1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
5 * Lustre Metadata Server (MDS) filesystem interface code
7 * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
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 # define EXPORT_SYMTAB
29 #define DEBUG_SUBSYSTEM S_MDS
31 #include <linux/module.h>
32 #include <linux/kmod.h>
33 #include <linux/version.h>
34 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
35 #include <linux/mount.h>
37 #include <linux/lustre_mds.h>
38 #include <linux/obd_class.h>
39 #include <linux/obd_support.h>
40 #include <linux/lustre_lib.h>
41 #include <linux/lustre_fsfilt.h>
42 #include <portals/list.h>
44 #include <linux/lustre_smfs.h>
45 #include "mds_internal.h"
47 /* This limit is arbitrary, but for now we fit it in 1 page (32k clients) */
48 #define MDS_MAX_CLIENTS (PAGE_SIZE * 8)
49 #define MDS_MAX_CLIENT_WORDS (MDS_MAX_CLIENTS / sizeof(unsigned long))
51 #define LAST_RCVD "last_rcvd"
52 #define LOV_OBJID "lov_objid"
53 #define LAST_FID "last_fid"
55 /* Add client data to the MDS. We use a bitmap to locate a free space
56 * in the last_rcvd file if cl_off is -1 (i.e. a new client).
57 * Otherwise, we have just read the data from the last_rcvd file and
60 int mds_client_add(struct obd_device *obd, struct mds_obd *mds,
61 struct mds_export_data *med, int cl_idx)
63 unsigned long *bitmap = mds->mds_client_bitmap;
64 int new_client = (cl_idx == -1);
67 LASSERT(bitmap != NULL);
69 /* XXX if mcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
70 if (!strcmp(med->med_mcd->mcd_uuid, obd->obd_uuid.uuid))
73 /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
74 * there's no need for extra complication here
77 cl_idx = find_first_zero_bit(bitmap, MDS_MAX_CLIENTS);
79 if (cl_idx >= MDS_MAX_CLIENTS) {
80 CERROR("no room for clients - fix MDS_MAX_CLIENTS\n");
83 if (test_and_set_bit(cl_idx, bitmap)) {
84 cl_idx = find_next_zero_bit(bitmap, MDS_MAX_CLIENTS,
89 if (test_and_set_bit(cl_idx, bitmap)) {
90 CERROR("MDS client %d: bit already set in bitmap!!\n",
96 CDEBUG(D_INFO, "client at idx %d with UUID '%s' added\n",
97 cl_idx, med->med_mcd->mcd_uuid);
99 med->med_idx = cl_idx;
100 med->med_off = le32_to_cpu(mds->mds_server_data->msd_client_start) +
101 (cl_idx * le16_to_cpu(mds->mds_server_data->msd_client_size));
104 struct lvfs_run_ctxt saved;
105 loff_t off = med->med_off;
106 struct file *file = mds->mds_rcvd_filp;
109 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
110 rc = fsfilt_write_record(obd, file, med->med_mcd,
111 sizeof(*med->med_mcd), &off, 1);
112 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
116 CDEBUG(D_INFO, "wrote client mcd at idx %u off %llu (len %u)\n",
117 med->med_idx, med->med_off,
118 (unsigned int)sizeof(*med->med_mcd));
123 int mds_client_free(struct obd_export *exp, int clear_client)
125 struct mds_export_data *med = &exp->exp_mds_data;
126 struct mds_obd *mds = &exp->exp_obd->u.mds;
127 struct obd_device *obd = exp->exp_obd;
128 struct mds_client_data zero_mcd;
129 struct lvfs_run_ctxt saved;
131 unsigned long *bitmap = mds->mds_client_bitmap;
136 /* XXX if mcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
137 if (!strcmp(med->med_mcd->mcd_uuid, obd->obd_uuid.uuid))
138 GOTO(free_and_out, 0);
