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 #include <linux/jbd.h>
45 #include <linux/ext3_fs.h>
46 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
47 # include <linux/mount.h>
48 # include <linux/buffer_head.h>
49 # include <linux/bio.h>
52 #include <linux/obd_class.h>
53 #include <linux/obd_lov.h>
54 #include <linux/obd_ost.h>
55 #include <linux/lustre_dlm.h>
56 #include <linux/lustre_fsfilt.h>
57 #include <linux/lprocfs_status.h>
58 #include <linux/lustre_log.h>
59 #include <linux/lustre_commit_confd.h>
60 #include <libcfs/list.h>
62 #include <linux/lustre_smfs.h>
63 #include "filter_internal.h"
65 /* Group 0 is no longer a legal group, to catch uninitialized IDs */
66 #define FILTER_MIN_GROUPS 3
68 static struct lvfs_callback_ops filter_lvfs_ops;
70 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
71 struct lov_stripe_md *ea, struct obd_trans_info *);
72 struct obd_llogs *filter_grab_llog_for_group(struct obd_device *,
73 int, struct obd_export *);
75 static void filter_commit_cb(struct obd_device *obd, __u64 transno,
76 void *cb_data, int error)
78 obd_transno_commit_cb(obd, transno, error);
81 /* Assumes caller has already pushed us into the kernel context. */
82 int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti,
85 struct filter_obd *filter = &exp->exp_obd->u.filter;
86 struct filter_export_data *fed = &exp->exp_filter_data;
87 struct filter_client_data *fcd = fed->fed_fcd;
90 int err, log_pri = D_HA;
92 /* Propagate error code. */
96 if (!exp->exp_obd->obd_replayable || oti == NULL)
99 /* we don't allocate new transnos for replayed requests */
100 if (oti->oti_transno == 0) {
101 spin_lock(&filter->fo_translock);
102 last_rcvd = le64_to_cpu(filter->fo_fsd->fsd_last_transno) + 1;
103 filter->fo_fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
104 spin_unlock(&filter->fo_translock);
105 oti->oti_transno = last_rcvd;
107 spin_lock(&filter->fo_translock);
108 last_rcvd = oti->oti_transno;
109 if (last_rcvd > le64_to_cpu(filter->fo_fsd->fsd_last_transno))
110 filter->fo_fsd->fsd_last_transno =
111 cpu_to_le64(last_rcvd);
112 spin_unlock(&filter->fo_translock);
115 fcd->fcd_last_rcvd = cpu_to_le64(last_rcvd);
117 /* could get xid from oti, if it's ever needed */
118 fcd->fcd_last_xid = 0;
120 off = fed->fed_lr_off;
122 fsfilt_add_journal_cb(exp->exp_obd, filter->fo_sb, last_rcvd,
123 oti->oti_handle, filter_commit_cb, NULL);
125 err = fsfilt_write_record(exp->exp_obd, filter->fo_rcvd_filp, fcd,
126 sizeof(*fcd), &off, 0);
133 CDEBUG(log_pri, "wrote trans "LPU64" for client %s at #%d: err = %d\n",
134 last_rcvd, fcd->fcd_uuid, fed->fed_lr_idx, err);
139 void f_dput(struct dentry *dentry)
141 /* Can't go inside filter_ddelete because it can block */
142 CDEBUG(D_INODE, "putting %s: %p, count = %d\n",
143 dentry->d_name.name, dentry, atomic_read(&dentry->d_count) - 1);
144 LASSERT(atomic_read(&dentry->d_count) > 0);
149 /* Add client data to the FILTER. We use a bitmap to locate a free space
150 * in the last_rcvd file if cl_idx is -1 (i.e. a new client).
151 * Otherwise, we have just read the data from the last_rcvd file and
152 * we know its offset. */
153 static int filter_client_add(struct obd_device *obd, struct filter_obd *filter,
154 struct filter_export_data *fed, int cl_idx)
156 unsigned long *bitmap = filter->fo_last_rcvd_slots;
157 int new_client = (cl_idx == -1);
160 LASSERT(bitmap != NULL);
162 /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
163 if (!strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid))
166 /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
167 * there's no need for extra complication here
170 cl_idx = find_first_zero_bit(bitmap, FILTER_LR_MAX_CLIENTS);
172 if (cl_idx >= FILTER_LR_MAX_CLIENTS) {
173 CERROR("no client slots - fix FILTER_LR_MAX_CLIENTS\n");
176 if (test_and_set_bit(cl_idx, bitmap)) {
177 CERROR("FILTER client %d: found bit is set in bitmap\n",
179 cl_idx = find_next_zero_bit(bitmap,
180 FILTER_LR_MAX_CLIENTS,
185 if (test_and_set_bit(cl_idx, bitmap)) {
186 CERROR("FILTER client %d: bit already set in bitmap!\n",
192 fed->fed_lr_idx = cl_idx;
193 fed->fed_lr_off = le32_to_cpu(filter->fo_fsd->fsd_client_start) +
194 cl_idx * le16_to_cpu(filter->fo_fsd->fsd_client_size);
196 CDEBUG(D_INFO, "client at index %d (%llu) with UUID '%s' added\n",
197 fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
200 struct lvfs_run_ctxt saved;
201 loff_t off = fed->fed_lr_off;
205 CDEBUG(D_INFO, "writing client fcd at idx %u (%llu) (len %u)\n",
206 fed->fed_lr_idx,off,(unsigned int)sizeof(*fed->fed_fcd));
208 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
209 /* Transaction needed to fix bug 1403 */
210 handle = fsfilt_start(obd,
211 filter->fo_rcvd_filp->f_dentry->d_inode,
212 FSFILT_OP_SETATTR, NULL);
213 if (IS_ERR(handle)) {
214 err = PTR_ERR(handle);
215 CERROR("unable to start transaction: rc %d\n", err);
217 err = fsfilt_write_record(obd, filter->fo_rcvd_filp,
219 sizeof(*fed->fed_fcd),
221 fsfilt_commit(obd, filter->fo_sb,
222 filter->fo_rcvd_filp->f_dentry->d_inode,
225 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
228 CERROR("error writing %s client idx %u: rc %d\n",
229 LAST_RCVD, fed->fed_lr_idx, err);
236 static int filter_client_free(struct obd_export *exp, int flags)
238 struct filter_export_data *fed = &exp->exp_filter_data;
239 struct filter_obd *filter = &exp->exp_obd->u.filter;
240 struct obd_device *obd = exp->exp_obd;
241 struct filter_client_data zero_fcd;
242 struct lvfs_run_ctxt saved;
247 if (fed->fed_fcd == NULL)
250 if (flags & OBD_OPT_FAILOVER)
253 /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
254 if (strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid ) == 0)
257 LASSERT(filter->fo_last_rcvd_slots != NULL);
259 off = fed->fed_lr_off;
261 CDEBUG(D_INFO, "freeing client at idx %u (%lld) with UUID '%s'\n",
262 fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
264 /* Clear the bit _after_ zeroing out the client so we don't
265 race with filter_client_add and zero out new clients.*/
266 if (!test_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
267 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
272 memset(&zero_fcd, 0, sizeof zero_fcd);
273 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
274 rc = fsfilt_write_record(obd, filter->fo_rcvd_filp, &zero_fcd,
275 sizeof(zero_fcd), &off, 1);
277 /* update server's transno */
278 filter_update_server_data(obd, filter->fo_rcvd_filp,
280 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
282 CDEBUG(rc == 0 ? D_INFO : D_ERROR,
283 "zeroing disconnecting client %s at idx %u (%llu) in %s rc %d\n",
284 fed->fed_fcd->fcd_uuid, fed->fed_lr_idx, fed->fed_lr_off,
287 if (!test_and_clear_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
288 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
294 OBD_FREE(fed->fed_fcd, sizeof(*fed->fed_fcd));
299 static int filter_free_server_data(struct filter_obd *filter)
301 OBD_FREE(filter->fo_fsd, sizeof(*filter->fo_fsd));
302 filter->fo_fsd = NULL;
303 OBD_FREE(filter->fo_last_rcvd_slots,
304 FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
305 filter->fo_last_rcvd_slots = NULL;
309 /* assumes caller is already in kernel ctxt */
310 int filter_update_server_data(struct obd_device *obd, struct file *filp,
311 struct filter_server_data *fsd, int force_sync)
317 CDEBUG(D_INODE, "server uuid : %s\n", fsd->fsd_uuid);
318 CDEBUG(D_INODE, "server last_rcvd : "LPU64"\n",
319 le64_to_cpu(fsd->fsd_last_transno));
320 CDEBUG(D_INODE, "server last_mount: "LPU64"\n",
321 le64_to_cpu(fsd->fsd_mount_count));
323 rc = fsfilt_write_record(obd, filp, fsd, sizeof(*fsd), &off,force_sync);
325 CERROR("error writing filter_server_data: rc = %d\n", rc);
330 int filter_update_last_objid(struct obd_device *obd, obd_gr group,
333 struct filter_obd *filter = &obd->u.filter;
339 if (filter->fo_last_objid_files[group] == NULL) {
340 CERROR("Object group "LPU64" not fully setup; not updating "
341 "last_objid\n", group);
345 CDEBUG(D_INODE, "server last_objid for group "LPU64": "LPU64"\n",
346 group, filter->fo_last_objids[group]);
348 tmp = cpu_to_le64(filter->fo_last_objids[group]);
349 rc = fsfilt_write_record(obd, filter->fo_last_objid_files[group],
350 &tmp, sizeof(tmp), &off, force_sync);
352 CERROR("error writing group "LPU64" last objid: rc = %d\n",
357 /* assumes caller has already in kernel ctxt */
358 static int filter_init_server_data(struct obd_device *obd, struct file * filp)
360 struct filter_obd *filter = &obd->u.filter;
361 struct filter_server_data *fsd;
362 struct filter_client_data *fcd = NULL;
363 struct inode *inode = filp->f_dentry->d_inode;
364 unsigned long last_rcvd_size = inode->i_size;
370 /* ensure padding in the struct is the correct size */
371 LASSERT (offsetof(struct filter_server_data, fsd_padding) +
372 sizeof(fsd->fsd_padding) == FILTER_LR_SERVER_SIZE);
373 LASSERT (offsetof(struct filter_client_data, fcd_padding) +
374 sizeof(fcd->fcd_padding) == FILTER_LR_CLIENT_SIZE);
376 OBD_ALLOC(fsd, sizeof(*fsd));
379 filter->fo_fsd = fsd;
381 OBD_ALLOC(filter->fo_last_rcvd_slots,
382 FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
383 if (filter->fo_last_rcvd_slots == NULL) {
384 OBD_FREE(fsd, sizeof(*fsd));
388 if (last_rcvd_size == 0) {
389 CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD);
391 memcpy(fsd->fsd_uuid, obd->obd_uuid.uuid,sizeof(fsd->fsd_uuid));
392 fsd->fsd_last_transno = 0;
393 mount_count = fsd->fsd_mount_count = 0;
394 fsd->fsd_server_size = cpu_to_le32(FILTER_LR_SERVER_SIZE);
395 fsd->fsd_client_start = cpu_to_le32(FILTER_LR_CLIENT_START);
396 fsd->fsd_client_size = cpu_to_le16(FILTER_LR_CLIENT_SIZE);
397 fsd->fsd_subdir_count = cpu_to_le16(FILTER_SUBDIR_COUNT);
398 filter->fo_subdir_count = FILTER_SUBDIR_COUNT;
400 rc = fsfilt_read_record(obd, filp, fsd, sizeof(*fsd), &off);
402 CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
406 if (strcmp(fsd->fsd_uuid, obd->obd_uuid.uuid) != 0) {
407 CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
408 obd->obd_uuid.uuid, fsd->fsd_uuid);
409 GOTO(err_fsd, rc = -EINVAL);
411 mount_count = le64_to_cpu(fsd->fsd_mount_count);
412 filter->fo_subdir_count = le16_to_cpu(fsd->fsd_subdir_count);
415 if (fsd->fsd_feature_incompat & ~cpu_to_le32(FILTER_INCOMPAT_SUPP)) {
416 CERROR("unsupported feature %x\n",
417 le32_to_cpu(fsd->fsd_feature_incompat) &
418 ~FILTER_INCOMPAT_SUPP);
419 GOTO(err_fsd, rc = -EINVAL);
421 if (fsd->fsd_feature_rocompat & ~cpu_to_le32(FILTER_ROCOMPAT_SUPP)) {
422 CERROR("read-only feature %x\n",
423 le32_to_cpu(fsd->fsd_feature_rocompat) &
424 ~FILTER_ROCOMPAT_SUPP);
425 /* Do something like remount filesystem read-only */
426 GOTO(err_fsd, rc = -EINVAL);
429 CDEBUG(D_INODE, "%s: server last_rcvd : "LPU64"\n",
430 obd->obd_name, le64_to_cpu(fsd->fsd_last_transno));
431 CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
432 obd->obd_name, mount_count + 1);
433 CDEBUG(D_INODE, "%s: server data size: %u\n",
434 obd->obd_name, le32_to_cpu(fsd->fsd_server_size));
435 CDEBUG(D_INODE, "%s: per-client data start: %u\n",
436 obd->obd_name, le32_to_cpu(fsd->fsd_client_start));
437 CDEBUG(D_INODE, "%s: per-client data size: %u\n",
438 obd->obd_name, le32_to_cpu(fsd->fsd_client_size));
439 CDEBUG(D_INODE, "%s: server subdir_count: %u\n",
440 obd->obd_name, le16_to_cpu(fsd->fsd_subdir_count));
441 CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
442 last_rcvd_size <= le32_to_cpu(fsd->fsd_client_start) ? 0 :
443 (last_rcvd_size - le32_to_cpu(fsd->fsd_client_start)) /
444 le16_to_cpu(fsd->fsd_client_size));
446 if (!obd->obd_replayable) {
447 CWARN("%s: recovery support OFF\n", obd->obd_name);
451 for (cl_idx = 0, off = le32_to_cpu(fsd->fsd_client_start);
452 off < last_rcvd_size; cl_idx++) {
454 struct obd_export *exp;
455 struct filter_export_data *fed;
458 OBD_ALLOC(fcd, sizeof(*fcd));
460 GOTO(err_client, rc = -ENOMEM);
463 /* Don't assume off is incremented properly by
464 * fsfilt_read_record(), in case sizeof(*fcd)
465 * isn't the same as fsd->fsd_client_size. */
466 off = le32_to_cpu(fsd->fsd_client_start) +
467 cl_idx * le16_to_cpu(fsd->fsd_client_size);
468 rc = fsfilt_read_record(obd, filp, fcd, sizeof(*fcd), &off);
470 CERROR("error reading FILT %s idx %d off %llu: rc %d\n",
471 LAST_RCVD, cl_idx, off, rc);
472 break; /* read error shouldn't cause startup to fail */
475 if (fcd->fcd_uuid[0] == '\0') {
476 CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
481 last_rcvd = le64_to_cpu(fcd->fcd_last_rcvd);
483 /* These exports are cleaned up by filter_disconnect(), so they
484 * need to be set up like real exports as filter_connect() does.
