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(struct ptlrpc_request *req, void *data, int rc)
66 struct osc_creator *oscc;
67 struct ost_body *body = NULL;
71 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
72 lustre_swab_ost_body);
73 if (body == NULL && rc == 0)
77 oscc = req->rq_async_args.pointer_arg[0];
78 LASSERT(oscc && (oscc->oscc_obd != LP_POISON));
80 spin_lock(&oscc->oscc_lock);
81 oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
85 int diff = body->oa.o_id - oscc->oscc_last_id;
87 if (diff < oscc->oscc_grow_count)
88 oscc->oscc_grow_count =
89 max(diff/3, OST_MIN_PRECREATE);
91 oscc->oscc_flags &= ~OSCC_FLAG_LOW;
92 oscc->oscc_last_id = body->oa.o_id;
94 spin_unlock(&oscc->oscc_lock);
98 /* valid race delorphan vs create, or somthing after resend */
99 spin_unlock(&oscc->oscc_lock);
100 DEBUG_REQ(D_INODE, req, "Got EGAIN - resend \n");
105 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
106 if (body && rc == -ENOSPC) {
107 oscc->oscc_grow_count = OST_MIN_PRECREATE;
108 oscc->oscc_last_id = body->oa.o_id;
110 spin_unlock(&oscc->oscc_lock);
111 DEBUG_REQ(D_INODE, req, "OST out of space, flagging");
115 /* filter always set body->oa.o_id as the last_id
116 * of filter (see filter_handle_precreate for detail)*/
117 if (body && body->oa.o_id > oscc->oscc_last_id)
118 oscc->oscc_last_id = body->oa.o_id;
119 spin_unlock(&oscc->oscc_lock);
123 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
124 oscc->oscc_grow_count = OST_MIN_PRECREATE;
125 spin_unlock(&oscc->oscc_lock);
126 DEBUG_REQ(D_ERROR, req,
127 "Unknown rc %d from async create: failing oscc", rc);
128 ptlrpc_fail_import(req->rq_import,
129 lustre_msg_get_conn_cnt(req->rq_reqmsg));
133 CDEBUG(D_HA, "preallocated through id "LPU64" (next to use "LPU64")\n",
134 oscc->oscc_last_id, oscc->oscc_next_id);
136 cfs_waitq_signal(&oscc->oscc_waitq);
140 static int oscc_internal_create(struct osc_creator *oscc)
142 struct ptlrpc_request *request;
143 struct ost_body *body;
144 __u32 size[] = { sizeof(struct ptlrpc_body), sizeof(*body) };
147 LASSERT_SPIN_LOCKED(&oscc->oscc_lock);
149 if (oscc->oscc_flags & OSCC_FLAG_CREATING ||
150 oscc->oscc_flags & OSCC_FLAG_RECOVERING) {
151 spin_unlock(&oscc->oscc_lock);
155 if (oscc->oscc_grow_count < oscc->oscc_max_grow_count &&
156 ((oscc->oscc_flags & OSCC_FLAG_LOW) == 0) &&
157 (__s64)(oscc->oscc_last_id - oscc->oscc_next_id) <=
158 (oscc->oscc_grow_count / 4 + 1)) {
159 oscc->oscc_flags |= OSCC_FLAG_LOW;
160 oscc->oscc_grow_count *= 2;
163 if (oscc->oscc_grow_count > oscc->oscc_max_grow_count / 2)
164 oscc->oscc_grow_count = oscc->oscc_max_grow_count / 2;
166 oscc->oscc_flags |= OSCC_FLAG_CREATING;
167 spin_unlock(&oscc->oscc_lock);
169 request = ptlrpc_prep_req(oscc->oscc_obd->u.cli.cl_import,
170 LUSTRE_OST_VERSION, OST_CREATE, 2,
172 if (request == NULL) {
173 spin_lock(&oscc->oscc_lock);
174 oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
175 spin_unlock(&oscc->oscc_lock);
179 request->rq_request_portal = OST_CREATE_PORTAL;
180 ptlrpc_at_set_req_timeout(request);
181 body = lustre_msg_buf(request->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
183 spin_lock(&oscc->oscc_lock);
184 body->oa.o_id = oscc->oscc_last_id + oscc->oscc_grow_count;
185 body->oa.o_gr = oscc->oscc_oa.o_gr;
186 LASSERT(body->oa.o_gr > 0);
