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 <linux/lustre_sec.h>
64 #include <linux/lustre_audit.h>
65 #include "filter_internal.h"
67 /* Group 0 is no longer a legal group, to catch uninitialized IDs */
68 #define FILTER_MIN_GROUPS 3
70 static struct lvfs_callback_ops filter_lvfs_ops;
72 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
73 struct lov_stripe_md *ea, struct obd_trans_info *);
74 struct obd_llogs *filter_grab_llog_for_group(struct obd_device *,
75 int, struct obd_export *);
77 static void filter_commit_cb(struct obd_device *obd, __u64 transno,
78 void *cb_data, int error)
80 obd_transno_commit_cb(obd, transno, error);
83 /* Assumes caller has already pushed us into the kernel context. */
84 int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti,
87 struct filter_obd *filter = &exp->exp_obd->u.filter;
88 struct filter_export_data *fed = &exp->exp_filter_data;
89 struct filter_client_data *fcd = fed->fed_fcd;
92 int err, log_pri = D_HA;
94 /* Propagate error code. */
98 if (!exp->exp_obd->obd_replayable || oti == NULL)
101 /* we don't allocate new transnos for replayed requests */
102 if (oti->oti_transno == 0) {
103 spin_lock(&filter->fo_translock);
104 last_rcvd = le64_to_cpu(filter->fo_fsd->fsd_last_transno) + 1;
105 filter->fo_fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
106 spin_unlock(&filter->fo_translock);
107 oti->oti_transno = last_rcvd;
109 spin_lock(&filter->fo_translock);
110 last_rcvd = oti->oti_transno;
111 if (last_rcvd > le64_to_cpu(filter->fo_fsd->fsd_last_transno))
112 filter->fo_fsd->fsd_last_transno =
113 cpu_to_le64(last_rcvd);
114 spin_unlock(&filter->fo_translock);
117 fcd->fcd_last_rcvd = cpu_to_le64(last_rcvd);
119 /* could get xid from oti, if it's ever needed */
120 fcd->fcd_last_xid = 0;
122 off = fed->fed_lr_off;
124 fsfilt_add_journal_cb(exp->exp_obd, filter->fo_sb, last_rcvd,
125 oti->oti_handle, filter_commit_cb, NULL);
127 err = fsfilt_write_record(exp->exp_obd, filter->fo_rcvd_filp, fcd,
128 sizeof(*fcd), &off, 0);
135 CDEBUG(log_pri, "wrote trans "LPU64" for client %s at #%d: err = %d\n",
136 last_rcvd, fcd->fcd_uuid, fed->fed_lr_idx, err);
141 void f_dput(struct dentry *dentry)
143 /* Can't go inside filter_ddelete because it can block */
144 CDEBUG(D_INODE, "putting %s: %p, count = %d\n",
145 dentry->d_name.name, dentry, atomic_read(&dentry->d_count) - 1);
146 LASSERT(atomic_read(&dentry->d_count) > 0);
151 /* Add client data to the FILTER. We use a bitmap to locate a free space
152 * in the last_rcvd file if cl_idx is -1 (i.e. a new client).
153 * Otherwise, we have just read the data from the last_rcvd file and
154 * we know its offset. */
155 static int filter_client_add(struct obd_device *obd, struct filter_obd *filter,
156 struct filter_export_data *fed, int cl_idx)
158 unsigned long *bitmap = filter->fo_last_rcvd_slots;
159 int new_client = (cl_idx == -1);
162 LASSERT(bitmap != NULL);
164 /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
165 if (!strcmp((char *)fed->fed_fcd->fcd_uuid, (char *)obd->obd_uuid.uuid))
168 /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
169 * there's no need for extra complication here
172 cl_idx = find_first_zero_bit(bitmap, FILTER_LR_MAX_CLIENTS);
174 if (cl_idx >= FILTER_LR_MAX_CLIENTS) {
175 CERROR("no client slots - fix FILTER_LR_MAX_CLIENTS\n");
178 if (test_and_set_bit(cl_idx, bitmap)) {
179 CERROR("FILTER client %d: found bit is set in bitmap\n",
181 cl_idx = find_next_zero_bit(bitmap,
182 FILTER_LR_MAX_CLIENTS,
187 if (test_and_set_bit(cl_idx, bitmap)) {
188 CERROR("FILTER client %d: bit already set in bitmap!\n",
194 fed->fed_lr_idx = cl_idx;
195 fed->fed_lr_off = le32_to_cpu(filter->fo_fsd->fsd_client_start) +
196 cl_idx * le16_to_cpu(filter->fo_fsd->fsd_client_size);
198 CDEBUG(D_INFO, "client at index %d (%llu) with UUID '%s' added\n",
199 fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
202 struct lvfs_run_ctxt saved;
203 loff_t off = fed->fed_lr_off;
207 CDEBUG(D_INFO, "writing client fcd at idx %u (%llu) (len %u)\n",
208 fed->fed_lr_idx,off,(unsigned int)sizeof(*fed->fed_fcd));
210 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
211 /* Transaction needed to fix bug 1403 */
212 handle = fsfilt_start(obd,
213 filter->fo_rcvd_filp->f_dentry->d_inode,
214 FSFILT_OP_SETATTR, NULL);
215 if (IS_ERR(handle)) {
216 err = PTR_ERR(handle);
217 CERROR("unable to start transaction: rc %d\n", err);
219 err = fsfilt_write_record(obd, filter->fo_rcvd_filp,
221 sizeof(*fed->fed_fcd),
223 fsfilt_commit(obd, filter->fo_sb,
224 filter->fo_rcvd_filp->f_dentry->d_inode,
227 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
230 CERROR("error writing %s client idx %u: rc %d\n",
231 LAST_RCVD, fed->fed_lr_idx, err);
238 static int filter_client_free(struct obd_export *exp, int flags)
240 struct filter_export_data *fed = &exp->exp_filter_data;
241 struct filter_obd *filter = &exp->exp_obd->u.filter;
242 struct obd_device *obd = exp->exp_obd;
243 struct filter_client_data zero_fcd;
244 struct lvfs_run_ctxt saved;
249 if (fed->fed_fcd == NULL)
252 if (flags & OBD_OPT_FAILOVER)
255 /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
256 if (!strcmp((char *)fed->fed_fcd->fcd_uuid, (char *)obd->obd_uuid.uuid))
259 LASSERT(filter->fo_last_rcvd_slots != NULL);
261 off = fed->fed_lr_off;
263 CDEBUG(D_INFO, "freeing client at idx %u (%lld) with UUID '%s'\n",
264 fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
266 /* Clear the bit _after_ zeroing out the client so we don't
267 race with filter_client_add and zero out new clients.*/
268 if (!test_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
269 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
274 memset(&zero_fcd, 0, sizeof zero_fcd);
275 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
276 rc = fsfilt_write_record(obd, filter->fo_rcvd_filp, &zero_fcd,
277 sizeof(zero_fcd), &off, 1);
279 /* update server's transno */
280 filter_update_server_data(obd, filter->fo_rcvd_filp,
282 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
284 CDEBUG(rc == 0 ? D_INFO : D_ERROR,
285 "zeroing disconnecting client %s at idx %u (%llu) in %s rc %d\n",
286 fed->fed_fcd->fcd_uuid, fed->fed_lr_idx, fed->fed_lr_off,
289 if (!test_and_clear_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
290 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
296 OBD_FREE(fed->fed_fcd, sizeof(*fed->fed_fcd));
301 static int filter_free_server_data(struct filter_obd *filter)
303 OBD_FREE(filter->fo_fsd, sizeof(*filter->fo_fsd));
304 filter->fo_fsd = NULL;
305 OBD_FREE(filter->fo_last_rcvd_slots, FILTER_LR_MAX_CLIENTS/8);
306 filter->fo_last_rcvd_slots = NULL;
310 /* assumes caller is already in kernel ctxt */
311 int filter_update_server_data(struct obd_device *obd, struct file *filp,
312 struct filter_server_data *fsd, int force_sync)
318 CDEBUG(D_INODE, "server uuid : %s\n", fsd->fsd_uuid);
319 CDEBUG(D_INODE, "server last_rcvd : "LPU64"\n",
320 le64_to_cpu(fsd->fsd_last_transno));
321 CDEBUG(D_INODE, "server last_mount: "LPU64"\n",
322 le64_to_cpu(fsd->fsd_mount_count));
324 rc = fsfilt_write_record(obd, filp, fsd, sizeof(*fsd), &off,force_sync);
326 CERROR("error writing filter_server_data: rc = %d\n", rc);
331 int filter_update_last_objid(struct obd_device *obd, obd_gr group,
334 struct filter_obd *filter = &obd->u.filter;
340 if (filter->fo_last_objid_files[group] == NULL) {
341 CERROR("Object group "LPU64" not fully setup; not updating "
342 "last_objid\n", group);
346 CDEBUG(D_INODE, "server last_objid for group "LPU64": "LPU64"\n",
347 group, filter->fo_last_objids[group]);
349 tmp = cpu_to_le64(filter->fo_last_objids[group]);
350 rc = fsfilt_write_record(obd, filter->fo_last_objid_files[group],
351 &tmp, sizeof(tmp), &off, force_sync);
353 CERROR("error writing group "LPU64" last objid: rc = %d\n",
358 /* assumes caller has already in kernel ctxt */
359 static int filter_init_server_data(struct obd_device *obd, struct file * filp)
361 struct filter_obd *filter = &obd->u.filter;
362 struct filter_server_data *fsd;
363 struct filter_client_data *fcd = NULL;
364 struct inode *inode = filp->f_dentry->d_inode;
365 unsigned long last_rcvd_size = inode->i_size;
371 /* ensure padding in the struct is the correct size */
372 LASSERT (offsetof(struct filter_server_data, fsd_padding) +
373 sizeof(fsd->fsd_padding) == FILTER_LR_SERVER_SIZE);
374 LASSERT (offsetof(struct filter_client_data, fcd_padding) +
375 sizeof(fcd->fcd_padding) == FILTER_LR_CLIENT_SIZE);
377 OBD_ALLOC(fsd, sizeof(*fsd));
380 filter->fo_fsd = fsd;
382 OBD_ALLOC(filter->fo_last_rcvd_slots, FILTER_LR_MAX_CLIENTS/8);
383 if (filter->fo_last_rcvd_slots == NULL) {
384 OBD_FREE(fsd, sizeof(*fsd));
388 if (last_rcvd_size == 0) {
389 CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD);
391 memcpy(fsd->fsd_uuid, obd->obd_uuid.uuid,sizeof(fsd->fsd_uuid));
392 fsd->fsd_last_transno = 0;
393 mount_count = fsd->fsd_mount_count = 0;
394 fsd->fsd_server_size = cpu_to_le32(FILTER_LR_SERVER_SIZE);
395 fsd->fsd_client_start = cpu_to_le32(FILTER_LR_CLIENT_START);
396 fsd->fsd_client_size = cpu_to_le16(FILTER_LR_CLIENT_SIZE);
397 fsd->fsd_subdir_count = cpu_to_le16(FILTER_SUBDIR_COUNT);
398 filter->fo_subdir_count = FILTER_SUBDIR_COUNT;
400 rc = fsfilt_read_record(obd, filp, fsd, sizeof(*fsd), &off);
402 CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
406 if (strcmp((char *)fsd->fsd_uuid, (char *)obd->obd_uuid.uuid)) {
407 CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
408 obd->obd_uuid.uuid, fsd->fsd_uuid);
409 GOTO(err_fsd, rc = -EINVAL);
411 mount_count = le64_to_cpu(fsd->fsd_mount_count);
412 filter->fo_subdir_count = le16_to_cpu(fsd->fsd_subdir_count);
415 if (fsd->fsd_feature_incompat & ~cpu_to_le32(FILTER_INCOMPAT_SUPP)) {
416 CERROR("unsupported feature %x\n",
417 le32_to_cpu(fsd->fsd_feature_incompat) &
418 ~FILTER_INCOMPAT_SUPP);
419 GOTO(err_fsd, rc = -EINVAL);
421 if (fsd->fsd_feature_rocompat & ~cpu_to_le32(FILTER_ROCOMPAT_SUPP)) {
422 CERROR("read-only feature %x\n",
423 le32_to_cpu(fsd->fsd_feature_rocompat) &
424 ~FILTER_ROCOMPAT_SUPP);
425 /* Do something like remount filesystem read-only */
426 GOTO(err_fsd, rc = -EINVAL);
429 CDEBUG(D_INODE, "%s: server last_rcvd : "LPU64"\n",
430 obd->obd_name, le64_to_cpu(fsd->fsd_last_transno));
431 CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
432 obd->obd_name, mount_count + 1);
433 CDEBUG(D_INODE, "%s: server data size: %u\n",
434 obd->obd_name, le32_to_cpu(fsd->fsd_server_size));
435 CDEBUG(D_INODE, "%s: per-client data start: %u\n",
436 obd->obd_name, le32_to_cpu(fsd->fsd_client_start));
437 CDEBUG(D_INODE, "%s: per-client data size: %u\n",
438 obd->obd_name, le32_to_cpu(fsd->fsd_client_size));
439 CDEBUG(D_INODE, "%s: server subdir_count: %u\n",
440 obd->obd_name, le16_to_cpu(fsd->fsd_subdir_count));
441 CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
442 last_rcvd_size <= le32_to_cpu(fsd->fsd_client_start) ? 0 :
443 (last_rcvd_size - le32_to_cpu(fsd->fsd_client_start)) /
444 le16_to_cpu(fsd->fsd_client_size));
446 if (!obd->obd_replayable) {
447 CWARN("%s: recovery support OFF\n", obd->obd_name);
451 for (cl_idx = 0, off = le32_to_cpu(fsd->fsd_client_start);
452 off < last_rcvd_size; cl_idx++) {
454 struct obd_export *exp;
455 struct filter_export_data *fed;
458 OBD_ALLOC(fcd, sizeof(*fcd));
460 GOTO(err_client, rc = -ENOMEM);
463 /* Don't assume off is incremented properly by
464 * fsfilt_read_record(), in case sizeof(*fcd)
465 * isn't the same as fsd->fsd_client_size. */
466 off = le32_to_cpu(fsd->fsd_client_start) +
467 cl_idx * le16_to_cpu(fsd->fsd_client_size);
468 rc = fsfilt_read_record(obd, filp, fcd, sizeof(*fcd), &off);
470 CERROR("error reading FILT %s idx %d off %llu: rc %d\n",
471 LAST_RCVD, cl_idx, off, rc);
472 break; /* read error shouldn't cause startup to fail */
475 if (fcd->fcd_uuid[0] == '\0') {
476 CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
481 last_rcvd = le64_to_cpu(fcd->fcd_last_rcvd);
483 /* These exports are cleaned up by filter_disconnect(), so they
484 * need to be set up like real exports as filter_connect() does.
