4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/mdt/mdt_lproc.c
34 * Author: Lai Siyao <lsy@clusterfs.com>
35 * Author: Fan Yong <fanyong@clusterfs.com>
38 #define DEBUG_SUBSYSTEM S_MDS
40 #include <linux/version.h>
41 #include <asm/statfs.h>
43 #include <linux/module.h>
44 #include <uapi/linux/lnet/nidstr.h>
45 /* LUSTRE_VERSION_CODE */
46 #include <uapi/linux/lustre/lustre_ver.h>
48 * struct OBD_{ALLOC,FREE}*()
51 #include <obd_support.h>
52 /* struct obd_export */
53 #include <lustre_export.h>
54 /* struct obd_device */
56 #include <obd_class.h>
57 #include <lustre_mds.h>
58 #include <lprocfs_status.h>
59 #include "mdt_internal.h"
62 * The rename stats output would be YAML formats, like
64 * - snapshot_time: 1234567890.123456
66 * 4kB: { samples: 1230, pct: 33, cum_pct: 45 }
67 * 8kB: { samples: 1242, pct: 33, cum_pct: 78 }
68 * 16kB: { samples: 132, pct: 3, cum_pct: 81 }
70 * 4kB: { samples: 123, pct: 33, cum_pct: 45 }
71 * 8kB: { samples: 124, pct: 33, cum_pct: 78 }
72 * 16kB: { samples: 12, pct: 3, cum_pct: 81 }
74 * 4kB: { samples: 123, pct: 33, cum_pct: 45 }
75 * 8kB: { samples: 124, pct: 33, cum_pct: 78 }
76 * 16kB: { samples: 12, pct: 3, cum_pct: 81 }
79 #define pct(a, b) (b ? a * 100 / b : 0)
81 static void display_rename_stats(struct seq_file *seq, char *name,
82 struct obd_histogram *hist)
84 unsigned long tot, t, cum = 0;
87 tot = lprocfs_oh_sum(hist);
89 seq_printf(seq, "- %-15s\n", name);
90 /* dir size start from 4K, start i from 10(2^10) here */
91 for (i = 0; i < OBD_HIST_MAX; i++) {
92 t = hist->oh_buckets[i];
98 seq_printf(seq, "%6s%d%s", " ", 1<< i, "bytes:");
100 seq_printf(seq, "%6s%d%s", " ", 1<<(i-10), "KB:");
102 seq_printf(seq, "%6s%d%s", " ", 1<<(i-20), "MB:");
104 seq_printf(seq, " { sample: %3lu, pct: %3lu, cum_pct: %3lu }\n",
105 t, pct(t, tot), pct(cum, tot));
112 static void rename_stats_show(struct seq_file *seq,
113 struct rename_stats *rename_stats)
115 struct timespec64 now;
117 /* this sampling races with updates */
118 ktime_get_real_ts64(&now);
119 seq_printf(seq, "rename_stats:\n");
120 seq_printf(seq, "- %-15s %llu.%9lu\n", "snapshot_time:",
121 (s64)now.tv_sec, now.tv_nsec);
123 display_rename_stats(seq, "same_dir",
124 &rename_stats->hist[RENAME_SAMEDIR_SIZE]);
125 display_rename_stats(seq, "crossdir_src",
126 &rename_stats->hist[RENAME_CROSSDIR_SRC_SIZE]);
127 display_rename_stats(seq, "crossdir_tgt",
128 &rename_stats->hist[RENAME_CROSSDIR_TGT_SIZE]);
133 static int mdt_rename_stats_seq_show(struct seq_file *seq, void *v)
135 struct mdt_device *mdt = seq->private;
137 rename_stats_show(seq, &mdt->mdt_rename_stats);
143 mdt_rename_stats_seq_write(struct file *file, const char __user *buf,
144 size_t len, loff_t *off)
146 struct seq_file *seq = file->private_data;
147 struct mdt_device *mdt = seq->private;
150 for (i = 0; i < RENAME_LAST; i++)
151 lprocfs_oh_clear(&mdt->mdt_rename_stats.hist[i]);
155 LPROC_SEQ_FOPS(mdt_rename_stats);
157 static int lproc_mdt_attach_rename_seqstat(struct mdt_device *mdt)
161 for (i = 0; i < RENAME_LAST; i++)
162 spin_lock_init(&mdt->mdt_rename_stats.hist[i].oh_lock);
164 return lprocfs_obd_seq_create(mdt2obd_dev(mdt), "rename_stats", 0644,
165 &mdt_rename_stats_fops, mdt);
168 void mdt_rename_counter_tally(struct mdt_thread_info *info,
169 struct mdt_device *mdt,
