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);
53 rc = snprintf(page, count, "%u\n", osfs.os_bsize);
59 static int ll_rd_kbytestotal(char *page, char **start, off_t off, int count,
62 struct super_block *sb = (struct super_block *)data;
63 struct obd_statfs osfs;
67 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ);
69 __u32 blk_size = osfs.os_bsize >> 10;
70 __u64 result = osfs.os_blocks;
72 while (blk_size >>= 1)
76 rc = snprintf(page, count, LPU64"\n", result);
82 static int ll_rd_kbytesfree(char *page, char **start, off_t off, int count,
85 struct super_block *sb = (struct super_block *)data;
86 struct obd_statfs osfs;
90 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ);
92 __u32 blk_size = osfs.os_bsize >> 10;
93 __u64 result = osfs.os_bfree;
95 while (blk_size >>= 1)
99 rc = snprintf(page, count, LPU64"\n", result);
104 static int ll_rd_kbytesavail(char *page, char **start, off_t off, int count,
105 int *eof, void *data)
107 struct super_block *sb = (struct super_block *)data;
108 struct obd_statfs osfs;
112 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ);
114 __u32 blk_size = osfs.os_bsize >> 10;
115 __u64 result = osfs.os_bavail;
117 while (blk_size >>= 1)
121 rc = snprintf(page, count, LPU64"\n", result);
126 static int ll_rd_filestotal(char *page, char **start, off_t off, int count,
127 int *eof, void *data)
129 struct super_block *sb = (struct super_block *)data;
130 struct obd_statfs osfs;
134 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ);
137 rc = snprintf(page, count, LPU64"\n", osfs.os_files);
142 static int ll_rd_filesfree(char *page, char **start, off_t off, int count,
143 int *eof, void *data)
145 struct super_block *sb = (struct super_block *)data;
146 struct obd_statfs osfs;
150 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ);
153 rc = snprintf(page, count, LPU64"\n", osfs.os_ffree);
159 static int ll_rd_client_type(char *page, char **start, off_t off, int count,
160 int *eof, void *data)
162 struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)data);
165 LASSERT(sbi != NULL);
168 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
169 rc = snprintf(page, count, "remote client\n");
171 rc = snprintf(page, count, "local client\n");
176 static int ll_rd_fstype(char *page, char **start, off_t off, int count,
177 int *eof, void *data)
179 struct super_block *sb = (struct super_block*)data;
183 return snprintf(page, count, "%s\n", sb->s_type->name);
186 static int ll_rd_sb_uuid(char *page, char **start, off_t off, int count,
187 int *eof, void *data)
189 struct super_block *sb = (struct super_block *)data;
193 return snprintf(page, count, "%s\n", ll_s2sbi(sb)->ll_sb_uuid.uuid);
196 static int ll_rd_max_readahead_mb(char *page, char **start, off_t off,
197 int count, int *eof, void *data)
199 struct super_block *sb = data;
200 struct ll_sb_info *sbi = ll_s2sbi(sb);
204 spin_lock(&sbi->ll_lock);
205 pages_number = sbi->ll_ra_info.ra_max_pages;
206 spin_unlock(&sbi->ll_lock);
208 mult = 1 << (20 - PAGE_CACHE_SHIFT);
209 return lprocfs_read_frac_helper(page, count, pages_number, mult);
212 static int ll_wr_max_readahead_mb(struct file *file, const char *buffer,
213 unsigned long count, void *data)
215 struct super_block *sb = data;
216 struct ll_sb_info *sbi = ll_s2sbi(sb);
217 int mult, rc, pages_number;
219 mult = 1 << (20 - CFS_PAGE_SHIFT);
220 rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
224 if (pages_number < 0 || pages_number > num_physpages / 2) {
225 CERROR("can't set file readahead more than %lu MB\n",
226 num_physpages >> (20 - CFS_PAGE_SHIFT + 1)); /*1/2 of RAM*/
230 spin_lock(&sbi->ll_lock);
231 sbi->ll_ra_info.ra_max_pages = pages_number;
232 spin_unlock(&sbi->ll_lock);
237 static int ll_rd_max_read_ahead_whole_mb(char *page, char **start, off_t off,
238 int count, int *eof, void *data)
240 struct super_block *sb = data;
241 struct ll_sb_info *sbi = ll_s2sbi(sb);
245 spin_lock(&sbi->ll_lock);
246 pages_number = sbi->ll_ra_info.ra_max_read_ahead_whole_pages;
247 spin_unlock(&sbi->ll_lock);
249 mult = 1 << (20 - CFS_PAGE_SHIFT);