140 CDEBUG(D_INFO, "freeing client at idx %u (%lld)with UUID '%s'\n",
141 med->med_idx, med->med_off, med->med_mcd->mcd_uuid);
145 /* Clear the bit _after_ zeroing out the client so we don't
146 race with mds_client_add and zero out new clients.*/
147 if (!test_bit(med->med_idx, bitmap)) {
148 CERROR("MDS client %u: bit already clear in bitmap!!\n",
154 memset(&zero_mcd, 0, sizeof zero_mcd);
155 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
156 rc = fsfilt_write_record(obd, mds->mds_rcvd_filp, &zero_mcd,
157 sizeof(zero_mcd), &med->med_off, 1);
158 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
160 CDEBUG(rc == 0 ? D_INFO : D_ERROR,
161 "zeroing out client %s idx %u in %s rc %d\n",
162 med->med_mcd->mcd_uuid, med->med_idx, LAST_RCVD, rc);
165 if (!test_and_clear_bit(med->med_idx, bitmap)) {
166 CERROR("MDS client %u: bit already clear in bitmap!!\n",
172 /* Make sure the server's last_transno is up to date. Do this
173 * after the client is freed so we know all the client's
174 * transactions have been committed. */
175 mds_update_server_data(exp->exp_obd, 1);
178 OBD_FREE(med->med_mcd, sizeof(*med->med_mcd));
182 static int mds_server_free_data(struct mds_obd *mds)
184 OBD_FREE(mds->mds_client_bitmap,
185 MDS_MAX_CLIENT_WORDS * sizeof(unsigned long));
186 OBD_FREE(mds->mds_server_data, sizeof(*mds->mds_server_data));
187 mds->mds_server_data = NULL;
192 static int mds_read_last_fid(struct obd_device *obd, struct file *file)
196 struct mds_obd *mds = &obd->u.mds;
197 unsigned long last_fid_size = file->f_dentry->d_inode->i_size;
200 if (last_fid_size == 0) {
201 CWARN("%s: initializing new %s\n", obd->obd_name,
202 file->f_dentry->d_name.name);
205 * as fid is used for forming res_id for locking, it should not
206 * be zero. This will keep us out of lots possible problems,
209 mds_set_last_fid(obd, 0);
213 rc = fsfilt_read_record(obd, file, &lastfid,
214 sizeof(lastfid), &off);
216 CERROR("error reading MDS %s: rc = %d\n",
217 file->f_dentry->d_name.name, rc);
222 * make sure, that fid is up-to-date.
224 mds_set_last_fid(obd, lastfid);
227 CDEBUG(D_INODE, "%s: server last_fid: "LPU64"\n",
228 obd->obd_name, mds->mds_last_fid);
230 rc = mds_update_last_fid(obd, NULL, 1);
234 static int mds_read_last_rcvd(struct obd_device *obd, struct file *file)
236 struct mds_obd *mds = &obd->u.mds;
237 struct mds_server_data *msd;
238 struct mds_client_data *mcd = NULL;
240 unsigned long last_rcvd_size = file->f_dentry->d_inode->i_size;
245 /* ensure padding in the struct is the correct size */
246 LASSERT(offsetof(struct mds_server_data, msd_padding) +
247 sizeof(msd->msd_padding) == MDS_LR_SERVER_SIZE);
248 LASSERT(offsetof(struct mds_client_data, mcd_padding) +
249 sizeof(mcd->mcd_padding) == MDS_LR_CLIENT_SIZE);
251 OBD_ALLOC_WAIT(msd, sizeof(*msd));
255 OBD_ALLOC_WAIT(mds->mds_client_bitmap,
256 MDS_MAX_CLIENT_WORDS * sizeof(unsigned long));
257 if (!mds->mds_client_bitmap) {
258 OBD_FREE(msd, sizeof(*msd));
262 mds->mds_server_data = msd;
264 if (last_rcvd_size == 0) {
265 CWARN("%s: initializing new %s\n", obd->obd_name,
266 file->f_dentry->d_name.name);
268 memcpy(msd->msd_uuid, obd->obd_uuid.uuid,sizeof(msd->msd_uuid));
269 msd->msd_last_transno = 0;
270 mount_count = msd->msd_mount_count = 0;
271 msd->msd_server_size = cpu_to_le32(MDS_LR_SERVER_SIZE);
