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, int rc)
61 struct osc_creator *oscc;
62 struct ost_body *body = NULL;
66 body = lustre_swab_repbuf(req, 0, sizeof(*body),
67 lustre_swab_ost_body);
68 if (body == NULL && rc == 0)
72 oscc = req->rq_async_args.pointer_arg[0];
73 LASSERT(oscc && (oscc->oscc_obd != LP_POISON));
75 spin_lock(&oscc->oscc_lock);
76 oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
77 if (rc == -ENOSPC || rc == -EROFS) {
78 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
79 if (body && rc == -ENOSPC) {
80 oscc->oscc_grow_count = OST_MIN_PRECREATE;
81 oscc->oscc_last_id = body->oa.o_id;
83 spin_unlock(&oscc->oscc_lock);
84 DEBUG_REQ(D_INODE, req, "OST out of space, flagging");
85 } else if (rc != 0 && rc != -EIO) {
86 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
87 oscc->oscc_grow_count = OST_MIN_PRECREATE;
88 spin_unlock(&oscc->oscc_lock);
89 DEBUG_REQ(D_ERROR, req,
90 "unknown rc %d from async create: failing oscc", rc);
91 ptlrpc_fail_import(req->rq_import, req->rq_import_generation);
94 oscc->oscc_flags &= ~OSCC_FLAG_LOW;
96 int diff = body->oa.o_id - oscc->oscc_last_id;
97 if (diff != oscc->oscc_grow_count)
98 oscc->oscc_grow_count =
99 max(diff/3, OST_MIN_PRECREATE);
100 oscc->oscc_last_id = body->oa.o_id;
103 spin_unlock(&oscc->oscc_lock);
106 CDEBUG(D_HA, "preallocated through id "LPU64" (last used "LPU64")\n",
107 oscc->oscc_last_id, oscc->oscc_next_id);
109 wake_up(&oscc->oscc_waitq);
113 static int oscc_internal_create(struct osc_creator *oscc)
115 struct ptlrpc_request *request;
116 struct ost_body *body;
117 int size = sizeof(*body);
120 spin_lock(&oscc->oscc_lock);
121 if (oscc->oscc_grow_count < OST_MAX_PRECREATE &&
122 !(oscc->oscc_flags & (OSCC_FLAG_LOW | OSCC_FLAG_RECOVERING)) &&
123 (__s64)(oscc->oscc_last_id - oscc->oscc_next_id) <=
124 (oscc->oscc_grow_count / 4 + 1)) {
125 oscc->oscc_flags |= OSCC_FLAG_LOW;
126 oscc->oscc_grow_count *= 2;
129 if (oscc->oscc_grow_count > OST_MAX_PRECREATE / 2)
130 oscc->oscc_grow_count = OST_MAX_PRECREATE / 2;
132 if (oscc->oscc_flags & OSCC_FLAG_CREATING ||
133 oscc->oscc_flags & OSCC_FLAG_RECOVERING) {
134 spin_unlock(&oscc->oscc_lock);
137 oscc->oscc_flags |= OSCC_FLAG_CREATING;
138 spin_unlock(&oscc->oscc_lock);
140 request = ptlrpc_prep_req(oscc->oscc_obd->u.cli.cl_import, OST_CREATE,
142 if (request == NULL) {
143 spin_lock(&oscc->oscc_lock);
144 oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
145 spin_unlock(&oscc->oscc_lock);
149 request->rq_request_portal = OST_CREATE_PORTAL; //XXX FIXME bug 249
150 body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof(*body));
152 spin_lock(&oscc->oscc_lock);
153 body->oa.o_id = oscc->oscc_last_id + oscc->oscc_grow_count;
154 body->oa.o_valid |= OBD_MD_FLID;
155 spin_unlock(&oscc->oscc_lock);
156 CDEBUG(D_HA, "preallocating through id "LPU64" (last used "LPU64")\n",
157 body->oa.o_id, oscc->oscc_next_id);
159 request->rq_replen = lustre_msg_size(1, &size);
161 request->rq_async_args.pointer_arg[0] = oscc;
162 request->rq_interpret_reply = osc_interpret_create;
163 ptlrpcd_add_req(request);
168 static int oscc_has_objects(struct osc_creator *oscc, int count)
171 spin_lock(&oscc->oscc_lock);
172 have_objs = ((__s64)(oscc->oscc_last_id - oscc->oscc_next_id) >= count);
173 spin_unlock(&oscc->oscc_lock);
176 oscc_internal_create(oscc);
181 static int oscc_wait_for_objects(struct osc_creator *oscc, int count)
187 have_objs = oscc_has_objects(oscc, count);
189 spin_lock(&oscc->oscc_lock);
190 ost_full = (oscc->oscc_flags & OSCC_FLAG_NOSPC);
191 spin_unlock(&oscc->oscc_lock);
193 osc_invalid = oscc->oscc_obd->u.cli.cl_import->imp_invalid;
195 return have_objs || ost_full || osc_invalid;
198 static int oscc_precreate(struct osc_creator *oscc, int wait)
200 struct l_wait_info lwi = { 0 };
204 if (oscc_has_objects(oscc, oscc->oscc_grow_count / 2))
210 /* no rc check -- a no-INTR, no-TIMEOUT wait can't fail */
