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 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_gr = FILTER_GROUP_FIRST_MDS + mds->mds_num;
135 oa.o_valid = OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
136 oa.o_flags = OBD_FL_DELORPHAN;
137 if (ost_uuid != NULL) {
138 memcpy(&oa.o_inline, ost_uuid, sizeof(*ost_uuid));
139 oa.o_valid |= OBD_MD_FLINLINE;
141 rc = obd_create(mds->mds_osc_exp, &oa, &empty_ea, &oti);
146 /* update the LOV-OSC knowledge of the last used object id's */
147 int mds_lov_set_nextid(struct obd_device *obd)
149 struct mds_obd *mds = &obd->u.mds;
153 LASSERT(!obd->obd_recovering);
155 LASSERT(mds->mds_lov_objids != NULL);
157 rc = obd_set_info(mds->mds_osc_exp, strlen("next_id"), "next_id",
158 mds->mds_lov_desc.ld_tgt_count, mds->mds_lov_objids);
162 /* tell the LOV-OSC by how much to pre-create */
163 int mds_lov_set_growth(struct mds_obd *mds, int count)
168 rc = obd_set_info(mds->mds_osc_exp, strlen("growth_count"),
169 "growth_count", sizeof(count), &count);
174 int mds_lov_connect(struct obd_device *obd, char * lov_name)
176 struct mds_obd *mds = &obd->u.mds;
177 struct lustre_handle conn = {0,};
178 char name[32] = "CATLIST";
183 if (IS_ERR(mds->mds_osc_obd))
184 RETURN(PTR_ERR(mds->mds_osc_obd));
186 if (mds->mds_osc_obd)
189 mds->mds_osc_obd = class_name2obd(lov_name);
190 if (!mds->mds_osc_obd) {
191 CERROR("MDS cannot locate LOV %s\n",
193 mds->mds_osc_obd = ERR_PTR(-ENOTCONN);
197 rc = obd_connect(&conn, mds->mds_osc_obd, &obd->obd_uuid);
199 CERROR("MDS cannot connect to LOV %s (%d)\n",
201 mds->mds_osc_obd = ERR_PTR(rc);
204 mds->mds_osc_exp = class_conn2export(&conn);
206 rc = obd_register_observer(mds->mds_osc_obd, obd);
208 CERROR("MDS cannot register as observer of LOV %s (%d)\n",
210 GOTO(err_discon, rc);
213 valsize = sizeof(mds->mds_lov_desc);
214 rc = obd_get_info(mds->mds_osc_exp, strlen("lovdesc") + 1, "lovdesc",
215 &valsize, &mds->mds_lov_desc);
219 i = lov_mds_md_size(mds->mds_lov_desc.ld_tgt_count);
220 if (i > mds->mds_max_mdsize)
221 mds->mds_max_mdsize = i;
222 mds->mds_max_cookiesize = mds->mds_lov_desc.ld_tgt_count*
223 sizeof(struct llog_cookie);
224 mds->mds_has_lov_desc = 1;
225 rc = mds_lov_read_objids(obd);
227 CERROR("cannot read %s: rc = %d\n", "lov_objids", rc);
231 rc = obd_llog_cat_initialize(obd, &obd->obd_llogs,
232 mds->mds_lov_desc.ld_tgt_count, name);
234 CERROR("failed to initialize catalog %d\n", rc);
238 /* FIXME before set info call is made, we must initialize logging */
239 group = FILTER_GROUP_FIRST_MDS + mds->mds_num;
240 valsize = sizeof(group);
241 rc = obd_set_info(mds->mds_osc_exp, strlen("mds_conn"), "mds_conn",
246 /* If we're mounting this code for the first time on an existing FS,
247 * we need to populate the objids array from the real OST values */
248 if (!mds->mds_lov_objids_valid) {
249 int size = sizeof(obd_id) * mds->mds_lov_desc.ld_tgt_count;
250 rc = obd_get_info(mds->mds_osc_exp, strlen("last_id"),
251 "last_id", &size, mds->mds_lov_objids);
253 for (i = 0; i < mds->mds_lov_desc.ld_tgt_count; i++)
254 CWARN("got last object "LPU64" from OST %d\n",
255 mds->mds_lov_objids[i], i);
256 mds->mds_lov_objids_valid = 1;
257 rc = mds_lov_write_objids(obd);
259 CERROR("got last objids from OSTs, but error "
260 "writing objids file: %d\n", rc);
264 /* I want to see a callback happen when the OBD moves to a
265 * "For General Use" state, and that's when we'll call
266 * set_nextid(). The class driver can help us here, because
267 * it can use the obd_recovering flag to determine when the
268 * the OBD is full available. */
269 if (!obd->obd_recovering)
270 rc = mds_postrecov(obd);
274 obd_register_observer(mds->mds_osc_obd, NULL);
276 obd_disconnect(mds->mds_osc_exp, 0);
277 mds->mds_osc_exp = NULL;
278 mds->mds_osc_obd = ERR_PTR(rc);
282 int mds_lov_disconnect(struct obd_device *obd, int flags)
284 struct mds_obd *mds = &obd->u.mds;
288 if (!IS_ERR(mds->mds_osc_obd) && mds->mds_osc_exp != NULL) {
289 /* cleanup all llogging subsystems */
290 rc = obd_llog_finish(obd, &obd->obd_llogs,
291 mds->mds_lov_desc.ld_tgt_count);
