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 (c) 2002, 2010, Oracle and/or its affiliates. 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,
64 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
68 rc = snprintf(page, count, "%u\n", osfs.os_bsize);
74 static int ll_rd_kbytestotal(char *page, char **start, off_t off, int count,
77 struct super_block *sb = (struct super_block *)data;
78 struct obd_statfs osfs;
82 rc = ll_statfs_internal(sb, &osfs,
83 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
86 __u32 blk_size = osfs.os_bsize >> 10;
87 __u64 result = osfs.os_blocks;
89 while (blk_size >>= 1)
93 rc = snprintf(page, count, LPU64"\n", result);
99 static int ll_rd_kbytesfree(char *page, char **start, off_t off, int count,
100 int *eof, void *data)
102 struct super_block *sb = (struct super_block *)data;
103 struct obd_statfs osfs;
107 rc = ll_statfs_internal(sb, &osfs,
108 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
111 __u32 blk_size = osfs.os_bsize >> 10;
112 __u64 result = osfs.os_bfree;
114 while (blk_size >>= 1)
118 rc = snprintf(page, count, LPU64"\n", result);
123 static int ll_rd_kbytesavail(char *page, char **start, off_t off, int count,
124 int *eof, void *data)
126 struct super_block *sb = (struct super_block *)data;
127 struct obd_statfs osfs;
131 rc = ll_statfs_internal(sb, &osfs,
132 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
135 __u32 blk_size = osfs.os_bsize >> 10;
136 __u64 result = osfs.os_bavail;
138 while (blk_size >>= 1)
142 rc = snprintf(page, count, LPU64"\n", result);
147 static int ll_rd_filestotal(char *page, char **start, off_t off, int count,
148 int *eof, void *data)
150 struct super_block *sb = (struct super_block *)data;
151 struct obd_statfs osfs;
155 rc = ll_statfs_internal(sb, &osfs,
156 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
160 rc = snprintf(page, count, LPU64"\n", osfs.os_files);
165 static int ll_rd_filesfree(char *page, char **start, off_t off, int count,
166 int *eof, void *data)
168 struct super_block *sb = (struct super_block *)data;
169 struct obd_statfs osfs;
173 rc = ll_statfs_internal(sb, &osfs,
174 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
178 rc = snprintf(page, count, LPU64"\n", osfs.os_ffree);
184 static int ll_rd_client_type(char *page, char **start, off_t off, int count,
185 int *eof, void *data)
187 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)data);
190 LASSERT(sbi != NULL);
193 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
194 rc = snprintf(page, count, "remote client\n");
196 rc = snprintf(page, count, "local client\n");
201 static int ll_rd_fstype(char *page, char **start, off_t off, int count,
202 int *eof, void *data)
204 struct super_block *sb = (struct super_block*)data;
208 return snprintf(page, count, "%s\n", sb->s_type->name);
211 static int ll_rd_sb_uuid(char *page, char **start, off_t off, int count,
212 int *eof, void *data)
214 struct super_block *sb = (struct super_block *)data;
218 return snprintf(page, count, "%s\n", ll_s2sbi(sb)->ll_sb_uuid.uuid);
221 static int ll_rd_site_stats(char *page, char **start, off_t off,
222 int count, int *eof, void *data)
224 struct super_block *sb = data;
227 * See description of statistical counters in struct cl_site, and
230 return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site),
234 static int ll_rd_max_readahead_mb(char *page, char **start, off_t off,
235 int count, int *eof, void *data)
237 struct super_block *sb = data;
238 struct ll_sb_info *sbi = ll_s2sbi(sb);
242 cfs_spin_lock(&sbi->ll_lock);
243 pages_number = sbi->ll_ra_info.ra_max_pages;
244 cfs_spin_unlock(&sbi->ll_lock);
246 mult = 1 << (20 - PAGE_CACHE_SHIFT);
247 return lprocfs_read_frac_helper(page, count, pages_number, mult);
250 static int ll_wr_max_readahead_mb(struct file *file, const char *buffer,
251 unsigned long count, void *data)
253 struct super_block *sb = data;
254 struct ll_sb_info *sbi = ll_s2sbi(sb);
255 int mult, rc, pages_number;
257 mult = 1 << (20 - CFS_PAGE_SHIFT);
258 rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
