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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_MDC
40 # include <linux/module.h>
41 # include <linux/kernel.h>
43 # include <liblustre.h>
46 #include <obd_class.h>
47 #include "mdc_internal.h"
48 #include <lustre_fid.h>
50 /* mdc_setattr does its own semaphore handling */
51 static int mdc_reint(struct ptlrpc_request *request,
52 struct mdc_rpc_lock *rpc_lock,
57 request->rq_send_state = level;
59 mdc_get_rpc_lock(rpc_lock, NULL);
60 rc = ptlrpc_queue_wait(request);
61 mdc_put_rpc_lock(rpc_lock, NULL);
63 CDEBUG(D_INFO, "error in handling %d\n", rc);
64 else if (!req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY)) {
70 /* Find and cancel locally locks matched by inode @bits & @mode in the resource
71 * found by @fid. Found locks are added into @cancel list. Returns the amount of
72 * locks added to @cancels list. */
73 int mdc_resource_get_unused(struct obd_export *exp, struct lu_fid *fid,
74 cfs_list_t *cancels, ldlm_mode_t mode,
77 struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
78 ldlm_policy_data_t policy = {{0}};
79 struct ldlm_res_id res_id;
80 struct ldlm_resource *res;
84 /* Return, i.e. cancel nothing, only if ELC is supported (flag in
85 * export) but disabled through procfs (flag in NS).
87 * This distinguishes from a case when ELC is not supported originally,
88 * when we still want to cancel locks in advance and just cancel them
89 * locally, without sending any RPC. */
90 if (exp_connect_cancelset(exp) && !ns_connect_cancelset(ns))
93 fid_build_reg_res_name(fid, &res_id);
94 res = ldlm_resource_get(exp->exp_obd->obd_namespace,
98 LDLM_RESOURCE_ADDREF(res);
99 /* Initialize ibits lock policy. */
100 policy.l_inodebits.bits = bits;
101 count = ldlm_cancel_resource_local(res, cancels, &policy,
103 LDLM_RESOURCE_DELREF(res);
104 ldlm_resource_putref(res);
108 static int mdc_prep_elc_req(struct obd_export *exp, struct ptlrpc_request *req,
109 cfs_list_t *cancels, int count)
111 return ldlm_prep_elc_req(exp, req, LUSTRE_MDS_VERSION, MDS_REINT,
115 int mdc_setattr(struct obd_export *exp, struct md_op_data *op_data,
116 void *ea, int ealen, void *ea2, int ea2len,
117 struct ptlrpc_request **request, struct md_open_data **mod)
119 CFS_LIST_HEAD(cancels);
120 struct ptlrpc_request *req;
121 struct mdc_rpc_lock *rpc_lock;
122 struct obd_device *obd = exp->exp_obd;
127 LASSERT(op_data != NULL);
129 bits = MDS_INODELOCK_UPDATE;
130 if (op_data->op_attr.ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID))
131 bits |= MDS_INODELOCK_LOOKUP;
132 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
133 (fid_is_sane(&op_data->op_fid1)) &&
134 !OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK_NET))
135 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
136 &cancels, LCK_EX, bits);
137 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
138 &RQF_MDS_REINT_SETATTR);
140 ldlm_lock_list_put(&cancels, l_bl_ast, count);
143 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
144 if ((op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) == 0)
145 req_capsule_set_size(&req->rq_pill, &RMF_MDT_EPOCH, RCL_CLIENT,
147 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, ealen);
148 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_CLIENT,
151 rc = mdc_prep_elc_req(exp, req, &cancels, count);
153 ptlrpc_request_free(req);
157 rpc_lock = obd->u.cli.cl_rpc_lock;
159 if (op_data->op_attr.ia_valid & (ATTR_MTIME | ATTR_CTIME))
160 CDEBUG(D_INODE, "setting mtime "CFS_TIME_T
161 ", ctime "CFS_TIME_T"\n",
162 LTIME_S(op_data->op_attr.ia_mtime),
163 LTIME_S(op_data->op_attr.ia_ctime));
164 mdc_setattr_pack(req, op_data, ea, ealen, ea2, ea2len);
166 ptlrpc_request_set_replen(req);
167 if (mod && (op_data->op_flags & MF_EPOCH_OPEN) &&
168 req->rq_import->imp_replayable)
170 LASSERT(*mod == NULL);
172 *mod = obd_mod_alloc();
174 DEBUG_REQ(D_ERROR, req, "Can't allocate "
178 req->rq_cb_data = *mod;
179 (*mod)->mod_open_req = req;
180 req->rq_commit_cb = mdc_commit_open;
182 * Take an extra reference on \var mod, it protects \var
183 * mod from being freed on eviction (commit callback is
184 * called despite rq_replay flag).
