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 if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
76 /* Self-export gets no proc entry */
79 rc = lprocfs_exp_setup(exp, client_nid);
81 /* Mask error for already created /proc entries */
82 RETURN(rc == -EALREADY ? 0 : rc);
84 stats = exp->exp_nid_stats;
85 stats->nid_stats = lprocfs_alloc_stats(LPROC_OFD_STATS_LAST,
86 LPROCFS_STATS_FLAG_NOPERCPU);
87 if (stats->nid_stats == NULL)
90 ofd_stats_counter_init(stats->nid_stats);
92 rc = lprocfs_register_stats(stats->nid_proc, "stats", stats->nid_stats);
94 lprocfs_free_stats(&stats->nid_stats);
98 rc = lprocfs_nid_ldlm_stats_init(stats);
107 * Decide which checksums both client and OST support, possibly forcing
108 * the use of T10PI checksums if the hardware supports this.
110 * The clients that have no T10-PI RPC checksum support will use the same
111 * mechanism to select checksum type as before, and will not be affected by
112 * the following logic.
114 * For the clients that have T10-PI RPC checksum support:
116 * If the OST supports T10-PI feature and T10-PI checksum is enforced, clients
117 * will have no other choice for RPC checksum type other than using the T10PI
118 * checksum type. This is useful for enforcing end-to-end integrity in the
121 * If the OST doesn't support T10-PI feature and T10-PI checksum is enforced,
122 * together with other checksum with reasonably good speeds (e.g. crc32,
123 * crc32c, adler, etc.), all T10-PI checksum types understood by the client
124 * (t10ip512, t10ip4K, t10crc512, t10crc4K) will be added to the available
125 * checksum types, regardless of the speeds of T10-PI checksums. This is
126 * useful for testing T10-PI checksum of RPC.
128 * If the OST supports T10-PI feature and T10-PI checksum is NOT enforced,
129 * the corresponding T10-PI checksum type will be added to the checksum type
130 * list, regardless of the speed of the T10-PI checksum. This provides clients
131 * the flexibility to choose whether to enable end-to-end integrity or not.
133 * If the OST does NOT supports T10-PI feature and T10-PI checksum is NOT
134 * enforced, together with other checksums with reasonably good speeds,
135 * all the T10-PI checksum types with good speeds will be added into the
136 * checksum type list. Note that a T10-PI checksum type with a speed worse
137 * than half of Alder will NOT be added as a option. In this circumstance,
138 * T10-PI checksum types has the same behavior like other normal checksum
143 ofd_mask_cksum_types(struct ofd_device *ofd, enum cksum_types *cksum_types)
145 bool enforce = ofd->ofd_checksum_t10pi_enforce;
146 enum cksum_types ofd_t10_cksum_type;
147 enum cksum_types client_t10_types = *cksum_types & OBD_CKSUM_T10_ALL;
148 enum cksum_types server_t10_types;
151 * The client set in ocd_cksum_types the checksum types it
152 * supports. We have to mask off the algorithms that we don't
153 * support. T10PI checksum types will be added later.
155 *cksum_types &= (ofd->ofd_cksum_types_supported & ~OBD_CKSUM_T10_ALL);
156 server_t10_types = ofd->ofd_cksum_types_supported & OBD_CKSUM_T10_ALL;
157 ofd_t10_cksum_type = ofd->ofd_lut.lut_dt_conf.ddp_t10_cksum_type;
159 /* Quick exit if no T10-PI support on client */
160 if (!client_t10_types)
164 * This OST has NO T10-PI feature. Add all supported T10-PI checksums
165 * as options if T10-PI checksum is enforced. If the T10-PI checksum is
166 * not enforced, only add them as options when speed is good.
168 if (ofd_t10_cksum_type == 0) {
170 * Server allows all T10PI checksums, and server_t10_types
171 * include quick ones.
174 *cksum_types |= client_t10_types;
176 *cksum_types |= client_t10_types & server_t10_types;
181 * This OST has T10-PI feature. Disable all other checksum types if
182 * T10-PI checksum is enforced. If the T10-PI checksum is not enforced,
183 * add the checksum type as an option.
185 if (client_t10_types & ofd_t10_cksum_type) {
187 *cksum_types = ofd_t10_cksum_type;
189 *cksum_types |= ofd_t10_cksum_type;
194 * Match client and OST server connection feature flags.
196 * Compute the compatibility flags for a connection request based on
197 * features mutually supported by client and server.
199 * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
200 * must not be updated here, otherwise a partially initialized value may
201 * be exposed. After the connection request is successfully processed,
202 * the top-level tgt_connect() request handler atomically updates the export
203 * connect flags from the obd_connect_data::ocd_connect_flags field of the
204 * reply. \see tgt_connect().
206 * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
207 * full struct obd_connect_data. So care must be taken when accessing fields
208 * that are not present in struct obd_connect_data_v1. See LU-16.