262 if (pages_number < 0 || pages_number > cfs_num_physpages / 2) {
263 CERROR("can't set file readahead more than %lu MB\n",
264 cfs_num_physpages >> (20 - CFS_PAGE_SHIFT + 1)); /*1/2 of RAM*/
268 cfs_spin_lock(&sbi->ll_lock);
269 sbi->ll_ra_info.ra_max_pages = pages_number;
270 cfs_spin_unlock(&sbi->ll_lock);
275 static int ll_rd_max_readahead_per_file_mb(char *page, char **start, off_t off,
276 int count, int *eof, void *data)
278 struct super_block *sb = data;
279 struct ll_sb_info *sbi = ll_s2sbi(sb);
283 cfs_spin_lock(&sbi->ll_lock);
284 pages_number = sbi->ll_ra_info.ra_max_pages_per_file;
285 cfs_spin_unlock(&sbi->ll_lock);
287 mult = 1 << (20 - CFS_PAGE_SHIFT);
288 return lprocfs_read_frac_helper(page, count, pages_number, mult);
291 static int ll_wr_max_readahead_per_file_mb(struct file *file, const char *buffer,
292 unsigned long count, void *data)
294 struct super_block *sb = data;
295 struct ll_sb_info *sbi = ll_s2sbi(sb);
296 int mult, rc, pages_number;
298 mult = 1 << (20 - CFS_PAGE_SHIFT);
299 rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
303 if (pages_number < 0 ||
304 pages_number > sbi->ll_ra_info.ra_max_pages) {
305 CERROR("can't set file readahead more than"
306 "max_read_ahead_mb %lu MB\n",
307 sbi->ll_ra_info.ra_max_pages);
311 cfs_spin_lock(&sbi->ll_lock);
312 sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
313 cfs_spin_unlock(&sbi->ll_lock);
318 static int ll_rd_max_read_ahead_whole_mb(char *page, char **start, off_t off,
319 int count, int *eof, void *data)
321 struct super_block *sb = data;
322 struct ll_sb_info *sbi = ll_s2sbi(sb);
326 cfs_spin_lock(&sbi->ll_lock);
327 pages_number = sbi->ll_ra_info.ra_max_read_ahead_whole_pages;
328 cfs_spin_unlock(&sbi->ll_lock);
330 mult = 1 << (20 - CFS_PAGE_SHIFT);
331 return lprocfs_read_frac_helper(page, count, pages_number, mult);
334 static int ll_wr_max_read_ahead_whole_mb(struct file *file, const char *buffer,
335 unsigned long count, void *data)
337 struct super_block *sb = data;
338 struct ll_sb_info *sbi = ll_s2sbi(sb);
339 int mult, rc, pages_number;
341 mult = 1 << (20 - CFS_PAGE_SHIFT);
342 rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
346 /* Cap this at the current max readahead window size, the readahead
347 * algorithm does this anyway so it's pointless to set it larger. */
348 if (pages_number < 0 ||
349 pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
350 CERROR("can't set max_read_ahead_whole_mb more than "
351 "max_read_ahead_per_file_mb: %lu\n",
352 sbi->ll_ra_info.ra_max_pages_per_file >> (20 - CFS_PAGE_SHIFT));
356 cfs_spin_lock(&sbi->ll_lock);
357 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
358 cfs_spin_unlock(&sbi->ll_lock);
363 static int ll_rd_max_cached_mb(char *page, char **start, off_t off,
364 int count, int *eof, void *data)
366 struct super_block *sb = data;
367 struct ll_sb_info *sbi = ll_s2sbi(sb);
371 cfs_spin_lock(&sbi->ll_lock);
372 pages_number = sbi->ll_async_page_max;
373 cfs_spin_unlock(&sbi->ll_lock);
375 mult = 1 << (20 - CFS_PAGE_SHIFT);
376 return lprocfs_read_frac_helper(page, count, pages_number, mult);;
379 static int ll_wr_max_cached_mb(struct file *file, const char *buffer,
380 unsigned long count, void *data)
382 struct super_block *sb = data;
383 struct ll_sb_info *sbi = ll_s2sbi(sb);
384 int mult, rc, pages_number;
386 mult = 1 << (20 - CFS_PAGE_SHIFT);
387 rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
391 if (pages_number < 0 || pages_number > cfs_num_physpages) {
392 CERROR("can't set max cache more than %lu MB\n",
393 cfs_num_physpages >> (20 - CFS_PAGE_SHIFT));
397 cfs_spin_lock(&sbi->ll_lock);
398 sbi->ll_async_page_max = pages_number ;
399 cfs_spin_unlock(&sbi->ll_lock);
402 /* Not set up yet, don't call llap_shrink_cache */
408 static int ll_rd_checksum(char *page, char **start, off_t off,
409 int count, int *eof, void *data)
411 struct super_block *sb = data;
412 struct ll_sb_info *sbi = ll_s2sbi(sb);
414 return snprintf(page, count, "%u\n",
415 (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
