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/lustre_dlm.h>
40 # if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
41 # include <linux/workqueue.h>
42 # include <linux/smp_lock.h>
44 # include <linux/locks.h>
46 #else /* __KERNEL__ */
47 # include <liblustre.h>
51 # include <linux/ctype.h>
52 # include <linux/init.h>
57 #include <linux/obd_class.h>
58 #include "osc_internal.h"
60 static int osc_interpret_create(struct ptlrpc_request *req, void *data,
63 struct osc_creator *oscc;
64 struct ost_body *body = NULL;
68 body = lustre_swab_repbuf(req, 0, sizeof(*body),
69 lustre_swab_ost_body);
70 if (body == NULL && rc == 0)
74 oscc = req->rq_async_args.pointer_arg[0];
75 spin_lock(&oscc->oscc_lock);
77 oscc->oscc_last_id = body->oa.o_id;
79 DEBUG_REQ(D_INODE, req, "OST out of space, flagging");
80 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
81 } else if (rc != 0 && rc != -EIO) {
82 DEBUG_REQ(D_ERROR, req,
83 "unknown rc %d from async create: failing oscc",
85 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
86 ptlrpc_fail_import(req->rq_import, req->rq_import_generation);
88 oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
89 spin_unlock(&oscc->oscc_lock);
91 CDEBUG(D_INFO, "preallocated through id "LPU64" (last used "LPU64")\n",
92 oscc->oscc_last_id, oscc->oscc_next_id);
94 wake_up(&oscc->oscc_waitq);
98 static int oscc_internal_create(struct osc_creator *oscc)
100 struct ptlrpc_request *request;
101 struct ost_body *body;
102 int size = sizeof(*body);
105 spin_lock(&oscc->oscc_lock);
106 if (oscc->oscc_flags & OSCC_FLAG_CREATING) {
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_valid |= OBD_MD_FLID;
128 CDEBUG(D_INFO, "preallocating through id "LPU64" (last used "LPU64")\n",
129 body->oa.o_id, oscc->oscc_next_id);
130 spin_unlock(&oscc->oscc_lock);
132 request->rq_replen = lustre_msg_size(1, &size);
134 request->rq_async_args.pointer_arg[0] = oscc;
135 request->rq_interpret_reply = osc_interpret_create;
136 ptlrpcd_add_req(request);
141 static int oscc_has_objects(struct osc_creator *oscc, int count)
144 spin_lock(&oscc->oscc_lock);
145 have_objs = ((__s64)(oscc->oscc_last_id - oscc->oscc_next_id) >= count);
146 spin_unlock(&oscc->oscc_lock);
149 oscc_internal_create(oscc);
154 static int oscc_wait_for_objects(struct osc_creator *oscc, int count)
160 have_objs = oscc_has_objects(oscc, count);
162 spin_lock(&oscc->oscc_lock);
163 ost_full = (oscc->oscc_flags & OSCC_FLAG_NOSPC);
164 spin_unlock(&oscc->oscc_lock);
166 osc_invalid = oscc->oscc_obd->u.cli.cl_import->imp_invalid;
168 return have_objs || ost_full || osc_invalid;
171 static int oscc_precreate(struct osc_creator *oscc, int wait)
173 struct l_wait_info lwi = { 0 };
177 if (oscc_has_objects(oscc, oscc->oscc_kick_barrier))
183 /* no rc check -- a no-INTR, no-TIMEOUT wait can't fail */
184 l_wait_event(oscc->oscc_waitq, oscc_wait_for_objects(oscc, 1), &lwi);
186 if (!oscc_has_objects(oscc, 1) && (oscc->oscc_flags & OSCC_FLAG_NOSPC))
189 if (oscc->oscc_obd->u.cli.cl_import->imp_invalid)
195 int oscc_recovering(struct osc_creator *oscc)
