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) 2012, 2013, 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, const struct lu_fid *fid,
74 struct list_head *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 int mdc_setattr(struct obd_export *exp, struct md_op_data *op_data,
109 void *ea, int ealen, void *ea2, int ea2len,
110 struct ptlrpc_request **request, struct md_open_data **mod)
112 struct list_head cancels = LIST_HEAD_INIT(cancels);
113 struct ptlrpc_request *req;
114 struct mdc_rpc_lock *rpc_lock;
115 struct obd_device *obd = exp->exp_obd;
120 LASSERT(op_data != NULL);
122 bits = MDS_INODELOCK_UPDATE;
123 if (op_data->op_attr.ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID))
124 bits |= MDS_INODELOCK_LOOKUP;
125 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
126 (fid_is_sane(&op_data->op_fid1)) &&
127 !OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK_NET))
128 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
129 &cancels, LCK_EX, bits);
130 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
131 &RQF_MDS_REINT_SETATTR);
133 ldlm_lock_list_put(&cancels, l_bl_ast, count);
136 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
137 if ((op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) == 0)
138 req_capsule_set_size(&req->rq_pill, &RMF_MDT_EPOCH, RCL_CLIENT,
140 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, ealen);
141 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_CLIENT,
144 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
146 ptlrpc_request_free(req);
150 rpc_lock = obd->u.cli.cl_rpc_lock;
152 if (op_data->op_attr.ia_valid & (ATTR_MTIME | ATTR_CTIME))
153 CDEBUG(D_INODE, "setting mtime "CFS_TIME_T
154 ", ctime "CFS_TIME_T"\n",
155 LTIME_S(op_data->op_attr.ia_mtime),
156 LTIME_S(op_data->op_attr.ia_ctime));
157 mdc_setattr_pack(req, op_data, ea, ealen, ea2, ea2len);
159 ptlrpc_request_set_replen(req);
160 if (mod && (op_data->op_flags & MF_EPOCH_OPEN) &&
161 req->rq_import->imp_replayable)
163 LASSERT(*mod == NULL);
165 *mod = obd_mod_alloc();
167 DEBUG_REQ(D_ERROR, req, "Can't allocate "
171 req->rq_cb_data = *mod;
172 (*mod)->mod_open_req = req;
173 req->rq_commit_cb = mdc_commit_open;
174 (*mod)->mod_is_create = true;
176 * Take an extra reference on \var mod, it protects \var
177 * mod from being freed on eviction (commit callback is
178 * called despite rq_replay flag).
179 * Will be put on mdc_done_writing().
185 rc = mdc_reint(req, rpc_lock, LUSTRE_IMP_FULL);
187 /* Save the obtained info in the original RPC for the replay case. */
188 if (rc == 0 && (op_data->op_flags & MF_EPOCH_OPEN)) {
189 struct mdt_ioepoch *epoch;
190 struct mdt_body *body;
192 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
193 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
194 LASSERT(epoch != NULL);
195 LASSERT(body != NULL);
196 epoch->handle = body->handle;
197 epoch->ioepoch = body->ioepoch;
198 req->rq_replay_cb = mdc_replay_open;
199 /** bug 3633, open may be committed and estale answer is not error */
200 } else if (rc == -ESTALE && (op_data->op_flags & MF_SOM_CHANGE)) {
202 } else if (rc == -ERESTARTSYS) {
206 if (rc && req->rq_commit_cb) {
207 /* Put an extra reference on \var mod on error case. */
208 if (mod != NULL && *mod != NULL)
210 req->rq_commit_cb(req);
215 int mdc_create(struct obd_export *exp, struct md_op_data *op_data,
216 const void *data, int datalen, int mode, __u32 uid, __u32 gid,
217 cfs_cap_t cap_effective, __u64 rdev,
218 struct ptlrpc_request **request)
220 struct ptlrpc_request *req;
222 int count, resends = 0;
223 struct obd_import *import = exp->exp_obd->u.cli.cl_import;
224 int generation = import->imp_generation;
225 struct list_head cancels = LIST_HEAD_INIT(cancels);
228 /* For case if upper layer did not alloc fid, do it now. */
229 if (!fid_is_sane(&op_data->op_fid2)) {
231 * mdc_fid_alloc() may return errno 1 in case of switch to new
232 * sequence, handle this.
234 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
236 CERROR("Can't alloc new fid, rc %d\n", rc);
243 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
244 (fid_is_sane(&op_data->op_fid1)))
245 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
247 MDS_INODELOCK_UPDATE);
249 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
250 &RQF_MDS_REINT_CREATE_RMT_ACL);
252 ldlm_lock_list_put(&cancels, l_bl_ast, count);
255 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
256 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
257 op_data->op_namelen + 1);
258 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
259 data && datalen ? datalen : 0);
261 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
263 ptlrpc_request_free(req);
268 * mdc_create_pack() fills msg->bufs[1] with name and msg->bufs[2] with
269 * tgt, for symlinks or lov MD data.
