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, 2015, 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
39 # include <linux/module.h>
40 # include <linux/kernel.h>
42 #include "../include/obd_class.h"
43 #include "mdc_internal.h"
44 #include "../include/lustre_fid.h"
46 /* mdc_setattr does its own semaphore handling */
47 static int mdc_reint(struct ptlrpc_request *request,
48 struct mdc_rpc_lock *rpc_lock,
53 request->rq_send_state = level;
55 mdc_get_rpc_lock(rpc_lock, NULL);
56 rc = ptlrpc_queue_wait(request);
57 mdc_put_rpc_lock(rpc_lock, NULL);
59 CDEBUG(D_INFO, "error in handling %d\n", rc);
60 else if (!req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY))
66 /* Find and cancel locally locks matched by inode @bits & @mode in the resource
67 * found by @fid. Found locks are added into @cancel list. Returns the amount of
68 * locks added to @cancels list. */
69 int mdc_resource_get_unused(struct obd_export *exp, const struct lu_fid *fid,
70 struct list_head *cancels, ldlm_mode_t mode,
73 struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
74 ldlm_policy_data_t policy = {};
75 struct ldlm_res_id res_id;
76 struct ldlm_resource *res;
79 /* Return, i.e. cancel nothing, only if ELC is supported (flag in
80 * export) but disabled through procfs (flag in NS).
82 * This distinguishes from a case when ELC is not supported originally,
83 * when we still want to cancel locks in advance and just cancel them
84 * locally, without sending any RPC. */
85 if (exp_connect_cancelset(exp) && !ns_connect_cancelset(ns))
88 fid_build_reg_res_name(fid, &res_id);
89 res = ldlm_resource_get(exp->exp_obd->obd_namespace,
93 LDLM_RESOURCE_ADDREF(res);
94 /* Initialize ibits lock policy. */
95 policy.l_inodebits.bits = bits;
96 count = ldlm_cancel_resource_local(res, cancels, &policy,
98 LDLM_RESOURCE_DELREF(res);
99 ldlm_resource_putref(res);
103 int mdc_setattr(struct obd_export *exp, struct md_op_data *op_data,
104 void *ea, int ealen, void *ea2, int ea2len,
105 struct ptlrpc_request **request, struct md_open_data **mod)
108 struct ptlrpc_request *req;
109 struct mdc_rpc_lock *rpc_lock;
110 struct obd_device *obd = exp->exp_obd;
114 bits = MDS_INODELOCK_UPDATE;
115 if (op_data->op_attr.ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID))
116 bits |= MDS_INODELOCK_LOOKUP;
117 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
118 (fid_is_sane(&op_data->op_fid1)) &&
119 !OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK_NET))
120 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
121 &cancels, LCK_EX, bits);
122 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
123 &RQF_MDS_REINT_SETATTR);
125 ldlm_lock_list_put(&cancels, l_bl_ast, count);
128 if ((op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) == 0)
129 req_capsule_set_size(&req->rq_pill, &RMF_MDT_EPOCH, RCL_CLIENT,
131 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, ealen);
132 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_CLIENT,
135 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
137 ptlrpc_request_free(req);
141 rpc_lock = obd->u.cli.cl_rpc_lock;
143 if (op_data->op_attr.ia_valid & (ATTR_MTIME | ATTR_CTIME))
144 CDEBUG(D_INODE, "setting mtime %ld, ctime %ld\n",
145 LTIME_S(op_data->op_attr.ia_mtime),
146 LTIME_S(op_data->op_attr.ia_ctime));
147 mdc_setattr_pack(req, op_data, ea, ealen, ea2, ea2len);
149 ptlrpc_request_set_replen(req);
150 if (mod && (op_data->op_flags & MF_EPOCH_OPEN) &&
151 req->rq_import->imp_replayable) {
154 *mod = obd_mod_alloc();
156 DEBUG_REQ(D_ERROR, req, "Can't allocate md_open_data");
159 req->rq_cb_data = *mod;
160 (*mod)->mod_open_req = req;
161 req->rq_commit_cb = mdc_commit_open;
162 (*mod)->mod_is_create = true;
164 * Take an extra reference on \var mod, it protects \var
165 * mod from being freed on eviction (commit callback is
166 * called despite rq_replay flag).
167 * Will be put on mdc_done_writing().
173 rc = mdc_reint(req, rpc_lock, LUSTRE_IMP_FULL);
175 /* Save the obtained info in the original RPC for the replay case. */
176 if (rc == 0 && (op_data->op_flags & MF_EPOCH_OPEN)) {
177 struct mdt_ioepoch *epoch;
178 struct mdt_body *body;
180 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
181 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
182 epoch->handle = body->handle;
183 epoch->ioepoch = body->ioepoch;
184 req->rq_replay_cb = mdc_replay_open;
185 /** bug 3633, open may be committed and estale answer is not error */
186 } else if (rc == -ESTALE && (op_data->op_flags & MF_SOM_CHANGE)) {
188 } else if (rc == -ERESTARTSYS) {
192 if (rc && req->rq_commit_cb) {
193 /* Put an extra reference on \var mod on error case. */
196 req->rq_commit_cb(req);
201 int mdc_create(struct obd_export *exp, struct md_op_data *op_data,
202 const void *data, int datalen, int mode, __u32 uid, __u32 gid,
203 cfs_cap_t cap_effective, __u64 rdev,
204 struct ptlrpc_request **request)
206 struct ptlrpc_request *req;
208 int count, resends = 0;
209 struct obd_import *import = exp->exp_obd->u.cli.cl_import;
210 int generation = import->imp_generation;
213 /* For case if upper layer did not alloc fid, do it now. */
214 if (!fid_is_sane(&op_data->op_fid2)) {
216 * mdc_fid_alloc() may return errno 1 in case of switch to new
217 * sequence, handle this.
