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 <libcfs/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"
54 #define VIRT_FID "virt_fid"
56 /* Add client data to the MDS. We use a bitmap to locate a free space
57 * in the last_rcvd file if cl_off is -1 (i.e. a new client).
58 * Otherwise, we have just read the data from the last_rcvd file and
61 int mds_client_add(struct obd_device *obd, struct mds_obd *mds,
62 struct mds_export_data *med, int cl_idx)
64 unsigned long *bitmap = mds->mds_client_bitmap;
65 int new_client = (cl_idx == -1);
68 LASSERT(bitmap != NULL);
70 /* XXX if mcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
71 if (!strcmp(med->med_mcd->mcd_uuid, obd->obd_uuid.uuid))
74 /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
75 * there's no need for extra complication here
78 cl_idx = find_first_zero_bit(bitmap, MDS_MAX_CLIENTS);
80 if (cl_idx >= MDS_MAX_CLIENTS) {
81 CERROR("no room for clients - fix MDS_MAX_CLIENTS\n");
84 if (test_and_set_bit(cl_idx, bitmap)) {
85 cl_idx = find_next_zero_bit(bitmap, MDS_MAX_CLIENTS,
90 if (test_and_set_bit(cl_idx, bitmap)) {
91 CERROR("MDS client %d: bit already set in bitmap!!\n",
97 CDEBUG(D_INFO, "client at idx %d with UUID '%s' added\n",
98 cl_idx, med->med_mcd->mcd_uuid);
100 med->med_idx = cl_idx;
101 med->med_off = le32_to_cpu(mds->mds_server_data->msd_client_start) +
102 (cl_idx * le16_to_cpu(mds->mds_server_data->msd_client_size));
105 struct file *file = mds->mds_rcvd_filp;
106 struct lvfs_run_ctxt saved;
107 loff_t off = med->med_off;
110 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
111 rc = fsfilt_write_record(obd, file, med->med_mcd,
112 sizeof(*med->med_mcd), &off, 1);
113 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
117 CDEBUG(D_INFO, "wrote client mcd at idx %u off %llu (len %u)\n",
118 med->med_idx, med->med_off,
119 (unsigned int)sizeof(*med->med_mcd));
124 int mds_client_free(struct obd_export *exp, int clear_client)
126 struct mds_export_data *med = &exp->exp_mds_data;
127 struct mds_obd *mds = &exp->exp_obd->u.mds;
128 unsigned long *bitmap = mds->mds_client_bitmap;
129 struct obd_device *obd = exp->exp_obd;
130 struct mds_client_data zero_mcd;
131 struct lvfs_run_ctxt saved;
137 /* XXX if mcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
138 if (!strcmp(med->med_mcd->mcd_uuid, obd->obd_uuid.uuid))
139 GOTO(free_and_out, 0);
141 CDEBUG(D_INFO, "freeing client at idx %u (%lld)with UUID '%s'\n",
142 med->med_idx, med->med_off, med->med_mcd->mcd_uuid);
146 /* Clear the bit _after_ zeroing out the client so we don't
147 race with mds_client_add and zero out new clients.*/
148 if (!test_bit(med->med_idx, bitmap)) {
149 CERROR("MDS client %u: bit already clear in bitmap!!\n",
155 memset(&zero_mcd, 0, sizeof zero_mcd);
156 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
157 rc = fsfilt_write_record(obd, mds->mds_rcvd_filp, &zero_mcd,
158 sizeof(zero_mcd), &med->med_off, 1);
159 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
161 CDEBUG(rc == 0 ? D_INFO : D_ERROR,
162 "zeroing out client %s idx %u in %s rc %d\n",
163 med->med_mcd->mcd_uuid, med->med_idx, LAST_RCVD, rc);
166 if (!test_and_clear_bit(med->med_idx, bitmap)) {
167 CERROR("MDS client %u: bit already clear in bitmap!!\n",
173 /* Make sure the server's last_transno is up to date. Do this
174 * after the client is freed so we know all the client's
175 * transactions have been committed. */
176 mds_update_server_data(exp->exp_obd, 1);
179 OBD_FREE(med->med_mcd, sizeof(*med->med_mcd));
183 static int mds_server_free_data(struct mds_obd *mds)
185 OBD_FREE(mds->mds_client_bitmap,
186 MDS_MAX_CLIENT_WORDS * sizeof(unsigned long));
187 OBD_FREE(mds->mds_server_data, sizeof(*mds->mds_server_data));
188 mds->mds_server_data = NULL;
193 static int mds_read_last_fid(struct obd_device *obd, struct file *file)
197 struct mds_obd *mds = &obd->u.mds;
198 unsigned long last_fid_size = file->f_dentry->d_inode->i_size;
201 if (last_fid_size == 0) {
202 CWARN("%s: initializing new %s\n", obd->obd_name,
203 file->f_dentry->d_name.name);
206 * as fid is used for forming res_id for locking, it should not
207 * be zero. This will keep us out of lots possible problems,
210 mds_set_last_fid(obd, 0);
214 rc = fsfilt_read_record(obd, file, &lastfid,
215 sizeof(lastfid), &off);
217 CERROR("error reading MDS %s: rc = %d\n",
218 file->f_dentry->d_name.name, rc);
223 * make sure, that fid is up-to-date.
