1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * linux/fs/obdfilter/filter.c
6 * Copyright (c) 2001-2003 Cluster File Systems, Inc.
7 * Author: Peter Braam <braam@clusterfs.com>
8 * Author: Andreas Dilger <adilger@clusterfs.com>
10 * This file is part of Lustre, http://www.lustre.org.
12 * Lustre is free software; you can redistribute it and/or
13 * modify it under the terms of version 2 of the GNU General Public
14 * License as published by the Free Software Foundation.
16 * Lustre is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with Lustre; if not, write to the Free Software
23 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
27 * Invariant: Get O/R i_sem for lookup, if needed, before any journal ops
28 * (which need to get journal_lock, may block if journal full).
30 * Invariant: Call filter_start_transno() before any journal ops to avoid the
31 * same deadlock problem. We can (and want) to get rid of the
32 * transno sem in favour of the dir/inode i_sem to avoid single
33 * threaded operation on the OST.
36 #define DEBUG_SUBSYSTEM S_FILTER
38 #include <linux/config.h>
39 #include <linux/module.h>
41 #include <linux/dcache.h>
42 #include <linux/init.h>
43 #include <linux/version.h>
44 #include <linux/jbd.h>
45 #include <linux/ext3_fs.h>
46 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
47 # include <linux/mount.h>
48 # include <linux/buffer_head.h>
49 # include <linux/bio.h>
52 #include <linux/obd_class.h>
53 #include <linux/obd_lov.h>
54 #include <linux/obd_ost.h>
55 #include <linux/lustre_dlm.h>
56 #include <linux/lustre_fsfilt.h>
57 #include <linux/lprocfs_status.h>
58 #include <linux/lustre_log.h>
59 #include <linux/lustre_commit_confd.h>
60 #include <libcfs/list.h>
62 #include <linux/lustre_smfs.h>
63 #include "filter_internal.h"
65 /* Group 0 is no longer a legal group, to catch uninitialized IDs */
66 #define FILTER_MIN_GROUPS 3
68 static struct lvfs_callback_ops filter_lvfs_ops;
70 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
71 struct lov_stripe_md *ea, struct obd_trans_info *);
72 struct obd_llogs *filter_grab_llog_for_group(struct obd_device *,
73 int, struct obd_export *);
75 static void filter_commit_cb(struct obd_device *obd, __u64 transno,
76 void *cb_data, int error)
78 obd_transno_commit_cb(obd, transno, error);
81 /* Assumes caller has already pushed us into the kernel context. */
82 int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti,
85 struct filter_obd *filter = &exp->exp_obd->u.filter;
86 struct filter_export_data *fed = &exp->exp_filter_data;
87 struct filter_client_data *fcd = fed->fed_fcd;
90 int err, log_pri = D_HA;
92 /* Propagate error code. */
96 if (!exp->exp_obd->obd_replayable || oti == NULL)
99 /* we don't allocate new transnos for replayed requests */
100 if (oti->oti_transno == 0) {
101 spin_lock(&filter->fo_translock);
102 last_rcvd = le64_to_cpu(filter->fo_fsd->fsd_last_transno) + 1;
103 filter->fo_fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
104 spin_unlock(&filter->fo_translock);
105 oti->oti_transno = last_rcvd;
107 spin_lock(&filter->fo_translock);
108 last_rcvd = oti->oti_transno;
109 if (last_rcvd > le64_to_cpu(filter->fo_fsd->fsd_last_transno))
110 filter->fo_fsd->fsd_last_transno =
111 cpu_to_le64(last_rcvd);
112 spin_unlock(&filter->fo_translock);
115 fcd->fcd_last_rcvd = cpu_to_le64(last_rcvd);
117 /* could get xid from oti, if it's ever needed */
118 fcd->fcd_last_xid = 0;
120 off = fed->fed_lr_off;
122 fsfilt_add_journal_cb(exp->exp_obd, filter->fo_sb, last_rcvd,
123 oti->oti_handle, filter_commit_cb, NULL);
125 err = fsfilt_write_record(exp->exp_obd, filter->fo_rcvd_filp, fcd,
126 sizeof(*fcd), &off, 0);
133 CDEBUG(log_pri, "wrote trans "LPU64" for client %s at #%d: err = %d\n",
134 last_rcvd, fcd->fcd_uuid, fed->fed_lr_idx, err);
139 void f_dput(struct dentry *dentry)
141 /* Can't go inside filter_ddelete because it can block */
142 CDEBUG(D_INODE, "putting %s: %p, count = %d\n",
143 dentry->d_name.name, dentry, atomic_read(&dentry->d_count) - 1);
144 LASSERT(atomic_read(&dentry->d_count) > 0);
149 /* Add client data to the FILTER. We use a bitmap to locate a free space
150 * in the last_rcvd file if cl_idx is -1 (i.e. a new client).
151 * Otherwise, we have just read the data from the last_rcvd file and
152 * we know its offset. */
153 static int filter_client_add(struct obd_device *obd, struct filter_obd *filter,
154 struct filter_export_data *fed, int cl_idx)
156 unsigned long *bitmap = filter->fo_last_rcvd_slots;
157 int new_client = (cl_idx == -1);
160 LASSERT(bitmap != NULL);
162 /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
163 if (!strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid))
166 /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
167 * there's no need for extra complication here
170 cl_idx = find_first_zero_bit(bitmap, FILTER_LR_MAX_CLIENTS);
172 if (cl_idx >= FILTER_LR_MAX_CLIENTS) {
173 CERROR("no client slots - fix FILTER_LR_MAX_CLIENTS\n");
176 if (test_and_set_bit(cl_idx, bitmap)) {
177 CERROR("FILTER client %d: found bit is set in bitmap\n",
179 cl_idx = find_next_zero_bit(bitmap,
180 FILTER_LR_MAX_CLIENTS,
185 if (test_and_set_bit(cl_idx, bitmap)) {
186 CERROR("FILTER client %d: bit already set in bitmap!\n",
192 fed->fed_lr_idx = cl_idx;
193 fed->fed_lr_off = le32_to_cpu(filter->fo_fsd->fsd_client_start) +
194 cl_idx * le16_to_cpu(filter->fo_fsd->fsd_client_size);
196 CDEBUG(D_INFO, "client at index %d (%llu) with UUID '%s' added\n",
197 fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
200 struct lvfs_run_ctxt saved;
201 loff_t off = fed->fed_lr_off;
205 CDEBUG(D_INFO, "writing client fcd at idx %u (%llu) (len %u)\n",
206 fed->fed_lr_idx,off,(unsigned int)sizeof(*fed->fed_fcd));
208 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
209 /* Transaction needed to fix bug 1403 */
210 handle = fsfilt_start(obd,
211 filter->fo_rcvd_filp->f_dentry->d_inode,
212 FSFILT_OP_SETATTR, NULL);
213 if (IS_ERR(handle)) {
214 err = PTR_ERR(handle);
215 CERROR("unable to start transaction: rc %d\n", err);
217 err = fsfilt_write_record(obd, filter->fo_rcvd_filp,
219 sizeof(*fed->fed_fcd),
221 fsfilt_commit(obd, filter->fo_sb,
222 filter->fo_rcvd_filp->f_dentry->d_inode,
225 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
228 CERROR("error writing %s client idx %u: rc %d\n",
229 LAST_RCVD, fed->fed_lr_idx, err);
236 static int filter_client_free(struct obd_export *exp, int flags)
238 struct filter_export_data *fed = &exp->exp_filter_data;
239 struct filter_obd *filter = &exp->exp_obd->u.filter;
240 struct obd_device *obd = exp->exp_obd;
241 struct filter_client_data zero_fcd;
242 struct lvfs_run_ctxt saved;
247 if (fed->fed_fcd == NULL)
250 if (flags & OBD_OPT_FAILOVER)
253 /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
254 if (strcmp(fed->fed_fcd->fcd_uuid, obd->obd_uuid.uuid ) == 0)
257 LASSERT(filter->fo_last_rcvd_slots != NULL);
259 off = fed->fed_lr_off;
261 CDEBUG(D_INFO, "freeing client at idx %u (%lld) with UUID '%s'\n",
262 fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
264 /* Clear the bit _after_ zeroing out the client so we don't
265 race with filter_client_add and zero out new clients.*/
266 if (!test_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
267 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
272 memset(&zero_fcd, 0, sizeof zero_fcd);
273 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
274 rc = fsfilt_write_record(obd, filter->fo_rcvd_filp, &zero_fcd,
275 sizeof(zero_fcd), &off, 1);
277 /* update server's transno */
278 filter_update_server_data(obd, filter->fo_rcvd_filp,
280 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
282 CDEBUG(rc == 0 ? D_INFO : D_ERROR,
283 "zeroing disconnecting client %s at idx %u (%llu) in %s rc %d\n",
284 fed->fed_fcd->fcd_uuid, fed->fed_lr_idx, fed->fed_lr_off,
287 if (!test_and_clear_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
288 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
294 OBD_FREE(fed->fed_fcd, sizeof(*fed->fed_fcd));
299 static int filter_free_server_data(struct filter_obd *filter)
301 OBD_FREE(filter->fo_fsd, sizeof(*filter->fo_fsd));
302 filter->fo_fsd = NULL;
303 OBD_FREE(filter->fo_last_rcvd_slots, FILTER_LR_MAX_CLIENTS/8);
304 filter->fo_last_rcvd_slots = NULL;
308 /* assumes caller is already in kernel ctxt */
309 int filter_update_server_data(struct obd_device *obd, struct file *filp,
310 struct filter_server_data *fsd, int force_sync)
316 CDEBUG(D_INODE, "server uuid : %s\n", fsd->fsd_uuid);
317 CDEBUG(D_INODE, "server last_rcvd : "LPU64"\n",
318 le64_to_cpu(fsd->fsd_last_transno));
319 CDEBUG(D_INODE, "server last_mount: "LPU64"\n",
320 le64_to_cpu(fsd->fsd_mount_count));
322 rc = fsfilt_write_record(obd, filp, fsd, sizeof(*fsd), &off,force_sync);
324 CERROR("error writing filter_server_data: rc = %d\n", rc);
329 int filter_update_last_objid(struct obd_device *obd, obd_gr group,
332 struct filter_obd *filter = &obd->u.filter;
338 if (filter->fo_last_objid_files[group] == NULL) {
339 CERROR("Object group "LPU64" not fully setup; not updating "
340 "last_objid\n", group);
344 CDEBUG(D_INODE, "server last_objid for group "LPU64": "LPU64"\n",
345 group, filter->fo_last_objids[group]);
347 tmp = cpu_to_le64(filter->fo_last_objids[group]);
348 rc = fsfilt_write_record(obd, filter->fo_last_objid_files[group],
349 &tmp, sizeof(tmp), &off, force_sync);
351 CERROR("error writing group "LPU64" last objid: rc = %d\n",
356 /* assumes caller has already in kernel ctxt */
357 static int filter_init_server_data(struct obd_device *obd, struct file * filp)
359 struct filter_obd *filter = &obd->u.filter;
360 struct filter_server_data *fsd;
361 struct filter_client_data *fcd = NULL;
362 struct inode *inode = filp->f_dentry->d_inode;
363 unsigned long last_rcvd_size = inode->i_size;
369 /* ensure padding in the struct is the correct size */
370 LASSERT (offsetof(struct filter_server_data, fsd_padding) +
371 sizeof(fsd->fsd_padding) == FILTER_LR_SERVER_SIZE);
372 LASSERT (offsetof(struct filter_client_data, fcd_padding) +
373 sizeof(fcd->fcd_padding) == FILTER_LR_CLIENT_SIZE);
375 OBD_ALLOC(fsd, sizeof(*fsd));
378 filter->fo_fsd = fsd;
380 OBD_ALLOC(filter->fo_last_rcvd_slots, FILTER_LR_MAX_CLIENTS/8);
381 if (filter->fo_last_rcvd_slots == NULL) {
382 OBD_FREE(fsd, sizeof(*fsd));
386 if (last_rcvd_size == 0) {
387 CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD);
389 memcpy(fsd->fsd_uuid, obd->obd_uuid.uuid,sizeof(fsd->fsd_uuid));
390 fsd->fsd_last_transno = 0;
391 mount_count = fsd->fsd_mount_count = 0;
392 fsd->fsd_server_size = cpu_to_le32(FILTER_LR_SERVER_SIZE);
393 fsd->fsd_client_start = cpu_to_le32(FILTER_LR_CLIENT_START);