418 static int ll_wr_checksum(struct file *file, const char *buffer,
419 unsigned long count, void *data)
421 struct super_block *sb = data;
422 struct ll_sb_info *sbi = ll_s2sbi(sb);
429 rc = lprocfs_write_helper(buffer, count, &val);
433 sbi->ll_flags |= LL_SBI_CHECKSUM;
435 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
437 rc = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
438 KEY_CHECKSUM, sizeof(val), &val, NULL);
440 CWARN("Failed to set OSC checksum flags: %d\n", rc);
445 static int ll_rd_max_rw_chunk(char *page, char **start, off_t off,
446 int count, int *eof, void *data)
448 struct super_block *sb = data;
450 return snprintf(page, count, "%lu\n", ll_s2sbi(sb)->ll_max_rw_chunk);
453 static int ll_wr_max_rw_chunk(struct file *file, const char *buffer,
454 unsigned long count, void *data)
456 struct super_block *sb = data;
459 rc = lprocfs_write_helper(buffer, count, &val);
462 ll_s2sbi(sb)->ll_max_rw_chunk = val;
466 static int ll_rd_track_id(char *page, int count, void *data,
467 enum stats_track_type type)
469 struct super_block *sb = data;
471 if (ll_s2sbi(sb)->ll_stats_track_type == type) {
472 return snprintf(page, count, "%d\n",
473 ll_s2sbi(sb)->ll_stats_track_id);
475 } else if (ll_s2sbi(sb)->ll_stats_track_type == STATS_TRACK_ALL) {
476 return snprintf(page, count, "0 (all)\n");
478 return snprintf(page, count, "untracked\n");
482 static int ll_wr_track_id(const char *buffer, unsigned long count, void *data,
483 enum stats_track_type type)
485 struct super_block *sb = data;
488 rc = lprocfs_write_helper(buffer, count, &pid);
491 ll_s2sbi(sb)->ll_stats_track_id = pid;
493 ll_s2sbi(sb)->ll_stats_track_type = STATS_TRACK_ALL;
495 ll_s2sbi(sb)->ll_stats_track_type = type;
496 lprocfs_clear_stats(ll_s2sbi(sb)->ll_stats);
500 static int ll_rd_track_pid(char *page, char **start, off_t off,
501 int count, int *eof, void *data)
503 return (ll_rd_track_id(page, count, data, STATS_TRACK_PID));
506 static int ll_wr_track_pid(struct file *file, const char *buffer,
507 unsigned long count, void *data)
509 return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PID));
512 static int ll_rd_track_ppid(char *page, char **start, off_t off,
513 int count, int *eof, void *data)
515 return (ll_rd_track_id(page, count, data, STATS_TRACK_PPID));
518 static int ll_wr_track_ppid(struct file *file, const char *buffer,
519 unsigned long count, void *data)
521 return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PPID));
524 static int ll_rd_track_gid(char *page, char **start, off_t off,
525 int count, int *eof, void *data)
527 return (ll_rd_track_id(page, count, data, STATS_TRACK_GID));
530 static int ll_wr_track_gid(struct file *file, const char *buffer,
531 unsigned long count, void *data)
533 return (ll_wr_track_id(buffer, count, data, STATS_TRACK_GID));
536 static int ll_rd_statahead_max(char *page, char **start, off_t off,
537 int count, int *eof, void *data)
539 struct super_block *sb = data;
540 struct ll_sb_info *sbi = ll_s2sbi(sb);
542 return snprintf(page, count, "%u\n", sbi->ll_sa_max);
545 static int ll_wr_statahead_max(struct file *file, const char *buffer,
546 unsigned long count, void *data)
548 struct super_block *sb = data;
549 struct ll_sb_info *sbi = ll_s2sbi(sb);
552 rc = lprocfs_write_helper(buffer, count, &val);
556 if (val >= 0 && val <= LL_SA_RPC_MAX)
557 sbi->ll_sa_max = val;
559 CERROR("Bad statahead_max value %d. Valid values are in the "
560 "range [0, %d]\n", val, LL_SA_RPC_MAX);
565 static int ll_rd_statahead_stats(char *page, char **start, off_t off,
566 int count, int *eof, void *data)
568 struct super_block *sb = data;
569 struct ll_sb_info *sbi = ll_s2sbi(sb);
571 return snprintf(page, count,
572 "statahead total: %u\n"
573 "statahead wrong: %u\n",
574 atomic_read(&sbi->ll_sa_total),
575 atomic_read(&sbi->ll_sa_wrong));
578 static int ll_rd_lazystatfs(char *page, char **start, off_t off,
579 int count, int *eof, void *data)
581 struct super_block *sb = data;
582 struct ll_sb_info *sbi = ll_s2sbi(sb);
584 return snprintf(page, count, "%u\n",
