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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * Author: Alexey Zhuravlev <bzzz@whamcloud.com>
39 * Author: Mikhail Pershin <tappro@whamcloud.com>
42 #define DEBUG_SUBSYSTEM S_FILTER
44 #include "ofd_internal.h"
46 int ofd_record_write(const struct lu_env *env, struct ofd_device *ofd,
47 struct dt_object *dt, struct lu_buf *buf, loff_t *off)
56 th = dt_trans_create(env, ofd->ofd_osd);
60 rc = dt_declare_record_write(env, dt, buf->lb_len, *off, th);
62 rc = dt_trans_start_local(env, ofd->ofd_osd, th);
64 rc = dt_record_write(env, dt, buf, off, th);
66 dt_trans_stop(env, ofd->ofd_osd, th);
71 int ofd_precreate_batch(struct ofd_device *ofd, int batch)
75 spin_lock(&ofd->ofd_batch_lock);
76 count = min(ofd->ofd_precreate_batch, batch);
77 spin_unlock(&ofd->ofd_batch_lock);
82 struct ofd_seq *ofd_seq_get(struct ofd_device *ofd, obd_seq seq)
86 read_lock(&ofd->ofd_seq_list_lock);
87 cfs_list_for_each_entry(oseq, &ofd->ofd_seq_list, os_list) {
88 if (oseq->os_seq == seq) {
89 cfs_atomic_inc(&oseq->os_refc);
90 read_unlock(&ofd->ofd_seq_list_lock);
94 read_unlock(&ofd->ofd_seq_list_lock);
98 static void ofd_seq_destroy(const struct lu_env *env,
101 LASSERT(cfs_list_empty(&oseq->os_list));
102 LASSERT(oseq->os_lastid_obj != NULL);
103 lu_object_put(env, &oseq->os_lastid_obj->do_lu);
107 void ofd_seq_put(const struct lu_env *env, struct ofd_seq *oseq)
109 if (cfs_atomic_dec_and_test(&oseq->os_refc))
110 ofd_seq_destroy(env, oseq);
113 static void ofd_seq_delete(const struct lu_env *env, struct ofd_seq *oseq)
115 cfs_list_del_init(&oseq->os_list);
116 ofd_seq_put(env, oseq);
120 * Add a new sequence to the OFD device.
122 * \param ofd OFD device
123 * \param new_seq new sequence to be added
125 * \retval the seq to be added or the existing seq
127 static struct ofd_seq *ofd_seq_add(const struct lu_env *env,
128 struct ofd_device *ofd,
129 struct ofd_seq *new_seq)
131 struct ofd_seq *os = NULL;
133 write_lock(&ofd->ofd_seq_list_lock);
134 cfs_list_for_each_entry(os, &ofd->ofd_seq_list, os_list) {
135 if (os->os_seq == new_seq->os_seq) {
136 cfs_atomic_inc(&os->os_refc);
137 write_unlock(&ofd->ofd_seq_list_lock);
138 /* The seq has not been added to the list */
139 ofd_seq_put(env, new_seq);
143 cfs_atomic_inc(&new_seq->os_refc);
144 cfs_list_add_tail(&new_seq->os_list, &ofd->ofd_seq_list);
145 write_unlock(&ofd->ofd_seq_list_lock);
149 obd_id ofd_seq_last_oid(struct ofd_seq *oseq)
153 spin_lock(&oseq->os_last_oid_lock);
154 id = oseq->os_last_oid;
155 spin_unlock(&oseq->os_last_oid_lock);
160 void ofd_seq_last_oid_set(struct ofd_seq *oseq, obd_id id)
162 spin_lock(&oseq->os_last_oid_lock);
163 if (likely(oseq->os_last_oid < id))
164 oseq->os_last_oid = id;
165 spin_unlock(&oseq->os_last_oid_lock);
