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;
134 mds_idmap_cleanup(med);
139 /* XXX if mcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
140 if (!strcmp(med->med_mcd->mcd_uuid, obd->obd_uuid.uuid))
141 GOTO(free_and_out, 0);
143 CDEBUG(D_INFO, "freeing client at idx %u (%lld)with UUID '%s'\n",
144 med->med_idx, med->med_off, med->med_mcd->mcd_uuid);
148 /* Clear the bit _after_ zeroing out the client so we don't
149 race with mds_client_add and zero out new clients.*/
150 if (!test_bit(med->med_idx, bitmap)) {
151 CERROR("MDS client %u: bit already clear in bitmap!!\n",
157 memset(&zero_mcd, 0, sizeof zero_mcd);
158 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
159 rc = fsfilt_write_record(obd, mds->mds_rcvd_filp, &zero_mcd,
160 sizeof(zero_mcd), &med->med_off, 1);
161 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
163 CDEBUG(rc == 0 ? D_INFO : D_ERROR,
164 "zeroing out client %s idx %u in %s rc %d\n",
165 med->med_mcd->mcd_uuid, med->med_idx, LAST_RCVD, rc);
168 if (!test_and_clear_bit(med->med_idx, bitmap)) {
169 CERROR("MDS client %u: bit already clear in bitmap!!\n",
175 /* Make sure the server's last_transno is up to date. Do this
176 * after the client is freed so we know all the client's
177 * transactions have been committed. */
178 mds_update_server_data(exp->exp_obd, 1);
181 OBD_FREE(med->med_mcd, sizeof(*med->med_mcd));
185 static int mds_server_free_data(struct mds_obd *mds)
187 OBD_FREE(mds->mds_client_bitmap,
188 MDS_MAX_CLIENT_WORDS * sizeof(unsigned long));
189 OBD_FREE(mds->mds_server_data, sizeof(*mds->mds_server_data));
190 mds->mds_server_data = NULL;
195 static int mds_read_last_fid(struct obd_device *obd, struct file *file)
199 struct mds_obd *mds = &obd->u.mds;
200 unsigned long last_fid_size = file->f_dentry->d_inode->i_size;
203 if (last_fid_size == 0) {
204 CWARN("%s: initializing new %s\n", obd->obd_name,
205 file->f_dentry->d_name.name);
208 * as fid is used for forming res_id for locking, it should not
209 * be zero. This will keep us out of lots possible problems,
212 mds_set_last_fid(obd, 0);
216 rc = fsfilt_read_record(obd, file, &lastfid,
217 sizeof(lastfid), &off);
219 CERROR("error reading MDS %s: rc = %d\n",
220 file->f_dentry->d_name.name, rc);
225 * make sure, that fid is up-to-date.
227 mds_set_last_fid(obd, lastfid);
230 CDEBUG(D_INODE, "%s: server last_fid: "LPU64"\n",
231 obd->obd_name, mds->mds_last_fid);
233 rc = mds_update_last_fid(obd, NULL, 1);
237 static int mds_read_last_rcvd(struct obd_device *obd, struct file *file)
239 unsigned long last_rcvd_size = file->f_dentry->d_inode->i_size;
240 struct mds_obd *mds = &obd->u.mds;
241 struct mds_server_data *msd = NULL;
242 struct mds_client_data *mcd = NULL;
248 /* ensure padding in the struct is the correct size */
249 LASSERT(offsetof(struct mds_server_data, msd_padding) +
250 sizeof(msd->msd_padding) == MDS_LR_SERVER_SIZE);
251 LASSERT(offsetof(struct mds_client_data, mcd_padding) +
252 sizeof(mcd->mcd_padding) == MDS_LR_CLIENT_SIZE);
254 OBD_ALLOC_WAIT(msd, sizeof(*msd));
258 OBD_ALLOC_WAIT(mds->mds_client_bitmap,
259 MDS_MAX_CLIENT_WORDS * sizeof(unsigned long));
260 if (!mds->mds_client_bitmap) {
261 OBD_FREE(msd, sizeof(*msd));
265 mds->mds_server_data = msd;
267 if (last_rcvd_size == 0) {
268 CWARN("%s: initializing new %s\n", obd->obd_name,
269 file->f_dentry->d_name.name);
271 memcpy(msd->msd_uuid, obd->obd_uuid.uuid,sizeof(msd->msd_uuid));
