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) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 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/ofd/ofd_obd.c
34 * This file contains OBD API methods for OBD Filter Device (OFD) which are
35 * used for export handling, configuration purposes and recovery.
36 * Several methods are used by ECHO client only since it still uses OBD API.
37 * Such methods have _echo_ prefix in name.
39 * Author: Andreas Dilger <andreas.dilger@intel.com>
40 * Author: Alexey Zhuravlev <alexey.zhuravlev@intel.com>
41 * Author: Mikhail Pershin <mike.pershin@intel.com>
44 #define DEBUG_SUBSYSTEM S_FILTER
46 #include "ofd_internal.h"
47 #include <obd_cksum.h>
48 #include <uapi/linux/lustre/lustre_ioctl.h>
49 #include <lustre_quota.h>
50 #include <lustre_lfsck.h>
51 #include <lustre_nodemap.h>
54 * Initialize OFD per-export statistics.
56 * This function sets up procfs entries for various OFD export counters. These
57 * counters are for per-client statistics tracked on the server.
59 * \param[in] ofd OFD device
60 * \param[in] exp OBD export
61 * \param[in] client_nid NID of client
63 * \retval 0 if successful
64 * \retval negative value on error
66 static int ofd_export_stats_init(struct ofd_device *ofd,
67 struct obd_export *exp,
68 lnet_nid_t *client_nid)
70 struct obd_device *obd = ofd_obd(ofd);
71 struct nid_stat *stats;
75 LASSERT(obd->obd_uses_nid_stats);
77 if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
78 /* Self-export gets no proc entry */
81 rc = lprocfs_exp_setup(exp, client_nid);
83 /* Mask error for already created /proc entries */
84 RETURN(rc == -EALREADY ? 0 : rc);
86 stats = exp->exp_nid_stats;
87 stats->nid_stats = lprocfs_alloc_stats(NUM_OBD_STATS +
89 LPROCFS_STATS_FLAG_NOPERCPU);
90 if (stats->nid_stats == NULL)
93 lprocfs_init_ops_stats(LPROC_OFD_STATS_LAST, stats->nid_stats);
95 ofd_stats_counter_init(stats->nid_stats);
97 rc = lprocfs_register_stats(stats->nid_proc, "stats", stats->nid_stats);
99 lprocfs_free_stats(&stats->nid_stats);
103 rc = lprocfs_nid_ldlm_stats_init(stats);
112 * Decide which checksums both client and OST support, possibly forcing
113 * the use of T10PI checksums if the hardware supports this.
115 * The clients that have no T10-PI RPC checksum support will use the same
116 * mechanism to select checksum type as before, and will not be affected by
117 * the following logic.
119 * For the clients that have T10-PI RPC checksum support:
121 * If the OST supports T10-PI feature and T10-PI checksum is enforced, clients
122 * will have no other choice for RPC checksum type other than using the T10PI
123 * checksum type. This is useful for enforcing end-to-end integrity in the
126 * If the OST doesn't support T10-PI feature and T10-PI checksum is enforced,
127 * together with other checksum with reasonably good speeds (e.g. crc32,
128 * crc32c, adler, etc.), all T10-PI checksum types understood by the client
129 * (t10ip512, t10ip4K, t10crc512, t10crc4K) will be added to the available
130 * checksum types, regardless of the speeds of T10-PI checksums. This is
131 * useful for testing T10-PI checksum of RPC.
133 * If the OST supports T10-PI feature and T10-PI checksum is NOT enforced,
134 * the corresponding T10-PI checksum type will be added to the checksum type
135 * list, regardless of the speed of the T10-PI checksum. This provides clients
136 * the flexibility to choose whether to enable end-to-end integrity or not.
138 * If the OST does NOT supports T10-PI feature and T10-PI checksum is NOT
139 * enforced, together with other checksums with reasonably good speeds,
140 * all the T10-PI checksum types with good speeds will be added into the
141 * checksum type list. Note that a T10-PI checksum type with a speed worse
142 * than half of Alder will NOT be added as a option. In this circumstance,
143 * T10-PI checksum types has the same behavior like other normal checksum
148 ofd_mask_cksum_types(struct ofd_device *ofd, enum cksum_types *cksum_types)
150 bool enforce = ofd->ofd_checksum_t10pi_enforce;
151 enum cksum_types ofd_t10_cksum_type;
152 enum cksum_types client_t10_types = *cksum_types & OBD_CKSUM_T10_ALL;
153 enum cksum_types server_t10_types;
156 * The client set in ocd_cksum_types the checksum types it
157 * supports. We have to mask off the algorithms that we don't
158 * support. T10PI checksum types will be added later.
160 *cksum_types &= (ofd->ofd_cksum_types_supported & ~OBD_CKSUM_T10_ALL);
161 server_t10_types = ofd->ofd_cksum_types_supported & OBD_CKSUM_T10_ALL;
162 ofd_t10_cksum_type = ofd->ofd_lut.lut_dt_conf.ddp_t10_cksum_type;
164 /* Quick exit if no T10-PI support on client */
165 if (!client_t10_types)
169 * This OST has NO T10-PI feature. Add all supported T10-PI checksums
170 * as options if T10-PI checksum is enforced. If the T10-PI checksum is
171 * not enforced, only add them as options when speed is good.
