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;
76 if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
77 /* Self-export gets no proc entry */
80 rc = lprocfs_exp_setup(exp, client_nid);
82 /* Mask error for already created /proc entries */
83 RETURN(rc == -EALREADY ? 0 : rc);
85 stats = exp->exp_nid_stats;
86 stats->nid_stats = lprocfs_alloc_stats(LPROC_OFD_STATS_LAST,
87 LPROCFS_STATS_FLAG_NOPERCPU);
88 if (!stats->nid_stats)
91 ofd_stats_counter_init(stats->nid_stats);
93 rc = lprocfs_register_stats(stats->nid_proc, "stats", stats->nid_stats);
95 lprocfs_free_stats(&stats->nid_stats);
99 rc = lprocfs_nid_ldlm_stats_init(stats);
108 * Decide which checksums both client and OST support, possibly forcing
109 * the use of T10PI checksums if the hardware supports this.
111 * The clients that have no T10-PI RPC checksum support will use the same
112 * mechanism to select checksum type as before, and will not be affected by
113 * the following logic.
115 * For the clients that have T10-PI RPC checksum support:
117 * If the OST supports T10-PI feature and T10-PI checksum is enforced, clients
118 * will have no other choice for RPC checksum type other than using the T10PI
119 * checksum type. This is useful for enforcing end-to-end integrity in the
122 * If the OST doesn't support T10-PI feature and T10-PI checksum is enforced,
123 * together with other checksum with reasonably good speeds (e.g. crc32,
124 * crc32c, adler, etc.), all T10-PI checksum types understood by the client
125 * (t10ip512, t10ip4K, t10crc512, t10crc4K) will be added to the available
126 * checksum types, regardless of the speeds of T10-PI checksums. This is
127 * useful for testing T10-PI checksum of RPC.
129 * If the OST supports T10-PI feature and T10-PI checksum is NOT enforced,
130 * the corresponding T10-PI checksum type will be added to the checksum type
131 * list, regardless of the speed of the T10-PI checksum. This provides clients
132 * the flexibility to choose whether to enable end-to-end integrity or not.
134 * If the OST does NOT supports T10-PI feature and T10-PI checksum is NOT
135 * enforced, together with other checksums with reasonably good speeds,
136 * all the T10-PI checksum types with good speeds will be added into the
137 * checksum type list. Note that a T10-PI checksum type with a speed worse
138 * than half of Alder will NOT be added as a option. In this circumstance,
139 * T10-PI checksum types has the same behavior like other normal checksum
144 ofd_mask_cksum_types(struct ofd_device *ofd, enum cksum_types *cksum_types)
146 bool enforce = ofd->ofd_checksum_t10pi_enforce;
147 enum cksum_types ofd_t10_cksum_type;
148 enum cksum_types client_t10_types = *cksum_types & OBD_CKSUM_T10_ALL;
149 enum cksum_types server_t10_types;
152 * The client set in ocd_cksum_types the checksum types it
153 * supports. We have to mask off the algorithms that we don't
154 * support. T10PI checksum types will be added later.
156 *cksum_types &= (ofd->ofd_cksum_types_supported & ~OBD_CKSUM_T10_ALL);
157 server_t10_types = ofd->ofd_cksum_types_supported & OBD_CKSUM_T10_ALL;
158 ofd_t10_cksum_type = ofd->ofd_lut.lut_dt_conf.ddp_t10_cksum_type;
160 /* Quick exit if no T10-PI support on client */
161 if (!client_t10_types)
165 * This OST has NO T10-PI feature. Add all supported T10-PI checksums
166 * as options if T10-PI checksum is enforced. If the T10-PI checksum is
167 * not enforced, only add them as options when speed is good.
169 if (ofd_t10_cksum_type == 0) {
171 * Server allows all T10PI checksums, and server_t10_types
172 * include quick ones.
175 *cksum_types |= client_t10_types;
177 *cksum_types |= client_t10_types & server_t10_types;
182 * This OST has T10-PI feature. Disable all other checksum types if
183 * T10-PI checksum is enforced. If the T10-PI checksum is not enforced,
184 * add the checksum type as an option.
186 if (client_t10_types & ofd_t10_cksum_type) {
188 *cksum_types = ofd_t10_cksum_type;
190 *cksum_types |= ofd_t10_cksum_type;
195 * Match client and OST server connection feature flags.
197 * Compute the compatibility flags for a connection request based on
198 * features mutually supported by client and server.
200 * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
201 * must not be updated here, otherwise a partially initialized value may
202 * be exposed. After the connection request is successfully processed,
203 * the top-level tgt_connect() request handler atomically updates the export
204 * connect flags from the obd_connect_data::ocd_connect_flags field of the
205 * reply. \see tgt_connect().
207 * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
208 * full struct obd_connect_data. So care must be taken when accessing fields
209 * that are not present in struct obd_connect_data_v1. See LU-16.