486 exp = class_new_export(obd);
487 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64" "
488 "srv lr: "LPU64" fcd_group %d \n", fcd->fcd_uuid, cl_idx,
489 last_rcvd, le64_to_cpu(fsd->fsd_last_transno),
490 le32_to_cpu(fcd->fcd_group));
492 GOTO(err_client, rc = -ENOMEM);
494 memcpy(&exp->exp_client_uuid.uuid, fcd->fcd_uuid,
495 sizeof exp->exp_client_uuid.uuid);
496 fed = &exp->exp_filter_data;
498 fed->fed_group = le32_to_cpu(fcd->fcd_group);
499 filter_client_add(obd, filter, fed, cl_idx);
500 /* create helper if export init gets more complex */
501 spin_lock_init(&fed->fed_lock);
504 exp->exp_replay_needed = 1;
505 obd->obd_recoverable_clients++;
506 obd->obd_max_recoverable_clients++;
507 class_export_put(exp);
509 CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
512 if (last_rcvd > le64_to_cpu(fsd->fsd_last_transno))
513 fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
517 obd->obd_last_committed = le64_to_cpu(fsd->fsd_last_transno);
519 if (obd->obd_recoverable_clients) {
520 CWARN("RECOVERY: service %s, %d recoverable clients, "
521 "last_transno "LPU64"\n", obd->obd_name,
522 obd->obd_recoverable_clients,
523 le64_to_cpu(fsd->fsd_last_transno));
524 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
525 target_start_recovery_thread(obd, ost_handle);
529 OBD_FREE(fcd, sizeof(*fcd));
532 filter->fo_mount_count = mount_count + 1;
533 fsd->fsd_mount_count = cpu_to_le64(filter->fo_mount_count);
535 /* save it, so mount count and last_transno is current */
536 rc = filter_update_server_data(obd, filp, filter->fo_fsd, 1);
538 GOTO(err_client, rc);
542 class_disconnect_exports(obd, 0);
544 filter_free_server_data(filter);
548 static int filter_cleanup_groups(struct obd_device *obd)
550 struct filter_obd *filter = &obd->u.filter;
551 struct dentry *dentry;
555 for (i = 0; i < filter->fo_group_count; i++) {
556 if (filter->fo_subdirs != NULL) {
557 for (k = 0; k < filter->fo_subdir_count; k++) {
558 dentry = filter->fo_subdirs[i].dentry[k];
562 filter->fo_subdirs[i].dentry[k] = NULL;
565 if (filter->fo_last_objid_files[i] != NULL) {
566 filp_close(filter->fo_last_objid_files[i], 0);
567 filter->fo_last_objid_files[i] = NULL;
569 if (filter->fo_groups[i] != NULL) {
570 dput(filter->fo_groups[i]);
571 filter->fo_groups[i] = NULL;
574 if (filter->fo_subdirs != NULL)
575 OBD_FREE(filter->fo_subdirs,
576 filter->fo_group_count * sizeof(*filter->fo_subdirs));
577 if (filter->fo_groups != NULL)
578 OBD_FREE(filter->fo_groups,
579 filter->fo_group_count * sizeof(*filter->fo_groups));
580 if (filter->fo_last_objids != NULL)
581 OBD_FREE(filter->fo_last_objids,
582 filter->fo_group_count * sizeof(__u64));
583 if (filter->fo_last_objid_files != NULL)
584 OBD_FREE(filter->fo_last_objid_files,
585 filter->fo_group_count * sizeof(struct file *));
586 f_dput(filter->fo_dentry_O);
590 static int filter_update_last_group(struct obd_device *obd, int group)
592 struct filter_obd *filter = &obd->u.filter;
593 struct file *filp = NULL;
594 int last_group = 0, rc;
598 if (group <= filter->fo_committed_group)
601 filp = filp_open("LAST_GROUP", O_RDWR, 0700);
605 CERROR("cannot open LAST_GROUP: rc = %d\n", rc);
609 rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
611 CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n",rc);
614 LASSERT(off == 0 || last_group >= FILTER_MIN_GROUPS);
615 CDEBUG(D_INODE, "%s: previous %d, new %d\n",
616 obd->obd_name, last_group, group);
620 /* must be sync: bXXXX */
621 rc = fsfilt_write_record(obd, filp, &last_group, sizeof(__u32), &off, 1);
623 CDEBUG(D_INODE, "error updating LAST_GROUP: rc %d\n", rc);
627 filter->fo_committed_group = group;
634 static int filter_read_group_internal(struct obd_device *obd, int group,
637 struct filter_obd *filter = &obd->u.filter;
638 __u64 *new_objids = NULL;
639 struct filter_subdirs *new_subdirs = NULL, *tmp_subdirs = NULL;
640 struct dentry **new_groups = NULL;
641 struct file **new_files = NULL;
642 struct dentry *dentry;
644 int old_count = filter->fo_group_count, rc, stage = 0, i;
649 snprintf(name, 24, "%d", group);
653 dentry = ll_lookup_one_len(name, filter->fo_dentry_O,
655 if (IS_ERR(dentry)) {
656 CERROR("Cannot lookup expected object group %d: %ld\n",
657 group, PTR_ERR(dentry));
658 RETURN(PTR_ERR(dentry));
661 dentry = simple_mkdir(filter->fo_dentry_O, name, 0700, 1);
662 if (IS_ERR(dentry)) {
663 CERROR("cannot lookup/create O/%s: rc = %ld\n", name,
665 RETURN(PTR_ERR(dentry));
670 snprintf(name, 24, "O/%d/LAST_ID", group);
672 filp = filp_open(name, O_CREAT | O_RDWR, 0700);
674 CERROR("cannot create %s: rc = %ld\n", name, PTR_ERR(filp));
675 GOTO(cleanup, rc = PTR_ERR(filp));
679 rc = fsfilt_read_record(obd, filp, &last_objid, sizeof(__u64), &off);
681 CDEBUG(D_INODE, "error reading %s: rc %d\n", name, rc);
685 if (filter->fo_subdir_count) {
686 OBD_ALLOC(tmp_subdirs, sizeof(*tmp_subdirs));
687 if (tmp_subdirs == NULL)
688 GOTO(cleanup, rc = -ENOMEM);
691 for (i = 0; i < filter->fo_subdir_count; i++) {
693 snprintf(dir, sizeof(dir), "d%u", i);
695 tmp_subdirs->dentry[i] =
696 simple_mkdir(dentry, dir, 0700, 1);
697 if (IS_ERR(tmp_subdirs->dentry[i])) {
698 rc = PTR_ERR(tmp_subdirs->dentry[i]);
699 CERROR("can't lookup/create O/%d/%s: rc = %d\n",
703 CDEBUG(D_INODE, "got/created O/%d/%s: %p\n", group, dir,
704 tmp_subdirs->dentry[i]);
708 /* 'group' is an index; we need an array of length 'group + 1' */
709 if (group + 1 > old_count) {
711 OBD_ALLOC(new_objids, len * sizeof(*new_objids));
712 OBD_ALLOC(new_subdirs, len * sizeof(*new_subdirs));
713 OBD_ALLOC(new_groups, len * sizeof(*new_groups));
714 OBD_ALLOC(new_files, len * sizeof(*new_files));
716 if (new_objids == NULL || new_subdirs == NULL ||
717 new_groups == NULL || new_files == NULL)
718 GOTO(cleanup, rc = -ENOMEM);
720 memcpy(new_objids, filter->fo_last_objids,
721 old_count * sizeof(*new_objids));
722 memcpy(new_subdirs, filter->fo_subdirs,
723 old_count * sizeof(*new_subdirs));
724 memcpy(new_groups, filter->fo_groups,
725 old_count * sizeof(*new_groups));
726 memcpy(new_files, filter->fo_last_objid_files,
727 old_count * sizeof(*new_files));
730 OBD_FREE(filter->fo_last_objids,
731 old_count * sizeof(*new_objids));
732 OBD_FREE(filter->fo_subdirs,
733 old_count * sizeof(*new_subdirs));
734 OBD_FREE(filter->fo_groups,
735 old_count * sizeof(*new_groups));
736 OBD_FREE(filter->fo_last_objid_files,
737 old_count * sizeof(*new_files));
739 filter->fo_last_objids = new_objids;
740 filter->fo_subdirs = new_subdirs;
741 filter->fo_groups = new_groups;
742 filter->fo_last_objid_files = new_files;
743 filter->fo_group_count = len;
746 filter->fo_groups[group] = dentry;
747 filter->fo_last_objid_files[group] = filp;
748 if (filter->fo_subdir_count) {
749 filter->fo_subdirs[group] = *tmp_subdirs;
750 OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
753 filter_update_last_group(obd, group);
755 if (filp->f_dentry->d_inode->i_size == 0) {
756 filter->fo_last_objids[group] = FILTER_INIT_OBJID;
760 filter->fo_last_objids[group] = le64_to_cpu(last_objid);
761 CDEBUG(D_INODE, "%s: server last_objid group %d: "LPU64"\n",
762 obd->obd_name, group, last_objid);
767 if (new_objids != NULL)
768 OBD_FREE(new_objids, group * sizeof(*new_objids));
769 if (new_subdirs != NULL)
770 OBD_FREE(new_subdirs, group * sizeof(*new_subdirs));
771 if (new_groups != NULL)
772 OBD_FREE(new_groups, group * sizeof(*new_groups));
773 if (new_files != NULL)
774 OBD_FREE(new_files, group * sizeof(*new_files));
776 if (filter->fo_subdir_count) {
777 for (i = 0; i < filter->fo_subdir_count; i++) {
778 if (tmp_subdirs->dentry[i] != NULL)
779 dput(tmp_subdirs->dentry[i]);
781 OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
791 static int filter_read_groups(struct obd_device *obd, int last_group,
794 struct filter_obd *filter = &obd->u.filter;
795 int old_count, group, rc = 0;
797 down(&filter->fo_init_lock);
798 old_count = filter->fo_group_count;
799 for (group = old_count; group <= last_group; group++) {
801 continue; /* no group zero */
803 rc = filter_read_group_internal(obd, group, create);
807 up(&filter->fo_init_lock);
811 static int filter_prep_groups(struct obd_device *obd)
813 struct filter_obd *filter = &obd->u.filter;
814 struct dentry *dentry, *O_dentry;
815 int rc = 0, cleanup_phase = 0;
816 struct file *filp = NULL;
821 O_dentry = simple_mkdir(current->fs->pwd, "O", 0700, 1);
822 CDEBUG(D_INODE, "got/created O: %p\n", O_dentry);
823 if (IS_ERR(O_dentry)) {
824 rc = PTR_ERR(O_dentry);
825 CERROR("cannot open/create O: rc = %d\n", rc);
828 filter->fo_dentry_O = O_dentry;
829 cleanup_phase = 1; /* O_dentry */
831 /* Lookup "R" to tell if we're on an old OST FS and need to convert
832 * from O/R/<dir>/<objid> to O/0/<dir>/<objid>. This can be removed
833 * some time post 1.0 when all old-style OSTs have converted along
834 * with the init_objid hack. */
835 dentry = ll_lookup_one_len("R", O_dentry, 1);
837 GOTO(cleanup, rc = PTR_ERR(dentry));
838 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
839 struct dentry *O0_dentry = lookup_one_len("0", O_dentry, 1);
842 CWARN("converting OST to new object layout\n");
843 if (IS_ERR(O0_dentry)) {
844 rc = PTR_ERR(O0_dentry);
