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) 2011, 2015, 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 "../include/obd_class.h"
39 #include "mdc_internal.h"
40 #include "../include/lustre_fid.h"
42 /* mdc_setattr does its own semaphore handling */
43 static int mdc_reint(struct ptlrpc_request *request,
44 struct mdc_rpc_lock *rpc_lock,
49 request->rq_send_state = level;
51 mdc_get_rpc_lock(rpc_lock, NULL);
52 rc = ptlrpc_queue_wait(request);
53 mdc_put_rpc_lock(rpc_lock, NULL);
55 CDEBUG(D_INFO, "error in handling %d\n", rc);
56 else if (!req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY))
62 /* Find and cancel locally locks matched by inode @bits & @mode in the resource
63 * found by @fid. Found locks are added into @cancel list. Returns the amount of
64 * locks added to @cancels list.
66 int mdc_resource_get_unused(struct obd_export *exp, const struct lu_fid *fid,
67 struct list_head *cancels, enum ldlm_mode mode,
70 struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
71 ldlm_policy_data_t policy = {};
72 struct ldlm_res_id res_id;
73 struct ldlm_resource *res;
76 /* Return, i.e. cancel nothing, only if ELC is supported (flag in
77 * export) but disabled through procfs (flag in NS).
79 * This distinguishes from a case when ELC is not supported originally,
80 * when we still want to cancel locks in advance and just cancel them
81 * locally, without sending any RPC.
83 if (exp_connect_cancelset(exp) && !ns_connect_cancelset(ns))
86 fid_build_reg_res_name(fid, &res_id);
87 res = ldlm_resource_get(exp->exp_obd->obd_namespace,
91 LDLM_RESOURCE_ADDREF(res);
92 /* Initialize ibits lock policy. */
93 policy.l_inodebits.bits = bits;
94 count = ldlm_cancel_resource_local(res, cancels, &policy,
96 LDLM_RESOURCE_DELREF(res);
97 ldlm_resource_putref(res);
101 int mdc_setattr(struct obd_export *exp, struct md_op_data *op_data,
102 void *ea, size_t ealen, void *ea2, size_t ea2len,
103 struct ptlrpc_request **request, struct md_open_data **mod)
106 struct ptlrpc_request *req;
107 struct mdc_rpc_lock *rpc_lock;
108 struct obd_device *obd = exp->exp_obd;
112 bits = MDS_INODELOCK_UPDATE;
113 if (op_data->op_attr.ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID))
114 bits |= MDS_INODELOCK_LOOKUP;
115 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
116 (fid_is_sane(&op_data->op_fid1)))
117 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
118 &cancels, LCK_EX, bits);
119 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
120 &RQF_MDS_REINT_SETATTR);
122 ldlm_lock_list_put(&cancels, l_bl_ast, count);
125 if ((op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) == 0)
126 req_capsule_set_size(&req->rq_pill, &RMF_MDT_EPOCH, RCL_CLIENT,
128 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, ealen);
129 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_CLIENT,
132 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
134 ptlrpc_request_free(req);
138 rpc_lock = obd->u.cli.cl_rpc_lock;
140 if (op_data->op_attr.ia_valid & (ATTR_MTIME | ATTR_CTIME))
141 CDEBUG(D_INODE, "setting mtime %ld, ctime %ld\n",
142 LTIME_S(op_data->op_attr.ia_mtime),
143 LTIME_S(op_data->op_attr.ia_ctime));
144 mdc_setattr_pack(req, op_data, ea, ealen, ea2, ea2len);
146 ptlrpc_request_set_replen(req);
147 if (mod && (op_data->op_flags & MF_EPOCH_OPEN) &&
148 req->rq_import->imp_replayable) {
151 *mod = obd_mod_alloc();
153 DEBUG_REQ(D_ERROR, req, "Can't allocate md_open_data");
156 req->rq_cb_data = *mod;
157 (*mod)->mod_open_req = req;
158 req->rq_commit_cb = mdc_commit_open;
159 (*mod)->mod_is_create = true;
161 * Take an extra reference on \var mod, it protects \var
162 * mod from being freed on eviction (commit callback is
163 * called despite rq_replay flag).
164 * Will be put on mdc_done_writing().
170 rc = mdc_reint(req, rpc_lock, LUSTRE_IMP_FULL);
172 /* Save the obtained info in the original RPC for the replay case. */
173 if (rc == 0 && (op_data->op_flags & MF_EPOCH_OPEN)) {
174 struct mdt_ioepoch *epoch;
175 struct mdt_body *body;
177 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
178 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
179 epoch->handle = body->mbo_handle;
180 epoch->ioepoch = body->mbo_ioepoch;
181 req->rq_replay_cb = mdc_replay_open;
182 /** bug 3633, open may be committed and estale answer is not error */
183 } else if (rc == -ESTALE && (op_data->op_flags & MF_SOM_CHANGE)) {
185 } else if (rc == -ERESTARTSYS) {
189 if (rc && req->rq_commit_cb) {
190 /* Put an extra reference on \var mod on error case. */
193 req->rq_commit_cb(req);
198 int mdc_create(struct obd_export *exp, struct md_op_data *op_data,
199 const void *data, size_t datalen, umode_t mode,
200 uid_t uid, gid_t gid, cfs_cap_t cap_effective,
201 __u64 rdev, struct ptlrpc_request **request)
203 struct ptlrpc_request *req;
205 int count, resends = 0;
206 struct obd_import *import = exp->exp_obd->u.cli.cl_import;
207 int generation = import->imp_generation;
210 /* For case if upper layer did not alloc fid, do it now. */
211 if (!fid_is_sane(&op_data->op_fid2)) {
213 * mdc_fid_alloc() may return errno 1 in case of switch to new
214 * sequence, handle this.