187 body->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
188 spin_unlock(&oscc->oscc_lock);
189 CDEBUG(D_RPCTRACE, "prealloc through id "LPU64" (last seen "LPU64")\n",
190 body->oa.o_id, oscc->oscc_last_id);
192 ptlrpc_req_set_repsize(request, 2, size);
194 request->rq_async_args.pointer_arg[0] = oscc;
195 request->rq_interpret_reply = osc_interpret_create;
196 ptlrpcd_add_req(request);
201 static int oscc_has_objects(struct osc_creator *oscc, int count)
204 spin_lock(&oscc->oscc_lock);
205 have_objs = ((__s64)(oscc->oscc_last_id - oscc->oscc_next_id) >= count);
208 oscc_internal_create(oscc);
210 spin_unlock(&oscc->oscc_lock);
216 static int oscc_wait_for_objects(struct osc_creator *oscc, int count)
222 have_objs = oscc_has_objects(oscc, count);
224 spin_lock(&oscc->oscc_lock);
225 ost_full = (oscc->oscc_flags & OSCC_FLAG_NOSPC);
226 spin_unlock(&oscc->oscc_lock);
228 osc_invalid = oscc->oscc_obd->u.cli.cl_import->imp_invalid;
230 return have_objs || ost_full || osc_invalid;
233 static int oscc_precreate(struct osc_creator *oscc, int wait)
235 struct l_wait_info lwi = { 0 };
239 if (oscc_has_objects(oscc, oscc->oscc_grow_count / 2))
245 /* no rc check -- a no-INTR, no-TIMEOUT wait can't fail */
246 l_wait_event(oscc->oscc_waitq, oscc_wait_for_objects(oscc, 1), &lwi);
248 if (!oscc_has_objects(oscc, 1) && (oscc->oscc_flags & OSCC_FLAG_NOSPC))
251 if (oscc->oscc_obd->u.cli.cl_import->imp_invalid)
257 int oscc_recovering(struct osc_creator *oscc)
261 spin_lock(&oscc->oscc_lock);
262 recov = oscc->oscc_flags & OSCC_FLAG_RECOVERING;
263 spin_unlock(&oscc->oscc_lock);
268 /* decide if the OST has remaining object, return value :
269 0 : the OST has remaining object, and don't need to do precreate.
270 1 : the OST has no remaining object, and will send a RPC for precreate.
271 2 : the OST has no remaining object, and will not get any for
272 a potentially very long time
275 int osc_precreate(struct obd_export *exp)
277 struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc;
278 struct obd_import *imp = exp->exp_imp_reverse;
281 LASSERT(oscc != NULL);
282 if (imp != NULL && imp->imp_deactive)
285 if (oscc->oscc_last_id < oscc->oscc_next_id) {
286 spin_lock(&oscc->oscc_lock);
287 if (oscc->oscc_flags & OSCC_FLAG_NOSPC) {
288 spin_unlock(&oscc->oscc_lock);
291 if (oscc->oscc_flags & OSCC_FLAG_SYNC_IN_PROGRESS) {
292 spin_unlock(&oscc->oscc_lock);
295 if (oscc->oscc_flags & OSCC_FLAG_RECOVERING) {
296 spin_unlock(&oscc->oscc_lock);
300 if (oscc->oscc_flags & OSCC_FLAG_CREATING) {
301 spin_unlock(&oscc->oscc_lock);
305 oscc_internal_create(oscc);
311 int osc_create(struct obd_export *exp, struct obdo *oa,
312 struct lov_stripe_md **ea, struct obd_trans_info *oti)
314 struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc;
315 struct lov_stripe_md *lsm;
316 int try_again = 1, rc = 0;
321 LASSERT(oa->o_gr > 0);
322 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
324 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
325 oa->o_flags == OBD_FL_RECREATE_OBJS) {
326 RETURN(osc_real_create(exp, oa, ea, oti));
329 if (oa->o_gr == FILTER_GROUP_LLOG || oa->o_gr == FILTER_GROUP_ECHO)
330 RETURN(osc_real_create(exp, oa, ea, oti));
332 /* this is the special case where create removes orphans */
333 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
334 oa->o_flags == OBD_FL_DELORPHAN) {
335 spin_lock(&oscc->oscc_lock);
