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);
75 oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
77 oscc->oscc_last_id = body->oa.o_id;
79 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
80 spin_unlock(&oscc->oscc_lock);
81 DEBUG_REQ(D_INODE, req, "OST out of space, flagging");
82 } else if (rc != 0 && rc != -EIO) {
83 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
84 spin_unlock(&oscc->oscc_lock);
85 DEBUG_REQ(D_ERROR, req,
86 "unknown rc %d from async create: failing oscc", rc);
87 ptlrpc_fail_import(req->rq_import, req->rq_import_generation);
89 spin_unlock(&oscc->oscc_lock);
92 CDEBUG(D_HA, "preallocated through id "LPU64" (last used "LPU64")\n",
93 oscc->oscc_last_id, oscc->oscc_next_id);
95 wake_up(&oscc->oscc_waitq);
99 static int oscc_internal_create(struct osc_creator *oscc)
101 struct ptlrpc_request *request;
102 struct ost_body *body;
103 int size = sizeof(*body);
106 spin_lock(&oscc->oscc_lock);
107 if (oscc->oscc_flags & OSCC_FLAG_CREATING ||
108 oscc->oscc_flags & OSCC_FLAG_RECOVERING) {
109 spin_unlock(&oscc->oscc_lock);
112 oscc->oscc_flags |= OSCC_FLAG_CREATING;
113 spin_unlock(&oscc->oscc_lock);
115 request = ptlrpc_prep_req(oscc->oscc_obd->u.cli.cl_import,
116 LUSTRE_OBD_VERSION, OST_CREATE,
118 if (request == NULL) {
119 spin_lock(&oscc->oscc_lock);
120 oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
121 spin_unlock(&oscc->oscc_lock);
125 request->rq_request_portal = OST_CREATE_PORTAL; //XXX FIXME bug 249
126 body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof(*body));
128 spin_lock(&oscc->oscc_lock);
129 body->oa.o_id = oscc->oscc_last_id + oscc->oscc_grow_count;
130 /* probably we should take frequence of request into account? -bzzz */
131 if (oscc->oscc_grow_count < oscc->oscc_max_grow_count) {
132 oscc->oscc_grow_count *= 2;
133 if (oscc->oscc_grow_count > oscc->oscc_max_grow_count)
134 oscc->oscc_grow_count = oscc->oscc_max_grow_count;
136 body->oa.o_gr = oscc->oscc_gr;
137 LASSERT(body->oa.o_gr > 0);
138 body->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
139 spin_unlock(&oscc->oscc_lock);
140 CDEBUG(D_INFO, "preallocating through id "LPU64" (last used "LPU64")\n",
141 body->oa.o_id, oscc->oscc_next_id);
143 request->rq_replen = lustre_msg_size(1, &size);
145 request->rq_async_args.pointer_arg[0] = oscc;
146 request->rq_interpret_reply = osc_interpret_create;
147 ptlrpcd_add_req(request);
152 static int oscc_has_objects(struct osc_creator *oscc, int count)
155 spin_lock(&oscc->oscc_lock);
156 have_objs = ((__s64)(oscc->oscc_last_id - oscc->oscc_next_id) >= count);
157 spin_unlock(&oscc->oscc_lock);
160 oscc_internal_create(oscc);
165 static int oscc_wait_for_objects(struct osc_creator *oscc, int count)
171 have_objs = oscc_has_objects(oscc, count);
173 spin_lock(&oscc->oscc_lock);
174 ost_full = (oscc->oscc_flags & OSCC_FLAG_NOSPC);
175 spin_unlock(&oscc->oscc_lock);
177 osc_invalid = oscc->oscc_obd->u.cli.cl_import->imp_invalid;
179 return have_objs || ost_full || osc_invalid;
182 static int oscc_precreate(struct osc_creator *oscc, int wait)
184 struct l_wait_info lwi = { 0 };
188 if (oscc_has_objects(oscc, oscc->oscc_kick_barrier))
194 /* no rc check -- a no-INTR, no-TIMEOUT wait can't fail */
195 l_wait_event(oscc->oscc_waitq, oscc_wait_for_objects(oscc, 1), &lwi);
197 if (!oscc_has_objects(oscc, 1) && (oscc->oscc_flags & OSCC_FLAG_NOSPC))
200 if (oscc->oscc_obd->u.cli.cl_import->imp_invalid)
206 int oscc_recovering(struct osc_creator *oscc)
210 spin_lock(&oscc->oscc_lock);
211 recov = oscc->oscc_flags & OSCC_FLAG_RECOVERING;
212 spin_unlock(&oscc->oscc_lock);
217 int osc_create(struct obd_export *exp, struct obdo *oa,
218 struct lov_stripe_md **ea, struct obd_trans_info *oti)
220 struct lov_stripe_md *lsm;
221 struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc;
222 int try_again = 1, rc = 0;
226 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
227 LASSERT(oa->o_gr > 0);
229 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
230 oa->o_flags == OBD_FL_RECREATE_OBJS) {
231 /* Exceptional case where we are trying to repair missing
232 * objects for various groups. We have already validated that
233 * this is a valid group for the file. Don't set oscc->oscc_gr.