199 spin_lock(&oscc->oscc_lock);
200 recov = oscc->oscc_flags & OSCC_FLAG_RECOVERING;
201 spin_unlock(&oscc->oscc_lock);
206 int osc_create(struct obd_export *exp, struct obdo *oa,
207 struct lov_stripe_md **ea, struct obd_trans_info *oti)
209 struct lov_stripe_md *lsm;
210 struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc;
211 int try_again = 1, rc = 0;
216 if ((oa->o_valid & OBD_MD_FLGROUP) && (oa->o_gr != 0))
217 RETURN(osc_real_create(exp, oa, ea, oti));
219 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
220 oa->o_flags == OBD_FL_RECREATE_OBJS) {
221 RETURN(osc_real_create(exp, oa, ea, oti));
226 rc = obd_alloc_memmd(exp, &lsm);
231 /* this is the special case where create removes orphans */
232 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
233 oa->o_flags == OBD_FL_DELORPHAN) {
234 CDEBUG(D_HA, "%p: oscc recovery started\n", oscc);
235 /* delete from next_id on up */
236 oa->o_valid |= OBD_MD_FLID;
237 oa->o_id = oscc->oscc_next_id - 1;
239 rc = osc_real_create(exp, oa, ea, NULL);
241 spin_lock(&oscc->oscc_lock);
243 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
244 oscc->oscc_flags &= ~OSCC_FLAG_RECOVERING;
245 oscc->oscc_last_id = oa->o_id;
246 wake_up(&oscc->oscc_waitq);
247 spin_unlock(&oscc->oscc_lock);
249 CDEBUG(D_HA, "%p: oscc recovery finished\n", oscc);
254 /* If orphans are being recovered, then we must wait until it is
255 finished before we can continue with create. */
256 if (oscc_recovering(oscc)) {
257 struct l_wait_info lwi;
259 CDEBUG(D_HA, "%p: oscc recovery in progress, waiting\n", oscc);
261 lwi = LWI_TIMEOUT(MAX(obd_timeout * HZ, 1), NULL, NULL);
262 rc = l_wait_event(oscc->oscc_waitq, !oscc_recovering(oscc),
264 LASSERT(rc == 0 || rc == -ETIMEDOUT);
265 if (rc == -ETIMEDOUT) {
266 CDEBUG(D_HA, "%p: timed out waiting for recovery\n", oscc);
269 CDEBUG(D_HA, "%p: oscc recovery over, waking up\n", oscc);
274 spin_lock(&oscc->oscc_lock);
275 if (oscc->oscc_last_id >= oscc->oscc_next_id) {
276 memcpy(oa, &oscc->oscc_oa, sizeof(*oa));
277 oa->o_id = oscc->oscc_next_id;
278 lsm->lsm_object_id = oscc->oscc_next_id;
280 oscc->oscc_next_id++;
282 } else if (oscc->oscc_flags & OSCC_FLAG_NOSPC) {
284 spin_unlock(&oscc->oscc_lock);
287 spin_unlock(&oscc->oscc_lock);
288 rc = oscc_precreate(oscc, try_again);
294 CDEBUG(D_INFO, "returning objid "LPU64"\n", lsm->lsm_object_id);
295 else if (*ea == NULL)
296 obd_free_memmd(exp, &lsm);
300 void oscc_init(struct obd_device *obd)
302 struct osc_creator *oscc;
307 oscc = &obd->u.cli.cl_oscc;
309 memset(oscc, 0, sizeof(*oscc));
310 INIT_LIST_HEAD(&oscc->oscc_list);
311 init_waitqueue_head(&oscc->oscc_waitq);
312 spin_lock_init(&oscc->oscc_lock);
313 oscc->oscc_obd = obd;
314 oscc->oscc_kick_barrier = 100;
315 oscc->oscc_grow_count = 2000;
316 oscc->oscc_initial_create_count = 2000;
318 oscc->oscc_next_id = 2;
319 oscc->oscc_last_id = 1;
320 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
321 /* XXX the export handle should give the oscc the last object */
322 /* oed->oed_oscc.oscc_last_id = exph->....; */