1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2002 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22 #define DEBUG_SUBSYSTEM S_LLITE
24 #include <linux/version.h>
25 #include <linux/lustre_lite.h>
26 #include <linux/lprocfs_status.h>
27 #include <linux/seq_file.h>
28 #include <linux/obd_support.h>
30 #include "llite_internal.h"
32 /* /proc/lustre/llite mount point registration */
33 struct proc_dir_entry *proc_lustre_fs_root;
34 struct file_operations llite_dump_pgcache_fops;
35 struct file_operations ll_ra_stats_fops;
38 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
39 struct super_block *sb, char *osc, char *mdc)
43 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi){}
46 long long mnt_instance;
48 static int ll_rd_blksize(char *page, char **start, off_t off, int count,
51 struct super_block *sb = (struct super_block *)data;
52 struct obd_statfs osfs;
56 rc = ll_statfs_internal(sb, &osfs, jiffies - HZ);
59 rc = snprintf(page, count, "%u\n", osfs.os_bsize);
65 static int ll_rd_kbytestotal(char *page, char **start, off_t off, int count,
68 struct super_block *sb = (struct super_block *)data;
69 struct obd_statfs osfs;
73 rc = ll_statfs_internal(sb, &osfs, jiffies - HZ);
75 __u32 blk_size = osfs.os_bsize >> 10;
76 __u64 result = osfs.os_blocks;
78 while (blk_size >>= 1)
82 rc = snprintf(page, count, LPU64"\n", result);
88 static int ll_rd_kbytesfree(char *page, char **start, off_t off, int count,
91 struct super_block *sb = (struct super_block *)data;
92 struct obd_statfs osfs;
96 rc = ll_statfs_internal(sb, &osfs, jiffies - HZ);
98 __u32 blk_size = osfs.os_bsize >> 10;
99 __u64 result = osfs.os_bfree;
101 while (blk_size >>= 1)
105 rc = snprintf(page, count, LPU64"\n", result);
110 static int ll_rd_kbytesavail(char *page, char **start, off_t off, int count,
111 int *eof, void *data)
113 struct super_block *sb = (struct super_block *)data;
114 struct obd_statfs osfs;
118 rc = ll_statfs_internal(sb, &osfs, jiffies - HZ);
120 __u32 blk_size = osfs.os_bsize >> 10;
121 __u64 result = osfs.os_bavail;
123 while (blk_size >>= 1)
127 rc = snprintf(page, count, LPU64"\n", result);
132 static int ll_rd_filestotal(char *page, char **start, off_t off, int count,
133 int *eof, void *data)
135 struct super_block *sb = (struct super_block *)data;
136 struct obd_statfs osfs;
140 rc = ll_statfs_internal(sb, &osfs, jiffies - HZ);
143 rc = snprintf(page, count, LPU64"\n", osfs.os_files);
148 static int ll_rd_filesfree(char *page, char **start, off_t off, int count,
149 int *eof, void *data)
151 struct super_block *sb = (struct super_block *)data;
152 struct obd_statfs osfs;
156 rc = ll_statfs_internal(sb, &osfs, jiffies - HZ);
159 rc = snprintf(page, count, LPU64"\n", osfs.os_ffree);
165 static int ll_rd_fstype(char *page, char **start, off_t off, int count,
166 int *eof, void *data)
168 struct super_block *sb = (struct super_block*)data;
172 return snprintf(page, count, "%s\n", sb->s_type->name);
175 static int ll_rd_sb_uuid(char *page, char **start, off_t off, int count,
176 int *eof, void *data)
178 struct super_block *sb = (struct super_block *)data;
182 return snprintf(page, count, "%s\n", ll_s2sbi(sb)->ll_sb_uuid.uuid);
185 static int ll_rd_max_readahead_mb(char *page, char **start, off_t off,
186 int count, int *eof, void *data)
188 struct super_block *sb = data;
189 struct ll_sb_info *sbi = ll_s2sbi(sb);
192 spin_lock(&sbi->ll_lock);
193 val = sbi->ll_ra_info.ra_max_pages >> (20 - PAGE_CACHE_SHIFT);
194 spin_unlock(&sbi->ll_lock);
196 return snprintf(page, count, "%u\n", val);
