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/obd_ost.h>
52 #include <linux/lustre_dlm.h>
53 #include <linux/lustre_fsfilt.h>
54 #include <linux/lprocfs_status.h>
55 #include <linux/lustre_log.h>
56 #include <linux/lustre_commit_confd.h>
57 #include <portals/list.h>
59 #include <linux/lustre_smfs.h>
60 #include "filter_internal.h"
62 /* Group 0 is no longer a legal group, to catch uninitialized IDs */
63 #define FILTER_MIN_GROUPS 3
65 static struct lvfs_callback_ops filter_lvfs_ops;
67 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
68 struct lov_stripe_md *ea, struct obd_trans_info *);
69 static struct obd_llogs *filter_grab_llog_for_group(struct obd_device *,
70 int, struct obd_export *);
72 static void filter_commit_cb(struct obd_device *obd, __u64 transno,
73 void *cb_data, int error)
75 obd_transno_commit_cb(obd, transno, error);
78 /* Assumes caller has already pushed us into the kernel context. */
79 int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti,
82 struct filter_obd *filter = &exp->exp_obd->u.filter;
83 struct filter_export_data *fed = &exp->exp_filter_data;
84 struct filter_client_data *fcd = fed->fed_fcd;
87 int err, log_pri = D_HA;
89 /* Propagate error code. */
93 if (!exp->exp_obd->obd_replayable || oti == NULL)
96 /* we don't allocate new transnos for replayed requests */
97 if (oti->oti_transno == 0) {
98 spin_lock(&filter->fo_translock);
99 last_rcvd = le64_to_cpu(filter->fo_fsd->fsd_last_transno) + 1;
100 filter->fo_fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
101 spin_unlock(&filter->fo_translock);
102 oti->oti_transno = last_rcvd;
104 spin_lock(&filter->fo_translock);
105 last_rcvd = oti->oti_transno;
106 if (last_rcvd > le64_to_cpu(filter->fo_fsd->fsd_last_transno))
107 filter->fo_fsd->fsd_last_transno =
108 cpu_to_le64(last_rcvd);
109 spin_unlock(&filter->fo_translock);
111 fcd->fcd_last_rcvd = cpu_to_le64(last_rcvd);
113 /* could get xid from oti, if it's ever needed */
114 fcd->fcd_last_xid = 0;
116 off = fed->fed_lr_off;
118 fsfilt_add_journal_cb(exp->exp_obd, filter->fo_sb, last_rcvd,
119 oti->oti_handle, filter_commit_cb, NULL);
121 err = fsfilt_write_record(exp->exp_obd, filter->fo_rcvd_filp, fcd,
122 sizeof(*fcd), &off, 0);
129 CDEBUG(log_pri, "wrote trans "LPU64" for client %s at #%d: err = %d\n",
130 last_rcvd, fcd->fcd_uuid, fed->fed_lr_idx, err);
135 void f_dput(struct dentry *dentry)
137 /* Can't go inside filter_ddelete because it can block */
138 CDEBUG(D_INODE, "putting %s: %p, count = %d\n",
139 dentry->d_name.name, dentry, atomic_read(&dentry->d_count) - 1);
140 LASSERT(atomic_read(&dentry->d_count) > 0);
145 /* Add client data to the FILTER. We use a bitmap to locate a free space
146 * in the last_rcvd file if cl_idx is -1 (i.e. a new client).
147 * Otherwise, we have just read the data from the last_rcvd file and
148 * we know its offset. */
149 static int filter_client_add(struct obd_device *obd, struct filter_obd *filter,
150 struct filter_export_data *fed, int cl_idx)
152 unsigned long *bitmap = filter->fo_last_rcvd_slots;
153 int new_client = (cl_idx == -1);
156 LASSERT(bitmap != NULL);
158 /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
159 if (!strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid))
162 /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
163 * there's no need for extra complication here
166 cl_idx = find_first_zero_bit(bitmap, FILTER_LR_MAX_CLIENTS);
168 if (cl_idx >= FILTER_LR_MAX_CLIENTS) {
169 CERROR("no client slots - fix FILTER_LR_MAX_CLIENTS\n");
172 if (test_and_set_bit(cl_idx, bitmap)) {
173 CERROR("FILTER client %d: found bit is set in bitmap\n",
175 cl_idx = find_next_zero_bit(bitmap,
176 FILTER_LR_MAX_CLIENTS,
181 if (test_and_set_bit(cl_idx, bitmap)) {
182 CERROR("FILTER client %d: bit already set in bitmap!\n",
188 fed->fed_lr_idx = cl_idx;
189 fed->fed_lr_off = le32_to_cpu(filter->fo_fsd->fsd_client_start) +
190 cl_idx * le16_to_cpu(filter->fo_fsd->fsd_client_size);
192 CDEBUG(D_INFO, "client at index %d (%llu) with UUID '%s' added\n",
193 fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
196 struct lvfs_run_ctxt saved;
197 loff_t off = fed->fed_lr_off;
201 CDEBUG(D_INFO, "writing client fcd at idx %u (%llu) (len %u)\n",
202 fed->fed_lr_idx,off,(unsigned int)sizeof(*fed->fed_fcd));
204 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
205 /* Transaction needed to fix bug 1403 */
206 handle = fsfilt_start(obd,
207 filter->fo_rcvd_filp->f_dentry->d_inode,
208 FSFILT_OP_SETATTR, NULL);
209 if (IS_ERR(handle)) {
210 err = PTR_ERR(handle);
211 CERROR("unable to start transaction: rc %d\n", err);
213 err = fsfilt_write_record(obd, filter->fo_rcvd_filp,
215 sizeof(*fed->fed_fcd),
217 fsfilt_commit(obd, filter->fo_sb,
218 filter->fo_rcvd_filp->f_dentry->d_inode,
221 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
224 CERROR("error writing %s client idx %u: rc %d\n",
225 LAST_RCVD, fed->fed_lr_idx, err);
232 static int filter_client_free(struct obd_export *exp, int flags)
234 struct filter_export_data *fed = &exp->exp_filter_data;
235 struct filter_obd *filter = &exp->exp_obd->u.filter;
236 struct obd_device *obd = exp->exp_obd;
237 struct filter_client_data zero_fcd;
238 struct lvfs_run_ctxt saved;
243 if (fed->fed_fcd == NULL)
246 if (flags & OBD_OPT_FAILOVER)
249 /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
250 if (strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid ) == 0)
253 LASSERT(filter->fo_last_rcvd_slots != NULL);
255 off = fed->fed_lr_off;
257 CDEBUG(D_INFO, "freeing client at idx %u (%lld) with UUID '%s'\n",
258 fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
260 /* Clear the bit _after_ zeroing out the client so we don't
261 race with filter_client_add and zero out new clients.*/
262 if (!test_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
263 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
268 memset(&zero_fcd, 0, sizeof zero_fcd);
269 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
270 rc = fsfilt_write_record(obd, filter->fo_rcvd_filp, &zero_fcd,
271 sizeof(zero_fcd), &off, 1);
273 /* update server's transno */
274 filter_update_server_data(obd, filter->fo_rcvd_filp,
276 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
278 CDEBUG(rc == 0 ? D_INFO : D_ERROR,
279 "zeroing disconnecting client %s at idx %u (%llu) in %s rc %d\n",
280 fed->fed_fcd->fcd_uuid, fed->fed_lr_idx, fed->fed_lr_off,
283 if (!test_and_clear_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
284 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
290 OBD_FREE(fed->fed_fcd, sizeof(*fed->fed_fcd));
295 static int filter_free_server_data(struct filter_obd *filter)
297 OBD_FREE(filter->fo_fsd, sizeof(*filter->fo_fsd));
298 filter->fo_fsd = NULL;
299 OBD_FREE(filter->fo_last_rcvd_slots,
300 FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
301 filter->fo_last_rcvd_slots = NULL;
305 /* assumes caller is already in kernel ctxt */
306 int filter_update_server_data(struct obd_device *obd, struct file *filp,
307 struct filter_server_data *fsd, int force_sync)
313 CDEBUG(D_INODE, "server uuid : %s\n", fsd->fsd_uuid);
314 CDEBUG(D_INODE, "server last_rcvd : "LPU64"\n",
315 le64_to_cpu(fsd->fsd_last_transno));
316 CDEBUG(D_INODE, "server last_mount: "LPU64"\n",
317 le64_to_cpu(fsd->fsd_mount_count));
319 rc = fsfilt_write_record(obd, filp, fsd, sizeof(*fsd), &off,force_sync);
321 CERROR("error writing filter_server_data: rc = %d\n", rc);
326 int filter_update_last_objid(struct obd_device *obd, obd_gr group,
329 struct filter_obd *filter = &obd->u.filter;
335 if (filter->fo_last_objid_files[group] == NULL) {
336 CERROR("Object group "LPU64" not fully setup; not updating "
337 "last_objid\n", group);
341 CDEBUG(D_INODE, "server last_objid for group "LPU64": "LPU64"\n",
342 group, filter->fo_last_objids[group]);
344 tmp = cpu_to_le64(filter->fo_last_objids[group]);
345 rc = fsfilt_write_record(obd, filter->fo_last_objid_files[group],
346 &tmp, sizeof(tmp), &off, force_sync);
348 CERROR("error writing group "LPU64" last objid: rc = %d\n",
353 /* assumes caller has already in kernel ctxt */
354 static int filter_init_server_data(struct obd_device *obd, struct file * filp)
356 struct filter_obd *filter = &obd->u.filter;
357 struct filter_server_data *fsd;
358 struct filter_client_data *fcd = NULL;
359 struct inode *inode = filp->f_dentry->d_inode;
360 unsigned long last_rcvd_size = inode->i_size;
366 /* ensure padding in the struct is the correct size */
367 LASSERT (offsetof(struct filter_server_data, fsd_padding) +
368 sizeof(fsd->fsd_padding) == FILTER_LR_SERVER_SIZE);
369 LASSERT (offsetof(struct filter_client_data, fcd_padding) +
370 sizeof(fcd->fcd_padding) == FILTER_LR_CLIENT_SIZE);
372 OBD_ALLOC(fsd, sizeof(*fsd));
375 filter->fo_fsd = fsd;
377 OBD_ALLOC(filter->fo_last_rcvd_slots,
378 FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
379 if (filter->fo_last_rcvd_slots == NULL) {
380 OBD_FREE(fsd, sizeof(*fsd));
384 if (last_rcvd_size == 0) {
385 CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD);
387 memcpy(fsd->fsd_uuid, obd->obd_uuid.uuid,sizeof(fsd->fsd_uuid));
388 fsd->fsd_last_transno = 0;
389 mount_count = fsd->fsd_mount_count = 0;
390 fsd->fsd_server_size = cpu_to_le32(FILTER_LR_SERVER_SIZE);
391 fsd->fsd_client_start = cpu_to_le32(FILTER_LR_CLIENT_START);
392 fsd->fsd_client_size = cpu_to_le16(FILTER_LR_CLIENT_SIZE);
393 fsd->fsd_subdir_count = cpu_to_le16(FILTER_SUBDIR_COUNT);
394 filter->fo_subdir_count = FILTER_SUBDIR_COUNT;
396 rc = fsfilt_read_record(obd, filp, fsd, sizeof(*fsd), &off);
398 CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
402 if (strcmp(fsd->fsd_uuid, obd->obd_uuid.uuid) != 0) {
403 CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
404 obd->obd_uuid.uuid, fsd->fsd_uuid);
405 GOTO(err_fsd, rc = -EINVAL);
407 mount_count = le64_to_cpu(fsd->fsd_mount_count);
408 filter->fo_subdir_count = le16_to_cpu(fsd->fsd_subdir_count);
411 if (fsd->fsd_feature_incompat & ~cpu_to_le32(FILTER_INCOMPAT_SUPP)) {
412 CERROR("unsupported feature %x\n",
413 le32_to_cpu(fsd->fsd_feature_incompat) &
414 ~FILTER_INCOMPAT_SUPP);
415 GOTO(err_fsd, rc = -EINVAL);
417 if (fsd->fsd_feature_rocompat & ~cpu_to_le32(FILTER_ROCOMPAT_SUPP)) {
418 CERROR("read-only feature %x\n",
419 le32_to_cpu(fsd->fsd_feature_rocompat) &
420 ~FILTER_ROCOMPAT_SUPP);
421 /* Do something like remount filesystem read-only */
422 GOTO(err_fsd, rc = -EINVAL);
425 CDEBUG(D_INODE, "%s: server last_rcvd : "LPU64"\n",
426 obd->obd_name, le64_to_cpu(fsd->fsd_last_transno));
427 CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
428 obd->obd_name, mount_count + 1);
429 CDEBUG(D_INODE, "%s: server data size: %u\n",
430 obd->obd_name, le32_to_cpu(fsd->fsd_server_size));
431 CDEBUG(D_INODE, "%s: per-client data start: %u\n",
432 obd->obd_name, le32_to_cpu(fsd->fsd_client_start));
433 CDEBUG(D_INODE, "%s: per-client data size: %u\n",
434 obd->obd_name, le32_to_cpu(fsd->fsd_client_size));
435 CDEBUG(D_INODE, "%s: server subdir_count: %u\n",
436 obd->obd_name, le16_to_cpu(fsd->fsd_subdir_count));
437 CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
438 last_rcvd_size <= le32_to_cpu(fsd->fsd_client_start) ? 0 :
439 (last_rcvd_size - le32_to_cpu(fsd->fsd_client_start)) /
440 le16_to_cpu(fsd->fsd_client_size));
442 if (!obd->obd_replayable) {
443 CWARN("%s: recovery support OFF\n", obd->obd_name);
447 for (cl_idx = 0, off = le32_to_cpu(fsd->fsd_client_start);
448 off < last_rcvd_size; cl_idx++) {
450 struct obd_export *exp;
451 struct filter_export_data *fed;
454 OBD_ALLOC(fcd, sizeof(*fcd));
456 GOTO(err_client, rc = -ENOMEM);
459 /* Don't assume off is incremented properly by
460 * fsfilt_read_record(), in case sizeof(*fcd)
461 * isn't the same as fsd->fsd_client_size. */
462 off = le32_to_cpu(fsd->fsd_client_start) +
463 cl_idx * le16_to_cpu(fsd->fsd_client_size);
464 rc = fsfilt_read_record(obd, filp, fcd, sizeof(*fcd), &off);
466 CERROR("error reading FILT %s idx %d off %llu: rc %d\n",
467 LAST_RCVD, cl_idx, off, rc);
468 break; /* read error shouldn't cause startup to fail */
471 if (fcd->fcd_uuid[0] == '\0') {
472 CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
477 last_rcvd = le64_to_cpu(fcd->fcd_last_rcvd);
479 /* These exports are cleaned up by filter_disconnect(), so they
480 * need to be set up like real exports as filter_connect() does.