185 * Will be put on mdc_done_writing().
191 rc = mdc_reint(req, rpc_lock, LUSTRE_IMP_FULL);
193 /* Save the obtained info in the original RPC for the replay case. */
194 if (rc == 0 && (op_data->op_flags & MF_EPOCH_OPEN)) {
195 struct mdt_ioepoch *epoch;
196 struct mdt_body *body;
198 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
199 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
200 LASSERT(epoch != NULL);
201 LASSERT(body != NULL);
202 epoch->handle = body->handle;
203 epoch->ioepoch = body->ioepoch;
204 req->rq_replay_cb = mdc_replay_open;
205 /** bug 3633, open may be committed and estale answer is not error */
206 } else if (rc == -ESTALE && (op_data->op_flags & MF_SOM_CHANGE)) {
208 } else if (rc == -ERESTARTSYS) {
212 if (rc && req->rq_commit_cb) {
213 /* Put an extra reference on \var mod on error case. */
215 req->rq_commit_cb(req);
220 int mdc_create(struct obd_export *exp, struct md_op_data *op_data,
221 const void *data, int datalen, int mode, __u32 uid, __u32 gid,
222 cfs_cap_t cap_effective, __u64 rdev,
223 struct ptlrpc_request **request)
225 struct ptlrpc_request *req;
227 int count, resends = 0;
228 struct obd_import *import = exp->exp_obd->u.cli.cl_import;
229 int generation = import->imp_generation;
230 CFS_LIST_HEAD(cancels);
233 /* For case if upper layer did not alloc fid, do it now. */
234 if (!fid_is_sane(&op_data->op_fid2)) {
236 * mdc_fid_alloc() may return errno 1 in case of switch to new
237 * sequence, handle this.
239 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
241 CERROR("Can't alloc new fid, rc %d\n", rc);
248 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
249 (fid_is_sane(&op_data->op_fid1)))
250 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
252 MDS_INODELOCK_UPDATE);
254 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
255 &RQF_MDS_REINT_CREATE_RMT_ACL);
257 ldlm_lock_list_put(&cancels, l_bl_ast, count);
260 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
261 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
262 op_data->op_namelen + 1);
263 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
264 data && datalen ? datalen : 0);
266 rc = mdc_prep_elc_req(exp, req, &cancels, count);
268 ptlrpc_request_free(req);
273 * mdc_create_pack() fills msg->bufs[1] with name and msg->bufs[2] with
274 * tgt, for symlinks or lov MD data.
276 mdc_create_pack(req, op_data, data, datalen, mode, uid,
277 gid, cap_effective, rdev);
279 ptlrpc_request_set_replen(req);
281 /* ask ptlrpc not to resend on EINPROGRESS since we have our own retry
283 req->rq_no_retry_einprogress = 1;
286 req->rq_generation_set = 1;
287 req->rq_import_generation = generation;
288 req->rq_sent = cfs_time_current_sec() + resends;
290 level = LUSTRE_IMP_FULL;
292 rc = mdc_reint(req, exp->exp_obd->u.cli.cl_rpc_lock, level);
294 /* Resend if we were told to. */
295 if (rc == -ERESTARTSYS) {
296 level = LUSTRE_IMP_RECOVER;
298 } else if (rc == -EINPROGRESS) {
299 /* Retry create infinitely until succeed or get other
301 ptlrpc_req_finished(req);
304 CDEBUG(D_HA, "%s: resend:%d create on "DFID"/"DFID"\n",
305 exp->exp_obd->obd_name, resends,
306 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2));
308 if (generation == import->imp_generation) {
311 CDEBUG(D_HA, "resend cross eviction\n");
314 } else if (rc == 0) {
315 struct mdt_body *body;
316 struct lustre_capa *capa;
318 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
320 if (body->valid & OBD_MD_FLMDSCAPA) {
321 capa = req_capsule_server_get(&req->rq_pill,
332 int mdc_unlink(struct obd_export *exp, struct md_op_data *op_data,
333 struct ptlrpc_request **request)
335 CFS_LIST_HEAD(cancels);
336 struct obd_device *obd = class_exp2obd(exp);
337 struct ptlrpc_request *req = *request;
341 LASSERT(req == NULL);
343 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
344 (fid_is_sane(&op_data->op_fid1)) &&
345 !OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK_NET))
346 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
