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;
69 if (rc == -ENOSPC || rc == -EROFS) {
70 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
71 if (body && rc == -ENOSPC) {
72 oscc->oscc_grow_count = OST_MIN_PRECREATE;
73 oscc->oscc_last_id = body->oa.o_id;
75 spin_unlock(&oscc->oscc_lock);
76 DEBUG_REQ(D_INODE, req, "OST out of space, flagging");
77 } else if (rc != 0 && rc != -EIO) {
78 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
79 oscc->oscc_grow_count = OST_MIN_PRECREATE;
80 spin_unlock(&oscc->oscc_lock);
81 DEBUG_REQ(D_ERROR, req,
82 "Unknown rc %d from async create: failing oscc", rc);
83 ptlrpc_fail_import(req->rq_import,
84 lustre_msg_get_conn_cnt(req->rq_reqmsg));
88 int diff = body->oa.o_id - oscc->oscc_last_id;
90 if (diff < oscc->oscc_grow_count)
91 oscc->oscc_grow_count =
92 max(diff/3, OST_MIN_PRECREATE);
94 oscc->oscc_flags &= ~OSCC_FLAG_LOW;
95 oscc->oscc_last_id = body->oa.o_id;
98 /* filter always set body->oa.o_id as the last_id
99 * of filter (see filter_handle_precreate for detail)*/
100 if (body && body->oa.o_id > oscc->oscc_last_id)
101 oscc->oscc_last_id = body->oa.o_id;
103 spin_unlock(&oscc->oscc_lock);
107 CDEBUG(D_HA, "preallocated through id "LPU64" (next to use "LPU64")\n",
108 oscc->oscc_last_id, oscc->oscc_next_id);
110 cfs_waitq_signal(&oscc->oscc_waitq);
114 static int oscc_internal_create(struct osc_creator *oscc)
116 struct ptlrpc_request *request;
117 struct ost_body *body;
118 int size[] = { sizeof(struct ptlrpc_body), sizeof(*body) };
121 spin_lock(&oscc->oscc_lock);
122 if (oscc->oscc_grow_count < OST_MAX_PRECREATE &&
123 !(oscc->oscc_flags & (OSCC_FLAG_LOW | OSCC_FLAG_RECOVERING)) &&
124 (__s64)(oscc->oscc_last_id - oscc->oscc_next_id) <=
125 (oscc->oscc_grow_count / 4 + 1)) {
126 oscc->oscc_flags |= OSCC_FLAG_LOW;
127 oscc->oscc_grow_count *= 2;
130 if (oscc->oscc_grow_count > OST_MAX_PRECREATE / 2)
131 oscc->oscc_grow_count = OST_MAX_PRECREATE / 2;
133 if (oscc->oscc_flags & OSCC_FLAG_CREATING ||
134 oscc->oscc_flags & OSCC_FLAG_RECOVERING) {
135 spin_unlock(&oscc->oscc_lock);
138 oscc->oscc_flags |= OSCC_FLAG_CREATING;
139 spin_unlock(&oscc->oscc_lock);
141 request = ptlrpc_prep_req(oscc->oscc_obd->u.cli.cl_import,
142 LUSTRE_OST_VERSION, OST_CREATE, 2,
144 if (request == NULL) {
145 spin_lock(&oscc->oscc_lock);
146 oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
147 spin_unlock(&oscc->oscc_lock);
151 request->rq_request_portal = OST_CREATE_PORTAL; //XXX FIXME bug 249
152 body = lustre_msg_buf(request->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
154 spin_lock(&oscc->oscc_lock);
155 body->oa.o_id = oscc->oscc_last_id + oscc->oscc_grow_count;
156 body->oa.o_gr = oscc->oscc_oa.o_gr;
157 LASSERT(body->oa.o_gr > 0);
158 body->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
159 spin_unlock(&oscc->oscc_lock);
160 CDEBUG(D_HA, "preallocating through id "LPU64" (last seen "LPU64")\n",
161 body->oa.o_id, oscc->oscc_last_id);
163 ptlrpc_req_set_repsize(request, 2, size);
165 request->rq_async_args.pointer_arg[0] = oscc;
