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_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 spin_lock(&sbi->ll_lock);
278 pages_number = sbi->ll_ra_info.ra_max_pages_per_file;
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_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", sbi->ll_ra_info.ra_max_pages);
304 spin_lock(&sbi->ll_lock);
305 sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
306 spin_unlock(&sbi->ll_lock);
311 static int ll_rd_max_read_ahead_whole_mb(char *page, char **start, off_t off,
312 int count, int *eof, void *data)
314 struct super_block *sb = data;
315 struct ll_sb_info *sbi = ll_s2sbi(sb);
319 spin_lock(&sbi->ll_lock);
320 pages_number = sbi->ll_ra_info.ra_max_read_ahead_whole_pages;
321 spin_unlock(&sbi->ll_lock);
323 mult = 1 << (20 - CFS_PAGE_SHIFT);
324 return lprocfs_read_frac_helper(page, count, pages_number, mult);
327 static int ll_wr_max_read_ahead_whole_mb(struct file *file, const char *buffer,
328 unsigned long count, void *data)
330 struct super_block *sb = data;
331 struct ll_sb_info *sbi = ll_s2sbi(sb);
332 int mult, rc, pages_number;
334 mult = 1 << (20 - CFS_PAGE_SHIFT);
335 rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
339 /* Cap this at the current max readahead window size, the readahead
340 * algorithm does this anyway so it's pointless to set it larger. */
341 if (pages_number < 0 ||
342 pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
343 CERROR("can't set max_read_ahead_whole_mb more than "
344 "max_read_ahead_per_file_mb: %lu\n",
345 sbi->ll_ra_info.ra_max_pages_per_file >> (20 - CFS_PAGE_SHIFT));
349 spin_lock(&sbi->ll_lock);
350 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
351 spin_unlock(&sbi->ll_lock);
356 static int ll_rd_max_cached_mb(char *page, char **start, off_t off,
357 int count, int *eof, void *data)
359 struct super_block *sb = data;
360 struct ll_sb_info *sbi = ll_s2sbi(sb);
364 spin_lock(&sbi->ll_lock);
365 pages_number = sbi->ll_async_page_max;
366 spin_unlock(&sbi->ll_lock);
368 mult = 1 << (20 - CFS_PAGE_SHIFT);
369 return lprocfs_read_frac_helper(page, count, pages_number, mult);;
372 static int ll_wr_max_cached_mb(struct file *file, const char *buffer,
373 unsigned long count, void *data)
375 struct super_block *sb = data;
376 struct ll_sb_info *sbi = ll_s2sbi(sb);
377 int mult, rc, pages_number;
379 mult = 1 << (20 - CFS_PAGE_SHIFT);
380 rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
384 if (pages_number < 0 || pages_number > num_physpages) {
385 CERROR("can't set max cache more than %lu MB\n",
386 num_physpages >> (20 - CFS_PAGE_SHIFT));
390 spin_lock(&sbi->ll_lock);
391 sbi->ll_async_page_max = pages_number ;
392 spin_unlock(&sbi->ll_lock);
395 /* Not set up yet, don't call llap_shrink_cache */
401 static int ll_rd_checksum(char *page, char **start, off_t off,
402 int count, int *eof, void *data)
404 struct super_block *sb = data;
405 struct ll_sb_info *sbi = ll_s2sbi(sb);
407 return snprintf(page, count, "%u\n",
408 (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
411 static int ll_wr_checksum(struct file *file, const char *buffer,
412 unsigned long count, void *data)
414 struct super_block *sb = data;
415 struct ll_sb_info *sbi = ll_s2sbi(sb);
422 rc = lprocfs_write_helper(buffer, count, &val);
426 sbi->ll_flags |= LL_SBI_CHECKSUM;
428 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
430 rc = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
431 KEY_CHECKSUM, sizeof(val), &val, NULL);
433 CWARN("Failed to set OSC checksum flags: %d\n", rc);
438 static int ll_rd_max_rw_chunk(char *page, char **start, off_t off,
439 int count, int *eof, void *data)
441 struct super_block *sb = data;
443 return snprintf(page, count, "%lu\n", ll_s2sbi(sb)->ll_max_rw_chunk);
446 static int ll_wr_max_rw_chunk(struct file *file, const char *buffer,
447 unsigned long count, void *data)
