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 <lustre_lite.h>
26 #include <lprocfs_status.h>
27 #include <linux/seq_file.h>
28 #include <obd_support.h>
30 #include "llite_internal.h"
32 struct proc_dir_entry *proc_lustre_fs_root;
35 /* /proc/lustre/llite mount point registration */
36 struct file_operations llite_dump_pgcache_fops;
37 struct file_operations ll_ra_stats_fops;
38 struct file_operations ll_rw_extents_stats_fops;
39 struct file_operations ll_rw_extents_stats_pp_fops;
40 struct file_operations ll_rw_offset_stats_fops;
42 static int ll_rd_blksize(char *page, char **start, off_t off, int count,
45 struct super_block *sb = (struct super_block *)data;
46 struct obd_statfs osfs;
50 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
54 rc = snprintf(page, count, "%u\n", osfs.os_bsize);
60 static int ll_rd_kbytestotal(char *page, char **start, off_t off, int count,
63 struct super_block *sb = (struct super_block *)data;
64 struct obd_statfs osfs;
68 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
71 __u32 blk_size = osfs.os_bsize >> 10;
72 __u64 result = osfs.os_blocks;
74 while (blk_size >>= 1)
78 rc = snprintf(page, count, LPU64"\n", result);
84 static int ll_rd_kbytesfree(char *page, char **start, off_t off, int count,
87 struct super_block *sb = (struct super_block *)data;
88 struct obd_statfs osfs;
92 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
95 __u32 blk_size = osfs.os_bsize >> 10;
96 __u64 result = osfs.os_bfree;
98 while (blk_size >>= 1)
102 rc = snprintf(page, count, LPU64"\n", result);
107 static int ll_rd_kbytesavail(char *page, char **start, off_t off, int count,
108 int *eof, void *data)
110 struct super_block *sb = (struct super_block *)data;
111 struct obd_statfs osfs;
115 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
118 __u32 blk_size = osfs.os_bsize >> 10;
119 __u64 result = osfs.os_bavail;
121 while (blk_size >>= 1)
125 rc = snprintf(page, count, LPU64"\n", result);
130 static int ll_rd_filestotal(char *page, char **start, off_t off, int count,
131 int *eof, void *data)
133 struct super_block *sb = (struct super_block *)data;
134 struct obd_statfs osfs;
138 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
142 rc = snprintf(page, count, LPU64"\n", osfs.os_files);
147 static int ll_rd_filesfree(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, cfs_time_current_64() - HZ,
159 rc = snprintf(page, count, LPU64"\n", osfs.os_ffree);
165 static int ll_rd_client_type(char *page, char **start, off_t off, int count,
166 int *eof, void *data)
168 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)data);
171 LASSERT(sbi != NULL);
174 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
175 rc = snprintf(page, count, "remote client\n");
177 rc = snprintf(page, count, "local client\n");
182 static int ll_rd_fstype(char *page, char **start, off_t off, int count,
183 int *eof, void *data)
185 struct super_block *sb = (struct super_block*)data;
189 return snprintf(page, count, "%s\n", sb->s_type->name);
192 static int ll_rd_sb_uuid(char *page, char **start, off_t off, int count,
193 int *eof, void *data)
195 struct super_block *sb = (struct super_block *)data;
199 return snprintf(page, count, "%s\n", ll_s2sbi(sb)->ll_sb_uuid.uuid);
202 static int ll_rd_max_readahead_mb(char *page, char **start, off_t off,
203 int count, int *eof, void *data)
205 struct super_block *sb = data;
206 struct ll_sb_info *sbi = ll_s2sbi(sb);
210 spin_lock(&sbi->ll_lock);
211 pages_number = sbi->ll_ra_info.ra_max_pages;
212 spin_unlock(&sbi->ll_lock);
214 mult = 1 << (20 - PAGE_CACHE_SHIFT);
215 return lprocfs_read_frac_helper(page, count, pages_number, mult);
218 static int ll_wr_max_readahead_mb(struct file *file, const char *buffer,
219 unsigned long count, void *data)
221 struct super_block *sb = data;
222 struct ll_sb_info *sbi = ll_s2sbi(sb);
223 int mult, rc, pages_number;
225 mult = 1 << (20 - CFS_PAGE_SHIFT);
226 rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
230 if (pages_number < 0 || pages_number > num_physpages / 2) {
231 CERROR("can't set file readahead more than %lu MB\n",
232 num_physpages >> (20 - CFS_PAGE_SHIFT + 1)); /*1/2 of RAM*/
236 spin_lock(&sbi->ll_lock);
237 sbi->ll_ra_info.ra_max_pages = pages_number;
