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 <linux/lustre_smfs.h>
59 #include "filter_internal.h"
61 static struct lvfs_callback_ops filter_lvfs_ops;
63 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
64 struct lov_stripe_md *ea, struct obd_trans_info *);
66 static void filter_commit_cb(struct obd_device *obd, __u64 transno,
67 void *cb_data, int error)
69 obd_transno_commit_cb(obd, transno, error);
72 /* Assumes caller has already pushed us into the kernel context. */
73 int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti,
76 struct filter_obd *filter = &exp->exp_obd->u.filter;
77 struct filter_export_data *fed = &exp->exp_filter_data;
78 struct filter_client_data *fcd = fed->fed_fcd;
81 int err, log_pri = D_HA;
83 /* Propagate error code. */
87 if (!exp->exp_obd->obd_replayable || oti == NULL)
90 /* we don't allocate new transnos for replayed requests */
91 if (oti->oti_transno == 0) {
92 spin_lock(&filter->fo_translock);
93 last_rcvd = le64_to_cpu(filter->fo_fsd->fsd_last_transno) + 1;
94 filter->fo_fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
95 spin_unlock(&filter->fo_translock);
96 oti->oti_transno = last_rcvd;
98 spin_lock(&filter->fo_translock);
99 last_rcvd = oti->oti_transno;
100 if (last_rcvd > le64_to_cpu(filter->fo_fsd->fsd_last_transno))
101 filter->fo_fsd->fsd_last_transno =
102 cpu_to_le64(last_rcvd);
103 spin_unlock(&filter->fo_translock);
105 fcd->fcd_last_rcvd = cpu_to_le64(last_rcvd);
107 /* could get xid from oti, if it's ever needed */
108 fcd->fcd_last_xid = 0;
110 off = fed->fed_lr_off;
112 fsfilt_add_journal_cb(exp->exp_obd, filter->fo_sb, last_rcvd,
113 oti->oti_handle, filter_commit_cb, NULL);
115 err = fsfilt_write_record(exp->exp_obd, filter->fo_rcvd_filp, fcd,
116 sizeof(*fcd), &off, 0);
123 CDEBUG(log_pri, "wrote trans "LPU64" for client %s at #%d: err = %d\n",
124 last_rcvd, fcd->fcd_uuid, fed->fed_lr_idx, err);
129 void f_dput(struct dentry *dentry)
131 /* Can't go inside filter_ddelete because it can block */
132 CDEBUG(D_INODE, "putting %s: %p, count = %d\n",
133 dentry->d_name.name, dentry, atomic_read(&dentry->d_count) - 1);
134 LASSERT(atomic_read(&dentry->d_count) > 0);
139 /* Add client data to the FILTER. We use a bitmap to locate a free space
140 * in the last_rcvd file if cl_idx is -1 (i.e. a new client).
141 * Otherwise, we have just read the data from the last_rcvd file and
142 * we know its offset. */
143 static int filter_client_add(struct obd_device *obd, struct filter_obd *filter,
144 struct filter_export_data *fed, int cl_idx)
146 unsigned long *bitmap = filter->fo_last_rcvd_slots;
147 int new_client = (cl_idx == -1);
150 LASSERT(bitmap != NULL);
152 /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
153 if (!strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid))
156 /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
157 * there's no need for extra complication here
160 cl_idx = find_first_zero_bit(bitmap, FILTER_LR_MAX_CLIENTS);
162 if (cl_idx >= FILTER_LR_MAX_CLIENTS) {
163 CERROR("no client slots - fix FILTER_LR_MAX_CLIENTS\n");
166 if (test_and_set_bit(cl_idx, bitmap)) {
167 CERROR("FILTER client %d: found bit is set in bitmap\n",
169 cl_idx = find_next_zero_bit(bitmap,
170 FILTER_LR_MAX_CLIENTS,
175 if (test_and_set_bit(cl_idx, bitmap)) {
176 CERROR("FILTER client %d: bit already set in bitmap!\n",
182 fed->fed_lr_idx = cl_idx;
183 fed->fed_lr_off = le32_to_cpu(filter->fo_fsd->fsd_client_start) +
184 cl_idx * le16_to_cpu(filter->fo_fsd->fsd_client_size);
186 CDEBUG(D_INFO, "client at index %d (%llu) with UUID '%s' added\n",
187 fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
190 struct lvfs_run_ctxt saved;
191 loff_t off = fed->fed_lr_off;
195 CDEBUG(D_INFO, "writing client fcd at idx %u (%llu) (len %u)\n",
196 fed->fed_lr_idx,off,(unsigned int)sizeof(*fed->fed_fcd));
198 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
199 /* Transaction needed to fix bug 1403 */
200 handle = fsfilt_start(obd,
201 filter->fo_rcvd_filp->f_dentry->d_inode,
202 FSFILT_OP_SETATTR, NULL);
203 if (IS_ERR(handle)) {
204 err = PTR_ERR(handle);
205 CERROR("unable to start transaction: rc %d\n", err);
207 err = fsfilt_write_record(obd, filter->fo_rcvd_filp,
209 sizeof(*fed->fed_fcd),
211 fsfilt_commit(obd, filter->fo_sb,
212 filter->fo_rcvd_filp->f_dentry->d_inode,
215 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
218 CERROR("error writing %s client idx %u: rc %d\n",
219 LAST_RCVD, fed->fed_lr_idx, err);
226 static int filter_client_free(struct obd_export *exp, int flags)
228 struct filter_export_data *fed = &exp->exp_filter_data;
229 struct filter_obd *filter = &exp->exp_obd->u.filter;
230 struct obd_device *obd = exp->exp_obd;
231 struct filter_client_data zero_fcd;
232 struct lvfs_run_ctxt saved;
237 if (fed->fed_fcd == NULL)
240 if (flags & OBD_OPT_FAILOVER)
243 /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
244 if (strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid ) == 0)
247 LASSERT(filter->fo_last_rcvd_slots != NULL);
249 off = fed->fed_lr_off;
251 CDEBUG(D_INFO, "freeing client at idx %u (%lld) with UUID '%s'\n",
252 fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
254 /* Clear the bit _after_ zeroing out the client so we don't
255 race with filter_client_add and zero out new clients.*/
256 if (!test_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
257 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
262 memset(&zero_fcd, 0, sizeof zero_fcd);
263 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
264 rc = fsfilt_write_record(obd, filter->fo_rcvd_filp, &zero_fcd,
265 sizeof(zero_fcd), &off, 1);
266 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
268 CDEBUG(rc == 0 ? D_INFO : D_ERROR,
269 "zeroing disconnecting client %s at idx %u (%llu) in %s rc %d\n",
270 fed->fed_fcd->fcd_uuid, fed->fed_lr_idx, fed->fed_lr_off,
273 if (!test_and_clear_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
274 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
280 OBD_FREE(fed->fed_fcd, sizeof(*fed->fed_fcd));
285 static int filter_free_server_data(struct filter_obd *filter)
287 OBD_FREE(filter->fo_fsd, sizeof(*filter->fo_fsd));
288 filter->fo_fsd = NULL;
289 OBD_FREE(filter->fo_last_rcvd_slots,
290 FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
291 filter->fo_last_rcvd_slots = NULL;
295 /* assumes caller is already in kernel ctxt */
296 int filter_update_server_data(struct obd_device *obd, struct file *filp,
297 struct filter_server_data *fsd, int force_sync)
303 CDEBUG(D_INODE, "server uuid : %s\n", fsd->fsd_uuid);
304 CDEBUG(D_INODE, "server last_rcvd : "LPU64"\n",
305 le64_to_cpu(fsd->fsd_last_transno));
306 CDEBUG(D_INODE, "server last_mount: "LPU64"\n",
307 le64_to_cpu(fsd->fsd_mount_count));
309 rc = fsfilt_write_record(obd, filp, fsd, sizeof(*fsd), &off,force_sync);
311 CERROR("error writing filter_server_data: rc = %d\n", rc);
316 int filter_update_last_objid(struct obd_device *obd, obd_gr group,
319 struct filter_obd *filter = &obd->u.filter;
325 if (filter->fo_last_objid_files[group] == NULL) {
326 CERROR("Object group "LPU64" not fully setup; not updating "
327 "last_objid\n", group);
331 CDEBUG(D_INODE, "server last_objid for group "LPU64": "LPU64"\n",
332 group, filter->fo_last_objids[group]);
334 tmp = cpu_to_le64(filter->fo_last_objids[group]);
335 rc = fsfilt_write_record(obd, filter->fo_last_objid_files[group],
336 &tmp, sizeof(tmp), &off, force_sync);
338 CERROR("error writing group "LPU64" last objid: rc = %d\n",
343 /* assumes caller has already in kernel ctxt */
344 static int filter_init_server_data(struct obd_device *obd, struct file * filp)
346 struct filter_obd *filter = &obd->u.filter;
347 struct filter_server_data *fsd;
348 struct filter_client_data *fcd = NULL;
349 struct inode *inode = filp->f_dentry->d_inode;
350 unsigned long last_rcvd_size = inode->i_size;
356 /* ensure padding in the struct is the correct size */
357 LASSERT (offsetof(struct filter_server_data, fsd_padding) +
358 sizeof(fsd->fsd_padding) == FILTER_LR_SERVER_SIZE);
359 LASSERT (offsetof(struct filter_client_data, fcd_padding) +
360 sizeof(fcd->fcd_padding) == FILTER_LR_CLIENT_SIZE);
362 OBD_ALLOC(fsd, sizeof(*fsd));
365 filter->fo_fsd = fsd;
367 OBD_ALLOC(filter->fo_last_rcvd_slots,
368 FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
369 if (filter->fo_last_rcvd_slots == NULL) {
370 OBD_FREE(fsd, sizeof(*fsd));
374 if (last_rcvd_size == 0) {
375 CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD);
377 memcpy(fsd->fsd_uuid, obd->obd_uuid.uuid,sizeof(fsd->fsd_uuid));
378 fsd->fsd_last_transno = 0;
379 mount_count = fsd->fsd_mount_count = 0;
380 fsd->fsd_server_size = cpu_to_le32(FILTER_LR_SERVER_SIZE);
381 fsd->fsd_client_start = cpu_to_le32(FILTER_LR_CLIENT_START);
382 fsd->fsd_client_size = cpu_to_le16(FILTER_LR_CLIENT_SIZE);
383 fsd->fsd_subdir_count = cpu_to_le16(FILTER_SUBDIR_COUNT);
384 filter->fo_subdir_count = FILTER_SUBDIR_COUNT;
386 rc = fsfilt_read_record(obd, filp, fsd, sizeof(*fsd), &off);
388 CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
392 if (strcmp(fsd->fsd_uuid, obd->obd_uuid.uuid) != 0) {
393 CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
394 obd->obd_uuid.uuid, fsd->fsd_uuid);
395 GOTO(err_fsd, rc = -EINVAL);
397 mount_count = le64_to_cpu(fsd->fsd_mount_count);
398 filter->fo_subdir_count = le16_to_cpu(fsd->fsd_subdir_count);
401 if (fsd->fsd_feature_incompat & ~cpu_to_le32(FILTER_INCOMPAT_SUPP)) {
402 CERROR("unsupported feature %x\n",
403 le32_to_cpu(fsd->fsd_feature_incompat) &
404 ~FILTER_INCOMPAT_SUPP);
405 GOTO(err_fsd, rc = -EINVAL);
407 if (fsd->fsd_feature_rocompat & ~cpu_to_le32(FILTER_ROCOMPAT_SUPP)) {
408 CERROR("read-only feature %x\n",
409 le32_to_cpu(fsd->fsd_feature_rocompat) &
410 ~FILTER_ROCOMPAT_SUPP);
411 /* Do something like remount filesystem read-only */
412 GOTO(err_fsd, rc = -EINVAL);
415 CDEBUG(D_INODE, "%s: server last_rcvd : "LPU64"\n",
416 obd->obd_name, le64_to_cpu(fsd->fsd_last_transno));
417 CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
418 obd->obd_name, mount_count + 1);
419 CDEBUG(D_INODE, "%s: server data size: %u\n",
420 obd->obd_name, le32_to_cpu(fsd->fsd_server_size));
421 CDEBUG(D_INODE, "%s: per-client data start: %u\n",
422 obd->obd_name, le32_to_cpu(fsd->fsd_client_start));
423 CDEBUG(D_INODE, "%s: per-client data size: %u\n",
424 obd->obd_name, le32_to_cpu(fsd->fsd_client_size));
425 CDEBUG(D_INODE, "%s: server subdir_count: %u\n",
426 obd->obd_name, le16_to_cpu(fsd->fsd_subdir_count));
427 CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
428 last_rcvd_size <= le32_to_cpu(fsd->fsd_client_start) ? 0 :
429 (last_rcvd_size - le32_to_cpu(fsd->fsd_client_start)) /
430 le16_to_cpu(fsd->fsd_client_size));
432 if (!obd->obd_replayable) {
433 CWARN("%s: recovery support OFF\n", obd->obd_name);
437 for (cl_idx = 0, off = le32_to_cpu(fsd->fsd_client_start);
438 off < last_rcvd_size; cl_idx++) {
440 struct obd_export *exp;
441 struct filter_export_data *fed;
444 OBD_ALLOC(fcd, sizeof(*fcd));
446 GOTO(err_client, rc = -ENOMEM);
449 /* Don't assume off is incremented properly by
450 * fsfilt_read_record(), in case sizeof(*fcd)
451 * isn't the same as fsd->fsd_client_size. */
452 off = le32_to_cpu(fsd->fsd_client_start) +
453 cl_idx * le16_to_cpu(fsd->fsd_client_size);
454 rc = fsfilt_read_record(obd, filp, fcd, sizeof(*fcd), &off);
456 CERROR("error reading FILT %s idx %d off %llu: rc %d\n",
457 LAST_RCVD, cl_idx, off, rc);
458 break; /* read error shouldn't cause startup to fail */
461 if (fcd->fcd_uuid[0] == '\0') {
462 CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
467 last_rcvd = le64_to_cpu(fcd->fcd_last_rcvd);
469 /* These exports are cleaned up by filter_disconnect(), so they
470 * need to be set up like real exports as filter_connect() does.