482 exp = class_new_export(obd);
483 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
484 " srv lr: "LPU64"\n", fcd->fcd_uuid, cl_idx,
485 last_rcvd, le64_to_cpu(fsd->fsd_last_transno));
487 GOTO(err_client, rc = -ENOMEM);
489 memcpy(&exp->exp_client_uuid.uuid, fcd->fcd_uuid,
490 sizeof exp->exp_client_uuid.uuid);
491 fed = &exp->exp_filter_data;
493 filter_client_add(obd, filter, fed, cl_idx);
494 /* create helper if export init gets more complex */
495 spin_lock_init(&fed->fed_lock);
498 exp->exp_replay_needed = 1;
499 obd->obd_recoverable_clients++;
500 obd->obd_max_recoverable_clients++;
501 class_export_put(exp);
503 CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
506 if (last_rcvd > le64_to_cpu(fsd->fsd_last_transno))
507 fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
511 obd->obd_last_committed = le64_to_cpu(fsd->fsd_last_transno);
513 if (obd->obd_recoverable_clients) {
514 CWARN("RECOVERY: %d recoverable clients, last_rcvd "
515 LPU64"\n", obd->obd_recoverable_clients,
516 le64_to_cpu(fsd->fsd_last_transno));
517 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
518 target_start_recovery_thread(obd, ost_handle);
522 OBD_FREE(fcd, sizeof(*fcd));
525 filter->fo_mount_count = mount_count + 1;
526 fsd->fsd_mount_count = cpu_to_le64(filter->fo_mount_count);
528 /* save it, so mount count and last_transno is current */
529 rc = filter_update_server_data(obd, filp, filter->fo_fsd, 1);
534 class_disconnect_exports(obd, 0);
536 filter_free_server_data(filter);
540 static int filter_cleanup_groups(struct obd_device *obd)
542 struct filter_obd *filter = &obd->u.filter;
543 struct dentry *dentry;
547 for (i = 0; i < filter->fo_group_count; i++) {
548 if (filter->fo_subdirs != NULL) {
549 for (k = 0; k < filter->fo_subdir_count; k++) {
550 dentry = filter->fo_subdirs[i].dentry[k];
554 filter->fo_subdirs[i].dentry[k] = NULL;
557 if (filter->fo_last_objid_files[i] != NULL) {
558 filp_close(filter->fo_last_objid_files[i], 0);
559 filter->fo_last_objid_files[i] = NULL;
561 if (filter->fo_groups[i] != NULL) {
562 dput(filter->fo_groups[i]);
563 filter->fo_groups[i] = NULL;
566 if (filter->fo_subdirs != NULL)
567 OBD_FREE(filter->fo_subdirs,
568 filter->fo_group_count * sizeof(*filter->fo_subdirs));
569 if (filter->fo_groups != NULL)
570 OBD_FREE(filter->fo_groups,
571 filter->fo_group_count * sizeof(*filter->fo_groups));
572 if (filter->fo_last_objids != NULL)
573 OBD_FREE(filter->fo_last_objids,
574 filter->fo_group_count * sizeof(__u64));
575 if (filter->fo_last_objid_files != NULL)
576 OBD_FREE(filter->fo_last_objid_files,
577 filter->fo_group_count * sizeof(struct file *));
578 f_dput(filter->fo_dentry_O);
582 static int filter_update_last_group(struct obd_device *obd, int group)
584 struct filter_obd *filter = &obd->u.filter;
585 struct file *filp = NULL;
586 int last_group = 0, rc;
590 if (group <= filter->fo_committed_group)
593 filp = filp_open("LAST_GROUP", O_RDWR, 0700);
597 CERROR("cannot open LAST_GROUP: rc = %d\n", rc);
601 rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
603 CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n",rc);
606 LASSERT(off == 0 || last_group >= FILTER_MIN_GROUPS);
607 CDEBUG(D_INODE, "%s: previous %d, new %d\n",
608 obd->obd_name, last_group, group);
612 /* must be sync: bXXXX */
613 rc = fsfilt_write_record(obd, filp, &last_group, sizeof(__u32), &off, 1);
615 CDEBUG(D_INODE, "error updating LAST_GROUP: rc %d\n", rc);
619 filter->fo_committed_group = group;
626 static int filter_read_group_internal(struct obd_device *obd, int group,
629 struct filter_obd *filter = &obd->u.filter;
630 __u64 *new_objids = NULL;
631 struct filter_subdirs *new_subdirs = NULL, *tmp_subdirs = NULL;
632 struct dentry **new_groups = NULL;
633 struct file **new_files = NULL;
634 struct dentry *dentry;
636 int old_count = filter->fo_group_count, rc, stage = 0, i;
641 snprintf(name, 24, "%d", group);
645 dentry = ll_lookup_one_len(name, filter->fo_dentry_O,
647 if (IS_ERR(dentry)) {
648 CERROR("Cannot lookup expected object group %d: %ld\n",
649 group, PTR_ERR(dentry));
650 RETURN(PTR_ERR(dentry));
653 dentry = simple_mkdir(filter->fo_dentry_O, name, 0700, 1);
654 if (IS_ERR(dentry)) {
655 CERROR("cannot lookup/create O/%s: rc = %ld\n", name,
657 RETURN(PTR_ERR(dentry));
662 snprintf(name, 24, "O/%d/LAST_ID", group);
664 filp = filp_open(name, O_CREAT | O_RDWR, 0700);
666 CERROR("cannot create %s: rc = %ld\n", name, PTR_ERR(filp));
667 GOTO(cleanup, rc = PTR_ERR(filp));
671 rc = fsfilt_read_record(obd, filp, &last_objid, sizeof(__u64), &off);
673 CDEBUG(D_INODE, "error reading %s: rc %d\n", name, rc);
677 if (filter->fo_subdir_count) {
678 OBD_ALLOC(tmp_subdirs, sizeof(*tmp_subdirs));
679 if (tmp_subdirs == NULL)
680 GOTO(cleanup, rc = -ENOMEM);
683 for (i = 0; i < filter->fo_subdir_count; i++) {
685 snprintf(dir, sizeof(dir), "d%u", i);
687 tmp_subdirs->dentry[i] =
688 simple_mkdir(dentry, dir, 0700, 1);
689 if (IS_ERR(tmp_subdirs->dentry[i])) {
690 rc = PTR_ERR(tmp_subdirs->dentry[i]);
691 CERROR("can't lookup/create O/%d/%s: rc = %d\n",
695 CDEBUG(D_INODE, "got/created O/%d/%s: %p\n", group, dir,
696 tmp_subdirs->dentry[i]);
700 /* 'group' is an index; we need an array of length 'group + 1' */
701 if (group + 1 > old_count) {
703 OBD_ALLOC(new_objids, len * sizeof(*new_objids));
704 OBD_ALLOC(new_subdirs, len * sizeof(*new_subdirs));
705 OBD_ALLOC(new_groups, len * sizeof(*new_groups));
706 OBD_ALLOC(new_files, len * sizeof(*new_files));
708 if (new_objids == NULL || new_subdirs == NULL ||
709 new_groups == NULL || new_files == NULL)
710 GOTO(cleanup, rc = -ENOMEM);
712 memcpy(new_objids, filter->fo_last_objids,
713 old_count * sizeof(*new_objids));
714 memcpy(new_subdirs, filter->fo_subdirs,
715 old_count * sizeof(*new_subdirs));
716 memcpy(new_groups, filter->fo_groups,
717 old_count * sizeof(*new_groups));
718 memcpy(new_files, filter->fo_last_objid_files,
719 old_count * sizeof(*new_files));
722 OBD_FREE(filter->fo_last_objids,
723 old_count * sizeof(*new_objids));
724 OBD_FREE(filter->fo_subdirs,
725 old_count * sizeof(*new_subdirs));
726 OBD_FREE(filter->fo_groups,
727 old_count * sizeof(*new_groups));
728 OBD_FREE(filter->fo_last_objid_files,
729 old_count * sizeof(*new_files));
731 filter->fo_last_objids = new_objids;
732 filter->fo_subdirs = new_subdirs;
733 filter->fo_groups = new_groups;
734 filter->fo_last_objid_files = new_files;
735 filter->fo_group_count = len;
738 filter->fo_groups[group] = dentry;
739 filter->fo_last_objid_files[group] = filp;
740 if (filter->fo_subdir_count) {
741 filter->fo_subdirs[group] = *tmp_subdirs;
742 OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
745 filter_update_last_group(obd, group);
747 if (filp->f_dentry->d_inode->i_size == 0) {
748 filter->fo_last_objids[group] = FILTER_INIT_OBJID;
752 filter->fo_last_objids[group] = le64_to_cpu(last_objid);
753 CDEBUG(D_INODE, "%s: server last_objid group %d: "LPU64"\n",
754 obd->obd_name, group, last_objid);
759 if (new_objids != NULL)
760 OBD_FREE(new_objids, group * sizeof(*new_objids));
761 if (new_subdirs != NULL)
762 OBD_FREE(new_subdirs, group * sizeof(*new_subdirs));
763 if (new_groups != NULL)
764 OBD_FREE(new_groups, group * sizeof(*new_groups));
765 if (new_files != NULL)
766 OBD_FREE(new_files, group * sizeof(*new_files));
768 if (filter->fo_subdir_count) {
769 for (i = 0; i < filter->fo_subdir_count; i++) {
770 if (tmp_subdirs->dentry[i] != NULL)
771 dput(tmp_subdirs->dentry[i]);
773 OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
783 static int filter_read_groups(struct obd_device *obd, int last_group,
786 struct filter_obd *filter = &obd->u.filter;
787 int old_count, group, rc = 0;
789 down(&filter->fo_init_lock);
790 old_count = filter->fo_group_count;
791 for (group = old_count; group <= last_group; group++) {
793 continue; /* no group zero */
795 rc = filter_read_group_internal(obd, group, create);
799 up(&filter->fo_init_lock);
803 static int filter_prep_groups(struct obd_device *obd)
805 struct filter_obd *filter = &obd->u.filter;
806 struct dentry *dentry, *O_dentry;
807 int rc = 0, cleanup_phase = 0;
808 struct file *filp = NULL;
813 O_dentry = simple_mkdir(current->fs->pwd, "O", 0700, 1);
814 CDEBUG(D_INODE, "got/created O: %p\n", O_dentry);
815 if (IS_ERR(O_dentry)) {
816 rc = PTR_ERR(O_dentry);
817 CERROR("cannot open/create O: rc = %d\n", rc);
820 filter->fo_dentry_O = O_dentry;
821 cleanup_phase = 1; /* O_dentry */
823 /* Lookup "R" to tell if we're on an old OST FS and need to convert
824 * from O/R/<dir>/<objid> to O/0/<dir>/<objid>. This can be removed
825 * some time post 1.0 when all old-style OSTs have converted along
826 * with the init_objid hack. */
827 dentry = ll_lookup_one_len("R", O_dentry, 1);
829 GOTO(cleanup, rc = PTR_ERR(dentry));
830 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
831 struct dentry *O0_dentry = lookup_one_len("0", O_dentry, 1);
834 CWARN("converting OST to new object layout\n");
835 if (IS_ERR(O0_dentry)) {
836 rc = PTR_ERR(O0_dentry);
