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 struct lprocfs_vars lprocfs_llite_obd_vars[] = {
464 { "uuid", ll_rd_sb_uuid, 0, 0 },
465 //{ "mntpt_path", ll_rd_path, 0, 0 },
466 { "fstype", ll_rd_fstype, 0, 0 },
467 { "blocksize", ll_rd_blksize, 0, 0 },
468 { "kbytestotal", ll_rd_kbytestotal, 0, 0 },
469 { "kbytesfree", ll_rd_kbytesfree, 0, 0 },
470 { "kbytesavail", ll_rd_kbytesavail, 0, 0 },
471 { "filestotal", ll_rd_filestotal, 0, 0 },
472 { "filesfree", ll_rd_filesfree, 0, 0 },
473 { "client_type", ll_rd_client_type, 0, 0 },
474 //{ "filegroups", lprocfs_rd_filegroups, 0, 0 },
475 { "max_read_ahead_mb", ll_rd_max_readahead_mb,
476 ll_wr_max_readahead_mb, 0 },
477 { "max_read_ahead_whole_mb", ll_rd_max_read_ahead_whole_mb,
478 ll_wr_max_read_ahead_whole_mb, 0 },
479 { "max_cached_mb", ll_rd_max_cached_mb, ll_wr_max_cached_mb, 0 },
480 { "checksum_pages", ll_rd_checksum, ll_wr_checksum, 0 },
481 { "max_rw_chunk", ll_rd_max_rw_chunk, ll_wr_max_rw_chunk, 0 },
482 { "stats_track_pid", ll_rd_track_pid, ll_wr_track_pid, 0 },
483 { "stats_track_ppid", ll_rd_track_ppid, ll_wr_track_ppid, 0 },
484 { "stats_track_gid", ll_rd_track_gid, ll_wr_track_gid, 0 },
488 #define MAX_STRING_SIZE 128
490 struct llite_file_opcode {
494 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
496 { LPROC_LL_DIRTY_HITS, LPROCFS_TYPE_REGS, "dirty_pages_hits" },
497 { LPROC_LL_DIRTY_MISSES, LPROCFS_TYPE_REGS, "dirty_pages_misses" },
498 { LPROC_LL_WB_WRITEPAGE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
499 "writeback_from_writepage" },
500 { LPROC_LL_WB_PRESSURE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
501 "writeback_from_pressure" },
502 { LPROC_LL_WB_OK, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
503 "writeback_ok_pages" },
504 { LPROC_LL_WB_FAIL, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
505 "writeback_failed_pages" },
506 { LPROC_LL_READ_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
508 { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
510 { LPROC_LL_BRW_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
512 { LPROC_LL_BRW_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
515 { LPROC_LL_IOCTL, LPROCFS_TYPE_REGS, "ioctl" },
516 { LPROC_LL_OPEN, LPROCFS_TYPE_REGS, "open" },
517 { LPROC_LL_RELEASE, LPROCFS_TYPE_REGS, "close" },
518 { LPROC_LL_MAP, LPROCFS_TYPE_REGS, "mmap" },
519 { LPROC_LL_LLSEEK, LPROCFS_TYPE_REGS, "seek" },
520 { LPROC_LL_FSYNC, LPROCFS_TYPE_REGS, "fsync" },
521 /* inode operation */
522 { LPROC_LL_SETATTR, LPROCFS_TYPE_REGS, "setattr" },
523 { LPROC_LL_TRUNC, LPROCFS_TYPE_REGS, "truncate" },
524 { LPROC_LL_FLOCK, LPROCFS_TYPE_REGS, "flock" },
525 { LPROC_LL_GETATTR, LPROCFS_TYPE_REGS, "getattr" },
526 /* special inode operation */
527 { LPROC_LL_STAFS, LPROCFS_TYPE_REGS, "statfs" },
528 { LPROC_LL_ALLOC_INODE, LPROCFS_TYPE_REGS, "alloc_inode" },
529 { LPROC_LL_SETXATTR, LPROCFS_TYPE_REGS, "setxattr" },
530 { LPROC_LL_GETXATTR, LPROCFS_TYPE_REGS, "getxattr" },
531 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_REGS, "listxattr" },
532 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_REGS, "removexattr" },
533 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_REGS, "inode_permission" },
534 { LPROC_LL_DIRECT_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
