1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osc/osc_create.c
37 * For testing and management it is treated as an obd_device,
38 * although * it does not export a full OBD method table (the
39 * requests are coming * in over the wire, so object target modules
40 * do not have a full * method table.)
42 * Author: Peter Braam <braam@clusterfs.com>
46 # define EXPORT_SYMTAB
48 #define DEBUG_SUBSYSTEM S_OSC
51 # include <libcfs/libcfs.h>
52 #else /* __KERNEL__ */
53 # include <liblustre.h>
60 # include <lustre_dlm.h>
61 #include <obd_class.h>
62 #include "osc_internal.h"
64 static int osc_interpret_create(const struct lu_env *env,
65 struct ptlrpc_request *req, void *data, int rc)
67 struct osc_creator *oscc;
68 struct ost_body *body = NULL;
72 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
73 lustre_swab_ost_body);
74 if (body == NULL && rc == 0)
78 oscc = req->rq_async_args.pointer_arg[0];
79 LASSERT(oscc && (oscc->oscc_obd != LP_POISON));
81 spin_lock(&oscc->oscc_lock);
82 oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
86 int diff = body->oa.o_id - oscc->oscc_last_id;
88 if (diff < oscc->oscc_grow_count)
89 oscc->oscc_grow_count =
90 max(diff/3, OST_MIN_PRECREATE);
92 oscc->oscc_flags &= ~OSCC_FLAG_LOW;
93 oscc->oscc_last_id = body->oa.o_id;
95 spin_unlock(&oscc->oscc_lock);
99 /* valid race delorphan vs create, or somthing after resend */
100 spin_unlock(&oscc->oscc_lock);
101 DEBUG_REQ(D_INODE, req, "Got EAGAIN - resend \n");
106 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
107 if (body && rc == -ENOSPC) {
108 oscc->oscc_grow_count = OST_MIN_PRECREATE;
109 oscc->oscc_last_id = body->oa.o_id;
111 spin_unlock(&oscc->oscc_lock);
112 DEBUG_REQ(D_INODE, req, "OST out of space, flagging");
116 /* filter always set body->oa.o_id as the last_id
117 * of filter (see filter_handle_precreate for detail)*/
118 if (body && body->oa.o_id > oscc->oscc_last_id)
119 oscc->oscc_last_id = body->oa.o_id;
120 spin_unlock(&oscc->oscc_lock);
124 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
125 oscc->oscc_grow_count = OST_MIN_PRECREATE;
126 spin_unlock(&oscc->oscc_lock);
127 DEBUG_REQ(D_ERROR, req,
128 "Unknown rc %d from async create: failing oscc", rc);
129 ptlrpc_fail_import(req->rq_import,
130 lustre_msg_get_conn_cnt(req->rq_reqmsg));
134 CDEBUG(D_HA, "preallocated through id "LPU64" (next to use "LPU64")\n",
135 oscc->oscc_last_id, oscc->oscc_next_id);
137 cfs_waitq_signal(&oscc->oscc_waitq);
141 static int oscc_internal_create(struct osc_creator *oscc)
143 struct ptlrpc_request *request;
144 struct ost_body *body;
145 __u32 size[] = { sizeof(struct ptlrpc_body), sizeof(*body) };
148 LASSERT_SPIN_LOCKED(&oscc->oscc_lock);
150 if (oscc->oscc_flags & OSCC_FLAG_CREATING ||
151 oscc->oscc_flags & OSCC_FLAG_RECOVERING) {
152 spin_unlock(&oscc->oscc_lock);
156 if (oscc->oscc_grow_count < oscc->oscc_max_grow_count &&
157 ((oscc->oscc_flags & OSCC_FLAG_LOW) == 0) &&
158 (__s64)(oscc->oscc_last_id - oscc->oscc_next_id) <=
159 (oscc->oscc_grow_count / 4 + 1)) {
160 oscc->oscc_flags |= OSCC_FLAG_LOW;
161 oscc->oscc_grow_count *= 2;
164 if (oscc->oscc_grow_count > oscc->oscc_max_grow_count / 2)
165 oscc->oscc_grow_count = oscc->oscc_max_grow_count / 2;
167 oscc->oscc_flags |= OSCC_FLAG_CREATING;
168 spin_unlock(&oscc->oscc_lock);
170 request = ptlrpc_prep_req(oscc->oscc_obd->u.cli.cl_import,
171 LUSTRE_OST_VERSION, OST_CREATE, 2,
173 if (request == NULL) {
174 spin_lock(&oscc->oscc_lock);
175 oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
176 spin_unlock(&oscc->oscc_lock);
180 request->rq_request_portal = OST_CREATE_PORTAL;
181 ptlrpc_at_set_req_timeout(request);
