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() - CFS_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() - CFS_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() - CFS_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() - CFS_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() - CFS_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() - CFS_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 cfs_spin_lock(&sbi->ll_lock);
237 pages_number = sbi->ll_ra_info.ra_max_pages;
238 cfs_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 > cfs_num_physpages / 2) {
257 CERROR("can't set file readahead more than %lu MB\n",
258 cfs_num_physpages >> (20 - CFS_PAGE_SHIFT + 1)); /*1/2 of RAM*/
262 cfs_spin_lock(&sbi->ll_lock);
263 sbi->ll_ra_info.ra_max_pages = pages_number;
264 cfs_spin_unlock(&sbi->ll_lock);
269 static int ll_rd_max_readahead_per_file_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 cfs_spin_lock(&sbi->ll_lock);
278 pages_number = sbi->ll_ra_info.ra_max_pages_per_file;
279 cfs_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_readahead_per_file_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 if (pages_number < 0 ||
298 pages_number > sbi->ll_ra_info.ra_max_pages) {
299 CERROR("can't set file readahead more than"
300 "max_read_ahead_mb %lu MB\n",
301 sbi->ll_ra_info.ra_max_pages);
305 cfs_spin_lock(&sbi->ll_lock);
306 sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
307 cfs_spin_unlock(&sbi->ll_lock);
312 static int ll_rd_max_read_ahead_whole_mb(char *page, char **start, off_t off,
313 int count, int *eof, void *data)
315 struct super_block *sb = data;
316 struct ll_sb_info *sbi = ll_s2sbi(sb);
320 cfs_spin_lock(&sbi->ll_lock);
321 pages_number = sbi->ll_ra_info.ra_max_read_ahead_whole_pages;
322 cfs_spin_unlock(&sbi->ll_lock);
324 mult = 1 << (20 - CFS_PAGE_SHIFT);
325 return lprocfs_read_frac_helper(page, count, pages_number, mult);
328 static int ll_wr_max_read_ahead_whole_mb(struct file *file, const char *buffer,
329 unsigned long count, void *data)
331 struct super_block *sb = data;
332 struct ll_sb_info *sbi = ll_s2sbi(sb);
333 int mult, rc, pages_number;
335 mult = 1 << (20 - CFS_PAGE_SHIFT);
336 rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
340 /* Cap this at the current max readahead window size, the readahead
341 * algorithm does this anyway so it's pointless to set it larger. */
342 if (pages_number < 0 ||
343 pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
344 CERROR("can't set max_read_ahead_whole_mb more than "
345 "max_read_ahead_per_file_mb: %lu\n",
346 sbi->ll_ra_info.ra_max_pages_per_file >> (20 - CFS_PAGE_SHIFT));
350 cfs_spin_lock(&sbi->ll_lock);
351 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
352 cfs_spin_unlock(&sbi->ll_lock);
357 static int ll_rd_max_cached_mb(char *page, char **start, off_t off,
358 int count, int *eof, void *data)
360 struct super_block *sb = data;
361 struct ll_sb_info *sbi = ll_s2sbi(sb);
365 cfs_spin_lock(&sbi->ll_lock);
366 pages_number = sbi->ll_async_page_max;
367 cfs_spin_unlock(&sbi->ll_lock);
369 mult = 1 << (20 - CFS_PAGE_SHIFT);
370 return lprocfs_read_frac_helper(page, count, pages_number, mult);;
373 static int ll_wr_max_cached_mb(struct file *file, const char *buffer,
374 unsigned long count, void *data)
376 struct super_block *sb = data;
377 struct ll_sb_info *sbi = ll_s2sbi(sb);
378 int mult, rc, pages_number;
380 mult = 1 << (20 - CFS_PAGE_SHIFT);
381 rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
385 if (pages_number < 0 || pages_number > cfs_num_physpages) {
