1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2001-2003 Cluster File Systems, Inc.
6 * This file is part of the Lustre file system, http://www.lustre.org
7 * Lustre is a trademark of Cluster File Systems, Inc.
9 * You may have signed or agreed to another license before downloading
10 * this software. If so, you are bound by the terms and conditions
11 * of that agreement, and the following does not apply to you. See the
12 * LICENSE file included with this distribution for more information.
14 * If you did not agree to a different license, then this copy of Lustre
15 * is open source software; you can redistribute it and/or modify it
16 * under the terms of version 2 of the GNU General Public License as
17 * published by the Free Software Foundation.
19 * In either case, Lustre is distributed in the hope that it will be
20 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
21 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
22 * license text for more details.
26 # define EXPORT_SYMTAB
28 #define DEBUG_SUBSYSTEM S_MDC
31 #ifndef AUTOCONF_INCLUDED
32 # include <linux/config.h>
34 # include <linux/module.h>
35 # include <linux/kernel.h>
37 # include <liblustre.h>
40 #include <obd_class.h>
41 #include "mdc_internal.h"
42 #include <lustre_fid.h>
44 /* mdc_setattr does its own semaphore handling */
45 static int mdc_reint(struct ptlrpc_request *request,
46 struct mdc_rpc_lock *rpc_lock, int level)
50 request->rq_send_state = level;
52 mdc_get_rpc_lock(rpc_lock, NULL);
53 rc = ptlrpc_queue_wait(request);
54 mdc_put_rpc_lock(rpc_lock, NULL);
56 CDEBUG(D_INFO, "error in handling %d\n", rc);
57 else if (!lustre_swab_repbuf(request, REPLY_REC_OFF,
58 sizeof(struct mdt_body),
59 lustre_swab_mdt_body)) {
60 CERROR ("Can't unpack mdt_body\n");
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, struct lu_fid *fid,
70 struct list_head *cancels, ldlm_mode_t mode,
73 ldlm_policy_data_t policy = {{0}};
74 struct ldlm_res_id res_id;
75 struct ldlm_resource *res;
79 fid_build_reg_res_name(fid, &res_id);
80 res = ldlm_resource_get(exp->exp_obd->obd_namespace,
85 /* Initialize ibits lock policy. */
86 policy.l_inodebits.bits = bits;
87 count = ldlm_cancel_resource_local(res, cancels, &policy,
89 ldlm_resource_putref(res);
93 /* If mdc_setattr is called with an 'iattr', then it is a normal RPC that
94 * should take the normal semaphore and go to the normal portal.
96 * If it is called with iattr->ia_valid & ATTR_FROM_OPEN, then it is a
97 * magic open-path setattr that should take the setattr semaphore and
98 * go to the setattr portal. */
99 int mdc_setattr(struct obd_export *exp, struct md_op_data *op_data,
100 void *ea, int ealen, void *ea2, int ea2len,
101 struct ptlrpc_request **request, struct md_open_data **mod)
103 CFS_LIST_HEAD(cancels);
104 struct ptlrpc_request *req;
105 struct mdt_rec_setattr *rec;
106 struct mdc_rpc_lock *rpc_lock;
107 struct obd_device *obd = exp->exp_obd;
108 int size[7] = { sizeof(struct ptlrpc_body),
109 sizeof(*rec), 0, 0, ealen, ea2len, 0 };
110 int count = 0, bufcount = 4, rc;
114 LASSERT(op_data != NULL);
116 size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
117 sizeof(struct lustre_capa) : 0;
119 if (op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN))
120 size[REQ_REC_OFF + 2] = sizeof(struct mdt_epoch);
128 bits = MDS_INODELOCK_UPDATE;
129 if (op_data->op_attr.ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID))
130 bits |= MDS_INODELOCK_LOOKUP;
131 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
132 (fid_is_sane(&op_data->op_fid1)))
133 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
134 &cancels, LCK_EX, bits);
135 if (exp_connect_cancelset(exp) && count) {
137 size[REQ_REC_OFF + 5] = ldlm_request_bufsize(count, MDS_REINT);
139 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
140 MDS_REINT, bufcount, size, NULL);
141 if (exp_connect_cancelset(exp) && req)
142 ldlm_cli_cancel_list(&cancels, count, req, REQ_REC_OFF + 5, 0);
144 ldlm_lock_list_put(&cancels, l_bl_ast, count);
149 if (op_data->op_attr.ia_valid & ATTR_FROM_OPEN) {
150 req->rq_request_portal = MDS_SETATTR_PORTAL; //XXX FIXME bug 249
151 rpc_lock = obd->u.cli.cl_setattr_lock;
153 rpc_lock = obd->u.cli.cl_rpc_lock;
156 if (op_data->op_attr.ia_valid & (ATTR_MTIME | ATTR_CTIME))
157 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu\n",
158 LTIME_S(op_data->op_attr.ia_mtime),
159 LTIME_S(op_data->op_attr.ia_ctime));
160 mdc_setattr_pack(req, REQ_REC_OFF, op_data, ea, ealen, ea2, ea2len);
162 size[REPLY_REC_OFF] = sizeof(struct mdt_body);
163 size[REPLY_REC_OFF + 1] = sizeof(struct lustre_capa);
164 ptlrpc_req_set_repsize(req, 3, size);
165 if (mod && (op_data->op_flags & MF_EPOCH_OPEN) &&
166 req->rq_import->imp_replayable)
168 LASSERT(*mod == NULL);
