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 static void display_rename_stats(struct seq_file *seq, char *name,
80 struct obd_histogram *hist)
82 unsigned long tot, t, cum = 0;
85 tot = lprocfs_oh_sum(hist);
87 seq_printf(seq, "- %-15s\n", name);
88 /* dir size start from 4K, start i from 10(2^10) here */
89 for (i = 0; i < OBD_HIST_MAX; i++) {
90 t = hist->oh_buckets[i];
96 seq_printf(seq, "%6s%d%s", " ", 1<< i, "bytes:");
98 seq_printf(seq, "%6s%d%s", " ", 1<<(i-10), "KB:");
100 seq_printf(seq, "%6s%d%s", " ", 1<<(i-20), "MB:");
102 seq_printf(seq, " { sample: %3lu, pct: %3u, cum_pct: %3u }\n",
103 t, pct(t, tot), pct(cum, tot));
110 static void rename_stats_show(struct seq_file *seq,
111 struct rename_stats *rename_stats)
113 struct timespec64 now;
115 /* this sampling races with updates */
116 ktime_get_real_ts64(&now);
117 seq_printf(seq, "rename_stats:\n");
118 seq_printf(seq, "- %-15s %llu.%9lu\n", "snapshot_time:",
119 (s64)now.tv_sec, now.tv_nsec);
121 display_rename_stats(seq, "same_dir",
122 &rename_stats->hist[RENAME_SAMEDIR_SIZE]);
123 display_rename_stats(seq, "crossdir_src",
124 &rename_stats->hist[RENAME_CROSSDIR_SRC_SIZE]);
125 display_rename_stats(seq, "crossdir_tgt",
126 &rename_stats->hist[RENAME_CROSSDIR_TGT_SIZE]);
129 static int mdt_rename_stats_seq_show(struct seq_file *seq, void *v)
131 struct mdt_device *mdt = seq->private;
133 rename_stats_show(seq, &mdt->mdt_rename_stats);
139 mdt_rename_stats_seq_write(struct file *file, const char __user *buf,
140 size_t len, loff_t *off)
142 struct seq_file *seq = file->private_data;
143 struct mdt_device *mdt = seq->private;
146 for (i = 0; i < RENAME_LAST; i++)
147 lprocfs_oh_clear(&mdt->mdt_rename_stats.hist[i]);
151 LPROC_SEQ_FOPS(mdt_rename_stats);
153 static int lproc_mdt_attach_rename_seqstat(struct mdt_device *mdt)
157 for (i = 0; i < RENAME_LAST; i++)
158 spin_lock_init(&mdt->mdt_rename_stats.hist[i].oh_lock);
160 return lprocfs_obd_seq_create(mdt2obd_dev(mdt), "rename_stats", 0644,
161 &mdt_rename_stats_fops, mdt);
164 void mdt_rename_counter_tally(struct mdt_thread_info *info,
165 struct mdt_device *mdt,
166 struct ptlrpc_request *req,
167 struct mdt_object *src,
168 struct mdt_object *tgt)
170 struct md_attr *ma = &info->mti_attr;
171 struct rename_stats *rstats = &mdt->mdt_rename_stats;
174 ma->ma_need = MA_INODE;
176 rc = mo_attr_get(info->mti_env, mdt_object_child(src), ma);
178 CERROR("%s: "DFID" attr_get, rc = %d\n",
179 mdt_obd_name(mdt), PFID(mdt_object_fid(src)), rc);
184 mdt_counter_incr(req, LPROC_MDT_SAMEDIR_RENAME);
185 lprocfs_oh_tally_log2(&rstats->hist[RENAME_SAMEDIR_SIZE],
186 (unsigned int)ma->ma_attr.la_size);
190 mdt_counter_incr(req, LPROC_MDT_CROSSDIR_RENAME);
191 lprocfs_oh_tally_log2(&rstats->hist[RENAME_CROSSDIR_SRC_SIZE],
192 (unsigned int)ma->ma_attr.la_size);
194 ma->ma_need = MA_INODE;
196 rc = mo_attr_get(info->mti_env, mdt_object_child(tgt), ma);
198 CERROR("%s: "DFID" attr_get, rc = %d\n",
199 mdt_obd_name(mdt), PFID(mdt_object_fid(tgt)), rc);
203 lprocfs_oh_tally_log2(&rstats->hist[RENAME_CROSSDIR_TGT_SIZE],
204 (unsigned int)ma->ma_attr.la_size);
207 static ssize_t identity_expire_show(struct kobject *kobj,
208 struct attribute *attr, char *buf)
210 struct obd_device *obd = container_of(kobj, struct obd_device,
212 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
214 return scnprintf(buf, PAGE_SIZE, "%lld\n",
215 mdt->mdt_identity_cache->uc_entry_expire);
218 static ssize_t identity_expire_store(struct kobject *kobj,
219 struct attribute *attr,
220 const char *buffer, size_t count)
222 struct obd_device *obd = container_of(kobj, struct obd_device,
224 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
228 rc = kstrtoll(buffer, 10, &val);
235 mdt->mdt_identity_cache->uc_entry_expire = val;
239 LUSTRE_RW_ATTR(identity_expire);
241 static ssize_t identity_acquire_expire_show(struct kobject *kobj,
242 struct attribute *attr, char *buf)