211 * \param[in] env execution environment
212 * \param[in] exp the obd_export associated with this
214 * \param[in] data stores data for this connect request
215 * \param[in] new_connection is this connection new or not
217 * \retval 0 if success
218 * \retval -EPROTO client and server feature requirements are
220 * \retval -EBADF OST index in connect request doesn't match
223 static int ofd_parse_connect_data(const struct lu_env *env,
224 struct obd_export *exp,
225 struct obd_connect_data *data,
228 struct ofd_device *ofd = ofd_exp(exp);
229 struct filter_export_data *fed = &exp->exp_filter_data;
235 "%s: cli %s/%p ocd_connect_flags: %#llx ocd_version: %x ocd_grant: %d ocd_index: %u ocd_group %u\n",
236 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
237 data->ocd_connect_flags, data->ocd_version,
238 data->ocd_grant, data->ocd_index, data->ocd_group);
240 if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) {
241 CWARN("!!! This export (nid %s) used object group %d earlier; now it's trying to use group %d! This could be a bug in the MDS. Please report to https://jira.whamcloud.com/\n",
242 obd_export_nid2str(exp), fed->fed_group,
246 fed->fed_group = data->ocd_group;
248 data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
250 if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
251 data->ocd_connect_flags2 &= OST_CONNECT_SUPPORTED2;
253 /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
254 if (data->ocd_connect_flags & OBD_CONNECT_MDS)
255 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
256 exp->exp_obd->obd_name, data->ocd_group);
257 else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
260 /* Determine optimal brw size before calculating grant */
261 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
262 data->ocd_brw_size = 65536;
263 } else if (OCD_HAS_FLAG(data, BRW_SIZE)) {
264 if (data->ocd_brw_size > ofd->ofd_brw_size)
265 data->ocd_brw_size = ofd->ofd_brw_size;
266 if (data->ocd_brw_size == 0) {
267 CERROR("%s: cli %s/%p ocd_connect_flags: %#llx ocd_version: %x ocd_grant: %d ocd_index: %u ocd_brw_size is unexpectedly zero, network data corruption? Refusing connection of this client\n",
268 exp->exp_obd->obd_name,
269 exp->exp_client_uuid.uuid,
270 exp, data->ocd_connect_flags, data->ocd_version,
271 data->ocd_grant, data->ocd_index);
276 if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
277 struct dt_device_param *ddp = &ofd->ofd_lut.lut_dt_conf;
279 /* client is reporting its page size, for future use */
280 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
281 data->ocd_grant_blkbits = ofd->ofd_lut.lut_tgd.tgd_blockbits;
283 * ddp_inodespace may not be power-of-two value, eg. for ldiskfs
284 * it's LDISKFS_DIR_REC_LEN(20) = 28.
286 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
287 /* ocd_grant_tax_kb is in 1K byte blocks */
288 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
289 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
293 * Save connect_data we have so far because tgt_grant_connect()
294 * uses it to calculate grant, and we want to save the client
295 * version before it is overwritten by LUSTRE_VERSION_CODE.
297 exp->exp_connect_data = *data;
298 if (OCD_HAS_FLAG(data, GRANT))
299 tgt_grant_connect(env, exp, data, new_connection);
301 if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
302 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
303 int index = lsd->lsd_osd_index;
305 if (index != data->ocd_index) {
306 LCONSOLE_ERROR_MSG(0x136,
307 "Connection from %s to index %u doesn't match actual OST index %u in last_rcvd file, bad configuration?\n",
308 obd_export_nid2str(exp), index,
312 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
313 /* this will only happen on the first connect */
314 lsd->lsd_feature_compat |= OBD_COMPAT_OST;
316 * sync is not needed here as tgt_client_new will
317 * set exp_need_sync flag
319 tgt_server_data_update(env, &ofd->ofd_lut, 0);
323 if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
324 __u32 cksum_types = data->ocd_cksum_types;
326 ofd_mask_cksum_types(ofd, &data->ocd_cksum_types);
328 if (unlikely(data->ocd_cksum_types == 0)) {
329 CERROR("%s: Connect with checksum support but no ocd_cksum_types is set\n",
330 exp->exp_obd->obd_name);
335 "%s: cli %s supports cksum type %x, return %x\n",
336 exp->exp_obd->obd_name, obd_export_nid2str(exp),
337 cksum_types, data->ocd_cksum_types);
340 * This client does not support OBD_CONNECT_CKSUM.
341 * Report failure to negotiate checksum at connect
344 "%s: cli %s does not support OBD_CONNECT_CKSUM\n",
345 exp->exp_obd->obd_name, obd_export_nid2str(exp));
348 if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
349 data->ocd_maxbytes = ofd->ofd_lut.lut_dt_conf.ddp_maxbytes;
351 data->ocd_version = LUSTRE_VERSION_CODE;
353 if (OCD_HAS_FLAG(data, PINGLESS)) {
354 if (ptlrpc_pinger_suppress_pings()) {
355 spin_lock(&exp->exp_obd->obd_dev_lock);
356 list_del_init(&exp->exp_obd_chain_timed);
357 spin_unlock(&exp->exp_obd->obd_dev_lock);
359 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
367 * Re-initialize export upon client reconnection.