225 mds_set_last_fid(obd, lastfid);
228 CDEBUG(D_INODE, "%s: server last_fid: "LPU64"\n",
229 obd->obd_name, mds->mds_last_fid);
231 rc = mds_update_last_fid(obd, NULL, 1);
235 static int mds_read_last_rcvd(struct obd_device *obd, struct file *file)
237 unsigned long last_rcvd_size = file->f_dentry->d_inode->i_size;
238 struct mds_obd *mds = &obd->u.mds;
239 struct mds_server_data *msd = NULL;
240 struct mds_client_data *mcd = NULL;
246 /* ensure padding in the struct is the correct size */
247 LASSERT(offsetof(struct mds_server_data, msd_padding) +
248 sizeof(msd->msd_padding) == MDS_LR_SERVER_SIZE);
249 LASSERT(offsetof(struct mds_client_data, mcd_padding) +
250 sizeof(mcd->mcd_padding) == MDS_LR_CLIENT_SIZE);
252 OBD_ALLOC_WAIT(msd, sizeof(*msd));
256 OBD_ALLOC_WAIT(mds->mds_client_bitmap,
257 MDS_MAX_CLIENT_WORDS * sizeof(unsigned long));
258 if (!mds->mds_client_bitmap) {
259 OBD_FREE(msd, sizeof(*msd));
263 mds->mds_server_data = msd;
265 if (last_rcvd_size == 0) {
266 CWARN("%s: initializing new %s\n", obd->obd_name,
267 file->f_dentry->d_name.name);
269 memcpy(msd->msd_uuid, obd->obd_uuid.uuid,sizeof(msd->msd_uuid));
270 msd->msd_last_transno = 0;
271 mount_count = msd->msd_mount_count = 0;
272 msd->msd_server_size = cpu_to_le32(MDS_LR_SERVER_SIZE);
273 msd->msd_client_start = cpu_to_le32(MDS_LR_CLIENT_START);
274 msd->msd_client_size = cpu_to_le16(MDS_LR_CLIENT_SIZE);
275 msd->msd_feature_rocompat = cpu_to_le32(MDS_ROCOMPAT_LOVOBJID);
277 rc = fsfilt_read_record(obd, file, msd, sizeof(*msd), &off);
279 CERROR("error reading MDS %s: rc = %d\n",
280 file->f_dentry->d_name.name, rc);
283 if (strcmp(msd->msd_uuid, obd->obd_uuid.uuid) != 0) {
284 CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
285 obd->obd_uuid.uuid, msd->msd_uuid);
286 GOTO(err_msd, rc = -EINVAL);
288 mount_count = le64_to_cpu(msd->msd_mount_count);
290 if (msd->msd_feature_incompat & ~cpu_to_le32(MDS_INCOMPAT_SUPP)) {
291 CERROR("unsupported incompat feature %x\n",
292 le32_to_cpu(msd->msd_feature_incompat) &
294 GOTO(err_msd, rc = -EINVAL);
296 /* XXX updating existing b_devel fs only, can be removed in future */
297 msd->msd_feature_rocompat = cpu_to_le32(MDS_ROCOMPAT_LOVOBJID);
298 if (msd->msd_feature_rocompat & ~cpu_to_le32(MDS_ROCOMPAT_SUPP)) {
299 CERROR("unsupported read-only feature %x\n",
300 le32_to_cpu(msd->msd_feature_rocompat) &
302 /* Do something like remount filesystem read-only */
303 GOTO(err_msd, rc = -EINVAL);
306 mds->mds_last_transno = le64_to_cpu(msd->msd_last_transno);
308 CDEBUG(D_INODE, "%s: server last_transno: "LPU64"\n",
309 obd->obd_name, mds->mds_last_transno);
310 CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
311 obd->obd_name, mount_count + 1);
312 CDEBUG(D_INODE, "%s: server data size: %u\n",
313 obd->obd_name, le32_to_cpu(msd->msd_server_size));
314 CDEBUG(D_INODE, "%s: per-client data start: %u\n",
315 obd->obd_name, le32_to_cpu(msd->msd_client_start));
316 CDEBUG(D_INODE, "%s: per-client data size: %u\n",
317 obd->obd_name, le32_to_cpu(msd->msd_client_size));
318 CDEBUG(D_INODE, "%s: last_rcvd size: %lu\n",
319 obd->obd_name, last_rcvd_size);