272 msd->msd_last_transno = 0;
273 mount_count = msd->msd_mount_count = 0;
274 msd->msd_server_size = cpu_to_le32(MDS_LR_SERVER_SIZE);
275 msd->msd_client_start = cpu_to_le32(MDS_LR_CLIENT_START);
276 msd->msd_client_size = cpu_to_le16(MDS_LR_CLIENT_SIZE);
277 msd->msd_feature_rocompat = cpu_to_le32(MDS_ROCOMPAT_LOVOBJID);
279 rc = fsfilt_read_record(obd, file, msd, sizeof(*msd), &off);
281 CERROR("error reading MDS %s: rc = %d\n",
282 file->f_dentry->d_name.name, rc);
285 if (strcmp(msd->msd_uuid, obd->obd_uuid.uuid) != 0) {
286 CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
287 obd->obd_uuid.uuid, msd->msd_uuid);
288 GOTO(err_msd, rc = -EINVAL);
290 mount_count = le64_to_cpu(msd->msd_mount_count);
292 if (msd->msd_feature_incompat & ~cpu_to_le32(MDS_INCOMPAT_SUPP)) {
293 CERROR("unsupported incompat feature %x\n",
294 le32_to_cpu(msd->msd_feature_incompat) &
296 GOTO(err_msd, rc = -EINVAL);
298 /* XXX updating existing b_devel fs only, can be removed in future */
299 msd->msd_feature_rocompat = cpu_to_le32(MDS_ROCOMPAT_LOVOBJID);
300 if (msd->msd_feature_rocompat & ~cpu_to_le32(MDS_ROCOMPAT_SUPP)) {
301 CERROR("unsupported read-only feature %x\n",
302 le32_to_cpu(msd->msd_feature_rocompat) &
304 /* Do something like remount filesystem read-only */
305 GOTO(err_msd, rc = -EINVAL);
308 mds->mds_last_transno = le64_to_cpu(msd->msd_last_transno);
310 CDEBUG(D_INODE, "%s: server last_transno: "LPU64"\n",
311 obd->obd_name, mds->mds_last_transno);
312 CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
313 obd->obd_name, mount_count + 1);
314 CDEBUG(D_INODE, "%s: server data size: %u\n",
315 obd->obd_name, le32_to_cpu(msd->msd_server_size));
316 CDEBUG(D_INODE, "%s: per-client data start: %u\n",
317 obd->obd_name, le32_to_cpu(msd->msd_client_start));
318 CDEBUG(D_INODE, "%s: per-client data size: %u\n",
319 obd->obd_name, le32_to_cpu(msd->msd_client_size));
320 CDEBUG(D_INODE, "%s: last_rcvd size: %lu\n",
321 obd->obd_name, last_rcvd_size);
322 CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
323 last_rcvd_size <= le32_to_cpu(msd->msd_client_start) ? 0 :
324 (last_rcvd_size - le32_to_cpu(msd->msd_client_start)) /
325 le16_to_cpu(msd->msd_client_size));
327 /* When we do a clean MDS shutdown, we save the last_transno into
328 * the header. If we find clients with higher last_transno values
329 * then those clients may need recovery done. */
330 for (cl_idx = 0, off = le32_to_cpu(msd->msd_client_start);
331 off < last_rcvd_size; cl_idx++) {
333 struct obd_export *exp;
334 struct mds_export_data *med;
337 OBD_ALLOC_WAIT(mcd, sizeof(*mcd));
339 GOTO(err_client, rc = -ENOMEM);
342 /* Don't assume off is incremented properly by
343 * fsfilt_read_record(), in case sizeof(*mcd)
344 * isn't the same as msd->msd_client_size. */
345 off = le32_to_cpu(msd->msd_client_start) +
346 cl_idx * le16_to_cpu(msd->msd_client_size);
347 rc = fsfilt_read_record(obd, file, mcd, sizeof(*mcd), &off);
349 CERROR("error reading MDS %s idx %d, off %llu: rc %d\n",
350 file->f_dentry->d_name.name, cl_idx, off, rc);
351 break; /* read error shouldn't cause startup to fail */
354 if (mcd->mcd_uuid[0] == '\0') {
355 CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
360 last_transno = le64_to_cpu(mcd->mcd_last_transno);
362 /* These exports are cleaned up by mds_disconnect(), so they
363 * need to be set up like real exports as mds_connect() does.
