1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * linux/fs/obdfilter/filter.c
6 * Copyright (c) 2001-2003 Cluster File Systems, Inc.
7 * Author: Peter Braam <braam@clusterfs.com>
8 * Author: Andreas Dilger <adilger@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 * Invariant: Get O/R i_sem for lookup, if needed, before any journal ops
28 * (which need to get journal_lock, may block if journal full).
30 * Invariant: Call filter_start_transno() before any journal ops to avoid the
31 * same deadlock problem. We can (and want) to get rid of the
32 * transno sem in favour of the dir/inode i_sem to avoid single
33 * threaded operation on the OST.
36 #define DEBUG_SUBSYSTEM S_FILTER
38 #include <linux/config.h>
39 #include <linux/module.h>
41 #include <linux/dcache.h>
42 #include <linux/init.h>
43 #include <linux/version.h>
44 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
45 # include <linux/mount.h>
46 # include <linux/buffer_head.h>
49 #include <linux/obd_class.h>
50 #include <linux/obd_lov.h>
51 #include <linux/lustre_dlm.h>
52 #include <linux/lustre_fsfilt.h>
53 #include <linux/lprocfs_status.h>
54 #include <linux/lustre_log.h>
55 #include <linux/lustre_commit_confd.h>
56 #include <portals/list.h>
58 #include "filter_internal.h"
60 static struct lvfs_callback_ops filter_lvfs_ops;
62 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
63 struct lov_stripe_md *ea, struct obd_trans_info *);
65 static void filter_commit_cb(struct obd_device *obd, __u64 transno,
66 void *cb_data, int error)
68 obd_transno_commit_cb(obd, transno, error);
71 /* Assumes caller has already pushed us into the kernel context. */
72 int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti,
75 struct filter_obd *filter = &exp->exp_obd->u.filter;
76 struct filter_export_data *fed = &exp->exp_filter_data;
77 struct filter_client_data *fcd = fed->fed_fcd;
80 int err, log_pri = D_HA;
82 /* Propagate error code. */
86 if (!exp->exp_obd->obd_replayable || oti == NULL)
89 /* we don't allocate new transnos for replayed requests */
90 if (oti->oti_transno == 0) {
91 spin_lock(&filter->fo_translock);
92 last_rcvd = le64_to_cpu(filter->fo_fsd->fsd_last_transno) + 1;
93 filter->fo_fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
94 spin_unlock(&filter->fo_translock);
95 oti->oti_transno = last_rcvd;
97 spin_lock(&filter->fo_translock);
98 last_rcvd = oti->oti_transno;
99 if (last_rcvd > le64_to_cpu(filter->fo_fsd->fsd_last_transno))
100 filter->fo_fsd->fsd_last_transno =
101 cpu_to_le64(last_rcvd);
102 spin_unlock(&filter->fo_translock);
104 fcd->fcd_last_rcvd = cpu_to_le64(last_rcvd);
106 /* could get xid from oti, if it's ever needed */
107 fcd->fcd_last_xid = 0;
109 off = fed->fed_lr_off;
111 fsfilt_add_journal_cb(exp->exp_obd, filter->fo_sb, last_rcvd,
112 oti->oti_handle, filter_commit_cb, NULL);
114 err = fsfilt_write_record(exp->exp_obd, filter->fo_rcvd_filp, fcd,
115 sizeof(*fcd), &off, 0);
122 CDEBUG(log_pri, "wrote trans "LPU64" for client %s at #%d: err = %d\n",
123 last_rcvd, fcd->fcd_uuid, fed->fed_lr_idx, err);
128 void f_dput(struct dentry *dentry)
130 /* Can't go inside filter_ddelete because it can block */
131 CDEBUG(D_INODE, "putting %s: %p, count = %d\n",
132 dentry->d_name.name, dentry, atomic_read(&dentry->d_count) - 1);
133 LASSERT(atomic_read(&dentry->d_count) > 0);
138 /* Add client data to the FILTER. We use a bitmap to locate a free space
139 * in the last_rcvd file if cl_idx is -1 (i.e. a new client).
140 * Otherwise, we have just read the data from the last_rcvd file and
141 * we know its offset. */
142 static int filter_client_add(struct obd_device *obd, struct filter_obd *filter,
143 struct filter_export_data *fed, int cl_idx)
145 unsigned long *bitmap = filter->fo_last_rcvd_slots;
146 int new_client = (cl_idx == -1);
149 LASSERT(bitmap != NULL);
151 /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
152 if (!strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid))
155 /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
156 * there's no need for extra complication here
159 cl_idx = find_first_zero_bit(bitmap, FILTER_LR_MAX_CLIENTS);
161 if (cl_idx >= FILTER_LR_MAX_CLIENTS) {
162 CERROR("no client slots - fix FILTER_LR_MAX_CLIENTS\n");
165 if (test_and_set_bit(cl_idx, bitmap)) {
166 CERROR("FILTER client %d: found bit is set in bitmap\n",
168 cl_idx = find_next_zero_bit(bitmap,
169 FILTER_LR_MAX_CLIENTS,
174 if (test_and_set_bit(cl_idx, bitmap)) {
175 CERROR("FILTER client %d: bit already set in bitmap!\n",
181 fed->fed_lr_idx = cl_idx;
182 fed->fed_lr_off = le32_to_cpu(filter->fo_fsd->fsd_client_start) +
183 cl_idx * le16_to_cpu(filter->fo_fsd->fsd_client_size);
185 CDEBUG(D_INFO, "client at index %d (%llu) with UUID '%s' added\n",
186 fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
189 struct lvfs_run_ctxt saved;
190 loff_t off = fed->fed_lr_off;
194 CDEBUG(D_INFO, "writing client fcd at idx %u (%llu) (len %u)\n",
195 fed->fed_lr_idx,off,(unsigned int)sizeof(*fed->fed_fcd));
197 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
198 /* Transaction needed to fix bug 1403 */
199 handle = fsfilt_start(obd,
200 filter->fo_rcvd_filp->f_dentry->d_inode,
201 FSFILT_OP_SETATTR, NULL);
202 if (IS_ERR(handle)) {
203 err = PTR_ERR(handle);
204 CERROR("unable to start transaction: rc %d\n", err);
206 err = fsfilt_write_record(obd, filter->fo_rcvd_filp,
208 sizeof(*fed->fed_fcd),
210 fsfilt_commit(obd, filter->fo_sb,
211 filter->fo_rcvd_filp->f_dentry->d_inode,
214 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
217 CERROR("error writing %s client idx %u: rc %d\n",
218 LAST_RCVD, fed->fed_lr_idx, err);
225 static int filter_client_free(struct obd_export *exp, int flags)
227 struct filter_export_data *fed = &exp->exp_filter_data;
228 struct filter_obd *filter = &exp->exp_obd->u.filter;
229 struct obd_device *obd = exp->exp_obd;
230 struct filter_client_data zero_fcd;
231 struct lvfs_run_ctxt saved;
236 if (fed->fed_fcd == NULL)
239 if (flags & OBD_OPT_FAILOVER)
242 /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
243 if (strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid ) == 0)
246 LASSERT(filter->fo_last_rcvd_slots != NULL);
248 off = fed->fed_lr_off;
250 CDEBUG(D_INFO, "freeing client at idx %u (%lld) with UUID '%s'\n",
251 fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
253 /* Clear the bit _after_ zeroing out the client so we don't
254 race with filter_client_add and zero out new clients.*/
255 if (!test_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
256 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
261 memset(&zero_fcd, 0, sizeof zero_fcd);
262 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
263 rc = fsfilt_write_record(obd, filter->fo_rcvd_filp, &zero_fcd,
264 sizeof(zero_fcd), &off, 1);
265 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
267 CDEBUG(rc == 0 ? D_INFO : D_ERROR,
268 "zeroing disconnecting client %s at idx %u (%llu) in %s rc %d\n",
269 fed->fed_fcd->fcd_uuid, fed->fed_lr_idx, fed->fed_lr_off,
272 if (!test_and_clear_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
273 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
279 OBD_FREE(fed->fed_fcd, sizeof(*fed->fed_fcd));
284 static int filter_free_server_data(struct filter_obd *filter)
286 OBD_FREE(filter->fo_fsd, sizeof(*filter->fo_fsd));
287 filter->fo_fsd = NULL;
288 OBD_FREE(filter->fo_last_rcvd_slots,
289 FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
290 filter->fo_last_rcvd_slots = NULL;
294 /* assumes caller is already in kernel ctxt */
295 int filter_update_server_data(struct obd_device *obd, struct file *filp,
296 struct filter_server_data *fsd, int force_sync)
302 CDEBUG(D_INODE, "server uuid : %s\n", fsd->fsd_uuid);
303 CDEBUG(D_INODE, "server last_rcvd : "LPU64"\n",
304 le64_to_cpu(fsd->fsd_last_transno));
305 CDEBUG(D_INODE, "server last_mount: "LPU64"\n",
306 le64_to_cpu(fsd->fsd_mount_count));
308 rc = fsfilt_write_record(obd, filp, fsd, sizeof(*fsd), &off,force_sync);
310 CERROR("error writing filter_server_data: rc = %d\n", rc);
315 int filter_update_last_objid(struct obd_device *obd, obd_gr group,
318 struct filter_obd *filter = &obd->u.filter;
324 if (filter->fo_last_objid_files[group] == NULL) {
325 CERROR("Object group "LPU64" not fully setup; not updating "
326 "last_objid\n", group);
330 CDEBUG(D_INODE, "server last_objid for group "LPU64": "LPU64"\n",
331 group, filter->fo_last_objids[group]);
333 tmp = cpu_to_le64(filter->fo_last_objids[group]);
334 rc = fsfilt_write_record(obd, filter->fo_last_objid_files[group],
335 &tmp, sizeof(tmp), &off, force_sync);
337 CERROR("error writing group "LPU64" last objid: rc = %d\n",
342 /* assumes caller has already in kernel ctxt */
343 static int filter_init_server_data(struct obd_device *obd, struct file * filp)
345 struct filter_obd *filter = &obd->u.filter;
346 struct filter_server_data *fsd;
347 struct filter_client_data *fcd = NULL;
348 struct inode *inode = filp->f_dentry->d_inode;
349 unsigned long last_rcvd_size = inode->i_size;
355 /* ensure padding in the struct is the correct size */
356 LASSERT (offsetof(struct filter_server_data, fsd_padding) +
357 sizeof(fsd->fsd_padding) == FILTER_LR_SERVER_SIZE);
358 LASSERT (offsetof(struct filter_client_data, fcd_padding) +
359 sizeof(fcd->fcd_padding) == FILTER_LR_CLIENT_SIZE);
361 OBD_ALLOC(fsd, sizeof(*fsd));
364 filter->fo_fsd = fsd;
366 OBD_ALLOC(filter->fo_last_rcvd_slots,
367 FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
368 if (filter->fo_last_rcvd_slots == NULL) {
369 OBD_FREE(fsd, sizeof(*fsd));
373 if (last_rcvd_size == 0) {
374 CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD);
