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 the Lustre file system, http://www.lustre.org
11 * Lustre is a trademark of Cluster File Systems, Inc.
13 * You may have signed or agreed to another license before downloading
14 * this software. If so, you are bound by the terms and conditions
15 * of that agreement, and the following does not apply to you. See the
16 * LICENSE file included with this distribution for more information.
18 * If you did not agree to a different license, then this copy of Lustre
19 * is open source software; you can redistribute it and/or modify it
20 * under the terms of version 2 of the GNU General Public License as
21 * published by the Free Software Foundation.
23 * In either case, Lustre is distributed in the hope that it will be
24 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
25 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
26 * license text for more details.
30 # define EXPORT_SYMTAB
32 #define DEBUG_SUBSYSTEM S_MDS
34 #include <linux/module.h>
35 #include <linux/kmod.h>
36 #include <linux/version.h>
37 #include <linux/sched.h>
38 #include <linux/lustre_quota.h>
39 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
40 #include <linux/mount.h>
42 #include <linux/lustre_mds.h>
43 #include <linux/obd_class.h>
44 #include <linux/obd_support.h>
45 #include <linux/lustre_lib.h>
46 #include <linux/lustre_fsfilt.h>
47 #include <linux/lustre_disk.h>
48 #include <libcfs/list.h>
50 #include "mds_internal.h"
52 #define HEALTH_CHECK "health_check"
54 /* Add client data to the MDS. We use a bitmap to locate a free space
55 * in the last_rcvd file if cl_off is -1 (i.e. a new client).
56 * Otherwise, we have just read the data from the last_rcvd file and
59 * It should not be possible to fail adding an existing client - otherwise
60 * mds_init_server_data() callsite needs to be fixed.
62 int mds_client_add(struct obd_device *obd, struct mds_obd *mds,
63 struct mds_export_data *med, int cl_idx)
65 unsigned long *bitmap = mds->mds_client_bitmap;
66 int new_client = (cl_idx == -1);
69 LASSERT(bitmap != NULL);
70 LASSERTF(cl_idx > -2, "%d\n", cl_idx);
72 /* XXX if mcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
73 if (!strcmp(med->med_mcd->mcd_uuid, obd->obd_uuid.uuid))
76 /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
77 * there's no need for extra complication here
80 cl_idx = find_first_zero_bit(bitmap, LR_MAX_CLIENTS);
82 if (cl_idx >= LR_MAX_CLIENTS ||
83 OBD_FAIL_CHECK_ONCE(OBD_FAIL_MDS_CLIENT_ADD)) {
84 CERROR("no room for clients - fix LR_MAX_CLIENTS\n");
87 if (test_and_set_bit(cl_idx, bitmap)) {
88 cl_idx = find_next_zero_bit(bitmap, LR_MAX_CLIENTS,
93 if (test_and_set_bit(cl_idx, bitmap)) {
94 CERROR("MDS client %d: bit already set in bitmap!!\n",
100 CDEBUG(D_INFO, "client at idx %d with UUID '%s' added\n",
101 cl_idx, med->med_mcd->mcd_uuid);
103 med->med_lr_idx = cl_idx;
104 med->med_lr_off = le32_to_cpu(mds->mds_server_data->lsd_client_start) +
105 (cl_idx * le16_to_cpu(mds->mds_server_data->lsd_client_size));
106 LASSERTF(med->med_lr_off > 0, "med_lr_off = %llu\n", med->med_lr_off);
109 struct lvfs_run_ctxt saved;
110 loff_t off = med->med_lr_off;
111 struct file *file = mds->mds_rcvd_filp;
114 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
115 rc = fsfilt_write_record(obd, file, med->med_mcd,
116 sizeof(*med->med_mcd), &off, 1);
117 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
121 CDEBUG(D_INFO, "wrote client mcd at idx %u off %llu (len %u)\n",
122 med->med_lr_idx, med->med_lr_off,
123 (unsigned int)sizeof(*med->med_mcd));
128 int mds_client_free(struct obd_export *exp)
130 struct mds_export_data *med = &exp->exp_mds_data;
131 struct mds_obd *mds = &exp->exp_obd->u.mds;
132 struct obd_device *obd = exp->exp_obd;
133 struct mds_client_data zero_mcd;
134 struct lvfs_run_ctxt saved;
142 /* XXX if mcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
143 if (!strcmp(med->med_mcd->mcd_uuid, obd->obd_uuid.uuid))
146 CDEBUG(D_INFO, "freeing client at idx %u, offset %lld with UUID '%s'\n",
147 med->med_lr_idx, med->med_lr_off, med->med_mcd->mcd_uuid);
149 LASSERT(mds->mds_client_bitmap != NULL);