170 struct ptlrpc_request *req,
171 struct mdt_object *src,
172 struct mdt_object *tgt)
174 struct md_attr *ma = &info->mti_attr;
175 struct rename_stats *rstats = &mdt->mdt_rename_stats;
178 ma->ma_need = MA_INODE;
180 rc = mo_attr_get(info->mti_env, mdt_object_child(src), ma);
182 CERROR("%s: "DFID" attr_get, rc = %d\n",
183 mdt_obd_name(mdt), PFID(mdt_object_fid(src)), rc);
188 mdt_counter_incr(req, LPROC_MDT_SAMEDIR_RENAME);
189 lprocfs_oh_tally_log2(&rstats->hist[RENAME_SAMEDIR_SIZE],
190 (unsigned int)ma->ma_attr.la_size);
194 mdt_counter_incr(req, LPROC_MDT_CROSSDIR_RENAME);
195 lprocfs_oh_tally_log2(&rstats->hist[RENAME_CROSSDIR_SRC_SIZE],
196 (unsigned int)ma->ma_attr.la_size);
198 ma->ma_need = MA_INODE;
200 rc = mo_attr_get(info->mti_env, mdt_object_child(tgt), ma);
202 CERROR("%s: "DFID" attr_get, rc = %d\n",
203 mdt_obd_name(mdt), PFID(mdt_object_fid(tgt)), rc);
207 lprocfs_oh_tally_log2(&rstats->hist[RENAME_CROSSDIR_TGT_SIZE],
208 (unsigned int)ma->ma_attr.la_size);
211 static int mdt_identity_expire_seq_show(struct seq_file *m, void *data)
213 struct obd_device *obd = m->private;
214 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
216 seq_printf(m, "%lld\n", mdt->mdt_identity_cache->uc_entry_expire);
221 mdt_identity_expire_seq_write(struct file *file, const char __user *buffer,
222 size_t count, loff_t *off)
224 struct seq_file *m = file->private_data;
225 struct obd_device *obd = m->private;
226 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
230 rc = kstrtoll_from_user(buffer, count, 0, &val);
237 mdt->mdt_identity_cache->uc_entry_expire = val;
241 LPROC_SEQ_FOPS(mdt_identity_expire);
243 static int mdt_identity_acquire_expire_seq_show(struct seq_file *m, void *data)
245 struct obd_device *obd = m->private;
246 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
248 seq_printf(m, "%lld\n", mdt->mdt_identity_cache->uc_acquire_expire);
253 mdt_identity_acquire_expire_seq_write(struct file *file,
254 const char __user *buffer,
255 size_t count, loff_t *off)
257 struct seq_file *m = file->private_data;
258 struct obd_device *obd = m->private;
259 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
263 rc = kstrtoll_from_user(buffer, count, 0, &val);
267 if (val < 0 || val > INT_MAX)
270 mdt->mdt_identity_cache->uc_acquire_expire = val;
274 LPROC_SEQ_FOPS(mdt_identity_acquire_expire);
276 static int mdt_identity_upcall_seq_show(struct seq_file *m, void *data)
278 struct obd_device *obd = m->private;
279 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
280 struct upcall_cache *hash = mdt->mdt_identity_cache;
282 down_read(&hash->uc_upcall_rwsem);
283 seq_printf(m, "%s\n", hash->uc_upcall);
284 up_read(&hash->uc_upcall_rwsem);
289 mdt_identity_upcall_seq_write(struct file *file, const char __user *buffer,
290 size_t count, loff_t *off)
292 struct seq_file *m = file->private_data;
293 struct obd_device *obd = m->private;
294 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
295 struct upcall_cache *hash = mdt->mdt_identity_cache;
299 if (count >= UC_CACHE_UPCALL_MAXPATH) {
300 CERROR("%s: identity upcall too long\n", mdt_obd_name(mdt));
303 OBD_ALLOC(kernbuf, count + 1);
305 GOTO(failed, rc = -ENOMEM);
306 if (copy_from_user(kernbuf, buffer, count))
307 GOTO(failed, rc = -EFAULT);
309 /* Remove any extraneous bits from the upcall (e.g. linefeeds) */