486 exp = class_new_export(obd);
487 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64" "
488 "srv lr: "LPU64" fcd_group %d \n", fcd->fcd_uuid, cl_idx,
489 last_rcvd, le64_to_cpu(fsd->fsd_last_transno),
490 le32_to_cpu(fcd->fcd_group));
492 GOTO(err_client, rc = -ENOMEM);
494 memcpy(&exp->exp_client_uuid.uuid, fcd->fcd_uuid,
495 sizeof exp->exp_client_uuid.uuid);
496 fed = &exp->exp_filter_data;
498 fed->fed_group = le32_to_cpu(fcd->fcd_group);
499 filter_client_add(obd, filter, fed, cl_idx);
500 /* create helper if export init gets more complex */
501 spin_lock_init(&fed->fed_lock);
504 exp->exp_connected = 0;
505 exp->exp_req_replay_needed = 1;
506 exp->exp_lock_replay_needed = 1;
507 atomic_inc(&obd->obd_req_replay_clients);
508 atomic_inc(&obd->obd_lock_replay_clients);
509 obd->obd_recoverable_clients++;
510 obd->obd_max_recoverable_clients++;
511 class_export_put(exp);
513 CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
516 if (last_rcvd > le64_to_cpu(fsd->fsd_last_transno))
517 fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
521 obd->obd_last_committed = le64_to_cpu(fsd->fsd_last_transno);
523 if (obd->obd_recoverable_clients) {
524 CWARN("RECOVERY: service %s, %d recoverable clients, "
525 "last_transno "LPU64"\n", obd->obd_name,
526 obd->obd_recoverable_clients,
527 le64_to_cpu(fsd->fsd_last_transno));
528 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
529 target_start_recovery_thread(obd, ost_handle);
533 OBD_FREE(fcd, sizeof(*fcd));
536 filter->fo_mount_count = mount_count + 1;
537 fsd->fsd_mount_count = cpu_to_le64(filter->fo_mount_count);
539 /* save it, so mount count and last_transno is current */
540 rc = filter_update_server_data(obd, filp, filter->fo_fsd, 1);
542 GOTO(err_client, rc);
546 class_disconnect_exports(obd, 0);
548 filter_free_server_data(filter);
552 static int filter_cleanup_groups(struct obd_device *obd)
554 struct filter_obd *filter = &obd->u.filter;
555 struct dentry *dentry;
559 for (i = 0; i < filter->fo_group_count; i++) {
560 if (filter->fo_subdirs != NULL) {
561 for (k = 0; k < filter->fo_subdir_count; k++) {
562 dentry = filter->fo_subdirs[i].dentry[k];
566 filter->fo_subdirs[i].dentry[k] = NULL;
569 if (filter->fo_last_objid_files[i] != NULL) {
570 filp_close(filter->fo_last_objid_files[i], 0);
571 filter->fo_last_objid_files[i] = NULL;
573 if (filter->fo_groups[i] != NULL) {
574 dput(filter->fo_groups[i]);
575 filter->fo_groups[i] = NULL;
578 if (filter->fo_subdirs != NULL)
579 OBD_FREE(filter->fo_subdirs,
580 filter->fo_group_count * sizeof(*filter->fo_subdirs));
581 if (filter->fo_groups != NULL)
582 OBD_FREE(filter->fo_groups,
583 filter->fo_group_count * sizeof(*filter->fo_groups));
584 if (filter->fo_last_objids != NULL)
585 OBD_FREE(filter->fo_last_objids,
586 filter->fo_group_count * sizeof(__u64));
587 if (filter->fo_last_objid_files != NULL)
588 OBD_FREE(filter->fo_last_objid_files,
589 filter->fo_group_count * sizeof(struct file *));
590 f_dput(filter->fo_dentry_O);
594 static int filter_update_last_group(struct obd_device *obd, int group)
596 struct filter_obd *filter = &obd->u.filter;
597 struct file *filp = NULL;
598 int last_group = 0, rc;
602 if (group <= filter->fo_committed_group)
605 filp = filp_open("LAST_GROUP", O_RDWR, 0700);
609 CERROR("cannot open LAST_GROUP: rc = %d\n", rc);
613 rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
615 CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n",rc);
618 LASSERT(off == 0 || last_group >= FILTER_MIN_GROUPS);
619 CDEBUG(D_INODE, "%s: previous %d, new %d\n",
620 obd->obd_name, last_group, group);
624 /* must be sync: bXXXX */
625 rc = fsfilt_write_record(obd, filp, &last_group, sizeof(__u32), &off, 1);
627 CDEBUG(D_INODE, "error updating LAST_GROUP: rc %d\n", rc);
631 filter->fo_committed_group = group;
638 static int filter_read_group_internal(struct obd_device *obd, int group,
641 struct filter_obd *filter = &obd->u.filter;
642 __u64 *new_objids = NULL;
643 struct filter_subdirs *new_subdirs = NULL, *tmp_subdirs = NULL;
644 struct dentry **new_groups = NULL;
645 struct file **new_files = NULL;
646 struct dentry *dentry;
648 int old_count = filter->fo_group_count, rc, stage = 0, i;
653 snprintf(name, 24, "%d", group);
657 dentry = ll_lookup_one_len(name, filter->fo_dentry_O,
659 if (IS_ERR(dentry)) {
660 CERROR("Cannot lookup expected object group %d: %ld\n",
661 group, PTR_ERR(dentry));
662 RETURN(PTR_ERR(dentry));
665 dentry = simple_mkdir(filter->fo_dentry_O, name, 0700, 1);
666 if (IS_ERR(dentry)) {
667 CERROR("cannot lookup/create O/%s: rc = %ld\n", name,
669 RETURN(PTR_ERR(dentry));
674 snprintf(name, 24, "O/%d/LAST_ID", group);
676 filp = filp_open(name, O_CREAT | O_RDWR, 0700);
678 CERROR("cannot create %s: rc = %ld\n", name, PTR_ERR(filp));
679 GOTO(cleanup, rc = PTR_ERR(filp));
683 rc = fsfilt_read_record(obd, filp, &last_objid, sizeof(__u64), &off);
685 CDEBUG(D_INODE, "error reading %s: rc %d\n", name, rc);
689 if (filter->fo_subdir_count) {
690 OBD_ALLOC(tmp_subdirs, sizeof(*tmp_subdirs));
691 if (tmp_subdirs == NULL)
692 GOTO(cleanup, rc = -ENOMEM);
695 for (i = 0; i < filter->fo_subdir_count; i++) {
697 snprintf(dir, sizeof(dir), "d%u", i);
699 tmp_subdirs->dentry[i] =
700 simple_mkdir(dentry, dir, 0700, 1);
701 if (IS_ERR(tmp_subdirs->dentry[i])) {
702 rc = PTR_ERR(tmp_subdirs->dentry[i]);
703 CERROR("can't lookup/create O/%d/%s: rc = %d\n",
707 CDEBUG(D_INODE, "got/created O/%d/%s: %p\n", group, dir,
708 tmp_subdirs->dentry[i]);
712 /* 'group' is an index; we need an array of length 'group + 1' */
713 if (group + 1 > old_count) {
715 OBD_ALLOC(new_objids, len * sizeof(*new_objids));
716 OBD_ALLOC(new_subdirs, len * sizeof(*new_subdirs));
717 OBD_ALLOC(new_groups, len * sizeof(*new_groups));
718 OBD_ALLOC(new_files, len * sizeof(*new_files));
720 if (new_objids == NULL || new_subdirs == NULL ||
721 new_groups == NULL || new_files == NULL)
722 GOTO(cleanup, rc = -ENOMEM);
724 memcpy(new_objids, filter->fo_last_objids,
725 old_count * sizeof(*new_objids));
726 memcpy(new_subdirs, filter->fo_subdirs,
727 old_count * sizeof(*new_subdirs));
728 memcpy(new_groups, filter->fo_groups,
729 old_count * sizeof(*new_groups));
730 memcpy(new_files, filter->fo_last_objid_files,
731 old_count * sizeof(*new_files));
734 OBD_FREE(filter->fo_last_objids,
735 old_count * sizeof(*new_objids));
736 OBD_FREE(filter->fo_subdirs,
737 old_count * sizeof(*new_subdirs));
738 OBD_FREE(filter->fo_groups,
739 old_count * sizeof(*new_groups));
740 OBD_FREE(filter->fo_last_objid_files,
741 old_count * sizeof(*new_files));
743 filter->fo_last_objids = new_objids;
744 filter->fo_subdirs = new_subdirs;
745 filter->fo_groups = new_groups;
746 filter->fo_last_objid_files = new_files;
747 filter->fo_group_count = len;
750 filter->fo_groups[group] = dentry;
751 filter->fo_last_objid_files[group] = filp;
752 if (filter->fo_subdir_count) {
753 filter->fo_subdirs[group] = *tmp_subdirs;
754 OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
757 filter_update_last_group(obd, group);
759 if (filp->f_dentry->d_inode->i_size == 0) {
760 filter->fo_last_objids[group] = FILTER_INIT_OBJID;
764 filter->fo_last_objids[group] = le64_to_cpu(last_objid);
765 CDEBUG(D_INODE, "%s: server last_objid group %d: "LPU64"\n",
766 obd->obd_name, group, last_objid);
771 if (new_objids != NULL)
772 OBD_FREE(new_objids, group * sizeof(*new_objids));
773 if (new_subdirs != NULL)
774 OBD_FREE(new_subdirs, group * sizeof(*new_subdirs));
775 if (new_groups != NULL)
776 OBD_FREE(new_groups, group * sizeof(*new_groups));
777 if (new_files != NULL)
778 OBD_FREE(new_files, group * sizeof(*new_files));
780 if (filter->fo_subdir_count) {
781 for (i = 0; i < filter->fo_subdir_count; i++) {
782 if (tmp_subdirs->dentry[i] != NULL)
783 dput(tmp_subdirs->dentry[i]);
785 OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
795 static int filter_read_groups(struct obd_device *obd, int last_group,
798 struct filter_obd *filter = &obd->u.filter;
799 int old_count, group, rc = 0;
801 down(&filter->fo_init_lock);
802 old_count = filter->fo_group_count;
803 for (group = old_count; group <= last_group; group++) {
805 continue; /* no group zero */
807 rc = filter_read_group_internal(obd, group, create);
811 up(&filter->fo_init_lock);
815 static int filter_prep_groups(struct obd_device *obd)
817 struct filter_obd *filter = &obd->u.filter;
818 struct dentry *dentry, *O_dentry;
819 int rc = 0, cleanup_phase = 0;
820 struct file *filp = NULL;
825 O_dentry = simple_mkdir(current->fs->pwd, "O", 0700, 1);
826 CDEBUG(D_INODE, "got/created O: %p\n", O_dentry);
827 if (IS_ERR(O_dentry)) {
828 rc = PTR_ERR(O_dentry);
829 CERROR("cannot open/create O: rc = %d\n", rc);
832 filter->fo_dentry_O = O_dentry;
833 cleanup_phase = 1; /* O_dentry */
835 /* Lookup "R" to tell if we're on an old OST FS and need to convert
836 * from O/R/<dir>/<objid> to O/0/<dir>/<objid>. This can be removed
837 * some time post 1.0 when all old-style OSTs have converted along
838 * with the init_objid hack. */
839 dentry = ll_lookup_one_len("R", O_dentry, 1);
841 GOTO(cleanup, rc = PTR_ERR(dentry));
842 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
843 struct dentry *O0_dentry = lookup_one_len("0", O_dentry, 1);
846 CWARN("converting OST to new object layout\n");
847 if (IS_ERR(O0_dentry)) {
848 rc = PTR_ERR(O0_dentry);
849 CERROR("error looking up O/0: rc %d\n", rc);
853 if (O0_dentry->d_inode) {