376 memcpy(fsd->fsd_uuid, obd->obd_uuid.uuid,sizeof(fsd->fsd_uuid));
377 fsd->fsd_last_transno = 0;
378 mount_count = fsd->fsd_mount_count = 0;
379 fsd->fsd_server_size = cpu_to_le32(FILTER_LR_SERVER_SIZE);
380 fsd->fsd_client_start = cpu_to_le32(FILTER_LR_CLIENT_START);
381 fsd->fsd_client_size = cpu_to_le16(FILTER_LR_CLIENT_SIZE);
382 fsd->fsd_subdir_count = cpu_to_le16(FILTER_SUBDIR_COUNT);
383 filter->fo_subdir_count = FILTER_SUBDIR_COUNT;
385 rc = fsfilt_read_record(obd, filp, fsd, sizeof(*fsd), &off);
387 CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
391 if (strcmp(fsd->fsd_uuid, obd->obd_uuid.uuid) != 0) {
392 CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
393 obd->obd_uuid.uuid, fsd->fsd_uuid);
394 GOTO(err_fsd, rc = -EINVAL);
396 mount_count = le64_to_cpu(fsd->fsd_mount_count);
397 filter->fo_subdir_count = le16_to_cpu(fsd->fsd_subdir_count);
400 if (fsd->fsd_feature_incompat & ~cpu_to_le32(FILTER_INCOMPAT_SUPP)) {
401 CERROR("unsupported feature %x\n",
402 le32_to_cpu(fsd->fsd_feature_incompat) &
403 ~FILTER_INCOMPAT_SUPP);
404 GOTO(err_fsd, rc = -EINVAL);
406 if (fsd->fsd_feature_rocompat & ~cpu_to_le32(FILTER_ROCOMPAT_SUPP)) {
407 CERROR("read-only feature %x\n",
408 le32_to_cpu(fsd->fsd_feature_rocompat) &
409 ~FILTER_ROCOMPAT_SUPP);
410 /* Do something like remount filesystem read-only */
411 GOTO(err_fsd, rc = -EINVAL);
414 CDEBUG(D_INODE, "%s: server last_rcvd : "LPU64"\n",
415 obd->obd_name, le64_to_cpu(fsd->fsd_last_transno));
416 CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
417 obd->obd_name, mount_count + 1);
418 CDEBUG(D_INODE, "%s: server data size: %u\n",
419 obd->obd_name, le32_to_cpu(fsd->fsd_server_size));
420 CDEBUG(D_INODE, "%s: per-client data start: %u\n",
421 obd->obd_name, le32_to_cpu(fsd->fsd_client_start));
422 CDEBUG(D_INODE, "%s: per-client data size: %u\n",
423 obd->obd_name, le32_to_cpu(fsd->fsd_client_size));
424 CDEBUG(D_INODE, "%s: server subdir_count: %u\n",
425 obd->obd_name, le16_to_cpu(fsd->fsd_subdir_count));
426 CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
427 last_rcvd_size <= le32_to_cpu(fsd->fsd_client_start) ? 0 :
428 (last_rcvd_size - le32_to_cpu(fsd->fsd_client_start)) /
429 le16_to_cpu(fsd->fsd_client_size));
431 if (!obd->obd_replayable) {
432 CWARN("%s: recovery support OFF\n", obd->obd_name);
436 for (cl_idx = 0, off = le32_to_cpu(fsd->fsd_client_start);
437 off < last_rcvd_size; cl_idx++) {
439 struct obd_export *exp;
440 struct filter_export_data *fed;
443 OBD_ALLOC(fcd, sizeof(*fcd));
445 GOTO(err_client, rc = -ENOMEM);
448 /* Don't assume off is incremented properly by
449 * fsfilt_read_record(), in case sizeof(*fcd)
450 * isn't the same as fsd->fsd_client_size. */
451 off = le32_to_cpu(fsd->fsd_client_start) +
452 cl_idx * le16_to_cpu(fsd->fsd_client_size);
453 rc = fsfilt_read_record(obd, filp, fcd, sizeof(*fcd), &off);
455 CERROR("error reading FILT %s idx %d off %llu: rc %d\n",
456 LAST_RCVD, cl_idx, off, rc);
457 break; /* read error shouldn't cause startup to fail */
460 if (fcd->fcd_uuid[0] == '\0') {
461 CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
466 last_rcvd = le64_to_cpu(fcd->fcd_last_rcvd);
468 /* These exports are cleaned up by filter_disconnect(), so they
469 * need to be set up like real exports as filter_connect() does.
471 exp = class_new_export(obd);
472 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
473 " srv lr: "LPU64"\n", fcd->fcd_uuid, cl_idx,
474 last_rcvd, le64_to_cpu(fsd->fsd_last_transno));
476 GOTO(err_client, rc = -ENOMEM);
478 memcpy(&exp->exp_client_uuid.uuid, fcd->fcd_uuid,
479 sizeof exp->exp_client_uuid.uuid);
480 fed = &exp->exp_filter_data;
482 filter_client_add(obd, filter, fed, cl_idx);
483 /* create helper if export init gets more complex */
484 spin_lock_init(&fed->fed_lock);
487 obd->obd_recoverable_clients++;
488 class_export_put(exp);
490 CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
493 if (last_rcvd > le64_to_cpu(fsd->fsd_last_transno))
494 fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
498 obd->obd_last_committed = le64_to_cpu(fsd->fsd_last_transno);
500 if (obd->obd_recoverable_clients) {
501 CWARN("RECOVERY: %d recoverable clients, last_rcvd "
502 LPU64"\n", obd->obd_recoverable_clients,
503 le64_to_cpu(fsd->fsd_last_transno));
504 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
505 obd->obd_recovering = 1;
509 OBD_FREE(fcd, sizeof(*fcd));
512 filter->fo_mount_count = mount_count + 1;
513 fsd->fsd_mount_count = cpu_to_le64(filter->fo_mount_count);
515 /* save it, so mount count and last_transno is current */
516 rc = filter_update_server_data(obd, filp, filter->fo_fsd, 1);
521 class_disconnect_exports(obd, 0);
523 filter_free_server_data(filter);
527 static int filter_cleanup_groups(struct obd_device *obd)
529 struct filter_obd *filter = &obd->u.filter;
530 struct dentry *dentry;
534 for (i = 0; i < filter->fo_group_count; i++) {
535 if (filter->fo_subdirs != NULL) {
536 for (k = 0; k < filter->fo_subdir_count; k++) {
537 dentry = filter->fo_subdirs[i].dentry[k];
541 filter->fo_subdirs[i].dentry[k] = NULL;
544 if (filter->fo_last_objid_files[i] != NULL) {
545 filp_close(filter->fo_last_objid_files[i], 0);
546 filter->fo_last_objid_files[i] = NULL;
548 if (filter->fo_groups[i] != NULL) {
549 dput(filter->fo_groups[i]);
550 filter->fo_groups[i] = NULL;
553 if (filter->fo_subdirs != NULL)
554 OBD_FREE(filter->fo_subdirs,
555 filter->fo_group_count * sizeof(*filter->fo_subdirs));
556 if (filter->fo_groups != NULL)
557 OBD_FREE(filter->fo_groups,
558 filter->fo_group_count * sizeof(*filter->fo_groups));
559 if (filter->fo_last_objids != NULL)
560 OBD_FREE(filter->fo_last_objids,
561 filter->fo_group_count * sizeof(__u64));
562 if (filter->fo_last_objid_files != NULL)
563 OBD_FREE(filter->fo_last_objid_files,
564 filter->fo_group_count * sizeof(struct file *));
565 f_dput(filter->fo_dentry_O);
569 static int filter_read_group_internal(struct obd_device *obd, int group,
572 struct filter_obd *filter = &obd->u.filter;
573 __u64 *new_objids = NULL;
574 struct filter_subdirs *new_subdirs = NULL, *tmp_subdirs;
575 struct dentry **new_groups = NULL;
576 struct file **new_files = NULL;
577 struct dentry *dentry;
579 int old_count = filter->fo_group_count, rc, stage = 0, i;
584 snprintf(name, 24, "%d", group);
588 dentry = ll_lookup_one_len(name, filter->fo_dentry_O,
590 if (IS_ERR(dentry)) {
591 CERROR("Cannot lookup expected object group %d: %ld\n",
592 group, PTR_ERR(dentry));
593 RETURN(PTR_ERR(dentry));
596 dentry = simple_mkdir(filter->fo_dentry_O, name, 0700, 1);
597 if (IS_ERR(dentry)) {
598 CERROR("cannot lookup/create O/%s: rc = %ld\n", name,
600 RETURN(PTR_ERR(dentry));
605 snprintf(name, 24, "O/%d/LAST_ID", group);
607 filp = filp_open(name, O_CREAT | O_RDWR, 0700);
609 CERROR("cannot create %s: rc = %ld\n", name, PTR_ERR(filp));
610 GOTO(cleanup, rc = PTR_ERR(filp));
614 rc = fsfilt_read_record(obd, filp, &last_objid, sizeof(__u64), &off);
616 CDEBUG(D_INODE, "error reading %s: rc %d\n", name, rc);
620 if (filter->fo_subdir_count) {
621 OBD_ALLOC(tmp_subdirs, sizeof(*tmp_subdirs));
622 if (tmp_subdirs == NULL)
623 GOTO(cleanup, rc = -ENOMEM);
626 for (i = 0; i < filter->fo_subdir_count; i++) {
628 snprintf(dir, sizeof(dir), "d%u", i);
630 tmp_subdirs->dentry[i] =
631 simple_mkdir(dentry, dir, 0700, 1);
632 if (IS_ERR(tmp_subdirs->dentry[i])) {
633 rc = PTR_ERR(tmp_subdirs->dentry[i]);
634 CERROR("can't lookup/create O/%d/%s: rc = %d\n",
638 CDEBUG(D_INODE, "got/created O/%d/%s: %p\n", group, dir,
639 tmp_subdirs->dentry[i]);
643 /* 'group' is an index; we need an array of length 'group + 1' */
644 if (group + 1 > old_count) {
646 OBD_ALLOC(new_objids, len * sizeof(*new_objids));
647 OBD_ALLOC(new_subdirs, len * sizeof(*new_subdirs));
648 OBD_ALLOC(new_groups, len * sizeof(*new_groups));
649 OBD_ALLOC(new_files, len * sizeof(*new_files));
651 if (new_objids == NULL || new_subdirs == NULL ||
652 new_groups == NULL || new_files == NULL)
653 GOTO(cleanup, rc = -ENOMEM);
655 memcpy(new_objids, filter->fo_last_objids,
656 old_count * sizeof(*new_objids));
657 memcpy(new_subdirs, filter->fo_subdirs,
658 old_count * sizeof(*new_subdirs));
659 memcpy(new_groups, filter->fo_groups,
660 old_count * sizeof(*new_groups));
661 memcpy(new_files, filter->fo_last_objid_files,
662 old_count * sizeof(*new_files));
665 OBD_FREE(filter->fo_last_objids,
666 old_count * sizeof(*new_objids));
667 OBD_FREE(filter->fo_subdirs,
668 old_count * sizeof(*new_subdirs));
669 OBD_FREE(filter->fo_groups,
670 old_count * sizeof(*new_groups));
671 OBD_FREE(filter->fo_last_objid_files,
672 old_count * sizeof(*new_files));
674 filter->fo_last_objids = new_objids;
675 filter->fo_subdirs = new_subdirs;
676 filter->fo_groups = new_groups;
677 filter->fo_last_objid_files = new_files;
678 filter->fo_group_count = len;
681 filter->fo_groups[group] = dentry;
682 filter->fo_last_objid_files[group] = filp;
683 if (filter->fo_subdir_count) {
684 filter->fo_subdirs[group] = *tmp_subdirs;
685 OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
688 if (filp->f_dentry->d_inode->i_size == 0) {
689 filter->fo_last_objids[group] = FILTER_INIT_OBJID;
693 filter->fo_last_objids[group] = le64_to_cpu(last_objid);
694 CDEBUG(D_INODE, "%s: server last_objid group %d: "LPU64"\n",
695 obd->obd_name, group, last_objid);
700 if (new_objids != NULL)
701 OBD_FREE(new_objids, group * sizeof(*new_objids));
702 if (new_subdirs != NULL)
703 OBD_FREE(new_subdirs, group * sizeof(*new_subdirs));
704 if (new_groups != NULL)
705 OBD_FREE(new_groups, group * sizeof(*new_groups));
706 if (new_files != NULL)
707 OBD_FREE(new_files, group * sizeof(*new_files));