173 if (ofd_t10_cksum_type == 0) {
175 * Server allows all T10PI checksums, and server_t10_types
176 * include quick ones.
179 *cksum_types |= client_t10_types;
181 *cksum_types |= client_t10_types & server_t10_types;
186 * This OST has T10-PI feature. Disable all other checksum types if
187 * T10-PI checksum is enforced. If the T10-PI checksum is not enforced,
188 * add the checksum type as an option.
190 if (client_t10_types & ofd_t10_cksum_type) {
192 *cksum_types = ofd_t10_cksum_type;
194 *cksum_types |= ofd_t10_cksum_type;
199 * Match client and OST server connection feature flags.
201 * Compute the compatibility flags for a connection request based on
202 * features mutually supported by client and server.
204 * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
205 * must not be updated here, otherwise a partially initialized value may
206 * be exposed. After the connection request is successfully processed,
207 * the top-level tgt_connect() request handler atomically updates the export
208 * connect flags from the obd_connect_data::ocd_connect_flags field of the
209 * reply. \see tgt_connect().
211 * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
212 * full struct obd_connect_data. So care must be taken when accessing fields
213 * that are not present in struct obd_connect_data_v1. See LU-16.
215 * \param[in] env execution environment
216 * \param[in] exp the obd_export associated with this
218 * \param[in] data stores data for this connect request
219 * \param[in] new_connection is this connection new or not
221 * \retval 0 if success
222 * \retval -EPROTO client and server feature requirements are
224 * \retval -EBADF OST index in connect request doesn't match
227 static int ofd_parse_connect_data(const struct lu_env *env,
228 struct obd_export *exp,
229 struct obd_connect_data *data,
232 struct ofd_device *ofd = ofd_exp(exp);
233 struct filter_export_data *fed = &exp->exp_filter_data;
238 CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: %#llx"
239 " ocd_version: %x ocd_grant: %d ocd_index: %u"
241 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
242 data->ocd_connect_flags, data->ocd_version,
243 data->ocd_grant, data->ocd_index, data->ocd_group);
245 if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
246 CWARN("!!! This export (nid %s) used object group %d "
247 "earlier; now it's trying to use group %d! This could "
248 "be a bug in the MDS. Please report to "
249 "https://jira.hpdd.intel.com/\n",
250 obd_export_nid2str(exp), fed->fed_group,
254 fed->fed_group = data->ocd_group;
256 data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
258 if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
259 data->ocd_connect_flags2 &= OST_CONNECT_SUPPORTED2;
261 /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
262 if (data->ocd_connect_flags & OBD_CONNECT_MDS)
263 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
264 exp->exp_obd->obd_name, data->ocd_group);
265 else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
268 /* Determine optimal brw size before calculating grant */
269 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
270 data->ocd_brw_size = 65536;
271 } else if (OCD_HAS_FLAG(data, BRW_SIZE)) {
272 if (data->ocd_brw_size > ofd->ofd_brw_size)
273 data->ocd_brw_size = ofd->ofd_brw_size;
274 if (data->ocd_brw_size == 0) {
275 CERROR("%s: cli %s/%p ocd_connect_flags: %#llx"
276 " ocd_version: %x ocd_grant: %d ocd_index: %u "
277 "ocd_brw_size is unexpectedly zero, "
278 "network data corruption?"
279 "Refusing connection of this client\n",
280 exp->exp_obd->obd_name,
281 exp->exp_client_uuid.uuid,
282 exp, data->ocd_connect_flags, data->ocd_version,
283 data->ocd_grant, data->ocd_index);
288 if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
289 struct dt_device_param *ddp = &ofd->ofd_lut.lut_dt_conf;
291 /* client is reporting its page size, for future use */
292 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
293 data->ocd_grant_blkbits = ofd->ofd_lut.lut_tgd.tgd_blockbits;
294 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
295 * it's LDISKFS_DIR_REC_LEN(20) = 28. */
296 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
297 /* ocd_grant_tax_kb is in 1K byte blocks */
298 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
299 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
302 /* Save connect_data we have so far because tgt_grant_connect()
303 * uses it to calculate grant, and we want to save the client
304 * version before it is overwritten by LUSTRE_VERSION_CODE. */
305 exp->exp_connect_data = *data;
306 if (OCD_HAS_FLAG(data, GRANT))
307 tgt_grant_connect(env, exp, data, new_connection);
309 if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
310 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
311 int index = lsd->lsd_osd_index;
313 if (index != data->ocd_index) {
314 LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
315 " %u doesn't match actual OST index"
316 " %u in last_rcvd file, bad "
318 obd_export_nid2str(exp), index,
322 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
323 /* this will only happen on the first connect */
324 lsd->lsd_feature_compat |= OBD_COMPAT_OST;
325 /* sync is not needed here as tgt_client_new will
326 * set exp_need_sync flag */
327 tgt_server_data_update(env, &ofd->ofd_lut, 0);
331 if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
332 __u32 cksum_types = data->ocd_cksum_types;
334 ofd_mask_cksum_types(ofd, &data->ocd_cksum_types);
336 if (unlikely(data->ocd_cksum_types == 0)) {
337 CERROR("%s: Connect with checksum support but no "
338 "ocd_cksum_types is set\n",
339 exp->exp_obd->obd_name);
343 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
344 "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
345 cksum_types, data->ocd_cksum_types);
347 /* This client does not support OBD_CONNECT_CKSUM.