854 CERROR("Both O/R and O/0 exist. Fix manually.\n");
855 GOTO(cleanup_O0, rc = -EEXIST);
858 down(&O_dentry->d_inode->i_sem);
859 rc = vfs_rename(O_dentry->d_inode, dentry,
860 O_dentry->d_inode, O0_dentry);
861 up(&O_dentry->d_inode->i_sem);
864 CERROR("error renaming O/R to O/0: rc %d\n", rc);
865 GOTO(cleanup_O0, rc);
867 filter->fo_fsd->fsd_feature_incompat |=
868 cpu_to_le32(FILTER_INCOMPAT_GROUPS);
869 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
871 GOTO(cleanup_O0, rc);
883 cleanup_phase = 2; /* groups */
885 /* we have to initialize all groups before first connections from
886 * clients because they may send create/destroy for any group -bzzz */
887 filp = filp_open("LAST_GROUP", O_CREAT | O_RDWR, 0700);
889 CERROR("cannot create LAST_GROUP: rc = %ld\n", PTR_ERR(filp));
890 GOTO(cleanup, rc = PTR_ERR(filp));
892 cleanup_phase = 3; /* filp */
894 rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
896 CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n", rc);
900 last_group = FILTER_MIN_GROUPS;
902 LASSERT(last_group >= FILTER_MIN_GROUPS);
905 CWARN("%s: initialize groups [%d,%d]\n", obd->obd_name,
906 FILTER_MIN_GROUPS, last_group);
907 filter->fo_committed_group = last_group;
908 rc = filter_read_groups(obd, last_group, 1);
916 switch (cleanup_phase) {
920 filter_cleanup_groups(obd);
922 f_dput(filter->fo_dentry_O);
923 filter->fo_dentry_O = NULL;
930 /* setup the object store with correct subdirectories */
931 static int filter_prep(struct obd_device *obd)
933 struct lvfs_run_ctxt saved;
934 struct filter_obd *filter = &obd->u.filter;
940 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
941 file = filp_open(LAST_RCVD, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
942 if (!file || IS_ERR(file)) {
944 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
949 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
950 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
951 file->f_dentry->d_inode->i_mode);
952 GOTO(err_filp, rc = -ENOENT);
955 /* steal operations */
956 inode = file->f_dentry->d_inode;
957 filter->fo_fop = file->f_op;
958 filter->fo_iop = inode->i_op;
959 filter->fo_aops = inode->i_mapping->a_ops;
961 rc = filter_init_server_data(obd, file);
963 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
966 filter->fo_rcvd_filp = file;
968 rc = filter_prep_groups(obd);
970 GOTO(err_server_data, rc);
973 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
978 //class_disconnect_exports(obd, 0);
979 filter_free_server_data(filter);
981 if (filp_close(file, 0))
982 CERROR("can't close %s after error\n", LAST_RCVD);
983 filter->fo_rcvd_filp = NULL;
987 /* cleanup the filter: write last used object id to status file */
988 static void filter_post(struct obd_device *obd)
990 struct lvfs_run_ctxt saved;
991 struct filter_obd *filter = &obd->u.filter;
994 /* XXX: filter_update_lastobjid used to call fsync_dev. It might be
995 * best to start a transaction with h_sync, because we removed this
998 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
999 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
1002 CERROR("error writing server data: rc = %d\n", rc);
1004 for (i = 1; i < filter->fo_group_count; i++) {
1005 rc = filter_update_last_objid(obd, i,
1006 (i == filter->fo_group_count - 1));
1008 CERROR("error writing group %d lastobjid: rc = %d\n",
1012 rc = filp_close(filter->fo_rcvd_filp, 0);
1013 filter->fo_rcvd_filp = NULL;
1015 CERROR("error closing %s: rc = %d\n", LAST_RCVD, rc);
1017 filter_cleanup_groups(obd);
1018 filter_free_server_data(filter);
1019 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1022 static void filter_set_last_id(struct filter_obd *filter, int group, obd_id id)
1024 LASSERT(filter->fo_fsd != NULL);
1026 LASSERT(group < filter->fo_group_count);
1028 spin_lock(&filter->fo_objidlock);
1029 filter->fo_last_objids[group] = id;
1030 spin_unlock(&filter->fo_objidlock);
1033 __u64 filter_last_id(struct filter_obd *filter, int group)
1036 LASSERT(filter->fo_fsd != NULL);
1038 LASSERT(group < filter->fo_group_count);
1040 spin_lock(&filter->fo_objidlock);
1041 id = filter->fo_last_objids[group];
1042 spin_unlock(&filter->fo_objidlock);
1047 static void filter_save_last_id(struct filter_obd *filter, int group, obd_id id)
1050 LASSERT(group < filter->fo_group_count);
1052 spin_lock(&filter->fo_lastidlock);
1053 if (id > filter_last_id(filter, group))
1054 filter_set_last_id(filter, group, id);
1055 spin_unlock(&filter->fo_lastidlock);
1058 /* direct cut-n-paste of mds_blocking_ast() */
1059 static int filter_blocking_ast(struct ldlm_lock *lock,
1060 struct ldlm_lock_desc *desc,
1061 void *data, int flag)
1066 if (flag == LDLM_CB_CANCELING) {
1067 /* Don't need to do anything here. */
1071 /* XXX layering violation! -phil */
1072 lock_res_and_lock(lock);
1073 /* Get this: if filter_blocking_ast is racing with ldlm_intent_policy,
1074 * such that filter_blocking_ast is called just before l_i_p takes the
1075 * ns_lock, then by the time we get the lock, we might not be the
1076 * correct blocking function anymore. So check, and return early, if
1078 if (lock->l_blocking_ast != filter_blocking_ast) {
1079 unlock_res_and_lock(lock);
1083 lock->l_flags |= LDLM_FL_CBPENDING;
1084 do_ast = (!lock->l_readers && !lock->l_writers);
1085 unlock_res_and_lock(lock);
1088 struct lustre_handle lockh;
1091 LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
1092 ldlm_lock2handle(lock, &lockh);
1093 rc = ldlm_cli_cancel(&lockh);
1095 CERROR("ldlm_cli_cancel: %d\n", rc);
1097 LDLM_DEBUG(lock, "Lock still has references, will be "
1103 extern void *lock_dir(struct inode *dir, struct qstr *name);
1104 extern void unlock_dir(struct inode *dir, void *lock);
1106 static void *filter_lock_dentry(struct obd_device *obd,
1107 struct dentry *dparent,
1115 len = sprintf(name, LPU64, id);
1116 qstr_assign(&qstr, (char *)name, len);
1117 return lock_dir(dparent->d_inode, &qstr);
1119 down(&dparent->d_inode->i_sem);
1124 /* We never dget the object parent, so DON'T dput it either */
1125 static void filter_parent_unlock(struct dentry *dparent, void *lock)
1128 LASSERT(lock != NULL);
1129 unlock_dir(dparent->d_inode, lock);
1131 up(&dparent->d_inode->i_sem);
1135 /* We never dget the object parent, so DON'T dput it either */
1136 struct dentry *filter_parent(struct obd_device *obd, obd_gr group, obd_id objid)
1138 struct filter_obd *filter = &obd->u.filter;
1139 LASSERT(group < filter->fo_group_count);
1142 if (filter->fo_subdir_count == 0)
1143 return filter->fo_groups[group];
1145 return filter->fo_subdirs[group].dentry[objid & (filter->fo_subdir_count - 1)];
1148 /* We never dget the object parent, so DON'T dput it either */
1149 struct dentry *filter_parent_lock(struct obd_device *obd, obd_gr group,
1150 obd_id objid, void **lock)
1152 unsigned long now = jiffies;
1153 struct dentry *dparent = filter_parent(obd, group, objid);
1155 if (IS_ERR(dparent))
1159 LASSERT(dparent->d_inode);
1161 *lock = filter_lock_dentry(obd, dparent, objid);
1162 fsfilt_check_slow(now, obd_timeout, "parent lock");
1166 /* How to get files, dentries, inodes from object id's.
1168 * If dir_dentry is passed, the caller has already locked the parent
1169 * appropriately for this operation (normally a write lock). If
1170 * dir_dentry is NULL, we do a read lock while we do the lookup to
1171 * avoid races with create/destroy and such changing the directory
1172 * internal to the filesystem code. */
1173 struct dentry *filter_id2dentry(struct obd_device *obd,
1174 struct dentry *dir_dentry,
1175 obd_gr group, obd_id id)
1177 struct dentry *dparent = dir_dentry;
1178 struct dentry *dchild;
1185 CERROR("fatal: invalid object id 0\n");
1186 RETURN(ERR_PTR(-ESTALE));
1189 len = sprintf(name, LPU64, id);
1190 if (dir_dentry == NULL) {
1191 dparent = filter_parent_lock(obd, group, id, &lock);
1192 if (IS_ERR(dparent)) {
1193 CERROR("%s: error getting object "LPU64":"LPU64
1194 " parent: rc %ld\n", obd->obd_name,
1195 id, group, PTR_ERR(dparent));
1199 CDEBUG(D_INODE, "looking up object O/%.*s/%s\n",
1200 dparent->d_name.len, dparent->d_name.name, name);
1201 dchild = /*ll_*/lookup_one_len(name, dparent, len);
1202 if (dir_dentry == NULL)
1203 filter_parent_unlock(dparent, lock);
1204 if (IS_ERR(dchild)) {
1205 CERROR("%s: child lookup error %ld\n", obd->obd_name,
1210 if (dchild->d_inode != NULL && is_bad_inode(dchild->d_inode)) {
1211 CERROR("%s: got bad inode "LPU64"\n", obd->obd_name, id);
1213 RETURN(ERR_PTR(-ENOENT));
1216 CDEBUG(D_INODE, "got child objid %s: %p, count = %d\n",
1217 name, dchild, atomic_read(&dchild->d_count));
1219 LASSERT(atomic_read(&dchild->d_count) > 0);
1224 static int filter_prepare_destroy(struct obd_device *obd, obd_id objid,
1227 struct lustre_handle lockh;
1228 int flags = LDLM_AST_DISCARD_DATA, rc;
1229 struct ldlm_res_id res_id = { .name = { objid, 0, group, 0 } };
1230 ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
1233 /* Tell the clients that the object is gone now and that they should
1234 * throw away any cached pages. */
1235 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, res_id,
1236 LDLM_EXTENT, &policy, LCK_PW,
1237 &flags, filter_blocking_ast, ldlm_completion_ast,
1238 NULL, NULL, NULL, 0, NULL, &lockh);
1240 /* We only care about the side-effects, just drop the lock. */
1242 ldlm_lock_decref(&lockh, LCK_PW);
1247 /* Caller must hold LCK_PW on parent and push us into kernel context.
