1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * linux/fs/obdfilter/filter.c
6 * Copyright (c) 2001-2003 Cluster File Systems, Inc.
7 * Author: Peter Braam <braam@clusterfs.com>
8 * Author: Andreas Dilger <adilger@clusterfs.com>
10 * This file is part of Lustre, http://www.lustre.org.
12 * Lustre is free software; you can redistribute it and/or
13 * modify it under the terms of version 2 of the GNU General Public
14 * License as published by the Free Software Foundation.
16 * Lustre is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with Lustre; if not, write to the Free Software
23 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
27 * Invariant: Get O/R i_sem for lookup, if needed, before any journal ops
28 * (which need to get journal_lock, may block if journal full).
30 * Invariant: Call filter_start_transno() before any journal ops to avoid the
31 * same deadlock problem. We can (and want) to get rid of the
32 * transno sem in favour of the dir/inode i_sem to avoid single
33 * threaded operation on the OST.
36 #define DEBUG_SUBSYSTEM S_FILTER
38 #include <linux/config.h>
39 #include <linux/module.h>
41 #include <linux/dcache.h>
42 #include <linux/init.h>
43 #include <linux/version.h>
44 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
45 # include <linux/mount.h>
46 # include <linux/buffer_head.h>
49 #include <linux/obd_class.h>
50 #include <linux/obd_lov.h>
51 #include <linux/lustre_dlm.h>
52 #include <linux/lustre_fsfilt.h>
53 #include <linux/lprocfs_status.h>
54 #include <linux/lustre_log.h>
55 #include <linux/lustre_commit_confd.h>
56 #include <portals/list.h>
58 #include "filter_internal.h"
60 static struct lvfs_callback_ops filter_lvfs_ops;
62 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
63 struct lov_stripe_md *ea, struct obd_trans_info *);
65 static void filter_commit_cb(struct obd_device *obd, __u64 transno,
66 void *cb_data, int error)
68 obd_transno_commit_cb(obd, transno, error);
72 /* Assumes caller has already pushed us into the kernel context. */
73 int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti,
76 struct filter_obd *filter = &exp->exp_obd->u.filter;
77 struct filter_export_data *fed = &exp->exp_filter_data;
78 struct filter_client_data *fcd = fed->fed_fcd;
81 int err, log_pri = D_HA;
83 /* Propagate error code. */
87 if (!exp->exp_obd->obd_replayable || oti == NULL)
90 /* we don't allocate new transnos for replayed requests */
91 if (oti->oti_transno == 0) {
92 spin_lock(&filter->fo_translock);
93 last_rcvd = le64_to_cpu(filter->fo_fsd->fsd_last_transno) + 1;
94 filter->fo_fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
95 spin_unlock(&filter->fo_translock);
96 oti->oti_transno = last_rcvd;
98 spin_lock(&filter->fo_translock);
99 last_rcvd = oti->oti_transno;
100 if (last_rcvd > le64_to_cpu(filter->fo_fsd->fsd_last_transno))
101 filter->fo_fsd->fsd_last_transno =
102 cpu_to_le64(last_rcvd);
103 spin_unlock(&filter->fo_translock);
105 fcd->fcd_last_rcvd = cpu_to_le64(last_rcvd);
107 /* could get xid from oti, if it's ever needed */
108 fcd->fcd_last_xid = 0;
110 off = fed->fed_lr_off;
111 fsfilt_add_journal_cb(exp->exp_obd, last_rcvd, oti->oti_handle,
112 filter_commit_cb, NULL);
113 err = fsfilt_write_record(exp->exp_obd, filter->fo_rcvd_filp, fcd,
114 sizeof(*fcd), &off, 0);
121 CDEBUG(log_pri, "wrote trans "LPU64" for client %s at #%d: err = %d\n",
122 last_rcvd, fcd->fcd_uuid, fed->fed_lr_idx, err);
127 void f_dput(struct dentry *dentry)
129 /* Can't go inside filter_ddelete because it can block */
130 CDEBUG(D_INODE, "putting %s: %p, count = %d\n",
131 dentry->d_name.name, dentry, atomic_read(&dentry->d_count) - 1);
132 LASSERT(atomic_read(&dentry->d_count) > 0);
137 /* Add client data to the FILTER. We use a bitmap to locate a free space
138 * in the last_rcvd file if cl_idx is -1 (i.e. a new client).
139 * Otherwise, we have just read the data from the last_rcvd file and
140 * we know its offset. */
141 static int filter_client_add(struct obd_device *obd, struct filter_obd *filter,
142 struct filter_export_data *fed, int cl_idx)
144 unsigned long *bitmap = filter->fo_last_rcvd_slots;
145 int new_client = (cl_idx == -1);
148 LASSERT(bitmap != NULL);
150 /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
151 if (!strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid))
154 /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
155 * there's no need for extra complication here
158 cl_idx = find_first_zero_bit(bitmap, FILTER_LR_MAX_CLIENTS);
160 if (cl_idx >= FILTER_LR_MAX_CLIENTS) {
161 CERROR("no client slots - fix FILTER_LR_MAX_CLIENTS\n");
164 if (test_and_set_bit(cl_idx, bitmap)) {
165 CERROR("FILTER client %d: found bit is set in bitmap\n",
167 cl_idx = find_next_zero_bit(bitmap,
168 FILTER_LR_MAX_CLIENTS,
173 if (test_and_set_bit(cl_idx, bitmap)) {
174 CERROR("FILTER client %d: bit already set in bitmap!\n",
180 fed->fed_lr_idx = cl_idx;
181 fed->fed_lr_off = le32_to_cpu(filter->fo_fsd->fsd_client_start) +
182 cl_idx * le16_to_cpu(filter->fo_fsd->fsd_client_size);
184 CDEBUG(D_INFO, "client at index %d (%llu) with UUID '%s' added\n",
185 fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
188 struct obd_run_ctxt saved;
189 loff_t off = fed->fed_lr_off;
193 CDEBUG(D_INFO, "writing client fcd at idx %u (%llu) (len %u)\n",
194 fed->fed_lr_idx,off,(unsigned int)sizeof(*fed->fed_fcd));
196 push_ctxt(&saved, &obd->obd_ctxt, NULL);
197 /* Transaction needed to fix bug 1403 */
198 handle = fsfilt_start(obd,
199 filter->fo_rcvd_filp->f_dentry->d_inode,
200 FSFILT_OP_SETATTR, NULL);
201 if (IS_ERR(handle)) {
202 err = PTR_ERR(handle);
203 CERROR("unable to start transaction: rc %d\n", err);
205 err = fsfilt_write_record(obd, filter->fo_rcvd_filp,
207 sizeof(*fed->fed_fcd),
210 filter->fo_rcvd_filp->f_dentry->d_inode,
213 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
216 CERROR("error writing %s client idx %u: rc %d\n",
217 LAST_RCVD, fed->fed_lr_idx, err);
224 static int filter_client_free(struct obd_export *exp, int flags)
226 struct filter_export_data *fed = &exp->exp_filter_data;
227 struct filter_obd *filter = &exp->exp_obd->u.filter;
228 struct obd_device *obd = exp->exp_obd;
229 struct filter_client_data zero_fcd;
230 struct obd_run_ctxt saved;
235 if (fed->fed_fcd == NULL)
238 if (flags & OBD_OPT_FAILOVER)
241 /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
242 if (strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid ) == 0)
245 LASSERT(filter->fo_last_rcvd_slots != NULL);
247 off = fed->fed_lr_off;
249 CDEBUG(D_INFO, "freeing client at idx %u (%lld) with UUID '%s'\n",
250 fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
252 /* Clear the bit _after_ zeroing out the client so we don't
253 race with filter_client_add and zero out new clients.*/
254 if (!test_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
255 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
260 memset(&zero_fcd, 0, sizeof zero_fcd);
261 push_ctxt(&saved, &obd->obd_ctxt, NULL);
262 rc = fsfilt_write_record(obd, filter->fo_rcvd_filp, &zero_fcd,
263 sizeof(zero_fcd), &off, 1);
264 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
266 CDEBUG(rc == 0 ? D_INFO : D_ERROR,
267 "zeroing disconnecting client %s at idx %u (%llu) in %s rc %d\n",
268 fed->fed_fcd->fcd_uuid, fed->fed_lr_idx, fed->fed_lr_off,
271 if (!test_and_clear_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
272 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
278 OBD_FREE(fed->fed_fcd, sizeof(*fed->fed_fcd));
283 static int filter_free_server_data(struct filter_obd *filter)
285 OBD_FREE(filter->fo_fsd, sizeof(*filter->fo_fsd));
286 filter->fo_fsd = NULL;
287 OBD_FREE(filter->fo_last_rcvd_slots,
288 FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
289 filter->fo_last_rcvd_slots = NULL;
293 /* assumes caller is already in kernel ctxt */
294 int filter_update_server_data(struct obd_device *obd, struct file *filp,
295 struct filter_server_data *fsd, int force_sync)
301 CDEBUG(D_INODE, "server uuid : %s\n", fsd->fsd_uuid);
302 CDEBUG(D_INODE, "server last_rcvd : "LPU64"\n",
303 le64_to_cpu(fsd->fsd_last_transno));
304 CDEBUG(D_INODE, "server last_mount: "LPU64"\n",
305 le64_to_cpu(fsd->fsd_mount_count));
307 rc = fsfilt_write_record(obd, filp, fsd, sizeof(*fsd), &off,force_sync);