238 spin_unlock(&sbi->ll_lock);
243 static int ll_rd_max_read_ahead_whole_mb(char *page, char **start, off_t off,
244 int count, int *eof, void *data)
246 struct super_block *sb = data;
247 struct ll_sb_info *sbi = ll_s2sbi(sb);
251 spin_lock(&sbi->ll_lock);
252 pages_number = sbi->ll_ra_info.ra_max_read_ahead_whole_pages;
253 spin_unlock(&sbi->ll_lock);
255 mult = 1 << (20 - CFS_PAGE_SHIFT);
256 return lprocfs_read_frac_helper(page, count, pages_number, mult);
259 static int ll_wr_max_read_ahead_whole_mb(struct file *file, const char *buffer,
260 unsigned long count, void *data)
262 struct super_block *sb = data;
263 struct ll_sb_info *sbi = ll_s2sbi(sb);
264 int mult, rc, pages_number;
266 mult = 1 << (20 - CFS_PAGE_SHIFT);
267 rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
271 /* Cap this at the current max readahead window size, the readahead
272 * algorithm does this anyway so it's pointless to set it larger. */
273 if (pages_number < 0 || pages_number > sbi->ll_ra_info.ra_max_pages) {
274 CERROR("can't set max_read_ahead_whole_mb more than "
275 "max_read_ahead_mb: %lu\n",
276 sbi->ll_ra_info.ra_max_pages >> (20 - CFS_PAGE_SHIFT));
280 spin_lock(&sbi->ll_lock);
281 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
282 spin_unlock(&sbi->ll_lock);
287 static int ll_rd_max_cached_mb(char *page, char **start, off_t off,
288 int count, int *eof, void *data)
290 struct super_block *sb = data;
291 struct ll_sb_info *sbi = ll_s2sbi(sb);
295 spin_lock(&sbi->ll_lock);
296 pages_number = sbi->ll_async_page_max;
297 spin_unlock(&sbi->ll_lock);
299 mult = 1 << (20 - CFS_PAGE_SHIFT);
300 return lprocfs_read_frac_helper(page, count, pages_number, mult);;
303 static int ll_wr_max_cached_mb(struct file *file, const char *buffer,
304 unsigned long count, void *data)
306 struct super_block *sb = data;
307 struct ll_sb_info *sbi = ll_s2sbi(sb);
308 int mult, rc, pages_number;
310 mult = 1 << (20 - CFS_PAGE_SHIFT);
311 rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
315 if (pages_number < 0 || pages_number > num_physpages) {
316 CERROR("can't set max cache more than %lu MB\n",
317 num_physpages >> (20 - CFS_PAGE_SHIFT));
321 spin_lock(&sbi->ll_lock);
322 sbi->ll_async_page_max = pages_number ;
323 spin_unlock(&sbi->ll_lock);
326 /* Not set up yet, don't call llap_shrink_cache */
329 if (sbi->ll_async_page_count >= sbi->ll_async_page_max)
330 llap_shrink_cache(sbi, 0);
335 static int ll_rd_checksum(char *page, char **start, off_t off,
336 int count, int *eof, void *data)
338 struct super_block *sb = data;
339 struct ll_sb_info *sbi = ll_s2sbi(sb);
341 return snprintf(page, count, "%u\n",
342 (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
345 static int ll_wr_checksum(struct file *file, const char *buffer,
346 unsigned long count, void *data)
348 struct super_block *sb = data;
349 struct ll_sb_info *sbi = ll_s2sbi(sb);
356 rc = lprocfs_write_helper(buffer, count, &val);
360 sbi->ll_flags |= LL_SBI_CHECKSUM;
362 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
364 rc = obd_set_info_async(sbi->ll_dt_exp, strlen("checksum"), "checksum",
365 sizeof(val), &val, NULL);
367 CWARN("Failed to set OSC checksum flags: %d\n", rc);
372 static int ll_rd_max_rw_chunk(char *page, char **start, off_t off,
373 int count, int *eof, void *data)
375 struct super_block *sb = data;
377 return snprintf(page, count, "%lu\n", ll_s2sbi(sb)->ll_max_rw_chunk);
380 static int ll_wr_max_rw_chunk(struct file *file, const char *buffer,
381 unsigned long count, void *data)
383 struct super_block *sb = data;
386 rc = lprocfs_write_helper(buffer, count, &val);
389 ll_s2sbi(sb)->ll_max_rw_chunk = val;
393 static int ll_rd_track_id(char *page, int count, void *data,
394 enum stats_track_type type)
396 struct super_block *sb = data;
398 if (ll_s2sbi(sb)->ll_stats_track_type == type) {
399 return snprintf(page, count, "%d\n",
400 ll_s2sbi(sb)->ll_stats_track_id);
402 } else if (ll_s2sbi(sb)->ll_stats_track_type == STATS_TRACK_ALL) {
403 return snprintf(page, count, "0 (all)\n");
405 return snprintf(page, count, "untracked\n");
409 static int ll_wr_track_id(const char *buffer, unsigned long count, void *data,
410 enum stats_track_type type)
412 struct super_block *sb = data;