386 CERROR("can't set max cache more than %lu MB\n",
387 cfs_num_physpages >> (20 - CFS_PAGE_SHIFT));
391 cfs_spin_lock(&sbi->ll_lock);
392 sbi->ll_async_page_max = pages_number ;
393 cfs_spin_unlock(&sbi->ll_lock);
396 /* Not set up yet, don't call llap_shrink_cache */
402 static int ll_rd_checksum(char *page, char **start, off_t off,
403 int count, int *eof, void *data)
405 struct super_block *sb = data;
406 struct ll_sb_info *sbi = ll_s2sbi(sb);
408 return snprintf(page, count, "%u\n",
409 (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
412 static int ll_wr_checksum(struct file *file, const char *buffer,
413 unsigned long count, void *data)
415 struct super_block *sb = data;
416 struct ll_sb_info *sbi = ll_s2sbi(sb);
423 rc = lprocfs_write_helper(buffer, count, &val);
427 sbi->ll_flags |= LL_SBI_CHECKSUM;
429 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
431 rc = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
432 KEY_CHECKSUM, sizeof(val), &val, NULL);
434 CWARN("Failed to set OSC checksum flags: %d\n", rc);
439 static int ll_rd_max_rw_chunk(char *page, char **start, off_t off,
440 int count, int *eof, void *data)
442 struct super_block *sb = data;
444 return snprintf(page, count, "%lu\n", ll_s2sbi(sb)->ll_max_rw_chunk);
447 static int ll_wr_max_rw_chunk(struct file *file, const char *buffer,
448 unsigned long count, void *data)
450 struct super_block *sb = data;
453 rc = lprocfs_write_helper(buffer, count, &val);
456 ll_s2sbi(sb)->ll_max_rw_chunk = val;
460 static int ll_rd_track_id(char *page, int count, void *data,
461 enum stats_track_type type)
463 struct super_block *sb = data;
465 if (ll_s2sbi(sb)->ll_stats_track_type == type) {
466 return snprintf(page, count, "%d\n",
467 ll_s2sbi(sb)->ll_stats_track_id);
469 } else if (ll_s2sbi(sb)->ll_stats_track_type == STATS_TRACK_ALL) {
470 return snprintf(page, count, "0 (all)\n");
472 return snprintf(page, count, "untracked\n");
476 static int ll_wr_track_id(const char *buffer, unsigned long count, void *data,
477 enum stats_track_type type)
479 struct super_block *sb = data;
482 rc = lprocfs_write_helper(buffer, count, &pid);
485 ll_s2sbi(sb)->ll_stats_track_id = pid;
487 ll_s2sbi(sb)->ll_stats_track_type = STATS_TRACK_ALL;
489 ll_s2sbi(sb)->ll_stats_track_type = type;
490 lprocfs_clear_stats(ll_s2sbi(sb)->ll_stats);
494 static int ll_rd_track_pid(char *page, char **start, off_t off,
495 int count, int *eof, void *data)
497 return (ll_rd_track_id(page, count, data, STATS_TRACK_PID));
500 static int ll_wr_track_pid(struct file *file, const char *buffer,
501 unsigned long count, void *data)
503 return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PID));
506 static int ll_rd_track_ppid(char *page, char **start, off_t off,
507 int count, int *eof, void *data)
509 return (ll_rd_track_id(page, count, data, STATS_TRACK_PPID));
512 static int ll_wr_track_ppid(struct file *file, const char *buffer,
513 unsigned long count, void *data)
515 return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PPID));
518 static int ll_rd_track_gid(char *page, char **start, off_t off,
519 int count, int *eof, void *data)
521 return (ll_rd_track_id(page, count, data, STATS_TRACK_GID));
524 static int ll_wr_track_gid(struct file *file, const char *buffer,
525 unsigned long count, void *data)
527 return (ll_wr_track_id(buffer, count, data, STATS_TRACK_GID));
530 static int ll_rd_statahead_max(char *page, char **start, off_t off,
531 int count, int *eof, void *data)
533 struct super_block *sb = data;