250 return lprocfs_read_frac_helper(page, count, pages_number, mult);
253 static int ll_wr_max_read_ahead_whole_mb(struct file *file, const char *buffer,
254 unsigned long count, void *data)
256 struct super_block *sb = data;
257 struct ll_sb_info *sbi = ll_s2sbi(sb);
258 int mult, rc, pages_number;
260 mult = 1 << (20 - CFS_PAGE_SHIFT);
261 rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
265 /* Cap this at the current max readahead window size, the readahead
266 * algorithm does this anyway so it's pointless to set it larger. */
267 if (pages_number < 0 || pages_number > sbi->ll_ra_info.ra_max_pages) {
268 CERROR("can't set max_read_ahead_whole_mb more than "
269 "max_read_ahead_mb: %lu\n",
270 sbi->ll_ra_info.ra_max_pages >> (20 - CFS_PAGE_SHIFT));
274 spin_lock(&sbi->ll_lock);
275 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
276 spin_unlock(&sbi->ll_lock);
281 static int ll_rd_max_cached_mb(char *page, char **start, off_t off,
282 int count, int *eof, void *data)
284 struct super_block *sb = data;
285 struct ll_sb_info *sbi = ll_s2sbi(sb);
289 spin_lock(&sbi->ll_lock);
290 pages_number = sbi->ll_async_page_max;
291 spin_unlock(&sbi->ll_lock);
293 mult = 1 << (20 - CFS_PAGE_SHIFT);
294 return lprocfs_read_frac_helper(page, count, pages_number, mult);;
297 static int ll_wr_max_cached_mb(struct file *file, const char *buffer,
298 unsigned long count, void *data)
300 struct super_block *sb = data;
301 struct ll_sb_info *sbi = ll_s2sbi(sb);
302 int mult, rc, pages_number;
304 mult = 1 << (20 - CFS_PAGE_SHIFT);
305 rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
309 if (pages_number < 0 || pages_number > num_physpages) {
310 CERROR("can't set max cache more than %lu MB\n",
311 num_physpages >> (20 - CFS_PAGE_SHIFT));
315 spin_lock(&sbi->ll_lock);
316 sbi->ll_async_page_max = pages_number ;
317 spin_unlock(&sbi->ll_lock);
320 /* Not set up yet, don't call llap_shrink_cache */
323 if (sbi->ll_async_page_count >= sbi->ll_async_page_max)
324 llap_shrink_cache(sbi, 0);
329 static int ll_rd_checksum(char *page, char **start, off_t off,
330 int count, int *eof, void *data)
332 struct super_block *sb = data;
333 struct ll_sb_info *sbi = ll_s2sbi(sb);
335 return snprintf(page, count, "%u\n",
336 (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
339 static int ll_wr_checksum(struct file *file, const char *buffer,
340 unsigned long count, void *data)
342 struct super_block *sb = data;
343 struct ll_sb_info *sbi = ll_s2sbi(sb);
350 rc = lprocfs_write_helper(buffer, count, &val);
354 sbi->ll_flags |= LL_SBI_CHECKSUM;
356 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
358 rc = obd_set_info_async(sbi->ll_dt_exp, strlen("checksum"), "checksum",
359 sizeof(val), &val, NULL);
361 CWARN("Failed to set OSC checksum flags: %d\n", rc);
366 static int ll_rd_max_rw_chunk(char *page, char **start, off_t off,
367 int count, int *eof, void *data)
369 struct super_block *sb = data;
371 return snprintf(page, count, "%lu\n", ll_s2sbi(sb)->ll_max_rw_chunk);
374 static int ll_wr_max_rw_chunk(struct file *file, const char *buffer,
375 unsigned long count, void *data)
377 struct super_block *sb = data;
380 rc = lprocfs_write_helper(buffer, count, &val);
383 ll_s2sbi(sb)->ll_max_rw_chunk = val;
387 static int ll_rd_track_id(char *page, int count, void *data,
388 enum stats_track_type type)
390 struct super_block *sb = data;
392 if (ll_s2sbi(sb)->ll_stats_track_type == type) {
393 return snprintf(page, count, "%d\n",
394 ll_s2sbi(sb)->ll_stats_track_id);
396 } else if (ll_s2sbi(sb)->ll_stats_track_type == STATS_TRACK_ALL) {
397 return snprintf(page, count, "0 (all)\n");
399 return snprintf(page, count, "untracked\n");
403 static int ll_wr_track_id(const char *buffer, unsigned long count, void *data,
404 enum stats_track_type type)
406 struct super_block *sb = data;
409 rc = lprocfs_write_helper(buffer, count, &pid);
412 ll_s2sbi(sb)->ll_stats_track_id = pid;
414 ll_s2sbi(sb)->ll_stats_track_type = STATS_TRACK_ALL;
416 ll_s2sbi(sb)->ll_stats_track_type = type;
417 lprocfs_clear_stats(ll_s2sbi(sb)->ll_stats);
421 static int ll_rd_track_pid(char *page, char **start, off_t off,