394 fsd->fsd_client_size = cpu_to_le16(FILTER_LR_CLIENT_SIZE);
395 fsd->fsd_subdir_count = cpu_to_le16(FILTER_SUBDIR_COUNT);
396 filter->fo_subdir_count = FILTER_SUBDIR_COUNT;
398 rc = fsfilt_read_record(obd, filp, fsd, sizeof(*fsd), &off);
400 CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
404 if (strcmp(fsd->fsd_uuid, obd->obd_uuid.uuid) != 0) {
405 CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
406 obd->obd_uuid.uuid, fsd->fsd_uuid);
407 GOTO(err_fsd, rc = -EINVAL);
409 mount_count = le64_to_cpu(fsd->fsd_mount_count);
410 filter->fo_subdir_count = le16_to_cpu(fsd->fsd_subdir_count);
413 if (fsd->fsd_feature_incompat & ~cpu_to_le32(FILTER_INCOMPAT_SUPP)) {
414 CERROR("unsupported feature %x\n",
415 le32_to_cpu(fsd->fsd_feature_incompat) &
416 ~FILTER_INCOMPAT_SUPP);
417 GOTO(err_fsd, rc = -EINVAL);
419 if (fsd->fsd_feature_rocompat & ~cpu_to_le32(FILTER_ROCOMPAT_SUPP)) {
420 CERROR("read-only feature %x\n",
421 le32_to_cpu(fsd->fsd_feature_rocompat) &
422 ~FILTER_ROCOMPAT_SUPP);
423 /* Do something like remount filesystem read-only */
424 GOTO(err_fsd, rc = -EINVAL);
427 CDEBUG(D_INODE, "%s: server last_rcvd : "LPU64"\n",
428 obd->obd_name, le64_to_cpu(fsd->fsd_last_transno));
429 CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
430 obd->obd_name, mount_count + 1);
431 CDEBUG(D_INODE, "%s: server data size: %u\n",
432 obd->obd_name, le32_to_cpu(fsd->fsd_server_size));
433 CDEBUG(D_INODE, "%s: per-client data start: %u\n",
434 obd->obd_name, le32_to_cpu(fsd->fsd_client_start));
435 CDEBUG(D_INODE, "%s: per-client data size: %u\n",
436 obd->obd_name, le32_to_cpu(fsd->fsd_client_size));
437 CDEBUG(D_INODE, "%s: server subdir_count: %u\n",
438 obd->obd_name, le16_to_cpu(fsd->fsd_subdir_count));
439 CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
440 last_rcvd_size <= le32_to_cpu(fsd->fsd_client_start) ? 0 :
441 (last_rcvd_size - le32_to_cpu(fsd->fsd_client_start)) /
442 le16_to_cpu(fsd->fsd_client_size));
444 if (!obd->obd_replayable) {
445 CWARN("%s: recovery support OFF\n", obd->obd_name);
449 for (cl_idx = 0, off = le32_to_cpu(fsd->fsd_client_start);
450 off < last_rcvd_size; cl_idx++) {
452 struct obd_export *exp;
453 struct filter_export_data *fed;
456 OBD_ALLOC(fcd, sizeof(*fcd));
458 GOTO(err_client, rc = -ENOMEM);
461 /* Don't assume off is incremented properly by
462 * fsfilt_read_record(), in case sizeof(*fcd)
463 * isn't the same as fsd->fsd_client_size. */
464 off = le32_to_cpu(fsd->fsd_client_start) +
465 cl_idx * le16_to_cpu(fsd->fsd_client_size);
466 rc = fsfilt_read_record(obd, filp, fcd, sizeof(*fcd), &off);
468 CERROR("error reading FILT %s idx %d off %llu: rc %d\n",
469 LAST_RCVD, cl_idx, off, rc);
470 break; /* read error shouldn't cause startup to fail */
473 if (fcd->fcd_uuid[0] == '\0') {
474 CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
479 last_rcvd = le64_to_cpu(fcd->fcd_last_rcvd);
481 /* These exports are cleaned up by filter_disconnect(), so they
482 * need to be set up like real exports as filter_connect() does.
484 exp = class_new_export(obd);
485 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64" "
486 "srv lr: "LPU64" fcd_group %d \n", fcd->fcd_uuid, cl_idx,
487 last_rcvd, le64_to_cpu(fsd->fsd_last_transno),
488 le32_to_cpu(fcd->fcd_group));
490 GOTO(err_client, rc = -ENOMEM);
492 memcpy(&exp->exp_client_uuid.uuid, fcd->fcd_uuid,
493 sizeof exp->exp_client_uuid.uuid);
494 fed = &exp->exp_filter_data;
496 fed->fed_group = le32_to_cpu(fcd->fcd_group);
497 filter_client_add(obd, filter, fed, cl_idx);
498 /* create helper if export init gets more complex */
499 spin_lock_init(&fed->fed_lock);
502 exp->exp_replay_needed = 1;
503 obd->obd_recoverable_clients++;
504 obd->obd_max_recoverable_clients++;
505 class_export_put(exp);
507 CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
510 if (last_rcvd > le64_to_cpu(fsd->fsd_last_transno))
511 fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
515 obd->obd_last_committed = le64_to_cpu(fsd->fsd_last_transno);
517 if (obd->obd_recoverable_clients) {
518 CWARN("RECOVERY: service %s, %d recoverable clients, "
519 "last_transno "LPU64"\n", obd->obd_name,
520 obd->obd_recoverable_clients,
521 le64_to_cpu(fsd->fsd_last_transno));
522 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
523 target_start_recovery_thread(obd, ost_handle);
527 OBD_FREE(fcd, sizeof(*fcd));
530 filter->fo_mount_count = mount_count + 1;
531 fsd->fsd_mount_count = cpu_to_le64(filter->fo_mount_count);
533 /* save it, so mount count and last_transno is current */
534 rc = filter_update_server_data(obd, filp, filter->fo_fsd, 1);
536 GOTO(err_client, rc);
540 class_disconnect_exports(obd, 0);
542 filter_free_server_data(filter);
546 static int filter_cleanup_groups(struct obd_device *obd)
548 struct filter_obd *filter = &obd->u.filter;
549 struct dentry *dentry;
553 for (i = 0; i < filter->fo_group_count; i++) {
554 if (filter->fo_subdirs != NULL) {
555 for (k = 0; k < filter->fo_subdir_count; k++) {
556 dentry = filter->fo_subdirs[i].dentry[k];
560 filter->fo_subdirs[i].dentry[k] = NULL;
563 if (filter->fo_last_objid_files[i] != NULL) {
564 filp_close(filter->fo_last_objid_files[i], 0);
565 filter->fo_last_objid_files[i] = NULL;
567 if (filter->fo_groups[i] != NULL) {
568 dput(filter->fo_groups[i]);
569 filter->fo_groups[i] = NULL;
572 if (filter->fo_subdirs != NULL)
573 OBD_FREE(filter->fo_subdirs,
574 filter->fo_group_count * sizeof(*filter->fo_subdirs));
575 if (filter->fo_groups != NULL)
576 OBD_FREE(filter->fo_groups,
577 filter->fo_group_count * sizeof(*filter->fo_groups));
578 if (filter->fo_last_objids != NULL)
579 OBD_FREE(filter->fo_last_objids,
580 filter->fo_group_count * sizeof(__u64));
581 if (filter->fo_last_objid_files != NULL)
582 OBD_FREE(filter->fo_last_objid_files,
583 filter->fo_group_count * sizeof(struct file *));
584 f_dput(filter->fo_dentry_O);
588 static int filter_update_last_group(struct obd_device *obd, int group)
590 struct filter_obd *filter = &obd->u.filter;
591 struct file *filp = NULL;
592 int last_group = 0, rc;
596 if (group <= filter->fo_committed_group)
599 filp = filp_open("LAST_GROUP", O_RDWR, 0700);
603 CERROR("cannot open LAST_GROUP: rc = %d\n", rc);
607 rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
609 CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n",rc);
612 LASSERT(off == 0 || last_group >= FILTER_MIN_GROUPS);
613 CDEBUG(D_INODE, "%s: previous %d, new %d\n",
614 obd->obd_name, last_group, group);
618 /* must be sync: bXXXX */
619 rc = fsfilt_write_record(obd, filp, &last_group, sizeof(__u32), &off, 1);
621 CDEBUG(D_INODE, "error updating LAST_GROUP: rc %d\n", rc);
625 filter->fo_committed_group = group;
632 static int filter_read_group_internal(struct obd_device *obd, int group,
635 struct filter_obd *filter = &obd->u.filter;
636 __u64 *new_objids = NULL;
637 struct filter_subdirs *new_subdirs = NULL, *tmp_subdirs = NULL;
638 struct dentry **new_groups = NULL;
639 struct file **new_files = NULL;
640 struct dentry *dentry;
642 int old_count = filter->fo_group_count, rc, stage = 0, i;
647 snprintf(name, 24, "%d", group);
651 dentry = ll_lookup_one_len(name, filter->fo_dentry_O,
653 if (IS_ERR(dentry)) {
654 CERROR("Cannot lookup expected object group %d: %ld\n",
655 group, PTR_ERR(dentry));
656 RETURN(PTR_ERR(dentry));
659 dentry = simple_mkdir(filter->fo_dentry_O, name, 0700, 1);
660 if (IS_ERR(dentry)) {
661 CERROR("cannot lookup/create O/%s: rc = %ld\n", name,
663 RETURN(PTR_ERR(dentry));
668 snprintf(name, 24, "O/%d/LAST_ID", group);
670 filp = filp_open(name, O_CREAT | O_RDWR, 0700);
672 CERROR("cannot create %s: rc = %ld\n", name, PTR_ERR(filp));
673 GOTO(cleanup, rc = PTR_ERR(filp));
677 rc = fsfilt_read_record(obd, filp, &last_objid, sizeof(__u64), &off);
679 CDEBUG(D_INODE, "error reading %s: rc %d\n", name, rc);
683 if (filter->fo_subdir_count) {
684 OBD_ALLOC(tmp_subdirs, sizeof(*tmp_subdirs));
685 if (tmp_subdirs == NULL)
686 GOTO(cleanup, rc = -ENOMEM);
689 for (i = 0; i < filter->fo_subdir_count; i++) {
691 snprintf(dir, sizeof(dir), "d%u", i);
693 tmp_subdirs->dentry[i] =
694 simple_mkdir(dentry, dir, 0700, 1);
695 if (IS_ERR(tmp_subdirs->dentry[i])) {
696 rc = PTR_ERR(tmp_subdirs->dentry[i]);
697 CERROR("can't lookup/create O/%d/%s: rc = %d\n",
701 CDEBUG(D_INODE, "got/created O/%d/%s: %p\n", group, dir,
702 tmp_subdirs->dentry[i]);
706 /* 'group' is an index; we need an array of length 'group + 1' */
707 if (group + 1 > old_count) {
709 OBD_ALLOC(new_objids, len * sizeof(*new_objids));
710 OBD_ALLOC(new_subdirs, len * sizeof(*new_subdirs));
711 OBD_ALLOC(new_groups, len * sizeof(*new_groups));
712 OBD_ALLOC(new_files, len * sizeof(*new_files));
714 if (new_objids == NULL || new_subdirs == NULL ||
715 new_groups == NULL || new_files == NULL)
716 GOTO(cleanup, rc = -ENOMEM);
718 memcpy(new_objids, filter->fo_last_objids,
719 old_count * sizeof(*new_objids));
720 memcpy(new_subdirs, filter->fo_subdirs,
721 old_count * sizeof(*new_subdirs));
722 memcpy(new_groups, filter->fo_groups,
723 old_count * sizeof(*new_groups));
724 memcpy(new_files, filter->fo_last_objid_files,
725 old_count * sizeof(*new_files));
728 OBD_FREE(filter->fo_last_objids,
729 old_count * sizeof(*new_objids));
730 OBD_FREE(filter->fo_subdirs,
731 old_count * sizeof(*new_subdirs));
732 OBD_FREE(filter->fo_groups,
733 old_count * sizeof(*new_groups));
734 OBD_FREE(filter->fo_last_objid_files,
735 old_count * sizeof(*new_files));
737 filter->fo_last_objids = new_objids;
738 filter->fo_subdirs = new_subdirs;
739 filter->fo_groups = new_groups;
740 filter->fo_last_objid_files = new_files;
741 filter->fo_group_count = len;
744 filter->fo_groups[group] = dentry;
745 filter->fo_last_objid_files[group] = filp;
746 if (filter->fo_subdir_count) {
747 filter->fo_subdirs[group] = *tmp_subdirs;
748 OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
751 filter_update_last_group(obd, group);
753 if (filp->f_dentry->d_inode->i_size == 0) {
754 filter->fo_last_objids[group] = FILTER_INIT_OBJID;
758 filter->fo_last_objids[group] = le64_to_cpu(last_objid);
759 CDEBUG(D_INODE, "%s: server last_objid group %d: "LPU64"\n",
760 obd->obd_name, group, last_objid);
765 if (new_objids != NULL)
766 OBD_FREE(new_objids, group * sizeof(*new_objids));
767 if (new_subdirs != NULL)
768 OBD_FREE(new_subdirs, group * sizeof(*new_subdirs));
769 if (new_groups != NULL)
770 OBD_FREE(new_groups, group * sizeof(*new_groups));
771 if (new_files != NULL)
772 OBD_FREE(new_files, group * sizeof(*new_files));
774 if (filter->fo_subdir_count) {
775 for (i = 0; i < filter->fo_subdir_count; i++) {
776 if (tmp_subdirs->dentry[i] != NULL)
777 dput(tmp_subdirs->dentry[i]);
779 OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