845 CERROR("error looking up O/0: rc %d\n", rc);
849 if (O0_dentry->d_inode) {
850 CERROR("Both O/R and O/0 exist. Fix manually.\n");
851 GOTO(cleanup_O0, rc = -EEXIST);
854 down(&O_dentry->d_inode->i_sem);
855 rc = vfs_rename(O_dentry->d_inode, dentry,
856 O_dentry->d_inode, O0_dentry);
857 up(&O_dentry->d_inode->i_sem);
860 CERROR("error renaming O/R to O/0: rc %d\n", rc);
861 GOTO(cleanup_O0, rc);
863 filter->fo_fsd->fsd_feature_incompat |=
864 cpu_to_le32(FILTER_INCOMPAT_GROUPS);
865 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
867 GOTO(cleanup_O0, rc);
879 cleanup_phase = 2; /* groups */
881 /* we have to initialize all groups before first connections from
882 * clients because they may send create/destroy for any group -bzzz */
883 filp = filp_open("LAST_GROUP", O_CREAT | O_RDWR, 0700);
885 CERROR("cannot create LAST_GROUP: rc = %ld\n", PTR_ERR(filp));
886 GOTO(cleanup, rc = PTR_ERR(filp));
888 cleanup_phase = 3; /* filp */
890 rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
892 CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n", rc);
896 last_group = FILTER_MIN_GROUPS;
898 LASSERT(last_group >= FILTER_MIN_GROUPS);
901 CWARN("%s: initialize groups [%d,%d]\n", obd->obd_name,
902 FILTER_MIN_GROUPS, last_group);
903 filter->fo_committed_group = last_group;
904 rc = filter_read_groups(obd, last_group, 1);
912 switch (cleanup_phase) {
916 filter_cleanup_groups(obd);
918 f_dput(filter->fo_dentry_O);
919 filter->fo_dentry_O = NULL;
926 /* setup the object store with correct subdirectories */
927 static int filter_prep(struct obd_device *obd)
929 struct lvfs_run_ctxt saved;
930 struct filter_obd *filter = &obd->u.filter;
936 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
937 file = filp_open(LAST_RCVD, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
938 if (!file || IS_ERR(file)) {
940 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
945 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
946 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
947 file->f_dentry->d_inode->i_mode);
948 GOTO(err_filp, rc = -ENOENT);
951 /* steal operations */
952 inode = file->f_dentry->d_inode;
953 filter->fo_fop = file->f_op;
954 filter->fo_iop = inode->i_op;
955 filter->fo_aops = inode->i_mapping->a_ops;
957 rc = filter_init_server_data(obd, file);
959 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
962 filter->fo_rcvd_filp = file;
964 rc = filter_prep_groups(obd);
966 GOTO(err_server_data, rc);
969 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
974 //class_disconnect_exports(obd, 0);
975 filter_free_server_data(filter);
977 if (filp_close(file, 0))
978 CERROR("can't close %s after error\n", LAST_RCVD);
979 filter->fo_rcvd_filp = NULL;
983 /* cleanup the filter: write last used object id to status file */
984 static void filter_post(struct obd_device *obd)
986 struct lvfs_run_ctxt saved;
987 struct filter_obd *filter = &obd->u.filter;
990 /* XXX: filter_update_lastobjid used to call fsync_dev. It might be
991 * best to start a transaction with h_sync, because we removed this
994 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
995 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
998 CERROR("error writing server data: rc = %d\n", rc);
1000 for (i = 1; i < filter->fo_group_count; i++) {
1001 rc = filter_update_last_objid(obd, i,
1002 (i == filter->fo_group_count - 1));
1004 CERROR("error writing group %d lastobjid: rc = %d\n",
1008 rc = filp_close(filter->fo_rcvd_filp, 0);
1009 filter->fo_rcvd_filp = NULL;
1011 CERROR("error closing %s: rc = %d\n", LAST_RCVD, rc);
1013 filter_cleanup_groups(obd);
1014 filter_free_server_data(filter);
1015 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1018 static void filter_set_last_id(struct filter_obd *filter, int group, obd_id id)
1020 LASSERT(filter->fo_fsd != NULL);
1022 LASSERT(group < filter->fo_group_count);
1024 spin_lock(&filter->fo_objidlock);
1025 filter->fo_last_objids[group] = id;
1026 spin_unlock(&filter->fo_objidlock);
1029 __u64 filter_last_id(struct filter_obd *filter, int group)
1032 LASSERT(filter->fo_fsd != NULL);
1034 LASSERT(group < filter->fo_group_count);
1036 spin_lock(&filter->fo_objidlock);
1037 id = filter->fo_last_objids[group];
1038 spin_unlock(&filter->fo_objidlock);
1043 /* direct cut-n-paste of mds_blocking_ast() */
1044 static int filter_blocking_ast(struct ldlm_lock *lock,
1045 struct ldlm_lock_desc *desc,
1046 void *data, int flag)
1051 if (flag == LDLM_CB_CANCELING) {
1052 /* Don't need to do anything here. */
1056 /* XXX layering violation! -phil */
1057 l_lock(&lock->l_resource->lr_namespace->ns_lock);
1058 /* Get this: if filter_blocking_ast is racing with ldlm_intent_policy,
1059 * such that filter_blocking_ast is called just before l_i_p takes the
1060 * ns_lock, then by the time we get the lock, we might not be the
1061 * correct blocking function anymore. So check, and return early, if
1063 if (lock->l_blocking_ast != filter_blocking_ast) {
1064 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
1068 lock->l_flags |= LDLM_FL_CBPENDING;
1069 do_ast = (!lock->l_readers && !lock->l_writers);
1070 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
1073 struct lustre_handle lockh;
1076 LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
1077 ldlm_lock2handle(lock, &lockh);
1078 rc = ldlm_cli_cancel(&lockh);
1080 CERROR("ldlm_cli_cancel: %d\n", rc);
1082 LDLM_DEBUG(lock, "Lock still has references, will be "
1088 extern void *lock_dir(struct inode *dir, struct qstr *name);
1089 extern void unlock_dir(struct inode *dir, void *lock);
1091 static void * filter_lock_dentry(struct obd_device *obd,
1092 struct dentry *dparent, obd_id id)
1098 qstr.len = sprintf(name, LPU64, id);
1099 return lock_dir(dparent->d_inode, &qstr);
1101 down(&dparent->d_inode->i_sem);
1106 /* We never dget the object parent, so DON'T dput it either */
1107 static void filter_parent_unlock(struct dentry *dparent, void *lock)
1110 LASSERT(lock != NULL);
1111 unlock_dir(dparent->d_inode, lock);
1113 up(&dparent->d_inode->i_sem);
1117 /* We never dget the object parent, so DON'T dput it either */
1118 struct dentry *filter_parent(struct obd_device *obd, obd_gr group, obd_id objid)
1120 struct filter_obd *filter = &obd->u.filter;
1121 LASSERT(group < filter->fo_group_count);
1124 if (filter->fo_subdir_count == 0)
1125 return filter->fo_groups[group];
1127 return filter->fo_subdirs[group].dentry[objid & (filter->fo_subdir_count - 1)];
1130 /* We never dget the object parent, so DON'T dput it either */
1131 struct dentry *filter_parent_lock(struct obd_device *obd, obd_gr group,
1132 obd_id objid, void **lock)
1134 unsigned long now = jiffies;
1135 struct dentry *dparent = filter_parent(obd, group, objid);
1137 if (IS_ERR(dparent))
1141 LASSERT(dparent->d_inode);
1143 *lock = filter_lock_dentry(obd, dparent, objid);
1144 if (time_after(jiffies, now + 15 * HZ))
1145 CERROR("slow parent lock %lus\n", (jiffies - now) / HZ);
1149 /* How to get files, dentries, inodes from object id's.
1151 * If dir_dentry is passed, the caller has already locked the parent
1152 * appropriately for this operation (normally a write lock). If
1153 * dir_dentry is NULL, we do a read lock while we do the lookup to
1154 * avoid races with create/destroy and such changing the directory
1155 * internal to the filesystem code. */
1156 struct dentry *filter_id2dentry(struct obd_device *obd,
1157 struct dentry *dir_dentry,
1158 obd_gr group, obd_id id)
1160 struct dentry *dparent = dir_dentry;
1161 struct dentry *dchild;
1168 CERROR("fatal: invalid object id 0\n");
1169 RETURN(ERR_PTR(-ESTALE));
1172 len = sprintf(name, LPU64, id);
1173 if (dir_dentry == NULL) {
1174 dparent = filter_parent_lock(obd, group, id, &lock);
1175 if (IS_ERR(dparent))
1178 CDEBUG(D_INODE, "looking up object O/%*s/%s\n",
1179 dparent->d_name.len, dparent->d_name.name, name);
1180 dchild = /*ll_*/lookup_one_len(name, dparent, len);
1181 if (dir_dentry == NULL)
1182 filter_parent_unlock(dparent, lock);
1183 if (IS_ERR(dchild)) {
1184 CERROR("%s: child lookup error %ld\n", obd->obd_name,
1189 if (dchild->d_inode != NULL && is_bad_inode(dchild->d_inode)) {
1190 CERROR("%s: got bad inode "LPU64"\n", obd->obd_name, id);
1192 RETURN(ERR_PTR(-ENOENT));
1195 CDEBUG(D_INODE, "got child objid %s: %p, count = %d\n",
1196 name, dchild, atomic_read(&dchild->d_count));
1198 LASSERT(atomic_read(&dchild->d_count) > 0);
1203 static int filter_prepare_destroy(struct obd_device *obd, obd_id objid,
1206 struct lustre_handle lockh;
1207 int flags = LDLM_AST_DISCARD_DATA, rc;
1208 struct ldlm_res_id res_id = { .name = { objid, 0, group, 0 } };
1209 ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
1212 /* Tell the clients that the object is gone now and that they should
1213 * throw away any cached pages. */
1214 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, res_id,
1215 LDLM_EXTENT, &policy, LCK_PW,
1216 &flags, filter_blocking_ast, ldlm_completion_ast,
1217 NULL, NULL, NULL, 0, NULL, &lockh);
1219 /* We only care about the side-effects, just drop the lock. */
1221 ldlm_lock_decref(&lockh, LCK_PW);
1226 /* Caller must hold LCK_PW on parent and push us into kernel context.