837 CERROR("error looking up O/0: rc %d\n", rc);
841 if (O0_dentry->d_inode) {
842 CERROR("Both O/R and O/0 exist. Fix manually.\n");
843 GOTO(cleanup_O0, rc = -EEXIST);
846 down(&O_dentry->d_inode->i_sem);
847 rc = vfs_rename(O_dentry->d_inode, dentry,
848 O_dentry->d_inode, O0_dentry);
849 up(&O_dentry->d_inode->i_sem);
852 CERROR("error renaming O/R to O/0: rc %d\n", rc);
853 GOTO(cleanup_O0, rc);
855 filter->fo_fsd->fsd_feature_incompat |=
856 cpu_to_le32(FILTER_INCOMPAT_GROUPS);
857 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
859 GOTO(cleanup_O0, rc);
871 cleanup_phase = 2; /* groups */
873 /* we have to initialize all groups before first connections from
874 * clients because they may send create/destroy for any group -bzzz */
875 filp = filp_open("LAST_GROUP", O_CREAT | O_RDWR, 0700);
877 CERROR("cannot create LAST_GROUP: rc = %ld\n", PTR_ERR(filp));
878 GOTO(cleanup, rc = PTR_ERR(filp));
880 cleanup_phase = 3; /* filp */
882 rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
884 CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n", rc);
888 last_group = FILTER_MIN_GROUPS;
890 LASSERT(last_group >= FILTER_MIN_GROUPS);
893 CWARN("%s: initialize groups [%d,%d]\n", obd->obd_name,
894 FILTER_MIN_GROUPS, last_group);
895 filter->fo_committed_group = last_group;
896 rc = filter_read_groups(obd, last_group, 1);
904 switch (cleanup_phase) {
908 filter_cleanup_groups(obd);
910 f_dput(filter->fo_dentry_O);
911 filter->fo_dentry_O = NULL;
918 /* setup the object store with correct subdirectories */
919 static int filter_prep(struct obd_device *obd)
921 struct lvfs_run_ctxt saved;
922 struct filter_obd *filter = &obd->u.filter;
928 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
929 file = filp_open(LAST_RCVD, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
930 if (!file || IS_ERR(file)) {
932 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
937 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
938 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
939 file->f_dentry->d_inode->i_mode);
940 GOTO(err_filp, rc = -ENOENT);
943 /* steal operations */
944 inode = file->f_dentry->d_inode;
945 filter->fo_fop = file->f_op;
946 filter->fo_iop = inode->i_op;
947 filter->fo_aops = inode->i_mapping->a_ops;
949 rc = filter_init_server_data(obd, file);
951 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
954 filter->fo_rcvd_filp = file;
956 rc = filter_prep_groups(obd);
958 GOTO(err_server_data, rc);
961 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
966 //class_disconnect_exports(obd, 0);
967 filter_free_server_data(filter);
969 if (filp_close(file, 0))
970 CERROR("can't close %s after error\n", LAST_RCVD);
971 filter->fo_rcvd_filp = NULL;
975 /* cleanup the filter: write last used object id to status file */
976 static void filter_post(struct obd_device *obd)
978 struct lvfs_run_ctxt saved;
979 struct filter_obd *filter = &obd->u.filter;
982 /* XXX: filter_update_lastobjid used to call fsync_dev. It might be
983 * best to start a transaction with h_sync, because we removed this
986 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
987 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
990 CERROR("error writing server data: rc = %d\n", rc);
992 for (i = 1; i < filter->fo_group_count; i++) {
993 rc = filter_update_last_objid(obd, i,
994 (i == filter->fo_group_count - 1));
996 CERROR("error writing group %d lastobjid: rc = %d\n",
1000 rc = filp_close(filter->fo_rcvd_filp, 0);
1001 filter->fo_rcvd_filp = NULL;
1003 CERROR("error closing %s: rc = %d\n", LAST_RCVD, rc);
1005 filter_cleanup_groups(obd);
1006 filter_free_server_data(filter);
1007 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1010 static void filter_set_last_id(struct filter_obd *filter, int group, obd_id id)
1012 LASSERT(filter->fo_fsd != NULL);
1014 LASSERT(group < filter->fo_group_count);
1016 spin_lock(&filter->fo_objidlock);
1017 filter->fo_last_objids[group] = id;
1018 spin_unlock(&filter->fo_objidlock);
1021 __u64 filter_last_id(struct filter_obd *filter, int group)
1024 LASSERT(filter->fo_fsd != NULL);
1026 LASSERT(group < filter->fo_group_count);
1028 spin_lock(&filter->fo_objidlock);
1029 id = filter->fo_last_objids[group];
1030 spin_unlock(&filter->fo_objidlock);
1035 /* direct cut-n-paste of mds_blocking_ast() */
1036 static int filter_blocking_ast(struct ldlm_lock *lock,
1037 struct ldlm_lock_desc *desc,
1038 void *data, int flag)
1043 if (flag == LDLM_CB_CANCELING) {
1044 /* Don't need to do anything here. */
1048 /* XXX layering violation! -phil */
1049 l_lock(&lock->l_resource->lr_namespace->ns_lock);
1050 /* Get this: if filter_blocking_ast is racing with ldlm_intent_policy,
1051 * such that filter_blocking_ast is called just before l_i_p takes the
1052 * ns_lock, then by the time we get the lock, we might not be the
1053 * correct blocking function anymore. So check, and return early, if
1055 if (lock->l_blocking_ast != filter_blocking_ast) {
1056 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
1060 lock->l_flags |= LDLM_FL_CBPENDING;
1061 do_ast = (!lock->l_readers && !lock->l_writers);
1062 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
1065 struct lustre_handle lockh;
1068 LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
1069 ldlm_lock2handle(lock, &lockh);
1070 rc = ldlm_cli_cancel(&lockh);
1072 CERROR("ldlm_cli_cancel: %d\n", rc);
1074 LDLM_DEBUG(lock, "Lock still has references, will be "
1080 extern void *lock_dir(struct inode *dir, struct qstr *name);
1081 extern void unlock_dir(struct inode *dir, void *lock);
1083 static void * filter_lock_dentry(struct obd_device *obd,
1084 struct dentry *dparent, obd_id id)
1090 qstr.len = sprintf(name, LPU64, id);
1091 return lock_dir(dparent->d_inode, &qstr);
1093 down(&dparent->d_inode->i_sem);
1098 /* We never dget the object parent, so DON'T dput it either */
1099 static void filter_parent_unlock(struct dentry *dparent, void *lock)
1102 LASSERT(lock != NULL);
1103 unlock_dir(dparent->d_inode, lock);
1105 up(&dparent->d_inode->i_sem);
1109 /* We never dget the object parent, so DON'T dput it either */
1110 struct dentry *filter_parent(struct obd_device *obd, obd_gr group, obd_id objid)
1112 struct filter_obd *filter = &obd->u.filter;
1113 LASSERT(group < filter->fo_group_count);
1116 if (filter->fo_subdir_count == 0)
1117 return filter->fo_groups[group];
1119 return filter->fo_subdirs[group].dentry[objid & (filter->fo_subdir_count - 1)];
1122 /* We never dget the object parent, so DON'T dput it either */
1123 struct dentry *filter_parent_lock(struct obd_device *obd, obd_gr group,
1124 obd_id objid, void **lock)
1126 unsigned long now = jiffies;
1127 struct dentry *dparent = filter_parent(obd, group, objid);
1129 if (IS_ERR(dparent))
1133 LASSERT(dparent->d_inode);
1135 *lock = filter_lock_dentry(obd, dparent, objid);
1136 if (time_after(jiffies, now + 15 * HZ))
1137 CERROR("slow parent lock %lus\n", (jiffies - now) / HZ);
1141 /* How to get files, dentries, inodes from object id's.
1143 * If dir_dentry is passed, the caller has already locked the parent
1144 * appropriately for this operation (normally a write lock). If
1145 * dir_dentry is NULL, we do a read lock while we do the lookup to
1146 * avoid races with create/destroy and such changing the directory
1147 * internal to the filesystem code. */
1148 struct dentry *filter_fid2dentry(struct obd_device *obd,
1149 struct dentry *dir_dentry,
1150 obd_gr group, obd_id id)
1152 struct dentry *dparent = dir_dentry;
1153 struct dentry *dchild;
1160 CERROR("fatal: invalid object id 0\n");
1161 RETURN(ERR_PTR(-ESTALE));
1164 len = sprintf(name, LPU64, id);
1165 if (dir_dentry == NULL) {
1166 dparent = filter_parent_lock(obd, group, id, &lock);
1167 if (IS_ERR(dparent))
1170 CDEBUG(D_INODE, "looking up object O/%*s/%s\n",
1171 dparent->d_name.len, dparent->d_name.name, name);
1172 dchild = /*ll_*/lookup_one_len(name, dparent, len);
1173 if (dir_dentry == NULL)
1174 filter_parent_unlock(dparent, lock);
1175 if (IS_ERR(dchild)) {
1176 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
1180 CDEBUG(D_INODE, "got child objid %s: %p, count = %d\n",
1181 name, dchild, atomic_read(&dchild->d_count));
1183 LASSERT(atomic_read(&dchild->d_count) > 0);
1188 static int filter_prepare_destroy(struct obd_device *obd, obd_id objid,
1191 struct lustre_handle lockh;
1192 int flags = LDLM_AST_DISCARD_DATA, rc;
1193 struct ldlm_res_id res_id = { .name = { objid, 0, group, 0 } };
1194 ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
1197 /* Tell the clients that the object is gone now and that they should
1198 * throw away any cached pages. */
1199 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, res_id,
1200 LDLM_EXTENT, &policy, LCK_PW,
1201 &flags, filter_blocking_ast, ldlm_completion_ast,
1202 NULL, NULL, NULL, 0, NULL, &lockh);
1204 /* We only care about the side-effects, just drop the lock. */
1206 ldlm_lock_decref(&lockh, LCK_PW);
1211 /* Caller must hold LCK_PW on parent and push us into kernel context.