1248 * Caller is also required to ensure that dchild->d_inode exists. */
1249 static int filter_destroy_internal(struct obd_device *obd, obd_id objid,
1250 struct dentry *dparent,
1251 struct dentry *dchild)
1253 struct inode *inode = dchild->d_inode;
1257 if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) {
1258 CERROR("destroying objid %.*s nlink = %lu, count = %d\n",
1259 dchild->d_name.len, dchild->d_name.name,
1260 (unsigned long)inode->i_nlink,
1261 atomic_read(&inode->i_count));
1264 rc = vfs_unlink(dparent->d_inode, dchild);
1267 CERROR("error unlinking objid %.*s: rc %d\n",
1268 dchild->d_name.len, dchild->d_name.name, rc);
1273 static int filter_intent_policy(struct ldlm_namespace *ns,
1274 struct ldlm_lock **lockp, void *req_cookie,
1275 ldlm_mode_t mode, int flags, void *data)
1277 struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1278 struct ptlrpc_request *req = req_cookie;
1279 struct ldlm_lock *lock = *lockp, *l = NULL;
1280 struct ldlm_resource *res = lock->l_resource;
1281 ldlm_processing_policy policy;
1282 struct ost_lvb *res_lvb, *reply_lvb;
1283 struct ldlm_reply *rep;
1284 struct list_head *tmp;
1286 int tmpflags = 0, rc, repsize[2] = {sizeof(*rep), sizeof(*reply_lvb)};
1289 policy = ldlm_get_processing_policy(res);
1290 LASSERT(policy != NULL);
1291 LASSERT(req != NULL);
1293 rc = lustre_pack_reply(req, 2, repsize, NULL);
1295 RETURN(req->rq_status = rc);
1297 rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*rep));
1298 LASSERT(rep != NULL);
1300 reply_lvb = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*reply_lvb));
1301 LASSERT(reply_lvb != NULL);
1303 //fixup_handle_for_resent_req(req, lock, &lockh);
1305 /* If we grant any lock at all, it will be a whole-file read lock.
1306 * Call the extent policy function to see if our request can be
1307 * granted, or is blocked. */
1308 lock->l_policy_data.l_extent.start = 0;
1309 lock->l_policy_data.l_extent.end = OBD_OBJECT_EOF;
1310 lock->l_req_mode = LCK_PR;
1312 lock_res_and_lock(lock);
1313 res = lock->l_resource;
1314 rc = policy(lock, &tmpflags, 0, &err, &rpc_list);
1316 /* FIXME: we should change the policy function slightly, to not make
1317 * this list at all, since we just turn around and free it */
1318 while (!list_empty(&rpc_list)) {
1319 struct ldlm_lock *wlock =
1320 list_entry(rpc_list.next, struct ldlm_lock, l_cp_ast);
1321 LASSERT((lock->l_flags & LDLM_FL_AST_SENT) == 0);
1322 LASSERT(lock->l_flags & LDLM_FL_CP_REQD);
1323 lock->l_flags &= ~LDLM_FL_CP_REQD;
1324 list_del_init(&wlock->l_cp_ast);
1325 LDLM_LOCK_PUT(wlock);
1328 if (rc == LDLM_ITER_CONTINUE) {
1329 /* The lock met with no resistance; we're finished. */
1330 unlock_res_and_lock(lock);
1331 RETURN(ELDLM_LOCK_REPLACED);
1334 /* Do not grant any lock, but instead send GL callbacks. The extent
1335 * policy nicely created a list of all PW locks for us. We will choose
1336 * the highest of those which are larger than the size in the LVB, if
1337 * any, and perform a glimpse callback. */
1338 res_lvb = res->lr_lvb_data;
1339 LASSERT(res_lvb != NULL);
1340 *reply_lvb = *res_lvb;
1342 list_for_each(tmp, &res->lr_granted) {
1343 struct ldlm_lock *tmplock =
1344 list_entry(tmp, struct ldlm_lock, l_res_link);
1346 if (tmplock->l_granted_mode == LCK_PR)
1349 if (tmplock->l_policy_data.l_extent.end <= reply_lvb->lvb_size)
1353 l = LDLM_LOCK_GET(tmplock);
1357 if (l->l_policy_data.l_extent.start >
1358 tmplock->l_policy_data.l_extent.start)
1362 l = LDLM_LOCK_GET(tmplock);
1366 /* There were no PW locks beyond the size in the LVB; finished. */
1368 RETURN(ELDLM_LOCK_ABORTED);
1370 if (l->l_glimpse_ast == NULL) {
1371 /* We are racing with unlink(); just return -ENOENT */
1372 CDEBUG(D_OTHER, "race with unlink: "LPX64"/"LPX64"\n",
1373 res->lr_name.name[0], res->lr_name.name[2]);
1374 rep->lock_policy_res1 = -ENOENT;
1378 LASSERTF(l->l_glimpse_ast != NULL, "l == %p", l);
1380 rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
1381 if (rc != 0 && res->lr_namespace->ns_lvbo &&
1382 res->lr_namespace->ns_lvbo->lvbo_update) {
1383 res->lr_namespace->ns_lvbo->lvbo_update(res, NULL, 0, 1);
1387 *reply_lvb = *res_lvb;
1392 RETURN(ELDLM_LOCK_ABORTED);
1395 static int filter_post_fs_cleanup(struct obd_device *obd)
1399 rc = fsfilt_post_cleanup(obd);
1405 static int filter_group_set_fs_flags(struct obd_device *obd, int group)
1407 struct filter_obd *filter = &obd->u.filter;
1411 /* zero group is not longer valid. */
1414 for (i = 0; i < filter->fo_subdir_count; i++) {
1415 struct dentry *dentry;
1416 dentry = (filter->fo_subdirs + group)->dentry[i];
1417 rc = fsfilt_set_fs_flags(obd, dentry->d_inode,
1418 SM_DO_REC | SM_DO_COW);
1426 static int filter_post_fs_setup(struct obd_device *obd)
1428 struct filter_obd *filter = &obd->u.filter;
1432 rc = fsfilt_post_setup(obd, filter->fo_dentry_O);
1437 /* mount the file system (secretly) */
1438 int filter_common_setup(struct obd_device *obd, obd_count len, void *buf,
1441 struct lustre_cfg *lcfg = buf;
1442 struct filter_obd *filter = &obd->u.filter;
1443 struct lvfs_obd_ctxt *lvfs_ctxt = NULL;
1444 struct vfsmount *mnt;
1450 if ((LUSTRE_CFG_BUFLEN(lcfg, 1)) < 1 ||
1451 (LUSTRE_CFG_BUFLEN(lcfg, 2) < 1))
1454 obd->obd_fsops = fsfilt_get_ops(lustre_cfg_string(lcfg, 2));
1455 if (IS_ERR(obd->obd_fsops))
1456 RETURN(PTR_ERR(obd->obd_fsops));
1458 rc = lvfs_mount_fs(lustre_cfg_string(lcfg, 1),
1459 lustre_cfg_string(lcfg, 2),
1460 option, MS_NOATIME | MS_NODIRATIME, &lvfs_ctxt);
1462 CERROR("lvfs_mount_fs failed: rc = %d\n", rc);
1467 mnt = lvfs_ctxt->loc_mnt;
1468 filter->fo_lvfs_ctxt = lvfs_ctxt;
1470 if (LUSTRE_CFG_BUFLEN(lcfg, 3) > 0 && lustre_cfg_buf(lcfg, 3)) {
1471 str = lustre_cfg_string(lcfg, 3);
1473 obd->obd_replayable = 1;
1474 obd_sync_filter = 1;
1475 CWARN("%s: recovery enabled\n", obd->obd_name);
1478 CERROR("unrecognised flag '%c'\n",
1481 // XXX Robert? Why do we get errors here
1482 // GOTO(err_mntput, rc = -EINVAL);
1486 filter->fo_vfsmnt = mnt;
1487 filter->fo_sb = mnt->mnt_sb;
1488 filter->fo_fstype = mnt->mnt_sb->s_type->name;
1489 CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
1491 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1492 obd->obd_lvfs_ctxt.pwdmnt = mnt;
1493 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
1494 obd->obd_lvfs_ctxt.fs = get_ds();
1495 obd->obd_lvfs_ctxt.cb_ops = filter_lvfs_ops;
1497 ll_clear_rdonly(ll_sbdev(filter->fo_sb));
1499 rc = fsfilt_setup(obd, mnt->mnt_sb);
1501 GOTO(err_mntput, rc);
1503 sema_init(&filter->fo_init_lock, 1);
1504 filter->fo_committed_group = 0;
1505 rc = filter_prep(obd);
1507 GOTO(err_mntput, rc);
1509 filter->fo_destroys_in_progress = 0;
1510 for (i = 0; i < 32; i++)
1511 sema_init(&filter->fo_create_locks[i], 1);
1513 spin_lock_init(&filter->fo_translock);
1514 spin_lock_init(&filter->fo_objidlock);
1515 spin_lock_init(&filter->fo_lastidlock);
1516 INIT_LIST_HEAD(&filter->fo_export_list);
1517 sema_init(&filter->fo_alloc_lock, 1);
1518 spin_lock_init(&filter->fo_r_pages.oh_lock);
1519 spin_lock_init(&filter->fo_w_pages.oh_lock);
1520 spin_lock_init(&filter->fo_r_discont_pages.oh_lock);
1521 spin_lock_init(&filter->fo_w_discont_pages.oh_lock);
1522 spin_lock_init(&filter->fo_r_discont_blocks.oh_lock);
1523 spin_lock_init(&filter->fo_w_discont_blocks.oh_lock);
1524 filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE;
1526 INIT_LIST_HEAD(&filter->fo_llog_list);
1527 spin_lock_init(&filter->fo_llog_list_lock);
1529 spin_lock_init(&filter->fo_denylist_lock);
1530 INIT_LIST_HEAD(&filter->fo_denylist);
1532 sprintf(ns_name, "filter-%s", obd->obd_uuid.uuid);
1533 obd->obd_namespace = ldlm_namespace_new(ns_name, LDLM_NAMESPACE_SERVER);
1535 if (obd->obd_namespace == NULL)
1536 GOTO(err_post, rc = -ENOMEM);
1537 obd->obd_namespace->ns_lvbp = obd;
1538 obd->obd_namespace->ns_lvbo = &filter_lvbo;
1539 ldlm_register_intent(obd->obd_namespace, filter_intent_policy);
1541 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1542 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
1544 rc = obd_llog_cat_initialize(obd, &obd->obd_llogs, 1, CATLIST);
1546 CERROR("failed to setup llogging subsystems\n");
1555 lvfs_umount_fs(filter->fo_lvfs_ctxt);
1559 fsfilt_put_ops(obd->obd_fsops);
1563 static int filter_attach(struct obd_device *obd, obd_count len, void *data)
1565 struct lprocfs_static_vars lvars;
1568 lprocfs_init_vars(filter, &lvars);
1569 rc = lprocfs_obd_attach(obd, lvars.obd_vars);
1573 rc = lprocfs_alloc_obd_stats(obd, LPROC_FILTER_LAST);
1577 /* Init obdfilter private stats here */
1578 lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_READ_BYTES,
1579 LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
1580 lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
1581 LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
1583 return lproc_filter_attach_seqstat(obd);
1586 static int filter_detach(struct obd_device *dev)
1588 lprocfs_free_obd_stats(dev);
1589 return lprocfs_obd_detach(dev);
1592 static int filter_setup(struct obd_device *obd, obd_count len, void *buf)
1594 struct lustre_cfg *lcfg = buf;