534 struct ll_sb_info *sbi = ll_s2sbi(sb);
536 return snprintf(page, count, "%u\n", sbi->ll_sa_max);
539 static int ll_wr_statahead_max(struct file *file, const char *buffer,
540 unsigned long count, void *data)
542 struct super_block *sb = data;
543 struct ll_sb_info *sbi = ll_s2sbi(sb);
546 rc = lprocfs_write_helper(buffer, count, &val);
550 if (val >= 0 && val <= LL_SA_RPC_MAX)
551 sbi->ll_sa_max = val;
553 CERROR("Bad statahead_max value %d. Valid values are in the "
554 "range [0, %d]\n", val, LL_SA_RPC_MAX);
559 static int ll_rd_statahead_stats(char *page, char **start, off_t off,
560 int count, int *eof, void *data)
562 struct super_block *sb = data;
563 struct ll_sb_info *sbi = ll_s2sbi(sb);
565 return snprintf(page, count,
566 "statahead wrong: %u\n"
567 "statahead total: %u\n"
571 "miss count: %llu\n",
580 static int ll_rd_lazystatfs(char *page, char **start, off_t off,
581 int count, int *eof, void *data)
583 struct super_block *sb = data;
584 struct ll_sb_info *sbi = ll_s2sbi(sb);
586 return snprintf(page, count, "%u\n",
587 (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
590 static int ll_wr_lazystatfs(struct file *file, const char *buffer,
591 unsigned long count, void *data)
593 struct super_block *sb = data;
594 struct ll_sb_info *sbi = ll_s2sbi(sb);
597 rc = lprocfs_write_helper(buffer, count, &val);
602 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
604 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
609 static struct lprocfs_vars lprocfs_llite_obd_vars[] = {
610 { "uuid", ll_rd_sb_uuid, 0, 0 },
611 //{ "mntpt_path", ll_rd_path, 0, 0 },
612 { "fstype", ll_rd_fstype, 0, 0 },
613 { "site", ll_rd_site_stats, 0, 0 },
614 { "blocksize", ll_rd_blksize, 0, 0 },
615 { "kbytestotal", ll_rd_kbytestotal, 0, 0 },
616 { "kbytesfree", ll_rd_kbytesfree, 0, 0 },
617 { "kbytesavail", ll_rd_kbytesavail, 0, 0 },
618 { "filestotal", ll_rd_filestotal, 0, 0 },
619 { "filesfree", ll_rd_filesfree, 0, 0 },
620 { "client_type", ll_rd_client_type, 0, 0 },
621 //{ "filegroups", lprocfs_rd_filegroups, 0, 0 },
622 { "max_read_ahead_mb", ll_rd_max_readahead_mb,
623 ll_wr_max_readahead_mb, 0 },
624 { "max_read_ahead_per_file_mb", ll_rd_max_readahead_per_file_mb,
625 ll_wr_max_readahead_per_file_mb, 0 },
626 { "max_read_ahead_whole_mb", ll_rd_max_read_ahead_whole_mb,
627 ll_wr_max_read_ahead_whole_mb, 0 },
628 { "max_cached_mb", ll_rd_max_cached_mb, ll_wr_max_cached_mb, 0 },
629 { "checksum_pages", ll_rd_checksum, ll_wr_checksum, 0 },
630 { "max_rw_chunk", ll_rd_max_rw_chunk, ll_wr_max_rw_chunk, 0 },
631 { "stats_track_pid", ll_rd_track_pid, ll_wr_track_pid, 0 },
632 { "stats_track_ppid", ll_rd_track_ppid, ll_wr_track_ppid, 0 },
633 { "stats_track_gid", ll_rd_track_gid, ll_wr_track_gid, 0 },
634 { "statahead_max", ll_rd_statahead_max, ll_wr_statahead_max, 0 },
635 { "statahead_stats", ll_rd_statahead_stats, 0, 0 },
636 { "lazystatfs", ll_rd_lazystatfs, ll_wr_lazystatfs, 0 },
640 #define MAX_STRING_SIZE 128
642 struct llite_file_opcode {
646 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
648 { LPROC_LL_DIRTY_HITS, LPROCFS_TYPE_REGS, "dirty_pages_hits" },
649 { LPROC_LL_DIRTY_MISSES, LPROCFS_TYPE_REGS, "dirty_pages_misses" },
650 { LPROC_LL_WB_WRITEPAGE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
651 "writeback_from_writepage" },
652 { LPROC_LL_WB_PRESSURE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
653 "writeback_from_pressure" },
654 { LPROC_LL_WB_OK, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
655 "writeback_ok_pages" },