789 static int filter_read_groups(struct obd_device *obd, int last_group,
792 struct filter_obd *filter = &obd->u.filter;
793 int old_count, group, rc = 0;
795 down(&filter->fo_init_lock);
796 old_count = filter->fo_group_count;
797 for (group = old_count; group <= last_group; group++) {
799 continue; /* no group zero */
801 rc = filter_read_group_internal(obd, group, create);
805 up(&filter->fo_init_lock);
809 static int filter_prep_groups(struct obd_device *obd)
811 struct filter_obd *filter = &obd->u.filter;
812 struct dentry *dentry, *O_dentry;
813 int rc = 0, cleanup_phase = 0;
814 struct file *filp = NULL;
819 O_dentry = simple_mkdir(current->fs->pwd, "O", 0700, 1);
820 CDEBUG(D_INODE, "got/created O: %p\n", O_dentry);
821 if (IS_ERR(O_dentry)) {
822 rc = PTR_ERR(O_dentry);
823 CERROR("cannot open/create O: rc = %d\n", rc);
826 filter->fo_dentry_O = O_dentry;
827 cleanup_phase = 1; /* O_dentry */
829 /* Lookup "R" to tell if we're on an old OST FS and need to convert
830 * from O/R/<dir>/<objid> to O/0/<dir>/<objid>. This can be removed
831 * some time post 1.0 when all old-style OSTs have converted along
832 * with the init_objid hack. */
833 dentry = ll_lookup_one_len("R", O_dentry, 1);
835 GOTO(cleanup, rc = PTR_ERR(dentry));
836 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
837 struct dentry *O0_dentry = lookup_one_len("0", O_dentry, 1);
840 CWARN("converting OST to new object layout\n");
841 if (IS_ERR(O0_dentry)) {
842 rc = PTR_ERR(O0_dentry);
843 CERROR("error looking up O/0: rc %d\n", rc);
847 if (O0_dentry->d_inode) {
848 CERROR("Both O/R and O/0 exist. Fix manually.\n");
849 GOTO(cleanup_O0, rc = -EEXIST);
852 down(&O_dentry->d_inode->i_sem);
853 rc = vfs_rename(O_dentry->d_inode, dentry,
854 O_dentry->d_inode, O0_dentry);
855 up(&O_dentry->d_inode->i_sem);
858 CERROR("error renaming O/R to O/0: rc %d\n", rc);
859 GOTO(cleanup_O0, rc);
861 filter->fo_fsd->fsd_feature_incompat |=
862 cpu_to_le32(FILTER_INCOMPAT_GROUPS);
863 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
865 GOTO(cleanup_O0, rc);
877 cleanup_phase = 2; /* groups */
879 /* we have to initialize all groups before first connections from
880 * clients because they may send create/destroy for any group -bzzz */
881 filp = filp_open("LAST_GROUP", O_CREAT | O_RDWR, 0700);
883 CERROR("cannot create LAST_GROUP: rc = %ld\n", PTR_ERR(filp));
884 GOTO(cleanup, rc = PTR_ERR(filp));
886 cleanup_phase = 3; /* filp */
888 rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
890 CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n", rc);
894 last_group = FILTER_MIN_GROUPS;
896 LASSERT(last_group >= FILTER_MIN_GROUPS);
899 CWARN("%s: initialize groups [%d,%d]\n", obd->obd_name,
900 FILTER_MIN_GROUPS, last_group);
901 filter->fo_committed_group = last_group;
902 rc = filter_read_groups(obd, last_group, 1);
910 switch (cleanup_phase) {
914 filter_cleanup_groups(obd);
916 f_dput(filter->fo_dentry_O);
917 filter->fo_dentry_O = NULL;
924 /* setup the object store with correct subdirectories */
925 static int filter_prep(struct obd_device *obd)
927 struct lvfs_run_ctxt saved;
928 struct filter_obd *filter = &obd->u.filter;
934 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
935 file = filp_open(LAST_RCVD, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
936 if (!file || IS_ERR(file)) {
938 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
943 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
944 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
945 file->f_dentry->d_inode->i_mode);
946 GOTO(err_filp, rc = -ENOENT);
949 /* steal operations */
950 inode = file->f_dentry->d_inode;
951 filter->fo_fop = file->f_op;
952 filter->fo_iop = inode->i_op;
953 filter->fo_aops = inode->i_mapping->a_ops;
955 rc = filter_init_server_data(obd, file);
957 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
960 filter->fo_rcvd_filp = file;
962 rc = filter_prep_groups(obd);
964 GOTO(err_server_data, rc);
967 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
972 //class_disconnect_exports(obd, 0);
973 filter_free_server_data(filter);
975 if (filp_close(file, 0))
976 CERROR("can't close %s after error\n", LAST_RCVD);
977 filter->fo_rcvd_filp = NULL;
981 /* cleanup the filter: write last used object id to status file */
982 static void filter_post(struct obd_device *obd)
984 struct lvfs_run_ctxt saved;
985 struct filter_obd *filter = &obd->u.filter;
988 /* XXX: filter_update_lastobjid used to call fsync_dev. It might be
989 * best to start a transaction with h_sync, because we removed this
992 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
993 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
996 CERROR("error writing server data: rc = %d\n", rc);
998 for (i = 1; i < filter->fo_group_count; i++) {
999 rc = filter_update_last_objid(obd, i,
1000 (i == filter->fo_group_count - 1));
1002 CERROR("error writing group %d lastobjid: rc = %d\n",
1006 rc = filp_close(filter->fo_rcvd_filp, 0);
1007 filter->fo_rcvd_filp = NULL;
1009 CERROR("error closing %s: rc = %d\n", LAST_RCVD, rc);
1011 filter_cleanup_groups(obd);
1012 filter_free_server_data(filter);
1013 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1016 static void filter_set_last_id(struct filter_obd *filter, int group, obd_id id)
1018 LASSERT(filter->fo_fsd != NULL);
1020 LASSERT(group < filter->fo_group_count);
1022 spin_lock(&filter->fo_objidlock);
1023 filter->fo_last_objids[group] = id;
1024 spin_unlock(&filter->fo_objidlock);
1027 __u64 filter_last_id(struct filter_obd *filter, int group)
1030 LASSERT(filter->fo_fsd != NULL);
1032 LASSERT(group < filter->fo_group_count);
1034 spin_lock(&filter->fo_objidlock);
1035 id = filter->fo_last_objids[group];
1036 spin_unlock(&filter->fo_objidlock);
1041 /* direct cut-n-paste of mds_blocking_ast() */
1042 static int filter_blocking_ast(struct ldlm_lock *lock,
1043 struct ldlm_lock_desc *desc,
1044 void *data, int flag)
1049 if (flag == LDLM_CB_CANCELING) {
1050 /* Don't need to do anything here. */
1054 /* XXX layering violation! -phil */
1055 l_lock(&lock->l_resource->lr_namespace->ns_lock);
1056 /* Get this: if filter_blocking_ast is racing with ldlm_intent_policy,
1057 * such that filter_blocking_ast is called just before l_i_p takes the
1058 * ns_lock, then by the time we get the lock, we might not be the
1059 * correct blocking function anymore. So check, and return early, if
1061 if (lock->l_blocking_ast != filter_blocking_ast) {
1062 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
1066 lock->l_flags |= LDLM_FL_CBPENDING;
1067 do_ast = (!lock->l_readers && !lock->l_writers);
1068 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
1071 struct lustre_handle lockh;
1074 LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
1075 ldlm_lock2handle(lock, &lockh);
1076 rc = ldlm_cli_cancel(&lockh);
1078 CERROR("ldlm_cli_cancel: %d\n", rc);
1080 LDLM_DEBUG(lock, "Lock still has references, will be "
1086 extern void *lock_dir(struct inode *dir, struct qstr *name);
1087 extern void unlock_dir(struct inode *dir, void *lock);
1089 static void * filter_lock_dentry(struct obd_device *obd,
1090 struct dentry *dparent, obd_id id)
1096 qstr.len = sprintf(name, LPU64, id);
1097 return lock_dir(dparent->d_inode, &qstr);
1099 down(&dparent->d_inode->i_sem);
1104 /* We never dget the object parent, so DON'T dput it either */
1105 static void filter_parent_unlock(struct dentry *dparent, void *lock)
1108 LASSERT(lock != NULL);
1109 unlock_dir(dparent->d_inode, lock);
1111 up(&dparent->d_inode->i_sem);
1115 /* We never dget the object parent, so DON'T dput it either */
1116 struct dentry *filter_parent(struct obd_device *obd, obd_gr group, obd_id objid)
1118 struct filter_obd *filter = &obd->u.filter;
1119 LASSERT(group < filter->fo_group_count);
1122 if (filter->fo_subdir_count == 0)
1123 return filter->fo_groups[group];
1125 return filter->fo_subdirs[group].dentry[objid & (filter->fo_subdir_count - 1)];
1128 /* We never dget the object parent, so DON'T dput it either */
1129 struct dentry *filter_parent_lock(struct obd_device *obd, obd_gr group,
1130 obd_id objid, void **lock)
1132 unsigned long now = jiffies;
1133 struct dentry *dparent = filter_parent(obd, group, objid);
1135 if (IS_ERR(dparent))
1139 LASSERT(dparent->d_inode);
1141 *lock = filter_lock_dentry(obd, dparent, objid);
1142 fsfilt_check_slow(now, obd_timeout, "parent lock");
1146 /* How to get files, dentries, inodes from object id's.
1148 * If dir_dentry is passed, the caller has already locked the parent
1149 * appropriately for this operation (normally a write lock). If
1150 * dir_dentry is NULL, we do a read lock while we do the lookup to
1151 * avoid races with create/destroy and such changing the directory
1152 * internal to the filesystem code. */
1153 struct dentry *filter_id2dentry(struct obd_device *obd,
1154 struct dentry *dir_dentry,
1155 obd_gr group, obd_id id)
1157 struct dentry *dparent = dir_dentry;
1158 struct dentry *dchild;
1165 CERROR("fatal: invalid object id 0\n");
1166 RETURN(ERR_PTR(-ESTALE));
1169 len = sprintf(name, LPU64, id);
1170 if (dir_dentry == NULL) {
1171 dparent = filter_parent_lock(obd, group, id, &lock);
1172 if (IS_ERR(dparent)) {
1173 CERROR("%s: error getting object "LPU64":"LPU64
1174 " parent: rc %ld\n", obd->obd_name,
1175 id, group, PTR_ERR(dparent));
1179 CDEBUG(D_INODE, "looking up object O/%.*s/%s\n",
1180 dparent->d_name.len, dparent->d_name.name, name);
1181 dchild = /*ll_*/lookup_one_len(name, dparent, len);
1182 if (dir_dentry == NULL)
1183 filter_parent_unlock(dparent, lock);
1184 if (IS_ERR(dchild)) {
1185 CERROR("%s: child lookup error %ld\n", obd->obd_name,
1190 if (dchild->d_inode != NULL && is_bad_inode(dchild->d_inode)) {
1191 CERROR("%s: got bad inode "LPU64"\n", obd->obd_name, id);
1193 RETURN(ERR_PTR(-ENOENT));
1196 CDEBUG(D_INODE, "got child objid %s: %p, count = %d\n",
1197 name, dchild, atomic_read(&dchild->d_count));
1199 LASSERT(atomic_read(&dchild->d_count) > 0);
1204 static int filter_prepare_destroy(struct obd_device *obd, obd_id objid,
1207 struct lustre_handle lockh;
1208 int flags = LDLM_AST_DISCARD_DATA, rc;
1209 struct ldlm_res_id res_id = { .name = { objid, 0, group, 0 } };
1210 ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
1213 /* Tell the clients that the object is gone now and that they should
1214 * throw away any cached pages. */
1215 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, res_id,
1216 LDLM_EXTENT, &policy, LCK_PW,
1217 &flags, filter_blocking_ast, ldlm_completion_ast,
1218 NULL, NULL, NULL, 0, NULL, &lockh);
1220 /* We only care about the side-effects, just drop the lock. */
1222 ldlm_lock_decref(&lockh, LCK_PW);
1227 /* Caller must hold LCK_PW on parent and push us into kernel context.