365 CDEBUG(D_HA|D_WARNING,"RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
366 " srv lr: "LPU64" lx: "LPU64"\n", mcd->mcd_uuid, cl_idx,
367 last_transno, le64_to_cpu(msd->msd_last_transno),
370 exp = class_new_export(obd);
372 GOTO(err_client, rc = -ENOMEM);
374 memcpy(&exp->exp_client_uuid.uuid, mcd->mcd_uuid,
375 sizeof exp->exp_client_uuid.uuid);
376 med = &exp->exp_mds_data;
378 mds_client_add(obd, mds, med, cl_idx);
379 /* create helper if export init gets more complex */
380 INIT_LIST_HEAD(&med->med_open_head);
381 spin_lock_init(&med->med_open_lock);
384 exp->exp_replay_needed = 1;
385 obd->obd_recoverable_clients++;
386 obd->obd_max_recoverable_clients++;
387 class_export_put(exp);
389 CDEBUG(D_OTHER, "client at idx %d has last_transno = "LPU64"\n",
390 cl_idx, last_transno);
392 if (last_transno > mds->mds_last_transno)
393 mds->mds_last_transno = last_transno;
396 OBD_FREE(mcd, sizeof(*mcd));
397 obd->obd_last_committed = mds->mds_last_transno;
398 if (obd->obd_recoverable_clients) {
399 CWARN("RECOVERY: service %s, %d recoverable clients, "
400 "last_transno "LPU64"\n", obd->obd_name,
401 obd->obd_recoverable_clients, mds->mds_last_transno);
402 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
403 target_start_recovery_thread(obd, mds_handle);
404 obd->obd_recovery_start = LTIME_S(CURRENT_TIME);
407 mds->mds_mount_count = mount_count + 1;
408 msd->msd_mount_count = cpu_to_le64(mds->mds_mount_count);
410 /* save it, so mount count and last_transno is current */
411 rc = mds_update_server_data(obd, 1);
413 GOTO(err_client, rc);
418 class_disconnect_exports(obd, 0);
420 mds_server_free_data(mds);
424 static int mds_fs_post_setup(struct obd_device *obd)
426 struct mds_obd *mds = &obd->u.mds;
427 struct dentry *dentry;
431 dentry = mds_id2dentry(obd, &mds->mds_rootid, NULL);
432 if (IS_ERR(dentry)) {
433 CERROR("Can't find ROOT, err = %d\n",
434 (int)PTR_ERR(dentry));
435 RETURN(PTR_ERR(dentry));
438 rc = fsfilt_post_setup(obd, dentry);
442 LASSERT(dentry->d_inode != NULL);
444 fsfilt_set_fs_flags(obd, dentry->d_inode,
445 SM_DO_REC | SM_DO_COW);
447 fsfilt_set_fs_flags(obd, mds->mds_pending_dir->d_inode,
448 SM_DO_REC | SM_DO_COW);
450 fsfilt_set_mds_flags(obd, mds->mds_sb);
458 * sets up root inode lustre_id. It tries to read it first from root inode and
459 * if it is not there, new rootid is allocated and saved there.