272 msd->msd_client_start = cpu_to_le32(MDS_LR_CLIENT_START);
273 msd->msd_client_size = cpu_to_le16(MDS_LR_CLIENT_SIZE);
274 msd->msd_feature_rocompat = cpu_to_le32(MDS_ROCOMPAT_LOVOBJID);
276 rc = fsfilt_read_record(obd, file, msd, sizeof(*msd), &off);
278 CERROR("error reading MDS %s: rc = %d\n",
279 file->f_dentry->d_name.name, rc);
282 if (strcmp(msd->msd_uuid, obd->obd_uuid.uuid) != 0) {
283 CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
284 obd->obd_uuid.uuid, msd->msd_uuid);
285 GOTO(err_msd, rc = -EINVAL);
287 mount_count = le64_to_cpu(msd->msd_mount_count);
289 if (msd->msd_feature_incompat & ~cpu_to_le32(MDS_INCOMPAT_SUPP)) {
290 CERROR("unsupported incompat feature %x\n",
291 le32_to_cpu(msd->msd_feature_incompat) &
293 GOTO(err_msd, rc = -EINVAL);
295 /* XXX updating existing b_devel fs only, can be removed in future */
296 msd->msd_feature_rocompat = cpu_to_le32(MDS_ROCOMPAT_LOVOBJID);
297 if (msd->msd_feature_rocompat & ~cpu_to_le32(MDS_ROCOMPAT_SUPP)) {
298 CERROR("unsupported read-only feature %x\n",
299 le32_to_cpu(msd->msd_feature_rocompat) &
301 /* Do something like remount filesystem read-only */
302 GOTO(err_msd, rc = -EINVAL);
305 mds->mds_last_transno = le64_to_cpu(msd->msd_last_transno);
307 CDEBUG(D_INODE, "%s: server last_transno: "LPU64"\n",
308 obd->obd_name, mds->mds_last_transno);
309 CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
310 obd->obd_name, mount_count + 1);
311 CDEBUG(D_INODE, "%s: server data size: %u\n",
312 obd->obd_name, le32_to_cpu(msd->msd_server_size));
313 CDEBUG(D_INODE, "%s: per-client data start: %u\n",
314 obd->obd_name, le32_to_cpu(msd->msd_client_start));
315 CDEBUG(D_INODE, "%s: per-client data size: %u\n",
316 obd->obd_name, le32_to_cpu(msd->msd_client_size));
317 CDEBUG(D_INODE, "%s: last_rcvd size: %lu\n",
318 obd->obd_name, last_rcvd_size);
319 CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
320 last_rcvd_size <= le32_to_cpu(msd->msd_client_start) ? 0 :
321 (last_rcvd_size - le32_to_cpu(msd->msd_client_start)) /
322 le16_to_cpu(msd->msd_client_size));
324 /* When we do a clean MDS shutdown, we save the last_transno into
325 * the header. If we find clients with higher last_transno values
326 * then those clients may need recovery done. */
327 for (cl_idx = 0, off = le32_to_cpu(msd->msd_client_start);
328 off < last_rcvd_size; cl_idx++) {
330 struct obd_export *exp;
331 struct mds_export_data *med;
334 OBD_ALLOC_WAIT(mcd, sizeof(*mcd));
336 GOTO(err_client, rc = -ENOMEM);
339 /* Don't assume off is incremented properly by
340 * fsfilt_read_record(), in case sizeof(*mcd)
341 * isn't the same as msd->msd_client_size. */
342 off = le32_to_cpu(msd->msd_client_start) +
343 cl_idx * le16_to_cpu(msd->msd_client_size);
344 rc = fsfilt_read_record(obd, file, mcd, sizeof(*mcd), &off);
346 CERROR("error reading MDS %s idx %d, off %llu: rc %d\n",
347 file->f_dentry->d_name.name, cl_idx, off, rc);
348 break; /* read error shouldn't cause startup to fail */
351 if (mcd->mcd_uuid[0] == '\0') {
352 CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
357 last_transno = le64_to_cpu(mcd->mcd_last_transno);
359 /* These exports are cleaned up by mds_disconnect(), so they
360 * need to be set up like real exports as mds_connect() does.
