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.
5 * Author Peter Braam <braam@clusterfs.com>
7 * This file is part of Lustre, http://www.lustre.org.
9 * Lustre is free software; you can redistribute it and/or
10 * modify it under the terms of version 2 of the GNU General Public
11 * License as published by the Free Software Foundation.
13 * Lustre is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with Lustre; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22 * For testing and management it is treated as an obd_device,
23 * although * it does not export a full OBD method table (the
24 * requests are coming * in over the wire, so object target modules
25 * do not have a full * method table.)
30 # define EXPORT_SYMTAB
32 #define DEBUG_SUBSYSTEM S_OSC
35 # include <linux/version.h>
36 # include <linux/module.h>
37 # include <linux/mm.h>
38 # include <linux/highmem.h>
39 # include <linux/ctype.h>
40 # include <linux/init.h>
41 # if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
42 # include <linux/workqueue.h>
43 # include <linux/smp_lock.h>
45 # include <linux/locks.h>
47 #else /* __KERNEL__ */
48 # include <liblustre.h>
55 # include <linux/lustre_dlm.h>
56 #include <linux/obd_class.h>
57 #include "osc_internal.h"
59 static int osc_interpret_create(struct ptlrpc_request *req, void *data,
62 struct osc_creator *oscc;
63 struct ost_body *body = NULL;
67 body = lustre_swab_repbuf(req, 0, sizeof(*body),
68 lustre_swab_ost_body);
69 if (body == NULL && rc == 0)
73 oscc = req->rq_async_args.pointer_arg[0];
74 spin_lock(&oscc->oscc_lock);
76 oscc->oscc_last_id = body->oa.o_id;
78 DEBUG_REQ(D_INODE, req, "OST out of space, flagging");
79 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
80 } else if (rc != 0 && rc != -EIO) {
81 DEBUG_REQ(D_ERROR, req,
82 "unknown rc %d from async create: failing oscc",
84 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
85 ptlrpc_fail_import(req->rq_import, req->rq_import_generation);
87 oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
88 spin_unlock(&oscc->oscc_lock);
90 CDEBUG(D_INFO, "preallocated through id "LPU64" (last used "LPU64")\n",
91 oscc->oscc_last_id, oscc->oscc_next_id);
93 wake_up(&oscc->oscc_waitq);
97 static int oscc_internal_create(struct osc_creator *oscc)
99 struct ptlrpc_request *request;
100 struct ost_body *body;
101 int size = sizeof(*body);
104 spin_lock(&oscc->oscc_lock);
105 if (oscc->oscc_flags & OSCC_FLAG_CREATING) {
106 spin_unlock(&oscc->oscc_lock);
109 oscc->oscc_flags |= OSCC_FLAG_CREATING;
110 spin_unlock(&oscc->oscc_lock);
112 request = ptlrpc_prep_req(oscc->oscc_obd->u.cli.cl_import, OST_CREATE,
114 if (request == NULL) {
115 spin_lock(&oscc->oscc_lock);
116 oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
117 spin_unlock(&oscc->oscc_lock);
121 request->rq_request_portal = OST_CREATE_PORTAL; //XXX FIXME bug 249
122 body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof(*body));
124 spin_lock(&oscc->oscc_lock);
125 body->oa.o_id = oscc->oscc_last_id + oscc->oscc_grow_count;
126 body->oa.o_gr = oscc->oscc_gr;
127 LASSERT(body->oa.o_gr > 0);
128 body->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
129 CDEBUG(D_INFO, "preallocating through id "LPU64" (last used "LPU64")\n",
130 body->oa.o_id, oscc->oscc_next_id);
131 spin_unlock(&oscc->oscc_lock);
133 request->rq_replen = lustre_msg_size(1, &size);
135 request->rq_async_args.pointer_arg[0] = oscc;
136 request->rq_interpret_reply = osc_interpret_create;
137 ptlrpcd_add_req(request);
142 static int oscc_has_objects(struct osc_creator *oscc, int count)
145 spin_lock(&oscc->oscc_lock);
146 have_objs = ((__s64)(oscc->oscc_last_id - oscc->oscc_next_id) >= count);
147 spin_unlock(&oscc->oscc_lock);
150 oscc_internal_create(oscc);
155 static int oscc_wait_for_objects(struct osc_creator *oscc, int count)
161 have_objs = oscc_has_objects(oscc, count);
163 spin_lock(&oscc->oscc_lock);
164 ost_full = (oscc->oscc_flags & OSCC_FLAG_NOSPC);