348 * Report failure to negotiate checksum at connect */
349 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
350 "OBD_CONNECT_CKSUM\n",
351 exp->exp_obd->obd_name, obd_export_nid2str(exp));
354 if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
355 data->ocd_maxbytes = ofd->ofd_lut.lut_dt_conf.ddp_maxbytes;
357 data->ocd_version = LUSTRE_VERSION_CODE;
359 if (OCD_HAS_FLAG(data, PINGLESS)) {
360 if (ptlrpc_pinger_suppress_pings()) {
361 spin_lock(&exp->exp_obd->obd_dev_lock);
362 list_del_init(&exp->exp_obd_chain_timed);
363 spin_unlock(&exp->exp_obd->obd_dev_lock);
365 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
373 * Re-initialize export upon client reconnection.
375 * This function parses connection data from reconnect and resets
378 * \param[in] env execution environment
379 * \param[in] exp OBD export
380 * \param[in] obd OFD device
381 * \param[in] cluuid NID of client
382 * \param[in] data connection data from request
383 * \param[in] localdata client NID
385 * \retval 0 if successful
386 * \retval negative value on error
388 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
389 struct obd_device *obd, struct obd_uuid *cluuid,
390 struct obd_connect_data *data,
393 struct ofd_device *ofd;
398 if (exp == NULL || obd == NULL || cluuid == NULL)
401 rc = nodemap_add_member(*(lnet_nid_t *)client_nid, exp);
402 if (rc != 0 && rc != -EEXIST)
405 ofd = ofd_dev(obd->obd_lu_dev);
407 rc = ofd_parse_connect_data(env, exp, data, false);
409 ofd_export_stats_init(ofd, exp, client_nid);
411 nodemap_del_member(exp);
417 * Initialize new client connection.
419 * This function handles new connection to the OFD. The new export is
420 * created (in context of class_connect()) and persistent client data is
421 * initialized on storage.
423 * \param[in] env execution environment
424 * \param[out] _exp stores pointer to new export
425 * \param[in] obd OFD device
426 * \param[in] cluuid client UUID
427 * \param[in] data connection data from request
428 * \param[in] localdata client NID
430 * \retval 0 if successful
431 * \retval negative value on error
433 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
434 struct obd_device *obd, struct obd_uuid *cluuid,
435 struct obd_connect_data *data, void *localdata)
437 struct obd_export *exp;
438 struct ofd_device *ofd;
439 struct lustre_handle conn = { 0 };
443 if (_exp == NULL || obd == NULL || cluuid == NULL)
446 ofd = ofd_dev(obd->obd_lu_dev);
448 rc = class_connect(&conn, obd, cluuid);
452 exp = class_conn2export(&conn);
453 LASSERT(exp != NULL);
455 if (localdata != NULL) {
456 rc = nodemap_add_member(*(lnet_nid_t *)localdata, exp);
457 if (rc != 0 && rc != -EEXIST)
460 CDEBUG(D_HA, "%s: cannot find nodemap for client %s: "
461 "nid is null\n", obd->obd_name, cluuid->uuid);
464 rc = ofd_parse_connect_data(env, exp, data, true);
468 if (obd->obd_replayable) {
469 struct tg_export_data *ted = &exp->exp_target_data;
471 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
472 sizeof(ted->ted_lcd->lcd_uuid));
473 rc = tgt_client_new(env, exp);
476 ofd_export_stats_init(ofd, exp, localdata);
479 CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
480 data ? data->ocd_group : -1);
484 class_disconnect(exp);
485 nodemap_del_member(exp);
494 * Disconnect a connected client.
496 * This function terminates the client connection. The client export is
497 * disconnected (cleaned up) and client data on persistent storage is removed.
499 * \param[in] exp OBD export
501 * \retval 0 if successful
502 * \retval negative value on error
504 int ofd_obd_disconnect(struct obd_export *exp)
506 struct ofd_device *ofd = ofd_exp(exp);
513 class_export_get(exp);
515 if (!(exp->exp_flags & OBD_OPT_FORCE))
516 tgt_grant_sanity_check(ofd_obd(ofd), __func__);
518 rc = server_disconnect_export(exp);
520 tgt_grant_discard(exp);
522 /* Do not erase record for recoverable client. */
523 if (exp->exp_obd->obd_replayable &&
524 (!exp->exp_obd->obd_fail || exp->exp_failed)) {
525 rc = lu_env_init(&env, LCT_DT_THREAD);
529 tgt_client_del(&env, exp);
533 nodemap_del_member(exp);
534 class_export_put(exp);
539 * Implementation of obd_ops::o_init_export.
541 * This function is called from class_new_export() and initializes
542 * the OFD-specific data for new export.
