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) 2002, 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.
33 #define DEBUG_SUBSYSTEM S_MDC
35 #include <linux/module.h>
36 #include <linux/kernel.h>
38 #include <obd_class.h>
39 #include "mdc_internal.h"
40 #include <lustre_fid.h>
42 /* mdc_setattr does its own semaphore handling */
43 static int mdc_reint(struct ptlrpc_request *request, int level)
47 request->rq_send_state = level;
49 mdc_get_mod_rpc_slot(request, NULL);
50 rc = ptlrpc_queue_wait(request);
51 mdc_put_mod_rpc_slot(request, NULL);
53 CDEBUG(D_INFO, "error in handling %d\n", rc);
54 else if (!req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY)) {
60 /* Find and cancel locally locks matched by inode @bits & @mode in the resource
61 * found by @fid. Found locks are added into @cancel list. Returns the amount of
62 * locks added to @cancels list. */
63 int mdc_resource_get_unused(struct obd_export *exp, const struct lu_fid *fid,
64 struct list_head *cancels, enum ldlm_mode mode,
67 struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
68 union ldlm_policy_data policy = { {0} };
69 struct ldlm_res_id res_id;
70 struct ldlm_resource *res;
74 /* Return, i.e. cancel nothing, only if ELC is supported (flag in
75 * export) but disabled through procfs (flag in NS).
77 * This distinguishes from a case when ELC is not supported originally,
78 * when we still want to cancel locks in advance and just cancel them
79 * locally, without sending any RPC. */
80 if (exp_connect_cancelset(exp) && !ns_connect_cancelset(ns))
83 fid_build_reg_res_name(fid, &res_id);
84 res = ldlm_resource_get(exp->exp_obd->obd_namespace,
88 LDLM_RESOURCE_ADDREF(res);
89 /* Initialize ibits lock policy. */
90 policy.l_inodebits.bits = bits;
91 count = ldlm_cancel_resource_local(res, cancels, &policy,
93 LDLM_RESOURCE_DELREF(res);
94 ldlm_resource_putref(res);
98 int mdc_setattr(struct obd_export *exp, struct md_op_data *op_data,
99 void *ea, size_t ealen, struct ptlrpc_request **request)
101 struct list_head cancels = LIST_HEAD_INIT(cancels);
102 struct ptlrpc_request *req;
107 LASSERT(op_data != NULL);
109 bits = MDS_INODELOCK_UPDATE;
110 if (op_data->op_attr.ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID))
111 bits |= MDS_INODELOCK_LOOKUP;
112 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
113 (fid_is_sane(&op_data->op_fid1)))
114 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
115 &cancels, LCK_EX, bits);
116 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
117 &RQF_MDS_REINT_SETATTR);
119 ldlm_lock_list_put(&cancels, l_bl_ast, count);
123 req_capsule_set_size(&req->rq_pill, &RMF_MDT_EPOCH, RCL_CLIENT, 0);
124 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, ealen);
125 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_CLIENT, 0);
127 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
129 ptlrpc_request_free(req);
133 if (op_data->op_attr.ia_valid & (ATTR_MTIME | ATTR_CTIME))
134 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld\n",
135 (s64)op_data->op_attr.ia_mtime.tv_sec,
136 (s64)op_data->op_attr.ia_ctime.tv_sec);
137 mdc_setattr_pack(req, op_data, ea, ealen);
139 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
140 MIN(req->rq_import->imp_connect_data.ocd_max_easize,
142 ptlrpc_request_set_replen(req);
144 rc = mdc_reint(req, LUSTRE_IMP_FULL);
145 if (rc == -ERESTARTSYS)
153 int mdc_create(struct obd_export *exp, struct md_op_data *op_data,
154 const void *data, size_t datalen,
155 umode_t mode, uid_t uid, gid_t gid,
156 cfs_cap_t cap_effective, __u64 rdev,
157 struct ptlrpc_request **request)
159 struct ptlrpc_request *req;
161 int count, resends = 0;
162 struct obd_import *import = exp->exp_obd->u.cli.cl_import;
163 int generation = import->imp_generation;
164 struct list_head cancels = LIST_HEAD_INIT(cancels);
167 /* For case if upper layer did not alloc fid, do it now. */
168 if (!fid_is_sane(&op_data->op_fid2)) {
170 * mdc_fid_alloc() may return errno 1 in case of switch to new
171 * sequence, handle this.