151 off = med->med_lr_off;
153 /* Don't clear med_lr_idx here as it is likely also unset. At worst
154 * we leak a client slot that will be cleaned on the next recovery. */
156 CERROR("%s: client idx %d has offset %lld\n",
157 obd->obd_name, med->med_lr_idx, off);
158 GOTO(free, rc = -EINVAL);
161 /* Clear the bit _after_ zeroing out the client so we don't
162 race with mds_client_add and zero out new clients.*/
163 if (!test_bit(med->med_lr_idx, mds->mds_client_bitmap)) {
164 CERROR("MDS client %u: bit already clear in bitmap!!\n",
169 if (!(exp->exp_flags & OBD_OPT_FAILOVER)) {
170 memset(&zero_mcd, 0, sizeof zero_mcd);
171 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
172 rc = fsfilt_write_record(obd, mds->mds_rcvd_filp, &zero_mcd,
173 sizeof(zero_mcd), &off, 1);
174 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
176 CDEBUG(rc == 0 ? D_INFO : D_ERROR,
177 "zeroing out client %s idx %u in %s rc %d\n",
178 med->med_mcd->mcd_uuid, med->med_lr_idx, LAST_RCVD, rc);
181 if (!test_and_clear_bit(med->med_lr_idx, mds->mds_client_bitmap)) {
182 CERROR("MDS client %u: bit already clear in bitmap!!\n",
188 /* Make sure the server's last_transno is up to date. Do this
189 * after the client is freed so we know all the client's
190 * transactions have been committed. */
191 mds_update_server_data(exp->exp_obd, 0);
195 OBD_FREE(med->med_mcd, sizeof(*med->med_mcd));
201 static int mds_server_free_data(struct mds_obd *mds)
203 OBD_FREE(mds->mds_client_bitmap, LR_MAX_CLIENTS / 8);
204 OBD_FREE(mds->mds_server_data, sizeof(*mds->mds_server_data));
205 mds->mds_server_data = NULL;
210 static int mds_init_server_data(struct obd_device *obd, struct file *file)
212 struct mds_obd *mds = &obd->u.mds;
213 struct lr_server_data *lsd;
214 struct mds_client_data *mcd = NULL;
216 unsigned long last_rcvd_size = file->f_dentry->d_inode->i_size;
221 /* ensure padding in the struct is the correct size */
222 LASSERT(offsetof(struct lr_server_data, lsd_padding) +
223 sizeof(lsd->lsd_padding) == LR_SERVER_SIZE);
224 LASSERT(offsetof(struct mds_client_data, mcd_padding) +
225 sizeof(mcd->mcd_padding) == LR_CLIENT_SIZE);
227 OBD_ALLOC_WAIT(lsd, sizeof(*lsd));
231 OBD_ALLOC_WAIT(mds->mds_client_bitmap, LR_MAX_CLIENTS / 8);
232 if (!mds->mds_client_bitmap) {
233 OBD_FREE(lsd, sizeof(*lsd));
237 mds->mds_server_data = lsd;
239 if (last_rcvd_size == 0) {
240 CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD);
242 memcpy(lsd->lsd_uuid, obd->obd_uuid.uuid,sizeof(lsd->lsd_uuid));
243 lsd->lsd_last_transno = 0;
244 mount_count = lsd->lsd_mount_count = 0;
245 lsd->lsd_server_size = cpu_to_le32(LR_SERVER_SIZE);
246 lsd->lsd_client_start = cpu_to_le32(LR_CLIENT_START);
247 lsd->lsd_client_size = cpu_to_le16(LR_CLIENT_SIZE);
248 lsd->lsd_feature_rocompat = cpu_to_le32(OBD_ROCOMPAT_LOVOBJID);
249 lsd->lsd_feature_incompat = cpu_to_le32(OBD_INCOMPAT_MDT |
250 OBD_INCOMPAT_COMMON_LR);
252 rc = fsfilt_read_record(obd, file, lsd, sizeof(*lsd), &off);
254 CERROR("error reading MDS %s: rc = %d\n", LAST_RCVD, rc);
257 if (strcmp(lsd->lsd_uuid, obd->obd_uuid.uuid) != 0) {
258 CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
259 obd->obd_uuid.uuid, lsd->lsd_uuid);
260 GOTO(err_msd, rc = -EINVAL);
262 mount_count = le64_to_cpu(lsd->lsd_mount_count);
264 if (lsd->lsd_feature_incompat & ~cpu_to_le32(MDT_INCOMPAT_SUPP)) {
265 CERROR("%s: unsupported incompat filesystem feature(s) %x\n",
266 obd->obd_name, le32_to_cpu(lsd->lsd_feature_incompat) &
268 GOTO(err_msd, rc = -EINVAL);
271 if (lsd->lsd_feature_rocompat & ~cpu_to_le32(MDT_ROCOMPAT_SUPP)) {
272 CERROR("%s: unsupported read-only filesystem feature(s) %x\n",
273 obd->obd_name, le32_to_cpu(lsd->lsd_feature_rocompat) &
275 /* Do something like remount filesystem read-only */
276 GOTO(err_msd, rc = -EINVAL);
279 if (!(lsd->lsd_feature_incompat & cpu_to_le32(OBD_INCOMPAT_COMMON_LR))){
280 CDEBUG(D_WARNING, "using old last_rcvd format\n");
281 lsd->lsd_mount_count = lsd->lsd_last_transno;
282 lsd->lsd_last_transno = lsd->lsd_unused;
283 /* If we update the last_rcvd, we can never go back to
284 an old install, so leave this in the old format for now.