536 { LPROC_LL_DIRECT_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
541 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
545 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
546 lprocfs_counter_add(sbi->ll_stats, op, count);
547 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
548 sbi->ll_stats_track_id == current->pid)
549 lprocfs_counter_add(sbi->ll_stats, op, count);
550 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
551 sbi->ll_stats_track_id == current->p_pptr->pid)
552 lprocfs_counter_add(sbi->ll_stats, op, count);
553 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
554 sbi->ll_stats_track_id == current->gid)
555 lprocfs_counter_add(sbi->ll_stats, op, count);
557 EXPORT_SYMBOL(ll_stats_ops_tally);
559 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
560 struct super_block *sb, char *osc, char *mdc)
562 struct lprocfs_vars lvars[2];
563 struct lustre_sb_info *lsi = s2lsi(sb);
564 struct ll_sb_info *sbi = ll_s2sbi(sb);
565 struct obd_device *obd;
566 char name[MAX_STRING_SIZE + 1], *ptr;
568 struct proc_dir_entry *entry;
571 memset(lvars, 0, sizeof(lvars));
573 name[MAX_STRING_SIZE] = '\0';
574 lvars[0].name = name;
576 LASSERT(sbi != NULL);
577 LASSERT(mdc != NULL);
578 LASSERT(osc != NULL);
581 len = strlen(lsi->lsi_lmd->lmd_profile);
582 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
583 if (ptr && (strcmp(ptr, "-client") == 0))
587 snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
588 lsi->lsi_lmd->lmd_profile, sb);
590 sbi->ll_proc_root = lprocfs_register(name, parent, NULL, NULL);
591 if (IS_ERR(sbi->ll_proc_root)) {
592 err = PTR_ERR(sbi->ll_proc_root);
593 sbi->ll_proc_root = NULL;
597 entry = create_proc_entry("dump_page_cache", 0444, sbi->ll_proc_root);
599 GOTO(out, err = -ENOMEM);
600 entry->proc_fops = &llite_dump_pgcache_fops;
603 entry = create_proc_entry("read_ahead_stats", 0644, sbi->ll_proc_root);
605 GOTO(out, err = -ENOMEM);
606 entry->proc_fops = &ll_ra_stats_fops;
609 entry = create_proc_entry("extents_stats", 0644, sbi->ll_proc_root);
611 GOTO(out, err = -ENOMEM);
612 entry->proc_fops = &ll_rw_extents_stats_fops;
615 entry = create_proc_entry("extents_stats_per_process", 0644,
618 GOTO(out, err = -ENOMEM);
619 entry->proc_fops = &ll_rw_extents_stats_pp_fops;
622 entry = create_proc_entry("offset_stats", 0644, sbi->ll_proc_root);
624 GOTO(out, err = -ENOMEM);
625 entry->proc_fops = &ll_rw_offset_stats_fops;
628 /* File operations stats */
629 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
630 LPROCFS_STATS_FLAG_PERCPU);
631 if (sbi->ll_stats == NULL)
632 GOTO(out, err = -ENOMEM);
633 /* do counter init */
634 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
635 __u32 type = llite_opcode_table[id].type;
637 if (type & LPROCFS_TYPE_REGS)
639 else if (type & LPROCFS_TYPE_BYTES)
641 else if (type & LPROCFS_TYPE_PAGES)
643 lprocfs_counter_init(sbi->ll_stats,
644 llite_opcode_table[id].opcode,
645 (type & LPROCFS_CNTR_AVGMINMAX),
646 llite_opcode_table[id].opname, ptr);
648 err = lprocfs_register_stats(sbi->ll_proc_root, "stats", sbi->ll_stats);
652 err = lprocfs_add_vars(sbi->ll_proc_root, lprocfs_llite_obd_vars, sb);
657 obd = class_name2obd(mdc);
659 LASSERT(obd != NULL);
660 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
661 LASSERT(obd->obd_type->typ_name != NULL);