1212 * Caller is also required to ensure that dchild->d_inode exists. */
1213 static int filter_destroy_internal(struct obd_device *obd, obd_id objid,
1214 struct dentry *dparent,
1215 struct dentry *dchild)
1217 struct inode *inode = dchild->d_inode;
1221 if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) {
1222 CERROR("destroying objid %*s nlink = %lu, count = %d\n",
1223 dchild->d_name.len, dchild->d_name.name,
1224 (unsigned long)inode->i_nlink,
1225 atomic_read(&inode->i_count));
1228 rc = vfs_unlink(dparent->d_inode, dchild);
1231 CERROR("error unlinking objid %*s: rc %d\n",
1232 dchild->d_name.len, dchild->d_name.name, rc);
1237 static int filter_intent_policy(struct ldlm_namespace *ns,
1238 struct ldlm_lock **lockp, void *req_cookie,
1239 ldlm_mode_t mode, int flags, void *data)
1241 struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1242 struct ptlrpc_request *req = req_cookie;
1243 struct ldlm_lock *lock = *lockp, *l = NULL;
1244 struct ldlm_resource *res = lock->l_resource;
1245 ldlm_processing_policy policy;
1246 struct ost_lvb *res_lvb, *reply_lvb;
1247 struct list_head *tmp;
1249 int tmpflags = 0, rc, repsize[2] = {sizeof(struct ldlm_reply),
1250 sizeof(struct ost_lvb) };
1253 policy = ldlm_get_processing_policy(res);
1254 LASSERT(policy != NULL);
1255 LASSERT(req != NULL);
1257 rc = lustre_pack_reply(req, 2, repsize, NULL);
1259 RETURN(req->rq_status = rc);
1261 reply_lvb = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*reply_lvb));
1262 LASSERT(reply_lvb != NULL);
1264 //fixup_handle_for_resent_req(req, lock, &lockh);
1266 /* If we grant any lock at all, it will be a whole-file read lock.
1267 * Call the extent policy function to see if our request can be
1268 * granted, or is blocked. */
1269 lock->l_policy_data.l_extent.start = 0;
1270 lock->l_policy_data.l_extent.end = OBD_OBJECT_EOF;
1271 lock->l_req_mode = LCK_PR;
1273 l_lock(&res->lr_namespace->ns_lock);
1275 res->lr_tmp = &rpc_list;
1276 rc = policy(lock, &tmpflags, 0, &err);
1279 /* FIXME: we should change the policy function slightly, to not make
1280 * this list at all, since we just turn around and free it */
1281 while (!list_empty(&rpc_list)) {
1282 struct ldlm_ast_work *w =
1283 list_entry(rpc_list.next, struct ldlm_ast_work, w_list);
1284 list_del(&w->w_list);
1285 LDLM_LOCK_PUT(w->w_lock);
1286 OBD_FREE(w, sizeof(*w));
1289 if (rc == LDLM_ITER_CONTINUE) {
1290 /* The lock met with no resistance; we're finished. */
1291 l_unlock(&res->lr_namespace->ns_lock);
1292 RETURN(ELDLM_LOCK_REPLACED);
1295 /* Do not grant any lock, but instead send GL callbacks. The extent
1296 * policy nicely created a list of all PW locks for us. We will choose
1297 * the highest of those which are larger than the size in the LVB, if
1298 * any, and perform a glimpse callback. */
1299 down(&res->lr_lvb_sem);
1300 res_lvb = res->lr_lvb_data;
1301 LASSERT(res_lvb != NULL);
1302 reply_lvb->lvb_size = res_lvb->lvb_size;
1303 reply_lvb->lvb_blocks = res_lvb->lvb_blocks;
1304 up(&res->lr_lvb_sem);
1306 list_for_each(tmp, &res->lr_granted) {
1307 struct ldlm_lock *tmplock =
1308 list_entry(tmp, struct ldlm_lock, l_res_link);
1310 if (tmplock->l_granted_mode == LCK_PR)
1313 if (tmplock->l_policy_data.l_extent.end <= reply_lvb->lvb_size)
1317 l = LDLM_LOCK_GET(tmplock);
1321 if (l->l_policy_data.l_extent.start >
1322 tmplock->l_policy_data.l_extent.start)
1326 l = LDLM_LOCK_GET(tmplock);
1328 l_unlock(&res->lr_namespace->ns_lock);
1330 /* There were no PW locks beyond the size in the LVB; finished. */
1332 RETURN(ELDLM_LOCK_ABORTED);
1334 LASSERT(l->l_glimpse_ast != NULL);
1335 rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
1336 if (rc != 0 && res->lr_namespace->ns_lvbo &&
1337 res->lr_namespace->ns_lvbo->lvbo_update) {
1338 res->lr_namespace->ns_lvbo->lvbo_update(res, NULL, 0, 1);
1341 down(&res->lr_lvb_sem);
1342 reply_lvb->lvb_size = res_lvb->lvb_size;
1343 reply_lvb->lvb_blocks = res_lvb->lvb_blocks;
1344 up(&res->lr_lvb_sem);
1348 RETURN(ELDLM_LOCK_ABORTED);
1351 static int filter_post_fs_cleanup(struct obd_device *obd)
1355 rc = fsfilt_post_cleanup(obd);
1360 static int filter_group_set_fs_flags(struct obd_device *obd, int group)
1362 struct filter_obd *filter = &obd->u.filter;
1366 /* zero group is not longer valid. */
1369 for (i = 0; i < filter->fo_subdir_count; i++) {
1370 struct dentry *dentry;
1371 dentry = (filter->fo_subdirs + group)->dentry[i];
1372 rc = fsfilt_set_fs_flags(obd, dentry->d_inode,
1373 SM_DO_REC | SM_DO_COW);
1379 static int filter_post_fs_setup(struct obd_device *obd)
1381 struct filter_obd *filter = &obd->u.filter;
1383 struct llog_ctxt *ctxt = NULL;
1385 rc = fsfilt_post_setup(obd, filter->fo_dentry_O);
1389 for (j = 0; j < filter->fo_group_count; j++) {
1390 rc = filter_group_set_fs_flags(obd, j);
1395 fsfilt_get_reint_log_ctxt(obd, filter->fo_sb, &ctxt);
1397 ctxt->loc_obd = obd;
1398 ctxt->loc_idx = LLOG_REINT_ORIG_CTXT;
1399 obd->obd_llog_ctxt[LLOG_REINT_ORIG_CTXT] = ctxt;
1401 fsfilt_set_ost_flags(obd, filter->fo_sb);
1405 /* mount the file system (secretly) */
1406 int filter_common_setup(struct obd_device *obd, obd_count len,
1407 void *buf, char *option)
1409 struct lustre_cfg* lcfg = buf;
1410 struct filter_obd *filter = &obd->u.filter;
1411 struct vfsmount *mnt;
1412 char name[32] = "CATLIST";
1416 dev_clear_rdonly(2);
1418 if (!lcfg->lcfg_inlbuf1 || !lcfg->lcfg_inlbuf2)
1421 obd->obd_fsops = fsfilt_get_ops(lcfg->lcfg_inlbuf2);
1422 if (IS_ERR(obd->obd_fsops))
1423 RETURN(PTR_ERR(obd->obd_fsops));
1425 mnt = do_kern_mount(lcfg->lcfg_inlbuf2, MS_NOATIME | MS_NODIRATIME,
1426 lcfg->lcfg_inlbuf1, option);
1431 if (lcfg->lcfg_inllen3 > 0 && lcfg->lcfg_inlbuf3) {
1432 if (*lcfg->lcfg_inlbuf3 == 'f') {
1433 obd->obd_replayable = 1;
1434 obd_sync_filter = 1;
1435 CWARN("%s: recovery enabled\n", obd->obd_name);
1437 if (*lcfg->lcfg_inlbuf3 != 'n') {
1438 CERROR("unrecognised flag '%c'\n",
1439 *lcfg->lcfg_inlbuf3);
1441 // XXX Robert? Why do we get errors here
1442 // GOTO(err_mntput, rc = -EINVAL);
1446 filter->fo_vfsmnt = mnt;
1447 filter->fo_sb = mnt->mnt_sb;
1448 filter->fo_fstype = mnt->mnt_sb->s_type->name;
1449 CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
1451 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1452 obd->obd_lvfs_ctxt.pwdmnt = mnt;
1453 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
1454 obd->obd_lvfs_ctxt.fs = get_ds();
1455 obd->obd_lvfs_ctxt.cb_ops = filter_lvfs_ops;
1457 rc = fsfilt_setup(obd, mnt->mnt_sb);
1459 GOTO(err_mntput, rc);
1461 sema_init(&filter->fo_init_lock, 1);
1462 filter->fo_committed_group = 0;
1463 rc = filter_prep(obd);
1465 GOTO(err_mntput, rc);
1467 filter->fo_destroy_in_progress = 0;
1468 sema_init(&filter->fo_create_lock, 1);
1470 spin_lock_init(&filter->fo_translock);
1471 spin_lock_init(&filter->fo_objidlock);
1472 INIT_LIST_HEAD(&filter->fo_export_list);
1473 sema_init(&filter->fo_alloc_lock, 1);
1474 spin_lock_init(&filter->fo_r_pages.oh_lock);
1475 spin_lock_init(&filter->fo_w_pages.oh_lock);
1476 spin_lock_init(&filter->fo_r_discont_pages.oh_lock);
1477 spin_lock_init(&filter->fo_w_discont_pages.oh_lock);
1478 spin_lock_init(&filter->fo_r_discont_blocks.oh_lock);
1479 spin_lock_init(&filter->fo_w_discont_blocks.oh_lock);
1480 filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE;
1482 INIT_LIST_HEAD(&filter->fo_llog_list);
1483 spin_lock_init(&filter->fo_llog_list_lock);
1485 obd->obd_namespace = ldlm_namespace_new("filter-tgt",
1486 LDLM_NAMESPACE_SERVER);
1487 if (obd->obd_namespace == NULL)
1488 GOTO(err_post, rc = -ENOMEM);
1489 obd->obd_namespace->ns_lvbp = obd;
1490 obd->obd_namespace->ns_lvbo = &filter_lvbo;
1491 ldlm_register_intent(obd->obd_namespace, filter_intent_policy);
1493 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1494 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
1496 rc = obd_llog_cat_initialize(obd, &obd->obd_llogs, 1, name);
1498 CERROR("failed to setup llogging subsystems\n");
1511 fsfilt_put_ops(obd->obd_fsops);
1515 static int filter_attach(struct obd_device *obd, obd_count len, void *data)
1517 struct lprocfs_static_vars lvars;
1520 lprocfs_init_vars(filter, &lvars);
1521 rc = lprocfs_obd_attach(obd, lvars.obd_vars);
1525 rc = lprocfs_alloc_obd_stats(obd, LPROC_FILTER_LAST);
1529 /* Init obdfilter private stats here */