415 rc = lprocfs_write_helper(buffer, count, &pid);
418 ll_s2sbi(sb)->ll_stats_track_id = pid;
420 ll_s2sbi(sb)->ll_stats_track_type = STATS_TRACK_ALL;
422 ll_s2sbi(sb)->ll_stats_track_type = type;
423 lprocfs_clear_stats(ll_s2sbi(sb)->ll_stats);
427 static int ll_rd_track_pid(char *page, char **start, off_t off,
428 int count, int *eof, void *data)
430 return (ll_rd_track_id(page, count, data, STATS_TRACK_PID));
433 static int ll_wr_track_pid(struct file *file, const char *buffer,
434 unsigned long count, void *data)
436 return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PID));
439 static int ll_rd_track_ppid(char *page, char **start, off_t off,
440 int count, int *eof, void *data)
442 return (ll_rd_track_id(page, count, data, STATS_TRACK_PPID));
445 static int ll_wr_track_ppid(struct file *file, const char *buffer,
446 unsigned long count, void *data)
448 return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PPID));
451 static int ll_rd_track_gid(char *page, char **start, off_t off,
452 int count, int *eof, void *data)
454 return (ll_rd_track_id(page, count, data, STATS_TRACK_GID));
457 static int ll_wr_track_gid(struct file *file, const char *buffer,
458 unsigned long count, void *data)
460 return (ll_wr_track_id(buffer, count, data, STATS_TRACK_GID));
463 static int ll_rd_contention_time(char *page, char **start, off_t off,
464 int count, int *eof, void *data)
466 struct super_block *sb = data;
469 return snprintf(page, count, "%u\n", ll_s2sbi(sb)->ll_contention_time);
473 static int ll_wr_contention_time(struct file *file, const char *buffer,
474 unsigned long count, void *data)
476 struct super_block *sb = data;
477 struct ll_sb_info *sbi = ll_s2sbi(sb);
479 return lprocfs_write_helper(buffer, count,&sbi->ll_contention_time) ?:
483 static int ll_rd_lockless_truncate(char *page, char **start, off_t off,
484 int count, int *eof, void *data)
486 struct super_block *sb = data;
489 return snprintf(page, count, "%u\n",
490 ll_s2sbi(sb)->ll_lockless_truncate_enable);
493 static int ll_wr_lockless_truncate(struct file *file, const char *buffer,
494 unsigned long count, void *data)
496 struct super_block *sb = data;
497 struct ll_sb_info *sbi = ll_s2sbi(sb);
499 return lprocfs_write_helper(buffer, count,
500 &sbi->ll_lockless_truncate_enable)
504 static struct lprocfs_vars lprocfs_llite_obd_vars[] = {
505 { "uuid", ll_rd_sb_uuid, 0, 0 },
506 //{ "mntpt_path", ll_rd_path, 0, 0 },
507 { "fstype", ll_rd_fstype, 0, 0 },
508 { "blocksize", ll_rd_blksize, 0, 0 },
509 { "kbytestotal", ll_rd_kbytestotal, 0, 0 },
510 { "kbytesfree", ll_rd_kbytesfree, 0, 0 },
511 { "kbytesavail", ll_rd_kbytesavail, 0, 0 },
512 { "filestotal", ll_rd_filestotal, 0, 0 },
513 { "filesfree", ll_rd_filesfree, 0, 0 },
514 { "client_type", ll_rd_client_type, 0, 0 },
515 //{ "filegroups", lprocfs_rd_filegroups, 0, 0 },
516 { "max_read_ahead_mb", ll_rd_max_readahead_mb,
517 ll_wr_max_readahead_mb, 0 },
518 { "max_read_ahead_whole_mb", ll_rd_max_read_ahead_whole_mb,
519 ll_wr_max_read_ahead_whole_mb, 0 },
520 { "max_cached_mb", ll_rd_max_cached_mb, ll_wr_max_cached_mb, 0 },
521 { "checksum_pages", ll_rd_checksum, ll_wr_checksum, 0 },
522 { "max_rw_chunk", ll_rd_max_rw_chunk, ll_wr_max_rw_chunk, 0 },
523 { "stats_track_pid", ll_rd_track_pid, ll_wr_track_pid, 0 },
524 { "stats_track_ppid", ll_rd_track_ppid, ll_wr_track_ppid, 0 },
525 { "stats_track_gid", ll_rd_track_gid, ll_wr_track_gid, 0 },
526 { "contention_seconds", ll_rd_contention_time, ll_wr_contention_time, 0},
527 { "lockless_truncate", ll_rd_lockless_truncate,
528 ll_wr_lockless_truncate, 0},
532 #define MAX_STRING_SIZE 128
534 struct llite_file_opcode {
538 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
540 { LPROC_LL_DIRTY_HITS, LPROCFS_TYPE_REGS, "dirty_pages_hits" },
541 { LPROC_LL_DIRTY_MISSES, LPROCFS_TYPE_REGS, "dirty_pages_misses" },
542 { LPROC_LL_WB_WRITEPAGE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
543 "writeback_from_writepage" },
544 { LPROC_LL_WB_PRESSURE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
545 "writeback_from_pressure" },
546 { LPROC_LL_WB_OK, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
547 "writeback_ok_pages" },
548 { LPROC_LL_WB_FAIL, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