210 * \param[in] env execution environment
211 * \param[in] exp the obd_export associated with this
213 * \param[in] data stores data for this connect request
214 * \param[in] new_connection is this connection new or not
216 * \retval 0 if success
217 * \retval -EPROTO client and server feature requirements are
219 * \retval -EBADF OST index in connect request doesn't match
222 static int ofd_parse_connect_data(const struct lu_env *env,
223 struct obd_export *exp,
224 struct obd_connect_data *data,
227 struct ofd_device *ofd = ofd_exp(exp);
228 struct filter_export_data *fed = &exp->exp_filter_data;
233 CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: %#llx"
234 " ocd_version: %x ocd_grant: %d ocd_index: %u"
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 "
242 "earlier; now it's trying to use group %d! This could "
243 "be a bug in the MDS. Please report to "
244 "https://jira.whamcloud.com/\n",
245 obd_export_nid2str(exp), fed->fed_group,
249 fed->fed_group = data->ocd_group;
251 data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
253 if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
254 data->ocd_connect_flags2 &= OST_CONNECT_SUPPORTED2;
256 /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
257 if (data->ocd_connect_flags & OBD_CONNECT_MDS)
258 CDEBUG(D_HA, "%s: Received MDS connection for group %u\n",
259 exp->exp_obd->obd_name, data->ocd_group);
260 else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN)
263 /* Determine optimal brw size before calculating grant */
264 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) {
265 data->ocd_brw_size = 65536;
266 } else if (OCD_HAS_FLAG(data, BRW_SIZE)) {
267 if (data->ocd_brw_size > ofd->ofd_brw_size)
268 data->ocd_brw_size = ofd->ofd_brw_size;
269 if (data->ocd_brw_size == 0) {
270 CERROR("%s: cli %s/%p ocd_connect_flags: %#llx"
271 " ocd_version: %x ocd_grant: %d ocd_index: %u "
272 "ocd_brw_size is unexpectedly zero, "
273 "network data corruption?"
274 "Refusing connection of this client\n",
275 exp->exp_obd->obd_name,
276 exp->exp_client_uuid.uuid,
277 exp, data->ocd_connect_flags, data->ocd_version,
278 data->ocd_grant, data->ocd_index);
283 if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
284 struct dt_device_param *ddp = &ofd->ofd_lut.lut_dt_conf;
286 /* client is reporting its page size, for future use */
287 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
288 data->ocd_grant_blkbits = ofd->ofd_lut.lut_tgd.tgd_blockbits;
289 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
290 * it's LDISKFS_DIR_REC_LEN(20) = 28. */
291 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
292 /* ocd_grant_tax_kb is in 1K byte blocks */
293 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
294 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
297 /* Save connect_data we have so far because tgt_grant_connect()
298 * uses it to calculate grant, and we want to save the client
299 * version before it is overwritten by LUSTRE_VERSION_CODE. */
300 exp->exp_connect_data = *data;
301 if (OCD_HAS_FLAG(data, GRANT))
302 tgt_grant_connect(env, exp, data, new_connection);
304 if (data->ocd_connect_flags & OBD_CONNECT_INDEX) {
305 struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd;
306 int index = lsd->lsd_osd_index;
308 if (index != data->ocd_index) {
309 LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index"
310 " %u doesn't match actual OST index"
311 " %u in last_rcvd file, bad "
313 obd_export_nid2str(exp), index,
317 if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) {
318 /* this will only happen on the first connect */
319 lsd->lsd_feature_compat |= OBD_COMPAT_OST;
320 /* sync is not needed here as tgt_client_new will
321 * set exp_need_sync flag */
322 tgt_server_data_update(env, &ofd->ofd_lut, 0);
326 if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) {
327 __u32 cksum_types = data->ocd_cksum_types;
329 ofd_mask_cksum_types(ofd, &data->ocd_cksum_types);
331 if (unlikely(data->ocd_cksum_types == 0)) {
332 CERROR("%s: Connect with checksum support but no "
333 "ocd_cksum_types is set\n",
334 exp->exp_obd->obd_name);
338 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
339 "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
340 cksum_types, data->ocd_cksum_types);
342 /* This client does not support OBD_CONNECT_CKSUM.
343 * Report failure to negotiate checksum at connect */
344 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
345 "OBD_CONNECT_CKSUM\n",
346 exp->exp_obd->obd_name, obd_export_nid2str(exp));
349 if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES)
350 data->ocd_maxbytes = ofd->ofd_lut.lut_dt_conf.ddp_maxbytes;
352 data->ocd_version = LUSTRE_VERSION_CODE;
354 if (OCD_HAS_FLAG(data, PINGLESS)) {
355 if (ptlrpc_pinger_suppress_pings()) {
356 spin_lock(&exp->exp_obd->obd_dev_lock);
357 list_del_init(&exp->exp_obd_chain_timed);
358 spin_unlock(&exp->exp_obd->obd_dev_lock);
360 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
368 * Re-initialize export upon client reconnection.