168 int ofd_seq_last_oid_write(const struct lu_env *env, struct ofd_device *ofd,
169 struct ofd_seq *oseq)
171 struct ofd_thread_info *info = ofd_info(env);
177 info->fti_buf.lb_buf = &tmp;
178 info->fti_buf.lb_len = sizeof(tmp);
181 CDEBUG(D_INODE, "%s: write last_objid for seq "LPX64" : "LPX64"\n",
182 ofd_name(ofd), oseq->os_seq, ofd_seq_last_oid(oseq));
184 tmp = cpu_to_le64(ofd_seq_last_oid(oseq));
186 rc = ofd_record_write(env, ofd, oseq->os_lastid_obj, &info->fti_buf,
191 static int ofd_seq_count_write(const struct lu_env *env, struct ofd_device *ofd)
193 struct ofd_thread_info *info = ofd_info(env);
199 info->fti_buf.lb_buf = &tmp;
200 info->fti_buf.lb_len = sizeof(tmp);
203 tmp = cpu_to_le32(ofd->ofd_seq_count);
205 rc = ofd_record_write(env, ofd, ofd->ofd_seq_count_file,
206 &info->fti_buf, &info->fti_off);
211 void ofd_seqs_fini(const struct lu_env *env, struct ofd_device *ofd)
213 struct ofd_seq *oseq;
217 CFS_INIT_LIST_HEAD(&dispose);
218 write_lock(&ofd->ofd_seq_list_lock);
219 cfs_list_for_each_entry_safe(oseq, tmp, &ofd->ofd_seq_list, os_list) {
220 cfs_list_move(&oseq->os_list, &dispose);
222 write_unlock(&ofd->ofd_seq_list_lock);
224 while (!cfs_list_empty(&dispose)) {
225 oseq = container_of0(dispose.next, struct ofd_seq, os_list);
226 ofd_seq_delete(env, oseq);
229 LASSERT(cfs_list_empty(&ofd->ofd_seq_list));
233 struct ofd_seq *ofd_seq_load(const struct lu_env *env, struct ofd_device *ofd,
236 struct ofd_thread_info *info = ofd_info(env);
237 struct ofd_seq *oseq = NULL;
238 struct dt_object *dob;
244 /* if seq is already initialized */
245 oseq = ofd_seq_get(ofd, seq);
251 RETURN(ERR_PTR(-ENOMEM));
253 lu_last_id_fid(&info->fti_fid, seq);
254 memset(&info->fti_attr, 0, sizeof(info->fti_attr));
255 info->fti_attr.la_valid = LA_MODE;
256 info->fti_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
257 info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
259 /* create object tracking per-seq last created
260 * id to be used by orphan recovery mechanism */
261 dob = dt_find_or_create(env, ofd->ofd_osd, &info->fti_fid,
262 &info->fti_dof, &info->fti_attr);
268 oseq->os_lastid_obj = dob;
270 CFS_INIT_LIST_HEAD(&oseq->os_list);
271 mutex_init(&oseq->os_create_lock);
272 spin_lock_init(&oseq->os_last_oid_lock);
275 cfs_atomic_set(&oseq->os_refc, 1);
277 rc = dt_attr_get(env, dob, &info->fti_attr, BYPASS_CAPA);
281 if (info->fti_attr.la_size == 0) {
282 /* object is just created, initialize last id */
283 oseq->os_last_oid = OFD_INIT_OBJID;
284 ofd_seq_last_oid_write(env, ofd, oseq);
285 ofd_seq_count_write(env, ofd);
286 } else if (info->fti_attr.la_size == sizeof(lastid)) {
288 info->fti_buf.lb_buf = &lastid;
289 info->fti_buf.lb_len = sizeof(lastid);
291 rc = dt_record_read(env, dob, &info->fti_buf, &info->fti_off);
