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 */);
167 /* tell the LOV-OSC by how much to pre-create */
168 int mds_lov_set_growth(struct mds_obd *mds, int count)
173 rc = obd_set_info(mds->mds_osc_exp, strlen("growth_count"),
174 "growth_count", sizeof(count), &count);
179 int mds_lov_connect(struct obd_device *obd, char * lov_name)
181 struct mds_obd *mds = &obd->u.mds;
182 struct lustre_handle conn = {0,};
187 if (IS_ERR(mds->mds_osc_obd))
188 RETURN(PTR_ERR(mds->mds_osc_obd));
190 if (mds->mds_osc_obd)
193 mds->mds_osc_obd = class_name2obd(lov_name);
194 if (!mds->mds_osc_obd) {
195 CERROR("MDS cannot locate LOV %s\n",
197 mds->mds_osc_obd = ERR_PTR(-ENOTCONN);
201 rc = obd_connect(&conn, mds->mds_osc_obd, &obd->obd_uuid);
203 CERROR("MDS cannot connect to LOV %s (%d)\n",
205 mds->mds_osc_obd = ERR_PTR(rc);
208 mds->mds_osc_exp = class_conn2export(&conn);
210 rc = obd_register_observer(mds->mds_osc_obd, obd);
212 CERROR("MDS cannot register as observer of LOV %s (%d)\n",
214 GOTO(err_discon, rc);
217 valsize = sizeof(mds->mds_lov_desc);
218 rc = obd_get_info(mds->mds_osc_exp, strlen("lovdesc") + 1, "lovdesc",
219 &valsize, &mds->mds_lov_desc);
223 mds->mds_max_mdsize = lov_mds_md_size(mds->mds_lov_desc.ld_tgt_count);
224 mds->mds_max_cookiesize = mds->mds_lov_desc.ld_tgt_count*
225 sizeof(struct llog_cookie);
226 mds->mds_has_lov_desc = 1;
227 rc = mds_lov_read_objids(obd);
229 CERROR("cannot read %s: rc = %d\n", "lov_objids", rc);
233 rc = llog_cat_initialize(obd, mds->mds_lov_desc.ld_tgt_count);
235 CERROR("failed to initialize catalog %d\n", rc);
239 /* FIXME before set info call is made, we must initialize logging */
240 rc = obd_set_info(mds->mds_osc_exp, strlen("mds_conn"), "mds_conn",
245 /* If we're mounting this code for the first time on an existing FS,
246 * we need to populate the objids array from the real OST values */
247 if (!mds->mds_lov_objids_valid) {
248 int size = sizeof(obd_id) * mds->mds_lov_desc.ld_tgt_count;
249 rc = obd_get_info(mds->mds_osc_exp, strlen("last_id"),
250 "last_id", &size, mds->mds_lov_objids);
252 for (i = 0; i < mds->mds_lov_desc.ld_tgt_count; i++)
253 CWARN("got last object "LPU64" from OST %d\n",
254 mds->mds_lov_objids[i], i);
255 mds->mds_lov_objids_valid = 1;
256 rc = mds_lov_write_objids(obd);
258 CERROR("got last objids from OSTs, but error "
259 "writing objids file: %d\n", rc);
263 /* I want to see a callback happen when the OBD moves to a
264 * "For General Use" state, and that's when we'll call
265 * set_nextid(). The class driver can help us here, because
266 * it can use the obd_recovering flag to determine when the
267 * the OBD is full available. */
268 if (!obd->obd_recovering) {
269 rc = llog_connect(llog_get_context(obd, LLOG_UNLINK_ORIG_CTXT),
270 obd->u.mds.mds_lov_desc.ld_tgt_count, NULL,
273 CERROR("faild at llog_origin_connect: %d\n", rc);
275 rc = mds_cleanup_orphans(obd);
277 CERROR("Cleanup %d orphans while MDS isn't recovering\n", rc);
279 rc = mds_lov_set_nextid(obd);
286 /* cleanup all llogging subsystems */
287 rc = obd_llog_finish(obd, mds->mds_lov_desc.ld_tgt_count);
289 CERROR("failed to cleanup llogging subsystems\n");
291 obd_register_observer(mds->mds_osc_obd, NULL);
293 obd_disconnect(mds->mds_osc_exp, 0);
294 mds->mds_osc_exp = NULL;
295 mds->mds_osc_obd = ERR_PTR(rc);
299 int mds_lov_disconnect(struct obd_device *obd, int flags)
301 struct mds_obd *mds = &obd->u.mds;
305 if (!IS_ERR(mds->mds_osc_obd) && mds->mds_osc_exp != NULL) {
306 /* cleanup all llogging subsystems */