585 (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
588 static int ll_wr_lazystatfs(struct file *file, const char *buffer,
589 unsigned long count, void *data)
591 struct super_block *sb = data;
592 struct ll_sb_info *sbi = ll_s2sbi(sb);
595 rc = lprocfs_write_helper(buffer, count, &val);
600 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
602 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
607 static struct lprocfs_vars lprocfs_llite_obd_vars[] = {
608 { "uuid", ll_rd_sb_uuid, 0, 0 },
609 //{ "mntpt_path", ll_rd_path, 0, 0 },
610 { "fstype", ll_rd_fstype, 0, 0 },
611 { "site", ll_rd_site_stats, 0, 0 },
612 { "blocksize", ll_rd_blksize, 0, 0 },
613 { "kbytestotal", ll_rd_kbytestotal, 0, 0 },
614 { "kbytesfree", ll_rd_kbytesfree, 0, 0 },
615 { "kbytesavail", ll_rd_kbytesavail, 0, 0 },
616 { "filestotal", ll_rd_filestotal, 0, 0 },
617 { "filesfree", ll_rd_filesfree, 0, 0 },
618 { "client_type", ll_rd_client_type, 0, 0 },
619 //{ "filegroups", lprocfs_rd_filegroups, 0, 0 },
620 { "max_read_ahead_mb", ll_rd_max_readahead_mb,
621 ll_wr_max_readahead_mb, 0 },
622 { "max_read_ahead_per_file_mb", ll_rd_max_readahead_per_file_mb,
623 ll_wr_max_readahead_per_file_mb, 0 },
624 { "max_read_ahead_whole_mb", ll_rd_max_read_ahead_whole_mb,
625 ll_wr_max_read_ahead_whole_mb, 0 },
626 { "max_cached_mb", ll_rd_max_cached_mb, ll_wr_max_cached_mb, 0 },
627 { "checksum_pages", ll_rd_checksum, ll_wr_checksum, 0 },
628 { "max_rw_chunk", ll_rd_max_rw_chunk, ll_wr_max_rw_chunk, 0 },
629 { "stats_track_pid", ll_rd_track_pid, ll_wr_track_pid, 0 },
630 { "stats_track_ppid", ll_rd_track_ppid, ll_wr_track_ppid, 0 },
631 { "stats_track_gid", ll_rd_track_gid, ll_wr_track_gid, 0 },
632 { "statahead_max", ll_rd_statahead_max, ll_wr_statahead_max, 0 },
633 { "statahead_stats", ll_rd_statahead_stats, 0, 0 },
634 { "lazystatfs", ll_rd_lazystatfs, ll_wr_lazystatfs, 0 },
638 #define MAX_STRING_SIZE 128
640 struct llite_file_opcode {
644 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
646 { LPROC_LL_DIRTY_HITS, LPROCFS_TYPE_REGS, "dirty_pages_hits" },
647 { LPROC_LL_DIRTY_MISSES, LPROCFS_TYPE_REGS, "dirty_pages_misses" },
648 { LPROC_LL_WB_WRITEPAGE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
649 "writeback_from_writepage" },
650 { LPROC_LL_WB_PRESSURE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
651 "writeback_from_pressure" },
652 { LPROC_LL_WB_OK, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
653 "writeback_ok_pages" },
654 { LPROC_LL_WB_FAIL, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
655 "writeback_failed_pages" },
656 { LPROC_LL_READ_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
658 { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
660 { LPROC_LL_BRW_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
662 { LPROC_LL_BRW_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
664 { LPROC_LL_OSC_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
666 { LPROC_LL_OSC_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
669 { LPROC_LL_IOCTL, LPROCFS_TYPE_REGS, "ioctl" },
670 { LPROC_LL_OPEN, LPROCFS_TYPE_REGS, "open" },
671 { LPROC_LL_RELEASE, LPROCFS_TYPE_REGS, "close" },
672 { LPROC_LL_MAP, LPROCFS_TYPE_REGS, "mmap" },
673 { LPROC_LL_LLSEEK, LPROCFS_TYPE_REGS, "seek" },
674 { LPROC_LL_FSYNC, LPROCFS_TYPE_REGS, "fsync" },
675 /* inode operation */
676 { LPROC_LL_SETATTR, LPROCFS_TYPE_REGS, "setattr" },
677 { LPROC_LL_TRUNC, LPROCFS_TYPE_REGS, "truncate" },
678 { LPROC_LL_LOCKLESS_TRUNC, LPROCFS_TYPE_REGS, "lockless_truncate"},
679 { LPROC_LL_FLOCK, LPROCFS_TYPE_REGS, "flock" },
680 { LPROC_LL_GETATTR, LPROCFS_TYPE_REGS, "getattr" },
681 /* special inode operation */
682 { LPROC_LL_STAFS, LPROCFS_TYPE_REGS, "statfs" },
683 { LPROC_LL_ALLOC_INODE, LPROCFS_TYPE_REGS, "alloc_inode" },
684 { LPROC_LL_SETXATTR, LPROCFS_TYPE_REGS, "setxattr" },
685 { LPROC_LL_GETXATTR, LPROCFS_TYPE_REGS, "getxattr" },
686 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_REGS, "listxattr" },
687 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_REGS, "removexattr" },
