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_HA, "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 oscc->oscc_flags & OSCC_FLAG_RECOVERING) {
107 spin_unlock(&oscc->oscc_lock);
110 oscc->oscc_flags |= OSCC_FLAG_CREATING;
111 spin_unlock(&oscc->oscc_lock);
113 request = ptlrpc_prep_req(oscc->oscc_obd->u.cli.cl_import, OST_CREATE,
115 if (request == NULL) {
116 spin_lock(&oscc->oscc_lock);
117 oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
118 spin_unlock(&oscc->oscc_lock);
122 request->rq_request_portal = OST_CREATE_PORTAL; //XXX FIXME bug 249
123 body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof(*body));
125 spin_lock(&oscc->oscc_lock);
126 body->oa.o_id = oscc->oscc_last_id + oscc->oscc_grow_count;
127 body->oa.o_gr = oscc->oscc_gr;
128 LASSERT(body->oa.o_gr > 0);
129 body->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
130 CDEBUG(D_INFO, "preallocating through id "LPU64" (last used "LPU64")\n",
131 body->oa.o_id, oscc->oscc_next_id);
132 spin_unlock(&oscc->oscc_lock);
134 request->rq_replen = lustre_msg_size(1, &size);
136 request->rq_async_args.pointer_arg[0] = oscc;
137 request->rq_interpret_reply = osc_interpret_create;
138 ptlrpcd_add_req(request);
143 static int oscc_has_objects(struct osc_creator *oscc, int count)
146 spin_lock(&oscc->oscc_lock);
147 have_objs = ((__s64)(oscc->oscc_last_id - oscc->oscc_next_id) >= count);
148 spin_unlock(&oscc->oscc_lock);
151 oscc_internal_create(oscc);
156 static int oscc_wait_for_objects(struct osc_creator *oscc, int count)
162 have_objs = oscc_has_objects(oscc, count);
164 spin_lock(&oscc->oscc_lock);
165 ost_full = (oscc->oscc_flags & OSCC_FLAG_NOSPC);
166 spin_unlock(&oscc->oscc_lock);
168 osc_invalid = oscc->oscc_obd->u.cli.cl_import->imp_invalid;
170 return have_objs || ost_full || osc_invalid;
173 static int oscc_precreate(struct osc_creator *oscc, int wait)
175 struct l_wait_info lwi = { 0 };
179 if (oscc_has_objects(oscc, oscc->oscc_kick_barrier))
185 /* no rc check -- a no-INTR, no-TIMEOUT wait can't fail */
186 l_wait_event(oscc->oscc_waitq, oscc_wait_for_objects(oscc, 1), &lwi);
188 if (!oscc_has_objects(oscc, 1) && (oscc->oscc_flags & OSCC_FLAG_NOSPC))
191 if (oscc->oscc_obd->u.cli.cl_import->imp_invalid)
197 int oscc_recovering(struct osc_creator *oscc)
201 spin_lock(&oscc->oscc_lock);
202 recov = oscc->oscc_flags & OSCC_FLAG_RECOVERING;
203 spin_unlock(&oscc->oscc_lock);
208 int osc_create(struct obd_export *exp, struct obdo *oa,
209 struct lov_stripe_md **ea, struct obd_trans_info *oti)
211 struct lov_stripe_md *lsm;
212 struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc;
213 int try_again = 1, rc = 0;
217 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
218 LASSERT(oa->o_gr > 0);
220 LASSERT(oscc->oscc_gr == 0 || oscc->oscc_gr == oa->o_gr);
221 oscc->oscc_gr = oa->o_gr;
223 if (oa->o_gr == FILTER_GROUP_LLOG || oa->o_gr == FILTER_GROUP_ECHO)
224 RETURN(osc_real_create(exp, oa, ea, oti));
226 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
227 oa->o_flags == OBD_FL_RECREATE_OBJS) {
228 RETURN(osc_real_create(exp, oa, ea, oti));
233 rc = obd_alloc_memmd(exp, &lsm);
238 /* this is the special case where create removes orphans */
239 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
240 oa->o_flags == OBD_FL_DELORPHAN) {
241 CDEBUG(D_HA, "%p: oscc recovery started\n", oscc);
242 /* delete from next_id on up */
243 oa->o_valid |= OBD_MD_FLID;
244 oa->o_id = oscc->oscc_next_id - 1;
246 CDEBUG(D_HA, "%s: deleting to next_id: "LPU64"\n",
247 oscc->oscc_obd->u.cli.cl_import->imp_target_uuid.uuid,
250 rc = osc_real_create(exp, oa, ea, NULL);
252 spin_lock(&oscc->oscc_lock);
254 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
255 oscc->oscc_flags &= ~OSCC_FLAG_RECOVERING;
256 oscc->oscc_last_id = oa->o_id;
257 wake_up(&oscc->oscc_waitq);
258 spin_unlock(&oscc->oscc_lock);
260 /*recover happen in mds_setup, before cobd_setup, so
261 *reset oscc_gr = 0 here, it sould be no harm to CMD
264 CDEBUG(D_HA, "%p: oscc recovery finished\n", oscc);
270 /* If orphans are being recovered, then we must wait until
271 it is finished before we can continue with create. */
272 if (oscc_recovering(oscc)) {
273 struct l_wait_info lwi;
275 CDEBUG(D_HA,"%p: oscc recovery in progress, waiting\n",
278 lwi = LWI_TIMEOUT(MAX(obd_timeout * HZ, 1), NULL, NULL);
279 rc = l_wait_event(oscc->oscc_waitq,
280 !oscc_recovering(oscc), &lwi);
281 LASSERT(rc == 0 || rc == -ETIMEDOUT);
282 if (rc == -ETIMEDOUT) {
283 CDEBUG(D_HA, "%p: timed out waiting for "
287 CDEBUG(D_HA, "%p: oscc recovery over, waking up\n",
291 spin_lock(&oscc->oscc_lock);
292 if (oscc->oscc_last_id >= oscc->oscc_next_id) {
293 memcpy(oa, &oscc->oscc_oa, sizeof(*oa));
294 oa->o_id = oscc->oscc_next_id;
295 oa->o_gr = oscc->oscc_gr;
296 lsm->lsm_object_id = oscc->oscc_next_id;
297 lsm->lsm_object_gr = oscc->oscc_gr;
299 oscc->oscc_next_id++;
301 } else if (oscc->oscc_flags & OSCC_FLAG_NOSPC) {
303 spin_unlock(&oscc->oscc_lock);
306 spin_unlock(&oscc->oscc_lock);
307 rc = oscc_precreate(oscc, try_again);
313 CDEBUG(D_INFO, "returning objid "LPU64"/"LPU64"\n",
314 lsm->lsm_object_id, lsm->lsm_object_gr);
315 else if (*ea == NULL)
316 obd_free_memmd(exp, &lsm);
320 void oscc_init(struct obd_device *obd)
322 struct osc_creator *oscc;
327 oscc = &obd->u.cli.cl_oscc;
329 memset(oscc, 0, sizeof(*oscc));
330 INIT_LIST_HEAD(&oscc->oscc_list);
331 init_waitqueue_head(&oscc->oscc_waitq);
332 spin_lock_init(&oscc->oscc_lock);
333 oscc->oscc_obd = obd;
334 oscc->oscc_kick_barrier = 100;
335 oscc->oscc_grow_count = 2000;
336 oscc->oscc_initial_create_count = 2000;
338 oscc->oscc_next_id = 2;
339 oscc->oscc_last_id = 1;
340 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
341 /* XXX the export handle should give the oscc the last object */
342 /* oed->oed_oscc.oscc_last_id = exph->....; */