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 spin_lock(&oscc->oscc_lock);
74 oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
75 if (rc == -ENOSPC || rc == -EROFS) {
76 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
77 if (body && rc == -ENOSPC) {
78 oscc->oscc_grow_count = OST_MIN_PRECREATE;
79 oscc->oscc_last_id = body->oa.o_id;
81 spin_unlock(&oscc->oscc_lock);
82 DEBUG_REQ(D_INODE, req, "OST out of space, flagging");
83 } else if (rc != 0 && rc != -EIO) {
84 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
85 oscc->oscc_grow_count = OST_MIN_PRECREATE;
86 spin_unlock(&oscc->oscc_lock);
87 DEBUG_REQ(D_ERROR, req,
88 "unknown rc %d from async create: failing oscc", rc);
89 ptlrpc_fail_import(req->rq_import, req->rq_import_generation);
92 oscc->oscc_flags &= ~OSCC_FLAG_LOW;
94 int diff = body->oa.o_id - oscc->oscc_last_id;
95 if (diff != oscc->oscc_grow_count)
96 oscc->oscc_grow_count =
97 max(diff/3, OST_MIN_PRECREATE);
98 oscc->oscc_last_id = body->oa.o_id;
101 spin_unlock(&oscc->oscc_lock);
104 CDEBUG(D_HA, "preallocated through id "LPU64" (last used "LPU64")\n",
105 oscc->oscc_last_id, oscc->oscc_next_id);
107 wake_up(&oscc->oscc_waitq);
111 static int oscc_internal_create(struct osc_creator *oscc)
113 struct ptlrpc_request *request;
114 struct ost_body *body;
115 int size = sizeof(*body);
118 spin_lock(&oscc->oscc_lock);
119 if (oscc->oscc_grow_count < OST_MAX_PRECREATE &&
120 !(oscc->oscc_flags & (OSCC_FLAG_LOW | OSCC_FLAG_RECOVERING)) &&
121 (__s64)(oscc->oscc_last_id - oscc->oscc_next_id) <=
122 (oscc->oscc_grow_count / 4 + 1)) {
123 oscc->oscc_flags |= OSCC_FLAG_LOW;
124 oscc->oscc_grow_count *= 2;
127 if (oscc->oscc_grow_count > OST_MAX_PRECREATE / 2)
128 oscc->oscc_grow_count = OST_MAX_PRECREATE / 2;
130 if (oscc->oscc_flags & OSCC_FLAG_CREATING ||
131 oscc->oscc_flags & OSCC_FLAG_RECOVERING) {
132 spin_unlock(&oscc->oscc_lock);
135 oscc->oscc_flags |= OSCC_FLAG_CREATING;
136 spin_unlock(&oscc->oscc_lock);
138 request = ptlrpc_prep_req(oscc->oscc_obd->u.cli.cl_import,
139 LUSTRE_OBD_VERSION, OST_CREATE,
141 if (request == NULL) {
142 spin_lock(&oscc->oscc_lock);
143 oscc->oscc_flags &= ~OSCC_FLAG_CREATING;
144 spin_unlock(&oscc->oscc_lock);
148 request->rq_request_portal = OST_CREATE_PORTAL; //XXX FIXME bug 249
149 body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof(*body));
151 spin_lock(&oscc->oscc_lock);
152 body->oa.o_id = oscc->oscc_last_id + oscc->oscc_grow_count;
153 /* probably we should take frequence of request into account? -bzzz */
154 if (oscc->oscc_grow_count < oscc->oscc_max_grow_count) {
155 oscc->oscc_grow_count *= 2;
156 if (oscc->oscc_grow_count > oscc->oscc_max_grow_count)
157 oscc->oscc_grow_count = oscc->oscc_max_grow_count;
159 body->oa.o_gr = oscc->oscc_gr;
160 LASSERT(body->oa.o_gr > 0);
161 body->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
162 spin_unlock(&oscc->oscc_lock);