544 * \param[in] exp OBD export
546 * \retval 0 if successful
547 * \retval negative value on error
549 static int ofd_init_export(struct obd_export *exp)
553 spin_lock_init(&exp->exp_filter_data.fed_lock);
554 INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
555 atomic_set(&exp->exp_filter_data.fed_soft_sync_count, 0);
556 spin_lock(&exp->exp_lock);
557 exp->exp_connecting = 1;
558 spin_unlock(&exp->exp_lock);
560 /* self-export doesn't need client data and ldlm initialization */
561 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
562 &exp->exp_client_uuid)))
565 rc = tgt_client_alloc(exp);
567 ldlm_init_export(exp);
569 CERROR("%s: Can't initialize export: rc %d\n",
570 exp->exp_obd->obd_name, rc);
575 * Implementation of obd_ops::o_destroy_export.
577 * This function is called from class_export_destroy() to cleanup
578 * the OFD-specific data for export being destroyed.
580 * \param[in] exp OBD export
582 * \retval 0 if successful
583 * \retval negative value on error
585 static int ofd_destroy_export(struct obd_export *exp)
587 struct ofd_device *ofd = ofd_exp(exp);
589 if (exp->exp_target_data.ted_pending)
590 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
591 "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
592 exp, exp->exp_target_data.ted_pending);
594 target_destroy_export(exp);
596 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
597 &exp->exp_client_uuid)))
600 ldlm_destroy_export(exp);
601 tgt_client_free(exp);
603 ofd_fmd_cleanup(exp);
606 * discard grants once we're sure no more
607 * interaction with the client is possible
609 tgt_grant_discard(exp);
610 ofd_fmd_cleanup(exp);
612 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
613 ofd->ofd_lut.lut_tgd.tgd_tot_granted_clients--;
615 if (!(exp->exp_flags & OBD_OPT_FORCE))
616 tgt_grant_sanity_check(exp->exp_obd, __func__);
618 LASSERT(list_empty(&exp->exp_filter_data.fed_mod_list));
623 * Notify all devices in server stack about recovery completion.
625 * This function calls ldo_recovery_complete() for all lower devices in the
626 * server stack so they will be prepared for normal operations.
628 * \param[in] env execution environment
629 * \param[in] ofd OFD device
631 * \retval 0 if successful
632 * \retval negative value on error
634 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
636 struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
639 CDEBUG(D_HA, "%s: recovery is over\n", ofd_name(ofd));
641 if (!ofd->ofd_skip_lfsck && !ofd->ofd_osd->dd_rdonly) {
642 struct lfsck_start_param lsp;
644 lsp.lsp_start = NULL;
645 lsp.lsp_index_valid = 0;
646 rc = lfsck_start(env, ofd->ofd_osd, &lsp);
647 if (rc != 0 && rc != -EALREADY)
648 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
652 return ldev->ld_ops->ldo_recovery_complete(env, ldev);
656 * Implementation of obd_ops::o_postrecov.
658 * This function is called from target_finish_recovery() upon recovery
661 * \param[in] obd OBD device of OFD
663 * \retval 0 if successful
664 * \retval negative value on error
666 static int ofd_obd_postrecov(struct obd_device *obd)
669 struct lu_device *ldev = obd->obd_lu_dev;
674 rc = lu_env_init(&env, LCT_DT_THREAD);
677 ofd_info_init(&env, obd->obd_self_export);
679 rc = ofd_postrecov(&env, ofd_dev(ldev));
686 * Implementation of obd_ops::o_set_info_async.
688 * This function is not called from request handler, it is only used by
689 * class_notify_sptlrpc_conf() locally by direct obd_set_info_async() call.
690 * \see ofd_set_info_hdl() for request handler function.
692 * \param[in] env execution environment
693 * \param[in] exp OBD export of OFD device
694 * \param[in] keylen length of \a key
695 * \param[in] key key name
696 * \param[in] vallen length of \a val
697 * \param[in] val the \a key value
698 * \param[in] set not used in OFD
700 * \retval 0 if successful
701 * \retval negative value on error
703 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
704 __u32 keylen, void *key, __u32 vallen, void *val,
705 struct ptlrpc_request_set *set)
711 if (exp->exp_obd == NULL) {
712 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
716 if (KEY_IS(KEY_SPTLRPC_CONF)) {
717 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
719 CERROR("%s: Unsupported key %s\n",
720 exp->exp_obd->obd_name, (char*)key);
727 * Implementation of obd_ops::o_get_info.
729 * This function is not called from request handler, it is only used by
730 * direct call from nrs_orr_range_fill_physical() in ptlrpc, see LU-3239.
732 * \see ofd_get_info_hdl() for request handler function.
734 * \param[in] env execution environment
735 * \param[in] exp OBD export of OFD device
736 * \param[in] keylen length of \a key
737 * \param[in] key key name
738 * \param[out] vallen length of key value
739 * \param[out] val the key value to return
741 * \retval 0 if successful
742 * \retval negative value on error
744 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
745 __u32 keylen, void *key, __u32 *vallen, void *val)
747 struct ofd_thread_info *info;
748 struct ofd_device *ofd;
749 struct ll_fiemap_info_key *fm_key = key;
750 struct fiemap *fiemap = val;
755 if (exp->exp_obd == NULL) {
756 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
762 if (KEY_IS(KEY_FIEMAP)) {
763 info = ofd_info_init(env, exp);
765 rc = ostid_to_fid(&info->fti_fid, &fm_key->lfik_oa.o_oi,
766 ofd->ofd_lut.lut_lsd.lsd_osd_index);
770 rc = ofd_fiemap_get(env, ofd, &info->fti_fid, fiemap);
772 CERROR("%s: not supported key %s\n", ofd_name(ofd), (char*)key);
780 * Implementation of obd_ops::o_statfs.