422 int count, int *eof, void *data)
424 return (ll_rd_track_id(page, count, data, STATS_TRACK_PID));
427 static int ll_wr_track_pid(struct file *file, const char *buffer,
428 unsigned long count, void *data)
430 return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PID));
433 static int ll_rd_track_ppid(char *page, char **start, off_t off,
434 int count, int *eof, void *data)
436 return (ll_rd_track_id(page, count, data, STATS_TRACK_PPID));
439 static int ll_wr_track_ppid(struct file *file, const char *buffer,
440 unsigned long count, void *data)
442 return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PPID));
445 static int ll_rd_track_gid(char *page, char **start, off_t off,
446 int count, int *eof, void *data)
448 return (ll_rd_track_id(page, count, data, STATS_TRACK_GID));
451 static int ll_wr_track_gid(struct file *file, const char *buffer,
452 unsigned long count, void *data)
454 return (ll_wr_track_id(buffer, count, data, STATS_TRACK_GID));
457 static struct lprocfs_vars lprocfs_obd_vars[] = {
458 { "uuid", ll_rd_sb_uuid, 0, 0 },
459 //{ "mntpt_path", ll_rd_path, 0, 0 },
460 { "fstype", ll_rd_fstype, 0, 0 },
461 { "blocksize", ll_rd_blksize, 0, 0 },
462 { "kbytestotal", ll_rd_kbytestotal, 0, 0 },
463 { "kbytesfree", ll_rd_kbytesfree, 0, 0 },
464 { "kbytesavail", ll_rd_kbytesavail, 0, 0 },
465 { "filestotal", ll_rd_filestotal, 0, 0 },
466 { "filesfree", ll_rd_filesfree, 0, 0 },
467 { "client_type", ll_rd_client_type, 0, 0 },
468 //{ "filegroups", lprocfs_rd_filegroups, 0, 0 },
469 { "max_read_ahead_mb", ll_rd_max_readahead_mb,
470 ll_wr_max_readahead_mb, 0 },
471 { "max_read_ahead_whole_mb", ll_rd_max_read_ahead_whole_mb,
472 ll_wr_max_read_ahead_whole_mb, 0 },
473 { "max_cached_mb", ll_rd_max_cached_mb, ll_wr_max_cached_mb, 0 },
474 { "checksum_pages", ll_rd_checksum, ll_wr_checksum, 0 },
475 { "max_rw_chunk", ll_rd_max_rw_chunk, ll_wr_max_rw_chunk, 0 },
476 { "stats_track_pid", ll_rd_track_pid, ll_wr_track_pid, 0 },
477 { "stats_track_ppid", ll_rd_track_ppid, ll_wr_track_ppid, 0 },
478 { "stats_track_gid", ll_rd_track_gid, ll_wr_track_gid, 0 },
482 #define MAX_STRING_SIZE 128
484 struct llite_file_opcode {
488 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
490 { LPROC_LL_DIRTY_HITS, LPROCFS_TYPE_REGS, "dirty_pages_hits" },
491 { LPROC_LL_DIRTY_MISSES, LPROCFS_TYPE_REGS, "dirty_pages_misses" },
492 { LPROC_LL_WB_WRITEPAGE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
493 "writeback_from_writepage" },
494 { LPROC_LL_WB_PRESSURE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
495 "writeback_from_pressure" },
496 { LPROC_LL_WB_OK, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
497 "writeback_ok_pages" },
498 { LPROC_LL_WB_FAIL, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
499 "writeback_failed_pages" },
500 { LPROC_LL_READ_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
502 { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
504 { LPROC_LL_BRW_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
506 { LPROC_LL_BRW_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
509 { LPROC_LL_IOCTL, LPROCFS_TYPE_REGS, "ioctl" },
510 { LPROC_LL_OPEN, LPROCFS_TYPE_REGS, "open" },
511 { LPROC_LL_RELEASE, LPROCFS_TYPE_REGS, "close" },
512 { LPROC_LL_MAP, LPROCFS_TYPE_REGS, "mmap" },
513 { LPROC_LL_LLSEEK, LPROCFS_TYPE_REGS, "seek" },
514 { LPROC_LL_FSYNC, LPROCFS_TYPE_REGS, "fsync" },
515 /* inode operation */
516 { LPROC_LL_SETATTR, LPROCFS_TYPE_REGS, "setattr" },
517 { LPROC_LL_TRUNC, LPROCFS_TYPE_REGS, "truncate" },
518 { LPROC_LL_FLOCK, LPROCFS_TYPE_REGS, "flock" },
519 { LPROC_LL_GETATTR, LPROCFS_TYPE_REGS, "getattr" },
520 /* special inode operation */
521 { LPROC_LL_STAFS, LPROCFS_TYPE_REGS, "statfs" },
522 { LPROC_LL_ALLOC_INODE, LPROCFS_TYPE_REGS, "alloc_inode" },
523 { LPROC_LL_SETXATTR, LPROCFS_TYPE_REGS, "setxattr" },
524 { LPROC_LL_GETXATTR, LPROCFS_TYPE_REGS, "getxattr" },
525 { LPROC_LL_LISTXATTR, LPROCFS_TYPE_REGS, "listxattr" },
526 { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_REGS, "removexattr" },
