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 struct ptlrpc_request *mdc_prep_elc_req(struct obd_export *exp,
94 int bufcount, int *size, int off,
95 struct list_head *cancels, int count)
97 return ldlm_prep_elc_req(exp, LUSTRE_MDS_VERSION, MDS_REINT,
98 bufcount, size, off, 0, cancels, count);
101 /* If mdc_setattr is called with an 'iattr', then it is a normal RPC that
102 * should take the normal semaphore and go to the normal portal.
104 * If it is called with iattr->ia_valid & ATTR_FROM_OPEN, then it is a
105 * magic open-path setattr that should take the setattr semaphore and
106 * go to the setattr portal. */
107 int mdc_setattr(struct obd_export *exp, struct md_op_data *op_data,
108 void *ea, int ealen, void *ea2, int ea2len,
109 struct ptlrpc_request **request, struct md_open_data **mod)
111 CFS_LIST_HEAD(cancels);
112 struct ptlrpc_request *req;
113 struct mdt_rec_setattr *rec;
114 struct mdc_rpc_lock *rpc_lock;
115 struct obd_device *obd = exp->exp_obd;
116 int size[7] = { sizeof(struct ptlrpc_body),
117 sizeof(*rec), 0, 0, ealen, ea2len, 0 };
118 int count = 0, bufcount = 4, rc;
122 LASSERT(op_data != NULL);
124 size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
125 sizeof(struct lustre_capa) : 0;
127 if (op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN))
128 size[REQ_REC_OFF + 2] = sizeof(struct mdt_epoch);
136 bits = MDS_INODELOCK_UPDATE;
137 if (op_data->op_attr.ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID))
138 bits |= MDS_INODELOCK_LOOKUP;
139 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
140 (fid_is_sane(&op_data->op_fid1)))
141 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
142 &cancels, LCK_EX, bits);
143 if (exp_connect_cancelset(exp) && count)
145 req = mdc_prep_elc_req(exp, bufcount, size,
146 REQ_REC_OFF + 5, &cancels, count);
151 if (op_data->op_attr.ia_valid & ATTR_FROM_OPEN) {
152 req->rq_request_portal = MDS_SETATTR_PORTAL; //XXX FIXME bug 249
153 rpc_lock = obd->u.cli.cl_setattr_lock;
155 rpc_lock = obd->u.cli.cl_rpc_lock;
158 if (op_data->op_attr.ia_valid & (ATTR_MTIME | ATTR_CTIME))
159 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu\n",
160 LTIME_S(op_data->op_attr.ia_mtime),
161 LTIME_S(op_data->op_attr.ia_ctime));
162 mdc_setattr_pack(req, REQ_REC_OFF, op_data, ea, ealen, ea2, ea2len);
164 size[REPLY_REC_OFF] = sizeof(struct mdt_body);
165 size[REPLY_REC_OFF + 1] = sizeof(struct lustre_capa);
166 ptlrpc_req_set_repsize(req, 3, size);
167 if (mod && (op_data->op_flags & MF_EPOCH_OPEN) &&
168 req->rq_import->imp_replayable)
170 LASSERT(*mod == NULL);
174 DEBUG_REQ(D_ERROR, req, "Can't allocate "
177 CFS_INIT_LIST_HEAD(&(*mod)->mod_replay_list);
181 req->rq_cb_data = *mod;
182 req->rq_commit_cb = mdc_commit_delayed;
183 list_add_tail(&req->rq_mod_list, &(*mod)->mod_replay_list);
184 /* This is not the last request in sequence for truncate. */
185 if (op_data->op_flags & MF_EPOCH_OPEN)
188 req->rq_sequence = 1;
191 rc = mdc_reint(req, rpc_lock, LUSTRE_IMP_FULL);
193 if (rc == -ERESTARTSYS)
195 if (rc && req->rq_commit_cb)
196 req->rq_commit_cb(req);
200 int mdc_create(struct obd_export *exp, struct md_op_data *op_data,
201 const void *data, int datalen, int mode, __u32 uid, __u32 gid,
202 __u32 cap_effective, __u64 rdev, struct ptlrpc_request **request)
204 int size[6] = { sizeof(struct ptlrpc_body),
205 sizeof(struct mdt_rec_create),
206 0, op_data->op_namelen + 1, 0, 0 };
207 struct obd_device *obd = exp->exp_obd;
208 int level, bufcount = 4, rc;
209 struct ptlrpc_request *req;
211 CFS_LIST_HEAD(cancels);
214 /* For case if upper layer did not alloc fid, do it now. */
215 if (!fid_is_sane(&op_data->op_fid2)) {
217 * mdc_fid_alloc() may return errno 1 in case of switch to new
218 * sequence, handle this.
220 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
222 CERROR("Can't alloc new fid, rc %d\n", rc);
227 size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
228 sizeof(struct lustre_capa) : 0;
230 if (data && datalen) {
231 size[bufcount] = datalen;
235 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
236 (fid_is_sane(&op_data->op_fid1)))
237 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
239 MDS_INODELOCK_UPDATE);
240 if (exp_connect_cancelset(exp) && count)
242 req = mdc_prep_elc_req(exp, bufcount, size,
243 REQ_REC_OFF + 4, &cancels, count);
249 * mdc_create_pack() fills msg->bufs[1] with name and msg->bufs[2] with
250 * tgt, for symlinks or lov MD data.