782 * This function returns information about a storage file system.
783 * It is called from several places by using the OBD API as well as
784 * by direct call, e.g. from request handler.
786 * \see ofd_statfs_hdl() for request handler function.
788 * Report also the state of the OST to the caller in osfs->os_state
789 * (OS_STATE_READONLY, OS_STATE_DEGRADED).
791 * \param[in] env execution environment
792 * \param[in] exp OBD export of OFD device
793 * \param[out] osfs statistic data to return
794 * \param[in] max_age maximum age for cached data
795 * \param[in] flags not used in OFD
797 * \retval 0 if successful
798 * \retval negative value on error
800 int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
801 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
803 struct obd_device *obd = class_exp2obd(exp);
804 struct ofd_device *ofd = ofd_exp(exp);
805 struct tg_grants_data *tgd = &ofd->ofd_lut.lut_tgd;
810 rc = tgt_statfs_internal(env, &ofd->ofd_lut, osfs, max_age, NULL);
814 /* at least try to account for cached pages. its still racy and
815 * might be under-reporting if clients haven't announced their
816 * caches with brw recently */
818 CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
819 " pending %llu free %llu avail %llu\n",
820 tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
821 tgd->tgd_tot_pending,
822 osfs->os_bfree << tgd->tgd_blockbits,
823 osfs->os_bavail << tgd->tgd_blockbits);
825 osfs->os_bavail -= min_t(u64, osfs->os_bavail,
826 ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
827 osfs->os_bsize - 1) >> tgd->tgd_blockbits));
829 /* The QoS code on the MDS does not care about space reserved for
830 * precreate, so take it out. */
831 if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
832 struct tg_export_data *ted;
834 ted = &obd->obd_self_export->exp_target_data;
835 osfs->os_bavail -= min_t(u64, osfs->os_bavail,
836 ted->ted_grant >> tgd->tgd_blockbits);
839 tgt_grant_sanity_check(obd, __func__);
840 CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
841 "%llu objects: %llu free; state %x\n",
842 osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
843 osfs->os_files, osfs->os_ffree, osfs->os_state);
845 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
846 ofd->ofd_lut.lut_lsd.lsd_osd_index)) {
847 /* Reduce free inode count to zero, but keep "used" intact */
848 osfs->os_files -= osfs->os_ffree;
849 osfs->os_ffree -= osfs->os_ffree;
852 /* OS_STATE_READONLY can be set by OSD already */
853 if (ofd->ofd_raid_degraded)
854 osfs->os_state |= OS_STATE_DEGRADED;
856 if (obd->obd_self_export != exp && !exp_grant_param_supp(exp) &&
857 tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
858 /* clients which don't support OBD_CONNECT_GRANT_PARAM
859 * should not see a block size > page size, otherwise
860 * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
861 * block size which is the biggest block size known to work
862 * with all client's page size. */
863 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
864 osfs->os_bfree <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
865 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
866 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
869 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
870 ofd->ofd_lut.lut_lsd.lsd_osd_index)) {
871 /* Reduce free blocks count near zero, but keep "used" intact */
872 osfs->os_bavail -= osfs->os_bavail - 2;
873 osfs->os_blocks -= osfs->os_bfree - 2;
874 osfs->os_bfree -= osfs->os_bfree - 2;
883 * Implementation of obd_ops::o_setattr.
885 * This function is only used by ECHO client when it is run on top of OFD,
886 * \see ofd_setattr_hdl() for request handler function.
888 * \param[in] env execution environment
889 * \param[in] exp OBD export of OFD device
890 * \param[in] oa setattr parameters
892 * \retval 0 if successful
893 * \retval negative value on error
895 static int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp,
898 struct ofd_thread_info *info;
899 struct ofd_device *ofd = ofd_exp(exp);
900 struct ldlm_namespace *ns = ofd->ofd_namespace;
901 struct ldlm_resource *res;
902 struct ofd_object *fo;
903 struct lu_fid *fid = &oa->o_oi.oi_fid;
908 info = ofd_info_init(env, exp);
910 ost_fid_build_resid(fid, &info->fti_resid);
912 /* This would be very bad - accidentally truncating a file when
913 * changing the time or similar - bug 12203. */
914 if (oa->o_valid & OBD_MD_FLSIZE) {
915 static char mdsinum[48];
917 if (oa->o_valid & OBD_MD_FLFID)
918 snprintf(mdsinum, sizeof(mdsinum) - 1,
919 "of parent "DFID, oa->o_parent_seq,
920 oa->o_parent_oid, 0);
924 CERROR("%s: setattr from %s trying to truncate object "DFID
925 " %s\n", ofd_name(ofd), obd_export_nid2str(exp),
927 GOTO(out, rc = -EPERM);
930 fo = ofd_object_find_exists(env, ofd, fid);
932 CERROR("%s: can't find object "DFID"\n",
933 ofd_name(ofd), PFID(fid));
934 GOTO(out, rc = PTR_ERR(fo));
937 la_from_obdo(&info->fti_attr, oa, oa->o_valid);
938 info->fti_attr.la_valid &= ~LA_TYPE;
940 /* setting objects attributes (including owner/group) */
941 rc = ofd_attr_set(env, fo, &info->fti_attr, oa);
943 GOTO(out_unlock, rc);
945 ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, NULL, 1);
948 ofd_object_put(env, fo);
951 /* we do not call this before to avoid lu_object_find() in
952 * ->lvbo_update() holding another reference on the object.
