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 "filter_internal.h"
66 /* Group 0 is no longer a legal group, to catch uninitialized IDs */
67 #define FILTER_MIN_GROUPS 3
69 static struct lvfs_callback_ops filter_lvfs_ops;
71 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
72 struct lov_stripe_md *ea, struct obd_trans_info *);
73 struct obd_llogs *filter_grab_llog_for_group(struct obd_device *,
74 int, struct obd_export *);
76 static void filter_commit_cb(struct obd_device *obd, __u64 transno,
77 void *cb_data, int error)
79 obd_transno_commit_cb(obd, transno, error);
82 /* Assumes caller has already pushed us into the kernel context. */
83 int filter_finish_transno(struct obd_export *exp, struct obd_trans_info *oti,
86 struct filter_obd *filter = &exp->exp_obd->u.filter;
87 struct filter_export_data *fed = &exp->exp_filter_data;
88 struct filter_client_data *fcd = fed->fed_fcd;
91 int err, log_pri = D_HA;
93 /* Propagate error code. */
97 if (!exp->exp_obd->obd_replayable || oti == NULL)
100 /* we don't allocate new transnos for replayed requests */
101 if (oti->oti_transno == 0) {
102 spin_lock(&filter->fo_translock);
103 last_rcvd = le64_to_cpu(filter->fo_fsd->fsd_last_transno) + 1;
104 filter->fo_fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
105 spin_unlock(&filter->fo_translock);
106 oti->oti_transno = last_rcvd;
108 spin_lock(&filter->fo_translock);
109 last_rcvd = oti->oti_transno;
110 if (last_rcvd > le64_to_cpu(filter->fo_fsd->fsd_last_transno))
111 filter->fo_fsd->fsd_last_transno =
112 cpu_to_le64(last_rcvd);
113 spin_unlock(&filter->fo_translock);
116 fcd->fcd_last_rcvd = cpu_to_le64(last_rcvd);
118 /* could get xid from oti, if it's ever needed */
119 fcd->fcd_last_xid = 0;
121 off = fed->fed_lr_off;
123 fsfilt_add_journal_cb(exp->exp_obd, filter->fo_sb, last_rcvd,
124 oti->oti_handle, filter_commit_cb, NULL);
126 err = fsfilt_write_record(exp->exp_obd, filter->fo_rcvd_filp, fcd,
127 sizeof(*fcd), &off, 0);
134 CDEBUG(log_pri, "wrote trans "LPU64" for client %s at #%d: err = %d\n",
135 last_rcvd, fcd->fcd_uuid, fed->fed_lr_idx, err);
140 void f_dput(struct dentry *dentry)
142 /* Can't go inside filter_ddelete because it can block */
143 CDEBUG(D_INODE, "putting %s: %p, count = %d\n",
144 dentry->d_name.name, dentry, atomic_read(&dentry->d_count) - 1);
145 LASSERT(atomic_read(&dentry->d_count) > 0);
150 /* Add client data to the FILTER. We use a bitmap to locate a free space
151 * in the last_rcvd file if cl_idx is -1 (i.e. a new client).
152 * Otherwise, we have just read the data from the last_rcvd file and
153 * we know its offset. */
154 static int filter_client_add(struct obd_device *obd, struct filter_obd *filter,
155 struct filter_export_data *fed, int cl_idx)
157 unsigned long *bitmap = filter->fo_last_rcvd_slots;
158 int new_client = (cl_idx == -1);
161 LASSERT(bitmap != NULL);
163 /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
164 if (!strcmp((char *)fed->fed_fcd->fcd_uuid, (char *)obd->obd_uuid.uuid))
167 /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
168 * there's no need for extra complication here
171 cl_idx = find_first_zero_bit(bitmap, FILTER_LR_MAX_CLIENTS);
173 if (cl_idx >= FILTER_LR_MAX_CLIENTS) {
174 CERROR("no client slots - fix FILTER_LR_MAX_CLIENTS\n");
177 if (test_and_set_bit(cl_idx, bitmap)) {
178 CERROR("FILTER client %d: found bit is set in bitmap\n",
180 cl_idx = find_next_zero_bit(bitmap,
181 FILTER_LR_MAX_CLIENTS,
186 if (test_and_set_bit(cl_idx, bitmap)) {
187 CERROR("FILTER client %d: bit already set in bitmap!\n",
193 fed->fed_lr_idx = cl_idx;
194 fed->fed_lr_off = le32_to_cpu(filter->fo_fsd->fsd_client_start) +
195 cl_idx * le16_to_cpu(filter->fo_fsd->fsd_client_size);
197 CDEBUG(D_INFO, "client at index %d (%llu) with UUID '%s' added\n",
198 fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
201 struct lvfs_run_ctxt saved;
202 loff_t off = fed->fed_lr_off;
206 CDEBUG(D_INFO, "writing client fcd at idx %u (%llu) (len %u)\n",
207 fed->fed_lr_idx,off,(unsigned int)sizeof(*fed->fed_fcd));
209 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
210 /* Transaction needed to fix bug 1403 */
211 handle = fsfilt_start(obd,
212 filter->fo_rcvd_filp->f_dentry->d_inode,
213 FSFILT_OP_SETATTR, NULL);
214 if (IS_ERR(handle)) {
215 err = PTR_ERR(handle);
216 CERROR("unable to start transaction: rc %d\n", err);
218 err = fsfilt_write_record(obd, filter->fo_rcvd_filp,
220 sizeof(*fed->fed_fcd),
222 fsfilt_commit(obd, filter->fo_sb,
223 filter->fo_rcvd_filp->f_dentry->d_inode,
226 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
229 CERROR("error writing %s client idx %u: rc %d\n",
230 LAST_RCVD, fed->fed_lr_idx, err);
237 static int filter_client_free(struct obd_export *exp, int flags)
239 struct filter_export_data *fed = &exp->exp_filter_data;
240 struct filter_obd *filter = &exp->exp_obd->u.filter;
241 struct obd_device *obd = exp->exp_obd;
242 struct filter_client_data zero_fcd;
243 struct lvfs_run_ctxt saved;
248 if (fed->fed_fcd == NULL)
251 if (flags & OBD_OPT_FAILOVER)
254 /* XXX if fcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
255 if (!strcmp((char *)fed->fed_fcd->fcd_uuid, (char *)obd->obd_uuid.uuid))
258 LASSERT(filter->fo_last_rcvd_slots != NULL);
260 off = fed->fed_lr_off;
262 CDEBUG(D_INFO, "freeing client at idx %u (%lld) with UUID '%s'\n",
263 fed->fed_lr_idx, fed->fed_lr_off, fed->fed_fcd->fcd_uuid);
265 /* Clear the bit _after_ zeroing out the client so we don't
266 race with filter_client_add and zero out new clients.*/
267 if (!test_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
268 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
273 memset(&zero_fcd, 0, sizeof zero_fcd);
274 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
275 rc = fsfilt_write_record(obd, filter->fo_rcvd_filp, &zero_fcd,
276 sizeof(zero_fcd), &off, 1);
278 /* update server's transno */
279 filter_update_server_data(obd, filter->fo_rcvd_filp,
281 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
283 CDEBUG(rc == 0 ? D_INFO : D_ERROR,
284 "zeroing disconnecting client %s at idx %u (%llu) in %s rc %d\n",
285 fed->fed_fcd->fcd_uuid, fed->fed_lr_idx, fed->fed_lr_off,
288 if (!test_and_clear_bit(fed->fed_lr_idx, filter->fo_last_rcvd_slots)) {
289 CERROR("FILTER client %u: bit already clear in bitmap!!\n",
295 OBD_FREE(fed->fed_fcd, sizeof(*fed->fed_fcd));
300 static int filter_free_server_data(struct filter_obd *filter)
302 OBD_FREE(filter->fo_fsd, sizeof(*filter->fo_fsd));
303 filter->fo_fsd = NULL;
304 OBD_FREE(filter->fo_last_rcvd_slots, FILTER_LR_MAX_CLIENTS/8);
305 filter->fo_last_rcvd_slots = NULL;
309 /* assumes caller is already in kernel ctxt */
310 int filter_update_server_data(struct obd_device *obd, struct file *filp,
311 struct filter_server_data *fsd, int force_sync)
317 CDEBUG(D_INODE, "server uuid : %s\n", fsd->fsd_uuid);
318 CDEBUG(D_INODE, "server last_rcvd : "LPU64"\n",
319 le64_to_cpu(fsd->fsd_last_transno));
320 CDEBUG(D_INODE, "server last_mount: "LPU64"\n",
321 le64_to_cpu(fsd->fsd_mount_count));
323 rc = fsfilt_write_record(obd, filp, fsd, sizeof(*fsd), &off,force_sync);
325 CERROR("error writing filter_server_data: rc = %d\n", rc);
330 int filter_update_last_objid(struct obd_device *obd, obd_gr group,
333 struct filter_obd *filter = &obd->u.filter;
339 if (filter->fo_last_objid_files[group] == NULL) {
340 CERROR("Object group "LPU64" not fully setup; not updating "
341 "last_objid\n", group);
345 CDEBUG(D_INODE, "server last_objid for group "LPU64": "LPU64"\n",
346 group, filter->fo_last_objids[group]);
348 tmp = cpu_to_le64(filter->fo_last_objids[group]);
349 rc = fsfilt_write_record(obd, filter->fo_last_objid_files[group],
350 &tmp, sizeof(tmp), &off, force_sync);
352 CERROR("error writing group "LPU64" last objid: rc = %d\n",
357 /* assumes caller has already in kernel ctxt */
358 static int filter_init_server_data(struct obd_device *obd, struct file * filp)
360 struct filter_obd *filter = &obd->u.filter;
361 struct filter_server_data *fsd;
362 struct filter_client_data *fcd = NULL;
363 struct inode *inode = filp->f_dentry->d_inode;
364 unsigned long last_rcvd_size = inode->i_size;
370 /* ensure padding in the struct is the correct size */
371 LASSERT (offsetof(struct filter_server_data, fsd_padding) +
372 sizeof(fsd->fsd_padding) == FILTER_LR_SERVER_SIZE);
373 LASSERT (offsetof(struct filter_client_data, fcd_padding) +
374 sizeof(fcd->fcd_padding) == FILTER_LR_CLIENT_SIZE);
376 OBD_ALLOC(fsd, sizeof(*fsd));
379 filter->fo_fsd = fsd;
381 OBD_ALLOC(filter->fo_last_rcvd_slots, FILTER_LR_MAX_CLIENTS/8);
382 if (filter->fo_last_rcvd_slots == NULL) {
383 OBD_FREE(fsd, sizeof(*fsd));
387 if (last_rcvd_size == 0) {
388 CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD);
390 memcpy(fsd->fsd_uuid, obd->obd_uuid.uuid,sizeof(fsd->fsd_uuid));
391 fsd->fsd_last_transno = 0;
392 mount_count = fsd->fsd_mount_count = 0;
393 fsd->fsd_server_size = cpu_to_le32(FILTER_LR_SERVER_SIZE);
394 fsd->fsd_client_start = cpu_to_le32(FILTER_LR_CLIENT_START);
395 fsd->fsd_client_size = cpu_to_le16(FILTER_LR_CLIENT_SIZE);
396 fsd->fsd_subdir_count = cpu_to_le16(FILTER_SUBDIR_COUNT);
397 filter->fo_subdir_count = FILTER_SUBDIR_COUNT;
399 rc = fsfilt_read_record(obd, filp, fsd, sizeof(*fsd), &off);
401 CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
405 if (strcmp((char *)fsd->fsd_uuid, (char *)obd->obd_uuid.uuid)) {
406 CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
407 obd->obd_uuid.uuid, fsd->fsd_uuid);
408 GOTO(err_fsd, rc = -EINVAL);
410 mount_count = le64_to_cpu(fsd->fsd_mount_count);
411 filter->fo_subdir_count = le16_to_cpu(fsd->fsd_subdir_count);
414 if (fsd->fsd_feature_incompat & ~cpu_to_le32(FILTER_INCOMPAT_SUPP)) {
415 CERROR("unsupported feature %x\n",
416 le32_to_cpu(fsd->fsd_feature_incompat) &
417 ~FILTER_INCOMPAT_SUPP);
418 GOTO(err_fsd, rc = -EINVAL);
420 if (fsd->fsd_feature_rocompat & ~cpu_to_le32(FILTER_ROCOMPAT_SUPP)) {
421 CERROR("read-only feature %x\n",
422 le32_to_cpu(fsd->fsd_feature_rocompat) &
423 ~FILTER_ROCOMPAT_SUPP);
424 /* Do something like remount filesystem read-only */
425 GOTO(err_fsd, rc = -EINVAL);
428 CDEBUG(D_INODE, "%s: server last_rcvd : "LPU64"\n",
429 obd->obd_name, le64_to_cpu(fsd->fsd_last_transno));
430 CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
431 obd->obd_name, mount_count + 1);
432 CDEBUG(D_INODE, "%s: server data size: %u\n",
433 obd->obd_name, le32_to_cpu(fsd->fsd_server_size));
434 CDEBUG(D_INODE, "%s: per-client data start: %u\n",
435 obd->obd_name, le32_to_cpu(fsd->fsd_client_start));
436 CDEBUG(D_INODE, "%s: per-client data size: %u\n",
437 obd->obd_name, le32_to_cpu(fsd->fsd_client_size));
438 CDEBUG(D_INODE, "%s: server subdir_count: %u\n",
439 obd->obd_name, le16_to_cpu(fsd->fsd_subdir_count));
440 CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
441 last_rcvd_size <= le32_to_cpu(fsd->fsd_client_start) ? 0 :
442 (last_rcvd_size - le32_to_cpu(fsd->fsd_client_start)) /
443 le16_to_cpu(fsd->fsd_client_size));
445 if (!obd->obd_replayable) {
446 CWARN("%s: recovery support OFF\n", obd->obd_name);
450 for (cl_idx = 0, off = le32_to_cpu(fsd->fsd_client_start);
451 off < last_rcvd_size; cl_idx++) {
453 struct obd_export *exp;
454 struct filter_export_data *fed;
457 OBD_ALLOC(fcd, sizeof(*fcd));
459 GOTO(err_client, rc = -ENOMEM);
462 /* Don't assume off is incremented properly by
463 * fsfilt_read_record(), in case sizeof(*fcd)
464 * isn't the same as fsd->fsd_client_size. */
465 off = le32_to_cpu(fsd->fsd_client_start) +
466 cl_idx * le16_to_cpu(fsd->fsd_client_size);
467 rc = fsfilt_read_record(obd, filp, fcd, sizeof(*fcd), &off);
469 CERROR("error reading FILT %s idx %d off %llu: rc %d\n",
470 LAST_RCVD, cl_idx, off, rc);
471 break; /* read error shouldn't cause startup to fail */
474 if (fcd->fcd_uuid[0] == '\0') {
475 CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
480 last_rcvd = le64_to_cpu(fcd->fcd_last_rcvd);
482 /* These exports are cleaned up by filter_disconnect(), so they
483 * need to be set up like real exports as filter_connect() does.