472 exp = class_new_export(obd);
473 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
474 " srv lr: "LPU64"\n", fcd->fcd_uuid, cl_idx,
475 last_rcvd, le64_to_cpu(fsd->fsd_last_transno));
477 GOTO(err_client, rc = -ENOMEM);
479 memcpy(&exp->exp_client_uuid.uuid, fcd->fcd_uuid,
480 sizeof exp->exp_client_uuid.uuid);
481 fed = &exp->exp_filter_data;
483 filter_client_add(obd, filter, fed, cl_idx);
484 /* create helper if export init gets more complex */
485 spin_lock_init(&fed->fed_lock);
488 obd->obd_recoverable_clients++;
489 class_export_put(exp);
491 CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
494 if (last_rcvd > le64_to_cpu(fsd->fsd_last_transno))
495 fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
499 obd->obd_last_committed = le64_to_cpu(fsd->fsd_last_transno);
501 if (obd->obd_recoverable_clients) {
502 CWARN("RECOVERY: %d recoverable clients, last_rcvd "
503 LPU64"\n", obd->obd_recoverable_clients,
504 le64_to_cpu(fsd->fsd_last_transno));
505 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
506 obd->obd_recovering = 1;
510 OBD_FREE(fcd, sizeof(*fcd));
513 filter->fo_mount_count = mount_count + 1;
514 fsd->fsd_mount_count = cpu_to_le64(filter->fo_mount_count);
516 /* save it, so mount count and last_transno is current */
517 rc = filter_update_server_data(obd, filp, filter->fo_fsd, 1);
522 class_disconnect_exports(obd, 0);
524 filter_free_server_data(filter);
528 static int filter_cleanup_groups(struct obd_device *obd)
530 struct filter_obd *filter = &obd->u.filter;
531 struct dentry *dentry;
535 for (i = 0; i < filter->fo_group_count; i++) {
536 if (filter->fo_subdirs != NULL) {
537 for (k = 0; k < filter->fo_subdir_count; k++) {
538 dentry = filter->fo_subdirs[i].dentry[k];
542 filter->fo_subdirs[i].dentry[k] = NULL;
545 if (filter->fo_last_objid_files[i] != NULL) {
546 filp_close(filter->fo_last_objid_files[i], 0);
547 filter->fo_last_objid_files[i] = NULL;
549 if (filter->fo_groups[i] != NULL) {
550 dput(filter->fo_groups[i]);
551 filter->fo_groups[i] = NULL;
554 if (filter->fo_subdirs != NULL)
555 OBD_FREE(filter->fo_subdirs,
556 filter->fo_group_count * sizeof(*filter->fo_subdirs));
557 if (filter->fo_groups != NULL)
558 OBD_FREE(filter->fo_groups,
559 filter->fo_group_count * sizeof(*filter->fo_groups));
560 if (filter->fo_last_objids != NULL)
561 OBD_FREE(filter->fo_last_objids,
562 filter->fo_group_count * sizeof(__u64));
563 if (filter->fo_last_objid_files != NULL)
564 OBD_FREE(filter->fo_last_objid_files,
565 filter->fo_group_count * sizeof(struct file *));
566 f_dput(filter->fo_dentry_O);
570 static int filter_read_group_internal(struct obd_device *obd, int group,
573 struct filter_obd *filter = &obd->u.filter;
574 __u64 *new_objids = NULL;
575 struct filter_subdirs *new_subdirs = NULL, *tmp_subdirs = NULL;
576 struct dentry **new_groups = NULL;
577 struct file **new_files = NULL;
578 struct dentry *dentry;
580 int old_count = filter->fo_group_count, rc, stage = 0, i;
585 snprintf(name, 24, "%d", group);
589 dentry = ll_lookup_one_len(name, filter->fo_dentry_O,
591 if (IS_ERR(dentry)) {
592 CERROR("Cannot lookup expected object group %d: %ld\n",
593 group, PTR_ERR(dentry));
594 RETURN(PTR_ERR(dentry));
597 dentry = simple_mkdir(filter->fo_dentry_O, name, 0700, 1);
598 if (IS_ERR(dentry)) {
599 CERROR("cannot lookup/create O/%s: rc = %ld\n", name,
601 RETURN(PTR_ERR(dentry));
606 snprintf(name, 24, "O/%d/LAST_ID", group);
608 filp = filp_open(name, O_CREAT | O_RDWR, 0700);
610 CERROR("cannot create %s: rc = %ld\n", name, PTR_ERR(filp));
611 GOTO(cleanup, rc = PTR_ERR(filp));
615 rc = fsfilt_read_record(obd, filp, &last_objid, sizeof(__u64), &off);
617 CDEBUG(D_INODE, "error reading %s: rc %d\n", name, rc);
621 if (filter->fo_subdir_count) {
622 OBD_ALLOC(tmp_subdirs, sizeof(*tmp_subdirs));
623 if (tmp_subdirs == NULL)
624 GOTO(cleanup, rc = -ENOMEM);
627 for (i = 0; i < filter->fo_subdir_count; i++) {
629 snprintf(dir, sizeof(dir), "d%u", i);
631 tmp_subdirs->dentry[i] =
632 simple_mkdir(dentry, dir, 0700, 1);
633 if (IS_ERR(tmp_subdirs->dentry[i])) {
634 rc = PTR_ERR(tmp_subdirs->dentry[i]);
635 CERROR("can't lookup/create O/%d/%s: rc = %d\n",
639 CDEBUG(D_INODE, "got/created O/%d/%s: %p\n", group, dir,
640 tmp_subdirs->dentry[i]);
644 /* 'group' is an index; we need an array of length 'group + 1' */
645 if (group + 1 > old_count) {
647 OBD_ALLOC(new_objids, len * sizeof(*new_objids));
648 OBD_ALLOC(new_subdirs, len * sizeof(*new_subdirs));
649 OBD_ALLOC(new_groups, len * sizeof(*new_groups));
650 OBD_ALLOC(new_files, len * sizeof(*new_files));
652 if (new_objids == NULL || new_subdirs == NULL ||
653 new_groups == NULL || new_files == NULL)
654 GOTO(cleanup, rc = -ENOMEM);
656 memcpy(new_objids, filter->fo_last_objids,
657 old_count * sizeof(*new_objids));
658 memcpy(new_subdirs, filter->fo_subdirs,
659 old_count * sizeof(*new_subdirs));
660 memcpy(new_groups, filter->fo_groups,
661 old_count * sizeof(*new_groups));
662 memcpy(new_files, filter->fo_last_objid_files,
663 old_count * sizeof(*new_files));
666 OBD_FREE(filter->fo_last_objids,
667 old_count * sizeof(*new_objids));
668 OBD_FREE(filter->fo_subdirs,
669 old_count * sizeof(*new_subdirs));
670 OBD_FREE(filter->fo_groups,
671 old_count * sizeof(*new_groups));
672 OBD_FREE(filter->fo_last_objid_files,
673 old_count * sizeof(*new_files));
675 filter->fo_last_objids = new_objids;
676 filter->fo_subdirs = new_subdirs;
677 filter->fo_groups = new_groups;
678 filter->fo_last_objid_files = new_files;
679 filter->fo_group_count = len;
682 filter->fo_groups[group] = dentry;
683 filter->fo_last_objid_files[group] = filp;
684 if (filter->fo_subdir_count) {
685 filter->fo_subdirs[group] = *tmp_subdirs;
686 OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
689 if (filp->f_dentry->d_inode->i_size == 0) {
690 filter->fo_last_objids[group] = FILTER_INIT_OBJID;
694 filter->fo_last_objids[group] = le64_to_cpu(last_objid);
695 CDEBUG(D_INODE, "%s: server last_objid group %d: "LPU64"\n",
696 obd->obd_name, group, last_objid);
701 if (new_objids != NULL)
702 OBD_FREE(new_objids, group * sizeof(*new_objids));
703 if (new_subdirs != NULL)
704 OBD_FREE(new_subdirs, group * sizeof(*new_subdirs));
705 if (new_groups != NULL)
706 OBD_FREE(new_groups, group * sizeof(*new_groups));
707 if (new_files != NULL)
708 OBD_FREE(new_files, group * sizeof(*new_files));
710 if (filter->fo_subdir_count) {
711 for (i = 0; i < filter->fo_subdir_count; i++) {
712 if (tmp_subdirs->dentry[i] != NULL)
713 dput(tmp_subdirs->dentry[i]);
715 OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
725 static int filter_read_groups(struct obd_device *obd, int last_group,
728 struct filter_obd *filter = &obd->u.filter;
729 int old_count = filter->fo_group_count, group = old_count, rc = 0;
731 for (group = old_count; group <= last_group; group++) {
733 continue; /* no group zero */
735 rc = filter_read_group_internal(obd, group, create);
742 static int filter_prep_groups(struct obd_device *obd)
744 struct filter_obd *filter = &obd->u.filter;
745 struct dentry *dentry, *O_dentry;
746 int rc = 0, cleanup_phase = 0;
749 O_dentry = simple_mkdir(current->fs->pwd, "O", 0700, 1);
750 CDEBUG(D_INODE, "got/created O: %p\n", O_dentry);
751 if (IS_ERR(O_dentry)) {
752 rc = PTR_ERR(O_dentry);
753 CERROR("cannot open/create O: rc = %d\n", rc);
756 filter->fo_dentry_O = O_dentry;
757 cleanup_phase = 1; /* O_dentry */
759 /* Lookup "R" to tell if we're on an old OST FS and need to convert
760 * from O/R/<dir>/<objid> to O/0/<dir>/<objid>. This can be removed
761 * some time post 1.0 when all old-style OSTs have converted along
762 * with the init_objid hack. */
763 dentry = ll_lookup_one_len("R", O_dentry, 1);
765 GOTO(cleanup, rc = PTR_ERR(dentry));
766 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
767 struct dentry *O0_dentry = lookup_one_len("0", O_dentry, 1);
770 CWARN("converting OST to new object layout\n");
771 if (IS_ERR(O0_dentry)) {
772 rc = PTR_ERR(O0_dentry);
773 CERROR("error looking up O/0: rc %d\n", rc);
777 if (O0_dentry->d_inode) {
778 CERROR("Both O/R and O/0 exist. Fix manually.\n");
779 GOTO(cleanup_O0, rc = -EEXIST);
782 down(&O_dentry->d_inode->i_sem);
