1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 # define EXPORT_SYMTAB
40 #define DEBUG_SUBSYSTEM S_MDC
43 #ifndef AUTOCONF_INCLUDED
44 # include <linux/config.h>
46 # include <linux/module.h>
47 # include <linux/kernel.h>
49 # include <liblustre.h>
52 #include <obd_class.h>
53 #include "mdc_internal.h"
54 #include <lustre_fid.h>
56 /* mdc_setattr does its own semaphore handling */
57 static int mdc_reint(struct ptlrpc_request *request,
58 struct mdc_rpc_lock *rpc_lock,
63 request->rq_send_state = level;
65 mdc_get_rpc_lock(rpc_lock, NULL);
66 rc = ptlrpc_queue_wait(request);
67 mdc_put_rpc_lock(rpc_lock, NULL);
69 CDEBUG(D_INFO, "error in handling %d\n", rc);
70 else if (!req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY)) {
76 /* Find and cancel locally locks matched by inode @bits & @mode in the resource
77 * found by @fid. Found locks are added into @cancel list. Returns the amount of
78 * locks added to @cancels list. */
79 int mdc_resource_get_unused(struct obd_export *exp, struct lu_fid *fid,
80 cfs_list_t *cancels, ldlm_mode_t mode,
83 ldlm_policy_data_t policy = {{0}};
84 struct ldlm_res_id res_id;
85 struct ldlm_resource *res;
89 fid_build_reg_res_name(fid, &res_id);
90 res = ldlm_resource_get(exp->exp_obd->obd_namespace,
94 LDLM_RESOURCE_ADDREF(res);
95 /* Initialize ibits lock policy. */
96 policy.l_inodebits.bits = bits;
97 count = ldlm_cancel_resource_local(res, cancels, &policy,
99 LDLM_RESOURCE_DELREF(res);
100 ldlm_resource_putref(res);
104 static int mdc_prep_elc_req(struct obd_export *exp, struct ptlrpc_request *req,
105 cfs_list_t *cancels, int count)
107 return ldlm_prep_elc_req(exp, req, LUSTRE_MDS_VERSION, MDS_REINT,
111 /* If mdc_setattr is called with an 'iattr', then it is a normal RPC that
112 * should take the normal semaphore and go to the normal portal.
114 * If it is called with iattr->ia_valid & ATTR_FROM_OPEN, then it is a
115 * magic open-path setattr that should take the setattr semaphore and
116 * go to the setattr portal. */
117 int mdc_setattr(struct obd_export *exp, struct md_op_data *op_data,
118 void *ea, int ealen, void *ea2, int ea2len,
119 struct ptlrpc_request **request, struct md_open_data **mod)
121 CFS_LIST_HEAD(cancels);
122 struct ptlrpc_request *req;
123 struct mdc_rpc_lock *rpc_lock;
124 struct obd_device *obd = exp->exp_obd;
129 LASSERT(op_data != NULL);
131 bits = MDS_INODELOCK_UPDATE;
132 if (op_data->op_attr.ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID))
133 bits |= MDS_INODELOCK_LOOKUP;
134 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
135 (fid_is_sane(&op_data->op_fid1)) &&
136 !OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK))
137 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
138 &cancels, LCK_EX, bits);
139 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
140 &RQF_MDS_REINT_SETATTR);
142 ldlm_lock_list_put(&cancels, l_bl_ast, count);
145 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
146 if ((op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) == 0)
147 req_capsule_set_size(&req->rq_pill, &RMF_MDT_EPOCH, RCL_CLIENT,
149 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, ealen);
150 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_CLIENT,
153 rc = mdc_prep_elc_req(exp, req, &cancels, count);
155 ptlrpc_request_free(req);
159 if (op_data->op_attr.ia_valid & ATTR_FROM_OPEN) {
160 req->rq_request_portal = MDS_SETATTR_PORTAL;
161 ptlrpc_at_set_req_timeout(req);
162 rpc_lock = obd->u.cli.cl_setattr_lock;
164 rpc_lock = obd->u.cli.cl_rpc_lock;
167 if (op_data->op_attr.ia_valid & (ATTR_MTIME | ATTR_CTIME))
168 CDEBUG(D_INODE, "setting mtime "CFS_TIME_T
169 ", ctime "CFS_TIME_T"\n",
170 LTIME_S(op_data->op_attr.ia_mtime),
171 LTIME_S(op_data->op_attr.ia_ctime));
172 mdc_setattr_pack(req, op_data, ea, ealen, ea2, ea2len);
174 ptlrpc_request_set_replen(req);
175 if (mod && (op_data->op_flags & MF_EPOCH_OPEN) &&
176 req->rq_import->imp_replayable)
178 LASSERT(*mod == NULL);
180 *mod = obd_mod_alloc();
182 DEBUG_REQ(D_ERROR, req, "Can't allocate "
186 req->rq_cb_data = *mod;
187 (*mod)->mod_open_req = req;
188 req->rq_commit_cb = mdc_commit_open;
190 * Take an extra reference on \var mod, it protects \var
191 * mod from being freed on eviction (commit callback is
192 * called despite rq_replay flag).