369 * This function parses connection data from reconnect and resets
372 * \param[in] env execution environment
373 * \param[in] exp OBD export
374 * \param[in] obd OFD device
375 * \param[in] cluuid NID of client
376 * \param[in] data connection data from request
377 * \param[in] localdata client NID
379 * \retval 0 if successful
380 * \retval negative value on error
382 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
383 struct obd_device *obd, struct obd_uuid *cluuid,
384 struct obd_connect_data *data,
387 struct ofd_device *ofd;
392 if (!exp || !obd || !cluuid)
395 rc = nodemap_add_member(*(lnet_nid_t *)client_nid, exp);
396 if (rc != 0 && rc != -EEXIST)
399 ofd = ofd_dev(obd->obd_lu_dev);
401 rc = ofd_parse_connect_data(env, exp, data, false);
403 ofd_export_stats_init(ofd, exp, client_nid);
405 nodemap_del_member(exp);
411 * Initialize new client connection.
413 * This function handles new connection to the OFD. The new export is
414 * created (in context of class_connect()) and persistent client data is
415 * initialized on storage.
417 * \param[in] env execution environment
418 * \param[out] _exp stores pointer to new export
419 * \param[in] obd OFD device
420 * \param[in] cluuid client UUID
421 * \param[in] data connection data from request
422 * \param[in] localdata client NID
424 * \retval 0 if successful
425 * \retval negative value on error
427 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
428 struct obd_device *obd, struct obd_uuid *cluuid,
429 struct obd_connect_data *data, void *localdata)
431 struct obd_export *exp;
432 struct ofd_device *ofd;
433 struct lustre_handle conn = { 0 };
438 if (!_exp || !obd || !cluuid)
441 ofd = ofd_dev(obd->obd_lu_dev);
443 rc = class_connect(&conn, obd, cluuid);
447 exp = class_conn2export(&conn);
448 LASSERT(exp != NULL);
451 rc = nodemap_add_member(*(lnet_nid_t *)localdata, exp);
452 if (rc != 0 && rc != -EEXIST)
456 "%s: cannot find nodemap for client %s: nid is null\n",
457 obd->obd_name, cluuid->uuid);
460 rc = ofd_parse_connect_data(env, exp, data, true);
464 if (obd->obd_replayable) {
465 struct tg_export_data *ted = &exp->exp_target_data;
467 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
468 sizeof(ted->ted_lcd->lcd_uuid));
469 rc = tgt_client_new(env, exp);
472 ofd_export_stats_init(ofd, exp, localdata);
475 CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
476 data ? data->ocd_group : -1);
480 class_disconnect(exp);
481 nodemap_del_member(exp);
490 * Disconnect a connected client.
492 * This function terminates the client connection. The client export is
493 * disconnected (cleaned up) and client data on persistent storage is removed.
495 * \param[in] exp OBD export
497 * \retval 0 if successful
498 * \retval negative value on error
500 int ofd_obd_disconnect(struct obd_export *exp)
502 struct ofd_device *ofd = ofd_exp(exp);
509 class_export_get(exp);
511 if (!(exp->exp_flags & OBD_OPT_FORCE))
512 tgt_grant_sanity_check(ofd_obd(ofd), __func__);
514 rc = server_disconnect_export(exp);
516 tgt_grant_discard(exp);
518 /* Do not erase record for recoverable client. */
519 if (exp->exp_obd->obd_replayable &&
520 (!exp->exp_obd->obd_fail || exp->exp_failed)) {
521 rc = lu_env_init(&env, LCT_DT_THREAD);
525 tgt_client_del(&env, exp);
529 nodemap_del_member(exp);
530 class_export_put(exp);
535 * Implementation of obd_ops::o_init_export.
537 * This function is called from class_new_export() and initializes
538 * the OFD-specific data for new export.
540 * \param[in] exp OBD export
542 * \retval 0 if successful
543 * \retval negative value on error
545 static int ofd_init_export(struct obd_export *exp)
549 spin_lock_init(&exp->exp_target_data.ted_fmd_lock);
550 INIT_LIST_HEAD(&exp->exp_target_data.ted_fmd_list);
551 atomic_set(&exp->exp_filter_data.fed_soft_sync_count, 0);
552 spin_lock(&exp->exp_lock);
553 exp->exp_connecting = 1;
554 spin_unlock(&exp->exp_lock);
556 /* self-export doesn't need client data and ldlm initialization */
557 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
558 &exp->exp_client_uuid)))
561 rc = tgt_client_alloc(exp);
563 ldlm_init_export(exp);
565 CERROR("%s: Can't initialize export: rc %d\n",
566 exp->exp_obd->obd_name, rc);
571 * Implementation of obd_ops::o_destroy_export.
573 * This function is called from class_export_destroy() to cleanup
574 * the OFD-specific data for export being destroyed.