527 { LPROC_LL_INODE_PERM, LPROCFS_TYPE_REGS, "inode_permission" },
528 { LPROC_LL_DIRECT_READ, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
530 { LPROC_LL_DIRECT_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
535 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
539 if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
540 lprocfs_counter_add(sbi->ll_stats, op, count);
541 else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
542 sbi->ll_stats_track_id == current->pid)
543 lprocfs_counter_add(sbi->ll_stats, op, count);
544 else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
545 sbi->ll_stats_track_id == current->p_pptr->pid)
546 lprocfs_counter_add(sbi->ll_stats, op, count);
547 else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
548 sbi->ll_stats_track_id == current->gid)
549 lprocfs_counter_add(sbi->ll_stats, op, count);
551 EXPORT_SYMBOL(ll_stats_ops_tally);
553 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
554 struct super_block *sb, char *osc, char *mdc)
556 struct lprocfs_vars lvars[2];
557 struct lustre_sb_info *lsi = s2lsi(sb);
558 struct ll_sb_info *sbi = ll_s2sbi(sb);
559 struct obd_device *obd;
560 char name[MAX_STRING_SIZE + 1], *ptr;
562 struct proc_dir_entry *entry;
565 memset(lvars, 0, sizeof(lvars));
567 name[MAX_STRING_SIZE] = '\0';
568 lvars[0].name = name;
570 LASSERT(sbi != NULL);
571 LASSERT(mdc != NULL);
572 LASSERT(osc != NULL);
575 len = strlen(lsi->lsi_lmd->lmd_profile);
576 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
577 if (ptr && (strcmp(ptr, "-client") == 0))
581 snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
582 lsi->lsi_lmd->lmd_profile, sb);
584 sbi->ll_proc_root = lprocfs_register(name, parent, NULL, NULL);
585 if (IS_ERR(sbi->ll_proc_root)) {
586 err = PTR_ERR(sbi->ll_proc_root);
587 sbi->ll_proc_root = NULL;
591 entry = create_proc_entry("dump_page_cache", 0444, sbi->ll_proc_root);
593 GOTO(out, err = -ENOMEM);
594 entry->proc_fops = &llite_dump_pgcache_fops;
597 entry = create_proc_entry("read_ahead_stats", 0644, sbi->ll_proc_root);
599 GOTO(out, err = -ENOMEM);
600 entry->proc_fops = &ll_ra_stats_fops;
603 entry = create_proc_entry("extents_stats", 0644, sbi->ll_proc_root);
605 GOTO(out, err = -ENOMEM);
606 entry->proc_fops = &ll_rw_extents_stats_fops;
609 entry = create_proc_entry("extents_stats_per_process", 0644,
612 GOTO(out, err = -ENOMEM);
613 entry->proc_fops = &ll_rw_extents_stats_pp_fops;
616 entry = create_proc_entry("offset_stats", 0644, sbi->ll_proc_root);
618 GOTO(out, err = -ENOMEM);
619 entry->proc_fops = &ll_rw_offset_stats_fops;
622 /* File operations stats */
623 sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES, 0);
624 if (sbi->ll_stats == NULL)
625 GOTO(out, err = -ENOMEM);
626 /* do counter init */
627 for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
628 __u32 type = llite_opcode_table[id].type;
630 if (type & LPROCFS_TYPE_REGS)
633 if (type & LPROCFS_TYPE_BYTES)
636 if (type & LPROCFS_TYPE_PAGES)
640 lprocfs_counter_init(sbi->ll_stats,
641 llite_opcode_table[id].opcode,
642 (type & LPROCFS_CNTR_AVGMINMAX),
643 llite_opcode_table[id].opname, ptr);
645 err = lprocfs_register_stats(sbi->ll_proc_root, "stats", sbi->ll_stats);
649 err = lprocfs_add_vars(sbi->ll_proc_root, lprocfs_obd_vars, sb);
654 obd = class_name2obd(mdc);
656 LASSERT(obd != NULL);
657 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
658 LASSERT(obd->obd_type->typ_name != NULL);
660 snprintf(name, MAX_STRING_SIZE, "%s/common_name",
661 obd->obd_type->typ_name);
662 lvars[0].read_fptr = lprocfs_rd_name;
663 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
667 snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
668 lvars[0].read_fptr = lprocfs_rd_uuid;
669 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
674 obd = class_name2obd(osc);
676 LASSERT(obd != NULL);
677 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
678 LASSERT(obd->obd_type->typ_name != NULL);
680 snprintf(name, MAX_STRING_SIZE, "%s/common_name",
681 obd->obd_type->typ_name);
682 lvars[0].read_fptr = lprocfs_rd_name;
683 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