173 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
180 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
181 (fid_is_sane(&op_data->op_fid1)))
182 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
184 MDS_INODELOCK_UPDATE);
186 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
187 &RQF_MDS_REINT_CREATE_ACL);
189 ldlm_lock_list_put(&cancels, l_bl_ast, count);
193 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
194 op_data->op_namelen + 1);
195 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
196 data && datalen ? datalen : 0);
198 req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
199 RCL_CLIENT, op_data->op_file_secctx_name != NULL ?
200 strlen(op_data->op_file_secctx_name) + 1 : 0);
202 req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX, RCL_CLIENT,
203 op_data->op_file_secctx_size);
205 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
207 ptlrpc_request_free(req);
212 * mdc_create_pack() fills msg->bufs[1] with name and msg->bufs[2] with
213 * tgt, for symlinks or lov MD data.
215 mdc_create_pack(req, op_data, data, datalen, mode, uid,
216 gid, cap_effective, rdev);
218 ptlrpc_request_set_replen(req);
220 /* ask ptlrpc not to resend on EINPROGRESS since we have our own retry
222 req->rq_no_retry_einprogress = 1;
225 req->rq_generation_set = 1;
226 req->rq_import_generation = generation;
227 req->rq_sent = ktime_get_real_seconds() + resends;
229 level = LUSTRE_IMP_FULL;
231 rc = mdc_reint(req, level);
233 /* Resend if we were told to. */
234 if (rc == -ERESTARTSYS) {
235 level = LUSTRE_IMP_RECOVER;
237 } else if (rc == -EINPROGRESS) {
238 /* Retry create infinitely until succeed or get other
239 * error code or interrupted. */
240 ptlrpc_req_finished(req);
241 if (generation == import->imp_generation) {
242 if (signal_pending(current))
246 CDEBUG(D_HA, "%s: resend:%d create on "DFID"/"DFID"\n",
247 exp->exp_obd->obd_name, resends,
248 PFID(&op_data->op_fid1),
249 PFID(&op_data->op_fid2));
252 CDEBUG(D_HA, "resend cross eviction\n");
261 int mdc_unlink(struct obd_export *exp, struct md_op_data *op_data,
262 struct ptlrpc_request **request)
264 struct list_head cancels = LIST_HEAD_INIT(cancels);
265 struct obd_device *obd = class_exp2obd(exp);
266 struct ptlrpc_request *req = *request;
270 LASSERT(req == NULL);
272 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
273 (fid_is_sane(&op_data->op_fid1)))
274 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
276 MDS_INODELOCK_UPDATE);
277 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
278 (fid_is_sane(&op_data->op_fid3)))
279 /* don't cancel DoM lock which may cause data flush */
280 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
283 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
284 &RQF_MDS_REINT_UNLINK);
286 ldlm_lock_list_put(&cancels, l_bl_ast, count);
290 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
291 op_data->op_namelen + 1);
293 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
295 ptlrpc_request_free(req);
299 mdc_unlink_pack(req, op_data);
301 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
302 obd->u.cli.cl_default_mds_easize);
303 ptlrpc_request_set_replen(req);
307 rc = mdc_reint(req, LUSTRE_IMP_FULL);
308 if (rc == -ERESTARTSYS)
313 int mdc_link(struct obd_export *exp, struct md_op_data *op_data,
314 struct ptlrpc_request **request)
316 struct list_head cancels = LIST_HEAD_INIT(cancels);
317 struct ptlrpc_request *req;
321 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
322 (fid_is_sane(&op_data->op_fid2)))
323 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
325 MDS_INODELOCK_UPDATE);
326 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
327 (fid_is_sane(&op_data->op_fid1)))
328 count += mdc_resource_get_unused(exp, &op_data->op_fid1,