663 snprintf(name, MAX_STRING_SIZE, "%s/common_name",
664 obd->obd_type->typ_name);
665 lvars[0].read_fptr = lprocfs_rd_name;
666 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
670 snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
671 lvars[0].read_fptr = lprocfs_rd_uuid;
672 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
677 obd = class_name2obd(osc);
679 LASSERT(obd != NULL);
680 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
681 LASSERT(obd->obd_type->typ_name != NULL);
683 snprintf(name, MAX_STRING_SIZE, "%s/common_name",
684 obd->obd_type->typ_name);
685 lvars[0].read_fptr = lprocfs_rd_name;
686 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
690 snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
691 lvars[0].read_fptr = lprocfs_rd_uuid;
692 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
695 lprocfs_remove(&sbi->ll_proc_root);
696 lprocfs_free_stats(&sbi->ll_stats);
701 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi)
703 if (sbi->ll_proc_root) {
704 lprocfs_remove(&sbi->ll_proc_root);
705 lprocfs_free_stats(&sbi->ll_stats);
708 #undef MAX_STRING_SIZE
710 #define seq_page_flag(seq, page, flag, has_flags) do { \
711 if (test_bit(PG_##flag, &(page)->flags)) { \
715 seq_putc(seq, '|'); \
716 seq_puts(seq, #flag); \
720 static void *llite_dump_pgcache_seq_start(struct seq_file *seq, loff_t *pos)
722 struct ll_async_page *dummy_llap = seq->private;
724 if (dummy_llap->llap_magic == 2)
730 static int llite_dump_pgcache_seq_show(struct seq_file *seq, void *v)
732 struct ll_async_page *llap, *dummy_llap = seq->private;
733 struct ll_sb_info *sbi = dummy_llap->llap_cookie;
735 /* 2.4 doesn't seem to have SEQ_START_TOKEN, so we implement
736 * it in our own state */
737 if (dummy_llap->llap_magic == 0) {
738 seq_printf(seq, "gener | llap cookie origin wq du wb | page "
739 "inode index count [ page flags ]\n");
743 spin_lock(&sbi->ll_lock);
745 llap = llite_pglist_next_llap(sbi, &dummy_llap->llap_pglist_item);
748 struct page *page = llap->llap_page;
750 LASSERTF(llap->llap_origin < LLAP__ORIGIN_MAX, "%u\n",
753 seq_printf(seq," %5lu | %p %p %s %s %s %s | %p %lu/%u(%p) "
756 llap, llap->llap_cookie,
757 llap_origins[llap->llap_origin],
758 llap->llap_write_queued ? "wq" : "- ",
759 llap->llap_defer_uptodate ? "du" : "- ",
760 PageWriteback(page) ? "wb" : "-",
761 page, page->mapping->host->i_ino,
762 page->mapping->host->i_generation,
763 page->mapping->host, page->index,
765 seq_page_flag(seq, page, locked, has_flags);
766 seq_page_flag(seq, page, error, has_flags);
767 seq_page_flag(seq, page, referenced, has_flags);
768 seq_page_flag(seq, page, uptodate, has_flags);
769 seq_page_flag(seq, page, dirty, has_flags);
770 #if (LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,12))
771 seq_page_flag(seq, page, highmem, has_flags);
773 seq_page_flag(seq, page, writeback, has_flags);
775 seq_puts(seq, "-]\n");
777 seq_puts(seq, "]\n");
780 spin_unlock(&sbi->ll_lock);
785 static void *llite_dump_pgcache_seq_next(struct seq_file *seq, void *v,
788 struct ll_async_page *llap, *dummy_llap = seq->private;
789 struct ll_sb_info *sbi = dummy_llap->llap_cookie;
791 /* bail if we just displayed the banner */
792 if (dummy_llap->llap_magic == 0) {
793 dummy_llap->llap_magic = 1;
797 /* we've just displayed the llap that is after us in the list.