449 struct super_block *sb = data;
452 rc = lprocfs_write_helper(buffer, count, &val);
455 ll_s2sbi(sb)->ll_max_rw_chunk = val;
459 static int ll_rd_track_id(char *page, int count, void *data,
460 enum stats_track_type type)
462 struct super_block *sb = data;
464 if (ll_s2sbi(sb)->ll_stats_track_type == type) {
465 return snprintf(page, count, "%d\n",
466 ll_s2sbi(sb)->ll_stats_track_id);
468 } else if (ll_s2sbi(sb)->ll_stats_track_type == STATS_TRACK_ALL) {
469 return snprintf(page, count, "0 (all)\n");
471 return snprintf(page, count, "untracked\n");
475 static int ll_wr_track_id(const char *buffer, unsigned long count, void *data,
476 enum stats_track_type type)
478 struct super_block *sb = data;
481 rc = lprocfs_write_helper(buffer, count, &pid);
484 ll_s2sbi(sb)->ll_stats_track_id = pid;
486 ll_s2sbi(sb)->ll_stats_track_type = STATS_TRACK_ALL;
488 ll_s2sbi(sb)->ll_stats_track_type = type;
489 lprocfs_clear_stats(ll_s2sbi(sb)->ll_stats);
493 static int ll_rd_track_pid(char *page, char **start, off_t off,
494 int count, int *eof, void *data)
496 return (ll_rd_track_id(page, count, data, STATS_TRACK_PID));
499 static int ll_wr_track_pid(struct file *file, const char *buffer,
500 unsigned long count, void *data)
502 return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PID));
505 static int ll_rd_track_ppid(char *page, char **start, off_t off,
506 int count, int *eof, void *data)
508 return (ll_rd_track_id(page, count, data, STATS_TRACK_PPID));
511 static int ll_wr_track_ppid(struct file *file, const char *buffer,
512 unsigned long count, void *data)
514 return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PPID));
517 static int ll_rd_track_gid(char *page, char **start, off_t off,
518 int count, int *eof, void *data)
520 return (ll_rd_track_id(page, count, data, STATS_TRACK_GID));
523 static int ll_wr_track_gid(struct file *file, const char *buffer,
524 unsigned long count, void *data)
526 return (ll_wr_track_id(buffer, count, data, STATS_TRACK_GID));
529 static int ll_rd_statahead_max(char *page, char **start, off_t off,
530 int count, int *eof, void *data)
532 struct super_block *sb = data;
533 struct ll_sb_info *sbi = ll_s2sbi(sb);
535 return snprintf(page, count, "%u\n", sbi->ll_sa_max);
538 static int ll_wr_statahead_max(struct file *file, const char *buffer,
539 unsigned long count, void *data)
541 struct super_block *sb = data;
542 struct ll_sb_info *sbi = ll_s2sbi(sb);
545 rc = lprocfs_write_helper(buffer, count, &val);
549 if (val >= 0 && val <= LL_SA_RPC_MAX)
550 sbi->ll_sa_max = val;
552 CERROR("Bad statahead_max value %d. Valid values are in the "
553 "range [0, %d]\n", val, LL_SA_RPC_MAX);
558 static int ll_rd_statahead_stats(char *page, char **start, off_t off,
559 int count, int *eof, void *data)
561 struct super_block *sb = data;
562 struct ll_sb_info *sbi = ll_s2sbi(sb);
564 return snprintf(page, count,
565 "statahead total: %u\n"
566 "statahead wrong: %u\n",
567 atomic_read(&sbi->ll_sa_total),
568 atomic_read(&sbi->ll_sa_wrong));
571 static int ll_rd_lazystatfs(char *page, char **start, off_t off,
572 int count, int *eof, void *data)
574 struct super_block *sb = data;
575 struct ll_sb_info *sbi = ll_s2sbi(sb);
577 return snprintf(page, count, "%u\n",
578 (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
581 static int ll_wr_lazystatfs(struct file *file, const char *buffer,
582 unsigned long count, void *data)
584 struct super_block *sb = data;
585 struct ll_sb_info *sbi = ll_s2sbi(sb);
588 rc = lprocfs_write_helper(buffer, count, &val);
593 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
595 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
600 static struct lprocfs_vars lprocfs_llite_obd_vars[] = {
601 { "uuid", ll_rd_sb_uuid, 0, 0 },
602 //{ "mntpt_path", ll_rd_path, 0, 0 },
603 { "fstype", ll_rd_fstype, 0, 0 },
604 { "site", ll_rd_site_stats, 0, 0 },
605 { "blocksize", ll_rd_blksize, 0, 0 },