199 static int ll_wr_max_readahead_mb(struct file *file, const char *buffer,
200 unsigned long count, void *data)
202 struct super_block *sb = data;
203 struct ll_sb_info *sbi = ll_s2sbi(sb);
206 rc = lprocfs_write_helper(buffer, count, &val);
210 if (val < 0 || val > (num_physpages >> (20 - PAGE_CACHE_SHIFT - 1))) {
211 CERROR("can't set readahead more than %lu MB\n",
212 num_physpages >> (20 - PAGE_CACHE_SHIFT - 1));
216 spin_lock(&sbi->ll_lock);
217 sbi->ll_ra_info.ra_max_pages = val << (20 - PAGE_CACHE_SHIFT);
218 spin_unlock(&sbi->ll_lock);
223 static int ll_rd_max_cached_mb(char *page, char **start, off_t off,
224 int count, int *eof, void *data)
226 struct super_block *sb = data;
227 struct ll_sb_info *sbi = ll_s2sbi(sb);
230 spin_lock(&sbi->ll_lock);
231 val = sbi->ll_async_page_max >> (20 - PAGE_CACHE_SHIFT);
232 spin_unlock(&sbi->ll_lock);
234 return snprintf(page, count, "%u\n", val);
237 static int ll_wr_max_cached_mb(struct file *file, const char *buffer,
238 unsigned long count, void *data)
240 struct super_block *sb = data;
241 struct ll_sb_info *sbi = ll_s2sbi(sb);
244 rc = lprocfs_write_helper(buffer, count, &val);
248 if (val < 0 || val > (num_physpages >> (20 - PAGE_CACHE_SHIFT))) {
249 CERROR("can't set max cache more than %lu MB\n",
250 num_physpages >> (20 - PAGE_CACHE_SHIFT));
254 spin_lock(&sbi->ll_lock);
255 sbi->ll_async_page_max = val << (20 - PAGE_CACHE_SHIFT);
256 spin_unlock(&sbi->ll_lock);
258 if (sbi->ll_async_page_count >= sbi->ll_async_page_max)
259 llap_shrink_cache(sbi, 0);
264 static struct lprocfs_vars lprocfs_obd_vars[] = {
265 { "uuid", ll_rd_sb_uuid, 0, 0 },
266 //{ "mntpt_path", ll_rd_path, 0, 0 },
267 { "fstype", ll_rd_fstype, 0, 0 },
268 { "blocksize", ll_rd_blksize, 0, 0 },
269 { "kbytestotal", ll_rd_kbytestotal, 0, 0 },
270 { "kbytesfree", ll_rd_kbytesfree, 0, 0 },
271 { "kbytesavail", ll_rd_kbytesavail, 0, 0 },
272 { "filestotal", ll_rd_filestotal, 0, 0 },
273 { "filesfree", ll_rd_filesfree, 0, 0 },
274 //{ "filegroups", lprocfs_rd_filegroups, 0, 0 },
275 { "max_read_ahead_mb", ll_rd_max_readahead_mb,
276 ll_wr_max_readahead_mb, 0 },
277 { "max_cached_mb", ll_rd_max_cached_mb, ll_wr_max_cached_mb, 0 },
281 #define MAX_STRING_SIZE 128
283 struct llite_file_opcode {
287 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
289 { LPROC_LL_DIRTY_HITS, LPROCFS_TYPE_REGS, "dirty_pages_hits" },
290 { LPROC_LL_DIRTY_MISSES, LPROCFS_TYPE_REGS, "dirty_pages_misses" },
291 { LPROC_LL_WB_WRITEPAGE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
292 "writeback_from_writepage" },
293 { LPROC_LL_WB_PRESSURE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
294 "writeback_from_pressure" },
295 { LPROC_LL_WB_OK, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
296 "writeback_ok_pages" },
297 { LPROC_LL_WB_FAIL, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
298 "writeback_failed_pages" },
299 { LPROC_LL_READ_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
301 { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
303 { LPROC_LL_BRW_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
305 { LPROC_LL_BRW_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
308 { LPROC_LL_IOCTL, LPROCFS_TYPE_REGS, "ioctl" },
309 { LPROC_LL_OPEN, LPROCFS_TYPE_REGS, "open" },
310 { LPROC_LL_RELEASE, LPROCFS_TYPE_REGS, "close" },
311 { LPROC_LL_MAP, LPROCFS_TYPE_REGS, "mmap" },
312 { LPROC_LL_LLSEEK, LPROCFS_TYPE_REGS, "seek" },
313 { LPROC_LL_FSYNC, LPROCFS_TYPE_REGS, "fsync" },
314 /* inode operation */
315 { LPROC_LL_SETATTR, LPROCFS_TYPE_REGS, "setattr" },