1599 /* 2.6.9 selinux wants a full option page for do_kern_mount (bug6471) */
1600 page = get_zeroed_page(GFP_KERNEL);
1604 memcpy((void *)page, lustre_cfg_buf(lcfg, 4),
1605 LUSTRE_CFG_BUFLEN(lcfg, 4));
1607 /* all mount options including errors=remount-ro and asyncdel are passed
1608 * using 4th lcfg param. And it is good, finally we have got rid of
1609 * hardcoded fs types in the code. */
1610 rc = filter_common_setup(obd, len, buf, (void *)page);
1615 rc = filter_post_fs_setup(obd);
1619 static int filter_cleanup(struct obd_device *obd, int flags)
1621 struct filter_obd *filter = &obd->u.filter;
1622 ll_sbdev_type save_dev;
1625 if (flags & OBD_OPT_FAILOVER)
1626 CERROR("%s: shutting down for failover; client state will"
1627 " be preserved.\n", obd->obd_name);
1629 if (!list_empty(&obd->obd_exports)) {
1630 CERROR("%s: still has clients!\n", obd->obd_name);
1631 class_disconnect_exports(obd, flags);
1632 if (!list_empty(&obd->obd_exports)) {
1633 CERROR("still has exports after forced cleanup?\n");
1638 target_cleanup_recovery(obd);
1640 ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
1642 if (filter->fo_sb == NULL)
1645 save_dev = ll_sbdev(filter->fo_sb);
1646 filter_post_fs_cleanup(obd);
1649 shrink_dcache_parent(filter->fo_sb->s_root);
1652 spin_lock(&filter->fo_denylist_lock);
1653 while (!list_empty(&filter->fo_denylist)) {
1654 deny_sec_t *p_deny_sec = list_entry(filter->fo_denylist.next,
1656 list_del(&p_deny_sec->list);
1657 OBD_FREE(p_deny_sec, sizeof(*p_deny_sec));
1659 spin_unlock(&filter->fo_denylist_lock);
1662 lvfs_umount_fs(filter->fo_lvfs_ctxt);
1663 //destroy_buffers(filter->fo_sb->s_dev);
1664 filter->fo_sb = NULL;
1665 fsfilt_put_ops(obd->obd_fsops);
1668 ll_clear_rdonly(save_dev);
1673 static int filter_process_config(struct obd_device *obd, obd_count len, void *buf)
1675 struct lustre_cfg *lcfg = buf;
1676 struct filter_obd *filter = &obd->u.filter;
1680 switch(lcfg->lcfg_command) {
1681 case LCFG_SET_SECURITY: {
1682 if ((LUSTRE_CFG_BUFLEN(lcfg, 1) == 0) ||
1683 (LUSTRE_CFG_BUFLEN(lcfg, 2) == 0))
1684 GOTO(out, rc = -EINVAL);
1686 if (!strcmp(lustre_cfg_string(lcfg, 1), "deny_sec")){
1687 spin_lock(&filter->fo_denylist_lock);
1688 rc = add_deny_security(lustre_cfg_string(lcfg, 2),
1689 &filter->fo_denylist);
1690 spin_unlock(&filter->fo_denylist_lock);
1692 CERROR("Unrecognized key\n");
1698 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1699 GOTO(out, rc = -EINVAL);
1707 static int filter_connect_post(struct obd_export *exp, unsigned initial,
1708 unsigned long connect_flags)
1710 struct obd_device *obd = exp->exp_obd;
1711 struct filter_export_data *fed;
1712 char str[PTL_NALFMT_SIZE];
1713 struct obd_llogs *llog;
1714 struct llog_ctxt *ctxt;
1718 fed = &exp->exp_filter_data;
1719 if (fed->fed_group < FILTER_MIN_GROUPS)
1722 /* initialize llogs for connections from MDS */
1723 llog = filter_grab_llog_for_group(obd, fed->fed_group, exp);
1724 LASSERT(llog != NULL);
1726 ctxt = llog_get_context(llog, LLOG_UNLINK_REPL_CTXT);
1727 LASSERT(ctxt != NULL);
1729 rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse);
1731 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
1732 exp->exp_connection->c_peer.peer_id.nid, str);
1734 CDEBUG(D_OTHER, "%s: init llog ctxt for export "LPX64"/%s, group %d\n",
1735 obd->obd_name, exp->exp_connection->c_peer.peer_id.nid,
1736 str, fed->fed_group);
1741 /* nearly identical to mds_connect */
1742 static int filter_connect(struct lustre_handle *conn, struct obd_device *obd,
1743 struct obd_uuid *cluuid,
1744 struct obd_connect_data *data,
1745 unsigned long connect_flags)
1747 struct obd_export *exp;
1748 struct filter_export_data *fed;
1749 struct filter_client_data *fcd = NULL;
1750 struct filter_obd *filter = &obd->u.filter;
1751 struct lvfs_run_ctxt saved;
1755 if (conn == NULL || obd == NULL || cluuid == NULL)
1758 rc = class_connect(conn, obd, cluuid);
1761 exp = class_conn2export(conn);
1762 LASSERT(exp != NULL);
1764 fed = &exp->exp_filter_data;
1766 spin_lock_init(&fed->fed_lock);
1768 /* connection from MDS */
1769 group = connect_flags;
1770 if (obd->obd_replayable) {
1771 OBD_ALLOC(fcd, sizeof(*fcd));
1773 CERROR("filter: out of memory for client data\n");
1774 GOTO(cleanup, rc = -ENOMEM);
1777 memcpy(fcd->fcd_uuid, cluuid, sizeof(fcd->fcd_uuid));
1779 fed->fed_fcd->fcd_group = group;
1780 rc = filter_client_add(obd, filter, fed, -1);
1784 CWARN("%s: Received MDS connection ("LPX64"); group %d\n",
1785 obd->obd_name, exp->exp_handle.h_cookie, group);
1790 if (fed->fed_group != 0 && fed->fed_group != group) {
1791 char str[PTL_NALFMT_SIZE];
1792 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
1793 exp->exp_connection->c_peer.peer_id.nid, str);
1794 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
1795 "earlier; now it's trying to use group %d! This could "
1796 "be a bug in the MDS. Tell CFS.\n",
1797 exp->exp_connection->c_peer.peer_id.nid, str,
1798 fed->fed_group, group);
1799 GOTO(cleanup, rc = -EPROTO);
1801 fed->fed_group = group;
1803 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1804 rc = filter_read_groups(obd, group, 1);
1805 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1807 CERROR("can't read group %u\n", group);
1811 rc = filter_group_set_fs_flags(obd, group);
1813 CERROR("can't set kml flags %u\n", group);
1820 OBD_FREE(fcd, sizeof(*fcd));
1821 class_disconnect(exp, 0);
1823 class_export_put(exp);
1828 static int filter_precleanup(struct obd_device *obd, int flags)
1830 struct filter_group_llog *log;
1831 struct filter_obd *filter;
1835 filter = &obd->u.filter;
1837 spin_lock(&filter->fo_llog_list_lock);
1838 while (!list_empty(&filter->fo_llog_list)) {
1839 log = list_entry(filter->fo_llog_list.next,
1840 struct filter_group_llog, list);
1841 list_del(&log->list);
1842 spin_unlock(&filter->fo_llog_list_lock);
1844 rc = obd_llog_finish(obd, log->llogs, 0);
1846 CERROR("failed to cleanup llogging subsystem for %u\n",
1848 OBD_FREE(log->llogs, sizeof(*(log->llogs)));
1849 OBD_FREE(log, sizeof(*log));
1850 spin_lock(&filter->fo_llog_list_lock);
1852 spin_unlock(&filter->fo_llog_list_lock);
1854 rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1856 CERROR("failed to cleanup llogging subsystem\n");
1861 /* Do extra sanity checks for grant accounting. We do this at connect,
1862 * disconnect, and statfs RPC time, so it shouldn't be too bad. We can
1863 * always get rid of it or turn it off when we know accounting is good. */
1864 static void filter_grant_sanity_check(struct obd_device *obd, const char *func)
1866 struct filter_export_data *fed;
1867 struct obd_export *exp;
1868 obd_size maxsize = obd->obd_osfs.os_blocks * obd->obd_osfs.os_bsize;
1869 obd_size tot_dirty = 0, tot_pending = 0, tot_granted = 0;
1870 obd_size fo_tot_dirty, fo_tot_pending, fo_tot_granted;
1871 int level = D_CACHE;
1873 if (list_empty(&obd->obd_exports))
1876 spin_lock(&obd->obd_osfs_lock);
1877 spin_lock(&obd->obd_dev_lock);
1878 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1879 fed = &exp->exp_filter_data;
1880 if (fed->fed_grant < 0 || fed->fed_pending < 0 ||
1883 if (maxsize > 0) { /* we may not have done a statfs yet */
1884 LASSERTF(fed->fed_grant + fed->fed_pending <= maxsize,
1885 "cli %s/%p %ld+%ld > "LPU64"\n",
1886 exp->exp_client_uuid.uuid, exp,
1887 fed->fed_grant, fed->fed_pending, maxsize);
1888 LASSERTF(fed->fed_dirty <= maxsize,
1889 "cli %s/%p %ld > "LPU64"\n",
1890 exp->exp_client_uuid.uuid, exp,
1891 fed->fed_dirty, maxsize);
1893 CDEBUG(level, "%s: cli %s/%p dirty %ld pend %ld grant %ld\n",
1894 obd->obd_name, exp->exp_client_uuid.uuid, exp,
1895 fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1896 tot_granted += fed->fed_grant + fed->fed_pending;
1897 tot_pending += fed->fed_pending;
1898 tot_dirty += fed->fed_dirty;
1900 fo_tot_granted = obd->u.filter.fo_tot_granted;
1901 fo_tot_pending = obd->u.filter.fo_tot_pending;
1902 fo_tot_dirty = obd->u.filter.fo_tot_dirty;
1903 spin_unlock(&obd->obd_dev_lock);
1904 spin_unlock(&obd->obd_osfs_lock);
1906 /* Do these assertions outside the spinlocks so we don't kill system */
1907 if (tot_granted != fo_tot_granted)
1908 CERROR("%s: tot_granted "LPU64" != fo_tot_granted "LPU64"\n",
1909 func, tot_granted, fo_tot_granted);
1910 if (tot_pending != fo_tot_pending)
1911 CERROR("%s: tot_pending "LPU64" != fo_tot_pending "LPU64"\n",
1912 func, tot_pending, fo_tot_pending);
1913 if (tot_dirty != fo_tot_dirty)
1914 CERROR("%s: tot_dirty "LPU64" != fo_tot_dirty "LPU64"\n",
1915 func, tot_dirty, fo_tot_dirty);
1916 if (tot_pending > tot_granted)
1917 CERROR("%s: tot_pending "LPU64" > tot_granted "LPU64"\n",
1918 func, tot_pending, tot_granted);
1919 if (tot_granted > maxsize)
1920 CERROR("%s: tot_granted "LPU64" > maxsize "LPU64"\n",
1921 func, tot_granted, maxsize);
1922 if (tot_dirty > maxsize)
1923 CERROR("%s: tot_dirty "LPU64" > maxsize "LPU64"\n",
1924 func, tot_dirty, maxsize);
1927 /* Remove this client from the grant accounting totals. We also remove
1928 * the export from the obd device under the osfs and dev locks to ensure
1929 * that the filter_grant_sanity_check() calculations are always valid.