606 { "kbytestotal", ll_rd_kbytestotal, 0, 0 },
607 { "kbytesfree", ll_rd_kbytesfree, 0, 0 },
608 { "kbytesavail", ll_rd_kbytesavail, 0, 0 },
609 { "filestotal", ll_rd_filestotal, 0, 0 },
610 { "filesfree", ll_rd_filesfree, 0, 0 },
611 { "client_type", ll_rd_client_type, 0, 0 },
612 //{ "filegroups", lprocfs_rd_filegroups, 0, 0 },
613 { "max_read_ahead_mb", ll_rd_max_readahead_mb,
614 ll_wr_max_readahead_mb, 0 },
615 { "max_read_ahead_per_file_mb", ll_rd_max_readahead_per_file_mb,
616 ll_wr_max_readahead_per_file_mb, 0 },
617 { "max_read_ahead_whole_mb", ll_rd_max_read_ahead_whole_mb,
618 ll_wr_max_read_ahead_whole_mb, 0 },
619 { "max_cached_mb", ll_rd_max_cached_mb, ll_wr_max_cached_mb, 0 },
620 { "checksum_pages", ll_rd_checksum, ll_wr_checksum, 0 },
621 { "max_rw_chunk", ll_rd_max_rw_chunk, ll_wr_max_rw_chunk, 0 },
622 { "stats_track_pid", ll_rd_track_pid, ll_wr_track_pid, 0 },
623 { "stats_track_ppid", ll_rd_track_ppid, ll_wr_track_ppid, 0 },
624 { "stats_track_gid", ll_rd_track_gid, ll_wr_track_gid, 0 },
625 { "statahead_max", ll_rd_statahead_max, ll_wr_statahead_max, 0 },
626 { "statahead_stats", ll_rd_statahead_stats, 0, 0 },
627 { "lazystatfs", ll_rd_lazystatfs, ll_wr_lazystatfs, 0 },
631 #define MAX_STRING_SIZE 128
633 struct llite_file_opcode {
637 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
639 { LPROC_LL_DIRTY_HITS, LPROCFS_TYPE_REGS, "dirty_pages_hits" },
640 { LPROC_LL_DIRTY_MISSES, LPROCFS_TYPE_REGS, "dirty_pages_misses" },
641 { LPROC_LL_WB_WRITEPAGE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
642 "writeback_from_writepage" },
643 { LPROC_LL_WB_PRESSURE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
644 "writeback_from_pressure" },
645 { LPROC_LL_WB_OK, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
646 "writeback_ok_pages" },
647 { LPROC_LL_WB_FAIL, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
648 "writeback_failed_pages" },
649 { LPROC_LL_READ_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
651 { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
653 { LPROC_LL_BRW_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
655 { LPROC_LL_BRW_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
658 { LPROC_LL_IOCTL, LPROCFS_TYPE_REGS, "ioctl" },
659 { LPROC_LL_OPEN, LPROCFS_TYPE_REGS, "open" },
660 { LPROC_LL_RELEASE, LPROCFS_TYPE_REGS, "close" },
661 { LPROC_LL_MAP, LPROCFS_TYPE_REGS, "mmap" },
662 { LPROC_LL_LLSEEK, LPROCFS_TYPE_REGS, "seek" },
663 { LPROC_LL_FSYNC, LPROCFS_TYPE_REGS, "fsync" },
664 /* inode operation */
665 { LPROC_LL_SETATTR, LPROCFS_TYPE_REGS, "setattr" },
666 { LPROC_LL_TRUNC, LPROCFS_TYPE_REGS, "truncate" },
667 { LPROC_LL_LOCKLESS_TRUNC, LPROCFS_TYPE_REGS, "lockless_truncate"},
668 { LPROC_LL_FLOCK, LPROCFS_TYPE_REGS, "flock" },
669 { LPROC_LL_GETATTR, LPROCFS_TYPE_REGS, "getattr" },
670 /* special inode operation */
671 { LPROC_LL_STAFS, LPROCFS_TYPE_REGS, "statfs" },
672 { LPROC_LL_ALLOC_INODE, LPROCFS_TYPE_REGS, "alloc_inode" },
673 { LPROC_LL_SETXATTR, LPROCFS_TYPE_REGS, "setxattr" },
674 { LPROC_LL_GETXATTR, LPROCFS_TYPE_REGS, "getxattr" },
675 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_REGS, "listxattr" },
676 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_REGS, "removexattr" },
677 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_REGS, "inode_permission" },
678 { LPROC_LL_DIRECT_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
680 { LPROC_LL_DIRECT_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
682 { LPROC_LL_LOCKLESS_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
683 "lockless_read_bytes" },
684 { LPROC_LL_LOCKLESS_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
685 "lockless_write_bytes" },
689 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
693 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
694 lprocfs_counter_add(sbi->ll_stats, op, count);