485 exp = class_new_export(obd);
486 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64" "
487 "srv lr: "LPU64" fcd_group %d \n", fcd->fcd_uuid, cl_idx,
488 last_rcvd, le64_to_cpu(fsd->fsd_last_transno),
489 le32_to_cpu(fcd->fcd_group));
491 GOTO(err_client, rc = -ENOMEM);
493 memcpy(&exp->exp_client_uuid.uuid, fcd->fcd_uuid,
494 sizeof exp->exp_client_uuid.uuid);
495 fed = &exp->exp_filter_data;
497 fed->fed_group = le32_to_cpu(fcd->fcd_group);
498 filter_client_add(obd, filter, fed, cl_idx);
499 /* create helper if export init gets more complex */
500 spin_lock_init(&fed->fed_lock);
503 exp->exp_connected = 0;
504 exp->exp_req_replay_needed = 1;
505 exp->exp_lock_replay_needed = 1;
506 atomic_inc(&obd->obd_req_replay_clients);
507 atomic_inc(&obd->obd_lock_replay_clients);
508 obd->obd_recoverable_clients++;
509 obd->obd_max_recoverable_clients++;
510 class_export_put(exp);
512 CDEBUG(D_OTHER, "client at idx %d has last_rcvd = "LPU64"\n",
515 if (last_rcvd > le64_to_cpu(fsd->fsd_last_transno))
516 fsd->fsd_last_transno = cpu_to_le64(last_rcvd);
520 obd->obd_last_committed = le64_to_cpu(fsd->fsd_last_transno);
522 if (obd->obd_recoverable_clients) {
523 CWARN("RECOVERY: service %s, %d recoverable clients, "
524 "last_transno "LPU64"\n", obd->obd_name,
525 obd->obd_recoverable_clients,
526 le64_to_cpu(fsd->fsd_last_transno));
527 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
528 target_start_recovery_thread(obd, ost_handle);
532 OBD_FREE(fcd, sizeof(*fcd));
535 filter->fo_mount_count = mount_count + 1;
536 fsd->fsd_mount_count = cpu_to_le64(filter->fo_mount_count);
538 /* save it, so mount count and last_transno is current */
539 rc = filter_update_server_data(obd, filp, filter->fo_fsd, 1);
541 GOTO(err_client, rc);
545 class_disconnect_exports(obd, 0);
547 filter_free_server_data(filter);
551 static int filter_cleanup_groups(struct obd_device *obd)
553 struct filter_obd *filter = &obd->u.filter;
554 struct dentry *dentry;
558 for (i = 0; i < filter->fo_group_count; i++) {
559 if (filter->fo_subdirs != NULL) {
560 for (k = 0; k < filter->fo_subdir_count; k++) {
561 dentry = filter->fo_subdirs[i].dentry[k];
565 filter->fo_subdirs[i].dentry[k] = NULL;
568 if (filter->fo_last_objid_files[i] != NULL) {
569 filp_close(filter->fo_last_objid_files[i], 0);
570 filter->fo_last_objid_files[i] = NULL;
572 if (filter->fo_groups[i] != NULL) {
573 dput(filter->fo_groups[i]);
574 filter->fo_groups[i] = NULL;
577 if (filter->fo_subdirs != NULL)
578 OBD_FREE(filter->fo_subdirs,
579 filter->fo_group_count * sizeof(*filter->fo_subdirs));
580 if (filter->fo_groups != NULL)
581 OBD_FREE(filter->fo_groups,
582 filter->fo_group_count * sizeof(*filter->fo_groups));
583 if (filter->fo_last_objids != NULL)
584 OBD_FREE(filter->fo_last_objids,
585 filter->fo_group_count * sizeof(__u64));
586 if (filter->fo_last_objid_files != NULL)
587 OBD_FREE(filter->fo_last_objid_files,
588 filter->fo_group_count * sizeof(struct file *));
589 f_dput(filter->fo_dentry_O);
593 static int filter_update_last_group(struct obd_device *obd, int group)
595 struct filter_obd *filter = &obd->u.filter;
596 struct file *filp = NULL;
597 int last_group = 0, rc;
601 if (group <= filter->fo_committed_group)
604 filp = filp_open("LAST_GROUP", O_RDWR, 0700);
608 CERROR("cannot open LAST_GROUP: rc = %d\n", rc);
612 rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
614 CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n",rc);
617 LASSERT(off == 0 || last_group >= FILTER_MIN_GROUPS);
618 CDEBUG(D_INODE, "%s: previous %d, new %d\n",
619 obd->obd_name, last_group, group);
623 /* must be sync: bXXXX */
624 rc = fsfilt_write_record(obd, filp, &last_group, sizeof(__u32), &off, 1);
626 CDEBUG(D_INODE, "error updating LAST_GROUP: rc %d\n", rc);
630 filter->fo_committed_group = group;
637 static int filter_read_group_internal(struct obd_device *obd, int group,
640 struct filter_obd *filter = &obd->u.filter;
641 __u64 *new_objids = NULL;
642 struct filter_subdirs *new_subdirs = NULL, *tmp_subdirs = NULL;
643 struct dentry **new_groups = NULL;
644 struct file **new_files = NULL;
645 struct dentry *dentry;
647 int old_count = filter->fo_group_count, rc, stage = 0, i;
652 snprintf(name, 24, "%d", group);
656 dentry = ll_lookup_one_len(name, filter->fo_dentry_O,
658 if (IS_ERR(dentry)) {
659 CERROR("Cannot lookup expected object group %d: %ld\n",
660 group, PTR_ERR(dentry));
661 RETURN(PTR_ERR(dentry));
664 dentry = simple_mkdir(filter->fo_dentry_O, name, 0700, 1);
665 if (IS_ERR(dentry)) {
666 CERROR("cannot lookup/create O/%s: rc = %ld\n", name,
668 RETURN(PTR_ERR(dentry));
673 snprintf(name, 24, "O/%d/LAST_ID", group);
675 filp = filp_open(name, O_CREAT | O_RDWR, 0700);
677 CERROR("cannot create %s: rc = %ld\n", name, PTR_ERR(filp));
678 GOTO(cleanup, rc = PTR_ERR(filp));
682 rc = fsfilt_read_record(obd, filp, &last_objid, sizeof(__u64), &off);
684 CDEBUG(D_INODE, "error reading %s: rc %d\n", name, rc);
688 if (filter->fo_subdir_count) {
689 OBD_ALLOC(tmp_subdirs, sizeof(*tmp_subdirs));
690 if (tmp_subdirs == NULL)
691 GOTO(cleanup, rc = -ENOMEM);
694 for (i = 0; i < filter->fo_subdir_count; i++) {
696 snprintf(dir, sizeof(dir), "d%u", i);
698 tmp_subdirs->dentry[i] = simple_mkdir(dentry, dir, 0700, 1);
699 if (IS_ERR(tmp_subdirs->dentry[i])) {
700 rc = PTR_ERR(tmp_subdirs->dentry[i]);
701 CERROR("can't lookup/create O/%d/%s: rc = %d\n",
706 /* turn on smfs plugins if any */
707 fsfilt_set_fs_flags(obd, tmp_subdirs->dentry[i]->d_inode,
710 CDEBUG(D_INODE, "got/created O/%d/%s: %p\n", group, dir,
711 tmp_subdirs->dentry[i]);
715 /* 'group' is an index; we need an array of length 'group + 1' */
716 if (group + 1 > old_count) {
718 OBD_ALLOC(new_objids, len * sizeof(*new_objids));
719 OBD_ALLOC(new_subdirs, len * sizeof(*new_subdirs));
720 OBD_ALLOC(new_groups, len * sizeof(*new_groups));
721 OBD_ALLOC(new_files, len * sizeof(*new_files));
723 if (new_objids == NULL || new_subdirs == NULL ||
724 new_groups == NULL || new_files == NULL)
725 GOTO(cleanup, rc = -ENOMEM);
727 memcpy(new_objids, filter->fo_last_objids,
728 old_count * sizeof(*new_objids));
729 memcpy(new_subdirs, filter->fo_subdirs,
730 old_count * sizeof(*new_subdirs));
731 memcpy(new_groups, filter->fo_groups,
732 old_count * sizeof(*new_groups));
733 memcpy(new_files, filter->fo_last_objid_files,
734 old_count * sizeof(*new_files));
737 OBD_FREE(filter->fo_last_objids,
738 old_count * sizeof(*new_objids));
739 OBD_FREE(filter->fo_subdirs,
740 old_count * sizeof(*new_subdirs));
741 OBD_FREE(filter->fo_groups,
742 old_count * sizeof(*new_groups));
743 OBD_FREE(filter->fo_last_objid_files,
744 old_count * sizeof(*new_files));
746 filter->fo_last_objids = new_objids;
747 filter->fo_subdirs = new_subdirs;
748 filter->fo_groups = new_groups;
749 filter->fo_last_objid_files = new_files;
750 filter->fo_group_count = len;
753 filter->fo_groups[group] = dentry;
754 filter->fo_last_objid_files[group] = filp;
755 if (filter->fo_subdir_count) {
756 filter->fo_subdirs[group] = *tmp_subdirs;
757 OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
760 filter_update_last_group(obd, group);
762 if (filp->f_dentry->d_inode->i_size == 0) {
763 filter->fo_last_objids[group] = FILTER_INIT_OBJID;
767 filter->fo_last_objids[group] = le64_to_cpu(last_objid);
768 CDEBUG(D_INODE, "%s: server last_objid group %d: "LPU64"\n",
769 obd->obd_name, group, last_objid);
774 if (new_objids != NULL)
775 OBD_FREE(new_objids, group * sizeof(*new_objids));
776 if (new_subdirs != NULL)
777 OBD_FREE(new_subdirs, group * sizeof(*new_subdirs));
778 if (new_groups != NULL)
779 OBD_FREE(new_groups, group * sizeof(*new_groups));
780 if (new_files != NULL)
781 OBD_FREE(new_files, group * sizeof(*new_files));
783 if (filter->fo_subdir_count) {
784 for (i = 0; i < filter->fo_subdir_count; i++) {
785 if (tmp_subdirs->dentry[i] != NULL)
786 dput(tmp_subdirs->dentry[i]);
788 OBD_FREE(tmp_subdirs, sizeof(*tmp_subdirs));
798 static int filter_read_groups(struct obd_device *obd, int last_group,
801 struct filter_obd *filter = &obd->u.filter;
802 int old_count, group, rc = 0;
804 down(&filter->fo_init_lock);
805 old_count = filter->fo_group_count;
806 for (group = old_count; group <= last_group; group++) {
808 continue; /* no group zero */
810 rc = filter_read_group_internal(obd, group, create);
814 up(&filter->fo_init_lock);
818 static int filter_prep_groups(struct obd_device *obd)
820 struct filter_obd *filter = &obd->u.filter;
821 struct dentry *dentry, *O_dentry;
822 int rc = 0, cleanup_phase = 0;
823 struct file *filp = NULL;
828 O_dentry = simple_mkdir(current->fs->pwd, "O", 0700, 1);
829 CDEBUG(D_INODE, "got/created O: %p\n", O_dentry);
830 if (IS_ERR(O_dentry)) {
831 rc = PTR_ERR(O_dentry);
832 CERROR("cannot open/create O: rc = %d\n", rc);
835 filter->fo_dentry_O = O_dentry;
836 cleanup_phase = 1; /* O_dentry */
838 /* Lookup "R" to tell if we're on an old OST FS and need to convert
839 * from O/R/<dir>/<objid> to O/0/<dir>/<objid>. This can be removed
840 * some time post 1.0 when all old-style OSTs have converted along
841 * with the init_objid hack. */
842 dentry = ll_lookup_one_len("R", O_dentry, 1);
844 GOTO(cleanup, rc = PTR_ERR(dentry));
845 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
846 struct dentry *O0_dentry = lookup_one_len("0", O_dentry, 1);
849 CWARN("converting OST to new object layout\n");
850 if (IS_ERR(O0_dentry)) {
851 rc = PTR_ERR(O0_dentry);
852 CERROR("error looking up O/0: rc %d\n", rc);
856 if (O0_dentry->d_inode) {
857 CERROR("Both O/R and O/0 exist. Fix manually.\n");