320 CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
321 last_rcvd_size <= le32_to_cpu(msd->msd_client_start) ? 0 :
322 (last_rcvd_size - le32_to_cpu(msd->msd_client_start)) /
323 le16_to_cpu(msd->msd_client_size));
325 /* When we do a clean MDS shutdown, we save the last_transno into
326 * the header. If we find clients with higher last_transno values
327 * then those clients may need recovery done. */
328 for (cl_idx = 0, off = le32_to_cpu(msd->msd_client_start);
329 off < last_rcvd_size; cl_idx++) {
331 struct obd_export *exp;
332 struct mds_export_data *med;
335 OBD_ALLOC_WAIT(mcd, sizeof(*mcd));
337 GOTO(err_client, rc = -ENOMEM);
340 /* Don't assume off is incremented properly by
341 * fsfilt_read_record(), in case sizeof(*mcd)
342 * isn't the same as msd->msd_client_size. */
343 off = le32_to_cpu(msd->msd_client_start) +
344 cl_idx * le16_to_cpu(msd->msd_client_size);
345 rc = fsfilt_read_record(obd, file, mcd, sizeof(*mcd), &off);
347 CERROR("error reading MDS %s idx %d, off %llu: rc %d\n",
348 file->f_dentry->d_name.name, cl_idx, off, rc);
349 break; /* read error shouldn't cause startup to fail */
352 if (mcd->mcd_uuid[0] == '\0') {
353 CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
358 last_transno = le64_to_cpu(mcd->mcd_last_transno);
360 /* These exports are cleaned up by mds_disconnect(), so they
361 * need to be set up like real exports as mds_connect() does.
363 CDEBUG(D_HA|D_WARNING,"RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
364 " srv lr: "LPU64" lx: "LPU64"\n", mcd->mcd_uuid, cl_idx,
365 last_transno, le64_to_cpu(msd->msd_last_transno),
368 exp = class_new_export(obd);
370 GOTO(err_client, rc = -ENOMEM);
372 memcpy(&exp->exp_client_uuid.uuid, mcd->mcd_uuid,
373 sizeof exp->exp_client_uuid.uuid);
374 med = &exp->exp_mds_data;
376 mds_client_add(obd, mds, med, cl_idx);
377 /* create helper if export init gets more complex */
378 INIT_LIST_HEAD(&med->med_open_head);
379 spin_lock_init(&med->med_open_lock);
382 exp->exp_replay_needed = 1;
383 obd->obd_recoverable_clients++;
384 obd->obd_max_recoverable_clients++;
385 class_export_put(exp);
387 CDEBUG(D_OTHER, "client at idx %d has last_transno = "LPU64"\n",
388 cl_idx, last_transno);
390 if (last_transno > mds->mds_last_transno)
391 mds->mds_last_transno = last_transno;
394 OBD_FREE(mcd, sizeof(*mcd));
395 obd->obd_last_committed = mds->mds_last_transno;
396 if (obd->obd_recoverable_clients) {
397 CWARN("RECOVERY: service %s, %d recoverable clients, "
398 "last_transno "LPU64"\n", obd->obd_name,
399 obd->obd_recoverable_clients, mds->mds_last_transno);
400 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
401 target_start_recovery_thread(obd, mds_handle);
402 obd->obd_recovery_start = LTIME_S(CURRENT_TIME);
405 mds->mds_mount_count = mount_count + 1;
406 msd->msd_mount_count = cpu_to_le64(mds->mds_mount_count);
408 /* save it, so mount count and last_transno is current */
409 rc = mds_update_server_data(obd, 1);
411 GOTO(err_client, rc);
416 class_disconnect_exports(obd, 0);
418 mds_server_free_data(mds);
422 static int mds_fs_post_setup(struct obd_device *obd)
424 struct mds_obd *mds = &obd->u.mds;
425 struct dentry *dentry;