309 CERROR("error writing filter_server_data: rc = %d\n", rc);
314 int filter_update_last_objid(struct obd_device *obd, obd_gr group,
317 struct filter_obd *filter = &obd->u.filter;
323 CDEBUG(D_INODE, "server last_objid for group "LPU64": "LPU64"\n",
324 group, filter->fo_last_objids[group]);
326 tmp = cpu_to_le64(filter->fo_last_objids[group]);
327 rc = fsfilt_write_record(obd, filter->fo_last_objid_files[group],
328 &tmp, sizeof(tmp), &off, force_sync);
330 CERROR("error writing group "LPU64" last objid: rc = %d\n",
335 /* assumes caller has already in kernel ctxt */
336 static int filter_init_server_data(struct obd_device *obd, struct file * filp)
338 struct filter_obd *filter = &obd->u.filter;
339 struct filter_server_data *fsd;
340 struct filter_client_data *fcd = NULL;
341 struct inode *inode = filp->f_dentry->d_inode;
342 unsigned long last_rcvd_size = inode->i_size;
348 /* ensure padding in the struct is the correct size */
349 LASSERT (offsetof(struct filter_server_data, fsd_padding) +
350 sizeof(fsd->fsd_padding) == FILTER_LR_SERVER_SIZE);
351 LASSERT (offsetof(struct filter_client_data, fcd_padding) +
352 sizeof(fcd->fcd_padding) == FILTER_LR_CLIENT_SIZE);
354 OBD_ALLOC(fsd, sizeof(*fsd));
357 filter->fo_fsd = fsd;
359 OBD_ALLOC(filter->fo_last_rcvd_slots,
360 FILTER_LR_MAX_CLIENT_WORDS * sizeof(unsigned long));
361 if (filter->fo_last_rcvd_slots == NULL) {
362 OBD_FREE(fsd, sizeof(*fsd));
366 if (last_rcvd_size == 0) {
367 CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD);
369 memcpy(fsd->fsd_uuid, obd->obd_uuid.uuid,sizeof(fsd->fsd_uuid));
370 fsd->fsd_last_transno = 0;
371 mount_count = fsd->fsd_mount_count = 0;
372 fsd->fsd_server_size = cpu_to_le32(FILTER_LR_SERVER_SIZE);
373 fsd->fsd_client_start = cpu_to_le32(FILTER_LR_CLIENT_START);
374 fsd->fsd_client_size = cpu_to_le16(FILTER_LR_CLIENT_SIZE);
375 fsd->fsd_subdir_count = cpu_to_le16(FILTER_SUBDIR_COUNT);
376 filter->fo_subdir_count = FILTER_SUBDIR_COUNT;
378 rc = fsfilt_read_record(obd, filp, fsd, sizeof(*fsd), &off);
380 CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
384 if (strcmp(fsd->fsd_uuid, obd->obd_uuid.uuid) != 0) {
385 CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
386 obd->obd_uuid.uuid, fsd->fsd_uuid);
387 GOTO(err_fsd, rc = -EINVAL);
389 mount_count = le64_to_cpu(fsd->fsd_mount_count);
390 filter->fo_subdir_count = le16_to_cpu(fsd->fsd_subdir_count);
393 if (fsd->fsd_feature_incompat & ~cpu_to_le32(FILTER_INCOMPAT_SUPP)) {
394 CERROR("unsupported feature %x\n",
395 le32_to_cpu(fsd->fsd_feature_incompat) &
396 ~FILTER_INCOMPAT_SUPP);
397 GOTO(err_fsd, rc = -EINVAL);
399 if (fsd->fsd_feature_rocompat & ~cpu_to_le32(FILTER_ROCOMPAT_SUPP)) {
400 CERROR("read-only feature %x\n",
401 le32_to_cpu(fsd->fsd_feature_rocompat) &
402 ~FILTER_ROCOMPAT_SUPP);
403 /* Do something like remount filesystem read-only */
404 GOTO(err_fsd, rc = -EINVAL);
407 CDEBUG(D_INODE, "%s: server last_rcvd : "LPU64"\n",
408 obd->obd_name, le64_to_cpu(fsd->fsd_last_transno));
409 CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
410 obd->obd_name, mount_count + 1);
411 CDEBUG(D_INODE, "%s: server data size: %u\n",
412 obd->obd_name, le32_to_cpu(fsd->fsd_server_size));
413 CDEBUG(D_INODE, "%s: per-client data start: %u\n",
414 obd->obd_name, le32_to_cpu(fsd->fsd_client_start));
415 CDEBUG(D_INODE, "%s: per-client data size: %u\n",
416 obd->obd_name, le32_to_cpu(fsd->fsd_client_size));
417 CDEBUG(D_INODE, "%s: server subdir_count: %u\n",
418 obd->obd_name, le16_to_cpu(fsd->fsd_subdir_count));
419 CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
420 last_rcvd_size <= le32_to_cpu(fsd->fsd_client_start) ? 0 :
421 (last_rcvd_size - le32_to_cpu(fsd->fsd_client_start)) /
422 le16_to_cpu(fsd->fsd_client_size));
424 if (!obd->obd_replayable) {
425 CWARN("%s: recovery support OFF\n", obd->obd_name);
429 for (cl_idx = 0, off = le32_to_cpu(fsd->fsd_client_start);
430 off < last_rcvd_size; cl_idx++) {
432 struct obd_export *exp;
433 struct filter_export_data *fed;
436 OBD_ALLOC(fcd, sizeof(*fcd));
438 GOTO(err_client, rc = -ENOMEM);
441 /* Don't assume off is incremented properly by
442 * fsfilt_read_record(), in case sizeof(*fcd)
443 * isn't the same as fsd->fsd_client_size. */
444 off = le32_to_cpu(fsd->fsd_client_start) +
445 cl_idx * le16_to_cpu(fsd->fsd_client_size);
446 rc = fsfilt_read_record(obd, filp, fcd, sizeof(*fcd), &off);
448 CERROR("error reading FILT %s idx %d off %llu: rc %d\n",
449 LAST_RCVD, cl_idx, off, rc);
450 break; /* read error shouldn't cause startup to fail */
453 if (fcd->fcd_uuid[0] == '\0') {
454 CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
459 last_rcvd = le64_to_cpu(fcd->fcd_last_rcvd);
461 /* These exports are cleaned up by filter_disconnect(), so they
462 * need to be set up like real exports as filter_connect() does.
464 exp = class_new_export(obd);
465 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
466 " srv lr: "LPU64"\n", fcd->fcd_uuid, cl_idx,
467 last_rcvd, le64_to_cpu(fsd->fsd_last_transno));
469 GOTO(err_client, rc = -ENOMEM);
471 memcpy(&exp->exp_client_uuid.uuid, fcd->fcd_uuid,
472 sizeof exp->exp_client_uuid.uuid);
473 fed = &exp->exp_filter_data;
475 filter_client_add(obd, filter, fed, cl_idx);
476 /* create helper if export init gets more complex */
477 spin_lock_init(&fed->fed_lock);
480 obd->obd_recoverable_clients++;
481 class_export_put(exp);
483 CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
486 if (last_rcvd > le64_to_cpu(fsd->fsd_last_transno))
487 fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
491 obd->obd_last_committed = le64_to_cpu(fsd->fsd_last_transno);
493 if (obd->obd_recoverable_clients) {
494 CWARN("RECOVERY: %d recoverable clients, last_rcvd "
495 LPU64"\n", obd->obd_recoverable_clients,
496 le64_to_cpu(fsd->fsd_last_transno));
497 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
498 obd->obd_recovering = 1;
502 OBD_FREE(fcd, sizeof(*fcd));
505 filter->fo_mount_count = mount_count + 1;
506 fsd->fsd_mount_count = cpu_to_le64(filter->fo_mount_count);
508 /* save it, so mount count and last_transno is current */
509 rc = filter_update_server_data(obd, filp, filter->fo_fsd, 1);
514 class_disconnect_exports(obd, 0);
516 filter_free_server_data(filter);
520 static int filter_cleanup_groups(struct obd_device *obd)
522 struct filter_obd *filter = &obd->u.filter;
524 struct dentry *dentry;
528 if (filter->fo_dentry_O_groups != NULL) {
529 for (i = 0; i < FILTER_GROUPS; i++) {
530 dentry = filter->fo_dentry_O_groups[i];
534 OBD_FREE(filter->fo_dentry_O_groups,
535 FILTER_GROUPS * sizeof(*filter->fo_dentry_O_groups));
536 filter->fo_dentry_O_groups = NULL;
538 if (filter->fo_last_objid_files != NULL) {
539 for (i = 0; i < FILTER_GROUPS; i++) {
540 filp = filter->fo_last_objid_files[i];
544 OBD_FREE(filter->fo_last_objid_files,
545 FILTER_GROUPS * sizeof(*filter->fo_last_objid_files));
546 filter->fo_last_objid_files = NULL;
548 if (filter->fo_dentry_O_sub != NULL) {
549 for (i = 0; i < filter->fo_subdir_count; i++) {
550 dentry = filter->fo_dentry_O_sub[i];
554 OBD_FREE(filter->fo_dentry_O_sub,
555 filter->fo_subdir_count *
556 sizeof(*filter->fo_dentry_O_sub));
557 filter->fo_dentry_O_sub = NULL;
559 if (filter->fo_last_objids != NULL) {
560 OBD_FREE(filter->fo_last_objids,
561 FILTER_GROUPS * sizeof(*filter->fo_last_objids));
562 filter->fo_last_objids = NULL;
564 if (filter->fo_dentry_O != NULL) {
565 f_dput(filter->fo_dentry_O);
566 filter->fo_dentry_O = NULL;
571 /* FIXME: object groups */
572 static int filter_prep_groups(struct obd_device *obd)
574 struct filter_obd *filter = &obd->u.filter;
575 struct dentry *dentry, *O_dentry;
577 int i, rc = 0, cleanup_phase = 0;
580 O_dentry = simple_mkdir(current->fs->pwd, "O", 0700, 1);
581 CDEBUG(D_INODE, "got/created O: %p\n", O_dentry);
582 if (IS_ERR(O_dentry)) {
583 rc = PTR_ERR(O_dentry);
584 CERROR("cannot open/create O: rc = %d\n", rc);
587 filter->fo_dentry_O = O_dentry;
588 cleanup_phase = 1; /* O_dentry */
590 /* Lookup "R" to tell if we're on an old OST FS and need to convert
591 * from O/R/<dir>/<objid> to O/0/<dir>/<objid>. This can be removed
592 * some time post 1.0 when all old-style OSTs have converted along
593 * with the init_objid hack. */
594 dentry = ll_lookup_one_len("R", O_dentry, 1);
