1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
5 * Lustre Metadata Server (mds) handling of striped file data
7 * Copyright (C) 2001-2003 Cluster File Systems, Inc.
8 * Author: Peter Braam <braam@clusterfs.com>
10 * This file is part of Lustre, http://www.lustre.org.
12 * Lustre is free software; you can redistribute it and/or
13 * modify it under the terms of version 2 of the GNU General Public
14 * License as published by the Free Software Foundation.
16 * Lustre is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with Lustre; if not, write to the Free Software
23 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
27 # define EXPORT_SYMTAB
29 #define DEBUG_SUBSYSTEM S_MDS
31 #include <linux/module.h>
32 #include <linux/lustre_mds.h>
33 #include <linux/lustre_idl.h>
34 #include <linux/obd_class.h>
35 #include <linux/obd_lov.h>
36 #include <linux/lustre_lib.h>
37 #include <linux/lustre_fsfilt.h>
39 #include "mds_internal.h"
41 void le_lov_desc_to_cpu (struct lov_desc *ld)
43 ld->ld_tgt_count = le32_to_cpu (ld->ld_tgt_count);
44 ld->ld_default_stripe_count = le32_to_cpu (ld->ld_default_stripe_count);
45 ld->ld_default_stripe_size = le32_to_cpu (ld->ld_default_stripe_size);
46 ld->ld_pattern = le32_to_cpu (ld->ld_pattern);
49 void cpu_to_le_lov_desc (struct lov_desc *ld)
51 ld->ld_tgt_count = cpu_to_le32 (ld->ld_tgt_count);
52 ld->ld_default_stripe_count = cpu_to_le32 (ld->ld_default_stripe_count);
53 ld->ld_default_stripe_size = cpu_to_le32 (ld->ld_default_stripe_size);
54 ld->ld_pattern = cpu_to_le32 (ld->ld_pattern);
57 void mds_lov_update_objids(struct obd_device *obd, obd_id *ids)
59 struct mds_obd *mds = &obd->u.mds;
64 for (i = 0; i < mds->mds_lov_desc.ld_tgt_count; i++)
65 if (ids[i] > (mds->mds_lov_objids)[i])
66 (mds->mds_lov_objids)[i] = ids[i];
71 static int mds_lov_read_objids(struct obd_device *obd)
73 struct mds_obd *mds = &obd->u.mds;
76 int i, rc, size = mds->mds_lov_desc.ld_tgt_count * sizeof(*ids);
79 if (mds->mds_lov_objids != NULL)
85 mds->mds_lov_objids = ids;
87 if (mds->mds_lov_objid_filp->f_dentry->d_inode->i_size == 0)
89 rc = fsfilt_read_record(obd, mds->mds_lov_objid_filp, ids, size, &off);
91 CERROR("Error reading objids %d\n", rc);
93 mds->mds_lov_objids_valid = 1;
97 for (i = 0; i < mds->mds_lov_desc.ld_tgt_count; i++)
98 CDEBUG(D_INFO, "read last object "LPU64" for idx %d\n",
99 mds->mds_lov_objids[i], i);
104 int mds_lov_write_objids(struct obd_device *obd)
106 struct mds_obd *mds = &obd->u.mds;
108 int i, rc, size = mds->mds_lov_desc.ld_tgt_count * sizeof(obd_id);
111 for (i = 0; i < mds->mds_lov_desc.ld_tgt_count; i++)
112 CDEBUG(D_INFO, "writing last object "LPU64" for idx %d\n",
113 mds->mds_lov_objids[i], i);
115 rc = fsfilt_write_record(obd, mds->mds_lov_objid_filp,
116 mds->mds_lov_objids, size, &off, 0);
120 static int mds_lov_clearorphans(struct mds_obd *mds, struct obd_uuid *ost_uuid)
124 struct obd_trans_info oti = {0};
125 struct lov_stripe_md *empty_ea = NULL;
128 LASSERT(mds->mds_lov_objids != NULL);
130 /* This create will in fact either create or destroy: If the OST is
131 * missing objects below this ID, they will be created. If it finds
132 * objects above this ID, they will be removed. */
133 memset(&oa, 0, sizeof(oa));
134 oa.o_valid = OBD_MD_FLFLAGS;
135 oa.o_flags = OBD_FL_DELORPHAN;
136 if (ost_uuid != NULL) {
137 memcpy(&oa.o_inline, ost_uuid, sizeof(*ost_uuid));
138 oa.o_valid |= OBD_MD_FLINLINE;
140 rc = obd_create(mds->mds_osc_exp, &oa, &empty_ea, &oti);
145 /* update the LOV-OSC knowledge of the last used object id's */
146 int mds_lov_set_nextid(struct obd_device *obd)
148 struct mds_obd *mds = &obd->u.mds;
152 LASSERT(!obd->obd_recovering);
154 LASSERT(mds->mds_lov_objids != NULL);
156 rc = obd_set_info(mds->mds_osc_exp, strlen("next_id"), "next_id",
157 mds->mds_lov_desc.ld_tgt_count, mds->mds_lov_objids);
161 rc = mds_lov_clearorphans(mds, NULL /* all OSTs */);
166 if (rc && mds->mds_lov_objids) {
167 /* Might as well crash here, until we figure out what to do.