709 if (filter->fo_subdir_count) {
710 for (i = 0; i < filter->fo_subdir_count; i++) {
711 if (tmp_subdirs->dentry[i] != NULL)
712 dput(tmp_subdirs->dentry[i]);
714 OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
724 static int filter_read_groups(struct obd_device *obd, int last_group,
727 struct filter_obd *filter = &obd->u.filter;
728 int old_count = filter->fo_group_count, group = old_count, rc = 0;
730 for (group = old_count; group <= last_group; group++) {
732 continue; /* no group zero */
734 rc = filter_read_group_internal(obd, group, create);
741 static int filter_prep_groups(struct obd_device *obd)
743 struct filter_obd *filter = &obd->u.filter;
744 struct dentry *dentry, *O_dentry;
745 int rc = 0, cleanup_phase = 0;
748 O_dentry = simple_mkdir(current->fs->pwd, "O", 0700, 1);
749 CDEBUG(D_INODE, "got/created O: %p\n", O_dentry);
750 if (IS_ERR(O_dentry)) {
751 rc = PTR_ERR(O_dentry);
752 CERROR("cannot open/create O: rc = %d\n", rc);
755 filter->fo_dentry_O = O_dentry;
756 cleanup_phase = 1; /* O_dentry */
758 /* Lookup "R" to tell if we're on an old OST FS and need to convert
759 * from O/R/<dir>/<objid> to O/0/<dir>/<objid>. This can be removed
760 * some time post 1.0 when all old-style OSTs have converted along
761 * with the init_objid hack. */
762 dentry = ll_lookup_one_len("R", O_dentry, 1);
764 GOTO(cleanup, rc = PTR_ERR(dentry));
765 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
766 struct dentry *O0_dentry = lookup_one_len("0", O_dentry, 1);
769 CWARN("converting OST to new object layout\n");
770 if (IS_ERR(O0_dentry)) {
771 rc = PTR_ERR(O0_dentry);
772 CERROR("error looking up O/0: rc %d\n", rc);
776 if (O0_dentry->d_inode) {
777 CERROR("Both O/R and O/0 exist. Fix manually.\n");
778 GOTO(cleanup_O0, rc = -EEXIST);
781 down(&O_dentry->d_inode->i_sem);
782 rc = vfs_rename(O_dentry->d_inode, dentry,
783 O_dentry->d_inode, O0_dentry);
784 up(&O_dentry->d_inode->i_sem);
787 CERROR("error renaming O/R to O/0: rc %d\n", rc);
788 GOTO(cleanup_O0, rc);
790 filter->fo_fsd->fsd_feature_incompat |=
791 cpu_to_le32(FILTER_INCOMPAT_GROUPS);
792 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
794 GOTO(cleanup_O0, rc);
806 cleanup_phase = 2; /* groups */
808 /* Group 0 is no longer a legal group, to catch uninitialized IDs */
809 #define FILTER_MIN_GROUPS 3
810 rc = filter_read_groups(obd, FILTER_MIN_GROUPS, 1);
817 switch (cleanup_phase) {
819 filter_cleanup_groups(obd);
821 f_dput(filter->fo_dentry_O);
822 filter->fo_dentry_O = NULL;
829 /* setup the object store with correct subdirectories */
830 static int filter_prep(struct obd_device *obd)
832 struct lvfs_run_ctxt saved;
833 struct filter_obd *filter = &obd->u.filter;
839 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
840 file = filp_open(LAST_RCVD, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
841 if (!file || IS_ERR(file)) {
843 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
848 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
849 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
850 file->f_dentry->d_inode->i_mode);
851 GOTO(err_filp, rc = -ENOENT);
854 /* steal operations */
855 inode = file->f_dentry->d_inode;
856 filter->fo_fop = file->f_op;
857 filter->fo_iop = inode->i_op;
858 filter->fo_aops = inode->i_mapping->a_ops;
860 rc = filter_init_server_data(obd, file);
862 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
865 filter->fo_rcvd_filp = file;
867 rc = filter_prep_groups(obd);
869 GOTO(err_server_data, rc);
872 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
877 //class_disconnect_exports(obd, 0);
878 filter_free_server_data(filter);
880 if (filp_close(file, 0))
881 CERROR("can't close %s after error\n", LAST_RCVD);
882 filter->fo_rcvd_filp = NULL;
886 /* cleanup the filter: write last used object id to status file */
887 static void filter_post(struct obd_device *obd)
889 struct lvfs_run_ctxt saved;
890 struct filter_obd *filter = &obd->u.filter;
893 /* XXX: filter_update_lastobjid used to call fsync_dev. It might be
894 * best to start a transaction with h_sync, because we removed this
897 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
898 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
901 CERROR("error writing server data: rc = %d\n", rc);
903 for (i = 1; i < filter->fo_group_count; i++) {
904 rc = filter_update_last_objid(obd, i,
905 (i == filter->fo_group_count - 1));
907 CERROR("error writing group %d lastobjid: rc = %d\n",
911 rc = filp_close(filter->fo_rcvd_filp, 0);
912 filter->fo_rcvd_filp = NULL;
914 CERROR("error closing %s: rc = %d\n", LAST_RCVD, rc);
916 filter_cleanup_groups(obd);
917 filter_free_server_data(filter);
918 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
921 static void filter_set_last_id(struct filter_obd *filter, int group, obd_id id)
923 LASSERT(filter->fo_fsd != NULL);
925 LASSERT(group < filter->fo_group_count);
927 spin_lock(&filter->fo_objidlock);
928 filter->fo_last_objids[group] = id;
929 spin_unlock(&filter->fo_objidlock);
932 __u64 filter_last_id(struct filter_obd *filter, int group)
935 LASSERT(filter->fo_fsd != NULL);
937 LASSERT(group < filter->fo_group_count);
939 spin_lock(&filter->fo_objidlock);
940 id = filter->fo_last_objids[group];
941 spin_unlock(&filter->fo_objidlock);
946 /* direct cut-n-paste of mds_blocking_ast() */
947 static int filter_blocking_ast(struct ldlm_lock *lock,
948 struct ldlm_lock_desc *desc,
949 void *data, int flag)
954 if (flag == LDLM_CB_CANCELING) {
955 /* Don't need to do anything here. */
959 /* XXX layering violation! -phil */
960 l_lock(&lock->l_resource->lr_namespace->ns_lock);
961 /* Get this: if filter_blocking_ast is racing with ldlm_intent_policy,
962 * such that filter_blocking_ast is called just before l_i_p takes the
963 * ns_lock, then by the time we get the lock, we might not be the
964 * correct blocking function anymore. So check, and return early, if
966 if (lock->l_blocking_ast != filter_blocking_ast) {
967 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
971 lock->l_flags |= LDLM_FL_CBPENDING;
972 do_ast = (!lock->l_readers && !lock->l_writers);
973 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
976 struct lustre_handle lockh;
979 LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
980 ldlm_lock2handle(lock, &lockh);
981 rc = ldlm_cli_cancel(&lockh);
983 CERROR("ldlm_cli_cancel: %d\n", rc);
985 LDLM_DEBUG(lock, "Lock still has references, will be "
991 static int filter_lock_dentry(struct obd_device *obd, struct dentry *dparent)
993 down(&dparent->d_inode->i_sem);
997 /* We never dget the object parent, so DON'T dput it either */
998 static void filter_parent_unlock(struct dentry *dparent)
1000 up(&dparent->d_inode->i_sem);
1003 /* We never dget the object parent, so DON'T dput it either */
1004 struct dentry *filter_parent(struct obd_device *obd, obd_gr group, obd_id objid)
1006 struct filter_obd *filter = &obd->u.filter;
1007 LASSERT(group < filter->fo_group_count);
1010 if (filter->fo_subdir_count == 0)
1011 return filter->fo_groups[group];
1013 return filter->fo_subdirs[group].dentry[objid & (filter->fo_subdir_count - 1)];
1016 /* We never dget the object parent, so DON'T dput it either */
1017 struct dentry *filter_parent_lock(struct obd_device *obd, obd_gr group,
1020 unsigned long now = jiffies;
1021 struct dentry *dparent = filter_parent(obd, group, objid);
1024 if (IS_ERR(dparent))
1028 LASSERT(dparent->d_inode);
1030 rc = filter_lock_dentry(obd, dparent);
1031 if (time_after(jiffies, now + 15 * HZ))
1032 CERROR("slow parent lock %lus\n", (jiffies - now) / HZ);
1033 return rc ? ERR_PTR(rc) : dparent;
1036 /* How to get files, dentries, inodes from object id's.
1038 * If dir_dentry is passed, the caller has already locked the parent
1039 * appropriately for this operation (normally a write lock). If
1040 * dir_dentry is NULL, we do a read lock while we do the lookup to
1041 * avoid races with create/destroy and such changing the directory
1042 * internal to the filesystem code. */
1043 struct dentry *filter_fid2dentry(struct obd_device *obd,
1044 struct dentry *dir_dentry,
1045 obd_gr group, obd_id id)
1047 struct dentry *dparent = dir_dentry;
1048 struct dentry *dchild;
1054 CERROR("fatal: invalid object id 0\n");
1055 RETURN(ERR_PTR(-ESTALE));
1058 len = sprintf(name, LPU64, id);
1059 if (dir_dentry == NULL) {
1060 dparent = filter_parent_lock(obd, group, id);
1061 if (IS_ERR(dparent))
1064 CDEBUG(D_INODE, "looking up object O/%*s/%s\n",
1065 dparent->d_name.len, dparent->d_name.name, name);
1066 dchild = /*ll_*/lookup_one_len(name, dparent, len);
1067 if (dir_dentry == NULL)
1068 filter_parent_unlock(dparent);
1069 if (IS_ERR(dchild)) {
1070 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
1074 CDEBUG(D_INODE, "got child objid %s: %p, count = %d\n",
1075 name, dchild, atomic_read(&dchild->d_count));
1077 LASSERT(atomic_read(&dchild->d_count) > 0);
1082 static int filter_prepare_destroy(struct obd_device *obd, obd_id objid,
1085 struct lustre_handle lockh;
1086 int flags = LDLM_AST_DISCARD_DATA, rc;
1087 struct ldlm_res_id res_id = { .name = { objid, 0, group, 0 } };
1088 ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
1091 /* Tell the clients that the object is gone now and that they should
1092 * throw away any cached pages. */
1093 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, res_id,
1094 LDLM_EXTENT, &policy, LCK_PW,
1095 &flags, filter_blocking_ast, ldlm_completion_ast,
1096 NULL, NULL, NULL, 0, NULL, &lockh);
1098 /* We only care about the side-effects, just drop the lock. */
1100 ldlm_lock_decref(&lockh, LCK_PW);
1105 /* Caller must hold LCK_PW on parent and push us into kernel context.