1227 * Caller is also required to ensure that dchild->d_inode exists. */
1228 static int filter_destroy_internal(struct obd_device *obd, obd_id objid,
1229 struct dentry *dparent,
1230 struct dentry *dchild)
1232 struct inode *inode = dchild->d_inode;
1236 if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) {
1237 CERROR("destroying objid %*s nlink = %lu, count = %d\n",
1238 dchild->d_name.len, dchild->d_name.name,
1239 (unsigned long)inode->i_nlink,
1240 atomic_read(&inode->i_count));
1243 rc = vfs_unlink(dparent->d_inode, dchild);
1246 CERROR("error unlinking objid %*s: rc %d\n",
1247 dchild->d_name.len, dchild->d_name.name, rc);
1252 static int filter_intent_policy(struct ldlm_namespace *ns,
1253 struct ldlm_lock **lockp, void *req_cookie,
1254 ldlm_mode_t mode, int flags, void *data)
1256 struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1257 struct ptlrpc_request *req = req_cookie;
1258 struct ldlm_lock *lock = *lockp, *l = NULL;
1259 struct ldlm_resource *res = lock->l_resource;
1260 ldlm_processing_policy policy;
1261 struct ost_lvb *res_lvb, *reply_lvb;
1262 struct ldlm_reply *rep;
1263 struct list_head *tmp;
1265 int tmpflags = 0, rc, repsize[2] = {sizeof(*rep), sizeof(*reply_lvb)};
1268 policy = ldlm_get_processing_policy(res);
1269 LASSERT(policy != NULL);
1270 LASSERT(req != NULL);
1272 rc = lustre_pack_reply(req, 2, repsize, NULL);
1274 RETURN(req->rq_status = rc);
1276 rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*rep));
1277 LASSERT(rep != NULL);
1279 reply_lvb = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*reply_lvb));
1280 LASSERT(reply_lvb != NULL);
1282 //fixup_handle_for_resent_req(req, lock, &lockh);
1284 /* If we grant any lock at all, it will be a whole-file read lock.
1285 * Call the extent policy function to see if our request can be
1286 * granted, or is blocked. */
1287 lock->l_policy_data.l_extent.start = 0;
1288 lock->l_policy_data.l_extent.end = OBD_OBJECT_EOF;
1289 lock->l_req_mode = LCK_PR;
1291 l_lock(&res->lr_namespace->ns_lock);
1293 res->lr_tmp = &rpc_list;
1294 rc = policy(lock, &tmpflags, 0, &err);
1297 /* FIXME: we should change the policy function slightly, to not make
1298 * this list at all, since we just turn around and free it */
1299 while (!list_empty(&rpc_list)) {
1300 struct ldlm_ast_work *w =
1301 list_entry(rpc_list.next, struct ldlm_ast_work, w_list);
1302 list_del(&w->w_list);
1303 LDLM_LOCK_PUT(w->w_lock);
1304 OBD_FREE(w, sizeof(*w));
1307 if (rc == LDLM_ITER_CONTINUE) {
1308 /* The lock met with no resistance; we're finished. */
1309 l_unlock(&res->lr_namespace->ns_lock);
1310 RETURN(ELDLM_LOCK_REPLACED);
1313 /* Do not grant any lock, but instead send GL callbacks. The extent
1314 * policy nicely created a list of all PW locks for us. We will choose
1315 * the highest of those which are larger than the size in the LVB, if
1316 * any, and perform a glimpse callback. */
1317 down(&res->lr_lvb_sem);
1318 res_lvb = res->lr_lvb_data;
1319 LASSERT(res_lvb != NULL);
1320 *reply_lvb = *res_lvb;
1321 up(&res->lr_lvb_sem);
1323 list_for_each(tmp, &res->lr_granted) {
1324 struct ldlm_lock *tmplock =
1325 list_entry(tmp, struct ldlm_lock, l_res_link);
1327 if (tmplock->l_granted_mode == LCK_PR)
1330 if (tmplock->l_policy_data.l_extent.end <= reply_lvb->lvb_size)
1334 l = LDLM_LOCK_GET(tmplock);
1338 if (l->l_policy_data.l_extent.start >
1339 tmplock->l_policy_data.l_extent.start)
1343 l = LDLM_LOCK_GET(tmplock);
1345 l_unlock(&res->lr_namespace->ns_lock);
1347 /* There were no PW locks beyond the size in the LVB; finished. */
1349 RETURN(ELDLM_LOCK_ABORTED);
1351 if (l->l_glimpse_ast == NULL) {
1352 /* We are racing with unlink(); just return -ENOENT */
1353 rep->lock_policy_res1 = -ENOENT;
1357 LASSERTF(l->l_glimpse_ast != NULL, "l == %p", l);
1359 rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
1360 if (rc != 0 && res->lr_namespace->ns_lvbo &&
1361 res->lr_namespace->ns_lvbo->lvbo_update) {
1362 res->lr_namespace->ns_lvbo->lvbo_update(res, NULL, 0, 1);
1365 down(&res->lr_lvb_sem);
1366 *reply_lvb = *res_lvb;
1367 up(&res->lr_lvb_sem);
1371 RETURN(ELDLM_LOCK_ABORTED);
1374 static int filter_post_fs_cleanup(struct obd_device *obd)
1378 rc = fsfilt_post_cleanup(obd);
1383 static int filter_group_set_fs_flags(struct obd_device *obd, int group)
1385 struct filter_obd *filter = &obd->u.filter;
1389 /* zero group is not longer valid. */
1392 for (i = 0; i < filter->fo_subdir_count; i++) {
1393 struct dentry *dentry;
1394 dentry = (filter->fo_subdirs + group)->dentry[i];
1395 rc = fsfilt_set_fs_flags(obd, dentry->d_inode,
1396 SM_DO_REC | SM_DO_COW);
1402 static int filter_post_fs_setup(struct obd_device *obd)
1404 struct filter_obd *filter = &obd->u.filter;
1406 struct llog_ctxt *ctxt = NULL;
1408 rc = fsfilt_post_setup(obd, filter->fo_dentry_O);
1412 for (j = 0; j < filter->fo_group_count; j++) {
1413 rc = filter_group_set_fs_flags(obd, j);
1418 fsfilt_get_reint_log_ctxt(obd, filter->fo_sb, &ctxt);
1420 ctxt->loc_obd = obd;
1421 ctxt->loc_idx = LLOG_REINT_ORIG_CTXT;
1422 obd->obd_llog_ctxt[LLOG_REINT_ORIG_CTXT] = ctxt;
1424 fsfilt_set_ost_flags(obd, filter->fo_sb);
1428 /* mount the file system (secretly) */
1429 int filter_common_setup(struct obd_device *obd, obd_count len, void *buf,
1432 struct lustre_cfg* lcfg = buf;
1433 struct filter_obd *filter = &obd->u.filter;
1434 struct vfsmount *mnt;
1439 dev_clear_rdonly(2);
1441 if (!lcfg->lcfg_inlbuf1 || !lcfg->lcfg_inlbuf2)
1444 obd->obd_fsops = fsfilt_get_ops(lcfg->lcfg_inlbuf2);
1445 if (IS_ERR(obd->obd_fsops))
1446 RETURN(PTR_ERR(obd->obd_fsops));
1448 mnt = do_kern_mount(lcfg->lcfg_inlbuf2, MS_NOATIME | MS_NODIRATIME,
1449 lcfg->lcfg_inlbuf1, option);
1454 if (lcfg->lcfg_inllen3 > 0 && lcfg->lcfg_inlbuf3) {
1455 if (*lcfg->lcfg_inlbuf3 == 'f') {
1456 obd->obd_replayable = 1;
1457 obd_sync_filter = 1;
1458 CWARN("%s: recovery enabled\n", obd->obd_name);
1460 if (*lcfg->lcfg_inlbuf3 != 'n') {
1461 CERROR("unrecognised flag '%c'\n",
1462 *lcfg->lcfg_inlbuf3);
1464 // XXX Robert? Why do we get errors here
1465 // GOTO(err_mntput, rc = -EINVAL);
1469 filter->fo_vfsmnt = mnt;
1470 filter->fo_sb = mnt->mnt_sb;
1471 filter->fo_fstype = mnt->mnt_sb->s_type->name;
1472 CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
1474 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1475 obd->obd_lvfs_ctxt.pwdmnt = mnt;
1476 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
1477 obd->obd_lvfs_ctxt.fs = get_ds();
1478 obd->obd_lvfs_ctxt.cb_ops = filter_lvfs_ops;
1480 rc = fsfilt_setup(obd, mnt->mnt_sb);
1482 GOTO(err_mntput, rc);
1484 sema_init(&filter->fo_init_lock, 1);
1485 filter->fo_committed_group = 0;
1486 rc = filter_prep(obd);
1488 GOTO(err_mntput, rc);
1490 filter->fo_destroys_in_progress = 0;
1491 for (i = 0; i < 32; i++)
1492 sema_init(&filter->fo_create_locks[i], 1);
1494 spin_lock_init(&filter->fo_translock);
1495 spin_lock_init(&filter->fo_objidlock);
1496 INIT_LIST_HEAD(&filter->fo_export_list);
1497 sema_init(&filter->fo_alloc_lock, 1);
1498 spin_lock_init(&filter->fo_r_pages.oh_lock);
1499 spin_lock_init(&filter->fo_w_pages.oh_lock);
1500 spin_lock_init(&filter->fo_r_discont_pages.oh_lock);
1501 spin_lock_init(&filter->fo_w_discont_pages.oh_lock);
1502 spin_lock_init(&filter->fo_r_discont_blocks.oh_lock);
1503 spin_lock_init(&filter->fo_w_discont_blocks.oh_lock);
1504 filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE;
1506 INIT_LIST_HEAD(&filter->fo_llog_list);
1507 spin_lock_init(&filter->fo_llog_list_lock);
1509 sprintf(ns_name, "filter-%s", obd->obd_uuid.uuid);
1510 obd->obd_namespace = ldlm_namespace_new(ns_name, LDLM_NAMESPACE_SERVER);
1512 if (obd->obd_namespace == NULL)
1513 GOTO(err_post, rc = -ENOMEM);
1514 obd->obd_namespace->ns_lvbp = obd;
1515 obd->obd_namespace->ns_lvbo = &filter_lvbo;
1516 ldlm_register_intent(obd->obd_namespace, filter_intent_policy);
1518 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1519 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
1521 rc = obd_llog_cat_initialize(obd, &obd->obd_llogs, 1, CATLIST);
1523 CERROR("failed to setup llogging subsystems\n");
1536 fsfilt_put_ops(obd->obd_fsops);
1540 static int filter_attach(struct obd_device *obd, obd_count len, void *data)
1542 struct lprocfs_static_vars lvars;
1545 lprocfs_init_vars(filter, &lvars);
1546 rc = lprocfs_obd_attach(obd, lvars.obd_vars);
1550 rc = lprocfs_alloc_obd_stats(obd, LPROC_FILTER_LAST);