293 CERROR("failed to cleanup llogging subsystems\n");
295 obd_register_observer(mds->mds_osc_obd, NULL);
297 rc = obd_disconnect(mds->mds_osc_exp, flags);
298 /* if obd_disconnect fails (probably because the
299 * export was disconnected by class_disconnect_exports)
300 * then we just need to drop our ref. */
302 class_export_put(mds->mds_osc_exp);
303 mds->mds_osc_exp = NULL;
304 mds->mds_osc_obd = NULL;
310 int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
311 void *karg, void *uarg)
313 static struct obd_uuid cfg_uuid = { .uuid = "config_uuid" };
314 struct obd_device *obd = exp->exp_obd;
315 struct mds_obd *mds = &obd->u.mds;
316 struct obd_ioctl_data *data = karg;
317 struct lvfs_run_ctxt saved;
321 case OBD_IOC_RECORD: {
322 char *name = data->ioc_inlbuf1;
323 if (mds->mds_cfg_llh)
326 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
327 rc = llog_create(llog_get_context(&obd->obd_llogs,
328 LLOG_CONFIG_ORIG_CTXT),
329 &mds->mds_cfg_llh, NULL, name);
331 llog_init_handle(mds->mds_cfg_llh, LLOG_F_IS_PLAIN,
334 mds->mds_cfg_llh = NULL;
335 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
340 case OBD_IOC_ENDRECORD: {
341 if (!mds->mds_cfg_llh)
344 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
345 rc = llog_close(mds->mds_cfg_llh);
346 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
348 mds->mds_cfg_llh = NULL;
352 case OBD_IOC_CLEAR_LOG: {
353 char *name = data->ioc_inlbuf1;
354 if (mds->mds_cfg_llh)
357 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
358 rc = llog_create(llog_get_context(&obd->obd_llogs,
359 LLOG_CONFIG_ORIG_CTXT),
360 &mds->mds_cfg_llh, NULL, name);
362 llog_init_handle(mds->mds_cfg_llh, LLOG_F_IS_PLAIN,
365 rc = llog_destroy(mds->mds_cfg_llh);
366 llog_free_handle(mds->mds_cfg_llh);
368 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
370 mds->mds_cfg_llh = NULL;
374 case OBD_IOC_DORECORD: {
376 struct llog_rec_hdr rec;
377 if (!mds->mds_cfg_llh)
380 rec.lrh_len = llog_data_len(data->ioc_plen1);
382 if (data->ioc_type == LUSTRE_CFG_TYPE) {
383 rec.lrh_type = OBD_CFG_REC;
384 } else if (data->ioc_type == PORTALS_CFG_TYPE) {
385 rec.lrh_type = PTL_CFG_REC;
387 CERROR("unknown cfg record type:%d \n", data->ioc_type);
391 OBD_ALLOC(cfg_buf, data->ioc_plen1);
394 rc = copy_from_user(cfg_buf, data->ioc_pbuf1, data->ioc_plen1);
396 OBD_FREE(cfg_buf, data->ioc_plen1);
400 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
401 rc = llog_write_rec(mds->mds_cfg_llh, &rec, NULL, 0,
403 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
405 OBD_FREE(cfg_buf, data->ioc_plen1);
409 case OBD_IOC_PARSE: {
410 struct llog_ctxt *ctxt =
411 llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT);
412 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
413 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
414 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
421 case OBD_IOC_DUMP_LOG: {
422 struct llog_ctxt *ctxt =
423 llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT);
424 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
425 rc = class_config_dump_llog(ctxt, data->ioc_inlbuf1, NULL);
426 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
433 case OBD_IOC_SET_READONLY: {
435 struct inode *inode = obd->u.mds.mds_sb->s_root->d_inode;
436 BDEVNAME_DECLARE_STORAGE(tmp);
437 CERROR("setting device %s read-only\n",
438 ll_bdevname(obd->u.mds.mds_sb, tmp));
440 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
442 rc = fsfilt_commit(obd, obd->u.mds.mds_sb, inode, handle, 1);
444 dev_set_rdonly(ll_sbdev(obd->u.mds.mds_sb), 2);
448 case OBD_IOC_CATLOGLIST: {
449 int count = mds->mds_lov_desc.ld_tgt_count;
450 rc = llog_catalog_list(obd, count, data);
454 case OBD_IOC_LLOG_CHECK:
455 case OBD_IOC_LLOG_CANCEL:
456 case OBD_IOC_LLOG_REMOVE: {
457 struct llog_ctxt *ctxt =
458 llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT);
459 char name[32] = "CATLIST";
462 obd_llog_finish(obd, &obd->obd_llogs,
463 mds->mds_lov_desc.ld_tgt_count);
464 push_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL);
465 rc = llog_ioctl(ctxt, cmd, data);
466 pop_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL);
