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 the Lustre file system, http://www.lustre.org
8 * Lustre is a trademark of Cluster File Systems, Inc.
10 * You may have signed or agreed to another license before downloading
11 * this software. If so, you are bound by the terms and conditions
12 * of that agreement, and the following does not apply to you. See the
13 * LICENSE file included with this distribution for more information.
15 * If you did not agree to a different license, then this copy of Lustre
16 * is open source software; you can redistribute it and/or modify it
17 * under the terms of version 2 of the GNU General Public License as
18 * published by the Free Software Foundation.
20 * In either case, Lustre is distributed in the hope that it will be
21 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
22 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
23 * license text for more details.
25 * For testing and management it is treated as an obd_device,
26 * although * it does not export a full OBD method table (the
27 * requests are coming * in over the wire, so object target modules
28 * do not have a full * method table.)
33 # define EXPORT_SYMTAB
35 #define DEBUG_SUBSYSTEM S_OSC
38 # include <libcfs/libcfs.h>
39 #else /* __KERNEL__ */
40 # include <liblustre.h>
47 # include <lustre_dlm.h>
48 #include <obd_class.h>
49 #include "osc_internal.h"
51 static int osc_interpret_create(struct ptlrpc_request *req, void *data, int rc)
53 struct osc_creator *oscc;
54 struct ost_body *body = NULL;
58 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
59 lustre_swab_ost_body);
60 if (body == NULL && rc == 0)
64 oscc = req->rq_async_args.pointer_arg[0];
65 LASSERT(oscc && (oscc->oscc_obd != LP_POISON));
67 spin_lock(&oscc->oscc_lock);
68 oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
72 int diff = body->oa.o_id - oscc->oscc_last_id;
74 if (diff < oscc->oscc_grow_count)
75 oscc->oscc_grow_count =
76 max(diff/3, OST_MIN_PRECREATE);
78 oscc->oscc_flags &= ~OSCC_FLAG_LOW;
79 oscc->oscc_last_id = body->oa.o_id;
81 spin_unlock(&oscc->oscc_lock);
85 /* valid race delorphan vs create, or somthing after resend */
86 spin_unlock(&oscc->oscc_lock);
87 DEBUG_REQ(D_INODE, req, "Got EGAIN - resend \n");
91 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
92 if (body && rc == -ENOSPC) {
93 oscc->oscc_grow_count = OST_MIN_PRECREATE;
94 oscc->oscc_last_id = body->oa.o_id;
96 spin_unlock(&oscc->oscc_lock);
97 DEBUG_REQ(D_INODE, req, "OST out of space, flagging");
101 /* filter always set body->oa.o_id as the last_id
102 * of filter (see filter_handle_precreate for detail)*/
103 if (body && body->oa.o_id > oscc->oscc_last_id)
104 oscc->oscc_last_id = body->oa.o_id;
105 spin_unlock(&oscc->oscc_lock);
109 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
110 oscc->oscc_grow_count = OST_MIN_PRECREATE;
111 spin_unlock(&oscc->oscc_lock);
112 DEBUG_REQ(D_ERROR, req,
113 "Unknown rc %d from async create: failing oscc", rc);
114 ptlrpc_fail_import(req->rq_import,
115 lustre_msg_get_conn_cnt(req->rq_reqmsg));
119 CDEBUG(D_HA, "preallocated through id "LPU64" (next to use "LPU64")\n",
120 oscc->oscc_last_id, oscc->oscc_next_id);
122 cfs_waitq_signal(&oscc->oscc_waitq);
126 static int oscc_internal_create(struct osc_creator *oscc)
128 struct ptlrpc_request *request;
129 struct ost_body *body;
130 int size[] = { sizeof(struct ptlrpc_body), sizeof(*body) };
133 LASSERT_SPIN_LOCKED(&oscc->oscc_lock);
135 if (oscc->oscc_grow_count < OST_MAX_PRECREATE &&
136 !(oscc->oscc_flags & (OSCC_FLAG_LOW | OSCC_FLAG_RECOVERING)) &&
137 (__s64)(oscc->oscc_last_id - oscc->oscc_next_id) <=
138 (oscc->oscc_grow_count / 4 + 1)) {
139 oscc->oscc_flags |= OSCC_FLAG_LOW;