858 GOTO(cleanup_O0, rc = -EEXIST);
861 down(&O_dentry->d_inode->i_sem);
862 rc = vfs_rename(O_dentry->d_inode, dentry,
863 O_dentry->d_inode, O0_dentry);
864 up(&O_dentry->d_inode->i_sem);
867 CERROR("error renaming O/R to O/0: rc %d\n", rc);
868 GOTO(cleanup_O0, rc);
870 filter->fo_fsd->fsd_feature_incompat |=
871 cpu_to_le32(FILTER_INCOMPAT_GROUPS);
872 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
874 GOTO(cleanup_O0, rc);
886 cleanup_phase = 2; /* groups */
888 /* we have to initialize all groups before first connections from
889 * clients because they may send create/destroy for any group -bzzz */
890 filp = filp_open("LAST_GROUP", O_CREAT | O_RDWR, 0700);
892 CERROR("cannot create LAST_GROUP: rc = %ld\n", PTR_ERR(filp));
893 GOTO(cleanup, rc = PTR_ERR(filp));
895 cleanup_phase = 3; /* filp */
897 rc = fsfilt_read_record(obd, filp, &last_group, sizeof(__u32), &off);
899 CDEBUG(D_INODE, "error reading LAST_GROUP: rc %d\n", rc);
903 last_group = FILTER_MIN_GROUPS;
905 LASSERT(last_group >= FILTER_MIN_GROUPS);
908 CWARN("%s: initialize groups [%d,%d]\n", obd->obd_name,
909 FILTER_MIN_GROUPS, last_group);
910 filter->fo_committed_group = last_group;
911 rc = filter_read_groups(obd, last_group, 1);
919 switch (cleanup_phase) {
923 filter_cleanup_groups(obd);
925 f_dput(filter->fo_dentry_O);
926 filter->fo_dentry_O = NULL;
934 /* setup the object store with correct subdirectories */
935 static int filter_prep(struct obd_device *obd)
937 struct lvfs_run_ctxt saved;
938 struct filter_obd *filter = &obd->u.filter;
944 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
945 file = filp_open(LAST_RCVD, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
946 if (!file || IS_ERR(file)) {
948 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
953 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
954 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
955 file->f_dentry->d_inode->i_mode);
956 GOTO(err_filp, rc = -ENOENT);
959 /* steal operations */
960 inode = file->f_dentry->d_inode;
961 filter->fo_fop = file->f_op;
962 filter->fo_iop = inode->i_op;
963 filter->fo_aops = inode->i_mapping->a_ops;
965 rc = filter_init_server_data(obd, file);
967 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
970 filter->fo_rcvd_filp = file;
972 rc = filter_prep_groups(obd);
974 GOTO(err_server_data, rc);
977 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
982 //class_disconnect_exports(obd, 0);
983 filter_free_server_data(filter);
985 if (filp_close(file, 0))
986 CERROR("can't close %s after error\n", LAST_RCVD);
987 filter->fo_rcvd_filp = NULL;
991 /* cleanup the filter: write last used object id to status file */
992 static void filter_post(struct obd_device *obd)
994 struct lvfs_run_ctxt saved;
995 struct filter_obd *filter = &obd->u.filter;
998 /* XXX: filter_update_lastobjid used to call fsync_dev. It might be
999 * best to start a transaction with h_sync, because we removed this
1002 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1003 rc = filter_update_server_data(obd, filter->fo_rcvd_filp,
1006 CERROR("error writing server data: rc = %d\n", rc);
1008 for (i = 1; i < filter->fo_group_count; i++) {
1009 rc = filter_update_last_objid(obd, i,
1010 (i == filter->fo_group_count - 1));
1012 CERROR("error writing group %d lastobjid: rc = %d\n",
1016 rc = filp_close(filter->fo_rcvd_filp, 0);
1017 filter->fo_rcvd_filp = NULL;
1019 CERROR("error closing %s: rc = %d\n", LAST_RCVD, rc);
1021 filter_cleanup_groups(obd);
1022 filter_free_server_data(filter);
1023 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1025 filter_free_capa_keys(filter);
1028 static void filter_set_last_id(struct filter_obd *filter, int group, obd_id id)
1030 LASSERT(filter->fo_fsd != NULL);
1032 LASSERT(group < filter->fo_group_count);
1034 spin_lock(&filter->fo_objidlock);
1035 filter->fo_last_objids[group] = id;
1036 spin_unlock(&filter->fo_objidlock);
1039 __u64 filter_last_id(struct filter_obd *filter, int group)
1042 LASSERT(filter->fo_fsd != NULL);
1044 LASSERT(group < filter->fo_group_count);
1046 spin_lock(&filter->fo_objidlock);
1047 id = filter->fo_last_objids[group];
1048 spin_unlock(&filter->fo_objidlock);
1053 static void filter_save_last_id(struct filter_obd *filter, int group, obd_id id)
1056 LASSERT(group < filter->fo_group_count);
1058 spin_lock(&filter->fo_lastidlock);
1059 if (id > filter_last_id(filter, group))
1060 filter_set_last_id(filter, group, id);
1061 spin_unlock(&filter->fo_lastidlock);
1064 /* direct cut-n-paste of mds_blocking_ast() */
1065 static int filter_blocking_ast(struct ldlm_lock *lock,
1066 struct ldlm_lock_desc *desc,
1067 void *data, int flag)
1072 if (flag == LDLM_CB_CANCELING) {
1073 /* Don't need to do anything here. */
1077 /* XXX layering violation! -phil */
1078 lock_res_and_lock(lock);
1079 /* Get this: if filter_blocking_ast is racing with ldlm_intent_policy,
1080 * such that filter_blocking_ast is called just before l_i_p takes the
1081 * ns_lock, then by the time we get the lock, we might not be the
1082 * correct blocking function anymore. So check, and return early, if
1084 if (lock->l_blocking_ast != filter_blocking_ast) {
1085 unlock_res_and_lock(lock);
1089 lock->l_flags |= LDLM_FL_CBPENDING;
1090 do_ast = (!lock->l_readers && !lock->l_writers);
1091 unlock_res_and_lock(lock);
1094 struct lustre_handle lockh;
1097 LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
1098 ldlm_lock2handle(lock, &lockh);
1099 rc = ldlm_cli_cancel(&lockh);
1101 CERROR("ldlm_cli_cancel: %d\n", rc);
1103 LDLM_DEBUG(lock, "Lock still has references, will be "
1109 extern void *lock_dir(struct inode *dir, struct qstr *name);
1110 extern void unlock_dir(struct inode *dir, void *lock);
1112 static void *filter_lock_dentry(struct obd_device *obd,
1113 struct dentry *dparent,
1121 len = sprintf(name, LPU64, id);
1122 qstr_assign(&qstr, (char *)name, len);
1123 return lock_dir(dparent->d_inode, &qstr);
1125 down(&dparent->d_inode->i_sem);
1130 /* We never dget the object parent, so DON'T dput it either */
1131 static void filter_parent_unlock(struct dentry *dparent, void *lock)
1134 LASSERT(lock != NULL);
1135 unlock_dir(dparent->d_inode, lock);
1137 up(&dparent->d_inode->i_sem);
1141 /* We never dget the object parent, so DON'T dput it either */
1142 struct dentry *filter_parent(struct obd_device *obd, obd_gr group, obd_id objid)
1144 struct filter_obd *filter = &obd->u.filter;
1145 LASSERTF(group < filter->fo_group_count,
1146 "group %d < filter->fo_group_count %d failed\n",
1147 (int)group, (int)filter->fo_group_count);
1148 LASSERTF(group > 0, "group %d > 0 failed\n", (int)group);
1150 if (filter->fo_subdir_count == 0)
1151 return filter->fo_groups[group];
1153 return filter->fo_subdirs[group].dentry[objid & (filter->fo_subdir_count - 1)];
1156 /* We never dget the object parent, so DON'T dput it either */
1157 struct dentry *filter_parent_lock(struct obd_device *obd, obd_gr group,
1158 obd_id objid, void **lock)
1160 unsigned long now = jiffies;
1161 struct dentry *dparent = filter_parent(obd, group, objid);
1163 if (IS_ERR(dparent))
1167 LASSERT(dparent->d_inode);
1169 *lock = filter_lock_dentry(obd, dparent, objid);
1170 fsfilt_check_slow(now, obd_timeout, "parent lock");
1174 /* How to get files, dentries, inodes from object id's.
1176 * If dir_dentry is passed, the caller has already locked the parent
1177 * appropriately for this operation (normally a write lock). If
1178 * dir_dentry is NULL, we do a read lock while we do the lookup to
1179 * avoid races with create/destroy and such changing the directory
1180 * internal to the filesystem code. */
1181 struct dentry *filter_id2dentry(struct obd_device *obd,
1182 struct dentry *dir_dentry,
1183 obd_gr group, obd_id id)
1185 struct dentry *dparent = dir_dentry;
1186 struct dentry *dchild;
1193 CERROR("fatal: invalid object id 0\n");
1194 RETURN(ERR_PTR(-ESTALE));
1197 len = sprintf(name, LPU64, id);
1198 if (dir_dentry == NULL) {
1199 dparent = filter_parent_lock(obd, group, id, &lock);
1200 if (IS_ERR(dparent)) {
1201 CERROR("%s: error getting object "LPU64":"LPU64
1202 " parent: rc %ld\n", obd->obd_name,
1203 id, group, PTR_ERR(dparent));
1207 CDEBUG(D_INODE, "looking up object O/%.*s/%s\n",
1208 dparent->d_name.len, dparent->d_name.name, name);
1209 dchild = /*ll_*/lookup_one_len(name, dparent, len);
1210 if (dir_dentry == NULL)
1211 filter_parent_unlock(dparent, lock);
1212 if (IS_ERR(dchild)) {
1213 CERROR("%s: child lookup error %ld\n", obd->obd_name,
1218 if (dchild->d_inode != NULL && is_bad_inode(dchild->d_inode)) {
1219 CERROR("%s: got bad inode "LPU64"\n", obd->obd_name, id);
1221 RETURN(ERR_PTR(-ENOENT));
1224 CDEBUG(D_INODE, "got child objid %s: %p, count = %d\n",
1225 name, dchild, atomic_read(&dchild->d_count));
1227 LASSERT(atomic_read(&dchild->d_count) > 0);
1232 static int filter_prepare_destroy(struct obd_device *obd, obd_id objid,
1235 struct lustre_handle lockh;
1236 int flags = LDLM_AST_DISCARD_DATA, rc;
1237 struct ldlm_res_id res_id = { .name = { objid, 0, group, 0 } };
1238 ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
1241 /* Tell the clients that the object is gone now and that they should
1242 * throw away any cached pages. */
1243 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, res_id,
1244 LDLM_EXTENT, &policy, LCK_PW,
1245 &flags, filter_blocking_ast, ldlm_completion_ast,
1246 NULL, NULL, NULL, 0, NULL, &lockh);
1248 /* We only care about the side-effects, just drop the lock. */
1250 ldlm_lock_decref(&lockh, LCK_PW);
1255 /* Caller must hold LCK_PW on parent and push us into kernel context.