316 { LPROC_LL_TRUNC, LPROCFS_TYPE_REGS, "punch" },
317 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
318 { LPROC_LL_GETATTR, LPROCFS_TYPE_REGS, "getattr" },
320 { LPROC_LL_REVALIDATE, LPROCFS_TYPE_REGS, "getattr" },
322 /* special inode operation */
323 { LPROC_LL_STAFS, LPROCFS_TYPE_REGS, "statfs" },
324 { LPROC_LL_ALLOC_INODE, LPROCFS_TYPE_REGS, "alloc_inode" },
325 { LPROC_LL_DIRECT_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
327 { LPROC_LL_DIRECT_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
332 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
333 struct super_block *sb, char *osc, char *mdc)
335 struct lprocfs_vars lvars[2];
336 struct ll_sb_info *sbi = ll_s2sbi(sb);
337 struct obd_device *obd;
338 char name[MAX_STRING_SIZE + 1];
340 struct lprocfs_stats *svc_stats = NULL;
341 struct proc_dir_entry *entry;
344 memset(lvars, 0, sizeof(lvars));
346 name[MAX_STRING_SIZE] = '\0';
347 lvars[0].name = name;
349 LASSERT(sbi != NULL);
350 LASSERT(mdc != NULL);
351 LASSERT(osc != NULL);
354 snprintf(name, MAX_STRING_SIZE, "fs%llu", mnt_instance);
357 sbi->ll_proc_root = lprocfs_register(name, parent, NULL, NULL);
358 if (IS_ERR(sbi->ll_proc_root)) {
359 err = PTR_ERR(sbi->ll_proc_root);
360 sbi->ll_proc_root = NULL;
364 entry = create_proc_entry("dump_page_cache", 0444, sbi->ll_proc_root);
366 GOTO(out, err = -ENOMEM);
367 entry->proc_fops = &llite_dump_pgcache_fops;
370 entry = create_proc_entry("read_ahead_stats", 0444, sbi->ll_proc_root);
372 GOTO(out, err = -ENOMEM);
373 entry->proc_fops = &ll_ra_stats_fops;
376 svc_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES);
377 if (svc_stats == NULL) {
381 /* do counter init */
382 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
383 __u32 type = llite_opcode_table[id].type;
385 if (type & LPROCFS_TYPE_REGS)
388 if (type & LPROCFS_TYPE_BYTES)
391 if (type & LPROCFS_TYPE_PAGES)
395 lprocfs_counter_init(svc_stats, llite_opcode_table[id].opcode,
396 (type & LPROCFS_CNTR_AVGMINMAX),
397 llite_opcode_table[id].opname, ptr);
399 err = lprocfs_register_stats(sbi->ll_proc_root, "stats", svc_stats);
403 sbi->ll_stats = svc_stats;
404 /* need place to keep svc_stats */
406 /* Static configuration info */
407 err = lprocfs_add_vars(sbi->ll_proc_root, lprocfs_obd_vars, sb);
412 obd = class_name2obd(mdc);
414 LASSERT(obd != NULL);
415 LASSERT(obd->obd_type != NULL);
416 LASSERT(obd->obd_type->typ_name != NULL);
418 snprintf(name, MAX_STRING_SIZE, "%s/common_name",
419 obd->obd_type->typ_name);
420 lvars[0].read_fptr = lprocfs_rd_name;
421 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
425 snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
426 lvars[0].read_fptr = lprocfs_rd_uuid;
427 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
432 obd = class_name2obd(osc);
434 LASSERT(obd != NULL);
435 LASSERT(obd->obd_type != NULL);
436 LASSERT(obd->obd_type->typ_name != NULL);
438 snprintf(name, MAX_STRING_SIZE, "%s/common_name",
439 obd->obd_type->typ_name);
440 lvars[0].read_fptr = lprocfs_rd_name;
441 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
445 snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
446 lvars[0].read_fptr = lprocfs_rd_uuid;
447 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
451 lprocfs_free_stats(svc_stats);
452 if (sbi->ll_proc_root)
453 lprocfs_remove(sbi->ll_proc_root);
458 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi)
460 if (sbi->ll_proc_root) {
461 struct proc_dir_entry *file_stats =
462 lprocfs_srch(sbi->ll_proc_root, "stats");
465 lprocfs_free_stats(sbi->ll_stats);