798 * we advance to a position beyond it, returning null if there
799 * isn't another llap in the list beyond that new position. */
800 spin_lock(&sbi->ll_lock);
801 llap = llite_pglist_next_llap(sbi, &dummy_llap->llap_pglist_item);
802 list_del_init(&dummy_llap->llap_pglist_item);
804 list_add(&dummy_llap->llap_pglist_item,&llap->llap_pglist_item);
805 llap =llite_pglist_next_llap(sbi,&dummy_llap->llap_pglist_item);
807 spin_unlock(&sbi->ll_lock);
811 dummy_llap->llap_magic = 2;
817 static void null_stop(struct seq_file *seq, void *v)
821 struct seq_operations llite_dump_pgcache_seq_sops = {
822 .start = llite_dump_pgcache_seq_start,
824 .next = llite_dump_pgcache_seq_next,
825 .show = llite_dump_pgcache_seq_show,
828 /* we're displaying llaps in a list_head list. we don't want to hold a lock
829 * while we walk the entire list, and we don't want to have to seek into
830 * the right position in the list as an app advances with many syscalls. we
831 * allocate a dummy llap and hang it off file->private. its position in
832 * the list records where the app is currently displaying. this way our
833 * seq .start and .stop don't actually do anything. .next returns null
834 * when the dummy hits the end of the list which eventually leads to .release
835 * where we tear down. this kind of displaying is super-racey, so we put
836 * a generation counter on the list so the output shows when the list
837 * changes between reads.
839 static int llite_dump_pgcache_seq_open(struct inode *inode, struct file *file)
841 struct proc_dir_entry *dp = PDE(inode);
842 struct ll_async_page *dummy_llap;
843 struct seq_file *seq;
844 struct ll_sb_info *sbi = dp->data;
847 LPROCFS_ENTRY_AND_CHECK(dp);
849 OBD_ALLOC_PTR_WAIT(dummy_llap);
850 if (dummy_llap == NULL)
852 dummy_llap->llap_page = NULL;
853 dummy_llap->llap_cookie = sbi;
854 dummy_llap->llap_magic = 0;
856 rc = seq_open(file, &llite_dump_pgcache_seq_sops);
858 OBD_FREE(dummy_llap, sizeof(*dummy_llap));
861 seq = file->private_data;
862 seq->private = dummy_llap;
864 spin_lock(&sbi->ll_lock);
865 list_add(&dummy_llap->llap_pglist_item, &sbi->ll_pglist);
866 spin_unlock(&sbi->ll_lock);
874 static int llite_dump_pgcache_seq_release(struct inode *inode,
877 struct seq_file *seq = file->private_data;
878 struct ll_async_page *dummy_llap = seq->private;
879 struct ll_sb_info *sbi = dummy_llap->llap_cookie;
881 spin_lock(&sbi->ll_lock);
882 if (!list_empty(&dummy_llap->llap_pglist_item))
883 list_del_init(&dummy_llap->llap_pglist_item);
884 spin_unlock(&sbi->ll_lock);
885 OBD_FREE(dummy_llap, sizeof(*dummy_llap));
887 return lprocfs_seq_release(inode, file);
890 struct file_operations llite_dump_pgcache_fops = {
891 .owner = THIS_MODULE,
892 .open = llite_dump_pgcache_seq_open,
894 .release = llite_dump_pgcache_seq_release,
897 static int ll_ra_stats_seq_show(struct seq_file *seq, void *v)
900 struct ll_sb_info *sbi = seq->private;
901 struct ll_ra_info *ra = &sbi->ll_ra_info;
903 static char *ra_stat_strings[] = {
904 [RA_STAT_HIT] = "hits",
905 [RA_STAT_MISS] = "misses",
906 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
907 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
908 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
909 [RA_STAT_FAILED_MATCH] = "failed lock match",
910 [RA_STAT_DISCARDED] = "read but discarded",
911 [RA_STAT_ZERO_LEN] = "zero length file",
912 [RA_STAT_ZERO_WINDOW] = "zero size window",
913 [RA_STAT_EOF] = "read-ahead to EOF",
914 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