1256 * Caller is also required to ensure that dchild->d_inode exists. */
1257 static int filter_destroy_internal(struct obd_device *obd, obd_id objid,
1258 struct dentry *dparent,
1259 struct dentry *dchild)
1261 struct inode *inode = dchild->d_inode;
1265 if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) {
1266 CERROR("destroying objid %.*s nlink = %lu, count = %d\n",
1267 dchild->d_name.len, dchild->d_name.name,
1268 (unsigned long)inode->i_nlink,
1269 atomic_read(&inode->i_count));
1272 rc = vfs_unlink(dparent->d_inode, dchild);
1275 CERROR("error unlinking objid %.*s: rc %d\n",
1276 dchild->d_name.len, dchild->d_name.name, rc);
1281 static int filter_intent_policy(struct ldlm_namespace *ns,
1282 struct ldlm_lock **lockp, void *req_cookie,
1283 ldlm_mode_t mode, int flags, void *data)
1285 struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
1286 struct ptlrpc_request *req = req_cookie;
1287 struct ldlm_lock *lock = *lockp, *l = NULL;
1288 struct ldlm_resource *res = lock->l_resource;
1289 ldlm_processing_policy policy;
1290 struct ost_lvb *res_lvb, *reply_lvb;
1291 struct ldlm_reply *rep;
1292 struct list_head *tmp;
1294 int tmpflags = 0, rc, repsize[2] = {sizeof(*rep), sizeof(*reply_lvb)};
1297 policy = ldlm_get_processing_policy(res);
1298 LASSERT(policy != NULL);
1299 LASSERT(req != NULL);
1301 rc = lustre_pack_reply(req, 2, repsize, NULL);
1303 RETURN(req->rq_status = rc);
1305 rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*rep));
1306 LASSERT(rep != NULL);
1308 reply_lvb = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*reply_lvb));
1309 LASSERT(reply_lvb != NULL);
1311 //fixup_handle_for_resent_req(req, lock, &lockh);
1313 /* If we grant any lock at all, it will be a whole-file read lock.
1314 * Call the extent policy function to see if our request can be
1315 * granted, or is blocked. */
1316 lock->l_policy_data.l_extent.start = 0;
1317 lock->l_policy_data.l_extent.end = OBD_OBJECT_EOF;
1318 lock->l_req_mode = LCK_PR;
1320 lock_res_and_lock(lock);
1321 res = lock->l_resource;
1322 rc = policy(lock, &tmpflags, 0, &err, &rpc_list);
1324 /* FIXME: we should change the policy function slightly, to not make
1325 * this list at all, since we just turn around and free it */
1326 while (!list_empty(&rpc_list)) {
1327 struct ldlm_lock *wlock =
1328 list_entry(rpc_list.next, struct ldlm_lock, l_cp_ast);
1329 LASSERT((lock->l_flags & LDLM_FL_AST_SENT) == 0);
1330 LASSERT(lock->l_flags & LDLM_FL_CP_REQD);
1331 lock->l_flags &= ~LDLM_FL_CP_REQD;
1332 list_del_init(&wlock->l_cp_ast);
1333 LDLM_LOCK_PUT(wlock);
1336 if (rc == LDLM_ITER_CONTINUE) {
1337 /* The lock met with no resistance; we're finished. */
1338 unlock_res_and_lock(lock);
1339 RETURN(ELDLM_LOCK_REPLACED);
1342 /* Do not grant any lock, but instead send GL callbacks. The extent
1343 * policy nicely created a list of all PW locks for us. We will choose
1344 * the highest of those which are larger than the size in the LVB, if
1345 * any, and perform a glimpse callback. */
1346 res_lvb = res->lr_lvb_data;
1347 LASSERT(res_lvb != NULL);
1348 *reply_lvb = *res_lvb;
1350 list_for_each(tmp, &res->lr_granted) {
1351 struct ldlm_lock *tmplock =
1352 list_entry(tmp, struct ldlm_lock, l_res_link);
1354 if (tmplock->l_granted_mode == LCK_PR)
1357 if (tmplock->l_policy_data.l_extent.end <= reply_lvb->lvb_size)
1361 l = LDLM_LOCK_GET(tmplock);
1365 if (l->l_policy_data.l_extent.start >
1366 tmplock->l_policy_data.l_extent.start)
1370 l = LDLM_LOCK_GET(tmplock);
1374 /* There were no PW locks beyond the size in the LVB; finished. */
1376 RETURN(ELDLM_LOCK_ABORTED);
1378 if (l->l_glimpse_ast == NULL) {
1379 /* We are racing with unlink(); just return -ENOENT */
1380 CDEBUG(D_OTHER, "race with unlink: "LPX64"/"LPX64"\n",
1381 res->lr_name.name[0], res->lr_name.name[2]);
1382 rep->lock_policy_res1 = -ENOENT;
1386 LASSERTF(l->l_glimpse_ast != NULL, "l == %p", l);
1388 rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
1389 if (rc != 0 && res->lr_namespace->ns_lvbo &&
1390 res->lr_namespace->ns_lvbo->lvbo_update) {
1391 res->lr_namespace->ns_lvbo->lvbo_update(res, NULL, 0, 1);
1395 *reply_lvb = *res_lvb;
1400 RETURN(ELDLM_LOCK_ABORTED);
1403 static int filter_post_fs_cleanup(struct obd_device *obd)
1407 rc = fsfilt_post_cleanup(obd);
1412 static int filter_post_fs_setup(struct obd_device *obd)
1414 struct filter_obd *filter = &obd->u.filter;
1418 rc = fsfilt_post_setup(obd, filter->fo_dentry_O);
1423 /* mount the file system (secretly) */
1424 int filter_common_setup(struct obd_device *obd, obd_count len, void *buf,
1427 struct lustre_cfg *lcfg = buf;
1428 struct filter_obd *filter = &obd->u.filter;
1429 struct lvfs_obd_ctxt *lvfs_ctxt = NULL;
1430 struct vfsmount *mnt;
1431 struct crypto_tfm *tfm;
1437 if ((LUSTRE_CFG_BUFLEN(lcfg, 1)) < 1 ||
1438 (LUSTRE_CFG_BUFLEN(lcfg, 2) < 1))
1441 obd->obd_fsops = fsfilt_get_ops(lustre_cfg_string(lcfg, 2));
1442 if (IS_ERR(obd->obd_fsops))
1443 RETURN(PTR_ERR(obd->obd_fsops));
1445 rc = lvfs_mount_fs(lustre_cfg_string(lcfg, 1),
1446 lustre_cfg_string(lcfg, 2),
1447 option, MS_NOATIME | MS_NODIRATIME, &lvfs_ctxt);
1449 CERROR("lvfs_mount_fs failed: rc = %d\n", rc);
1454 mnt = lvfs_ctxt->loc_mnt;
1455 filter->fo_lvfs_ctxt = lvfs_ctxt;
1457 if (LUSTRE_CFG_BUFLEN(lcfg, 3) > 0 && lustre_cfg_buf(lcfg, 3)) {
1458 str = lustre_cfg_string(lcfg, 3);
1460 obd->obd_replayable = 1;
1461 obd_sync_filter = 1;
1462 CWARN("%s: recovery enabled\n", obd->obd_name);
1465 CERROR("unrecognised flag '%c'\n",
1468 // XXX Robert? Why do we get errors here
1469 // GOTO(err_mntput, rc = -EINVAL);
1473 filter->fo_vfsmnt = mnt;
1474 filter->fo_sb = mnt->mnt_sb;
1475 filter->fo_fstype = mnt->mnt_sb->s_type->name;
1476 CDEBUG(D_SUPER, "%s: mnt = %p\n", filter->fo_fstype, mnt);
1478 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1479 obd->obd_lvfs_ctxt.pwdmnt = mnt;
1480 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
1481 obd->obd_lvfs_ctxt.fs = get_ds();
1482 obd->obd_lvfs_ctxt.cb_ops = filter_lvfs_ops;
1484 ll_clear_rdonly(ll_sbdev(filter->fo_sb));
1486 rc = fsfilt_setup(obd, mnt->mnt_sb);
1488 GOTO(err_mntput, rc);
1490 sema_init(&filter->fo_init_lock, 1);
1491 filter->fo_committed_group = 0;
1493 tfm = crypto_alloc_tfm(CAPA_HMAC_ALG, 0);
1495 GOTO(err_mntput, rc = -ENOSYS);
1497 filter->fo_capa_hmac = tfm;
1498 INIT_LIST_HEAD(&filter->fo_capa_keys);
1499 spin_lock_init(&filter->fo_capa_lock);
1501 rc = filter_prep(obd);
1505 filter->fo_destroys_in_progress = 0;
1506 for (i = 0; i < 32; i++)
1507 sema_init(&filter->fo_create_locks[i], 1);
1509 spin_lock_init(&filter->fo_translock);
1510 spin_lock_init(&filter->fo_objidlock);
1511 spin_lock_init(&filter->fo_lastidlock);
1512 INIT_LIST_HEAD(&filter->fo_export_list);
1513 sema_init(&filter->fo_alloc_lock, 1);
1514 spin_lock_init(&filter->fo_r_pages.oh_lock);
1515 spin_lock_init(&filter->fo_w_pages.oh_lock);
1516 spin_lock_init(&filter->fo_r_discont_pages.oh_lock);
1517 spin_lock_init(&filter->fo_w_discont_pages.oh_lock);
1518 spin_lock_init(&filter->fo_r_discont_blocks.oh_lock);
1519 spin_lock_init(&filter->fo_w_discont_blocks.oh_lock);
1520 filter->fo_readcache_max_filesize = FILTER_MAX_CACHE_SIZE;
1522 INIT_LIST_HEAD(&filter->fo_llog_list);
1523 spin_lock_init(&filter->fo_llog_list_lock);
1525 spin_lock_init(&filter->fo_denylist_lock);
1526 INIT_LIST_HEAD(&filter->fo_denylist);
1528 sprintf(ns_name, "filter-%s", obd->obd_uuid.uuid);
1529 obd->obd_namespace = ldlm_namespace_new(ns_name, LDLM_NAMESPACE_SERVER);
1531 if (obd->obd_namespace == NULL)
1532 GOTO(err_post, rc = -ENOMEM);
1533 obd->obd_namespace->ns_lvbp = obd;
1534 obd->obd_namespace->ns_lvbo = &filter_lvbo;
1535 ldlm_register_intent(obd->obd_namespace, filter_intent_policy);
1537 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1538 "filter_ldlm_cb_client", &obd->obd_ldlm_client);
1540 rc = obd_llog_cat_initialize(obd, &obd->obd_llogs, 1, CATLIST);
1542 CERROR("failed to setup llogging subsystems\n");
1550 crypto_free_tfm(filter->fo_capa_hmac);
1553 lvfs_umount_fs(filter->fo_lvfs_ctxt);
1557 fsfilt_put_ops(obd->obd_fsops);
1561 static int filter_attach(struct obd_device *obd, obd_count len, void *data)
1563 struct lprocfs_static_vars lvars;
1566 lprocfs_init_vars(filter, &lvars);
1567 rc = lprocfs_obd_attach(obd, lvars.obd_vars);
1571 rc = lprocfs_alloc_obd_stats(obd, LPROC_FILTER_LAST);
1575 /* Init obdfilter private stats here */
1576 lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_READ_BYTES,
1577 LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
1578 lprocfs_counter_init(obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
1579 LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
1581 return lproc_filter_attach_seqstat(obd);
1584 static int filter_detach(struct obd_device *dev)
1586 lprocfs_free_obd_stats(dev);
1587 return lprocfs_obd_detach(dev);
1590 static int filter_setup(struct obd_device *obd, obd_count len, void *buf)
1592 struct lustre_cfg *lcfg = buf;
1597 /* 2.6.9 selinux wants a full option page for do_kern_mount (bug6471) */