461 int mds_fs_setup_rootid(struct obd_device *obd)
466 struct dentry *dentry;
467 struct mds_obd *mds = &obd->u.mds;
470 /* getting root directory and setup its fid. */
471 dentry = mds_id2dentry(obd, &mds->mds_rootid, NULL);
472 if (IS_ERR(dentry)) {
473 CERROR("Can't find ROOT by "DLID4", err = %d\n",
474 OLID4(&mds->mds_rootid), (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, &mds->mds_rootid, inode, 1);
484 GOTO(out_dentry, rc);
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(&mds->mds_rootid) == mds->mds_num);
492 mds_set_last_fid(obd, id_fid(&mds->mds_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, &mds->mds_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);
524 CWARN("%s: rootid: "DLID4"\n", obd->obd_name,
525 OLID4(&mds->mds_rootid));
529 static int mds_update_virtid_fid(struct obd_device *obd,
530 void *handle, int force_sync)
532 struct mds_obd *mds = &obd->u.mds;
533 struct file *filp = mds->mds_virtid_filp;
534 struct lvfs_run_ctxt saved;
539 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
540 rc = fsfilt_write_record(obd, filp, &mds->mds_virtid_fid,
541 sizeof(mds->mds_virtid_fid),
544 CERROR("error writing MDS virtid_fid #"LPU64
545 ", err = %d\n", mds->mds_virtid_fid, rc);
548 CDEBUG(D_SUPER, "wrote virtid fid #"LPU64" at idx "
549 "%llu: err = %d\n", mds->mds_virtid_fid,
551 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
556 static int mds_read_virtid_fid(struct obd_device *obd,
561 struct mds_obd *mds = &obd->u.mds;
562 unsigned long virtid_fid_size = file->f_dentry->d_inode->i_size;
565 if (virtid_fid_size == 0) {
566 mds->mds_virtid_fid = mds_alloc_fid(obd);
568 rc = fsfilt_read_record(obd, file, &mds->mds_virtid_fid,
569 sizeof(mds->mds_virtid_fid), &off);
571 CERROR("error reading MDS %s: rc = %d\n",
572 file->f_dentry->d_name.name, rc);
576 rc = mds_update_virtid_fid(obd, NULL, 1);
582 * initializes lustre_id for virtual id directory, it is needed sometimes, as it
583 * is possible that it will be the parent for object an operations is going to
586 int mds_fs_setup_virtid(struct obd_device *obd)
590 struct lustre_id sid;
591 struct mds_obd *mds = &obd->u.mds;
592 struct inode *inode = mds->mds_id_dir->d_inode;
595 handle = fsfilt_start(obd, inode,
596 FSFILT_OP_SETATTR, NULL);
598 if (IS_ERR(handle)) {
599 rc = PTR_ERR(handle);
600 CERROR("fsfilt_start() failed, rc = %d\n", rc);
604 id_group(&sid) = mds->mds_num;
605 id_fid(&sid) = mds->mds_virtid_fid;
607 id_ino(&sid) = inode->i_ino;
608 id_gen(&sid) = inode->i_generation;
609 id_type(&sid) = (S_IFMT & inode->i_mode);
612 rc = mds_update_inode_sid(obd, inode, handle, &sid);
616 CERROR("mds_update_inode_sid() failed, rc = %d\n",
621 rc = fsfilt_commit(obd, mds->mds_sb, inode, handle, 0);
623 CERROR("fsfilt_commit() failed, rc = %d\n", rc);
630 int mds_fs_setup(struct obd_device *obd, struct vfsmount *mnt)
632 struct mds_obd *mds = &obd->u.mds;
633 struct lvfs_run_ctxt saved;
634 struct dentry *dentry;
639 rc = cleanup_group_info();
643 mds->mds_vfsmnt = mnt;
644 mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
646 fsfilt_setup(obd, mds->mds_sb);
648 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
649 obd->obd_lvfs_ctxt.pwdmnt = mnt;
650 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
651 obd->obd_lvfs_ctxt.fs = get_ds();
652 obd->obd_lvfs_ctxt.cb_ops = mds_lvfs_ops;
654 /* setup the directory tree */
655 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
656 dentry = simple_mkdir(current->fs->pwd, "ROOT", 0755, 0);
657 if (IS_ERR(dentry)) {
658 rc = PTR_ERR(dentry);
659 CERROR("cannot create ROOT directory: rc = %d\n", rc);