168 * If we OBD_FREE, we'll just LASSERT the next time through this
171 OBD_FREE(mds->mds_lov_objids,
172 mds->mds_lov_desc.ld_tgt_count * sizeof(obd_id));
173 mds->mds_lov_objids = NULL;
179 /* tell the LOV-OSC by how much to pre-create */
180 int mds_lov_set_growth(struct mds_obd *mds, int count)
185 rc = obd_set_info(mds->mds_osc_exp, strlen("growth_count"),
186 "growth_count", sizeof(count), &count);
191 int mds_lov_connect(struct obd_device *obd, char * lov_name)
193 struct mds_obd *mds = &obd->u.mds;
194 struct lustre_handle conn = {0,};
199 if (IS_ERR(mds->mds_osc_obd))
200 RETURN(PTR_ERR(mds->mds_osc_obd));
202 if (mds->mds_osc_obd)
205 mds->mds_osc_obd = class_name2obd(lov_name);
206 if (!mds->mds_osc_obd) {
207 CERROR("MDS cannot locate LOV %s\n",
209 mds->mds_osc_obd = ERR_PTR(-ENOTCONN);
213 rc = obd_connect(&conn, mds->mds_osc_obd, &obd->obd_uuid);
215 CERROR("MDS cannot connect to LOV %s (%d)\n",
217 mds->mds_osc_obd = ERR_PTR(rc);
220 mds->mds_osc_exp = class_conn2export(&conn);
222 rc = obd_register_observer(mds->mds_osc_obd, obd);
224 CERROR("MDS cannot register as observer of LOV %s (%d)\n",
226 GOTO(err_discon, rc);
229 valsize = sizeof(mds->mds_lov_desc);
230 rc = obd_get_info(mds->mds_osc_exp, strlen("lovdesc") + 1, "lovdesc",
231 &valsize, &mds->mds_lov_desc);
235 mds->mds_max_mdsize = lov_mds_md_size(mds->mds_lov_desc.ld_tgt_count);
236 mds->mds_max_cookiesize = mds->mds_lov_desc.ld_tgt_count*
237 sizeof(struct llog_cookie);
238 mds->mds_has_lov_desc = 1;
239 rc = mds_lov_read_objids(obd);
241 CERROR("cannot read %s: rc = %d\n", "lov_objids", rc);
245 #ifdef ENABLE_ORPHANS
246 rc = llog_cat_initialize(obd, mds->mds_lov_desc.ld_tgt_count);
248 CERROR("failed to initialize catalog %d\n", rc);
252 /* FIXME before this set info call is made, we must initialize the logging */
253 rc = obd_set_info(mds->mds_osc_exp, strlen("mds_conn"), "mds_conn",
258 /* If we're mounting this code for the first time on an existing FS,
259 * we need to populate the objids array from the real OST values */
260 if (!mds->mds_lov_objids_valid) {
261 int size = sizeof(obd_id) * mds->mds_lov_desc.ld_tgt_count;
262 rc = obd_get_info(mds->mds_osc_exp, strlen("last_id"),
263 "last_id", &size, mds->mds_lov_objids);
265 for (i = 0; i < mds->mds_lov_desc.ld_tgt_count; i++)
266 CWARN("got last object "LPU64" from OST %d\n",
267 mds->mds_lov_objids[i], i);
268 mds->mds_lov_objids_valid = 1;
269 rc = mds_lov_write_objids(obd);
271 CERROR("got last objids from OSTs, but error "
272 "writing objids file: %d\n", rc);
276 /* I want to see a callback happen when the OBD moves to a
277 * "For General Use" state, and that's when we'll call
278 * set_nextid(). The class driver can help us here, because
279 * it can use the obd_recovering flag to determine when the
280 * the OBD is full available. */
281 if (!obd->obd_recovering) {
282 #ifdef ENABLE_ORPHANS
283 rc = llog_connect(llog_get_context(obd, LLOG_UNLINK_ORIG_CTXT),
284 obd->u.mds.mds_lov_desc.ld_tgt_count, NULL, NULL);
286 CERROR("faild at llog_origin_connect: %d\n", rc);
289 rc = mds_cleanup_orphans(obd);
291 CERROR("Cleanup %d orphans while MDS isn't recovering\n", rc);
293 rc = mds_lov_set_nextid(obd);
300 #ifdef ENABLE_ORPHANS
301 /* cleanup all llogging subsystems */