362 CDEBUG(D_HA|D_WARNING,"RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
363 " srv lr: "LPU64" lx: "LPU64"\n", mcd->mcd_uuid, cl_idx,
364 last_transno, le64_to_cpu(msd->msd_last_transno),
367 exp = class_new_export(obd);
369 GOTO(err_client, rc = -ENOMEM);
371 memcpy(&exp->exp_client_uuid.uuid, mcd->mcd_uuid,
372 sizeof exp->exp_client_uuid.uuid);
373 med = &exp->exp_mds_data;
375 mds_client_add(obd, mds, med, cl_idx);
376 /* create helper if export init gets more complex */
377 INIT_LIST_HEAD(&med->med_open_head);
378 spin_lock_init(&med->med_open_lock);
381 exp->exp_replay_needed = 1;
382 obd->obd_recoverable_clients++;
383 obd->obd_max_recoverable_clients++;
384 class_export_put(exp);
386 CDEBUG(D_OTHER, "client at idx %d has last_transno = "LPU64"\n",
387 cl_idx, last_transno);
389 if (last_transno > mds->mds_last_transno)
390 mds->mds_last_transno = last_transno;
393 OBD_FREE(mcd, sizeof(*mcd));
394 obd->obd_last_committed = mds->mds_last_transno;
395 if (obd->obd_recoverable_clients) {
396 CWARN("RECOVERY: service %s, %d recoverable clients, "
397 "last_transno "LPU64"\n", obd->obd_name,
398 obd->obd_recoverable_clients, mds->mds_last_transno);
399 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
400 target_start_recovery_thread(obd, mds_handle);
401 obd->obd_recovery_start = LTIME_S(CURRENT_TIME);
404 mds->mds_mount_count = mount_count + 1;
405 msd->msd_mount_count = cpu_to_le64(mds->mds_mount_count);
407 /* save it, so mount count and last_transno is current */
408 rc = mds_update_server_data(obd, 1);
410 GOTO(err_client, rc);
415 class_disconnect_exports(obd, 0);
417 mds_server_free_data(mds);
421 static int mds_fs_post_setup(struct obd_device *obd)
423 struct mds_obd *mds = &obd->u.mds;
424 struct dentry *dentry;
428 dentry = mds_id2dentry(obd, &mds->mds_rootid, NULL);
429 if (IS_ERR(dentry)) {
430 CERROR("Can't find ROOT, err = %d\n",
431 (int)PTR_ERR(dentry));
432 RETURN(PTR_ERR(dentry));
435 rc = fsfilt_post_setup(obd, dentry);
439 LASSERT(dentry->d_inode != NULL);
441 fsfilt_set_fs_flags(obd, dentry->d_inode,
442 SM_DO_REC | SM_DO_COW);
444 fsfilt_set_fs_flags(obd, mds->mds_pending_dir->d_inode,
445 SM_DO_REC | SM_DO_COW);
447 fsfilt_set_mds_flags(obd, mds->mds_sb);
455 * sets up root inode lustre_id. It tries to read it first from root inode and
456 * if it is not there, new rootid is allocated and saved there.