235 RETURN(osc_real_create(exp, oa, ea, oti));
238 LASSERT(oscc->oscc_gr == 0 || oscc->oscc_gr == oa->o_gr);
239 oscc->oscc_gr = oa->o_gr;
241 if (oa->o_gr == FILTER_GROUP_LLOG || oa->o_gr == FILTER_GROUP_ECHO)
242 RETURN(osc_real_create(exp, oa, ea, oti));
244 /* this is the special case where create removes orphans */
245 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
246 oa->o_flags == OBD_FL_DELORPHAN) {
247 spin_lock(&oscc->oscc_lock);
248 if (oscc->oscc_flags & OSCC_FLAG_SYNC_IN_PROGRESS) {
249 spin_unlock(&oscc->oscc_lock);
252 if (!(oscc->oscc_flags & OSCC_FLAG_RECOVERING)) {
253 spin_unlock(&oscc->oscc_lock);
256 oscc->oscc_flags |= OSCC_FLAG_SYNC_IN_PROGRESS;
257 spin_unlock(&oscc->oscc_lock);
258 CDEBUG(D_HA, "%s; oscc recovery started\n",
259 exp->exp_obd->obd_name);
260 LASSERT(oscc->oscc_flags & OSCC_FLAG_RECOVERING);
262 /* delete from next_id on up */
263 oa->o_valid |= OBD_MD_FLID;
264 oa->o_id = oscc->oscc_next_id - 1;
266 CDEBUG(D_HA, "%s: deleting to next_id: "LPU64"\n",
267 exp->exp_obd->obd_name, oa->o_id);
269 rc = osc_real_create(exp, oa, ea, NULL);
270 if (oscc->oscc_obd == NULL) {
271 CWARN("the obd for oscc %p has been freed\n", oscc);
275 spin_lock(&oscc->oscc_lock);
276 oscc->oscc_flags &= ~OSCC_FLAG_SYNC_IN_PROGRESS;
277 if (rc == 0 || rc == -ENOSPC) {
279 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
280 oscc->oscc_flags &= ~OSCC_FLAG_RECOVERING;
281 oscc->oscc_last_id = oa->o_id;
283 CDEBUG(D_HA, "%s: oscc recovery finished: %d\n",
284 exp->exp_obd->obd_name, rc);
285 wake_up(&oscc->oscc_waitq);
288 CDEBUG(D_ERROR, "%s: oscc recovery failed: %d\n",
289 exp->exp_obd->obd_name, rc);
291 spin_unlock(&oscc->oscc_lock);
298 rc = obd_alloc_memmd(exp, &lsm);
304 /* If orphans are being recovered, then we must wait until
305 it is finished before we can continue with create. */
306 if (oscc_recovering(oscc)) {
307 struct l_wait_info lwi;
309 CDEBUG(D_HA,"%p: oscc recovery in progress, waiting\n",
311 lwi = LWI_TIMEOUT(MAX(obd_timeout*HZ/4, 1), NULL, NULL);
312 rc = l_wait_event(oscc->oscc_waitq,
313 !oscc_recovering(oscc), &lwi);
315 LASSERT(rc == 0 || rc == -ETIMEDOUT);
316 if (rc == -ETIMEDOUT) {
317 CDEBUG(D_HA,"%p: timeout waiting on recovery\n",
321 CDEBUG(D_HA, "%s: oscc recovery over, waking up\n",
322 exp->exp_obd->obd_name);
325 spin_lock(&oscc->oscc_lock);
326 if (oscc->oscc_last_id >= oscc->oscc_next_id) {
327 memcpy(oa, &oscc->oscc_oa, sizeof(*oa));
328 oa->o_id = oscc->oscc_next_id;
329 oa->o_gr = oscc->oscc_gr;
330 lsm->lsm_object_id = oscc->oscc_next_id;
331 lsm->lsm_object_gr = oscc->oscc_gr;
333 oscc->oscc_next_id++;
335 } else if (oscc->oscc_flags & OSCC_FLAG_NOSPC) {
337 spin_unlock(&oscc->oscc_lock);
340 spin_unlock(&oscc->oscc_lock);
341 rc = oscc_precreate(oscc, try_again);
347 CDEBUG(D_HA, "%s: returning objid "LPU64"\n",
348 oscc->oscc_obd->u.cli.cl_import->imp_target_uuid.uuid,
350 else if (*ea == NULL)
351 obd_free_memmd(exp, &lsm);
355 void oscc_init(struct obd_device *obd)
357 struct osc_creator *oscc;
362 oscc = &obd->u.cli.cl_oscc;
364 memset(oscc, 0, sizeof(*oscc));
365 INIT_LIST_HEAD(&oscc->oscc_list);
366 init_waitqueue_head(&oscc->oscc_waitq);
367 spin_lock_init(&oscc->oscc_lock);
368 oscc->oscc_obd = obd;
369 oscc->oscc_kick_barrier = 100;
370 oscc->oscc_grow_count = 36;
371 oscc->oscc_max_grow_count = 2000;
373 oscc->oscc_next_id = 2;
374 oscc->oscc_last_id = 1;
375 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
376 /* XXX the export handle should give the oscc the last object */
377 /* oed->oed_oscc.oscc_last_id = exph->....; */