302 rc = obd_llog_finish(obd, mds->mds_lov_desc.ld_tgt_count);
304 CERROR("failed to cleanup llogging subsystems\n");
307 obd_register_observer(mds->mds_osc_obd, NULL);
309 obd_disconnect(mds->mds_osc_exp, 0);
310 mds->mds_osc_exp = NULL;
311 mds->mds_osc_obd = ERR_PTR(rc);
315 int mds_lov_disconnect(struct obd_device *obd, int flags)
317 struct mds_obd *mds = &obd->u.mds;
321 if (!IS_ERR(mds->mds_osc_obd) && mds->mds_osc_exp != NULL) {
322 #ifdef ENABLE_ORPHANS
323 /* cleanup all llogging subsystems */
324 rc = obd_llog_finish(obd, mds->mds_lov_desc.ld_tgt_count);
326 CERROR("failed to cleanup llogging subsystems\n");
329 obd_register_observer(mds->mds_osc_obd, NULL);
331 rc = obd_disconnect(mds->mds_osc_exp, flags);
332 /* if obd_disconnect fails (probably because the
333 * export was disconnected by class_disconnect_exports)
334 * then we just need to drop our ref. */
336 class_export_put(mds->mds_osc_exp);
337 mds->mds_osc_exp = NULL;
338 mds->mds_osc_obd = NULL;
344 int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
345 void *karg, void *uarg)
347 static struct obd_uuid cfg_uuid = { .uuid = "config_uuid" };
348 struct obd_device *obd = exp->exp_obd;
349 struct mds_obd *mds = &obd->u.mds;
350 struct obd_ioctl_data *data = karg;
351 struct obd_run_ctxt saved;
355 case OBD_IOC_RECORD: {
356 char *name = data->ioc_inlbuf1;
357 if (mds->mds_cfg_llh)
360 push_ctxt(&saved, &obd->obd_ctxt, NULL);
361 rc = llog_create(llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT),
362 &mds->mds_cfg_llh, NULL, name);
364 llog_init_handle(mds->mds_cfg_llh, LLOG_F_IS_PLAIN,
367 mds->mds_cfg_llh = NULL;
368 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
373 case OBD_IOC_ENDRECORD: {
374 if (!mds->mds_cfg_llh)
377 push_ctxt(&saved, &obd->obd_ctxt, NULL);
378 rc = llog_close(mds->mds_cfg_llh);
379 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
381 mds->mds_cfg_llh = NULL;
385 case OBD_IOC_DORECORD: {
387 struct llog_rec_hdr rec;
388 if (!mds->mds_cfg_llh)
391 rec.lrh_len = llog_data_len(data->ioc_plen1);
393 if (data->ioc_type == LUSTRE_CFG_TYPE) {
394 rec.lrh_type = OBD_CFG_REC;
395 } else if (data->ioc_type == PORTALS_CFG_TYPE) {
396 rec.lrh_type = PTL_CFG_REC;
398 CERROR("unknown cfg record type:%d \n", data->ioc_type);
402 OBD_ALLOC(cfg_buf, data->ioc_plen1);
405 rc = copy_from_user(cfg_buf, data->ioc_pbuf1, data->ioc_plen1);
407 OBD_FREE(cfg_buf, data->ioc_plen1);
411 push_ctxt(&saved, &obd->obd_ctxt, NULL);
412 rc = llog_write_rec(mds->mds_cfg_llh, &rec, NULL, 0,
414 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
416 OBD_FREE(cfg_buf, data->ioc_plen1);
420 case OBD_IOC_PARSE: {
421 struct llog_ctxt *ctxt =
422 llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
423 push_ctxt(&saved, &obd->obd_ctxt, NULL);
424 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
425 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
432 case OBD_IOC_DUMP_LOG: {
433 struct llog_ctxt *ctxt =
434 llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
435 push_ctxt(&saved, &obd->obd_ctxt, NULL);
436 rc = class_config_dump_llog(ctxt, data->ioc_inlbuf1, NULL);
437 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
444 case OBD_IOC_SET_READONLY: {
446 struct inode *inode = obd->u.mds.mds_sb->s_root->d_inode;