466 lprocfs_remove(file_stats);
470 #undef MAX_STRING_SIZE
472 #define seq_page_flag(seq, page, flag, has_flags) do { \
473 if (test_bit(PG_##flag, &(page)->flags)) { \
477 seq_putc(seq, '|'); \
478 seq_puts(seq, #flag); \
482 static void *llite_dump_pgcache_seq_start(struct seq_file *seq, loff_t *pos)
484 struct ll_async_page *dummy_llap = seq->private;
486 if (dummy_llap->llap_magic == 2)
492 static int llite_dump_pgcache_seq_show(struct seq_file *seq, void *v)
494 struct ll_async_page *llap, *dummy_llap = seq->private;
495 struct ll_sb_info *sbi = dummy_llap->llap_cookie;
497 /* 2.4 doesn't seem to have SEQ_START_TOKEN, so we implement
498 * it in our own state */
499 if (dummy_llap->llap_magic == 0) {
500 seq_printf(seq, "gener | llap cookie origin wq du | page "
501 "inode index count [ page flags ]\n");
505 spin_lock(&sbi->ll_lock);
507 llap = llite_pglist_next_llap(sbi, &dummy_llap->llap_pglist_item);
510 struct page *page = llap->llap_page;
512 LASSERTF(llap->llap_origin < LLAP__ORIGIN_MAX, "%u\n",
515 seq_printf(seq, "%5lu | %p %p %s %s %s | %p %p %lu %u [",
517 llap, llap->llap_cookie,
518 llap_origins[llap->llap_origin],
519 llap->llap_write_queued ? "wq" : "- ",
520 llap->llap_defer_uptodate ? "du" : "- ",
521 page, page->mapping->host, page->index,
523 seq_page_flag(seq, page, locked, has_flags);
524 seq_page_flag(seq, page, error, has_flags);
525 seq_page_flag(seq, page, referenced, has_flags);
526 seq_page_flag(seq, page, uptodate, has_flags);
527 seq_page_flag(seq, page, dirty, has_flags);
528 seq_page_flag(seq, page, highmem, has_flags);
530 seq_puts(seq, "-]\n");
532 seq_puts(seq, "]\n");
535 spin_unlock(&sbi->ll_lock);
540 static void *llite_dump_pgcache_seq_next(struct seq_file *seq, void *v,
543 struct ll_async_page *llap, *dummy_llap = seq->private;
544 struct ll_sb_info *sbi = dummy_llap->llap_cookie;
546 /* bail if we just displayed the banner */
547 if (dummy_llap->llap_magic == 0) {
548 dummy_llap->llap_magic = 1;
552 /* we've just displayed the llap that is after us in the list.
553 * we advance to a position beyond it, returning null if there
554 * isn't another llap in the list beyond that new position. */
555 spin_lock(&sbi->ll_lock);
556 llap = llite_pglist_next_llap(sbi, &dummy_llap->llap_pglist_item);
557 list_del_init(&dummy_llap->llap_pglist_item);
559 list_add(&dummy_llap->llap_pglist_item,&llap->llap_pglist_item);
560 llap =llite_pglist_next_llap(sbi,&dummy_llap->llap_pglist_item);
562 spin_unlock(&sbi->ll_lock);
566 dummy_llap->llap_magic = 2;
572 static void llite_dump_pgcache_seq_stop(struct seq_file *seq, void *v)
576 struct seq_operations llite_dump_pgcache_seq_sops = {
577 .start = llite_dump_pgcache_seq_start,
578 .stop = llite_dump_pgcache_seq_stop,
579 .next = llite_dump_pgcache_seq_next,
580 .show = llite_dump_pgcache_seq_show,
583 /* we're displaying llaps in a list_head list. we don't want to hold a lock
584 * while we walk the entire list, and we don't want to have to seek into
585 * the right position in the list as an app advances with many syscalls. we
586 * allocate a dummy llap and hang it off file->private. its position in
587 * the list records where the app is currently displaying. this way our
588 * seq .start and .stop don't actually do anything. .next returns null
589 * when the dummy hits the end of the list which eventually leads to .release
590 * where we tear down. this kind of displaying is super-racey, so we put
591 * a generation counter on the list so the output shows when the list
592 * changes between reads.