1106 * Caller is also required to ensure that dchild->d_inode exists. */
1107 static int filter_destroy_internal(struct obd_device *obd, obd_id objid,
1108 struct dentry *dparent,
1109 struct dentry *dchild)
1111 struct inode *inode = dchild->d_inode;
1115 if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) {
1116 CERROR("destroying objid %*s nlink = %lu, count = %d\n",
1117 dchild->d_name.len, dchild->d_name.name,
1118 (unsigned long)inode->i_nlink,
1119 atomic_read(&inode->i_count));
1122 rc = vfs_unlink(dparent->d_inode, dchild);
1125 CERROR("error unlinking objid %*s: rc %d\n",
1126 dchild->d_name.len, dchild->d_name.name, rc);
1131 static int filter_intent_policy(struct ldlm_namespace *ns,
1132 struct ldlm_lock **lockp, void *req_cookie,
1133 ldlm_mode_t mode, int flags, void *data)
1135 struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1136 struct ptlrpc_request *req = req_cookie;
1137 struct ldlm_lock *lock = *lockp, *l = NULL;
1138 struct ldlm_resource *res = lock->l_resource;
1139 ldlm_processing_policy policy;
1140 struct ost_lvb *res_lvb, *reply_lvb;
1141 struct list_head *tmp;
1143 int tmpflags = 0, rc, repsize[2] = {sizeof(struct ldlm_reply),
1144 sizeof(struct ost_lvb) };
1147 policy = ldlm_get_processing_policy(res);
1148 LASSERT(policy != NULL);
1149 LASSERT(req != NULL);
1151 rc = lustre_pack_reply(req, 2, repsize, NULL);
1153 RETURN(req->rq_status = rc);
1155 reply_lvb = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*reply_lvb));
1156 LASSERT(reply_lvb != NULL);
1158 //fixup_handle_for_resent_req(req, lock, &lockh);
1160 /* If we grant any lock at all, it will be a whole-file read lock.
1161 * Call the extent policy function to see if our request can be
1162 * granted, or is blocked. */
1163 lock->l_policy_data.l_extent.start = 0;
1164 lock->l_policy_data.l_extent.end = OBD_OBJECT_EOF;
1165 lock->l_req_mode = LCK_PR;
1167 l_lock(&res->lr_namespace->ns_lock);
1169 res->lr_tmp = &rpc_list;
1170 rc = policy(lock, &tmpflags, 0, &err);
1173 /* FIXME: we should change the policy function slightly, to not make
1174 * this list at all, since we just turn around and free it */
1175 while (!list_empty(&rpc_list)) {
1176 struct ldlm_ast_work *w =
1177 list_entry(rpc_list.next, struct ldlm_ast_work, w_list);
1178 list_del(&w->w_list);
1179 LDLM_LOCK_PUT(w->w_lock);
1180 OBD_FREE(w, sizeof(*w));
1183 if (rc == LDLM_ITER_CONTINUE) {
1184 /* The lock met with no resistance; we're finished. */
1185 l_unlock(&res->lr_namespace->ns_lock);
1186 RETURN(ELDLM_LOCK_REPLACED);
1189 /* Do not grant any lock, but instead send GL callbacks. The extent
1190 * policy nicely created a list of all PW locks for us. We will choose
1191 * the highest of those which are larger than the size in the LVB, if
1192 * any, and perform a glimpse callback. */
1193 down(&res->lr_lvb_sem);
1194 res_lvb = res->lr_lvb_data;
1195 LASSERT(res_lvb != NULL);
1196 reply_lvb->lvb_size = res_lvb->lvb_size;
1197 reply_lvb->lvb_blocks = res_lvb->lvb_blocks;
1198 up(&res->lr_lvb_sem);
1200 list_for_each(tmp, &res->lr_granted) {
1201 struct ldlm_lock *tmplock =
1202 list_entry(tmp, struct ldlm_lock, l_res_link);
1204 if (tmplock->l_granted_mode == LCK_PR)
1207 if (tmplock->l_policy_data.l_extent.end <= reply_lvb->lvb_size)
1211 l = LDLM_LOCK_GET(tmplock);
1215 if (l->l_policy_data.l_extent.start >
1216 tmplock->l_policy_data.l_extent.start)
1220 l = LDLM_LOCK_GET(tmplock);
1222 l_unlock(&res->lr_namespace->ns_lock);
1224 /* There were no PW locks beyond the size in the LVB; finished. */
1226 RETURN(ELDLM_LOCK_ABORTED);
1228 LASSERT(l->l_glimpse_ast != NULL);
1229 rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
1230 if (rc != 0 && res->lr_namespace->ns_lvbo &&
1231 res->lr_namespace->ns_lvbo->lvbo_update) {
1232 res->lr_namespace->ns_lvbo->lvbo_update(res, NULL, 0, 1);
1235 down(&res->lr_lvb_sem);
1236 reply_lvb->lvb_size = res_lvb->lvb_size;
1237 reply_lvb->lvb_blocks = res_lvb->lvb_blocks;
1238 up(&res->lr_lvb_sem);
1242 RETURN(ELDLM_LOCK_ABORTED);
1245 static int filter_post_fs_cleanup(struct obd_device *obd)
1249 rc = fsfilt_post_cleanup(obd);
1254 static int filter_group_set_kml_flags(struct obd_device *obd, int group)
1256 struct filter_obd *filter = &obd->u.filter;
1260 /* zero group is not longer valid. */
1263 for (i = 0; i < filter->fo_subdir_count; i++) {
1264 struct dentry *dentry;
1265 dentry = (filter->fo_subdirs + group)->dentry[i];
1266 rc = fsfilt_set_kml_flags(obd, dentry->d_inode);
1272 static int filter_post_fs_setup(struct obd_device *obd)
1274 struct filter_obd *filter = &obd->u.filter;
1276 struct llog_ctxt *ctxt = NULL;
1278 rc = fsfilt_post_setup(obd);
1282 for (j = 0; j < filter->fo_group_count; j++) {
1283 rc = filter_group_set_kml_flags(obd, j);
1288 fsfilt_get_reint_log_ctxt(obd, filter->fo_sb, &ctxt);
1290 ctxt->loc_obd = obd;
1291 ctxt->loc_idx = LLOG_REINT_ORIG_CTXT;
1292 obd->obd_llog_ctxt[LLOG_REINT_ORIG_CTXT] = ctxt;
1294 fsfilt_set_ost_flags(obd, filter->fo_sb);
1298 /* mount the file system (secretly) */
1299 int filter_common_setup(struct obd_device *obd, obd_count len,
1300 void *buf, char *option)
1302 struct lustre_cfg* lcfg = buf;
1303 struct filter_obd *filter = &obd->u.filter;
1304 struct vfsmount *mnt;
1305 char name[32] = "CATLIST";
1309 dev_clear_rdonly(2);
1311 if (!lcfg->lcfg_inlbuf1 || !lcfg->lcfg_inlbuf2)
1314 obd->obd_fsops = fsfilt_get_ops(lcfg->lcfg_inlbuf2);
1315 if (IS_ERR(obd->obd_fsops))
1316 RETURN(PTR_ERR(obd->obd_fsops));
1318 mnt = do_kern_mount(lcfg->lcfg_inlbuf2, MS_NOATIME | MS_NODIRATIME,
1319 lcfg->lcfg_inlbuf1, option);
1324 if (lcfg->lcfg_inllen3 > 0 && lcfg->lcfg_inlbuf3) {
1325 if (*lcfg->lcfg_inlbuf3 == 'f') {
1326 obd->obd_replayable = 1;
1327 obd_sync_filter = 1;
1328 CWARN("%s: recovery enabled\n", obd->obd_name);
1330 if (*lcfg->lcfg_inlbuf3 != 'n') {
1331 CERROR("unrecognised flag '%c'\n",
1332 *lcfg->lcfg_inlbuf3);
1334 // XXX Robert? Why do we get errors here
1335 // GOTO(err_mntput, rc = -EINVAL);
1339 filter->fo_vfsmnt = mnt;
1340 filter->fo_sb = mnt->mnt_sb;
1341 filter->fo_fstype = mnt->mnt_sb->s_type->name;
1342 CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
1344 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1345 obd->obd_lvfs_ctxt.pwdmnt = mnt;
1346 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
1347 obd->obd_lvfs_ctxt.fs = get_ds();
1348 obd->obd_lvfs_ctxt.cb_ops = filter_lvfs_ops;
1350 rc = fsfilt_setup(obd, mnt->mnt_sb);
1352 GOTO(err_mntput, rc);
1354 rc = filter_prep(obd);
1356 GOTO(err_mntput, rc);
1359 filter->fo_destroy_in_progress = 0;
1360 sema_init(&filter->fo_create_lock, 1);
1362 spin_lock_init(&filter->fo_translock);
1363 spin_lock_init(&filter->fo_objidlock);
1364 INIT_LIST_HEAD(&filter->fo_export_list);
1365 sema_init(&filter->fo_alloc_lock, 1);
1366 spin_lock_init(&filter->fo_r_pages.oh_lock);
1367 spin_lock_init(&filter->fo_w_pages.oh_lock);
1368 spin_lock_init(&filter->fo_r_discont_pages.oh_lock);
1369 spin_lock_init(&filter->fo_w_discont_pages.oh_lock);
1370 spin_lock_init(&filter->fo_r_discont_blocks.oh_lock);
1371 spin_lock_init(&filter->fo_w_discont_blocks.oh_lock);
1372 filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE;
1374 INIT_LIST_HEAD(&filter->fo_llog_list);
1375 spin_lock_init(&filter->fo_llog_list_lock);
1377 obd->obd_namespace = ldlm_namespace_new("filter-tgt",
1378 LDLM_NAMESPACE_SERVER);
1379 if (obd->obd_namespace == NULL)
1380 GOTO(err_post, rc = -ENOMEM);
1381 obd->obd_namespace->ns_lvbp = obd;
1382 obd->obd_namespace->ns_lvbo = &filter_lvbo;
1383 ldlm_register_intent(obd->obd_namespace, filter_intent_policy);
1385 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1386 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
1388 rc = obd_llog_cat_initialize(obd, &obd->obd_llogs, 1, name);
1390 CERROR("failed to setup llogging subsystems\n");
1403 fsfilt_put_ops(obd->obd_fsops);
1407 static int filter_attach(struct obd_device *obd, obd_count len, void *data)
1409 struct lprocfs_static_vars lvars;
1412 lprocfs_init_vars(filter, &lvars);
1413 rc = lprocfs_obd_attach(obd, lvars.obd_vars);
1417 rc = lprocfs_alloc_obd_stats(obd, LPROC_FILTER_LAST);
1421 /* Init obdfilter private stats here */
1422 lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_READ_BYTES,