1554 /* Init obdfilter private stats here */
1555 lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_READ_BYTES,
1556 LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
1557 lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
1558 LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
1560 return lproc_filter_attach_seqstat(obd);
1563 static int filter_detach(struct obd_device *dev)
1565 lprocfs_free_obd_stats(dev);
1566 return lprocfs_obd_detach(dev);
1569 static int filter_setup(struct obd_device *obd, obd_count len, void *buf)
1571 struct lustre_cfg* lcfg = buf;
1574 /* all mount options including errors=remount-ro and asyncdel are passed
1575 * using 4th lcfg param. And it is good, finally we have got rid of
1576 * hardcoded fs types in the code. */
1577 rc = filter_common_setup(obd, len, buf, lcfg->lcfg_inlbuf4);
1580 rc = filter_post_fs_setup(obd);
1584 static int filter_cleanup(struct obd_device *obd, int flags)
1586 struct filter_obd *filter = &obd->u.filter;
1589 if (flags & OBD_OPT_FAILOVER)
1590 CERROR("%s: shutting down for failover; client state will"
1591 " be preserved.\n", obd->obd_name);
1593 if (!list_empty(&obd->obd_exports)) {
1594 CERROR("%s: still has clients!\n", obd->obd_name);
1595 class_disconnect_exports(obd, flags);
1596 if (!list_empty(&obd->obd_exports)) {
1597 CERROR("still has exports after forced cleanup?\n");
1602 target_cleanup_recovery(obd);
1604 ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
1606 if (filter->fo_sb == NULL)
1609 filter_post_fs_cleanup(obd);
1612 shrink_dcache_parent(filter->fo_sb->s_root);
1615 if (atomic_read(&filter->fo_vfsmnt->mnt_count) > 1)
1616 CERROR("%s: mount point %p busy, mnt_count: %d\n",
1617 obd->obd_name, filter->fo_vfsmnt,
1618 atomic_read(&filter->fo_vfsmnt->mnt_count));
1621 mntput(filter->fo_vfsmnt);
1622 //destroy_buffers(filter->fo_sb->s_dev);
1623 filter->fo_sb = NULL;
1624 fsfilt_put_ops(obd->obd_fsops);
1627 dev_clear_rdonly(2);
1632 static int filter_connect_post(struct obd_export *exp, unsigned long connect_flags)
1634 struct obd_device *obd = exp->exp_obd;
1635 struct filter_export_data *fed;
1636 char str[PTL_NALFMT_SIZE];
1637 struct obd_llogs *llog;
1638 struct llog_ctxt *ctxt;
1642 fed = &exp->exp_filter_data;
1643 if (fed->fed_group < FILTER_MIN_GROUPS)
1646 /* initialize llogs for connections from MDS */
1647 llog = filter_grab_llog_for_group(obd, fed->fed_group, exp);
1648 LASSERT(llog != NULL);
1650 ctxt = llog_get_context(llog, LLOG_UNLINK_REPL_CTXT);
1651 LASSERT(ctxt != NULL);
1653 rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse);
1654 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
1655 exp->exp_connection->c_peer.peer_id.nid, str);
1656 CDEBUG(D_OTHER, "%s: init llog ctxt for export "LPX64"/%s, group %d\n",
1657 obd->obd_name, exp->exp_connection->c_peer.peer_id.nid,
1658 str, fed->fed_group);
1663 /* nearly identical to mds_connect */
1664 static int filter_connect(struct lustre_handle *conn, struct obd_device *obd,
1665 struct obd_uuid *cluuid, unsigned long connect_flags)
1667 struct obd_export *exp;
1668 struct filter_export_data *fed;
1669 struct filter_client_data *fcd = NULL;
1670 struct filter_obd *filter = &obd->u.filter;
1671 struct lvfs_run_ctxt saved;
1675 if (conn == NULL || obd == NULL || cluuid == NULL)
1678 rc = class_connect(conn, obd, cluuid);
1681 exp = class_conn2export(conn);
1682 LASSERT(exp != NULL);
1684 fed = &exp->exp_filter_data;
1686 spin_lock_init(&fed->fed_lock);
1688 /* connection from MDS */
1689 group = connect_flags;
1690 if (obd->obd_replayable) {
1691 OBD_ALLOC(fcd, sizeof(*fcd));
1693 CERROR("filter: out of memory for client data\n");
1694 GOTO(cleanup, rc = -ENOMEM);
1697 memcpy(fcd->fcd_uuid, cluuid, sizeof(fcd->fcd_uuid));
1699 fed->fed_fcd->fcd_group = group;
1700 rc = filter_client_add(obd, filter, fed, -1);
1704 CWARN("%s: Received MDS connection ("LPX64"); group %d\n",
1705 obd->obd_name, exp->exp_handle.h_cookie, group);
1710 if (fed->fed_group != 0 && fed->fed_group != group) {
1711 char str[PTL_NALFMT_SIZE];
1712 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
1713 exp->exp_connection->c_peer.peer_id.nid, str);
1714 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
1715 "earlier; now it's trying to use group %d! This could "
1716 "be a bug in the MDS. Tell CFS.\n",
1717 exp->exp_connection->c_peer.peer_id.nid, str,
1718 fed->fed_group, group);
1719 GOTO(cleanup, rc = -EPROTO);
1721 fed->fed_group = group;
1723 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1724 rc = filter_read_groups(obd, group, 1);
1725 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1727 CERROR("can't read group %u\n", group);
1730 rc = filter_group_set_fs_flags(obd, group);
1732 CERROR("can't set kml flags %u\n", group);
1739 OBD_FREE(fcd, sizeof(*fcd));
1740 class_disconnect(exp, 0);
1742 class_export_put(exp);
1747 static int filter_precleanup(struct obd_device *obd, int flags)
1749 struct filter_group_llog *log;
1750 struct filter_obd *filter;
1754 filter = &obd->u.filter;
1756 spin_lock(&filter->fo_llog_list_lock);
1757 while (!list_empty(&filter->fo_llog_list)) {
1758 log = list_entry(filter->fo_llog_list.next,
1759 struct filter_group_llog, list);
1760 list_del(&log->list);
1761 spin_unlock(&filter->fo_llog_list_lock);
1763 rc = obd_llog_finish(obd, log->llogs, 0);
1765 CERROR("failed to cleanup llogging subsystem for %u\n",
1767 OBD_FREE(log->llogs, sizeof(*(log->llogs)));
1768 OBD_FREE(log, sizeof(*log));
1769 spin_lock(&filter->fo_llog_list_lock);
1771 spin_unlock(&filter->fo_llog_list_lock);
1773 rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1775 CERROR("failed to cleanup llogging subsystem\n");
1780 /* Do extra sanity checks for grant accounting. We do this at connect,
1781 * disconnect, and statfs RPC time, so it shouldn't be too bad. We can
1782 * always get rid of it or turn it off when we know accounting is good. */
1783 static void filter_grant_sanity_check(struct obd_device *obd, const char *func)
1785 struct filter_export_data *fed;
1786 struct obd_export *exp;
1787 obd_size maxsize = obd->obd_osfs.os_blocks * obd->obd_osfs.os_bsize;
1788 obd_size tot_dirty = 0, tot_pending = 0, tot_granted = 0;
1789 obd_size fo_tot_dirty, fo_tot_pending, fo_tot_granted;
1791 if (list_empty(&obd->obd_exports))
1794 spin_lock(&obd->obd_osfs_lock);
1795 spin_lock(&obd->obd_dev_lock);
1796 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1797 fed = &exp->exp_filter_data;
1798 LASSERTF(fed->fed_grant + fed->fed_pending <= maxsize,
1799 "cli %s/%p %lu+%lu > "LPU64"\n",
1800 exp->exp_client_uuid.uuid, exp,
1801 fed->fed_grant, fed->fed_pending, maxsize);
1802 LASSERTF(fed->fed_dirty <= maxsize, "cli %s/%p %lu > "LPU64"\n",
1803 exp->exp_client_uuid.uuid, exp,fed->fed_dirty,maxsize);
1804 CDEBUG(D_CACHE,"%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1805 obd->obd_name, exp->exp_client_uuid.uuid, exp,
1806 fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1807 tot_granted += fed->fed_grant + fed->fed_pending;
1808 tot_pending += fed->fed_pending;
1809 tot_dirty += fed->fed_dirty;
1811 fo_tot_granted = obd->u.filter.fo_tot_granted;
1812 fo_tot_pending = obd->u.filter.fo_tot_pending;
1813 fo_tot_dirty = obd->u.filter.fo_tot_dirty;
1814 spin_unlock(&obd->obd_dev_lock);
1815 spin_unlock(&obd->obd_osfs_lock);
1817 /* Do these assertions outside the spinlocks so we don't kill system */
1818 if (tot_granted != fo_tot_granted)
1819 CERROR("%s: tot_granted "LPU64" != fo_tot_granted "LPU64"\n",
1820 func, tot_granted, fo_tot_granted);
1821 if (tot_pending != fo_tot_pending)
1822 CERROR("%s: tot_pending "LPU64" != fo_tot_pending "LPU64"\n",
1823 func, tot_pending, fo_tot_pending);
1824 if (tot_dirty != fo_tot_dirty)
1825 CERROR("%s: tot_dirty "LPU64" != fo_tot_dirty "LPU64"\n",
1826 func, tot_dirty, fo_tot_dirty);
1827 if (tot_pending > tot_granted)
1828 CERROR("%s: tot_pending "LPU64" > tot_granted "LPU64"\n",
1829 func, tot_pending, tot_granted);
1830 if (tot_granted > maxsize)
1831 CERROR("%s: tot_granted "LPU64" > maxsize "LPU64"\n",
1832 func, tot_granted, maxsize);
1833 if (tot_dirty > maxsize)
1834 CERROR("%s: tot_dirty "LPU64" > maxsize "LPU64"\n",
1835 func, tot_dirty, maxsize);
1838 /* Remove this client from the grant accounting totals. We also remove
1839 * the export from the obd device under the osfs and dev locks to ensure
1840 * that the filter_grant_sanity_check() calculations are always valid.