596 GOTO(cleanup, rc = PTR_ERR(dentry));
597 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
598 struct dentry *O0_dentry = lookup_one_len("0", O_dentry, 1);
601 CWARN("converting OST to new object layout\n");
602 if (IS_ERR(O0_dentry)) {
603 rc = PTR_ERR(O0_dentry);
604 CERROR("error looking up O/0: rc %d\n", rc);
608 if (O0_dentry->d_inode) {
609 CERROR("Both O/R and O/0 exist. Fix manually.\n");
610 GOTO(cleanup_O0, rc = -EEXIST);
613 down(&O_dentry->d_inode->i_sem);
614 rc = vfs_rename(O_dentry->d_inode, dentry,
615 O_dentry->d_inode, O0_dentry);
616 up(&O_dentry->d_inode->i_sem);
619 CERROR("error renaming O/R to O/0: rc %d\n", rc);
620 GOTO(cleanup_O0, rc);
622 filter->fo_fsd->fsd_feature_incompat |=
623 cpu_to_le32(FILTER_INCOMPAT_GROUPS);
624 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
626 GOTO(cleanup_O0, rc);
638 OBD_ALLOC(filter->fo_last_objids, FILTER_GROUPS * sizeof(__u64));
639 if (filter->fo_last_objids == NULL)
640 GOTO(cleanup, rc = -ENOMEM);
641 cleanup_phase = 2; /* groups */
643 OBD_ALLOC(filter->fo_dentry_O_groups, FILTER_GROUPS * sizeof(dentry));
644 if (filter->fo_dentry_O_groups == NULL)
645 GOTO(cleanup, rc = -ENOMEM);
646 OBD_ALLOC(filter->fo_last_objid_files, FILTER_GROUPS * sizeof(filp));
647 if (filter->fo_last_objid_files == NULL)
648 GOTO(cleanup, rc = -ENOMEM);
650 for (i = 0; i < FILTER_GROUPS; i++) {
654 sprintf(name, "%d", i);
655 dentry = simple_mkdir(O_dentry, name, 0700, 1);
656 CDEBUG(D_INODE, "got/created O/%s: %p\n", name, dentry);
657 if (IS_ERR(dentry)) {
658 rc = PTR_ERR(dentry);
659 CERROR("cannot lookup/create O/%s: rc = %d\n",
663 filter->fo_dentry_O_groups[i] = dentry;
665 sprintf(name, "O/%d/LAST_ID", i);
666 filp = filp_open(name, O_CREAT | O_RDWR, 0700);
669 CERROR("cannot create %s: rc = %d\n", name, rc);
672 filter->fo_last_objid_files[i] = filp;
674 if (filp->f_dentry->d_inode->i_size == 0) {
675 if (i == 0 && filter->fo_fsd->fsd_unused != 0) {
676 /* OST conversion, remove sometime post 1.0 */
677 filter->fo_last_objids[i] =
678 le64_to_cpu(filter->fo_fsd->fsd_unused);
679 CWARN("saving old objid "LPU64" to LAST_ID\n",
680 filter->fo_last_objids[i]);
681 rc = filter_update_last_objid(obd, 0, 1);
685 filter->fo_last_objids[i] = FILTER_INIT_OBJID;
690 rc = fsfilt_read_record(obd, filp, &filter->fo_last_objids[i],
691 sizeof(__u64), &off);
693 CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
697 filter->fo_last_objids[i] =
698 le64_to_cpu(filter->fo_last_objids[i]);
699 CDEBUG(D_INODE, "%s: server last_objid group %d: "LPU64"\n",
700 obd->obd_name, i, filter->fo_last_objids[i]);
703 if (filter->fo_subdir_count) {
704 O_dentry = filter->fo_dentry_O_groups[0];
705 OBD_ALLOC(filter->fo_dentry_O_sub,
706 filter->fo_subdir_count * sizeof(dentry));
707 if (filter->fo_dentry_O_sub == NULL)
708 GOTO(cleanup, rc = -ENOMEM);
710 for (i = 0; i < filter->fo_subdir_count; i++) {
712 snprintf(dir, sizeof(dir), "d%u", i);
714 dentry = simple_mkdir(O_dentry, dir, 0700, 1);
715 CDEBUG(D_INODE, "got/created O/0/%s: %p\n", dir,dentry);
716 if (IS_ERR(dentry)) {
717 rc = PTR_ERR(dentry);
718 CERROR("can't lookup/create O/0/%s: rc = %d\n",
722 filter->fo_dentry_O_sub[i] = dentry;
728 filter_cleanup_groups(obd);
732 /* setup the object store with correct subdirectories */
733 static int filter_prep(struct obd_device *obd)
735 struct obd_run_ctxt saved;
736 struct filter_obd *filter = &obd->u.filter;
742 push_ctxt(&saved, &obd->obd_ctxt, NULL);
743 file = filp_open(LAST_RCVD, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
744 if (!file || IS_ERR(file)) {
746 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
751 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
752 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
753 file->f_dentry->d_inode->i_mode);
754 GOTO(err_filp, rc = -ENOENT);
757 /* steal operations */
758 inode = file->f_dentry->d_inode;
759 filter->fo_fop = file->f_op;
760 filter->fo_iop = inode->i_op;
761 filter->fo_aops = inode->i_mapping->a_ops;
763 rc = filter_init_server_data(obd, file);
765 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
768 filter->fo_rcvd_filp = file;
770 rc = filter_prep_groups(obd);
772 GOTO(err_server_data, rc);
775 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
780 //class_disconnect_exports(obd, 0);
781 filter_free_server_data(filter);
783 if (filp_close(file, 0))
784 CERROR("can't close %s after error\n", LAST_RCVD);
785 filter->fo_rcvd_filp = NULL;
789 /* cleanup the filter: write last used object id to status file */
790 static void filter_post(struct obd_device *obd)
792 struct obd_run_ctxt saved;
793 struct filter_obd *filter = &obd->u.filter;
796 /* XXX: filter_update_lastobjid used to call fsync_dev. It might be
797 * best to start a transaction with h_sync, because we removed this
800 push_ctxt(&saved, &obd->obd_ctxt, NULL);
801 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
804 CERROR("error writing server data: rc = %d\n", rc);
806 for (i = 0; i < FILTER_GROUPS; i++) {
807 rc = filter_update_last_objid(obd, i, (i == FILTER_GROUPS - 1));
809 CERROR("error writing group %d lastobjid: rc = %d\n",
813 rc = filp_close(filter->fo_rcvd_filp, 0);
814 filter->fo_rcvd_filp = NULL;
816 CERROR("error closing %s: rc = %d\n", LAST_RCVD, rc);
818 filter_cleanup_groups(obd);
819 filter_free_server_data(filter);
820 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
823 static void filter_set_last_id(struct filter_obd *filter, struct obdo *oa,
827 LASSERT(filter->fo_fsd != NULL);
830 LASSERT(oa->o_gr <= FILTER_GROUPS);
834 spin_lock(&filter->fo_objidlock);
835 filter->fo_last_objids[group] = id;
836 spin_unlock(&filter->fo_objidlock);
839 __u64 filter_last_id(struct filter_obd *filter, struct obdo *oa)
843 LASSERT(filter->fo_fsd != NULL);
846 LASSERT(oa->o_gr <= FILTER_GROUPS);
850 /* FIXME: object groups */
851 spin_lock(&filter->fo_objidlock);
852 id = filter->fo_last_objids[group];
853 spin_unlock(&filter->fo_objidlock);
858 /* direct cut-n-paste of mds_blocking_ast() */
859 static int filter_blocking_ast(struct ldlm_lock *lock,
860 struct ldlm_lock_desc *desc,
861 void *data, int flag)
866 if (flag == LDLM_CB_CANCELING) {
867 /* Don't need to do anything here. */
871 /* XXX layering violation! -phil */
872 l_lock(&lock->l_resource->lr_namespace->ns_lock);
873 /* Get this: if filter_blocking_ast is racing with ldlm_intent_policy,
874 * such that filter_blocking_ast is called just before l_i_p takes the
875 * ns_lock, then by the time we get the lock, we might not be the
876 * correct blocking function anymore. So check, and return early, if
878 if (lock->l_blocking_ast != filter_blocking_ast) {
879 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
883 lock->l_flags |= LDLM_FL_CBPENDING;
884 do_ast = (!lock->l_readers && !lock->l_writers);
885 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
888 struct lustre_handle lockh;
891 LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
892 ldlm_lock2handle(lock, &lockh);
893 rc = ldlm_cli_cancel(&lockh);
895 CERROR("ldlm_cli_cancel: %d\n", rc);
897 LDLM_DEBUG(lock, "Lock still has references, will be "
903 static int filter_lock_dentry(struct obd_device *obd, struct dentry *dparent)
905 down(&dparent->d_inode->i_sem);
909 /* We never dget the object parent, so DON'T dput it either */
910 static void filter_parent_unlock(struct dentry *dparent)
912 up(&dparent->d_inode->i_sem);
915 /* We never dget the object parent, so DON'T dput it either */
916 struct dentry *filter_parent(struct obd_device *obd, obd_gr group, obd_id objid)
918 struct filter_obd *filter = &obd->u.filter;
919 LASSERT(group < FILTER_GROUPS); /* FIXME: object groups */
921 if (group > 0 || filter->fo_subdir_count == 0)
922 return filter->fo_dentry_O_groups[group];
924 return filter->fo_dentry_O_sub[objid & (filter->fo_subdir_count - 1)];
927 /* We never dget the object parent, so DON'T dput it either */
928 struct dentry *filter_parent_lock(struct obd_device *obd, obd_gr group,
931 unsigned long now = jiffies;
932 struct dentry *dparent = filter_parent(obd, group, objid);
938 rc = filter_lock_dentry(obd, dparent);
939 if (time_after(jiffies, now + 15 * HZ))
940 CERROR("slow parent lock %lus\n", (jiffies - now) / HZ);
941 return rc ? ERR_PTR(rc) : dparent;
944 /* How to get files, dentries, inodes from object id's.