576 * \param[in] exp OBD export
578 * \retval 0 if successful
579 * \retval negative value on error
581 static int ofd_destroy_export(struct obd_export *exp)
583 struct ofd_device *ofd = ofd_exp(exp);
585 if (exp->exp_target_data.ted_pending)
586 CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
587 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
588 exp, exp->exp_target_data.ted_pending);
590 target_destroy_export(exp);
592 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
593 &exp->exp_client_uuid)))
596 ldlm_destroy_export(exp);
597 tgt_client_free(exp);
599 ofd_fmd_cleanup(exp);
602 * discard grants once we're sure no more
603 * interaction with the client is possible
605 tgt_grant_discard(exp);
607 if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
608 ofd->ofd_lut.lut_tgd.tgd_tot_granted_clients--;
610 if (!(exp->exp_flags & OBD_OPT_FORCE))
611 tgt_grant_sanity_check(exp->exp_obd, __func__);
617 * Notify all devices in server stack about recovery completion.
619 * This function calls ldo_recovery_complete() for all lower devices in the
620 * server stack so they will be prepared for normal operations.
622 * \param[in] env execution environment
623 * \param[in] ofd OFD device
625 * \retval 0 if successful
626 * \retval negative value on error
628 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
630 struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
633 CDEBUG(D_HA, "%s: recovery is over\n", ofd_name(ofd));
635 if (!ofd->ofd_skip_lfsck && !ofd->ofd_osd->dd_rdonly) {
636 struct lfsck_start_param lsp;
638 lsp.lsp_start = NULL;
639 lsp.lsp_index_valid = 0;
640 rc = lfsck_start(env, ofd->ofd_osd, &lsp);
641 if (rc != 0 && rc != -EALREADY)
642 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
646 return ldev->ld_ops->ldo_recovery_complete(env, ldev);
650 * Implementation of obd_ops::o_postrecov.
652 * This function is called from target_finish_recovery() upon recovery
655 * \param[in] obd OBD device of OFD
657 * \retval 0 if successful
658 * \retval negative value on error
660 static int ofd_obd_postrecov(struct obd_device *obd)
663 struct lu_device *ldev = obd->obd_lu_dev;
668 rc = lu_env_init(&env, LCT_DT_THREAD);
671 ofd_info_init(&env, obd->obd_self_export);
673 rc = ofd_postrecov(&env, ofd_dev(ldev));
680 * Implementation of obd_ops::o_set_info_async.
682 * This function is not called from request handler, it is only used by
683 * class_notify_sptlrpc_conf() locally by direct obd_set_info_async() call.
684 * \see ofd_set_info_hdl() for request handler function.
686 * \param[in] env execution environment
687 * \param[in] exp OBD export of OFD device
688 * \param[in] keylen length of \a key
689 * \param[in] key key name
690 * \param[in] vallen length of \a val
691 * \param[in] val the \a key value
692 * \param[in] set not used in OFD
694 * \retval 0 if successful
695 * \retval negative value on error
697 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
698 __u32 keylen, void *key, __u32 vallen, void *val,
699 struct ptlrpc_request_set *set)
706 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
710 if (KEY_IS(KEY_SPTLRPC_CONF)) {
711 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
713 CERROR("%s: Unsupported key %s\n",
714 exp->exp_obd->obd_name, (char *)key);
721 * Implementation of obd_ops::o_get_info.
723 * This function is not called from request handler, it is only used by
724 * direct call from nrs_orr_range_fill_physical() in ptlrpc, see LU-3239.
726 * \see ofd_get_info_hdl() for request handler function.
728 * \param[in] env execution environment
729 * \param[in] exp OBD export of OFD device
730 * \param[in] keylen length of \a key
731 * \param[in] key key name
732 * \param[out] vallen length of key value
733 * \param[out] val the key value to return
735 * \retval 0 if successful
736 * \retval negative value on error
738 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
739 __u32 keylen, void *key, __u32 *vallen, void *val)
741 struct ofd_thread_info *info;
742 struct ofd_device *ofd;
743 struct ll_fiemap_info_key *fm_key = key;
744 struct fiemap *fiemap = val;
750 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
756 if (KEY_IS(KEY_FIEMAP)) {
757 info = ofd_info_init(env, exp);
759 rc = ostid_to_fid(&info->fti_fid, &fm_key->lfik_oa.o_oi,
760 ofd->ofd_lut.lut_lsd.lsd_osd_index);
764 rc = ofd_fiemap_get(env, ofd, &info->fti_fid, fiemap);
766 CERROR("%s: not supported key %s\n",
767 ofd_name(ofd), (char *)key);
775 * Implementation of obd_ops::o_statfs.
777 * This function returns information about a storage file system.
778 * It is called from several places by using the OBD API as well as
779 * by direct call, e.g. from request handler.
781 * \see ofd_statfs_hdl() for request handler function.
783 * Report also the state of the OST to the caller in osfs->os_state
784 * (OS_STATE_READONLY, OS_STATE_DEGRADED).
