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 ofd->ofd_seq_count++;
146 write_unlock(&ofd->ofd_seq_list_lock);
150 obd_id ofd_seq_last_oid(struct ofd_seq *oseq)
154 spin_lock(&oseq->os_last_oid_lock);
155 id = oseq->os_last_oid;
156 spin_unlock(&oseq->os_last_oid_lock);
161 void ofd_seq_last_oid_set(struct ofd_seq *oseq, obd_id id)
163 spin_lock(&oseq->os_last_oid_lock);
164 if (likely(oseq->os_last_oid < id))
165 oseq->os_last_oid = id;
166 spin_unlock(&oseq->os_last_oid_lock);
169 int ofd_seq_last_oid_write(const struct lu_env *env, struct ofd_device *ofd,
170 struct ofd_seq *oseq)
172 struct ofd_thread_info *info = ofd_info(env);
178 info->fti_buf.lb_buf = &tmp;
179 info->fti_buf.lb_len = sizeof(tmp);
182 CDEBUG(D_INODE, "%s: write last_objid for seq "LPX64" : "LPX64"\n",
183 ofd_name(ofd), oseq->os_seq, ofd_seq_last_oid(oseq));
185 tmp = cpu_to_le64(ofd_seq_last_oid(oseq));
187 rc = ofd_record_write(env, ofd, oseq->os_lastid_obj, &info->fti_buf,
192 void ofd_seqs_fini(const struct lu_env *env, struct ofd_device *ofd)
194 struct ofd_seq *oseq;
198 CFS_INIT_LIST_HEAD(&dispose);
199 write_lock(&ofd->ofd_seq_list_lock);
200 cfs_list_for_each_entry_safe(oseq, tmp, &ofd->ofd_seq_list, os_list) {
201 cfs_list_move(&oseq->os_list, &dispose);
203 write_unlock(&ofd->ofd_seq_list_lock);
205 while (!cfs_list_empty(&dispose)) {
206 oseq = container_of0(dispose.next, struct ofd_seq, os_list);
207 ofd_seq_delete(env, oseq);
210 LASSERT(cfs_list_empty(&ofd->ofd_seq_list));
214 struct ofd_seq *ofd_seq_load(const struct lu_env *env, struct ofd_device *ofd,
217 struct ofd_thread_info *info = ofd_info(env);
218 struct ofd_seq *oseq = NULL;
219 struct dt_object *dob;
225 /* if seq is already initialized */
226 oseq = ofd_seq_get(ofd, seq);
232 RETURN(ERR_PTR(-ENOMEM));
234 lu_last_id_fid(&info->fti_fid, seq);
235 memset(&info->fti_attr, 0, sizeof(info->fti_attr));
236 info->fti_attr.la_valid = LA_MODE;
237 info->fti_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
238 info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
240 /* create object tracking per-seq last created
241 * id to be used by orphan recovery mechanism */
242 dob = dt_find_or_create(env, ofd->ofd_osd, &info->fti_fid,
243 &info->fti_dof, &info->fti_attr);
249 oseq->os_lastid_obj = dob;
251 CFS_INIT_LIST_HEAD(&oseq->os_list);
252 mutex_init(&oseq->os_create_lock);
253 spin_lock_init(&oseq->os_last_oid_lock);
256 cfs_atomic_set(&oseq->os_refc, 1);
258 rc = dt_attr_get(env, dob, &info->fti_attr, BYPASS_CAPA);
262 if (info->fti_attr.la_size == 0) {
263 /* object is just created, initialize last id */
264 oseq->os_last_oid = OFD_INIT_OBJID;
265 ofd_seq_last_oid_write(env, ofd, oseq);
266 } else if (info->fti_attr.la_size == sizeof(lastid)) {
268 info->fti_buf.lb_buf = &lastid;
269 info->fti_buf.lb_len = sizeof(lastid);
271 rc = dt_record_read(env, dob, &info->fti_buf, &info->fti_off);
273 CERROR("%s: can't read last_id: rc = %d\n",
277 oseq->os_last_oid = le64_to_cpu(lastid);
279 CERROR("%s: corrupted size "LPU64" LAST_ID of seq "LPX64"\n",
280 ofd_name(ofd), (__u64)info->fti_attr.la_size, seq);
281 GOTO(cleanup, rc = -EINVAL);
284 oseq = ofd_seq_add(env, ofd, oseq);
287 ofd_seq_put(env, oseq);
291 /* object sequence management */
292 int ofd_seqs_init(const struct lu_env *env, struct ofd_device *ofd)
294 rwlock_init(&ofd->ofd_seq_list_lock);
295 CFS_INIT_LIST_HEAD(&ofd->ofd_seq_list);
296 ofd->ofd_seq_count = 0;
300 int ofd_clients_data_init(const struct lu_env *env, struct ofd_device *ofd,
303 struct obd_device *obd = ofd_obd(ofd);
304 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
305 struct lsd_client_data *lcd = NULL;
306 struct filter_export_data *fed;
309 loff_t off = lsd->lsd_client_start;
311 CLASSERT(offsetof(struct lsd_client_data, lcd_padding) +
312 sizeof(lcd->lcd_padding) == LR_CLIENT_SIZE);
318 for (cl_idx = 0; off < fsize; cl_idx++) {
319 struct obd_export *exp;
322 /* Don't assume off is incremented properly by
323 * fsfilt_read_record(), in case sizeof(*lcd)
324 * isn't the same as fsd->lsd_client_size. */
325 off = lsd->lsd_client_start + cl_idx * lsd->lsd_client_size;
326 rc = tgt_client_data_read(env, &ofd->ofd_lut, lcd, &off, cl_idx);
328 CERROR("%s: error reading FILT %s idx %d off %llu: "
329 "rc = %d\n", ofd_name(ofd), LAST_RCVD, cl_idx,
332 break; /* read error shouldn't cause startup to fail */
335 if (lcd->lcd_uuid[0] == '\0') {
336 CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
341 last_rcvd = lcd->lcd_last_transno;
343 /* These exports are cleaned up by ofd_disconnect(), so they
344 * need to be set up like real exports as ofd_connect() does.