429 dentry = mds_id2dentry(obd, &mds->mds_rootid, NULL);
430 if (IS_ERR(dentry)) {
431 CERROR("Can't find ROOT, err = %d\n",
432 (int)PTR_ERR(dentry));
433 RETURN(PTR_ERR(dentry));
436 rc = fsfilt_post_setup(obd, dentry);
440 LASSERT(dentry->d_inode != NULL);
442 fsfilt_set_fs_flags(obd, dentry->d_inode,
443 SM_DO_REC | SM_DO_COW);
445 fsfilt_set_fs_flags(obd, mds->mds_pending_dir->d_inode,
446 SM_DO_REC | SM_DO_COW);
448 fsfilt_set_mds_flags(obd, mds->mds_sb);
456 * sets up root inode lustre_id. It tries to read it first from root inode and
457 * if it is not there, new rootid is allocated and saved there.
459 int mds_fs_setup_rootid(struct obd_device *obd)
464 struct dentry *dentry;
465 struct mds_obd *mds = &obd->u.mds;
468 /* getting root directory and setup its fid. */
469 dentry = mds_id2dentry(obd, &mds->mds_rootid, NULL);
470 if (IS_ERR(dentry)) {
471 CERROR("Can't find ROOT by "DLID4", err = %d\n",
472 OLID4(&mds->mds_rootid), (int)PTR_ERR(dentry));
473 RETURN(PTR_ERR(dentry));
476 inode = dentry->d_inode;
477 LASSERT(dentry->d_inode);
479 rc = mds_pack_inode2id(obd, &mds->mds_rootid, inode, 1);
482 GOTO(out_dentry, rc);
485 * rootid is filled by mds_read_inode_sid(), so we do not need
486 * to allocate it and update. The only thing we need to check is
489 LASSERT(id_group(&mds->mds_rootid) == mds->mds_num);
490 mds_set_last_fid(obd, id_fid(&mds->mds_rootid));
491 GOTO(out_dentry, rc);
494 /* allocating new one, as it is not found in root inode. */
495 handle = fsfilt_start(obd, inode,
496 FSFILT_OP_SETATTR, NULL);
498 if (IS_ERR(handle)) {
499 rc = PTR_ERR(handle);
500 CERROR("fsfilt_start() failed, rc = %d\n", rc);
501 GOTO(out_dentry, rc);
505 rc = mds_alloc_inode_sid(obd, inode, handle, &mds->mds_rootid);
509 CERROR("mds_alloc_inode_sid() failed, rc = %d\n",
511 GOTO(out_dentry, rc);
514 rc = fsfilt_commit(obd, mds->mds_sb, inode, handle, 0);
516 CERROR("fsfilt_commit() failed, rc = %d\n", rc);
522 CWARN("%s: rootid: "DLID4"\n", obd->obd_name,
523 OLID4(&mds->mds_rootid));
527 static int mds_update_virtid_fid(struct obd_device *obd,
528 void *handle, int force_sync)
530 struct mds_obd *mds = &obd->u.mds;
531 struct file *filp = mds->mds_virtid_filp;
532 struct lvfs_run_ctxt saved;
537 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
538 rc = fsfilt_write_record(obd, filp, &mds->mds_virtid_fid,
539 sizeof(mds->mds_virtid_fid),
542 CERROR("error writing MDS virtid_fid #"LPU64
543 ", err = %d\n", mds->mds_virtid_fid, rc);
546 CDEBUG(D_SUPER, "wrote virtid fid #"LPU64" at idx "
547 "%llu: err = %d\n", mds->mds_virtid_fid,
549 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
554 static int mds_read_virtid_fid(struct obd_device *obd,
559 struct mds_obd *mds = &obd->u.mds;
560 unsigned long virtid_fid_size = file->f_dentry->d_inode->i_size;
563 if (virtid_fid_size == 0) {
564 mds->mds_virtid_fid = mds_alloc_fid(obd);
566 rc = fsfilt_read_record(obd, file, &mds->mds_virtid_fid,
567 sizeof(mds->mds_virtid_fid), &off);
569 CERROR("error reading MDS %s: rc = %d\n",
570 file->f_dentry->d_name.name, rc);
574 rc = mds_update_virtid_fid(obd, NULL, 1);