549 "writeback_failed_pages" },
550 { LPROC_LL_READ_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
552 { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
554 { LPROC_LL_BRW_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
556 { LPROC_LL_BRW_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
559 { LPROC_LL_IOCTL, LPROCFS_TYPE_REGS, "ioctl" },
560 { LPROC_LL_OPEN, LPROCFS_TYPE_REGS, "open" },
561 { LPROC_LL_RELEASE, LPROCFS_TYPE_REGS, "close" },
562 { LPROC_LL_MAP, LPROCFS_TYPE_REGS, "mmap" },
563 { LPROC_LL_LLSEEK, LPROCFS_TYPE_REGS, "seek" },
564 { LPROC_LL_FSYNC, LPROCFS_TYPE_REGS, "fsync" },
565 /* inode operation */
566 { LPROC_LL_SETATTR, LPROCFS_TYPE_REGS, "setattr" },
567 { LPROC_LL_TRUNC, LPROCFS_TYPE_REGS, "truncate" },
568 { LPROC_LL_LOCKLESS_TRUNC, LPROCFS_TYPE_REGS, "lockless_truncate"},
569 { LPROC_LL_FLOCK, LPROCFS_TYPE_REGS, "flock" },
570 { LPROC_LL_GETATTR, LPROCFS_TYPE_REGS, "getattr" },
571 /* special inode operation */
572 { LPROC_LL_STAFS, LPROCFS_TYPE_REGS, "statfs" },
573 { LPROC_LL_ALLOC_INODE, LPROCFS_TYPE_REGS, "alloc_inode" },
574 { LPROC_LL_SETXATTR, LPROCFS_TYPE_REGS, "setxattr" },
575 { LPROC_LL_GETXATTR, LPROCFS_TYPE_REGS, "getxattr" },
576 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_REGS, "listxattr" },
577 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_REGS, "removexattr" },
578 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_REGS, "inode_permission" },
579 { LPROC_LL_DIRECT_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
581 { LPROC_LL_DIRECT_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
583 { LPROC_LL_LOCKLESS_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
584 "lockless_read_bytes" },
585 { LPROC_LL_LOCKLESS_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
586 "lockless_write_bytes" },
590 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
594 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
595 lprocfs_counter_add(sbi->ll_stats, op, count);
596 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
597 sbi->ll_stats_track_id == current->pid)
598 lprocfs_counter_add(sbi->ll_stats, op, count);
599 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
600 sbi->ll_stats_track_id == current->p_pptr->pid)
601 lprocfs_counter_add(sbi->ll_stats, op, count);
602 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
603 sbi->ll_stats_track_id == current->gid)
604 lprocfs_counter_add(sbi->ll_stats, op, count);
606 EXPORT_SYMBOL(ll_stats_ops_tally);
608 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
609 struct super_block *sb, char *osc, char *mdc)
611 struct lprocfs_vars lvars[2];
612 struct lustre_sb_info *lsi = s2lsi(sb);
613 struct ll_sb_info *sbi = ll_s2sbi(sb);
614 struct obd_device *obd;
615 char name[MAX_STRING_SIZE + 1], *ptr;
616 int err, id, len, rc;
619 memset(lvars, 0, sizeof(lvars));
621 name[MAX_STRING_SIZE] = '\0';
622 lvars[0].name = name;
624 LASSERT(sbi != NULL);
625 LASSERT(mdc != NULL);
626 LASSERT(osc != NULL);
629 len = strlen(lsi->lsi_lmd->lmd_profile);
630 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
631 if (ptr && (strcmp(ptr, "-client") == 0))
635 snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
636 lsi->lsi_lmd->lmd_profile, sb);
638 sbi->ll_proc_root = lprocfs_register(name, parent, NULL, NULL);
639 if (IS_ERR(sbi->ll_proc_root)) {
640 err = PTR_ERR(sbi->ll_proc_root);
641 sbi->ll_proc_root = NULL;
646 rc = lprocfs_seq_create(sbi->ll_proc_root, "dump_page_cache", 0444,
647 &llite_dump_pgcache_fops, sbi);
649 CWARN("Error adding the dump_page_cache file\n");
651 rc = lprocfs_seq_create(sbi->ll_proc_root, "read_ahead_stats", 0644,
652 &ll_ra_stats_fops, sbi);
654 CWARN("Error adding the read_ahead_stats file\n");
656 rc = lprocfs_seq_create(sbi->ll_proc_root, "extents_stats", 0644,
657 &ll_rw_extents_stats_fops, sbi);
659 CWARN("Error adding the extent_stats file\n");
661 rc = lprocfs_seq_create(sbi->ll_proc_root, "extents_stats_per_process",
662 0644, &ll_rw_extents_stats_pp_fops, sbi);
664 CWARN("Error adding the extents_stats_per_process file\n");
666 rc = lprocfs_seq_create(sbi->ll_proc_root, "offset_stats", 0644,
667 &ll_rw_offset_stats_fops, sbi);
669 CWARN("Error adding the offset_stats file\n");