656 { LPROC_LL_WB_FAIL, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
657 "writeback_failed_pages" },
658 { LPROC_LL_READ_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
660 { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
662 { LPROC_LL_BRW_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
664 { LPROC_LL_BRW_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
667 { LPROC_LL_IOCTL, LPROCFS_TYPE_REGS, "ioctl" },
668 { LPROC_LL_OPEN, LPROCFS_TYPE_REGS, "open" },
669 { LPROC_LL_RELEASE, LPROCFS_TYPE_REGS, "close" },
670 { LPROC_LL_MAP, LPROCFS_TYPE_REGS, "mmap" },
671 { LPROC_LL_LLSEEK, LPROCFS_TYPE_REGS, "seek" },
672 { LPROC_LL_FSYNC, LPROCFS_TYPE_REGS, "fsync" },
673 /* inode operation */
674 { LPROC_LL_SETATTR, LPROCFS_TYPE_REGS, "setattr" },
675 { LPROC_LL_TRUNC, LPROCFS_TYPE_REGS, "truncate" },
676 { LPROC_LL_LOCKLESS_TRUNC, LPROCFS_TYPE_REGS, "lockless_truncate"},
677 { LPROC_LL_FLOCK, LPROCFS_TYPE_REGS, "flock" },
678 { LPROC_LL_GETATTR, LPROCFS_TYPE_REGS, "getattr" },
679 /* special inode operation */
680 { LPROC_LL_STAFS, LPROCFS_TYPE_REGS, "statfs" },
681 { LPROC_LL_ALLOC_INODE, LPROCFS_TYPE_REGS, "alloc_inode" },
682 { LPROC_LL_SETXATTR, LPROCFS_TYPE_REGS, "setxattr" },
683 { LPROC_LL_GETXATTR, LPROCFS_TYPE_REGS, "getxattr" },
684 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_REGS, "listxattr" },
685 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_REGS, "removexattr" },
686 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_REGS, "inode_permission" },
687 { LPROC_LL_DIRECT_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
689 { LPROC_LL_DIRECT_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
691 { LPROC_LL_LOCKLESS_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
692 "lockless_read_bytes" },
693 { LPROC_LL_LOCKLESS_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
694 "lockless_write_bytes" },
698 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
702 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
703 lprocfs_counter_add(sbi->ll_stats, op, count);
704 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
705 sbi->ll_stats_track_id == current->pid)
706 lprocfs_counter_add(sbi->ll_stats, op, count);
707 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
708 sbi->ll_stats_track_id == current->parent->pid)
709 lprocfs_counter_add(sbi->ll_stats, op, count);
710 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
711 sbi->ll_stats_track_id == cfs_curproc_gid())
712 lprocfs_counter_add(sbi->ll_stats, op, count);
714 EXPORT_SYMBOL(ll_stats_ops_tally);
716 static const char *ra_stat_string[] = {
717 [RA_STAT_HIT] = "hits",
718 [RA_STAT_MISS] = "misses",
719 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
720 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
721 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
722 [RA_STAT_FAILED_MATCH] = "failed lock match",
723 [RA_STAT_DISCARDED] = "read but discarded",
724 [RA_STAT_ZERO_LEN] = "zero length file",
725 [RA_STAT_ZERO_WINDOW] = "zero size window",
726 [RA_STAT_EOF] = "read-ahead to EOF",
727 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
728 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
732 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
733 struct super_block *sb, char *osc, char *mdc)
735 struct lprocfs_vars lvars[2];
736 struct lustre_sb_info *lsi = s2lsi(sb);
737 struct ll_sb_info *sbi = ll_s2sbi(sb);
738 struct obd_device *obd;
739 char name[MAX_STRING_SIZE + 1], *ptr;
740 int err, id, len, rc;
743 memset(lvars, 0, sizeof(lvars));