688 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_REGS, "inode_permission" },
689 { LPROC_LL_DIRECT_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
691 { LPROC_LL_DIRECT_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
693 { LPROC_LL_LOCKLESS_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
694 "lockless_read_bytes" },
695 { LPROC_LL_LOCKLESS_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
696 "lockless_write_bytes" },
700 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
704 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
705 lprocfs_counter_add(sbi->ll_stats, op, count);
706 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
707 sbi->ll_stats_track_id == current->pid)
708 lprocfs_counter_add(sbi->ll_stats, op, count);
709 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
710 sbi->ll_stats_track_id == current->parent->pid)
711 lprocfs_counter_add(sbi->ll_stats, op, count);
712 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
713 sbi->ll_stats_track_id == cfs_curproc_gid())
714 lprocfs_counter_add(sbi->ll_stats, op, count);
716 EXPORT_SYMBOL(ll_stats_ops_tally);
718 static const char *ra_stat_string[] = {
719 [RA_STAT_HIT] = "hits",
720 [RA_STAT_MISS] = "misses",
721 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
722 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
723 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
724 [RA_STAT_FAILED_MATCH] = "failed lock match",
725 [RA_STAT_DISCARDED] = "read but discarded",
726 [RA_STAT_ZERO_LEN] = "zero length file",
727 [RA_STAT_ZERO_WINDOW] = "zero size window",
728 [RA_STAT_EOF] = "read-ahead to EOF",
729 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
730 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
734 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
735 struct super_block *sb, char *osc, char *mdc)
737 struct lprocfs_vars lvars[2];
738 struct lustre_sb_info *lsi = s2lsi(sb);
739 struct ll_sb_info *sbi = ll_s2sbi(sb);
740 struct obd_device *obd;
741 char name[MAX_STRING_SIZE + 1], *ptr;
742 int err, id, len, rc;
745 memset(lvars, 0, sizeof(lvars));
747 name[MAX_STRING_SIZE] = '\0';
748 lvars[0].name = name;
750 LASSERT(sbi != NULL);
751 LASSERT(mdc != NULL);
752 LASSERT(osc != NULL);
755 len = strlen(lsi->lsi_lmd->lmd_profile);
756 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
757 if (ptr && (strcmp(ptr, "-client") == 0))
761 snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
762 lsi->lsi_lmd->lmd_profile, sb);
764 sbi->ll_proc_root = lprocfs_register(name, parent, NULL, NULL);
765 if (IS_ERR(sbi->ll_proc_root)) {
766 err = PTR_ERR(sbi->ll_proc_root);
767 sbi->ll_proc_root = NULL;
771 rc = lprocfs_seq_create(sbi->ll_proc_root, "dump_page_cache", 0444,
772 &vvp_dump_pgcache_file_ops, sbi);
774 CWARN("Error adding the dump_page_cache file\n");
776 rc = lprocfs_seq_create(sbi->ll_proc_root, "extents_stats", 0644,
777 &ll_rw_extents_stats_fops, sbi);
779 CWARN("Error adding the extent_stats file\n");
781 rc = lprocfs_seq_create(sbi->ll_proc_root, "extents_stats_per_process",
782 0644, &ll_rw_extents_stats_pp_fops, sbi);
784 CWARN("Error adding the extents_stats_per_process file\n");
786 rc = lprocfs_seq_create(sbi->ll_proc_root, "offset_stats", 0644,
787 &ll_rw_offset_stats_fops, sbi);
789 CWARN("Error adding the offset_stats file\n");
791 /* File operations stats */
792 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
793 LPROCFS_STATS_FLAG_NONE);
794 if (sbi->ll_stats == NULL)
795 GOTO(out, err = -ENOMEM);
796 /* do counter init */
797 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
798 __u32 type = llite_opcode_table[id].type;
800 if (type & LPROCFS_TYPE_REGS)
802 else if (type & LPROCFS_TYPE_BYTES)
804 else if (type & LPROCFS_TYPE_PAGES)
806 lprocfs_counter_init(sbi->ll_stats,
807 llite_opcode_table[id].opcode,
808 (type & LPROCFS_CNTR_AVGMINMAX),
809 llite_opcode_table[id].opname, ptr);
811 err = lprocfs_register_stats(sbi->ll_proc_root, "stats", sbi->ll_stats);
815 sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
816 LPROCFS_STATS_FLAG_NONE);