783 rc = vfs_rename(O_dentry->d_inode, dentry,
784 O_dentry->d_inode, O0_dentry);
785 up(&O_dentry->d_inode->i_sem);
788 CERROR("error renaming O/R to O/0: rc %d\n", rc);
789 GOTO(cleanup_O0, rc);
791 filter->fo_fsd->fsd_feature_incompat |=
792 cpu_to_le32(FILTER_INCOMPAT_GROUPS);
793 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
795 GOTO(cleanup_O0, rc);
807 cleanup_phase = 2; /* groups */
809 /* Group 0 is no longer a legal group, to catch uninitialized IDs */
810 #define FILTER_MIN_GROUPS 3
811 rc = filter_read_groups(obd, FILTER_MIN_GROUPS, 1);
818 switch (cleanup_phase) {
820 filter_cleanup_groups(obd);
822 f_dput(filter->fo_dentry_O);
823 filter->fo_dentry_O = NULL;
830 /* setup the object store with correct subdirectories */
831 static int filter_prep(struct obd_device *obd)
833 struct lvfs_run_ctxt saved;
834 struct filter_obd *filter = &obd->u.filter;
840 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
841 file = filp_open(LAST_RCVD, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
842 if (!file || IS_ERR(file)) {
844 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
849 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
850 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
851 file->f_dentry->d_inode->i_mode);
852 GOTO(err_filp, rc = -ENOENT);
855 /* steal operations */
856 inode = file->f_dentry->d_inode;
857 filter->fo_fop = file->f_op;
858 filter->fo_iop = inode->i_op;
859 filter->fo_aops = inode->i_mapping->a_ops;
861 rc = filter_init_server_data(obd, file);
863 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
866 filter->fo_rcvd_filp = file;
868 rc = filter_prep_groups(obd);
870 GOTO(err_server_data, rc);
873 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
878 //class_disconnect_exports(obd, 0);
879 filter_free_server_data(filter);
881 if (filp_close(file, 0))
882 CERROR("can't close %s after error\n", LAST_RCVD);
883 filter->fo_rcvd_filp = NULL;
887 /* cleanup the filter: write last used object id to status file */
888 static void filter_post(struct obd_device *obd)
890 struct lvfs_run_ctxt saved;
891 struct filter_obd *filter = &obd->u.filter;
894 /* XXX: filter_update_lastobjid used to call fsync_dev. It might be
895 * best to start a transaction with h_sync, because we removed this
898 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
899 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
902 CERROR("error writing server data: rc = %d\n", rc);
904 for (i = 1; i < filter->fo_group_count; i++) {
905 rc = filter_update_last_objid(obd, i,
906 (i == filter->fo_group_count - 1));
908 CERROR("error writing group %d lastobjid: rc = %d\n",
912 rc = filp_close(filter->fo_rcvd_filp, 0);
913 filter->fo_rcvd_filp = NULL;
915 CERROR("error closing %s: rc = %d\n", LAST_RCVD, rc);
917 filter_cleanup_groups(obd);
918 filter_free_server_data(filter);
919 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
922 static void filter_set_last_id(struct filter_obd *filter, int group, obd_id id)
924 LASSERT(filter->fo_fsd != NULL);
926 LASSERT(group < filter->fo_group_count);
928 spin_lock(&filter->fo_objidlock);
929 filter->fo_last_objids[group] = id;
930 spin_unlock(&filter->fo_objidlock);
933 __u64 filter_last_id(struct filter_obd *filter, int group)
936 LASSERT(filter->fo_fsd != NULL);
938 LASSERT(group < filter->fo_group_count);
940 spin_lock(&filter->fo_objidlock);
941 id = filter->fo_last_objids[group];
942 spin_unlock(&filter->fo_objidlock);
947 /* direct cut-n-paste of mds_blocking_ast() */
948 static int filter_blocking_ast(struct ldlm_lock *lock,
949 struct ldlm_lock_desc *desc,
950 void *data, int flag)
955 if (flag == LDLM_CB_CANCELING) {
956 /* Don't need to do anything here. */
960 /* XXX layering violation! -phil */
961 l_lock(&lock->l_resource->lr_namespace->ns_lock);
962 /* Get this: if filter_blocking_ast is racing with ldlm_intent_policy,
963 * such that filter_blocking_ast is called just before l_i_p takes the
964 * ns_lock, then by the time we get the lock, we might not be the
965 * correct blocking function anymore. So check, and return early, if
967 if (lock->l_blocking_ast != filter_blocking_ast) {
968 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
972 lock->l_flags |= LDLM_FL_CBPENDING;
973 do_ast = (!lock->l_readers && !lock->l_writers);
974 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
977 struct lustre_handle lockh;
980 LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
981 ldlm_lock2handle(lock, &lockh);
982 rc = ldlm_cli_cancel(&lockh);
984 CERROR("ldlm_cli_cancel: %d\n", rc);
986 LDLM_DEBUG(lock, "Lock still has references, will be "
992 extern void *lock_dir(struct inode *dir, struct qstr *name);
993 extern void unlock_dir(struct inode *dir, void *lock);
995 static void * filter_lock_dentry(struct obd_device *obd,
996 struct dentry *dparent, obd_id id)
1002 qstr.len = sprintf(name, LPU64, id);
1003 return lock_dir(dparent->d_inode, &qstr);
1005 down(&dparent->d_inode->i_sem);
1010 /* We never dget the object parent, so DON'T dput it either */
1011 static void filter_parent_unlock(struct dentry *dparent, void *lock)
1014 LASSERT(lock != NULL);
1015 unlock_dir(dparent->d_inode, lock);
1017 up(&dparent->d_inode->i_sem);
1021 /* We never dget the object parent, so DON'T dput it either */
1022 struct dentry *filter_parent(struct obd_device *obd, obd_gr group, obd_id objid)
1024 struct filter_obd *filter = &obd->u.filter;
1025 LASSERT(group < filter->fo_group_count);
1028 if (filter->fo_subdir_count == 0)
1029 return filter->fo_groups[group];
1031 return filter->fo_subdirs[group].dentry[objid & (filter->fo_subdir_count - 1)];
1034 /* We never dget the object parent, so DON'T dput it either */
1035 struct dentry *filter_parent_lock(struct obd_device *obd, obd_gr group,
1036 obd_id objid, void **lock)
1038 unsigned long now = jiffies;
1039 struct dentry *dparent = filter_parent(obd, group, objid);
1041 if (IS_ERR(dparent))
1045 LASSERT(dparent->d_inode);
1047 *lock = filter_lock_dentry(obd, dparent, objid);
1048 if (time_after(jiffies, now + 15 * HZ))
1049 CERROR("slow parent lock %lus\n", (jiffies - now) / HZ);
1053 /* How to get files, dentries, inodes from object id's.
1055 * If dir_dentry is passed, the caller has already locked the parent
1056 * appropriately for this operation (normally a write lock). If
1057 * dir_dentry is NULL, we do a read lock while we do the lookup to
1058 * avoid races with create/destroy and such changing the directory
1059 * internal to the filesystem code. */
1060 struct dentry *filter_fid2dentry(struct obd_device *obd,
1061 struct dentry *dir_dentry,
1062 obd_gr group, obd_id id)
1064 struct dentry *dparent = dir_dentry;
1065 struct dentry *dchild;
1072 CERROR("fatal: invalid object id 0\n");
1073 RETURN(ERR_PTR(-ESTALE));
1076 len = sprintf(name, LPU64, id);
1077 if (dir_dentry == NULL) {
1078 dparent = filter_parent_lock(obd, group, id, &lock);
1079 if (IS_ERR(dparent))
1082 CDEBUG(D_INODE, "looking up object O/%*s/%s\n",
1083 dparent->d_name.len, dparent->d_name.name, name);
1084 dchild = /*ll_*/lookup_one_len(name, dparent, len);
1085 if (dir_dentry == NULL)
1086 filter_parent_unlock(dparent, lock);
1087 if (IS_ERR(dchild)) {
1088 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
1092 CDEBUG(D_INODE, "got child objid %s: %p, count = %d\n",
1093 name, dchild, atomic_read(&dchild->d_count));
1095 LASSERT(atomic_read(&dchild->d_count) > 0);
1100 static int filter_prepare_destroy(struct obd_device *obd, obd_id objid,
1103 struct lustre_handle lockh;
1104 int flags = LDLM_AST_DISCARD_DATA, rc;
1105 struct ldlm_res_id res_id = { .name = { objid, 0, group, 0 } };
1106 ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
1109 /* Tell the clients that the object is gone now and that they should
1110 * throw away any cached pages. */
1111 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, res_id,
1112 LDLM_EXTENT, &policy, LCK_PW,
1113 &flags, filter_blocking_ast, ldlm_completion_ast,
1114 NULL, NULL, NULL, 0, NULL, &lockh);
1116 /* We only care about the side-effects, just drop the lock. */
1118 ldlm_lock_decref(&lockh, LCK_PW);
1123 /* Caller must hold LCK_PW on parent and push us into kernel context.