252 mdc_create_pack(req, REQ_REC_OFF, op_data, data, datalen, mode, uid,
253 gid, cap_effective, rdev);
255 size[REPLY_REC_OFF] = sizeof(struct mdt_body);
256 size[REPLY_REC_OFF + 1] = sizeof(struct lustre_capa);
257 ptlrpc_req_set_repsize(req, 3, size);
259 level = LUSTRE_IMP_FULL;
261 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, level);
263 /* Resend if we were told to. */
264 if (rc == -ERESTARTSYS) {
265 level = LUSTRE_IMP_RECOVER;
267 } else if (rc == 0) {
268 struct mdt_body *body;
269 struct lustre_capa *capa;
271 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
274 if (body->valid & OBD_MD_FLMDSCAPA) {
275 capa = lustre_unpack_capa(req->rq_repmsg,
278 CERROR("Missing/short MDS capability\n");
288 int mdc_unlink(struct obd_export *exp, struct md_op_data *op_data,
289 struct ptlrpc_request **request)
291 CFS_LIST_HEAD(cancels);
292 struct obd_device *obd = class_exp2obd(exp);
293 struct ptlrpc_request *req = *request;
294 int size[5] = { sizeof(struct ptlrpc_body),
295 sizeof(struct mdt_rec_unlink),
296 0, op_data->op_namelen + 1, 0 };
297 int count = 0, rc, bufcount = 4;
300 LASSERT(req == NULL);
302 size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
303 sizeof(struct lustre_capa) : 0;
305 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
306 (fid_is_sane(&op_data->op_fid1)))
307 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
309 MDS_INODELOCK_UPDATE);
310 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
311 (fid_is_sane(&op_data->op_fid3)))
312 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
315 if (exp_connect_cancelset(exp) && count)
318 req = mdc_prep_elc_req(exp, bufcount, size,
319 REQ_REC_OFF + 3, &cancels, count);
325 size[REPLY_REC_OFF] = sizeof(struct mdt_body);
326 size[REPLY_REC_OFF + 1] = obd->u.cli.cl_max_mds_easize;
327 size[REPLY_REC_OFF + 2] = obd->u.cli.cl_max_mds_cookiesize;
328 ptlrpc_req_set_repsize(req, 4, size);
330 mdc_unlink_pack(req, REQ_REC_OFF, op_data);
332 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
333 if (rc == -ERESTARTSYS)
338 int mdc_link(struct obd_export *exp, struct md_op_data *op_data,
339 struct ptlrpc_request **request)
341 CFS_LIST_HEAD(cancels);
342 struct obd_device *obd = exp->exp_obd;
343 struct ptlrpc_request *req;
344 int size[6] = { sizeof(struct ptlrpc_body),
345 sizeof(struct mdt_rec_link),
346 0, 0, op_data->op_namelen + 1, 0 };
347 int count = 0, rc, bufcount = 5;
350 size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
351 sizeof(struct lustre_capa) : 0;
352 size[REQ_REC_OFF + 2] = op_data->op_capa2 ?
353 sizeof(struct lustre_capa) : 0;
355 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
356 (fid_is_sane(&op_data->op_fid2)))
357 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
359 MDS_INODELOCK_UPDATE);
360 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
361 (fid_is_sane(&op_data->op_fid1)))
362 count += mdc_resource_get_unused(exp, &op_data->op_fid1,
364 MDS_INODELOCK_UPDATE);
365 if (exp_connect_cancelset(exp) && count)
367 req = mdc_prep_elc_req(exp, bufcount, size,
368 REQ_REC_OFF + 4, &cancels, count);
373 mdc_link_pack(req, REQ_REC_OFF, op_data);
375 size[REPLY_REC_OFF] = sizeof(struct mdt_body);
376 ptlrpc_req_set_repsize(req, 2, size);
378 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
380 if (rc == -ERESTARTSYS)
386 int mdc_rename(struct obd_export *exp, struct md_op_data *op_data,
387 const char *old, int oldlen, const char *new, int newlen,
388 struct ptlrpc_request **request)
390 CFS_LIST_HEAD(cancels);
391 struct obd_device *obd = exp->exp_obd;
392 struct ptlrpc_request *req;
393 int size[7] = { sizeof(struct ptlrpc_body),
394 sizeof(struct mdt_rec_rename),
395 0, 0, oldlen + 1, newlen + 1, 0 };
396 int count = 0, rc, bufcount = 6;
399 size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
400 sizeof(struct lustre_capa) : 0;
401 size[REQ_REC_OFF + 2] = op_data->op_capa2 ?
402 sizeof(struct lustre_capa) : 0;
404 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
405 (fid_is_sane(&op_data->op_fid1)))
406 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
408 MDS_INODELOCK_UPDATE);
409 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
410 (fid_is_sane(&op_data->op_fid2)))
411 count += mdc_resource_get_unused(exp, &op_data->op_fid2,
413 MDS_INODELOCK_UPDATE);
414 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
415 (fid_is_sane(&op_data->op_fid3)))
416 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
418 MDS_INODELOCK_LOOKUP);
419 if ((op_data->op_flags & MF_MDC_CANCEL_FID4) &&
420 (fid_is_sane(&op_data->op_fid4)))
421 count += mdc_resource_get_unused(exp, &op_data->op_fid4,
424 if (exp_connect_cancelset(exp) && count)
426 req = mdc_prep_elc_req(exp, bufcount, size,
427 REQ_REC_OFF + 5, &cancels, count);
432 mdc_rename_pack(req, REQ_REC_OFF, op_data, old, oldlen, new, newlen);
434 size[REPLY_REC_OFF] = sizeof(struct mdt_body);
435 size[REPLY_REC_OFF + 1] = obd->u.cli.cl_max_mds_easize;
436 size[REPLY_REC_OFF + 2] = obd->u.cli.cl_max_mds_cookiesize;
437 ptlrpc_req_set_repsize(req, 4, size);
439 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
441 if (rc == -ERESTARTSYS)