193 * Will be put on mdc_done_writing().
199 rc = mdc_reint(req, rpc_lock, LUSTRE_IMP_FULL);
201 /* Save the obtained info in the original RPC for the replay case. */
202 if (rc == 0 && (op_data->op_flags & MF_EPOCH_OPEN)) {
203 struct mdt_ioepoch *epoch;
204 struct mdt_body *body;
206 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
207 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
208 LASSERT(epoch != NULL);
209 LASSERT(body != NULL);
210 epoch->handle = body->handle;
211 epoch->ioepoch = body->ioepoch;
212 req->rq_replay_cb = mdc_replay_open;
213 /** bug 3633, open may be committed and estale answer is not error */
214 } else if (rc == -ESTALE && (op_data->op_flags & MF_SOM_CHANGE)) {
216 } else if (rc == -ERESTARTSYS) {
220 if (rc && req->rq_commit_cb) {
221 /* Put an extra reference on \var mod on error case. */
223 req->rq_commit_cb(req);
228 int mdc_create(struct obd_export *exp, struct md_op_data *op_data,
229 const void *data, int datalen, int mode, __u32 uid, __u32 gid,
230 cfs_cap_t cap_effective, __u64 rdev,
231 struct ptlrpc_request **request)
233 struct ptlrpc_request *req;
236 CFS_LIST_HEAD(cancels);
239 /* For case if upper layer did not alloc fid, do it now. */
240 if (!fid_is_sane(&op_data->op_fid2)) {
242 * mdc_fid_alloc() may return errno 1 in case of switch to new
243 * sequence, handle this.
245 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
247 CERROR("Can't alloc new fid, rc %d\n", rc);
252 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
253 (fid_is_sane(&op_data->op_fid1)))
254 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
256 MDS_INODELOCK_UPDATE);
258 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
259 &RQF_MDS_REINT_CREATE_RMT_ACL);
261 ldlm_lock_list_put(&cancels, l_bl_ast, count);
264 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
265 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
266 op_data->op_namelen + 1);
267 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
268 data && datalen ? datalen : 0);
270 rc = mdc_prep_elc_req(exp, req, &cancels, count);
272 ptlrpc_request_free(req);
277 * mdc_create_pack() fills msg->bufs[1] with name and msg->bufs[2] with
278 * tgt, for symlinks or lov MD data.