695 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
696 sbi->ll_stats_track_id == current->pid)
697 lprocfs_counter_add(sbi->ll_stats, op, count);
698 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
699 sbi->ll_stats_track_id == current->parent->pid)
700 lprocfs_counter_add(sbi->ll_stats, op, count);
701 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
702 sbi->ll_stats_track_id == cfs_curproc_gid())
703 lprocfs_counter_add(sbi->ll_stats, op, count);
705 EXPORT_SYMBOL(ll_stats_ops_tally);
707 static const char *ra_stat_string[] = {
708 [RA_STAT_HIT] = "hits",
709 [RA_STAT_MISS] = "misses",
710 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
711 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
712 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
713 [RA_STAT_FAILED_MATCH] = "failed lock match",
714 [RA_STAT_DISCARDED] = "read but discarded",
715 [RA_STAT_ZERO_LEN] = "zero length file",
716 [RA_STAT_ZERO_WINDOW] = "zero size window",
717 [RA_STAT_EOF] = "read-ahead to EOF",
718 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
719 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
723 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
724 struct super_block *sb, char *osc, char *mdc)
726 struct lprocfs_vars lvars[2];
727 struct lustre_sb_info *lsi = s2lsi(sb);
728 struct ll_sb_info *sbi = ll_s2sbi(sb);
729 struct obd_device *obd;
730 char name[MAX_STRING_SIZE + 1], *ptr;
731 int err, id, len, rc;
734 memset(lvars, 0, sizeof(lvars));
736 name[MAX_STRING_SIZE] = '\0';
737 lvars[0].name = name;
739 LASSERT(sbi != NULL);
740 LASSERT(mdc != NULL);
741 LASSERT(osc != NULL);
744 len = strlen(lsi->lsi_lmd->lmd_profile);
745 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
746 if (ptr && (strcmp(ptr, "-client") == 0))
750 snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
751 lsi->lsi_lmd->lmd_profile, sb);
753 sbi->ll_proc_root = lprocfs_register(name, parent, NULL, NULL);
754 if (IS_ERR(sbi->ll_proc_root)) {
755 err = PTR_ERR(sbi->ll_proc_root);
756 sbi->ll_proc_root = NULL;
760 rc = lprocfs_seq_create(sbi->ll_proc_root, "dump_page_cache", 0444,
761 &vvp_dump_pgcache_file_ops, sbi);
763 CWARN("Error adding the dump_page_cache file\n");
765 rc = lprocfs_seq_create(sbi->ll_proc_root, "extents_stats", 0644,
766 &ll_rw_extents_stats_fops, sbi);
768 CWARN("Error adding the extent_stats file\n");
770 rc = lprocfs_seq_create(sbi->ll_proc_root, "extents_stats_per_process",
771 0644, &ll_rw_extents_stats_pp_fops, sbi);
773 CWARN("Error adding the extents_stats_per_process file\n");
775 rc = lprocfs_seq_create(sbi->ll_proc_root, "offset_stats", 0644,
776 &ll_rw_offset_stats_fops, sbi);
778 CWARN("Error adding the offset_stats file\n");
780 /* File operations stats */
781 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
782 LPROCFS_STATS_FLAG_NONE);
783 if (sbi->ll_stats == NULL)
784 GOTO(out, err = -ENOMEM);
785 /* do counter init */
786 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
787 __u32 type = llite_opcode_table[id].type;
789 if (type & LPROCFS_TYPE_REGS)
791 else if (type & LPROCFS_TYPE_BYTES)
793 else if (type & LPROCFS_TYPE_PAGES)
795 lprocfs_counter_init(sbi->ll_stats,
796 llite_opcode_table[id].opcode,
797 (type & LPROCFS_CNTR_AVGMINMAX),
798 llite_opcode_table[id].opname, ptr);
800 err = lprocfs_register_stats(sbi->ll_proc_root, "stats", sbi->ll_stats);
804 sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
805 LPROCFS_STATS_FLAG_NONE);
806 if (sbi->ll_ra_stats == NULL)
807 GOTO(out, err = -ENOMEM);
809 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
810 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
811 ra_stat_string[id], "pages");
812 err = lprocfs_register_stats(sbi->ll_proc_root, "read_ahead_stats",
818 err = lprocfs_add_vars(sbi->ll_proc_root, lprocfs_llite_obd_vars, sb);
823 obd = class_name2obd(mdc);
825 LASSERT(obd != NULL);