1530 lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_READ_BYTES,
1531 LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
1532 lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
1533 LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
1535 return lproc_filter_attach_seqstat(obd);
1538 static int filter_detach(struct obd_device *dev)
1540 lprocfs_free_obd_stats(dev);
1541 return lprocfs_obd_detach(dev);
1544 static int filter_setup(struct obd_device *obd, obd_count len, void *buf)
1546 struct lustre_cfg* lcfg = buf;
1549 /* all mount options including errors=remount-ro and asyncdel are passed
1550 * using 4th lcfg param. And it is good, finally we have got rid of
1551 * hardcoded fs types in the code. */
1552 rc = filter_common_setup(obd, len, buf, lcfg->lcfg_inlbuf4);
1555 rc = filter_post_fs_setup(obd);
1559 static int filter_cleanup(struct obd_device *obd, int flags)
1561 struct filter_obd *filter = &obd->u.filter;
1564 if (flags & OBD_OPT_FAILOVER)
1565 CERROR("%s: shutting down for failover; client state will"
1566 " be preserved.\n", obd->obd_name);
1568 if (!list_empty(&obd->obd_exports)) {
1569 CERROR("%s: still has clients!\n", obd->obd_name);
1570 class_disconnect_exports(obd, flags);
1571 if (!list_empty(&obd->obd_exports)) {
1572 CERROR("still has exports after forced cleanup?\n");
1577 ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
1579 if (filter->fo_sb == NULL)
1582 filter_post_fs_cleanup(obd);
1585 shrink_dcache_parent(filter->fo_sb->s_root);
1588 if (atomic_read(&filter->fo_vfsmnt->mnt_count) > 1)
1589 CERROR("%s: mount point %p busy, mnt_count: %d\n",
1590 obd->obd_name, filter->fo_vfsmnt,
1591 atomic_read(&filter->fo_vfsmnt->mnt_count));
1594 mntput(filter->fo_vfsmnt);
1595 //destroy_buffers(filter->fo_sb->s_dev);
1596 filter->fo_sb = NULL;
1597 fsfilt_put_ops(obd->obd_fsops);
1600 dev_clear_rdonly(2);
1605 static int filter_connect_post(struct obd_export *exp)
1607 struct obd_device *obd = exp->exp_obd;
1608 struct filter_export_data *fed;
1609 char str[PTL_NALFMT_SIZE];
1610 struct obd_llogs *llog;
1611 struct llog_ctxt *ctxt;
1615 fed = &exp->exp_filter_data;
1616 if (fed->fed_group < FILTER_MIN_GROUPS)
1619 /* initialize llogs for connections from MDS */
1620 llog = filter_grab_llog_for_group(obd, fed->fed_group, exp);
1621 LASSERT(llog != NULL);
1623 ctxt = llog_get_context(llog, LLOG_UNLINK_REPL_CTXT);
1624 LASSERT(ctxt != NULL);
1626 rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse);
1627 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
1628 exp->exp_connection->c_peer.peer_nid, str);
1629 CDEBUG(D_OTHER, "%s: init llog ctxt for export "LPX64"/%s, group %d\n",
1630 obd->obd_name, exp->exp_connection->c_peer.peer_nid,
1631 str, fed->fed_group);
1636 /* nearly identical to mds_connect */
1637 static int filter_connect(struct lustre_handle *conn, struct obd_device *obd,
1638 struct obd_uuid *cluuid, unsigned long connect_flags)
1640 struct obd_export *exp;
1641 struct filter_export_data *fed;
1642 struct filter_client_data *fcd = NULL;
1643 struct filter_obd *filter = &obd->u.filter;
1644 struct lvfs_run_ctxt saved;
1648 if (conn == NULL || obd == NULL || cluuid == NULL)
1651 rc = class_connect(conn, obd, cluuid);
1654 exp = class_conn2export(conn);
1655 LASSERT(exp != NULL);
1657 fed = &exp->exp_filter_data;
1659 spin_lock_init(&fed->fed_lock);
1661 if (obd->obd_replayable) {
1662 OBD_ALLOC(fcd, sizeof(*fcd));
1664 CERROR("filter: out of memory for client data\n");
1665 GOTO(cleanup, rc = -ENOMEM);
1668 memcpy(fcd->fcd_uuid, cluuid, sizeof(fcd->fcd_uuid));
1671 rc = filter_client_add(obd, filter, fed, -1);
1676 if (connect_flags == 0)
1679 /* connection from MDS */
1680 group = connect_flags;
1681 CWARN("%s: Received MDS connection ("LPX64"); group %d\n",
1682 obd->obd_name, exp->exp_handle.h_cookie, group);
1684 if (fed->fed_group != 0 && fed->fed_group != group) {
1685 char str[PTL_NALFMT_SIZE];
1686 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
1687 exp->exp_connection->c_peer.peer_nid, str);
1688 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
1689 "earlier; now it's trying to use group %d! This could "
1690 "be a bug in the MDS. Tell CFS.\n",
1691 exp->exp_connection->c_peer.peer_nid, str,
1692 fed->fed_group, group);
1693 GOTO(cleanup, rc = -EPROTO);
1695 fed->fed_group = group;
1697 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1698 rc = filter_read_groups(obd, group, 1);
1699 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1701 CERROR("can't read group %u\n", group);
1704 rc = filter_group_set_fs_flags(obd, group);
1706 CERROR("can't set kml flags %u\n", group);
1713 OBD_FREE(fcd, sizeof(*fcd));
1714 class_disconnect(exp, 0);
1716 class_export_put(exp);
1721 static int filter_precleanup(struct obd_device *obd, int flags)
1723 struct filter_group_llog *log;
1724 struct filter_obd *filter;
1728 filter = &obd->u.filter;
1730 spin_lock(&filter->fo_llog_list_lock);
1731 while (!list_empty(&filter->fo_llog_list)) {
1732 log = list_entry(filter->fo_llog_list.next,
1733 struct filter_group_llog, list);
1734 list_del(&log->list);
1735 spin_unlock(&filter->fo_llog_list_lock);
1737 rc = obd_llog_finish(obd, log->llogs, 0);
1739 CERROR("failed to cleanup llogging subsystem for %u\n",
1741 OBD_FREE(log->llogs, sizeof(*(log->llogs)));
1742 OBD_FREE(log, sizeof(*log));
1743 spin_lock(&filter->fo_llog_list_lock);
1745 spin_unlock(&filter->fo_llog_list_lock);
1747 rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1749 CERROR("failed to cleanup llogging subsystem\n");
1754 /* Do extra sanity checks for grant accounting. We do this at connect,
1755 * disconnect, and statfs RPC time, so it shouldn't be too bad. We can
1756 * always get rid of it or turn it off when we know accounting is good. */
1757 static void filter_grant_sanity_check(struct obd_device *obd, const char *func)
1759 struct filter_export_data *fed;
1760 struct obd_export *exp;
1761 obd_size maxsize = obd->obd_osfs.os_blocks * obd->obd_osfs.os_bsize;
1762 obd_size tot_dirty = 0, tot_pending = 0, tot_granted = 0;
1763 obd_size fo_tot_dirty, fo_tot_pending, fo_tot_granted;
1765 if (list_empty(&obd->obd_exports))
1768 spin_lock(&obd->obd_osfs_lock);
1769 spin_lock(&obd->obd_dev_lock);
1770 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1771 fed = &exp->exp_filter_data;
1772 LASSERTF(fed->fed_grant + fed->fed_pending <= maxsize,
1773 "cli %s/%p %lu+%lu > "LPU64"\n",
1774 exp->exp_client_uuid.uuid, exp,
1775 fed->fed_grant, fed->fed_pending, maxsize);
1776 LASSERTF(fed->fed_dirty <= maxsize, "cli %s/%p %lu > "LPU64"\n",
1777 exp->exp_client_uuid.uuid, exp,fed->fed_dirty,maxsize);
1778 CDEBUG(D_CACHE,"%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1779 obd->obd_name, exp->exp_client_uuid.uuid, exp,
1780 fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1781 tot_granted += fed->fed_grant + fed->fed_pending;
1782 tot_pending += fed->fed_pending;
1783 tot_dirty += fed->fed_dirty;
1785 fo_tot_granted = obd->u.filter.fo_tot_granted;
1786 fo_tot_pending = obd->u.filter.fo_tot_pending;
1787 fo_tot_dirty = obd->u.filter.fo_tot_dirty;
1788 spin_unlock(&obd->obd_dev_lock);
1789 spin_unlock(&obd->obd_osfs_lock);
1791 /* Do these assertions outside the spinlocks so we don't kill system */
1792 if (tot_granted != fo_tot_granted)
1793 CERROR("%s: tot_granted "LPU64" != fo_tot_granted "LPU64"\n",
1794 func, tot_granted, fo_tot_granted);
1795 if (tot_pending != fo_tot_pending)
1796 CERROR("%s: tot_pending "LPU64" != fo_tot_pending "LPU64"\n",
1797 func, tot_pending, fo_tot_pending);
1798 if (tot_dirty != fo_tot_dirty)
1799 CERROR("%s: tot_dirty "LPU64" != fo_tot_dirty "LPU64"\n",
1800 func, tot_dirty, fo_tot_dirty);
1801 if (tot_pending > tot_granted)
1802 CERROR("%s: tot_pending "LPU64" > tot_granted "LPU64"\n",
1803 func, tot_pending, tot_granted);
1804 if (tot_granted > maxsize)
1805 CERROR("%s: tot_granted "LPU64" > maxsize "LPU64"\n",
1806 func, tot_granted, maxsize);
1807 if (tot_dirty > maxsize)
1808 CERROR("%s: tot_dirty "LPU64" > maxsize "LPU64"\n",
1809 func, tot_dirty, maxsize);
1812 /* Remove this client from the grant accounting totals. We also remove
1813 * the export from the obd device under the osfs and dev locks to ensure
1814 * that the filter_grant_sanity_check() calculations are always valid.