244 struct obd_device *obd = container_of(kobj, struct obd_device,
246 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
248 return scnprintf(buf, PAGE_SIZE, "%lld\n",
249 mdt->mdt_identity_cache->uc_acquire_expire);
252 static ssize_t identity_acquire_expire_store(struct kobject *kobj,
253 struct attribute *attr,
254 const char *buffer, size_t count)
256 struct obd_device *obd = container_of(kobj, struct obd_device,
258 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
262 rc = kstrtoll(buffer, 0, &val);
266 if (val < 0 || val > INT_MAX)
269 mdt->mdt_identity_cache->uc_acquire_expire = val;
273 LUSTRE_RW_ATTR(identity_acquire_expire);
275 static ssize_t identity_upcall_show(struct kobject *kobj,
276 struct attribute *attr, char *buf)
278 struct obd_device *obd = container_of(kobj, struct obd_device,
280 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
281 struct upcall_cache *hash = mdt->mdt_identity_cache;
284 down_read(&hash->uc_upcall_rwsem);
285 rc = scnprintf(buf, PAGE_SIZE, "%s\n", hash->uc_upcall);
286 up_read(&hash->uc_upcall_rwsem);
290 static ssize_t identity_upcall_store(struct kobject *kobj,
291 struct attribute *attr,
292 const char *buffer, size_t count)
294 struct obd_device *obd = container_of(kobj, struct obd_device,
296 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
297 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));
304 /* Remove any extraneous bits from the upcall (e.g. linefeeds) */
305 down_write(&hash->uc_upcall_rwsem);
306 sscanf(buffer, "%s", hash->uc_upcall);
307 up_write(&hash->uc_upcall_rwsem);
309 if (strcmp(hash->uc_name, mdt_obd_name(mdt)) != 0)
310 CWARN("%s: write to upcall name %s\n",
311 mdt_obd_name(mdt), hash->uc_upcall);
313 if (strcmp(hash->uc_upcall, "NONE") == 0 && mdt->mdt_opts.mo_acl)
314 CWARN("%s: disable \"identity_upcall\" with ACL enabled maybe "
315 "cause unexpected \"EACCESS\"\n", mdt_obd_name(mdt));
317 CDEBUG(D_CONFIG, "%s: identity upcall set to %s\n", mdt_obd_name(mdt),
321 LUSTRE_RW_ATTR(identity_upcall);
323 static ssize_t identity_flush_store(struct kobject *kobj,
324 struct attribute *attr,
325 const char *buffer, size_t count)
327 struct obd_device *obd = container_of(kobj, struct obd_device,
329 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
333 rc = kstrtoint(buffer, 0, &uid);
337 mdt_flush_identity(mdt->mdt_identity_cache, uid);
340 LUSTRE_WO_ATTR(identity_flush);
343 lprocfs_identity_info_seq_write(struct file *file, const char __user *buffer,
344 size_t count, void *data)
346 struct seq_file *m = file->private_data;
347 struct obd_device *obd = m->private;
348 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
349 struct identity_downcall_data *param;
350 int size = sizeof(*param), rc, checked = 0;
354 CERROR("%s: invalid data count = %lu, size = %d\n",
355 mdt_obd_name(mdt), (unsigned long) count, size);
359 OBD_ALLOC(param, size);
363 if (copy_from_user(param, buffer, size)) {
364 CERROR("%s: bad identity data\n", mdt_obd_name(mdt));
365 GOTO(out, rc = -EFAULT);
370 if (param->idd_magic != IDENTITY_DOWNCALL_MAGIC) {
371 CERROR("%s: MDS identity downcall bad params\n",
373 GOTO(out, rc = -EINVAL);
376 if (param->idd_nperms > N_PERMS_MAX) {
377 CERROR("%s: perm count %d more than maximum %d\n",
378 mdt_obd_name(mdt), param->idd_nperms,
380 GOTO(out, rc = -EINVAL);
383 if (param->idd_ngroups > NGROUPS_MAX) {
384 CERROR("%s: group count %d more than maximum %d\n",
385 mdt_obd_name(mdt), param->idd_ngroups,
387 GOTO(out, rc = -EINVAL);
390 if (param->idd_ngroups) {
391 rc = param->idd_ngroups; /* save idd_ngroups */
392 OBD_FREE(param, size);
393 size = offsetof(struct identity_downcall_data,
399 rc = upcall_cache_downcall(mdt->mdt_identity_cache, param->idd_err,
400 param->idd_uid, param);
404 OBD_FREE(param, size);
406 return rc ? rc : count;
408 LPROC_SEQ_FOPS_WR_ONLY(mdt, identity_info);
410 static int mdt_site_stats_seq_show(struct seq_file *m, void *data)