946 * If dir_dentry is passed, the caller has already locked the parent
947 * appropriately for this operation (normally a write lock). If
948 * dir_dentry is NULL, we do a read lock while we do the lookup to
949 * avoid races with create/destroy and such changing the directory
950 * internal to the filesystem code. */
951 struct dentry *filter_fid2dentry(struct obd_device *obd,
952 struct dentry *dir_dentry,
953 obd_gr group, obd_id id)
955 struct dentry *dparent = dir_dentry;
956 struct dentry *dchild;
962 CERROR("fatal: invalid object id 0\n");
963 RETURN(ERR_PTR(-ESTALE));
966 len = sprintf(name, LPU64, id);
967 if (dir_dentry == NULL) {
968 dparent = filter_parent_lock(obd, group, id);
972 CDEBUG(D_INODE, "looking up object O/%*s/%s\n",
973 dparent->d_name.len, dparent->d_name.name, name);
974 dchild = /*ll_*/lookup_one_len(name, dparent, len);
975 if (dir_dentry == NULL)
976 filter_parent_unlock(dparent);
977 if (IS_ERR(dchild)) {
978 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
982 CDEBUG(D_INODE, "got child objid %s: %p, count = %d\n",
983 name, dchild, atomic_read(&dchild->d_count));
985 LASSERT(atomic_read(&dchild->d_count) > 0);
990 static int filter_prepare_destroy(struct obd_device *obd, obd_id objid)
992 struct lustre_handle lockh;
993 int flags = LDLM_AST_DISCARD_DATA, rc;
994 struct ldlm_res_id res_id = { .name = { objid } };
995 ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
998 /* Tell the clients that the object is gone now and that they should
999 * throw away any cached pages. */
1000 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, res_id,
1001 LDLM_EXTENT, &policy, LCK_PW,
1002 &flags, filter_blocking_ast, ldlm_completion_ast,
1003 NULL, NULL, NULL, 0, NULL, &lockh);
1005 /* We only care about the side-effects, just drop the lock. */
1007 ldlm_lock_decref(&lockh, LCK_PW);
1012 /* Caller must hold LCK_PW on parent and push us into kernel context.
1013 * Caller is also required to ensure that dchild->d_inode exists. */
1014 static int filter_destroy_internal(struct obd_device *obd, obd_id objid,
1015 struct dentry *dparent,
1016 struct dentry *dchild)
1018 struct inode *inode = dchild->d_inode;
1022 if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) {
1023 CERROR("destroying objid %*s nlink = %lu, count = %d\n",
1024 dchild->d_name.len, dchild->d_name.name,
1025 (unsigned long)inode->i_nlink,
1026 atomic_read(&inode->i_count));
1029 rc = vfs_unlink(dparent->d_inode, dchild);
1032 CERROR("error unlinking objid %*s: rc %d\n",
1033 dchild->d_name.len, dchild->d_name.name, rc);
1038 static int filter_intent_policy(struct ldlm_namespace *ns,
1039 struct ldlm_lock **lockp, void *req_cookie,
1040 ldlm_mode_t mode, int flags, void *data)
1042 struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1043 struct ptlrpc_request *req = req_cookie;
1044 struct ldlm_lock *lock = *lockp, *l = NULL;
1045 struct ldlm_resource *res = lock->l_resource;
1046 ldlm_processing_policy policy;
1047 struct ost_lvb *res_lvb, *reply_lvb;
1048 struct list_head *tmp;
1050 int tmpflags = 0, rc, repsize[2] = {sizeof(struct ldlm_reply),
1051 sizeof(struct ost_lvb) };
1054 policy = ldlm_get_processing_policy(res);
1055 LASSERT(policy != NULL);
1056 LASSERT(req != NULL);
1058 rc = lustre_pack_reply(req, 2, repsize, NULL);
1060 RETURN(req->rq_status = rc);
1062 reply_lvb = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*reply_lvb));
1063 LASSERT(reply_lvb != NULL);
1065 //fixup_handle_for_resent_req(req, lock, &lockh);
1067 /* If we grant any lock at all, it will be a whole-file read lock.
1068 * Call the extent policy function to see if our request can be
1069 * granted, or is blocked. */
1070 lock->l_policy_data.l_extent.start = 0;
1071 lock->l_policy_data.l_extent.end = OBD_OBJECT_EOF;
1072 lock->l_req_mode = LCK_PR;
1074 l_lock(&res->lr_namespace->ns_lock);
1076 res->lr_tmp = &rpc_list;
1077 rc = policy(lock, &tmpflags, 0, &err);
1080 /* FIXME: we should change the policy function slightly, to not make
1081 * this list at all, since we just turn around and free it */
1082 while (!list_empty(&rpc_list)) {
1083 struct ldlm_ast_work *w =
1084 list_entry(rpc_list.next, struct ldlm_ast_work, w_list);
1085 list_del(&w->w_list);
1086 LDLM_LOCK_PUT(w->w_lock);
1087 OBD_FREE(w, sizeof(*w));
1090 if (rc == LDLM_ITER_CONTINUE) {
1091 /* The lock met with no resistance; we're finished. */
1092 l_unlock(&res->lr_namespace->ns_lock);
1093 RETURN(ELDLM_LOCK_REPLACED);
1096 /* Do not grant any lock, but instead send GL callbacks. The extent
1097 * policy nicely created a list of all PW locks for us. We will choose
1098 * the highest of those which are larger than the size in the LVB, if
1099 * any, and perform a glimpse callback. */
1100 down(&res->lr_lvb_sem);
1101 res_lvb = res->lr_lvb_data;
1102 LASSERT(res_lvb != NULL);
1103 reply_lvb->lvb_size = res_lvb->lvb_size;
1104 reply_lvb->lvb_blocks = res_lvb->lvb_blocks;
1105 up(&res->lr_lvb_sem);
1107 list_for_each(tmp, &res->lr_granted) {
1108 struct ldlm_lock *tmplock =
1109 list_entry(tmp, struct ldlm_lock, l_res_link);
1111 if (tmplock->l_granted_mode == LCK_PR)
1114 if (tmplock->l_policy_data.l_extent.end <= reply_lvb->lvb_size)
1118 l = LDLM_LOCK_GET(tmplock);
1122 if (l->l_policy_data.l_extent.start >
1123 tmplock->l_policy_data.l_extent.start)
1127 l = LDLM_LOCK_GET(tmplock);
1129 l_unlock(&res->lr_namespace->ns_lock);
1131 /* There were no PW locks beyond the size in the LVB; finished. */
1133 RETURN(ELDLM_LOCK_ABORTED);
1135 LASSERT(l->l_glimpse_ast != NULL);
1136 rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
1137 if (rc != 0 && res->lr_namespace->ns_lvbo &&
1138 res->lr_namespace->ns_lvbo->lvbo_update) {
1139 res->lr_namespace->ns_lvbo->lvbo_update(res, NULL, 0, 1);
1142 down(&res->lr_lvb_sem);
1143 reply_lvb->lvb_size = res_lvb->lvb_size;
1144 reply_lvb->lvb_blocks = res_lvb->lvb_blocks;
1145 up(&res->lr_lvb_sem);
1149 RETURN(ELDLM_LOCK_ABORTED);
1152 /* mount the file system (secretly) */
1153 int filter_common_setup(struct obd_device *obd, obd_count len, void *buf,
1156 struct lustre_cfg* lcfg = buf;
1157 struct filter_obd *filter = &obd->u.filter;
1158 struct vfsmount *mnt;
1162 dev_clear_rdonly(2);
1164 if (!lcfg->lcfg_inlbuf1 || !lcfg->lcfg_inlbuf2)
1167 obd->obd_fsops = fsfilt_get_ops(lcfg->lcfg_inlbuf2);
1168 if (IS_ERR(obd->obd_fsops))
1169 RETURN(PTR_ERR(obd->obd_fsops));
1171 mnt = do_kern_mount(lcfg->lcfg_inlbuf2, MS_NOATIME | MS_NODIRATIME,
1172 lcfg->lcfg_inlbuf1, option);
1177 if (lcfg->lcfg_inllen3 > 0 && lcfg->lcfg_inlbuf3) {
1178 if (*lcfg->lcfg_inlbuf3 == 'f') {
1179 obd->obd_replayable = 1;
1180 obd_sync_filter = 1;
1181 CWARN("%s: recovery enabled\n", obd->obd_name);
1183 if (*lcfg->lcfg_inlbuf3 != 'n') {
1184 CERROR("unrecognised flag '%c'\n",
1185 *lcfg->lcfg_inlbuf3);
1187 // XXX Robert? Why do we get errors here
1188 // GOTO(err_mntput, rc = -EINVAL);
1192 filter->fo_vfsmnt = mnt;
1193 filter->fo_sb = mnt->mnt_sb;
1194 filter->fo_fstype = mnt->mnt_sb->s_type->name;
1195 CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
1197 OBD_SET_CTXT_MAGIC(&obd->obd_ctxt);
1198 obd->obd_ctxt.pwdmnt = mnt;
1199 obd->obd_ctxt.pwd = mnt->mnt_root;
1200 obd->obd_ctxt.fs = get_ds();
1201 obd->obd_ctxt.cb_ops = filter_lvfs_ops;
1203 rc = filter_prep(obd);