745 name[MAX_STRING_SIZE] = '\0';
746 lvars[0].name = name;
748 LASSERT(sbi != NULL);
749 LASSERT(mdc != NULL);
750 LASSERT(osc != NULL);
753 len = strlen(lsi->lsi_lmd->lmd_profile);
754 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
755 if (ptr && (strcmp(ptr, "-client") == 0))
759 snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
760 lsi->lsi_lmd->lmd_profile, sb);
762 sbi->ll_proc_root = lprocfs_register(name, parent, NULL, NULL);
763 if (IS_ERR(sbi->ll_proc_root)) {
764 err = PTR_ERR(sbi->ll_proc_root);
765 sbi->ll_proc_root = NULL;
769 rc = lprocfs_seq_create(sbi->ll_proc_root, "dump_page_cache", 0444,
770 &vvp_dump_pgcache_file_ops, sbi);
772 CWARN("Error adding the dump_page_cache file\n");
774 rc = lprocfs_seq_create(sbi->ll_proc_root, "extents_stats", 0644,
775 &ll_rw_extents_stats_fops, sbi);
777 CWARN("Error adding the extent_stats file\n");
779 rc = lprocfs_seq_create(sbi->ll_proc_root, "extents_stats_per_process",
780 0644, &ll_rw_extents_stats_pp_fops, sbi);
782 CWARN("Error adding the extents_stats_per_process file\n");
784 rc = lprocfs_seq_create(sbi->ll_proc_root, "offset_stats", 0644,
785 &ll_rw_offset_stats_fops, sbi);
787 CWARN("Error adding the offset_stats file\n");
789 /* File operations stats */
790 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
791 LPROCFS_STATS_FLAG_NONE);
792 if (sbi->ll_stats == NULL)
793 GOTO(out, err = -ENOMEM);
794 /* do counter init */
795 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
796 __u32 type = llite_opcode_table[id].type;
798 if (type & LPROCFS_TYPE_REGS)
800 else if (type & LPROCFS_TYPE_BYTES)
802 else if (type & LPROCFS_TYPE_PAGES)
804 lprocfs_counter_init(sbi->ll_stats,
805 llite_opcode_table[id].opcode,
806 (type & LPROCFS_CNTR_AVGMINMAX),
807 llite_opcode_table[id].opname, ptr);
809 err = lprocfs_register_stats(sbi->ll_proc_root, "stats", sbi->ll_stats);
813 sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
814 LPROCFS_STATS_FLAG_NONE);
815 if (sbi->ll_ra_stats == NULL)
816 GOTO(out, err = -ENOMEM);
818 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
819 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
820 ra_stat_string[id], "pages");
821 err = lprocfs_register_stats(sbi->ll_proc_root, "read_ahead_stats",
827 err = lprocfs_add_vars(sbi->ll_proc_root, lprocfs_llite_obd_vars, sb);
832 obd = class_name2obd(mdc);
834 LASSERT(obd != NULL);
835 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
836 LASSERT(obd->obd_type->typ_name != NULL);
838 snprintf(name, MAX_STRING_SIZE, "%s/common_name",
839 obd->obd_type->typ_name);
840 lvars[0].read_fptr = lprocfs_rd_name;
841 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
845 snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
846 lvars[0].read_fptr = lprocfs_rd_uuid;
847 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
852 obd = class_name2obd(osc);
854 LASSERT(obd != NULL);
855 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
856 LASSERT(obd->obd_type->typ_name != NULL);
858 snprintf(name, MAX_STRING_SIZE, "%s/common_name",
859 obd->obd_type->typ_name);
860 lvars[0].read_fptr = lprocfs_rd_name;
861 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
865 snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
866 lvars[0].read_fptr = lprocfs_rd_uuid;
867 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
870 lprocfs_remove(&sbi->ll_proc_root);
871 lprocfs_free_stats(&sbi->ll_ra_stats);
872 lprocfs_free_stats(&sbi->ll_stats);
877 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi)
879 if (sbi->ll_proc_root) {