663 mdc_pack_id(&mds->mds_rootid, dentry->d_inode->i_ino,
664 dentry->d_inode->i_generation, S_IFDIR, 0, 0);
668 dentry = lookup_one_len("__iopen__", current->fs->pwd,
669 strlen("__iopen__"));
670 if (IS_ERR(dentry)) {
671 rc = PTR_ERR(dentry);
672 CERROR("cannot lookup __iopen__ directory: rc = %d\n", rc);
675 mds->mds_id_de = dentry;
677 if (!dentry->d_inode) {
679 CERROR("__iopen__ directory has no inode? rc = %d\n", rc);
683 dentry = simple_mkdir(current->fs->pwd, "PENDING", 0777, 1);
684 if (IS_ERR(dentry)) {
685 rc = PTR_ERR(dentry);
686 CERROR("cannot create PENDING directory: rc = %d\n", rc);
689 mds->mds_pending_dir = dentry;
691 dentry = simple_mkdir(current->fs->pwd, "LOGS", 0777, 1);
692 if (IS_ERR(dentry)) {
693 rc = PTR_ERR(dentry);
694 CERROR("cannot create LOGS directory: rc = %d\n", rc);
695 GOTO(err_pending, rc);
697 mds->mds_logs_dir = dentry;
699 dentry = simple_mkdir(current->fs->pwd, "OBJECTS", 0777, 1);
700 if (IS_ERR(dentry)) {
701 rc = PTR_ERR(dentry);
702 CERROR("cannot create OBJECTS directory: rc = %d\n", rc);
705 mds->mds_objects_dir = dentry;
707 dentry = simple_mkdir(current->fs->pwd, "FIDS", 0777, 1);
708 if (IS_ERR(dentry)) {
709 rc = PTR_ERR(dentry);
710 CERROR("cannot create FIDS directory: rc = %d\n", rc);
711 GOTO(err_objects, rc);
713 mds->mds_id_dir = dentry;
715 dentry = simple_mkdir(current->fs->pwd, "UNNAMED", 0777, 1);
716 if (IS_ERR(dentry)) {
717 rc = PTR_ERR(dentry);
718 CERROR("cannot create UNNAMED directory: rc = %d\n", rc);
719 GOTO(err_unnamed, rc);
721 mds->mds_unnamed_dir = dentry;
723 /* open and test the last rcvd file */
724 file = filp_open(LAST_RCVD, O_RDWR | O_CREAT, 0644);
727 CERROR("cannot open/create %s file: rc = %d\n", LAST_RCVD, rc);
728 GOTO(err_id_dir, rc = PTR_ERR(file));
730 mds->mds_rcvd_filp = file;
732 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
733 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
734 file->f_dentry->d_inode->i_mode);
735 GOTO(err_last_rcvd, rc = -ENOENT);
738 rc = mds_read_last_rcvd(obd, file);
740 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
741 GOTO(err_last_rcvd, rc);
744 /* open and test last fid file */
745 file = filp_open(LAST_FID, O_RDWR | O_CREAT, 0644);
748 CERROR("cannot open/create %s file: rc = %d\n",
750 GOTO(err_client, rc = PTR_ERR(file));
752 mds->mds_fid_filp = file;
753 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
754 CERROR("%s is not a regular file!: mode = %o\n",
755 LAST_FID, file->f_dentry->d_inode->i_mode);
756 GOTO(err_last_fid, rc = -ENOENT);
759 rc = mds_read_last_fid(obd, file);
761 CERROR("cannot read %s: rc = %d\n", LAST_FID, rc);
762 GOTO(err_last_fid, rc);
765 /* open and test virtid fid file */
766 file = filp_open(VIRT_FID, O_RDWR | O_CREAT, 0644);
769 CERROR("cannot open/create %s file: rc = %d\n",
771 GOTO(err_last_fid, rc = PTR_ERR(file));
773 mds->mds_virtid_filp = file;
774 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
775 CERROR("%s is not a regular file!: mode = %o\n",
776 VIRT_FID, file->f_dentry->d_inode->i_mode);
777 GOTO(err_virtid_fid, rc = -ENOENT);
780 rc = mds_read_virtid_fid(obd, file);
782 CERROR("cannot read %s: rc = %d\n", VIRT_FID, rc);
783 GOTO(err_virtid_fid, rc);
786 /* open and test the lov objid file */
787 file = filp_open(LOV_OBJID, O_RDWR | O_CREAT, 0644);
790 CERROR("cannot open/create %s file: rc = %d\n", LOV_OBJID, rc);
791 GOTO(err_last_fid, rc = PTR_ERR(file));
793 mds->mds_dt_objid_filp = file;
794 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
795 CERROR("%s is not a regular file!: mode = %o\n", LOV_OBJID,
796 file->f_dentry->d_inode->i_mode);