1205 GOTO(err_mntput, rc);
1207 spin_lock_init(&filter->fo_translock);
1208 spin_lock_init(&filter->fo_objidlock);
1209 INIT_LIST_HEAD(&filter->fo_export_list);
1210 sema_init(&filter->fo_alloc_lock, 1);
1211 spin_lock_init(&filter->fo_r_pages.oh_lock);
1212 spin_lock_init(&filter->fo_w_pages.oh_lock);
1213 spin_lock_init(&filter->fo_r_discont_pages.oh_lock);
1214 spin_lock_init(&filter->fo_w_discont_pages.oh_lock);
1215 spin_lock_init(&filter->fo_r_discont_blocks.oh_lock);
1216 spin_lock_init(&filter->fo_w_discont_blocks.oh_lock);
1217 filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE;
1219 obd->obd_namespace = ldlm_namespace_new("filter-tgt",
1220 LDLM_NAMESPACE_SERVER);
1221 if (obd->obd_namespace == NULL)
1222 GOTO(err_post, rc = -ENOMEM);
1223 obd->obd_namespace->ns_lvbp = obd;
1224 obd->obd_namespace->ns_lvbo = &filter_lvbo;
1225 ldlm_register_intent(obd->obd_namespace, filter_intent_policy);
1227 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1228 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
1230 rc = llog_cat_initialize(obd, 1);
1232 CERROR("failed to setup llogging subsystems\n");
1246 fsfilt_put_ops(obd->obd_fsops);
1250 static int filter_setup(struct obd_device *obd, obd_count len, void *buf)
1252 struct lprocfs_static_vars lvars;
1253 struct lustre_cfg* lcfg = buf;
1254 const char *str = NULL;
1255 char *option = NULL;
1259 if (!lcfg->lcfg_inlbuf1 || !lcfg->lcfg_inlbuf2)
1262 if (!strcmp(lcfg->lcfg_inlbuf2, "ext3") ||
1263 !strcmp(lcfg->lcfg_inlbuf2, "ldiskfs")) {
1264 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1265 /* bug 1577: implement async-delete for 2.5 */
1266 str = "errors=remount-ro,asyncdel";
1268 str = "errors=remount-ro";
1270 n = strlen(str) + 1;
1271 OBD_ALLOC(option, n);
1274 strcpy(option, str);
1277 rc = filter_common_setup(obd, len, buf, option);
1279 OBD_FREE(option, n);
1281 lprocfs_init_vars(filter, &lvars);
1282 if (rc == 0 && lprocfs_obd_setup(obd, lvars.obd_vars) == 0 &&
1283 lprocfs_alloc_obd_stats(obd, LPROC_FILTER_LAST) == 0) {
1284 /* Init obdfilter private stats here */
1285 lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_READ_BYTES,
1286 LPROCFS_CNTR_AVGMINMAX,
1287 "read_bytes", "bytes");
1288 lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
1289 LPROCFS_CNTR_AVGMINMAX,
1290 "write_bytes", "bytes");
1292 lproc_filter_attach_seqstat(obd);
1298 static int filter_cleanup(struct obd_device *obd, int flags)
1300 struct filter_obd *filter = &obd->u.filter;
1303 if (flags & OBD_OPT_FAILOVER)
1304 CERROR("%s: shutting down for failover; client state will"
1305 " be preserved.\n", obd->obd_name);
1307 if (!list_empty(&obd->obd_exports)) {
1308 CERROR("%s: still has clients!\n", obd->obd_name);
1309 class_disconnect_exports(obd, flags);
1310 if (!list_empty(&obd->obd_exports)) {
1311 CERROR("still has exports after forced cleanup?\n");
1316 ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
1318 if (filter->fo_sb == NULL)
1321 lprocfs_free_obd_stats(obd);
1322 lprocfs_obd_cleanup(obd);
1326 shrink_dcache_parent(filter->fo_sb->s_root);
1329 if (atomic_read(&filter->fo_vfsmnt->mnt_count) > 1)
1330 CERROR("%s: mount point %p busy, mnt_count: %d\n",
1331 obd->obd_name, filter->fo_vfsmnt,
1332 atomic_read(&filter->fo_vfsmnt->mnt_count));
1335 mntput(filter->fo_vfsmnt);
1336 //destroy_buffers(filter->fo_sb->s_dev);
1337 filter->fo_sb = NULL;
1338 fsfilt_put_ops(obd->obd_fsops);
1341 dev_clear_rdonly(2);
1346 /* nearly identical to mds_connect */
1347 static int filter_connect(struct lustre_handle *conn, struct obd_device *obd,
1348 struct obd_uuid *cluuid)
1350 struct obd_export *exp;
1351 struct filter_export_data *fed;
1352 struct filter_client_data *fcd = NULL;
1353 struct filter_obd *filter = &obd->u.filter;
1357 if (conn == NULL || obd == NULL || cluuid == NULL)
1360 rc = class_connect(conn, obd, cluuid);
1363 exp = class_conn2export(conn);
1364 LASSERT(exp != NULL);
1366 fed = &exp->exp_filter_data;
1368 spin_lock_init(&fed->fed_lock);
1370 if (!obd->obd_replayable)
1371 GOTO(cleanup, rc = 0);
1373 OBD_ALLOC(fcd, sizeof(*fcd));
1375 CERROR("filter: out of memory for client data\n");
1376 GOTO(cleanup, rc = -ENOMEM);
1379 memcpy(fcd->fcd_uuid, cluuid, sizeof(fcd->fcd_uuid));
1382 rc = filter_client_add(obd, filter, fed, -1);
1387 OBD_FREE(fcd, sizeof(*fcd));
1388 class_disconnect(exp, 0);
1390 class_export_put(exp);
1395 static int filter_precleanup(struct obd_device *obd, int flags)
1400 rc = obd_llog_finish(obd, 0);
1402 CERROR("failed to cleanup llogging subsystem\n");
1407 /* Do extra sanity checks for grant accounting. We do this at connect,
1408 * disconnect, and statfs RPC time, so it shouldn't be too bad. We can
1409 * always get rid of it or turn it off when we know accounting is good. */
1410 static void filter_grant_sanity_check(struct obd_device *obd, const char *func)
1412 struct filter_export_data *fed;
1413 struct obd_export *exp;
1414 obd_size maxsize = obd->obd_osfs.os_blocks * obd->obd_osfs.os_bsize;
1415 obd_size tot_dirty = 0, tot_pending = 0, tot_granted = 0;
1416 obd_size fo_tot_dirty, fo_tot_pending, fo_tot_granted;
1418 if (list_empty(&obd->obd_exports))
1421 spin_lock(&obd->obd_osfs_lock);
1422 spin_lock(&obd->obd_dev_lock);
1423 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1424 fed = &exp->exp_filter_data;
1425 LASSERTF(fed->fed_grant + fed->fed_pending <= maxsize,
1426 "cli %s/%p %lu+%lu > "LPU64"\n",
1427 exp->exp_client_uuid.uuid, exp,
1428 fed->fed_grant, fed->fed_pending, maxsize);
1429 LASSERTF(fed->fed_dirty <= maxsize, "cli %s/%p %lu > "LPU64"\n",
1430 exp->exp_client_uuid.uuid, exp,fed->fed_dirty,maxsize);
1431 CDEBUG(D_CACHE,"%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1432 obd->obd_name, exp->exp_client_uuid.uuid, exp,
1433 fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1434 tot_granted += fed->fed_grant + fed->fed_pending;
1435 tot_pending += fed->fed_pending;
1436 tot_dirty += fed->fed_dirty;
1438 fo_tot_granted = obd->u.filter.fo_tot_granted;
1439 fo_tot_pending = obd->u.filter.fo_tot_pending;
1440 fo_tot_dirty = obd->u.filter.fo_tot_dirty;
1441 spin_unlock(&obd->obd_dev_lock);
1442 spin_unlock(&obd->obd_osfs_lock);
1444 /* Do these assertions outside the spinlocks so we don't kill system */
1445 if (tot_granted != fo_tot_granted)
1446 CERROR("%s: tot_granted "LPU64" != fo_tot_granted "LPU64"\n",
1447 func, tot_granted, fo_tot_granted);
1448 if (tot_pending != fo_tot_pending)
1449 CERROR("%s: tot_pending "LPU64" != fo_tot_pending "LPU64"\n",
1450 func, tot_pending, fo_tot_pending);
1451 if (tot_dirty != fo_tot_dirty)
1452 CERROR("%s: tot_dirty "LPU64" != fo_tot_dirty "LPU64"\n",
1453 func, tot_dirty, fo_tot_dirty);
1454 if (tot_pending > tot_granted)
1455 CERROR("%s: tot_pending "LPU64" > tot_granted "LPU64"\n",
1456 func, tot_pending, tot_granted);
1457 if (tot_granted > maxsize)
1458 CERROR("%s: tot_granted "LPU64" > maxsize "LPU64"\n",
1459 func, tot_granted, maxsize);
1460 if (tot_dirty > maxsize)
1461 CERROR("%s: tot_dirty "LPU64" > maxsize "LPU64"\n",
1462 func, tot_dirty, maxsize);
1465 /* Remove this client from the grant accounting totals. We also remove
1466 * the export from the obd device under the osfs and dev locks to ensure
1467 * that the filter_grant_sanity_check() calculations are always valid.