172 DEBUG_REQ(D_ERROR, req, "Can't allocate "
175 CFS_INIT_LIST_HEAD(&(*mod)->mod_replay_list);
179 req->rq_cb_data = *mod;
180 req->rq_commit_cb = mdc_commit_delayed;
181 list_add_tail(&req->rq_mod_list, &(*mod)->mod_replay_list);
182 /* This is not the last request in sequence for truncate. */
183 if (op_data->op_flags & MF_EPOCH_OPEN)
186 req->rq_sequence = 1;
189 rc = mdc_reint(req, rpc_lock, LUSTRE_IMP_FULL);
191 if (rc == -ERESTARTSYS)
193 if (rc && req->rq_commit_cb)
194 req->rq_commit_cb(req);
198 int mdc_create(struct obd_export *exp, struct md_op_data *op_data,
199 const void *data, int datalen, int mode, __u32 uid, __u32 gid,
200 __u32 cap_effective, __u64 rdev, struct ptlrpc_request **request)
202 int size[6] = { sizeof(struct ptlrpc_body),
203 sizeof(struct mdt_rec_create),
204 0, op_data->op_namelen + 1, 0, 0 };
205 struct obd_device *obd = exp->exp_obd;
206 int level, bufcount = 4, rc;
207 struct ptlrpc_request *req;
209 CFS_LIST_HEAD(cancels);
212 /* For case if upper layer did not alloc fid, do it now. */
213 if (!fid_is_sane(&op_data->op_fid2)) {
215 * mdc_fid_alloc() may return errno 1 in case of switch to new
216 * sequence, handle this.
218 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
220 CERROR("Can't alloc new fid, rc %d\n", rc);
225 size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
226 sizeof(struct lustre_capa) : 0;
228 if (data && datalen) {
229 size[bufcount] = datalen;
233 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
234 (fid_is_sane(&op_data->op_fid1)))
235 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
237 MDS_INODELOCK_UPDATE);
238 if (exp_connect_cancelset(exp) && count) {
240 size[REQ_REC_OFF + 4] = ldlm_request_bufsize(count, MDS_REINT);
242 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
243 MDS_REINT, bufcount, size, NULL);
244 if (exp_connect_cancelset(exp) && req)
245 ldlm_cli_cancel_list(&cancels, count, req, REQ_REC_OFF + 4, 0);
247 ldlm_lock_list_put(&cancels, l_bl_ast, count);
253 * mdc_create_pack() fills msg->bufs[1] with name and msg->bufs[2] with
254 * tgt, for symlinks or lov MD data.
256 mdc_create_pack(req, REQ_REC_OFF, op_data, data, datalen, mode, uid,
257 gid, cap_effective, rdev);
259 size[REPLY_REC_OFF] = sizeof(struct mdt_body);
260 size[REPLY_REC_OFF + 1] = sizeof(struct lustre_capa);
261 ptlrpc_req_set_repsize(req, 3, size);
263 level = LUSTRE_IMP_FULL;
265 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, level);
267 /* Resend if we were told to. */
268 if (rc == -ERESTARTSYS) {
269 level = LUSTRE_IMP_RECOVER;
271 } else if (rc == 0) {
272 struct mdt_body *body;
273 struct lustre_capa *capa;
275 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
278 if (body->valid & OBD_MD_FLMDSCAPA) {
279 capa = lustre_unpack_capa(req->rq_repmsg,
282 CERROR("Missing/short MDS capability\n");
292 int mdc_unlink(struct obd_export *exp, struct md_op_data *op_data,
293 struct ptlrpc_request **request)
295 CFS_LIST_HEAD(cancels);
296 struct obd_device *obd = class_exp2obd(exp);
297 struct ptlrpc_request *req = *request;
298 int size[5] = { sizeof(struct ptlrpc_body),
299 sizeof(struct mdt_rec_unlink),
300 0, op_data->op_namelen + 1, 0 };
301 int count = 0, rc, bufcount = 4;
304 LASSERT(req == NULL);
306 size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
307 sizeof(struct lustre_capa) : 0;
309 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
310 (fid_is_sane(&op_data->op_fid1)))
311 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
313 MDS_INODELOCK_UPDATE);
314 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
315 (fid_is_sane(&op_data->op_fid3)))
316 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
319 if (exp_connect_cancelset(exp) && count) {
321 size[REQ_REC_OFF + 3] = ldlm_request_bufsize(count, MDS_REINT);
323 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
324 MDS_REINT, bufcount, size, NULL);
325 if (exp_connect_cancelset(exp) && req)
326 ldlm_cli_cancel_list(&cancels, count, req, REQ_REC_OFF + 3, 0);
328 ldlm_lock_list_put(&cancels, l_bl_ast, count);
334 size[REPLY_REC_OFF] = sizeof(struct mdt_body);
335 size[REPLY_REC_OFF + 1] = obd->u.cli.cl_max_mds_easize;
336 size[REPLY_REC_OFF + 2] = obd->u.cli.cl_max_mds_cookiesize;
337 ptlrpc_req_set_repsize(req, 4, size);
339 mdc_unlink_pack(req, REQ_REC_OFF, op_data);
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)
350 CFS_LIST_HEAD(cancels);
351 struct obd_device *obd = exp->exp_obd;
352 struct ptlrpc_request *req;
353 int size[6] = { sizeof(struct ptlrpc_body),
354 sizeof(struct mdt_rec_link),
355 0, 0, op_data->op_namelen + 1, 0 };
356 int count = 0, rc, bufcount = 5;
359 size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
360 sizeof(struct lustre_capa) : 0;
361 size[REQ_REC_OFF + 2] = op_data->op_capa2 ?