1124 * Caller is also required to ensure that dchild->d_inode exists. */
1125 static int filter_destroy_internal(struct obd_device *obd, obd_id objid,
1126 struct dentry *dparent,
1127 struct dentry *dchild)
1129 struct inode *inode = dchild->d_inode;
1133 if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) {
1134 CERROR("destroying objid %*s nlink = %lu, count = %d\n",
1135 dchild->d_name.len, dchild->d_name.name,
1136 (unsigned long)inode->i_nlink,
1137 atomic_read(&inode->i_count));
1140 rc = vfs_unlink(dparent->d_inode, dchild);
1143 CERROR("error unlinking objid %*s: rc %d\n",
1144 dchild->d_name.len, dchild->d_name.name, rc);
1149 static int filter_intent_policy(struct ldlm_namespace *ns,
1150 struct ldlm_lock **lockp, void *req_cookie,
1151 ldlm_mode_t mode, int flags, void *data)
1153 struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1154 struct ptlrpc_request *req = req_cookie;
1155 struct ldlm_lock *lock = *lockp, *l = NULL;
1156 struct ldlm_resource *res = lock->l_resource;
1157 ldlm_processing_policy policy;
1158 struct ost_lvb *res_lvb, *reply_lvb;
1159 struct list_head *tmp;
1161 int tmpflags = 0, rc, repsize[2] = {sizeof(struct ldlm_reply),
1162 sizeof(struct ost_lvb) };
1165 policy = ldlm_get_processing_policy(res);
1166 LASSERT(policy != NULL);
1167 LASSERT(req != NULL);
1169 rc = lustre_pack_reply(req, 2, repsize, NULL);
1171 RETURN(req->rq_status = rc);
1173 reply_lvb = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*reply_lvb));
1174 LASSERT(reply_lvb != NULL);
1176 //fixup_handle_for_resent_req(req, lock, &lockh);
1178 /* If we grant any lock at all, it will be a whole-file read lock.
1179 * Call the extent policy function to see if our request can be
1180 * granted, or is blocked. */
1181 lock->l_policy_data.l_extent.start = 0;
1182 lock->l_policy_data.l_extent.end = OBD_OBJECT_EOF;
1183 lock->l_req_mode = LCK_PR;
1185 l_lock(&res->lr_namespace->ns_lock);
1187 res->lr_tmp = &rpc_list;
1188 rc = policy(lock, &tmpflags, 0, &err);
1191 /* FIXME: we should change the policy function slightly, to not make
1192 * this list at all, since we just turn around and free it */
1193 while (!list_empty(&rpc_list)) {
1194 struct ldlm_ast_work *w =
1195 list_entry(rpc_list.next, struct ldlm_ast_work, w_list);
1196 list_del(&w->w_list);
1197 LDLM_LOCK_PUT(w->w_lock);
1198 OBD_FREE(w, sizeof(*w));
1201 if (rc == LDLM_ITER_CONTINUE) {
1202 /* The lock met with no resistance; we're finished. */
1203 l_unlock(&res->lr_namespace->ns_lock);
1204 RETURN(ELDLM_LOCK_REPLACED);
1207 /* Do not grant any lock, but instead send GL callbacks. The extent
1208 * policy nicely created a list of all PW locks for us. We will choose
1209 * the highest of those which are larger than the size in the LVB, if
1210 * any, and perform a glimpse callback. */
1211 down(&res->lr_lvb_sem);
1212 res_lvb = res->lr_lvb_data;
1213 LASSERT(res_lvb != NULL);
1214 reply_lvb->lvb_size = res_lvb->lvb_size;
1215 reply_lvb->lvb_blocks = res_lvb->lvb_blocks;
1216 up(&res->lr_lvb_sem);
1218 list_for_each(tmp, &res->lr_granted) {
1219 struct ldlm_lock *tmplock =
1220 list_entry(tmp, struct ldlm_lock, l_res_link);
1222 if (tmplock->l_granted_mode == LCK_PR)
1225 if (tmplock->l_policy_data.l_extent.end <= reply_lvb->lvb_size)
1229 l = LDLM_LOCK_GET(tmplock);
1233 if (l->l_policy_data.l_extent.start >
1234 tmplock->l_policy_data.l_extent.start)
1238 l = LDLM_LOCK_GET(tmplock);
1240 l_unlock(&res->lr_namespace->ns_lock);
1242 /* There were no PW locks beyond the size in the LVB; finished. */
1244 RETURN(ELDLM_LOCK_ABORTED);
1246 LASSERT(l->l_glimpse_ast != NULL);
1247 rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
1248 if (rc != 0 && res->lr_namespace->ns_lvbo &&
1249 res->lr_namespace->ns_lvbo->lvbo_update) {
1250 res->lr_namespace->ns_lvbo->lvbo_update(res, NULL, 0, 1);
1253 down(&res->lr_lvb_sem);
1254 reply_lvb->lvb_size = res_lvb->lvb_size;
1255 reply_lvb->lvb_blocks = res_lvb->lvb_blocks;
1256 up(&res->lr_lvb_sem);
1260 RETURN(ELDLM_LOCK_ABORTED);
1263 static int filter_post_fs_cleanup(struct obd_device *obd)
1267 rc = fsfilt_post_cleanup(obd);
1272 static int filter_group_set_fs_flags(struct obd_device *obd, int group)
1274 struct filter_obd *filter = &obd->u.filter;
1278 /* zero group is not longer valid. */
1281 for (i = 0; i < filter->fo_subdir_count; i++) {
1282 struct dentry *dentry;
1283 dentry = (filter->fo_subdirs + group)->dentry[i];
1284 rc = fsfilt_set_fs_flags(obd, dentry->d_inode,
1285 SM_DO_REC | SM_DO_COW);
1291 static int filter_post_fs_setup(struct obd_device *obd)
1293 struct filter_obd *filter = &obd->u.filter;
1295 struct llog_ctxt *ctxt = NULL;
1297 rc = fsfilt_post_setup(obd);
1301 for (j = 0; j < filter->fo_group_count; j++) {
1302 rc = filter_group_set_fs_flags(obd, j);
1307 fsfilt_get_reint_log_ctxt(obd, filter->fo_sb, &ctxt);
1309 ctxt->loc_obd = obd;
1310 ctxt->loc_idx = LLOG_REINT_ORIG_CTXT;
1311 obd->obd_llog_ctxt[LLOG_REINT_ORIG_CTXT] = ctxt;
1313 fsfilt_set_ost_flags(obd, filter->fo_sb);
1317 /* mount the file system (secretly) */
1318 int filter_common_setup(struct obd_device *obd, obd_count len,
1319 void *buf, char *option)
1321 struct lustre_cfg* lcfg = buf;
1322 struct filter_obd *filter = &obd->u.filter;
1323 struct vfsmount *mnt;
1324 char name[32] = "CATLIST";
1328 dev_clear_rdonly(2);
1330 if (!lcfg->lcfg_inlbuf1 || !lcfg->lcfg_inlbuf2)
1333 obd->obd_fsops = fsfilt_get_ops(lcfg->lcfg_inlbuf2);
1334 if (IS_ERR(obd->obd_fsops))
1335 RETURN(PTR_ERR(obd->obd_fsops));
1337 mnt = do_kern_mount(lcfg->lcfg_inlbuf2, MS_NOATIME | MS_NODIRATIME,
1338 lcfg->lcfg_inlbuf1, option);
1343 if (lcfg->lcfg_inllen3 > 0 && lcfg->lcfg_inlbuf3) {
1344 if (*lcfg->lcfg_inlbuf3 == 'f') {
1345 obd->obd_replayable = 1;
1346 obd_sync_filter = 1;
1347 CWARN("%s: recovery enabled\n", obd->obd_name);
1349 if (*lcfg->lcfg_inlbuf3 != 'n') {
1350 CERROR("unrecognised flag '%c'\n",
1351 *lcfg->lcfg_inlbuf3);
1353 // XXX Robert? Why do we get errors here
1354 // GOTO(err_mntput, rc = -EINVAL);
1358 filter->fo_vfsmnt = mnt;
1359 filter->fo_sb = mnt->mnt_sb;
1360 filter->fo_fstype = mnt->mnt_sb->s_type->name;
1361 CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
1363 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1364 obd->obd_lvfs_ctxt.pwdmnt = mnt;
1365 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
1366 obd->obd_lvfs_ctxt.fs = get_ds();
1367 obd->obd_lvfs_ctxt.cb_ops = filter_lvfs_ops;
1369 rc = fsfilt_setup(obd, mnt->mnt_sb);
1371 GOTO(err_mntput, rc);
1373 rc = filter_prep(obd);
1375 GOTO(err_mntput, rc);
1378 filter->fo_destroy_in_progress = 0;
1379 sema_init(&filter->fo_create_lock, 1);
1381 spin_lock_init(&filter->fo_translock);
1382 spin_lock_init(&filter->fo_objidlock);
1383 INIT_LIST_HEAD(&filter->fo_export_list);
1384 sema_init(&filter->fo_alloc_lock, 1);
1385 spin_lock_init(&filter->fo_r_pages.oh_lock);
1386 spin_lock_init(&filter->fo_w_pages.oh_lock);
1387 spin_lock_init(&filter->fo_r_discont_pages.oh_lock);
1388 spin_lock_init(&filter->fo_w_discont_pages.oh_lock);
1389 spin_lock_init(&filter->fo_r_discont_blocks.oh_lock);
1390 spin_lock_init(&filter->fo_w_discont_blocks.oh_lock);
1391 filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE;
1393 INIT_LIST_HEAD(&filter->fo_llog_list);
1394 spin_lock_init(&filter->fo_llog_list_lock);
1396 obd->obd_namespace = ldlm_namespace_new("filter-tgt",
1397 LDLM_NAMESPACE_SERVER);
1398 if (obd->obd_namespace == NULL)
1399 GOTO(err_post, rc = -ENOMEM);
1400 obd->obd_namespace->ns_lvbp = obd;
1401 obd->obd_namespace->ns_lvbo = &filter_lvbo;
1402 ldlm_register_intent(obd->obd_namespace, filter_intent_policy);
1404 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1405 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
1407 rc = obd_llog_cat_initialize(obd, &obd->obd_llogs, 1, name);
1409 CERROR("failed to setup llogging subsystems\n");
1422 fsfilt_put_ops(obd->obd_fsops);
1426 static int filter_attach(struct obd_device *obd, obd_count len, void *data)
1428 struct lprocfs_static_vars lvars;
1431 lprocfs_init_vars(filter, &lvars);
1432 rc = lprocfs_obd_attach(obd, lvars.obd_vars);
1436 rc = lprocfs_alloc_obd_stats(obd, LPROC_FILTER_LAST);
1440 /* Init obdfilter private stats here */
1441 lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_READ_BYTES,
1442 LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
1443 lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
1444 LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
1446 return lproc_filter_attach_seqstat(obd);
1449 static int filter_detach(struct obd_device *dev)
1451 lprocfs_free_obd_stats(dev);
1452 return lprocfs_obd_detach(dev);
1455 static int filter_setup(struct obd_device *obd, obd_count len, void *buf)
1457 struct lustre_cfg* lcfg = buf;
1460 /* all mount options including errors=remount-ro and asyncdel are passed
1461 * using 4th lcfg param. And it is good, finally we have got rid of
1462 * hardcoded fs types in the code. */
1463 rc = filter_common_setup(obd, len, buf, lcfg->lcfg_inlbuf4);
1466 rc = filter_post_fs_setup(obd);
1470 static int filter_cleanup(struct obd_device *obd, int flags)
1472 struct filter_obd *filter = &obd->u.filter;
1475 if (flags & OBD_OPT_FAILOVER)
1476 CERROR("%s: shutting down for failover; client state will"
1477 " be preserved.\n", obd->obd_name);
1479 if (!list_empty(&obd->obd_exports)) {