1468 * The client should do something similar when it invalidates its import. */
1469 static void filter_grant_discard(struct obd_export *exp)
1471 struct obd_device *obd = exp->exp_obd;
1472 struct filter_obd *filter = &obd->u.filter;
1473 struct filter_export_data *fed = &exp->exp_filter_data;
1475 spin_lock(&obd->obd_osfs_lock);
1476 spin_lock(&exp->exp_obd->obd_dev_lock);
1477 list_del_init(&exp->exp_obd_chain);
1478 spin_unlock(&exp->exp_obd->obd_dev_lock);
1480 CDEBUG(D_CACHE, "%s: cli %s/%p dirty %lu pend %lu grant %lu\n",
1481 obd->obd_name, exp->exp_client_uuid.uuid, exp,
1482 fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1484 LASSERTF(filter->fo_tot_granted >= fed->fed_grant,
1485 "%s: tot_granted "LPU64" cli %s/%p fed_grant %lu\n",
1486 obd->obd_name, filter->fo_tot_granted,
1487 exp->exp_client_uuid.uuid, exp, fed->fed_grant);
1488 filter->fo_tot_granted -= fed->fed_grant;
1489 LASSERTF(exp->exp_obd->u.filter.fo_tot_pending >= fed->fed_pending,
1490 "%s: tot_pending "LPU64" cli %s/%p fed_pending %lu\n",
1491 obd->obd_name, filter->fo_tot_pending,
1492 exp->exp_client_uuid.uuid, exp, fed->fed_pending);
1493 LASSERTF(filter->fo_tot_dirty >= fed->fed_dirty,
1494 "%s: tot_dirty "LPU64" cli %s/%p fed_dirty %lu\n",
1495 obd->obd_name, filter->fo_tot_dirty,
1496 exp->exp_client_uuid.uuid, exp, fed->fed_dirty);
1497 filter->fo_tot_dirty -= fed->fed_dirty;
1501 spin_unlock(&obd->obd_osfs_lock);
1504 static int filter_destroy_export(struct obd_export *exp)
1508 if (exp->exp_filter_data.fed_pending)
1509 CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
1510 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
1511 exp, exp->exp_filter_data.fed_pending);
1513 target_destroy_export(exp);
1515 if (exp->exp_obd->obd_replayable)
1516 filter_client_free(exp, exp->exp_flags);
1518 filter_grant_discard(exp);
1519 if (!(exp->exp_flags & OBD_OPT_FORCE))
1520 filter_grant_sanity_check(exp->exp_obd, __FUNCTION__);
1525 /* also incredibly similar to mds_disconnect */
1526 static int filter_disconnect(struct obd_export *exp, int flags)
1528 struct obd_device *obd = exp->exp_obd;
1529 unsigned long irqflags;
1530 struct llog_ctxt *ctxt;
1535 class_export_get(exp);
1537 spin_lock_irqsave(&exp->exp_lock, irqflags);
1538 exp->exp_flags = flags;
1539 spin_unlock_irqrestore(&exp->exp_lock, irqflags);
1541 if (!(flags & OBD_OPT_FORCE))
1542 filter_grant_sanity_check(obd, __FUNCTION__);
1543 filter_grant_discard(exp);
1545 /* Disconnect early so that clients can't keep using export */
1546 rc = class_disconnect(exp, flags);
1548 ldlm_cancel_locks_for_export(exp);
1550 fsfilt_sync(obd, obd->u.filter.fo_sb);
1552 /* flush any remaining cancel messages out to the target */
1553 ctxt = llog_get_context(obd, LLOG_UNLINK_REPL_CTXT);
1554 llog_sync(ctxt, exp);
1556 class_export_put(exp);
1560 struct dentry *__filter_oa2dentry(struct obd_device *obd,
1561 struct obdo *oa, const char *what)
1563 struct dentry *dchild = NULL;
1566 if (oa->o_valid & OBD_MD_FLGROUP)
1569 dchild = filter_fid2dentry(obd, NULL, group, oa->o_id);
1571 if (IS_ERR(dchild)) {
1572 CERROR("%s error looking up object: "LPU64"\n", what, oa->o_id);
1576 if (dchild->d_inode == NULL) {
1577 CERROR("%s on non-existent object: "LPU64"\n", what, oa->o_id);
1579 RETURN(ERR_PTR(-ENOENT));
1585 static int filter_getattr(struct obd_export *exp, struct obdo *oa,
1586 struct lov_stripe_md *md)
1588 struct dentry *dentry = NULL;
1589 struct obd_device *obd;
1593 obd = class_exp2obd(exp);
1595 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1596 exp->exp_handle.h_cookie);
1600 dentry = filter_oa2dentry(obd, oa);
1602 RETURN(PTR_ERR(dentry));
1604 /* Limit the valid bits in the return data to what we actually use */
1605 oa->o_valid = OBD_MD_FLID;
1606 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
1612 /* this is called from filter_truncate() until we have filter_punch() */
1613 static int filter_setattr(struct obd_export *exp, struct obdo *oa,
1614 struct lov_stripe_md *md, struct obd_trans_info *oti)
1616 struct obd_run_ctxt saved;
1617 struct filter_obd *filter;
1618 struct dentry *dentry;
1620 struct ldlm_res_id res_id = { .name = { oa->o_id } };
1621 struct ldlm_resource *res;
1626 LASSERT(oti != NULL);
1628 dentry = filter_oa2dentry(exp->exp_obd, oa);
1630 RETURN(PTR_ERR(dentry));
1632 filter = &exp->exp_obd->u.filter;
1634 iattr_from_obdo(&iattr, oa, oa->o_valid);
1636 push_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
1639 if (iattr.ia_valid & ATTR_SIZE)
1640 down(&dentry->d_inode->i_sem);
1641 handle = fsfilt_start(exp->exp_obd, dentry->d_inode, FSFILT_OP_SETATTR,
1644 GOTO(out_unlock, rc = PTR_ERR(handle));
1646 /* XXX this could be a rwsem instead, if filter_preprw played along */
1647 if (iattr.ia_valid & ATTR_ATTR_FLAG)
1648 rc = fsfilt_iocontrol(exp->exp_obd, dentry->d_inode, NULL,
1650 (long)&iattr.ia_attr_flags);
1652 rc = fsfilt_setattr(exp->exp_obd, dentry, handle, &iattr, 1);
1653 rc = filter_finish_transno(exp, oti, rc);
1654 rc2 = fsfilt_commit(exp->exp_obd, dentry->d_inode, handle, 0);
1656 CERROR("error on commit, err = %d\n", rc2);
1661 if (iattr.ia_valid & ATTR_SIZE) {
1662 res = ldlm_resource_get(exp->exp_obd->obd_namespace, NULL,
1663 res_id, LDLM_EXTENT, 0);
1665 CERROR("!!! resource_get failed for object "LPU64" -- "
1666 "filter_setattr with no lock?\n", oa->o_id);
1668 if (res->lr_namespace->ns_lvbo &&
1669 res->lr_namespace->ns_lvbo->lvbo_update) {
1670 rc = res->lr_namespace->ns_lvbo->lvbo_update
1673 ldlm_resource_putref(res);
1677 oa->o_valid = OBD_MD_FLID;
1678 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
1681 if (iattr.ia_valid & ATTR_SIZE)
1682 up(&dentry->d_inode->i_sem);
1684 pop_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
1690 /* XXX identical to osc_unpackmd */
1691 static int filter_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
1692 struct lov_mds_md *lmm, int lmm_bytes)
1698 if (lmm_bytes < sizeof (*lmm)) {
1699 CERROR("lov_mds_md too small: %d, need %d\n",
1700 lmm_bytes, (int)sizeof(*lmm));
1703 /* XXX LOV_MAGIC etc check? */
1705 if (lmm->lmm_object_id == cpu_to_le64(0)) {
1706 CERROR("lov_mds_md: zero lmm_object_id\n");
1711 lsm_size = lov_stripe_md_size(1);
1715 if (*lsmp != NULL && lmm == NULL) {
1716 OBD_FREE(*lsmp, lsm_size);
1721 if (*lsmp == NULL) {
1722 OBD_ALLOC(*lsmp, lsm_size);
1726 loi_init((*lsmp)->lsm_oinfo);
1730 /* XXX zero *lsmp? */
1731 (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
1732 LASSERT((*lsmp)->lsm_object_id);
1735 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
1740 static void filter_destroy_precreated(struct obd_export *exp, struct obdo *oa,
1741 struct filter_obd *filter)
1743 struct obdo doa; /* XXX obdo on stack */
1748 memset(&doa, 0, sizeof(doa));
1749 if (oa->o_valid & OBD_MD_FLGROUP) {
1750 doa.o_valid |= OBD_MD_FLGROUP;
1751 doa.o_gr = oa->o_gr;
1755 doa.o_mode = S_IFREG;
1757 last = filter_last_id(filter, &doa);
1758 CWARN("deleting orphan objects from "LPU64" to "LPU64"\n",
1759 oa->o_id + 1, last);
1760 for (id = oa->o_id + 1; id <= last; id++) {
1762 filter_destroy(exp, &doa, NULL, NULL);
1764 spin_lock(&filter->fo_objidlock);
1765 filter->fo_last_objids[doa.o_gr] = oa->o_id;
1766 spin_unlock(&filter->fo_objidlock);
1770 /* returns a negative error or a nonnegative number of files to create */
1771 static int filter_should_precreate(struct obd_export *exp, struct obdo *oa,
1774 struct obd_device *obd = exp->exp_obd;
1775 struct filter_obd *filter = &obd->u.filter;
1779 diff = oa->o_id - filter_last_id(filter, oa);
1780 CDEBUG(D_INFO, "filter_last_id() = "LPU64" -> diff = %d\n",
1781 filter_last_id(filter, oa), diff);
1783 /* delete orphans request */
1784 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1785 (oa->o_flags & OBD_FL_DELORPHAN)) {
1788 if (-diff > 10000) { /* XXX make this smarter */
1789 CERROR("ignoring bogus orphan destroy request: obdid "
1790 LPU64" last_id "LPU64"\n",
1791 oa->o_id, filter_last_id(filter, oa));
1794 filter_destroy_precreated(exp, oa, filter);
1795 rc = filter_update_last_objid(obd, group, 0);
1797 CERROR("unable to write lastobjid, but orphans"
1801 /* only precreate if group == 0 and o_id is specfied */
1802 if (!(oa->o_valid & OBD_FL_DELORPHAN) &&
1803 (group != 0 || oa->o_id == 0))
1811 /* We rely on the fact that only one thread will be creating files in a given
1812 * group at a time, which is why we don't need an atomic filter_get_new_id.