797 GOTO(err_lov_objid, rc = -ENOENT);
801 rc = mds_fs_post_setup(obd);
803 CERROR("can not post setup fsfilt\n");
805 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
809 if (mds->mds_dt_objid_filp && filp_close(mds->mds_dt_objid_filp, 0))
810 CERROR("can't close %s after error\n", LOV_OBJID);
812 if (mds->mds_virtid_filp && filp_close(mds->mds_virtid_filp, 0))
813 CERROR("can't close %s after error\n", VIRT_FID);
815 if (mds->mds_fid_filp && filp_close(mds->mds_fid_filp, 0))
816 CERROR("can't close %s after error\n", LAST_FID);
818 class_disconnect_exports(obd, 0);
820 if (mds->mds_rcvd_filp && filp_close(mds->mds_rcvd_filp, 0))
821 CERROR("can't close %s after error\n", LAST_RCVD);
823 dput(mds->mds_unnamed_dir);
825 dput(mds->mds_id_dir);
827 dput(mds->mds_objects_dir);
829 dput(mds->mds_logs_dir);
831 dput(mds->mds_pending_dir);
833 dput(mds->mds_id_de);
837 static int mds_fs_post_cleanup(struct obd_device *obd)
840 rc = fsfilt_post_cleanup(obd);
844 int mds_fs_cleanup(struct obd_device *obd, int flags)
846 struct mds_obd *mds = &obd->u.mds;
847 struct lvfs_run_ctxt saved;
850 if (flags & OBD_OPT_FAILOVER)
851 CERROR("%s: shutting down for failover; client state will"
852 " be preserved.\n", obd->obd_name);
854 class_disconnect_exports(obd, flags); /* cleans up client info too */
855 target_cleanup_recovery(obd);
856 mds_server_free_data(mds);
858 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
859 if (mds->mds_virtid_filp) {
860 rc = filp_close(mds->mds_virtid_filp, 0);
861 mds->mds_virtid_filp = NULL;
863 CERROR("%s file won't close, rc = %d\n", VIRT_FID, rc);
865 if (mds->mds_fid_filp) {
866 rc = filp_close(mds->mds_fid_filp, 0);
867 mds->mds_fid_filp = NULL;
869 CERROR("%s file won't close, rc = %d\n", LAST_FID, rc);
871 if (mds->mds_rcvd_filp) {
872 rc = filp_close(mds->mds_rcvd_filp, 0);
873 mds->mds_rcvd_filp = NULL;
875 CERROR("%s file won't close, rc = %d\n", LAST_RCVD, rc);
877 if (mds->mds_dt_objid_filp) {
878 rc = filp_close(mds->mds_dt_objid_filp, 0);
879 mds->mds_dt_objid_filp = NULL;
881 CERROR("%s file won't close, rc=%d\n", LOV_OBJID, rc);
883 if (mds->mds_unnamed_dir != NULL) {
884 l_dput(mds->mds_unnamed_dir);
885 mds->mds_unnamed_dir = NULL;
887 if (mds->mds_id_dir != NULL) {
888 l_dput(mds->mds_id_dir);
889 mds->mds_id_dir = NULL;
891 if (mds->mds_objects_dir != NULL) {
892 l_dput(mds->mds_objects_dir);
893 mds->mds_objects_dir = NULL;
895 if (mds->mds_logs_dir) {
896 l_dput(mds->mds_logs_dir);
897 mds->mds_logs_dir = NULL;
899 if (mds->mds_pending_dir) {
900 l_dput(mds->mds_pending_dir);
901 mds->mds_pending_dir = NULL;
903 rc = mds_fs_post_cleanup(obd);
905 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
906 shrink_dcache_parent(mds->mds_id_de);
907 dput(mds->mds_id_de);
912 /* Creates an object with the same name as its id. Because this is not at all
913 * performance sensitive, it is accomplished by creating a file, checking the
914 * id, and renaming it. */
915 int mds_obd_create(struct obd_export *exp, struct obdo *oa,
916 struct lov_stripe_md **ea, struct obd_trans_info *oti)
918 struct mds_obd *mds = &exp->exp_obd->u.mds;
919 struct inode *parent_inode = mds->mds_objects_dir->d_inode;
921 struct dentry *dchild;
922 struct lvfs_run_ctxt saved;
923 char idname[LL_ID_NAMELEN];
924 int rc = 0, err, idlen;
928 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
929 down(&parent_inode->i_sem);
931 idlen = ll_id2str(idname, oa->o_id, oa->o_generation);
932 dchild = lookup_one_len(idname, mds->mds_objects_dir, idlen);
934 GOTO(out_pop, rc = PTR_ERR(dchild));
936 if (dchild->d_inode == NULL) {