307 rc = obd_llog_finish(obd, mds->mds_lov_desc.ld_tgt_count);
309 CERROR("failed to cleanup llogging subsystems\n");
311 obd_register_observer(mds->mds_osc_obd, NULL);
313 rc = obd_disconnect(mds->mds_osc_exp, flags);
314 /* if obd_disconnect fails (probably because the
315 * export was disconnected by class_disconnect_exports)
316 * then we just need to drop our ref. */
318 class_export_put(mds->mds_osc_exp);
319 mds->mds_osc_exp = NULL;
320 mds->mds_osc_obd = NULL;
326 int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
327 void *karg, void *uarg)
329 static struct obd_uuid cfg_uuid = { .uuid = "config_uuid" };
330 struct obd_device *obd = exp->exp_obd;
331 struct mds_obd *mds = &obd->u.mds;
332 struct obd_ioctl_data *data = karg;
333 struct obd_run_ctxt saved;
337 case OBD_IOC_RECORD: {
338 char *name = data->ioc_inlbuf1;
339 if (mds->mds_cfg_llh)
342 push_ctxt(&saved, &obd->obd_ctxt, NULL);
343 rc = llog_create(llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT),
344 &mds->mds_cfg_llh, NULL, name);
346 llog_init_handle(mds->mds_cfg_llh, LLOG_F_IS_PLAIN,
349 mds->mds_cfg_llh = NULL;
350 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
355 case OBD_IOC_ENDRECORD: {
356 if (!mds->mds_cfg_llh)
359 push_ctxt(&saved, &obd->obd_ctxt, NULL);
360 rc = llog_close(mds->mds_cfg_llh);
361 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
363 mds->mds_cfg_llh = NULL;
367 case OBD_IOC_CLEAR_LOG: {
368 char *name = data->ioc_inlbuf1;
369 if (mds->mds_cfg_llh)
372 push_ctxt(&saved, &obd->obd_ctxt, NULL);
373 rc = llog_create(llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT),
374 &mds->mds_cfg_llh, NULL, name);
376 llog_init_handle(mds->mds_cfg_llh, LLOG_F_IS_PLAIN,
379 rc = llog_destroy(mds->mds_cfg_llh);
380 llog_free_handle(mds->mds_cfg_llh);
382 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
384 mds->mds_cfg_llh = NULL;
388 case OBD_IOC_DORECORD: {
390 struct llog_rec_hdr rec;
391 if (!mds->mds_cfg_llh)
394 rec.lrh_len = llog_data_len(data->ioc_plen1);
396 if (data->ioc_type == LUSTRE_CFG_TYPE) {
397 rec.lrh_type = OBD_CFG_REC;
398 } else if (data->ioc_type == PORTALS_CFG_TYPE) {
399 rec.lrh_type = PTL_CFG_REC;
401 CERROR("unknown cfg record type:%d \n", data->ioc_type);
405 OBD_ALLOC(cfg_buf, data->ioc_plen1);
408 rc = copy_from_user(cfg_buf, data->ioc_pbuf1, data->ioc_plen1);
410 OBD_FREE(cfg_buf, data->ioc_plen1);
414 push_ctxt(&saved, &obd->obd_ctxt, NULL);
415 rc = llog_write_rec(mds->mds_cfg_llh, &rec, NULL, 0,
417 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
419 OBD_FREE(cfg_buf, data->ioc_plen1);
423 case OBD_IOC_PARSE: {
424 struct llog_ctxt *ctxt =
425 llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
426 push_ctxt(&saved, &obd->obd_ctxt, NULL);
427 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
428 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
435 case OBD_IOC_DUMP_LOG: {
436 struct llog_ctxt *ctxt =
437 llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
438 push_ctxt(&saved, &obd->obd_ctxt, NULL);
439 rc = class_config_dump_llog(ctxt, data->ioc_inlbuf1, NULL);
440 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
447 case OBD_IOC_SET_READONLY: {
449 struct inode *inode = obd->u.mds.mds_sb->s_root->d_inode;
450 BDEVNAME_DECLARE_STORAGE(tmp);
451 CERROR("setting device %s read-only\n",
452 ll_bdevname(obd->u.mds.mds_sb, tmp));
454 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
456 rc = fsfilt_commit(obd, inode, handle, 1);
458 dev_set_rdonly(ll_sbdev(obd->u.mds.mds_sb), 2);