1813 * Even if we had that atomic function, the following race would exist:
1815 * thread 1: gets id x from filter_next_id
1816 * thread 2: gets id (x + 1) from filter_next_id
1817 * thread 2: creates object (x + 1)
1818 * thread 1: tries to create object x, gets -ENOSPC
1820 static int filter_precreate(struct obd_device *obd, struct obdo *oa,
1821 obd_gr group, int *num)
1823 struct dentry *dchild = NULL;
1824 struct filter_obd *filter;
1825 struct dentry *dparent;
1826 int err = 0, rc = 0, i;
1828 int recreate_obj = 0;
1829 void *handle = NULL;
1832 filter = &obd->u.filter;
1834 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1835 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1839 for (i = 0; i < *num && err == 0; i++) {
1840 int cleanup_phase = 0;
1845 last_id = filter_last_id(filter, oa);
1846 if (next_id > last_id) {
1847 CERROR("Error: Trying to recreate obj greater"
1848 "than last id "LPD64" > "LPD64"\n",
1853 next_id = filter_last_id(filter, oa) + 1;
1855 CDEBUG(D_INFO, "precreate objid "LPU64"\n", next_id);
1857 dparent = filter_parent_lock(obd, group, next_id);
1858 if (IS_ERR(dparent))
1859 GOTO(cleanup, rc = PTR_ERR(dparent));
1862 dchild = filter_fid2dentry(obd, dparent, group, next_id);
1864 GOTO(cleanup, rc = PTR_ERR(dchild));
1867 if (dchild->d_inode != NULL) {
1868 /* This would only happen if lastobjid was bad on disk*/
1869 /* Could also happen if recreating missing obj but
1873 CERROR("Serious error: recreating obj %*s but "
1874 "obj already exists \n",
1875 dchild->d_name.len, dchild->d_name.name);
1877 CERROR("Serious error: objid %*s already "
1878 "exists; is this filesystem corrupt?\n",
1879 dchild->d_name.len, dchild->d_name.name);
1881 GOTO(cleanup, rc = -EEXIST);
1884 handle = fsfilt_start_log(obd, dparent->d_inode,
1885 FSFILT_OP_CREATE, NULL, 1);
1887 GOTO(cleanup, rc = PTR_ERR(handle));
1890 rc = ll_vfs_create(dparent->d_inode, dchild, S_IFREG, NULL);
1892 CERROR("create failed rc = %d\n", rc);
1896 if (!recreate_obj) {
1897 filter_set_last_id(filter, oa, next_id);
1898 err = filter_update_last_objid(obd, group, 0);
1900 CERROR("unable to write lastobjid "
1901 "but file created\n");
1905 switch(cleanup_phase) {
1907 err = fsfilt_commit(obd, dparent->d_inode, handle, 0);
1909 CERROR("error on commit, err = %d\n", err);
1916 filter_parent_unlock(dparent);
1926 CDEBUG(D_INFO, "filter_precreate() created %d objects\n", i);
1930 static int filter_create(struct obd_export *exp, struct obdo *oa,
1931 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1933 struct obd_device *obd = NULL;
1934 struct obd_run_ctxt saved;
1935 struct lov_stripe_md *lsm = NULL;
1940 if (oa->o_valid & OBD_MD_FLGROUP)
1943 CDEBUG(D_INFO, "filter_create(od->o_gr="LPU64",od->o_id="LPU64")\n",
1948 rc = obd_alloc_memmd(exp, &lsm);
1955 push_ctxt(&saved, &obd->obd_ctxt, NULL);
1957 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
1958 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1959 if (oa->o_id > filter_last_id(&obd->u.filter, oa)) {
1960 CERROR("recreate objid "LPU64" > last id "LPU64"\n",
1961 oa->o_id, filter_last_id(&obd->u.filter, oa));
1965 rc = filter_precreate(obd, oa, group, &diff);
1968 diff = filter_should_precreate(exp, oa, group);
1970 oa->o_id = filter_last_id(&obd->u.filter, oa);
1971 rc = filter_precreate(obd, oa, group, &diff);
1973 oa->o_valid = OBD_MD_FLID;
1977 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
1978 if (rc && ea != NULL && *ea != lsm) {
1979 obd_free_memmd(exp, &lsm);
1980 } else if (rc == 0 && ea != NULL) {
1981 /* XXX LOV STACKING: the lsm that is passed to us from
1982 * LOV does not have valid lsm_oinfo data structs, so
1983 * don't go touching that. This needs to be fixed in a
1985 lsm->lsm_object_id = oa->o_id;
1992 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
1993 struct lov_stripe_md *ea, struct obd_trans_info *oti)
1995 struct obd_device *obd;
1996 struct filter_obd *filter;
1997 struct dentry *dchild = NULL, *dparent = NULL;
1998 struct obd_run_ctxt saved;
1999 void *handle = NULL;
2000 struct llog_cookie *fcc = NULL;
2001 int rc, rc2, cleanup_phase = 0, have_prepared = 0;
2005 if (oa->o_valid & OBD_MD_FLGROUP)
2009 filter = &obd->u.filter;
2011 push_ctxt(&saved, &obd->obd_ctxt, NULL);
2014 dparent = filter_parent_lock(obd, group, oa->o_id);
2015 if (IS_ERR(dparent))
2016 GOTO(cleanup, rc = PTR_ERR(dparent));
2019 dchild = filter_fid2dentry(obd, dparent, group, oa->o_id);
2021 GOTO(cleanup, rc = -ENOENT);
2024 if (dchild->d_inode == NULL) {
2025 CDEBUG(D_INODE, "destroying non-existent object "LPU64"\n",
2027 GOTO(cleanup, rc = -ENOENT);
2030 if (!have_prepared) {
2031 /* If we're really going to destroy the object, get ready
2032 * by getting the clients to discard their cached data.
2034 * We have to drop the parent lock, because
2035 * filter_prepare_destroy will acquire a PW on the object, and
2036 * we don't want to deadlock with an incoming write to the
2037 * object, which has the extent PW and then wants to get the
2038 * parent dentry to do the lookup.
2040 * We dput the child because it's not worth the extra
2041 * complication of condition the above code to skip it on the
2042 * second time through. */
2044 filter_parent_unlock(dparent);
2046 filter_prepare_destroy(obd, oa->o_id);
2051 handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_UNLINK,oti,1);
2053 GOTO(cleanup, rc = PTR_ERR(handle));
2056 /* Our MDC connection is established by the MDS to us */
2057 if (oa->o_valid & OBD_MD_FLCOOKIE) {
2058 OBD_ALLOC(fcc, sizeof(*fcc));
2060 memcpy(fcc, obdo_logcookie(oa), sizeof(*fcc));
2063 rc = filter_destroy_internal(obd, oa->o_id, dparent, dchild);
2066 switch(cleanup_phase) {
2070 fsfilt_add_journal_cb(obd, 0, oti->oti_handle,
2071 filter_cancel_cookies_cb,
2074 fsfilt_add_journal_cb(obd, 0, handle,
2075 filter_cancel_cookies_cb,
2078 rc = filter_finish_transno(exp, oti, rc);
2079 rc2 = fsfilt_commit(obd, dparent->d_inode, handle, 0);
2081 CERROR("error on commit, err = %d\n", rc2);
2088 filter_parent_unlock(dparent);
2090 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
2093 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
2100 /* NB start and end are used for punch, but not truncate */
2101 static int filter_truncate(struct obd_export *exp, struct obdo *oa,
2102 struct lov_stripe_md *lsm,
2103 obd_off start, obd_off end,
2104 struct obd_trans_info *oti)
2109 if (end != OBD_OBJECT_EOF)
2110 CERROR("PUNCH not supported, only truncate: end = "LPX64"\n",
2113 CDEBUG(D_INODE, "calling truncate for object "LPU64", valid = %x, "
2114 "o_size = "LPD64"\n", oa->o_id, oa->o_valid, start);
2116 error = filter_setattr(exp, oa, NULL, oti);
2120 static int filter_sync(struct obd_export *exp, struct obdo *oa,
2121 struct lov_stripe_md *lsm, obd_off start, obd_off end)
2123 struct obd_run_ctxt saved;
2124 struct filter_obd *filter;
2125 struct dentry *dentry;
2126 struct llog_ctxt *ctxt;
2130 filter = &exp->exp_obd->u.filter;
2132 /* an objid of zero is taken to mean "sync whole filesystem" */
2133 if (!oa || !(oa->o_valid & OBD_MD_FLID)) {
2134 rc = fsfilt_sync(exp->exp_obd, filter->fo_sb);
2135 /* flush any remaining cancel messages out to the target */
2136 ctxt = llog_get_context(exp->exp_obd, LLOG_UNLINK_REPL_CTXT);
2137 llog_sync(ctxt, exp);
2141 dentry = filter_oa2dentry(exp->exp_obd, oa);
2143 RETURN(PTR_ERR(dentry));
2145 push_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
2147 down(&dentry->d_inode->i_sem);
2148 rc = filemap_fdatasync(dentry->d_inode->i_mapping);
2150 /* just any file to grab fsync method - "file" arg unused */
2151 struct file *file = filter->fo_rcvd_filp;
2153 if (file->f_op && file->f_op->fsync)
2154 rc = file->f_op->fsync(NULL, dentry, 1);
2156 rc2 = filemap_fdatawait(dentry->d_inode->i_mapping);
2160 up(&dentry->d_inode->i_sem);
2162 oa->o_valid = OBD_MD_FLID;