1598 page = get_zeroed_page(GFP_KERNEL);
1602 memcpy((void *)page, lustre_cfg_buf(lcfg, 4),
1603 LUSTRE_CFG_BUFLEN(lcfg, 4));
1605 /* all mount options including errors=remount-ro and asyncdel are passed
1606 * using 4th lcfg param. And it is good, finally we have got rid of
1607 * hardcoded fs types in the code. */
1608 rc = filter_common_setup(obd, len, buf, (void *)page);
1613 rc = filter_post_fs_setup(obd);
1617 static int filter_cleanup(struct obd_device *obd, int flags)
1619 struct filter_obd *filter = &obd->u.filter;
1620 ll_sbdev_type save_dev;
1623 if (flags & OBD_OPT_FAILOVER)
1624 CERROR("%s: shutting down for failover; client state will"
1625 " be preserved.\n", obd->obd_name);
1627 if (!list_empty(&obd->obd_exports)) {
1628 CERROR("%s: still has clients!\n", obd->obd_name);
1629 class_disconnect_exports(obd, flags);
1630 if (!list_empty(&obd->obd_exports)) {
1631 CERROR("still has exports after forced cleanup?\n");
1636 target_cleanup_recovery(obd);
1638 ldlm_namespace_free(obd->obd_namespace, flags & OBD_OPT_FORCE);
1640 if (filter->fo_sb == NULL)
1643 save_dev = ll_sbdev(filter->fo_sb);
1644 filter_post_fs_cleanup(obd);
1647 shrink_dcache_parent(filter->fo_sb->s_root);
1650 spin_lock(&filter->fo_denylist_lock);
1651 while (!list_empty(&filter->fo_denylist)) {
1652 deny_sec_t *p_deny_sec = list_entry(filter->fo_denylist.next,
1654 list_del(&p_deny_sec->list);
1655 OBD_FREE(p_deny_sec, sizeof(*p_deny_sec));
1657 spin_unlock(&filter->fo_denylist_lock);
1660 lvfs_umount_fs(filter->fo_lvfs_ctxt);
1661 //destroy_buffers(filter->fo_sb->s_dev);
1662 filter->fo_sb = NULL;
1663 fsfilt_put_ops(obd->obd_fsops);
1666 ll_clear_rdonly(save_dev);
1671 static int filter_process_config(struct obd_device *obd, obd_count len, void *buf)
1673 struct lustre_cfg *lcfg = buf;
1674 struct filter_obd *filter = &obd->u.filter;
1678 switch(lcfg->lcfg_command) {
1679 case LCFG_SET_SECURITY: {
1680 if ((LUSTRE_CFG_BUFLEN(lcfg, 1) == 0) ||
1681 (LUSTRE_CFG_BUFLEN(lcfg, 2) == 0))
1682 GOTO(out, rc = -EINVAL);
1684 if (!strcmp(lustre_cfg_string(lcfg, 1), "deny_sec")){
1685 spin_lock(&filter->fo_denylist_lock);
1686 rc = add_deny_security(lustre_cfg_string(lcfg, 2),
1687 &filter->fo_denylist);
1688 spin_unlock(&filter->fo_denylist_lock);
1690 CERROR("Unrecognized key\n");
1696 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1697 GOTO(out, rc = -EINVAL);
1705 static int filter_connect_post(struct obd_export *exp, unsigned initial,
1706 unsigned long connect_flags)
1708 struct obd_device *obd = exp->exp_obd;
1709 struct filter_export_data *fed;
1710 char str[PTL_NALFMT_SIZE];
1711 struct obd_llogs *llog;
1712 struct llog_ctxt *ctxt;
1716 fed = &exp->exp_filter_data;
1717 if (fed->fed_group < FILTER_MIN_GROUPS)
1720 /* initialize llogs for connections from MDS */
1721 llog = filter_grab_llog_for_group(obd, fed->fed_group, exp);
1722 LASSERT(llog != NULL);
1724 ctxt = llog_get_context(llog, LLOG_UNLINK_REPL_CTXT);
1725 LASSERT(ctxt != NULL);
1727 rc = llog_receptor_accept(ctxt, exp->exp_imp_reverse);
1729 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
1730 exp->exp_connection->c_peer.peer_id.nid, str);
1732 CDEBUG(D_OTHER, "%s: init llog ctxt for export "LPX64"/%s, group %d\n",
1733 obd->obd_name, exp->exp_connection->c_peer.peer_id.nid,
1734 str, fed->fed_group);
1739 /* nearly identical to mds_connect */
1740 static int filter_connect(struct lustre_handle *conn, struct obd_device *obd,
1741 struct obd_uuid *cluuid,
1742 struct obd_connect_data *data,
1743 unsigned long connect_flags)
1745 struct obd_export *exp;
1746 struct filter_export_data *fed;
1747 struct filter_client_data *fcd = NULL;
1748 struct filter_obd *filter = &obd->u.filter;
1749 struct lvfs_run_ctxt saved;
1753 if (conn == NULL || obd == NULL || cluuid == NULL)
1756 rc = class_connect(conn, obd, cluuid);
1759 exp = class_conn2export(conn);
1760 LASSERT(exp != NULL);
1762 fed = &exp->exp_filter_data;
1764 spin_lock_init(&fed->fed_lock);
1766 /* connection from MDS */
1767 group = connect_flags;
1768 if (obd->obd_replayable) {
1769 OBD_ALLOC(fcd, sizeof(*fcd));
1771 CERROR("filter: out of memory for client data\n");
1772 GOTO(cleanup, rc = -ENOMEM);
1775 memcpy(fcd->fcd_uuid, cluuid, sizeof(fcd->fcd_uuid));
1777 fed->fed_fcd->fcd_group = group;
1778 rc = filter_client_add(obd, filter, fed, -1);
1782 CWARN("%s: Received MDS connection ("LPX64"); group %d\n",
1783 obd->obd_name, exp->exp_handle.h_cookie, group);
1788 if (fed->fed_group != 0 && fed->fed_group != group) {
1789 char str[PTL_NALFMT_SIZE];
1790 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
1791 exp->exp_connection->c_peer.peer_id.nid, str);
1792 CERROR("!!! This export (nid "LPX64"/%s) used object group %d "
1793 "earlier; now it's trying to use group %d! This could "
1794 "be a bug in the MDS. Tell CFS.\n",
1795 exp->exp_connection->c_peer.peer_id.nid, str,
1796 fed->fed_group, group);
1797 GOTO(cleanup, rc = -EPROTO);
1799 fed->fed_group = group;
1801 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1802 rc = filter_read_groups(obd, group, 1);
1803 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1805 CERROR("can't read group %u\n", group);
1811 OBD_FREE(fcd, sizeof(*fcd));
1812 class_disconnect(exp, 0);
1814 class_export_put(exp);
1819 static int filter_precleanup(struct obd_device *obd, int flags)
1821 struct filter_group_llog *log;
1822 struct filter_obd *filter;
1826 filter = &obd->u.filter;
1828 spin_lock(&filter->fo_llog_list_lock);
1829 while (!list_empty(&filter->fo_llog_list)) {
1830 log = list_entry(filter->fo_llog_list.next,
1831 struct filter_group_llog, list);
1832 list_del(&log->list);
1833 spin_unlock(&filter->fo_llog_list_lock);
1835 rc = obd_llog_finish(obd, log->llogs, 0);
1837 CERROR("failed to cleanup llogging subsystem for %u\n",
1839 OBD_FREE(log->llogs, sizeof(*(log->llogs)));
1840 OBD_FREE(log, sizeof(*log));
1841 spin_lock(&filter->fo_llog_list_lock);
1843 spin_unlock(&filter->fo_llog_list_lock);
1845 rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1847 CERROR("failed to cleanup llogging subsystem\n");
1852 /* Do extra sanity checks for grant accounting. We do this at connect,
1853 * disconnect, and statfs RPC time, so it shouldn't be too bad. We can
1854 * always get rid of it or turn it off when we know accounting is good. */
1855 static void filter_grant_sanity_check(struct obd_device *obd, const char *func)
1857 struct filter_export_data *fed;
1858 struct obd_export *exp;
1859 obd_size maxsize = obd->obd_osfs.os_blocks * obd->obd_osfs.os_bsize;
1860 obd_size tot_dirty = 0, tot_pending = 0, tot_granted = 0;
1861 obd_size fo_tot_dirty, fo_tot_pending, fo_tot_granted;
1862 int level = D_CACHE;
1864 if (list_empty(&obd->obd_exports))
1867 spin_lock(&obd->obd_osfs_lock);
1868 spin_lock(&obd->obd_dev_lock);
1869 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
1870 fed = &exp->exp_filter_data;
1871 if (fed->fed_grant < 0 || fed->fed_pending < 0 ||
1874 if (maxsize > 0) { /* we may not have done a statfs yet */
1875 LASSERTF(fed->fed_grant + fed->fed_pending <= maxsize,
1876 "cli %s/%p %ld+%ld > "LPU64"\n",
1877 exp->exp_client_uuid.uuid, exp,
1878 fed->fed_grant, fed->fed_pending, maxsize);
1879 LASSERTF(fed->fed_dirty <= maxsize,
1880 "cli %s/%p %ld > "LPU64"\n",
1881 exp->exp_client_uuid.uuid, exp,
1882 fed->fed_dirty, maxsize);
1884 CDEBUG(level, "%s: cli %s/%p dirty %ld pend %ld grant %ld\n",
1885 obd->obd_name, exp->exp_client_uuid.uuid, exp,
1886 fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1887 tot_granted += fed->fed_grant + fed->fed_pending;
1888 tot_pending += fed->fed_pending;
1889 tot_dirty += fed->fed_dirty;
1891 fo_tot_granted = obd->u.filter.fo_tot_granted;
1892 fo_tot_pending = obd->u.filter.fo_tot_pending;
1893 fo_tot_dirty = obd->u.filter.fo_tot_dirty;
1894 spin_unlock(&obd->obd_dev_lock);
1895 spin_unlock(&obd->obd_osfs_lock);
1897 /* Do these assertions outside the spinlocks so we don't kill system */
1898 if (tot_granted != fo_tot_granted)
1899 CERROR("%s: tot_granted "LPU64" != fo_tot_granted "LPU64"\n",
1900 func, tot_granted, fo_tot_granted);
1901 if (tot_pending != fo_tot_pending)
1902 CERROR("%s: tot_pending "LPU64" != fo_tot_pending "LPU64"\n",
1903 func, tot_pending, fo_tot_pending);
1904 if (tot_dirty != fo_tot_dirty)
1905 CERROR("%s: tot_dirty "LPU64" != fo_tot_dirty "LPU64"\n",
1906 func, tot_dirty, fo_tot_dirty);
1907 if (tot_pending > tot_granted)
1908 CERROR("%s: tot_pending "LPU64" > tot_granted "LPU64"\n",
1909 func, tot_pending, tot_granted);
1910 if (tot_granted > maxsize)
1911 CERROR("%s: tot_granted "LPU64" > maxsize "LPU64"\n",
1912 func, tot_granted, maxsize);
1913 if (tot_dirty > maxsize)
1914 CERROR("%s: tot_dirty "LPU64" > maxsize "LPU64"\n",
1915 func, tot_dirty, maxsize);
1918 /* Remove this client from the grant accounting totals. We also remove
1919 * the export from the obd device under the osfs and dev locks to ensure
1920 * that the filter_grant_sanity_check() calculations are always valid.