915 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
918 do_gettimeofday(&now);
920 spin_lock(&sbi->ll_lock);
922 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
923 now.tv_sec, now.tv_usec);
924 seq_printf(seq, "pending issued pages: %lu\n",
927 for(i = 0; i < _NR_RA_STAT; i++)
928 seq_printf(seq, "%-25s %lu\n", ra_stat_strings[i],
931 spin_unlock(&sbi->ll_lock);
936 static ssize_t ll_ra_stats_seq_write(struct file *file, const char *buf,
937 size_t len, loff_t *off)
939 struct seq_file *seq = file->private_data;
940 struct ll_sb_info *sbi = seq->private;
941 struct ll_ra_info *ra = &sbi->ll_ra_info;
943 spin_lock(&sbi->ll_lock);
944 memset(ra->ra_stats, 0, sizeof(ra->ra_stats));
945 spin_unlock(&sbi->ll_lock);
950 LPROC_SEQ_FOPS(ll_ra_stats);
952 #define pct(a,b) (b ? a * 100 / b : 0)
954 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
955 struct seq_file *seq, int which)
957 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
958 unsigned long start, end, r, w;
959 char *unitp = "KMGTPEZY";
961 struct per_process_info *pp_info = &io_extents->pp_extents[which];
967 for(i = 0; i < LL_HIST_MAX; i++) {
968 read_tot += pp_info->pp_r_hist.oh_buckets[i];
969 write_tot += pp_info->pp_w_hist.oh_buckets[i];
972 for(i = 0; i < LL_HIST_MAX; i++) {
973 r = pp_info->pp_r_hist.oh_buckets[i];
974 w = pp_info->pp_w_hist.oh_buckets[i];
977 end = 1 << (i + LL_HIST_START - units);
978 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4lu %4lu | "
979 "%14lu %4lu %4lu\n", start, *unitp, end, *unitp,
980 (i == LL_HIST_MAX - 1) ? '+' : ' ',
981 r, pct(r, read_tot), pct(read_cum, read_tot),
982 w, pct(w, write_tot), pct(write_cum, write_tot));
984 if (start == 1<<10) {
989 if (read_cum == read_tot && write_cum == write_tot)
994 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
997 struct ll_sb_info *sbi = seq->private;
998 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1001 do_gettimeofday(&now);
1003 if (!sbi->ll_rw_stats_on) {
1004 seq_printf(seq, "Disabled\n"
1005 "Write anything in this file to activate\n");
1008 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
1009 now.tv_sec, now.tv_usec);
1010 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1011 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1012 "extents", "calls", "%", "cum%",
1013 "calls", "%", "cum%");
1014 spin_lock(&sbi->ll_pp_extent_lock);
1015 for(k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1016 if(io_extents->pp_extents[k].pid != 0) {
1017 seq_printf(seq, "\nPID: %d\n",
1018 io_extents->pp_extents[k].pid);
1019 ll_display_extents_info(io_extents, seq, k);
1022 spin_unlock(&sbi->ll_pp_extent_lock);
1026 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1027 const char *buf, size_t len,
1030 struct seq_file *seq = file->private_data;
1031 struct ll_sb_info *sbi = seq->private;
1032 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1035 sbi->ll_rw_stats_on = 1;
1036 spin_lock(&sbi->ll_pp_extent_lock);
1037 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1038 io_extents->pp_extents[i].pid = 0;
1039 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1040 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1042 spin_unlock(&sbi->ll_pp_extent_lock);
1046 LPROC_SEQ_FOPS(ll_rw_extents_stats_pp);
1048 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1051 struct ll_sb_info *sbi = seq->private;
1052 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1054 do_gettimeofday(&now);
1056 if (!sbi->ll_rw_stats_on) {
1057 seq_printf(seq, "Disabled\n"