447 BDEVNAME_DECLARE_STORAGE(tmp);
448 CERROR("setting device %s read-only\n",
449 ll_bdevname(obd->u.mds.mds_sb, tmp));
451 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
453 rc = fsfilt_commit(obd, inode, handle, 1);
455 dev_set_rdonly(ll_sbdev(obd->u.mds.mds_sb), 2);
459 case OBD_IOC_CATLOGLIST: {
460 int count = mds->mds_lov_desc.ld_tgt_count;
461 rc = llog_catlog_list(obd, count, data);
465 case OBD_IOC_LLOG_CHECK:
466 case OBD_IOC_LLOG_CANCEL:
467 case OBD_IOC_LLOG_REMOVE: {
468 struct llog_ctxt *ctxt =
469 llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
471 #ifdef ENABLE_ORPHANS
472 obd_llog_finish(obd, mds->mds_lov_desc.ld_tgt_count);
474 push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
475 rc = llog_ioctl(ctxt, cmd, data);
476 pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
478 #ifdef ENABLE_ORPHANS
479 llog_cat_initialize(obd, mds->mds_lov_desc.ld_tgt_count);
483 case OBD_IOC_LLOG_INFO:
484 case OBD_IOC_LLOG_PRINT: {
485 struct llog_ctxt *ctxt =
486 llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
488 push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
489 rc = llog_ioctl(ctxt, cmd, data);
490 pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
495 case OBD_IOC_ABORT_RECOVERY:
496 CERROR("aborting recovery for device %s\n", obd->obd_name);
497 target_abort_recovery(obd);
506 int mds_notify(struct obd_device *obd, struct obd_device *watched,
509 struct obd_uuid *uuid;
514 if (strcmp(watched->obd_type->typ_name, "osc")) {
515 CERROR("unexpected notification of %s %s!\n",
516 watched->obd_type->typ_name,
521 uuid = &watched->u.cli.cl_import->imp_target_uuid;
522 CWARN("MDS %s: %s now active, resetting orphans\n",
523 obd->obd_name, uuid->uuid);
524 RETURN(mds_lov_clearorphans(&obd->u.mds, uuid));
527 /* Convert the on-disk LOV EA structre.
528 * We always try to convert from an old LOV EA format to the common in-memory
529 * (lsm) format (obd_unpackmd() understands the old on-disk (lmm) format) and
530 * then convert back to the new on-disk format and save it back to disk
531 * (obd_packmd() only ever saves to the new on-disk format) so we don't have
532 * to convert it each time this inode is accessed.
534 * This function is a bit interesting in the error handling. We can safely
535 * ship the old lmm to the client in case of failure, since it uses the same
536 * obd_unpackmd() code and can do the conversion if the MDS fails for some
537 * reason. We will not delete the old lmm data until we have written the
538 * new format lmm data in fsfilt_set_md(). */
539 int mds_convert_lov_ea(struct obd_device *obd, struct inode *inode,
540 struct lov_mds_md *lmm, int lmm_size)
542 struct lov_stripe_md *lsm = NULL;
547 if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC)
550 CWARN("converting LOV EA on %lu/%u from V0 to V1\n",
551 inode->i_ino, inode->i_generation);
552 rc = obd_unpackmd(obd->u.mds.mds_osc_exp, &lsm, lmm, lmm_size);
556 rc = obd_packmd(obd->u.mds.mds_osc_exp, &lmm, lsm);
561 handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
562 if (IS_ERR(handle)) {
563 rc = PTR_ERR(handle);
567 rc = fsfilt_set_md(obd, inode, handle, lmm, lmm_size);
569 err = fsfilt_commit(obd, inode, handle, 0);
571 rc = err ? err : lmm_size;
574 obd_free_memmd(obd->u.mds.mds_osc_exp, &lsm);