348 MDS_INODELOCK_UPDATE);
349 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
350 (fid_is_sane(&op_data->op_fid3)) &&
351 !OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK_NET))
352 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
355 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
356 &RQF_MDS_REINT_UNLINK);
358 ldlm_lock_list_put(&cancels, l_bl_ast, count);
361 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
362 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
363 op_data->op_namelen + 1);
365 rc = mdc_prep_elc_req(exp, req, &cancels, count);
367 ptlrpc_request_free(req);
371 mdc_unlink_pack(req, op_data);
373 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
374 obd->u.cli.cl_max_mds_easize);
375 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
376 obd->u.cli.cl_max_mds_cookiesize);
377 ptlrpc_request_set_replen(req);
381 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
382 if (rc == -ERESTARTSYS)
387 int mdc_link(struct obd_export *exp, struct md_op_data *op_data,
388 struct ptlrpc_request **request)
390 CFS_LIST_HEAD(cancels);
391 struct obd_device *obd = exp->exp_obd;
392 struct ptlrpc_request *req;
396 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
397 (fid_is_sane(&op_data->op_fid2)))
398 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
400 MDS_INODELOCK_UPDATE);
401 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
402 (fid_is_sane(&op_data->op_fid1)))
403 count += mdc_resource_get_unused(exp, &op_data->op_fid1,
405 MDS_INODELOCK_UPDATE);
407 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_REINT_LINK);
409 ldlm_lock_list_put(&cancels, l_bl_ast, count);
412 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
413 mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
414 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
415 op_data->op_namelen + 1);
417 rc = mdc_prep_elc_req(exp, req, &cancels, count);
419 ptlrpc_request_free(req);
423 mdc_link_pack(req, op_data);
424 ptlrpc_request_set_replen(req);
426 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
428 if (rc == -ERESTARTSYS)
434 int mdc_rename(struct obd_export *exp, struct md_op_data *op_data,
435 const char *old, int oldlen, const char *new, int newlen,
436 struct ptlrpc_request **request)
438 CFS_LIST_HEAD(cancels);
439 struct obd_device *obd = exp->exp_obd;
440 struct ptlrpc_request *req;
444 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
445 (fid_is_sane(&op_data->op_fid1)))
446 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
448 MDS_INODELOCK_UPDATE);
449 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
450 (fid_is_sane(&op_data->op_fid2)))
451 count += mdc_resource_get_unused(exp, &op_data->op_fid2,
453 MDS_INODELOCK_UPDATE);
454 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
455 (fid_is_sane(&op_data->op_fid3)))
456 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
458 MDS_INODELOCK_LOOKUP);
459 if ((op_data->op_flags & MF_MDC_CANCEL_FID4) &&
460 (fid_is_sane(&op_data->op_fid4)))
461 count += mdc_resource_get_unused(exp, &op_data->op_fid4,
465 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
466 &RQF_MDS_REINT_RENAME);
468 ldlm_lock_list_put(&cancels, l_bl_ast, count);
472 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
473 mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
474 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT, oldlen + 1);
475 req_capsule_set_size(&req->rq_pill, &RMF_SYMTGT, RCL_CLIENT, newlen+1);
477 rc = mdc_prep_elc_req(exp, req, &cancels, count);
479 ptlrpc_request_free(req);
483 if (exp_connect_cancelset(exp) && req)
484 ldlm_cli_cancel_list(&cancels, count, req, 0);
486 mdc_rename_pack(req, op_data, old, oldlen, new, newlen);
488 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
489 obd->u.cli.cl_max_mds_easize);
490 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
491 obd->u.cli.cl_max_mds_cookiesize);
492 ptlrpc_request_set_replen(req);
494 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
496 if (rc == -ERESTARTSYS)