412 struct obd_device *obd = m->private;
413 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
415 return lu_site_stats_seq_print(mdt_lu_site(mdt), m);
417 LPROC_SEQ_FOPS_RO(mdt_site_stats);
419 #define BUFLEN (UUID_MAX + 4)
422 lprocfs_mds_evict_client_seq_write(struct file *file, const char __user *buf,
423 size_t count, loff_t *off)
425 struct seq_file *m = file->private_data;
426 struct obd_device *obd = m->private;
427 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
432 OBD_ALLOC(kbuf, BUFLEN);
437 * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
438 * bytes into kbuf, to ensure that the string is NUL-terminated.
439 * UUID_MAX should include a trailing NUL already.
441 if (copy_from_user(kbuf, buf, min_t(unsigned long, BUFLEN - 1, count)))
442 GOTO(out, rc = -EFAULT);
443 tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
445 if (strncmp(tmpbuf, "nid:", 4) != 0) {
446 count = lprocfs_evict_client_seq_write(file, buf, count, off);
450 if (mdt->mdt_opts.mo_evict_tgt_nids) {
451 rc = obd_set_info_async(NULL, mdt->mdt_child_exp,
452 sizeof(KEY_EVICT_BY_NID),
454 strlen(tmpbuf + 4) + 1,
457 CERROR("Failed to evict nid %s from OSTs: rc %d\n",
461 /* See the comments in function lprocfs_wr_evict_client()
462 * in ptlrpc/lproc_ptlrpc.c for details. - jay */
463 class_incref(obd, __func__, current);
464 obd_export_evict_by_nid(obd, tmpbuf + 4);
465 class_decref(obd, __func__, current);
469 OBD_FREE(kbuf, BUFLEN);
470 return rc < 0 ? rc : count;
475 static ssize_t evict_tgt_nids_show(struct kobject *kobj,
476 struct attribute *attr, char *buf)
478 struct obd_device *obd = container_of(kobj, struct obd_device,
480 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
482 return scnprintf(buf, PAGE_SIZE, "%u\n",
483 mdt->mdt_opts.mo_evict_tgt_nids);
486 static ssize_t evict_tgt_nids_store(struct kobject *kobj,
487 struct attribute *attr, const char *buffer,
490 struct obd_device *obd = container_of(kobj, struct obd_device,
492 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
496 rc = kstrtobool(buffer, &val);
500 mdt->mdt_opts.mo_evict_tgt_nids = val;
503 LUSTRE_RW_ATTR(evict_tgt_nids);
505 static ssize_t commit_on_sharing_show(struct kobject *kobj,
506 struct attribute *attr, char *buf)
508 struct obd_device *obd = container_of(kobj, struct obd_device,
510 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
512 return scnprintf(buf, PAGE_SIZE, "%u\n", mdt_cos_is_enabled(mdt));
515 static ssize_t commit_on_sharing_store(struct kobject *kobj,
516 struct attribute *attr,
517 const char *buffer, size_t count)
519 struct obd_device *obd = container_of(kobj, struct obd_device,
521 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
525 rc = kstrtobool(buffer, &val);
529 mdt_enable_cos(mdt, val);
532 LUSTRE_RW_ATTR(commit_on_sharing);
534 static int mdt_root_squash_seq_show(struct seq_file *m, void *data)
536 struct obd_device *obd = m->private;
537 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
538 struct root_squash_info *squash = &mdt->mdt_squash;
540 seq_printf(m, "%u:%u\n", squash->rsi_uid,
546 mdt_root_squash_seq_write(struct file *file, const char __user *buffer,
547 size_t count, loff_t *off)
549 struct seq_file *m = file->private_data;
550 struct obd_device *obd = m->private;
551 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
552 struct root_squash_info *squash = &mdt->mdt_squash;
554 return lprocfs_wr_root_squash(buffer, count, squash,
557 LPROC_SEQ_FOPS(mdt_root_squash);
559 static int mdt_nosquash_nids_seq_show(struct seq_file *m, void *data)
561 struct obd_device *obd = m->private;
562 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
563 struct root_squash_info *squash = &mdt->mdt_squash;
566 spin_lock(&squash->rsi_lock);
567 if (!list_empty(&squash->rsi_nosquash_nids)) {
568 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
569 &squash->rsi_nosquash_nids);