953 * otherwise concurrent destroy can make the object unavailable
954 * for 2nd lu_object_find() waiting for the first reference
955 * to go... deadlock! */
956 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
958 ldlm_res_lvbo_update(res, NULL, 0);
959 ldlm_resource_putref(res);
967 * Destroy OFD object by its FID.
969 * Supplemental function to destroy object by FID, it is used by request
970 * handler and by ofd_echo_destroy() below to find object by FID, lock it
971 * and call ofd_destroy() finally.
973 * \param[in] env execution environment
974 * \param[in] ofd OFD device
975 * \param[in] fid FID of object
976 * \param[in] orphan set if object being destroyed is an orphan
978 * \retval 0 if successful
979 * \retval negative value on error
981 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
982 const struct lu_fid *fid, int orphan)
984 struct ofd_thread_info *info = ofd_info(env);
985 struct lustre_handle lockh;
986 union ldlm_policy_data policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
987 struct ofd_object *fo;
988 __u64 flags = LDLM_FL_AST_DISCARD_DATA;
993 fo = ofd_object_find_exists(env, ofd, fid);
997 /* Tell the clients that the object is gone now and that they should
998 * throw away any cached pages. */
999 ost_fid_build_resid(fid, &info->fti_resid);
1000 rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid,
1001 LDLM_EXTENT, &policy, LCK_PW, &flags,
1002 ldlm_blocking_ast, ldlm_completion_ast,
1003 NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1005 /* We only care about the side-effects, just drop the lock. */
1007 ldlm_lock_decref(&lockh, LCK_PW);
1009 LASSERT(fo != NULL);
1011 rc = ofd_destroy(env, fo, orphan);
1014 ofd_object_put(env, fo);
1019 * Implementation of obd_ops::o_destroy.
1021 * This function is only used by ECHO client when it is run on top of OFD,
1022 * \see ofd_destroy_hdl() for request handler function.
1024 * \param[in] env execution environment
1025 * \param[in] exp OBD export of OFD device
1026 * \param[in] oa obdo structure with FID
1028 * Note: this is OBD API method which is common API for server OBDs and
1029 * client OBDs. Thus some parameters used in client OBDs may not be used
1030 * on server OBDs and vice versa.
1032 * \retval 0 if successful
1033 * \retval negative value on error
1035 static int ofd_echo_destroy(const struct lu_env *env, struct obd_export *exp,
1038 struct ofd_device *ofd = ofd_exp(exp);
1039 struct lu_fid *fid = &oa->o_oi.oi_fid;
1044 ofd_info_init(env, exp);
1046 rc = ofd_validate_seq(exp, ostid_seq(&oa->o_oi));
1050 CDEBUG(D_HA, "%s: Destroy object "DFID"\n", ofd_name(ofd), PFID(fid));
1052 rc = ofd_destroy_by_fid(env, ofd, fid, 0);
1053 if (rc == -ENOENT) {
1054 CDEBUG(D_INODE, "%s: destroying non-existent object "DFID"\n",
1055 ofd_name(ofd), PFID(fid));
1057 } else if (rc != 0) {
1058 CERROR("%s: error destroying object "DFID": %d\n",
1059 ofd_name(ofd), PFID(fid), rc);
1068 * Implementation of obd_ops::o_create.
1070 * This function is only used by ECHO client when it is run on top of OFD
1071 * and just creates an object.
1072 * \see ofd_create_hdl() for request handler function.
1074 * \param[in] env execution environment
1075 * \param[in] exp OBD export of OFD device
1076 * \param[in] oa obdo structure with FID sequence to use
1078 * Note: this is OBD API method which is common API for server OBDs and
1079 * client OBDs. Thus some parameters used in client OBDs may not be used
1080 * on server OBDs and vice versa.
1082 * \retval 0 if successful
1083 * \retval negative value on error
1085 static int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
1088 struct ofd_device *ofd = ofd_exp(exp);
1089 u64 seq = ostid_seq(&oa->o_oi);
1090 struct ofd_seq *oseq;
1099 ofd_info_init(env, exp);
1101 LASSERT(seq == FID_SEQ_ECHO);
1102 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1104 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1106 down_read(&ofd->ofd_lastid_rwsem);
1107 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1108 * we may do that in the future.