1815 * The client should do something similar when it invalidates its import. */
1816 static void filter_grant_discard(struct obd_export *exp)
1818 struct obd_device *obd = exp->exp_obd;
1819 struct filter_obd *filter = &obd->u.filter;
1820 struct filter_export_data *fed = &exp->exp_filter_data;
1822 spin_lock(&obd->obd_osfs_lock);
1823 spin_lock(&exp->exp_obd->obd_dev_lock);
1824 list_del_init(&exp->exp_obd_chain);
1825 spin_unlock(&exp->exp_obd->obd_dev_lock);
1827 CDEBUG(D_CACHE, "%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1828 obd->obd_name, exp->exp_client_uuid.uuid, exp,
1829 fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1831 LASSERTF(filter->fo_tot_granted >= fed->fed_grant,
1832 "%s: tot_granted "LPU64" cli %s/%p fed_grant %lu\n",
1833 obd->obd_name, filter->fo_tot_granted,
1834 exp->exp_client_uuid.uuid, exp, fed->fed_grant);
1835 filter->fo_tot_granted -= fed->fed_grant;
1836 LASSERTF(exp->exp_obd->u.filter.fo_tot_pending >= fed->fed_pending,
1837 "%s: tot_pending "LPU64" cli %s/%p fed_pending %lu\n",
1838 obd->obd_name, filter->fo_tot_pending,
1839 exp->exp_client_uuid.uuid, exp, fed->fed_pending);
1840 LASSERTF(filter->fo_tot_dirty >= fed->fed_dirty,
1841 "%s: tot_dirty "LPU64" cli %s/%p fed_dirty %lu\n",
1842 obd->obd_name, filter->fo_tot_dirty,
1843 exp->exp_client_uuid.uuid, exp, fed->fed_dirty);
1844 filter->fo_tot_dirty -= fed->fed_dirty;
1848 spin_unlock(&obd->obd_osfs_lock);
1851 static int filter_destroy_export(struct obd_export *exp)
1855 if (exp->exp_filter_data.fed_pending)
1856 CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
1857 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
1858 exp, exp->exp_filter_data.fed_pending);
1860 target_destroy_export(exp);
1862 if (exp->exp_obd->obd_replayable)
1863 filter_client_free(exp, exp->exp_flags);
1865 filter_grant_discard(exp);
1866 if (!(exp->exp_flags & OBD_OPT_FORCE))
1867 filter_grant_sanity_check(exp->exp_obd, __FUNCTION__);
1872 static void filter_sync_llogs(struct obd_device *obd, struct obd_export *dexp)
1874 struct filter_group_llog *fglog, *nlog;
1875 struct filter_obd *filter;
1876 int worked = 0, group;
1877 struct llog_ctxt *ctxt;
1880 filter = &obd->u.filter;
1882 /* we can't sync log holding spinlock. also, we do not want to get
1883 * into livelock. so we do following: loop over MDS's exports in
1884 * group order and skip already synced llogs -bzzz */
1886 /* look for group with min. number, but > worked */
1889 spin_lock(&filter->fo_llog_list_lock);
1890 list_for_each_entry(nlog, &filter->fo_llog_list, list) {
1892 if (nlog->group <= worked) {
1893 /* this group is already synced */
1897 if (group < nlog->group) {
1898 /* we have group with smaller number to sync */
1902 /* store current minimal group */
1904 group = nlog->group;
1906 spin_unlock(&filter->fo_llog_list_lock);
1911 worked = fglog->group;
1912 if (fglog->exp && (dexp == fglog->exp || dexp == NULL)) {
1913 ctxt = llog_get_context(fglog->llogs,
1914 LLOG_UNLINK_REPL_CTXT);
1915 LASSERT(ctxt != NULL);
1916 llog_sync(ctxt, fglog->exp);
1918 } while (fglog != NULL);
1921 /* also incredibly similar to mds_disconnect */
1922 static int filter_disconnect(struct obd_export *exp, int flags)
1924 struct obd_device *obd = exp->exp_obd;
1925 unsigned long irqflags;
1930 class_export_get(exp);
1932 spin_lock_irqsave(&exp->exp_lock, irqflags);
1933 exp->exp_flags = flags;
1934 spin_unlock_irqrestore(&exp->exp_lock, irqflags);
1936 if (!(flags & OBD_OPT_FORCE))
1937 filter_grant_sanity_check(obd, __FUNCTION__);
1938 filter_grant_discard(exp);
1940 /* Disconnect early so that clients can't keep using export */
1941 rc = class_disconnect(exp, flags);
1943 ldlm_cancel_locks_for_export(exp);
1945 fsfilt_sync(obd, obd->u.filter.fo_sb);
1947 /* flush any remaining cancel messages out to the target */
1948 filter_sync_llogs(obd, exp);
1950 class_export_put(exp);
1954 struct dentry *__filter_oa2dentry(struct obd_device *obd,
1955 struct obdo *oa, const char *what)
1957 struct dentry *dchild = NULL;
1960 if (oa->o_valid & OBD_MD_FLGROUP)
1963 dchild = filter_fid2dentry(obd, NULL, group, oa->o_id);
1965 if (IS_ERR(dchild)) {
1966 CERROR("%s error looking up object: "LPU64"\n", what, oa->o_id);
1970 if (dchild->d_inode == NULL) {
1971 CERROR("%s: %s on non-existent object: "LPU64"\n",
1972 obd->obd_name, what, oa->o_id);
1974 RETURN(ERR_PTR(-ENOENT));
1980 static int filter_getattr(struct obd_export *exp, struct obdo *oa,
1981 struct lov_stripe_md *md)
1983 struct dentry *dentry = NULL;
1984 struct obd_device *obd;
1988 obd = class_exp2obd(exp);
1990 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1991 exp->exp_handle.h_cookie);
1995 dentry = filter_oa2dentry(obd, oa);
1997 RETURN(PTR_ERR(dentry));
1999 /* Limit the valid bits in the return data to what we actually use */
2000 oa->o_valid = OBD_MD_FLID;
2001 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2007 /* this is called from filter_truncate() until we have filter_punch() */
2008 static int filter_setattr(struct obd_export *exp, struct obdo *oa,
2009 struct lov_stripe_md *md, struct obd_trans_info *oti)
2011 struct lvfs_run_ctxt saved;
2012 struct filter_obd *filter;
2013 struct dentry *dentry;
2015 struct ldlm_res_id res_id = { .name = { oa->o_id, 0, oa->o_gr, 0 } };
2016 struct ldlm_resource *res;
2021 LASSERT(oti != NULL);
2023 dentry = filter_oa2dentry(exp->exp_obd, oa);
2025 RETURN(PTR_ERR(dentry));
2027 filter = &exp->exp_obd->u.filter;
2029 iattr_from_obdo(&iattr, oa, oa->o_valid);
2031 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2034 if (iattr.ia_valid & ATTR_SIZE)
2035 down(&dentry->d_inode->i_sem);
2036 handle = fsfilt_start(exp->exp_obd, dentry->d_inode, FSFILT_OP_SETATTR,
2039 GOTO(out_unlock, rc = PTR_ERR(handle));
2041 /* XXX this could be a rwsem instead, if filter_preprw played along */
2042 if (iattr.ia_valid & ATTR_ATTR_FLAG)
2043 rc = fsfilt_iocontrol(exp->exp_obd, dentry->d_inode, NULL,
2045 (long)&iattr.ia_attr_flags);
2047 rc = fsfilt_setattr(exp->exp_obd, dentry, handle, &iattr, 1);
2048 rc = filter_finish_transno(exp, oti, rc);
2049 rc2 = fsfilt_commit(exp->exp_obd, filter->fo_sb, dentry->d_inode, handle, 0);
2051 CERROR("error on commit, err = %d\n", rc2);
2056 if (iattr.ia_valid & ATTR_SIZE) {
2057 res = ldlm_resource_get(exp->exp_obd->obd_namespace, NULL,
2058 res_id, LDLM_EXTENT, 0);
2060 CERROR("!!! resource_get failed for object "LPU64" -- "
2061 "filter_setattr with no lock?\n", oa->o_id);
2063 if (res->lr_namespace->ns_lvbo &&
2064 res->lr_namespace->ns_lvbo->lvbo_update) {
2065 rc = res->lr_namespace->ns_lvbo->lvbo_update
2068 ldlm_resource_putref(res);
2072 oa->o_valid = OBD_MD_FLID;
2073 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2076 if (iattr.ia_valid & ATTR_SIZE)
2077 up(&dentry->d_inode->i_sem);
2079 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2085 /* XXX identical to osc_unpackmd */
2086 static int filter_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2087 struct lov_mds_md *lmm, int lmm_bytes)
2093 if (lmm_bytes < sizeof (*lmm)) {
2094 CERROR("lov_mds_md too small: %d, need %d\n",
2095 lmm_bytes, (int)sizeof(*lmm));
2098 /* XXX LOV_MAGIC etc check? */
2100 if (lmm->lmm_object_id == cpu_to_le64(0)) {
2101 CERROR("lov_mds_md: zero lmm_object_id\n");
2106 lsm_size = lov_stripe_md_size(1);
2110 if (*lsmp != NULL && lmm == NULL) {
2111 OBD_FREE(*lsmp, lsm_size);
2116 if (*lsmp == NULL) {
2117 OBD_ALLOC(*lsmp, lsm_size);
2121 loi_init((*lsmp)->lsm_oinfo);
2125 /* XXX zero *lsmp? */
2126 (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
2127 LASSERT((*lsmp)->lsm_object_id);
2130 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
2135 static void filter_destroy_precreated(struct obd_export *exp, struct obdo *oa,
2136 struct filter_obd *filter)
2138 struct obdo doa; /* XXX obdo on stack */
2143 memset(&doa, 0, sizeof(doa));
2144 if (oa->o_valid & OBD_MD_FLGROUP) {
2145 doa.o_valid |= OBD_MD_FLGROUP;
2146 doa.o_gr = oa->o_gr;
2150 doa.o_mode = S_IFREG;
2151 doa.o_gr = oa->o_gr;
2152 doa.o_valid = oa->o_valid & OBD_MD_FLGROUP;
2154 filter->fo_destroy_in_progress = 1;
2155 down(&filter->fo_create_lock);
2156 if (!filter->fo_destroy_in_progress) {
2157 CERROR("%s: destroy_in_progress already cleared\n",
2158 exp->exp_obd->obd_name);
2159 up(&filter->fo_create_lock);
2164 last = filter_last_id(filter, doa.o_gr);
2165 CWARN("%s: deleting orphan objects from "LPU64" to "LPU64"\n",
2166 exp->exp_obd->obd_name, oa->o_id + 1, last);
2167 for (id = oa->o_id + 1; id <= last; id++) {
2169 filter_destroy(exp, &doa, NULL, NULL);
2172 CDEBUG(D_HA, "%s: after destroy: set last_objids["LPU64"] = "LPU64"\n",
2173 exp->exp_obd->obd_name, doa.o_gr, oa->o_id);
2175 spin_lock(&filter->fo_objidlock);
2176 filter->fo_last_objids[doa.o_gr] = oa->o_id;
2177 spin_unlock(&filter->fo_objidlock);
2179 filter->fo_destroy_in_progress = 0;
2180 up(&filter->fo_create_lock);
2185 /* returns a negative error or a nonnegative number of files to create */
2186 static int filter_should_precreate(struct obd_export *exp, struct obdo *oa,
2189 struct obd_device *obd = exp->exp_obd;
2190 struct filter_obd *filter = &obd->u.filter;
2194 diff = oa->o_id - filter_last_id(filter, oa->o_gr);
2195 CDEBUG(D_INFO, "filter_last_id() = "LPU64" -> diff = %d\n",
2196 filter_last_id(filter, oa->o_gr), diff);
2198 /* delete orphans request */
2199 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2200 (oa->o_flags & OBD_FL_DELORPHAN)) {
2203 if (-diff > 10000) { /* XXX make this smarter */
2204 CERROR("ignoring bogus orphan destroy request: obdid "
2205 LPU64" last_id "LPU64"\n",
2206 oa->o_id, filter_last_id(filter, oa->o_gr));
2209 filter_destroy_precreated(exp, oa, filter);
2210 rc = filter_update_last_objid(obd, group, 0);
2212 CERROR("unable to write lastobjid, but orphans"
2216 /* only precreate if group == 0 and o_id is specfied */
2217 if (!(oa->o_valid & OBD_FL_DELORPHAN) &&
2218 (/*group != 0 ||*/ oa->o_id == 0))
2225 static int filter_precreate_rec(struct obd_device *obd, struct dentry *dentry,
2226 int *number, struct obdo *oa)
2231 rc = fsfilt_precreate_rec(obd, dentry, number, oa);
2236 /* We rely on the fact that only one thread will be creating files in a given
2237 * group at a time, which is why we don't need an atomic filter_get_new_id.