458 int mds_fs_setup_rootid(struct obd_device *obd)
463 struct dentry *dentry;
464 struct lustre_id rootid;
465 struct mds_obd *mds = &obd->u.mds;
468 memcpy(&rootid, &mds->mds_rootid, sizeof(rootid));
470 /* getting root directory and setup its fid. */
471 dentry = mds_id2dentry(obd, &rootid, NULL);
472 if (IS_ERR(dentry)) {
473 CERROR("Can't find ROOT, err = %d\n",
474 (int)PTR_ERR(dentry));
475 RETURN(PTR_ERR(dentry));
478 inode = dentry->d_inode;
479 LASSERT(dentry->d_inode);
481 rc = mds_pack_inode2id(obd, &rootid, inode, 1);
487 * rootid is filled by mds_read_inode_sid(), so we do not need
488 * to allocate it and update. The only thing we need to check is
491 LASSERT(id_group(&rootid) == mds->mds_num);
492 mds_set_last_fid(obd, id_fid(&rootid));
493 GOTO(out_dentry, rc);
496 /* allocating new one, as it is not found in root inode. */
497 handle = fsfilt_start(obd, inode,
498 FSFILT_OP_SETATTR, NULL);
500 if (IS_ERR(handle)) {
501 rc = PTR_ERR(handle);
502 CERROR("fsfilt_start() failed, rc = %d\n", rc);
503 GOTO(out_dentry, rc);
507 rc = mds_alloc_inode_sid(obd, inode, handle, &rootid);
511 CERROR("mds_alloc_inode_sid() failed, rc = %d\n",
513 GOTO(out_dentry, rc);
516 rc = fsfilt_commit(obd, mds->mds_sb, inode, handle, 0);
518 CERROR("fsfilt_commit() failed, rc = %d\n", rc);
523 memcpy(&mds->mds_rootid, &rootid, sizeof(rootid));
524 CWARN("%s: rootid: "DLID4"\n", obd->obd_name,
531 * initializes lustre_id for virtual id directory, it is needed sometimes, as it
532 * is possible that it will be the parent for object an operations is going to
535 int mds_fs_setup_virtid(struct obd_device *obd)
540 struct mds_obd *mds = &obd->u.mds;
541 struct inode *inode = mds->mds_id_dir->d_inode;
544 handle = fsfilt_start(obd, inode,
545 FSFILT_OP_SETATTR, NULL);
547 if (IS_ERR(handle)) {
548 rc = PTR_ERR(handle);
549 CERROR("fsfilt_start() failed, rc = %d\n", rc);
554 rc = mds_alloc_inode_sid(obd, inode, handle, &id);
558 CERROR("mds_alloc_inode_sid() failed, rc = %d\n",
563 rc = fsfilt_commit(obd, mds->mds_sb, inode, handle, 0);
565 CERROR("fsfilt_commit() failed, rc = %d\n", rc);
572 int mds_fs_setup(struct obd_device *obd, struct vfsmount *mnt)
574 struct mds_obd *mds = &obd->u.mds;
575 struct lvfs_run_ctxt saved;
576 struct dentry *dentry;
581 rc = cleanup_group_info();
585 mds->mds_vfsmnt = mnt;
586 mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
588 fsfilt_setup(obd, mds->mds_sb);
590 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
591 obd->obd_lvfs_ctxt.pwdmnt = mnt;
592 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
593 obd->obd_lvfs_ctxt.fs = get_ds();
594 obd->obd_lvfs_ctxt.cb_ops = mds_lvfs_ops;
596 /* setup the directory tree */
597 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
598 dentry = simple_mkdir(current->fs->pwd, "ROOT", 0755, 0);
599 if (IS_ERR(dentry)) {
600 rc = PTR_ERR(dentry);
601 CERROR("cannot create ROOT directory: rc = %d\n", rc);
605 mdc_pack_id(&mds->mds_rootid, dentry->d_inode->i_ino,
606 dentry->d_inode->i_generation, S_IFDIR, 0, 0);
610 dentry = lookup_one_len("__iopen__", current->fs->pwd,
611 strlen("__iopen__"));
612 if (IS_ERR(dentry)) {
613 rc = PTR_ERR(dentry);
614 CERROR("cannot lookup __iopen__ directory: rc = %d\n", rc);