786 * \param[in] env execution environment
787 * \param[in] exp OBD export of OFD device
788 * \param[out] osfs statistic data to return
789 * \param[in] max_age maximum age for cached data
790 * \param[in] flags not used in OFD
792 * \retval 0 if successful
793 * \retval negative value on error
795 int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
796 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
798 struct obd_device *obd = class_exp2obd(exp);
799 struct ofd_device *ofd = ofd_exp(exp);
800 struct tg_grants_data *tgd = &ofd->ofd_lut.lut_tgd;
805 rc = tgt_statfs_internal(env, &ofd->ofd_lut, osfs, max_age, NULL);
810 * at least try to account for cached pages. its still racy and
811 * might be under-reporting if clients haven't announced their
812 * caches with brw recently
815 CDEBUG(D_SUPER | D_CACHE,
816 "blocks cached %llu granted %llu pending %llu free %llu avail %llu\n",
817 tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
818 tgd->tgd_tot_pending,
819 osfs->os_bfree << tgd->tgd_blockbits,
820 osfs->os_bavail << tgd->tgd_blockbits);
822 osfs->os_bavail -= min_t(u64, osfs->os_bavail,
823 ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
824 osfs->os_bsize - 1) >> tgd->tgd_blockbits));
827 * The QoS code on the MDS does not care about space reserved for
828 * precreate, so take it out.
830 if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
831 struct tg_export_data *ted;
833 ted = &obd->obd_self_export->exp_target_data;
834 osfs->os_granted = min_t(u64, osfs->os_bavail,
835 ted->ted_grant >> tgd->tgd_blockbits);
836 osfs->os_bavail -= osfs->os_granted;
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) {
859 * clients which don't support OBD_CONNECT_GRANT_PARAM
860 * should not see a block size > page size, otherwise
861 * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
862 * block size which is the biggest block size known to work
863 * with all client's page size.
865 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
866 osfs->os_bfree <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
867 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
868 osfs->os_granted <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
869 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
872 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
873 ofd->ofd_lut.lut_lsd.lsd_osd_index)) {
874 /* Reduce free blocks count near zero, but keep "used" intact */
875 osfs->os_bavail -= osfs->os_bavail - 2;
876 osfs->os_blocks -= osfs->os_bfree - 2;
877 osfs->os_bfree -= osfs->os_bfree - 2;
886 * Implementation of obd_ops::o_setattr.
888 * This function is only used by ECHO client when it is run on top of OFD,
889 * \see ofd_setattr_hdl() for request handler function.
891 * \param[in] env execution environment
892 * \param[in] exp OBD export of OFD device
893 * \param[in] oa setattr parameters
895 * \retval 0 if successful
896 * \retval negative value on error
898 static int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp,
901 struct ofd_thread_info *info;
902 struct ofd_device *ofd = ofd_exp(exp);
903 struct ldlm_namespace *ns = ofd->ofd_namespace;
904 struct ldlm_resource *res;
905 struct ofd_object *fo;
906 struct lu_fid *fid = &oa->o_oi.oi_fid;
911 info = ofd_info_init(env, exp);
913 ost_fid_build_resid(fid, &info->fti_resid);
916 * This would be very bad - accidentally truncating a file when
917 * changing the time or similar - bug 12203.
919 if (oa->o_valid & OBD_MD_FLSIZE) {
920 static char mdsinum[48];
922 if (oa->o_valid & OBD_MD_FLFID)
923 snprintf(mdsinum, sizeof(mdsinum) - 1,
924 "of parent "DFID, oa->o_parent_seq,
925 oa->o_parent_oid, 0);
929 CERROR("%s: setattr from %s trying to truncate object "DFID
930 " %s\n", ofd_name(ofd), obd_export_nid2str(exp),
932 GOTO(out, rc = -EPERM);
935 fo = ofd_object_find_exists(env, ofd, fid);
937 CERROR("%s: can't find object "DFID"\n",
938 ofd_name(ofd), PFID(fid));
939 GOTO(out, rc = PTR_ERR(fo));
942 la_from_obdo(&info->fti_attr, oa, oa->o_valid);
943 info->fti_attr.la_valid &= ~LA_TYPE;
945 /* setting objects attributes (including owner/group) */
946 rc = ofd_attr_set(env, fo, &info->fti_attr, oa);
948 GOTO(out_unlock, rc);
950 ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, NULL, 1);
953 ofd_object_put(env, fo);
957 * we do not call this before to avoid lu_object_find() in
958 * ->lvbo_update() holding another reference on the object.
959 * otherwise concurrent destroy can make the object unavailable
960 * for 2nd lu_object_find() waiting for the first reference
963 res = ldlm_resource_get(ns, NULL, &info->fti_resid,
966 ldlm_res_lvbo_update(env, res, NULL, 0);
967 ldlm_resource_putref(res);
975 * Destroy OFD object by its FID.