163 CDEBUG(D_INFO, "preallocating through id "LPU64" (last used "LPU64")\n",
164 body->oa.o_id, oscc->oscc_next_id);
166 request->rq_replen = lustre_msg_size(1, &size);
168 request->rq_async_args.pointer_arg[0] = oscc;
169 request->rq_interpret_reply = osc_interpret_create;
170 ptlrpcd_add_req(request);
175 static int oscc_has_objects(struct osc_creator *oscc, int count)
178 spin_lock(&oscc->oscc_lock);
179 have_objs = ((__s64)(oscc->oscc_last_id - oscc->oscc_next_id) >= count);
180 spin_unlock(&oscc->oscc_lock);
183 oscc_internal_create(oscc);
188 static int oscc_wait_for_objects(struct osc_creator *oscc, int count)
194 have_objs = oscc_has_objects(oscc, count);
196 spin_lock(&oscc->oscc_lock);
197 ost_full = (oscc->oscc_flags & OSCC_FLAG_NOSPC);
198 spin_unlock(&oscc->oscc_lock);
200 osc_invalid = oscc->oscc_obd->u.cli.cl_import->imp_invalid;
202 return have_objs || ost_full || osc_invalid;
205 static int oscc_precreate(struct osc_creator *oscc, int wait)
207 struct l_wait_info lwi = { 0 };
211 if (oscc_has_objects(oscc, oscc->oscc_grow_count / 2))
217 /* no rc check -- a no-INTR, no-TIMEOUT wait can't fail */
218 l_wait_event(oscc->oscc_waitq, oscc_wait_for_objects(oscc, 1), &lwi);
220 if (!oscc_has_objects(oscc, 1) && (oscc->oscc_flags & OSCC_FLAG_NOSPC))
223 if (oscc->oscc_obd->u.cli.cl_import->imp_invalid)
229 int oscc_recovering(struct osc_creator *oscc)
233 spin_lock(&oscc->oscc_lock);
234 recov = oscc->oscc_flags & OSCC_FLAG_RECOVERING;
235 spin_unlock(&oscc->oscc_lock);
240 int osc_create(struct obd_export *exp, struct obdo *oa,
241 void *acl, int acl_size,
242 struct lov_stripe_md **ea, struct obd_trans_info *oti)
244 struct lov_stripe_md *lsm;
245 struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc;
246 int try_again = 1, rc = 0;
250 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
251 LASSERT(oa->o_gr > 0);
252 LASSERT(acl == NULL && acl_size == 0);
254 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
255 oa->o_flags == OBD_FL_RECREATE_OBJS) {
256 /* Exceptional case where we are trying to repair missing
257 * objects for various groups. We have already validated that
258 * this is a valid group for the file. Don't set oscc->oscc_gr.
260 RETURN(osc_real_create(exp, oa, ea, oti));
263 LASSERT(oscc->oscc_gr == 0 || oscc->oscc_gr == oa->o_gr);
264 oscc->oscc_gr = oa->o_gr;
266 if (oa->o_gr == FILTER_GROUP_LLOG || oa->o_gr == FILTER_GROUP_ECHO)
267 RETURN(osc_real_create(exp, oa, ea, oti));
269 /* this is the special case where create removes orphans */
270 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
271 oa->o_flags == OBD_FL_DELORPHAN) {
272 spin_lock(&oscc->oscc_lock);
273 if (oscc->oscc_flags & OSCC_FLAG_SYNC_IN_PROGRESS) {
274 spin_unlock(&oscc->oscc_lock);
277 if (!(oscc->oscc_flags & OSCC_FLAG_RECOVERING)) {
278 spin_unlock(&oscc->oscc_lock);
281 oscc->oscc_flags |= OSCC_FLAG_SYNC_IN_PROGRESS;
282 spin_unlock(&oscc->oscc_lock);