617 mds->mds_id_de = dentry;
619 if (!dentry->d_inode) {
621 CERROR("__iopen__ directory has no inode? rc = %d\n", rc);
625 dentry = simple_mkdir(current->fs->pwd, "PENDING", 0777, 1);
626 if (IS_ERR(dentry)) {
627 rc = PTR_ERR(dentry);
628 CERROR("cannot create PENDING directory: rc = %d\n", rc);
631 mds->mds_pending_dir = dentry;
633 dentry = simple_mkdir(current->fs->pwd, "LOGS", 0777, 1);
634 if (IS_ERR(dentry)) {
635 rc = PTR_ERR(dentry);
636 CERROR("cannot create LOGS directory: rc = %d\n", rc);
637 GOTO(err_pending, rc);
639 mds->mds_logs_dir = dentry;
641 dentry = simple_mkdir(current->fs->pwd, "OBJECTS", 0777, 1);
642 if (IS_ERR(dentry)) {
643 rc = PTR_ERR(dentry);
644 CERROR("cannot create OBJECTS directory: rc = %d\n", rc);
647 mds->mds_objects_dir = dentry;
649 dentry = simple_mkdir(current->fs->pwd, "FIDS", 0777, 1);
650 if (IS_ERR(dentry)) {
651 rc = PTR_ERR(dentry);
652 CERROR("cannot create FIDS directory: rc = %d\n", rc);
653 GOTO(err_objects, rc);
655 mds->mds_id_dir = dentry;
657 dentry = simple_mkdir(current->fs->pwd, "UNNAMED", 0777, 1);
658 if (IS_ERR(dentry)) {
659 rc = PTR_ERR(dentry);
660 CERROR("cannot create UNNAMED directory: rc = %d\n", rc);
661 GOTO(err_unnamed, rc);
663 mds->mds_unnamed_dir = dentry;
665 /* open and test the last rcvd file */
666 file = filp_open(LAST_RCVD, O_RDWR | O_CREAT, 0644);
669 CERROR("cannot open/create %s file: rc = %d\n", LAST_RCVD, rc);
670 GOTO(err_id_dir, rc = PTR_ERR(file));
672 mds->mds_rcvd_filp = file;
674 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
675 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
676 file->f_dentry->d_inode->i_mode);
677 GOTO(err_last_rcvd, rc = -ENOENT);
680 rc = mds_read_last_rcvd(obd, file);
682 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
683 GOTO(err_last_rcvd, rc);
686 /* open and test the last fid file */
687 file = filp_open(LAST_FID, O_RDWR | O_CREAT, 0644);
690 CERROR("cannot open/create %s file: rc = %d\n",
692 GOTO(err_client, rc = PTR_ERR(file));
694 mds->mds_fid_filp = file;
695 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
696 CERROR("%s is not a regular file!: mode = %o\n",
697 LAST_FID, file->f_dentry->d_inode->i_mode);
698 GOTO(err_last_fid, rc = -ENOENT);
701 rc = mds_read_last_fid(obd, file);
703 CERROR("cannot read %s: rc = %d\n", LAST_FID, rc);
704 GOTO(err_last_fid, rc);
707 /* open and test the lov objid file */
708 file = filp_open(LOV_OBJID, O_RDWR | O_CREAT, 0644);
711 CERROR("cannot open/create %s file: rc = %d\n", LOV_OBJID, rc);
712 GOTO(err_last_fid, rc = PTR_ERR(file));
714 mds->mds_lov_objid_filp = file;
715 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
716 CERROR("%s is not a regular file!: mode = %o\n", LOV_OBJID,
717 file->f_dentry->d_inode->i_mode);
718 GOTO(err_lov_objid, rc = -ENOENT);
722 rc = mds_fs_post_setup(obd);
724 CERROR("can not post setup fsfilt\n");
726 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
730 if (mds->mds_lov_objid_filp && filp_close(mds->mds_lov_objid_filp, 0))
731 CERROR("can't close %s after error\n", LOV_OBJID);
733 if (mds->mds_fid_filp && filp_close(mds->mds_fid_filp, 0))
734 CERROR("can't close %s after error\n", LAST_FID);
736 class_disconnect_exports(obd, 0);