346 exp = class_new_export(obd, (struct obd_uuid *)lcd->lcd_uuid);
348 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
349 " srv lr: "LPU64"\n", lcd->lcd_uuid, cl_idx,
350 last_rcvd, lsd->lsd_last_transno);
353 if (PTR_ERR(exp) == -EALREADY) {
354 /* export already exists, zero out this one */
355 CERROR("%s: Duplicate export %s!\n",
356 ofd_name(ofd), lcd->lcd_uuid);
359 GOTO(err_out, rc = PTR_ERR(exp));
362 fed = &exp->exp_filter_data;
363 *fed->fed_ted.ted_lcd = *lcd;
365 rc = tgt_client_add(env, exp, cl_idx);
366 LASSERTF(rc == 0, "rc = %d\n", rc); /* can't fail existing */
367 /* VBR: set export last committed version */
368 exp->exp_last_committed = last_rcvd;
369 spin_lock(&exp->exp_lock);
370 exp->exp_connecting = 0;
371 exp->exp_in_recovery = 0;
372 spin_unlock(&exp->exp_lock);
373 obd->obd_max_recoverable_clients++;
374 class_export_put(exp);
376 /* Need to check last_rcvd even for duplicated exports. */
377 CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
380 spin_lock(&ofd->ofd_lut.lut_translock);
381 if (last_rcvd > lsd->lsd_last_transno)
382 lsd->lsd_last_transno = last_rcvd;
383 spin_unlock(&ofd->ofd_lut.lut_translock);
391 int ofd_server_data_init(const struct lu_env *env, struct ofd_device *ofd)
393 struct ofd_thread_info *info = ofd_info(env);
394 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
395 struct obd_device *obd = ofd_obd(ofd);
396 unsigned long last_rcvd_size;
400 rc = dt_attr_get(env, ofd->ofd_lut.lut_last_rcvd, &info->fti_attr,
405 last_rcvd_size = (unsigned long)info->fti_attr.la_size;
407 /* ensure padding in the struct is the correct size */
408 CLASSERT (offsetof(struct lr_server_data, lsd_padding) +
409 sizeof(lsd->lsd_padding) == LR_SERVER_SIZE);
411 rc = server_name2index(obd->obd_name, &index, NULL);
413 CERROR("%s: Can not get index from obd_name: rc = %d\n",
418 if (last_rcvd_size == 0) {
419 LCONSOLE_WARN("%s: new disk, initializing\n", obd->obd_name);
421 memcpy(lsd->lsd_uuid, obd->obd_uuid.uuid,
422 sizeof(lsd->lsd_uuid));
423 lsd->lsd_last_transno = 0;
424 lsd->lsd_mount_count = 0;
425 lsd->lsd_server_size = LR_SERVER_SIZE;
426 lsd->lsd_client_start = LR_CLIENT_START;
427 lsd->lsd_client_size = LR_CLIENT_SIZE;
428 lsd->lsd_subdir_count = FILTER_SUBDIR_COUNT;
429 lsd->lsd_feature_incompat = OBD_INCOMPAT_OST;
430 lsd->lsd_osd_index = index;
432 rc = tgt_server_data_read(env, &ofd->ofd_lut);
434 CDEBUG(D_INODE,"OBD ofd: error reading %s: rc %d\n",
438 if (strcmp((char *)lsd->lsd_uuid,
439 (char *)obd->obd_uuid.uuid)) {
440 LCONSOLE_ERROR("Trying to start OBD %s using the wrong"
441 " disk %s. Were the /dev/ assignments "
443 obd->obd_uuid.uuid, lsd->lsd_uuid);
444 GOTO(err_fsd, rc = -EINVAL);
447 if (lsd->lsd_osd_index == 0) {
448 lsd->lsd_osd_index = index;
449 } else if (lsd->lsd_osd_index != index) {
450 LCONSOLE_ERROR("%s: index %d in last rcvd is different"
451 " with the index %d in config log."
