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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2014, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/ofd/ofd_lvb.c
34 * This file contains methods for OBD Filter Device (OFD)
35 * Lock Value Block (LVB) operations.
37 * LVB is special opaque (to LDLM) data that is associated with an LDLM lock
38 * and transferred from client to server and back. OFD LVBs are used to
39 * maintain current object size/times.
41 * Author: Andreas Dilger <andreas.dilger@intel.com>
42 * Author: Mikhail Pershin <mike.pershin@intel.com>
43 * Author: Alexey Zhuravlev <alexey.zhuravlev@intel.com>
46 #define DEBUG_SUBSYSTEM S_FILTER
48 #include <lustre_swab.h>
49 #include "ofd_internal.h"
52 * Implementation of ldlm_valblock_ops::lvbo_free for OFD.
54 * This function frees allocated LVB data if it associated with the given
57 * \param[in] res LDLM resource
59 * \retval 0 on successful setup
60 * \retval negative value on error
62 static int ofd_lvbo_free(struct ldlm_resource *res)
65 OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
71 * Implementation of ldlm_valblock_ops::lvbo_init for OFD.
73 * This function allocates and initializes new LVB data for the given
74 * LDLM resource if it is not allocated yet. New LVB is filled with attributes
75 * of the object associated with that resource. Function does nothing if LVB
76 * for the given LDLM resource is allocated already.
78 * Called with res->lr_lvb_sem held.
80 * \param[in] res LDLM resource
82 * \retval 0 on successful setup
83 * \retval negative value on error
85 static int ofd_lvbo_init(struct ldlm_resource *res)
88 struct ofd_device *ofd;
89 struct ofd_object *fo;
90 struct ofd_thread_info *info;
97 LASSERT(mutex_is_locked(&res->lr_lvb_mutex));
99 if (res->lr_lvb_data != NULL)
102 ofd = ldlm_res_to_ns(res)->ns_lvbp;
103 LASSERT(ofd != NULL);
105 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_OST_LVB))
108 rc = lu_env_init(&env, LCT_DT_THREAD);
114 GOTO(out_env, rc = -ENOMEM);
116 res->lr_lvb_data = lvb;
117 res->lr_lvb_len = sizeof(*lvb);
119 info = ofd_info_init(&env, NULL);
120 ost_fid_from_resid(&info->fti_fid, &res->lr_name,
121 ofd->ofd_lut.lut_lsd.lsd_osd_index);
122 fo = ofd_object_find(&env, ofd, &info->fti_fid);
124 GOTO(out_lvb, rc = PTR_ERR(fo));
126 rc = ofd_attr_get(&env, fo, &info->fti_attr);
130 lvb->lvb_size = info->fti_attr.la_size;
131 lvb->lvb_blocks = info->fti_attr.la_blocks;
132 lvb->lvb_mtime = info->fti_attr.la_mtime;
133 lvb->lvb_atime = info->fti_attr.la_atime;
134 lvb->lvb_ctime = info->fti_attr.la_ctime;
136 CDEBUG(D_DLMTRACE, "res: "DFID" initial lvb size: %llu, "
137 "mtime: %#llx, blocks: %#llx\n",
138 PFID(&info->fti_fid), lvb->lvb_size,
139 lvb->lvb_mtime, lvb->lvb_blocks);
143 ofd_object_put(&env, fo);
146 OST_LVB_SET_ERR(lvb->lvb_blocks, rc);
149 /* Don't free lvb data on lookup error */
154 * Implementation of ldlm_valblock_ops::lvbo_update for OFD.