594 static int llite_dump_pgcache_seq_open(struct inode *inode, struct file *file)
596 struct proc_dir_entry *dp = PDE(inode);
597 struct ll_async_page *dummy_llap;
598 struct seq_file *seq;
599 struct ll_sb_info *sbi = dp->data;
602 OBD_ALLOC_GFP(dummy_llap, sizeof(*dummy_llap), GFP_KERNEL);
603 if (dummy_llap == NULL)
605 dummy_llap->llap_page = NULL;
606 dummy_llap->llap_cookie = sbi;
607 dummy_llap->llap_magic = 0;
609 rc = seq_open(file, &llite_dump_pgcache_seq_sops);
611 OBD_FREE(dummy_llap, sizeof(*dummy_llap));
614 seq = file->private_data;
615 seq->private = dummy_llap;
617 spin_lock(&sbi->ll_lock);
618 list_add(&dummy_llap->llap_pglist_item, &sbi->ll_pglist);
619 spin_unlock(&sbi->ll_lock);
624 static int llite_dump_pgcache_seq_release(struct inode *inode,
627 struct seq_file *seq = file->private_data;
628 struct ll_async_page *dummy_llap = seq->private;
629 struct ll_sb_info *sbi = dummy_llap->llap_cookie;
631 spin_lock(&sbi->ll_lock);
632 if (!list_empty(&dummy_llap->llap_pglist_item))
633 list_del_init(&dummy_llap->llap_pglist_item);
634 spin_unlock(&sbi->ll_lock);
635 OBD_FREE(dummy_llap, sizeof(*dummy_llap));
637 return seq_release(inode, file);
640 struct file_operations llite_dump_pgcache_fops = {
641 .owner = THIS_MODULE,
642 .open = llite_dump_pgcache_seq_open,
644 .release = llite_dump_pgcache_seq_release,
647 static int ll_ra_stats_seq_show(struct seq_file *seq, void *v)
650 struct ll_sb_info *sbi = seq->private;
651 struct ll_ra_info *ra = &sbi->ll_ra_info;
653 static char *ra_stat_strings[] = {
654 [RA_STAT_HIT] = "hits",
655 [RA_STAT_MISS] = "misses",
656 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
657 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
658 [RA_STAT_FAILED_MATCH] = "failed lock match",
659 [RA_STAT_DISCARDED] = "read but discarded",
660 [RA_STAT_ZERO_LEN] = "zero length file",
661 [RA_STAT_ZERO_WINDOW] = "zero size window",
662 [RA_STAT_EOF] = "read-ahead to EOF",
663 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
666 do_gettimeofday(&now);
668 spin_lock(&sbi->ll_lock);
670 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
671 now.tv_sec, now.tv_usec);
672 seq_printf(seq, "pending issued pages: %lu\n",
675 for(i = 0; i < _NR_RA_STAT; i++)
676 seq_printf(seq, "%-25s %lu\n", ra_stat_strings[i],
679 spin_unlock(&sbi->ll_lock);
684 static void *ll_ra_stats_seq_start(struct seq_file *p, loff_t *pos)
690 static void *ll_ra_stats_seq_next(struct seq_file *p, void *v, loff_t *pos)
695 static void ll_ra_stats_seq_stop(struct seq_file *p, void *v)
698 struct seq_operations ll_ra_stats_seq_sops = {
699 .start = ll_ra_stats_seq_start,
700 .stop = ll_ra_stats_seq_stop,
701 .next = ll_ra_stats_seq_next,
702 .show = ll_ra_stats_seq_show,
705 static int ll_ra_stats_seq_open(struct inode *inode, struct file *file)
707 struct proc_dir_entry *dp = PDE(inode);
708 struct seq_file *seq;
711 rc = seq_open(file, &ll_ra_stats_seq_sops);
714 seq = file->private_data;
715 seq->private = dp->data;
719 static ssize_t ll_ra_stats_seq_write(struct file *file, const char *buf,
720 size_t len, loff_t *off)
722 struct seq_file *seq = file->private_data;
723 struct ll_sb_info *sbi = seq->private;
724 struct ll_ra_info *ra = &sbi->ll_ra_info;
726 spin_lock(&sbi->ll_lock);
727 memset(ra->ra_stats, 0, sizeof(ra->ra_stats));
728 spin_unlock(&sbi->ll_lock);
733 struct file_operations ll_ra_stats_fops = {
734 .owner = THIS_MODULE,
735 .open = ll_ra_stats_seq_open,
737 .write = ll_ra_stats_seq_write,
739 .release = seq_release,