283 CDEBUG(D_HA, "%s; oscc recovery started\n",
284 exp->exp_obd->obd_name);
285 LASSERT(oscc->oscc_flags & OSCC_FLAG_RECOVERING);
287 /* delete from next_id on up */
288 oa->o_valid |= OBD_MD_FLID;
289 oa->o_id = oscc->oscc_next_id - 1;
291 CDEBUG(D_HA, "%s: deleting to next_id: "LPU64"\n",
292 oscc->oscc_obd->obd_name, oa->o_id);
294 rc = osc_real_create(exp, oa, ea, NULL);
295 if (oscc->oscc_obd == NULL) {
296 CWARN("the obd for oscc %p has been freed\n", oscc);
300 spin_lock(&oscc->oscc_lock);
301 oscc->oscc_flags &= ~OSCC_FLAG_SYNC_IN_PROGRESS;
302 if (rc == 0 || rc == -ENOSPC) {
304 oscc->oscc_flags |= OSCC_FLAG_NOSPC;
305 oscc->oscc_flags &= ~OSCC_FLAG_RECOVERING;
306 oscc->oscc_last_id = oa->o_id;
308 CDEBUG(D_HA, "%s: oscc recovery finished: %d\n",
309 oscc->oscc_obd->obd_name, rc);
310 wake_up(&oscc->oscc_waitq);
313 CDEBUG(D_ERROR, "%s: oscc recovery failed: %d\n",
314 oscc->oscc_obd->obd_name, rc);
316 spin_unlock(&oscc->oscc_lock);
323 rc = obd_alloc_memmd(exp, &lsm);
329 /* If orphans are being recovered, then we must wait until
330 it is finished before we can continue with create. */
331 if (oscc_recovering(oscc)) {
332 struct l_wait_info lwi;
334 CDEBUG(D_HA,"%p: oscc recovery in progress, waiting\n",
336 lwi = LWI_TIMEOUT(MAX(obd_timeout*HZ/4, 1), NULL, NULL);
337 rc = l_wait_event(oscc->oscc_waitq,
338 !oscc_recovering(oscc), &lwi);
340 LASSERT(rc == 0 || rc == -ETIMEDOUT);
341 if (rc == -ETIMEDOUT) {
342 CDEBUG(D_HA,"%p: timeout waiting on recovery\n",
346 CDEBUG(D_HA, "%s: oscc recovery over, waking up\n",
347 exp->exp_obd->obd_name);
350 spin_lock(&oscc->oscc_lock);
351 if (oscc->oscc_flags & OSCC_FLAG_EXITING) {
352 spin_unlock(&oscc->oscc_lock);
356 if (oscc->oscc_last_id >= oscc->oscc_next_id) {
357 memcpy(oa, &oscc->oscc_oa, sizeof(*oa));
358 oa->o_id = oscc->oscc_next_id;
359 oa->o_gr = oscc->oscc_gr;
360 lsm->lsm_object_id = oscc->oscc_next_id;
361 lsm->lsm_object_gr = oscc->oscc_gr;
363 oscc->oscc_next_id++;
365 } else if (oscc->oscc_flags & OSCC_FLAG_NOSPC) {
367 spin_unlock(&oscc->oscc_lock);
370 spin_unlock(&oscc->oscc_lock);
371 rc = oscc_precreate(oscc, try_again);
377 CDEBUG(D_HA, "%s: returning objid "LPU64"\n",
378 oscc->oscc_obd->u.cli.cl_import->imp_target_uuid.uuid,
380 else if (*ea == NULL)
381 obd_free_memmd(exp, &lsm);
385 void oscc_init(struct obd_device *obd)
387 struct osc_creator *oscc;
392 oscc = &obd->u.cli.cl_oscc;
394 memset(oscc, 0, sizeof(*oscc));
395 INIT_LIST_HEAD(&oscc->oscc_list);
396 init_waitqueue_head(&oscc->oscc_waitq);
397 spin_lock_init(&oscc->oscc_lock);
398 oscc->oscc_obd = obd;
399 oscc->oscc_kick_barrier = 100;
400 oscc->oscc_max_grow_count = 2000;
401 oscc->oscc_grow_count = OST_MIN_PRECREATE;
403 oscc->oscc_next_id = 2;
404 oscc->oscc_last_id = 1;
405 oscc->oscc_flags |= OSCC_FLAG_RECOVERING;
406 /* XXX the export handle should give the oscc the last object */
407 /* oed->oed_oscc.oscc_last_id = exph->....; */