370 * This function parses connection data from reconnect and resets
373 * \param[in] env execution environment
374 * \param[in] exp OBD export
375 * \param[in] obd OFD device
376 * \param[in] cluuid NID of client
377 * \param[in] data connection data from request
378 * \param[in] localdata client NID
380 * \retval 0 if successful
381 * \retval negative value on error
383 static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp,
384 struct obd_device *obd, struct obd_uuid *cluuid,
385 struct obd_connect_data *data,
388 struct ofd_device *ofd;
393 if (exp == NULL || obd == NULL || cluuid == NULL)
396 rc = nodemap_add_member(*(lnet_nid_t *)client_nid, exp);
397 if (rc != 0 && rc != -EEXIST)
400 ofd = ofd_dev(obd->obd_lu_dev);
402 rc = ofd_parse_connect_data(env, exp, data, false);
404 ofd_export_stats_init(ofd, exp, client_nid);
406 nodemap_del_member(exp);
412 * Initialize new client connection.
414 * This function handles new connection to the OFD. The new export is
415 * created (in context of class_connect()) and persistent client data is
416 * initialized on storage.
418 * \param[in] env execution environment
419 * \param[out] _exp stores pointer to new export
420 * \param[in] obd OFD device
421 * \param[in] cluuid client UUID
422 * \param[in] data connection data from request
423 * \param[in] localdata client NID
425 * \retval 0 if successful
426 * \retval negative value on error
428 static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp,
429 struct obd_device *obd, struct obd_uuid *cluuid,
430 struct obd_connect_data *data, void *localdata)
432 struct obd_export *exp;
433 struct ofd_device *ofd;
434 struct lustre_handle conn = { 0 };
438 if (_exp == NULL || obd == NULL || cluuid == NULL)
441 ofd = ofd_dev(obd->obd_lu_dev);
443 rc = class_connect(&conn, obd, cluuid);
447 exp = class_conn2export(&conn);
448 LASSERT(exp != NULL);
450 if (localdata != NULL) {
451 rc = nodemap_add_member(*(lnet_nid_t *)localdata, exp);
452 if (rc != 0 && rc != -EEXIST)
455 CDEBUG(D_HA, "%s: cannot find nodemap for client %s: "
456 "nid is null\n", obd->obd_name, cluuid->uuid);
459 rc = ofd_parse_connect_data(env, exp, data, true);
463 if (obd->obd_replayable) {
464 struct tg_export_data *ted = &exp->exp_target_data;
466 memcpy(ted->ted_lcd->lcd_uuid, cluuid,
467 sizeof(ted->ted_lcd->lcd_uuid));
468 rc = tgt_client_new(env, exp);
471 ofd_export_stats_init(ofd, exp, localdata);
474 CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
475 data ? data->ocd_group : -1);
479 class_disconnect(exp);
480 nodemap_del_member(exp);
489 * Disconnect a connected client.
491 * This function terminates the client connection. The client export is
492 * disconnected (cleaned up) and client data on persistent storage is removed.
494 * \param[in] exp OBD export
496 * \retval 0 if successful
497 * \retval negative value on error
499 int ofd_obd_disconnect(struct obd_export *exp)
501 struct ofd_device *ofd = ofd_exp(exp);
508 class_export_get(exp);
510 if (!(exp->exp_flags & OBD_OPT_FORCE))
511 tgt_grant_sanity_check(ofd_obd(ofd), __func__);
513 rc = server_disconnect_export(exp);
515 tgt_grant_discard(exp);
517 /* Do not erase record for recoverable client. */
518 if (exp->exp_obd->obd_replayable &&
519 (!exp->exp_obd->obd_fail || exp->exp_failed)) {
520 rc = lu_env_init(&env, LCT_DT_THREAD);
524 tgt_client_del(&env, exp);
528 nodemap_del_member(exp);
529 class_export_put(exp);
534 * Implementation of obd_ops::o_init_export.
536 * This function is called from class_new_export() and initializes
537 * the OFD-specific data for new export.
539 * \param[in] exp OBD export
541 * \retval 0 if successful
542 * \retval negative value on error
544 static int ofd_init_export(struct obd_export *exp)
548 spin_lock_init(&exp->exp_filter_data.fed_lock);
549 INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list);
550 atomic_set(&exp->exp_filter_data.fed_soft_sync_count, 0);
551 spin_lock(&exp->exp_lock);
552 exp->exp_connecting = 1;
553 spin_unlock(&exp->exp_lock);
555 /* self-export doesn't need client data and ldlm initialization */
556 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
557 &exp->exp_client_uuid)))
560 rc = tgt_client_alloc(exp);
562 ldlm_init_export(exp);
564 CERROR("%s: Can't initialize export: rc %d\n",
565 exp->exp_obd->obd_name, rc);
570 * Implementation of obd_ops::o_destroy_export.
572 * This function is called from class_export_destroy() to cleanup
573 * the OFD-specific data for export being destroyed.