1058 "Write anything in this file to activate\n");
1061 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
1062 now.tv_sec, now.tv_usec);
1064 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1065 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1066 "extents", "calls", "%", "cum%",
1067 "calls", "%", "cum%");
1068 spin_lock(&sbi->ll_lock);
1069 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1070 spin_unlock(&sbi->ll_lock);
1075 static ssize_t ll_rw_extents_stats_seq_write(struct file *file, const char *buf,
1076 size_t len, loff_t *off)
1078 struct seq_file *seq = file->private_data;
1079 struct ll_sb_info *sbi = seq->private;
1080 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1083 sbi->ll_rw_stats_on = 1;
1084 spin_lock(&sbi->ll_pp_extent_lock);
1085 for(i = 0; i <= LL_PROCESS_HIST_MAX; i++)
1087 io_extents->pp_extents[i].pid = 0;
1088 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1089 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1091 spin_unlock(&sbi->ll_pp_extent_lock);
1096 LPROC_SEQ_FOPS(ll_rw_extents_stats);
1098 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid, struct file
1099 *file, size_t count, int rw)
1102 struct ll_rw_process_info *process;
1103 struct ll_rw_process_info *offset;
1104 int *off_count = &sbi->ll_rw_offset_entry_count;
1105 int *process_count = &sbi->ll_offset_process_count;
1106 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1108 if(!sbi->ll_rw_stats_on)
1110 process = sbi->ll_rw_process_info;
1111 offset = sbi->ll_rw_offset_info;
1113 spin_lock(&sbi->ll_pp_extent_lock);
1114 /* Extent statistics */
1115 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1116 if(io_extents->pp_extents[i].pid == pid) {
1124 sbi->ll_extent_process_count =
1125 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1126 cur = sbi->ll_extent_process_count;
1127 io_extents->pp_extents[cur].pid = pid;
1128 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1129 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1132 for(i = 0; (count >= (1 << LL_HIST_START << i)) &&
1133 (i < (LL_HIST_MAX - 1)); i++);
1135 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1136 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1138 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1139 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1141 spin_unlock(&sbi->ll_pp_extent_lock);
1143 spin_lock(&sbi->ll_process_lock);
1144 /* Offset statistics */
1145 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1146 if (process[i].rw_pid == pid) {
1147 if (process[i].rw_last_file != file) {
1148 process[i].rw_range_start = file->f_pos;
1149 process[i].rw_last_file_pos =
1150 file->f_pos + count;
1151 process[i].rw_smallest_extent = count;
1152 process[i].rw_largest_extent = count;
1153 process[i].rw_offset = 0;
1154 process[i].rw_last_file = file;
1155 spin_unlock(&sbi->ll_process_lock);
1158 if (process[i].rw_last_file_pos != file->f_pos) {
1160 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1161 offset[*off_count].rw_op = process[i].rw_op;
1162 offset[*off_count].rw_pid = pid;
1163 offset[*off_count].rw_range_start =
1164 process[i].rw_range_start;
1165 offset[*off_count].rw_range_end =
1166 process[i].rw_last_file_pos;
1167 offset[*off_count].rw_smallest_extent =
1168 process[i].rw_smallest_extent;
1169 offset[*off_count].rw_largest_extent =
1170 process[i].rw_largest_extent;
1171 offset[*off_count].rw_offset =
1172 process[i].rw_offset;
1173 process[i].rw_op = rw;
1174 process[i].rw_range_start = file->f_pos;