310 down_write(&hash->uc_upcall_rwsem);
311 sscanf(kernbuf, "%s", hash->uc_upcall);
312 up_write(&hash->uc_upcall_rwsem);
314 if (strcmp(hash->uc_name, mdt_obd_name(mdt)) != 0)
315 CWARN("%s: write to upcall name %s\n",
316 mdt_obd_name(mdt), hash->uc_upcall);
318 if (strcmp(hash->uc_upcall, "NONE") == 0 && mdt->mdt_opts.mo_acl)
319 CWARN("%s: disable \"identity_upcall\" with ACL enabled maybe "
320 "cause unexpected \"EACCESS\"\n", mdt_obd_name(mdt));
322 CDEBUG(D_CONFIG, "%s: identity upcall set to %s\n", mdt_obd_name(mdt),
324 OBD_FREE(kernbuf, count + 1);
329 OBD_FREE(kernbuf, count + 1);
332 LPROC_SEQ_FOPS(mdt_identity_upcall);
335 lprocfs_identity_flush_seq_write(struct file *file, const char __user *buffer,
336 size_t count, void *data)
338 struct seq_file *m = file->private_data;
339 struct obd_device *obd = m->private;
340 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
344 rc = kstrtoint_from_user(buffer, count, 0, &uid);
348 mdt_flush_identity(mdt->mdt_identity_cache, uid);
351 LPROC_SEQ_FOPS_WR_ONLY(mdt, identity_flush);
354 lprocfs_identity_info_seq_write(struct file *file, const char __user *buffer,
355 size_t count, void *data)
357 struct seq_file *m = file->private_data;
358 struct obd_device *obd = m->private;
359 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
360 struct identity_downcall_data *param;
361 int size = sizeof(*param), rc, checked = 0;
365 CERROR("%s: invalid data count = %lu, size = %d\n",
366 mdt_obd_name(mdt), (unsigned long) count, size);
370 OBD_ALLOC(param, size);
374 if (copy_from_user(param, buffer, size)) {
375 CERROR("%s: bad identity data\n", mdt_obd_name(mdt));
376 GOTO(out, rc = -EFAULT);
381 if (param->idd_magic != IDENTITY_DOWNCALL_MAGIC) {
382 CERROR("%s: MDS identity downcall bad params\n",
384 GOTO(out, rc = -EINVAL);
387 if (param->idd_nperms > N_PERMS_MAX) {
388 CERROR("%s: perm count %d more than maximum %d\n",
389 mdt_obd_name(mdt), param->idd_nperms,
391 GOTO(out, rc = -EINVAL);
394 if (param->idd_ngroups > NGROUPS_MAX) {
395 CERROR("%s: group count %d more than maximum %d\n",
396 mdt_obd_name(mdt), param->idd_ngroups,
398 GOTO(out, rc = -EINVAL);
401 if (param->idd_ngroups) {
402 rc = param->idd_ngroups; /* save idd_ngroups */
403 OBD_FREE(param, size);
404 size = offsetof(struct identity_downcall_data,
410 rc = upcall_cache_downcall(mdt->mdt_identity_cache, param->idd_err,
411 param->idd_uid, param);
415 OBD_FREE(param, size);
417 return rc ? rc : count;
419 LPROC_SEQ_FOPS_WR_ONLY(mdt, identity_info);
421 static int mdt_site_stats_seq_show(struct seq_file *m, void *data)
423 struct obd_device *obd = m->private;
424 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
426 return lu_site_stats_seq_print(mdt_lu_site(mdt), m);
428 LPROC_SEQ_FOPS_RO(mdt_site_stats);
430 #define BUFLEN (UUID_MAX + 4)
433 lprocfs_mds_evict_client_seq_write(struct file *file, const char __user *buf,
434 size_t count, loff_t *off)
436 struct seq_file *m = file->private_data;
437 struct obd_device *obd = m->private;
438 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
443 OBD_ALLOC(kbuf, BUFLEN);
448 * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
449 * bytes into kbuf, to ensure that the string is NUL-terminated.
450 * UUID_MAX should include a trailing NUL already.