1841 * The client should do something similar when it invalidates its import. */
1842 static void filter_grant_discard(struct obd_export *exp)
1844 struct obd_device *obd = exp->exp_obd;
1845 struct filter_obd *filter = &obd->u.filter;
1846 struct filter_export_data *fed = &exp->exp_filter_data;
1848 spin_lock(&obd->obd_osfs_lock);
1849 spin_lock(&exp->exp_obd->obd_dev_lock);
1850 list_del_init(&exp->exp_obd_chain);
1851 spin_unlock(&exp->exp_obd->obd_dev_lock);
1853 CDEBUG(D_CACHE, "%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1854 obd->obd_name, exp->exp_client_uuid.uuid, exp,
1855 fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1857 LASSERTF(filter->fo_tot_granted >= fed->fed_grant,
1858 "%s: tot_granted "LPU64" cli %s/%p fed_grant %lu\n",
1859 obd->obd_name, filter->fo_tot_granted,
1860 exp->exp_client_uuid.uuid, exp, fed->fed_grant);
1861 filter->fo_tot_granted -= fed->fed_grant;
1862 LASSERTF(exp->exp_obd->u.filter.fo_tot_pending >= fed->fed_pending,
1863 "%s: tot_pending "LPU64" cli %s/%p fed_pending %lu\n",
1864 obd->obd_name, filter->fo_tot_pending,
1865 exp->exp_client_uuid.uuid, exp, fed->fed_pending);
1866 LASSERTF(filter->fo_tot_dirty >= fed->fed_dirty,
1867 "%s: tot_dirty "LPU64" cli %s/%p fed_dirty %lu\n",
1868 obd->obd_name, filter->fo_tot_dirty,
1869 exp->exp_client_uuid.uuid, exp, fed->fed_dirty);
1870 filter->fo_tot_dirty -= fed->fed_dirty;
1874 spin_unlock(&obd->obd_osfs_lock);
1877 static int filter_destroy_export(struct obd_export *exp)
1881 if (exp->exp_filter_data.fed_pending)
1882 CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
1883 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
1884 exp, exp->exp_filter_data.fed_pending);
1886 target_destroy_export(exp);
1888 if (exp->exp_obd->obd_replayable)
1889 filter_client_free(exp, exp->exp_flags);
1891 filter_grant_discard(exp);
1892 if (!(exp->exp_flags & OBD_OPT_FORCE))
1893 filter_grant_sanity_check(exp->exp_obd, __FUNCTION__);
1898 static void filter_sync_llogs(struct obd_device *obd, struct obd_export *dexp)
1900 struct filter_group_llog *fglog, *nlog;
1901 struct filter_obd *filter;
1902 int worked = 0, group;
1903 struct llog_ctxt *ctxt;
1906 filter = &obd->u.filter;
1908 /* we can't sync log holding spinlock. also, we do not want to get
1909 * into livelock. so we do following: loop over MDS's exports in
1910 * group order and skip already synced llogs -bzzz */
1912 /* look for group with min. number, but > worked */
1915 spin_lock(&filter->fo_llog_list_lock);
1916 list_for_each_entry(nlog, &filter->fo_llog_list, list) {
1918 if (nlog->group <= worked) {
1919 /* this group is already synced */
1923 if (group < nlog->group) {
1924 /* we have group with smaller number to sync */
1928 /* store current minimal group */
1930 group = nlog->group;
1932 spin_unlock(&filter->fo_llog_list_lock);
1937 worked = fglog->group;
1938 if (fglog->exp && (dexp == fglog->exp || dexp == NULL)) {
1939 ctxt = llog_get_context(fglog->llogs,
1940 LLOG_UNLINK_REPL_CTXT);
1941 LASSERT(ctxt != NULL);
1942 llog_sync(ctxt, fglog->exp);
1944 } while (fglog != NULL);
1947 /* also incredibly similar to mds_disconnect */
1948 static int filter_disconnect(struct obd_export *exp, unsigned long flags)
1950 struct obd_device *obd = exp->exp_obd;
1951 unsigned long irqflags;
1956 class_export_get(exp);
1958 spin_lock_irqsave(&exp->exp_lock, irqflags);
1959 exp->exp_flags = flags;
1960 spin_unlock_irqrestore(&exp->exp_lock, irqflags);
1962 if (!(flags & OBD_OPT_FORCE))
1963 filter_grant_sanity_check(obd, __FUNCTION__);
1964 filter_grant_discard(exp);
1966 /* Disconnect early so that clients can't keep using export */
1967 rc = class_disconnect(exp, flags);
1969 ldlm_cancel_locks_for_export(exp);
1971 fsfilt_sync(obd, obd->u.filter.fo_sb);
1973 /* flush any remaining cancel messages out to the target */
1974 filter_sync_llogs(obd, exp);
1976 class_export_put(exp);
1980 struct dentry *__filter_oa2dentry(struct obd_device *obd,
1981 struct obdo *oa, const char *what)
1983 struct dentry *dchild = NULL;
1986 if (oa->o_valid & OBD_MD_FLGROUP)
1989 dchild = filter_id2dentry(obd, NULL, group, oa->o_id);
1991 if (IS_ERR(dchild)) {
1992 CERROR("%s error looking up object: "LPU64"\n", what, oa->o_id);
1996 if (dchild->d_inode == NULL) {
1997 CERROR("%s: %s on non-existent object: "LPU64"\n",
1998 obd->obd_name, what, oa->o_id);
2000 RETURN(ERR_PTR(-ENOENT));
2006 static int filter_getattr(struct obd_export *exp, struct obdo *oa,
2007 struct lov_stripe_md *md)
2009 struct dentry *dentry = NULL;
2010 struct obd_device *obd;
2014 obd = class_exp2obd(exp);
2016 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2017 exp->exp_handle.h_cookie);
2021 dentry = filter_oa2dentry(obd, oa);
2023 RETURN(PTR_ERR(dentry));
2025 /* Limit the valid bits in the return data to what we actually use */
2026 oa->o_valid = OBD_MD_FLID;
2027 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2033 /* this is called from filter_truncate() until we have filter_punch() */
2034 static int filter_setattr(struct obd_export *exp, struct obdo *oa,
2035 struct lov_stripe_md *md, struct obd_trans_info *oti)
2037 struct lvfs_run_ctxt saved;
2038 struct filter_obd *filter;
2039 struct dentry *dentry;
2041 struct ldlm_res_id res_id = { .name = { oa->o_id, 0, oa->o_gr, 0 } };
2042 struct ldlm_resource *res;
2047 LASSERT(oti != NULL);
2049 dentry = filter_oa2dentry(exp->exp_obd, oa);
2051 RETURN(PTR_ERR(dentry));
2053 filter = &exp->exp_obd->u.filter;
2055 iattr_from_obdo(&iattr, oa, oa->o_valid);
2057 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2060 if (iattr.ia_valid & ATTR_SIZE)
2061 down(&dentry->d_inode->i_sem);
2062 handle = fsfilt_start(exp->exp_obd, dentry->d_inode, FSFILT_OP_SETATTR,
2065 GOTO(out_unlock, rc = PTR_ERR(handle));
2067 /* XXX this could be a rwsem instead, if filter_preprw played along */
2068 if (iattr.ia_valid & ATTR_ATTR_FLAG)
2069 rc = fsfilt_iocontrol(exp->exp_obd, dentry->d_inode, NULL,
2071 (long)&iattr.ia_attr_flags);
2073 rc = fsfilt_setattr(exp->exp_obd, dentry, handle, &iattr, 1);
2074 rc = filter_finish_transno(exp, oti, rc);
2075 rc2 = fsfilt_commit(exp->exp_obd, filter->fo_sb, dentry->d_inode, handle, 0);
2077 CERROR("error on commit, err = %d\n", rc2);
2082 if (iattr.ia_valid & ATTR_SIZE) {
2083 res = ldlm_resource_get(exp->exp_obd->obd_namespace, NULL,
2084 res_id, LDLM_EXTENT, 0);
2086 CERROR("!!! resource_get failed for object "LPU64" -- "
2087 "filter_setattr with no lock?\n", oa->o_id);
2089 if (res->lr_namespace->ns_lvbo &&
2090 res->lr_namespace->ns_lvbo->lvbo_update) {
2091 rc = res->lr_namespace->ns_lvbo->lvbo_update
2094 ldlm_resource_putref(res);
2098 oa->o_valid = OBD_MD_FLID;
2099 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2102 if (iattr.ia_valid & ATTR_SIZE)
2103 up(&dentry->d_inode->i_sem);
2105 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2111 /* XXX identical to osc_unpackmd */
2112 static int filter_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2113 struct lov_mds_md *lmm, int lmm_bytes)
2119 if (lmm_bytes < sizeof (*lmm)) {
2120 CERROR("lov_mds_md too small: %d, need %d\n",
2121 lmm_bytes, (int)sizeof(*lmm));
2124 /* XXX LOV_MAGIC etc check? */
2126 if (lmm->lmm_object_id == cpu_to_le64(0)) {
2127 CERROR("lov_mds_md: zero lmm_object_id\n");
2132 lsm_size = lov_stripe_md_size(1);
2136 if (*lsmp != NULL && lmm == NULL) {
2137 OBD_FREE(*lsmp, lsm_size);
2142 if (*lsmp == NULL) {
2143 OBD_ALLOC(*lsmp, lsm_size);
2147 loi_init((*lsmp)->lsm_oinfo);
2151 /* XXX zero *lsmp? */
2152 (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
2153 LASSERT((*lsmp)->lsm_object_id);
2156 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
2161 static void filter_destroy_precreated(struct obd_export *exp, struct obdo *oa,
2162 struct filter_obd *filter)
2164 struct obdo *doa = NULL;
2169 LASSERT(oa->o_gr != 0);
2170 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2174 CERROR("cannot allocate doa, error %d\n",
2180 doa->o_mode = S_IFREG;
2181 doa->o_gr = oa->o_gr;
2182 doa->o_valid = oa->o_valid & (OBD_MD_FLGROUP | OBD_MD_FLID);
2184 set_bit(doa->o_gr, &filter->fo_destroys_in_progress);
2185 down(&filter->fo_create_locks[doa->o_gr]);
2186 if (!test_bit(doa->o_gr, &filter->fo_destroys_in_progress)) {
2187 CERROR("%s:["LPU64"] destroy_in_progress already cleared\n",
2188 exp->exp_obd->obd_name, doa->o_gr);
2189 up(&filter->fo_create_locks[doa->o_gr]);
2190 GOTO(out_free_doa, 0);
2193 last = filter_last_id(filter, doa->o_gr);
2194 CWARN("%s:["LPU64"] deleting orphan objects from "LPU64" to "LPU64"\n",
2195 exp->exp_obd->obd_name, doa->o_gr, oa->o_id + 1, last);
2196 for (id = oa->o_id + 1; id <= last; id++) {
2198 filter_destroy(exp, doa, NULL, NULL);
2201 CDEBUG(D_HA, "%s:["LPU64"] after destroy: set last_objids = "LPU64"\n",
2202 exp->exp_obd->obd_name, doa->o_gr, oa->o_id);
2204 filter_set_last_id(filter, doa->o_gr, oa->o_id);
2206 clear_bit(doa->o_gr, &filter->fo_destroys_in_progress);
2207 up(&filter->fo_create_locks[doa->o_gr]);
2214 /* returns a negative error or a nonnegative number of files to create */
2215 static int filter_should_precreate(struct obd_export *exp, struct obdo *oa,
2218 struct obd_device *obd = exp->exp_obd;
2219 struct filter_obd *filter = &obd->u.filter;
2223 diff = oa->o_id - filter_last_id(filter, oa->o_gr);
2224 CDEBUG(D_INFO, "filter_last_id() = "LPU64" -> diff = %d\n",
2225 filter_last_id(filter, oa->o_gr), diff);
2227 /* delete orphans request */
2228 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2229 (oa->o_flags & OBD_FL_DELORPHAN)) {
2232 if (-diff > 10000) { /* XXX make this smarter */
2233 CERROR("ignoring bogus orphan destroy request: obdid "
2234 LPU64" last_id "LPU64"\n",
2235 oa->o_id, filter_last_id(filter, oa->o_gr));
2238 filter_destroy_precreated(exp, oa, filter);
2239 rc = filter_update_last_objid(obd, group, 0);
2241 CERROR("unable to write lastobjid, but orphans"
2245 /* only precreate if group == 0 and o_id is specfied */
2246 if (!(oa->o_valid & OBD_FL_DELORPHAN) &&
2247 (/*group != 0 ||*/ oa->o_id == 0))
2254 static int filter_precreate_rec(struct obd_device *obd, struct dentry *dentry,
2255 int *number, struct obdo *oa)
2260 rc = fsfilt_precreate_rec(obd, dentry, number, oa);
2265 /* We rely on the fact that only one thread will be creating files in a given
2266 * group at a time, which is why we don't need an atomic filter_get_new_id.