182 body = lustre_msg_buf(request->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
184 spin_lock(&oscc->oscc_lock);
185 body->oa.o_id = oscc->oscc_last_id + oscc->oscc_grow_count;
186 body->oa.o_gr = oscc->oscc_oa.o_gr;
187 LASSERT_MDS_GROUP(body->oa.o_gr);
188 body->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
189 spin_unlock(&oscc->oscc_lock);
190 CDEBUG(D_RPCTRACE, "prealloc through id "LPU64" (last seen "LPU64")\n",
191 body->oa.o_id, oscc->oscc_last_id);
193 ptlrpc_req_set_repsize(request, 2, size);
195 request->rq_async_args.pointer_arg[0] = oscc;
196 request->rq_interpret_reply = osc_interpret_create;
197 ptlrpcd_add_req(request, PSCOPE_OTHER);
202 static int oscc_has_objects(struct osc_creator *oscc, int count)
205 spin_lock(&oscc->oscc_lock);
206 have_objs = ((__s64)(oscc->oscc_last_id - oscc->oscc_next_id) >= count);
209 oscc_internal_create(oscc);
211 spin_unlock(&oscc->oscc_lock);
217 static int oscc_wait_for_objects(struct osc_creator *oscc, int count)
223 have_objs = oscc_has_objects(oscc, count);
225 spin_lock(&oscc->oscc_lock);
226 ost_full = (oscc->oscc_flags & OSCC_FLAG_NOSPC);
227 spin_unlock(&oscc->oscc_lock);
229 osc_invalid = oscc->oscc_obd->u.cli.cl_import->imp_invalid;
231 return have_objs || ost_full || osc_invalid;
234 static int oscc_precreate(struct osc_creator *oscc, int wait)
236 struct l_wait_info lwi = { 0 };
240 if (oscc_has_objects(oscc, oscc->oscc_grow_count / 2))
246 /* no rc check -- a no-INTR, no-TIMEOUT wait can't fail */
247 l_wait_event(oscc->oscc_waitq, oscc_wait_for_objects(oscc, 1), &lwi);
249 if (!oscc_has_objects(oscc, 1) && (oscc->oscc_flags & OSCC_FLAG_NOSPC))
252 if (oscc->oscc_obd->u.cli.cl_import->imp_invalid)
258 int oscc_recovering(struct osc_creator *oscc)
262 spin_lock(&oscc->oscc_lock);
263 recov = oscc->oscc_flags & OSCC_FLAG_RECOVERING;
264 spin_unlock(&oscc->oscc_lock);
269 /* decide if the OST has remaining object, return value :
270 0 : the OST has remaining object, and don't need to do precreate.
271 1 : the OST has no remaining object, and will send a RPC for precreate.
272 2 : the OST has no remaining object, and will not get any for
273 a potentially very long time
276 int osc_precreate(struct obd_export *exp)
278 struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc;
279 struct obd_import *imp = exp->exp_imp_reverse;
282 LASSERT(oscc != NULL);
283 if (imp != NULL && imp->imp_deactive)
286 if (oscc_recovering(oscc))
289 if (oscc->oscc_flags & OSCC_FLAG_NOSPC)
292 if (oscc->oscc_last_id < oscc->oscc_next_id) {
293 if (oscc->oscc_flags & OSCC_FLAG_SYNC_IN_PROGRESS)
296 spin_lock(&oscc->oscc_lock);
297 if (oscc->oscc_flags & OSCC_FLAG_CREATING) {
298 spin_unlock(&oscc->oscc_lock);
302 oscc_internal_create(oscc);
308 int osc_create(struct obd_export *exp, struct obdo *oa,
309 struct lov_stripe_md **ea, struct obd_trans_info *oti)
311 struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc;
312 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
313 struct lov_stripe_md *lsm;
314 int try_again = 1, rc = 0;
319 LASSERT_MDS_GROUP(oa->o_gr);
320 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
322 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
323 oa->o_flags == OBD_FL_RECREATE_OBJS) {
324 RETURN(osc_real_create(exp, oa, ea, oti));
327 if (oa->o_gr == FILTER_GROUP_LLOG || oa->o_gr == FILTER_GROUP_ECHO)
328 RETURN(osc_real_create(exp, oa, ea, oti));
330 /* this is the special case where create removes orphans */
331 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
332 oa->o_flags == OBD_FL_DELORPHAN) {
333 spin_lock(&oscc->oscc_lock);
334 if (oscc->oscc_flags & OSCC_FLAG_SYNC_IN_PROGRESS) {
