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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_LOV
39 #include <libcfs/libcfs.h>
41 #include <obd_class.h>
42 #include <lustre/lustre_idl.h>
44 #include "lov_internal.h"
46 static void lov_init_set(struct lov_request_set *set)
49 atomic_set(&set->set_completes, 0);
50 atomic_set(&set->set_success, 0);
51 atomic_set(&set->set_finish_checked, 0);
52 INIT_LIST_HEAD(&set->set_list);
53 atomic_set(&set->set_refcount, 1);
54 init_waitqueue_head(&set->set_waitq);
57 void lov_finish_set(struct lov_request_set *set)
59 struct list_head *pos, *n;
60 struct lov_request *req;
64 list_for_each_safe(pos, n, &set->set_list) {
65 req = list_entry(pos, struct lov_request, rq_link);
66 list_del_init(&req->rq_link);
68 if (req->rq_oi.oi_osfs != NULL)
69 OBD_FREE_PTR(req->rq_oi.oi_osfs);
78 int lov_set_finished(struct lov_request_set *set, int idempotent)
80 int completes = atomic_read(&set->set_completes);
82 CDEBUG(D_INFO, "check set %d/%d\n", completes, set->set_count);
84 if (completes == set->set_count) {
87 if (atomic_inc_return(&set->set_finish_checked) == 1)
93 void lov_update_set(struct lov_request_set *set,
94 struct lov_request *req, int rc)
99 atomic_inc(&set->set_completes);
101 atomic_inc(&set->set_success);
103 wake_up(&set->set_waitq);
106 void lov_set_add_req(struct lov_request *req, struct lov_request_set *set)
108 list_add_tail(&req->rq_link, &set->set_list);
113 static int lov_check_set(struct lov_obd *lov, int idx)
116 mutex_lock(&lov->lov_lock);
118 if (lov->lov_tgts[idx] == NULL ||
119 lov->lov_tgts[idx]->ltd_active ||
120 (lov->lov_tgts[idx]->ltd_exp != NULL &&
121 class_exp2cliimp(lov->lov_tgts[idx]->ltd_exp)->imp_connect_tried))
124 mutex_unlock(&lov->lov_lock);
128 /* Check if the OSC connection exists and is active.
129 * If the OSC has not yet had a chance to connect to the OST the first time,
130 * wait once for it to connect instead of returning an error.
132 int lov_check_and_wait_active(struct lov_obd *lov, int ost_idx)
134 wait_queue_head_t waitq;
135 struct l_wait_info lwi;
136 struct lov_tgt_desc *tgt;
139 mutex_lock(&lov->lov_lock);
141 tgt = lov->lov_tgts[ost_idx];
143 if (unlikely(tgt == NULL))
146 if (likely(tgt->ltd_active))
149 if (tgt->ltd_exp && class_exp2cliimp(tgt->ltd_exp)->imp_connect_tried)
152 mutex_unlock(&lov->lov_lock);
154 init_waitqueue_head(&waitq);
155 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(obd_timeout),
156 cfs_time_seconds(1), NULL, NULL);
158 rc = l_wait_event(waitq, lov_check_set(lov, ost_idx), &lwi);
165 mutex_unlock(&lov->lov_lock);
169 #define LOV_U64_MAX ((__u64)~0ULL)
170 #define LOV_SUM_MAX(tot, add) \
172 if ((tot) + (add) < (tot)) \
173 (tot) = LOV_U64_MAX; \
178 int lov_fini_statfs(struct obd_device *obd, struct obd_statfs *osfs,int success)
183 __u32 expected_stripes = lov_get_stripecnt(&obd->u.lov,
185 if (osfs->os_files != LOV_U64_MAX)
186 lov_do_div64(osfs->os_files, expected_stripes);
187 if (osfs->os_ffree != LOV_U64_MAX)
188 lov_do_div64(osfs->os_ffree, expected_stripes);
190 spin_lock(&obd->obd_osfs_lock);
191 memcpy(&obd->obd_osfs, osfs, sizeof(*osfs));
192 obd->obd_osfs_age = cfs_time_current_64();
193 spin_unlock(&obd->obd_osfs_lock);
200 int lov_fini_statfs_set(struct lov_request_set *set)
208 if (atomic_read(&set->set_completes)) {
209 rc = lov_fini_statfs(set->set_obd, set->set_oi->oi_osfs,
210 atomic_read(&set->set_success));
216 void lov_update_statfs(struct obd_statfs *osfs, struct obd_statfs *lov_sfs,
219 int shift = 0, quit = 0;
223 memcpy(osfs, lov_sfs, sizeof(*lov_sfs));
225 if (osfs->os_bsize != lov_sfs->os_bsize) {
226 /* assume all block sizes are always powers of 2 */
227 /* get the bits difference */
228 tmp = osfs->os_bsize | lov_sfs->os_bsize;
229 for (shift = 0; shift <= 64; ++shift) {
241 if (osfs->os_bsize < lov_sfs->os_bsize) {
242 osfs->os_bsize = lov_sfs->os_bsize;
244 osfs->os_bfree >>= shift;
245 osfs->os_bavail >>= shift;
246 osfs->os_blocks >>= shift;
247 } else if (shift != 0) {
248 lov_sfs->os_bfree >>= shift;
249 lov_sfs->os_bavail >>= shift;
250 lov_sfs->os_blocks >>= shift;
253 /* Sandia requested that df (and so, statfs) only
254 returned minimal available space on
255 a single OST, so people would be able to
256 write this much data guaranteed. */
257 if (osfs->os_bavail > lov_sfs->os_bavail) {
258 /* Presumably if new bavail is smaller,
259 new bfree is bigger as well */
260 osfs->os_bfree = lov_sfs->os_bfree;
261 osfs->os_bavail = lov_sfs->os_bavail;
264 osfs->os_bfree += lov_sfs->os_bfree;
265 osfs->os_bavail += lov_sfs->os_bavail;
267 osfs->os_blocks += lov_sfs->os_blocks;
268 /* XXX not sure about this one - depends on policy.
