1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
5 * Lustre Metadata Target (mdt) open/close file handling
7 * Copyright (C) 2002-2006 Cluster File Systems, Inc.
8 * Author: Huang Hua <huanghua@clusterfs.com>
10 * This file is part of the Lustre file system, http://www.lustre.org
11 * Lustre is a trademark of Cluster File Systems, Inc.
13 * You may have signed or agreed to another license before downloading
14 * this software. If so, you are bound by the terms and conditions
15 * of that agreement, and the following does not apply to you. See the
16 * LICENSE file included with this distribution for more information.
18 * If you did not agree to a different license, then this copy of Lustre
19 * is open source software; you can redistribute it and/or modify it
20 * under the terms of version 2 of the GNU General Public License as
21 * published by the Free Software Foundation.
23 * In either case, Lustre is distributed in the hope that it will be
24 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
25 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
26 * license text for more details.
30 # define EXPORT_SYMTAB
32 #define DEBUG_SUBSYSTEM S_MDS
34 #include "mdt_internal.h"
37 * MDS file data handling: file data holds a handle for a file opened
41 static void mdt_mfd_get(void *mfdp)
43 struct mdt_file_data *mfd = mfdp;
45 atomic_inc(&mfd->mfd_refcount);
46 CDEBUG(D_INFO, "GETting mfd %p : new refcount %d\n", mfd,
47 atomic_read(&mfd->mfd_refcount));
50 /* Create a new mdt_file_data struct.
51 * reference is set to 1 */
52 static struct mdt_file_data *mdt_mfd_new(void)
54 struct mdt_file_data *mfd;
58 CERROR("mds: out of memory\n");
62 atomic_set(&mfd->mfd_refcount, 1);
64 INIT_LIST_HEAD(&mfd->mfd_handle.h_link);
65 INIT_LIST_HEAD(&mfd->mfd_list);
66 class_handle_hash(&mfd->mfd_handle, mdt_mfd_get);
71 /* Get a new reference on the mfd pointed to by handle, if handle is still
72 * valid. Caller must drop reference with mdt_mfd_put(). */
73 static struct mdt_file_data *mdt_handle2mfd(const struct lustre_handle *handle)
76 LASSERT(handle != NULL);
77 RETURN(class_handle2object(handle->cookie));
80 /* Drop mfd reference, freeing struct if this is the last one. */
81 static void mdt_mfd_put(struct mdt_file_data *mfd)
83 CDEBUG(D_INFO, "PUTting mfd %p : new refcount %d\n", mfd,
84 atomic_read(&mfd->mfd_refcount) - 1);
85 LASSERT(atomic_read(&mfd->mfd_refcount) > 0 &&
86 atomic_read(&mfd->mfd_refcount) < 0x5a5a);
87 if (atomic_dec_and_test(&mfd->mfd_refcount)) {
88 LASSERT(list_empty(&mfd->mfd_handle.h_link));
93 static int mdt_object_open(struct mdt_thread_info *info,
97 struct mdt_export_data *med;
98 struct mdt_file_data *mfd;
99 struct mdt_body *repbody;
100 struct lov_mds_md *lmm;
104 med = &mdt_info_req(info)->rq_export->exp_mdt_data;
105 repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
106 lmm = req_capsule_server_get(&info->mti_pill, &RMF_MDT_MD);
108 rc = mo_attr_get(info->mti_ctxt, mdt_object_child(o),
110 if (rc == -EREMOTE) {
111 repbody->fid1 = *mdt_object_fid(o);
112 repbody->valid |= OBD_MD_FLID;
118 mdt_pack_attr2body(repbody, &info->mti_attr, mdt_object_fid(o));
121 rc = mo_xattr_get(info->mti_ctxt, mdt_object_child(o),
122 lmm, info->mti_mdt->mdt_max_mdsize, "lov");
125 if (S_ISDIR(info->mti_attr.la_mode))
126 repbody->valid |= OBD_MD_FLDIREA;