166 request->rq_interpret_reply = osc_interpret_create;
167 ptlrpcd_add_req(request);
172 static int oscc_has_objects(struct osc_creator *oscc, int count)
175 spin_lock(&oscc->oscc_lock);
176 have_objs = ((__s64)(oscc->oscc_last_id - oscc->oscc_next_id) >= count);
177 spin_unlock(&oscc->oscc_lock);
180 oscc_internal_create(oscc);
185 static int oscc_wait_for_objects(struct osc_creator *oscc, int count)
191 have_objs = oscc_has_objects(oscc, count);
193 spin_lock(&oscc->oscc_lock);
194 ost_full = (oscc->oscc_flags & OSCC_FLAG_NOSPC);
195 spin_unlock(&oscc->oscc_lock);
197 osc_invalid = oscc->oscc_obd->u.cli.cl_import->imp_invalid;
199 return have_objs || ost_full || osc_invalid;
202 static int oscc_precreate(struct osc_creator *oscc, int wait)
204 struct l_wait_info lwi = { 0 };
208 if (oscc_has_objects(oscc, oscc->oscc_grow_count / 2))
214 /* no rc check -- a no-INTR, no-TIMEOUT wait can't fail */
215 l_wait_event(oscc->oscc_waitq, oscc_wait_for_objects(oscc, 1), &lwi);
217 if (!oscc_has_objects(oscc, 1) && (oscc->oscc_flags & OSCC_FLAG_NOSPC))
220 if (oscc->oscc_obd->u.cli.cl_import->imp_invalid)
226 int oscc_recovering(struct osc_creator *oscc)
230 spin_lock(&oscc->oscc_lock);
231 recov = oscc->oscc_flags & OSCC_FLAG_RECOVERING;
232 spin_unlock(&oscc->oscc_lock);
237 /* decide if the OST has remaining object, return value :
238 0 : the OST has remaining object, and don't need to do precreate.
239 1 : the OST has no remaining object, and will send a RPC for precreate.
240 2 : the OST has no remaining object, and will not get any for
241 a potentially very long time
244 int osc_precreate(struct obd_export *exp, int need_create)
246 struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc;
247 struct obd_import *imp = exp->exp_imp_reverse;
250 LASSERT(oscc != NULL);
251 if (imp != NULL && imp->imp_deactive)
254 if (oscc->oscc_last_id < oscc->oscc_next_id) {
255 spin_lock(&oscc->oscc_lock);
256 if (oscc->oscc_flags & OSCC_FLAG_NOSPC) {
257 spin_unlock(&oscc->oscc_lock);
260 if (oscc->oscc_flags & OSCC_FLAG_SYNC_IN_PROGRESS) {
261 spin_unlock(&oscc->oscc_lock);
264 if (oscc->oscc_flags & OSCC_FLAG_RECOVERING) {
265 spin_unlock(&oscc->oscc_lock);
268 spin_unlock(&oscc->oscc_lock);
270 if (oscc->oscc_flags & OSCC_FLAG_CREATING)
276 oscc_internal_create(oscc);
282 int osc_create(struct obd_export *exp, struct obdo *oa,
283 struct lov_stripe_md **ea, struct obd_trans_info *oti)
285 struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc;
286 struct lov_stripe_md *lsm;
287 int try_again = 1, rc = 0;
292 LASSERT(oa->o_gr > 0);
293 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
295 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
296 oa->o_flags == OBD_FL_RECREATE_OBJS) {
297 RETURN(osc_real_create(exp, oa, ea, oti));
300 if (oa->o_gr == FILTER_GROUP_LLOG || oa->o_gr == FILTER_GROUP_ECHO)
301 RETURN(osc_real_create(exp, oa, ea, oti));
303 /* this is the special case where create removes orphans */
304 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
305 oa->o_flags == OBD_FL_DELORPHAN) {
306 spin_lock(&oscc->oscc_lock);