580 * initializes lustre_id for virtual id directory, it is needed sometimes, as it
581 * is possible that it will be the parent for object an operations is going to
584 int mds_fs_setup_virtid(struct obd_device *obd)
588 struct lustre_id sid;
589 struct mds_obd *mds = &obd->u.mds;
590 struct inode *inode = mds->mds_id_dir->d_inode;
593 handle = fsfilt_start(obd, inode,
594 FSFILT_OP_SETATTR, NULL);
596 if (IS_ERR(handle)) {
597 rc = PTR_ERR(handle);
598 CERROR("fsfilt_start() failed, rc = %d\n", rc);
602 id_group(&sid) = mds->mds_num;
603 id_fid(&sid) = mds->mds_virtid_fid;
605 id_ino(&sid) = inode->i_ino;
606 id_gen(&sid) = inode->i_generation;
607 id_type(&sid) = (S_IFMT & inode->i_mode);
610 rc = mds_update_inode_sid(obd, inode, handle, &sid);
614 CERROR("mds_update_inode_sid() failed, rc = %d\n",
619 rc = fsfilt_commit(obd, mds->mds_sb, inode, handle, 0);
621 CERROR("fsfilt_commit() failed, rc = %d\n", rc);
628 int mds_fs_setup(struct obd_device *obd, struct vfsmount *mnt)
630 struct mds_obd *mds = &obd->u.mds;
631 struct lvfs_run_ctxt saved;
632 struct dentry *dentry;
637 rc = cleanup_group_info();
641 mds->mds_vfsmnt = mnt;
642 mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
644 fsfilt_setup(obd, mds->mds_sb);
646 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
647 obd->obd_lvfs_ctxt.pwdmnt = mnt;
648 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
649 obd->obd_lvfs_ctxt.fs = get_ds();
650 obd->obd_lvfs_ctxt.cb_ops = mds_lvfs_ops;
652 /* setup the directory tree */
653 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
654 dentry = simple_mkdir(current->fs->pwd, "ROOT", 0755, 0);
655 if (IS_ERR(dentry)) {
656 rc = PTR_ERR(dentry);
657 CERROR("cannot create ROOT directory: rc = %d\n", rc);
661 mdc_pack_id(&mds->mds_rootid, dentry->d_inode->i_ino,
662 dentry->d_inode->i_generation, S_IFDIR, 0, 0);
666 dentry = lookup_one_len("__iopen__", current->fs->pwd,
667 strlen("__iopen__"));
668 if (IS_ERR(dentry)) {
669 rc = PTR_ERR(dentry);
670 CERROR("cannot lookup __iopen__ directory: rc = %d\n", rc);
673 mds->mds_id_de = dentry;
675 if (!dentry->d_inode) {
677 CERROR("__iopen__ directory has no inode? rc = %d\n", rc);
681 dentry = simple_mkdir(current->fs->pwd, "PENDING", 0777, 1);
682 if (IS_ERR(dentry)) {
683 rc = PTR_ERR(dentry);
684 CERROR("cannot create PENDING directory: rc = %d\n", rc);
687 mds->mds_pending_dir = dentry;
689 dentry = simple_mkdir(current->fs->pwd, "LOGS", 0777, 1);
690 if (IS_ERR(dentry)) {
691 rc = PTR_ERR(dentry);
692 CERROR("cannot create LOGS directory: rc = %d\n", rc);
693 GOTO(err_pending, rc);
695 mds->mds_logs_dir = dentry;
697 dentry = simple_mkdir(current->fs->pwd, "OBJECTS", 0777, 1);
698 if (IS_ERR(dentry)) {
699 rc = PTR_ERR(dentry);
700 CERROR("cannot create OBJECTS directory: rc = %d\n", rc);
703 mds->mds_objects_dir = dentry;
705 dentry = simple_mkdir(current->fs->pwd, "FIDS", 0777, 1);
706 if (IS_ERR(dentry)) {
707 rc = PTR_ERR(dentry);
708 CERROR("cannot create FIDS directory: rc = %d\n", rc);
709 GOTO(err_objects, rc);
711 mds->mds_id_dir = dentry;
713 dentry = simple_mkdir(current->fs->pwd, "UNNAMED", 0777, 1);
714 if (IS_ERR(dentry)) {