1930 * The client should do something similar when it invalidates its import. */
1931 static void filter_grant_discard(struct obd_export *exp)
1933 struct obd_device *obd = exp->exp_obd;
1934 struct filter_obd *filter = &obd->u.filter;
1935 struct filter_export_data *fed = &exp->exp_filter_data;
1936 int level = D_CACHE;
1938 spin_lock(&obd->obd_osfs_lock);
1939 spin_lock(&exp->exp_obd->obd_dev_lock);
1940 list_del_init(&exp->exp_obd_chain);
1941 spin_unlock(&exp->exp_obd->obd_dev_lock);
1943 if (fed->fed_dirty < 0 || fed->fed_grant < 0 || fed->fed_pending < 0)
1945 CDEBUG(level, "%s: cli %s/%p dirty %ld pend %ld grant %ld\n",
1946 obd->obd_name, exp->exp_client_uuid.uuid, exp,
1947 fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1949 LASSERTF(filter->fo_tot_granted >= fed->fed_grant,
1950 "%s: tot_granted "LPU64" cli %s/%p fed_grant %ld\n",
1951 obd->obd_name, filter->fo_tot_granted,
1952 exp->exp_client_uuid.uuid, exp, fed->fed_grant);
1953 filter->fo_tot_granted -= fed->fed_grant;
1954 LASSERTF(filter->fo_tot_pending >= fed->fed_pending,
1955 "%s: tot_pending "LPU64" cli %s/%p fed_pending %ld\n",
1956 obd->obd_name, filter->fo_tot_pending,
1957 exp->exp_client_uuid.uuid, exp, fed->fed_pending);
1958 LASSERTF(filter->fo_tot_dirty >= fed->fed_dirty,
1959 "%s: tot_dirty "LPU64" cli %s/%p fed_dirty %ld\n",
1960 obd->obd_name, filter->fo_tot_dirty,
1961 exp->exp_client_uuid.uuid, exp, fed->fed_dirty);
1962 filter->fo_tot_dirty -= fed->fed_dirty;
1966 spin_unlock(&obd->obd_osfs_lock);
1969 static int filter_destroy_export(struct obd_export *exp)
1973 if (exp->exp_filter_data.fed_pending)
1974 CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
1975 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
1976 exp, exp->exp_filter_data.fed_pending);
1978 target_destroy_export(exp);
1980 if (exp->exp_obd->obd_replayable)
1981 filter_client_free(exp, exp->exp_flags);
1983 filter_grant_discard(exp);
1984 if (!(exp->exp_flags & OBD_OPT_FORCE))
1985 filter_grant_sanity_check(exp->exp_obd, __FUNCTION__);
1990 static void filter_sync_llogs(struct obd_device *obd, struct obd_export *dexp)
1992 struct filter_group_llog *fglog, *nlog;
1993 struct filter_obd *filter;
1994 int worked = 0, group;
1995 struct llog_ctxt *ctxt;
1998 filter = &obd->u.filter;
2000 /* we can't sync log holding spinlock. also, we do not want to get
2001 * into livelock. so we do following: loop over MDS's exports in
2002 * group order and skip already synced llogs -bzzz */
2004 /* look for group with min. number, but > worked */
2007 spin_lock(&filter->fo_llog_list_lock);
2008 list_for_each_entry(nlog, &filter->fo_llog_list, list) {
2010 if (nlog->group <= worked) {
2011 /* this group is already synced */
2015 if (group < nlog->group) {
2016 /* we have group with smaller number to sync */
2020 /* store current minimal group */
2022 group = nlog->group;
2024 spin_unlock(&filter->fo_llog_list_lock);
2029 worked = fglog->group;
2030 if (fglog->exp && (dexp == fglog->exp || dexp == NULL)) {
2031 ctxt = llog_get_context(fglog->llogs,
2032 LLOG_UNLINK_REPL_CTXT);
2033 LASSERT(ctxt != NULL);
2034 llog_sync(ctxt, fglog->exp);
2036 } while (fglog != NULL);
2039 /* also incredibly similar to mds_disconnect */
2040 static int filter_disconnect(struct obd_export *exp, unsigned long flags)
2042 struct obd_device *obd = exp->exp_obd;
2043 unsigned long irqflags;
2048 class_export_get(exp);
2050 spin_lock_irqsave(&exp->exp_lock, irqflags);
2051 exp->exp_flags = flags;
2052 spin_unlock_irqrestore(&exp->exp_lock, irqflags);
2054 if (!(flags & OBD_OPT_FORCE))
2055 filter_grant_sanity_check(obd, __FUNCTION__);
2056 filter_grant_discard(exp);
2058 /* Disconnect early so that clients can't keep using export */
2059 rc = class_disconnect(exp, flags);
2061 ldlm_cancel_locks_for_export(exp);
2063 fsfilt_sync(obd, obd->u.filter.fo_sb);
2065 /* flush any remaining cancel messages out to the target */
2066 filter_sync_llogs(obd, exp);
2067 class_export_put(exp);
2071 struct dentry *__filter_oa2dentry(struct obd_device *obd,
2072 struct obdo *oa, const char *what)
2074 struct dentry *dchild = NULL;
2078 if (oa->o_valid & OBD_MD_FLGROUP)
2081 dchild = filter_id2dentry(obd, NULL, group, oa->o_id);
2083 if (IS_ERR(dchild)) {
2084 CERROR("%s error looking up object: "LPU64"\n",
2089 if (dchild->d_inode == NULL) {
2090 CDEBUG(D_INFO, "%s: %s on non-existent object: "
2091 LPU64"\n", obd->obd_name, what, oa->o_id);
2093 RETURN(ERR_PTR(-ENOENT));
2099 static int filter_getattr(struct obd_export *exp, struct obdo *oa,
2100 struct lov_stripe_md *md)
2102 struct dentry *dentry = NULL;
2103 struct obd_device *obd;
2107 obd = class_exp2obd(exp);
2109 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2110 exp->exp_handle.h_cookie);
2114 dentry = filter_oa2dentry(obd, oa);
2116 RETURN(PTR_ERR(dentry));
2118 /* Limit the valid bits in the return data to what we actually use */
2119 oa->o_valid = OBD_MD_FLID;
2120 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2126 int filter_setattr_internal(struct obd_export *exp, struct dentry *dentry,
2127 struct obdo *oa, struct obd_trans_info *oti)
2129 struct filter_obd *filter;
2135 LASSERT(dentry != NULL);
2136 LASSERT(!IS_ERR(dentry));
2137 LASSERT(dentry->d_inode != NULL);
2139 filter = &exp->exp_obd->u.filter;
2140 iattr_from_obdo(&iattr, oa, oa->o_valid);
2142 if (iattr.ia_valid & ATTR_SIZE)
2143 down(&dentry->d_inode->i_sem);
2144 handle = fsfilt_start(exp->exp_obd, dentry->d_inode,
2145 FSFILT_OP_SETATTR, oti);
2147 GOTO(out_unlock, rc = PTR_ERR(handle));
2149 /* XXX this could be a rwsem instead, if filter_preprw played along */
2150 if (iattr.ia_valid & ATTR_ATTR_FLAG)
2151 rc = fsfilt_iocontrol(exp->exp_obd, dentry->d_inode,
2152 NULL, EXT3_IOC_SETFLAGS,
2153 (long)&iattr.ia_attr_flags);
2155 rc = fsfilt_setattr(exp->exp_obd, dentry, handle,
2158 rc = filter_finish_transno(exp, oti, rc);
2160 err = fsfilt_commit(exp->exp_obd, filter->fo_sb,
2161 dentry->d_inode, handle,
2164 CERROR("error on commit, err = %d\n", err);
2170 if (iattr.ia_valid & ATTR_SIZE)
2171 up(&dentry->d_inode->i_sem);
2175 /* this is called from filter_truncate() until we have filter_punch() */
2176 int filter_setattr(struct obd_export *exp, struct obdo *oa,
2177 struct lov_stripe_md *md, struct obd_trans_info *oti)
2179 struct ldlm_res_id res_id = { .name = { oa->o_id, 0, oa->o_gr, 0 } };
2180 struct ldlm_valblock_ops *ns_lvbo;
2181 struct lvfs_run_ctxt saved;
2182 struct filter_obd *filter;
2183 struct ldlm_resource *res;
2184 struct dentry *dentry;
2190 LASSERT(oti != NULL);
2192 filter = &exp->exp_obd->u.filter;
2193 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2195 /* make sure that object is allocated. */
2196 dentry = filter_crow_object(exp->exp_obd, oa);
2198 GOTO(out_pop, rc = PTR_ERR(dentry));
2202 /* setting objects attributes (including owner/group) */
2203 rc = filter_setattr_internal(exp, dentry, oa, oti);
2205 GOTO(out_unlock, rc);
2207 res = ldlm_resource_get(exp->exp_obd->obd_namespace, NULL,
2208 res_id, LDLM_EXTENT, 0);
2211 ns_lvbo = res->lr_namespace->ns_lvbo;
2212 if (ns_lvbo && ns_lvbo->lvbo_update)
2213 rc = ns_lvbo->lvbo_update(res, NULL, 0, 0);
2214 ldlm_resource_putref(res);
2217 oa->o_valid = OBD_MD_FLID;
2218 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2225 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2229 /* XXX identical to osc_unpackmd */
2230 static int filter_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2231 struct lov_mds_md *lmm, int lmm_bytes)
2237 if (lmm_bytes < sizeof (*lmm)) {
2238 CERROR("lov_mds_md too small: %d, need %d\n",
2239 lmm_bytes, (int)sizeof(*lmm));
2242 /* XXX LOV_MAGIC etc check? */
2244 if (lmm->lmm_object_id == cpu_to_le64(0)) {
2245 CERROR("lov_mds_md: zero lmm_object_id\n");
2250 lsm_size = lov_stripe_md_size(1);
2254 if (*lsmp != NULL && lmm == NULL) {
2255 OBD_FREE(*lsmp, lsm_size);
2260 if (*lsmp == NULL) {
2261 OBD_ALLOC(*lsmp, lsm_size);
2265 loi_init((*lsmp)->lsm_oinfo);
2269 /* XXX zero *lsmp? */
2270 (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
2271 LASSERT((*lsmp)->lsm_object_id);
2274 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
2279 static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2280 unsigned long max_age)
2282 struct filter_obd *filter = &obd->u.filter;
2283 int blockbits = filter->fo_sb->s_blocksize_bits;
2287 /* at least try to account for cached pages. its still racey and
2288 * might be under-reporting if clients haven't announced their
2289 * caches with brw recently */
2290 spin_lock(&obd->obd_osfs_lock);
2291 rc = fsfilt_statfs(obd, filter->fo_sb, max_age);
2292 memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
2293 spin_unlock(&obd->obd_osfs_lock);
2295 CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
2296 " pending "LPU64" free "LPU64" avail "LPU64"\n",
2297 filter->fo_tot_dirty, filter->fo_tot_granted,
2298 filter->fo_tot_pending,
2299 osfs->os_bfree << blockbits, osfs->os_bavail << blockbits);
2301 filter_grant_sanity_check(obd, __FUNCTION__);
2303 osfs->os_bavail -= min(osfs->os_bavail,
2304 (filter->fo_tot_dirty + filter->fo_tot_pending +
2305 osfs->os_bsize -1) >> blockbits);
2310 int filter_create_object(struct obd_device *obd, struct obdo *oa)
2312 struct dentry *dparent = NULL;
2313 struct dentry *dchild = NULL;
2314 struct lvfs_ucred uc = {0,};
2315 struct lvfs_run_ctxt saved;
2316 struct filter_obd *filter;
2317 int cleanup_phase = 0;
2318 int err = 0, rc = 0;
2319 void *handle = NULL;
2323 filter = &obd->u.filter;
2325 down(&filter->fo_create_locks[oa->o_gr]);
2327 if (test_bit(oa->o_gr, &filter->fo_destroys_in_progress)) {
2328 CWARN("%s: create aborted by destroy\n",
2330 GOTO(out, rc = -EALREADY);
2333 CDEBUG(D_INFO, "create objid "LPU64"\n", oa->o_id);
2335 dparent = filter_parent_lock(obd, oa->o_gr, oa->o_id, &lock);
2336 if (IS_ERR(dparent))
2337 GOTO(cleanup, rc = PTR_ERR(dparent));
2340 dchild = filter_id2dentry(obd, dparent, oa->o_gr, oa->o_id);
2342 GOTO(cleanup, rc = PTR_ERR(dchild));
2345 if (dchild->d_inode != NULL)
2346 GOTO(cleanup, rc = 0);
2348 handle = fsfilt_start_log(obd, dparent->d_inode,
2349 FSFILT_OP_CREATE, NULL, 1);
2351 GOTO(cleanup, rc = PTR_ERR(handle));
2354 uc.luc_uid = oa->o_uid;
2355 uc.luc_gid = oa->o_gid;
2356 uc.luc_fsuid = oa->o_uid;
2357 uc.luc_fsgid = oa->o_gid;
2358 uc.luc_cap = current->cap_effective;
2359 cap_raise(uc.luc_cap, CAP_SYS_RESOURCE);
2361 push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
2362 rc = ll_vfs_create(dparent->d_inode, dchild, S_IFREG, NULL);
2363 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
2366 CERROR("create failed rc = %d\n", rc);