330 MDS_INODELOCK_UPDATE);
332 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_REINT_LINK);
334 ldlm_lock_list_put(&cancels, l_bl_ast, count);
338 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
339 op_data->op_namelen + 1);
341 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
343 ptlrpc_request_free(req);
347 mdc_link_pack(req, op_data);
348 ptlrpc_request_set_replen(req);
350 rc = mdc_reint(req, LUSTRE_IMP_FULL);
352 if (rc == -ERESTARTSYS)
358 int mdc_rename(struct obd_export *exp, struct md_op_data *op_data,
359 const char *old, size_t oldlen, const char *new, size_t newlen,
360 struct ptlrpc_request **request)
362 struct list_head cancels = LIST_HEAD_INIT(cancels);
363 struct obd_device *obd = exp->exp_obd;
364 struct ptlrpc_request *req;
369 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
370 (fid_is_sane(&op_data->op_fid1)))
371 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
373 MDS_INODELOCK_UPDATE);
374 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
375 (fid_is_sane(&op_data->op_fid2)))
376 count += mdc_resource_get_unused(exp, &op_data->op_fid2,
378 MDS_INODELOCK_UPDATE);
379 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
380 (fid_is_sane(&op_data->op_fid3)))
381 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
383 MDS_INODELOCK_LOOKUP);
384 if ((op_data->op_flags & MF_MDC_CANCEL_FID4) &&
385 (fid_is_sane(&op_data->op_fid4)))
386 count += mdc_resource_get_unused(exp, &op_data->op_fid4,
390 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
391 op_data->op_cli_flags & CLI_MIGRATE ?
392 &RQF_MDS_REINT_MIGRATE : &RQF_MDS_REINT_RENAME);
394 ldlm_lock_list_put(&cancels, l_bl_ast, count);
398 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT, oldlen + 1);
399 req_capsule_set_size(&req->rq_pill, &RMF_SYMTGT, RCL_CLIENT, newlen+1);
400 if (op_data->op_cli_flags & CLI_MIGRATE)
401 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
402 op_data->op_data_size);
404 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
406 ptlrpc_request_free(req);
410 if (exp_connect_cancelset(exp) && req)
411 ldlm_cli_cancel_list(&cancels, count, req, 0);
413 if (op_data->op_cli_flags & CLI_MIGRATE)
414 mdc_migrate_pack(req, op_data, old, oldlen);
416 mdc_rename_pack(req, op_data, old, oldlen, new, newlen);
418 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
419 obd->u.cli.cl_default_mds_easize);
420 ptlrpc_request_set_replen(req);
422 rc = mdc_reint(req, LUSTRE_IMP_FULL);
424 if (rc == -ERESTARTSYS)
430 int mdc_file_resync(struct obd_export *exp, struct md_op_data *op_data)
432 struct list_head cancels = LIST_HEAD_INIT(cancels);
433 struct ptlrpc_request *req;
434 struct ldlm_lock *lock;
435 struct mdt_rec_resync *rec;
439 if (op_data->op_flags & MF_MDC_CANCEL_FID1 &&
440 fid_is_sane(&op_data->op_fid1))
441 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
443 MDS_INODELOCK_LAYOUT);
445 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
446 &RQF_MDS_REINT_RESYNC);
448 ldlm_lock_list_put(&cancels, l_bl_ast, count);
452 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
454 ptlrpc_request_free(req);
458 CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
459 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
460 rec->rs_opcode = REINT_RESYNC;
461 rec->rs_fsuid = op_data->op_fsuid;
462 rec->rs_fsgid = op_data->op_fsgid;
463 rec->rs_cap = op_data->op_cap;
464 rec->rs_fid = op_data->op_fid1;
465 rec->rs_bias = op_data->op_bias;
466 rec->rs_mirror_id = op_data->op_mirror_id;
468 lock = ldlm_handle2lock(&op_data->op_lease_handle);
470 rec->rs_lease_handle = lock->l_remote_handle;
474 ptlrpc_request_set_replen(req);
476 rc = mdc_reint(req, LUSTRE_IMP_FULL);
477 if (rc == -ERESTARTSYS)
480 ptlrpc_req_finished(req);