1109 * Return -ENOSPC until the LAST_ID rebuilt. */
1110 if (unlikely(ofd->ofd_lastid_rebuilding))
1111 GOTO(out_sem, rc = -ENOSPC);
1113 rc = ofd_validate_seq(exp, seq);
1117 oseq = ofd_seq_load(env, ofd, seq);
1119 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1120 ofd_name(ofd), seq, PTR_ERR(oseq));
1121 GOTO(out_sem, rc = -EINVAL);
1124 mutex_lock(&oseq->os_create_lock);
1125 granted = tgt_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
1129 CDEBUG(D_HA, "%s: failed to acquire grant space for "
1130 "precreate (%lld): rc = %d\n", ofd_name(ofd), diff, rc);
1135 next_id = ofd_seq_last_oid(oseq) + 1;
1136 count = ofd_precreate_batch(ofd, (int)diff);
1138 rc = ofd_precreate_objects(env, ofd, next_id, oseq, count, 0);
1140 CERROR("%s: unable to precreate: rc = %d\n",
1143 rc = ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1145 CERROR("%s: Bad %llu to set " DOSTID " : rc %d\n",
1147 (unsigned long long)ofd_seq_last_oid(oseq),
1148 POSTID(&oa->o_oi), rc);
1150 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1153 tgt_grant_commit(ofd_obd(ofd)->obd_self_export, granted, rc);
1155 mutex_unlock(&oseq->os_create_lock);
1156 ofd_seq_put(env, oseq);
1159 up_read(&ofd->ofd_lastid_rwsem);
1164 * Implementation of obd_ops::o_getattr.
1166 * This function is only used by ECHO client when it is run on top of OFD
1167 * and returns attributes of object.
1168 * \see ofd_getattr_hdl() for request handler function.
1170 * \param[in] env execution environment
1171 * \param[in] exp OBD export of OFD device
1172 * \param[in,out] oa contains FID of object to get attributes from and
1173 * is used to return attributes back
1175 * \retval 0 if successful
1176 * \retval negative value on error
1178 static int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp,
1181 struct ofd_device *ofd = ofd_exp(exp);
1182 struct ofd_thread_info *info;
1183 struct lu_fid *fid = &oa->o_oi.oi_fid;
1184 struct ofd_object *fo;
1189 info = ofd_info_init(env, exp);
1191 fo = ofd_object_find_exists(env, ofd, fid);
1193 GOTO(out, rc = PTR_ERR(fo));
1195 LASSERT(fo != NULL);
1196 rc = ofd_attr_get(env, fo, &info->fti_attr);
1197 oa->o_valid = OBD_MD_FLID;
1201 obdo_from_la(oa, &info->fti_attr,
1202 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1204 /* Store object version in reply */
1205 curr_version = dt_version_get(env, ofd_object_child(fo));
1206 if ((__s64)curr_version != -EOPNOTSUPP) {
1207 oa->o_valid |= OBD_MD_FLDATAVERSION;
1208 oa->o_data_version = curr_version;
1212 ofd_object_put(env, fo);
1218 * Get object version for OBD_IOC_GET_OBJ_VERSION ioctl.
1220 * This is supplemental function for ofd_iocontrol() to return object
1221 * version for lctl tool.
1223 * \param[in] env execution environment
1224 * \param[in] ofd OFD device
1225 * \param[out] karg ioctl data
1227 * \retval 0 if successful
1228 * \retval negative value on error
1230 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1231 struct ofd_device *ofd, void *karg)
1233 struct obd_ioctl_data *data = karg;
1235 struct ofd_object *fo;
1236 dt_obj_version_t version;
1241 if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
1242 GOTO(out, rc = -EINVAL);
1244 if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
1245 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1246 } else if (data->ioc_inlbuf3 != NULL &&
1247 data->ioc_inllen3 == sizeof(__u64) &&
1248 data->ioc_inlbuf4 != NULL &&
1249 data->ioc_inllen4 == sizeof(__u64)) {
1250 struct ost_id ostid = { };
1252 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1253 rc = ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1256 rc = ostid_to_fid(&fid, &ostid,
1257 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1261 GOTO(out, rc = -EINVAL);
1264 if (!fid_is_sane(&fid))
1265 GOTO(out, rc = -EINVAL);
1267 fo = ofd_object_find(env, ofd, &fid);
1269 GOTO(out, rc = PTR_ERR(fo));
1271 if (!ofd_object_exists(fo))
1272 GOTO(out_fo, rc = -ENOENT);
1274 if (lu_object_remote(&fo->ofo_obj.do_lu))
1275 GOTO(out_fo, rc = -EREMOTE);
1277 version = dt_version_get(env, ofd_object_child(fo));
1279 GOTO(out_fo, rc = -EIO);
1281 *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1285 ofd_object_put(env, fo);
1291 * Implementation of obd_ops::o_iocontrol.
1293 * This is OFD ioctl handling function which is primary interface for
1294 * Lustre tools like lfs, lctl and lfsck.
