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)
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);
166 rc = mdc_reint(req, rpc_lock, LUSTRE_IMP_FULL);
168 if (rc == -ERESTARTSYS)
174 int mdc_create(struct obd_export *exp, struct md_op_data *op_data,
175 const void *data, int datalen, int mode, __u32 uid, __u32 gid,
176 __u32 cap_effective, __u64 rdev, struct ptlrpc_request **request)
178 int size[6] = { sizeof(struct ptlrpc_body),
179 sizeof(struct mdt_rec_create),
180 0, op_data->op_namelen + 1, 0, 0 };
181 struct obd_device *obd = exp->exp_obd;
182 int level, bufcount = 4, rc;
183 struct ptlrpc_request *req;
185 CFS_LIST_HEAD(cancels);
188 /* For case if upper layer did not alloc fid, do it now. */
189 if (!fid_is_sane(&op_data->op_fid2)) {
191 * mdc_fid_alloc() may return errno 1 in case of switch to new
192 * sequence, handle this.
194 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
196 CERROR("Can't alloc new fid, rc %d\n", rc);
201 size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
202 sizeof(struct lustre_capa) : 0;
204 if (data && datalen) {
205 size[bufcount] = datalen;
209 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
210 (fid_is_sane(&op_data->op_fid1)))
211 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
213 MDS_INODELOCK_UPDATE);
214 if (exp_connect_cancelset(exp) && count) {
216 size[REQ_REC_OFF + 4] = ldlm_request_bufsize(count, MDS_REINT);
218 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
219 MDS_REINT, bufcount, size, NULL);
220 if (exp_connect_cancelset(exp) && req)
221 ldlm_cli_cancel_list(&cancels, count, req, REQ_REC_OFF + 4, 0);
223 ldlm_lock_list_put(&cancels, l_bl_ast, count);
229 * mdc_create_pack() fills msg->bufs[1] with name and msg->bufs[2] with
230 * tgt, for symlinks or lov MD data.
232 mdc_create_pack(req, REQ_REC_OFF, op_data, data, datalen, mode, uid,
233 gid, cap_effective, rdev);
235 size[REPLY_REC_OFF] = sizeof(struct mdt_body);
236 size[REPLY_REC_OFF + 1] = sizeof(struct lustre_capa);
237 ptlrpc_req_set_repsize(req, 3, size);
239 level = LUSTRE_IMP_FULL;
241 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, level);
243 /* Resend if we were told to. */
244 if (rc == -ERESTARTSYS) {
245 level = LUSTRE_IMP_RECOVER;
247 } else if (rc == 0) {
248 struct mdt_body *body;
249 struct lustre_capa *capa;
251 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
254 if (body->valid & OBD_MD_FLMDSCAPA) {
255 capa = lustre_unpack_capa(req->rq_repmsg,
258 CERROR("Missing/short MDS capability\n");
268 int mdc_unlink(struct obd_export *exp, struct md_op_data *op_data,
269 struct ptlrpc_request **request)
271 CFS_LIST_HEAD(cancels);
272 struct obd_device *obd = class_exp2obd(exp);
273 struct ptlrpc_request *req = *request;
274 int size[5] = { sizeof(struct ptlrpc_body),
275 sizeof(struct mdt_rec_unlink),
276 0, op_data->op_namelen + 1, 0 };
277 int count = 0, rc, bufcount = 4;
280 LASSERT(req == NULL);
282 size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
283 sizeof(struct lustre_capa) : 0;
285 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
286 (fid_is_sane(&op_data->op_fid1)))
287 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
289 MDS_INODELOCK_UPDATE);
290 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
291 (fid_is_sane(&op_data->op_fid3)))
292 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
295 if (exp_connect_cancelset(exp) && count) {
297 size[REQ_REC_OFF + 3] = ldlm_request_bufsize(count, MDS_REINT);
299 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
300 MDS_REINT, bufcount, size, NULL);
301 if (exp_connect_cancelset(exp) && req)
302 ldlm_cli_cancel_list(&cancels, count, req, REQ_REC_OFF + 3, 0);
304 ldlm_lock_list_put(&cancels, l_bl_ast, count);
310 size[REPLY_REC_OFF] = sizeof(struct mdt_body);
311 size[REPLY_REC_OFF + 1] = obd->u.cli.cl_max_mds_easize;
312 size[REPLY_REC_OFF + 2] = obd->u.cli.cl_max_mds_cookiesize;
313 ptlrpc_req_set_repsize(req, 4, size);
315 mdc_unlink_pack(req, REQ_REC_OFF, op_data);
317 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
318 if (rc == -ERESTARTSYS)
323 int mdc_link(struct obd_export *exp, struct md_op_data *op_data,
324 struct ptlrpc_request **request)
326 CFS_LIST_HEAD(cancels);
327 struct obd_device *obd = exp->exp_obd;
328 struct ptlrpc_request *req;
329 int size[6] = { sizeof(struct ptlrpc_body),
330 sizeof(struct mdt_rec_link),
331 0, 0, op_data->op_namelen + 1, 0 };
332 int count = 0, rc, bufcount = 5;
335 size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
336 sizeof(struct lustre_capa) : 0;
337 size[REQ_REC_OFF + 2] = op_data->op_capa2 ?