1423 LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
1424 lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
1425 LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
1427 return lproc_filter_attach_seqstat(obd);
1430 static int filter_detach(struct obd_device *dev)
1432 lprocfs_free_obd_stats(dev);
1433 return lprocfs_obd_detach(dev);
1436 static int filter_setup(struct obd_device *obd, obd_count len, void *buf)
1438 struct lustre_cfg* lcfg = buf;
1441 /* all mount options including errors=remount-ro and asyncdel are passed
1442 * using 4th lcfg param. And it is good, finally we have got rid of
1443 * hardcoded fs types in the code. */
1444 rc = filter_common_setup(obd, len, buf, lcfg->lcfg_inlbuf4);
1447 rc = filter_post_fs_setup(obd);
1451 static int filter_cleanup(struct obd_device *obd, int flags)
1453 struct filter_obd *filter = &obd->u.filter;
1456 if (flags & OBD_OPT_FAILOVER)
1457 CERROR("%s: shutting down for failover; client state will"
1458 " be preserved.\n", obd->obd_name);
1460 if (!list_empty(&obd->obd_exports)) {
1461 CERROR("%s: still has clients!\n", obd->obd_name);
1462 class_disconnect_exports(obd, flags);
1463 if (!list_empty(&obd->obd_exports)) {
1464 CERROR("still has exports after forced cleanup?\n");
1469 ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
1471 if (filter->fo_sb == NULL)
1474 filter_post_fs_cleanup(obd);
1477 shrink_dcache_parent(filter->fo_sb->s_root);
1480 if (atomic_read(&filter->fo_vfsmnt->mnt_count) > 1)
1481 CERROR("%s: mount point %p busy, mnt_count: %d\n",
1482 obd->obd_name, filter->fo_vfsmnt,
1483 atomic_read(&filter->fo_vfsmnt->mnt_count));
1486 mntput(filter->fo_vfsmnt);
1487 //destroy_buffers(filter->fo_sb->s_dev);
1488 filter->fo_sb = NULL;
1489 fsfilt_put_ops(obd->obd_fsops);
1492 dev_clear_rdonly(2);
1497 /* nearly identical to mds_connect */
1498 static int filter_connect(struct lustre_handle *conn, struct obd_device *obd,
1499 struct obd_uuid *cluuid)
1501 struct obd_export *exp;
1502 struct filter_export_data *fed;
1503 struct filter_client_data *fcd = NULL;
1504 struct filter_obd *filter = &obd->u.filter;
1508 if (conn == NULL || obd == NULL || cluuid == NULL)
1511 rc = class_connect(conn, obd, cluuid);
1514 exp = class_conn2export(conn);
1515 LASSERT(exp != NULL);
1517 fed = &exp->exp_filter_data;
1519 spin_lock_init(&fed->fed_lock);
1521 if (!obd->obd_replayable)
1522 GOTO(cleanup, rc = 0);
1524 OBD_ALLOC(fcd, sizeof(*fcd));
1526 CERROR("filter: out of memory for client data\n");
1527 GOTO(cleanup, rc = -ENOMEM);
1530 memcpy(fcd->fcd_uuid, cluuid, sizeof(fcd->fcd_uuid));
1533 rc = filter_client_add(obd, filter, fed, -1);
1538 OBD_FREE(fcd, sizeof(*fcd));
1539 class_disconnect(exp, 0);
1541 class_export_put(exp);
1546 static int filter_precleanup(struct obd_device *obd, int flags)
1548 struct filter_group_llog *log;
1549 struct filter_obd *filter;
1553 filter = &obd->u.filter;
1555 spin_lock(&filter->fo_llog_list_lock);
1556 while (!list_empty(&filter->fo_llog_list)) {
1557 log = list_entry(filter->fo_llog_list.next,
1558 struct filter_group_llog, list);
1559 list_del(&log->list);
1560 spin_unlock(&filter->fo_llog_list_lock);
1562 rc = obd_llog_finish(obd, log->llogs, 0);
1564 CERROR("failed to cleanup llogging subsystem for %u\n",
1566 OBD_FREE(log->llogs, sizeof(*(log->llogs)));
1567 OBD_FREE(log, sizeof(*log));
1568 spin_lock(&filter->fo_llog_list_lock);
1570 spin_unlock(&filter->fo_llog_list_lock);
1572 rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1574 CERROR("failed to cleanup llogging subsystem\n");
1579 /* Do extra sanity checks for grant accounting. We do this at connect,
1580 * disconnect, and statfs RPC time, so it shouldn't be too bad. We can
1581 * always get rid of it or turn it off when we know accounting is good. */
1582 static void filter_grant_sanity_check(struct obd_device *obd, const char *func)
1584 struct filter_export_data *fed;
1585 struct obd_export *exp;
1586 obd_size maxsize = obd->obd_osfs.os_blocks * obd->obd_osfs.os_bsize;
1587 obd_size tot_dirty = 0, tot_pending = 0, tot_granted = 0;
1588 obd_size fo_tot_dirty, fo_tot_pending, fo_tot_granted;
1590 if (list_empty(&obd->obd_exports))
1593 spin_lock(&obd->obd_osfs_lock);
1594 spin_lock(&obd->obd_dev_lock);
1595 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1596 fed = &exp->exp_filter_data;
1597 LASSERTF(fed->fed_grant + fed->fed_pending <= maxsize,
1598 "cli %s/%p %lu+%lu > "LPU64"\n",
1599 exp->exp_client_uuid.uuid, exp,
1600 fed->fed_grant, fed->fed_pending, maxsize);
1601 LASSERTF(fed->fed_dirty <= maxsize, "cli %s/%p %lu > "LPU64"\n",
1602 exp->exp_client_uuid.uuid, exp,fed->fed_dirty,maxsize);
1603 CDEBUG(D_CACHE,"%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1604 obd->obd_name, exp->exp_client_uuid.uuid, exp,
1605 fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1606 tot_granted += fed->fed_grant + fed->fed_pending;
1607 tot_pending += fed->fed_pending;
1608 tot_dirty += fed->fed_dirty;
1610 fo_tot_granted = obd->u.filter.fo_tot_granted;
1611 fo_tot_pending = obd->u.filter.fo_tot_pending;
1612 fo_tot_dirty = obd->u.filter.fo_tot_dirty;
1613 spin_unlock(&obd->obd_dev_lock);
1614 spin_unlock(&obd->obd_osfs_lock);
1616 /* Do these assertions outside the spinlocks so we don't kill system */
1617 if (tot_granted != fo_tot_granted)
1618 CERROR("%s: tot_granted "LPU64" != fo_tot_granted "LPU64"\n",
1619 func, tot_granted, fo_tot_granted);
1620 if (tot_pending != fo_tot_pending)
1621 CERROR("%s: tot_pending "LPU64" != fo_tot_pending "LPU64"\n",
1622 func, tot_pending, fo_tot_pending);
1623 if (tot_dirty != fo_tot_dirty)
1624 CERROR("%s: tot_dirty "LPU64" != fo_tot_dirty "LPU64"\n",
1625 func, tot_dirty, fo_tot_dirty);
1626 if (tot_pending > tot_granted)
1627 CERROR("%s: tot_pending "LPU64" > tot_granted "LPU64"\n",
1628 func, tot_pending, tot_granted);
1629 if (tot_granted > maxsize)
1630 CERROR("%s: tot_granted "LPU64" > maxsize "LPU64"\n",
1631 func, tot_granted, maxsize);
1632 if (tot_dirty > maxsize)
1633 CERROR("%s: tot_dirty "LPU64" > maxsize "LPU64"\n",
1634 func, tot_dirty, maxsize);
1637 /* Remove this client from the grant accounting totals. We also remove
1638 * the export from the obd device under the osfs and dev locks to ensure
1639 * that the filter_grant_sanity_check() calculations are always valid.
1640 * The client should do something similar when it invalidates its import. */
1641 static void filter_grant_discard(struct obd_export *exp)
1643 struct obd_device *obd = exp->exp_obd;
1644 struct filter_obd *filter = &obd->u.filter;
1645 struct filter_export_data *fed = &exp->exp_filter_data;
1647 spin_lock(&obd->obd_osfs_lock);
1648 spin_lock(&exp->exp_obd->obd_dev_lock);
1649 list_del_init(&exp->exp_obd_chain);
1650 spin_unlock(&exp->exp_obd->obd_dev_lock);
1652 CDEBUG(D_CACHE, "%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1653 obd->obd_name, exp->exp_client_uuid.uuid, exp,
1654 fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1656 LASSERTF(filter->fo_tot_granted >= fed->fed_grant,
1657 "%s: tot_granted "LPU64" cli %s/%p fed_grant %lu\n",
1658 obd->obd_name, filter->fo_tot_granted,
1659 exp->exp_client_uuid.uuid, exp, fed->fed_grant);
1660 filter->fo_tot_granted -= fed->fed_grant;
1661 LASSERTF(exp->exp_obd->u.filter.fo_tot_pending >= fed->fed_pending,
1662 "%s: tot_pending "LPU64" cli %s/%p fed_pending %lu\n",
1663 obd->obd_name, filter->fo_tot_pending,
1664 exp->exp_client_uuid.uuid, exp, fed->fed_pending);
1665 LASSERTF(filter->fo_tot_dirty >= fed->fed_dirty,
1666 "%s: tot_dirty "LPU64" cli %s/%p fed_dirty %lu\n",
1667 obd->obd_name, filter->fo_tot_dirty,
1668 exp->exp_client_uuid.uuid, exp, fed->fed_dirty);
1669 filter->fo_tot_dirty -= fed->fed_dirty;
1673 spin_unlock(&obd->obd_osfs_lock);
1676 static int filter_destroy_export(struct obd_export *exp)
1680 if (exp->exp_filter_data.fed_pending)
1681 CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
1682 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
1683 exp, exp->exp_filter_data.fed_pending);
1685 target_destroy_export(exp);
1687 if (exp->exp_obd->obd_replayable)
1688 filter_client_free(exp, exp->exp_flags);
1690 filter_grant_discard(exp);
1691 if (!(exp->exp_flags & OBD_OPT_FORCE))
1692 filter_grant_sanity_check(exp->exp_obd, __FUNCTION__);