937 struct dentry_params dp;
940 CWARN("creating log with ID "LPU64"\n", oa->o_id);
942 dchild->d_fsdata = (void *) &dp;
944 dp.p_inum = oa->o_id;
945 rc = ll_vfs_create(parent_inode, dchild, S_IFREG, NULL);
946 if (dchild->d_fsdata == (void *)(unsigned long)oa->o_id)
947 dchild->d_fsdata = NULL;
949 CDEBUG(D_INODE, "err during create: %d\n", rc);
953 inode = dchild->d_inode;
954 LASSERT(inode->i_ino == oa->o_id);
955 inode->i_generation = oa->o_generation;
956 CDEBUG(D_HA, "recreated ino %lu with gen %u\n",
957 inode->i_ino, inode->i_generation);
958 mark_inode_dirty(inode);
960 CWARN("it should be here!\n");
965 sprintf(idname, "OBJECTS/%u.%u", ll_insecure_random_int(), current->pid);
966 filp = filp_open(idname, O_CREAT | O_EXCL, 0644);
970 CERROR("impossible object name collision %s\n",
974 CERROR("error creating tmp object %s: rc %d\n",
979 LASSERT(mds->mds_objects_dir == filp->f_dentry->d_parent);
981 oa->o_id = filp->f_dentry->d_inode->i_ino;
982 oa->o_generation = filp->f_dentry->d_inode->i_generation;
983 idlen = ll_id2str(idname, oa->o_id, oa->o_generation);
985 CWARN("created log anonymous "LPU64"/%u\n",
986 oa->o_id, oa->o_generation);
988 dchild = lookup_one_len(idname, mds->mds_objects_dir, idlen);
989 if (IS_ERR(dchild)) {
990 CERROR("getting neg dentry for obj rename: %d\n", rc);
991 GOTO(out_close, rc = PTR_ERR(dchild));
993 if (dchild->d_inode != NULL) {
994 CERROR("impossible non-negative obj dentry " LPU64":%u!\n",
995 oa->o_id, oa->o_generation);
999 handle = fsfilt_start(exp->exp_obd, mds->mds_objects_dir->d_inode,
1000 FSFILT_OP_RENAME, NULL);
1002 GOTO(out_dput, rc = PTR_ERR(handle));
1005 rc = vfs_rename(mds->mds_objects_dir->d_inode, filp->f_dentry,
1006 mds->mds_objects_dir->d_inode, dchild);
1009 CERROR("error renaming new object "LPU64":%u: rc %d\n",
1010 oa->o_id, oa->o_generation, rc);
1012 err = fsfilt_commit(exp->exp_obd, mds->mds_sb,
1013 mds->mds_objects_dir->d_inode, handle, 0);
1015 oa->o_gr = FILTER_GROUP_FIRST_MDS + mds->mds_num;
1016 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGENER | OBD_MD_FLGROUP;
1022 err = filp_close(filp, 0);
1024 CERROR("closing tmpfile %s: rc %d\n", idname, rc);
1029 up(&parent_inode->i_sem);
1030 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
1034 int mds_obd_destroy(struct obd_export *exp, struct obdo *oa,
1035 struct lov_stripe_md *ea, struct obd_trans_info *oti)
1037 struct mds_obd *mds = &exp->exp_obd->u.mds;
1038 struct inode *parent_inode = mds->mds_objects_dir->d_inode;
1039 struct obd_device *obd = exp->exp_obd;
1040 struct lvfs_run_ctxt saved;
1041 char idname[LL_ID_NAMELEN];
1044 int err, idlen, rc = 0;
1047 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1049 idlen = ll_id2str(idname, oa->o_id, oa->o_generation);
1051 down(&parent_inode->i_sem);
1052 de = lookup_one_len(idname, mds->mds_objects_dir, idlen);
1053 if (IS_ERR(de) || de->d_inode == NULL) {
1054 rc = IS_ERR(de) ? PTR_ERR(de) : -ENOENT;
1055 CERROR("destroying non-existent object "LPU64" %s: rc %d\n",
1056 oa->o_id, idname, rc);
1059 /* Stripe count is 1 here since this is some MDS specific stuff
1060 that is unlinked, not spanned across multiple OSTs */
1061 handle = fsfilt_start_log(obd, mds->mds_objects_dir->d_inode,
1062 FSFILT_OP_UNLINK, oti, 1);
1065 GOTO(out_dput, rc = PTR_ERR(handle));
1067 rc = vfs_unlink(mds->mds_objects_dir->d_inode, de);
1069 CERROR("error destroying object "LPU64":%u: rc %d\n",
1070 oa->o_id, oa->o_generation, rc);
1072 err = fsfilt_commit(obd, mds->mds_sb, mds->mds_objects_dir->d_inode,
1079 up(&parent_inode->i_sem);
1080 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);