671 /* File operations stats */
672 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
673 LPROCFS_STATS_FLAG_PERCPU);
674 if (sbi->ll_stats == NULL)
675 GOTO(out, err = -ENOMEM);
676 /* do counter init */
677 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
678 __u32 type = llite_opcode_table[id].type;
680 if (type & LPROCFS_TYPE_REGS)
682 else if (type & LPROCFS_TYPE_BYTES)
684 else if (type & LPROCFS_TYPE_PAGES)
686 lprocfs_counter_init(sbi->ll_stats,
687 llite_opcode_table[id].opcode,
688 (type & LPROCFS_CNTR_AVGMINMAX),
689 llite_opcode_table[id].opname, ptr);
691 err = lprocfs_register_stats(sbi->ll_proc_root, "stats", sbi->ll_stats);
695 err = lprocfs_add_vars(sbi->ll_proc_root, lprocfs_llite_obd_vars, sb);
700 obd = class_name2obd(mdc);
702 LASSERT(obd != NULL);
703 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
704 LASSERT(obd->obd_type->typ_name != NULL);
706 snprintf(name, MAX_STRING_SIZE, "%s/common_name",
707 obd->obd_type->typ_name);
708 lvars[0].read_fptr = lprocfs_rd_name;
709 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
713 snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
714 lvars[0].read_fptr = lprocfs_rd_uuid;
715 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
720 obd = class_name2obd(osc);
722 LASSERT(obd != NULL);
723 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
724 LASSERT(obd->obd_type->typ_name != NULL);
726 snprintf(name, MAX_STRING_SIZE, "%s/common_name",
727 obd->obd_type->typ_name);
728 lvars[0].read_fptr = lprocfs_rd_name;
729 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
733 snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
734 lvars[0].read_fptr = lprocfs_rd_uuid;
735 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
738 lprocfs_remove(&sbi->ll_proc_root);
739 lprocfs_free_stats(&sbi->ll_stats);
744 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi)
746 if (sbi->ll_proc_root) {
747 lprocfs_remove(&sbi->ll_proc_root);
748 lprocfs_free_stats(&sbi->ll_stats);
751 #undef MAX_STRING_SIZE
753 #define seq_page_flag(seq, page, flag, has_flags) do { \
754 if (test_bit(PG_##flag, &(page)->flags)) { \
758 seq_putc(seq, '|'); \
759 seq_puts(seq, #flag); \
763 static void *llite_dump_pgcache_seq_start(struct seq_file *seq, loff_t *pos)
765 struct ll_async_page *dummy_llap = seq->private;
767 if (dummy_llap->llap_magic == 2)
773 static int llite_dump_pgcache_seq_show(struct seq_file *seq, void *v)
775 struct ll_async_page *llap, *dummy_llap = seq->private;
776 struct ll_sb_info *sbi = dummy_llap->llap_cookie;
778 /* 2.4 doesn't seem to have SEQ_START_TOKEN, so we implement
779 * it in our own state */
780 if (dummy_llap->llap_magic == 0) {
781 seq_printf(seq, "gener | llap cookie origin wq du wb | page "
782 "inode index count [ page flags ]\n");
786 spin_lock(&sbi->ll_lock);
788 llap = llite_pglist_next_llap(sbi, &dummy_llap->llap_pglist_item);
791 struct page *page = llap->llap_page;
793 LASSERTF(llap->llap_origin < LLAP__ORIGIN_MAX, "%u\n",
796 seq_printf(seq," %5lu | %p %p %s %s %s %s | %p %lu/%u(%p) "
799 llap, llap->llap_cookie,
800 llap_origins[llap->llap_origin],
801 llap->llap_write_queued ? "wq" : "- ",
802 llap->llap_defer_uptodate ? "du" : "- ",
803 PageWriteback(page) ? "wb" : "-",
804 page, page->mapping->host->i_ino,
805 page->mapping->host->i_generation,
806 page->mapping->host, page->index,
808 seq_page_flag(seq, page, locked, has_flags);
809 seq_page_flag(seq, page, error, has_flags);
810 seq_page_flag(seq, page, referenced, has_flags);
811 seq_page_flag(seq, page, uptodate, has_flags);
812 seq_page_flag(seq, page, dirty, has_flags);
813 #if (LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,12))
814 seq_page_flag(seq, page, highmem, has_flags);
816 seq_page_flag(seq, page, writeback, has_flags);
818 seq_puts(seq, "-]\n");
820 seq_puts(seq, "]\n");
823 spin_unlock(&sbi->ll_lock);
828 static void *llite_dump_pgcache_seq_next(struct seq_file *seq, void *v,
831 struct ll_async_page *llap, *dummy_llap = seq->private;
832 struct ll_sb_info *sbi = dummy_llap->llap_cookie;
834 /* bail if we just displayed the banner */
835 if (dummy_llap->llap_magic == 0) {
836 dummy_llap->llap_magic = 1;
840 /* we've just displayed the llap that is after us in the list.