293 CERROR("%s: can't read last_id: rc = %d\n",
297 oseq->os_last_oid = le64_to_cpu(lastid);
299 CERROR("%s: corrupted size "LPU64" LAST_ID of seq "LPX64"\n",
300 ofd_name(ofd), (__u64)info->fti_attr.la_size, seq);
301 GOTO(cleanup, rc = -EINVAL);
304 oseq = ofd_seq_add(env, ofd, oseq);
307 ofd_seq_put(env, oseq);
311 /* object sequence management */
312 int ofd_seqs_init(const struct lu_env *env, struct ofd_device *ofd)
314 struct ofd_thread_info *info = ofd_info(env);
315 unsigned long seq_count_size;
322 rwlock_init(&ofd->ofd_seq_list_lock);
323 CFS_INIT_LIST_HEAD(&ofd->ofd_seq_list);
325 rc = dt_attr_get(env, ofd->ofd_seq_count_file,
326 &info->fti_attr, BYPASS_CAPA);
330 seq_count_size = (unsigned long)info->fti_attr.la_size;
332 if (seq_count_size == sizeof(seq_count)) {
334 info->fti_buf.lb_buf = &seq_count;
335 info->fti_buf.lb_len = sizeof(seq_count);
337 rc = dt_record_read(env, ofd->ofd_seq_count_file,
338 &info->fti_buf, &info->fti_off);
340 CERROR("%s: can't read LAST_GROUP: rc = %d\n",
345 ofd->ofd_seq_count = le64_to_cpu(seq_count);
346 } else if (seq_count_size == 0) {
347 ofd->ofd_seq_count = 0;
349 CERROR("%s: seqs file is corrupted? size = %lu\n",
350 ofd_name(ofd), seq_count_size);
351 GOTO(cleanup, rc = -EIO);
354 for (i = 0; i <= ofd->ofd_seq_count; i++) {
355 struct ofd_seq *oseq;
357 oseq = ofd_seq_load(env, ofd, i);
359 CERROR("%s: can't load seq %d: rc = %d\n",
360 ofd_name(ofd), i, rc);
361 /* Clean all previously set seqs */
362 ofd_seqs_fini(env, ofd);
365 ofd_seq_put(env, oseq);
369 CDEBUG(D_OTHER, "%s: %u seqs initialized\n", ofd_name(ofd),
370 ofd->ofd_seq_count + 1);
375 int ofd_clients_data_init(const struct lu_env *env, struct ofd_device *ofd,
378 struct obd_device *obd = ofd_obd(ofd);
379 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
380 struct lsd_client_data *lcd = NULL;
381 struct filter_export_data *fed;
384 loff_t off = lsd->lsd_client_start;
386 CLASSERT(offsetof(struct lsd_client_data, lcd_padding) +
387 sizeof(lcd->lcd_padding) == LR_CLIENT_SIZE);
393 for (cl_idx = 0; off < fsize; cl_idx++) {
394 struct obd_export *exp;
397 /* Don't assume off is incremented properly by
398 * fsfilt_read_record(), in case sizeof(*lcd)
399 * isn't the same as fsd->lsd_client_size. */
400 off = lsd->lsd_client_start + cl_idx * lsd->lsd_client_size;
401 rc = tgt_client_data_read(env, &ofd->ofd_lut, lcd, &off, cl_idx);
403 CERROR("%s: error reading FILT %s idx %d off %llu: "
404 "rc = %d\n", ofd_name(ofd), LAST_RCVD, cl_idx,
407 break; /* read error shouldn't cause startup to fail */
410 if (lcd->lcd_uuid[0] == '\0') {
411 CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
416 last_rcvd = lcd->lcd_last_transno;
418 /* These exports are cleaned up by ofd_disconnect(), so they
419 * need to be set up like real exports as ofd_connect() does.