573 seq_puts(m, "NONE\n");
574 spin_unlock(&squash->rsi_lock);
580 mdt_nosquash_nids_seq_write(struct file *file, const char __user *buffer,
581 size_t count, loff_t *off)
583 struct seq_file *m = file->private_data;
584 struct obd_device *obd = m->private;
585 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
586 struct root_squash_info *squash = &mdt->mdt_squash;
588 return lprocfs_wr_nosquash_nids(buffer, count, squash,
591 LPROC_SEQ_FOPS(mdt_nosquash_nids);
593 static ssize_t enable_remote_dir_show(struct kobject *kobj,
594 struct attribute *attr, char *buf)
596 struct obd_device *obd = container_of(kobj, struct obd_device,
598 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
600 return scnprintf(buf, PAGE_SIZE, "%u\n", mdt->mdt_enable_remote_dir);
603 static ssize_t enable_remote_dir_store(struct kobject *kobj,
604 struct attribute *attr,
605 const char *buffer, size_t count)
607 struct obd_device *obd = container_of(kobj, struct obd_device,
609 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
613 rc = kstrtobool(buffer, &val);
617 mdt->mdt_enable_remote_dir = val;
620 LUSTRE_RW_ATTR(enable_remote_dir);
622 static ssize_t enable_remote_dir_gid_show(struct kobject *kobj,
623 struct attribute *attr, char *buf)
625 struct obd_device *obd = container_of(kobj, struct obd_device,
627 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
629 return scnprintf(buf, PAGE_SIZE, "%d\n",
630 (int)mdt->mdt_enable_remote_dir_gid);
633 static ssize_t enable_remote_dir_gid_store(struct kobject *kobj,
634 struct attribute *attr,
635 const char *buffer, size_t count)
637 struct obd_device *obd = container_of(kobj, struct obd_device,
639 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
643 rc = kstrtoint(buffer, 0, &val);
647 mdt->mdt_enable_remote_dir_gid = val;
650 LUSTRE_RW_ATTR(enable_remote_dir_gid);
652 static ssize_t enable_chprojid_gid_show(struct kobject *kobj,
653 struct attribute *attr, char *buf)
655 struct obd_device *obd = container_of(kobj, struct obd_device,
657 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
659 return scnprintf(buf, PAGE_SIZE, "%d\n",
660 (int)mdt->mdt_enable_chprojid_gid);
663 static ssize_t enable_chprojid_gid_store(struct kobject *kobj,
664 struct attribute *attr,
665 const char *buffer, size_t count)
667 struct obd_device *obd = container_of(kobj, struct obd_device,
669 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
673 rc = kstrtoint(buffer, 0, &val);
677 mdt->mdt_enable_chprojid_gid = val;
680 LUSTRE_RW_ATTR(enable_chprojid_gid);
682 static ssize_t enable_striped_dir_show(struct kobject *kobj,
683 struct attribute *attr, char *buf)
685 struct obd_device *obd = container_of(kobj, struct obd_device,
687 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
689 return scnprintf(buf, PAGE_SIZE, "%u\n", mdt->mdt_enable_striped_dir);
692 static ssize_t enable_striped_dir_store(struct kobject *kobj,
693 struct attribute *attr,
694 const char *buffer, size_t count)
696 struct obd_device *obd = container_of(kobj, struct obd_device,
698 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
702 rc = kstrtobool(buffer, &val);
706 mdt->mdt_enable_striped_dir = val;
709 LUSTRE_RW_ATTR(enable_striped_dir);
711 static ssize_t enable_dir_migration_show(struct kobject *kobj,
712 struct attribute *attr, char *buf)
714 struct obd_device *obd = container_of(kobj, struct obd_device,
716 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
718 return scnprintf(buf, PAGE_SIZE, "%u\n", mdt->mdt_enable_dir_migration);
721 static ssize_t enable_dir_migration_store(struct kobject *kobj,
722 struct attribute *attr,
723 const char *buffer, size_t count)
725 struct obd_device *obd = container_of(kobj, struct obd_device,
727 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
731 rc = kstrtobool(buffer, &val);
735 mdt->mdt_enable_dir_migration = val;
738 LUSTRE_RW_ATTR(enable_dir_migration);
741 * Show MDT async commit count.