1697 static void filter_sync_llogs(struct obd_export *dexp)
1699 struct filter_group_llog *fglog, *nlog;
1700 struct obd_device *obd = dexp->exp_obd;
1701 struct filter_obd *filter;
1702 int worked = 0, group;
1703 struct llog_ctxt *ctxt;
1705 filter = &obd->u.filter;
1707 /* we can't sync log holding spinlock. also, we do not want to get
1708 * into livelock. so we do following: loop over MDS's exports in
1709 * group order and skip already synced llogs -bzzz */
1711 /* look for group with min. number, but > worked */
1714 spin_lock(&filter->fo_llog_list_lock);
1715 list_for_each_entry(nlog, &filter->fo_llog_list, list) {
1717 if (nlog->group <= worked) {
1718 /* this group is already synced */
1722 if (group < nlog->group) {
1723 /* we have group with smaller number to sync */
1727 /* store current minimal group */
1729 group = nlog->group;
1731 spin_unlock(&filter->fo_llog_list_lock);
1734 worked = fglog->group;
1735 ctxt = llog_get_context(fglog->llogs,
1736 LLOG_UNLINK_REPL_CTXT);
1737 llog_sync(ctxt, dexp);
1739 } while (fglog != NULL);
1742 /* also incredibly similar to mds_disconnect */
1743 static int filter_disconnect(struct obd_export *exp, int flags)
1745 struct obd_device *obd = exp->exp_obd;
1746 unsigned long irqflags;
1751 class_export_get(exp);
1753 spin_lock_irqsave(&exp->exp_lock, irqflags);
1754 exp->exp_flags = flags;
1755 spin_unlock_irqrestore(&exp->exp_lock, irqflags);
1757 if (!(flags & OBD_OPT_FORCE))
1758 filter_grant_sanity_check(obd, __FUNCTION__);
1759 filter_grant_discard(exp);
1761 /* Disconnect early so that clients can't keep using export */
1762 rc = class_disconnect(exp, flags);
1764 ldlm_cancel_locks_for_export(exp);
1766 fsfilt_sync(obd, obd->u.filter.fo_sb);
1768 /* flush any remaining cancel messages out to the target */
1769 filter_sync_llogs(exp);
1771 class_export_put(exp);
1775 struct dentry *__filter_oa2dentry(struct obd_device *obd,
1776 struct obdo *oa, const char *what)
1778 struct dentry *dchild = NULL;
1781 if (oa->o_valid & OBD_MD_FLGROUP)
1784 dchild = filter_fid2dentry(obd, NULL, group, oa->o_id);
1786 if (IS_ERR(dchild)) {
1787 CERROR("%s error looking up object: "LPU64"\n", what, oa->o_id);
1791 if (dchild->d_inode == NULL) {
1792 CERROR("%s: %s on non-existent object: "LPU64"\n",
1793 obd->obd_name, what, oa->o_id);
1795 RETURN(ERR_PTR(-ENOENT));
1801 static int filter_getattr(struct obd_export *exp, struct obdo *oa,
1802 struct lov_stripe_md *md)
1804 struct dentry *dentry = NULL;
1805 struct obd_device *obd;
1809 obd = class_exp2obd(exp);
1811 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1812 exp->exp_handle.h_cookie);
1816 dentry = filter_oa2dentry(obd, oa);
1818 RETURN(PTR_ERR(dentry));
1820 /* Limit the valid bits in the return data to what we actually use */
1821 oa->o_valid = OBD_MD_FLID;
1822 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
1828 /* this is called from filter_truncate() until we have filter_punch() */
1829 static int filter_setattr(struct obd_export *exp, struct obdo *oa,
1830 struct lov_stripe_md *md, struct obd_trans_info *oti)
1832 struct lvfs_run_ctxt saved;
1833 struct filter_obd *filter;
1834 struct dentry *dentry;
1836 struct ldlm_res_id res_id = { .name = { oa->o_id, 0, oa->o_gr, 0 } };
1837 struct ldlm_resource *res;
1842 LASSERT(oti != NULL);
1844 dentry = filter_oa2dentry(exp->exp_obd, oa);
1846 RETURN(PTR_ERR(dentry));
1848 filter = &exp->exp_obd->u.filter;
1850 iattr_from_obdo(&iattr, oa, oa->o_valid);
1852 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
1855 if (iattr.ia_valid & ATTR_SIZE)
1856 down(&dentry->d_inode->i_sem);
1857 handle = fsfilt_start(exp->exp_obd, dentry->d_inode, FSFILT_OP_SETATTR,
1860 GOTO(out_unlock, rc = PTR_ERR(handle));
1862 /* XXX this could be a rwsem instead, if filter_preprw played along */
1863 if (iattr.ia_valid & ATTR_ATTR_FLAG)
1864 rc = fsfilt_iocontrol(exp->exp_obd, dentry->d_inode, NULL,
1866 (long)&iattr.ia_attr_flags);
1868 rc = fsfilt_setattr(exp->exp_obd, dentry, handle, &iattr, 1);
1869 rc = filter_finish_transno(exp, oti, rc);
1870 rc2 = fsfilt_commit(exp->exp_obd, filter->fo_sb, dentry->d_inode, handle, 0);
1872 CERROR("error on commit, err = %d\n", rc2);
1877 if (iattr.ia_valid & ATTR_SIZE) {
1878 res = ldlm_resource_get(exp->exp_obd->obd_namespace, NULL,
1879 res_id, LDLM_EXTENT, 0);
1881 CERROR("!!! resource_get failed for object "LPU64" -- "
1882 "filter_setattr with no lock?\n", oa->o_id);
1884 if (res->lr_namespace->ns_lvbo &&
1885 res->lr_namespace->ns_lvbo->lvbo_update) {
1886 rc = res->lr_namespace->ns_lvbo->lvbo_update
1889 ldlm_resource_putref(res);
1893 oa->o_valid = OBD_MD_FLID;
1894 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
1897 if (iattr.ia_valid & ATTR_SIZE)
1898 up(&dentry->d_inode->i_sem);
1900 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
1906 /* XXX identical to osc_unpackmd */
1907 static int filter_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
1908 struct lov_mds_md *lmm, int lmm_bytes)
1914 if (lmm_bytes < sizeof (*lmm)) {
1915 CERROR("lov_mds_md too small: %d, need %d\n",
1916 lmm_bytes, (int)sizeof(*lmm));
1919 /* XXX LOV_MAGIC etc check? */
1921 if (lmm->lmm_object_id == cpu_to_le64(0)) {
1922 CERROR("lov_mds_md: zero lmm_object_id\n");
1927 lsm_size = lov_stripe_md_size(1);
1931 if (*lsmp != NULL && lmm == NULL) {
1932 OBD_FREE(*lsmp, lsm_size);
1937 if (*lsmp == NULL) {
1938 OBD_ALLOC(*lsmp, lsm_size);
1942 loi_init((*lsmp)->lsm_oinfo);
1946 /* XXX zero *lsmp? */
1947 (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
1948 LASSERT((*lsmp)->lsm_object_id);
1951 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
1956 static void filter_destroy_precreated(struct obd_export *exp, struct obdo *oa,
1957 struct filter_obd *filter)
1959 struct obdo doa; /* XXX obdo on stack */
1964 memset(&doa, 0, sizeof(doa));
1965 if (oa->o_valid & OBD_MD_FLGROUP) {
1966 doa.o_valid |= OBD_MD_FLGROUP;
1967 doa.o_gr = oa->o_gr;
1971 doa.o_mode = S_IFREG;
1972 doa.o_gr = oa->o_gr;
1973 doa.o_valid = oa->o_valid & OBD_MD_FLGROUP;
1975 filter->fo_destroy_in_progress = 1;
1976 down(&filter->fo_create_lock);
1977 if (!filter->fo_destroy_in_progress) {
1978 CERROR("%s: destroy_in_progress already cleared\n",
1979 exp->exp_obd->obd_name);
1980 up(&filter->fo_create_lock);
1985 last = filter_last_id(filter, doa.o_gr);
1986 CWARN("%s: deleting orphan objects from "LPU64" to "LPU64"\n",
1987 exp->exp_obd->obd_name, oa->o_id + 1, last);
1988 for (id = oa->o_id + 1; id <= last; id++) {
1990 filter_destroy(exp, &doa, NULL, NULL);
1993 CDEBUG(D_HA, "%s: after destroy: set last_objids["LPU64"] = "LPU64"\n",
1994 exp->exp_obd->obd_name, doa.o_gr, oa->o_id);
1996 spin_lock(&filter->fo_objidlock);
1997 filter->fo_last_objids[doa.o_gr] = oa->o_id;
1998 spin_unlock(&filter->fo_objidlock);
2000 filter->fo_destroy_in_progress = 0;
2001 up(&filter->fo_create_lock);
2006 /* returns a negative error or a nonnegative number of files to create */
2007 static int filter_should_precreate(struct obd_export *exp, struct obdo *oa,
2010 struct obd_device *obd = exp->exp_obd;
2011 struct filter_obd *filter = &obd->u.filter;
2015 diff = oa->o_id - filter_last_id(filter, oa->o_gr);
2016 CDEBUG(D_INFO, "filter_last_id() = "LPU64" -> diff = %d\n",
2017 filter_last_id(filter, oa->o_gr), diff);
2019 /* delete orphans request */
2020 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2021 (oa->o_flags & OBD_FL_DELORPHAN)) {
2024 if (-diff > 10000) { /* XXX make this smarter */
2025 CERROR("ignoring bogus orphan destroy request: obdid "
2026 LPU64" last_id "LPU64"\n",
2027 oa->o_id, filter_last_id(filter, oa->o_gr));
2030 filter_destroy_precreated(exp, oa, filter);
2031 rc = filter_update_last_objid(obd, group, 0);
2033 CERROR("unable to write lastobjid, but orphans"
2037 /* only precreate if group == 0 and o_id is specfied */
2038 if (!(oa->o_valid & OBD_FL_DELORPHAN) &&
2039 (group != 0 || oa->o_id == 0))
2046 static int filter_precreate_rec(struct obd_device *obd, struct dentry *dentry,
2047 int *number, struct obdo *oa)
2052 rc = fsfilt_precreate_rec(obd, dentry, number, oa);
2057 /* We rely on the fact that only one thread will be creating files in a given
2058 * group at a time, which is why we don't need an atomic filter_get_new_id.