452 if (copy_from_user(kbuf, buf, min_t(unsigned long, BUFLEN - 1, count)))
453 GOTO(out, rc = -EFAULT);
454 tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
456 if (strncmp(tmpbuf, "nid:", 4) != 0) {
457 count = lprocfs_evict_client_seq_write(file, buf, count, off);
461 if (mdt->mdt_opts.mo_evict_tgt_nids) {
462 rc = obd_set_info_async(NULL, mdt->mdt_child_exp,
463 sizeof(KEY_EVICT_BY_NID),
465 strlen(tmpbuf + 4) + 1,
468 CERROR("Failed to evict nid %s from OSTs: rc %d\n",
472 /* See the comments in function lprocfs_wr_evict_client()
473 * in ptlrpc/lproc_ptlrpc.c for details. - jay */
474 class_incref(obd, __func__, current);
475 obd_export_evict_by_nid(obd, tmpbuf + 4);
476 class_decref(obd, __func__, current);
480 OBD_FREE(kbuf, BUFLEN);
481 return rc < 0 ? rc : count;
486 static int mdt_evict_tgt_nids_seq_show(struct seq_file *m, void *data)
488 struct obd_device *obd = m->private;
489 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
491 seq_printf(m, "%u\n", mdt->mdt_opts.mo_evict_tgt_nids);
496 mdt_evict_tgt_nids_seq_write(struct file *file, const char __user *buffer,
497 size_t count, loff_t *off)
499 struct seq_file *m = file->private_data;
500 struct obd_device *obd = m->private;
501 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
505 rc = kstrtobool_from_user(buffer, count, &val);
509 mdt->mdt_opts.mo_evict_tgt_nids = val;
512 LPROC_SEQ_FOPS(mdt_evict_tgt_nids);
514 static int mdt_cos_seq_show(struct seq_file *m, void *data)
516 struct obd_device *obd = m->private;
517 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
519 seq_printf(m, "%u\n", mdt_cos_is_enabled(mdt));
524 mdt_cos_seq_write(struct file *file, const char __user *buffer,
525 size_t count, loff_t *off)
527 struct seq_file *m = file->private_data;
528 struct obd_device *obd = m->private;
529 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
533 rc = kstrtobool_from_user(buffer, count, &val);
537 mdt_enable_cos(mdt, val);
540 LPROC_SEQ_FOPS(mdt_cos);
542 static int mdt_root_squash_seq_show(struct seq_file *m, void *data)
544 struct obd_device *obd = m->private;
545 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
546 struct root_squash_info *squash = &mdt->mdt_squash;
548 seq_printf(m, "%u:%u\n", squash->rsi_uid,
554 mdt_root_squash_seq_write(struct file *file, const char __user *buffer,
555 size_t count, loff_t *off)
557 struct seq_file *m = file->private_data;
558 struct obd_device *obd = m->private;
559 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
560 struct root_squash_info *squash = &mdt->mdt_squash;
562 return lprocfs_wr_root_squash(buffer, count, squash,
565 LPROC_SEQ_FOPS(mdt_root_squash);
567 static int mdt_nosquash_nids_seq_show(struct seq_file *m, void *data)
569 struct obd_device *obd = m->private;
570 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
571 struct root_squash_info *squash = &mdt->mdt_squash;
574 down_read(&squash->rsi_sem);
575 if (!list_empty(&squash->rsi_nosquash_nids)) {
576 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
577 &squash->rsi_nosquash_nids);
581 seq_puts(m, "NONE\n");
582 up_read(&squash->rsi_sem);
588 mdt_nosquash_nids_seq_write(struct file *file, const char __user *buffer,
589 size_t count, loff_t *off)
591 struct seq_file *m = file->private_data;
592 struct obd_device *obd = m->private;
593 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
594 struct root_squash_info *squash = &mdt->mdt_squash;
596 return lprocfs_wr_nosquash_nids(buffer, count, squash,
599 LPROC_SEQ_FOPS(mdt_nosquash_nids);
601 static int mdt_enable_remote_dir_seq_show(struct seq_file *m, void *data)
603 struct obd_device *obd = m->private;
604 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
606 seq_printf(m, "%u\n", mdt->mdt_enable_remote_dir);
611 mdt_enable_remote_dir_seq_write(struct file *file, const char __user *buffer,
612 size_t count, loff_t *off)
614 struct seq_file *m = file->private_data;
615 struct obd_device *obd = m->private;
616 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
620 rc = kstrtobool_from_user(buffer, count, &val);
624 mdt->mdt_enable_remote_dir = val;
627 LPROC_SEQ_FOPS(mdt_enable_remote_dir);
629 static int mdt_enable_remote_dir_gid_seq_show(struct seq_file *m, void *data)
631 struct obd_device *obd = m->private;
632 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
634 seq_printf(m, "%d\n",
635 (int)mdt->mdt_enable_remote_dir_gid);
640 mdt_enable_remote_dir_gid_seq_write(struct file *file,
641 const char __user *buffer,
642 size_t count, loff_t *off)
644 struct seq_file *m = file->private_data;
645 struct obd_device *obd = m->private;
646 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
650 rc = kstrtoint_from_user(buffer, count, 0, &val);
654 mdt->mdt_enable_remote_dir_gid = val;
657 LPROC_SEQ_FOPS(mdt_enable_remote_dir_gid);
659 static int mdt_enable_striped_dir_seq_show(struct seq_file *m, void *data)
661 struct obd_device *obd = m->private;
662 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
664 seq_printf(m, "%u\n", mdt->mdt_enable_striped_dir);
669 mdt_enable_striped_dir_seq_write(struct file *file, const char __user *buffer,
670 size_t count, loff_t *off)
672 struct seq_file *m = file->private_data;
673 struct obd_device *obd = m->private;
674 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
678 rc = kstrtobool_from_user(buffer, count, &val);
682 mdt->mdt_enable_striped_dir = val;
685 LPROC_SEQ_FOPS(mdt_enable_striped_dir);
687 static int mdt_enable_dir_migration_seq_show(struct seq_file *m, void *data)
689 struct obd_device *obd = m->private;
690 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
692 seq_printf(m, "%u\n", mdt->mdt_enable_dir_migration);
697 mdt_enable_dir_migration_seq_write(struct file *file, const char __user *buffer,
698 size_t count, loff_t *off)
700 struct seq_file *m = file->private_data;
701 struct obd_device *obd = m->private;
702 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
706 rc = kstrtobool_from_user(buffer, count, &val);
710 mdt->mdt_enable_dir_migration = val;
713 LPROC_SEQ_FOPS(mdt_enable_dir_migration);
717 * Show MDT policy for handling dirty metadata under a lock being cancelled.