575 * \param[in] exp OBD export
577 * \retval 0 if successful
578 * \retval negative value on error
580 static int ofd_destroy_export(struct obd_export *exp)
582 struct ofd_device *ofd = ofd_exp(exp);
584 if (exp->exp_target_data.ted_pending)
585 CERROR("%s: cli %s/%p has %lu pending on destroyed export"
586 "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
587 exp, exp->exp_target_data.ted_pending);
589 target_destroy_export(exp);
591 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
592 &exp->exp_client_uuid)))
595 ldlm_destroy_export(exp);
596 tgt_client_free(exp);
598 ofd_fmd_cleanup(exp);
601 * discard grants once we're sure no more
602 * interaction with the client is possible
604 tgt_grant_discard(exp);
605 ofd_fmd_cleanup(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__);
613 LASSERT(list_empty(&exp->exp_filter_data.fed_mod_list));
618 * Notify all devices in server stack about recovery completion.
620 * This function calls ldo_recovery_complete() for all lower devices in the
621 * server stack so they will be prepared for normal operations.
623 * \param[in] env execution environment
624 * \param[in] ofd OFD device
626 * \retval 0 if successful
627 * \retval negative value on error
629 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
631 struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
634 CDEBUG(D_HA, "%s: recovery is over\n", ofd_name(ofd));
636 if (!ofd->ofd_skip_lfsck && !ofd->ofd_osd->dd_rdonly) {
637 struct lfsck_start_param lsp;
639 lsp.lsp_start = NULL;
640 lsp.lsp_index_valid = 0;
641 rc = lfsck_start(env, ofd->ofd_osd, &lsp);
642 if (rc != 0 && rc != -EALREADY)
643 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
647 return ldev->ld_ops->ldo_recovery_complete(env, ldev);
651 * Implementation of obd_ops::o_postrecov.
653 * This function is called from target_finish_recovery() upon recovery
656 * \param[in] obd OBD device of OFD
658 * \retval 0 if successful
659 * \retval negative value on error
661 static int ofd_obd_postrecov(struct obd_device *obd)
664 struct lu_device *ldev = obd->obd_lu_dev;
669 rc = lu_env_init(&env, LCT_DT_THREAD);
672 ofd_info_init(&env, obd->obd_self_export);
674 rc = ofd_postrecov(&env, ofd_dev(ldev));
681 * Implementation of obd_ops::o_set_info_async.
683 * This function is not called from request handler, it is only used by
684 * class_notify_sptlrpc_conf() locally by direct obd_set_info_async() call.
685 * \see ofd_set_info_hdl() for request handler function.
687 * \param[in] env execution environment
688 * \param[in] exp OBD export of OFD device
689 * \param[in] keylen length of \a key
690 * \param[in] key key name
691 * \param[in] vallen length of \a val
692 * \param[in] val the \a key value
693 * \param[in] set not used in OFD
695 * \retval 0 if successful
696 * \retval negative value on error
698 static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp,
699 __u32 keylen, void *key, __u32 vallen, void *val,
700 struct ptlrpc_request_set *set)
706 if (exp->exp_obd == NULL) {
707 CDEBUG(D_IOCTL, "invalid export %p\n", exp);
711 if (KEY_IS(KEY_SPTLRPC_CONF)) {
712 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
714 CERROR("%s: Unsupported key %s\n",
715 exp->exp_obd->obd_name, (char*)key);
722 * Implementation of obd_ops::o_get_info.
724 * This function is not called from request handler, it is only used by
725 * direct call from nrs_orr_range_fill_physical() in ptlrpc, see LU-3239.
727 * \see ofd_get_info_hdl() for request handler function.
729 * \param[in] env execution environment
730 * \param[in] exp OBD export of OFD device
731 * \param[in] keylen length of \a key
732 * \param[in] key key name
733 * \param[out] vallen length of key value
734 * \param[out] val the key value to return
736 * \retval 0 if successful
737 * \retval negative value on error
739 static int ofd_get_info(const struct lu_env *env, struct obd_export *exp,
740 __u32 keylen, void *key, __u32 *vallen, void *val)
742 struct ofd_thread_info *info;
743 struct ofd_device *ofd;
744 struct ll_fiemap_info_key *fm_key = key;
745 struct fiemap *fiemap = val;
750 if (exp->exp_obd == NULL) {
751 CDEBUG(D_IOCTL, "invalid client export %p\n", exp);
757 if (KEY_IS(KEY_FIEMAP)) {
758 info = ofd_info_init(env, exp);
760 rc = ostid_to_fid(&info->fti_fid, &fm_key->lfik_oa.o_oi,
761 ofd->ofd_lut.lut_lsd.lsd_osd_index);
765 rc = ofd_fiemap_get(env, ofd, &info->fti_fid, fiemap);
767 CERROR("%s: not supported key %s\n", 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);
809 /* at least try to account for cached pages. its still racy and
810 * might be under-reporting if clients haven't announced their
811 * caches with brw recently */
813 CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
814 " pending %llu free %llu avail %llu\n",