2059 * Even if we had that atomic function, the following race would exist:
2061 * thread 1: gets id x from filter_next_id
2062 * thread 2: gets id (x + 1) from filter_next_id
2063 * thread 2: creates object (x + 1)
2064 * thread 1: tries to create object x, gets -ENOSPC
2066 static int filter_precreate(struct obd_device *obd, struct obdo *oa,
2067 obd_gr group, int *num)
2069 struct dentry *dchild = NULL, *dparent = NULL;
2070 struct filter_obd *filter;
2071 int err = 0, rc = 0, recreate_obj = 0, i;
2073 void *handle = NULL;
2076 filter = &obd->u.filter;
2078 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2079 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
2083 CDEBUG(D_HA, "%s: precreating %d objects\n", obd->obd_name, *num);
2085 down(&filter->fo_create_lock);
2087 for (i = 0; i < *num && err == 0; i++) {
2088 int cleanup_phase = 0;
2090 if (filter->fo_destroy_in_progress) {
2091 CWARN("%s: precreate aborted by destroy\n",
2099 last_id = filter_last_id(filter, group);
2100 if (next_id > last_id) {
2101 CERROR("Error: Trying to recreate obj greater"
2102 "than last id "LPD64" > "LPD64"\n",
2104 GOTO(cleanup, rc = -EINVAL);
2107 next_id = filter_last_id(filter, group) + 1;
2110 CDEBUG(D_INFO, "precreate objid "LPU64"\n", next_id);
2112 dparent = filter_parent_lock(obd, group, next_id);
2113 if (IS_ERR(dparent))
2114 GOTO(cleanup, rc = PTR_ERR(dparent));
2117 /*only do precreate rec record. so clean kml flags here*/
2118 fsfilt_clear_kml_flags(obd, dparent->d_inode);
2120 dchild = filter_fid2dentry(obd, dparent, group, next_id);
2122 GOTO(cleanup, rc = PTR_ERR(dchild));
2125 if (dchild->d_inode != NULL) {
2126 /* This would only happen if lastobjid was bad on disk*/
2127 /* Could also happen if recreating missing obj but
2131 CERROR("%s: Serious error: recreating obj %*s "
2132 "but obj already exists \n",
2133 obd->obd_name, dchild->d_name.len,
2134 dchild->d_name.name);
2137 CERROR("%s: Serious error: objid %*s already "
2138 "exists; is this filesystem corrupt?\n",
2139 obd->obd_name, dchild->d_name.len,
2140 dchild->d_name.name);
2143 GOTO(cleanup, rc = -EEXIST);
2146 handle = fsfilt_start_log(obd, dparent->d_inode,
2147 FSFILT_OP_CREATE, NULL, 1);
2149 GOTO(cleanup, rc = PTR_ERR(handle));
2152 rc = ll_vfs_create(dparent->d_inode, dchild, S_IFREG, NULL);
2154 CERROR("create failed rc = %d\n", rc);
2158 if (!recreate_obj) {
2159 filter_set_last_id(filter, group, next_id);
2160 err = filter_update_last_objid(obd, group, 0);
2162 CERROR("unable to write lastobjid "
2163 "but file created\n");
2165 fsfilt_set_kml_flags(obd, dparent->d_inode);
2168 switch(cleanup_phase) {
2170 err = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode, handle, 0);
2172 CERROR("error on commit, err = %d\n", err);
2179 filter_parent_unlock(dparent);
2190 /* check if we have an error after ll_vfs_create(). It is possible that
2191 * there will be say -ENOSPC and we will leak it. */
2193 rc = filter_precreate_rec(obd, dparent, num, oa);
2195 up(&filter->fo_create_lock);
2197 CDEBUG(D_HA, "%s: server last_objid for group "LPU64": "LPU64"\n",
2198 obd->obd_name, group, filter->fo_last_objids[group]);
2200 CDEBUG(D_HA, "%s: filter_precreate() created %d objects\n",
2206 static int filter_create(struct obd_export *exp, struct obdo *oa,
2207 struct lov_stripe_md **ea, struct obd_trans_info *oti)
2209 struct obd_device *obd = NULL;
2210 struct filter_obd *filter;
2211 struct lvfs_run_ctxt saved;
2212 struct lov_stripe_md *lsm = NULL;
2213 struct filter_export_data *fed;
2214 char str[PTL_NALFMT_SIZE];
2215 int group = oa->o_gr, rc = 0, diff, recreate_objs = 0;
2218 if (!(oa->o_valid & OBD_MD_FLGROUP) || group == 0) {
2219 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2220 exp->exp_connection->c_peer.peer_nid, str);
2221 CERROR("!!! nid "LPX64"/%s sent invalid object group %d\n",
2222 exp->exp_connection->c_peer.peer_nid, str, group);
2226 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2227 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
2232 fed = &exp->exp_filter_data;
2233 filter = &obd->u.filter;
2235 if (fed->fed_group != group && !recreate_objs &&
2236 !(oa->o_valid & OBD_MD_REINT)) {
2237 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2238 exp->exp_connection->c_peer.peer_nid, str);
2239 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
2240 "earlier; now it's trying to use group %d! This could "
2241 "be a bug in the MDS. Tell CFS.\n",
2242 exp->exp_connection->c_peer.peer_nid, str,
2243 fed->fed_group, group);
2247 CDEBUG(D_INFO, "filter_create(od->o_gr=%d,od->o_id="LPU64")\n",
2252 rc = obd_alloc_memmd(exp, &lsm);
2259 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2261 if (oa->o_valid & OBD_MD_REINT) {
2262 int num = *((int*)oa->o_inline);
2263 rc = filter_precreate(obd, oa, oa->o_gr, &num);
2264 } else if (recreate_objs) {
2265 if (oa->o_id > filter_last_id(&obd->u.filter, group)) {
2266 CERROR("recreate objid "LPU64" > last id "LPU64"\n",
2267 oa->o_id, filter_last_id(&obd->u.filter, group));
2271 rc = filter_precreate(obd, oa, group, &diff);
2274 diff = filter_should_precreate(exp, oa, group);
2276 oa->o_id = filter_last_id(&obd->u.filter, group);
2277 rc = filter_precreate(obd, oa, group, &diff);
2279 oa->o_valid = OBD_MD_FLID;
2283 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2284 if (rc && ea != NULL && *ea != lsm) {
2285 obd_free_memmd(exp, &lsm);
2286 } else if (rc == 0 && ea != NULL) {
2287 /* XXX LOV STACKING: the lsm that is passed to us from
2288 * LOV does not have valid lsm_oinfo data structs, so
2289 * don't go touching that. This needs to be fixed in a
2291 lsm->lsm_object_id = oa->o_id;
2292 lsm->lsm_object_gr = oa->o_gr;
2299 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
2300 struct lov_stripe_md *ea, struct obd_trans_info *oti)
2302 struct obd_device *obd;
2303 struct filter_obd *filter;
2304 struct dentry *dchild = NULL, *dparent = NULL;
2305 struct lvfs_run_ctxt saved;
2306 void *handle = NULL;
2307 struct llog_cookie *fcc = NULL;
2308 int rc, rc2, cleanup_phase = 0, have_prepared = 0;
2311 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2314 filter = &obd->u.filter;
2316 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2319 dparent = filter_parent_lock(obd, oa->o_gr, oa->o_id);
2320 if (IS_ERR(dparent))
2321 GOTO(cleanup, rc = PTR_ERR(dparent));
2324 dchild = filter_fid2dentry(obd, dparent, oa->o_gr, oa->o_id);
2326 GOTO(cleanup, rc = -ENOENT);
2329 if (dchild->d_inode == NULL) {
2330 CDEBUG(D_INODE, "destroying non-existent object "LPU64"\n",
2332 GOTO(cleanup, rc = -ENOENT);
2335 if (!have_prepared) {
2336 /* If we're really going to destroy the object, get ready
2337 * by getting the clients to discard their cached data.
2339 * We have to drop the parent lock, because
2340 * filter_prepare_destroy will acquire a PW on the object, and
2341 * we don't want to deadlock with an incoming write to the
2342 * object, which has the extent PW and then wants to get the
2343 * parent dentry to do the lookup.
2345 * We dput the child because it's not worth the extra
2346 * complication of condition the above code to skip it on the
2347 * second time through. */
2349 filter_parent_unlock(dparent);
2351 filter_prepare_destroy(obd, oa->o_id, oa->o_gr);
2356 handle = fsfilt_start_log(obd, dparent->d_inode, FSFILT_OP_UNLINK, oti, 1);
2358 GOTO(cleanup, rc = PTR_ERR(handle));
2362 /* Our MDC connection is established by the MDS to us */
2363 if (oa->o_valid & OBD_MD_FLCOOKIE) {
2364 OBD_ALLOC(fcc, sizeof(*fcc));
2366 memcpy(fcc, obdo_logcookie(oa), sizeof(*fcc));
2369 rc = filter_destroy_internal(obd, oa->o_id, dparent, dchild);
2372 switch(cleanup_phase) {
2376 fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
2378 filter_cancel_cookies_cb,
2381 fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
2383 filter_cancel_cookies_cb,
2386 rc = filter_finish_transno(exp, oti, rc);
2387 rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode,
2390 CERROR("error on commit, err = %d\n", rc2);
2397 filter_parent_unlock(dparent);
2399 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2402 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
2409 /* NB start and end are used for punch, but not truncate */
2410 static int filter_truncate(struct obd_export *exp, struct obdo *oa,
2411 struct lov_stripe_md *lsm,
2412 obd_off start, obd_off end,
2413 struct obd_trans_info *oti)
2418 if (end != OBD_OBJECT_EOF)
2419 CERROR("PUNCH not supported, only truncate: end = "LPX64"\n",
2422 CDEBUG(D_INODE, "calling truncate for object "LPU64", valid = %x, "
2423 "o_size = "LPD64"\n", oa->o_id, oa->o_valid, start);
2425 error = filter_setattr(exp, oa, NULL, oti);
2429 static int filter_sync(struct obd_export *exp, struct obdo *oa,
2430 struct lov_stripe_md *lsm, obd_off start, obd_off end)
2432 struct obd_device *obd = exp->exp_obd;
2433 struct lvfs_run_ctxt saved;
2434 struct filter_obd *filter;
2435 struct dentry *dentry;
2436 struct llog_ctxt *ctxt;
2440 filter = &obd->u.filter;
2442 /* an objid of zero is taken to mean "sync whole filesystem" */
2443 if (!oa || !(oa->o_valid & OBD_MD_FLID)) {
2444 rc = fsfilt_sync(obd, filter->fo_sb);
2445 /* flush any remaining cancel messages out to the target */
2446 ctxt = llog_get_context(&obd->obd_llogs, LLOG_UNLINK_REPL_CTXT);
2447 llog_sync(ctxt, exp);
2451 dentry = filter_oa2dentry(obd, oa);
2453 RETURN(PTR_ERR(dentry));
2455 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2457 down(&dentry->d_inode->i_sem);
2458 rc = filemap_fdatasync(dentry->d_inode->i_mapping);
2460 /* just any file to grab fsync method - "file" arg unused */
2461 struct file *file = filter->fo_rcvd_filp;
2463 if (file->f_op && file->f_op->fsync)
2464 rc = file->f_op->fsync(NULL, dentry, 1);
2466 rc2 = filemap_fdatawait(dentry->d_inode->i_mapping);
2470 up(&dentry->d_inode->i_sem);
2472 oa->o_valid = OBD_MD_FLID;
2473 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2475 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2481 static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2482 unsigned long max_age)
2484 struct filter_obd *filter = &obd->u.filter;
2485 int blockbits = filter->fo_sb->s_blocksize_bits;
2489 /* at least try to account for cached pages. its still racey and
2490 * might be under-reporting if clients haven't announced their
2491 * caches with brw recently */
2492 spin_lock(&obd->obd_osfs_lock);
2493 rc = fsfilt_statfs(obd, filter->fo_sb, max_age);
2494 memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
2495 spin_unlock(&obd->obd_osfs_lock);
2497 CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
2498 " pending "LPU64" free "LPU64" avail "LPU64"\n",
2499 filter->fo_tot_dirty, filter->fo_tot_granted,
2500 filter->fo_tot_pending,
2501 osfs->os_bfree << blockbits, osfs->os_bavail << blockbits);