2238 * Even if we had that atomic function, the following race would exist:
2240 * thread 1: gets id x from filter_next_id
2241 * thread 2: gets id (x + 1) from filter_next_id
2242 * thread 2: creates object (x + 1)
2243 * thread 1: tries to create object x, gets -ENOSPC
2245 static int filter_precreate(struct obd_device *obd, struct obdo *oa,
2246 obd_gr group, int *num)
2248 struct dentry *dchild = NULL, *dparent = NULL;
2249 struct filter_obd *filter;
2250 int err = 0, rc = 0, recreate_obj = 0, i;
2252 void *handle = NULL;
2256 filter = &obd->u.filter;
2258 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2259 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
2263 CDEBUG(D_HA, "%s: precreating %d objects\n", obd->obd_name, *num);
2265 down(&filter->fo_create_lock);
2267 for (i = 0; i < *num && err == 0; i++) {
2268 int cleanup_phase = 0;
2270 if (filter->fo_destroy_in_progress) {
2271 CWARN("%s: precreate aborted by destroy\n",
2279 last_id = filter_last_id(filter, group);
2280 if (next_id > last_id) {
2281 CERROR("Error: Trying to recreate obj greater"
2282 "than last id "LPD64" > "LPD64"\n",
2284 GOTO(cleanup, rc = -EINVAL);
2287 next_id = filter_last_id(filter, group) + 1;
2290 CDEBUG(D_INFO, "precreate objid "LPU64"\n", next_id);
2292 dparent = filter_parent_lock(obd, group, next_id, &lock);
2293 if (IS_ERR(dparent))
2294 GOTO(cleanup, rc = PTR_ERR(dparent));
2297 /*only do precreate rec record. so clean kml flags here*/
2298 fsfilt_clear_fs_flags(obd, dparent->d_inode,
2301 dchild = filter_fid2dentry(obd, dparent, group, next_id);
2303 GOTO(cleanup, rc = PTR_ERR(dchild));
2306 if (dchild->d_inode != NULL) {
2307 /* This would only happen if lastobjid was bad on disk*/
2308 /* Could also happen if recreating missing obj but
2312 CERROR("%s: Serious error: recreating obj %*s "
2313 "but obj already exists \n",
2314 obd->obd_name, dchild->d_name.len,
2315 dchild->d_name.name);
2318 CERROR("%s: Serious error: objid %*s already "
2319 "exists; is this filesystem corrupt?\n",
2320 obd->obd_name, dchild->d_name.len,
2321 dchild->d_name.name);
2324 GOTO(cleanup, rc = -EEXIST);
2327 handle = fsfilt_start_log(obd, dparent->d_inode,
2328 FSFILT_OP_CREATE, NULL, 1);
2330 GOTO(cleanup, rc = PTR_ERR(handle));
2333 rc = ll_vfs_create(dparent->d_inode, dchild, S_IFREG, NULL);
2335 CERROR("create failed rc = %d\n", rc);
2339 if (!recreate_obj) {
2340 filter_set_last_id(filter, group, next_id);
2341 err = filter_update_last_objid(obd, group, 0);
2343 CERROR("unable to write lastobjid "
2344 "but file created\n");
2346 fsfilt_set_fs_flags(obd, dparent->d_inode, SM_DO_REC);
2349 switch(cleanup_phase) {
2351 err = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode, handle, 0);
2353 CERROR("error on commit, err = %d\n", err);
2360 filter_parent_unlock(dparent, lock);
2371 /* check if we have an error after ll_vfs_create(). It is possible that
2372 * there will be say -ENOSPC and we will leak it. */
2374 rc = filter_precreate_rec(obd, dparent, num, oa);
2376 up(&filter->fo_create_lock);
2378 CDEBUG(D_HA, "%s: server last_objid for group "LPU64": "LPU64"\n",
2379 obd->obd_name, group, filter->fo_last_objids[group]);
2381 CDEBUG(D_HA, "%s: filter_precreate() created %d objects\n",
2387 static int filter_create(struct obd_export *exp, struct obdo *oa,
2388 struct lov_stripe_md **ea, struct obd_trans_info *oti)
2390 struct obd_device *obd = NULL;
2391 struct filter_obd *filter;
2392 struct lvfs_run_ctxt saved;
2393 struct lov_stripe_md *lsm = NULL;
2394 struct filter_export_data *fed;
2395 char str[PTL_NALFMT_SIZE];
2396 int group = oa->o_gr, rc = 0, diff, recreate_objs = 0;
2399 if (!(oa->o_valid & OBD_MD_FLGROUP) || group == 0) {
2400 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2401 exp->exp_connection->c_peer.peer_nid, str);
2402 CERROR("!!! nid "LPX64"/%s sent invalid object group %d\n",
2403 exp->exp_connection->c_peer.peer_nid, str, group);
2407 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2408 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
2413 fed = &exp->exp_filter_data;
2414 filter = &obd->u.filter;
2416 if (fed->fed_group != group && !recreate_objs &&
2417 !(oa->o_valid & OBD_MD_REINT)) {
2418 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2419 exp->exp_connection->c_peer.peer_nid, str);
2420 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
2421 "earlier; now it's trying to use group %d! This could "
2422 "be a bug in the MDS. Tell CFS.\n",
2423 exp->exp_connection->c_peer.peer_nid, str,
2424 fed->fed_group, group);
2428 CDEBUG(D_INFO, "filter_create(od->o_gr=%d,od->o_id="LPU64")\n",
2433 rc = obd_alloc_memmd(exp, &lsm);
2440 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2442 if (oa->o_valid & OBD_MD_REINT) {
2443 int num = *((int*)oa->o_inline);
2444 rc = filter_precreate(obd, oa, oa->o_gr, &num);
2445 } else if (recreate_objs) {
2446 if (oa->o_id > filter_last_id(&obd->u.filter, group)) {
2447 CERROR("recreate objid "LPU64" > last id "LPU64"\n",
2448 oa->o_id, filter_last_id(&obd->u.filter, group));
2452 rc = filter_precreate(obd, oa, group, &diff);
2455 diff = filter_should_precreate(exp, oa, group);
2457 oa->o_id = filter_last_id(&obd->u.filter, group);
2458 rc = filter_precreate(obd, oa, group, &diff);
2460 oa->o_valid = OBD_MD_FLID;
2464 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2465 if (rc && ea != NULL && *ea != lsm) {
2466 obd_free_memmd(exp, &lsm);
2467 } else if (rc == 0 && ea != NULL) {
2468 /* XXX LOV STACKING: the lsm that is passed to us from
2469 * LOV does not have valid lsm_oinfo data structs, so
2470 * don't go touching that. This needs to be fixed in a
2472 lsm->lsm_object_id = oa->o_id;
2473 lsm->lsm_object_gr = oa->o_gr;
2480 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
2481 struct lov_stripe_md *ea, struct obd_trans_info *oti)
2483 struct obd_device *obd;
2484 struct filter_obd *filter;
2485 struct dentry *dchild = NULL, *dparent = NULL;
2486 struct lvfs_run_ctxt saved;
2487 void *handle = NULL;
2488 struct llog_cookie *fcc = NULL;
2489 int rc, rc2, cleanup_phase = 0, have_prepared = 0;
2493 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2496 filter = &obd->u.filter;
2498 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2501 dparent = filter_parent_lock(obd, oa->o_gr, oa->o_id, &lock);
2502 if (IS_ERR(dparent))
2503 GOTO(cleanup, rc = PTR_ERR(dparent));
2506 dchild = filter_fid2dentry(obd, dparent, oa->o_gr, oa->o_id);
2508 GOTO(cleanup, rc = -ENOENT);
2511 if (dchild->d_inode == NULL) {
2512 CDEBUG(D_INODE, "destroying non-existent object "LPU64"\n",
2514 GOTO(cleanup, rc = -ENOENT);
2517 if (!have_prepared) {
2518 /* If we're really going to destroy the object, get ready
2519 * by getting the clients to discard their cached data.
2521 * We have to drop the parent lock, because
2522 * filter_prepare_destroy will acquire a PW on the object, and
2523 * we don't want to deadlock with an incoming write to the
2524 * object, which has the extent PW and then wants to get the
2525 * parent dentry to do the lookup.
2527 * We dput the child because it's not worth the extra
2528 * complication of condition the above code to skip it on the
2529 * second time through. */
2531 filter_parent_unlock(dparent, lock);
2533 filter_prepare_destroy(obd, oa->o_id, oa->o_gr);
2538 handle = fsfilt_start_log(obd, dparent->d_inode, FSFILT_OP_UNLINK, oti, 1);
2540 GOTO(cleanup, rc = PTR_ERR(handle));
2544 /* Our MDC connection is established by the MDS to us */
2545 if (oa->o_valid & OBD_MD_FLCOOKIE) {
2546 OBD_ALLOC(fcc, sizeof(*fcc));
2548 memcpy(fcc, obdo_logcookie(oa), sizeof(*fcc));
2551 rc = filter_destroy_internal(obd, oa->o_id, dparent, dchild);
2554 switch(cleanup_phase) {
2558 fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
2560 filter_cancel_cookies_cb,
2563 fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
2565 filter_cancel_cookies_cb,
2568 rc = filter_finish_transno(exp, oti, rc);
2569 rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode,
2572 CERROR("error on commit, err = %d\n", rc2);
2579 filter_parent_unlock(dparent, lock);
2581 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2584 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
2591 /* NB start and end are used for punch, but not truncate */
2592 static int filter_truncate(struct obd_export *exp, struct obdo *oa,
2593 struct lov_stripe_md *lsm,
2594 obd_off start, obd_off end,
2595 struct obd_trans_info *oti)
2600 if (end != OBD_OBJECT_EOF)
2601 CERROR("PUNCH not supported, only truncate: end = "LPX64"\n",
2604 CDEBUG(D_INODE, "calling truncate for object "LPU64", valid = %x, "
2605 "o_size = "LPD64"\n", oa->o_id, oa->o_valid, start);
2607 error = filter_setattr(exp, oa, NULL, oti);
2611 static int filter_sync(struct obd_export *exp, struct obdo *oa,
2612 struct lov_stripe_md *lsm, obd_off start, obd_off end)
2614 struct obd_device *obd = exp->exp_obd;
2615 struct lvfs_run_ctxt saved;
2616 struct filter_obd *filter;
2617 struct dentry *dentry;
2621 filter = &obd->u.filter;
2623 /* an objid of zero is taken to mean "sync whole filesystem" */
2624 if (!oa || !(oa->o_valid & OBD_MD_FLID)) {
2625 rc = fsfilt_sync(obd, filter->fo_sb);
2626 /* flush any remaining cancel messages out to the target */
2627 filter_sync_llogs(obd, NULL);
2631 dentry = filter_oa2dentry(obd, oa);
2633 RETURN(PTR_ERR(dentry));
2635 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2637 down(&dentry->d_inode->i_sem);
2638 rc = filemap_fdatasync(dentry->d_inode->i_mapping);
2640 /* just any file to grab fsync method - "file" arg unused */
2641 struct file *file = filter->fo_rcvd_filp;
2643 if (file->f_op && file->f_op->fsync)
2644 rc = file->f_op->fsync(NULL, dentry, 1);
2646 rc2 = filemap_fdatawait(dentry->d_inode->i_mapping);
2650 up(&dentry->d_inode->i_sem);
2652 oa->o_valid = OBD_MD_FLID;
2653 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2655 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2661 static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2662 unsigned long max_age)
2664 struct filter_obd *filter = &obd->u.filter;
2665 int blockbits = filter->fo_sb->s_blocksize_bits;
2669 /* at least try to account for cached pages. its still racey and
2670 * might be under-reporting if clients haven't announced their