744 * @data unused for single entry
746 * Return: 0 on success
747 * negative value on error
749 static ssize_t async_commit_count_show(struct kobject *kobj,
750 struct attribute *attr, char *buf)
752 struct obd_device *obd = container_of(kobj, struct obd_device,
754 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
756 return scnprintf(buf, PAGE_SIZE, "%d\n",
757 atomic_read(&mdt->mdt_async_commit_count));
760 static ssize_t async_commit_count_store(struct kobject *kobj,
761 struct attribute *attr,
762 const char *buffer, size_t count)
764 struct obd_device *obd = container_of(kobj, struct obd_device,
766 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
770 rc = kstrtoint(buffer, 10, &val);
774 atomic_set(&mdt->mdt_async_commit_count, val);
778 LUSTRE_RW_ATTR(async_commit_count);
781 * Show MDT sync count.
783 * \param[in] m seq_file handle
784 * \param[in] data unused for single entry
786 * \retval 0 on success
787 * \retval negative value on error
789 static ssize_t sync_count_show(struct kobject *kobj, struct attribute *attr,
792 struct obd_device *obd = container_of(kobj, struct obd_device,
794 struct lu_target *tgt = obd->u.obt.obt_lut;
796 return scnprintf(buf, PAGE_SIZE, "%d\n",
797 atomic_read(&tgt->lut_sync_count));
800 static ssize_t sync_count_store(struct kobject *kobj, struct attribute *attr,
801 const char *buffer, size_t count)
803 struct obd_device *obd = container_of(kobj, struct obd_device,
805 struct lu_target *tgt = obd->u.obt.obt_lut;
809 rc = kstrtoint(buffer, 0, &val);
813 atomic_set(&tgt->lut_sync_count, val);
817 LUSTRE_RW_ATTR(sync_count);
819 static char *dom_open_lock_modes[NUM_DOM_LOCK_ON_OPEN_MODES] = {
820 [NO_DOM_LOCK_ON_OPEN] = "never",
821 [TRYLOCK_DOM_ON_OPEN] = "trylock",
822 [ALWAYS_DOM_LOCK_ON_OPEN] = "always",
825 /* This must be longer than the longest string above */
826 #define DOM_LOCK_MODES_MAXLEN 16
829 * Show MDT policy for data prefetch on open for DoM files..
831 * \param[in] m seq_file handle
832 * \param[in] data unused
834 * \retval 0 on success
835 * \retval negative value on error
837 static ssize_t dom_lock_show(struct kobject *kobj, struct attribute *attr,
840 struct obd_device *obd = container_of(kobj, struct obd_device,
842 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
844 return scnprintf(buf, PAGE_SIZE, "%s\n",
845 dom_open_lock_modes[mdt->mdt_opts.mo_dom_lock]);
849 * Change MDT policy for data prefetch on open for DoM files.
851 * This variable defines how DOM lock is taken at open enqueue.
852 * There are three possible modes:
853 * 1) never - never take DoM lock on open. DoM lock will be taken as separate
854 * IO lock with own enqueue.
855 * 2) trylock - DoM lock will be taken only if non-blocked.
856 * 3) always - DoM lock will be taken always even if it is blocking lock.
858 * If dom_read_open is enabled too then DoM lock is taken in PR mode and
859 * is paired with LAYOUT lock when possible.
861 * \param[in] file proc file
862 * \param[in] buffer string which represents policy
863 * \param[in] count \a buffer length
864 * \param[in] off unused for single entry
866 * \retval \a count on success
867 * \retval negative number on error
869 static ssize_t dom_lock_store(struct kobject *kobj, struct attribute *attr,
870 const char *buffer, size_t count)
872 struct obd_device *obd = container_of(kobj, struct obd_device,
874 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
878 if (count == 0 || count >= DOM_LOCK_MODES_MAXLEN)
881 for (i = 0 ; i < NUM_DOM_LOCK_ON_OPEN_MODES; i++) {
882 /* buffer might have '\n' but using strlen() avoids it */
883 if (strncmp(buffer, dom_open_lock_modes[i],
884 strlen(dom_open_lock_modes[i])) == 0) {
890 /* Legacy numeric codes */
892 rc = kstrtoint(buffer, 0, &val);
897 if (val < 0 || val >= NUM_DOM_LOCK_ON_OPEN_MODES)
900 mdt->mdt_opts.mo_dom_lock = val;
903 LUSTRE_RW_ATTR(dom_lock);
906 * Show MDT policy for data prefetch on open for DoM files..