715 rc = PTR_ERR(dentry);
716 CERROR("cannot create UNNAMED directory: rc = %d\n", rc);
717 GOTO(err_unnamed, rc);
719 mds->mds_unnamed_dir = dentry;
721 /* open and test the last rcvd file */
722 file = filp_open(LAST_RCVD, O_RDWR | O_CREAT, 0644);
725 CERROR("cannot open/create %s file: rc = %d\n", LAST_RCVD, rc);
726 GOTO(err_id_dir, rc = PTR_ERR(file));
728 mds->mds_rcvd_filp = file;
730 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
731 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
732 file->f_dentry->d_inode->i_mode);
733 GOTO(err_last_rcvd, rc = -ENOENT);
736 rc = mds_read_last_rcvd(obd, file);
738 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
739 GOTO(err_last_rcvd, rc);
742 /* open and test last fid file */
743 file = filp_open(LAST_FID, O_RDWR | O_CREAT, 0644);
746 CERROR("cannot open/create %s file: rc = %d\n",
748 GOTO(err_client, rc = PTR_ERR(file));
750 mds->mds_fid_filp = file;
751 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
752 CERROR("%s is not a regular file!: mode = %o\n",
753 LAST_FID, file->f_dentry->d_inode->i_mode);
754 GOTO(err_last_fid, rc = -ENOENT);
757 rc = mds_read_last_fid(obd, file);
759 CERROR("cannot read %s: rc = %d\n", LAST_FID, rc);
760 GOTO(err_last_fid, rc);
763 /* open and test virtid fid file */
764 file = filp_open(VIRT_FID, O_RDWR | O_CREAT, 0644);
767 CERROR("cannot open/create %s file: rc = %d\n",
769 GOTO(err_last_fid, rc = PTR_ERR(file));
771 mds->mds_virtid_filp = file;
772 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
773 CERROR("%s is not a regular file!: mode = %o\n",
774 VIRT_FID, file->f_dentry->d_inode->i_mode);
775 GOTO(err_virtid_fid, rc = -ENOENT);
778 rc = mds_read_virtid_fid(obd, file);
780 CERROR("cannot read %s: rc = %d\n", VIRT_FID, rc);
781 GOTO(err_virtid_fid, rc);
784 /* open and test the lov objid file */
785 file = filp_open(LOV_OBJID, O_RDWR | O_CREAT, 0644);
788 CERROR("cannot open/create %s file: rc = %d\n", LOV_OBJID, rc);
789 GOTO(err_last_fid, rc = PTR_ERR(file));
791 mds->mds_dt_objid_filp = file;
792 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
793 CERROR("%s is not a regular file!: mode = %o\n", LOV_OBJID,
794 file->f_dentry->d_inode->i_mode);
795 GOTO(err_lov_objid, rc = -ENOENT);
799 rc = mds_fs_post_setup(obd);
801 CERROR("can not post setup fsfilt\n");
803 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
807 if (mds->mds_dt_objid_filp && filp_close(mds->mds_dt_objid_filp, 0))
808 CERROR("can't close %s after error\n", LOV_OBJID);
810 if (mds->mds_virtid_filp && filp_close(mds->mds_virtid_filp, 0))
811 CERROR("can't close %s after error\n", VIRT_FID);
813 if (mds->mds_fid_filp && filp_close(mds->mds_fid_filp, 0))
814 CERROR("can't close %s after error\n", LAST_FID);
816 class_disconnect_exports(obd, 0);
818 if (mds->mds_rcvd_filp && filp_close(mds->mds_rcvd_filp, 0))
819 CERROR("can't close %s after error\n", LAST_RCVD);
821 dput(mds->mds_unnamed_dir);
823 dput(mds->mds_id_dir);
825 dput(mds->mds_objects_dir);
827 dput(mds->mds_logs_dir);
829 dput(mds->mds_pending_dir);
831 dput(mds->mds_id_de);
835 static int mds_fs_post_cleanup(struct obd_device *obd)