216 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
223 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
224 (fid_is_sane(&op_data->op_fid1)))
225 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
227 MDS_INODELOCK_UPDATE);
229 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
230 &RQF_MDS_REINT_CREATE_ACL);
232 ldlm_lock_list_put(&cancels, l_bl_ast, count);
235 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
236 op_data->op_namelen + 1);
237 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
238 data && datalen ? datalen : 0);
240 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
242 ptlrpc_request_free(req);
247 * mdc_create_pack() fills msg->bufs[1] with name and msg->bufs[2] with
248 * tgt, for symlinks or lov MD data.
250 mdc_create_pack(req, op_data, data, datalen, mode, uid,
251 gid, cap_effective, rdev);
253 ptlrpc_request_set_replen(req);
255 /* ask ptlrpc not to resend on EINPROGRESS since we have our own retry
258 req->rq_no_retry_einprogress = 1;
261 req->rq_generation_set = 1;
262 req->rq_import_generation = generation;
263 req->rq_sent = ktime_get_real_seconds() + resends;
265 level = LUSTRE_IMP_FULL;
267 rc = mdc_reint(req, exp->exp_obd->u.cli.cl_rpc_lock, level);
269 /* Resend if we were told to. */
270 if (rc == -ERESTARTSYS) {
271 level = LUSTRE_IMP_RECOVER;
273 } else if (rc == -EINPROGRESS) {
274 /* Retry create infinitely until succeed or get other
277 ptlrpc_req_finished(req);
280 CDEBUG(D_HA, "%s: resend:%d create on "DFID"/"DFID"\n",
281 exp->exp_obd->obd_name, resends,
282 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2));
284 if (generation == import->imp_generation) {
287 CDEBUG(D_HA, "resend cross eviction\n");
296 int mdc_unlink(struct obd_export *exp, struct md_op_data *op_data,
297 struct ptlrpc_request **request)
300 struct obd_device *obd = class_exp2obd(exp);
301 struct ptlrpc_request *req = *request;
306 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
307 (fid_is_sane(&op_data->op_fid1)))
308 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
310 MDS_INODELOCK_UPDATE);
311 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
312 (fid_is_sane(&op_data->op_fid3)))
313 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
316 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
317 &RQF_MDS_REINT_UNLINK);
319 ldlm_lock_list_put(&cancels, l_bl_ast, count);
322 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
323 op_data->op_namelen + 1);
325 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
327 ptlrpc_request_free(req);
331 mdc_unlink_pack(req, op_data);
333 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
334 obd->u.cli.cl_default_mds_easize);
335 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
336 obd->u.cli.cl_default_mds_cookiesize);
337 ptlrpc_request_set_replen(req);
341 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
342 if (rc == -ERESTARTSYS)
347 int mdc_link(struct obd_export *exp, struct md_op_data *op_data,
348 struct ptlrpc_request **request)
351 struct obd_device *obd = exp->exp_obd;
352 struct ptlrpc_request *req;
355 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
356 (fid_is_sane(&op_data->op_fid2)))
357 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
359 MDS_INODELOCK_UPDATE);
360 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
361 (fid_is_sane(&op_data->op_fid1)))
362 count += mdc_resource_get_unused(exp, &op_data->op_fid1,
364 MDS_INODELOCK_UPDATE);
366 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_REINT_LINK);
368 ldlm_lock_list_put(&cancels, l_bl_ast, count);
371 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
372 op_data->op_namelen + 1);
374 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
376 ptlrpc_request_free(req);
380 mdc_link_pack(req, op_data);
381 ptlrpc_request_set_replen(req);
383 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
385 if (rc == -ERESTARTSYS)
391 int mdc_rename(struct obd_export *exp, struct md_op_data *op_data,
392 const char *old, size_t oldlen, const char *new, size_t newlen,
393 struct ptlrpc_request **request)
396 struct obd_device *obd = exp->exp_obd;
397 struct ptlrpc_request *req;
400 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
401 (fid_is_sane(&op_data->op_fid1)))
402 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
404 MDS_INODELOCK_UPDATE);
405 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
406 (fid_is_sane(&op_data->op_fid2)))
407 count += mdc_resource_get_unused(exp, &op_data->op_fid2,
409 MDS_INODELOCK_UPDATE);
410 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
411 (fid_is_sane(&op_data->op_fid3)))
412 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
414 MDS_INODELOCK_LOOKUP);
415 if ((op_data->op_flags & MF_MDC_CANCEL_FID4) &&
416 (fid_is_sane(&op_data->op_fid4)))
417 count += mdc_resource_get_unused(exp, &op_data->op_fid4,
421 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
422 &RQF_MDS_REINT_RENAME);
424 ldlm_lock_list_put(&cancels, l_bl_ast, count);
428 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT, oldlen + 1);
429 req_capsule_set_size(&req->rq_pill, &RMF_SYMTGT, RCL_CLIENT,
432 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
434 ptlrpc_request_free(req);
438 if (exp_connect_cancelset(exp) && req)
439 ldlm_cli_cancel_list(&cancels, count, req, 0);
441 mdc_rename_pack(req, op_data, old, oldlen, new, newlen);
443 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
444 obd->u.cli.cl_default_mds_easize);
445 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
446 obd->u.cli.cl_default_mds_cookiesize);
447 ptlrpc_request_set_replen(req);
449 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
451 if (rc == -ERESTARTSYS)