2370 /* nobody else is touching this newly created object */
2371 LASSERT(dchild->d_inode);
2372 if (oa->o_valid & OBD_MD_FLIFID) {
2373 rc = fsfilt_set_md(obd, dchild->d_inode, handle, obdo_id(oa),
2374 sizeof(struct lustre_id), EA_SID);
2376 CERROR("store fid in object failed! rc:%d\n", rc);
2380 CWARN("create object without lustre id!\n");
2384 fsfilt_set_fs_flags(obd, dparent->d_inode, SM_DO_REC);
2386 /* save last created object id */
2387 filter_save_last_id(filter, oa->o_gr, oa->o_id);
2389 rc = filter_update_last_objid(obd, oa->o_gr, 0);
2391 CERROR("unable to write lastobjid, but "
2392 "orphans were deleted, err = %d\n",
2397 switch(cleanup_phase) {
2399 err = fsfilt_commit(obd, filter->fo_sb,
2400 dparent->d_inode, handle, 0);
2402 CERROR("error on commit, err = %d\n", err);
2409 filter_parent_unlock(dparent, lock);
2418 up(&filter->fo_create_locks[oa->o_gr]);
2423 filter_crow_object(struct obd_device *obd, struct obdo *oa)
2425 struct dentry *dentry;
2431 /* check if object is already allocated */
2432 dentry = filter_id2dentry(obd, NULL, oa->o_gr, oa->o_id);
2436 if (dentry->d_inode)
2440 CDEBUG(D_INODE, "OSS object "LPU64"/"LPU64
2441 " does not exists - allocate it now\n",
2442 oa->o_id, oa->o_gr);
2444 uid = oa->o_valid & OBD_MD_FLUID ? oa->o_uid : 0;
2445 gid = oa->o_valid & OBD_MD_FLGID ? oa->o_gid : 0;
2447 rc = filter_create_object(obd, oa);
2449 CERROR("cannot create OSS object "LPU64"/"LPU64
2450 ", err = %d\n", oa->o_id, oa->o_gr, rc);
2451 RETURN(ERR_PTR(rc));
2454 /* lookup for just created object and return it to caller */
2455 dentry = filter_id2dentry(obd, NULL, oa->o_gr, oa->o_id);
2459 if (dentry->d_inode == NULL) {
2461 dentry = ERR_PTR(-ENOENT);
2462 CERROR("cannot find just created OSS object "
2463 LPU64"/"LPU64" err = %d\n", oa->o_id,
2464 oa->o_gr, (int)PTR_ERR(dentry));
2471 filter_clear_orphans(struct obd_export *exp, struct obdo *oa)
2473 struct obd_device *obd = NULL;
2474 struct filter_obd *filter;
2475 struct obdo *doa = NULL;
2476 int rc = 0, orphans;
2481 LASSERT(oa->o_gr != 0);
2482 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2485 filter = &obd->u.filter;
2487 last = filter_last_id(filter, oa->o_gr);
2488 orphans = last - oa->o_id;
2497 doa->o_gr = oa->o_gr;
2498 doa->o_mode = S_IFREG;
2499 doa->o_valid = oa->o_valid & (OBD_MD_FLGROUP | OBD_MD_FLID);
2501 set_bit(doa->o_gr, &filter->fo_destroys_in_progress);
2502 down(&filter->fo_create_locks[doa->o_gr]);
2503 if (!test_bit(doa->o_gr, &filter->fo_destroys_in_progress)) {
2504 CERROR("%s:["LPU64"] destroy_in_progress already cleared\n",
2505 exp->exp_obd->obd_name, doa->o_gr);
2506 up(&filter->fo_create_locks[doa->o_gr]);
2507 GOTO(out_free_doa, 0);
2510 CDEBUG(D_ERROR, "%s:["LPU64"] deleting orphan objects from "LPU64" to "
2511 LPU64"\n", exp->exp_obd->obd_name, doa->o_gr,
2512 oa->o_id + 1, last);
2514 for (id = oa->o_id + 1; id <= last; id++) {
2516 filter_destroy(exp, doa, NULL, NULL);
2519 CDEBUG(D_ERROR, "%s:["LPU64"] after destroy: set last_objids = "
2520 LPU64"\n", exp->exp_obd->obd_name, doa->o_gr, last + 1);
2522 /* return next free id to be used as a new start of sequence -bzzz */
2523 oa->o_id = last + 1;
2525 filter_set_last_id(filter, oa->o_gr, oa->o_id);
2526 clear_bit(doa->o_gr, &filter->fo_destroys_in_progress);
2527 up(&filter->fo_create_locks[oa->o_gr]);
2536 * by now this function is only needed as entry point for deleting orphanes on
2537 * OSS as objects are created on first write attempt. --umka
2540 filter_create(struct obd_export *exp, struct obdo *oa, void *acl,
2541 int acl_size, struct lov_stripe_md **ea,
2542 struct obd_trans_info *oti)
2544 struct filter_export_data *fed;
2545 struct obd_device *obd = NULL;
2546 int group = oa->o_gr, rc = 0;
2547 struct lvfs_run_ctxt saved;
2548 struct filter_obd *filter;
2549 char str[PTL_NALFMT_SIZE];
2552 LASSERT(acl == NULL && acl_size == 0);
2554 if (!(oa->o_valid & OBD_MD_FLGROUP) || group == 0) {
2555 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2556 exp->exp_connection->c_peer.peer_id.nid, str);
2557 CERROR("!!! nid "LPX64"/%s sent invalid object group %d\n",
2558 exp->exp_connection->c_peer.peer_id.nid, str, group);
2563 fed = &exp->exp_filter_data;
2564 filter = &obd->u.filter;
2566 if (fed->fed_group != group) {
2567 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2568 exp->exp_connection->c_peer.peer_id.nid, str);
2569 CERROR("!!! this export (nid "LPX64"/%s) used object group %d "
2570 "earlier; now it's trying to use group %d! This could "
2571 "be a bug in the MDS. Tell CFS.\n",
2572 exp->exp_connection->c_peer.peer_id.nid, str,
2573 fed->fed_group, group);
2577 CDEBUG(D_INFO, "filter_create(od->o_gr=%d,od->o_id="LPU64")\n",
2581 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2583 LASSERT((oa->o_valid & OBD_MD_FLFLAGS) &&
2584 (oa->o_flags == OBD_FL_DELORPHAN));
2586 rc = filter_clear_orphans(exp, oa);
2588 CERROR("cannot clear orphanes starting from "
2589 LPU64", err = %d\n", oa->o_id, rc);
2591 rc = filter_update_last_objid(obd, group, 0);
2593 CERROR("unable to write lastobjid, but "
2594 "orphans were deleted, err = %d\n",
2598 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2603 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
2604 struct lov_stripe_md *ea, struct obd_trans_info *oti)
2606 int rc, rc2, cleanup_phase = 0, have_prepared = 0;
2607 struct dentry *dchild = NULL, *dparent = NULL;
2608 struct llog_cookie *fcc = NULL;
2609 struct lvfs_run_ctxt saved;
2610 struct filter_obd *filter;
2611 struct obd_device *obd;
2612 void *handle = NULL;
2617 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2620 filter = &obd->u.filter;
2622 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2625 dparent = filter_parent_lock(obd, oa->o_gr, oa->o_id, &lock);
2626 if (IS_ERR(dparent))
2627 GOTO(cleanup, rc = PTR_ERR(dparent));
2630 dchild = filter_id2dentry(obd, dparent, oa->o_gr, oa->o_id);
2632 GOTO(cleanup, rc = PTR_ERR(dchild));
2635 if (dchild->d_inode == NULL) {
2636 CDEBUG(D_INODE, "destroying non-existent object "LPU64"\n",
2639 GOTO(cleanup, rc = -ENOENT);
2642 if (!have_prepared) {
2643 /* If we're really going to destroy the object, get ready
2644 * by getting the clients to discard their cached data.
2646 * We have to drop the parent lock, because
2647 * filter_prepare_destroy will acquire a PW on the object, and
2648 * we don't want to deadlock with an incoming write to the
2649 * object, which has the extent PW and then wants to get the
2650 * parent dentry to do the lookup.
2652 * We dput the child because it's not worth the extra
2653 * complication of condition the above code to skip it on the
2654 * second time through. */
2656 filter_parent_unlock(dparent, lock);
2658 filter_prepare_destroy(obd, oa->o_id, oa->o_gr);
2663 /* Our MDC connection is established by the MDS to us */
2664 if (oa->o_valid & OBD_MD_FLCOOKIE) {
2665 OBD_ALLOC(fcc, sizeof(*fcc));
2667 memcpy(fcc, obdo_logcookie(oa), sizeof(*fcc));
2670 /* we're gonna truncate it first in order to avoid possible
2673 * open trasaction open transaction
2674 * down(i_zombie) down(i_zombie)
2675 * restart transaction
2676 * (see BUG 4180) -bzzz
2678 down(&dchild->d_inode->i_sem);
2679 handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_SETATTR,NULL,1);
2680 if (IS_ERR(handle)) {
2681 up(&dchild->d_inode->i_sem);
2682 GOTO(cleanup, rc = PTR_ERR(handle));
2685 iattr.ia_valid = ATTR_SIZE;
2687 rc = fsfilt_setattr(obd, dchild, handle, &iattr, 1);
2689 rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode, handle, 0);
2690 up(&dchild->d_inode->i_sem);
2694 GOTO(cleanup, rc = rc2);
2696 handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_UNLINK,oti,1);
2698 GOTO(cleanup, rc = PTR_ERR(handle));
2702 rc = filter_destroy_internal(obd, oa->o_id, dparent, dchild);
2705 switch(cleanup_phase) {
2708 fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
2709 oti ? oti->oti_handle : handle,
2710 filter_cancel_cookies_cb, fcc);
2712 rc = filter_finish_transno(exp, oti, rc);
2713 rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode,
2714 handle, exp->exp_sync);
2716 CERROR("error on commit, err = %d\n", rc2);
2723 filter_parent_unlock(dparent, lock);
2725 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2728 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
2735 /* NB start and end are used for punch, but not truncate */
2736 static int filter_truncate(struct obd_export *exp, struct obdo *oa,
2737 struct lov_stripe_md *lsm,
2738 obd_off start, obd_off end,
2739 struct obd_trans_info *oti)
2744 if (end != OBD_OBJECT_EOF)
2745 CERROR("PUNCH not supported, only truncate: end = "LPX64"\n",
2748 CDEBUG(D_INODE, "calling truncate for object "LPU64", valid = "LPU64", "
2749 "o_size = "LPD64"\n", oa->o_id, oa->o_valid, start);
2751 error = filter_setattr(exp, oa, NULL, oti);
2755 static int filter_sync(struct obd_export *exp, struct obdo *oa,
2756 struct lov_stripe_md *lsm, obd_off start, obd_off end)
2758 struct obd_device *obd = exp->exp_obd;
2759 struct lvfs_run_ctxt saved;
2760 struct filter_obd *filter;
2761 struct dentry *dentry;
2765 filter = &obd->u.filter;
2767 /* an objid of zero is taken to mean "sync whole filesystem" */
2768 if (!oa || !(oa->o_valid & OBD_MD_FLID)) {
2769 rc = fsfilt_sync(obd, filter->fo_sb);
2770 /* flush any remaining cancel messages out to the target */
2771 filter_sync_llogs(obd, NULL);
2775 dentry = filter_oa2dentry(obd, oa);
2777 RETURN(PTR_ERR(dentry));
2779 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2781 down(&dentry->d_inode->i_sem);
2782 rc = filemap_fdatawrite(dentry->d_inode->i_mapping);
2784 /* just any file to grab fsync method - "file" arg unused */
2785 struct file *file = filter->fo_rcvd_filp;
2787 if (file->f_op && file->f_op->fsync)
2788 rc = file->f_op->fsync(NULL, dentry, 1);
2790 rc2 = filemap_fdatawait(dentry->d_inode->i_mapping);
2794 up(&dentry->d_inode->i_sem);
2796 oa->o_valid = OBD_MD_FLID;
2797 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2799 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2805 static int filter_set_info(struct obd_export *exp, __u32 keylen,
2806 void *key, __u32 vallen, void *val)
2808 struct obd_device *obd;
2812 obd = class_exp2obd(exp);
2814 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2815 exp->exp_handle.h_cookie);
2819 if (keylen == 8 && memcmp(key, "auditlog", 8) == 0) {
2821 rc = fsfilt_set_info(obd, obd->u.filter.fo_sb, NULL,
2822 8, "auditlog", vallen, val);
2824 } else if (keylen == 5 && strcmp(key, "audit") == 0) {
2825 //set audit for whole FS on OSS
2826 struct audit_attr_msg * msg = val;
2828 rc = fsfilt_set_info(obd, obd->u.filter.fo_sb, NULL,