719 * \param[in] m seq_file handle
720 * \param[in] data unused for single entry
722 * \retval 0 on success
723 * \retval negative value on error
725 static int mdt_slc_seq_show(struct seq_file *m, void *data)
727 struct obd_device *obd = m->private;
728 struct lu_target *tgt = obd->u.obt.obt_lut;
729 char *slc_states[] = {"never", "blocking", "always" };
731 seq_printf(m, "%s\n", slc_states[tgt->lut_sync_lock_cancel]);
734 LPROC_SEQ_FOPS_RO(mdt_slc);
737 * Show MDT async commit count.
739 * \param[in] m seq_file handle
740 * \param[in] data unused for single entry
742 * \retval 0 on success
743 * \retval negative value on error
745 static int mdt_async_commit_count_seq_show(struct seq_file *m, void *data)
747 struct obd_device *obd = m->private;
748 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
750 seq_printf(m, "%d\n", atomic_read(&mdt->mdt_async_commit_count));
755 mdt_async_commit_count_seq_write(struct file *file, const char __user *buffer,
756 size_t count, loff_t *off)
758 struct seq_file *m = file->private_data;
759 struct obd_device *obd = m->private;
760 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
764 rc = kstrtoint_from_user(buffer, count, 0, &val);
768 atomic_set(&mdt->mdt_async_commit_count, val);
772 LPROC_SEQ_FOPS(mdt_async_commit_count);
775 * Show MDT sync count.
777 * \param[in] m seq_file handle
778 * \param[in] data unused for single entry
780 * \retval 0 on success
781 * \retval negative value on error
783 static int mdt_sync_count_seq_show(struct seq_file *m, void *data)
785 struct obd_device *obd = m->private;
786 struct lu_target *tgt = obd->u.obt.obt_lut;
788 seq_printf(m, "%d\n", atomic_read(&tgt->lut_sync_count));
793 mdt_sync_count_seq_write(struct file *file, const char __user *buffer,
794 size_t count, loff_t *off)
796 struct seq_file *m = file->private_data;
797 struct obd_device *obd = m->private;
798 struct lu_target *tgt = obd->u.obt.obt_lut;
802 rc = kstrtoint_from_user(buffer, count, 0, &val);
806 atomic_set(&tgt->lut_sync_count, val);
810 LPROC_SEQ_FOPS(mdt_sync_count);
812 static char *dom_open_lock_modes[NUM_DOM_LOCK_ON_OPEN_MODES] = {
813 [NO_DOM_LOCK_ON_OPEN] = "never",
814 [TRYLOCK_DOM_ON_OPEN] = "trylock",
815 [ALWAYS_DOM_LOCK_ON_OPEN] = "always",
818 /* This must be longer than the longest string above */
819 #define DOM_LOCK_MODES_MAXLEN 16
822 * Show MDT policy for data prefetch on open for DoM files..
824 * \param[in] m seq_file handle
825 * \param[in] data unused
827 * \retval 0 on success
828 * \retval negative value on error
830 static int mdt_dom_lock_seq_show(struct seq_file *m, void *data)
832 struct obd_device *obd = m->private;
833 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
835 seq_printf(m, "%s\n", dom_open_lock_modes[mdt->mdt_opts.mo_dom_lock]);
840 * Change MDT policy for data prefetch on open for DoM files.
842 * This variable defines how DOM lock is taken at open enqueue.
843 * There are three possible modes:
844 * 1) never - never take DoM lock on open. DoM lock will be taken as separate
845 * IO lock with own enqueue.
846 * 2) trylock - DoM lock will be taken only if non-blocked.
847 * 3) always - DoM lock will be taken always even if it is blocking lock.
849 * If dom_read_open is enabled too then DoM lock is taken in PR mode and
850 * is paired with LAYOUT lock when possible.