2503 filter_grant_sanity_check(obd, __FUNCTION__);
2505 osfs->os_bavail -= min(osfs->os_bavail,
2506 (filter->fo_tot_dirty + filter->fo_tot_pending +
2507 osfs->os_bsize -1) >> blockbits);
2512 static int filter_get_info(struct obd_export *exp, __u32 keylen,
2513 void *key, __u32 *vallen, void *val)
2515 struct filter_export_data *fed = &exp->exp_filter_data;
2516 struct obd_device *obd;
2519 obd = class_exp2obd(exp);
2521 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2522 exp->exp_handle.h_cookie);
2526 if (keylen == strlen("blocksize") &&
2527 memcmp(key, "blocksize", keylen) == 0) {
2528 __u32 *blocksize = val;
2529 *vallen = sizeof(*blocksize);
2530 *blocksize = obd->u.filter.fo_sb->s_blocksize;
2534 if (keylen == strlen("blocksize_bits") &&
2535 memcmp(key, "blocksize_bits", keylen) == 0) {
2536 __u32 *blocksize_bits = val;
2537 *vallen = sizeof(*blocksize_bits);
2538 *blocksize_bits = obd->u.filter.fo_sb->s_blocksize_bits;
2542 if (keylen >= strlen("last_id") && memcmp(key, "last_id", 7) == 0) {
2543 obd_id *last_id = val;
2544 *last_id = filter_last_id(&obd->u.filter, fed->fed_group);
2547 if (keylen >= strlen("reint_log") && memcmp(key, "reint_log", 9) == 0) {
2548 /*Get log_context handle*/
2549 unsigned long *llh_handle = val;
2550 *vallen = sizeof(unsigned long);
2551 *llh_handle = (unsigned long)obd->obd_llog_ctxt[LLOG_REINT_ORIG_CTXT];
2554 if (keylen >= strlen("cache_sb") && memcmp(key, "cache_sb", 8) == 0) {
2555 /*Get log_context handle*/
2556 unsigned long *sb = val;
2557 *vallen = sizeof(unsigned long);
2558 *sb = (unsigned long)obd->u.filter.fo_sb;
2562 CDEBUG(D_IOCTL, "invalid key\n");
2566 struct obd_llogs *filter_grab_llog_for_group(struct obd_device *obd, int group)
2568 struct filter_group_llog *fglog, *nlog;
2569 char name[32] = "CATLIST";
2570 struct filter_obd *filter;
2571 struct list_head *cur;
2574 filter = &obd->u.filter;
2576 spin_lock(&filter->fo_llog_list_lock);
2577 list_for_each(cur, &filter->fo_llog_list) {
2578 fglog = list_entry(cur, struct filter_group_llog, list);
2579 if (fglog->group == group) {
2580 spin_unlock(&filter->fo_llog_list_lock);
2581 RETURN(fglog->llogs);
2584 spin_unlock(&filter->fo_llog_list_lock);
2586 OBD_ALLOC(fglog, sizeof(*fglog));
2589 fglog->group = group;
2591 OBD_ALLOC(fglog->llogs, sizeof(struct obd_llogs));
2592 if (fglog->llogs == NULL) {
2593 OBD_FREE(fglog, sizeof(*fglog));
2597 spin_lock(&filter->fo_llog_list_lock);
2598 list_for_each(cur, &filter->fo_llog_list) {
2599 nlog = list_entry(cur, struct filter_group_llog, list);
2600 LASSERT(nlog->group != group);
2602 list_add(&fglog->list, &filter->fo_llog_list);
2603 spin_unlock(&filter->fo_llog_list_lock);
2605 rc = obd_llog_cat_initialize(obd, fglog->llogs, 1, name);
2607 OBD_FREE(fglog->llogs, sizeof(*(fglog->llogs)));
2608 OBD_FREE(fglog, sizeof(*fglog));
2612 CDEBUG(D_OTHER, "%s: new llog 0x%p for group %u\n", obd->obd_name,
2613 fglog->llogs, group);
2615 RETURN(fglog->llogs);
2618 static int filter_set_info(struct obd_export *exp, __u32 keylen,
2619 void *key, __u32 vallen, void *val)
2621 struct lvfs_run_ctxt saved;
2622 struct filter_export_data *fed = &exp->exp_filter_data;
2623 struct obd_device *obd;
2624 struct lustre_handle conn;
2625 struct obd_llogs *llog;
2626 struct llog_ctxt *ctxt;
2631 conn.cookie = exp->exp_handle.h_cookie;
2635 CDEBUG(D_IOCTL, "invalid exp %p cookie "LPX64"\n",
2640 if (keylen < strlen("mds_conn") ||
2641 memcmp(key, "mds_conn", keylen) != 0)
2644 group = *((__u32 *)val);
2645 if (fed->fed_group != 0 && fed->fed_group != group) {
2646 char str[PTL_NALFMT_SIZE];
2647 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2648 exp->exp_connection->c_peer.peer_nid, str);
2649 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
2650 "earlier; now it's trying to use group %d! This could "
2651 "be a bug in the MDS. Tell CFS.\n",
2652 exp->exp_connection->c_peer.peer_nid, str,
2653 fed->fed_group, group);
2656 fed->fed_group = group;
2657 CWARN("Received MDS connection ("LPX64"); group %d\n", conn.cookie,
2662 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2663 rc = filter_read_groups(obd, group, 1);
2664 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2666 CERROR("can't read group %u\n", group);
2669 rc = filter_group_set_kml_flags(obd, group);
2671 CERROR("can't set kml flags %u\n", group);
2674 llog = filter_grab_llog_for_group(obd, group);
2675 LASSERT(llog != NULL);
2677 ctxt = llog_get_context(llog, LLOG_UNLINK_REPL_CTXT);
2678 LASSERT(ctxt != NULL);
2679 rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse);
2683 int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
2684 int len, void *karg, void *uarg)
2686 struct obd_device *obd = exp->exp_obd;
2687 struct obd_ioctl_data *data = karg;
2691 case OBD_IOC_ABORT_RECOVERY:
2692 CERROR("aborting recovery for device %s\n", obd->obd_name);
2693 target_abort_recovery(obd);
2696 case OBD_IOC_SET_READONLY: {
2698 struct super_block *sb = obd->u.filter.fo_sb;
2699 struct inode *inode = sb->s_root->d_inode;
2700 BDEVNAME_DECLARE_STORAGE(tmp);
2701 CERROR("setting device %s read-only\n",
2702 ll_bdevname(sb, tmp));
2704 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
2706 (void)fsfilt_commit(obd, sb, inode, handle, 1);
2708 dev_set_rdonly(ll_sbdev(obd->u.filter.fo_sb), 2);
2712 case OBD_IOC_CATLOGLIST: {
2713 rc = llog_catalog_list(obd, 1, data);
2717 case OBD_IOC_LLOG_CANCEL:
2718 case OBD_IOC_LLOG_REMOVE:
2719 case OBD_IOC_LLOG_INFO:
2720 case OBD_IOC_LLOG_PRINT: {
2721 /* FIXME to be finished */
2722 RETURN(-EOPNOTSUPP);
2724 struct llog_ctxt *ctxt = NULL;
2726 push_ctxt(&saved, &ctxt->loc_ctxt, NULL);
2727 rc = llog_ioctl(ctxt, cmd, data);
2728 pop_ctxt(&saved, &ctxt->loc_ctxt, NULL);
2741 static struct llog_operations filter_unlink_repl_logops;
2742 static struct llog_operations filter_size_orig_logops = {
2743 lop_setup: llog_obd_origin_setup,
2744 lop_cleanup: llog_catalog_cleanup,
2745 lop_add: llog_catalog_add,
2748 static int filter_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
2749 struct obd_device *tgt, int count,
2750 struct llog_catid *catid)
2752 struct llog_ctxt *ctxt;
2756 filter_unlink_repl_logops = llog_client_ops;
2757 filter_unlink_repl_logops.lop_cancel = llog_obd_repl_cancel;
2758 filter_unlink_repl_logops.lop_connect = llog_repl_connect;
2759 filter_unlink_repl_logops.lop_sync = llog_obd_repl_sync;
2761 rc = obd_llog_setup(obd, llogs, LLOG_UNLINK_REPL_CTXT, tgt, 0, NULL,
2762 &filter_unlink_repl_logops);
2765 /* FIXME - assign unlink_cb for filter's recovery */
2766 ctxt = llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT);
2767 ctxt->llog_proc_cb = filter_recov_log_unlink_cb;
2769 /* FIXME - count should be 1 to setup size log */
2770 rc = obd_llog_setup(obd, llogs, LLOG_SIZE_ORIG_CTXT, tgt, 0,
2771 &catid->lci_logid, &filter_size_orig_logops);
2775 static int filter_llog_finish(struct obd_device *obd,
2776 struct obd_llogs *llogs, int count)
2781 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT));
2785 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_SIZE_ORIG_CTXT));
2789 static int filter_llog_connect(struct obd_device *obd,
2790 struct llogd_conn_body *body)
2792 struct llog_ctxt *ctxt;
2793 struct obd_llogs *llog;
2797 CDEBUG(D_OTHER, "handle connect for %s: %u/%u/%u\n", obd->obd_name,
2798 (unsigned) body->lgdc_logid.lgl_ogr,
2799 (unsigned) body->lgdc_logid.lgl_oid,
2800 (unsigned) body->lgdc_logid.lgl_ogen);
2801 llog = filter_grab_llog_for_group(obd, body->lgdc_logid.lgl_ogr);
2802 LASSERT(llog != NULL);
2803 ctxt = llog_get_context(llog, body->lgdc_ctxt_idx);
2804 rc = llog_connect(ctxt, 1, &body->lgdc_logid,
2805 &body->lgdc_gen, NULL);
2807 CERROR("failed to connect\n");
2812 static struct dentry *filter_lvfs_fid2dentry(__u64 id, __u32 gen, __u64 gr,
2815 return filter_fid2dentry(data, NULL, gr, id);
2818 static struct lvfs_callback_ops filter_lvfs_ops = {
2819 l_fid2dentry: filter_lvfs_fid2dentry,
2822 static struct obd_ops filter_obd_ops = {
2823 .o_owner = THIS_MODULE,
2824 .o_attach = filter_attach,
2825 .o_detach = filter_detach,
2826 .o_get_info = filter_get_info,
2827 .o_set_info = filter_set_info,
2828 .o_setup = filter_setup,
2829 .o_precleanup = filter_precleanup,
2830 .o_cleanup = filter_cleanup,
2831 .o_connect = filter_connect,
2832 .o_disconnect = filter_disconnect,
2833 .o_statfs = filter_statfs,
2834 .o_getattr = filter_getattr,
2835 .o_unpackmd = filter_unpackmd,
2836 .o_create = filter_create,
2837 .o_setattr = filter_setattr,
2838 .o_destroy = filter_destroy,
2839 .o_brw = filter_brw,
2840 .o_punch = filter_truncate,
2841 .o_sync = filter_sync,
2842 .o_preprw = filter_preprw,
2843 .o_commitrw = filter_commitrw,
2844 .o_write_extents = filter_write_extents,
2845 .o_destroy_export = filter_destroy_export,
2846 .o_llog_init = filter_llog_init,
2847 .o_llog_finish = filter_llog_finish,
2848 .o_llog_connect = filter_llog_connect,
2849 .o_iocontrol = filter_iocontrol,
2852 static struct obd_ops filter_sanobd_ops = {
2853 .o_owner = THIS_MODULE,
2854 .o_attach = filter_attach,
2855 .o_detach = filter_detach,
2856 .o_get_info = filter_get_info,
2857 .o_set_info = filter_set_info,
2858 .o_setup = filter_san_setup,
2859 .o_precleanup = filter_precleanup,
2860 .o_cleanup = filter_cleanup,
2861 .o_connect = filter_connect,
2862 .o_disconnect = filter_disconnect,
2863 .o_statfs = filter_statfs,
2864 .o_getattr = filter_getattr,
2865 .o_unpackmd = filter_unpackmd,
2866 .o_create = filter_create,
2867 .o_setattr = filter_setattr,
2868 .o_destroy = filter_destroy,
2869 .o_brw = filter_brw,
2870 .o_punch = filter_truncate,
2871 .o_sync = filter_sync,
2872 .o_preprw = filter_preprw,
2873 .o_commitrw = filter_commitrw,
2874 .o_write_extents = filter_write_extents,
2875 .o_san_preprw = filter_san_preprw,
2876 .o_destroy_export = filter_destroy_export,
2877 .o_llog_init = filter_llog_init,
2878 .o_llog_finish = filter_llog_finish,
2879 .o_llog_connect = filter_llog_connect,
2880 .o_iocontrol = filter_iocontrol,
2883 static int __init obdfilter_init(void)
2885 struct lprocfs_static_vars lvars;
2888 printk(KERN_INFO "Lustre: Filtering OBD driver; info@clusterfs.com\n");
2890 lprocfs_init_vars(filter, &lvars);
2892 rc = class_register_type(&filter_obd_ops, NULL, lvars.module_vars,
2893 OBD_FILTER_DEVICENAME);
2897 rc = class_register_type(&filter_sanobd_ops, NULL, lvars.module_vars,
2898 OBD_FILTER_SAN_DEVICENAME);
2900 class_unregister_type(OBD_FILTER_DEVICENAME);
2904 static void __exit obdfilter_exit(void)
2906 class_unregister_type(OBD_FILTER_SAN_DEVICENAME);
2907 class_unregister_type(OBD_FILTER_DEVICENAME);
2910 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2911 MODULE_DESCRIPTION("Lustre Filtering OBD driver");
2912 MODULE_LICENSE("GPL");
2914 module_init(obdfilter_init);
2915 module_exit(obdfilter_exit);