421 exp = class_new_export(obd, (struct obd_uuid *)lcd->lcd_uuid);
423 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
424 " srv lr: "LPU64"\n", lcd->lcd_uuid, cl_idx,
425 last_rcvd, lsd->lsd_last_transno);
428 if (PTR_ERR(exp) == -EALREADY) {
429 /* export already exists, zero out this one */
430 CERROR("%s: Duplicate export %s!\n",
431 ofd_name(ofd), lcd->lcd_uuid);
434 GOTO(err_out, rc = PTR_ERR(exp));
437 fed = &exp->exp_filter_data;
438 *fed->fed_ted.ted_lcd = *lcd;
440 rc = tgt_client_add(env, exp, cl_idx);
441 LASSERTF(rc == 0, "rc = %d\n", rc); /* can't fail existing */
442 /* VBR: set export last committed version */
443 exp->exp_last_committed = last_rcvd;
444 spin_lock(&exp->exp_lock);
445 exp->exp_connecting = 0;
446 exp->exp_in_recovery = 0;
447 spin_unlock(&exp->exp_lock);
448 obd->obd_max_recoverable_clients++;
449 class_export_put(exp);
451 /* Need to check last_rcvd even for duplicated exports. */
452 CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
455 spin_lock(&ofd->ofd_lut.lut_translock);
456 if (last_rcvd > lsd->lsd_last_transno)
457 lsd->lsd_last_transno = last_rcvd;
458 spin_unlock(&ofd->ofd_lut.lut_translock);
466 int ofd_server_data_init(const struct lu_env *env, struct ofd_device *ofd)
468 struct ofd_thread_info *info = ofd_info(env);
469 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
470 struct obd_device *obd = ofd_obd(ofd);
471 unsigned long last_rcvd_size;
474 rc = dt_attr_get(env, ofd->ofd_lut.lut_last_rcvd, &info->fti_attr,
479 last_rcvd_size = (unsigned long)info->fti_attr.la_size;
481 /* ensure padding in the struct is the correct size */
482 CLASSERT (offsetof(struct lr_server_data, lsd_padding) +
483 sizeof(lsd->lsd_padding) == LR_SERVER_SIZE);
485 if (last_rcvd_size == 0) {
486 LCONSOLE_WARN("%s: new disk, initializing\n", obd->obd_name);
488 memcpy(lsd->lsd_uuid, obd->obd_uuid.uuid,
489 sizeof(lsd->lsd_uuid));
490 lsd->lsd_last_transno = 0;
491 lsd->lsd_mount_count = 0;
492 lsd->lsd_server_size = LR_SERVER_SIZE;
493 lsd->lsd_client_start = LR_CLIENT_START;
494 lsd->lsd_client_size = LR_CLIENT_SIZE;
495 lsd->lsd_subdir_count = FILTER_SUBDIR_COUNT;
496 lsd->lsd_feature_incompat = OBD_INCOMPAT_OST;
498 rc = tgt_server_data_read(env, &ofd->ofd_lut);
500 CDEBUG(D_INODE,"OBD ofd: error reading %s: rc %d\n",
504 if (strcmp((char *)lsd->lsd_uuid,
505 (char *)obd->obd_uuid.uuid)) {
506 LCONSOLE_ERROR("Trying to start OBD %s using the wrong"
507 " disk %s. Were the /dev/ assignments "
509 obd->obd_uuid.uuid, lsd->lsd_uuid);
510 GOTO(err_fsd, rc = -EINVAL);
514 lsd->lsd_mount_count++;
515 obd->u.obt.obt_mount_count = lsd->lsd_mount_count;
516 obd->u.obt.obt_instance = (__u32)obd->u.obt.obt_mount_count;
517 ofd->ofd_subdir_count = lsd->lsd_subdir_count;
519 if (lsd->lsd_feature_incompat & ~OFD_INCOMPAT_SUPP) {
520 CERROR("%s: unsupported incompat filesystem feature(s) %x\n",
522 lsd->lsd_feature_incompat & ~OFD_INCOMPAT_SUPP);
523 GOTO(err_fsd, rc = -EINVAL);
525 if (lsd->lsd_feature_rocompat & ~OFD_ROCOMPAT_SUPP) {
526 CERROR("%s: unsupported read-only filesystem feature(s) %x\n",
528 lsd->lsd_feature_rocompat & ~OFD_ROCOMPAT_SUPP);
529 /* Do something like remount filesystem read-only */
530 GOTO(err_fsd, rc = -EINVAL);
533 CDEBUG(D_INODE, "%s: server last_transno : "LPU64"\n",
534 obd->obd_name, lsd->lsd_last_transno);
535 CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
536 obd->obd_name, lsd->lsd_mount_count);