738 if (mds->mds_rcvd_filp && filp_close(mds->mds_rcvd_filp, 0))
739 CERROR("can't close %s after error\n", LAST_RCVD);
741 dput(mds->mds_unnamed_dir);
743 dput(mds->mds_id_dir);
745 dput(mds->mds_objects_dir);
747 dput(mds->mds_logs_dir);
749 dput(mds->mds_pending_dir);
751 dput(mds->mds_id_de);
755 static int mds_fs_post_cleanup(struct obd_device *obd)
758 rc = fsfilt_post_cleanup(obd);
762 int mds_fs_cleanup(struct obd_device *obd, int flags)
764 struct mds_obd *mds = &obd->u.mds;
765 struct lvfs_run_ctxt saved;
768 if (flags & OBD_OPT_FAILOVER)
769 CERROR("%s: shutting down for failover; client state will"
770 " be preserved.\n", obd->obd_name);
772 class_disconnect_exports(obd, flags); /* cleans up client info too */
773 target_cleanup_recovery(obd);
774 mds_server_free_data(mds);
776 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
777 if (mds->mds_fid_filp) {
778 rc = filp_close(mds->mds_fid_filp, 0);
779 mds->mds_fid_filp = NULL;
781 CERROR("%s file won't close, rc = %d\n", LAST_FID, rc);
783 if (mds->mds_rcvd_filp) {
784 rc = filp_close(mds->mds_rcvd_filp, 0);
785 mds->mds_rcvd_filp = NULL;
787 CERROR("%s file won't close, rc = %d\n", LAST_RCVD, rc);
789 if (mds->mds_lov_objid_filp) {
790 rc = filp_close(mds->mds_lov_objid_filp, 0);
791 mds->mds_lov_objid_filp = NULL;
793 CERROR("%s file won't close, rc=%d\n", LOV_OBJID, rc);
795 if (mds->mds_unnamed_dir != NULL) {
796 l_dput(mds->mds_unnamed_dir);
797 mds->mds_unnamed_dir = NULL;
799 if (mds->mds_id_dir != NULL) {
800 l_dput(mds->mds_id_dir);
801 mds->mds_id_dir = NULL;
803 if (mds->mds_objects_dir != NULL) {
804 l_dput(mds->mds_objects_dir);
805 mds->mds_objects_dir = NULL;
807 if (mds->mds_logs_dir) {
808 l_dput(mds->mds_logs_dir);
809 mds->mds_logs_dir = NULL;
811 if (mds->mds_pending_dir) {
812 l_dput(mds->mds_pending_dir);
813 mds->mds_pending_dir = NULL;
815 rc = mds_fs_post_cleanup(obd);
817 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
818 shrink_dcache_parent(mds->mds_id_de);
819 dput(mds->mds_id_de);
824 /* Creates an object with the same name as its id. Because this is not at all
825 * performance sensitive, it is accomplished by creating a file, checking the
826 * id, and renaming it. */
827 int mds_obd_create(struct obd_export *exp, struct obdo *oa,
828 struct lov_stripe_md **ea, struct obd_trans_info *oti)
830 struct mds_obd *mds = &exp->exp_obd->u.mds;
831 struct inode *parent_inode = mds->mds_objects_dir->d_inode;
833 struct dentry *dchild;
834 struct lvfs_run_ctxt saved;
835 char idname[LL_ID_NAMELEN];
836 int rc = 0, err, idlen;
840 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
841 down(&parent_inode->i_sem);
843 idlen = ll_id2str(idname, oa->o_id, oa->o_generation);
844 dchild = lookup_one_len(idname, mds->mds_objects_dir, idlen);
846 GOTO(out_pop, rc = PTR_ERR(dchild));
848 if (dchild->d_inode == NULL) {
849 struct dentry_params dp;
852 CWARN("creating log with ID "LPU64"\n", oa->o_id);
854 dchild->d_fsdata = (void *) &dp;
856 dp.p_inum = oa->o_id;
857 rc = ll_vfs_create(parent_inode, dchild, S_IFREG, NULL);
858 if (dchild->d_fsdata == (void *)(unsigned long)oa->o_id)
859 dchild->d_fsdata = NULL;
861 CDEBUG(D_INODE, "err during create: %d\n", rc);