140 oscc->oscc_grow_count *= 2;
143 if (oscc->oscc_grow_count > OST_MAX_PRECREATE / 2)
144 oscc->oscc_grow_count = OST_MAX_PRECREATE / 2;
146 if (oscc->oscc_flags & OSCC_FLAG_CREATING ||
147 oscc->oscc_flags & OSCC_FLAG_RECOVERING) {
148 spin_unlock(&oscc->oscc_lock);
151 oscc->oscc_flags |= OSCC_FLAG_CREATING;
152 spin_unlock(&oscc->oscc_lock);
154 request = ptlrpc_prep_req(oscc->oscc_obd->u.cli.cl_import,
155 LUSTRE_OST_VERSION, OST_CREATE, 2,
157 if (request == NULL) {
158 spin_lock(&oscc->oscc_lock);
159 oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
160 spin_unlock(&oscc->oscc_lock);
164 request->rq_request_portal = OST_CREATE_PORTAL; //XXX FIXME bug 249
165 body = lustre_msg_buf(request->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
167 spin_lock(&oscc->oscc_lock);
168 body->oa.o_id = oscc->oscc_last_id + oscc->oscc_grow_count;
169 body->oa.o_gr = oscc->oscc_oa.o_gr;
170 LASSERT(body->oa.o_gr > 0);
171 body->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
172 spin_unlock(&oscc->oscc_lock);
173 CDEBUG(D_RPCTRACE, "prealloc through id "LPU64" (last seen "LPU64")\n",
174 body->oa.o_id, oscc->oscc_last_id);
176 ptlrpc_req_set_repsize(request, 2, size);
178 request->rq_async_args.pointer_arg[0] = oscc;
179 request->rq_interpret_reply = osc_interpret_create;
180 ptlrpcd_add_req(request);
185 static int oscc_has_objects(struct osc_creator *oscc, int count)
188 spin_lock(&oscc->oscc_lock);
189 have_objs = ((__s64)(oscc->oscc_last_id - oscc->oscc_next_id) >= count);
192 oscc_internal_create(oscc);
194 spin_unlock(&oscc->oscc_lock);
200 static int oscc_wait_for_objects(struct osc_creator *oscc, int count)
206 have_objs = oscc_has_objects(oscc, count);
208 spin_lock(&oscc->oscc_lock);
209 ost_full = (oscc->oscc_flags & OSCC_FLAG_NOSPC);
210 spin_unlock(&oscc->oscc_lock);
212 osc_invalid = oscc->oscc_obd->u.cli.cl_import->imp_invalid;
214 return have_objs || ost_full || osc_invalid;
217 static int oscc_precreate(struct osc_creator *oscc, int wait)
219 struct l_wait_info lwi = { 0 };
223 if (oscc_has_objects(oscc, oscc->oscc_grow_count / 2))
229 /* no rc check -- a no-INTR, no-TIMEOUT wait can't fail */
230 l_wait_event(oscc->oscc_waitq, oscc_wait_for_objects(oscc, 1), &lwi);
232 if (!oscc_has_objects(oscc, 1) && (oscc->oscc_flags & OSCC_FLAG_NOSPC))
235 if (oscc->oscc_obd->u.cli.cl_import->imp_invalid)
241 int oscc_recovering(struct osc_creator *oscc)
245 spin_lock(&oscc->oscc_lock);
246 recov = oscc->oscc_flags & OSCC_FLAG_RECOVERING;
247 spin_unlock(&oscc->oscc_lock);
252 /* decide if the OST has remaining object, return value :
253 0 : the OST has remaining object, and don't need to do precreate.
254 1 : the OST has no remaining object, and will send a RPC for precreate.
255 2 : the OST has no remaining object, and will not get any for
256 a potentially very long time
259 int osc_precreate(struct obd_export *exp)
261 struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc;
262 struct obd_import *imp = exp->exp_imp_reverse;
265 LASSERT(oscc != NULL);
266 if (imp != NULL && imp->imp_deactive)
269 if (oscc->oscc_last_id < oscc->oscc_next_id) {
270 spin_lock(&oscc->oscc_lock);
271 if (oscc->oscc_flags & OSCC_FLAG_NOSPC) {
272 spin_unlock(&oscc->oscc_lock);
275 if (oscc->oscc_flags & OSCC_FLAG_SYNC_IN_PROGRESS) {
276 spin_unlock(&oscc->oscc_lock);
279 if (oscc->oscc_flags & OSCC_FLAG_RECOVERING) {
280 spin_unlock(&oscc->oscc_lock);
284 if (oscc->oscc_flags & OSCC_FLAG_CREATING) {
285 spin_unlock(&oscc->oscc_lock);
289 oscc_internal_create(oscc);
295 int osc_create(struct obd_export *exp, struct obdo *oa,
296 struct lov_stripe_md **ea, struct obd_trans_info *oti)
298 struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc;