815 tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
816 tgd->tgd_tot_pending,
817 osfs->os_bfree << tgd->tgd_blockbits,
818 osfs->os_bavail << tgd->tgd_blockbits);
820 osfs->os_bavail -= min_t(u64, osfs->os_bavail,
821 ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
822 osfs->os_bsize - 1) >> tgd->tgd_blockbits));
824 /* The QoS code on the MDS does not care about space reserved for
825 * precreate, so take it out. */
826 if (exp_connect_flags(exp) & OBD_CONNECT_MDS) {
827 struct tg_export_data *ted;
829 ted = &obd->obd_self_export->exp_target_data;
830 osfs->os_granted = min_t(u64, osfs->os_bavail,
831 ted->ted_grant >> tgd->tgd_blockbits);
832 osfs->os_bavail -= osfs->os_granted;
835 tgt_grant_sanity_check(obd, __func__);
836 CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
837 "%llu objects: %llu free; state %x\n",
838 osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
839 osfs->os_files, osfs->os_ffree, osfs->os_state);
841 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
842 ofd->ofd_lut.lut_lsd.lsd_osd_index)) {
843 /* Reduce free inode count to zero, but keep "used" intact */
844 osfs->os_files -= osfs->os_ffree;
845 osfs->os_ffree -= osfs->os_ffree;
848 /* OS_STATE_READONLY can be set by OSD already */
849 if (ofd->ofd_raid_degraded)
850 osfs->os_state |= OS_STATE_DEGRADED;
852 if (obd->obd_self_export != exp && !exp_grant_param_supp(exp) &&
853 tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
854 /* clients which don't support OBD_CONNECT_GRANT_PARAM
855 * should not see a block size > page size, otherwise
856 * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
857 * block size which is the biggest block size known to work
858 * with all client's page size. */
859 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
860 osfs->os_bfree <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
861 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
862 osfs->os_granted <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
863 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
866 if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
867 ofd->ofd_lut.lut_lsd.lsd_osd_index)) {
868 /* Reduce free blocks count near zero, but keep "used" intact */
869 osfs->os_bavail -= osfs->os_bavail - 2;
870 osfs->os_blocks -= osfs->os_bfree - 2;
871 osfs->os_bfree -= osfs->os_bfree - 2;
880 * Implementation of obd_ops::o_setattr.
882 * This function is only used by ECHO client when it is run on top of OFD,
883 * \see ofd_setattr_hdl() for request handler function.
885 * \param[in] env execution environment
886 * \param[in] exp OBD export of OFD device
887 * \param[in] oa setattr parameters
889 * \retval 0 if successful
890 * \retval negative value on error
892 static int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp,
895 struct ofd_thread_info *info;
896 struct ofd_device *ofd = ofd_exp(exp);
897 struct ldlm_namespace *ns = ofd->ofd_namespace;
898 struct ldlm_resource *res;
899 struct ofd_object *fo;
900 struct lu_fid *fid = &oa->o_oi.oi_fid;
905 info = ofd_info_init(env, exp);
907 ost_fid_build_resid(fid, &info->fti_resid);
909 /* This would be very bad - accidentally truncating a file when
910 * changing the time or similar - bug 12203. */
911 if (oa->o_valid & OBD_MD_FLSIZE) {
912 static char mdsinum[48];
914 if (oa->o_valid & OBD_MD_FLFID)
915 snprintf(mdsinum, sizeof(mdsinum) - 1,
916 "of parent "DFID, oa->o_parent_seq,
917 oa->o_parent_oid, 0);
921 CERROR("%s: setattr from %s trying to truncate object "DFID
922 " %s\n", ofd_name(ofd), obd_export_nid2str(exp),
924 GOTO(out, rc = -EPERM);
927 fo = ofd_object_find_exists(env, ofd, fid);
929 CERROR("%s: can't find object "DFID"\n",
930 ofd_name(ofd), PFID(fid));
931 GOTO(out, rc = PTR_ERR(fo));
934 la_from_obdo(&info->fti_attr, oa, oa->o_valid);
935 info->fti_attr.la_valid &= ~LA_TYPE;
937 /* setting objects attributes (including owner/group) */
938 rc = ofd_attr_set(env, fo, &info->fti_attr, oa);
940 GOTO(out_unlock, rc);
942 ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, NULL, 1);
945 ofd_object_put(env, fo);
948 /* we do not call this before to avoid lu_object_find() in
949 * ->lvbo_update() holding another reference on the object.
950 * otherwise concurrent destroy can make the object unavailable
951 * for 2nd lu_object_find() waiting for the first reference
952 * to go... deadlock! */
953 res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
955 ldlm_res_lvbo_update(env, res, NULL, 0);
956 ldlm_resource_putref(res);
964 * Destroy OFD object by its FID.
966 * Supplemental function to destroy object by FID, it is used by request
967 * handler and by ofd_echo_destroy() below to find object by FID, lock it
968 * and call ofd_destroy() finally.