1921 * The client should do something similar when it invalidates its import. */
1922 static void filter_grant_discard(struct obd_export *exp)
1924 struct obd_device *obd = exp->exp_obd;
1925 struct filter_obd *filter = &obd->u.filter;
1926 struct filter_export_data *fed = &exp->exp_filter_data;
1927 int level = D_CACHE;
1929 spin_lock(&obd->obd_osfs_lock);
1930 spin_lock(&exp->exp_obd->obd_dev_lock);
1931 list_del_init(&exp->exp_obd_chain);
1932 spin_unlock(&exp->exp_obd->obd_dev_lock);
1934 if (fed->fed_dirty < 0 || fed->fed_grant < 0 || fed->fed_pending < 0)
1936 CDEBUG(level, "%s: cli %s/%p dirty %ld pend %ld grant %ld\n",
1937 obd->obd_name, exp->exp_client_uuid.uuid, exp,
1938 fed->fed_dirty, fed->fed_pending, fed->fed_grant);
1940 LASSERTF(filter->fo_tot_granted >= fed->fed_grant,
1941 "%s: tot_granted "LPU64" cli %s/%p fed_grant %ld\n",
1942 obd->obd_name, filter->fo_tot_granted,
1943 exp->exp_client_uuid.uuid, exp, fed->fed_grant);
1944 filter->fo_tot_granted -= fed->fed_grant;
1945 LASSERTF(filter->fo_tot_pending >= fed->fed_pending,
1946 "%s: tot_pending "LPU64" cli %s/%p fed_pending %ld\n",
1947 obd->obd_name, filter->fo_tot_pending,
1948 exp->exp_client_uuid.uuid, exp, fed->fed_pending);
1949 LASSERTF(filter->fo_tot_dirty >= fed->fed_dirty,
1950 "%s: tot_dirty "LPU64" cli %s/%p fed_dirty %ld\n",
1951 obd->obd_name, filter->fo_tot_dirty,
1952 exp->exp_client_uuid.uuid, exp, fed->fed_dirty);
1953 filter->fo_tot_dirty -= fed->fed_dirty;
1957 spin_unlock(&obd->obd_osfs_lock);
1960 static int filter_destroy_export(struct obd_export *exp)
1964 if (exp->exp_filter_data.fed_pending)
1965 CERROR("%s: cli %s/%p has %lu pending on destroyed export\n",
1966 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
1967 exp, exp->exp_filter_data.fed_pending);
1969 target_destroy_export(exp);
1971 if (exp->exp_obd->obd_replayable)
1972 filter_client_free(exp, exp->exp_flags);
1974 filter_grant_discard(exp);
1975 if (!(exp->exp_flags & OBD_OPT_FORCE))
1976 filter_grant_sanity_check(exp->exp_obd, __FUNCTION__);
1981 static void filter_sync_llogs(struct obd_device *obd, struct obd_export *dexp)
1983 struct filter_group_llog *fglog, *nlog;
1984 struct filter_obd *filter;
1985 int worked = 0, group;
1986 struct llog_ctxt *ctxt;
1989 filter = &obd->u.filter;
1991 /* we can't sync log holding spinlock. also, we do not want to get
1992 * into livelock. so we do following: loop over MDS's exports in
1993 * group order and skip already synced llogs -bzzz */
1995 /* look for group with min. number, but > worked */
1998 spin_lock(&filter->fo_llog_list_lock);
1999 list_for_each_entry(nlog, &filter->fo_llog_list, list) {
2001 if (nlog->group <= worked) {
2002 /* this group is already synced */
2006 if (group < nlog->group) {
2007 /* we have group with smaller number to sync */
2011 /* store current minimal group */
2013 group = nlog->group;
2015 spin_unlock(&filter->fo_llog_list_lock);
2020 worked = fglog->group;
2021 if (fglog->exp && (dexp == fglog->exp || dexp == NULL)) {
2022 ctxt = llog_get_context(fglog->llogs,
2023 LLOG_UNLINK_REPL_CTXT);
2024 LASSERT(ctxt != NULL);
2025 llog_sync(ctxt, fglog->exp);
2027 } while (fglog != NULL);
2030 /* also incredibly similar to mds_disconnect */
2031 static int filter_disconnect(struct obd_export *exp, unsigned long flags)
2033 struct obd_device *obd = exp->exp_obd;
2034 unsigned long irqflags;
2039 class_export_get(exp);
2041 spin_lock_irqsave(&exp->exp_lock, irqflags);
2042 exp->exp_flags = flags;
2043 spin_unlock_irqrestore(&exp->exp_lock, irqflags);
2045 if (!(flags & OBD_OPT_FORCE))
2046 filter_grant_sanity_check(obd, __FUNCTION__);
2047 filter_grant_discard(exp);
2049 /* Disconnect early so that clients can't keep using export */
2050 rc = class_disconnect(exp, flags);
2052 ldlm_cancel_locks_for_export(exp);
2054 fsfilt_sync(obd, obd->u.filter.fo_sb);
2056 /* flush any remaining cancel messages out to the target */
2057 filter_sync_llogs(obd, exp);
2058 class_export_put(exp);
2062 struct dentry *__filter_oa2dentry(struct obd_device *obd,
2063 struct obdo *oa, const char *what)
2065 struct dentry *dchild = NULL;
2069 if (oa->o_valid & OBD_MD_FLGROUP)
2072 dchild = filter_id2dentry(obd, NULL, group, oa->o_id);
2074 if (IS_ERR(dchild)) {
2075 CERROR("%s error looking up object: "LPU64"\n",
2080 if (dchild->d_inode == NULL) {
2081 CDEBUG(D_INFO, "%s: %s on non-existent object: "
2082 LPU64"\n", obd->obd_name, what, oa->o_id);
2084 RETURN(ERR_PTR(-ENOENT));
2090 static int filter_getattr(struct obd_export *exp, struct obdo *oa,
2091 struct lov_stripe_md *md)
2093 struct dentry *dentry = NULL;
2094 struct obd_device *obd;
2098 obd = class_exp2obd(exp);
2100 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2101 exp->exp_handle.h_cookie);
2105 dentry = filter_oa2dentry(obd, oa);
2107 RETURN(PTR_ERR(dentry));
2109 /* Limit the valid bits in the return data to what we actually use */
2110 oa->o_valid = OBD_MD_FLID;
2111 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2117 int filter_setattr_internal(struct obd_export *exp, struct dentry *dentry,
2118 struct obdo *oa, struct obd_trans_info *oti)
2120 struct filter_obd *filter;
2126 LASSERT(dentry != NULL);
2127 LASSERT(!IS_ERR(dentry));
2128 LASSERT(dentry->d_inode != NULL);
2130 filter = &exp->exp_obd->u.filter;
2131 iattr_from_obdo(&iattr, oa, oa->o_valid);
2133 if (iattr.ia_valid & ATTR_SIZE)
2134 down(&dentry->d_inode->i_sem);
2135 handle = fsfilt_start(exp->exp_obd, dentry->d_inode,
2136 FSFILT_OP_SETATTR, oti);
2138 GOTO(out_unlock, rc = PTR_ERR(handle));
2140 /* XXX this could be a rwsem instead, if filter_preprw played along */
2141 if (iattr.ia_valid & ATTR_ATTR_FLAG)
2142 rc = fsfilt_iocontrol(exp->exp_obd, dentry->d_inode,
2143 NULL, EXT3_IOC_SETFLAGS,
2144 (long)&iattr.ia_attr_flags);
2146 rc = fsfilt_setattr(exp->exp_obd, dentry, handle,
2149 rc = filter_finish_transno(exp, oti, rc);
2151 err = fsfilt_commit(exp->exp_obd, filter->fo_sb,
2152 dentry->d_inode, handle,
2155 CERROR("error on commit, err = %d\n", err);
2161 if (iattr.ia_valid & ATTR_SIZE)
2162 up(&dentry->d_inode->i_sem);
2166 /* this is called from filter_truncate() until we have filter_punch() */
2167 int filter_setattr(struct obd_export *exp, struct obdo *oa,
2168 struct lov_stripe_md *md, struct obd_trans_info *oti,
2169 struct lustre_capa *capa)
2171 struct ldlm_res_id res_id = { .name = { oa->o_id, 0, oa->o_gr, 0 } };
2172 struct ldlm_valblock_ops *ns_lvbo;
2173 struct lvfs_run_ctxt saved;
2174 struct filter_obd *filter;
2175 struct ldlm_resource *res;
2176 struct dentry *dentry;
2182 LASSERT(oti != NULL);
2184 filter = &exp->exp_obd->u.filter;
2185 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2186 /* pass nid through */
2187 current->user->nid = oti->oti_nid;
2189 /* make sure that object is allocated. */
2190 dentry = filter_crow_object(exp->exp_obd, oa);
2192 GOTO(out_pop, rc = PTR_ERR(dentry));
2194 rc = filter_verify_fid(exp, dentry->d_inode, capa);
2200 /* setting objects attributes (including owner/group) */
2201 rc = filter_setattr_internal(exp, dentry, oa, oti);
2203 GOTO(out_unlock, rc);
2205 res = ldlm_resource_get(exp->exp_obd->obd_namespace, NULL,
2206 res_id, LDLM_EXTENT, 0);
2209 ns_lvbo = res->lr_namespace->ns_lvbo;
2210 if (ns_lvbo && ns_lvbo->lvbo_update)
2211 rc = ns_lvbo->lvbo_update(res, NULL, 0, 0);
2212 ldlm_resource_putref(res);
2215 oa->o_valid = OBD_MD_FLID;
2216 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2223 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2227 /* XXX identical to osc_unpackmd */
2228 static int filter_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2229 struct lov_mds_md *lmm, int lmm_bytes)
2235 if (lmm_bytes < sizeof (*lmm)) {
2236 CERROR("lov_mds_md too small: %d, need %d\n",
2237 lmm_bytes, (int)sizeof(*lmm));
2240 /* XXX LOV_MAGIC etc check? */
2242 if (lmm->lmm_object_id == cpu_to_le64(0)) {
2243 CERROR("lov_mds_md: zero lmm_object_id\n");
2248 lsm_size = lov_stripe_md_size(1);
2252 if (*lsmp != NULL && lmm == NULL) {
2253 OBD_FREE(*lsmp, lsm_size);
2258 if (*lsmp == NULL) {
2259 OBD_ALLOC(*lsmp, lsm_size);
2263 loi_init((*lsmp)->lsm_oinfo);
2267 /* XXX zero *lsmp? */
2268 (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
2269 LASSERT((*lsmp)->lsm_object_id);
2272 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
2277 static int filter_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2278 unsigned long max_age)
2280 struct filter_obd *filter = &obd->u.filter;
2281 int blockbits = filter->fo_sb->s_blocksize_bits;
2285 /* at least try to account for cached pages. its still racey and
2286 * might be under-reporting if clients haven't announced their
2287 * caches with brw recently */
2288 spin_lock(&obd->obd_osfs_lock);
2289 rc = fsfilt_statfs(obd, filter->fo_sb, max_age);
2290 memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
2291 spin_unlock(&obd->obd_osfs_lock);
2293 CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64
2294 " pending "LPU64" free "LPU64" avail "LPU64"\n",
2295 filter->fo_tot_dirty, filter->fo_tot_granted,
2296 filter->fo_tot_pending,
2297 osfs->os_bfree << blockbits, osfs->os_bavail << blockbits);
2299 filter_grant_sanity_check(obd, __FUNCTION__);
2301 osfs->os_bavail -= min(osfs->os_bavail,
2302 (filter->fo_tot_dirty + filter->fo_tot_pending +
2303 osfs->os_bsize -1) >> blockbits);
2308 int filter_create_object(struct obd_device *obd, struct obdo *oa)
2310 struct dentry *dparent = NULL;
2311 struct dentry *dchild = NULL;
2312 struct lvfs_ucred uc = {0,};
2313 struct lvfs_run_ctxt saved;
2314 struct filter_obd *filter;
2315 int cleanup_phase = 0;
2316 int err = 0, rc = 0;
2317 void *handle = NULL;
2321 filter = &obd->u.filter;
2323 down(&filter->fo_create_locks[oa->o_gr]);
2325 if (test_bit(oa->o_gr, &filter->fo_destroys_in_progress)) {
2326 CWARN("%s: create aborted by destroy\n",
2328 GOTO(out, rc = -EALREADY);
2331 CDEBUG(D_INFO, "create objid "LPU64"\n", oa->o_id);
2333 dparent = filter_parent_lock(obd, oa->o_gr, oa->o_id, &lock);
2334 if (IS_ERR(dparent))
2335 GOTO(cleanup, rc = PTR_ERR(dparent));
2338 dchild = filter_id2dentry(obd, dparent, oa->o_gr, oa->o_id);
2340 GOTO(cleanup, rc = PTR_ERR(dchild));
2343 if (dchild->d_inode != NULL)
2344 GOTO(cleanup, rc = 0);
2346 handle = fsfilt_start_log(obd, dparent->d_inode,
2347 FSFILT_OP_CREATE, NULL, 1);
2349 GOTO(cleanup, rc = PTR_ERR(handle));
2352 uc.luc_uid = oa->o_uid;
2353 uc.luc_gid = oa->o_gid;
2354 uc.luc_fsuid = oa->o_uid;
2355 uc.luc_fsgid = oa->o_gid;
2356 uc.luc_cap = current->cap_effective;
2357 cap_raise(uc.luc_cap, CAP_SYS_RESOURCE);
2359 push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
2360 rc = ll_vfs_create(dparent->d_inode, dchild, S_IFREG, NULL);
2361 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
2364 CERROR("create failed rc = %d\n", rc);