362 sizeof(struct lustre_capa) : 0;
364 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
365 (fid_is_sane(&op_data->op_fid2)))
366 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
368 MDS_INODELOCK_UPDATE);
369 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
370 (fid_is_sane(&op_data->op_fid1)))
371 count += mdc_resource_get_unused(exp, &op_data->op_fid1,
373 MDS_INODELOCK_UPDATE);
374 if (exp_connect_cancelset(exp) && count) {
376 size[REQ_REC_OFF + 4] = ldlm_request_bufsize(count, MDS_REINT);
379 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
380 MDS_REINT, bufcount, size, NULL);
381 if (exp_connect_cancelset(exp) && req)
382 ldlm_cli_cancel_list(&cancels, count, req, REQ_REC_OFF + 4, 0);
384 ldlm_lock_list_put(&cancels, l_bl_ast, count);
389 mdc_link_pack(req, REQ_REC_OFF, op_data);
391 size[REPLY_REC_OFF] = sizeof(struct mdt_body);
392 ptlrpc_req_set_repsize(req, 2, size);
394 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
396 if (rc == -ERESTARTSYS)
402 int mdc_rename(struct obd_export *exp, struct md_op_data *op_data,
403 const char *old, int oldlen, const char *new, int newlen,
404 struct ptlrpc_request **request)
406 CFS_LIST_HEAD(cancels);
407 struct obd_device *obd = exp->exp_obd;
408 struct ptlrpc_request *req;
409 int size[7] = { sizeof(struct ptlrpc_body),
410 sizeof(struct mdt_rec_rename),
411 0, 0, oldlen + 1, newlen + 1, 0 };
412 int count = 0, rc, bufcount = 6;
415 size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
416 sizeof(struct lustre_capa) : 0;
417 size[REQ_REC_OFF + 2] = op_data->op_capa2 ?
418 sizeof(struct lustre_capa) : 0;
420 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
421 (fid_is_sane(&op_data->op_fid1)))
422 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
424 MDS_INODELOCK_UPDATE);
425 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
426 (fid_is_sane(&op_data->op_fid2)))
427 count += mdc_resource_get_unused(exp, &op_data->op_fid2,
429 MDS_INODELOCK_UPDATE);
430 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
431 (fid_is_sane(&op_data->op_fid3)))
432 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
434 MDS_INODELOCK_LOOKUP);
435 if ((op_data->op_flags & MF_MDC_CANCEL_FID4) &&
436 (fid_is_sane(&op_data->op_fid4)))
437 count += mdc_resource_get_unused(exp, &op_data->op_fid4,
440 if (exp_connect_cancelset(exp) && count) {
442 size[REQ_REC_OFF + 5] = ldlm_request_bufsize(count, MDS_REINT);
444 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
445 MDS_REINT, bufcount, size, NULL);
446 if (exp_connect_cancelset(exp) && req)
447 ldlm_cli_cancel_list(&cancels, count, req, REQ_REC_OFF + 5, 0);
449 ldlm_lock_list_put(&cancels, l_bl_ast, count);
454 mdc_rename_pack(req, REQ_REC_OFF, op_data, old, oldlen, new, newlen);
456 size[REPLY_REC_OFF] = sizeof(struct mdt_body);
457 size[REPLY_REC_OFF + 1] = obd->u.cli.cl_max_mds_easize;
458 size[REPLY_REC_OFF + 2] = obd->u.cli.cl_max_mds_cookiesize;
459 ptlrpc_req_set_repsize(req, 4, size);
461 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
463 if (rc == -ERESTARTSYS)