1228 * Caller is also required to ensure that dchild->d_inode exists. */
1229 static int filter_destroy_internal(struct obd_device *obd, obd_id objid,
1230 struct dentry *dparent,
1231 struct dentry *dchild)
1233 struct inode *inode = dchild->d_inode;
1237 if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) {
1238 CERROR("destroying objid %.*s nlink = %lu, count = %d\n",
1239 dchild->d_name.len, dchild->d_name.name,
1240 (unsigned long)inode->i_nlink,
1241 atomic_read(&inode->i_count));
1244 rc = vfs_unlink(dparent->d_inode, dchild);
1247 CERROR("error unlinking objid %.*s: rc %d\n",
1248 dchild->d_name.len, dchild->d_name.name, rc);
1253 static int filter_intent_policy(struct ldlm_namespace *ns,
1254 struct ldlm_lock **lockp, void *req_cookie,
1255 ldlm_mode_t mode, int flags, void *data)
1257 struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1258 struct ptlrpc_request *req = req_cookie;
1259 struct ldlm_lock *lock = *lockp, *l = NULL;
1260 struct ldlm_resource *res = lock->l_resource;
1261 ldlm_processing_policy policy;
1262 struct ost_lvb *res_lvb, *reply_lvb;
1263 struct ldlm_reply *rep;
1264 struct list_head *tmp;
1266 int tmpflags = 0, rc, repsize[2] = {sizeof(*rep), sizeof(*reply_lvb)};
1269 policy = ldlm_get_processing_policy(res);
1270 LASSERT(policy != NULL);
1271 LASSERT(req != NULL);
1273 rc = lustre_pack_reply(req, 2, repsize, NULL);
1275 RETURN(req->rq_status = rc);
1277 rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*rep));
1278 LASSERT(rep != NULL);
1280 reply_lvb = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*reply_lvb));
1281 LASSERT(reply_lvb != NULL);
1283 //fixup_handle_for_resent_req(req, lock, &lockh);
1285 /* If we grant any lock at all, it will be a whole-file read lock.
1286 * Call the extent policy function to see if our request can be
1287 * granted, or is blocked. */
1288 lock->l_policy_data.l_extent.start = 0;
1289 lock->l_policy_data.l_extent.end = OBD_OBJECT_EOF;
1290 lock->l_req_mode = LCK_PR;
1292 l_lock(&res->lr_namespace->ns_lock);
1294 res->lr_tmp = &rpc_list;
1295 rc = policy(lock, &tmpflags, 0, &err);
1298 /* FIXME: we should change the policy function slightly, to not make
1299 * this list at all, since we just turn around and free it */
1300 while (!list_empty(&rpc_list)) {
1301 struct ldlm_ast_work *w =
1302 list_entry(rpc_list.next, struct ldlm_ast_work, w_list);
1303 list_del(&w->w_list);
1304 LDLM_LOCK_PUT(w->w_lock);
1305 OBD_FREE(w, sizeof(*w));
1308 if (rc == LDLM_ITER_CONTINUE) {
1309 /* The lock met with no resistance; we're finished. */
1310 l_unlock(&res->lr_namespace->ns_lock);
1311 RETURN(ELDLM_LOCK_REPLACED);
1314 /* Do not grant any lock, but instead send GL callbacks. The extent
1315 * policy nicely created a list of all PW locks for us. We will choose
1316 * the highest of those which are larger than the size in the LVB, if
1317 * any, and perform a glimpse callback. */
1318 down(&res->lr_lvb_sem);
1319 res_lvb = res->lr_lvb_data;
1320 LASSERT(res_lvb != NULL);
1321 *reply_lvb = *res_lvb;
1322 up(&res->lr_lvb_sem);
1324 list_for_each(tmp, &res->lr_granted) {
1325 struct ldlm_lock *tmplock =
1326 list_entry(tmp, struct ldlm_lock, l_res_link);
1328 if (tmplock->l_granted_mode == LCK_PR)
1331 if (tmplock->l_policy_data.l_extent.end <= reply_lvb->lvb_size)
1335 l = LDLM_LOCK_GET(tmplock);
1339 if (l->l_policy_data.l_extent.start >
1340 tmplock->l_policy_data.l_extent.start)
1344 l = LDLM_LOCK_GET(tmplock);
1346 l_unlock(&res->lr_namespace->ns_lock);
1348 /* There were no PW locks beyond the size in the LVB; finished. */
1350 RETURN(ELDLM_LOCK_ABORTED);
1352 if (l->l_glimpse_ast == NULL) {
1353 /* We are racing with unlink(); just return -ENOENT */
1354 rep->lock_policy_res1 = -ENOENT;
1358 LASSERTF(l->l_glimpse_ast != NULL, "l == %p", l);
1360 rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
1361 if (rc != 0 && res->lr_namespace->ns_lvbo &&
1362 res->lr_namespace->ns_lvbo->lvbo_update) {
1363 res->lr_namespace->ns_lvbo->lvbo_update(res, NULL, 0, 1);
1366 down(&res->lr_lvb_sem);
1367 *reply_lvb = *res_lvb;
1368 up(&res->lr_lvb_sem);
1372 RETURN(ELDLM_LOCK_ABORTED);
1375 static int filter_post_fs_cleanup(struct obd_device *obd)
1379 rc = fsfilt_post_cleanup(obd);
1384 static int filter_group_set_fs_flags(struct obd_device *obd, int group)
1386 struct filter_obd *filter = &obd->u.filter;
1390 /* zero group is not longer valid. */
1393 for (i = 0; i < filter->fo_subdir_count; i++) {
1394 struct dentry *dentry;
1395 dentry = (filter->fo_subdirs + group)->dentry[i];
1396 rc = fsfilt_set_fs_flags(obd, dentry->d_inode,
1397 SM_DO_REC | SM_DO_COW);
1403 static int filter_post_fs_setup(struct obd_device *obd)
1405 struct filter_obd *filter = &obd->u.filter;
1407 struct llog_ctxt *ctxt = NULL;
1409 rc = fsfilt_post_setup(obd, filter->fo_dentry_O);
1413 for (j = 0; j < filter->fo_group_count; j++) {
1414 rc = filter_group_set_fs_flags(obd, j);
1419 fsfilt_get_reint_log_ctxt(obd, filter->fo_sb, &ctxt);
1421 ctxt->loc_obd = obd;
1422 ctxt->loc_idx = LLOG_REINT_ORIG_CTXT;
1423 obd->obd_llog_ctxt[LLOG_REINT_ORIG_CTXT] = ctxt;
1425 fsfilt_set_ost_flags(obd, filter->fo_sb);
1429 /* mount the file system (secretly) */
1430 int filter_common_setup(struct obd_device *obd, obd_count len, void *buf,
1433 struct lustre_cfg* lcfg = buf;
1434 struct filter_obd *filter = &obd->u.filter;
1435 struct lvfs_obd_ctxt *lvfs_ctxt = NULL;
1436 struct vfsmount *mnt;
1442 if (!lcfg->lcfg_inlbuf1 || !lcfg->lcfg_inlbuf2)
1445 obd->obd_fsops = fsfilt_get_ops(lcfg->lcfg_inlbuf2);
1446 if (IS_ERR(obd->obd_fsops))
1447 RETURN(PTR_ERR(obd->obd_fsops));
1449 rc = lvfs_mount_fs(lcfg->lcfg_inlbuf1, lcfg->lcfg_inlbuf2,
1450 option, MS_NOATIME | MS_NODIRATIME, &lvfs_ctxt);
1452 CERROR("lvfs_mount_fs failed: rc = %d\n", rc);
1457 mnt = lvfs_ctxt->loc_mnt;
1458 filter->fo_lvfs_ctxt = lvfs_ctxt;
1460 if (lcfg->lcfg_inllen3 > 0 && lcfg->lcfg_inlbuf3) {
1461 if (*lcfg->lcfg_inlbuf3 == 'f') {
1462 obd->obd_replayable = 1;
1463 obd_sync_filter = 1;
1464 CWARN("%s: recovery enabled\n", obd->obd_name);
1466 if (*lcfg->lcfg_inlbuf3 != 'n') {
1467 CERROR("unrecognised flag '%c'\n",
1468 *lcfg->lcfg_inlbuf3);
1470 // XXX Robert? Why do we get errors here
1471 // GOTO(err_mntput, rc = -EINVAL);
1475 filter->fo_vfsmnt = mnt;
1476 filter->fo_sb = mnt->mnt_sb;
1477 filter->fo_fstype = mnt->mnt_sb->s_type->name;
1478 CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
1480 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1481 obd->obd_lvfs_ctxt.pwdmnt = mnt;
1482 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
1483 obd->obd_lvfs_ctxt.fs = get_ds();
1484 obd->obd_lvfs_ctxt.cb_ops = filter_lvfs_ops;
1486 ll_clear_rdonly(ll_sbdev(filter->fo_sb));
1488 rc = fsfilt_setup(obd, mnt->mnt_sb);
1490 GOTO(err_mntput, rc);
1492 sema_init(&filter->fo_init_lock, 1);
1493 filter->fo_committed_group = 0;
1494 rc = filter_prep(obd);
1496 GOTO(err_mntput, rc);
1498 filter->fo_destroys_in_progress = 0;
1499 for (i = 0; i < 32; i++)
1500 sema_init(&filter->fo_create_locks[i], 1);
1502 spin_lock_init(&filter->fo_translock);
1503 spin_lock_init(&filter->fo_objidlock);
1504 INIT_LIST_HEAD(&filter->fo_export_list);
1505 sema_init(&filter->fo_alloc_lock, 1);
1506 spin_lock_init(&filter->fo_r_pages.oh_lock);
1507 spin_lock_init(&filter->fo_w_pages.oh_lock);
1508 spin_lock_init(&filter->fo_r_discont_pages.oh_lock);
1509 spin_lock_init(&filter->fo_w_discont_pages.oh_lock);
1510 spin_lock_init(&filter->fo_r_discont_blocks.oh_lock);
1511 spin_lock_init(&filter->fo_w_discont_blocks.oh_lock);
1512 filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE;
1514 INIT_LIST_HEAD(&filter->fo_llog_list);
1515 spin_lock_init(&filter->fo_llog_list_lock);
1517 sprintf(ns_name, "filter-%s", obd->obd_uuid.uuid);
1518 obd->obd_namespace = ldlm_namespace_new(ns_name, LDLM_NAMESPACE_SERVER);
1520 if (obd->obd_namespace == NULL)
1521 GOTO(err_post, rc = -ENOMEM);
1522 obd->obd_namespace->ns_lvbp = obd;
1523 obd->obd_namespace->ns_lvbo = &filter_lvbo;
1524 ldlm_register_intent(obd->obd_namespace, filter_intent_policy);
1526 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1527 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
1529 rc = obd_llog_cat_initialize(obd, &obd->obd_llogs, 1, CATLIST);
1531 CERROR("failed to setup llogging subsystems\n");
1540 lvfs_umount_fs(filter->fo_lvfs_ctxt);
1544 fsfilt_put_ops(obd->obd_fsops);
1548 static int filter_attach(struct obd_device *obd, obd_count len, void *data)
1550 struct lprocfs_static_vars lvars;
1553 lprocfs_init_vars(filter, &lvars);
1554 rc = lprocfs_obd_attach(obd, lvars.obd_vars);