2671 * caches with brw recently */
2672 spin_lock(&obd->obd_osfs_lock);
2673 rc = fsfilt_statfs(obd, filter->fo_sb, max_age);
2674 memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
2675 spin_unlock(&obd->obd_osfs_lock);
2677 CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
2678 " pending "LPU64" free "LPU64" avail "LPU64"\n",
2679 filter->fo_tot_dirty, filter->fo_tot_granted,
2680 filter->fo_tot_pending,
2681 osfs->os_bfree << blockbits, osfs->os_bavail << blockbits);
2683 filter_grant_sanity_check(obd, __FUNCTION__);
2685 osfs->os_bavail -= min(osfs->os_bavail,
2686 (filter->fo_tot_dirty + filter->fo_tot_pending +
2687 osfs->os_bsize -1) >> blockbits);
2692 static int filter_get_info(struct obd_export *exp, __u32 keylen,
2693 void *key, __u32 *vallen, void *val)
2695 struct filter_export_data *fed = &exp->exp_filter_data;
2696 struct obd_device *obd;
2699 obd = class_exp2obd(exp);
2701 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2702 exp->exp_handle.h_cookie);
2706 if (keylen == strlen("blocksize") &&
2707 memcmp(key, "blocksize", keylen) == 0) {
2708 __u32 *blocksize = val;
2709 *vallen = sizeof(*blocksize);
2710 *blocksize = obd->u.filter.fo_sb->s_blocksize;
2714 if (keylen == strlen("blocksize_bits") &&
2715 memcmp(key, "blocksize_bits", keylen) == 0) {
2716 __u32 *blocksize_bits = val;
2717 *vallen = sizeof(*blocksize_bits);
2718 *blocksize_bits = obd->u.filter.fo_sb->s_blocksize_bits;
2722 if (keylen >= strlen("last_id") && memcmp(key, "last_id", 7) == 0) {
2723 obd_id *last_id = val;
2724 *last_id = filter_last_id(&obd->u.filter, fed->fed_group);
2727 if (keylen >= strlen("reint_log") && memcmp(key, "reint_log", 9) == 0) {
2728 /*Get log_context handle*/
2729 unsigned long *llh_handle = val;
2730 *vallen = sizeof(unsigned long);
2731 *llh_handle = (unsigned long)obd->obd_llog_ctxt[LLOG_REINT_ORIG_CTXT];
2734 if (keylen >= strlen("cache_sb") && memcmp(key, "cache_sb", 8) == 0) {
2735 /*Get log_context handle*/
2736 unsigned long *sb = val;
2737 *vallen = sizeof(unsigned long);
2738 *sb = (unsigned long)obd->u.filter.fo_sb;
2742 CDEBUG(D_IOCTL, "invalid key\n");
2746 struct obd_llogs *filter_grab_llog_for_group(struct obd_device *obd, int group,
2747 struct obd_export *export)
2749 struct filter_group_llog *fglog, *nlog;
2750 char name[32] = "CATLIST";
2751 struct filter_obd *filter;
2752 struct llog_ctxt *ctxt;
2753 struct list_head *cur;
2756 filter = &obd->u.filter;
2758 spin_lock(&filter->fo_llog_list_lock);
2759 list_for_each(cur, &filter->fo_llog_list) {
2760 fglog = list_entry(cur, struct filter_group_llog, list);
2761 if (fglog->group == group) {
2762 if (!(fglog->exp == NULL || fglog->exp == export || export == NULL))
2763 CWARN("%s: export for group %d changes: 0x%p -> 0x%p\n",
2764 obd->obd_name, group, fglog->exp, export);
2765 spin_unlock(&filter->fo_llog_list_lock);
2769 spin_unlock(&filter->fo_llog_list_lock);
2774 OBD_ALLOC(fglog, sizeof(*fglog));
2777 fglog->group = group;
2779 OBD_ALLOC(fglog->llogs, sizeof(struct obd_llogs));
2780 if (fglog->llogs == NULL) {
2781 OBD_FREE(fglog, sizeof(*fglog));
2785 spin_lock(&filter->fo_llog_list_lock);
2786 list_for_each(cur, &filter->fo_llog_list) {
2787 nlog = list_entry(cur, struct filter_group_llog, list);
2788 LASSERT(nlog->group != group);
2790 list_add(&fglog->list, &filter->fo_llog_list);
2791 spin_unlock(&filter->fo_llog_list_lock);
2793 rc = obd_llog_cat_initialize(obd, fglog->llogs, 1, name);
2795 OBD_FREE(fglog->llogs, sizeof(*(fglog->llogs)));
2796 OBD_FREE(fglog, sizeof(*fglog));
2802 fglog->exp = export;
2803 ctxt = llog_get_context(fglog->llogs, LLOG_UNLINK_REPL_CTXT);
2804 LASSERT(ctxt != NULL);
2806 llog_receptor_accept(ctxt, export->exp_imp_reverse);
2808 CDEBUG(D_OTHER, "%s: new llog 0x%p for group %u\n",
2809 obd->obd_name, fglog->llogs, group);
2811 RETURN(fglog->llogs);
2814 int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
2815 int len, void *karg, void *uarg)
2817 struct obd_device *obd = exp->exp_obd;
2818 struct obd_ioctl_data *data = karg;
2822 case OBD_IOC_ABORT_RECOVERY:
2823 target_stop_recovery_thread(obd);
2826 case OBD_IOC_SET_READONLY: {
2828 struct super_block *sb = obd->u.filter.fo_sb;
2829 struct inode *inode = sb->s_root->d_inode;
2830 BDEVNAME_DECLARE_STORAGE(tmp);
2831 CERROR("setting device %s read-only\n",
2832 ll_bdevname(sb, tmp));
2834 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
2836 (void)fsfilt_commit(obd, sb, inode, handle, 1);
2838 dev_set_rdonly(ll_sbdev(obd->u.filter.fo_sb), 2);
2842 case OBD_IOC_CATLOGLIST: {
2843 rc = llog_catalog_list(obd, 1, data);
2847 case OBD_IOC_SNAP_ADD: {
2848 char *name = data->ioc_inlbuf1;
2850 rc = fsfilt_set_snap_item(obd, obd->u.filter.fo_sb, name);
2854 case OBD_IOC_LLOG_CANCEL:
2855 case OBD_IOC_LLOG_REMOVE:
2856 case OBD_IOC_LLOG_INFO:
2857 case OBD_IOC_LLOG_PRINT: {
2858 /* FIXME to be finished */
2859 RETURN(-EOPNOTSUPP);
2861 struct llog_ctxt *ctxt = NULL;
2863 push_ctxt(&saved, &ctxt->loc_ctxt, NULL);
2864 rc = llog_ioctl(ctxt, cmd, data);
2865 pop_ctxt(&saved, &ctxt->loc_ctxt, NULL);
2878 static struct llog_operations filter_unlink_repl_logops;
2879 static struct llog_operations filter_size_orig_logops = {
2880 lop_setup: llog_obd_origin_setup,
2881 lop_cleanup: llog_catalog_cleanup,
2882 lop_add: llog_catalog_add,
2885 static int filter_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
2886 struct obd_device *tgt, int count,
2887 struct llog_catid *catid)
2889 struct llog_ctxt *ctxt;
2893 filter_unlink_repl_logops = llog_client_ops;
2894 filter_unlink_repl_logops.lop_cancel = llog_obd_repl_cancel;
2895 filter_unlink_repl_logops.lop_connect = llog_repl_connect;
2896 filter_unlink_repl_logops.lop_sync = llog_obd_repl_sync;
2898 rc = obd_llog_setup(obd, llogs, LLOG_UNLINK_REPL_CTXT, tgt, 0, NULL,
2899 &filter_unlink_repl_logops);
2902 /* FIXME - assign unlink_cb for filter's recovery */
2903 ctxt = llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT);
2904 ctxt->llog_proc_cb = filter_recov_log_unlink_cb;
2906 /* FIXME - count should be 1 to setup size log */
2907 rc = obd_llog_setup(obd, llogs, LLOG_SIZE_ORIG_CTXT, tgt, 0,
2908 &catid->lci_logid, &filter_size_orig_logops);
2912 static int filter_llog_finish(struct obd_device *obd,
2913 struct obd_llogs *llogs, int count)
2918 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT));
2922 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_SIZE_ORIG_CTXT));
2926 static int filter_llog_connect(struct obd_export *exp,
2927 struct llogd_conn_body *body)
2929 struct obd_device *obd = exp->exp_obd;
2930 struct llog_ctxt *ctxt;
2931 struct obd_llogs *llog;
2935 CDEBUG(D_OTHER, "handle connect for %s: %u/%u/%u\n", obd->obd_name,
2936 (unsigned) body->lgdc_logid.lgl_ogr,
2937 (unsigned) body->lgdc_logid.lgl_oid,
2938 (unsigned) body->lgdc_logid.lgl_ogen);
2939 llog = filter_grab_llog_for_group(obd, body->lgdc_logid.lgl_ogr, exp);
2940 LASSERT(llog != NULL);
2941 ctxt = llog_get_context(llog, body->lgdc_ctxt_idx);
2942 rc = llog_connect(ctxt, 1, &body->lgdc_logid,
2943 &body->lgdc_gen, NULL);
2945 CERROR("failed to connect\n");
2950 static struct dentry *filter_lvfs_fid2dentry(__u64 id, __u32 gen, __u64 gr,
2953 return filter_fid2dentry(data, NULL, gr, id);
2956 static struct lvfs_callback_ops filter_lvfs_ops = {
2957 l_fid2dentry: filter_lvfs_fid2dentry,
2960 static struct obd_ops filter_obd_ops = {
2961 .o_owner = THIS_MODULE,
2962 .o_attach = filter_attach,
2963 .o_detach = filter_detach,
2964 .o_get_info = filter_get_info,
2965 .o_setup = filter_setup,
2966 .o_precleanup = filter_precleanup,
2967 .o_cleanup = filter_cleanup,
2968 .o_connect = filter_connect,
2969 .o_connect_post = filter_connect_post,
2970 .o_disconnect = filter_disconnect,
2971 .o_statfs = filter_statfs,
2972 .o_getattr = filter_getattr,
2973 .o_unpackmd = filter_unpackmd,
2974 .o_create = filter_create,
2975 .o_setattr = filter_setattr,
2976 .o_destroy = filter_destroy,
2977 .o_brw = filter_brw,
2978 .o_punch = filter_truncate,
2979 .o_sync = filter_sync,
2980 .o_preprw = filter_preprw,
2981 .o_commitrw = filter_commitrw,
2982 .o_do_cow = filter_do_cow,
2983 .o_write_extents = filter_write_extents,
2984 .o_destroy_export = filter_destroy_export,
2985 .o_llog_init = filter_llog_init,
2986 .o_llog_finish = filter_llog_finish,
2987 .o_llog_connect = filter_llog_connect,
2988 .o_iocontrol = filter_iocontrol,
2991 static struct obd_ops filter_sanobd_ops = {
2992 .o_owner = THIS_MODULE,
2993 .o_attach = filter_attach,
2994 .o_detach = filter_detach,
2995 .o_get_info = filter_get_info,
2996 .o_setup = filter_san_setup,
2997 .o_precleanup = filter_precleanup,
2998 .o_cleanup = filter_cleanup,
2999 .o_connect = filter_connect,
3000 .o_connect_post = filter_connect_post,
3001 .o_disconnect = filter_disconnect,
3002 .o_statfs = filter_statfs,
3003 .o_getattr = filter_getattr,
3004 .o_unpackmd = filter_unpackmd,
3005 .o_create = filter_create,
3006 .o_setattr = filter_setattr,
3007 .o_destroy = filter_destroy,
3008 .o_brw = filter_brw,
3009 .o_punch = filter_truncate,
3010 .o_sync = filter_sync,
3011 .o_preprw = filter_preprw,
3012 .o_commitrw = filter_commitrw,
3013 .o_do_cow = filter_do_cow,
3014 .o_write_extents = filter_write_extents,
3015 .o_san_preprw = filter_san_preprw,
3016 .o_destroy_export = filter_destroy_export,
3017 .o_llog_init = filter_llog_init,
3018 .o_llog_finish = filter_llog_finish,
3019 .o_llog_connect = filter_llog_connect,
3020 .o_iocontrol = filter_iocontrol,
3023 static int __init obdfilter_init(void)
3025 struct lprocfs_static_vars lvars;
3028 printk(KERN_INFO "Lustre: Filtering OBD driver; info@clusterfs.com\n");
3030 lprocfs_init_vars(filter, &lvars);
3032 rc = class_register_type(&filter_obd_ops, NULL, lvars.module_vars,
3033 OBD_FILTER_DEVICENAME);
3037 rc = class_register_type(&filter_sanobd_ops, NULL, lvars.module_vars,
3038 OBD_FILTER_SAN_DEVICENAME);
3040 class_unregister_type(OBD_FILTER_DEVICENAME);
3044 static void __exit obdfilter_exit(void)
3046 class_unregister_type(OBD_FILTER_SAN_DEVICENAME);
3047 class_unregister_type(OBD_FILTER_DEVICENAME);
3050 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
3051 MODULE_DESCRIPTION("Lustre Filtering OBD driver");
3052 MODULE_LICENSE("GPL");
3054 module_init(obdfilter_init);
3055 module_exit(obdfilter_exit);