841 * we advance to a position beyond it, returning null if there
842 * isn't another llap in the list beyond that new position. */
843 spin_lock(&sbi->ll_lock);
844 llap = llite_pglist_next_llap(sbi, &dummy_llap->llap_pglist_item);
845 list_del_init(&dummy_llap->llap_pglist_item);
847 list_add(&dummy_llap->llap_pglist_item,&llap->llap_pglist_item);
848 llap =llite_pglist_next_llap(sbi,&dummy_llap->llap_pglist_item);
850 spin_unlock(&sbi->ll_lock);
854 dummy_llap->llap_magic = 2;
860 static void null_stop(struct seq_file *seq, void *v)
864 struct seq_operations llite_dump_pgcache_seq_sops = {
865 .start = llite_dump_pgcache_seq_start,
867 .next = llite_dump_pgcache_seq_next,
868 .show = llite_dump_pgcache_seq_show,
871 /* we're displaying llaps in a list_head list. we don't want to hold a lock
872 * while we walk the entire list, and we don't want to have to seek into
873 * the right position in the list as an app advances with many syscalls. we
874 * allocate a dummy llap and hang it off file->private. its position in
875 * the list records where the app is currently displaying. this way our
876 * seq .start and .stop don't actually do anything. .next returns null
877 * when the dummy hits the end of the list which eventually leads to .release
878 * where we tear down. this kind of displaying is super-racey, so we put
879 * a generation counter on the list so the output shows when the list
880 * changes between reads.
882 static int llite_dump_pgcache_seq_open(struct inode *inode, struct file *file)
884 struct proc_dir_entry *dp = PDE(inode);
885 struct ll_async_page *dummy_llap;
886 struct seq_file *seq;
887 struct ll_sb_info *sbi = dp->data;
890 LPROCFS_ENTRY_AND_CHECK(dp);
892 OBD_ALLOC_PTR_WAIT(dummy_llap);
893 if (dummy_llap == NULL)
895 dummy_llap->llap_page = NULL;
896 dummy_llap->llap_cookie = sbi;
897 dummy_llap->llap_magic = 0;
899 rc = seq_open(file, &llite_dump_pgcache_seq_sops);
901 OBD_FREE(dummy_llap, sizeof(*dummy_llap));
904 seq = file->private_data;
905 seq->private = dummy_llap;
907 spin_lock(&sbi->ll_lock);
908 list_add(&dummy_llap->llap_pglist_item, &sbi->ll_pglist);
909 spin_unlock(&sbi->ll_lock);
917 static int llite_dump_pgcache_seq_release(struct inode *inode,
920 struct seq_file *seq = file->private_data;
921 struct ll_async_page *dummy_llap = seq->private;
922 struct ll_sb_info *sbi = dummy_llap->llap_cookie;
924 spin_lock(&sbi->ll_lock);
925 if (!list_empty(&dummy_llap->llap_pglist_item))
926 list_del_init(&dummy_llap->llap_pglist_item);
927 spin_unlock(&sbi->ll_lock);
928 OBD_FREE(dummy_llap, sizeof(*dummy_llap));
930 return lprocfs_seq_release(inode, file);
933 struct file_operations llite_dump_pgcache_fops = {
934 .owner = THIS_MODULE,
935 .open = llite_dump_pgcache_seq_open,
937 .release = llite_dump_pgcache_seq_release,
940 static int ll_ra_stats_seq_show(struct seq_file *seq, void *v)
943 struct ll_sb_info *sbi = seq->private;
944 struct ll_ra_info *ra = &sbi->ll_ra_info;
946 static char *ra_stat_strings[] = {
947 [RA_STAT_HIT] = "hits",
948 [RA_STAT_MISS] = "misses",
949 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
950 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
951 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
952 [RA_STAT_FAILED_MATCH] = "failed lock match",
953 [RA_STAT_DISCARDED] = "read but discarded",
954 [RA_STAT_ZERO_LEN] = "zero length file",
955 [RA_STAT_ZERO_WINDOW] = "zero size window",
956 [RA_STAT_EOF] = "read-ahead to EOF",
957 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
958 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
961 do_gettimeofday(&now);
963 spin_lock(&sbi->ll_lock);
965 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
966 now.tv_sec, now.tv_usec);
967 seq_printf(seq, "pending issued pages: %lu\n",
970 for(i = 0; i < _NR_RA_STAT; i++)
971 seq_printf(seq, "%-25s %lu\n", ra_stat_strings[i],
974 spin_unlock(&sbi->ll_lock);
979 static ssize_t ll_ra_stats_seq_write(struct file *file, const char *buf,
980 size_t len, loff_t *off)
982 struct seq_file *seq = file->private_data;
983 struct ll_sb_info *sbi = seq->private;
984 struct ll_ra_info *ra = &sbi->ll_ra_info;
986 spin_lock(&sbi->ll_lock);
987 memset(ra->ra_stats, 0, sizeof(ra->ra_stats));
988 spin_unlock(&sbi->ll_lock);
993 LPROC_SEQ_FOPS(ll_ra_stats);
995 #define pct(a,b) (b ? a * 100 / b : 0)
997 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
998 struct seq_file *seq, int which)
1000 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1001 unsigned long start, end, r, w;
1002 char *unitp = "KMGTPEZY";
1004 struct per_process_info *pp_info = &io_extents->pp_extents[which];
1010 for(i = 0; i < LL_HIST_MAX; i++) {