269 * - could be minimum if we always stripe on all OBDs
270 * (but that would be wrong for any other policy,
271 * if one of the OBDs has no more objects left)
272 * - could be sum if we stripe whole objects
273 * - could be average, just to give a nice number
275 * To give a "reasonable" (if not wholly accurate)
276 * number, we divide the total number of free objects
277 * by expected stripe count (watch out for overflow).
279 LOV_SUM_MAX(osfs->os_files, lov_sfs->os_files);
280 LOV_SUM_MAX(osfs->os_ffree, lov_sfs->os_ffree);
284 /* The callback for osc_statfs_async that finilizes a request info when a
285 * response is received. */
286 static int cb_statfs_update(void *cookie, int rc)
288 struct obd_info *oinfo = cookie;
289 struct lov_request *lovreq;
290 struct lov_request_set *set;
291 struct obd_statfs *osfs, *lov_sfs;
293 struct lov_tgt_desc *tgt;
294 struct obd_device *lovobd, *tgtobd;
298 lovreq = container_of(oinfo, struct lov_request, rq_oi);
299 set = lovreq->rq_rqset;
300 lovobd = set->set_obd;
301 lov = &lovobd->u.lov;
302 osfs = set->set_oi->oi_osfs;
303 lov_sfs = oinfo->oi_osfs;
304 success = atomic_read(&set->set_success);
305 /* XXX: the same is done in lov_update_common_set, however
306 lovset->set_exp is not initialized. */
307 lov_update_set(set, lovreq, rc);
312 tgt = lov->lov_tgts[lovreq->rq_idx];
313 if (!tgt || !tgt->ltd_active)
314 GOTO(out_update, rc);
316 tgtobd = class_exp2obd(tgt->ltd_exp);
317 spin_lock(&tgtobd->obd_osfs_lock);
318 memcpy(&tgtobd->obd_osfs, lov_sfs, sizeof(*lov_sfs));
319 if ((oinfo->oi_flags & OBD_STATFS_FROM_CACHE) == 0)
320 tgtobd->obd_osfs_age = cfs_time_current_64();
321 spin_unlock(&tgtobd->obd_osfs_lock);
324 lov_update_statfs(osfs, lov_sfs, success);
328 if (set->set_oi->oi_flags & OBD_STATFS_PTLRPCD &&
329 lov_set_finished(set, 0)) {
330 lov_statfs_interpret(NULL, set, set->set_count !=
331 atomic_read(&set->set_success));
337 int lov_prep_statfs_set(struct obd_device *obd, struct obd_info *oinfo,
338 struct lov_request_set **reqset)
340 struct lov_request_set *set;
341 struct lov_obd *lov = &obd->u.lov;
345 OBD_ALLOC(set, sizeof(*set));
353 /* We only get block data from the OBD */
354 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
355 struct lov_request *req;
357 if (lov->lov_tgts[i] == NULL ||
358 (oinfo->oi_flags & OBD_STATFS_NODELAY &&
359 !lov->lov_tgts[i]->ltd_active)) {
360 CDEBUG(D_HA, "lov idx %d inactive\n", i);
364 if (!lov->lov_tgts[i]->ltd_active)
365 lov_check_and_wait_active(lov, i);
367 /* skip targets that have been explicitely disabled by the
369 if (!lov->lov_tgts[i]->ltd_exp) {
370 CDEBUG(D_HA, "lov idx %d administratively disabled\n", i);
374 OBD_ALLOC(req, sizeof(*req));
376 GOTO(out_set, rc = -ENOMEM);
378 OBD_ALLOC(req->rq_oi.oi_osfs, sizeof(*req->rq_oi.oi_osfs));
379 if (req->rq_oi.oi_osfs == NULL) {
380 OBD_FREE(req, sizeof(*req));
381 GOTO(out_set, rc = -ENOMEM);
385 req->rq_oi.oi_cb_up = cb_statfs_update;
386 req->rq_oi.oi_flags = oinfo->oi_flags;
388 lov_set_add_req(req, set);
391 GOTO(out_set, rc = -EIO);
395 lov_fini_statfs_set(set);