970 * \param[in] env execution environment
971 * \param[in] ofd OFD device
972 * \param[in] fid FID of object
973 * \param[in] orphan set if object being destroyed is an orphan
975 * \retval 0 if successful
976 * \retval negative value on error
978 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
979 const struct lu_fid *fid, int orphan)
981 struct ofd_thread_info *info = ofd_info(env);
982 struct lustre_handle lockh;
983 union ldlm_policy_data policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
984 struct ofd_object *fo;
985 __u64 flags = LDLM_FL_AST_DISCARD_DATA;
990 fo = ofd_object_find_exists(env, ofd, fid);
994 /* Tell the clients that the object is gone now and that they should
995 * throw away any cached pages. */
996 ost_fid_build_resid(fid, &info->fti_resid);
997 rc = ldlm_cli_enqueue_local(env, ofd->ofd_namespace, &info->fti_resid,
998 LDLM_EXTENT, &policy, LCK_PW, &flags,
999 ldlm_blocking_ast, ldlm_completion_ast,
1000 NULL, NULL, 0, LVB_T_NONE, NULL, &lockh);
1002 /* We only care about the side-effects, just drop the lock. */
1004 ldlm_lock_decref(&lockh, LCK_PW);
1006 LASSERT(fo != NULL);
1008 rc = ofd_destroy(env, fo, orphan);
1011 ofd_object_put(env, fo);
1016 * Implementation of obd_ops::o_destroy.
1018 * This function is only used by ECHO client when it is run on top of OFD,
1019 * \see ofd_destroy_hdl() for request handler function.
1021 * \param[in] env execution environment
1022 * \param[in] exp OBD export of OFD device
1023 * \param[in] oa obdo structure with FID
1025 * Note: this is OBD API method which is common API for server OBDs and
1026 * client OBDs. Thus some parameters used in client OBDs may not be used
1027 * on server OBDs and vice versa.
1029 * \retval 0 if successful
1030 * \retval negative value on error
1032 static int ofd_echo_destroy(const struct lu_env *env, struct obd_export *exp,
1035 struct ofd_device *ofd = ofd_exp(exp);
1036 struct lu_fid *fid = &oa->o_oi.oi_fid;
1041 ofd_info_init(env, exp);
1043 rc = ofd_validate_seq(exp, ostid_seq(&oa->o_oi));
1047 CDEBUG(D_HA, "%s: Destroy object "DFID"\n", ofd_name(ofd), PFID(fid));
1049 rc = ofd_destroy_by_fid(env, ofd, fid, 0);
1050 if (rc == -ENOENT) {
1051 CDEBUG(D_INODE, "%s: destroying non-existent object "DFID"\n",
1052 ofd_name(ofd), PFID(fid));
1054 } else if (rc != 0) {
1055 CERROR("%s: error destroying object "DFID": %d\n",
1056 ofd_name(ofd), PFID(fid), rc);
1065 * Implementation of obd_ops::o_create.
1067 * This function is only used by ECHO client when it is run on top of OFD
1068 * and just creates an object.
1069 * \see ofd_create_hdl() for request handler function.
1071 * \param[in] env execution environment
1072 * \param[in] exp OBD export of OFD device
1073 * \param[in] oa obdo structure with FID sequence to use
1075 * Note: this is OBD API method which is common API for server OBDs and
1076 * client OBDs. Thus some parameters used in client OBDs may not be used
1077 * on server OBDs and vice versa.
1079 * \retval 0 if successful
1080 * \retval negative value on error
1082 static int ofd_echo_create(const struct lu_env *env, struct obd_export *exp,
1085 struct ofd_device *ofd = ofd_exp(exp);
1086 u64 seq = ostid_seq(&oa->o_oi);
1087 struct ofd_seq *oseq;
1096 ofd_info_init(env, exp);
1098 LASSERT(seq == FID_SEQ_ECHO);
1099 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
1101 CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi));
1103 down_read(&ofd->ofd_lastid_rwsem);
1104 /* Currently, for safe, we do not distinguish which LAST_ID is broken,
1105 * we may do that in the future.
1106 * Return -ENOSPC until the LAST_ID rebuilt. */
1107 if (unlikely(ofd->ofd_lastid_rebuilding))
1108 GOTO(out_sem, rc = -ENOSPC);
1110 rc = ofd_validate_seq(exp, seq);
1114 oseq = ofd_seq_load(env, ofd, seq);
1116 CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n",
1117 ofd_name(ofd), seq, PTR_ERR(oseq));
1118 GOTO(out_sem, rc = -EINVAL);
1121 mutex_lock(&oseq->os_create_lock);
1122 granted = tgt_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
1126 CDEBUG(D_HA, "%s: failed to acquire grant space for "
1127 "precreate (%lld): rc = %d\n", ofd_name(ofd), diff, rc);
1132 next_id = ofd_seq_last_oid(oseq) + 1;
1133 count = ofd_precreate_batch(ofd, (int)diff);
1135 rc = ofd_precreate_objects(env, ofd, next_id, oseq, count, 0);
1137 CERROR("%s: unable to precreate: rc = %d\n",
1140 rc = ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq));
1142 CERROR("%s: Bad %llu to set " DOSTID " : rc %d\n",
1144 (unsigned long long)ofd_seq_last_oid(oseq),
1145 POSTID(&oa->o_oi), rc);
1147 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
1150 tgt_grant_commit(ofd_obd(ofd)->obd_self_export, granted, rc);
1152 mutex_unlock(&oseq->os_create_lock);
1153 ofd_seq_put(env, oseq);
1156 up_read(&ofd->ofd_lastid_rwsem);
1161 * Implementation of obd_ops::o_getattr.
