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, 2017, 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] lock LDLM lock on resource
82 * \retval 0 on successful setup
83 * \retval negative value on error
85 static int ofd_lvbo_init(const struct lu_env *env, struct ldlm_resource *res)
88 struct ofd_device *ofd;
89 struct ofd_object *fo;
90 struct ofd_thread_info *info;
96 LASSERT(mutex_is_locked(&res->lr_lvb_mutex));
98 if (res->lr_lvb_data != NULL)
101 ofd = ldlm_res_to_ns(res)->ns_lvbp;
102 LASSERT(ofd != NULL);
104 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_OST_LVB))
108 rc = lu_env_init(&_env, LCT_DT_THREAD);
116 GOTO(out, rc = -ENOMEM);
118 info = ofd_info(env);
119 res->lr_lvb_data = lvb;
120 res->lr_lvb_len = sizeof(*lvb);
122 ost_fid_from_resid(&info->fti_fid, &res->lr_name,
123 ofd->ofd_lut.lut_lsd.lsd_osd_index);
124 fo = ofd_object_find(env, ofd, &info->fti_fid);
126 GOTO(out_lvb, rc = PTR_ERR(fo));
128 rc = ofd_attr_get(env, fo, &info->fti_attr);
132 lvb->lvb_size = info->fti_attr.la_size;
133 lvb->lvb_blocks = info->fti_attr.la_blocks;
134 lvb->lvb_mtime = info->fti_attr.la_mtime;
135 lvb->lvb_atime = info->fti_attr.la_atime;
136 lvb->lvb_ctime = info->fti_attr.la_ctime;
138 CDEBUG(D_DLMTRACE, "res: "DFID" initial lvb size: %llu, "
139 "mtime: %#llx, blocks: %#llx\n",
140 PFID(&info->fti_fid), lvb->lvb_size,
141 lvb->lvb_mtime, lvb->lvb_blocks);
143 info->fti_attr.la_valid = 0;
147 ofd_object_put(env, fo);
150 OST_LVB_SET_ERR(lvb->lvb_blocks, rc);
152 /* Don't free lvb data on lookup error */
153 if (env && env == &_env)
159 * Implementation of ldlm_valblock_ops::lvbo_update for OFD.
161 * When a client generates a glimpse enqueue, it wants to get the current
162 * file size and updated attributes for a stat() type operation, but these
163 * attributes may be writeback cached on another client. The client with
164 * the DLM extent lock at the highest offset is asked for its current
165 * attributes via a glimpse callback on its extent lock, on the assumption
166 * that it has the highest file size and the newest timestamps. The timestamps
167 * are guaranteed to be correct if there is only a single writer on the file,
168 * but may be slightly inaccurate if there are multiple concurrent writers on
169 * the same object. In order to avoid race conditions between the glimpse AST
170 * and the client cancelling the lock, ofd_lvbo_update() also updates
171 * the attributes from the local object. If the last client hasn't done any
172 * writes yet, or has already written its data and cancelled its lock before
173 * it processed the glimpse, then the local inode will have more uptodate
176 * This is called in two ways:
177 * \a req != NULL : called by the DLM itself after a glimpse callback
178 * \a req == NULL : called by the OFD after a disk write
180 * \param[in] lock LDLM lock
181 * \param[in] req PTLRPC request
182 * \param[in] increase_only don't allow LVB values to decrease
184 * \retval 0 on successful setup
185 * \retval negative value on error
187 static int ofd_lvbo_update(const struct lu_env *env, struct ldlm_resource *res,
188 struct ldlm_lock *lock, struct ptlrpc_request *req,
191 struct ofd_thread_info *info;
192 struct ofd_device *ofd;
193 struct ofd_object *fo;
200 info = ofd_info(env);
201 LASSERT(res != NULL);
203 ofd = ldlm_res_to_ns(res)->ns_lvbp;
204 LASSERT(ofd != NULL);
206 fid_extract_from_res_name(&info->fti_fid, &res->lr_name);
208 lvb = res->lr_lvb_data;
210 CERROR("%s: no LVB data for "DFID"\n",
211 ofd_name(ofd), PFID(&info->fti_fid));
215 /* Update the LVB from the network message */
217 struct ost_lvb *rpc_lvb;
220 if (req->rq_import != NULL)
221 lvb_type = imp_connect_lvb_type(req->rq_import);
223 lvb_type = exp_connect_lvb_type(req->rq_export);
226 struct ost_lvb_v1 *lvb_v1;
228 lvb_v1 = req_capsule_server_swab_get(&req->rq_pill,
229 &RMF_DLM_LVB, lustre_swab_ost_lvb_v1);
233 rpc_lvb = &info->fti_lvb;
234 memcpy(rpc_lvb, lvb_v1, sizeof *lvb_v1);
235 rpc_lvb->lvb_mtime_ns = 0;
236 rpc_lvb->lvb_atime_ns = 0;
237 rpc_lvb->lvb_ctime_ns = 0;
239 rpc_lvb = req_capsule_server_swab_get(&req->rq_pill,
241 lustre_swab_ost_lvb);
247 if (rpc_lvb->lvb_size > lvb->lvb_size || !increase_only) {
248 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb size: "
249 "%llu -> %llu\n", PFID(&info->fti_fid),