908 * \param[in] m seq_file handle
909 * \param[in] data unused
911 * \retval 0 on success
912 * \retval negative value on error
914 static ssize_t dom_read_open_show(struct kobject *kobj,
915 struct attribute *attr, char *buf)
917 struct obd_device *obd = container_of(kobj, struct obd_device,
919 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
921 return scnprintf(buf, PAGE_SIZE, "%u\n",
922 !!mdt->mdt_opts.mo_dom_read_open);
926 * Modify MDT policy for data prefetch on open for DoM files.
928 * If enabled then Data-on-MDT file data may be read during open and
929 * returned back in reply. It works only with mo_dom_lock enabled.
931 * \param[in] file proc file
932 * \param[in] buffer string which represents policy
933 * \param[in] count \a buffer length
934 * \param[in] off unused for single entry
936 * \retval \a count on success
937 * \retval negative number on error
939 static ssize_t dom_read_open_store(struct kobject *kobj,
940 struct attribute *attr, const char *buffer,
943 struct obd_device *obd = container_of(kobj, struct obd_device,
945 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
949 rc = kstrtobool(buffer, &val);
953 mdt->mdt_opts.mo_dom_read_open = !!val;
956 LUSTRE_RW_ATTR(dom_read_open);
958 static ssize_t migrate_hsm_allowed_show(struct kobject *kobj,
959 struct attribute *attr, char *buf)
961 struct obd_device *obd = container_of(kobj, struct obd_device,
963 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
965 return scnprintf(buf, PAGE_SIZE, "%u\n",
966 mdt->mdt_opts.mo_migrate_hsm_allowed);
969 static ssize_t migrate_hsm_allowed_store(struct kobject *kobj,
970 struct attribute *attr,
971 const char *buffer, size_t count)
973 struct obd_device *obd = container_of(kobj, struct obd_device,
975 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
979 rc = kstrtobool(buffer, &val);
983 mdt->mdt_opts.mo_migrate_hsm_allowed = val;
986 LUSTRE_RW_ATTR(migrate_hsm_allowed);
988 static ssize_t readonly_show(struct kobject *kobj, struct attribute *attr,
991 struct obd_device *obd = container_of(kobj, struct obd_device,
993 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
995 return scnprintf(buf, PAGE_SIZE, "%u\n", mdt->mdt_readonly);
998 static ssize_t readonly_store(struct kobject *kobj, struct attribute *attr,
999 const char *buffer, size_t count)
1001 struct obd_device *obd = container_of(kobj, struct obd_device,
1003 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
1007 rc = kstrtobool(buffer, &val);
1011 mdt->mdt_readonly = val;
1014 LUSTRE_RW_ATTR(readonly);
1016 static ssize_t enable_remote_rename_show(struct kobject *kobj,
1017 struct attribute *attr,
1020 struct obd_device *obd = container_of(kobj, struct obd_device,
1022 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
1024 return scnprintf(buf, PAGE_SIZE, "%u\n",
1025 mdt->mdt_enable_remote_rename);
1028 static ssize_t enable_remote_rename_store(struct kobject *kobj,
1029 struct attribute *attr,
1030 const char *buffer, size_t count)
1032 struct obd_device *obd = container_of(kobj, struct obd_device,
1034 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
1038 rc = kstrtobool(buffer, &val);
1042 mdt->mdt_enable_remote_rename = val;
1045 LUSTRE_RW_ATTR(enable_remote_rename);
1047 LPROC_SEQ_FOPS_RO_TYPE(mdt, hash);
1048 LPROC_SEQ_FOPS_WR_ONLY(mdt, mds_evict_client);
1049 LUSTRE_RW_ATTR(job_cleanup_interval);
1050 LPROC_SEQ_FOPS_RW_TYPE(mdt, nid_stats_clear);
1051 LUSTRE_RW_ATTR(hsm_control);