977 * Supplemental function to destroy object by FID, it is used by request
978 * handler and by ofd_echo_destroy() below to find object by FID, lock it
979 * and call ofd_destroy() finally.
981 * \param[in] env execution environment
982 * \param[in] ofd OFD device
983 * \param[in] fid FID of object
984 * \param[in] orphan set if object being destroyed is an orphan
986 * \retval 0 if successful
987 * \retval negative value on error
989 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
990 const struct lu_fid *fid, int orphan)
992 struct ofd_thread_info *info = ofd_info(env);
993 struct lustre_handle lockh;
994 union ldlm_policy_data policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
995 struct ofd_object *fo;
996 __u64 flags = LDLM_FL_AST_DISCARD_DATA;
1001 fo = ofd_object_find_exists(env, ofd, fid);
1003 RETURN(PTR_ERR(fo));
1006 * Tell the clients that the object is gone now and that they should
1007 * throw away any cached pages.
1009 ost_fid_build_resid(fid, &info->fti_resid);
1010 rc = ldlm_cli_enqueue_local(env, ofd->ofd_namespace, &info->fti_resid,
1011 LDLM_EXTENT, &policy, LCK_PW, &flags,
1012 ldlm_blocking_ast, ldlm_completion_ast,
1013 NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1015 /* We only care about the side-effects, just drop the lock. */
1017 ldlm_lock_decref(&lockh, LCK_PW);
1019 LASSERT(fo != NULL);
1021 rc = ofd_destroy(env, fo, orphan);
1024 ofd_object_put(env, fo);
1029 * Implementation of obd_ops::o_destroy.
1031 * This function is only used by ECHO client when it is run on top of OFD,
1032 * \see ofd_destroy_hdl() for request handler function.
1034 * \param[in] env execution environment
1035 * \param[in] exp OBD export of OFD device
1036 * \param[in] oa obdo structure with FID
1038 * Note: this is OBD API method which is common API for server OBDs and
1039 * client OBDs. Thus some parameters used in client OBDs may not be used
1040 * on server OBDs and vice versa.
1042 * \retval 0 if successful
1043 * \retval negative value on error
1045 static int ofd_echo_destroy(const struct lu_env *env, struct obd_export *exp,
1048 struct ofd_device *ofd = ofd_exp(exp);
1049 struct lu_fid *fid = &oa->o_oi.oi_fid;
1054 ofd_info_init(env, exp);
1056 rc = ofd_validate_seq(exp, ostid_seq(&oa->o_oi));
1060 CDEBUG(D_HA, "%s: Destroy object "DFID"\n", ofd_name(ofd), PFID(fid));
1062 rc = ofd_destroy_by_fid(env, ofd, fid, 0);
1063 if (rc == -ENOENT) {
1064 CDEBUG(D_INODE, "%s: destroying non-existent object "DFID"\n",
1065 ofd_name(ofd), PFID(fid));
1067 } else if (rc != 0) {
1068 CERROR("%s: error destroying object "DFID": %d\n",
1069 ofd_name(ofd), PFID(fid), rc);
1078 * Implementation of obd_ops::o_create.
1080 * This function is only used by ECHO client when it is run on top of OFD
1081 * and just creates an object.
1082 * \see ofd_create_hdl() for request handler function.
1084 * \param[in] env execution environment
1085 * \param[in] exp OBD export of OFD device
1086 * \param[in] oa obdo structure with FID sequence to use
1088 * Note: this is OBD API method which is common API for server OBDs and
1089 * client OBDs. Thus some parameters used in client OBDs may not be used
1090 * on server OBDs and vice versa.
1092 * \retval 0 if successful
1093 * \retval negative value on error
1095 static int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
1098 struct ofd_device *ofd = ofd_exp(exp);
1099 u64 seq = ostid_seq(&oa->o_oi);
1100 struct ofd_seq *oseq;
1109 ofd_info_init(env, exp);
1111 LASSERT(seq == FID_SEQ_ECHO);
1112 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1114 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1116 down_read(&ofd->ofd_lastid_rwsem);
1118 * Currently, for safe, we do not distinguish which LAST_ID is broken,
1119 * we may do that in the future.
1120 * Return -ENOSPC until the LAST_ID rebuilt.
1122 if (unlikely(ofd->ofd_lastid_rebuilding))
1123 GOTO(out_sem, rc = -ENOSPC);
1125 rc = ofd_validate_seq(exp, seq);
1129 oseq = ofd_seq_load(env, ofd, seq);
1131 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1132 ofd_name(ofd), seq, PTR_ERR(oseq));
1133 GOTO(out_sem, rc = -EINVAL);
1136 mutex_lock(&oseq->os_create_lock);
1137 granted = tgt_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
1142 "%s: failed to acquire grant space for precreate (%lld): rc = %d\n",
1143 ofd_name(ofd), diff, rc);
1148 next_id = ofd_seq_last_oid(oseq) + 1;
1149 count = ofd_precreate_batch(ofd, (int)diff);
1151 rc = ofd_precreate_objects(env, ofd, next_id, oseq, count, 0);
1153 CERROR("%s: unable to precreate: rc = %d\n",
1156 rc = ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1158 CERROR("%s: Bad %llu to set " DOSTID " : rc %d\n",
1160 (unsigned long long)ofd_seq_last_oid(oseq),
1161 POSTID(&oa->o_oi), rc);
1163 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1166 tgt_grant_commit(ofd_obd(ofd)->obd_self_export, granted, rc);
1168 mutex_unlock(&oseq->os_create_lock);
1169 ofd_seq_put(env, oseq);
1172 up_read(&ofd->ofd_lastid_rwsem);
1177 * Implementation of obd_ops::o_getattr.