1480 CERROR("%s: still has clients!\n", obd->obd_name);
1481 class_disconnect_exports(obd, flags);
1482 if (!list_empty(&obd->obd_exports)) {
1483 CERROR("still has exports after forced cleanup?\n");
1488 ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
1490 if (filter->fo_sb == NULL)
1493 filter_post_fs_cleanup(obd);
1496 shrink_dcache_parent(filter->fo_sb->s_root);
1499 if (atomic_read(&filter->fo_vfsmnt->mnt_count) > 1)
1500 CERROR("%s: mount point %p busy, mnt_count: %d\n",
1501 obd->obd_name, filter->fo_vfsmnt,
1502 atomic_read(&filter->fo_vfsmnt->mnt_count));
1505 mntput(filter->fo_vfsmnt);
1506 //destroy_buffers(filter->fo_sb->s_dev);
1507 filter->fo_sb = NULL;
1508 fsfilt_put_ops(obd->obd_fsops);
1511 dev_clear_rdonly(2);
1516 /* nearly identical to mds_connect */
1517 static int filter_connect(struct lustre_handle *conn, struct obd_device *obd,
1518 struct obd_uuid *cluuid)
1520 struct obd_export *exp;
1521 struct filter_export_data *fed;
1522 struct filter_client_data *fcd = NULL;
1523 struct filter_obd *filter = &obd->u.filter;
1527 if (conn == NULL || obd == NULL || cluuid == NULL)
1530 rc = class_connect(conn, obd, cluuid);
1533 exp = class_conn2export(conn);
1534 LASSERT(exp != NULL);
1536 fed = &exp->exp_filter_data;
1538 spin_lock_init(&fed->fed_lock);
1540 if (!obd->obd_replayable)
1541 GOTO(cleanup, rc = 0);
1543 OBD_ALLOC(fcd, sizeof(*fcd));
1545 CERROR("filter: out of memory for client data\n");
1546 GOTO(cleanup, rc = -ENOMEM);
1549 memcpy(fcd->fcd_uuid, cluuid, sizeof(fcd->fcd_uuid));
1552 rc = filter_client_add(obd, filter, fed, -1);
1557 OBD_FREE(fcd, sizeof(*fcd));
1558 class_disconnect(exp, 0);
1560 class_export_put(exp);
1565 static int filter_precleanup(struct obd_device *obd, int flags)
1567 struct filter_group_llog *log;
1568 struct filter_obd *filter;
1572 filter = &obd->u.filter;
1574 spin_lock(&filter->fo_llog_list_lock);
1575 while (!list_empty(&filter->fo_llog_list)) {
1576 log = list_entry(filter->fo_llog_list.next,
1577 struct filter_group_llog, list);
1578 list_del(&log->list);
1579 spin_unlock(&filter->fo_llog_list_lock);
1581 rc = obd_llog_finish(obd, log->llogs, 0);
1583 CERROR("failed to cleanup llogging subsystem for %u\n",
1585 OBD_FREE(log->llogs, sizeof(*(log->llogs)));
1586 OBD_FREE(log, sizeof(*log));
1587 spin_lock(&filter->fo_llog_list_lock);
1589 spin_unlock(&filter->fo_llog_list_lock);
1591 rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1593 CERROR("failed to cleanup llogging subsystem\n");
1598 /* Do extra sanity checks for grant accounting. We do this at connect,
1599 * disconnect, and statfs RPC time, so it shouldn't be too bad. We can
1600 * always get rid of it or turn it off when we know accounting is good. */
1601 static void filter_grant_sanity_check(struct obd_device *obd, const char *func)
1603 struct filter_export_data *fed;
1604 struct obd_export *exp;
1605 obd_size maxsize = obd->obd_osfs.os_blocks * obd->obd_osfs.os_bsize;
1606 obd_size tot_dirty = 0, tot_pending = 0, tot_granted = 0;
1607 obd_size fo_tot_dirty, fo_tot_pending, fo_tot_granted;
1609 if (list_empty(&obd->obd_exports))
1612 spin_lock(&obd->obd_osfs_lock);
1613 spin_lock(&obd->obd_dev_lock);
1614 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1615 fed = &exp->exp_filter_data;
1616 LASSERTF(fed->fed_grant + fed->fed_pending <= maxsize,
1617 "cli %s/%p %lu+%lu > "LPU64"\n",
1618 exp->exp_client_uuid.uuid, exp,
1619 fed->fed_grant, fed->fed_pending, maxsize);
1620 LASSERTF(fed->fed_dirty <= maxsize, "cli %s/%p %lu > "LPU64"\n",
1621 exp->exp_client_uuid.uuid, exp,fed->fed_dirty,maxsize);
1622 CDEBUG(D_CACHE,"%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1623 obd->obd_name, exp->exp_client_uuid.uuid, exp,
1624 fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1625 tot_granted += fed->fed_grant + fed->fed_pending;
1626 tot_pending += fed->fed_pending;
1627 tot_dirty += fed->fed_dirty;
1629 fo_tot_granted = obd->u.filter.fo_tot_granted;
1630 fo_tot_pending = obd->u.filter.fo_tot_pending;
1631 fo_tot_dirty = obd->u.filter.fo_tot_dirty;
1632 spin_unlock(&obd->obd_dev_lock);
1633 spin_unlock(&obd->obd_osfs_lock);
1635 /* Do these assertions outside the spinlocks so we don't kill system */
1636 if (tot_granted != fo_tot_granted)
1637 CERROR("%s: tot_granted "LPU64" != fo_tot_granted "LPU64"\n",
1638 func, tot_granted, fo_tot_granted);
1639 if (tot_pending != fo_tot_pending)
1640 CERROR("%s: tot_pending "LPU64" != fo_tot_pending "LPU64"\n",
1641 func, tot_pending, fo_tot_pending);
1642 if (tot_dirty != fo_tot_dirty)
1643 CERROR("%s: tot_dirty "LPU64" != fo_tot_dirty "LPU64"\n",
1644 func, tot_dirty, fo_tot_dirty);
1645 if (tot_pending > tot_granted)
1646 CERROR("%s: tot_pending "LPU64" > tot_granted "LPU64"\n",
1647 func, tot_pending, tot_granted);
1648 if (tot_granted > maxsize)
1649 CERROR("%s: tot_granted "LPU64" > maxsize "LPU64"\n",
1650 func, tot_granted, maxsize);
1651 if (tot_dirty > maxsize)
1652 CERROR("%s: tot_dirty "LPU64" > maxsize "LPU64"\n",
1653 func, tot_dirty, maxsize);
1656 /* Remove this client from the grant accounting totals. We also remove
1657 * the export from the obd device under the osfs and dev locks to ensure
1658 * that the filter_grant_sanity_check() calculations are always valid.
1659 * The client should do something similar when it invalidates its import. */
1660 static void filter_grant_discard(struct obd_export *exp)
1662 struct obd_device *obd = exp->exp_obd;
1663 struct filter_obd *filter = &obd->u.filter;
1664 struct filter_export_data *fed = &exp->exp_filter_data;
1666 spin_lock(&obd->obd_osfs_lock);
1667 spin_lock(&exp->exp_obd->obd_dev_lock);
1668 list_del_init(&exp->exp_obd_chain);
1669 spin_unlock(&exp->exp_obd->obd_dev_lock);
1671 CDEBUG(D_CACHE, "%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1672 obd->obd_name, exp->exp_client_uuid.uuid, exp,
1673 fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1675 LASSERTF(filter->fo_tot_granted >= fed->fed_grant,
1676 "%s: tot_granted "LPU64" cli %s/%p fed_grant %lu\n",
1677 obd->obd_name, filter->fo_tot_granted,
1678 exp->exp_client_uuid.uuid, exp, fed->fed_grant);
1679 filter->fo_tot_granted -= fed->fed_grant;
1680 LASSERTF(exp->exp_obd->u.filter.fo_tot_pending >= fed->fed_pending,
1681 "%s: tot_pending "LPU64" cli %s/%p fed_pending %lu\n",
1682 obd->obd_name, filter->fo_tot_pending,
1683 exp->exp_client_uuid.uuid, exp, fed->fed_pending);
1684 LASSERTF(filter->fo_tot_dirty >= fed->fed_dirty,
1685 "%s: tot_dirty "LPU64" cli %s/%p fed_dirty %lu\n",
1686 obd->obd_name, filter->fo_tot_dirty,
1687 exp->exp_client_uuid.uuid, exp, fed->fed_dirty);
1688 filter->fo_tot_dirty -= fed->fed_dirty;
1692 spin_unlock(&obd->obd_osfs_lock);
1695 static int filter_destroy_export(struct obd_export *exp)
1699 if (exp->exp_filter_data.fed_pending)
1700 CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
1701 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
1702 exp, exp->exp_filter_data.fed_pending);
1704 target_destroy_export(exp);
1706 if (exp->exp_obd->obd_replayable)
1707 filter_client_free(exp, exp->exp_flags);
1709 filter_grant_discard(exp);
1710 if (!(exp->exp_flags & OBD_OPT_FORCE))
1711 filter_grant_sanity_check(exp->exp_obd, __FUNCTION__);
1716 static void filter_sync_llogs(struct obd_export *dexp)
1718 struct filter_group_llog *fglog, *nlog;
1719 struct obd_device *obd = dexp->exp_obd;
1720 struct filter_obd *filter;
1721 int worked = 0, group;
1722 struct llog_ctxt *ctxt;
1724 filter = &obd->u.filter;
1726 /* we can't sync log holding spinlock. also, we do not want to get
1727 * into livelock. so we do following: loop over MDS's exports in
1728 * group order and skip already synced llogs -bzzz */
1730 /* look for group with min. number, but > worked */
1733 spin_lock(&filter->fo_llog_list_lock);
1734 list_for_each_entry(nlog, &filter->fo_llog_list, list) {
1736 if (nlog->group <= worked) {
1737 /* this group is already synced */
1741 if (group < nlog->group) {
1742 /* we have group with smaller number to sync */
1746 /* store current minimal group */
1748 group = nlog->group;
1750 spin_unlock(&filter->fo_llog_list_lock);
1753 worked = fglog->group;
1754 ctxt = llog_get_context(fglog->llogs,
1755 LLOG_UNLINK_REPL_CTXT);
1756 llog_sync(ctxt, dexp);
1758 } while (fglog != NULL);
1761 /* also incredibly similar to mds_disconnect */
1762 static int filter_disconnect(struct obd_export *exp, int flags)
1764 struct obd_device *obd = exp->exp_obd;
1765 unsigned long irqflags;
1770 class_export_get(exp);
1772 spin_lock_irqsave(&exp->exp_lock, irqflags);
1773 exp->exp_flags = flags;
1774 spin_unlock_irqrestore(&exp->exp_lock, irqflags);
1776 if (!(flags & OBD_OPT_FORCE))
1777 filter_grant_sanity_check(obd, __FUNCTION__);
1778 filter_grant_discard(exp);
1780 /* Disconnect early so that clients can't keep using export */
1781 rc = class_disconnect(exp, flags);
1783 ldlm_cancel_locks_for_export(exp);
1785 fsfilt_sync(obd, obd->u.filter.fo_sb);
1787 /* flush any remaining cancel messages out to the target */
1788 filter_sync_llogs(exp);
1790 class_export_put(exp);
1794 struct dentry *__filter_oa2dentry(struct obd_device *obd,
1795 struct obdo *oa, const char *what)
1797 struct dentry *dchild = NULL;
1800 if (oa->o_valid & OBD_MD_FLGROUP)
1803 dchild = filter_fid2dentry(obd, NULL, group, oa->o_id);
1805 if (IS_ERR(dchild)) {