280 mdc_create_pack(req, op_data, data, datalen, mode, uid,
281 gid, cap_effective, rdev);
283 ptlrpc_request_set_replen(req);
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 == 0) {
294 struct mdt_body *body;
295 struct lustre_capa *capa;
297 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
299 if (body->valid & OBD_MD_FLMDSCAPA) {
300 capa = req_capsule_server_get(&req->rq_pill,
311 int mdc_unlink(struct obd_export *exp, struct md_op_data *op_data,
312 struct ptlrpc_request **request)
314 CFS_LIST_HEAD(cancels);
315 struct obd_device *obd = class_exp2obd(exp);
316 struct ptlrpc_request *req = *request;
320 LASSERT(req == NULL);
322 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
323 (fid_is_sane(&op_data->op_fid1)) &&
324 !OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK))
325 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
327 MDS_INODELOCK_UPDATE);
328 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
329 (fid_is_sane(&op_data->op_fid3)) &&
330 !OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK))
331 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
334 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
335 &RQF_MDS_REINT_UNLINK);
337 ldlm_lock_list_put(&cancels, l_bl_ast, count);
340 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
341 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
342 op_data->op_namelen + 1);
344 rc = mdc_prep_elc_req(exp, req, &cancels, count);
346 ptlrpc_request_free(req);
350 mdc_unlink_pack(req, op_data);
352 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
353 obd->u.cli.cl_max_mds_easize);
354 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
355 obd->u.cli.cl_max_mds_cookiesize);
356 ptlrpc_request_set_replen(req);
360 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
361 if (rc == -ERESTARTSYS)
366 int mdc_link(struct obd_export *exp, struct md_op_data *op_data,
367 struct ptlrpc_request **request)
369 CFS_LIST_HEAD(cancels);
370 struct obd_device *obd = exp->exp_obd;
371 struct ptlrpc_request *req;
375 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
376 (fid_is_sane(&op_data->op_fid2)))
377 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
379 MDS_INODELOCK_UPDATE);
380 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
381 (fid_is_sane(&op_data->op_fid1)))
382 count += mdc_resource_get_unused(exp, &op_data->op_fid1,
384 MDS_INODELOCK_UPDATE);
386 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_REINT_LINK);
388 ldlm_lock_list_put(&cancels, l_bl_ast, count);
391 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
392 mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
393 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
394 op_data->op_namelen + 1);
396 rc = mdc_prep_elc_req(exp, req, &cancels, count);
398 ptlrpc_request_free(req);
402 mdc_link_pack(req, op_data);
403 ptlrpc_request_set_replen(req);
405 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
407 if (rc == -ERESTARTSYS)
413 int mdc_rename(struct obd_export *exp, struct md_op_data *op_data,
414 const char *old, int oldlen, const char *new, int newlen,
415 struct ptlrpc_request **request)
417 CFS_LIST_HEAD(cancels);
418 struct obd_device *obd = exp->exp_obd;
419 struct ptlrpc_request *req;
423 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
424 (fid_is_sane(&op_data->op_fid1)))
425 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
427 MDS_INODELOCK_UPDATE);
428 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
429 (fid_is_sane(&op_data->op_fid2)))
430 count += mdc_resource_get_unused(exp, &op_data->op_fid2,
432 MDS_INODELOCK_UPDATE);
433 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
434 (fid_is_sane(&op_data->op_fid3)))
435 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
437 MDS_INODELOCK_LOOKUP);
438 if ((op_data->op_flags & MF_MDC_CANCEL_FID4) &&
439 (fid_is_sane(&op_data->op_fid4)))
440 count += mdc_resource_get_unused(exp, &op_data->op_fid4,
444 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
445 &RQF_MDS_REINT_RENAME);
447 ldlm_lock_list_put(&cancels, l_bl_ast, count);
451 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
452 mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
453 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT, oldlen + 1);
454 req_capsule_set_size(&req->rq_pill, &RMF_SYMTGT, RCL_CLIENT, newlen+1);
456 rc = mdc_prep_elc_req(exp, req, &cancels, count);
458 ptlrpc_request_free(req);
462 if (exp_connect_cancelset(exp) && req)
463 ldlm_cli_cancel_list(&cancels, count, req, 0);
465 mdc_rename_pack(req, op_data, old, oldlen, new, newlen);
467 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
468 obd->u.cli.cl_max_mds_easize);
469 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
470 obd->u.cli.cl_max_mds_cookiesize);
471 ptlrpc_request_set_replen(req);
473 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
475 if (rc == -ERESTARTSYS)