826 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
827 LASSERT(obd->obd_type->typ_name != NULL);
829 snprintf(name, MAX_STRING_SIZE, "%s/common_name",
830 obd->obd_type->typ_name);
831 lvars[0].read_fptr = lprocfs_rd_name;
832 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
836 snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
837 lvars[0].read_fptr = lprocfs_rd_uuid;
838 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
843 obd = class_name2obd(osc);
845 LASSERT(obd != NULL);
846 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
847 LASSERT(obd->obd_type->typ_name != NULL);
849 snprintf(name, MAX_STRING_SIZE, "%s/common_name",
850 obd->obd_type->typ_name);
851 lvars[0].read_fptr = lprocfs_rd_name;
852 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
856 snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
857 lvars[0].read_fptr = lprocfs_rd_uuid;
858 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
861 lprocfs_remove(&sbi->ll_proc_root);
862 lprocfs_free_stats(&sbi->ll_ra_stats);
863 lprocfs_free_stats(&sbi->ll_stats);
868 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi)
870 if (sbi->ll_proc_root) {
871 lprocfs_remove(&sbi->ll_proc_root);
872 lprocfs_free_stats(&sbi->ll_ra_stats);
873 lprocfs_free_stats(&sbi->ll_stats);
876 #undef MAX_STRING_SIZE
878 #define pct(a,b) (b ? a * 100 / b : 0)
880 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
881 struct seq_file *seq, int which)
883 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
884 unsigned long start, end, r, w;
885 char *unitp = "KMGTPEZY";
887 struct per_process_info *pp_info = &io_extents->pp_extents[which];
893 for(i = 0; i < LL_HIST_MAX; i++) {
894 read_tot += pp_info->pp_r_hist.oh_buckets[i];
895 write_tot += pp_info->pp_w_hist.oh_buckets[i];
898 for(i = 0; i < LL_HIST_MAX; i++) {
899 r = pp_info->pp_r_hist.oh_buckets[i];
900 w = pp_info->pp_w_hist.oh_buckets[i];
903 end = 1 << (i + LL_HIST_START - units);
904 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4lu %4lu | "
905 "%14lu %4lu %4lu\n", start, *unitp, end, *unitp,
906 (i == LL_HIST_MAX - 1) ? '+' : ' ',
907 r, pct(r, read_tot), pct(read_cum, read_tot),
908 w, pct(w, write_tot), pct(write_cum, write_tot));
910 if (start == 1<<10) {
915 if (read_cum == read_tot && write_cum == write_tot)
920 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
923 struct ll_sb_info *sbi = seq->private;
924 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
927 do_gettimeofday(&now);
929 if (!sbi->ll_rw_stats_on) {
930 seq_printf(seq, "disabled\n"
931 "write anything in this file to activate, "
932 "then 0 or \"[D/d]isabled\" to deactivate\n");
935 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
936 now.tv_sec, now.tv_usec);
937 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
938 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
939 "extents", "calls", "%", "cum%",
940 "calls", "%", "cum%");
941 spin_lock(&sbi->ll_pp_extent_lock);
942 for(k = 0; k < LL_PROCESS_HIST_MAX; k++) {
943 if(io_extents->pp_extents[k].pid != 0) {
944 seq_printf(seq, "\nPID: %d\n",
945 io_extents->pp_extents[k].pid);
946 ll_display_extents_info(io_extents, seq, k);
949 spin_unlock(&sbi->ll_pp_extent_lock);
953 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
954 const char *buf, size_t len,
957 struct seq_file *seq = file->private_data;
958 struct ll_sb_info *sbi = seq->private;
959 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
961 int value = 1, rc = 0;
963 rc = lprocfs_write_helper(buf, len, &value);
964 if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
965 strcmp(buf, "Disabled") == 0))
969 sbi->ll_rw_stats_on = 0;
971 sbi->ll_rw_stats_on = 1;
973 spin_lock(&sbi->ll_pp_extent_lock);
974 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
975 io_extents->pp_extents[i].pid = 0;
976 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
977 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