537 CDEBUG(D_INODE, "%s: server data size: %u\n",
538 obd->obd_name, lsd->lsd_server_size);
539 CDEBUG(D_INODE, "%s: per-client data start: %u\n",
540 obd->obd_name, lsd->lsd_client_start);
541 CDEBUG(D_INODE, "%s: per-client data size: %u\n",
542 obd->obd_name, lsd->lsd_client_size);
543 CDEBUG(D_INODE, "%s: server subdir_count: %u\n",
544 obd->obd_name, lsd->lsd_subdir_count);
545 CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
546 last_rcvd_size <= lsd->lsd_client_start ? 0 :
547 (last_rcvd_size - lsd->lsd_client_start) /
548 lsd->lsd_client_size);
550 if (!obd->obd_replayable)
551 CWARN("%s: recovery support OFF\n", obd->obd_name);
553 rc = ofd_clients_data_init(env, ofd, last_rcvd_size);
555 spin_lock(&ofd->ofd_lut.lut_translock);
556 obd->obd_last_committed = lsd->lsd_last_transno;
557 ofd->ofd_lut.lut_last_transno = lsd->lsd_last_transno;
558 spin_unlock(&ofd->ofd_lut.lut_translock);
560 /* save it, so mount count and last_transno is current */
561 rc = tgt_server_data_update(env, &ofd->ofd_lut, 0);
568 class_disconnect_exports(obd);
572 int ofd_fs_setup(const struct lu_env *env, struct ofd_device *ofd,
573 struct obd_device *obd)
575 struct ofd_thread_info *info = ofd_info(env);
576 struct dt_object *fo;
581 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FS_SETUP))
584 /* prepare transactions callbacks */
585 ofd->ofd_txn_cb.dtc_txn_start = NULL;
586 ofd->ofd_txn_cb.dtc_txn_stop = ofd_txn_stop_cb;
587 ofd->ofd_txn_cb.dtc_txn_commit = NULL;
588 ofd->ofd_txn_cb.dtc_cookie = ofd;
589 ofd->ofd_txn_cb.dtc_tag = LCT_DT_THREAD;
590 CFS_INIT_LIST_HEAD(&ofd->ofd_txn_cb.dtc_linkage);
592 dt_txn_callback_add(ofd->ofd_osd, &ofd->ofd_txn_cb);
594 rc = ofd_server_data_init(env, ofd);
598 lu_local_obj_fid(&info->fti_fid, OFD_HEALTH_CHECK_OID);
599 memset(&info->fti_attr, 0, sizeof(info->fti_attr));
600 info->fti_attr.la_valid = LA_MODE;
601 info->fti_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
602 info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
604 fo = dt_find_or_create(env, ofd->ofd_osd, &info->fti_fid,
605 &info->fti_dof, &info->fti_attr);
607 GOTO(out, rc = PTR_ERR(fo));
609 ofd->ofd_health_check_file = fo;
611 lu_local_obj_fid(&info->fti_fid, OFD_LAST_GROUP_OID);
612 memset(&info->fti_attr, 0, sizeof(info->fti_attr));
613 info->fti_attr.la_valid = LA_MODE;
614 info->fti_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
615 info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
617 fo = dt_find_or_create(env, ofd->ofd_osd, &info->fti_fid,
618 &info->fti_dof, &info->fti_attr);
620 GOTO(out_hc, rc = PTR_ERR(fo));
622 ofd->ofd_seq_count_file = fo;
624 rc = ofd_seqs_init(env, ofd);
630 lu_object_put(env, &ofd->ofd_seq_count_file->do_lu);
632 lu_object_put(env, &ofd->ofd_health_check_file->do_lu);
634 dt_txn_callback_del(ofd->ofd_osd, &ofd->ofd_txn_cb);
638 void ofd_fs_cleanup(const struct lu_env *env, struct ofd_device *ofd)
644 ofd_info_init(env, NULL);
646 ofd_seqs_fini(env, ofd);
648 i = dt_sync(env, ofd->ofd_osd);
650 CERROR("can't sync: %d\n", i);
652 /* Remove transaction callback */
653 dt_txn_callback_del(ofd->ofd_osd, &ofd->ofd_txn_cb);
655 if (ofd->ofd_seq_count_file) {
656 lu_object_put(env, &ofd->ofd_seq_count_file->do_lu);
657 ofd->ofd_seq_count_file = NULL;
660 if (ofd->ofd_health_check_file) {
661 lu_object_put(env, &ofd->ofd_health_check_file->do_lu);
662 ofd->ofd_health_check_file = NULL;