1558 rc = lprocfs_alloc_obd_stats(obd, LPROC_FILTER_LAST);
1562 /* Init obdfilter private stats here */
1563 lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_READ_BYTES,
1564 LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
1565 lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
1566 LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
1568 return lproc_filter_attach_seqstat(obd);
1571 static int filter_detach(struct obd_device *dev)
1573 lprocfs_free_obd_stats(dev);
1574 return lprocfs_obd_detach(dev);
1577 static int filter_setup(struct obd_device *obd, obd_count len, void *buf)
1579 struct lustre_cfg* lcfg = buf;
1582 /* all mount options including errors=remount-ro and asyncdel are passed
1583 * using 4th lcfg param. And it is good, finally we have got rid of
1584 * hardcoded fs types in the code. */
1585 rc = filter_common_setup(obd, len, buf, lcfg->lcfg_inlbuf4);
1588 rc = filter_post_fs_setup(obd);
1592 static int filter_cleanup(struct obd_device *obd, int flags)
1594 struct filter_obd *filter = &obd->u.filter;
1595 ll_sbdev_type save_dev;
1598 if (flags & OBD_OPT_FAILOVER)
1599 CERROR("%s: shutting down for failover; client state will"
1600 " be preserved.\n", obd->obd_name);
1602 if (!list_empty(&obd->obd_exports)) {
1603 CERROR("%s: still has clients!\n", obd->obd_name);
1604 class_disconnect_exports(obd, flags);
1605 if (!list_empty(&obd->obd_exports)) {
1606 CERROR("still has exports after forced cleanup?\n");
1611 target_cleanup_recovery(obd);
1613 ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
1615 if (filter->fo_sb == NULL)
1618 save_dev = ll_sbdev(filter->fo_sb);
1619 filter_post_fs_cleanup(obd);
1622 shrink_dcache_parent(filter->fo_sb->s_root);
1626 lvfs_umount_fs(filter->fo_lvfs_ctxt);
1627 //destroy_buffers(filter->fo_sb->s_dev);
1628 filter->fo_sb = NULL;
1629 fsfilt_put_ops(obd->obd_fsops);
1632 ll_clear_rdonly(save_dev);
1637 static int filter_connect_post(struct obd_export *exp, unsigned initial,
1638 unsigned long connect_flags)
1640 struct obd_device *obd = exp->exp_obd;
1641 struct filter_export_data *fed;
1642 char str[PTL_NALFMT_SIZE];
1643 struct obd_llogs *llog;
1644 struct llog_ctxt *ctxt;
1648 fed = &exp->exp_filter_data;
1649 if (fed->fed_group < FILTER_MIN_GROUPS)
1652 /* initialize llogs for connections from MDS */
1653 llog = filter_grab_llog_for_group(obd, fed->fed_group, exp);
1654 LASSERT(llog != NULL);
1656 ctxt = llog_get_context(llog, LLOG_UNLINK_REPL_CTXT);
1657 LASSERT(ctxt != NULL);
1659 rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse);
1660 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
1661 exp->exp_connection->c_peer.peer_id.nid, str);
1662 CDEBUG(D_OTHER, "%s: init llog ctxt for export "LPX64"/%s, group %d\n",
1663 obd->obd_name, exp->exp_connection->c_peer.peer_id.nid,
1664 str, fed->fed_group);
1669 /* nearly identical to mds_connect */
1670 static int filter_connect(struct lustre_handle *conn, struct obd_device *obd,
1671 struct obd_uuid *cluuid,
1672 struct obd_connect_data *data,
1673 unsigned long connect_flags)
1675 struct obd_export *exp;
1676 struct filter_export_data *fed;
1677 struct filter_client_data *fcd = NULL;
1678 struct filter_obd *filter = &obd->u.filter;
1679 struct lvfs_run_ctxt saved;
1683 if (conn == NULL || obd == NULL || cluuid == NULL)
1686 rc = class_connect(conn, obd, cluuid);
1689 exp = class_conn2export(conn);
1690 LASSERT(exp != NULL);
1692 fed = &exp->exp_filter_data;
1694 spin_lock_init(&fed->fed_lock);
1696 /* connection from MDS */
1697 group = connect_flags;
1698 if (obd->obd_replayable) {
1699 OBD_ALLOC(fcd, sizeof(*fcd));
1701 CERROR("filter: out of memory for client data\n");
1702 GOTO(cleanup, rc = -ENOMEM);
1705 memcpy(fcd->fcd_uuid, cluuid, sizeof(fcd->fcd_uuid));
1707 fed->fed_fcd->fcd_group = group;
1708 rc = filter_client_add(obd, filter, fed, -1);
1712 CWARN("%s: Received MDS connection ("LPX64"); group %d\n",
1713 obd->obd_name, exp->exp_handle.h_cookie, group);
1718 if (fed->fed_group != 0 && fed->fed_group != group) {
1719 char str[PTL_NALFMT_SIZE];
1720 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
1721 exp->exp_connection->c_peer.peer_id.nid, str);
1722 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
1723 "earlier; now it's trying to use group %d! This could "
1724 "be a bug in the MDS. Tell CFS.\n",
1725 exp->exp_connection->c_peer.peer_id.nid, str,
1726 fed->fed_group, group);
1727 GOTO(cleanup, rc = -EPROTO);
1729 fed->fed_group = group;
1731 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1732 rc = filter_read_groups(obd, group, 1);
1733 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1735 CERROR("can't read group %u\n", group);
1738 rc = filter_group_set_fs_flags(obd, group);
1740 CERROR("can't set kml flags %u\n", group);
1747 OBD_FREE(fcd, sizeof(*fcd));
1748 class_disconnect(exp, 0);
1750 class_export_put(exp);
1755 static int filter_precleanup(struct obd_device *obd, int flags)
1757 struct filter_group_llog *log;
1758 struct filter_obd *filter;
1762 filter = &obd->u.filter;
1764 spin_lock(&filter->fo_llog_list_lock);
1765 while (!list_empty(&filter->fo_llog_list)) {
1766 log = list_entry(filter->fo_llog_list.next,
1767 struct filter_group_llog, list);
1768 list_del(&log->list);
1769 spin_unlock(&filter->fo_llog_list_lock);
1771 rc = obd_llog_finish(obd, log->llogs, 0);
1773 CERROR("failed to cleanup llogging subsystem for %u\n",
1775 OBD_FREE(log->llogs, sizeof(*(log->llogs)));
1776 OBD_FREE(log, sizeof(*log));
1777 spin_lock(&filter->fo_llog_list_lock);
1779 spin_unlock(&filter->fo_llog_list_lock);
1781 rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1783 CERROR("failed to cleanup llogging subsystem\n");
1788 /* Do extra sanity checks for grant accounting. We do this at connect,
1789 * disconnect, and statfs RPC time, so it shouldn't be too bad. We can
1790 * always get rid of it or turn it off when we know accounting is good. */
1791 static void filter_grant_sanity_check(struct obd_device *obd, const char *func)
1793 struct filter_export_data *fed;
1794 struct obd_export *exp;
1795 obd_size maxsize = obd->obd_osfs.os_blocks * obd->obd_osfs.os_bsize;
1796 obd_size tot_dirty = 0, tot_pending = 0, tot_granted = 0;
1797 obd_size fo_tot_dirty, fo_tot_pending, fo_tot_granted;
1798 int level = D_CACHE;
1800 if (list_empty(&obd->obd_exports))
1803 spin_lock(&obd->obd_osfs_lock);
1804 spin_lock(&obd->obd_dev_lock);
1805 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1806 fed = &exp->exp_filter_data;
1807 if (fed->fed_grant < 0 || fed->fed_pending < 0 ||
1810 if (maxsize > 0) { /* we may not have done a statfs yet */
1811 LASSERTF(fed->fed_grant + fed->fed_pending <= maxsize,
1812 "cli %s/%p %ld+%ld > "LPU64"\n",
1813 exp->exp_client_uuid.uuid, exp,
1814 fed->fed_grant, fed->fed_pending, maxsize);
1815 LASSERTF(fed->fed_dirty <= maxsize,
1816 "cli %s/%p %ld > "LPU64"\n",
1817 exp->exp_client_uuid.uuid, exp,
1818 fed->fed_dirty, maxsize);
1820 CDEBUG(level, "%s: cli %s/%p dirty %ld pend %ld grant %ld\n",
1821 obd->obd_name, exp->exp_client_uuid.uuid, exp,
1822 fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1823 tot_granted += fed->fed_grant + fed->fed_pending;
1824 tot_pending += fed->fed_pending;
1825 tot_dirty += fed->fed_dirty;
1827 fo_tot_granted = obd->u.filter.fo_tot_granted;
1828 fo_tot_pending = obd->u.filter.fo_tot_pending;
1829 fo_tot_dirty = obd->u.filter.fo_tot_dirty;
1830 spin_unlock(&obd->obd_dev_lock);
1831 spin_unlock(&obd->obd_osfs_lock);
1833 /* Do these assertions outside the spinlocks so we don't kill system */
1834 if (tot_granted != fo_tot_granted)
1835 CERROR("%s: tot_granted "LPU64" != fo_tot_granted "LPU64"\n",
1836 func, tot_granted, fo_tot_granted);
1837 if (tot_pending != fo_tot_pending)
1838 CERROR("%s: tot_pending "LPU64" != fo_tot_pending "LPU64"\n",
1839 func, tot_pending, fo_tot_pending);
1840 if (tot_dirty != fo_tot_dirty)
1841 CERROR("%s: tot_dirty "LPU64" != fo_tot_dirty "LPU64"\n",
1842 func, tot_dirty, fo_tot_dirty);
1843 if (tot_pending > tot_granted)
1844 CERROR("%s: tot_pending "LPU64" > tot_granted "LPU64"\n",
1845 func, tot_pending, tot_granted);
1846 if (tot_granted > maxsize)
1847 CERROR("%s: tot_granted "LPU64" > maxsize "LPU64"\n",
1848 func, tot_granted, maxsize);
1849 if (tot_dirty > maxsize)
1850 CERROR("%s: tot_dirty "LPU64" > maxsize "LPU64"\n",
1851 func, tot_dirty, maxsize);
1854 /* Remove this client from the grant accounting totals. We also remove
1855 * the export from the obd device under the osfs and dev locks to ensure
1856 * that the filter_grant_sanity_check() calculations are always valid.