1053 LPROC_SEQ_FOPS_RO_TYPE(mdt, recovery_status);
1054 LUSTRE_RW_ATTR(recovery_time_hard);
1055 LUSTRE_RW_ATTR(recovery_time_soft);
1056 LUSTRE_RW_ATTR(ir_factor);
1058 LUSTRE_RO_ATTR(tot_dirty);
1059 LUSTRE_RO_ATTR(tot_granted);
1060 LUSTRE_RO_ATTR(tot_pending);
1061 LUSTRE_RW_ATTR(grant_compat_disable);
1062 LUSTRE_RO_ATTR(instance);
1064 LUSTRE_RO_ATTR(num_exports);
1066 static struct attribute *mdt_attrs[] = {
1067 &lustre_attr_tot_dirty.attr,
1068 &lustre_attr_tot_granted.attr,
1069 &lustre_attr_tot_pending.attr,
1070 &lustre_attr_grant_compat_disable.attr,
1071 &lustre_attr_instance.attr,
1072 &lustre_attr_recovery_time_hard.attr,
1073 &lustre_attr_recovery_time_soft.attr,
1074 &lustre_attr_ir_factor.attr,
1075 &lustre_attr_num_exports.attr,
1076 &lustre_attr_identity_expire.attr,
1077 &lustre_attr_identity_acquire_expire.attr,
1078 &lustre_attr_identity_upcall.attr,
1079 &lustre_attr_identity_flush.attr,
1080 &lustre_attr_evict_tgt_nids.attr,
1081 &lustre_attr_enable_remote_dir.attr,
1082 &lustre_attr_enable_remote_dir_gid.attr,
1083 &lustre_attr_enable_chprojid_gid.attr,
1084 &lustre_attr_enable_striped_dir.attr,
1085 &lustre_attr_enable_dir_migration.attr,
1086 &lustre_attr_enable_remote_rename.attr,
1087 &lustre_attr_commit_on_sharing.attr,
1088 &lustre_attr_async_commit_count.attr,
1089 &lustre_attr_sync_count.attr,
1090 &lustre_attr_dom_lock.attr,
1091 &lustre_attr_dom_read_open.attr,
1092 &lustre_attr_migrate_hsm_allowed.attr,
1093 &lustre_attr_hsm_control.attr,
1094 &lustre_attr_job_cleanup_interval.attr,
1095 &lustre_attr_readonly.attr,
1099 static struct lprocfs_vars lprocfs_mdt_obd_vars[] = {
1100 { .name = "recovery_status",
1101 .fops = &mdt_recovery_status_fops },
1102 { .name = "identity_info",
1103 .fops = &mdt_identity_info_fops },
1104 { .name = "site_stats",
1105 .fops = &mdt_site_stats_fops },
1106 { .name = "evict_client",
1107 .fops = &mdt_mds_evict_client_fops },
1108 { .name = "hash_stats",
1109 .fops = &mdt_hash_fops },
1110 { .name = "root_squash",
1111 .fops = &mdt_root_squash_fops },
1112 { .name = "nosquash_nids",
1113 .fops = &mdt_nosquash_nids_fops },
1118 lprocfs_mdt_print_open_files(struct cfs_hash *hs, struct cfs_hash_bd *bd,
1119 struct hlist_node *hnode, void *v)
1121 struct obd_export *exp = cfs_hash_object(hs, hnode);
1122 struct seq_file *seq = v;
1124 if (exp->exp_lock_hash != NULL) {
1125 struct mdt_export_data *med = &exp->exp_mdt_data;
1126 struct mdt_file_data *mfd;
1128 spin_lock(&med->med_open_lock);
1129 list_for_each_entry(mfd, &med->med_open_head, mfd_list) {
1130 seq_printf(seq, DFID"\n",
1131 PFID(mdt_object_fid(mfd->mfd_object)));
1133 spin_unlock(&med->med_open_lock);
1139 static int lprocfs_mdt_open_files_seq_show(struct seq_file *seq, void *v)
1141 struct nid_stat *stats = seq->private;
1142 struct obd_device *obd = stats->nid_obd;
1144 cfs_hash_for_each_key(obd->obd_nid_hash, &stats->nid,
1145 lprocfs_mdt_print_open_files, seq);
1150 int lprocfs_mdt_open_files_seq_open(struct inode *inode, struct file *file)
1152 struct seq_file *seq;
1155 rc = single_open(file, &lprocfs_mdt_open_files_seq_show, NULL);
1159 seq = file->private_data;
1160 seq->private = PDE_DATA(inode);
1165 void mdt_counter_incr(struct ptlrpc_request *req, int opcode)
1167 struct obd_export *exp = req->rq_export;
1169 if (exp->exp_obd && exp->exp_obd->obd_md_stats)
1170 lprocfs_counter_incr(exp->exp_obd->obd_md_stats,
1171 opcode + LPROC_MD_LAST_OPC);
1172 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