462 case OBD_IOC_CATLOGLIST: {
463 int count = mds->mds_lov_desc.ld_tgt_count;
464 rc = llog_catlog_list(obd, count, data);
468 case OBD_IOC_LLOG_CHECK:
469 case OBD_IOC_LLOG_CANCEL:
470 case OBD_IOC_LLOG_REMOVE: {
471 struct llog_ctxt *ctxt =
472 llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
475 obd_llog_finish(obd, mds->mds_lov_desc.ld_tgt_count);
476 push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
477 rc = llog_ioctl(ctxt, cmd, data);
478 pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
479 llog_cat_initialize(obd, mds->mds_lov_desc.ld_tgt_count);
480 rc2 = obd_set_info(mds->mds_osc_exp, strlen("mds_conn"), "mds_conn",
486 case OBD_IOC_LLOG_INFO:
487 case OBD_IOC_LLOG_PRINT: {
488 struct llog_ctxt *ctxt =
489 llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
491 push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
492 rc = llog_ioctl(ctxt, cmd, data);
493 pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
498 case OBD_IOC_ABORT_RECOVERY:
499 CERROR("aborting recovery for device %s\n", obd->obd_name);
500 target_abort_recovery(obd);
509 int mds_notify(struct obd_device *obd, struct obd_device *watched, int active)
511 struct obd_uuid *uuid;
518 if (strcmp(watched->obd_type->typ_name, "osc")) {
519 CERROR("unexpected notification of %s %s!\n",
520 watched->obd_type->typ_name, watched->obd_name);
524 uuid = &watched->u.cli.cl_import->imp_target_uuid;
525 if (obd->obd_recovering) {
526 CWARN("MDS %s: in recovery, not resetting orphans on %s\n",
527 obd->obd_name, uuid->uuid);
529 LASSERT(llog_get_context(obd, LLOG_UNLINK_ORIG_CTXT) != NULL);
531 rc = obd_set_info(obd->u.mds.mds_osc_exp, strlen("mds_conn"), "mds_conn",
536 rc = llog_connect(llog_get_context(obd, LLOG_UNLINK_ORIG_CTXT),
537 obd->u.mds.mds_lov_desc.ld_tgt_count,
540 CERROR("faild at llog_origin_connect: %d\n", rc);
544 CWARN("MDS %s: %s now active, resetting orphans\n",
545 obd->obd_name, uuid->uuid);
546 rc = mds_lov_clearorphans(&obd->u.mds, uuid);
551 /* Convert the on-disk LOV EA structre.
552 * We always try to convert from an old LOV EA format to the common in-memory
553 * (lsm) format (obd_unpackmd() understands the old on-disk (lmm) format) and
554 * then convert back to the new on-disk format and save it back to disk
555 * (obd_packmd() only ever saves to the new on-disk format) so we don't have
556 * to convert it each time this inode is accessed.
558 * This function is a bit interesting in the error handling. We can safely
559 * ship the old lmm to the client in case of failure, since it uses the same
560 * obd_unpackmd() code and can do the conversion if the MDS fails for some
561 * reason. We will not delete the old lmm data until we have written the
562 * new format lmm data in fsfilt_set_md(). */
563 int mds_convert_lov_ea(struct obd_device *obd, struct inode *inode,
564 struct lov_mds_md *lmm, int lmm_size)
566 struct lov_stripe_md *lsm = NULL;
571 if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC)
574 CWARN("converting LOV EA on %lu/%u from V0 to V1\n",
575 inode->i_ino, inode->i_generation);
576 rc = obd_unpackmd(obd->u.mds.mds_osc_exp, &lsm, lmm, lmm_size);
580 rc = obd_packmd(obd->u.mds.mds_osc_exp, &lmm, lsm);
585 handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
586 if (IS_ERR(handle)) {
587 rc = PTR_ERR(handle);
591 rc = fsfilt_set_md(obd, inode, handle, lmm, lmm_size);
593 err = fsfilt_commit(obd, inode, handle, 0);
595 rc = err ? err : lmm_size;
598 obd_free_memmd(obd->u.mds.mds_osc_exp, &lsm);