1011 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1012 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1015 for(i = 0; i < LL_HIST_MAX; i++) {
1016 r = pp_info->pp_r_hist.oh_buckets[i];
1017 w = pp_info->pp_w_hist.oh_buckets[i];
1020 end = 1 << (i + LL_HIST_START - units);
1021 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4lu %4lu | "
1022 "%14lu %4lu %4lu\n", start, *unitp, end, *unitp,
1023 (i == LL_HIST_MAX - 1) ? '+' : ' ',
1024 r, pct(r, read_tot), pct(read_cum, read_tot),
1025 w, pct(w, write_tot), pct(write_cum, write_tot));
1027 if (start == 1<<10) {
1032 if (read_cum == read_tot && write_cum == write_tot)
1037 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1040 struct ll_sb_info *sbi = seq->private;
1041 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1044 do_gettimeofday(&now);
1046 if (!sbi->ll_rw_stats_on) {
1047 seq_printf(seq, "Disabled\n"
1048 "Write anything in this file to activate\n");
1051 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
1052 now.tv_sec, now.tv_usec);
1053 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1054 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1055 "extents", "calls", "%", "cum%",
1056 "calls", "%", "cum%");
1057 spin_lock(&sbi->ll_pp_extent_lock);
1058 for(k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1059 if(io_extents->pp_extents[k].pid != 0) {
1060 seq_printf(seq, "\nPID: %d\n",
1061 io_extents->pp_extents[k].pid);
1062 ll_display_extents_info(io_extents, seq, k);
1065 spin_unlock(&sbi->ll_pp_extent_lock);
1069 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1070 const char *buf, size_t len,
1073 struct seq_file *seq = file->private_data;
1074 struct ll_sb_info *sbi = seq->private;
1075 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1078 sbi->ll_rw_stats_on = 1;
1079 spin_lock(&sbi->ll_pp_extent_lock);
1080 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1081 io_extents->pp_extents[i].pid = 0;
1082 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1083 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1085 spin_unlock(&sbi->ll_pp_extent_lock);
1089 LPROC_SEQ_FOPS(ll_rw_extents_stats_pp);
1091 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1094 struct ll_sb_info *sbi = seq->private;
1095 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1097 do_gettimeofday(&now);
1099 if (!sbi->ll_rw_stats_on) {
1100 seq_printf(seq, "Disabled\n"
1101 "Write anything in this file to activate\n");
1104 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
1105 now.tv_sec, now.tv_usec);
1107 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1108 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1109 "extents", "calls", "%", "cum%",
1110 "calls", "%", "cum%");
1111 spin_lock(&sbi->ll_lock);
1112 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1113 spin_unlock(&sbi->ll_lock);
1118 static ssize_t ll_rw_extents_stats_seq_write(struct file *file, const char *buf,
1119 size_t len, loff_t *off)
1121 struct seq_file *seq = file->private_data;
1122 struct ll_sb_info *sbi = seq->private;
1123 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1126 sbi->ll_rw_stats_on = 1;
1127 spin_lock(&sbi->ll_pp_extent_lock);
1128 for(i = 0; i <= LL_PROCESS_HIST_MAX; i++)
1130 io_extents->pp_extents[i].pid = 0;
1131 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1132 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1134 spin_unlock(&sbi->ll_pp_extent_lock);
1139 LPROC_SEQ_FOPS(ll_rw_extents_stats);
1141 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid, struct file
1142 *file, size_t count, int rw)
1145 struct ll_rw_process_info *process;
1146 struct ll_rw_process_info *offset;
1147 int *off_count = &sbi->ll_rw_offset_entry_count;
1148 int *process_count = &sbi->ll_offset_process_count;
1149 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1151 if(!sbi->ll_rw_stats_on)
1153 process = sbi->ll_rw_process_info;
1154 offset = sbi->ll_rw_offset_info;
1156 spin_lock(&sbi->ll_pp_extent_lock);
1157 /* Extent statistics */
1158 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1159 if(io_extents->pp_extents[i].pid == pid) {
1167 sbi->ll_extent_process_count =
1168 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1169 cur = sbi->ll_extent_process_count;
1170 io_extents->pp_extents[cur].pid = pid;
1171 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1172 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1175 for(i = 0; (count >= (1 << LL_HIST_START << i)) &&
1176 (i < (LL_HIST_MAX - 1)); i++);
1178 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1179 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1181 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1182 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1184 spin_unlock(&sbi->ll_pp_extent_lock);