1857 * The client should do something similar when it invalidates its import. */
1858 static void filter_grant_discard(struct obd_export *exp)
1860 struct obd_device *obd = exp->exp_obd;
1861 struct filter_obd *filter = &obd->u.filter;
1862 struct filter_export_data *fed = &exp->exp_filter_data;
1863 int level = D_CACHE;
1865 spin_lock(&obd->obd_osfs_lock);
1866 spin_lock(&exp->exp_obd->obd_dev_lock);
1867 list_del_init(&exp->exp_obd_chain);
1868 spin_unlock(&exp->exp_obd->obd_dev_lock);
1870 if (fed->fed_dirty < 0 || fed->fed_grant < 0 || fed->fed_pending < 0)
1872 CDEBUG(level, "%s: cli %s/%p dirty %ld pend %ld grant %ld\n",
1873 obd->obd_name, exp->exp_client_uuid.uuid, exp,
1874 fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1876 LASSERTF(filter->fo_tot_granted >= fed->fed_grant,
1877 "%s: tot_granted "LPU64" cli %s/%p fed_grant %ld\n",
1878 obd->obd_name, filter->fo_tot_granted,
1879 exp->exp_client_uuid.uuid, exp, fed->fed_grant);
1880 filter->fo_tot_granted -= fed->fed_grant;
1881 LASSERTF(filter->fo_tot_pending >= fed->fed_pending,
1882 "%s: tot_pending "LPU64" cli %s/%p fed_pending %ld\n",
1883 obd->obd_name, filter->fo_tot_pending,
1884 exp->exp_client_uuid.uuid, exp, fed->fed_pending);
1885 LASSERTF(filter->fo_tot_dirty >= fed->fed_dirty,
1886 "%s: tot_dirty "LPU64" cli %s/%p fed_dirty %ld\n",
1887 obd->obd_name, filter->fo_tot_dirty,
1888 exp->exp_client_uuid.uuid, exp, fed->fed_dirty);
1889 filter->fo_tot_dirty -= fed->fed_dirty;
1893 spin_unlock(&obd->obd_osfs_lock);
1896 static int filter_destroy_export(struct obd_export *exp)
1900 if (exp->exp_filter_data.fed_pending)
1901 CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
1902 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
1903 exp, exp->exp_filter_data.fed_pending);
1905 target_destroy_export(exp);
1907 if (exp->exp_obd->obd_replayable)
1908 filter_client_free(exp, exp->exp_flags);
1910 filter_grant_discard(exp);
1911 if (!(exp->exp_flags & OBD_OPT_FORCE))
1912 filter_grant_sanity_check(exp->exp_obd, __FUNCTION__);
1917 static void filter_sync_llogs(struct obd_device *obd, struct obd_export *dexp)
1919 struct filter_group_llog *fglog, *nlog;
1920 struct filter_obd *filter;
1921 int worked = 0, group;
1922 struct llog_ctxt *ctxt;
1925 filter = &obd->u.filter;
1927 /* we can't sync log holding spinlock. also, we do not want to get
1928 * into livelock. so we do following: loop over MDS's exports in
1929 * group order and skip already synced llogs -bzzz */
1931 /* look for group with min. number, but > worked */
1934 spin_lock(&filter->fo_llog_list_lock);
1935 list_for_each_entry(nlog, &filter->fo_llog_list, list) {
1937 if (nlog->group <= worked) {
1938 /* this group is already synced */
1942 if (group < nlog->group) {
1943 /* we have group with smaller number to sync */
1947 /* store current minimal group */
1949 group = nlog->group;
1951 spin_unlock(&filter->fo_llog_list_lock);
1956 worked = fglog->group;
1957 if (fglog->exp && (dexp == fglog->exp || dexp == NULL)) {
1958 ctxt = llog_get_context(fglog->llogs,
1959 LLOG_UNLINK_REPL_CTXT);
1960 LASSERT(ctxt != NULL);
1961 llog_sync(ctxt, fglog->exp);
1963 } while (fglog != NULL);
1966 /* also incredibly similar to mds_disconnect */
1967 static int filter_disconnect(struct obd_export *exp, unsigned long flags)
1969 struct obd_device *obd = exp->exp_obd;
1970 unsigned long irqflags;
1975 class_export_get(exp);
1977 spin_lock_irqsave(&exp->exp_lock, irqflags);
1978 exp->exp_flags = flags;
1979 spin_unlock_irqrestore(&exp->exp_lock, irqflags);
1981 if (!(flags & OBD_OPT_FORCE))
1982 filter_grant_sanity_check(obd, __FUNCTION__);
1983 filter_grant_discard(exp);
1985 /* Disconnect early so that clients can't keep using export */
1986 rc = class_disconnect(exp, flags);
1988 ldlm_cancel_locks_for_export(exp);
1990 fsfilt_sync(obd, obd->u.filter.fo_sb);
1992 /* flush any remaining cancel messages out to the target */
1993 filter_sync_llogs(obd, exp);
1994 class_export_put(exp);
1998 struct dentry *__filter_oa2dentry(struct obd_device *obd,
1999 struct obdo *oa, const char *what)
2001 struct dentry *dchild = NULL;
2004 if (oa->o_valid & OBD_MD_FLGROUP)
2007 dchild = filter_id2dentry(obd, NULL, group, oa->o_id);
2009 if (IS_ERR(dchild)) {
2010 CERROR("%s error looking up object: "LPU64"\n",
2015 if (dchild->d_inode == NULL) {
2016 CERROR("%s: %s on non-existent object: "LPU64"\n",
2017 obd->obd_name, what, oa->o_id);
2019 RETURN(ERR_PTR(-ENOENT));
2025 static int filter_getattr(struct obd_export *exp, struct obdo *oa,
2026 struct lov_stripe_md *md)
2028 struct dentry *dentry = NULL;
2029 struct obd_device *obd;
2033 obd = class_exp2obd(exp);
2035 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2036 exp->exp_handle.h_cookie);
2040 dentry = filter_oa2dentry(obd, oa);
2042 RETURN(PTR_ERR(dentry));
2044 /* Limit the valid bits in the return data to what we actually use */
2045 oa->o_valid = OBD_MD_FLID;
2046 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2052 /* this is called from filter_truncate() until we have filter_punch() */
2053 static int filter_setattr(struct obd_export *exp, struct obdo *oa,
2054 struct lov_stripe_md *md, struct obd_trans_info *oti)
2056 struct lvfs_run_ctxt saved;
2057 struct filter_obd *filter;
2058 struct dentry *dentry;
2060 struct ldlm_res_id res_id = { .name = { oa->o_id, 0, oa->o_gr, 0 } };
2061 struct ldlm_resource *res;
2066 LASSERT(oti != NULL);
2068 dentry = filter_oa2dentry(exp->exp_obd, oa);
2070 RETURN(PTR_ERR(dentry));
2072 filter = &exp->exp_obd->u.filter;
2074 iattr_from_obdo(&iattr, oa, oa->o_valid);
2076 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2079 if (iattr.ia_valid & ATTR_SIZE)
2080 down(&dentry->d_inode->i_sem);
2081 handle = fsfilt_start(exp->exp_obd, dentry->d_inode, FSFILT_OP_SETATTR,
2084 GOTO(out_unlock, rc = PTR_ERR(handle));
2086 /* XXX this could be a rwsem instead, if filter_preprw played along */
2087 if (iattr.ia_valid & ATTR_ATTR_FLAG)
2088 rc = fsfilt_iocontrol(exp->exp_obd, dentry->d_inode, NULL,
2090 (long)&iattr.ia_attr_flags);
2092 rc = fsfilt_setattr(exp->exp_obd, dentry, handle, &iattr, 1);
2093 rc = filter_finish_transno(exp, oti, rc);
2094 rc2 = fsfilt_commit(exp->exp_obd, filter->fo_sb, dentry->d_inode,
2095 handle, exp->exp_sync);
2097 CERROR("error on commit, err = %d\n", rc2);
2102 res = ldlm_resource_get(exp->exp_obd->obd_namespace, NULL,
2103 res_id, LDLM_EXTENT, 0);
2105 if (res->lr_namespace->ns_lvbo &&
2106 res->lr_namespace->ns_lvbo->lvbo_update)
2107 rc = res->lr_namespace->ns_lvbo->lvbo_update(res, NULL,
2109 ldlm_resource_putref(res);
2110 } else if (iattr.ia_valid & ATTR_SIZE) {
2111 /* called from MDS. */
2114 oa->o_valid = OBD_MD_FLID;
2115 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2118 if (iattr.ia_valid & ATTR_SIZE)
2119 up(&dentry->d_inode->i_sem);
2121 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2127 /* XXX identical to osc_unpackmd */
2128 static int filter_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2129 struct lov_mds_md *lmm, int lmm_bytes)
2135 if (lmm_bytes < sizeof (*lmm)) {
2136 CERROR("lov_mds_md too small: %d, need %d\n",
2137 lmm_bytes, (int)sizeof(*lmm));
2140 /* XXX LOV_MAGIC etc check? */
2142 if (lmm->lmm_object_id == cpu_to_le64(0)) {
2143 CERROR("lov_mds_md: zero lmm_object_id\n");
2148 lsm_size = lov_stripe_md_size(1);
2152 if (*lsmp != NULL && lmm == NULL) {
2153 OBD_FREE(*lsmp, lsm_size);
2158 if (*lsmp == NULL) {
2159 OBD_ALLOC(*lsmp, lsm_size);
2163 loi_init((*lsmp)->lsm_oinfo);
2167 /* XXX zero *lsmp? */
2168 (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
2169 LASSERT((*lsmp)->lsm_object_id);
2172 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
2177 static void filter_destroy_precreated(struct obd_export *exp, struct obdo *oa,
2178 struct filter_obd *filter)
2180 struct obdo *doa = NULL;
2185 LASSERT(oa->o_gr != 0);
2186 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2190 CERROR("cannot allocate doa, error %d\n",
2196 doa->o_mode = S_IFREG;
2197 doa->o_gr = oa->o_gr;
2198 doa->o_valid = oa->o_valid & (OBD_MD_FLGROUP | OBD_MD_FLID);
2200 set_bit(doa->o_gr, &filter->fo_destroys_in_progress);
2201 down(&filter->fo_create_locks[doa->o_gr]);
2202 if (!test_bit(doa->o_gr, &filter->fo_destroys_in_progress)) {
2203 CERROR("%s:["LPU64"] destroy_in_progress already cleared\n",
2204 exp->exp_obd->obd_name, doa->o_gr);
2205 up(&filter->fo_create_locks[doa->o_gr]);
2206 GOTO(out_free_doa, 0);
2209 last = filter_last_id(filter, doa->o_gr);
2210 CWARN("%s:["LPU64"] deleting orphan objects from "LPU64" to "LPU64"\n",
2211 exp->exp_obd->obd_name, doa->o_gr, oa->o_id + 1, last);
2212 for (id = oa->o_id + 1; id <= last; id++) {
2214 filter_destroy(exp, doa, NULL, NULL);
2217 CDEBUG(D_HA, "%s:["LPU64"] after destroy: set last_objids = "LPU64"\n",
2218 exp->exp_obd->obd_name, doa->o_gr, oa->o_id);
2220 filter_set_last_id(filter, doa->o_gr, oa->o_id);
2222 clear_bit(doa->o_gr, &filter->fo_destroys_in_progress);
2223 up(&filter->fo_create_locks[doa->o_gr]);
2230 /* returns a negative error or a nonnegative number of files to create */
2231 static int filter_should_precreate(struct obd_export *exp, struct obdo *oa,
2234 struct obd_device *obd = exp->exp_obd;
2235 struct filter_obd *filter = &obd->u.filter;
2239 diff = oa->o_id - filter_last_id(filter, oa->o_gr);
2240 CDEBUG(D_INFO, "filter_last_id() = "LPU64" -> diff = %d\n",
2241 filter_last_id(filter, oa->o_gr), diff);
2243 /* delete orphans request */
2244 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2245 (oa->o_flags & OBD_FL_DELORPHAN)) {
2248 if (-diff > OST_MAX_PRECREATE) {
2249 CERROR("ignoring bogus orphan destroy request: obdid "
2250 LPU64" last_id "LPU64"\n",
2251 oa->o_id, filter_last_id(filter, oa->o_gr));
2254 filter_destroy_precreated(exp, oa, filter);
2255 rc = filter_update_last_objid(obd, group, 0);
2257 CERROR("unable to write lastobjid, but orphans"
2261 /* only precreate if group == 0 and o_id is specfied */
2262 if (!(oa->o_valid & OBD_FL_DELORPHAN) &&
2263 (/*group != 0 ||*/ oa->o_id == 0))
2266 LASSERTF(diff >= 0, LPU64" - "LPU64" = %d\n", oa->o_id,
2267 filter_last_id(filter, oa->o_gr), diff);
2271 static int filter_precreate_rec(struct obd_device *obd, struct dentry *dentry,
2272 int *number, struct obdo *oa)
2277 rc = fsfilt_precreate_rec(obd, dentry, number, oa);
2282 static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2283 unsigned long max_age)
2285 struct filter_obd *filter = &obd->u.filter;
2286 int blockbits = filter->fo_sb->s_blocksize_bits;
2290 /* at least try to account for cached pages. its still racey and
2291 * might be under-reporting if clients haven't announced their
2292 * caches with brw recently */
2293 spin_lock(&obd->obd_osfs_lock);
2294 rc = fsfilt_statfs(obd, filter->fo_sb, max_age);
2295 memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
2296 spin_unlock(&obd->obd_osfs_lock);
2298 CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
2299 " pending "LPU64" free "LPU64" avail "LPU64"\n",
2300 filter->fo_tot_dirty, filter->fo_tot_granted,
2301 filter->fo_tot_pending,
2302 osfs->os_bfree << blockbits, osfs->os_bavail << blockbits);
2304 filter_grant_sanity_check(obd, __FUNCTION__);
2306 osfs->os_bavail -= min(osfs->os_bavail,
2307 (filter->fo_tot_dirty + filter->fo_tot_pending +
2308 osfs->os_bsize -1) >> blockbits);
2313 /* We rely on the fact that only one thread will be creating files in a given
2314 * group at a time, which is why we don't need an atomic filter_get_new_id.