452 " It might be disk corruption!\n",
453 obd->obd_name, lsd->lsd_osd_index,
455 GOTO(err_fsd, rc = -EINVAL);
459 lsd->lsd_mount_count++;
460 obd->u.obt.obt_mount_count = lsd->lsd_mount_count;
461 obd->u.obt.obt_instance = (__u32)obd->u.obt.obt_mount_count;
462 ofd->ofd_subdir_count = lsd->lsd_subdir_count;
464 if (lsd->lsd_feature_incompat & ~OFD_INCOMPAT_SUPP) {
465 CERROR("%s: unsupported incompat filesystem feature(s) %x\n",
467 lsd->lsd_feature_incompat & ~OFD_INCOMPAT_SUPP);
468 GOTO(err_fsd, rc = -EINVAL);
470 if (lsd->lsd_feature_rocompat & ~OFD_ROCOMPAT_SUPP) {
471 CERROR("%s: unsupported read-only filesystem feature(s) %x\n",
473 lsd->lsd_feature_rocompat & ~OFD_ROCOMPAT_SUPP);
474 /* Do something like remount filesystem read-only */
475 GOTO(err_fsd, rc = -EINVAL);
478 CDEBUG(D_INODE, "%s: server last_transno : "LPU64"\n",
479 obd->obd_name, lsd->lsd_last_transno);
480 CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
481 obd->obd_name, lsd->lsd_mount_count);
482 CDEBUG(D_INODE, "%s: server data size: %u\n",
483 obd->obd_name, lsd->lsd_server_size);
484 CDEBUG(D_INODE, "%s: per-client data start: %u\n",
485 obd->obd_name, lsd->lsd_client_start);
486 CDEBUG(D_INODE, "%s: per-client data size: %u\n",
487 obd->obd_name, lsd->lsd_client_size);
488 CDEBUG(D_INODE, "%s: server subdir_count: %u\n",
489 obd->obd_name, lsd->lsd_subdir_count);
490 CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
491 last_rcvd_size <= lsd->lsd_client_start ? 0 :
492 (last_rcvd_size - lsd->lsd_client_start) /
493 lsd->lsd_client_size);
495 if (!obd->obd_replayable)
496 CWARN("%s: recovery support OFF\n", obd->obd_name);
498 rc = ofd_clients_data_init(env, ofd, last_rcvd_size);
500 spin_lock(&ofd->ofd_lut.lut_translock);
501 obd->obd_last_committed = lsd->lsd_last_transno;
502 ofd->ofd_lut.lut_last_transno = lsd->lsd_last_transno;
503 spin_unlock(&ofd->ofd_lut.lut_translock);
505 /* save it, so mount count and last_transno is current */
506 rc = tgt_server_data_update(env, &ofd->ofd_lut, 0);
513 class_disconnect_exports(obd);
517 int ofd_fs_setup(const struct lu_env *env, struct ofd_device *ofd,
518 struct obd_device *obd)
520 struct ofd_thread_info *info = ofd_info(env);
521 struct dt_object *fo;
526 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FS_SETUP))
529 /* prepare transactions callbacks */
530 ofd->ofd_txn_cb.dtc_txn_start = NULL;
531 ofd->ofd_txn_cb.dtc_txn_stop = ofd_txn_stop_cb;
532 ofd->ofd_txn_cb.dtc_txn_commit = NULL;
533 ofd->ofd_txn_cb.dtc_cookie = ofd;
534 ofd->ofd_txn_cb.dtc_tag = LCT_DT_THREAD;
535 CFS_INIT_LIST_HEAD(&ofd->ofd_txn_cb.dtc_linkage);
537 dt_txn_callback_add(ofd->ofd_osd, &ofd->ofd_txn_cb);
539 rc = ofd_server_data_init(env, ofd);
543 lu_local_obj_fid(&info->fti_fid, OFD_HEALTH_CHECK_OID);
544 memset(&info->fti_attr, 0, sizeof(info->fti_attr));
545 info->fti_attr.la_valid = LA_MODE;
546 info->fti_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
547 info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
549 fo = dt_find_or_create(env, ofd->ofd_osd, &info->fti_fid,
550 &info->fti_dof, &info->fti_attr);
552 GOTO(out, rc = PTR_ERR(fo));
554 ofd->ofd_health_check_file = fo;
556 rc = ofd_seqs_init(env, ofd);
562 lu_object_put(env, &ofd->ofd_health_check_file->do_lu);
564 dt_txn_callback_del(ofd->ofd_osd, &ofd->ofd_txn_cb);
568 void ofd_fs_cleanup(const struct lu_env *env, struct ofd_device *ofd)
574 ofd_info_init(env, NULL);
576 ofd_seqs_fini(env, ofd);
578 i = dt_sync(env, ofd->ofd_osd);
580 CERROR("can't sync: %d\n", i);
582 /* Remove transaction callback */
583 dt_txn_callback_del(ofd->ofd_osd, &ofd->ofd_txn_cb);
585 if (ofd->ofd_health_check_file) {
586 lu_object_put(env, &ofd->ofd_health_check_file->do_lu);
587 ofd->ofd_health_check_file = NULL;