336 if (oscc->oscc_flags & OSCC_FLAG_SYNC_IN_PROGRESS) {
337 spin_unlock(&oscc->oscc_lock);
340 if (!(oscc->oscc_flags & OSCC_FLAG_RECOVERING)) {
341 spin_unlock(&oscc->oscc_lock);
344 oscc->oscc_flags |= OSCC_FLAG_SYNC_IN_PROGRESS;
345 /* seting flag LOW we prevent extra grow precreate size
346 * and enforce use last assigned size */
347 oscc->oscc_flags |= OSCC_FLAG_LOW;
348 spin_unlock(&oscc->oscc_lock);
349 CDEBUG(D_HA, "%s: oscc recovery started - delete to "LPU64"\n",
350 oscc->oscc_obd->obd_name, oscc->oscc_next_id - 1);
352 /* delete from next_id on up */
353 oa->o_valid |= OBD_MD_FLID;
354 oa->o_id = oscc->oscc_next_id - 1;
356 rc = osc_real_create(exp, oa, ea, NULL);
358 spin_lock(&oscc->oscc_lock);
359 oscc->oscc_flags &= ~OSCC_FLAG_SYNC_IN_PROGRESS;
360 if (rc == 0 || rc == -ENOSPC) {
362 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
363 oscc->oscc_flags &= ~OSCC_FLAG_RECOVERING;
364 oscc->oscc_last_id = oa->o_id;
365 CDEBUG(D_HA, "%s: oscc recovery finished, last_id: "
366 LPU64", rc: %d\n", oscc->oscc_obd->obd_name,
367 oscc->oscc_last_id, rc);
368 cfs_waitq_signal(&oscc->oscc_waitq);
370 CDEBUG(D_ERROR, "%s: oscc recovery failed: %d\n",
371 oscc->oscc_obd->obd_name, rc);
373 spin_unlock(&oscc->oscc_lock);
381 rc = obd_alloc_memmd(exp, &lsm);
387 /* If orphans are being recovered, then we must wait until
388 it is finished before we can continue with create. */
389 if (oscc_recovering(oscc)) {
390 struct l_wait_info lwi;
392 CDEBUG(D_HA,"%s: oscc recovery in progress, waiting\n",
393 oscc->oscc_obd->obd_name);
395 lwi = LWI_TIMEOUT(cfs_timeout_cap(cfs_time_seconds(
396 obd_timeout / 4)), NULL, NULL);
397 rc = l_wait_event(oscc->oscc_waitq,
398 !oscc_recovering(oscc), &lwi);
399 LASSERT(rc == 0 || rc == -ETIMEDOUT);
400 if (rc == -ETIMEDOUT) {
401 CDEBUG(D_HA,"%s: timeout waiting on recovery\n",
402 oscc->oscc_obd->obd_name);
405 CDEBUG(D_HA, "%s: oscc recovery over, waking up\n",
406 oscc->oscc_obd->obd_name);
409 spin_lock(&oscc->oscc_lock);
410 if (oscc->oscc_flags & OSCC_FLAG_EXITING) {
411 spin_unlock(&oscc->oscc_lock);
415 if (oscc->oscc_last_id >= oscc->oscc_next_id) {
416 memcpy(oa, &oscc->oscc_oa, sizeof(*oa));
417 oa->o_id = oscc->oscc_next_id;
418 lsm->lsm_object_id = oscc->oscc_next_id;
420 oscc->oscc_next_id++;
423 CDEBUG(D_RPCTRACE, "%s: set oscc_next_id = "LPU64"\n",
424 exp->exp_obd->obd_name, oscc->oscc_next_id);
425 } else if (oscc->oscc_flags & OSCC_FLAG_NOSPC) {
427 spin_unlock(&oscc->oscc_lock);
430 spin_unlock(&oscc->oscc_lock);
431 rc = oscc_precreate(oscc, try_again);
437 CDEBUG(D_INFO, "%s: returning objid "LPU64"\n",
438 obd2cli_tgt(oscc->oscc_obd), lsm->lsm_object_id);
439 else if (*ea == NULL)
440 obd_free_memmd(exp, &lsm);
444 void oscc_init(struct obd_device *obd)
446 struct osc_creator *oscc;
451 oscc = &obd->u.cli.cl_oscc;
453 memset(oscc, 0, sizeof(*oscc));
454 CFS_INIT_LIST_HEAD(&oscc->oscc_list);
455 cfs_waitq_init(&oscc->oscc_waitq);
456 spin_lock_init(&oscc->oscc_lock);
457 oscc->oscc_obd = obd;
458 oscc->oscc_grow_count = OST_MIN_PRECREATE;
459 oscc->oscc_max_grow_count = OST_MAX_PRECREATE;
461 oscc->oscc_next_id = 2;
462 oscc->oscc_last_id = 1;
463 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
464 /* XXX the export handle should give the oscc the last object */
465 /* oed->oed_oscc.oscc_last_id = exph->....; */