1806 CERROR("%s error looking up object: "LPU64"\n", what, oa->o_id);
1810 if (dchild->d_inode == NULL) {
1811 CERROR("%s: %s on non-existent object: "LPU64"\n",
1812 obd->obd_name, what, oa->o_id);
1814 RETURN(ERR_PTR(-ENOENT));
1820 static int filter_getattr(struct obd_export *exp, struct obdo *oa,
1821 struct lov_stripe_md *md)
1823 struct dentry *dentry = NULL;
1824 struct obd_device *obd;
1828 obd = class_exp2obd(exp);
1830 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1831 exp->exp_handle.h_cookie);
1835 dentry = filter_oa2dentry(obd, oa);
1837 RETURN(PTR_ERR(dentry));
1839 /* Limit the valid bits in the return data to what we actually use */
1840 oa->o_valid = OBD_MD_FLID;
1841 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
1847 /* this is called from filter_truncate() until we have filter_punch() */
1848 static int filter_setattr(struct obd_export *exp, struct obdo *oa,
1849 struct lov_stripe_md *md, struct obd_trans_info *oti)
1851 struct lvfs_run_ctxt saved;
1852 struct filter_obd *filter;
1853 struct dentry *dentry;
1855 struct ldlm_res_id res_id = { .name = { oa->o_id, 0, oa->o_gr, 0 } };
1856 struct ldlm_resource *res;
1861 LASSERT(oti != NULL);
1863 dentry = filter_oa2dentry(exp->exp_obd, oa);
1865 RETURN(PTR_ERR(dentry));
1867 filter = &exp->exp_obd->u.filter;
1869 iattr_from_obdo(&iattr, oa, oa->o_valid);
1871 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
1874 if (iattr.ia_valid & ATTR_SIZE)
1875 down(&dentry->d_inode->i_sem);
1876 handle = fsfilt_start(exp->exp_obd, dentry->d_inode, FSFILT_OP_SETATTR,
1879 GOTO(out_unlock, rc = PTR_ERR(handle));
1881 /* XXX this could be a rwsem instead, if filter_preprw played along */
1882 if (iattr.ia_valid & ATTR_ATTR_FLAG)
1883 rc = fsfilt_iocontrol(exp->exp_obd, dentry->d_inode, NULL,
1885 (long)&iattr.ia_attr_flags);
1887 rc = fsfilt_setattr(exp->exp_obd, dentry, handle, &iattr, 1);
1888 rc = filter_finish_transno(exp, oti, rc);
1889 rc2 = fsfilt_commit(exp->exp_obd, filter->fo_sb, dentry->d_inode, handle, 0);
1891 CERROR("error on commit, err = %d\n", rc2);
1896 if (iattr.ia_valid & ATTR_SIZE) {
1897 res = ldlm_resource_get(exp->exp_obd->obd_namespace, NULL,
1898 res_id, LDLM_EXTENT, 0);
1900 CERROR("!!! resource_get failed for object "LPU64" -- "
1901 "filter_setattr with no lock?\n", oa->o_id);
1903 if (res->lr_namespace->ns_lvbo &&
1904 res->lr_namespace->ns_lvbo->lvbo_update) {
1905 rc = res->lr_namespace->ns_lvbo->lvbo_update
1908 ldlm_resource_putref(res);
1912 oa->o_valid = OBD_MD_FLID;
1913 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
1916 if (iattr.ia_valid & ATTR_SIZE)
1917 up(&dentry->d_inode->i_sem);
1919 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
1925 /* XXX identical to osc_unpackmd */
1926 static int filter_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
1927 struct lov_mds_md *lmm, int lmm_bytes)
1933 if (lmm_bytes < sizeof (*lmm)) {
1934 CERROR("lov_mds_md too small: %d, need %d\n",
1935 lmm_bytes, (int)sizeof(*lmm));
1938 /* XXX LOV_MAGIC etc check? */
1940 if (lmm->lmm_object_id == cpu_to_le64(0)) {
1941 CERROR("lov_mds_md: zero lmm_object_id\n");
1946 lsm_size = lov_stripe_md_size(1);
1950 if (*lsmp != NULL && lmm == NULL) {
1951 OBD_FREE(*lsmp, lsm_size);
1956 if (*lsmp == NULL) {
1957 OBD_ALLOC(*lsmp, lsm_size);
1961 loi_init((*lsmp)->lsm_oinfo);
1965 /* XXX zero *lsmp? */
1966 (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
1967 LASSERT((*lsmp)->lsm_object_id);
1970 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
1975 static void filter_destroy_precreated(struct obd_export *exp, struct obdo *oa,
1976 struct filter_obd *filter)
1978 struct obdo doa; /* XXX obdo on stack */
1983 memset(&doa, 0, sizeof(doa));
1984 if (oa->o_valid & OBD_MD_FLGROUP) {
1985 doa.o_valid |= OBD_MD_FLGROUP;
1986 doa.o_gr = oa->o_gr;
1990 doa.o_mode = S_IFREG;
1991 doa.o_gr = oa->o_gr;
1992 doa.o_valid = oa->o_valid & OBD_MD_FLGROUP;
1994 filter->fo_destroy_in_progress = 1;
1995 down(&filter->fo_create_lock);
1996 if (!filter->fo_destroy_in_progress) {
1997 CERROR("%s: destroy_in_progress already cleared\n",
1998 exp->exp_obd->obd_name);
1999 up(&filter->fo_create_lock);
2004 last = filter_last_id(filter, doa.o_gr);
2005 CWARN("%s: deleting orphan objects from "LPU64" to "LPU64"\n",
2006 exp->exp_obd->obd_name, oa->o_id + 1, last);
2007 for (id = oa->o_id + 1; id <= last; id++) {
2009 filter_destroy(exp, &doa, NULL, NULL);
2012 CDEBUG(D_HA, "%s: after destroy: set last_objids["LPU64"] = "LPU64"\n",
2013 exp->exp_obd->obd_name, doa.o_gr, oa->o_id);
2015 spin_lock(&filter->fo_objidlock);
2016 filter->fo_last_objids[doa.o_gr] = oa->o_id;
2017 spin_unlock(&filter->fo_objidlock);
2019 filter->fo_destroy_in_progress = 0;
2020 up(&filter->fo_create_lock);
2025 /* returns a negative error or a nonnegative number of files to create */
2026 static int filter_should_precreate(struct obd_export *exp, struct obdo *oa,
2029 struct obd_device *obd = exp->exp_obd;
2030 struct filter_obd *filter = &obd->u.filter;
2034 diff = oa->o_id - filter_last_id(filter, oa->o_gr);
2035 CDEBUG(D_INFO, "filter_last_id() = "LPU64" -> diff = %d\n",
2036 filter_last_id(filter, oa->o_gr), diff);
2038 /* delete orphans request */
2039 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2040 (oa->o_flags & OBD_FL_DELORPHAN)) {
2043 if (-diff > 10000) { /* XXX make this smarter */
2044 CERROR("ignoring bogus orphan destroy request: obdid "
2045 LPU64" last_id "LPU64"\n",
2046 oa->o_id, filter_last_id(filter, oa->o_gr));
2049 filter_destroy_precreated(exp, oa, filter);
2050 rc = filter_update_last_objid(obd, group, 0);
2052 CERROR("unable to write lastobjid, but orphans"
2056 /* only precreate if group == 0 and o_id is specfied */
2057 if (!(oa->o_valid & OBD_FL_DELORPHAN) &&
2058 (group != 0 || oa->o_id == 0))
2065 static int filter_precreate_rec(struct obd_device *obd, struct dentry *dentry,
2066 int *number, struct obdo *oa)
2071 rc = fsfilt_precreate_rec(obd, dentry, number, oa);
2076 /* We rely on the fact that only one thread will be creating files in a given
2077 * group at a time, which is why we don't need an atomic filter_get_new_id.
2078 * Even if we had that atomic function, the following race would exist:
2080 * thread 1: gets id x from filter_next_id
2081 * thread 2: gets id (x + 1) from filter_next_id
2082 * thread 2: creates object (x + 1)
2083 * thread 1: tries to create object x, gets -ENOSPC
2085 static int filter_precreate(struct obd_device *obd, struct obdo *oa,
2086 obd_gr group, int *num)
2088 struct dentry *dchild = NULL, *dparent = NULL;
2089 struct filter_obd *filter;
2090 int err = 0, rc = 0, recreate_obj = 0, i;
2092 void *handle = NULL;
2096 filter = &obd->u.filter;
2098 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2099 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
2103 CDEBUG(D_HA, "%s: precreating %d objects\n", obd->obd_name, *num);
2105 down(&filter->fo_create_lock);
2107 for (i = 0; i < *num && err == 0; i++) {
2108 int cleanup_phase = 0;
2110 if (filter->fo_destroy_in_progress) {
2111 CWARN("%s: precreate aborted by destroy\n",
2119 last_id = filter_last_id(filter, group);
2120 if (next_id > last_id) {
2121 CERROR("Error: Trying to recreate obj greater"
2122 "than last id "LPD64" > "LPD64"\n",
2124 GOTO(cleanup, rc = -EINVAL);
2127 next_id = filter_last_id(filter, group) + 1;
2130 CDEBUG(D_INFO, "precreate objid "LPU64"\n", next_id);
2132 dparent = filter_parent_lock(obd, group, next_id, &lock);
2133 if (IS_ERR(dparent))
2134 GOTO(cleanup, rc = PTR_ERR(dparent));
2137 /*only do precreate rec record. so clean kml flags here*/
2138 fsfilt_clear_fs_flags(obd, dparent->d_inode,
2141 dchild = filter_fid2dentry(obd, dparent, group, next_id);
2143 GOTO(cleanup, rc = PTR_ERR(dchild));
2146 if (dchild->d_inode != NULL) {
2147 /* This would only happen if lastobjid was bad on disk*/
2148 /* Could also happen if recreating missing obj but
2152 CERROR("%s: Serious error: recreating obj %*s "
2153 "but obj already exists \n",
2154 obd->obd_name, dchild->d_name.len,
2155 dchild->d_name.name);
2158 CERROR("%s: Serious error: objid %*s already "
2159 "exists; is this filesystem corrupt?\n",
2160 obd->obd_name, dchild->d_name.len,
2161 dchild->d_name.name);
2164 GOTO(cleanup, rc = -EEXIST);
2167 handle = fsfilt_start_log(obd, dparent->d_inode,
2168 FSFILT_OP_CREATE, NULL, 1);
2170 GOTO(cleanup, rc = PTR_ERR(handle));
2173 rc = ll_vfs_create(dparent->d_inode, dchild, S_IFREG, NULL);
2175 CERROR("create failed rc = %d\n", rc);
2179 if (!recreate_obj) {
2180 filter_set_last_id(filter, group, next_id);
2181 err = filter_update_last_objid(obd, group, 0);
2183 CERROR("unable to write lastobjid "
2184 "but file created\n");
2186 fsfilt_set_fs_flags(obd, dparent->d_inode, SM_DO_REC);
2189 switch(cleanup_phase) {
2191 err = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode, handle, 0);
2193 CERROR("error on commit, err = %d\n", err);
2200 filter_parent_unlock(dparent, lock);
2211 /* check if we have an error after ll_vfs_create(). It is possible that
2212 * there will be say -ENOSPC and we will leak it. */
2214 rc = filter_precreate_rec(obd, dparent, num, oa);
2216 up(&filter->fo_create_lock);
2218 CDEBUG(D_HA, "%s: server last_objid for group "LPU64": "LPU64"\n",
2219 obd->obd_name, group, filter->fo_last_objids[group]);