335 spin_unlock(&oscc->oscc_lock);
338 if (!(oscc->oscc_flags & OSCC_FLAG_RECOVERING)) {
339 spin_unlock(&oscc->oscc_lock);
342 oscc->oscc_flags |= OSCC_FLAG_SYNC_IN_PROGRESS;
343 /* seting flag LOW we prevent extra grow precreate size
344 * and enforce use last assigned size */
345 oscc->oscc_flags |= OSCC_FLAG_LOW;
346 spin_unlock(&oscc->oscc_lock);
347 CDEBUG(D_HA, "%s: oscc recovery started - delete to "LPU64"\n",
348 oscc->oscc_obd->obd_name, oscc->oscc_next_id - 1);
350 /* delete from next_id on up */
351 oa->o_valid |= OBD_MD_FLID;
352 oa->o_id = oscc->oscc_next_id - 1;
354 rc = osc_real_create(exp, oa, ea, NULL);
356 spin_lock(&oscc->oscc_lock);
357 oscc->oscc_flags &= ~OSCC_FLAG_SYNC_IN_PROGRESS;
358 if (rc == 0 || rc == -ENOSPC) {
359 struct obd_connect_data *ocd;
362 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
363 oscc->oscc_flags &= ~OSCC_FLAG_RECOVERING;
365 oscc->oscc_last_id = oa->o_id;
366 ocd = &imp->imp_connect_data;
367 if (ocd->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN) {
368 CWARN("Skip orphan set, reset the last objid\n");
369 oscc->oscc_next_id = oa->o_id + 1;
372 /* sanity check for next objid. see bug 17025 */
373 LASSERT(oscc->oscc_next_id == oa->o_id + 1);
375 CDEBUG(D_HA, "%s: oscc recovery finished, last_id: "
376 LPU64", rc: %d\n", oscc->oscc_obd->obd_name,
377 oscc->oscc_last_id, rc);
378 cfs_waitq_signal(&oscc->oscc_waitq);
380 CDEBUG(D_ERROR, "%s: oscc recovery failed: %d\n",
381 oscc->oscc_obd->obd_name, rc);
383 spin_unlock(&oscc->oscc_lock);
391 rc = obd_alloc_memmd(exp, &lsm);
397 /* If orphans are being recovered, then we must wait until
398 it is finished before we can continue with create. */
399 if (oscc_recovering(oscc)) {
400 struct l_wait_info lwi;
402 CDEBUG(D_HA,"%s: oscc recovery in progress, waiting\n",
403 oscc->oscc_obd->obd_name);
405 lwi = LWI_TIMEOUT(cfs_timeout_cap(cfs_time_seconds(
406 obd_timeout / 4)), NULL, NULL);
407 rc = l_wait_event(oscc->oscc_waitq,
408 !oscc_recovering(oscc), &lwi);
409 LASSERT(rc == 0 || rc == -ETIMEDOUT);
410 if (rc == -ETIMEDOUT) {
411 CDEBUG(D_HA,"%s: timeout waiting on recovery\n",
412 oscc->oscc_obd->obd_name);
415 CDEBUG(D_HA, "%s: oscc recovery over, waking up\n",
416 oscc->oscc_obd->obd_name);
419 spin_lock(&oscc->oscc_lock);
420 if (oscc->oscc_flags & OSCC_FLAG_EXITING) {
421 spin_unlock(&oscc->oscc_lock);
425 if (oscc->oscc_last_id >= oscc->oscc_next_id) {
426 memcpy(oa, &oscc->oscc_oa, sizeof(*oa));
427 oa->o_id = oscc->oscc_next_id;
428 lsm->lsm_object_id = oscc->oscc_next_id;
430 oscc->oscc_next_id++;
433 CDEBUG(D_RPCTRACE, "%s: set oscc_next_id = "LPU64"\n",
434 exp->exp_obd->obd_name, oscc->oscc_next_id);
435 } else if (oscc->oscc_flags & OSCC_FLAG_NOSPC) {
437 spin_unlock(&oscc->oscc_lock);
440 spin_unlock(&oscc->oscc_lock);
441 rc = oscc_precreate(oscc, try_again);
447 CDEBUG(D_INFO, "%s: returning objid "LPU64"\n",
448 obd2cli_tgt(oscc->oscc_obd), lsm->lsm_object_id);
449 else if (*ea == NULL)
450 obd_free_memmd(exp, &lsm);
454 void oscc_init(struct obd_device *obd)
456 struct osc_creator *oscc;
461 oscc = &obd->u.cli.cl_oscc;
463 memset(oscc, 0, sizeof(*oscc));
464 CFS_INIT_LIST_HEAD(&oscc->oscc_list);
465 cfs_waitq_init(&oscc->oscc_waitq);
466 spin_lock_init(&oscc->oscc_lock);
467 oscc->oscc_obd = obd;
468 oscc->oscc_grow_count = OST_MIN_PRECREATE;
469 oscc->oscc_max_grow_count = OST_MAX_PRECREATE;
471 oscc->oscc_next_id = 2;
472 oscc->oscc_last_id = 1;
473 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
474 /* XXX the export handle should give the oscc the last object */
475 /* oed->oed_oscc.oscc_last_id = exph->....; */