299 struct lov_stripe_md *lsm;
300 int try_again = 1, rc = 0;
305 LASSERT(oa->o_gr > 0);
306 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
308 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
309 oa->o_flags == OBD_FL_RECREATE_OBJS) {
310 RETURN(osc_real_create(exp, oa, ea, oti));
313 if (oa->o_gr == FILTER_GROUP_LLOG || oa->o_gr == FILTER_GROUP_ECHO)
314 RETURN(osc_real_create(exp, oa, ea, oti));
316 /* this is the special case where create removes orphans */
317 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
318 oa->o_flags == OBD_FL_DELORPHAN) {
319 spin_lock(&oscc->oscc_lock);
320 if (oscc->oscc_flags & OSCC_FLAG_SYNC_IN_PROGRESS) {
321 spin_unlock(&oscc->oscc_lock);
324 if (!(oscc->oscc_flags & OSCC_FLAG_RECOVERING)) {
325 spin_unlock(&oscc->oscc_lock);
328 oscc->oscc_flags |= OSCC_FLAG_SYNC_IN_PROGRESS;
329 spin_unlock(&oscc->oscc_lock);
330 CDEBUG(D_HA, "%s: oscc recovery started - delete to "LPU64"\n",
331 oscc->oscc_obd->obd_name, oscc->oscc_next_id - 1);
333 /* delete from next_id on up */
334 oa->o_valid |= OBD_MD_FLID;
335 oa->o_id = oscc->oscc_next_id - 1;
337 rc = osc_real_create(exp, oa, ea, NULL);
339 spin_lock(&oscc->oscc_lock);
340 oscc->oscc_flags &= ~OSCC_FLAG_SYNC_IN_PROGRESS;
341 if (rc == 0 || rc == -ENOSPC) {
343 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
344 oscc->oscc_flags &= ~OSCC_FLAG_RECOVERING;
345 oscc->oscc_last_id = oa->o_id;
346 CDEBUG(D_HA, "%s: oscc recovery finished, last_id: "
347 LPU64", rc: %d\n", oscc->oscc_obd->obd_name,
348 oscc->oscc_last_id, rc);
349 cfs_waitq_signal(&oscc->oscc_waitq);
351 CDEBUG(D_ERROR, "%s: oscc recovery failed: %d\n",
352 oscc->oscc_obd->obd_name, rc);
354 spin_unlock(&oscc->oscc_lock);
362 rc = obd_alloc_memmd(exp, &lsm);
368 /* If orphans are being recovered, then we must wait until
369 it is finished before we can continue with create. */
370 if (oscc_recovering(oscc)) {
371 struct l_wait_info lwi;
373 CDEBUG(D_HA,"%s: oscc recovery in progress, waiting\n",
374 oscc->oscc_obd->obd_name);
376 lwi = LWI_TIMEOUT(cfs_timeout_cap(cfs_time_seconds(obd_timeout/4)),
378 rc = l_wait_event(oscc->oscc_waitq,
379 !oscc_recovering(oscc), &lwi);
380 LASSERT(rc == 0 || rc == -ETIMEDOUT);
381 if (rc == -ETIMEDOUT) {
382 CDEBUG(D_HA,"%s: timeout waiting on recovery\n",
383 oscc->oscc_obd->obd_name);
386 CDEBUG(D_HA, "%s: oscc recovery over, waking up\n",
387 oscc->oscc_obd->obd_name);
390 spin_lock(&oscc->oscc_lock);
391 if (oscc->oscc_flags & OSCC_FLAG_EXITING) {
392 spin_unlock(&oscc->oscc_lock);
396 if (oscc->oscc_last_id >= oscc->oscc_next_id) {
397 memcpy(oa, &oscc->oscc_oa, sizeof(*oa));
398 oa->o_id = oscc->oscc_next_id;
399 lsm->lsm_object_id = oscc->oscc_next_id;
401 oscc->oscc_next_id++;
404 CDEBUG(D_RPCTRACE, "%s: set oscc_next_id = "LPU64"\n",
405 exp->exp_obd->obd_name, oscc->oscc_next_id);
406 } else if (oscc->oscc_flags & OSCC_FLAG_NOSPC) {
408 spin_unlock(&oscc->oscc_lock);
411 spin_unlock(&oscc->oscc_lock);
412 rc = oscc_precreate(oscc, try_again);
418 CDEBUG(D_INFO, "%s: returning objid "LPU64"\n",
419 obd2cli_tgt(oscc->oscc_obd), lsm->lsm_object_id);
420 else if (*ea == NULL)
421 obd_free_memmd(exp, &lsm);
425 void oscc_init(struct obd_device *obd)
427 struct osc_creator *oscc;
432 oscc = &obd->u.cli.cl_oscc;
434 memset(oscc, 0, sizeof(*oscc));
435 CFS_INIT_LIST_HEAD(&oscc->oscc_list);
436 cfs_waitq_init(&oscc->oscc_waitq);
437 spin_lock_init(&oscc->oscc_lock);
438 oscc->oscc_obd = obd;
439 oscc->oscc_grow_count = OST_MIN_PRECREATE;
441 oscc->oscc_next_id = 2;
442 oscc->oscc_last_id = 1;
443 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
444 /* XXX the export handle should give the oscc the last object */
445 /* oed->oed_oscc.oscc_last_id = exph->....; */