1163 * This function is only used by ECHO client when it is run on top of OFD
1164 * and returns attributes of object.
1165 * \see ofd_getattr_hdl() for request handler function.
1167 * \param[in] env execution environment
1168 * \param[in] exp OBD export of OFD device
1169 * \param[in,out] oa contains FID of object to get attributes from and
1170 * is used to return attributes back
1172 * \retval 0 if successful
1173 * \retval negative value on error
1175 static int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp,
1178 struct ofd_device *ofd = ofd_exp(exp);
1179 struct ofd_thread_info *info;
1180 struct lu_fid *fid = &oa->o_oi.oi_fid;
1181 struct ofd_object *fo;
1186 info = ofd_info_init(env, exp);
1188 fo = ofd_object_find_exists(env, ofd, fid);
1190 GOTO(out, rc = PTR_ERR(fo));
1192 LASSERT(fo != NULL);
1193 rc = ofd_attr_get(env, fo, &info->fti_attr);
1194 oa->o_valid = OBD_MD_FLID;
1198 obdo_from_la(oa, &info->fti_attr,
1199 OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID);
1201 /* Store object version in reply */
1202 curr_version = dt_version_get(env, ofd_object_child(fo));
1203 if ((__s64)curr_version != -EOPNOTSUPP) {
1204 oa->o_valid |= OBD_MD_FLDATAVERSION;
1205 oa->o_data_version = curr_version;
1209 ofd_object_put(env, fo);
1215 * Get object version for OBD_IOC_GET_OBJ_VERSION ioctl.
1217 * This is supplemental function for ofd_iocontrol() to return object
1218 * version for lctl tool.
1220 * \param[in] env execution environment
1221 * \param[in] ofd OFD device
1222 * \param[out] karg ioctl data
1224 * \retval 0 if successful
1225 * \retval negative value on error
1227 static int ofd_ioc_get_obj_version(const struct lu_env *env,
1228 struct ofd_device *ofd, void *karg)
1230 struct obd_ioctl_data *data = karg;
1232 struct ofd_object *fo;
1233 dt_obj_version_t version;
1238 if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
1239 GOTO(out, rc = -EINVAL);
1241 if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) {
1242 fid = *(struct lu_fid *)data->ioc_inlbuf1;
1243 } else if (data->ioc_inlbuf3 != NULL &&
1244 data->ioc_inllen3 == sizeof(__u64) &&
1245 data->ioc_inlbuf4 != NULL &&
1246 data->ioc_inllen4 == sizeof(__u64)) {
1247 struct ost_id ostid = { };
1249 ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4);
1250 rc = ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3);
1253 rc = ostid_to_fid(&fid, &ostid,
1254 ofd->ofd_lut.lut_lsd.lsd_osd_index);
1258 GOTO(out, rc = -EINVAL);
1261 if (!fid_is_sane(&fid))
1262 GOTO(out, rc = -EINVAL);
1264 fo = ofd_object_find(env, ofd, &fid);
1266 GOTO(out, rc = PTR_ERR(fo));
1268 if (!ofd_object_exists(fo))
1269 GOTO(out_fo, rc = -ENOENT);
1271 if (lu_object_remote(&fo->ofo_obj.do_lu))
1272 GOTO(out_fo, rc = -EREMOTE);
1274 version = dt_version_get(env, ofd_object_child(fo));
1276 GOTO(out_fo, rc = -EIO);
1278 *(dt_obj_version_t *)data->ioc_inlbuf2 = version;
1282 ofd_object_put(env, fo);
1288 * Implementation of obd_ops::o_iocontrol.
1290 * This is OFD ioctl handling function which is primary interface for
1291 * Lustre tools like lfs, lctl and lfsck.
1293 * \param[in] cmd ioctl command
1294 * \param[in] exp OBD export of OFD
1295 * \param[in] len not used
1296 * \param[in,out] karg buffer with data
1297 * \param[in] uarg not used
1299 * \retval 0 if successful
1300 * \retval negative value on error
1302 static int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1303 void *karg, void __user *uarg)
1306 struct ofd_device *ofd = ofd_exp(exp);
1307 struct obd_device *obd = ofd_obd(ofd);
1312 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
1313 rc = lu_env_init(&env, LCT_DT_THREAD);
1318 case OBD_IOC_ABORT_RECOVERY:
1319 CERROR("%s: aborting recovery\n", obd->obd_name);
1320 obd->obd_abort_recovery = 1;
1321 target_stop_recovery_thread(obd);
1324 CDEBUG(D_RPCTRACE, "syncing ost %s\n", obd->obd_name);
1325 rc = dt_sync(&env, ofd->ofd_osd);
1327 case OBD_IOC_SET_READONLY:
1328 rc = dt_sync(&env, ofd->ofd_osd);
1330 rc = dt_ro(&env, ofd->ofd_osd);
1332 case OBD_IOC_START_LFSCK: {
1333 struct obd_ioctl_data *data = karg;
1334 struct lfsck_start_param lsp;
1336 if (unlikely(data == NULL)) {
1341 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
1342 lsp.lsp_index_valid = 0;
1343 rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
1346 case OBD_IOC_STOP_LFSCK: {
1347 struct obd_ioctl_data *data = karg;
1348 struct lfsck_stop stop;
1350 stop.ls_status = LS_STOPPED;
1351 /* Old lfsck utils may pass NULL @stop. */
1352 if (data->ioc_inlbuf1 == NULL)
1356 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
1358 rc = lfsck_stop(&env, ofd->ofd_osd, &stop);
1361 case OBD_IOC_GET_OBJ_VERSION:
1362 rc = ofd_ioc_get_obj_version(&env, ofd, karg);
1365 CERROR("%s: not supported cmd = %#x\n", obd->obd_name, cmd);
1374 * Implementation of obd_ops::o_precleanup.