852 * \param[in] file proc file
853 * \param[in] buffer string which represents policy
854 * \param[in] count \a buffer length
855 * \param[in] off unused for single entry
857 * \retval \a count on success
858 * \retval negative number on error
861 mdt_dom_lock_seq_write(struct file *file, const char __user *buffer,
862 size_t count, loff_t *off)
864 struct seq_file *m = file->private_data;
865 struct obd_device *obd = m->private;
866 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
867 char kernbuf[DOM_LOCK_MODES_MAXLEN];
871 if (count == 0 || count >= sizeof(kernbuf))
874 if (copy_from_user(kernbuf, buffer, count))
878 if (kernbuf[count - 1] == '\n')
879 kernbuf[count - 1] = 0;
881 for (i = 0 ; i < NUM_DOM_LOCK_ON_OPEN_MODES; i++) {
882 if (strcmp(kernbuf, dom_open_lock_modes[i]) == 0) {
888 /* Legacy numeric codes */
890 rc = kstrtoint_from_user(buffer, count, 0, &val);
895 if (val < 0 || val >= NUM_DOM_LOCK_ON_OPEN_MODES)
898 mdt->mdt_opts.mo_dom_lock = val;
901 LPROC_SEQ_FOPS(mdt_dom_lock);
904 * Show MDT policy for data prefetch on open for DoM files..
906 * \param[in] m seq_file handle
907 * \param[in] data unused
909 * \retval 0 on success
910 * \retval negative value on error
912 static int mdt_dom_read_open_seq_show(struct seq_file *m, void *data)
914 struct obd_device *obd = m->private;
915 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
917 seq_printf(m, "%u\n", !!mdt->mdt_opts.mo_dom_read_open);
922 * Modify MDT policy for data prefetch on open for DoM files.
924 * If enabled then Data-on-MDT file data may be read during open and
925 * returned back in reply. It works only with mo_dom_lock enabled.
927 * \param[in] file proc file
928 * \param[in] buffer string which represents policy
929 * \param[in] count \a buffer length
930 * \param[in] off unused for single entry
932 * \retval \a count on success
933 * \retval negative number on error
936 mdt_dom_read_open_seq_write(struct file *file, const char __user *buffer,
937 size_t count, loff_t *off)
939 struct seq_file *m = file->private_data;
940 struct obd_device *obd = m->private;
941 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
945 rc = kstrtobool_from_user(buffer, count, &val);
949 mdt->mdt_opts.mo_dom_read_open = !!val;
952 LPROC_SEQ_FOPS(mdt_dom_read_open);
954 static int mdt_migrate_hsm_allowed_seq_show(struct seq_file *m, void *data)
956 struct obd_device *obd = m->private;
957 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
959 seq_printf(m, "%u\n", (mdt->mdt_opts.mo_migrate_hsm_allowed != 0));
964 mdt_migrate_hsm_allowed_seq_write(struct file *file, const char __user *buffer,
965 size_t count, loff_t *off)
967 struct seq_file *m = file->private_data;
968 struct obd_device *obd = m->private;
969 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
973 rc = kstrtobool_from_user(buffer, count, &val);
977 mdt->mdt_opts.mo_migrate_hsm_allowed = val;
980 LPROC_SEQ_FOPS(mdt_migrate_hsm_allowed);
982 LPROC_SEQ_FOPS_RO_TYPE(mdt, recovery_status);
983 LPROC_SEQ_FOPS_RO_TYPE(mdt, num_exports);
984 LPROC_SEQ_FOPS_RO_TYPE(mdt, target_instance);
985 LPROC_SEQ_FOPS_RO_TYPE(mdt, hash);
986 LPROC_SEQ_FOPS_WR_ONLY(mdt, mds_evict_client);
987 LPROC_SEQ_FOPS_RW_TYPE(mdt, job_interval);
988 LPROC_SEQ_FOPS_RW_TYPE(mdt, ir_factor);
989 LPROC_SEQ_FOPS_RW_TYPE(mdt, nid_stats_clear);
990 LPROC_SEQ_FOPS(mdt_hsm_cdt_control);
992 LPROC_SEQ_FOPS_RW_TYPE(mdt, recovery_time_hard);
993 LPROC_SEQ_FOPS_RW_TYPE(mdt, recovery_time_soft);
995 LPROC_SEQ_FOPS_RO(tgt_tot_dirty);
996 LPROC_SEQ_FOPS_RO(tgt_tot_granted);
997 LPROC_SEQ_FOPS_RO(tgt_tot_pending);