2829 5, "audit", sizeof(msg->attr), &msg->attr);
2831 } else if (keylen == 9 && strcmp(key, "audit_obj") == 0) {
2832 struct obdo * oa = val;
2833 struct dentry * dentry;
2834 __u64 mask = oa->o_fid;
2836 dentry = filter_crow_object(obd, oa);
2838 RETURN(PTR_ERR(dentry));
2840 rc = fsfilt_set_info(obd, obd->u.filter.fo_sb, dentry->d_inode,
2841 5, "audit", sizeof(mask), &mask);
2850 CDEBUG(D_IOCTL, "invalid key\n");
2855 static int filter_get_info(struct obd_export *exp, __u32 keylen,
2856 void *key, __u32 *vallen, void *val)
2858 struct filter_export_data *fed = &exp->exp_filter_data;
2859 struct obd_device *obd;
2862 obd = class_exp2obd(exp);
2864 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2865 exp->exp_handle.h_cookie);
2869 if (keylen == strlen("blocksize") &&
2870 memcmp(key, "blocksize", keylen) == 0) {
2871 __u32 *blocksize = val;
2872 *vallen = sizeof(*blocksize);
2873 *blocksize = obd->u.filter.fo_sb->s_blocksize;
2877 if (keylen == strlen("blocksize_bits") &&
2878 memcmp(key, "blocksize_bits", keylen) == 0) {
2879 __u32 *blocksize_bits = val;
2880 *vallen = sizeof(*blocksize_bits);
2881 *blocksize_bits = obd->u.filter.fo_sb->s_blocksize_bits;
2885 if (keylen >= strlen("last_id") && memcmp(key, "last_id", 7) == 0) {
2886 obd_id *last_id = val;
2887 *last_id = filter_last_id(&obd->u.filter, fed->fed_group);
2890 if (keylen >= strlen("reint_log") && memcmp(key, "reint_log", 9) == 0) {
2891 /*Get log_context handle*/
2892 unsigned long *llh_handle = val;
2893 *vallen = sizeof(unsigned long);
2894 *llh_handle = (unsigned long)llog_get_context(&obd->obd_llogs,
2895 LLOG_REINT_ORIG_CTXT);
2898 if (keylen >= strlen("cache_sb") && memcmp(key, "cache_sb", 8) == 0) {
2899 /*Get log_context handle*/
2900 unsigned long *sb = val;
2901 *vallen = sizeof(unsigned long);
2902 *sb = (unsigned long)obd->u.filter.fo_sb;
2906 CDEBUG(D_IOCTL, "invalid key\n");
2910 struct obd_llogs *filter_grab_llog_for_group(struct obd_device *obd, int group,
2911 struct obd_export *export)
2913 struct filter_group_llog *fglog, *nlog;
2914 char name[32] = "CATLIST";
2915 struct filter_obd *filter;
2916 struct llog_ctxt *ctxt;
2917 struct list_head *cur;
2920 filter = &obd->u.filter;
2922 spin_lock(&filter->fo_llog_list_lock);
2923 list_for_each(cur, &filter->fo_llog_list) {
2924 fglog = list_entry(cur, struct filter_group_llog, list);
2925 if (fglog->group == group) {
2926 if (!(fglog->exp == NULL || fglog->exp == export || export == NULL))
2927 CWARN("%s: export for group %d changes: 0x%p -> 0x%p\n",
2928 obd->obd_name, group, fglog->exp, export);
2929 spin_unlock(&filter->fo_llog_list_lock);
2933 spin_unlock(&filter->fo_llog_list_lock);
2938 OBD_ALLOC(fglog, sizeof(*fglog));
2941 fglog->group = group;
2943 OBD_ALLOC(fglog->llogs, sizeof(struct obd_llogs));
2944 if (fglog->llogs == NULL) {
2945 OBD_FREE(fglog, sizeof(*fglog));
2949 spin_lock(&filter->fo_llog_list_lock);
2950 list_for_each(cur, &filter->fo_llog_list) {
2951 nlog = list_entry(cur, struct filter_group_llog, list);
2952 LASSERT(nlog->group != group);
2954 list_add(&fglog->list, &filter->fo_llog_list);
2955 spin_unlock(&filter->fo_llog_list_lock);
2957 rc = obd_llog_cat_initialize(obd, fglog->llogs, 1, name);
2959 OBD_FREE(fglog->llogs, sizeof(*(fglog->llogs)));
2960 OBD_FREE(fglog, sizeof(*fglog));
2966 fglog->exp = export;
2967 ctxt = llog_get_context(fglog->llogs, LLOG_UNLINK_REPL_CTXT);
2968 LASSERT(ctxt != NULL);
2970 llog_receptor_accept(ctxt, export->exp_imp_reverse);
2972 CDEBUG(D_OTHER, "%s: new llog 0x%p for group %u\n",
2973 obd->obd_name, fglog->llogs, group);
2975 RETURN(fglog->llogs);
2978 int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
2979 int len, void *karg, void *uarg)
2981 struct obd_device *obd = exp->exp_obd;
2982 struct obd_ioctl_data *data = karg;
2986 case OBD_IOC_ABORT_RECOVERY:
2987 target_stop_recovery_thread(obd);
2990 case OBD_IOC_SET_READONLY: {
2992 struct super_block *sb = obd->u.filter.fo_sb;
2993 struct inode *inode = sb->s_root->d_inode;
2994 BDEVNAME_DECLARE_STORAGE(tmp);
2995 CERROR("setting device %s read-only\n",
2996 ll_bdevname(sb, tmp));
2998 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
3000 (void)fsfilt_commit(obd, sb, inode, handle, 1);
3002 ll_set_rdonly(ll_sbdev(obd->u.filter.fo_sb));
3006 case OBD_IOC_CATLOGLIST: {
3007 rc = llog_catalog_list(obd, 1, data);
3010 case OBD_IOC_LLOG_CANCEL:
3011 case OBD_IOC_LLOG_REMOVE:
3012 case OBD_IOC_LLOG_INFO:
3013 case OBD_IOC_LLOG_PRINT: {
3014 /* FIXME to be finished */
3015 RETURN(-EOPNOTSUPP);
3017 struct llog_ctxt *ctxt = NULL;
3019 push_ctxt(&saved, &ctxt->loc_ctxt, NULL);
3020 rc = llog_ioctl(ctxt, cmd, data);
3021 pop_ctxt(&saved, &ctxt->loc_ctxt, NULL);
3034 static struct llog_operations filter_unlink_repl_logops;
3035 static struct llog_operations filter_size_orig_logops = {
3036 lop_setup: llog_obd_origin_setup,
3037 lop_cleanup: llog_catalog_cleanup,
3038 lop_add: llog_catalog_add,
3041 static int filter_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
3042 struct obd_device *tgt, int count,
3043 struct llog_catid *catid)
3045 struct llog_ctxt *ctxt;
3049 filter_unlink_repl_logops = llog_client_ops;
3050 filter_unlink_repl_logops.lop_cancel = llog_obd_repl_cancel;
3051 filter_unlink_repl_logops.lop_connect = llog_repl_connect;
3052 filter_unlink_repl_logops.lop_sync = llog_obd_repl_sync;
3054 rc = obd_llog_setup(obd, llogs, LLOG_UNLINK_REPL_CTXT, tgt, 0, NULL,
3055 &filter_unlink_repl_logops);
3058 /* FIXME - assign unlink_cb for filter's recovery */
3059 ctxt = llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT);
3060 ctxt->llog_proc_cb = filter_recov_log_unlink_cb;
3062 filter_size_orig_logops = llog_lvfs_ops;
3064 filter_size_orig_logops.lop_setup = llog_obd_origin_setup;
3065 filter_size_orig_logops.lop_cleanup = llog_catalog_cleanup;
3066 filter_size_orig_logops.lop_add = llog_catalog_add;
3068 rc = obd_llog_setup(obd, llogs, LLOG_SIZE_ORIG_CTXT, tgt, 1,
3069 &catid->lci_logid, &filter_size_orig_logops);
3073 static int filter_llog_finish(struct obd_device *obd,
3074 struct obd_llogs *llogs, int count)
3079 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT));
3083 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_SIZE_ORIG_CTXT));
3087 static int filter_llog_connect(struct obd_export *exp,
3088 struct llogd_conn_body *body)
3090 struct obd_device *obd = exp->exp_obd;
3091 struct llog_ctxt *ctxt;
3092 struct obd_llogs *llog;
3096 CDEBUG(D_OTHER, "handle connect for %s: %u/%u/%u\n", obd->obd_name,
3097 (unsigned) body->lgdc_logid.lgl_ogr,
3098 (unsigned) body->lgdc_logid.lgl_oid,
3099 (unsigned) body->lgdc_logid.lgl_ogen);
3100 llog = filter_grab_llog_for_group(obd, body->lgdc_logid.lgl_ogr, exp);
3101 LASSERT(llog != NULL);
3102 ctxt = llog_get_context(llog, body->lgdc_ctxt_idx);
3103 rc = llog_connect(ctxt, 1, &body->lgdc_logid,
3104 &body->lgdc_gen, NULL);
3106 CERROR("failed to connect\n");
3111 static struct dentry *filter_lvfs_id2dentry(__u64 id, __u32 gen,
3112 __u64 gr, void *data)
3114 return filter_id2dentry(data, NULL, gr, id);
3117 static struct lvfs_callback_ops filter_lvfs_ops = {
3118 l_id2dentry: filter_lvfs_id2dentry,
3121 static struct obd_ops filter_obd_ops = {
3122 .o_owner = THIS_MODULE,
3123 .o_attach = filter_attach,
3124 .o_detach = filter_detach,
3125 .o_set_info = filter_set_info,
3126 .o_get_info = filter_get_info,
3127 .o_setup = filter_setup,
3128 .o_precleanup = filter_precleanup,
3129 .o_cleanup = filter_cleanup,
3130 .o_process_config = filter_process_config,
3131 .o_connect = filter_connect,
3132 .o_connect_post = filter_connect_post,
3133 .o_disconnect = filter_disconnect,
3134 .o_statfs = filter_statfs,
3135 .o_getattr = filter_getattr,
3136 .o_unpackmd = filter_unpackmd,
3137 .o_create = filter_create,
3138 .o_setattr = filter_setattr,
3139 .o_destroy = filter_destroy,
3140 .o_brw = filter_brw,
3141 .o_punch = filter_truncate,
3142 .o_sync = filter_sync,
3143 .o_preprw = filter_preprw,
3144 .o_commitrw = filter_commitrw,
3145 .o_do_cow = filter_do_cow,
3146 .o_write_extents = filter_write_extents,
3147 .o_destroy_export = filter_destroy_export,
3148 .o_llog_init = filter_llog_init,
3149 .o_llog_finish = filter_llog_finish,
3150 .o_llog_connect = filter_llog_connect,
3151 .o_iocontrol = filter_iocontrol,
3154 static struct obd_ops filter_sanobd_ops = {
3155 .o_owner = THIS_MODULE,
3156 .o_attach = filter_attach,
3157 .o_detach = filter_detach,
3158 .o_get_info = filter_get_info,
3159 .o_setup = filter_san_setup,
3160 .o_precleanup = filter_precleanup,
3161 .o_cleanup = filter_cleanup,
3162 .o_connect = filter_connect,
3163 .o_connect_post = filter_connect_post,
3164 .o_disconnect = filter_disconnect,
3165 .o_statfs = filter_statfs,
3166 .o_getattr = filter_getattr,
3167 .o_unpackmd = filter_unpackmd,
3168 .o_create = filter_create,
3169 .o_setattr = filter_setattr,
3170 .o_destroy = filter_destroy,
3171 .o_brw = filter_brw,
3172 .o_punch = filter_truncate,
3173 .o_sync = filter_sync,
3174 .o_preprw = filter_preprw,
3175 .o_commitrw = filter_commitrw,
3176 .o_do_cow = filter_do_cow,
3177 .o_write_extents = filter_write_extents,
3178 .o_san_preprw = filter_san_preprw,
3179 .o_destroy_export = filter_destroy_export,
3180 .o_llog_init = filter_llog_init,
3181 .o_llog_finish = filter_llog_finish,
3182 .o_llog_connect = filter_llog_connect,
3183 .o_iocontrol = filter_iocontrol,
3186 static int __init obdfilter_init(void)
3188 struct lprocfs_static_vars lvars;
3191 printk(KERN_INFO "Lustre: Filtering OBD driver; info@clusterfs.com\n");
3193 lprocfs_init_vars(filter, &lvars);
3195 size = OBDFILTER_CREATED_SCRATCHPAD_ENTRIES *
3196 sizeof(*obdfilter_created_scratchpad);
3198 OBD_ALLOC(obdfilter_created_scratchpad, size);
3199 if (obdfilter_created_scratchpad == NULL) {
3200 CERROR ("Can't allocate scratchpad\n");
3204 rc = class_register_type(&filter_obd_ops, NULL, lvars.module_vars,
3205 OBD_FILTER_DEVICENAME);
3207 OBD_FREE(obdfilter_created_scratchpad, size);
3211 rc = class_register_type(&filter_sanobd_ops, NULL, lvars.module_vars,
3212 OBD_FILTER_SAN_DEVICENAME);
3214 class_unregister_type(OBD_FILTER_DEVICENAME);
3215 OBD_FREE(obdfilter_created_scratchpad, size);
3220 static void __exit obdfilter_exit(void)
3222 class_unregister_type(OBD_FILTER_SAN_DEVICENAME);
3223 class_unregister_type(OBD_FILTER_DEVICENAME);
3224 OBD_FREE(obdfilter_created_scratchpad,
3225 OBDFILTER_CREATED_SCRATCHPAD_ENTRIES *
3226 sizeof(*obdfilter_created_scratchpad));
3229 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
3230 MODULE_DESCRIPTION("Lustre Filtering OBD driver");
3231 MODULE_LICENSE("GPL");
3233 module_init(obdfilter_init);
3234 module_exit(obdfilter_exit);