2267 * Even if we had that atomic function, the following race would exist:
2269 * thread 1: gets id x from filter_next_id
2270 * thread 2: gets id (x + 1) from filter_next_id
2271 * thread 2: creates object (x + 1)
2272 * thread 1: tries to create object x, gets -ENOSPC
2274 static int filter_precreate(struct obd_device *obd, struct obdo *oa,
2275 obd_gr group, int *num)
2277 struct dentry *dchild = NULL, *dparent = NULL;
2278 int err = 0, rc = 0, recreate_obj = 0, i;
2279 struct filter_obd *filter;
2280 void *handle = NULL;
2285 filter = &obd->u.filter;
2287 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2288 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
2292 CDEBUG(D_HA, "%s: precreating %d objects\n", obd->obd_name, *num);
2294 down(&filter->fo_create_locks[group]);
2296 for (i = 0; i < *num && err == 0; i++) {
2297 int cleanup_phase = 0;
2299 if (test_bit(group, &filter->fo_destroys_in_progress)) {
2300 CWARN("%s: precreate aborted by destroy\n",
2308 last_id = filter_last_id(filter, group);
2309 if (next_id > last_id) {
2310 CERROR("Error: Trying to recreate obj greater"
2311 "than last id "LPD64" > "LPD64"\n",
2313 GOTO(cleanup, rc = -EINVAL);
2316 next_id = filter_last_id(filter, group) + 1;
2319 CDEBUG(D_INFO, "precreate objid "LPU64"\n", next_id);
2321 dparent = filter_parent_lock(obd, group, next_id, &lock);
2322 if (IS_ERR(dparent))
2323 GOTO(cleanup, rc = PTR_ERR(dparent));
2326 /*only do precreate rec record. so clean kml flags here*/
2327 fsfilt_clear_fs_flags(obd, dparent->d_inode, SM_DO_REC);
2329 dchild = filter_id2dentry(obd, dparent, group, next_id);
2331 GOTO(cleanup, rc = PTR_ERR(dchild));
2334 if (dchild->d_inode != NULL) {
2335 /* This would only happen if lastobjid was bad on disk*/
2336 /* Could also happen if recreating missing obj but
2340 CERROR("%s: Serious error: recreating obj %*s "
2341 "but obj already exists \n",
2342 obd->obd_name, dchild->d_name.len,
2343 dchild->d_name.name);
2346 CERROR("%s: Serious error: objid %*s already "
2347 "exists; is this filesystem corrupt?\n",
2348 obd->obd_name, dchild->d_name.len,
2349 dchild->d_name.name);
2352 GOTO(cleanup, rc = -EEXIST);
2355 handle = fsfilt_start_log(obd, dparent->d_inode,
2356 FSFILT_OP_CREATE, NULL, 1);
2358 GOTO(cleanup, rc = PTR_ERR(handle));
2361 rc = ll_vfs_create(dparent->d_inode, dchild, S_IFREG, NULL);
2363 CERROR("create failed rc = %d\n", rc);
2367 if (!recreate_obj) {
2368 filter_set_last_id(filter, group, next_id);
2369 err = filter_update_last_objid(obd, group, 0);
2371 CERROR("unable to write lastobjid "
2372 "but file created\n");
2374 fsfilt_set_fs_flags(obd, dparent->d_inode, SM_DO_REC);
2377 switch(cleanup_phase) {
2379 err = fsfilt_commit(obd, filter->fo_sb,
2380 dparent->d_inode, handle, 0);
2382 CERROR("error on commit, err = %d\n", err);
2389 filter_parent_unlock(dparent, lock);
2400 /* check if we have an error after ll_vfs_create(). It is possible that
2401 * there will be say -ENOSPC and we will leak it. */
2403 rc = filter_precreate_rec(obd, dparent, num, oa);
2405 up(&filter->fo_create_locks[group]);
2407 CDEBUG(D_HA, "%s: server last_objid for group "LPU64": "LPU64"\n",
2408 obd->obd_name, group, filter->fo_last_objids[group]);
2410 CDEBUG(D_HA, "%s: filter_precreate() created %d objects\n",
2416 static int filter_create(struct obd_export *exp, struct obdo *oa,
2417 struct lov_stripe_md **ea, struct obd_trans_info *oti)
2419 struct obd_device *obd = NULL;
2420 struct filter_obd *filter;
2421 struct lvfs_run_ctxt saved;
2422 struct lov_stripe_md *lsm = NULL;
2423 struct filter_export_data *fed;
2424 char str[PTL_NALFMT_SIZE];
2425 int group = oa->o_gr, rc = 0, diff, recreate_objs = 0;
2428 if (!(oa->o_valid & OBD_MD_FLGROUP) || group == 0) {
2429 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2430 exp->exp_connection->c_peer.peer_id.nid, str);
2431 CERROR("!!! nid "LPX64"/%s sent invalid object group %d\n",
2432 exp->exp_connection->c_peer.peer_id.nid, str, group);
2436 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2437 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
2442 fed = &exp->exp_filter_data;
2443 filter = &obd->u.filter;
2445 if (fed->fed_group != group && !recreate_objs &&
2446 !(oa->o_valid & OBD_MD_REINT)) {
2447 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2448 exp->exp_connection->c_peer.peer_id.nid, str);
2449 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
2450 "earlier; now it's trying to use group %d! This could "
2451 "be a bug in the MDS. Tell CFS.\n",
2452 exp->exp_connection->c_peer.peer_id.nid, str,
2453 fed->fed_group, group);
2457 CDEBUG(D_INFO, "filter_create(od->o_gr=%d,od->o_id="LPU64")\n",
2462 rc = obd_alloc_memmd(exp, &lsm);
2469 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2471 if (oa->o_valid & OBD_MD_REINT) {
2472 int num = *((int*)oa->o_inline);
2473 rc = filter_precreate(obd, oa, oa->o_gr, &num);
2474 } else if (recreate_objs) {
2475 if (oa->o_id > filter_last_id(&obd->u.filter, group)) {
2476 CERROR("recreate objid "LPU64" > last id "LPU64"\n",
2477 oa->o_id, filter_last_id(&obd->u.filter, group));
2481 rc = filter_precreate(obd, oa, group, &diff);
2484 diff = filter_should_precreate(exp, oa, group);
2486 oa->o_id = filter_last_id(&obd->u.filter, group);
2487 rc = filter_precreate(obd, oa, group, &diff);
2489 oa->o_valid = OBD_MD_FLID;
2493 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2494 if (rc && ea != NULL && *ea != lsm) {
2495 obd_free_memmd(exp, &lsm);
2496 } else if (rc == 0 && ea != NULL) {
2497 /* XXX LOV STACKING: the lsm that is passed to us from
2498 * LOV does not have valid lsm_oinfo data structs, so
2499 * don't go touching that. This needs to be fixed in a
2501 lsm->lsm_object_id = oa->o_id;
2502 lsm->lsm_object_gr = oa->o_gr;
2509 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
2510 struct lov_stripe_md *ea, struct obd_trans_info *oti)
2512 int rc, rc2, cleanup_phase = 0, have_prepared = 0;
2513 struct dentry *dchild = NULL, *dparent = NULL;
2514 struct llog_cookie *fcc = NULL;
2515 struct lvfs_run_ctxt saved;
2516 struct filter_obd *filter;
2517 struct obd_device *obd;
2518 void *handle = NULL;
2523 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2526 filter = &obd->u.filter;
2528 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2531 dparent = filter_parent_lock(obd, oa->o_gr, oa->o_id, &lock);
2532 if (IS_ERR(dparent))
2533 GOTO(cleanup, rc = PTR_ERR(dparent));
2536 dchild = filter_id2dentry(obd, dparent, oa->o_gr, oa->o_id);
2538 GOTO(cleanup, rc = -ENOENT);
2541 if (dchild->d_inode == NULL) {
2542 CDEBUG(D_INODE, "destroying non-existent object "LPU64"\n",
2544 GOTO(cleanup, rc = -ENOENT);
2547 if (!have_prepared) {
2548 /* If we're really going to destroy the object, get ready
2549 * by getting the clients to discard their cached data.
2551 * We have to drop the parent lock, because
2552 * filter_prepare_destroy will acquire a PW on the object, and
2553 * we don't want to deadlock with an incoming write to the
2554 * object, which has the extent PW and then wants to get the
2555 * parent dentry to do the lookup.
2557 * We dput the child because it's not worth the extra
2558 * complication of condition the above code to skip it on the
2559 * second time through. */
2561 filter_parent_unlock(dparent, lock);
2563 filter_prepare_destroy(obd, oa->o_id, oa->o_gr);
2568 /* Our MDC connection is established by the MDS to us */
2569 if (oa->o_valid & OBD_MD_FLCOOKIE) {
2570 OBD_ALLOC(fcc, sizeof(*fcc));
2572 memcpy(fcc, obdo_logcookie(oa), sizeof(*fcc));
2575 /* we're gonna truncate it first in order to avoid possible
2578 * open trasaction open transaction
2579 * down(i_zombie) down(i_zombie)
2580 * restart transaction
2581 * (see BUG 4180) -bzzz
2583 down(&dchild->d_inode->i_sem);
2584 handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_SETATTR,NULL,1);
2585 if (IS_ERR(handle)) {
2586 up(&dchild->d_inode->i_sem);
2587 GOTO(cleanup, rc = PTR_ERR(handle));
2590 iattr.ia_valid = ATTR_SIZE;
2592 rc = fsfilt_setattr(obd, dchild, handle, &iattr, 1);
2594 rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode, handle, 0);
2595 up(&dchild->d_inode->i_sem);
2599 GOTO(cleanup, rc = rc2);
2601 handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_UNLINK,oti,1);
2603 GOTO(cleanup, rc = PTR_ERR(handle));
2607 rc = filter_destroy_internal(obd, oa->o_id, dparent, dchild);
2610 switch(cleanup_phase) {
2614 fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
2616 filter_cancel_cookies_cb,
2619 fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
2621 filter_cancel_cookies_cb,
2624 rc = filter_finish_transno(exp, oti, rc);
2625 rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode,
2628 CERROR("error on commit, err = %d\n", rc2);
2635 filter_parent_unlock(dparent, lock);
2637 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2640 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
2647 /* NB start and end are used for punch, but not truncate */
2648 static int filter_truncate(struct obd_export *exp, struct obdo *oa,
2649 struct lov_stripe_md *lsm,
2650 obd_off start, obd_off end,
2651 struct obd_trans_info *oti)
2656 if (end != OBD_OBJECT_EOF)
2657 CERROR("PUNCH not supported, only truncate: end = "LPX64"\n",
2660 CDEBUG(D_INODE, "calling truncate for object "LPU64", valid = "LPU64", "
2661 "o_size = "LPD64"\n", oa->o_id, oa->o_valid, start);
2663 error = filter_setattr(exp, oa, NULL, oti);
2667 static int filter_sync(struct obd_export *exp, struct obdo *oa,
2668 struct lov_stripe_md *lsm, obd_off start, obd_off end)
2670 struct obd_device *obd = exp->exp_obd;
2671 struct lvfs_run_ctxt saved;
2672 struct filter_obd *filter;
2673 struct dentry *dentry;
2677 filter = &obd->u.filter;
2679 /* an objid of zero is taken to mean "sync whole filesystem" */
2680 if (!oa || !(oa->o_valid & OBD_MD_FLID)) {
2681 rc = fsfilt_sync(obd, filter->fo_sb);
2682 /* flush any remaining cancel messages out to the target */
2683 filter_sync_llogs(obd, NULL);
2687 dentry = filter_oa2dentry(obd, oa);
2689 RETURN(PTR_ERR(dentry));
2691 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2693 down(&dentry->d_inode->i_sem);
2694 rc = filemap_fdatasync(dentry->d_inode->i_mapping);
2696 /* just any file to grab fsync method - "file" arg unused */
2697 struct file *file = filter->fo_rcvd_filp;
2699 if (file->f_op && file->f_op->fsync)
2700 rc = file->f_op->fsync(NULL, dentry, 1);
2702 rc2 = filemap_fdatawait(dentry->d_inode->i_mapping);
2706 up(&dentry->d_inode->i_sem);
2708 oa->o_valid = OBD_MD_FLID;