998 LPROC_SEQ_FOPS(tgt_grant_compat_disable);
1000 static struct lprocfs_vars lprocfs_mdt_obd_vars[] = {
1001 { .name = "tot_dirty",
1002 .fops = &tgt_tot_dirty_fops },
1003 { .name = "tot_pending",
1004 .fops = &tgt_tot_pending_fops },
1005 { .name = "tot_granted",
1006 .fops = &tgt_tot_granted_fops },
1007 { .name = "grant_compat_disable",
1008 .fops = &tgt_grant_compat_disable_fops },
1009 { .name = "recovery_status",
1010 .fops = &mdt_recovery_status_fops },
1011 { .name = "num_exports",
1012 .fops = &mdt_num_exports_fops },
1013 { .name = "identity_expire",
1014 .fops = &mdt_identity_expire_fops },
1015 { .name = "identity_acquire_expire",
1016 .fops = &mdt_identity_acquire_expire_fops },
1017 { .name = "identity_upcall",
1018 .fops = &mdt_identity_upcall_fops },
1019 { .name = "identity_flush",
1020 .fops = &mdt_identity_flush_fops },
1021 { .name = "identity_info",
1022 .fops = &mdt_identity_info_fops },
1023 { .name = "site_stats",
1024 .fops = &mdt_site_stats_fops },
1025 { .name = "evict_client",
1026 .fops = &mdt_mds_evict_client_fops },
1027 { .name = "evict_tgt_nids",
1028 .fops = &mdt_evict_tgt_nids_fops },
1029 { .name = "hash_stats",
1030 .fops = &mdt_hash_fops },
1031 { .name = "commit_on_sharing",
1032 .fops = &mdt_cos_fops },
1033 { .name = "root_squash",
1034 .fops = &mdt_root_squash_fops },
1035 { .name = "nosquash_nids",
1036 .fops = &mdt_nosquash_nids_fops },
1037 { .name = "instance",
1038 .fops = &mdt_target_instance_fops },
1039 { .name = "ir_factor",
1040 .fops = &mdt_ir_factor_fops },
1041 { .name = "job_cleanup_interval",
1042 .fops = &mdt_job_interval_fops },
1043 { .name = "enable_remote_dir",
1044 .fops = &mdt_enable_remote_dir_fops },
1045 { .name = "enable_remote_dir_gid",
1046 .fops = &mdt_enable_remote_dir_gid_fops },
1047 { .name = "enable_striped_dir",
1048 .fops = &mdt_enable_striped_dir_fops },
1049 { .name = "enable_dir_migration",
1050 .fops = &mdt_enable_dir_migration_fops },
1051 { .name = "hsm_control",
1052 .fops = &mdt_hsm_cdt_control_fops },
1053 { .name = "recovery_time_hard",
1054 .fops = &mdt_recovery_time_hard_fops },
1055 { .name = "recovery_time_soft",
1056 .fops = &mdt_recovery_time_soft_fops },
1057 { .name = "sync_lock_cancel",
1058 .fops = &mdt_slc_fops },
1059 { .name = "async_commit_count",
1060 .fops = &mdt_async_commit_count_fops },
1061 { .name = "sync_count",
1062 .fops = &mdt_sync_count_fops },
1063 { .name = "dom_lock",
1064 .fops = &mdt_dom_lock_fops },
1065 { .name = "dom_read_open",
1066 .fops = &mdt_dom_read_open_fops },
1067 { .name = "migrate_hsm_allowed",
1068 .fops = &mdt_migrate_hsm_allowed_fops },
1073 lprocfs_mdt_print_open_files(struct cfs_hash *hs, struct cfs_hash_bd *bd,
1074 struct hlist_node *hnode, void *v)
1076 struct obd_export *exp = cfs_hash_object(hs, hnode);
1077 struct seq_file *seq = v;
1079 if (exp->exp_lock_hash != NULL) {
1080 struct mdt_export_data *med = &exp->exp_mdt_data;
1081 struct mdt_file_data *mfd;
1083 spin_lock(&med->med_open_lock);
1084 list_for_each_entry(mfd, &med->med_open_head, mfd_list) {
1085 seq_printf(seq, DFID"\n",
1086 PFID(mdt_object_fid(mfd->mfd_object)));
1088 spin_unlock(&med->med_open_lock);
1094 static int lprocfs_mdt_open_files_seq_show(struct seq_file *seq, void *v)
1096 struct nid_stat *stats = seq->private;
1097 struct obd_device *obd = stats->nid_obd;
1099 cfs_hash_for_each_key(obd->obd_nid_hash, &stats->nid,
1100 lprocfs_mdt_print_open_files, seq);
1105 int lprocfs_mdt_open_files_seq_open(struct inode *inode, struct file *file)
1107 struct seq_file *seq;