687 snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
688 lvars[0].read_fptr = lprocfs_rd_uuid;
689 err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
692 lprocfs_remove(&sbi->ll_proc_root);
693 lprocfs_free_stats(&sbi->ll_stats);
698 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi)
700 if (sbi->ll_proc_root) {
701 lprocfs_remove(&sbi->ll_proc_root);
702 lprocfs_free_stats(&sbi->ll_stats);
705 #undef MAX_STRING_SIZE
707 #define seq_page_flag(seq, page, flag, has_flags) do { \
708 if (test_bit(PG_##flag, &(page)->flags)) { \
712 seq_putc(seq, '|'); \
713 seq_puts(seq, #flag); \
717 static void *llite_dump_pgcache_seq_start(struct seq_file *seq, loff_t *pos)
719 struct ll_async_page *dummy_llap = seq->private;
721 if (dummy_llap->llap_magic == 2)
727 static int llite_dump_pgcache_seq_show(struct seq_file *seq, void *v)
729 struct ll_async_page *llap, *dummy_llap = seq->private;
730 struct ll_sb_info *sbi = dummy_llap->llap_cookie;
732 /* 2.4 doesn't seem to have SEQ_START_TOKEN, so we implement
733 * it in our own state */
734 if (dummy_llap->llap_magic == 0) {
735 seq_printf(seq, "gener | llap cookie origin wq du wb | page "
736 "inode index count [ page flags ]\n");
740 spin_lock(&sbi->ll_lock);
742 llap = llite_pglist_next_llap(sbi, &dummy_llap->llap_pglist_item);
745 struct page *page = llap->llap_page;
747 LASSERTF(llap->llap_origin < LLAP__ORIGIN_MAX, "%u\n",
750 seq_printf(seq," %5lu | %p %p %s %s %s %s | %p %lu/%u(%p) "
753 llap, llap->llap_cookie,
754 llap_origins[llap->llap_origin],
755 llap->llap_write_queued ? "wq" : "- ",
756 llap->llap_defer_uptodate ? "du" : "- ",
757 PageWriteback(page) ? "wb" : "-",
758 page, page->mapping->host->i_ino,
759 page->mapping->host->i_generation,
760 page->mapping->host, page->index,
762 seq_page_flag(seq, page, locked, has_flags);
763 seq_page_flag(seq, page, error, has_flags);
764 seq_page_flag(seq, page, referenced, has_flags);
765 seq_page_flag(seq, page, uptodate, has_flags);
766 seq_page_flag(seq, page, dirty, has_flags);
767 #if (LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,12))
768 seq_page_flag(seq, page, highmem, has_flags);
770 seq_page_flag(seq, page, writeback, has_flags);
772 seq_puts(seq, "-]\n");
774 seq_puts(seq, "]\n");
777 spin_unlock(&sbi->ll_lock);
782 static void *llite_dump_pgcache_seq_next(struct seq_file *seq, void *v,
785 struct ll_async_page *llap, *dummy_llap = seq->private;
786 struct ll_sb_info *sbi = dummy_llap->llap_cookie;
788 /* bail if we just displayed the banner */
789 if (dummy_llap->llap_magic == 0) {
790 dummy_llap->llap_magic = 1;
794 /* we've just displayed the llap that is after us in the list.
795 * we advance to a position beyond it, returning null if there
796 * isn't another llap in the list beyond that new position. */
797 spin_lock(&sbi->ll_lock);
798 llap = llite_pglist_next_llap(sbi, &dummy_llap->llap_pglist_item);
799 list_del_init(&dummy_llap->llap_pglist_item);
801 list_add(&dummy_llap->llap_pglist_item,&llap->llap_pglist_item);
802 llap =llite_pglist_next_llap(sbi,&dummy_llap->llap_pglist_item);
804 spin_unlock(&sbi->ll_lock);
808 dummy_llap->llap_magic = 2;
814 static void null_stop(struct seq_file *seq, void *v)
818 struct seq_operations llite_dump_pgcache_seq_sops = {
819 .start = llite_dump_pgcache_seq_start,
821 .next = llite_dump_pgcache_seq_next,
822 .show = llite_dump_pgcache_seq_show,
825 /* we're displaying llaps in a list_head list. we don't want to hold a lock
826 * while we walk the entire list, and we don't want to have to seek into
827 * the right position in the list as an app advances with many syscalls. we
828 * allocate a dummy llap and hang it off file->private. its position in
829 * the list records where the app is currently displaying. this way our
830 * seq .start and .stop don't actually do anything. .next returns null
831 * when the dummy hits the end of the list which eventually leads to .release
832 * where we tear down. this kind of displaying is super-racey, so we put
833 * a generation counter on the list so the output shows when the list
834 * changes between reads.