880 lprocfs_remove(&sbi->ll_proc_root);
881 lprocfs_free_stats(&sbi->ll_ra_stats);
882 lprocfs_free_stats(&sbi->ll_stats);
885 #undef MAX_STRING_SIZE
887 #define pct(a,b) (b ? a * 100 / b : 0)
889 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
890 struct seq_file *seq, int which)
892 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
893 unsigned long start, end, r, w;
894 char *unitp = "KMGTPEZY";
896 struct per_process_info *pp_info = &io_extents->pp_extents[which];
902 for(i = 0; i < LL_HIST_MAX; i++) {
903 read_tot += pp_info->pp_r_hist.oh_buckets[i];
904 write_tot += pp_info->pp_w_hist.oh_buckets[i];
907 for(i = 0; i < LL_HIST_MAX; i++) {
908 r = pp_info->pp_r_hist.oh_buckets[i];
909 w = pp_info->pp_w_hist.oh_buckets[i];
912 end = 1 << (i + LL_HIST_START - units);
913 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4lu %4lu | "
914 "%14lu %4lu %4lu\n", start, *unitp, end, *unitp,
915 (i == LL_HIST_MAX - 1) ? '+' : ' ',
916 r, pct(r, read_tot), pct(read_cum, read_tot),
917 w, pct(w, write_tot), pct(write_cum, write_tot));
919 if (start == 1<<10) {
924 if (read_cum == read_tot && write_cum == write_tot)
929 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
932 struct ll_sb_info *sbi = seq->private;
933 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
936 cfs_gettimeofday(&now);
938 if (!sbi->ll_rw_stats_on) {
939 seq_printf(seq, "disabled\n"
940 "write anything in this file to activate, "
941 "then 0 or \"[D/d]isabled\" to deactivate\n");
944 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
945 now.tv_sec, now.tv_usec);
946 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
947 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
948 "extents", "calls", "%", "cum%",
949 "calls", "%", "cum%");
950 cfs_spin_lock(&sbi->ll_pp_extent_lock);
951 for(k = 0; k < LL_PROCESS_HIST_MAX; k++) {
952 if(io_extents->pp_extents[k].pid != 0) {
953 seq_printf(seq, "\nPID: %d\n",
954 io_extents->pp_extents[k].pid);
955 ll_display_extents_info(io_extents, seq, k);
958 cfs_spin_unlock(&sbi->ll_pp_extent_lock);
962 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
963 const char *buf, size_t len,
966 struct seq_file *seq = file->private_data;
967 struct ll_sb_info *sbi = seq->private;
968 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
970 int value = 1, rc = 0;
972 rc = lprocfs_write_helper(buf, len, &value);
973 if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
974 strcmp(buf, "Disabled") == 0))
978 sbi->ll_rw_stats_on = 0;
980 sbi->ll_rw_stats_on = 1;
982 cfs_spin_lock(&sbi->ll_pp_extent_lock);
983 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
984 io_extents->pp_extents[i].pid = 0;
985 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
986 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
988 cfs_spin_unlock(&sbi->ll_pp_extent_lock);
992 LPROC_SEQ_FOPS(ll_rw_extents_stats_pp);
994 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
997 struct ll_sb_info *sbi = seq->private;
998 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1000 cfs_gettimeofday(&now);
1002 if (!sbi->ll_rw_stats_on) {
1003 seq_printf(seq, "disabled\n"
1004 "write anything in this file to activate, "
1005 "then 0 or \"[D/d]isabled\" to deactivate\n");
1008 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
1009 now.tv_sec, now.tv_usec);
1011 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1012 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1013 "extents", "calls", "%", "cum%",
1014 "calls", "%", "cum%");
1015 cfs_spin_lock(&sbi->ll_lock);