307 if (oscc->oscc_flags & OSCC_FLAG_SYNC_IN_PROGRESS) {
308 spin_unlock(&oscc->oscc_lock);
311 if (!(oscc->oscc_flags & OSCC_FLAG_RECOVERING)) {
312 spin_unlock(&oscc->oscc_lock);
315 oscc->oscc_flags |= OSCC_FLAG_SYNC_IN_PROGRESS;
316 spin_unlock(&oscc->oscc_lock);
317 CDEBUG(D_HA, "%s: oscc recovery started\n",
318 oscc->oscc_obd->obd_name);
320 /* delete from next_id on up */
321 oa->o_valid |= OBD_MD_FLID;
322 oa->o_id = oscc->oscc_next_id - 1;
324 CDEBUG(D_HA, "%s: deleting to next_id: "LPU64"\n",
325 oscc->oscc_obd->obd_name, oa->o_id);
327 rc = osc_real_create(exp, oa, ea, NULL);
329 spin_lock(&oscc->oscc_lock);
330 oscc->oscc_flags &= ~OSCC_FLAG_SYNC_IN_PROGRESS;
331 if (rc == 0 || rc == -ENOSPC) {
333 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
334 oscc->oscc_flags &= ~OSCC_FLAG_RECOVERING;
335 oscc->oscc_last_id = oa->o_id;
336 CDEBUG(D_HA, "%s: oscc recovery finished, last_id: "
337 LPU64", rc: %d\n", oscc->oscc_obd->obd_name,
338 oscc->oscc_last_id, rc);
339 cfs_waitq_signal(&oscc->oscc_waitq);
341 CDEBUG(D_ERROR, "%s: oscc recovery failed: %d\n",
342 oscc->oscc_obd->obd_name, rc);
344 spin_unlock(&oscc->oscc_lock);
352 rc = obd_alloc_memmd(exp, &lsm);
358 /* If orphans are being recovered, then we must wait until
359 it is finished before we can continue with create. */
360 if (oscc_recovering(oscc)) {
361 struct l_wait_info lwi;
363 CDEBUG(D_HA,"%p: oscc recovery in progress, waiting\n",
366 lwi = LWI_TIMEOUT(cfs_timeout_cap(cfs_time_seconds(obd_timeout/4)),
368 rc = l_wait_event(oscc->oscc_waitq,
369 !oscc_recovering(oscc), &lwi);
370 LASSERT(rc == 0 || rc == -ETIMEDOUT);
371 if (rc == -ETIMEDOUT) {
372 CDEBUG(D_HA,"%p: timeout waiting on recovery\n",
376 CDEBUG(D_HA, "%p: oscc recovery over, waking up\n",
380 spin_lock(&oscc->oscc_lock);
381 if (oscc->oscc_flags & OSCC_FLAG_EXITING) {
382 spin_unlock(&oscc->oscc_lock);
386 if (oscc->oscc_last_id >= oscc->oscc_next_id) {
387 memcpy(oa, &oscc->oscc_oa, sizeof(*oa));
388 oa->o_id = oscc->oscc_next_id;
389 lsm->lsm_object_id = oscc->oscc_next_id;
391 oscc->oscc_next_id++;
394 CDEBUG(D_HA, "%s: set oscc_next_id = "LPU64"\n",
395 exp->exp_obd->obd_name, oscc->oscc_next_id);
396 } else if (oscc->oscc_flags & OSCC_FLAG_NOSPC) {
398 spin_unlock(&oscc->oscc_lock);
401 spin_unlock(&oscc->oscc_lock);
402 rc = oscc_precreate(oscc, try_again);
408 CDEBUG(D_HA, "%s: returning objid "LPU64"\n",
409 obd2cli_tgt(oscc->oscc_obd), lsm->lsm_object_id);
410 else if (*ea == NULL)
411 obd_free_memmd(exp, &lsm);
415 void oscc_init(struct obd_device *obd)
417 struct osc_creator *oscc;
422 oscc = &obd->u.cli.cl_oscc;
424 memset(oscc, 0, sizeof(*oscc));
425 CFS_INIT_LIST_HEAD(&oscc->oscc_list);
426 cfs_waitq_init(&oscc->oscc_waitq);
427 spin_lock_init(&oscc->oscc_lock);
428 oscc->oscc_obd = obd;
429 oscc->oscc_grow_count = OST_MIN_PRECREATE;
431 oscc->oscc_next_id = 2;
432 oscc->oscc_last_id = 1;
433 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
434 /* XXX the export handle should give the oscc the last object */
435 /* oed->oed_oscc.oscc_last_id = exph->....; */