836 static int llite_dump_pgcache_seq_open(struct inode *inode, struct file *file)
838 struct proc_dir_entry *dp = PDE(inode);
839 struct ll_async_page *dummy_llap;
840 struct seq_file *seq;
841 struct ll_sb_info *sbi = dp->data;
844 LPROCFS_ENTRY_AND_CHECK(dp);
846 OBD_ALLOC_PTR_WAIT(dummy_llap);
847 if (dummy_llap == NULL)
849 dummy_llap->llap_page = NULL;
850 dummy_llap->llap_cookie = sbi;
851 dummy_llap->llap_magic = 0;
853 rc = seq_open(file, &llite_dump_pgcache_seq_sops);
855 OBD_FREE(dummy_llap, sizeof(*dummy_llap));
858 seq = file->private_data;
859 seq->private = dummy_llap;
861 spin_lock(&sbi->ll_lock);
862 list_add(&dummy_llap->llap_pglist_item, &sbi->ll_pglist);
863 spin_unlock(&sbi->ll_lock);
871 static int llite_dump_pgcache_seq_release(struct inode *inode,
874 struct seq_file *seq = file->private_data;
875 struct ll_async_page *dummy_llap = seq->private;
876 struct ll_sb_info *sbi = dummy_llap->llap_cookie;
878 spin_lock(&sbi->ll_lock);
879 if (!list_empty(&dummy_llap->llap_pglist_item))
880 list_del_init(&dummy_llap->llap_pglist_item);
881 spin_unlock(&sbi->ll_lock);
882 OBD_FREE(dummy_llap, sizeof(*dummy_llap));
884 return lprocfs_seq_release(inode, file);
887 struct file_operations llite_dump_pgcache_fops = {
888 .owner = THIS_MODULE,
889 .open = llite_dump_pgcache_seq_open,
891 .release = llite_dump_pgcache_seq_release,
894 static int ll_ra_stats_seq_show(struct seq_file *seq, void *v)
897 struct ll_sb_info *sbi = seq->private;
898 struct ll_ra_info *ra = &sbi->ll_ra_info;
900 static char *ra_stat_strings[] = {
901 [RA_STAT_HIT] = "hits",
902 [RA_STAT_MISS] = "misses",
903 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
904 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
905 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
906 [RA_STAT_FAILED_MATCH] = "failed lock match",
907 [RA_STAT_DISCARDED] = "read but discarded",
908 [RA_STAT_ZERO_LEN] = "zero length file",
909 [RA_STAT_ZERO_WINDOW] = "zero size window",
910 [RA_STAT_EOF] = "read-ahead to EOF",
911 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
912 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
915 do_gettimeofday(&now);
917 spin_lock(&sbi->ll_lock);
919 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
920 now.tv_sec, now.tv_usec);
921 seq_printf(seq, "pending issued pages: %lu\n",
924 for(i = 0; i < _NR_RA_STAT; i++)
925 seq_printf(seq, "%-25s %lu\n", ra_stat_strings[i],
928 spin_unlock(&sbi->ll_lock);
933 static ssize_t ll_ra_stats_seq_write(struct file *file, const char *buf,
934 size_t len, loff_t *off)
936 struct seq_file *seq = file->private_data;
937 struct ll_sb_info *sbi = seq->private;
938 struct ll_ra_info *ra = &sbi->ll_ra_info;
940 spin_lock(&sbi->ll_lock);
941 memset(ra->ra_stats, 0, sizeof(ra->ra_stats));
942 spin_unlock(&sbi->ll_lock);
947 LPROC_SEQ_FOPS(ll_ra_stats);
949 #define pct(a,b) (b ? a * 100 / b : 0)
951 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
952 struct seq_file *seq, int which)
954 unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
955 unsigned long start, end, r, w;
956 char *unitp = "KMGTPEZY";
958 struct per_process_info *pp_info = &io_extents->pp_extents[which];
964 for(i = 0; i < LL_HIST_MAX; i++) {
965 read_tot += pp_info->pp_r_hist.oh_buckets[i];
966 write_tot += pp_info->pp_w_hist.oh_buckets[i];
969 for(i = 0; i < LL_HIST_MAX; i++) {
970 r = pp_info->pp_r_hist.oh_buckets[i];
971 w = pp_info->pp_w_hist.oh_buckets[i];
974 end = 1 << (i + LL_HIST_START - units);
975 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4lu %4lu | "
976 "%14lu %4lu %4lu\n", start, *unitp, end, *unitp,
977 (i == LL_HIST_MAX - 1) ? '+' : ' ',
978 r, pct(r, read_tot), pct(read_cum, read_tot),
979 w, pct(w, write_tot), pct(write_cum, write_tot));
981 if (start == 1<<10) {
986 if (read_cum == read_tot && write_cum == write_tot)
991 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
994 struct ll_sb_info *sbi = seq->private;
995 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
998 do_gettimeofday(&now);
1000 if (!sbi->ll_rw_stats_on) {
1001 seq_printf(seq, "Disabled\n"
1002 "Write anything in this file to activate\n");