219 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
221 CERROR("Can't alloc new fid, rc %d\n", rc);
228 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
229 (fid_is_sane(&op_data->op_fid1)))
230 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
232 MDS_INODELOCK_UPDATE);
234 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
235 &RQF_MDS_REINT_CREATE_RMT_ACL);
237 ldlm_lock_list_put(&cancels, l_bl_ast, count);
240 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
241 op_data->op_namelen + 1);
242 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
243 data && datalen ? datalen : 0);
245 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
247 ptlrpc_request_free(req);
252 * mdc_create_pack() fills msg->bufs[1] with name and msg->bufs[2] with
253 * tgt, for symlinks or lov MD data.
255 mdc_create_pack(req, op_data, data, datalen, mode, uid,
256 gid, cap_effective, rdev);
258 ptlrpc_request_set_replen(req);
260 /* ask ptlrpc not to resend on EINPROGRESS since we have our own retry
262 req->rq_no_retry_einprogress = 1;
265 req->rq_generation_set = 1;
266 req->rq_import_generation = generation;
267 req->rq_sent = ktime_get_real_seconds() + resends;
269 level = LUSTRE_IMP_FULL;
271 rc = mdc_reint(req, exp->exp_obd->u.cli.cl_rpc_lock, level);
273 /* Resend if we were told to. */
274 if (rc == -ERESTARTSYS) {
275 level = LUSTRE_IMP_RECOVER;
277 } else if (rc == -EINPROGRESS) {
278 /* Retry create infinitely until succeed or get other
280 ptlrpc_req_finished(req);
283 CDEBUG(D_HA, "%s: resend:%d create on "DFID"/"DFID"\n",
284 exp->exp_obd->obd_name, resends,
285 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2));
287 if (generation == import->imp_generation) {
290 CDEBUG(D_HA, "resend cross eviction\n");
299 int mdc_unlink(struct obd_export *exp, struct md_op_data *op_data,
300 struct ptlrpc_request **request)
303 struct obd_device *obd = class_exp2obd(exp);
304 struct ptlrpc_request *req = *request;
309 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
310 (fid_is_sane(&op_data->op_fid1)) &&
311 !OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK_NET))
312 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
314 MDS_INODELOCK_UPDATE);
315 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
316 (fid_is_sane(&op_data->op_fid3)) &&
317 !OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK_NET))
318 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
321 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
322 &RQF_MDS_REINT_UNLINK);
324 ldlm_lock_list_put(&cancels, l_bl_ast, count);
327 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
328 op_data->op_namelen + 1);
330 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
332 ptlrpc_request_free(req);
336 mdc_unlink_pack(req, op_data);
338 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
339 obd->u.cli.cl_default_mds_easize);
340 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
341 obd->u.cli.cl_default_mds_cookiesize);
342 ptlrpc_request_set_replen(req);
346 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
347 if (rc == -ERESTARTSYS)
352 int mdc_link(struct obd_export *exp, struct md_op_data *op_data,
353 struct ptlrpc_request **request)
356 struct obd_device *obd = exp->exp_obd;
357 struct ptlrpc_request *req;
360 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
361 (fid_is_sane(&op_data->op_fid2)))
362 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
364 MDS_INODELOCK_UPDATE);
365 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
366 (fid_is_sane(&op_data->op_fid1)))
367 count += mdc_resource_get_unused(exp, &op_data->op_fid1,
369 MDS_INODELOCK_UPDATE);
371 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_REINT_LINK);
373 ldlm_lock_list_put(&cancels, l_bl_ast, count);
376 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
377 op_data->op_namelen + 1);
379 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
381 ptlrpc_request_free(req);
385 mdc_link_pack(req, op_data);
386 ptlrpc_request_set_replen(req);
388 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
390 if (rc == -ERESTARTSYS)
396 int mdc_rename(struct obd_export *exp, struct md_op_data *op_data,
397 const char *old, int oldlen, const char *new, int newlen,
398 struct ptlrpc_request **request)
401 struct obd_device *obd = exp->exp_obd;
402 struct ptlrpc_request *req;
405 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
406 (fid_is_sane(&op_data->op_fid1)))
407 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
409 MDS_INODELOCK_UPDATE);
410 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
411 (fid_is_sane(&op_data->op_fid2)))
412 count += mdc_resource_get_unused(exp, &op_data->op_fid2,
414 MDS_INODELOCK_UPDATE);
415 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
416 (fid_is_sane(&op_data->op_fid3)))
417 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
419 MDS_INODELOCK_LOOKUP);
420 if ((op_data->op_flags & MF_MDC_CANCEL_FID4) &&
421 (fid_is_sane(&op_data->op_fid4)))
422 count += mdc_resource_get_unused(exp, &op_data->op_fid4,
426 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
427 &RQF_MDS_REINT_RENAME);
429 ldlm_lock_list_put(&cancels, l_bl_ast, count);
433 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT, oldlen + 1);
434 req_capsule_set_size(&req->rq_pill, &RMF_SYMTGT, RCL_CLIENT, newlen+1);
436 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
438 ptlrpc_request_free(req);
442 if (exp_connect_cancelset(exp) && req)
443 ldlm_cli_cancel_list(&cancels, count, req, 0);
445 mdc_rename_pack(req, op_data, old, oldlen, new, newlen);
447 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
448 obd->u.cli.cl_default_mds_easize);
449 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
450 obd->u.cli.cl_default_mds_cookiesize);
451 ptlrpc_request_set_replen(req);
453 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
455 if (rc == -ERESTARTSYS)