338 sizeof(struct lustre_capa) : 0;
340 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
341 (fid_is_sane(&op_data->op_fid2)))
342 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
344 MDS_INODELOCK_UPDATE);
345 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
346 (fid_is_sane(&op_data->op_fid1)))
347 count += mdc_resource_get_unused(exp, &op_data->op_fid1,
349 MDS_INODELOCK_UPDATE);
350 if (exp_connect_cancelset(exp) && count) {
352 size[REQ_REC_OFF + 4] = ldlm_request_bufsize(count, MDS_REINT);
355 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
356 MDS_REINT, bufcount, size, NULL);
357 if (exp_connect_cancelset(exp) && req)
358 ldlm_cli_cancel_list(&cancels, count, req, REQ_REC_OFF + 4, 0);
360 ldlm_lock_list_put(&cancels, l_bl_ast, count);
365 mdc_link_pack(req, REQ_REC_OFF, op_data);
367 size[REPLY_REC_OFF] = sizeof(struct mdt_body);
368 ptlrpc_req_set_repsize(req, 2, size);
370 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
372 if (rc == -ERESTARTSYS)
378 int mdc_rename(struct obd_export *exp, struct md_op_data *op_data,
379 const char *old, int oldlen, const char *new, int newlen,
380 struct ptlrpc_request **request)
382 CFS_LIST_HEAD(cancels);
383 struct obd_device *obd = exp->exp_obd;
384 struct ptlrpc_request *req;
385 int size[7] = { sizeof(struct ptlrpc_body),
386 sizeof(struct mdt_rec_rename),
387 0, 0, oldlen + 1, newlen + 1, 0 };
388 int count = 0, rc, bufcount = 6;
391 size[REQ_REC_OFF + 1] = op_data->op_capa1 ?
392 sizeof(struct lustre_capa) : 0;
393 size[REQ_REC_OFF + 2] = op_data->op_capa2 ?
394 sizeof(struct lustre_capa) : 0;
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);
401 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
402 (fid_is_sane(&op_data->op_fid2)))
403 count += mdc_resource_get_unused(exp, &op_data->op_fid2,
405 MDS_INODELOCK_UPDATE);
406 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
407 (fid_is_sane(&op_data->op_fid3)))
408 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
410 MDS_INODELOCK_LOOKUP);
411 if ((op_data->op_flags & MF_MDC_CANCEL_FID4) &&
412 (fid_is_sane(&op_data->op_fid4)))
413 count += mdc_resource_get_unused(exp, &op_data->op_fid4,
416 if (exp_connect_cancelset(exp) && count) {
418 size[REQ_REC_OFF + 5] = ldlm_request_bufsize(count, MDS_REINT);
420 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
421 MDS_REINT, bufcount, size, NULL);
422 if (exp_connect_cancelset(exp) && req)
423 ldlm_cli_cancel_list(&cancels, count, req, REQ_REC_OFF + 5, 0);
425 ldlm_lock_list_put(&cancels, l_bl_ast, count);
430 mdc_rename_pack(req, REQ_REC_OFF, op_data, old, oldlen, new, newlen);
432 size[REPLY_REC_OFF] = sizeof(struct mdt_body);
433 size[REPLY_REC_OFF + 1] = obd->u.cli.cl_max_mds_easize;
434 size[REPLY_REC_OFF + 2] = obd->u.cli.cl_max_mds_cookiesize;
435 ptlrpc_req_set_repsize(req, 4, size);
437 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
439 if (rc == -ERESTARTSYS)