838 rc = fsfilt_post_cleanup(obd);
842 int mds_fs_cleanup(struct obd_device *obd, int flags)
844 struct mds_obd *mds = &obd->u.mds;
845 struct lvfs_run_ctxt saved;
848 if (flags & OBD_OPT_FAILOVER)
849 CERROR("%s: shutting down for failover; client state will"
850 " be preserved.\n", obd->obd_name);
852 class_disconnect_exports(obd, flags); /* cleans up client info too */
853 target_cleanup_recovery(obd);
854 mds_server_free_data(mds);
856 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
857 if (mds->mds_virtid_filp) {
858 rc = filp_close(mds->mds_virtid_filp, 0);
859 mds->mds_virtid_filp = NULL;
861 CERROR("%s file won't close, rc = %d\n", VIRT_FID, rc);
863 if (mds->mds_fid_filp) {
864 rc = filp_close(mds->mds_fid_filp, 0);
865 mds->mds_fid_filp = NULL;
867 CERROR("%s file won't close, rc = %d\n", LAST_FID, rc);
869 if (mds->mds_rcvd_filp) {
870 rc = filp_close(mds->mds_rcvd_filp, 0);
871 mds->mds_rcvd_filp = NULL;
873 CERROR("%s file won't close, rc = %d\n", LAST_RCVD, rc);
875 if (mds->mds_dt_objid_filp) {
876 rc = filp_close(mds->mds_dt_objid_filp, 0);
877 mds->mds_dt_objid_filp = NULL;
879 CERROR("%s file won't close, rc=%d\n", LOV_OBJID, rc);
881 if (mds->mds_unnamed_dir != NULL) {
882 l_dput(mds->mds_unnamed_dir);
883 mds->mds_unnamed_dir = NULL;
885 if (mds->mds_id_dir != NULL) {
886 l_dput(mds->mds_id_dir);
887 mds->mds_id_dir = NULL;
889 if (mds->mds_objects_dir != NULL) {
890 l_dput(mds->mds_objects_dir);
891 mds->mds_objects_dir = NULL;
893 if (mds->mds_logs_dir) {
894 l_dput(mds->mds_logs_dir);
895 mds->mds_logs_dir = NULL;
897 if (mds->mds_pending_dir) {
898 l_dput(mds->mds_pending_dir);
899 mds->mds_pending_dir = NULL;
901 rc = mds_fs_post_cleanup(obd);
903 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
904 shrink_dcache_parent(mds->mds_id_de);
905 dput(mds->mds_id_de);
910 /* Creates an object with the same name as its id. Because this is not at all
911 * performance sensitive, it is accomplished by creating a file, checking the
912 * id, and renaming it. */
913 int mds_obd_create(struct obd_export *exp, struct obdo *oa,
914 struct lov_stripe_md **ea, struct obd_trans_info *oti)
916 struct mds_obd *mds = &exp->exp_obd->u.mds;
917 struct inode *parent_inode = mds->mds_objects_dir->d_inode;
919 struct dentry *dchild;
920 struct lvfs_run_ctxt saved;
921 char idname[LL_ID_NAMELEN];
922 int rc = 0, err, idlen;
926 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
927 down(&parent_inode->i_sem);
929 idlen = ll_id2str(idname, oa->o_id, oa->o_generation);
930 dchild = lookup_one_len(idname, mds->mds_objects_dir, idlen);
932 GOTO(out_pop, rc = PTR_ERR(dchild));
934 if (dchild->d_inode == NULL) {
935 struct dentry_params dp;
938 CWARN("creating log with ID "LPU64"\n", oa->o_id);
940 dchild->d_fsdata = (void *) &dp;
942 dp.p_inum = oa->o_id;
943 rc = ll_vfs_create(parent_inode, dchild, S_IFREG, NULL);
944 if (dchild->d_fsdata == (void *)(unsigned long)oa->o_id)
945 dchild->d_fsdata = NULL;
947 CDEBUG(D_INODE, "err during create: %d\n", rc);
951 inode = dchild->d_inode;
952 LASSERT(inode->i_ino == oa->o_id);
953 inode->i_generation = oa->o_generation;