1173 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
1174 if (exp->exp_obd && exp->exp_obd->u.obt.obt_jobstats.ojs_hash &&
1175 (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS))
1176 lprocfs_job_stats_log(exp->exp_obd,
1177 lustre_msg_get_jobid(req->rq_reqmsg),
1181 static const char * const mdt_stats[] = {
1182 [LPROC_MDT_OPEN] = "open",
1183 [LPROC_MDT_CLOSE] = "close",
1184 [LPROC_MDT_MKNOD] = "mknod",
1185 [LPROC_MDT_LINK] = "link",
1186 [LPROC_MDT_UNLINK] = "unlink",
1187 [LPROC_MDT_MKDIR] = "mkdir",
1188 [LPROC_MDT_RMDIR] = "rmdir",
1189 [LPROC_MDT_RENAME] = "rename",
1190 [LPROC_MDT_GETATTR] = "getattr",
1191 [LPROC_MDT_SETATTR] = "setattr",
1192 [LPROC_MDT_GETXATTR] = "getxattr",
1193 [LPROC_MDT_SETXATTR] = "setxattr",
1194 [LPROC_MDT_STATFS] = "statfs",
1195 [LPROC_MDT_SYNC] = "sync",
1196 [LPROC_MDT_SAMEDIR_RENAME] = "samedir_rename",
1197 [LPROC_MDT_CROSSDIR_RENAME] = "crossdir_rename",
1198 [LPROC_MDT_IO_READ] = "read_bytes",
1199 [LPROC_MDT_IO_WRITE] = "write_bytes",
1200 [LPROC_MDT_IO_PUNCH] = "punch",
1203 void mdt_stats_counter_init(struct lprocfs_stats *stats)
1207 LASSERT(stats && stats->ls_num >= ARRAY_SIZE(mdt_stats));
1209 for (idx = 0; idx < ARRAY_SIZE(mdt_stats); idx++) {
1212 if (idx == LPROC_MDT_IO_WRITE || idx == LPROC_MDT_IO_READ)
1213 flags = LPROCFS_CNTR_AVGMINMAX;
1215 lprocfs_counter_init(stats, idx, flags, mdt_stats[idx], "reqs");
1219 int mdt_tunables_init(struct mdt_device *mdt, const char *name)
1221 struct obd_device *obd = mdt2obd_dev(mdt);
1226 LASSERT(name != NULL);
1228 obd->obd_ktype.default_attrs = mdt_attrs;
1229 obd->obd_vars = lprocfs_mdt_obd_vars;
1230 rc = lprocfs_obd_setup(obd, true);
1232 CERROR("%s: cannot create proc entries: rc = %d\n",
1233 mdt_obd_name(mdt), rc);
1237 rc = tgt_tunables_init(&mdt->mdt_lut);
1239 CERROR("%s: failed to init target tunables: rc = %d\n",
1240 mdt_obd_name(mdt), rc);
1244 rc = hsm_cdt_tunables_init(mdt);
1246 CERROR("%s: cannot create hsm proc entries: rc = %d\n",
1247 mdt_obd_name(mdt), rc);
1251 obd->obd_proc_exports_entry = proc_mkdir("exports",
1252 obd->obd_proc_entry);
1253 if (obd->obd_proc_exports_entry)
1254 lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
1255 obd, &mdt_nid_stats_clear_fops);
1257 rc = lprocfs_alloc_md_stats(obd, ARRAY_SIZE(mdt_stats));
1261 /* add additional MDT md_stats after the default ones */
1262 for (i = 0; i < ARRAY_SIZE(mdt_stats); i++) {
1263 int idx = i + LPROC_MD_LAST_OPC;
1266 if (idx == LPROC_MDT_IO_WRITE || idx == LPROC_MDT_IO_READ)
1267 flags = LPROCFS_CNTR_AVGMINMAX;
1269 lprocfs_counter_init(obd->obd_md_stats, idx, flags,
1270 mdt_stats[i], "reqs");
1273 rc = lprocfs_job_stats_init(obd, ARRAY_SIZE(mdt_stats),
1274 mdt_stats_counter_init);
1276 rc = lproc_mdt_attach_rename_seqstat(mdt);
1278 CERROR("%s: MDT can not create rename stats rc = %d\n",
1279 mdt_obd_name(mdt), rc);
1284 void mdt_tunables_fini(struct mdt_device *mdt)
1286 struct obd_device *obd = mdt2obd_dev(mdt);
1288 if (obd->obd_proc_exports_entry != NULL) {
1289 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
1290 obd->obd_proc_exports_entry = NULL;
1293 lprocfs_free_per_client_stats(obd);
1294 /* hsm_cdt_tunables is disabled earlier than this to avoid
1295 * coordinator restart.
1297 hsm_cdt_tunables_fini(mdt);
1298 tgt_tunables_fini(&mdt->mdt_lut);
1299 lprocfs_obd_cleanup(obd);
1300 lprocfs_free_md_stats(obd);
1301 lprocfs_free_obd_stats(obd);
1302 lprocfs_job_stats_fini(obd);