165 spin_unlock(&oscc->oscc_lock);
167 osc_invalid = oscc->oscc_obd->u.cli.cl_import->imp_invalid;
169 return have_objs || ost_full || osc_invalid;
172 static int oscc_precreate(struct osc_creator *oscc, int wait)
174 struct l_wait_info lwi = { 0 };
178 if (oscc_has_objects(oscc, oscc->oscc_kick_barrier))
184 /* no rc check -- a no-INTR, no-TIMEOUT wait can't fail */
185 l_wait_event(oscc->oscc_waitq, oscc_wait_for_objects(oscc, 1), &lwi);
187 if (!oscc_has_objects(oscc, 1) && (oscc->oscc_flags & OSCC_FLAG_NOSPC))
190 if (oscc->oscc_obd->u.cli.cl_import->imp_invalid)
196 int oscc_recovering(struct osc_creator *oscc)
200 spin_lock(&oscc->oscc_lock);
201 recov = oscc->oscc_flags & OSCC_FLAG_RECOVERING;
202 spin_unlock(&oscc->oscc_lock);
207 int osc_create(struct obd_export *exp, struct obdo *oa,
208 struct lov_stripe_md **ea, struct obd_trans_info *oti)
210 struct lov_stripe_md *lsm;
211 struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc;
212 int try_again = 1, rc = 0;
216 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
217 LASSERT(oa->o_gr > 0);
219 LASSERT(oscc->oscc_gr == 0 || oscc->oscc_gr == oa->o_gr);
220 oscc->oscc_gr = oa->o_gr;
222 if (oa->o_gr == FILTER_GROUP_LLOG || oa->o_gr == FILTER_GROUP_ECHO)
223 RETURN(osc_real_create(exp, oa, ea, oti));
225 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
226 oa->o_flags == OBD_FL_RECREATE_OBJS) {
227 RETURN(osc_real_create(exp, oa, ea, oti));
232 rc = obd_alloc_memmd(exp, &lsm);
237 /* this is the special case where create removes orphans */
238 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
239 oa->o_flags == OBD_FL_DELORPHAN) {
240 CDEBUG(D_HA, "%p: oscc recovery started\n", oscc);
241 /* delete from next_id on up */
242 oa->o_valid |= OBD_MD_FLID;
243 oa->o_id = oscc->oscc_next_id - 1;
245 rc = osc_real_create(exp, oa, ea, NULL);
247 spin_lock(&oscc->oscc_lock);
249 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
250 oscc->oscc_flags &= ~OSCC_FLAG_RECOVERING;
251 oscc->oscc_last_id = oa->o_id;
252 wake_up(&oscc->oscc_waitq);
253 spin_unlock(&oscc->oscc_lock);
255 CDEBUG(D_HA, "%p: oscc recovery finished\n", oscc);
260 /* If orphans are being recovered, then we must wait until it is
261 finished before we can continue with create. */
262 if (oscc_recovering(oscc)) {
263 struct l_wait_info lwi;
265 CDEBUG(D_HA, "%p: oscc recovery in progress, waiting\n", oscc);
267 lwi = LWI_TIMEOUT(MAX(obd_timeout * HZ, 1), NULL, NULL);
268 rc = l_wait_event(oscc->oscc_waitq, !oscc_recovering(oscc),
270 LASSERT(rc == 0 || rc == -ETIMEDOUT);
271 if (rc == -ETIMEDOUT) {
272 CDEBUG(D_HA, "%p: timed out waiting for recovery\n", oscc);
275 CDEBUG(D_HA, "%p: oscc recovery over, waking up\n", oscc);
280 spin_lock(&oscc->oscc_lock);
281 if (oscc->oscc_last_id >= oscc->oscc_next_id) {
282 memcpy(oa, &oscc->oscc_oa, sizeof(*oa));
283 oa->o_id = oscc->oscc_next_id;
284 oa->o_gr = oscc->oscc_gr;
285 lsm->lsm_object_id = oscc->oscc_next_id;
286 lsm->lsm_object_gr = oscc->oscc_gr;
288 oscc->oscc_next_id++;
290 } else if (oscc->oscc_flags & OSCC_FLAG_NOSPC) {
292 spin_unlock(&oscc->oscc_lock);
295 spin_unlock(&oscc->oscc_lock);
296 rc = oscc_precreate(oscc, try_again);
302 CDEBUG(D_INFO, "returning objid "LPU64"/"LPU64"\n",
303 lsm->lsm_object_id, lsm->lsm_object_gr);
304 else if (*ea == NULL)
305 obd_free_memmd(exp, &lsm);
309 void oscc_init(struct obd_device *obd)
311 struct osc_creator *oscc;
316 oscc = &obd->u.cli.cl_oscc;
318 memset(oscc, 0, sizeof(*oscc));
319 INIT_LIST_HEAD(&oscc->oscc_list);
320 init_waitqueue_head(&oscc->oscc_waitq);
321 spin_lock_init(&oscc->oscc_lock);
322 oscc->oscc_obd = obd;
323 oscc->oscc_kick_barrier = 100;
324 oscc->oscc_grow_count = 2000;
325 oscc->oscc_initial_create_count = 2000;
327 oscc->oscc_next_id = 2;
328 oscc->oscc_last_id = 1;
329 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
330 /* XXX the export handle should give the oscc the last object */
331 /* oed->oed_oscc.oscc_last_id = exph->....; */