2315 * Even if we had that atomic function, the following race would exist:
2317 * thread 1: gets id x from filter_next_id
2318 * thread 2: gets id (x + 1) from filter_next_id
2319 * thread 2: creates object (x + 1)
2320 * thread 1: tries to create object x, gets -ENOSPC
2322 static int filter_precreate(struct obd_device *obd, struct obdo *oa,
2323 obd_gr group, int *num)
2325 struct dentry *dchild = NULL, *dparent = NULL;
2326 int err = 0, rc = 0, recreate_obj = 0, i;
2327 struct filter_obd *filter;
2328 void *handle = NULL;
2330 struct obd_statfs *osfs;
2331 unsigned long enough_time = jiffies + (obd_timeout * HZ) / 3;
2335 filter = &obd->u.filter;
2337 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2338 (oa->o_flags & OBD_FL_RECREATE_OBJS)) {
2341 OBD_ALLOC(osfs, sizeof(*osfs));
2344 rc = filter_statfs(obd, osfs, jiffies-HZ);
2345 if (rc == 0 && osfs->os_bavail < (osfs->os_blocks >> 10)) {
2346 CDEBUG(D_HA, "OST out of space! avail "LPU64"\n",
2347 osfs->os_bavail<<filter->fo_sb->s_blocksize_bits);
2351 OBD_FREE(osfs, sizeof(*osfs));
2357 CDEBUG(D_HA, "%s: precreating %d objects\n", obd->obd_name, *num);
2359 down(&filter->fo_create_locks[group]);
2361 for (i = 0; i < *num && err == 0; i++) {
2362 int cleanup_phase = 0;
2364 if (test_bit(group, &filter->fo_destroys_in_progress)) {
2365 CWARN("%s: precreate aborted by destroy\n",
2373 last_id = filter_last_id(filter, group);
2374 if (next_id > last_id) {
2375 CERROR("Error: Trying to recreate obj greater"
2376 "than last id "LPD64" > "LPD64"\n",
2378 GOTO(cleanup, rc = -EINVAL);
2381 next_id = filter_last_id(filter, group) + 1;
2384 CDEBUG(D_INFO, "precreate objid "LPU64"\n", next_id);
2386 dparent = filter_parent_lock(obd, group, next_id, &lock);
2387 if (IS_ERR(dparent))
2388 GOTO(cleanup, rc = PTR_ERR(dparent));
2391 /*only do precreate rec record. so clean kml flags here*/
2392 fsfilt_clear_fs_flags(obd, dparent->d_inode, SM_DO_REC);
2394 dchild = filter_id2dentry(obd, dparent, group, next_id);
2396 GOTO(cleanup, rc = PTR_ERR(dchild));
2399 if (dchild->d_inode != NULL) {
2400 /* This would only happen if lastobjid was bad on disk*/
2401 /* Could also happen if recreating missing obj but
2405 CERROR("%s: recreating existing object %.*s?\n",
2406 obd->obd_name, dchild->d_name.len,
2407 dchild->d_name.name);
2409 CERROR("%s: Serious error: objid %.*s already "
2410 "exists; is this filesystem corrupt?\n",
2411 obd->obd_name, dchild->d_name.len,
2412 dchild->d_name.name);
2415 GOTO(cleanup, rc = -EEXIST);
2418 handle = fsfilt_start_log(obd, dparent->d_inode,
2419 FSFILT_OP_CREATE, NULL, 1);
2421 GOTO(cleanup, rc = PTR_ERR(handle));
2424 rc = ll_vfs_create(dparent->d_inode, dchild, S_IFREG, NULL);
2426 CERROR("create failed rc = %d\n", rc);
2430 if (!recreate_obj) {
2431 filter_set_last_id(filter, group, next_id);
2432 err = filter_update_last_objid(obd, group, 0);
2434 CERROR("unable to write lastobjid "
2435 "but file created\n");
2437 fsfilt_set_fs_flags(obd, dparent->d_inode, SM_DO_REC);
2440 switch(cleanup_phase) {
2442 err = fsfilt_commit(obd, filter->fo_sb,
2443 dparent->d_inode, handle, 0);
2445 CERROR("error on commit, err = %d\n", err);
2452 filter_parent_unlock(dparent, lock);
2459 if (time_after(jiffies, enough_time)) {
2460 CDEBUG(D_INODE,"%s: precreate slow - want %d got %d \n",
2461 obd->obd_name, *num, i);
2468 /* check if we have an error after ll_vfs_create(). It is possible that
2469 * there will be say -ENOSPC and we will leak it. */
2471 rc = filter_precreate_rec(obd, dparent, num, oa);
2473 up(&filter->fo_create_locks[group]);
2475 CDEBUG(D_HA, "%s: server last_objid for group "LPU64": "LPU64"\n",
2476 obd->obd_name, group, filter->fo_last_objids[group]);
2478 CDEBUG(D_HA, "%s: filter_precreate() created %d objects\n",
2484 static int filter_create(struct obd_export *exp, struct obdo *oa,
2485 struct lov_stripe_md **ea, struct obd_trans_info *oti)
2487 struct obd_device *obd = NULL;
2488 struct filter_obd *filter;
2489 struct lvfs_run_ctxt saved;
2490 struct lov_stripe_md *lsm = NULL;
2491 struct filter_export_data *fed;
2492 char str[PTL_NALFMT_SIZE];
2493 int group = oa->o_gr, rc = 0, diff, recreate_objs = 0;
2496 if (!(oa->o_valid & OBD_MD_FLGROUP) || group == 0) {
2497 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2498 exp->exp_connection->c_peer.peer_id.nid, str);
2499 CERROR("!!! nid "LPX64"/%s sent invalid object group %d\n",
2500 exp->exp_connection->c_peer.peer_id.nid, str, group);
2504 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
2505 (oa->o_flags & OBD_FL_RECREATE_OBJS))
2509 fed = &exp->exp_filter_data;
2510 filter = &obd->u.filter;
2512 if (fed->fed_group != group && !recreate_objs &&
2513 !(oa->o_valid & OBD_MD_REINT)) {
2514 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2515 exp->exp_connection->c_peer.peer_id.nid, str);
2516 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
2517 "earlier; now it's trying to use group %d! This could "
2518 "be a bug in the MDS. Tell CFS.\n",
2519 exp->exp_connection->c_peer.peer_id.nid, str,
2520 fed->fed_group, group);
2524 CDEBUG(D_INFO, "filter_create(od->o_gr=%d,od->o_id="LPU64")\n",
2529 rc = obd_alloc_memmd(exp, &lsm);
2536 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2538 if (oa->o_valid & OBD_MD_REINT) {
2539 int num = *((int*)oa->o_inline);
2540 rc = filter_precreate(obd, oa, oa->o_gr, &num);
2541 } else if (recreate_objs) {
2542 if (oa->o_id > filter_last_id(&obd->u.filter, group)) {
2543 CERROR("recreate objid "LPU64" > last id "LPU64"\n",
2544 oa->o_id, filter_last_id(&obd->u.filter, group));
2548 rc = filter_precreate(obd, oa, group, &diff);
2551 diff = filter_should_precreate(exp, oa, group);
2553 oa->o_id = filter_last_id(&obd->u.filter, group);
2554 rc = filter_precreate(obd, oa, group, &diff);
2555 oa->o_id = filter_last_id(&obd->u.filter, oa->o_gr);
2556 oa->o_valid = OBD_MD_FLID;
2560 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2561 if (rc && ea != NULL && *ea != lsm) {
2562 obd_free_memmd(exp, &lsm);
2563 } else if (rc == 0 && ea != NULL) {
2564 /* XXX LOV STACKING: the lsm that is passed to us from
2565 * LOV does not have valid lsm_oinfo data structs, so
2566 * don't go touching that. This needs to be fixed in a
2568 lsm->lsm_object_id = oa->o_id;
2569 lsm->lsm_object_gr = oa->o_gr;
2576 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
2577 struct lov_stripe_md *ea, struct obd_trans_info *oti)
2579 int rc, rc2, cleanup_phase = 0, have_prepared = 0;
2580 struct dentry *dchild = NULL, *dparent = NULL;
2581 struct llog_cookie *fcc = NULL;
2582 struct lvfs_run_ctxt saved;
2583 struct filter_obd *filter;
2584 struct obd_device *obd;
2585 void *handle = NULL;
2590 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2593 filter = &obd->u.filter;
2595 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2598 dparent = filter_parent_lock(obd, oa->o_gr, oa->o_id, &lock);
2599 if (IS_ERR(dparent))
2600 GOTO(cleanup, rc = PTR_ERR(dparent));
2603 dchild = filter_id2dentry(obd, dparent, oa->o_gr, oa->o_id);
2605 GOTO(cleanup, rc = PTR_ERR(dchild));
2608 if (dchild->d_inode == NULL) {
2609 CDEBUG(D_INODE, "destroying non-existent object "LPU64"\n",
2611 GOTO(cleanup, rc = -ENOENT);
2614 if (!have_prepared) {
2615 /* If we're really going to destroy the object, get ready
2616 * by getting the clients to discard their cached data.
2618 * We have to drop the parent lock, because
2619 * filter_prepare_destroy will acquire a PW on the object, and
2620 * we don't want to deadlock with an incoming write to the
2621 * object, which has the extent PW and then wants to get the
2622 * parent dentry to do the lookup.
2624 * We dput the child because it's not worth the extra
2625 * complication of condition the above code to skip it on the
2626 * second time through. */
2628 filter_parent_unlock(dparent, lock);
2630 filter_prepare_destroy(obd, oa->o_id, oa->o_gr);
2635 /* Our MDC connection is established by the MDS to us */
2636 if (oa->o_valid & OBD_MD_FLCOOKIE) {
2637 OBD_ALLOC(fcc, sizeof(*fcc));
2639 memcpy(fcc, obdo_logcookie(oa), sizeof(*fcc));
2642 /* we're gonna truncate it first in order to avoid possible
2645 * open trasaction open transaction
2646 * down(i_zombie) down(i_zombie)
2647 * restart transaction
2648 * (see BUG 4180) -bzzz
2650 down(&dchild->d_inode->i_sem);
2651 handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_SETATTR,NULL,1);
2652 if (IS_ERR(handle)) {
2653 up(&dchild->d_inode->i_sem);
2654 GOTO(cleanup, rc = PTR_ERR(handle));
2657 iattr.ia_valid = ATTR_SIZE;
2659 rc = fsfilt_setattr(obd, dchild, handle, &iattr, 1);
2661 rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode, handle, 0);
2662 up(&dchild->d_inode->i_sem);
2666 GOTO(cleanup, rc = rc2);
2668 handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_UNLINK,oti,1);
2670 GOTO(cleanup, rc = PTR_ERR(handle));
2674 rc = filter_destroy_internal(obd, oa->o_id, dparent, dchild);
2677 switch(cleanup_phase) {
2681 fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
2683 filter_cancel_cookies_cb,
2686 fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
2688 filter_cancel_cookies_cb,
2691 rc = filter_finish_transno(exp, oti, rc);
2692 rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode,
2693 handle, exp->exp_sync);
2695 CERROR("error on commit, err = %d\n", rc2);
2702 filter_parent_unlock(dparent, lock);
2704 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2707 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
2714 /* NB start and end are used for punch, but not truncate */
2715 static int filter_truncate(struct obd_export *exp, struct obdo *oa,
2716 struct lov_stripe_md *lsm,
2717 obd_off start, obd_off end,
2718 struct obd_trans_info *oti)
2723 if (end != OBD_OBJECT_EOF)
2724 CERROR("PUNCH not supported, only truncate: end = "LPX64"\n",
2727 CDEBUG(D_INODE, "calling truncate for object "LPU64", valid = "LPU64", "
2728 "o_size = "LPD64"\n", oa->o_id, oa->o_valid, start);
2730 error = filter_setattr(exp, oa, NULL, oti);
2734 static int filter_sync(struct obd_export *exp, struct obdo *oa,
2735 struct lov_stripe_md *lsm, obd_off start, obd_off end)
2737 struct obd_device *obd = exp->exp_obd;
2738 struct lvfs_run_ctxt saved;
2739 struct filter_obd *filter;
2740 struct dentry *dentry;
2744 filter = &obd->u.filter;
2746 /* an objid of zero is taken to mean "sync whole filesystem" */
2747 if (!oa || !(oa->o_valid & OBD_MD_FLID)) {
2748 rc = fsfilt_sync(obd, filter->fo_sb);
2749 /* flush any remaining cancel messages out to the target */
2750 filter_sync_llogs(obd, NULL);
2754 dentry = filter_oa2dentry(obd, oa);