2221 CDEBUG(D_HA, "%s: filter_precreate() created %d objects\n",
2227 static int filter_create(struct obd_export *exp, struct obdo *oa,
2228 struct lov_stripe_md **ea, struct obd_trans_info *oti)
2230 struct obd_device *obd = NULL;
2231 struct filter_obd *filter;
2232 struct lvfs_run_ctxt saved;
2233 struct lov_stripe_md *lsm = NULL;
2234 struct filter_export_data *fed;
2235 char str[PTL_NALFMT_SIZE];
2236 int group = oa->o_gr, rc = 0, diff, recreate_objs = 0;
2239 if (!(oa->o_valid & OBD_MD_FLGROUP) || group == 0) {
2240 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2241 exp->exp_connection->c_peer.peer_nid, str);
2242 CERROR("!!! nid "LPX64"/%s sent invalid object group %d\n",
2243 exp->exp_connection->c_peer.peer_nid, str, group);
2247 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2248 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
2253 fed = &exp->exp_filter_data;
2254 filter = &obd->u.filter;
2256 if (fed->fed_group != group && !recreate_objs &&
2257 !(oa->o_valid & OBD_MD_REINT)) {
2258 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2259 exp->exp_connection->c_peer.peer_nid, str);
2260 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
2261 "earlier; now it's trying to use group %d! This could "
2262 "be a bug in the MDS. Tell CFS.\n",
2263 exp->exp_connection->c_peer.peer_nid, str,
2264 fed->fed_group, group);
2268 CDEBUG(D_INFO, "filter_create(od->o_gr=%d,od->o_id="LPU64")\n",
2273 rc = obd_alloc_memmd(exp, &lsm);
2280 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2282 if (oa->o_valid & OBD_MD_REINT) {
2283 int num = *((int*)oa->o_inline);
2284 rc = filter_precreate(obd, oa, oa->o_gr, &num);
2285 } else if (recreate_objs) {
2286 if (oa->o_id > filter_last_id(&obd->u.filter, group)) {
2287 CERROR("recreate objid "LPU64" > last id "LPU64"\n",
2288 oa->o_id, filter_last_id(&obd->u.filter, group));
2292 rc = filter_precreate(obd, oa, group, &diff);
2295 diff = filter_should_precreate(exp, oa, group);
2297 oa->o_id = filter_last_id(&obd->u.filter, group);
2298 rc = filter_precreate(obd, oa, group, &diff);
2300 oa->o_valid = OBD_MD_FLID;
2304 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2305 if (rc && ea != NULL && *ea != lsm) {
2306 obd_free_memmd(exp, &lsm);
2307 } else if (rc == 0 && ea != NULL) {
2308 /* XXX LOV STACKING: the lsm that is passed to us from
2309 * LOV does not have valid lsm_oinfo data structs, so
2310 * don't go touching that. This needs to be fixed in a
2312 lsm->lsm_object_id = oa->o_id;
2313 lsm->lsm_object_gr = oa->o_gr;
2320 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
2321 struct lov_stripe_md *ea, struct obd_trans_info *oti)
2323 struct obd_device *obd;
2324 struct filter_obd *filter;
2325 struct dentry *dchild = NULL, *dparent = NULL;
2326 struct lvfs_run_ctxt saved;
2327 void *handle = NULL;
2328 struct llog_cookie *fcc = NULL;
2329 int rc, rc2, cleanup_phase = 0, have_prepared = 0;
2333 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2336 filter = &obd->u.filter;
2338 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2341 dparent = filter_parent_lock(obd, oa->o_gr, oa->o_id, &lock);
2342 if (IS_ERR(dparent))
2343 GOTO(cleanup, rc = PTR_ERR(dparent));
2346 dchild = filter_fid2dentry(obd, dparent, oa->o_gr, oa->o_id);
2348 GOTO(cleanup, rc = -ENOENT);
2351 if (dchild->d_inode == NULL) {
2352 CDEBUG(D_INODE, "destroying non-existent object "LPU64"\n",
2354 GOTO(cleanup, rc = -ENOENT);
2357 if (!have_prepared) {
2358 /* If we're really going to destroy the object, get ready
2359 * by getting the clients to discard their cached data.
2361 * We have to drop the parent lock, because
2362 * filter_prepare_destroy will acquire a PW on the object, and
2363 * we don't want to deadlock with an incoming write to the
2364 * object, which has the extent PW and then wants to get the
2365 * parent dentry to do the lookup.
2367 * We dput the child because it's not worth the extra
2368 * complication of condition the above code to skip it on the
2369 * second time through. */
2371 filter_parent_unlock(dparent, lock);
2373 filter_prepare_destroy(obd, oa->o_id, oa->o_gr);
2378 handle = fsfilt_start_log(obd, dparent->d_inode, FSFILT_OP_UNLINK, oti, 1);
2380 GOTO(cleanup, rc = PTR_ERR(handle));
2384 /* Our MDC connection is established by the MDS to us */
2385 if (oa->o_valid & OBD_MD_FLCOOKIE) {
2386 OBD_ALLOC(fcc, sizeof(*fcc));
2388 memcpy(fcc, obdo_logcookie(oa), sizeof(*fcc));
2391 rc = filter_destroy_internal(obd, oa->o_id, dparent, dchild);
2394 switch(cleanup_phase) {
2398 fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
2400 filter_cancel_cookies_cb,
2403 fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
2405 filter_cancel_cookies_cb,
2408 rc = filter_finish_transno(exp, oti, rc);
2409 rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode,
2412 CERROR("error on commit, err = %d\n", rc2);
2419 filter_parent_unlock(dparent, lock);
2421 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2424 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
2431 /* NB start and end are used for punch, but not truncate */
2432 static int filter_truncate(struct obd_export *exp, struct obdo *oa,
2433 struct lov_stripe_md *lsm,
2434 obd_off start, obd_off end,
2435 struct obd_trans_info *oti)
2440 if (end != OBD_OBJECT_EOF)
2441 CERROR("PUNCH not supported, only truncate: end = "LPX64"\n",
2444 CDEBUG(D_INODE, "calling truncate for object "LPU64", valid = %x, "
2445 "o_size = "LPD64"\n", oa->o_id, oa->o_valid, start);
2447 error = filter_setattr(exp, oa, NULL, oti);
2451 static int filter_sync(struct obd_export *exp, struct obdo *oa,
2452 struct lov_stripe_md *lsm, obd_off start, obd_off end)
2454 struct obd_device *obd = exp->exp_obd;
2455 struct lvfs_run_ctxt saved;
2456 struct filter_obd *filter;
2457 struct dentry *dentry;
2458 struct llog_ctxt *ctxt;
2462 filter = &obd->u.filter;
2464 /* an objid of zero is taken to mean "sync whole filesystem" */
2465 if (!oa || !(oa->o_valid & OBD_MD_FLID)) {
2466 rc = fsfilt_sync(obd, filter->fo_sb);
2467 /* flush any remaining cancel messages out to the target */
2468 ctxt = llog_get_context(&obd->obd_llogs, LLOG_UNLINK_REPL_CTXT);
2469 llog_sync(ctxt, exp);
2473 dentry = filter_oa2dentry(obd, oa);
2475 RETURN(PTR_ERR(dentry));
2477 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2479 down(&dentry->d_inode->i_sem);
2480 rc = filemap_fdatasync(dentry->d_inode->i_mapping);
2482 /* just any file to grab fsync method - "file" arg unused */
2483 struct file *file = filter->fo_rcvd_filp;
2485 if (file->f_op && file->f_op->fsync)
2486 rc = file->f_op->fsync(NULL, dentry, 1);
2488 rc2 = filemap_fdatawait(dentry->d_inode->i_mapping);
2492 up(&dentry->d_inode->i_sem);
2494 oa->o_valid = OBD_MD_FLID;
2495 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2497 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2503 static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2504 unsigned long max_age)
2506 struct filter_obd *filter = &obd->u.filter;
2507 int blockbits = filter->fo_sb->s_blocksize_bits;
2511 /* at least try to account for cached pages. its still racey and
2512 * might be under-reporting if clients haven't announced their
2513 * caches with brw recently */
2514 spin_lock(&obd->obd_osfs_lock);
2515 rc = fsfilt_statfs(obd, filter->fo_sb, max_age);
2516 memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
2517 spin_unlock(&obd->obd_osfs_lock);
2519 CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
2520 " pending "LPU64" free "LPU64" avail "LPU64"\n",
2521 filter->fo_tot_dirty, filter->fo_tot_granted,
2522 filter->fo_tot_pending,
2523 osfs->os_bfree << blockbits, osfs->os_bavail << blockbits);
2525 filter_grant_sanity_check(obd, __FUNCTION__);
2527 osfs->os_bavail -= min(osfs->os_bavail,
2528 (filter->fo_tot_dirty + filter->fo_tot_pending +
2529 osfs->os_bsize -1) >> blockbits);
2534 static int filter_get_info(struct obd_export *exp, __u32 keylen,
2535 void *key, __u32 *vallen, void *val)
2537 struct filter_export_data *fed = &exp->exp_filter_data;
2538 struct obd_device *obd;
2541 obd = class_exp2obd(exp);
2543 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2544 exp->exp_handle.h_cookie);
2548 if (keylen == strlen("blocksize") &&
2549 memcmp(key, "blocksize", keylen) == 0) {
2550 __u32 *blocksize = val;
2551 *vallen = sizeof(*blocksize);
2552 *blocksize = obd->u.filter.fo_sb->s_blocksize;
2556 if (keylen == strlen("blocksize_bits") &&
2557 memcmp(key, "blocksize_bits", keylen) == 0) {
2558 __u32 *blocksize_bits = val;
2559 *vallen = sizeof(*blocksize_bits);
2560 *blocksize_bits = obd->u.filter.fo_sb->s_blocksize_bits;
2564 if (keylen >= strlen("last_id") && memcmp(key, "last_id", 7) == 0) {
2565 obd_id *last_id = val;
2566 *last_id = filter_last_id(&obd->u.filter, fed->fed_group);
2569 if (keylen >= strlen("reint_log") && memcmp(key, "reint_log", 9) == 0) {
2570 /*Get log_context handle*/
2571 unsigned long *llh_handle = val;
2572 *vallen = sizeof(unsigned long);
2573 *llh_handle = (unsigned long)obd->obd_llog_ctxt[LLOG_REINT_ORIG_CTXT];
2576 if (keylen >= strlen("cache_sb") && memcmp(key, "cache_sb", 8) == 0) {
2577 /*Get log_context handle*/
2578 unsigned long *sb = val;
2579 *vallen = sizeof(unsigned long);
2580 *sb = (unsigned long)obd->u.filter.fo_sb;
2584 CDEBUG(D_IOCTL, "invalid key\n");