156 * When a client generates a glimpse enqueue, it wants to get the current
157 * file size and updated attributes for a stat() type operation, but these
158 * attributes may be writeback cached on another client. The client with
159 * the DLM extent lock at the highest offset is asked for its current
160 * attributes via a glimpse callback on its extent lock, on the assumption
161 * that it has the highest file size and the newest timestamps. The timestamps
162 * are guaranteed to be correct if there is only a single writer on the file,
163 * but may be slightly inaccurate if there are multiple concurrent writers on
164 * the same object. In order to avoid race conditions between the glimpse AST
165 * and the client cancelling the lock, ofd_lvbo_update() also updates
166 * the attributes from the local object. If the last client hasn't done any
167 * writes yet, or has already written its data and cancelled its lock before
168 * it processed the glimpse, then the local inode will have more uptodate
171 * This is called in two ways:
172 * \a req != NULL : called by the DLM itself after a glimpse callback
173 * \a req == NULL : called by the OFD after a disk write
175 * \param[in] res LDLM resource
176 * \param[in] req PTLRPC request
177 * \param[in] increase_only don't allow LVB values to decrease
179 * \retval 0 on successful setup
180 * \retval negative value on error
182 static int ofd_lvbo_update(struct ldlm_resource *res,
183 struct ptlrpc_request *req, int increase_only)
185 struct ofd_device *ofd;
186 struct ofd_object *fo;
187 struct ofd_thread_info *info;
194 LASSERT(res != NULL);
196 ofd = ldlm_res_to_ns(res)->ns_lvbp;
197 LASSERT(ofd != NULL);
199 rc = lu_env_init(&env, LCT_DT_THREAD);
203 info = ofd_info_init(&env, NULL);
204 fid_extract_from_res_name(&info->fti_fid, &res->lr_name);
206 lvb = res->lr_lvb_data;
208 CERROR("%s: no LVB data for "DFID"\n",
209 ofd_name(ofd), PFID(&info->fti_fid));
210 GOTO(out_env, rc = 0);
213 /* Update the LVB from the network message */
215 struct ost_lvb *rpc_lvb;
218 if (req->rq_import != NULL)
219 lvb_type = imp_connect_lvb_type(req->rq_import);
221 lvb_type = exp_connect_lvb_type(req->rq_export);
224 struct ost_lvb_v1 *lvb_v1;
226 lvb_v1 = req_capsule_server_swab_get(&req->rq_pill,
227 &RMF_DLM_LVB, lustre_swab_ost_lvb_v1);
231 rpc_lvb = &info->fti_lvb;
232 memcpy(rpc_lvb, lvb_v1, sizeof *lvb_v1);
233 rpc_lvb->lvb_mtime_ns = 0;
234 rpc_lvb->lvb_atime_ns = 0;
235 rpc_lvb->lvb_ctime_ns = 0;
237 rpc_lvb = req_capsule_server_swab_get(&req->rq_pill,
239 lustre_swab_ost_lvb);
245 if (rpc_lvb->lvb_size > lvb->lvb_size || !increase_only) {
246 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb size: "
247 "%llu -> %llu\n", PFID(&info->fti_fid),
248 lvb->lvb_size, rpc_lvb->lvb_size);
249 lvb->lvb_size = rpc_lvb->lvb_size;
251 if (rpc_lvb->lvb_mtime > lvb->lvb_mtime || !increase_only) {
252 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb mtime: "
253 "%llu -> %llu\n", PFID(&info->fti_fid),
254 lvb->lvb_mtime, rpc_lvb->lvb_mtime);
255 lvb->lvb_mtime = rpc_lvb->lvb_mtime;
257 if (rpc_lvb->lvb_atime > lvb->lvb_atime || !increase_only) {
258 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb atime: "
259 "%llu -> %llu\n", PFID(&info->fti_fid),
260 lvb->lvb_atime, rpc_lvb->lvb_atime);
261 lvb->lvb_atime = rpc_lvb->lvb_atime;
263 if (rpc_lvb->lvb_ctime > lvb->lvb_ctime || !increase_only) {
264 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb ctime: "
265 "%llu -> %llu\n", PFID(&info->fti_fid),