1016 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1017 cfs_spin_unlock(&sbi->ll_lock);
1022 static ssize_t ll_rw_extents_stats_seq_write(struct file *file, const char *buf,
1023 size_t len, loff_t *off)
1025 struct seq_file *seq = file->private_data;
1026 struct ll_sb_info *sbi = seq->private;
1027 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1029 int value = 1, rc = 0;
1031 rc = lprocfs_write_helper(buf, len, &value);
1032 if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1033 strcmp(buf, "Disabled") == 0))
1037 sbi->ll_rw_stats_on = 0;
1039 sbi->ll_rw_stats_on = 1;
1040 cfs_spin_lock(&sbi->ll_pp_extent_lock);
1041 for(i = 0; i <= LL_PROCESS_HIST_MAX; i++)
1043 io_extents->pp_extents[i].pid = 0;
1044 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1045 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1047 cfs_spin_unlock(&sbi->ll_pp_extent_lock);
1052 LPROC_SEQ_FOPS(ll_rw_extents_stats);
1054 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1055 struct ll_file_data *file, loff_t pos,
1056 size_t count, int rw)
1059 struct ll_rw_process_info *process;
1060 struct ll_rw_process_info *offset;
1061 int *off_count = &sbi->ll_rw_offset_entry_count;
1062 int *process_count = &sbi->ll_offset_process_count;
1063 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1065 if(!sbi->ll_rw_stats_on)
1067 process = sbi->ll_rw_process_info;
1068 offset = sbi->ll_rw_offset_info;
1070 cfs_spin_lock(&sbi->ll_pp_extent_lock);
1071 /* Extent statistics */
1072 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1073 if(io_extents->pp_extents[i].pid == pid) {
1081 sbi->ll_extent_process_count =
1082 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1083 cur = sbi->ll_extent_process_count;
1084 io_extents->pp_extents[cur].pid = pid;
1085 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1086 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1089 for(i = 0; (count >= (1 << LL_HIST_START << i)) &&
1090 (i < (LL_HIST_MAX - 1)); i++);
1092 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1093 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1095 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1096 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1098 cfs_spin_unlock(&sbi->ll_pp_extent_lock);
1100 cfs_spin_lock(&sbi->ll_process_lock);
1101 /* Offset statistics */
1102 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1103 if (process[i].rw_pid == pid) {
1104 if (process[i].rw_last_file != file) {
1105 process[i].rw_range_start = pos;
1106 process[i].rw_last_file_pos = pos + count;
1107 process[i].rw_smallest_extent = count;
1108 process[i].rw_largest_extent = count;
1109 process[i].rw_offset = 0;
1110 process[i].rw_last_file = file;
1111 cfs_spin_unlock(&sbi->ll_process_lock);
1114 if (process[i].rw_last_file_pos != pos) {
1116 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1117 offset[*off_count].rw_op = process[i].rw_op;
1118 offset[*off_count].rw_pid = pid;
1119 offset[*off_count].rw_range_start =
1120 process[i].rw_range_start;
1121 offset[*off_count].rw_range_end =
1122 process[i].rw_last_file_pos;
1123 offset[*off_count].rw_smallest_extent =
1124 process[i].rw_smallest_extent;
1125 offset[*off_count].rw_largest_extent =
1126 process[i].rw_largest_extent;
1127 offset[*off_count].rw_offset =
1128 process[i].rw_offset;
1129 process[i].rw_op = rw;
1130 process[i].rw_range_start = pos;
1131 process[i].rw_smallest_extent = count;
1132 process[i].rw_largest_extent = count;
1133 process[i].rw_offset = pos -
1134 process[i].rw_last_file_pos;