1296 * \param[in] cmd ioctl command
1297 * \param[in] exp OBD export of OFD
1298 * \param[in] len not used
1299 * \param[in,out] karg buffer with data
1300 * \param[in] uarg not used
1302 * \retval 0 if successful
1303 * \retval negative value on error
1305 static int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1306 void *karg, void __user *uarg)
1309 struct ofd_device *ofd = ofd_exp(exp);
1310 struct obd_device *obd = ofd_obd(ofd);
1315 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1316 rc = lu_env_init(&env, LCT_DT_THREAD);
1321 case OBD_IOC_ABORT_RECOVERY:
1322 CERROR("%s: aborting recovery\n", obd->obd_name);
1323 obd->obd_abort_recovery = 1;
1324 target_stop_recovery_thread(obd);
1327 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1328 rc = dt_sync(&env, ofd->ofd_osd);
1330 case OBD_IOC_SET_READONLY:
1331 rc = dt_sync(&env, ofd->ofd_osd);
1333 rc = dt_ro(&env, ofd->ofd_osd);
1335 case OBD_IOC_START_LFSCK: {
1336 struct obd_ioctl_data *data = karg;
1337 struct lfsck_start_param lsp;
1339 if (unlikely(data == NULL)) {
1344 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
1345 lsp.lsp_index_valid = 0;
1346 rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
1349 case OBD_IOC_STOP_LFSCK: {
1350 struct obd_ioctl_data *data = karg;
1351 struct lfsck_stop stop;
1353 stop.ls_status = LS_STOPPED;
1354 /* Old lfsck utils may pass NULL @stop. */
1355 if (data->ioc_inlbuf1 == NULL)
1359 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
1361 rc = lfsck_stop(&env, ofd->ofd_osd, &stop);
1364 case OBD_IOC_GET_OBJ_VERSION:
1365 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1368 CERROR("%s: not supported cmd = %#x\n", obd->obd_name, cmd);
1377 * Implementation of obd_ops::o_precleanup.
1379 * This function stops device activity before shutting it down. It is called
1380 * from a cleanup function upon forceful device cleanup. For OFD there are no
1381 * special actions, it just invokes target_recovery_cleanup().
1383 * \param[in] obd OBD device of OFD
1387 static int ofd_precleanup(struct obd_device *obd)
1390 target_cleanup_recovery(obd);
1395 * Implementation of obd_ops::o_ping.
1397 * This is OFD-specific part of OBD_PING request handling.
1398 * It controls Filter Modification Data (FMD) expiration each time PING is
1401 * \see ofd_fmd_expire() and ofd_fmd.c for details
1403 * \param[in] env execution environment
1404 * \param[in] exp OBD export of client
1408 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1410 ofd_fmd_expire(exp);
1415 * Implementation of obd_ops::o_health_check.
1417 * This function checks the OFD device health - ability to respond on
1418 * incoming requests. There are two health_check methods:
1419 * - get statfs from the OSD. It checks just responsiveness of
1421 * - do write attempt on bottom device to check it is fully operational and
1422 * is not stuck. This is expensive method and requires special configuration
1423 * option --enable-health-write while building Lustre, it is turned off
1426 * \param[in] nul not used
1427 * \param[in] obd OBD device of OFD
1429 * \retval 0 if successful
1430 * \retval negative value in case of error
1432 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1434 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1435 struct ofd_thread_info *info;
1437 #ifdef USE_HEALTH_CHECK_WRITE
1442 /* obd_proc_read_health pass NULL env, we need real one */
1443 rc = lu_env_init(&env, LCT_DT_THREAD);
1447 info = ofd_info_init(&env, NULL);
1448 rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1452 if (info->fti_u.osfs.os_state & OS_STATE_READONLY)
1453 GOTO(out, rc = -EROFS);
1455 #ifdef USE_HEALTH_CHECK_WRITE
1456 OBD_ALLOC(info->fti_buf.lb_buf, PAGE_SIZE);
1457 if (info->fti_buf.lb_buf == NULL)
1458 GOTO(out, rc = -ENOMEM);
1460 info->fti_buf.lb_len = PAGE_SIZE;
1463 th = dt_trans_create(&env, ofd->ofd_osd);
1465 GOTO(out, rc = PTR_ERR(th));
1467 rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1468 &info->fti_buf, info->fti_off, th);
1470 th->th_sync = 1; /* sync IO is needed */
1471 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1473 rc = dt_record_write(&env, ofd->ofd_health_check_file,
1474 &info->fti_buf, &info->fti_off,
1477 dt_trans_stop(&env, ofd->ofd_osd, th);
1479 OBD_FREE(info->fti_buf.lb_buf, PAGE_SIZE);
1481 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1488 struct obd_ops ofd_obd_ops = {
1489 .o_owner = THIS_MODULE,
1490 .o_connect = ofd_obd_connect,
1491 .o_reconnect = ofd_obd_reconnect,
1492 .o_disconnect = ofd_obd_disconnect,
1493 .o_create = ofd_echo_create,
1494 .o_statfs = ofd_statfs,
1495 .o_setattr = ofd_echo_setattr,
1496 .o_preprw = ofd_preprw,
1497 .o_commitrw = ofd_commitrw,
1498 .o_destroy = ofd_echo_destroy,
1499 .o_init_export = ofd_init_export,
1500 .o_destroy_export = ofd_destroy_export,
1501 .o_postrecov = ofd_obd_postrecov,
1502 .o_getattr = ofd_echo_getattr,
1503 .o_iocontrol = ofd_iocontrol,
1504 .o_precleanup = ofd_precleanup,
1506 .o_health_check = ofd_health_check,
1507 .o_set_info_async = ofd_set_info_async,
1508 .o_get_info = ofd_get_info,