266 lvb->lvb_ctime, rpc_lvb->lvb_ctime);
267 lvb->lvb_ctime = rpc_lvb->lvb_ctime;
269 if (rpc_lvb->lvb_blocks > lvb->lvb_blocks || !increase_only) {
270 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb blocks: "
271 "%llu -> %llu\n", PFID(&info->fti_fid),
272 lvb->lvb_blocks, rpc_lvb->lvb_blocks);
273 lvb->lvb_blocks = rpc_lvb->lvb_blocks;
279 /* Update the LVB from the disk inode */
280 ost_fid_from_resid(&info->fti_fid, &res->lr_name,
281 ofd->ofd_lut.lut_lsd.lsd_osd_index);
282 fo = ofd_object_find(&env, ofd, &info->fti_fid);
284 GOTO(out_env, rc = PTR_ERR(fo));
286 rc = ofd_attr_get(&env, fo, &info->fti_attr);
291 if (info->fti_attr.la_size > lvb->lvb_size || !increase_only) {
292 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb size from disk: "
293 "%llu -> %llu\n", PFID(&info->fti_fid),
294 lvb->lvb_size, info->fti_attr.la_size);
295 lvb->lvb_size = info->fti_attr.la_size;
298 if (info->fti_attr.la_mtime >lvb->lvb_mtime || !increase_only) {
299 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb mtime from disk: "
300 "%llu -> %llu\n", PFID(&info->fti_fid),
301 lvb->lvb_mtime, info->fti_attr.la_mtime);
302 lvb->lvb_mtime = info->fti_attr.la_mtime;
304 if (info->fti_attr.la_atime >lvb->lvb_atime || !increase_only) {
305 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb atime from disk: "
306 "%llu -> %llu\n", PFID(&info->fti_fid),
307 lvb->lvb_atime, info->fti_attr.la_atime);
308 lvb->lvb_atime = info->fti_attr.la_atime;
310 if (info->fti_attr.la_ctime >lvb->lvb_ctime || !increase_only) {
311 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb ctime from disk: "
312 "%llu -> %llu\n", PFID(&info->fti_fid),
313 lvb->lvb_ctime, info->fti_attr.la_ctime);
314 lvb->lvb_ctime = info->fti_attr.la_ctime;
316 if (info->fti_attr.la_blocks > lvb->lvb_blocks || !increase_only) {
317 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb blocks from disk: "
318 "%llu -> %llu\n", PFID(&info->fti_fid), lvb->lvb_blocks,
319 (unsigned long long)info->fti_attr.la_blocks);
320 lvb->lvb_blocks = info->fti_attr.la_blocks;
325 ofd_object_put(&env, fo);
332 * Implementation of ldlm_valblock_ops::lvbo_size for OFD.
334 * This function returns size of LVB data so appropriate RPC size will be
335 * reserved. This is used for compatibility needs between server and client
336 * of different Lustre versions.
338 * \param[in] lock LDLM lock
340 * \retval size of LVB data
342 static int ofd_lvbo_size(struct ldlm_lock *lock)
344 if (lock->l_export != NULL && exp_connect_lvb_type(lock->l_export))
345 return sizeof(struct ost_lvb);
347 return sizeof(struct ost_lvb_v1);
351 * Implementation of ldlm_valblock_ops::lvbo_fill for OFD.
353 * This function is called to fill the given RPC buffer \a buf with LVB data
355 * \param[in] lock LDLM lock
356 * \param[in] buf RPC buffer to fill
357 * \param[in] buflen buffer length
359 * \retval size of LVB data written into \a buf buffer
361 static int ofd_lvbo_fill(struct ldlm_lock *lock, void *buf, int buflen)
363 struct ldlm_resource *res = lock->l_resource;
366 /* Former lvbo_init not allocate the "LVB". */
367 if (unlikely(res->lr_lvb_len == 0))
370 lvb_len = ofd_lvbo_size(lock);
371 LASSERT(lvb_len <= res->lr_lvb_len);
373 if (lvb_len > buflen)
377 memcpy(buf, res->lr_lvb_data, lvb_len);
383 struct ldlm_valblock_ops ofd_lvbo = {
384 .lvbo_init = ofd_lvbo_init,
385 .lvbo_update = ofd_lvbo_update,
386 .lvbo_free = ofd_lvbo_free,
387 .lvbo_size = ofd_lvbo_size,
388 .lvbo_fill = ofd_lvbo_fill