865 inode = dchild->d_inode;
866 LASSERT(inode->i_ino == oa->o_id);
867 inode->i_generation = oa->o_generation;
868 CDEBUG(D_HA, "recreated ino %lu with gen %u\n",
869 inode->i_ino, inode->i_generation);
870 mark_inode_dirty(inode);
872 CWARN("it should be here!\n");
877 sprintf(idname, "OBJECTS/%u.%u", ll_insecure_random_int(), current->pid);
878 filp = filp_open(idname, O_CREAT | O_EXCL, 0644);
882 CERROR("impossible object name collision %s\n",
886 CERROR("error creating tmp object %s: rc %d\n",
891 LASSERT(mds->mds_objects_dir == filp->f_dentry->d_parent);
893 oa->o_id = filp->f_dentry->d_inode->i_ino;
894 oa->o_generation = filp->f_dentry->d_inode->i_generation;
895 idlen = ll_id2str(idname, oa->o_id, oa->o_generation);
897 CWARN("created log anonymous "LPU64"/%u\n",
898 oa->o_id, oa->o_generation);
900 dchild = lookup_one_len(idname, mds->mds_objects_dir, idlen);
901 if (IS_ERR(dchild)) {
902 CERROR("getting neg dentry for obj rename: %d\n", rc);
903 GOTO(out_close, rc = PTR_ERR(dchild));
905 if (dchild->d_inode != NULL) {
906 CERROR("impossible non-negative obj dentry " LPU64":%u!\n",
907 oa->o_id, oa->o_generation);
911 handle = fsfilt_start(exp->exp_obd, mds->mds_objects_dir->d_inode,
912 FSFILT_OP_RENAME, NULL);
914 GOTO(out_dput, rc = PTR_ERR(handle));
917 rc = vfs_rename(mds->mds_objects_dir->d_inode, filp->f_dentry,
918 mds->mds_objects_dir->d_inode, dchild);
921 CERROR("error renaming new object "LPU64":%u: rc %d\n",
922 oa->o_id, oa->o_generation, rc);
924 err = fsfilt_commit(exp->exp_obd, mds->mds_sb,
925 mds->mds_objects_dir->d_inode, handle, 0);
927 oa->o_gr = FILTER_GROUP_FIRST_MDS + mds->mds_num;
928 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGENER | OBD_MD_FLGROUP;
934 err = filp_close(filp, 0);
936 CERROR("closing tmpfile %s: rc %d\n", idname, rc);
941 up(&parent_inode->i_sem);
942 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
946 int mds_obd_destroy(struct obd_export *exp, struct obdo *oa,
947 struct lov_stripe_md *ea, struct obd_trans_info *oti)
949 struct mds_obd *mds = &exp->exp_obd->u.mds;
950 struct inode *parent_inode = mds->mds_objects_dir->d_inode;
951 struct obd_device *obd = exp->exp_obd;
952 struct lvfs_run_ctxt saved;
953 char idname[LL_ID_NAMELEN];
956 int err, idlen, rc = 0;
959 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
961 idlen = ll_id2str(idname, oa->o_id, oa->o_generation);
963 down(&parent_inode->i_sem);
964 de = lookup_one_len(idname, mds->mds_objects_dir, idlen);
965 if (IS_ERR(de) || de->d_inode == NULL) {
966 rc = IS_ERR(de) ? PTR_ERR(de) : -ENOENT;
967 CERROR("destroying non-existent object "LPU64" %s: rc %d\n",
968 oa->o_id, idname, rc);
971 /* Stripe count is 1 here since this is some MDS specific stuff
972 that is unlinked, not spanned across multiple OSTs */
973 handle = fsfilt_start_log(obd, mds->mds_objects_dir->d_inode,
974 FSFILT_OP_UNLINK, oti, 1);
977 GOTO(out_dput, rc = PTR_ERR(handle));
979 rc = vfs_unlink(mds->mds_objects_dir->d_inode, de);
981 CERROR("error destroying object "LPU64":%u: rc %d\n",
982 oa->o_id, oa->o_generation, rc);
984 err = fsfilt_commit(obd, mds->mds_sb, mds->mds_objects_dir->d_inode,
991 up(&parent_inode->i_sem);
992 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);