211 l_wait_event(oscc->oscc_waitq, oscc_wait_for_objects(oscc, 1), &lwi);
213 if (!oscc_has_objects(oscc, 1) && (oscc->oscc_flags & OSCC_FLAG_NOSPC))
216 if (oscc->oscc_obd->u.cli.cl_import->imp_invalid)
222 int oscc_recovering(struct osc_creator *oscc)
226 spin_lock(&oscc->oscc_lock);
227 recov = oscc->oscc_flags & OSCC_FLAG_RECOVERING;
228 spin_unlock(&oscc->oscc_lock);
233 int osc_create(struct obd_export *exp, struct obdo *oa,
234 struct lov_stripe_md **ea, struct obd_trans_info *oti)
236 struct lov_stripe_md *lsm;
237 struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc;
238 int try_again = 1, rc = 0;
243 if ((oa->o_valid & OBD_MD_FLGROUP) && (oa->o_gr != 0))
244 RETURN(osc_real_create(exp, oa, ea, oti));
246 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
247 oa->o_flags == OBD_FL_RECREATE_OBJS) {
248 RETURN(osc_real_create(exp, oa, ea, oti));
251 /* this is the special case where create removes orphans */
252 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
253 oa->o_flags == OBD_FL_DELORPHAN) {
254 spin_lock(&oscc->oscc_lock);
255 if (oscc->oscc_flags & OSCC_FLAG_SYNC_IN_PROGRESS) {
256 spin_unlock(&oscc->oscc_lock);
259 if (!(oscc->oscc_flags & OSCC_FLAG_RECOVERING)) {
260 spin_unlock(&oscc->oscc_lock);
263 oscc->oscc_flags |= OSCC_FLAG_SYNC_IN_PROGRESS;
264 spin_unlock(&oscc->oscc_lock);
265 CDEBUG(D_HA, "%s: oscc recovery started\n",
266 oscc->oscc_obd->obd_name);
268 /* delete from next_id on up */
269 oa->o_valid |= OBD_MD_FLID;
270 oa->o_id = oscc->oscc_next_id - 1;
272 CDEBUG(D_HA, "%s: deleting to next_id: "LPU64"\n",
273 oscc->oscc_obd->obd_name, oa->o_id);
275 rc = osc_real_create(exp, oa, ea, NULL);
277 spin_lock(&oscc->oscc_lock);
278 oscc->oscc_flags &= ~OSCC_FLAG_SYNC_IN_PROGRESS;
279 if (rc == 0 || rc == -ENOSPC) {
281 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
282 oscc->oscc_flags &= ~OSCC_FLAG_RECOVERING;
283 oscc->oscc_last_id = oa->o_id;
284 CDEBUG(D_HA, "%s: oscc recovery finished: %d\n",
285 oscc->oscc_obd->obd_name, rc);
286 wake_up(&oscc->oscc_waitq);
288 CDEBUG(D_ERROR, "%s: oscc recovery failed: %d\n",
289 oscc->oscc_obd->obd_name, rc);
291 spin_unlock(&oscc->oscc_lock);
299 rc = obd_alloc_memmd(exp, &lsm);
305 /* If orphans are being recovered, then we must wait until
306 it is finished before we can continue with create. */
307 if (oscc_recovering(oscc)) {
308 struct l_wait_info lwi;
310 CDEBUG(D_HA,"%p: oscc recovery in progress, waiting\n",
313 lwi = LWI_TIMEOUT(MAX(obd_timeout*HZ/4, 1), NULL, NULL);
314 rc = l_wait_event(oscc->oscc_waitq,
315 !oscc_recovering(oscc), &lwi);
316 LASSERT(rc == 0 || rc == -ETIMEDOUT);
317 if (rc == -ETIMEDOUT) {
318 CDEBUG(D_HA,"%p: timeout waiting on recovery\n",
322 CDEBUG(D_HA, "%p: oscc recovery over, waking up\n",
326 spin_lock(&oscc->oscc_lock);
327 if (oscc->oscc_flags & OSCC_FLAG_EXITING) {
328 spin_unlock(&oscc->oscc_lock);
332 if (oscc->oscc_last_id >= oscc->oscc_next_id) {
333 memcpy(oa, &oscc->oscc_oa, sizeof(*oa));
334 oa->o_id = oscc->oscc_next_id;
335 lsm->lsm_object_id = oscc->oscc_next_id;
337 oscc->oscc_next_id++;
339 } else if (oscc->oscc_flags & OSCC_FLAG_NOSPC) {
341 spin_unlock(&oscc->oscc_lock);
344 spin_unlock(&oscc->oscc_lock);
345 rc = oscc_precreate(oscc, try_again);
351 CDEBUG(D_HA, "%s: returning objid "LPU64"\n",
352 oscc->oscc_obd->u.cli.cl_import->imp_target_uuid.uuid,
354 else if (*ea == NULL)
355 obd_free_memmd(exp, &lsm);
359 void oscc_init(struct obd_device *obd)
361 struct osc_creator *oscc;
366 oscc = &obd->u.cli.cl_oscc;
368 memset(oscc, 0, sizeof(*oscc));
369 INIT_LIST_HEAD(&oscc->oscc_list);
370 init_waitqueue_head(&oscc->oscc_waitq);
371 spin_lock_init(&oscc->oscc_lock);
372 oscc->oscc_obd = obd;
373 oscc->oscc_grow_count = OST_MIN_PRECREATE;
375 oscc->oscc_next_id = 2;
376 oscc->oscc_last_id = 1;
377 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
378 /* XXX the export handle should give the oscc the last object */
379 /* oed->oed_oscc.oscc_last_id = exph->....; */