1110 rc = single_open(file, &lprocfs_mdt_open_files_seq_show, NULL);
1114 seq = file->private_data;
1115 seq->private = PDE_DATA(inode);
1120 void mdt_counter_incr(struct ptlrpc_request *req, int opcode)
1122 struct obd_export *exp = req->rq_export;
1124 if (exp->exp_obd && exp->exp_obd->obd_md_stats)
1125 lprocfs_counter_incr(exp->exp_obd->obd_md_stats,
1126 opcode + LPROC_MD_LAST_OPC);
1127 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
1128 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
1129 if (exp->exp_obd && exp->exp_obd->u.obt.obt_jobstats.ojs_hash &&
1130 (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS))
1131 lprocfs_job_stats_log(exp->exp_obd,
1132 lustre_msg_get_jobid(req->rq_reqmsg),
1136 static const char * const mdt_stats[] = {
1137 [LPROC_MDT_OPEN] = "open",
1138 [LPROC_MDT_CLOSE] = "close",
1139 [LPROC_MDT_MKNOD] = "mknod",
1140 [LPROC_MDT_LINK] = "link",
1141 [LPROC_MDT_UNLINK] = "unlink",
1142 [LPROC_MDT_MKDIR] = "mkdir",
1143 [LPROC_MDT_RMDIR] = "rmdir",
1144 [LPROC_MDT_RENAME] = "rename",
1145 [LPROC_MDT_GETATTR] = "getattr",
1146 [LPROC_MDT_SETATTR] = "setattr",
1147 [LPROC_MDT_GETXATTR] = "getxattr",
1148 [LPROC_MDT_SETXATTR] = "setxattr",
1149 [LPROC_MDT_STATFS] = "statfs",
1150 [LPROC_MDT_SYNC] = "sync",
1151 [LPROC_MDT_SAMEDIR_RENAME] = "samedir_rename",
1152 [LPROC_MDT_CROSSDIR_RENAME] = "crossdir_rename",
1153 [LPROC_MDT_IO_READ] = "read_bytes",
1154 [LPROC_MDT_IO_WRITE] = "write_bytes",
1155 [LPROC_MDT_IO_PUNCH] = "punch",
1158 void mdt_stats_counter_init(struct lprocfs_stats *stats)
1162 LASSERT(stats && stats->ls_num >= ARRAY_SIZE(mdt_stats));
1164 for (idx = 0; idx < ARRAY_SIZE(mdt_stats); idx++) {
1167 if (idx == LPROC_MDT_IO_WRITE || idx == LPROC_MDT_IO_READ)
1168 flags = LPROCFS_CNTR_AVGMINMAX;
1170 lprocfs_counter_init(stats, idx, flags, mdt_stats[idx], "reqs");
1174 int mdt_procfs_init(struct mdt_device *mdt, const char *name)
1176 struct obd_device *obd = mdt2obd_dev(mdt);
1181 LASSERT(name != NULL);
1183 obd->obd_vars = lprocfs_mdt_obd_vars;
1184 rc = lprocfs_obd_setup(obd, true);
1186 CERROR("%s: cannot create proc entries: rc = %d\n",
1187 mdt_obd_name(mdt), rc);
1191 rc = hsm_cdt_procfs_init(mdt);
1193 CERROR("%s: cannot create hsm proc entries: rc = %d\n",
1194 mdt_obd_name(mdt), rc);
1198 obd->obd_proc_exports_entry = proc_mkdir("exports",
1199 obd->obd_proc_entry);
1200 if (obd->obd_proc_exports_entry)
1201 lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
1202 obd, &mdt_nid_stats_clear_fops);
1204 rc = lprocfs_alloc_md_stats(obd, ARRAY_SIZE(mdt_stats));
1208 /* add additional MDT md_stats after the default ones */
1209 for (i = 0; i < ARRAY_SIZE(mdt_stats); i++) {
1210 int idx = i + LPROC_MD_LAST_OPC;
1213 if (idx == LPROC_MDT_IO_WRITE || idx == LPROC_MDT_IO_READ)
1214 flags = LPROCFS_CNTR_AVGMINMAX;
1216 lprocfs_counter_init(obd->obd_md_stats, idx, flags,
1217 mdt_stats[i], "reqs");
1220 rc = lprocfs_job_stats_init(obd, ARRAY_SIZE(mdt_stats),
1221 mdt_stats_counter_init);
1223 rc = lproc_mdt_attach_rename_seqstat(mdt);
1225 CERROR("%s: MDT can not create rename stats rc = %d\n",
1226 mdt_obd_name(mdt), rc);
1231 void mdt_procfs_fini(struct mdt_device *mdt)
1233 struct obd_device *obd = mdt2obd_dev(mdt);
1235 if (obd->obd_proc_exports_entry != NULL) {
1236 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
1237 obd->obd_proc_exports_entry = NULL;
1240 lprocfs_free_per_client_stats(obd);
1241 hsm_cdt_procfs_fini(mdt);
1242 lprocfs_obd_cleanup(obd);
1243 lprocfs_free_md_stats(obd);
1244 lprocfs_free_obd_stats(obd);
1245 lprocfs_job_stats_fini(obd);