128 repbody->valid |= OBD_MD_FLEASIZE;
129 repbody->eadatasize = rc;
134 CERROR("mds: out of memory\n");
138 if (flags & FMODE_WRITE) {
139 /*mds_get_write_access*/
140 } else if (flags & MDS_FMODE_EXEC) {
141 /*mds_deny_write_access*/
144 /* keep a reference on this object for this open,
145 * and is released by mdt_mfd_close() */
146 mdt_object_get(info->mti_ctxt, o);
148 mfd->mfd_mode = flags;
150 mfd->mfd_xid = mdt_info_req(info)->rq_xid;
152 spin_lock(&med->med_open_lock);
153 list_add(&mfd->mfd_list, &med->med_open_head);
154 spin_unlock(&med->med_open_lock);
156 repbody->handle.cookie = mfd->mfd_handle.h_cookie;
161 int mdt_open_by_fid(struct mdt_thread_info* info, const struct lu_fid *fid,
162 __u32 flags, struct ldlm_reply *rep)
164 struct mdt_object *o;
168 o = mdt_object_find(info->mti_ctxt, info->mti_mdt, fid);
170 if (mdt_object_exists(info->mti_ctxt, &o->mot_obj.mo_lu)) {
171 intent_set_disposition(rep, DISP_LOOKUP_EXECD);
172 intent_set_disposition(rep, DISP_LOOKUP_POS);
173 rc = mdt_object_open(info, o, flags);
174 intent_set_disposition(rep, DISP_OPEN_OPEN);
175 mdt_object_put(info->mti_ctxt, o);
177 intent_set_disposition(rep, DISP_LOOKUP_EXECD);
178 intent_set_disposition(rep, DISP_LOOKUP_NEG);
187 int mdt_pin(struct mdt_thread_info* info)
191 rc = mdt_open_by_fid(info, &info->mti_body->fid1,
192 info->mti_body->flags, NULL);
196 /* Get an internal lock on the inode number (but not generation) to sync
197 * new inode creation with inode unlink (bug 2029). If child_lockh is NULL
198 * we just get the lock as a barrier to wait for other holders of this lock,
199 * and drop it right away again. */
200 int mdt_lock_new_child(struct mdt_thread_info *info,
201 struct mdt_object *o,
202 struct mdt_lock_handle *child_lockh)
204 struct mdt_lock_handle lockh;
207 if (child_lockh == NULL)
208 child_lockh = &lockh;
210 mdt_lock_handle_init(&lockh);
211 lockh.mlh_mode = LCK_EX;
212 rc = mdt_object_lock(info, o, &lockh, MDS_INODELOCK_UPDATE);
215 CERROR("can not mdt_object_lock: %d\n", rc);
216 else if (child_lockh == &lockh)
217 mdt_object_unlock(info, o, &lockh);
222 int mdt_reint_open(struct mdt_thread_info *info)
224 struct mdt_device *mdt = info->mti_mdt;
225 struct mdt_object *parent;
226 struct mdt_object *child;
227 struct mdt_lock_handle *lh;
228 struct ldlm_reply *ldlm_rep;
229 struct ptlrpc_request *req = mdt_info_req(info);
230 struct mdt_body *body;
231 struct lu_fid *child_fid = &info->mti_tmp_fid1;
234 struct mdt_reint_record *rr = &info->mti_rr;
237 /* we now have no resent message, so it must be an intent */
238 LASSERT(info->mti_pill.rc_fmt == &RQF_LDLM_INTENT_OPEN);
240 /*TODO: MDS_CHECK_RESENT */;
241 ldlm_rep = req_capsule_server_get(&info->mti_pill, &RMF_DLM_REP);
242 body = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
244 if (strlen(rr->rr_name) == 0) {
245 /* partial remote open */
246 RETURN(mdt_open_by_fid(info, rr->rr_fid1,
247 info->mti_attr.la_flags, ldlm_rep));
250 intent_set_disposition(ldlm_rep, DISP_LOOKUP_EXECD);
251 lh = &info->mti_lh[MDT_LH_PARENT];
252 lh->mlh_mode = LCK_PW;
253 parent = mdt_object_find_lock(info, rr->rr_fid1, lh,
254 MDS_INODELOCK_UPDATE);
255 if (IS_ERR(parent)) {
256 /* FIXME: just simulate child not exist */
257 intent_set_disposition(ldlm_rep, DISP_LOOKUP_NEG);
258 GOTO(out, result = PTR_ERR(parent));