817 if (sbi->ll_ra_stats == NULL)
818 GOTO(out, err = -ENOMEM);
820 for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
821 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
822 ra_stat_string[id], "pages");
823 err = lprocfs_register_stats(sbi->ll_proc_root, "read_ahead_stats",
829 err = lprocfs_add_vars(sbi->ll_proc_root, lprocfs_llite_obd_vars, sb);
834 obd = class_name2obd(mdc);
836 LASSERT(obd != NULL);
837 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
838 LASSERT(obd->obd_type->typ_name != NULL);
840 snprintf(name, MAX_STRING_SIZE, "%s/common_name",
841 obd->obd_type->typ_name);
842 lvars[0].read_fptr = lprocfs_rd_name;
843 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
847 snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
848 lvars[0].read_fptr = lprocfs_rd_uuid;
849 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
854 obd = class_name2obd(osc);
856 LASSERT(obd != NULL);
857 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
858 LASSERT(obd->obd_type->typ_name != NULL);
860 snprintf(name, MAX_STRING_SIZE, "%s/common_name",
861 obd->obd_type->typ_name);
862 lvars[0].read_fptr = lprocfs_rd_name;
863 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
867 snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
868 lvars[0].read_fptr = lprocfs_rd_uuid;
869 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
872 lprocfs_remove(&sbi->ll_proc_root);
873 lprocfs_free_stats(&sbi->ll_ra_stats);
874 lprocfs_free_stats(&sbi->ll_stats);
879 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi)
881 if (sbi->ll_proc_root) {
882 lprocfs_remove(&sbi->ll_proc_root);
883 lprocfs_free_stats(&sbi->ll_ra_stats);
884 lprocfs_free_stats(&sbi->ll_stats);
887 #undef MAX_STRING_SIZE
889 #define pct(a,b) (b ? a * 100 / b : 0)
891 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
892 struct seq_file *seq, int which)
894 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
895 unsigned long start, end, r, w;
896 char *unitp = "KMGTPEZY";
898 struct per_process_info *pp_info = &io_extents->pp_extents[which];
904 for(i = 0; i < LL_HIST_MAX; i++) {
905 read_tot += pp_info->pp_r_hist.oh_buckets[i];
906 write_tot += pp_info->pp_w_hist.oh_buckets[i];
909 for(i = 0; i < LL_HIST_MAX; i++) {
910 r = pp_info->pp_r_hist.oh_buckets[i];
911 w = pp_info->pp_w_hist.oh_buckets[i];
914 end = 1 << (i + LL_HIST_START - units);
915 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4lu %4lu | "
916 "%14lu %4lu %4lu\n", start, *unitp, end, *unitp,
917 (i == LL_HIST_MAX - 1) ? '+' : ' ',
918 r, pct(r, read_tot), pct(read_cum, read_tot),
919 w, pct(w, write_tot), pct(write_cum, write_tot));
921 if (start == 1<<10) {
926 if (read_cum == read_tot && write_cum == write_tot)
931 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
934 struct ll_sb_info *sbi = seq->private;
935 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
938 cfs_gettimeofday(&now);
940 if (!sbi->ll_rw_stats_on) {
941 seq_printf(seq, "disabled\n"
942 "write anything in this file to activate, "
943 "then 0 or \"[D/d]isabled\" to deactivate\n");
946 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
947 now.tv_sec, now.tv_usec);
948 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
949 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
950 "extents", "calls", "%", "cum%",
951 "calls", "%", "cum%");
952 cfs_spin_lock(&sbi->ll_pp_extent_lock);
953 for(k = 0; k < LL_PROCESS_HIST_MAX; k++) {
954 if(io_extents->pp_extents[k].pid != 0) {
955 seq_printf(seq, "\nPID: %d\n",
956 io_extents->pp_extents[k].pid);
957 ll_display_extents_info(io_extents, seq, k);
960 cfs_spin_unlock(&sbi->ll_pp_extent_lock);
964 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
965 const char *buf, size_t len,
968 struct seq_file *seq = file->private_data;
969 struct ll_sb_info *sbi = seq->private;
970 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
972 int value = 1, rc = 0;
974 rc = lprocfs_write_helper(buf, len, &value);
975 if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
976 strcmp(buf, "Disabled") == 0))