467 obd_llog_cat_initialize(obd, &obd->obd_llogs,
468 mds->mds_lov_desc.ld_tgt_count,
470 rc2 = obd_set_info(mds->mds_osc_exp, strlen("mds_conn"),
471 "mds_conn", 0, NULL);
476 case OBD_IOC_LLOG_INFO:
477 case OBD_IOC_LLOG_PRINT: {
478 struct llog_ctxt *ctxt =
479 llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT);
481 push_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL);
482 rc = llog_ioctl(ctxt, cmd, data);
483 pop_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL);
488 case OBD_IOC_ABORT_RECOVERY:
489 CERROR("aborting recovery for device %s\n", obd->obd_name);
490 target_abort_recovery(obd);
500 struct mds_lov_sync_info {
501 struct obd_device *mlsi_obd; /* the lov device to sync */
502 struct obd_uuid *mlsi_uuid; /* target to sync */
505 int mds_lov_synchronize(void *data)
507 struct mds_lov_sync_info *mlsi = data;
508 struct llog_ctxt *ctxt;
509 struct obd_device *obd;
510 struct obd_uuid *uuid;
517 SIGNAL_MASK_LOCK(current, flags);
518 sigfillset(¤t->blocked);
520 SIGNAL_MASK_UNLOCK(current, flags);
522 obd = mlsi->mlsi_obd;
523 uuid = mlsi->mlsi_uuid;
525 OBD_FREE(mlsi, sizeof(*mlsi));
527 LASSERT(obd != NULL);
528 LASSERT(uuid != NULL);
530 rc = obd_set_info(obd->u.mds.mds_osc_exp, strlen("mds_conn"),
531 "mds_conn", 0, uuid);
535 ctxt = llog_get_context(&obd->obd_llogs, LLOG_UNLINK_ORIG_CTXT);
536 LASSERT(ctxt != NULL);
538 rc = llog_connect(ctxt, obd->u.mds.mds_lov_desc.ld_tgt_count,
541 CERROR("%s: failed at llog_origin_connect: %d\n",
546 CWARN("MDS %s: %s now active, resetting orphans\n",
547 obd->obd_name, uuid->uuid);
548 rc = mds_lov_clearorphans(&obd->u.mds, uuid);
550 CERROR("%s: failed at mds_lov_clearorphans: %d\n",
558 int mds_lov_start_synchronize(struct obd_device *obd, struct obd_uuid *uuid)
560 struct mds_lov_sync_info *mlsi;
565 OBD_ALLOC(mlsi, sizeof(*mlsi));
569 mlsi->mlsi_obd = obd;
570 mlsi->mlsi_uuid = uuid;
572 rc = kernel_thread(mds_lov_synchronize, mlsi, CLONE_VM | CLONE_FILES);
574 CERROR("%s: error starting mds_lov_synchronize: %d\n",
577 CDEBUG(D_HA, "%s: mds_lov_synchronize thread: %d\n",
585 int mds_notify(struct obd_device *obd, struct obd_device *watched, int active)
587 struct obd_uuid *uuid;
594 if (strcmp(watched->obd_type->typ_name, "osc")) {
595 CERROR("unexpected notification of %s %s!\n",
596 watched->obd_type->typ_name, watched->obd_name);
600 uuid = &watched->u.cli.cl_import->imp_target_uuid;
601 if (obd->obd_recovering) {
602 CWARN("MDS %s: in recovery, not resetting orphans on %s\n",
603 obd->obd_name, uuid->uuid);
605 rc = mds_lov_start_synchronize(obd, uuid);
610 /* Convert the on-disk LOV EA structre.
611 * We always try to convert from an old LOV EA format to the common in-memory
612 * (lsm) format (obd_unpackmd() understands the old on-disk (lmm) format) and
613 * then convert back to the new on-disk format and save it back to disk
614 * (obd_packmd() only ever saves to the new on-disk format) so we don't have
615 * to convert it each time this inode is accessed.
617 * This function is a bit interesting in the error handling. We can safely
618 * ship the old lmm to the client in case of failure, since it uses the same
619 * obd_unpackmd() code and can do the conversion if the MDS fails for some
620 * reason. We will not delete the old lmm data until we have written the
621 * new format lmm data in fsfilt_set_md(). */
622 int mds_convert_lov_ea(struct obd_device *obd, struct inode *inode,
623 struct lov_mds_md *lmm, int lmm_size)
625 struct lov_stripe_md *lsm = NULL;
630 if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC)
633 CWARN("converting LOV EA on %lu/%u from V0 to V1\n",
634 inode->i_ino, inode->i_generation);
635 rc = obd_unpackmd(obd->u.mds.mds_osc_exp, &lsm, lmm, lmm_size);
639 rc = obd_packmd(obd->u.mds.mds_osc_exp, &lmm, lsm);
644 handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
645 if (IS_ERR(handle)) {
646 rc = PTR_ERR(handle);
650 rc = fsfilt_set_md(obd, inode, handle, lmm, lmm_size);
652 err = fsfilt_commit(obd, obd->u.mds.mds_sb, inode, handle, 0);
654 rc = err ? err : lmm_size;
657 obd_free_memmd(obd->u.mds.mds_osc_exp, &lsm);