1179 * This function is only used by ECHO client when it is run on top of OFD
1180 * and returns attributes of object.
1181 * \see ofd_getattr_hdl() for request handler function.
1183 * \param[in] env execution environment
1184 * \param[in] exp OBD export of OFD device
1185 * \param[in,out] oa contains FID of object to get attributes from and
1186 * is used to return attributes back
1188 * \retval 0 if successful
1189 * \retval negative value on error
1191 static int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp,
1194 struct ofd_device *ofd = ofd_exp(exp);
1195 struct ofd_thread_info *info;
1196 struct lu_fid *fid = &oa->o_oi.oi_fid;
1197 struct ofd_object *fo;
1202 info = ofd_info_init(env, exp);
1204 fo = ofd_object_find_exists(env, ofd, fid);
1206 GOTO(out, rc = PTR_ERR(fo));
1208 LASSERT(fo != NULL);
1209 rc = ofd_attr_get(env, fo, &info->fti_attr);
1210 oa->o_valid = OBD_MD_FLID;
1214 obdo_from_la(oa, &info->fti_attr,
1215 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1217 /* Store object version in reply */
1218 curr_version = dt_version_get(env, ofd_object_child(fo));
1219 if ((__s64)curr_version != -EOPNOTSUPP) {
1220 oa->o_valid |= OBD_MD_FLDATAVERSION;
1221 oa->o_data_version = curr_version;
1225 ofd_object_put(env, fo);
1231 * Get object version for OBD_IOC_GET_OBJ_VERSION ioctl.
1233 * This is supplemental function for ofd_iocontrol() to return object
1234 * version for lctl tool.
1236 * \param[in] env execution environment
1237 * \param[in] ofd OFD device
1238 * \param[out] karg ioctl data
1240 * \retval 0 if successful
1241 * \retval negative value on error
1243 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1244 struct ofd_device *ofd, void *karg)
1246 struct obd_ioctl_data *data = karg;
1248 struct ofd_object *fo;
1249 dt_obj_version_t version;
1254 if (!data->ioc_inlbuf2 || data->ioc_inllen2 != sizeof(version))
1255 GOTO(out, rc = -EINVAL);
1257 if (data->ioc_inlbuf1 && data->ioc_inllen1 == sizeof(fid)) {
1258 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1259 } else if (data->ioc_inlbuf3 &&
1260 data->ioc_inllen3 == sizeof(__u64) &&
1261 data->ioc_inlbuf4 &&
1262 data->ioc_inllen4 == sizeof(__u64)) {
1263 struct ost_id ostid = { };
1265 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1266 rc = ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1269 rc = ostid_to_fid(&fid, &ostid,
1270 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1274 GOTO(out, rc = -EINVAL);
1277 if (!fid_is_sane(&fid))
1278 GOTO(out, rc = -EINVAL);
1280 fo = ofd_object_find(env, ofd, &fid);
1282 GOTO(out, rc = PTR_ERR(fo));
1284 if (!ofd_object_exists(fo))
1285 GOTO(out_fo, rc = -ENOENT);
1287 if (lu_object_remote(&fo->ofo_obj.do_lu))
1288 GOTO(out_fo, rc = -EREMOTE);
1290 version = dt_version_get(env, ofd_object_child(fo));
1292 GOTO(out_fo, rc = -EIO);
1294 *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1298 ofd_object_put(env, fo);
1304 * Implementation of obd_ops::o_iocontrol.
1306 * This is OFD ioctl handling function which is primary interface for
1307 * Lustre tools like lfs, lctl and lfsck.