954 CDEBUG(D_HA, "recreated ino %lu with gen %u\n",
955 inode->i_ino, inode->i_generation);
956 mark_inode_dirty(inode);
958 CWARN("it should be here!\n");
963 sprintf(idname, "OBJECTS/%u.%u", ll_insecure_random_int(), current->pid);
964 filp = filp_open(idname, O_CREAT | O_EXCL, 0644);
968 CERROR("impossible object name collision %s\n",
972 CERROR("error creating tmp object %s: rc %d\n",
977 LASSERT(mds->mds_objects_dir == filp->f_dentry->d_parent);
979 oa->o_id = filp->f_dentry->d_inode->i_ino;
980 oa->o_generation = filp->f_dentry->d_inode->i_generation;
981 idlen = ll_id2str(idname, oa->o_id, oa->o_generation);
983 CWARN("created log anonymous "LPU64"/%u\n",
984 oa->o_id, oa->o_generation);
986 dchild = lookup_one_len(idname, mds->mds_objects_dir, idlen);
987 if (IS_ERR(dchild)) {
988 CERROR("getting neg dentry for obj rename: %d\n", rc);
989 GOTO(out_close, rc = PTR_ERR(dchild));
991 if (dchild->d_inode != NULL) {
992 CERROR("impossible non-negative obj dentry " LPU64":%u!\n",
993 oa->o_id, oa->o_generation);
997 handle = fsfilt_start(exp->exp_obd, mds->mds_objects_dir->d_inode,
998 FSFILT_OP_RENAME, NULL);
1000 GOTO(out_dput, rc = PTR_ERR(handle));
1003 rc = vfs_rename(mds->mds_objects_dir->d_inode, filp->f_dentry,
1004 mds->mds_objects_dir->d_inode, dchild);
1007 CERROR("error renaming new object "LPU64":%u: rc %d\n",
1008 oa->o_id, oa->o_generation, rc);
1010 err = fsfilt_commit(exp->exp_obd, mds->mds_sb,
1011 mds->mds_objects_dir->d_inode, handle, 0);
1013 oa->o_gr = FILTER_GROUP_FIRST_MDS + mds->mds_num;
1014 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGENER | OBD_MD_FLGROUP;
1020 err = filp_close(filp, 0);
1022 CERROR("closing tmpfile %s: rc %d\n", idname, rc);
1027 up(&parent_inode->i_sem);
1028 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
1032 int mds_obd_destroy(struct obd_export *exp, struct obdo *oa,
1033 struct lov_stripe_md *ea, struct obd_trans_info *oti)
1035 struct mds_obd *mds = &exp->exp_obd->u.mds;
1036 struct inode *parent_inode = mds->mds_objects_dir->d_inode;
1037 struct obd_device *obd = exp->exp_obd;
1038 struct lvfs_run_ctxt saved;
1039 char idname[LL_ID_NAMELEN];
1042 int err, idlen, rc = 0;
1045 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1047 idlen = ll_id2str(idname, oa->o_id, oa->o_generation);
1049 down(&parent_inode->i_sem);
1050 de = lookup_one_len(idname, mds->mds_objects_dir, idlen);
1051 if (IS_ERR(de) || de->d_inode == NULL) {
1052 rc = IS_ERR(de) ? PTR_ERR(de) : -ENOENT;
1053 CERROR("destroying non-existent object "LPU64" %s: rc %d\n",
1054 oa->o_id, idname, rc);
1057 /* Stripe count is 1 here since this is some MDS specific stuff
1058 that is unlinked, not spanned across multiple OSTs */
1059 handle = fsfilt_start_log(obd, mds->mds_objects_dir->d_inode,
1060 FSFILT_OP_UNLINK, oti, 1);
1063 GOTO(out_dput, rc = PTR_ERR(handle));
1065 rc = vfs_unlink(mds->mds_objects_dir->d_inode, de);
1067 CERROR("error destroying object "LPU64":%u: rc %d\n",
1068 oa->o_id, oa->o_generation, rc);
1070 err = fsfilt_commit(obd, mds->mds_sb, mds->mds_objects_dir->d_inode,
1077 up(&parent_inode->i_sem);
1078 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);