2163 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2165 pop_ctxt(&saved, &exp->exp_obd->obd_ctxt, NULL);
2171 static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2172 unsigned long max_age)
2174 struct filter_obd *filter = &obd->u.filter;
2175 int blockbits = filter->fo_sb->s_blocksize_bits;
2179 /* at least try to account for cached pages. its still racey and
2180 * might be under-reporting if clients haven't announced their
2181 * caches with brw recently */
2182 spin_lock(&obd->obd_osfs_lock);
2183 rc = fsfilt_statfs(obd, filter->fo_sb, max_age);
2184 memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
2185 spin_unlock(&obd->obd_osfs_lock);
2187 CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
2188 " pending "LPU64" free "LPU64" avail "LPU64"\n",
2189 filter->fo_tot_dirty, filter->fo_tot_granted,
2190 filter->fo_tot_pending,
2191 osfs->os_bfree << blockbits, osfs->os_bavail << blockbits);
2193 filter_grant_sanity_check(obd, __FUNCTION__);
2195 osfs->os_bavail -= min(osfs->os_bavail,
2196 (filter->fo_tot_dirty + filter->fo_tot_pending +
2197 osfs->os_bsize -1) >> blockbits);
2202 static int filter_get_info(struct obd_export *exp, __u32 keylen,
2203 void *key, __u32 *vallen, void *val)
2205 struct obd_device *obd;
2208 obd = class_exp2obd(exp);
2210 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2211 exp->exp_handle.h_cookie);
2215 if (keylen == strlen("blocksize") &&
2216 memcmp(key, "blocksize", keylen) == 0) {
2217 __u32 *blocksize = val;
2218 *vallen = sizeof(*blocksize);
2219 *blocksize = obd->u.filter.fo_sb->s_blocksize;
2223 if (keylen == strlen("blocksize_bits") &&
2224 memcmp(key, "blocksize_bits", keylen) == 0) {
2225 __u32 *blocksize_bits = val;
2226 *vallen = sizeof(*blocksize_bits);
2227 *blocksize_bits = obd->u.filter.fo_sb->s_blocksize_bits;
2231 if (keylen >= strlen("last_id") && memcmp(key, "last_id", 7) == 0) {
2232 obd_id *last_id = val;
2233 /* FIXME: object groups */
2234 *last_id = filter_last_id(&obd->u.filter, 0);
2237 CDEBUG(D_IOCTL, "invalid key\n");
2241 static int filter_set_info(struct obd_export *exp, __u32 keylen,
2242 void *key, __u32 vallen, void *val)
2244 struct obd_device *obd;
2245 struct lustre_handle conn;
2246 struct llog_ctxt *ctxt;
2250 conn.cookie = exp->exp_handle.h_cookie;
2254 CDEBUG(D_IOCTL, "invalid exp %p cookie "LPX64"\n",
2259 if (keylen < strlen("mds_conn") ||
2260 memcmp(key, "mds_conn", keylen) != 0)
2263 CWARN("Received MDS connection ("LPX64")\n", conn.cookie);
2264 memcpy(&obd->u.filter.fo_mdc_conn, &conn, sizeof(conn));
2265 ctxt = llog_get_context(obd, LLOG_UNLINK_REPL_CTXT);
2266 rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse);
2270 int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
2271 int len, void *karg, void *uarg)
2273 struct obd_device *obd = exp->exp_obd;
2274 struct obd_ioctl_data *data = karg;
2278 case OBD_IOC_ABORT_RECOVERY:
2279 CERROR("aborting recovery for device %s\n", obd->obd_name);
2280 target_abort_recovery(obd);
2283 case OBD_IOC_SET_READONLY: {
2285 struct super_block *sb = obd->u.filter.fo_sb;
2286 struct inode *inode = sb->s_root->d_inode;
2287 BDEVNAME_DECLARE_STORAGE(tmp);
2288 CERROR("setting device %s read-only\n",
2289 ll_bdevname(sb, tmp));
2291 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
2293 (void)fsfilt_commit(obd, inode, handle, 1);
2295 dev_set_rdonly(ll_sbdev(obd->u.filter.fo_sb), 2);
2299 case OBD_IOC_CATLOGLIST: {
2300 rc = llog_catlog_list(obd, 1, data);
2304 case OBD_IOC_LLOG_CANCEL:
2305 case OBD_IOC_LLOG_REMOVE:
2306 case OBD_IOC_LLOG_INFO:
2307 case OBD_IOC_LLOG_PRINT: {
2308 /* FIXME to be finished */
2309 RETURN(-EOPNOTSUPP);
2311 struct llog_ctxt *ctxt = NULL;
2313 push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
2314 rc = llog_ioctl(ctxt, cmd, data);
2315 pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL);
2328 static struct llog_operations filter_unlink_repl_logops;
2329 static struct llog_operations filter_size_orig_logops = {
2330 lop_setup: llog_obd_origin_setup,
2331 lop_cleanup: llog_obd_origin_cleanup,
2332 lop_add: llog_obd_origin_add
2335 static int filter_llog_init(struct obd_device *obd, struct obd_device *tgt,
2336 int count, struct llog_catid *logid)
2338 struct llog_ctxt *ctxt;
2342 filter_unlink_repl_logops = llog_client_ops;
2343 filter_unlink_repl_logops.lop_cancel = llog_obd_repl_cancel;
2344 filter_unlink_repl_logops.lop_connect = llog_repl_connect;
2345 filter_unlink_repl_logops.lop_sync = llog_obd_repl_sync;
2347 rc = llog_setup(obd, LLOG_UNLINK_REPL_CTXT, tgt, 0, NULL,
2348 &filter_unlink_repl_logops);
2351 /* FIXME - assign unlink_cb for filter's recovery */
2352 ctxt = llog_get_context(obd, LLOG_UNLINK_REPL_CTXT);
2353 ctxt->llog_proc_cb = filter_recov_log_unlink_cb;
2355 rc = llog_setup(obd, LLOG_SIZE_ORIG_CTXT, tgt, 0, NULL,
2356 &filter_size_orig_logops);
2360 static int filter_llog_finish(struct obd_device *obd, int count)
2365 rc = llog_cleanup(llog_get_context(obd, LLOG_UNLINK_REPL_CTXT));
2369 rc = llog_cleanup(llog_get_context(obd, LLOG_SIZE_ORIG_CTXT));
2373 static struct dentry *filter_lvfs_fid2dentry(__u64 id, __u32 gen, __u64 gr,
2376 return filter_fid2dentry(data, NULL, gr, id);
2379 static struct lvfs_callback_ops filter_lvfs_ops = {
2380 l_fid2dentry: filter_lvfs_fid2dentry,
2383 static struct obd_ops filter_obd_ops = {
2384 .o_owner = THIS_MODULE,
2385 .o_get_info = filter_get_info,
2386 .o_set_info = filter_set_info,
2387 .o_setup = filter_setup,
2388 .o_precleanup = filter_precleanup,
2389 .o_cleanup = filter_cleanup,
2390 .o_connect = filter_connect,
2391 .o_disconnect = filter_disconnect,
2392 .o_statfs = filter_statfs,
2393 .o_getattr = filter_getattr,
2394 .o_unpackmd = filter_unpackmd,
2395 .o_create = filter_create,
2396 .o_setattr = filter_setattr,
2397 .o_destroy = filter_destroy,
2398 .o_brw = filter_brw,
2399 .o_punch = filter_truncate,
2400 .o_sync = filter_sync,
2401 .o_preprw = filter_preprw,
2402 .o_commitrw = filter_commitrw,
2403 .o_destroy_export = filter_destroy_export,
2404 .o_llog_init = filter_llog_init,
2405 .o_llog_finish = filter_llog_finish,
2406 .o_iocontrol = filter_iocontrol,
2409 static struct obd_ops filter_sanobd_ops = {
2410 .o_owner = THIS_MODULE,
2411 .o_get_info = filter_get_info,
2412 .o_set_info = filter_set_info,
2413 .o_setup = filter_san_setup,
2414 .o_precleanup = filter_precleanup,
2415 .o_cleanup = filter_cleanup,
2416 .o_connect = filter_connect,
2417 .o_disconnect = filter_disconnect,
2418 .o_statfs = filter_statfs,
2419 .o_getattr = filter_getattr,
2420 .o_unpackmd = filter_unpackmd,
2421 .o_create = filter_create,
2422 .o_setattr = filter_setattr,
2423 .o_destroy = filter_destroy,
2424 .o_brw = filter_brw,
2425 .o_punch = filter_truncate,
2426 .o_sync = filter_sync,
2427 .o_preprw = filter_preprw,
2428 .o_commitrw = filter_commitrw,
2429 .o_san_preprw = filter_san_preprw,
2430 .o_destroy_export = filter_destroy_export,
2431 .o_llog_init = filter_llog_init,
2432 .o_llog_finish = filter_llog_finish,
2433 .o_iocontrol = filter_iocontrol,
2436 static int __init obdfilter_init(void)
2438 struct lprocfs_static_vars lvars;
2441 printk(KERN_INFO "Lustre: Filtering OBD driver; info@clusterfs.com\n");
2443 lprocfs_init_vars(filter, &lvars);
2445 rc = class_register_type(&filter_obd_ops, lvars.module_vars,
2446 OBD_FILTER_DEVICENAME);
2450 rc = class_register_type(&filter_sanobd_ops, lvars.module_vars,
2451 OBD_FILTER_SAN_DEVICENAME);
2453 class_unregister_type(OBD_FILTER_DEVICENAME);
2457 static void __exit obdfilter_exit(void)
2459 class_unregister_type(OBD_FILTER_SAN_DEVICENAME);
2460 class_unregister_type(OBD_FILTER_DEVICENAME);
2463 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2464 MODULE_DESCRIPTION("Lustre Filtering OBD driver");
2465 MODULE_LICENSE("GPL");
2467 module_init(obdfilter_init);
2468 module_exit(obdfilter_exit);