2368 /* nobody else is touching this newly created object */
2369 LASSERT(dchild->d_inode);
2370 if (oa->o_valid & OBD_MD_FLIFID) {
2371 rc = fsfilt_set_md(obd, dchild->d_inode, handle, obdo_id(oa),
2372 sizeof(struct lustre_id), EA_SID);
2374 CERROR("store fid in object failed! rc:%d\n", rc);
2378 CWARN("create object without lustre id!\n");
2379 portals_debug_dumpstack(NULL);
2382 fsfilt_set_fs_flags(obd, dparent->d_inode, SM_DO_REC);
2384 /* save last created object id */
2385 filter_save_last_id(filter, oa->o_gr, oa->o_id);
2387 rc = filter_update_last_objid(obd, oa->o_gr, 0);
2389 CERROR("unable to write lastobjid, but "
2390 "orphans were deleted, err = %d\n",
2395 switch(cleanup_phase) {
2397 err = fsfilt_commit(obd, filter->fo_sb,
2398 dparent->d_inode, handle, 0);
2400 CERROR("error on commit, err = %d\n", err);
2407 filter_parent_unlock(dparent, lock);
2416 up(&filter->fo_create_locks[oa->o_gr]);
2421 filter_crow_object(struct obd_device *obd, struct obdo *oa)
2423 struct dentry *dentry;
2429 /* check if object is already allocated */
2430 dentry = filter_id2dentry(obd, NULL, oa->o_gr, oa->o_id);
2434 if (dentry->d_inode)
2438 CDEBUG(D_INODE, "OSS object "LPU64"/"LPU64
2439 " does not exists - allocate it now\n",
2440 oa->o_id, oa->o_gr);
2442 uid = oa->o_valid & OBD_MD_FLUID ? oa->o_uid : 0;
2443 gid = oa->o_valid & OBD_MD_FLGID ? oa->o_gid : 0;
2445 rc = filter_create_object(obd, oa);
2447 CERROR("cannot create OSS object "LPU64"/"LPU64
2448 ", err = %d\n", oa->o_id, oa->o_gr, rc);
2449 RETURN(ERR_PTR(rc));
2452 /* lookup for just created object and return it to caller */
2453 dentry = filter_id2dentry(obd, NULL, oa->o_gr, oa->o_id);
2457 if (dentry->d_inode == NULL) {
2459 dentry = ERR_PTR(-ENOENT);
2460 CERROR("cannot find just created OSS object "
2461 LPU64"/"LPU64" err = %d\n", oa->o_id,
2462 oa->o_gr, (int)PTR_ERR(dentry));
2469 filter_clear_orphans(struct obd_export *exp, struct obdo *oa)
2471 struct obd_device *obd = NULL;
2472 struct filter_obd *filter;
2473 struct obdo *doa = NULL;
2474 int rc = 0, orphans;
2479 LASSERT(oa->o_gr != 0);
2480 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2483 filter = &obd->u.filter;
2485 last = filter_last_id(filter, oa->o_gr);
2486 orphans = last - oa->o_id;
2495 doa->o_gr = oa->o_gr;
2496 doa->o_mode = S_IFREG;
2497 doa->o_valid = oa->o_valid & (OBD_MD_FLGROUP | OBD_MD_FLID);
2499 set_bit(doa->o_gr, &filter->fo_destroys_in_progress);
2500 down(&filter->fo_create_locks[doa->o_gr]);
2501 if (!test_bit(doa->o_gr, &filter->fo_destroys_in_progress)) {
2502 CERROR("%s:["LPU64"] destroy_in_progress already cleared\n",
2503 exp->exp_obd->obd_name, doa->o_gr);
2504 up(&filter->fo_create_locks[doa->o_gr]);
2505 GOTO(out_free_doa, 0);
2508 CDEBUG(D_ERROR, "%s:["LPU64"] deleting orphan objects from "LPU64" to "
2509 LPU64"\n", exp->exp_obd->obd_name, doa->o_gr,
2510 oa->o_id + 1, last);
2512 for (id = oa->o_id + 1; id <= last; id++) {
2514 filter_destroy(exp, doa, NULL, NULL);
2517 CDEBUG(D_ERROR, "%s:["LPU64"] after destroy: set last_objids = "
2518 LPU64"\n", exp->exp_obd->obd_name, doa->o_gr, last + 1);
2520 /* return next free id to be used as a new start of sequence -bzzz */
2521 oa->o_id = last + 1;
2523 filter_set_last_id(filter, oa->o_gr, oa->o_id);
2524 clear_bit(doa->o_gr, &filter->fo_destroys_in_progress);
2525 up(&filter->fo_create_locks[oa->o_gr]);
2534 * by now this function is only needed as entry point for deleting orphanes on
2535 * OSS as objects are created on first write attempt. --umka
2538 filter_create(struct obd_export *exp, struct obdo *oa, void *acl,
2539 int acl_size, struct lov_stripe_md **ea,
2540 struct obd_trans_info *oti)
2542 struct filter_export_data *fed;
2543 struct obd_device *obd = NULL;
2544 int group = oa->o_gr, rc = 0;
2545 struct lvfs_run_ctxt saved;
2546 struct filter_obd *filter;
2547 char str[PTL_NALFMT_SIZE];
2550 LASSERT(acl == NULL && acl_size == 0);
2552 if (!(oa->o_valid & OBD_MD_FLGROUP) || group == 0) {
2553 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2554 exp->exp_connection->c_peer.peer_id.nid, str);
2555 CERROR("!!! nid "LPX64"/%s sent invalid object group %d\n",
2556 exp->exp_connection->c_peer.peer_id.nid, str, group);
2561 fed = &exp->exp_filter_data;
2562 filter = &obd->u.filter;
2564 if (fed->fed_group != group) {
2565 portals_nid2str(exp->exp_connection->c_peer.peer_ni->pni_number,
2566 exp->exp_connection->c_peer.peer_id.nid, str);
2567 CERROR("!!! this export (nid "LPX64"/%s) used object group %d "
2568 "earlier; now it's trying to use group %d! This could "
2569 "be a bug in the MDS. Tell CFS.\n",
2570 exp->exp_connection->c_peer.peer_id.nid, str,
2571 fed->fed_group, group);
2575 CDEBUG(D_INFO, "filter_create(od->o_gr=%d,od->o_id="LPU64")\n",
2579 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2581 LASSERT((oa->o_valid & OBD_MD_FLFLAGS) &&
2582 (oa->o_flags == OBD_FL_DELORPHAN));
2584 rc = filter_clear_orphans(exp, oa);
2586 CERROR("cannot clear orphanes starting from "
2587 LPU64", err = %d\n", oa->o_id, rc);
2589 rc = filter_update_last_objid(obd, group, 0);
2591 CERROR("unable to write lastobjid, but "
2592 "orphans were deleted, err = %d\n",
2596 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2601 static int filter_destroy(struct obd_export *exp, struct obdo *oa,
2602 struct lov_stripe_md *ea, struct obd_trans_info *oti)
2604 int rc, rc2, cleanup_phase = 0, have_prepared = 0;
2605 struct dentry *dchild = NULL, *dparent = NULL;
2606 struct llog_cookie *fcc = NULL;
2607 struct lvfs_run_ctxt saved;
2608 struct filter_obd *filter;
2609 struct obd_device *obd;
2610 void *handle = NULL;
2615 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2618 filter = &obd->u.filter;
2620 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2623 dparent = filter_parent_lock(obd, oa->o_gr, oa->o_id, &lock);
2624 if (IS_ERR(dparent))
2625 GOTO(cleanup, rc = PTR_ERR(dparent));
2628 dchild = filter_id2dentry(obd, dparent, oa->o_gr, oa->o_id);
2630 GOTO(cleanup, rc = PTR_ERR(dchild));
2633 if (dchild->d_inode == NULL) {
2634 CDEBUG(D_INODE, "destroying non-existent object "LPU64"\n",
2637 GOTO(cleanup, rc = -ENOENT);
2640 if (!have_prepared) {
2641 /* If we're really going to destroy the object, get ready
2642 * by getting the clients to discard their cached data.
2644 * We have to drop the parent lock, because
2645 * filter_prepare_destroy will acquire a PW on the object, and
2646 * we don't want to deadlock with an incoming write to the
2647 * object, which has the extent PW and then wants to get the
2648 * parent dentry to do the lookup.
2650 * We dput the child because it's not worth the extra
2651 * complication of condition the above code to skip it on the
2652 * second time through. */
2654 filter_parent_unlock(dparent, lock);
2656 filter_prepare_destroy(obd, oa->o_id, oa->o_gr);
2661 /* Our MDC connection is established by the MDS to us */
2662 if (oa->o_valid & OBD_MD_FLCOOKIE) {
2663 OBD_ALLOC(fcc, sizeof(*fcc));
2665 memcpy(fcc, obdo_logcookie(oa), sizeof(*fcc));
2668 /* we're gonna truncate it first in order to avoid possible
2671 * open trasaction open transaction
2672 * down(i_zombie) down(i_zombie)
2673 * restart transaction
2674 * (see BUG 4180) -bzzz
2676 down(&dchild->d_inode->i_sem);
2677 handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_SETATTR,NULL,1);
2678 if (IS_ERR(handle)) {
2679 up(&dchild->d_inode->i_sem);
2680 GOTO(cleanup, rc = PTR_ERR(handle));
2683 iattr.ia_valid = ATTR_SIZE;
2685 rc = fsfilt_setattr(obd, dchild, handle, &iattr, 1);
2687 rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode, handle, 0);
2688 up(&dchild->d_inode->i_sem);
2692 GOTO(cleanup, rc = rc2);
2694 handle = fsfilt_start_log(obd, dparent->d_inode,FSFILT_OP_UNLINK,oti,1);
2696 GOTO(cleanup, rc = PTR_ERR(handle));
2700 rc = filter_destroy_internal(obd, oa->o_id, dparent, dchild);
2703 switch(cleanup_phase) {
2706 fsfilt_add_journal_cb(obd, filter->fo_sb, 0,
2707 oti ? oti->oti_handle : handle,
2708 filter_cancel_cookies_cb, fcc);
2710 rc = filter_finish_transno(exp, oti, rc);
2711 rc2 = fsfilt_commit(obd, filter->fo_sb, dparent->d_inode,
2712 handle, exp->exp_sync);
2714 CERROR("error on commit, err = %d\n", rc2);
2721 filter_parent_unlock(dparent, lock);
2723 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2726 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
2733 /* NB start and end are used for punch, but not truncate */
2734 static int filter_truncate(struct obd_export *exp, struct obdo *oa,
2735 struct lov_stripe_md *lsm, obd_off start,
2736 obd_off end, struct obd_trans_info *oti,
2737 struct lustre_capa *capa)
2743 if (end != OBD_OBJECT_EOF)
2744 CERROR("PUNCH not supported, only truncate: end = "LPX64"\n",
2747 rc = filter_verify_capa(OBD_BRW_WRITE, exp, capa);
2751 CDEBUG(D_INODE, "calling truncate for object "LPU64", valid = "LPU64", "
2752 "o_size = "LPD64"\n", oa->o_id, oa->o_valid, start);
2755 error = filter_setattr(exp, oa, NULL, oti, capa);
2759 static int filter_sync(struct obd_export *exp, struct obdo *oa,
2760 struct lov_stripe_md *lsm, obd_off start, obd_off end)
2762 struct obd_device *obd = exp->exp_obd;
2763 struct lvfs_run_ctxt saved;
2764 struct filter_obd *filter;
2765 struct dentry *dentry;
2769 filter = &obd->u.filter;
2771 /* an objid of zero is taken to mean "sync whole filesystem" */
2772 if (!oa || !(oa->o_valid & OBD_MD_FLID)) {
2773 rc = fsfilt_sync(obd, filter->fo_sb);
2774 /* flush any remaining cancel messages out to the target */
2775 filter_sync_llogs(obd, NULL);
2779 dentry = filter_oa2dentry(obd, oa);
2781 RETURN(PTR_ERR(dentry));
2783 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2785 down(&dentry->d_inode->i_sem);
2786 rc = filemap_fdatawrite(dentry->d_inode->i_mapping);
2788 /* just any file to grab fsync method - "file" arg unused */
2789 struct file *file = filter->fo_rcvd_filp;
2791 if (file->f_op && file->f_op->fsync)
2792 rc = file->f_op->fsync(NULL, dentry, 1);
2794 rc2 = filemap_fdatawait(dentry->d_inode->i_mapping);
2798 up(&dentry->d_inode->i_sem);
2800 oa->o_valid = OBD_MD_FLID;
2801 obdo_from_inode(oa, dentry->d_inode, FILTER_VALID_FLAGS);
2803 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
2809 static int filter_set_info(struct obd_export *exp, __u32 keylen,
2810 void *key, __u32 vallen, void *val)
2812 struct obd_device *obd;
2816 obd = class_exp2obd(exp);
2818 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2819 exp->exp_handle.h_cookie);
2823 if (keylen == 8 && memcmp(key, "auditlog", 8) == 0) {
2824 rc = fsfilt_set_info(obd, obd->u.filter.fo_sb, NULL,
2825 8, "auditlog", vallen, val);
2827 } else if (keylen == 5 && strcmp(key, "audit") == 0) {
2828 /* set audit for whole FS on OSS */
2829 struct audit_attr_msg * msg = val;
2831 rc = fsfilt_set_info(obd, obd->u.filter.fo_sb, NULL,