1175 process[i].rw_smallest_extent = count;
1176 process[i].rw_largest_extent = count;
1177 process[i].rw_offset = file->f_pos -
1178 process[i].rw_last_file_pos;
1180 if(process[i].rw_smallest_extent > count)
1181 process[i].rw_smallest_extent = count;
1182 if(process[i].rw_largest_extent < count)
1183 process[i].rw_largest_extent = count;
1184 process[i].rw_last_file_pos = file->f_pos + count;
1185 spin_unlock(&sbi->ll_process_lock);
1189 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1190 process[*process_count].rw_pid = pid;
1191 process[*process_count].rw_op = rw;
1192 process[*process_count].rw_range_start = file->f_pos;
1193 process[*process_count].rw_last_file_pos = file->f_pos + count;
1194 process[*process_count].rw_smallest_extent = count;
1195 process[*process_count].rw_largest_extent = count;
1196 process[*process_count].rw_offset = 0;
1197 process[*process_count].rw_last_file = file;
1198 spin_unlock(&sbi->ll_process_lock);
1201 char lpszt[] = LPSZ;
1203 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1206 struct ll_sb_info *sbi = seq->private;
1207 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1208 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1212 do_gettimeofday(&now);
1214 if (!sbi->ll_rw_stats_on) {
1215 seq_printf(seq, "Disabled\n"
1216 "Write anything in this file to activate\n");
1219 spin_lock(&sbi->ll_process_lock);
1221 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
1222 now.tv_sec, now.tv_usec);
1223 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1224 "R/W", "PID", "RANGE START", "RANGE END",
1225 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1226 sprintf(format, "%s%s%s%s%s\n",
1227 "%3c %10d %14Lu %14Lu %17", lpszt+1, " %17", lpszt+1, " %14Ld");
1228 /* We stored the discontiguous offsets here; print them first */
1229 for(i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1230 if (offset[i].rw_pid != 0)
1231 /* Is there a way to snip the '%' off of LPSZ? */
1232 seq_printf(seq, format,
1233 offset[i].rw_op ? 'W' : 'R',
1235 offset[i].rw_range_start,
1236 offset[i].rw_range_end,
1237 offset[i].rw_smallest_extent,
1238 offset[i].rw_largest_extent,
1239 offset[i].rw_offset);
1241 /* Then print the current offsets for each process */
1242 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1243 if (process[i].rw_pid != 0)
1244 seq_printf(seq, format,
1245 process[i].rw_op ? 'W' : 'R',
1247 process[i].rw_range_start,
1248 process[i].rw_last_file_pos,
1249 process[i].rw_smallest_extent,
1250 process[i].rw_largest_extent,
1251 process[i].rw_offset);
1253 spin_unlock(&sbi->ll_process_lock);
1258 static ssize_t ll_rw_offset_stats_seq_write(struct file *file, const char *buf,
1259 size_t len, loff_t *off)
1261 struct seq_file *seq = file->private_data;
1262 struct ll_sb_info *sbi = seq->private;
1263 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1264 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1266 sbi->ll_rw_stats_on = 1;
1268 spin_lock(&sbi->ll_process_lock);
1269 sbi->ll_offset_process_count = 0;
1270 sbi->ll_rw_offset_entry_count = 0;
1271 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1272 LL_PROCESS_HIST_MAX);
1273 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1274 LL_OFFSET_HIST_MAX);
1275 spin_unlock(&sbi->ll_process_lock);
1280 LPROC_SEQ_FOPS(ll_rw_offset_stats);
1282 void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars)
1284 lvars->module_vars = NULL;
1285 lvars->obd_vars = lprocfs_llite_obd_vars;
1287 #endif /* LPROCFS */