1136 if(process[i].rw_smallest_extent > count)
1137 process[i].rw_smallest_extent = count;
1138 if(process[i].rw_largest_extent < count)
1139 process[i].rw_largest_extent = count;
1140 process[i].rw_last_file_pos = pos + count;
1141 cfs_spin_unlock(&sbi->ll_process_lock);
1145 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1146 process[*process_count].rw_pid = pid;
1147 process[*process_count].rw_op = rw;
1148 process[*process_count].rw_range_start = pos;
1149 process[*process_count].rw_last_file_pos = pos + count;
1150 process[*process_count].rw_smallest_extent = count;
1151 process[*process_count].rw_largest_extent = count;
1152 process[*process_count].rw_offset = 0;
1153 process[*process_count].rw_last_file = file;
1154 cfs_spin_unlock(&sbi->ll_process_lock);
1157 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1160 struct ll_sb_info *sbi = seq->private;
1161 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1162 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1165 cfs_gettimeofday(&now);
1167 if (!sbi->ll_rw_stats_on) {
1168 seq_printf(seq, "disabled\n"
1169 "write anything in this file to activate, "
1170 "then 0 or \"[D/d]isabled\" to deactivate\n");
1173 cfs_spin_lock(&sbi->ll_process_lock);
1175 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
1176 now.tv_sec, now.tv_usec);
1177 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1178 "R/W", "PID", "RANGE START", "RANGE END",
1179 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1180 /* We stored the discontiguous offsets here; print them first */
1181 for(i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1182 if (offset[i].rw_pid != 0)
1183 seq_printf(seq,"%3c %10d %14Lu %14Lu %17lu %17lu %14Lu",
1184 offset[i].rw_op ? 'W' : 'R',
1186 offset[i].rw_range_start,
1187 offset[i].rw_range_end,
1188 (unsigned long)offset[i].rw_smallest_extent,
1189 (unsigned long)offset[i].rw_largest_extent,
1190 offset[i].rw_offset);
1192 /* Then print the current offsets for each process */
1193 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1194 if (process[i].rw_pid != 0)
1195 seq_printf(seq,"%3c %10d %14Lu %14Lu %17lu %17lu %14Lu",
1196 process[i].rw_op ? 'W' : 'R',
1198 process[i].rw_range_start,
1199 process[i].rw_last_file_pos,
1200 (unsigned long)process[i].rw_smallest_extent,
1201 (unsigned long)process[i].rw_largest_extent,
1202 process[i].rw_offset);
1204 cfs_spin_unlock(&sbi->ll_process_lock);
1209 static ssize_t ll_rw_offset_stats_seq_write(struct file *file, const char *buf,
1210 size_t len, loff_t *off)
1212 struct seq_file *seq = file->private_data;
1213 struct ll_sb_info *sbi = seq->private;
1214 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1215 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1216 int value = 1, rc = 0;
1218 rc = lprocfs_write_helper(buf, len, &value);
1220 if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1221 strcmp(buf, "Disabled") == 0))
1225 sbi->ll_rw_stats_on = 0;
1227 sbi->ll_rw_stats_on = 1;
1229 cfs_spin_lock(&sbi->ll_process_lock);
1230 sbi->ll_offset_process_count = 0;
1231 sbi->ll_rw_offset_entry_count = 0;
1232 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1233 LL_PROCESS_HIST_MAX);
1234 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1235 LL_OFFSET_HIST_MAX);
1236 cfs_spin_unlock(&sbi->ll_process_lock);
1241 LPROC_SEQ_FOPS(ll_rw_offset_stats);
1243 void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars)
1245 lvars->module_vars = NULL;
1246 lvars->obd_vars = lprocfs_llite_obd_vars;
1248 #endif /* LPROCFS */