1186 spin_lock(&sbi->ll_process_lock);
1187 /* Offset statistics */
1188 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1189 if (process[i].rw_pid == pid) {
1190 if (process[i].rw_last_file != file) {
1191 process[i].rw_range_start = file->f_pos;
1192 process[i].rw_last_file_pos =
1193 file->f_pos + count;
1194 process[i].rw_smallest_extent = count;
1195 process[i].rw_largest_extent = count;
1196 process[i].rw_offset = 0;
1197 process[i].rw_last_file = file;
1198 spin_unlock(&sbi->ll_process_lock);
1201 if (process[i].rw_last_file_pos != file->f_pos) {
1203 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1204 offset[*off_count].rw_op = process[i].rw_op;
1205 offset[*off_count].rw_pid = pid;
1206 offset[*off_count].rw_range_start =
1207 process[i].rw_range_start;
1208 offset[*off_count].rw_range_end =
1209 process[i].rw_last_file_pos;
1210 offset[*off_count].rw_smallest_extent =
1211 process[i].rw_smallest_extent;
1212 offset[*off_count].rw_largest_extent =
1213 process[i].rw_largest_extent;
1214 offset[*off_count].rw_offset =
1215 process[i].rw_offset;
1216 process[i].rw_op = rw;
1217 process[i].rw_range_start = file->f_pos;
1218 process[i].rw_smallest_extent = count;
1219 process[i].rw_largest_extent = count;
1220 process[i].rw_offset = file->f_pos -
1221 process[i].rw_last_file_pos;
1223 if(process[i].rw_smallest_extent > count)
1224 process[i].rw_smallest_extent = count;
1225 if(process[i].rw_largest_extent < count)
1226 process[i].rw_largest_extent = count;
1227 process[i].rw_last_file_pos = file->f_pos + count;
1228 spin_unlock(&sbi->ll_process_lock);
1232 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1233 process[*process_count].rw_pid = pid;
1234 process[*process_count].rw_op = rw;
1235 process[*process_count].rw_range_start = file->f_pos;
1236 process[*process_count].rw_last_file_pos = file->f_pos + count;
1237 process[*process_count].rw_smallest_extent = count;
1238 process[*process_count].rw_largest_extent = count;
1239 process[*process_count].rw_offset = 0;
1240 process[*process_count].rw_last_file = file;
1241 spin_unlock(&sbi->ll_process_lock);
1244 char lpszt[] = LPSZ;
1246 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1249 struct ll_sb_info *sbi = seq->private;
1250 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1251 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1255 do_gettimeofday(&now);
1257 if (!sbi->ll_rw_stats_on) {
1258 seq_printf(seq, "Disabled\n"
1259 "Write anything in this file to activate\n");
1262 spin_lock(&sbi->ll_process_lock);
1264 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
1265 now.tv_sec, now.tv_usec);
1266 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1267 "R/W", "PID", "RANGE START", "RANGE END",
1268 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1269 sprintf(format, "%s%s%s%s%s\n",
1270 "%3c %10d %14Lu %14Lu %17", lpszt+1, " %17", lpszt+1, " %14Ld");
1271 /* We stored the discontiguous offsets here; print them first */
1272 for(i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1273 if (offset[i].rw_pid != 0)
1274 /* Is there a way to snip the '%' off of LPSZ? */
1275 seq_printf(seq, format,
1276 offset[i].rw_op ? 'W' : 'R',
1278 offset[i].rw_range_start,
1279 offset[i].rw_range_end,
1280 offset[i].rw_smallest_extent,
1281 offset[i].rw_largest_extent,
1282 offset[i].rw_offset);
1284 /* Then print the current offsets for each process */
1285 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1286 if (process[i].rw_pid != 0)
1287 seq_printf(seq, format,
1288 process[i].rw_op ? 'W' : 'R',
1290 process[i].rw_range_start,
1291 process[i].rw_last_file_pos,
1292 process[i].rw_smallest_extent,
1293 process[i].rw_largest_extent,
1294 process[i].rw_offset);
1296 spin_unlock(&sbi->ll_process_lock);
1301 static ssize_t ll_rw_offset_stats_seq_write(struct file *file, const char *buf,
1302 size_t len, loff_t *off)
1304 struct seq_file *seq = file->private_data;
1305 struct ll_sb_info *sbi = seq->private;
1306 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1307 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1309 sbi->ll_rw_stats_on = 1;
1311 spin_lock(&sbi->ll_process_lock);
1312 sbi->ll_offset_process_count = 0;
1313 sbi->ll_rw_offset_entry_count = 0;
1314 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1315 LL_PROCESS_HIST_MAX);
1316 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1317 LL_OFFSET_HIST_MAX);
1318 spin_unlock(&sbi->ll_process_lock);
1323 LPROC_SEQ_FOPS(ll_rw_offset_stats);
1325 void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars)
1327 lvars->module_vars = NULL;
1328 lvars->obd_vars = lprocfs_llite_obd_vars;
1330 #endif /* LPROCFS */