980 sbi->ll_rw_stats_on = 0;
982 sbi->ll_rw_stats_on = 1;
984 cfs_spin_lock(&sbi->ll_pp_extent_lock);
985 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
986 io_extents->pp_extents[i].pid = 0;
987 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
988 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
990 cfs_spin_unlock(&sbi->ll_pp_extent_lock);
994 LPROC_SEQ_FOPS(ll_rw_extents_stats_pp);
996 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
999 struct ll_sb_info *sbi = seq->private;
1000 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1002 cfs_gettimeofday(&now);
1004 if (!sbi->ll_rw_stats_on) {
1005 seq_printf(seq, "disabled\n"
1006 "write anything in this file to activate, "
1007 "then 0 or \"[D/d]isabled\" to deactivate\n");
1010 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
1011 now.tv_sec, now.tv_usec);
1013 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1014 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1015 "extents", "calls", "%", "cum%",
1016 "calls", "%", "cum%");
1017 cfs_spin_lock(&sbi->ll_lock);
1018 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1019 cfs_spin_unlock(&sbi->ll_lock);
1024 static ssize_t ll_rw_extents_stats_seq_write(struct file *file, const char *buf,
1025 size_t len, loff_t *off)
1027 struct seq_file *seq = file->private_data;
1028 struct ll_sb_info *sbi = seq->private;
1029 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1031 int value = 1, rc = 0;
1033 rc = lprocfs_write_helper(buf, len, &value);
1034 if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1035 strcmp(buf, "Disabled") == 0))
1039 sbi->ll_rw_stats_on = 0;
1041 sbi->ll_rw_stats_on = 1;
1042 cfs_spin_lock(&sbi->ll_pp_extent_lock);
1043 for(i = 0; i <= LL_PROCESS_HIST_MAX; i++)
1045 io_extents->pp_extents[i].pid = 0;
1046 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1047 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1049 cfs_spin_unlock(&sbi->ll_pp_extent_lock);
1054 LPROC_SEQ_FOPS(ll_rw_extents_stats);
1056 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1057 struct ll_file_data *file, loff_t pos,
1058 size_t count, int rw)
1061 struct ll_rw_process_info *process;
1062 struct ll_rw_process_info *offset;
1063 int *off_count = &sbi->ll_rw_offset_entry_count;
1064 int *process_count = &sbi->ll_offset_process_count;
1065 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1067 if(!sbi->ll_rw_stats_on)
1069 process = sbi->ll_rw_process_info;
1070 offset = sbi->ll_rw_offset_info;
1072 cfs_spin_lock(&sbi->ll_pp_extent_lock);
1073 /* Extent statistics */
1074 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1075 if(io_extents->pp_extents[i].pid == pid) {
1083 sbi->ll_extent_process_count =
1084 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1085 cur = sbi->ll_extent_process_count;
1086 io_extents->pp_extents[cur].pid = pid;
1087 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1088 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1091 for(i = 0; (count >= (1 << LL_HIST_START << i)) &&
1092 (i < (LL_HIST_MAX - 1)); i++);
1094 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1095 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1097 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1098 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1100 cfs_spin_unlock(&sbi->ll_pp_extent_lock);
1102 cfs_spin_lock(&sbi->ll_process_lock);
1103 /* Offset statistics */
1104 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1105 if (process[i].rw_pid == pid) {
1106 if (process[i].rw_last_file != file) {
1107 process[i].rw_range_start = pos;
1108 process[i].rw_last_file_pos = pos + count;
1109 process[i].rw_smallest_extent = count;
1110 process[i].rw_largest_extent = count;
1111 process[i].rw_offset = 0;
1112 process[i].rw_last_file = file;
1113 cfs_spin_unlock(&sbi->ll_process_lock);
1116 if (process[i].rw_last_file_pos != pos) {
1118 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1119 offset[*off_count].rw_op = process[i].rw_op;
1120 offset[*off_count].rw_pid = pid;
1121 offset[*off_count].rw_range_start =