1005 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
1006 now.tv_sec, now.tv_usec);
1007 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1008 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1009 "extents", "calls", "%", "cum%",
1010 "calls", "%", "cum%");
1011 spin_lock(&sbi->ll_pp_extent_lock);
1012 for(k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1013 if(io_extents->pp_extents[k].pid != 0) {
1014 seq_printf(seq, "\nPID: %d\n",
1015 io_extents->pp_extents[k].pid);
1016 ll_display_extents_info(io_extents, seq, k);
1019 spin_unlock(&sbi->ll_pp_extent_lock);
1023 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1024 const char *buf, size_t len,
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;
1032 sbi->ll_rw_stats_on = 1;
1033 spin_lock(&sbi->ll_pp_extent_lock);
1034 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1035 io_extents->pp_extents[i].pid = 0;
1036 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1037 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1039 spin_unlock(&sbi->ll_pp_extent_lock);
1043 LPROC_SEQ_FOPS(ll_rw_extents_stats_pp);
1045 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1048 struct ll_sb_info *sbi = seq->private;
1049 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1051 do_gettimeofday(&now);
1053 if (!sbi->ll_rw_stats_on) {
1054 seq_printf(seq, "Disabled\n"
1055 "Write anything in this file to activate\n");
1058 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
1059 now.tv_sec, now.tv_usec);
1061 seq_printf(seq, "%15s %19s | %20s\n", " ", "read", "write");
1062 seq_printf(seq, "%13s %14s %4s %4s | %14s %4s %4s\n",
1063 "extents", "calls", "%", "cum%",
1064 "calls", "%", "cum%");
1065 spin_lock(&sbi->ll_lock);
1066 ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1067 spin_unlock(&sbi->ll_lock);
1072 static ssize_t ll_rw_extents_stats_seq_write(struct file *file, const char *buf,
1073 size_t len, loff_t *off)
1075 struct seq_file *seq = file->private_data;
1076 struct ll_sb_info *sbi = seq->private;
1077 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1080 sbi->ll_rw_stats_on = 1;
1081 spin_lock(&sbi->ll_pp_extent_lock);
1082 for(i = 0; i <= LL_PROCESS_HIST_MAX; i++)
1084 io_extents->pp_extents[i].pid = 0;
1085 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1086 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1088 spin_unlock(&sbi->ll_pp_extent_lock);
1093 LPROC_SEQ_FOPS(ll_rw_extents_stats);
1095 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid, struct file
1096 *file, size_t count, int rw)
1099 struct ll_rw_process_info *process;
1100 struct ll_rw_process_info *offset;
1101 int *off_count = &sbi->ll_rw_offset_entry_count;
1102 int *process_count = &sbi->ll_offset_process_count;
1103 struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1105 if(!sbi->ll_rw_stats_on)
1107 process = sbi->ll_rw_process_info;
1108 offset = sbi->ll_rw_offset_info;
1110 spin_lock(&sbi->ll_pp_extent_lock);
1111 /* Extent statistics */
1112 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1113 if(io_extents->pp_extents[i].pid == pid) {
1121 sbi->ll_extent_process_count =
1122 (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1123 cur = sbi->ll_extent_process_count;
1124 io_extents->pp_extents[cur].pid = pid;
1125 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1126 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1129 for(i = 0; (count >= (1 << LL_HIST_START << i)) &&
1130 (i < (LL_HIST_MAX - 1)); i++);
1132 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1133 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1135 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1136 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1138 spin_unlock(&sbi->ll_pp_extent_lock);
1140 spin_lock(&sbi->ll_process_lock);
1141 /* Offset statistics */
1142 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1143 if (process[i].rw_pid == pid) {
1144 if (process[i].rw_last_file != file) {
1145 process[i].rw_range_start = file->f_pos;
1146 process[i].rw_last_file_pos =
1147 file->f_pos + count;
1148 process[i].rw_smallest_extent = count;
1149 process[i].rw_largest_extent = count;
1150 process[i].rw_offset = 0;