261 result = mdo_lookup(info->mti_ctxt, mdt_object_child(parent),
262 rr->rr_name, child_fid);
263 if (result != 0 && result != -ENOENT) {
264 GOTO(out_parent, result);
267 if (result == -ENOENT) {
268 intent_set_disposition(ldlm_rep, DISP_LOOKUP_NEG);
269 if (!(info->mti_attr.la_flags & MDS_OPEN_CREAT))
270 GOTO(out_parent, result);
271 if (req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
272 GOTO(out_parent, result = -EROFS);
273 *child_fid = *info->mti_rr.rr_fid2;
275 intent_set_disposition(ldlm_rep, DISP_LOOKUP_POS);
276 if (info->mti_attr.la_flags & MDS_OPEN_EXCL &&
277 info->mti_attr.la_flags & MDS_OPEN_CREAT)
278 GOTO(out_parent, result = -EEXIST);
279 /* child_fid is filled by mdo_lookup(). */
280 LASSERT(lu_fid_eq(child_fid, info->mti_rr.rr_fid2));
283 child = mdt_object_find(info->mti_ctxt, mdt, child_fid);
285 GOTO(out_parent, result = PTR_ERR(child));
287 if (result == -ENOENT) {
288 /* not found and with MDS_OPEN_CREAT: let's create something */
289 result = mdo_create(info->mti_ctxt,
290 mdt_object_child(parent),
292 mdt_object_child(child),
294 intent_set_disposition(ldlm_rep, DISP_OPEN_CREATE);
296 GOTO(out_child, result);
301 result = mdt_object_open(info, child, info->mti_attr.la_flags);
302 intent_set_disposition(ldlm_rep, DISP_OPEN_OPEN);
303 GOTO(destroy_child, result);
306 if (created && result != 0 && result != -EREMOTE) {
307 mdo_unlink(info->mti_ctxt, mdt_object_child(parent),
308 mdt_object_child(child), rr->rr_name);
309 } else if (created) {
310 /* barrier with other thread */
311 mdt_lock_new_child(info, child, NULL);
314 mdt_object_put(info->mti_ctxt, child);
316 mdt_object_unlock_put(info, parent, lh);
321 int mdt_mfd_close(const struct lu_context *ctxt,
322 struct mdt_file_data *mfd,
327 if (mfd->mfd_mode & FMODE_WRITE) {
328 /*mdt_put_write_access*/
329 } else if (mfd->mfd_mode & MDS_FMODE_EXEC) {
330 /*mdt_allow_write_access*/
333 /* release reference on this object.
334 * it will be destroyed by lower layer if necessary.
336 mdt_object_put(ctxt, mfd->mfd_object);
342 int mdt_close(struct mdt_thread_info *info)
344 struct mdt_export_data *med;
345 struct mdt_body *repbody;
346 struct mdt_file_data *mfd;
347 struct mdt_object *o;
348 struct lov_mds_md *lmm;
352 med = &mdt_info_req(info)->rq_export->exp_mdt_data;
355 spin_lock(&med->med_open_lock);
356 mfd = mdt_handle2mfd(&(info->mti_body->handle));
358 spin_unlock(&med->med_open_lock);
359 CDEBUG(D_INODE, "no handle for file close ino "DFID3
360 ": cookie "LPX64, PFID3(&info->mti_body->fid1),
361 info->mti_body->handle.cookie);
364 class_handle_unhash(&mfd->mfd_handle);
365 list_del_init(&mfd->mfd_list);
366 spin_unlock(&med->med_open_lock);
367 /* mdt_handle2mfd increase reference count, we must drop it here */
371 if (lu_object_is_dying(&o->mot_header)) {
372 repbody = req_capsule_server_get(&info->mti_pill,
374 lmm = req_capsule_server_get(&info->mti_pill, &RMF_MDT_MD);
376 rc = mo_attr_get(info->mti_ctxt, mdt_object_child(o),
379 mdt_pack_attr2body(repbody, &info->mti_attr,
382 rc = mo_xattr_get(info->mti_ctxt, mdt_object_child(o),
383 lmm, info->mti_mdt->mdt_max_mdsize,
386 if (S_ISDIR(info->mti_attr.la_mode))
387 repbody->valid |= OBD_MD_FLDIREA;
389 repbody->valid |= OBD_MD_FLEASIZE;
390 repbody->eadatasize = rc;
396 rc = mdt_mfd_close(info->mti_ctxt, mfd, 1);
401 int mdt_done_writing(struct mdt_thread_info *info)