1376 * This function stops device activity before shutting it down. It is called
1377 * from a cleanup function upon forceful device cleanup. For OFD there are no
1378 * special actions, it just invokes target_recovery_cleanup().
1380 * \param[in] obd OBD device of OFD
1384 static int ofd_precleanup(struct obd_device *obd)
1387 target_cleanup_recovery(obd);
1392 * Implementation of obd_ops::o_ping.
1394 * This is OFD-specific part of OBD_PING request handling.
1395 * It controls Filter Modification Data (FMD) expiration each time PING is
1398 * \see ofd_fmd_expire() and ofd_fmd.c for details
1400 * \param[in] env execution environment
1401 * \param[in] exp OBD export of client
1405 static int ofd_ping(const struct lu_env *env, struct obd_export *exp)
1407 ofd_fmd_expire(exp);
1412 * Implementation of obd_ops::o_health_check.
1414 * This function checks the OFD device health - ability to respond on
1415 * incoming requests. There are two health_check methods:
1416 * - get statfs from the OSD. It checks just responsiveness of
1418 * - do write attempt on bottom device to check it is fully operational and
1419 * is not stuck. This is expensive method and requires special configuration
1420 * option --enable-health-write while building Lustre, it is turned off
1423 * \param[in] nul not used
1424 * \param[in] obd OBD device of OFD
1426 * \retval 0 if successful
1427 * \retval negative value in case of error
1429 static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd)
1431 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
1432 struct ofd_thread_info *info;
1434 #ifdef USE_HEALTH_CHECK_WRITE
1439 /* obd_proc_read_health pass NULL env, we need real one */
1440 rc = lu_env_init(&env, LCT_DT_THREAD);
1444 info = ofd_info_init(&env, NULL);
1445 rc = dt_statfs(&env, ofd->ofd_osd, &info->fti_u.osfs);
1449 if (info->fti_u.osfs.os_state & OS_STATE_READONLY)
1450 GOTO(out, rc = -EROFS);
1452 #ifdef USE_HEALTH_CHECK_WRITE
1453 OBD_ALLOC(info->fti_buf.lb_buf, PAGE_SIZE);
1454 if (info->fti_buf.lb_buf == NULL)
1455 GOTO(out, rc = -ENOMEM);
1457 info->fti_buf.lb_len = PAGE_SIZE;
1460 th = dt_trans_create(&env, ofd->ofd_osd);
1462 GOTO(out, rc = PTR_ERR(th));
1464 rc = dt_declare_record_write(&env, ofd->ofd_health_check_file,
1465 &info->fti_buf, info->fti_off, th);
1467 th->th_sync = 1; /* sync IO is needed */
1468 rc = dt_trans_start_local(&env, ofd->ofd_osd, th);
1470 rc = dt_record_write(&env, ofd->ofd_health_check_file,
1471 &info->fti_buf, &info->fti_off,
1474 dt_trans_stop(&env, ofd->ofd_osd, th);
1476 OBD_FREE(info->fti_buf.lb_buf, PAGE_SIZE);
1478 CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
1485 struct obd_ops ofd_obd_ops = {
1486 .o_owner = THIS_MODULE,
1487 .o_connect = ofd_obd_connect,
1488 .o_reconnect = ofd_obd_reconnect,
1489 .o_disconnect = ofd_obd_disconnect,
1490 .o_create = ofd_echo_create,
1491 .o_statfs = ofd_statfs,
1492 .o_setattr = ofd_echo_setattr,
1493 .o_preprw = ofd_preprw,
1494 .o_commitrw = ofd_commitrw,
1495 .o_destroy = ofd_echo_destroy,
1496 .o_init_export = ofd_init_export,
1497 .o_destroy_export = ofd_destroy_export,
1498 .o_postrecov = ofd_obd_postrecov,
1499 .o_getattr = ofd_echo_getattr,
1500 .o_iocontrol = ofd_iocontrol,
1501 .o_precleanup = ofd_precleanup,
1503 .o_health_check = ofd_health_check,
1504 .o_set_info_async = ofd_set_info_async,
1505 .o_get_info = ofd_get_info,