271 mdc_create_pack(req, op_data, data, datalen, mode, uid,
272 gid, cap_effective, rdev);
274 ptlrpc_request_set_replen(req);
276 /* ask ptlrpc not to resend on EINPROGRESS since we have our own retry
278 req->rq_no_retry_einprogress = 1;
281 req->rq_generation_set = 1;
282 req->rq_import_generation = generation;
283 req->rq_sent = cfs_time_current_sec() + resends;
285 level = LUSTRE_IMP_FULL;
287 rc = mdc_reint(req, exp->exp_obd->u.cli.cl_rpc_lock, level);
289 /* Resend if we were told to. */
290 if (rc == -ERESTARTSYS) {
291 level = LUSTRE_IMP_RECOVER;
293 } else if (rc == -EINPROGRESS) {
294 /* Retry create infinitely until succeed or get other
296 ptlrpc_req_finished(req);
299 CDEBUG(D_HA, "%s: resend:%d create on "DFID"/"DFID"\n",
300 exp->exp_obd->obd_name, resends,
301 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2));
303 if (generation == import->imp_generation) {
306 CDEBUG(D_HA, "resend cross eviction\n");
309 } else if (rc == 0) {
310 struct mdt_body *body;
311 struct lustre_capa *capa;
313 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
315 if (body->valid & OBD_MD_FLMDSCAPA) {
316 capa = req_capsule_server_get(&req->rq_pill,
327 int mdc_unlink(struct obd_export *exp, struct md_op_data *op_data,
328 struct ptlrpc_request **request)
330 struct list_head cancels = LIST_HEAD_INIT(cancels);
331 struct obd_device *obd = class_exp2obd(exp);
332 struct ptlrpc_request *req = *request;
336 LASSERT(req == NULL);
338 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
339 (fid_is_sane(&op_data->op_fid1)) &&
340 !OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK_NET))
341 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
343 MDS_INODELOCK_UPDATE);
344 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
345 (fid_is_sane(&op_data->op_fid3)) &&
346 !OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK_NET))
347 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
350 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
351 &RQF_MDS_REINT_UNLINK);
353 ldlm_lock_list_put(&cancels, l_bl_ast, count);
356 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
357 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
358 op_data->op_namelen + 1);
360 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
362 ptlrpc_request_free(req);
366 mdc_unlink_pack(req, op_data);
368 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
369 obd->u.cli.cl_default_mds_easize);
370 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
371 obd->u.cli.cl_default_mds_cookiesize);
372 ptlrpc_request_set_replen(req);
376 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
377 if (rc == -ERESTARTSYS)
382 int mdc_link(struct obd_export *exp, struct md_op_data *op_data,
383 struct ptlrpc_request **request)
385 struct list_head cancels = LIST_HEAD_INIT(cancels);
386 struct obd_device *obd = exp->exp_obd;
387 struct ptlrpc_request *req;
391 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
392 (fid_is_sane(&op_data->op_fid2)))
393 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
395 MDS_INODELOCK_UPDATE);
396 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
397 (fid_is_sane(&op_data->op_fid1)))
398 count += mdc_resource_get_unused(exp, &op_data->op_fid1,
400 MDS_INODELOCK_UPDATE);
402 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_REINT_LINK);
404 ldlm_lock_list_put(&cancels, l_bl_ast, count);
407 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
408 mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
409 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
410 op_data->op_namelen + 1);
412 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
414 ptlrpc_request_free(req);
418 mdc_link_pack(req, op_data);
419 ptlrpc_request_set_replen(req);
421 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
423 if (rc == -ERESTARTSYS)
429 int mdc_rename(struct obd_export *exp, struct md_op_data *op_data,
430 const char *old, int oldlen, const char *new, int newlen,
431 struct ptlrpc_request **request)
433 struct list_head cancels = LIST_HEAD_INIT(cancels);
434 struct obd_device *obd = exp->exp_obd;
435 struct ptlrpc_request *req;
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_UPDATE);
444 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
445 (fid_is_sane(&op_data->op_fid2)))
446 count += mdc_resource_get_unused(exp, &op_data->op_fid2,
448 MDS_INODELOCK_UPDATE);
449 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
450 (fid_is_sane(&op_data->op_fid3)))
451 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
453 MDS_INODELOCK_LOOKUP);
454 if ((op_data->op_flags & MF_MDC_CANCEL_FID4) &&
455 (fid_is_sane(&op_data->op_fid4)))
456 count += mdc_resource_get_unused(exp, &op_data->op_fid4,
460 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
461 &RQF_MDS_REINT_RENAME);
463 ldlm_lock_list_put(&cancels, l_bl_ast, count);
467 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
468 mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
469 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT, oldlen + 1);
470 req_capsule_set_size(&req->rq_pill, &RMF_SYMTGT, RCL_CLIENT, newlen+1);
472 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
474 ptlrpc_request_free(req);
478 if (exp_connect_cancelset(exp) && req)
479 ldlm_cli_cancel_list(&cancels, count, req, 0);
481 mdc_rename_pack(req, op_data, old, oldlen, new, newlen);
483 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
484 obd->u.cli.cl_default_mds_easize);
485 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
486 obd->u.cli.cl_default_mds_cookiesize);
487 ptlrpc_request_set_replen(req);
489 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
491 if (rc == -ERESTARTSYS)