2709 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2711 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2717 static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2718 unsigned long max_age)
2720 struct filter_obd *filter = &obd->u.filter;
2721 int blockbits = filter->fo_sb->s_blocksize_bits;
2725 /* at least try to account for cached pages. its still racey and
2726 * might be under-reporting if clients haven't announced their
2727 * caches with brw recently */
2728 spin_lock(&obd->obd_osfs_lock);
2729 rc = fsfilt_statfs(obd, filter->fo_sb, max_age);
2730 memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
2731 spin_unlock(&obd->obd_osfs_lock);
2733 CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
2734 " pending "LPU64" free "LPU64" avail "LPU64"\n",
2735 filter->fo_tot_dirty, filter->fo_tot_granted,
2736 filter->fo_tot_pending,
2737 osfs->os_bfree << blockbits, osfs->os_bavail << blockbits);
2739 filter_grant_sanity_check(obd, __FUNCTION__);
2741 osfs->os_bavail -= min(osfs->os_bavail,
2742 (filter->fo_tot_dirty + filter->fo_tot_pending +
2743 osfs->os_bsize -1) >> blockbits);
2748 static int filter_get_info(struct obd_export *exp, __u32 keylen,
2749 void *key, __u32 *vallen, void *val)
2751 struct filter_export_data *fed = &exp->exp_filter_data;
2752 struct obd_device *obd;
2755 obd = class_exp2obd(exp);
2757 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2758 exp->exp_handle.h_cookie);
2762 if (keylen == strlen("blocksize") &&
2763 memcmp(key, "blocksize", keylen) == 0) {
2764 __u32 *blocksize = val;
2765 *vallen = sizeof(*blocksize);
2766 *blocksize = obd->u.filter.fo_sb->s_blocksize;
2770 if (keylen == strlen("blocksize_bits") &&
2771 memcmp(key, "blocksize_bits", keylen) == 0) {
2772 __u32 *blocksize_bits = val;
2773 *vallen = sizeof(*blocksize_bits);
2774 *blocksize_bits = obd->u.filter.fo_sb->s_blocksize_bits;
2778 if (keylen >= strlen("last_id") && memcmp(key, "last_id", 7) == 0) {
2779 obd_id *last_id = val;
2780 *last_id = filter_last_id(&obd->u.filter, fed->fed_group);
2783 if (keylen >= strlen("reint_log") && memcmp(key, "reint_log", 9) == 0) {
2784 /*Get log_context handle*/
2785 unsigned long *llh_handle = val;
2786 *vallen = sizeof(unsigned long);
2787 *llh_handle = (unsigned long)obd->obd_llog_ctxt[LLOG_REINT_ORIG_CTXT];
2790 if (keylen >= strlen("cache_sb") && memcmp(key, "cache_sb", 8) == 0) {
2791 /*Get log_context handle*/
2792 unsigned long *sb = val;
2793 *vallen = sizeof(unsigned long);
2794 *sb = (unsigned long)obd->u.filter.fo_sb;
2798 CDEBUG(D_IOCTL, "invalid key\n");
2802 struct obd_llogs *filter_grab_llog_for_group(struct obd_device *obd, int group,
2803 struct obd_export *export)
2805 struct filter_group_llog *fglog, *nlog;
2806 char name[32] = "CATLIST";
2807 struct filter_obd *filter;
2808 struct llog_ctxt *ctxt;
2809 struct list_head *cur;
2812 filter = &obd->u.filter;
2814 spin_lock(&filter->fo_llog_list_lock);
2815 list_for_each(cur, &filter->fo_llog_list) {
2816 fglog = list_entry(cur, struct filter_group_llog, list);
2817 if (fglog->group == group) {
2818 if (!(fglog->exp == NULL || fglog->exp == export || export == NULL))
2819 CWARN("%s: export for group %d changes: 0x%p -> 0x%p\n",
2820 obd->obd_name, group, fglog->exp, export);
2821 spin_unlock(&filter->fo_llog_list_lock);
2825 spin_unlock(&filter->fo_llog_list_lock);
2830 OBD_ALLOC(fglog, sizeof(*fglog));
2833 fglog->group = group;
2835 OBD_ALLOC(fglog->llogs, sizeof(struct obd_llogs));
2836 if (fglog->llogs == NULL) {
2837 OBD_FREE(fglog, sizeof(*fglog));
2841 spin_lock(&filter->fo_llog_list_lock);
2842 list_for_each(cur, &filter->fo_llog_list) {
2843 nlog = list_entry(cur, struct filter_group_llog, list);
2844 LASSERT(nlog->group != group);
2846 list_add(&fglog->list, &filter->fo_llog_list);
2847 spin_unlock(&filter->fo_llog_list_lock);
2849 rc = obd_llog_cat_initialize(obd, fglog->llogs, 1, name);
2851 OBD_FREE(fglog->llogs, sizeof(*(fglog->llogs)));
2852 OBD_FREE(fglog, sizeof(*fglog));
2858 fglog->exp = export;
2859 ctxt = llog_get_context(fglog->llogs, LLOG_UNLINK_REPL_CTXT);
2860 LASSERT(ctxt != NULL);
2862 llog_receptor_accept(ctxt, export->exp_imp_reverse);
2864 CDEBUG(D_OTHER, "%s: new llog 0x%p for group %u\n",
2865 obd->obd_name, fglog->llogs, group);
2867 RETURN(fglog->llogs);
2870 int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
2871 int len, void *karg, void *uarg)
2873 struct obd_device *obd = exp->exp_obd;
2874 struct obd_ioctl_data *data = karg;
2878 case OBD_IOC_ABORT_RECOVERY:
2879 target_stop_recovery_thread(obd);
2882 case OBD_IOC_SET_READONLY: {
2884 struct super_block *sb = obd->u.filter.fo_sb;
2885 struct inode *inode = sb->s_root->d_inode;
2886 BDEVNAME_DECLARE_STORAGE(tmp);
2887 CERROR("setting device %s read-only\n",
2888 ll_bdevname(sb, tmp));
2890 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
2892 (void)fsfilt_commit(obd, sb, inode, handle, 1);
2894 dev_set_rdonly(ll_sbdev(obd->u.filter.fo_sb), 2);
2898 case OBD_IOC_CATLOGLIST: {
2899 rc = llog_catalog_list(obd, 1, data);
2902 case OBD_IOC_LLOG_CANCEL:
2903 case OBD_IOC_LLOG_REMOVE:
2904 case OBD_IOC_LLOG_INFO:
2905 case OBD_IOC_LLOG_PRINT: {
2906 /* FIXME to be finished */
2907 RETURN(-EOPNOTSUPP);
2909 struct llog_ctxt *ctxt = NULL;
2911 push_ctxt(&saved, &ctxt->loc_ctxt, NULL);
2912 rc = llog_ioctl(ctxt, cmd, data);
2913 pop_ctxt(&saved, &ctxt->loc_ctxt, NULL);
2926 static struct llog_operations filter_unlink_repl_logops;
2927 static struct llog_operations filter_size_orig_logops = {
2928 lop_setup: llog_obd_origin_setup,
2929 lop_cleanup: llog_catalog_cleanup,
2930 lop_add: llog_catalog_add,
2933 static int filter_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
2934 struct obd_device *tgt, int count,
2935 struct llog_catid *catid)
2937 struct llog_ctxt *ctxt;
2941 filter_unlink_repl_logops = llog_client_ops;
2942 filter_unlink_repl_logops.lop_cancel = llog_obd_repl_cancel;
2943 filter_unlink_repl_logops.lop_connect = llog_repl_connect;
2944 filter_unlink_repl_logops.lop_sync = llog_obd_repl_sync;
2946 rc = obd_llog_setup(obd, llogs, LLOG_UNLINK_REPL_CTXT, tgt, 0, NULL,
2947 &filter_unlink_repl_logops);
2950 /* FIXME - assign unlink_cb for filter's recovery */
2951 ctxt = llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT);
2952 ctxt->llog_proc_cb = filter_recov_log_unlink_cb;
2954 /* FIXME - count should be 1 to setup size log */
2955 rc = obd_llog_setup(obd, llogs, LLOG_SIZE_ORIG_CTXT, tgt, 0,
2956 &catid->lci_logid, &filter_size_orig_logops);
2960 static int filter_llog_finish(struct obd_device *obd,
2961 struct obd_llogs *llogs, int count)
2966 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT));
2970 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_SIZE_ORIG_CTXT));
2974 static int filter_llog_connect(struct obd_export *exp,
2975 struct llogd_conn_body *body)
2977 struct obd_device *obd = exp->exp_obd;
2978 struct llog_ctxt *ctxt;
2979 struct obd_llogs *llog;
2983 CDEBUG(D_OTHER, "handle connect for %s: %u/%u/%u\n", obd->obd_name,
2984 (unsigned) body->lgdc_logid.lgl_ogr,
2985 (unsigned) body->lgdc_logid.lgl_oid,
2986 (unsigned) body->lgdc_logid.lgl_ogen);
2987 llog = filter_grab_llog_for_group(obd, body->lgdc_logid.lgl_ogr, exp);
2988 LASSERT(llog != NULL);
2989 ctxt = llog_get_context(llog, body->lgdc_ctxt_idx);
2990 rc = llog_connect(ctxt, 1, &body->lgdc_logid,
2991 &body->lgdc_gen, NULL);
2993 CERROR("failed to connect\n");
2998 static struct dentry *filter_lvfs_id2dentry(__u64 id, __u32 gen,
2999 __u64 gr, void *data)
3001 return filter_id2dentry(data, NULL, gr, id);
3004 static struct lvfs_callback_ops filter_lvfs_ops = {
3005 l_id2dentry: filter_lvfs_id2dentry,
3008 static struct obd_ops filter_obd_ops = {
3009 .o_owner = THIS_MODULE,
3010 .o_attach = filter_attach,
3011 .o_detach = filter_detach,
3012 .o_get_info = filter_get_info,
3013 .o_setup = filter_setup,
3014 .o_precleanup = filter_precleanup,
3015 .o_cleanup = filter_cleanup,
3016 .o_connect = filter_connect,
3017 .o_connect_post = filter_connect_post,
3018 .o_disconnect = filter_disconnect,
3019 .o_statfs = filter_statfs,
3020 .o_getattr = filter_getattr,
3021 .o_unpackmd = filter_unpackmd,
3022 .o_create = filter_create,
3023 .o_setattr = filter_setattr,
3024 .o_destroy = filter_destroy,
3025 .o_brw = filter_brw,
3026 .o_punch = filter_truncate,
3027 .o_sync = filter_sync,
3028 .o_preprw = filter_preprw,
3029 .o_commitrw = filter_commitrw,
3030 .o_do_cow = filter_do_cow,
3031 .o_write_extents = filter_write_extents,
3032 .o_destroy_export = filter_destroy_export,
3033 .o_llog_init = filter_llog_init,
3034 .o_llog_finish = filter_llog_finish,
3035 .o_llog_connect = filter_llog_connect,
3036 .o_iocontrol = filter_iocontrol,
3039 static struct obd_ops filter_sanobd_ops = {
3040 .o_owner = THIS_MODULE,
3041 .o_attach = filter_attach,
3042 .o_detach = filter_detach,
3043 .o_get_info = filter_get_info,
3044 .o_setup = filter_san_setup,
3045 .o_precleanup = filter_precleanup,
3046 .o_cleanup = filter_cleanup,
3047 .o_connect = filter_connect,
3048 .o_connect_post = filter_connect_post,
3049 .o_disconnect = filter_disconnect,
3050 .o_statfs = filter_statfs,
3051 .o_getattr = filter_getattr,
3052 .o_unpackmd = filter_unpackmd,
3053 .o_create = filter_create,
3054 .o_setattr = filter_setattr,
3055 .o_destroy = filter_destroy,
3056 .o_brw = filter_brw,
3057 .o_punch = filter_truncate,
3058 .o_sync = filter_sync,
3059 .o_preprw = filter_preprw,
3060 .o_commitrw = filter_commitrw,
3061 .o_do_cow = filter_do_cow,
3062 .o_write_extents = filter_write_extents,
3063 .o_san_preprw = filter_san_preprw,
3064 .o_destroy_export = filter_destroy_export,
3065 .o_llog_init = filter_llog_init,
3066 .o_llog_finish = filter_llog_finish,
3067 .o_llog_connect = filter_llog_connect,
3068 .o_iocontrol = filter_iocontrol,
3071 static int __init obdfilter_init(void)
3073 struct lprocfs_static_vars lvars;
3076 printk(KERN_INFO "Lustre: Filtering OBD driver; info@clusterfs.com\n");
3078 lprocfs_init_vars(filter, &lvars);
3080 rc = class_register_type(&filter_obd_ops, NULL, lvars.module_vars,
3081 OBD_FILTER_DEVICENAME);
3085 rc = class_register_type(&filter_sanobd_ops, NULL, lvars.module_vars,
3086 OBD_FILTER_SAN_DEVICENAME);
3088 class_unregister_type(OBD_FILTER_DEVICENAME);
3092 static void __exit obdfilter_exit(void)
3094 class_unregister_type(OBD_FILTER_SAN_DEVICENAME);
3095 class_unregister_type(OBD_FILTER_DEVICENAME);
3098 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
3099 MODULE_DESCRIPTION("Lustre Filtering OBD driver");
3100 MODULE_LICENSE("GPL");
3102 module_init(obdfilter_init);
3103 module_exit(obdfilter_exit);