1151 process[i].rw_last_file = file;
1152 spin_unlock(&sbi->ll_process_lock);
1155 if (process[i].rw_last_file_pos != file->f_pos) {
1157 (*off_count + 1) % LL_OFFSET_HIST_MAX;
1158 offset[*off_count].rw_op = process[i].rw_op;
1159 offset[*off_count].rw_pid = pid;
1160 offset[*off_count].rw_range_start =
1161 process[i].rw_range_start;
1162 offset[*off_count].rw_range_end =
1163 process[i].rw_last_file_pos;
1164 offset[*off_count].rw_smallest_extent =
1165 process[i].rw_smallest_extent;
1166 offset[*off_count].rw_largest_extent =
1167 process[i].rw_largest_extent;
1168 offset[*off_count].rw_offset =
1169 process[i].rw_offset;
1170 process[i].rw_op = rw;
1171 process[i].rw_range_start = file->f_pos;
1172 process[i].rw_smallest_extent = count;
1173 process[i].rw_largest_extent = count;
1174 process[i].rw_offset = file->f_pos -
1175 process[i].rw_last_file_pos;
1177 if(process[i].rw_smallest_extent > count)
1178 process[i].rw_smallest_extent = count;
1179 if(process[i].rw_largest_extent < count)
1180 process[i].rw_largest_extent = count;
1181 process[i].rw_last_file_pos = file->f_pos + count;
1182 spin_unlock(&sbi->ll_process_lock);
1186 *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1187 process[*process_count].rw_pid = pid;
1188 process[*process_count].rw_op = rw;
1189 process[*process_count].rw_range_start = file->f_pos;
1190 process[*process_count].rw_last_file_pos = file->f_pos + count;
1191 process[*process_count].rw_smallest_extent = count;
1192 process[*process_count].rw_largest_extent = count;
1193 process[*process_count].rw_offset = 0;
1194 process[*process_count].rw_last_file = file;
1195 spin_unlock(&sbi->ll_process_lock);
1198 char lpszt[] = LPSZ;
1200 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1203 struct ll_sb_info *sbi = seq->private;
1204 struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1205 struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1209 do_gettimeofday(&now);
1211 if (!sbi->ll_rw_stats_on) {
1212 seq_printf(seq, "Disabled\n"
1213 "Write anything in this file to activate\n");
1216 spin_lock(&sbi->ll_process_lock);
1218 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
1219 now.tv_sec, now.tv_usec);
1220 seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1221 "R/W", "PID", "RANGE START", "RANGE END",
1222 "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1223 sprintf(format, "%s%s%s%s%s\n",
1224 "%3c %10d %14Lu %14Lu %17", lpszt+1, " %17", lpszt+1, " %14Ld");
1225 /* We stored the discontiguous offsets here; print them first */
1226 for(i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1227 if (offset[i].rw_pid != 0)
1228 /* Is there a way to snip the '%' off of LPSZ? */
1229 seq_printf(seq, format,
1230 offset[i].rw_op ? 'W' : 'R',
1232 offset[i].rw_range_start,
1233 offset[i].rw_range_end,
1234 offset[i].rw_smallest_extent,
1235 offset[i].rw_largest_extent,
1236 offset[i].rw_offset);
1238 /* Then print the current offsets for each process */
1239 for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1240 if (process[i].rw_pid != 0)
1241 seq_printf(seq, format,
1242 process[i].rw_op ? 'W' : 'R',
1244 process[i].rw_range_start,
1245 process[i].rw_last_file_pos,
1246 process[i].rw_smallest_extent,
1247 process[i].rw_largest_extent,
1248 process[i].rw_offset);
1250 spin_unlock(&sbi->ll_process_lock);
1255 static ssize_t ll_rw_offset_stats_seq_write(struct file *file, const char *buf,
1256 size_t len, loff_t *off)
1258 struct seq_file *seq = file->private_data;
1259 struct ll_sb_info *sbi = seq->private;
1260 struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1261 struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1263 sbi->ll_rw_stats_on = 1;
1265 spin_lock(&sbi->ll_process_lock);
1266 sbi->ll_offset_process_count = 0;
1267 sbi->ll_rw_offset_entry_count = 0;
1268 memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1269 LL_PROCESS_HIST_MAX);
1270 memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1271 LL_OFFSET_HIST_MAX);
1272 spin_unlock(&sbi->ll_process_lock);
1277 LPROC_SEQ_FOPS(ll_rw_offset_stats);
1279 LPROCFS_INIT_VARS(llite, NULL, lprocfs_obd_vars)
1280 #endif /* LPROCFS */