1309 * \param[in] cmd ioctl command
1310 * \param[in] exp OBD export of OFD
1311 * \param[in] len not used
1312 * \param[in,out] karg buffer with data
1313 * \param[in] uarg not used
1315 * \retval 0 if successful
1316 * \retval negative value on error
1318 static int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1319 void *karg, void __user *uarg)
1322 struct ofd_device *ofd = ofd_exp(exp);
1323 struct obd_device *obd = ofd_obd(ofd);
1328 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1329 rc = lu_env_init(&env, LCT_DT_THREAD);
1334 case OBD_IOC_ABORT_RECOVERY:
1335 CERROR("%s: aborting recovery\n", obd->obd_name);
1336 obd->obd_abort_recovery = 1;
1337 target_stop_recovery_thread(obd);
1340 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1341 rc = dt_sync(&env, ofd->ofd_osd);
1343 case OBD_IOC_SET_READONLY:
1344 rc = dt_sync(&env, ofd->ofd_osd);
1346 rc = dt_ro(&env, ofd->ofd_osd);
1348 case OBD_IOC_START_LFSCK: {
1349 struct obd_ioctl_data *data = karg;
1350 struct lfsck_start_param lsp;
1352 if (unlikely(!data)) {
1357 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
1358 lsp.lsp_index_valid = 0;
1359 rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
1362 case OBD_IOC_STOP_LFSCK: {
1363 struct obd_ioctl_data *data = karg;
1364 struct lfsck_stop stop;
1366 stop.ls_status = LS_STOPPED;
1367 /* Old lfsck utils may pass NULL @stop. */
1368 if (!data->ioc_inlbuf1)
1372 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
1374 rc = lfsck_stop(&env, ofd->ofd_osd, &stop);
1377 case OBD_IOC_GET_OBJ_VERSION:
1378 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1381 CERROR("%s: not supported cmd = %#x\n", obd->obd_name, cmd);
1390 * Implementation of obd_ops::o_precleanup.
1392 * This function stops device activity before shutting it down. It is called
1393 * from a cleanup function upon forceful device cleanup. For OFD there are no
1394 * special actions, it just invokes target_recovery_cleanup().
1396 * \param[in] obd OBD device of OFD
1400 static int ofd_precleanup(struct obd_device *obd)
1403 target_cleanup_recovery(obd);
1408 * Implementation of obd_ops::o_ping.
1410 * This is OFD-specific part of OBD_PING request handling.
1411 * It controls Filter Modification Data (FMD) expiration each time PING is
1414 * \see ofd_fmd_expire() and ofd_fmd.c for details
1416 * \param[in] env execution environment
1417 * \param[in] exp OBD export of client
1421 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1423 ofd_fmd_expire(exp);
1428 * Implementation of obd_ops::o_health_check.
1430 * This function checks the OFD device health - ability to respond on
1431 * incoming requests. There are two health_check methods:
1432 * - get statfs from the OSD. It checks just responsiveness of
1434 * - do write attempt on bottom device to check it is fully operational and
1435 * is not stuck. This is expensive method and requires special configuration
1436 * option --enable-health-write while building Lustre, it is turned off
1439 * \param[in] nul not used
1440 * \param[in] obd OBD device of OFD
1442 * \retval 0 if successful
1443 * \retval negative value in case of error
1445 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1447 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1448 struct ofd_thread_info *info;
1450 #ifdef USE_HEALTH_CHECK_WRITE
1455 /* obd_proc_read_health pass NULL env, we need real one */
1456 rc = lu_env_init(&env, LCT_DT_THREAD);
1460 info = ofd_info_init(&env, NULL);
1461 rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1465 if (info->fti_u.osfs.os_state & OS_STATE_READONLY)
1466 GOTO(out, rc = -EROFS);
1468 #ifdef USE_HEALTH_CHECK_WRITE
1469 OBD_ALLOC(info->fti_buf.lb_buf, PAGE_SIZE);
1470 if (!info->fti_buf.lb_buf)
1471 GOTO(out, rc = -ENOMEM);
1473 info->fti_buf.lb_len = PAGE_SIZE;
1476 th = dt_trans_create(&env, ofd->ofd_osd);
1478 GOTO(out, rc = PTR_ERR(th));
1480 rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1481 &info->fti_buf, info->fti_off, th);
1483 th->th_sync = 1; /* sync IO is needed */
1484 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1486 rc = dt_record_write(&env, ofd->ofd_health_check_file,
1487 &info->fti_buf, &info->fti_off,
1490 dt_trans_stop(&env, ofd->ofd_osd, th);
1492 OBD_FREE(info->fti_buf.lb_buf, PAGE_SIZE);
1494 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",
1502 struct obd_ops ofd_obd_ops = {
1503 .o_owner = THIS_MODULE,
1504 .o_connect = ofd_obd_connect,
1505 .o_reconnect = ofd_obd_reconnect,
1506 .o_disconnect = ofd_obd_disconnect,
1507 .o_create = ofd_echo_create,
1508 .o_statfs = ofd_statfs,
1509 .o_setattr = ofd_echo_setattr,
1510 .o_preprw = ofd_preprw,
1511 .o_commitrw = ofd_commitrw,
1512 .o_destroy = ofd_echo_destroy,
1513 .o_init_export = ofd_init_export,
1514 .o_destroy_export = ofd_destroy_export,
1515 .o_postrecov = ofd_obd_postrecov,
1516 .o_getattr = ofd_echo_getattr,
1517 .o_iocontrol = ofd_iocontrol,
1518 .o_precleanup = ofd_precleanup,
1520 .o_health_check = ofd_health_check,
1521 .o_set_info_async = ofd_set_info_async,
1522 .o_get_info = ofd_get_info,