979 spin_unlock(&sbi->ll_pp_extent_lock);
983 LPROC_SEQ_FOPS(ll_rw_extents_stats_pp);
985 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
988 struct ll_sb_info *sbi = seq->private;
989 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
991 do_gettimeofday(&now);
993 if (!sbi->ll_rw_stats_on) {
994 seq_printf(seq, "disabled\n"
995 "write anything in this file to activate, "
996 "then 0 or \"[D/d]isabled\" to deactivate\n");
999 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
1000 now.tv_sec, now.tv_usec);
1002 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1003 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1004 "extents", "calls", "%", "cum%",
1005 "calls", "%", "cum%");
1006 spin_lock(&sbi->ll_lock);
1007 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1008 spin_unlock(&sbi->ll_lock);
1013 static ssize_t ll_rw_extents_stats_seq_write(struct file *file, const char *buf,
1014 size_t len, loff_t *off)
1016 struct seq_file *seq = file->private_data;
1017 struct ll_sb_info *sbi = seq->private;
1018 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1020 int value = 1, rc = 0;
1022 rc = lprocfs_write_helper(buf, len, &value);
1023 if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1024 strcmp(buf, "Disabled") == 0))
1028 sbi->ll_rw_stats_on = 0;
1030 sbi->ll_rw_stats_on = 1;
1031 spin_lock(&sbi->ll_pp_extent_lock);
1032 for(i = 0; i <= LL_PROCESS_HIST_MAX; i++)
1034 io_extents->pp_extents[i].pid = 0;
1035 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1036 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1038 spin_unlock(&sbi->ll_pp_extent_lock);
1043 LPROC_SEQ_FOPS(ll_rw_extents_stats);
1045 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1046 struct ll_file_data *file, loff_t pos,
1047 size_t count, int rw)
1050 struct ll_rw_process_info *process;
1051 struct ll_rw_process_info *offset;
1052 int *off_count = &sbi->ll_rw_offset_entry_count;
1053 int *process_count = &sbi->ll_offset_process_count;
1054 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1056 if(!sbi->ll_rw_stats_on)
1058 process = sbi->ll_rw_process_info;
1059 offset = sbi->ll_rw_offset_info;
1061 spin_lock(&sbi->ll_pp_extent_lock);
1062 /* Extent statistics */
1063 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1064 if(io_extents->pp_extents[i].pid == pid) {
1072 sbi->ll_extent_process_count =
1073 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1074 cur = sbi->ll_extent_process_count;
1075 io_extents->pp_extents[cur].pid = pid;
1076 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1077 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1080 for(i = 0; (count >= (1 << LL_HIST_START << i)) &&
1081 (i < (LL_HIST_MAX - 1)); i++);
1083 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1084 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1086 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1087 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1089 spin_unlock(&sbi->ll_pp_extent_lock);
1091 spin_lock(&sbi->ll_process_lock);
1092 /* Offset statistics */
1093 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1094 if (process[i].rw_pid == pid) {
1095 if (process[i].rw_last_file != file) {
1096 process[i].rw_range_start = pos;
1097 process[i].rw_last_file_pos = pos + count;
1098 process[i].rw_smallest_extent = count;
1099 process[i].rw_largest_extent = count;
1100 process[i].rw_offset = 0;
1101 process[i].rw_last_file = file;
1102 spin_unlock(&sbi->ll_process_lock);
1105 if (process[i].rw_last_file_pos != pos) {
1107 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1108 offset[*off_count].rw_op = process[i].rw_op;
1109 offset[*off_count].rw_pid = pid;
1110 offset[*off_count].rw_range_start =
1111 process[i].rw_range_start;
1112 offset[*off_count].rw_range_end =
1113 process[i].rw_last_file_pos;
1114 offset[*off_count].rw_smallest_extent =