2756 RETURN(PTR_ERR(dentry));
2758 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2760 down(&dentry->d_inode->i_sem);
2761 rc = filemap_fdatasync(dentry->d_inode->i_mapping);
2763 /* just any file to grab fsync method - "file" arg unused */
2764 struct file *file = filter->fo_rcvd_filp;
2766 if (file->f_op && file->f_op->fsync)
2767 rc = file->f_op->fsync(NULL, dentry, 1);
2769 rc2 = filemap_fdatawait(dentry->d_inode->i_mapping);
2773 up(&dentry->d_inode->i_sem);
2775 oa->o_valid = OBD_MD_FLID;
2776 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2778 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2784 static int filter_get_info(struct obd_export *exp, __u32 keylen,
2785 void *key, __u32 *vallen, void *val)
2787 struct filter_export_data *fed = &exp->exp_filter_data;
2788 struct obd_device *obd;
2791 obd = class_exp2obd(exp);
2793 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2794 exp->exp_handle.h_cookie);
2798 if (keylen == strlen("blocksize") &&
2799 memcmp(key, "blocksize", keylen) == 0) {
2800 __u32 *blocksize = val;
2801 *vallen = sizeof(*blocksize);
2802 *blocksize = obd->u.filter.fo_sb->s_blocksize;
2806 if (keylen == strlen("blocksize_bits") &&
2807 memcmp(key, "blocksize_bits", keylen) == 0) {
2808 __u32 *blocksize_bits = val;
2809 *vallen = sizeof(*blocksize_bits);
2810 *blocksize_bits = obd->u.filter.fo_sb->s_blocksize_bits;
2814 if (keylen >= strlen("last_id") && memcmp(key, "last_id", 7) == 0) {
2815 obd_id *last_id = val;
2816 *last_id = filter_last_id(&obd->u.filter, fed->fed_group);
2819 if (keylen >= strlen("reint_log") && memcmp(key, "reint_log", 9) == 0) {
2820 /*Get log_context handle*/
2821 unsigned long *llh_handle = val;
2822 *vallen = sizeof(unsigned long);
2823 *llh_handle = (unsigned long)obd->obd_llog_ctxt[LLOG_REINT_ORIG_CTXT];
2826 if (keylen >= strlen("cache_sb") && memcmp(key, "cache_sb", 8) == 0) {
2827 /*Get log_context handle*/
2828 unsigned long *sb = val;
2829 *vallen = sizeof(unsigned long);
2830 *sb = (unsigned long)obd->u.filter.fo_sb;
2834 CDEBUG(D_IOCTL, "invalid key\n");
2838 struct obd_llogs *filter_grab_llog_for_group(struct obd_device *obd, int group,
2839 struct obd_export *export)
2841 struct filter_group_llog *fglog, *nlog;
2842 char name[32] = "CATLIST";
2843 struct filter_obd *filter;
2844 struct llog_ctxt *ctxt;
2845 struct list_head *cur;
2848 filter = &obd->u.filter;
2850 spin_lock(&filter->fo_llog_list_lock);
2851 list_for_each(cur, &filter->fo_llog_list) {
2852 fglog = list_entry(cur, struct filter_group_llog, list);
2853 if (fglog->group == group) {
2854 if (!(fglog->exp == NULL || fglog->exp == export || export == NULL))
2855 CWARN("%s: export for group %d changes: 0x%p -> 0x%p\n",
2856 obd->obd_name, group, fglog->exp, export);
2857 spin_unlock(&filter->fo_llog_list_lock);
2861 spin_unlock(&filter->fo_llog_list_lock);
2866 OBD_ALLOC(fglog, sizeof(*fglog));
2869 fglog->group = group;
2871 OBD_ALLOC(fglog->llogs, sizeof(struct obd_llogs));
2872 if (fglog->llogs == NULL) {
2873 OBD_FREE(fglog, sizeof(*fglog));
2877 spin_lock(&filter->fo_llog_list_lock);
2878 list_for_each(cur, &filter->fo_llog_list) {
2879 nlog = list_entry(cur, struct filter_group_llog, list);
2880 LASSERT(nlog->group != group);
2882 list_add(&fglog->list, &filter->fo_llog_list);
2883 spin_unlock(&filter->fo_llog_list_lock);
2885 rc = obd_llog_cat_initialize(obd, fglog->llogs, 1, name);
2887 OBD_FREE(fglog->llogs, sizeof(*(fglog->llogs)));
2888 OBD_FREE(fglog, sizeof(*fglog));
2894 fglog->exp = export;
2895 ctxt = llog_get_context(fglog->llogs, LLOG_UNLINK_REPL_CTXT);
2896 LASSERT(ctxt != NULL);
2898 llog_receptor_accept(ctxt, export->exp_imp_reverse);
2900 CDEBUG(D_OTHER, "%s: new llog 0x%p for group %u\n",
2901 obd->obd_name, fglog->llogs, group);
2903 RETURN(fglog->llogs);
2906 int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
2907 int len, void *karg, void *uarg)
2909 struct obd_device *obd = exp->exp_obd;
2910 struct obd_ioctl_data *data = karg;
2914 case OBD_IOC_ABORT_RECOVERY:
2915 target_stop_recovery_thread(obd);
2918 case OBD_IOC_SET_READONLY: {
2920 struct super_block *sb = obd->u.filter.fo_sb;
2921 struct inode *inode = sb->s_root->d_inode;
2922 BDEVNAME_DECLARE_STORAGE(tmp);
2923 CERROR("setting device %s read-only\n",
2924 ll_bdevname(sb, tmp));
2926 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
2928 (void)fsfilt_commit(obd, sb, inode, handle, 1);
2930 ll_set_rdonly(ll_sbdev(obd->u.filter.fo_sb));
2934 case OBD_IOC_CATLOGLIST: {
2935 rc = llog_catalog_list(obd, 1, data);
2938 case OBD_IOC_LLOG_CANCEL:
2939 case OBD_IOC_LLOG_REMOVE:
2940 case OBD_IOC_LLOG_INFO:
2941 case OBD_IOC_LLOG_PRINT: {
2942 /* FIXME to be finished */
2943 RETURN(-EOPNOTSUPP);
2945 struct llog_ctxt *ctxt = NULL;
2947 push_ctxt(&saved, &ctxt->loc_ctxt, NULL);
2948 rc = llog_ioctl(ctxt, cmd, data);
2949 pop_ctxt(&saved, &ctxt->loc_ctxt, NULL);
2962 static struct llog_operations filter_unlink_repl_logops;
2963 static struct llog_operations filter_size_orig_logops = {
2964 lop_setup: llog_obd_origin_setup,
2965 lop_cleanup: llog_catalog_cleanup,
2966 lop_add: llog_catalog_add,
2969 static int filter_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
2970 struct obd_device *tgt, int count,
2971 struct llog_catid *catid)
2973 struct llog_ctxt *ctxt;
2977 filter_unlink_repl_logops = llog_client_ops;
2978 filter_unlink_repl_logops.lop_cancel = llog_obd_repl_cancel;
2979 filter_unlink_repl_logops.lop_connect = llog_repl_connect;
2980 filter_unlink_repl_logops.lop_sync = llog_obd_repl_sync;
2982 rc = obd_llog_setup(obd, llogs, LLOG_UNLINK_REPL_CTXT, tgt, 0, NULL,
2983 &filter_unlink_repl_logops);
2986 /* FIXME - assign unlink_cb for filter's recovery */
2987 ctxt = llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT);
2988 ctxt->llog_proc_cb = filter_recov_log_unlink_cb;
2990 /* FIXME - count should be 1 to setup size log */
2991 rc = obd_llog_setup(obd, llogs, LLOG_SIZE_ORIG_CTXT, tgt, 0,
2992 &catid->lci_logid, &filter_size_orig_logops);
2996 static int filter_llog_finish(struct obd_device *obd,
2997 struct obd_llogs *llogs, int count)
3002 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT));
3006 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_SIZE_ORIG_CTXT));
3010 static int filter_llog_connect(struct obd_export *exp,
3011 struct llogd_conn_body *body)
3013 struct obd_device *obd = exp->exp_obd;
3014 struct llog_ctxt *ctxt;
3015 struct obd_llogs *llog;
3019 CDEBUG(D_OTHER, "handle connect for %s: %u/%u/%u\n", obd->obd_name,
3020 (unsigned) body->lgdc_logid.lgl_ogr,
3021 (unsigned) body->lgdc_logid.lgl_oid,
3022 (unsigned) body->lgdc_logid.lgl_ogen);
3023 llog = filter_grab_llog_for_group(obd, body->lgdc_logid.lgl_ogr, exp);
3024 LASSERT(llog != NULL);
3025 ctxt = llog_get_context(llog, body->lgdc_ctxt_idx);
3026 rc = llog_connect(ctxt, 1, &body->lgdc_logid,
3027 &body->lgdc_gen, NULL);
3029 CERROR("failed to connect\n");
3034 static struct dentry *filter_lvfs_id2dentry(__u64 id, __u32 gen,
3035 __u64 gr, void *data)
3037 return filter_id2dentry(data, NULL, gr, id);
3040 static struct lvfs_callback_ops filter_lvfs_ops = {
3041 l_id2dentry: filter_lvfs_id2dentry,
3044 static struct obd_ops filter_obd_ops = {
3045 .o_owner = THIS_MODULE,
3046 .o_attach = filter_attach,
3047 .o_detach = filter_detach,
3048 .o_get_info = filter_get_info,
3049 .o_setup = filter_setup,
3050 .o_precleanup = filter_precleanup,
3051 .o_cleanup = filter_cleanup,
3052 .o_connect = filter_connect,
3053 .o_connect_post = filter_connect_post,
3054 .o_disconnect = filter_disconnect,
3055 .o_statfs = filter_statfs,
3056 .o_getattr = filter_getattr,
3057 .o_unpackmd = filter_unpackmd,
3058 .o_create = filter_create,
3059 .o_setattr = filter_setattr,
3060 .o_destroy = filter_destroy,
3061 .o_brw = filter_brw,
3062 .o_punch = filter_truncate,
3063 .o_sync = filter_sync,
3064 .o_preprw = filter_preprw,
3065 .o_commitrw = filter_commitrw,
3066 .o_do_cow = filter_do_cow,
3067 .o_write_extents = filter_write_extents,
3068 .o_destroy_export = filter_destroy_export,
3069 .o_llog_init = filter_llog_init,
3070 .o_llog_finish = filter_llog_finish,
3071 .o_llog_connect = filter_llog_connect,
3072 .o_iocontrol = filter_iocontrol,
3075 static struct obd_ops filter_sanobd_ops = {
3076 .o_owner = THIS_MODULE,
3077 .o_attach = filter_attach,
3078 .o_detach = filter_detach,
3079 .o_get_info = filter_get_info,
3080 .o_setup = filter_san_setup,
3081 .o_precleanup = filter_precleanup,
3082 .o_cleanup = filter_cleanup,
3083 .o_connect = filter_connect,
3084 .o_connect_post = filter_connect_post,
3085 .o_disconnect = filter_disconnect,
3086 .o_statfs = filter_statfs,
3087 .o_getattr = filter_getattr,
3088 .o_unpackmd = filter_unpackmd,
3089 .o_create = filter_create,
3090 .o_setattr = filter_setattr,
3091 .o_destroy = filter_destroy,
3092 .o_brw = filter_brw,
3093 .o_punch = filter_truncate,
3094 .o_sync = filter_sync,
3095 .o_preprw = filter_preprw,
3096 .o_commitrw = filter_commitrw,
3097 .o_do_cow = filter_do_cow,
3098 .o_write_extents = filter_write_extents,
3099 .o_san_preprw = filter_san_preprw,
3100 .o_destroy_export = filter_destroy_export,
3101 .o_llog_init = filter_llog_init,
3102 .o_llog_finish = filter_llog_finish,
3103 .o_llog_connect = filter_llog_connect,
3104 .o_iocontrol = filter_iocontrol,
3107 static int __init obdfilter_init(void)
3109 struct lprocfs_static_vars lvars;
3112 printk(KERN_INFO "Lustre: Filtering OBD driver; info@clusterfs.com\n");
3114 lprocfs_init_vars(filter, &lvars);
3116 size = OBDFILTER_CREATED_SCRATCHPAD_ENTRIES *
3117 sizeof(*obdfilter_created_scratchpad);
3119 OBD_ALLOC(obdfilter_created_scratchpad, size);
3120 if (obdfilter_created_scratchpad == NULL) {
3121 CERROR ("Can't allocate scratchpad\n");
3125 rc = class_register_type(&filter_obd_ops, NULL, lvars.module_vars,
3126 OBD_FILTER_DEVICENAME);
3128 OBD_FREE(obdfilter_created_scratchpad, size);
3132 rc = class_register_type(&filter_sanobd_ops, NULL, lvars.module_vars,
3133 OBD_FILTER_SAN_DEVICENAME);
3135 class_unregister_type(OBD_FILTER_DEVICENAME);
3136 OBD_FREE(obdfilter_created_scratchpad, size);
3141 static void __exit obdfilter_exit(void)
3143 class_unregister_type(OBD_FILTER_SAN_DEVICENAME);
3144 class_unregister_type(OBD_FILTER_DEVICENAME);
3145 OBD_FREE(obdfilter_created_scratchpad,
3146 OBDFILTER_CREATED_SCRATCHPAD_ENTRIES *
3147 sizeof(*obdfilter_created_scratchpad));
3150 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
3151 MODULE_DESCRIPTION("Lustre Filtering OBD driver");
3152 MODULE_LICENSE("GPL");
3154 module_init(obdfilter_init);
3155 module_exit(obdfilter_exit);