2588 struct obd_llogs *filter_grab_llog_for_group(struct obd_device *obd, int group)
2590 struct filter_group_llog *fglog, *nlog;
2591 char name[32] = "CATLIST";
2592 struct filter_obd *filter;
2593 struct list_head *cur;
2596 filter = &obd->u.filter;
2598 spin_lock(&filter->fo_llog_list_lock);
2599 list_for_each(cur, &filter->fo_llog_list) {
2600 fglog = list_entry(cur, struct filter_group_llog, list);
2601 if (fglog->group == group) {
2602 spin_unlock(&filter->fo_llog_list_lock);
2603 RETURN(fglog->llogs);
2606 spin_unlock(&filter->fo_llog_list_lock);
2608 OBD_ALLOC(fglog, sizeof(*fglog));
2611 fglog->group = group;
2613 OBD_ALLOC(fglog->llogs, sizeof(struct obd_llogs));
2614 if (fglog->llogs == NULL) {
2615 OBD_FREE(fglog, sizeof(*fglog));
2619 spin_lock(&filter->fo_llog_list_lock);
2620 list_for_each(cur, &filter->fo_llog_list) {
2621 nlog = list_entry(cur, struct filter_group_llog, list);
2622 LASSERT(nlog->group != group);
2624 list_add(&fglog->list, &filter->fo_llog_list);
2625 spin_unlock(&filter->fo_llog_list_lock);
2627 rc = obd_llog_cat_initialize(obd, fglog->llogs, 1, name);
2629 OBD_FREE(fglog->llogs, sizeof(*(fglog->llogs)));
2630 OBD_FREE(fglog, sizeof(*fglog));
2634 CDEBUG(D_OTHER, "%s: new llog 0x%p for group %u\n", obd->obd_name,
2635 fglog->llogs, group);
2637 RETURN(fglog->llogs);
2640 static int filter_set_info(struct obd_export *exp, __u32 keylen,
2641 void *key, __u32 vallen, void *val)
2643 struct lvfs_run_ctxt saved;
2644 struct filter_export_data *fed = &exp->exp_filter_data;
2645 struct obd_device *obd;
2646 struct lustre_handle conn;
2647 struct obd_llogs *llog;
2648 struct llog_ctxt *ctxt;
2653 conn.cookie = exp->exp_handle.h_cookie;
2657 CDEBUG(D_IOCTL, "invalid exp %p cookie "LPX64"\n",
2662 if (keylen < strlen("mds_conn") ||
2663 memcmp(key, "mds_conn", keylen) != 0)
2666 group = *((__u32 *)val);
2667 if (fed->fed_group != 0 && fed->fed_group != group) {
2668 char str[PTL_NALFMT_SIZE];
2669 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2670 exp->exp_connection->c_peer.peer_nid, str);
2671 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
2672 "earlier; now it's trying to use group %d! This could "
2673 "be a bug in the MDS. Tell CFS.\n",
2674 exp->exp_connection->c_peer.peer_nid, str,
2675 fed->fed_group, group);
2678 fed->fed_group = group;
2679 CWARN("Received MDS connection ("LPX64"); group %d\n", conn.cookie,
2684 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2685 rc = filter_read_groups(obd, group, 1);
2686 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2688 CERROR("can't read group %u\n", group);
2691 rc = filter_group_set_fs_flags(obd, group);
2693 CERROR("can't set kml flags %u\n", group);
2696 llog = filter_grab_llog_for_group(obd, group);
2697 LASSERT(llog != NULL);
2699 ctxt = llog_get_context(llog, LLOG_UNLINK_REPL_CTXT);
2700 LASSERT(ctxt != NULL);
2701 rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse);
2705 int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
2706 int len, void *karg, void *uarg)
2708 struct obd_device *obd = exp->exp_obd;
2709 struct obd_ioctl_data *data = karg;
2713 case OBD_IOC_ABORT_RECOVERY:
2714 CERROR("aborting recovery for device %s\n", obd->obd_name);
2715 target_abort_recovery(obd);
2718 case OBD_IOC_SET_READONLY: {
2720 struct super_block *sb = obd->u.filter.fo_sb;
2721 struct inode *inode = sb->s_root->d_inode;
2722 BDEVNAME_DECLARE_STORAGE(tmp);
2723 CERROR("setting device %s read-only\n",
2724 ll_bdevname(sb, tmp));
2726 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
2728 (void)fsfilt_commit(obd, sb, inode, handle, 1);
2730 dev_set_rdonly(ll_sbdev(obd->u.filter.fo_sb), 2);
2734 case OBD_IOC_CATLOGLIST: {
2735 rc = llog_catalog_list(obd, 1, data);
2739 case OBD_IOC_SNAP_ADD: {
2740 char *name = data->ioc_inlbuf1;
2742 rc = fsfilt_set_snap_item(obd, obd->u.filter.fo_sb, name);
2746 case OBD_IOC_LLOG_CANCEL:
2747 case OBD_IOC_LLOG_REMOVE:
2748 case OBD_IOC_LLOG_INFO:
2749 case OBD_IOC_LLOG_PRINT: {
2750 /* FIXME to be finished */
2751 RETURN(-EOPNOTSUPP);
2753 struct llog_ctxt *ctxt = NULL;
2755 push_ctxt(&saved, &ctxt->loc_ctxt, NULL);
2756 rc = llog_ioctl(ctxt, cmd, data);
2757 pop_ctxt(&saved, &ctxt->loc_ctxt, NULL);
2770 static struct llog_operations filter_unlink_repl_logops;
2771 static struct llog_operations filter_size_orig_logops = {
2772 lop_setup: llog_obd_origin_setup,
2773 lop_cleanup: llog_catalog_cleanup,
2774 lop_add: llog_catalog_add,
2777 static int filter_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
2778 struct obd_device *tgt, int count,
2779 struct llog_catid *catid)
2781 struct llog_ctxt *ctxt;
2785 filter_unlink_repl_logops = llog_client_ops;
2786 filter_unlink_repl_logops.lop_cancel = llog_obd_repl_cancel;
2787 filter_unlink_repl_logops.lop_connect = llog_repl_connect;
2788 filter_unlink_repl_logops.lop_sync = llog_obd_repl_sync;
2790 rc = obd_llog_setup(obd, llogs, LLOG_UNLINK_REPL_CTXT, tgt, 0, NULL,
2791 &filter_unlink_repl_logops);
2794 /* FIXME - assign unlink_cb for filter's recovery */
2795 ctxt = llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT);
2796 ctxt->llog_proc_cb = filter_recov_log_unlink_cb;
2798 /* FIXME - count should be 1 to setup size log */
2799 rc = obd_llog_setup(obd, llogs, LLOG_SIZE_ORIG_CTXT, tgt, 0,
2800 &catid->lci_logid, &filter_size_orig_logops);
2804 static int filter_llog_finish(struct obd_device *obd,
2805 struct obd_llogs *llogs, int count)
2810 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT));
2814 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_SIZE_ORIG_CTXT));
2818 static int filter_llog_connect(struct obd_device *obd,
2819 struct llogd_conn_body *body)
2821 struct llog_ctxt *ctxt;
2822 struct obd_llogs *llog;
2826 CDEBUG(D_OTHER, "handle connect for %s: %u/%u/%u\n", obd->obd_name,
2827 (unsigned) body->lgdc_logid.lgl_ogr,
2828 (unsigned) body->lgdc_logid.lgl_oid,
2829 (unsigned) body->lgdc_logid.lgl_ogen);
2830 llog = filter_grab_llog_for_group(obd, body->lgdc_logid.lgl_ogr);
2831 LASSERT(llog != NULL);
2832 ctxt = llog_get_context(llog, body->lgdc_ctxt_idx);
2833 rc = llog_connect(ctxt, 1, &body->lgdc_logid,
2834 &body->lgdc_gen, NULL);
2836 CERROR("failed to connect\n");
2841 static struct dentry *filter_lvfs_fid2dentry(__u64 id, __u32 gen, __u64 gr,
2844 return filter_fid2dentry(data, NULL, gr, id);
2847 static struct lvfs_callback_ops filter_lvfs_ops = {
2848 l_fid2dentry: filter_lvfs_fid2dentry,
2851 static struct obd_ops filter_obd_ops = {
2852 .o_owner = THIS_MODULE,
2853 .o_attach = filter_attach,
2854 .o_detach = filter_detach,
2855 .o_get_info = filter_get_info,
2856 .o_set_info = filter_set_info,
2857 .o_setup = filter_setup,
2858 .o_precleanup = filter_precleanup,
2859 .o_cleanup = filter_cleanup,
2860 .o_connect = filter_connect,
2861 .o_disconnect = filter_disconnect,
2862 .o_statfs = filter_statfs,
2863 .o_getattr = filter_getattr,
2864 .o_unpackmd = filter_unpackmd,
2865 .o_create = filter_create,
2866 .o_setattr = filter_setattr,
2867 .o_destroy = filter_destroy,
2868 .o_brw = filter_brw,
2869 .o_punch = filter_truncate,
2870 .o_sync = filter_sync,
2871 .o_preprw = filter_preprw,
2872 .o_commitrw = filter_commitrw,
2873 .o_do_cow = filter_do_cow,
2874 .o_write_extents = filter_write_extents,
2875 .o_destroy_export = filter_destroy_export,
2876 .o_llog_init = filter_llog_init,
2877 .o_llog_finish = filter_llog_finish,
2878 .o_llog_connect = filter_llog_connect,
2879 .o_iocontrol = filter_iocontrol,
2882 static struct obd_ops filter_sanobd_ops = {
2883 .o_owner = THIS_MODULE,
2884 .o_attach = filter_attach,
2885 .o_detach = filter_detach,
2886 .o_get_info = filter_get_info,
2887 .o_set_info = filter_set_info,
2888 .o_setup = filter_san_setup,
2889 .o_precleanup = filter_precleanup,
2890 .o_cleanup = filter_cleanup,
2891 .o_connect = filter_connect,
2892 .o_disconnect = filter_disconnect,
2893 .o_statfs = filter_statfs,
2894 .o_getattr = filter_getattr,
2895 .o_unpackmd = filter_unpackmd,
2896 .o_create = filter_create,
2897 .o_setattr = filter_setattr,
2898 .o_destroy = filter_destroy,
2899 .o_brw = filter_brw,
2900 .o_punch = filter_truncate,
2901 .o_sync = filter_sync,
2902 .o_preprw = filter_preprw,
2903 .o_commitrw = filter_commitrw,
2904 .o_do_cow = filter_do_cow,
2905 .o_write_extents = filter_write_extents,
2906 .o_san_preprw = filter_san_preprw,
2907 .o_destroy_export = filter_destroy_export,
2908 .o_llog_init = filter_llog_init,
2909 .o_llog_finish = filter_llog_finish,
2910 .o_llog_connect = filter_llog_connect,
2911 .o_iocontrol = filter_iocontrol,
2914 static int __init obdfilter_init(void)
2916 struct lprocfs_static_vars lvars;
2919 printk(KERN_INFO "Lustre: Filtering OBD driver; info@clusterfs.com\n");
2921 lprocfs_init_vars(filter, &lvars);
2923 rc = class_register_type(&filter_obd_ops, NULL, lvars.module_vars,
2924 OBD_FILTER_DEVICENAME);
2928 rc = class_register_type(&filter_sanobd_ops, NULL, lvars.module_vars,
2929 OBD_FILTER_SAN_DEVICENAME);
2931 class_unregister_type(OBD_FILTER_DEVICENAME);
2935 static void __exit obdfilter_exit(void)
2937 class_unregister_type(OBD_FILTER_SAN_DEVICENAME);
2938 class_unregister_type(OBD_FILTER_DEVICENAME);
2941 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2942 MODULE_DESCRIPTION("Lustre Filtering OBD driver");
2943 MODULE_LICENSE("GPL");
2945 module_init(obdfilter_init);
2946 module_exit(obdfilter_exit);