1115 process[i].rw_smallest_extent;
1116 offset[*off_count].rw_largest_extent =
1117 process[i].rw_largest_extent;
1118 offset[*off_count].rw_offset =
1119 process[i].rw_offset;
1120 process[i].rw_op = rw;
1121 process[i].rw_range_start = pos;
1122 process[i].rw_smallest_extent = count;
1123 process[i].rw_largest_extent = count;
1124 process[i].rw_offset = pos -
1125 process[i].rw_last_file_pos;
1127 if(process[i].rw_smallest_extent > count)
1128 process[i].rw_smallest_extent = count;
1129 if(process[i].rw_largest_extent < count)
1130 process[i].rw_largest_extent = count;
1131 process[i].rw_last_file_pos = pos + count;
1132 spin_unlock(&sbi->ll_process_lock);
1136 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1137 process[*process_count].rw_pid = pid;
1138 process[*process_count].rw_op = rw;
1139 process[*process_count].rw_range_start = pos;
1140 process[*process_count].rw_last_file_pos = pos + count;
1141 process[*process_count].rw_smallest_extent = count;
1142 process[*process_count].rw_largest_extent = count;
1143 process[*process_count].rw_offset = 0;
1144 process[*process_count].rw_last_file = file;
1145 spin_unlock(&sbi->ll_process_lock);
1148 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1151 struct ll_sb_info *sbi = seq->private;
1152 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1153 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1156 do_gettimeofday(&now);
1158 if (!sbi->ll_rw_stats_on) {
1159 seq_printf(seq, "disabled\n"
1160 "write anything in this file to activate, "
1161 "then 0 or \"[D/d]isabled\" to deactivate\n");
1164 spin_lock(&sbi->ll_process_lock);
1166 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
1167 now.tv_sec, now.tv_usec);
1168 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1169 "R/W", "PID", "RANGE START", "RANGE END",
1170 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1171 /* We stored the discontiguous offsets here; print them first */
1172 for(i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1173 if (offset[i].rw_pid != 0)
1174 seq_printf(seq,"%3c %10d %14Lu %14Lu %17lu %17lu %14Lu",
1175 offset[i].rw_op ? 'W' : 'R',
1177 offset[i].rw_range_start,
1178 offset[i].rw_range_end,
1179 (unsigned long)offset[i].rw_smallest_extent,
1180 (unsigned long)offset[i].rw_largest_extent,
1181 offset[i].rw_offset);
1183 /* Then print the current offsets for each process */
1184 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1185 if (process[i].rw_pid != 0)
1186 seq_printf(seq,"%3c %10d %14Lu %14Lu %17lu %17lu %14Lu",
1187 process[i].rw_op ? 'W' : 'R',
1189 process[i].rw_range_start,
1190 process[i].rw_last_file_pos,
1191 (unsigned long)process[i].rw_smallest_extent,
1192 (unsigned long)process[i].rw_largest_extent,
1193 process[i].rw_offset);
1195 spin_unlock(&sbi->ll_process_lock);
1200 static ssize_t ll_rw_offset_stats_seq_write(struct file *file, const char *buf,
1201 size_t len, loff_t *off)
1203 struct seq_file *seq = file->private_data;
1204 struct ll_sb_info *sbi = seq->private;
1205 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1206 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1207 int value = 1, rc = 0;
1209 rc = lprocfs_write_helper(buf, len, &value);
1211 if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1212 strcmp(buf, "Disabled") == 0))
1216 sbi->ll_rw_stats_on = 0;
1218 sbi->ll_rw_stats_on = 1;
1220 spin_lock(&sbi->ll_process_lock);
1221 sbi->ll_offset_process_count = 0;
1222 sbi->ll_rw_offset_entry_count = 0;
1223 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1224 LL_PROCESS_HIST_MAX);
1225 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1226 LL_OFFSET_HIST_MAX);
1227 spin_unlock(&sbi->ll_process_lock);
1232 LPROC_SEQ_FOPS(ll_rw_offset_stats);
1234 void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars)
1236 lvars->module_vars = NULL;
1237 lvars->obd_vars = lprocfs_llite_obd_vars;
1239 #endif /* LPROCFS */