250 lvb->lvb_size, rpc_lvb->lvb_size);
251 lvb->lvb_size = rpc_lvb->lvb_size;
253 if (rpc_lvb->lvb_mtime > lvb->lvb_mtime || !increase_only) {
254 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb mtime: "
255 "%llu -> %llu\n", PFID(&info->fti_fid),
256 lvb->lvb_mtime, rpc_lvb->lvb_mtime);
257 lvb->lvb_mtime = rpc_lvb->lvb_mtime;
259 if (rpc_lvb->lvb_atime > lvb->lvb_atime || !increase_only) {
260 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb atime: "
261 "%llu -> %llu\n", PFID(&info->fti_fid),
262 lvb->lvb_atime, rpc_lvb->lvb_atime);
263 lvb->lvb_atime = rpc_lvb->lvb_atime;
265 if (rpc_lvb->lvb_ctime > lvb->lvb_ctime || !increase_only) {
266 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb ctime: "
267 "%llu -> %llu\n", PFID(&info->fti_fid),
268 lvb->lvb_ctime, rpc_lvb->lvb_ctime);
269 lvb->lvb_ctime = rpc_lvb->lvb_ctime;
271 if (rpc_lvb->lvb_blocks > lvb->lvb_blocks || !increase_only) {
272 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb blocks: "
273 "%llu -> %llu\n", PFID(&info->fti_fid),
274 lvb->lvb_blocks, rpc_lvb->lvb_blocks);
275 lvb->lvb_blocks = rpc_lvb->lvb_blocks;
281 /* Update the LVB from the disk inode */
282 ost_fid_from_resid(&info->fti_fid, &res->lr_name,
283 ofd->ofd_lut.lut_lsd.lsd_osd_index);
284 fo = ofd_object_find(env, ofd, &info->fti_fid);
286 GOTO(out, rc = PTR_ERR(fo));
288 rc = ofd_attr_get(env, fo, &info->fti_attr);
293 if (info->fti_attr.la_size > lvb->lvb_size || !increase_only) {
294 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb size from disk: "
295 "%llu -> %llu\n", PFID(&info->fti_fid),
296 lvb->lvb_size, info->fti_attr.la_size);
297 lvb->lvb_size = info->fti_attr.la_size;
300 if (info->fti_attr.la_mtime >lvb->lvb_mtime || !increase_only) {
301 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb mtime from disk: "
302 "%llu -> %llu\n", PFID(&info->fti_fid),
303 lvb->lvb_mtime, info->fti_attr.la_mtime);
304 lvb->lvb_mtime = info->fti_attr.la_mtime;
306 if (info->fti_attr.la_atime >lvb->lvb_atime || !increase_only) {
307 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb atime from disk: "
308 "%llu -> %llu\n", PFID(&info->fti_fid),
309 lvb->lvb_atime, info->fti_attr.la_atime);
310 lvb->lvb_atime = info->fti_attr.la_atime;
312 if (info->fti_attr.la_ctime >lvb->lvb_ctime || !increase_only) {
313 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb ctime from disk: "
314 "%llu -> %llu\n", PFID(&info->fti_fid),
315 lvb->lvb_ctime, info->fti_attr.la_ctime);
316 lvb->lvb_ctime = info->fti_attr.la_ctime;
318 if (info->fti_attr.la_blocks > lvb->lvb_blocks || !increase_only) {
319 CDEBUG(D_DLMTRACE, "res: "DFID" updating lvb blocks from disk: "
320 "%llu -> %llu\n", PFID(&info->fti_fid), lvb->lvb_blocks,
321 (unsigned long long)info->fti_attr.la_blocks);
322 lvb->lvb_blocks = info->fti_attr.la_blocks;
327 ofd_object_put(env, fo);
333 * Implementation of ldlm_valblock_ops::lvbo_size for OFD.
335 * This function returns size of LVB data so appropriate RPC size will be
336 * reserved. This is used for compatibility needs between server and client
337 * of different Lustre versions.
339 * \param[in] lock LDLM lock
341 * \retval size of LVB data
343 static int ofd_lvbo_size(struct ldlm_lock *lock)
345 if (lock->l_export != NULL && exp_connect_lvb_type(lock->l_export))
346 return sizeof(struct ost_lvb);
348 return sizeof(struct ost_lvb_v1);
352 * Implementation of ldlm_valblock_ops::lvbo_fill for OFD.
354 * This function is called to fill the given RPC buffer \a buf with LVB data
356 * \param[in] env execution environment
357 * \param[in] lock LDLM lock
358 * \param[in] buf RPC buffer to fill
359 * \param[in] buflen buffer length
361 * \retval size of LVB data written into \a buf buffer
363 static int ofd_lvbo_fill(const struct lu_env *env, struct ldlm_lock *lock,
364 void *buf, int *buflen)
366 struct ldlm_resource *res = lock->l_resource;
369 /* Former lvbo_init not allocate the "LVB". */
370 if (unlikely(res->lr_lvb_len == 0))
373 lvb_len = ofd_lvbo_size(lock);
374 LASSERT(lvb_len <= res->lr_lvb_len);
376 if (lvb_len > *buflen)
380 memcpy(buf, res->lr_lvb_data, lvb_len);
386 struct ldlm_valblock_ops ofd_lvbo = {
387 .lvbo_init = ofd_lvbo_init,
388 .lvbo_update = ofd_lvbo_update,
389 .lvbo_free = ofd_lvbo_free,
390 .lvbo_size = ofd_lvbo_size,
391 .lvbo_fill = ofd_lvbo_fill