last_file != file) {
1029 process[i].rw_range_start = pos;
1030 process[i].rw_last_file_pos = pos + count;
1031 process[i].rw_smallest_extent = count;
1032 process[i].rw_largest_extent = count;
1033 process[i].rw_offset = 0;
1034 process[i].rw_last_file = file;
1035 spin_unlock(&sbi->ll_process_lock);
1038 if (process[i].rw_last_file_pos != pos) {
1040 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1041 offset[*off_count].rw_op = process[i].rw_op;
1042 offset[*off_count].rw_pid = pid;
1043 offset[*off_count].rw_range_start =
1044 process[i].rw_range_start;
1045 offset[*off_count].rw_range_end =
1046 process[i].rw_last_file_pos;
1047 offset[*off_count].rw_smallest_extent =
1048 process[i].rw_smallest_extent;
1049 offset[*off_count].rw_largest_extent =
1050 process[i].rw_largest_extent;
1051 offset[*off_count].rw_offset =
1052 process[i].rw_offset;
1053 process[i].rw_op = rw;
1054 process[i].rw_range_start = pos;
1055 process[i].rw_smallest_extent = count;
1056 process[i].rw_largest_extent = count;
1057 process[i].rw_offset = pos -
1058 process[i].rw_last_file_pos;
1060 if(process[i].rw_smallest_extent > count)
1061 process[i].rw_smallest_extent = count;
1062 if(process[i].rw_largest_extent < count)
1063 process[i].rw_largest_extent = count;
1064 process[i].rw_last_file_pos = pos + count;
1065 spin_unlock(&sbi->ll_process_lock);
1069 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1070 process[*process_count].rw_pid = pid;
1071 process[*process_count].rw_op = rw;
1072 process[*process_count].rw_range_start = pos;
1073 process[*process_count].rw_last_file_pos = pos + count;
1074 process[*process_count].rw_smallest_extent = count;
1075 process[*process_count].rw_largest_extent = count;
1076 process[*process_count].rw_offset = 0;
1077 process[*process_count].rw_last_file = file;
1078 spin_unlock(&sbi->ll_process_lock);
1081 char lpszt[] = LPSZ;
1083 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1086 struct ll_sb_info *sbi = seq->private;
1087 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1088 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1092 do_gettimeofday(&now);
1094 if (!sbi->ll_rw_stats_on) {
1095 seq_printf(seq, "disabled\n"
1096 "write anything in this file to activate, "
1097 "then 0 or \"[D/d]isabled\" to deactivate\n");
1100 spin_lock(&sbi->ll_process_lock);
1102 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
1103 now.tv_sec, now.tv_usec);
1104 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1105 "R/W", "PID", "RANGE START", "RANGE END",
1106 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1107 sprintf(format, "%s%s%s%s%s\n",
1108 "%3c %10d %14Lu %14Lu %17", lpszt+1, " %17", lpszt+1, " %14Ld");
1109 /* We stored the discontiguous offsets here; print them first */
1110 for(i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1111 if (offset[i].rw_pid != 0)
1112 /* Is there a way to snip the '%' off of LPSZ? */
1113 seq_printf(seq, format,
1114 offset[i].rw_op ? 'W' : 'R',
1116 offset[i].rw_range_start,
1117 offset[i].rw_range_end,
1118 offset[i].rw_smallest_extent,
1119 offset[i].rw_largest_extent,
1120 offset[i].rw_offset);
1122 /* Then print the current offsets for each process */
1123 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1124 if (process[i].rw_pid != 0)
1125 seq_printf(seq, format,
1126 process[i].rw_op ? 'W' : 'R',
1128 process[i].rw_range_start,
1129 process[i].rw_last_file_pos,
1130 process[i].rw_smallest_extent,
1131 process[i].rw_largest_extent,
1132 process[i].rw_offset);
1134 spin_unlock(&sbi->ll_process_lock);
1139 static ssize_t ll_rw_offset_stats_seq_write(struct file *file, const char *buf,
1140 size_t len, loff_t *off)
1142 struct seq_file *seq = file->private_data;
1143 struct ll_sb_info *sbi = seq->private;
1144 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1145 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1146 int value = 1, rc = 0;
1148 rc = lprocfs_write_helper(buf, len, &value);
1150 if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1151 strcmp(buf, "Disabled") == 0))
1155 sbi->ll_rw_stats_on = 0;
1157 sbi->ll_rw_stats_on = 1;
1159 spin_lock(&sbi->ll_process_lock);
1160 sbi->ll_offset_process_count = 0;
1161 sbi->ll_rw_offset_entry_count = 0;
1162 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1163 LL_PROCESS_HIST_MAX);
1164 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1165 LL_OFFSET_HIST_MAX);
1166 spin_unlock(&sbi->ll_process_lock);
1171 LPROC_SEQ_FOPS(ll_rw_offset_stats);
1173 void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars)
1175 lvars->module_vars = NULL;
1176 lvars->obd_vars = lprocfs_llite_obd_vars;
1178 #endif /* LPROCFS */