2832 5, "audit", sizeof(msg->attr), &msg->attr);
2834 } else if (keylen == 9 && strcmp(key, "audit_obj") == 0) {
2835 struct obdo * oa = val;
2836 struct dentry * dentry;
2837 __u64 mask = oa->o_fid;
2839 dentry = filter_crow_object(obd, oa);
2841 RETURN(PTR_ERR(dentry));
2843 rc = fsfilt_set_info(obd, obd->u.filter.fo_sb, dentry->d_inode,
2844 5, "audit", sizeof(mask), &mask);
2849 } else if (keylen == 8 && memcmp(key, "capa_key", 8) == 0) {
2850 struct lustre_capa_key *lkey = val;
2851 rc = filter_update_capa_key(obd, lkey);
2856 CDEBUG(D_IOCTL, "invalid key '%s'\n", (char *)key);
2861 static int filter_get_info(struct obd_export *exp, __u32 keylen,
2862 void *key, __u32 *vallen, void *val)
2864 struct filter_export_data *fed = &exp->exp_filter_data;
2865 struct obd_device *obd;
2868 obd = class_exp2obd(exp);
2870 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2871 exp->exp_handle.h_cookie);
2875 if (keylen == strlen("blocksize") &&
2876 memcmp(key, "blocksize", keylen) == 0) {
2877 __u32 *blocksize = val;
2878 *vallen = sizeof(*blocksize);
2879 *blocksize = obd->u.filter.fo_sb->s_blocksize;
2883 if (keylen == strlen("blocksize_bits") &&
2884 memcmp(key, "blocksize_bits", keylen) == 0) {
2885 __u32 *blocksize_bits = val;
2886 *vallen = sizeof(*blocksize_bits);
2887 *blocksize_bits = obd->u.filter.fo_sb->s_blocksize_bits;
2891 if (keylen >= strlen("last_id") && memcmp(key, "last_id", 7) == 0) {
2892 obd_id *last_id = val;
2893 *last_id = filter_last_id(&obd->u.filter, fed->fed_group);
2896 if (keylen >= strlen("reint_log") && memcmp(key, "reint_log", 9) == 0) {
2897 /*Get log_context handle*/
2898 unsigned long *llh_handle = val;
2899 *vallen = sizeof(unsigned long);
2900 *llh_handle = (unsigned long)llog_get_context(&obd->obd_llogs,
2901 LLOG_REINT_ORIG_CTXT);
2904 if (keylen >= strlen("cache_sb") && memcmp(key, "cache_sb", 8) == 0) {
2905 unsigned long *sb = val;
2906 *vallen = sizeof(unsigned long);
2907 *sb = (unsigned long)obd->u.filter.fo_sb;
2911 CDEBUG(D_IOCTL, "invalid key\n");
2915 struct obd_llogs *filter_grab_llog_for_group(struct obd_device *obd, int group,
2916 struct obd_export *export)
2918 struct filter_group_llog *fglog, *nlog;
2919 char name[32] = "CATLIST";
2920 struct filter_obd *filter;
2921 struct llog_ctxt *ctxt;
2922 struct list_head *cur;
2925 filter = &obd->u.filter;
2927 spin_lock(&filter->fo_llog_list_lock);
2928 list_for_each(cur, &filter->fo_llog_list) {
2929 fglog = list_entry(cur, struct filter_group_llog, list);
2930 if (fglog->group == group) {
2931 if (!(fglog->exp == NULL || fglog->exp == export || export == NULL))
2932 CWARN("%s: export for group %d changes: 0x%p -> 0x%p\n",
2933 obd->obd_name, group, fglog->exp, export);
2934 spin_unlock(&filter->fo_llog_list_lock);
2938 spin_unlock(&filter->fo_llog_list_lock);
2943 OBD_ALLOC(fglog, sizeof(*fglog));
2946 fglog->group = group;
2948 OBD_ALLOC(fglog->llogs, sizeof(struct obd_llogs));
2949 if (fglog->llogs == NULL) {
2950 OBD_FREE(fglog, sizeof(*fglog));
2954 spin_lock(&filter->fo_llog_list_lock);
2955 list_for_each(cur, &filter->fo_llog_list) {
2956 nlog = list_entry(cur, struct filter_group_llog, list);
2957 LASSERT(nlog->group != group);
2959 list_add(&fglog->list, &filter->fo_llog_list);
2960 spin_unlock(&filter->fo_llog_list_lock);
2962 rc = obd_llog_cat_initialize(obd, fglog->llogs, 1, name);
2964 OBD_FREE(fglog->llogs, sizeof(*(fglog->llogs)));
2965 OBD_FREE(fglog, sizeof(*fglog));
2971 fglog->exp = export;
2972 ctxt = llog_get_context(fglog->llogs, LLOG_UNLINK_REPL_CTXT);
2973 LASSERT(ctxt != NULL);
2975 llog_receptor_accept(ctxt, export->exp_imp_reverse);
2977 CDEBUG(D_OTHER, "%s: new llog 0x%p for group %u\n",
2978 obd->obd_name, fglog->llogs, group);
2980 RETURN(fglog->llogs);
2983 int filter_iocontrol(unsigned int cmd, struct obd_export *exp,
2984 int len, void *karg, void *uarg)
2986 struct obd_device *obd = exp->exp_obd;
2987 struct obd_ioctl_data *data = karg;
2991 case OBD_IOC_ABORT_RECOVERY:
2992 target_stop_recovery_thread(obd);
2995 case OBD_IOC_SET_READONLY: {
2997 struct super_block *sb = obd->u.filter.fo_sb;
2998 struct inode *inode = sb->s_root->d_inode;
2999 BDEVNAME_DECLARE_STORAGE(tmp);
3000 CERROR("setting device %s read-only\n",
3001 ll_bdevname(sb, tmp));
3003 handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL);
3005 (void)fsfilt_commit(obd, sb, inode, handle, 1);
3007 ll_set_rdonly(ll_sbdev(obd->u.filter.fo_sb));
3011 case OBD_IOC_CATLOGLIST: {
3012 rc = llog_catalog_list(obd, 1, data);
3015 case OBD_IOC_LLOG_CANCEL:
3016 case OBD_IOC_LLOG_REMOVE:
3017 case OBD_IOC_LLOG_INFO:
3018 case OBD_IOC_LLOG_PRINT: {
3019 /* FIXME to be finished */
3020 RETURN(-EOPNOTSUPP);
3022 struct llog_ctxt *ctxt = NULL;
3024 push_ctxt(&saved, &ctxt->loc_ctxt, NULL);
3025 rc = llog_ioctl(ctxt, cmd, data);
3026 pop_ctxt(&saved, &ctxt->loc_ctxt, NULL);
3039 static struct llog_operations filter_unlink_repl_logops;
3040 static struct llog_operations filter_size_orig_logops = {
3041 lop_setup: llog_obd_origin_setup,
3042 lop_cleanup: llog_catalog_cleanup,
3043 lop_add: llog_catalog_add,
3046 static int filter_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
3047 struct obd_device *tgt, int count,
3048 struct llog_catid *catid)
3050 struct llog_ctxt *ctxt;
3054 filter_unlink_repl_logops = llog_client_ops;
3055 filter_unlink_repl_logops.lop_cancel = llog_obd_repl_cancel;
3056 filter_unlink_repl_logops.lop_connect = llog_repl_connect;
3057 filter_unlink_repl_logops.lop_sync = llog_obd_repl_sync;
3059 rc = obd_llog_setup(obd, llogs, LLOG_UNLINK_REPL_CTXT, tgt, 0, NULL,
3060 &filter_unlink_repl_logops);
3063 /* FIXME - assign unlink_cb for filter's recovery */
3064 ctxt = llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT);
3065 ctxt->llog_proc_cb = filter_recov_log_unlink_cb;
3067 filter_size_orig_logops = llog_lvfs_ops;
3069 filter_size_orig_logops.lop_setup = llog_obd_origin_setup;
3070 filter_size_orig_logops.lop_cleanup = llog_catalog_cleanup;
3071 filter_size_orig_logops.lop_add = llog_catalog_add;
3073 rc = obd_llog_setup(obd, llogs, LLOG_SIZE_ORIG_CTXT, tgt, 1,
3074 &catid->lci_logid, &filter_size_orig_logops);
3078 static int filter_llog_finish(struct obd_device *obd,
3079 struct obd_llogs *llogs, int count)
3084 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_UNLINK_REPL_CTXT));
3088 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_SIZE_ORIG_CTXT));
3092 static int filter_llog_connect(struct obd_export *exp,
3093 struct llogd_conn_body *body)
3095 struct obd_device *obd = exp->exp_obd;
3096 struct llog_ctxt *ctxt;
3097 struct obd_llogs *llog;
3101 CDEBUG(D_OTHER, "handle connect for %s: %u/%u/%u\n", obd->obd_name,
3102 (unsigned) body->lgdc_logid.lgl_ogr,
3103 (unsigned) body->lgdc_logid.lgl_oid,
3104 (unsigned) body->lgdc_logid.lgl_ogen);
3105 llog = filter_grab_llog_for_group(obd, body->lgdc_logid.lgl_ogr, exp);
3106 LASSERT(llog != NULL);
3107 ctxt = llog_get_context(llog, body->lgdc_ctxt_idx);
3108 rc = llog_connect(ctxt, 1, &body->lgdc_logid,
3109 &body->lgdc_gen, NULL);
3111 CERROR("failed to connect\n");
3116 static struct dentry *filter_lvfs_id2dentry(__u64 id, __u32 gen,
3117 __u64 gr, void *data)
3119 return filter_id2dentry(data, NULL, gr, id);
3122 static struct lvfs_callback_ops filter_lvfs_ops = {
3123 l_id2dentry: filter_lvfs_id2dentry,
3126 static struct obd_ops filter_obd_ops = {
3127 .o_owner = THIS_MODULE,
3128 .o_attach = filter_attach,
3129 .o_detach = filter_detach,
3130 .o_set_info = filter_set_info,
3131 .o_get_info = filter_get_info,
3132 .o_setup = filter_setup,
3133 .o_precleanup = filter_precleanup,
3134 .o_cleanup = filter_cleanup,
3135 .o_process_config = filter_process_config,
3136 .o_connect = filter_connect,
3137 .o_connect_post = filter_connect_post,
3138 .o_disconnect = filter_disconnect,
3139 .o_statfs = filter_statfs,
3140 .o_getattr = filter_getattr,
3141 .o_unpackmd = filter_unpackmd,
3142 .o_create = filter_create,
3143 .o_setattr = filter_setattr,
3144 .o_destroy = filter_destroy,
3145 .o_brw = filter_brw,
3146 .o_punch = filter_truncate,
3147 .o_sync = filter_sync,
3148 .o_preprw = filter_preprw,
3149 .o_commitrw = filter_commitrw,
3150 .o_do_cow = filter_do_cow,
3151 .o_write_extents = filter_write_extents,
3152 .o_destroy_export = filter_destroy_export,
3153 .o_llog_init = filter_llog_init,
3154 .o_llog_finish = filter_llog_finish,
3155 .o_llog_connect = filter_llog_connect,
3156 .o_iocontrol = filter_iocontrol,
3159 static struct obd_ops filter_sanobd_ops = {
3160 .o_owner = THIS_MODULE,
3161 .o_attach = filter_attach,
3162 .o_detach = filter_detach,
3163 .o_get_info = filter_get_info,
3164 .o_setup = filter_san_setup,
3165 .o_precleanup = filter_precleanup,
3166 .o_cleanup = filter_cleanup,
3167 .o_connect = filter_connect,
3168 .o_connect_post = filter_connect_post,
3169 .o_disconnect = filter_disconnect,
3170 .o_statfs = filter_statfs,
3171 .o_getattr = filter_getattr,
3172 .o_unpackmd = filter_unpackmd,
3173 .o_create = filter_create,
3174 .o_setattr = filter_setattr,
3175 .o_destroy = filter_destroy,
3176 .o_brw = filter_brw,
3177 .o_punch = filter_truncate,
3178 .o_sync = filter_sync,
3179 .o_preprw = filter_preprw,
3180 .o_commitrw = filter_commitrw,
3181 .o_do_cow = filter_do_cow,
3182 .o_write_extents = filter_write_extents,
3183 .o_san_preprw = filter_san_preprw,
3184 .o_destroy_export = filter_destroy_export,
3185 .o_llog_init = filter_llog_init,
3186 .o_llog_finish = filter_llog_finish,
3187 .o_llog_connect = filter_llog_connect,
3188 .o_iocontrol = filter_iocontrol,
3191 static int __init obdfilter_init(void)
3193 struct lprocfs_static_vars lvars;
3196 printk(KERN_INFO "Lustre: Filtering OBD driver; info@clusterfs.com\n");
3198 lprocfs_init_vars(filter, &lvars);
3200 size = OBDFILTER_CREATED_SCRATCHPAD_ENTRIES *
3201 sizeof(*obdfilter_created_scratchpad);
3203 OBD_ALLOC(obdfilter_created_scratchpad, size);
3204 if (obdfilter_created_scratchpad == NULL) {
3205 CERROR ("Can't allocate scratchpad\n");
3209 rc = class_register_type(&filter_obd_ops, NULL, lvars.module_vars,
3210 OBD_FILTER_DEVICENAME);
3212 OBD_FREE(obdfilter_created_scratchpad, size);
3216 rc = class_register_type(&filter_sanobd_ops, NULL, lvars.module_vars,
3217 OBD_FILTER_SAN_DEVICENAME);
3219 class_unregister_type(OBD_FILTER_DEVICENAME);
3220 OBD_FREE(obdfilter_created_scratchpad, size);
3225 static void __exit obdfilter_exit(void)
3227 class_unregister_type(OBD_FILTER_SAN_DEVICENAME);
3228 class_unregister_type(OBD_FILTER_DEVICENAME);
3229 OBD_FREE(obdfilter_created_scratchpad,
3230 OBDFILTER_CREATED_SCRATCHPAD_ENTRIES *
3231 sizeof(*obdfilter_created_scratchpad));
3234 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
3235 MODULE_DESCRIPTION("Lustre Filtering OBD driver");
3236 MODULE_LICENSE("GPL");
3238 module_init(obdfilter_init);
3239 module_exit(obdfilter_exit);