1122 process[i].rw_range_start;
1123 offset[*off_count].rw_range_end =
1124 process[i].rw_last_file_pos;
1125 offset[*off_count].rw_smallest_extent =
1126 process[i].rw_smallest_extent;
1127 offset[*off_count].rw_largest_extent =
1128 process[i].rw_largest_extent;
1129 offset[*off_count].rw_offset =
1130 process[i].rw_offset;
1131 process[i].rw_op = rw;
1132 process[i].rw_range_start = pos;
1133 process[i].rw_smallest_extent = count;
1134 process[i].rw_largest_extent = count;
1135 process[i].rw_offset = pos -
1136 process[i].rw_last_file_pos;
1138 if(process[i].rw_smallest_extent > count)
1139 process[i].rw_smallest_extent = count;
1140 if(process[i].rw_largest_extent < count)
1141 process[i].rw_largest_extent = count;
1142 process[i].rw_last_file_pos = pos + count;
1143 cfs_spin_unlock(&sbi->ll_process_lock);
1147 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1148 process[*process_count].rw_pid = pid;
1149 process[*process_count].rw_op = rw;
1150 process[*process_count].rw_range_start = pos;
1151 process[*process_count].rw_last_file_pos = pos + count;
1152 process[*process_count].rw_smallest_extent = count;
1153 process[*process_count].rw_largest_extent = count;
1154 process[*process_count].rw_offset = 0;
1155 process[*process_count].rw_last_file = file;
1156 cfs_spin_unlock(&sbi->ll_process_lock);
1159 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1162 struct ll_sb_info *sbi = seq->private;
1163 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1164 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1167 cfs_gettimeofday(&now);
1169 if (!sbi->ll_rw_stats_on) {
1170 seq_printf(seq, "disabled\n"
1171 "write anything in this file to activate, "
1172 "then 0 or \"[D/d]isabled\" to deactivate\n");
1175 cfs_spin_lock(&sbi->ll_process_lock);
1177 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
1178 now.tv_sec, now.tv_usec);
1179 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1180 "R/W", "PID", "RANGE START", "RANGE END",
1181 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1182 /* We stored the discontiguous offsets here; print them first */
1183 for(i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1184 if (offset[i].rw_pid != 0)
1185 seq_printf(seq,"%3c %10d %14Lu %14Lu %17lu %17lu %14Lu",
1186 offset[i].rw_op ? 'W' : 'R',
1188 offset[i].rw_range_start,
1189 offset[i].rw_range_end,
1190 (unsigned long)offset[i].rw_smallest_extent,
1191 (unsigned long)offset[i].rw_largest_extent,
1192 offset[i].rw_offset);
1194 /* Then print the current offsets for each process */
1195 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1196 if (process[i].rw_pid != 0)
1197 seq_printf(seq,"%3c %10d %14Lu %14Lu %17lu %17lu %14Lu",
1198 process[i].rw_op ? 'W' : 'R',
1200 process[i].rw_range_start,
1201 process[i].rw_last_file_pos,
1202 (unsigned long)process[i].rw_smallest_extent,
1203 (unsigned long)process[i].rw_largest_extent,
1204 process[i].rw_offset);
1206 cfs_spin_unlock(&sbi->ll_process_lock);
1211 static ssize_t ll_rw_offset_stats_seq_write(struct file *file, const char *buf,
1212 size_t len, loff_t *off)
1214 struct seq_file *seq = file->private_data;
1215 struct ll_sb_info *sbi = seq->private;
1216 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1217 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1218 int value = 1, rc = 0;
1220 rc = lprocfs_write_helper(buf, len, &value);
1222 if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1223 strcmp(buf, "Disabled") == 0))
1227 sbi->ll_rw_stats_on = 0;
1229 sbi->ll_rw_stats_on = 1;
1231 cfs_spin_lock(&sbi->ll_process_lock);
1232 sbi->ll_offset_process_count = 0;
1233 sbi->ll_rw_offset_entry_count = 0;
1234 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1235 LL_PROCESS_HIST_MAX);
1236 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1237 LL_OFFSET_HIST_MAX);
1238 cfs_spin_unlock(&sbi->ll_process_lock);
1243 LPROC_SEQ_FOPS(ll_rw_offset_stats);
1245 void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars)
1247 lvars->module_vars = NULL;
1248 lvars->obd_vars = lprocfs_llite_obd_vars;
1250 #endif /* LPROCFS */