285 lsd->lsd_feature_incompat |= cpu_to_le32(LR_INCOMPAT_COMMON_LR);
289 mds->mds_last_transno = le64_to_cpu(lsd->lsd_last_transno);
291 CDEBUG(D_INODE, "%s: server last_transno: "LPU64"\n",
292 obd->obd_name, mds->mds_last_transno);
293 CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
294 obd->obd_name, mount_count + 1);
295 CDEBUG(D_INODE, "%s: server data size: %u\n",
296 obd->obd_name, le32_to_cpu(lsd->lsd_server_size));
297 CDEBUG(D_INODE, "%s: per-client data start: %u\n",
298 obd->obd_name, le32_to_cpu(lsd->lsd_client_start));
299 CDEBUG(D_INODE, "%s: per-client data size: %u\n",
300 obd->obd_name, le32_to_cpu(lsd->lsd_client_size));
301 CDEBUG(D_INODE, "%s: last_rcvd size: %lu\n",
302 obd->obd_name, last_rcvd_size);
303 CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
304 last_rcvd_size <= le32_to_cpu(lsd->lsd_client_start) ? 0 :
305 (last_rcvd_size - le32_to_cpu(lsd->lsd_client_start)) /
306 le16_to_cpu(lsd->lsd_client_size));
308 if (!lsd->lsd_server_size || !lsd->lsd_client_start ||
309 !lsd->lsd_client_size) {
310 CERROR("Bad last_rcvd contents!\n");
311 GOTO(err_msd, rc = -EINVAL);
314 /* When we do a clean MDS shutdown, we save the last_transno into
315 * the header. If we find clients with higher last_transno values
316 * then those clients may need recovery done. */
317 for (cl_idx = 0, off = le32_to_cpu(lsd->lsd_client_start);
318 off < last_rcvd_size; cl_idx++) {
320 struct obd_export *exp;
321 struct mds_export_data *med;
324 OBD_ALLOC_WAIT(mcd, sizeof(*mcd));
326 GOTO(err_client, rc = -ENOMEM);
329 /* Don't assume off is incremented properly by
330 * fsfilt_read_record(), in case sizeof(*mcd)
331 * isn't the same as lsd->lsd_client_size. */
332 off = le32_to_cpu(lsd->lsd_client_start) +
333 cl_idx * le16_to_cpu(lsd->lsd_client_size);
334 rc = fsfilt_read_record(obd, file, mcd, sizeof(*mcd), &off);
336 CERROR("error reading MDS %s idx %d, off %llu: rc %d\n",
337 LAST_RCVD, cl_idx, off, rc);
338 break; /* read error shouldn't cause startup to fail */
341 if (mcd->mcd_uuid[0] == '\0') {
342 CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
347 last_transno = le64_to_cpu(mcd->mcd_last_transno);
349 /* These exports are cleaned up by mds_disconnect(), so they
350 * need to be set up like real exports as mds_connect() does.
352 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
353 " srv lr: "LPU64" lx: "LPU64"\n", mcd->mcd_uuid, cl_idx,
354 last_transno, le64_to_cpu(lsd->lsd_last_transno),
355 le64_to_cpu(mcd->mcd_last_xid));
357 exp = class_new_export(obd);
359 GOTO(err_client, rc = -ENOMEM);
361 memcpy(&exp->exp_client_uuid.uuid, mcd->mcd_uuid,
362 sizeof exp->exp_client_uuid.uuid);
363 med = &exp->exp_mds_data;
365 rc = mds_client_add(obd, mds, med, cl_idx);
366 LASSERTF(rc == 0, "rc = %d\n", rc); /* can't fail existing */
368 /* create helper if export init gets more complex */
369 INIT_LIST_HEAD(&med->med_open_head);
370 spin_lock_init(&med->med_open_lock);
373 exp->exp_replay_needed = 1;
374 obd->obd_recoverable_clients++;
375 obd->obd_max_recoverable_clients++;
376 class_export_put(exp);
378 CDEBUG(D_OTHER, "client at idx %d has last_transno = "LPU64"\n",
379 cl_idx, last_transno);
381 if (last_transno > mds->mds_last_transno)
382 mds->mds_last_transno = last_transno;
386 OBD_FREE(mcd, sizeof(*mcd));
388 obd->obd_last_committed = mds->mds_last_transno;
390 if (obd->obd_recoverable_clients) {
391 CWARN("RECOVERY: service %s, %d recoverable clients, "
392 "last_transno "LPU64"\n", obd->obd_name,
393 obd->obd_recoverable_clients, mds->mds_last_transno);
394 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
395 obd->obd_recovering = 1;
396 obd->obd_recovery_start = CURRENT_SECONDS;
397 /* Only used for lprocfs_status */
398 obd->obd_recovery_end = obd->obd_recovery_start +
399 OBD_RECOVERY_TIMEOUT / HZ;
402 mds->mds_mount_count = mount_count + 1;
403 lsd->lsd_mount_count = cpu_to_le64(mds->mds_mount_count);
405 /* save it, so mount count and last_transno is current */
406 rc = mds_update_server_data(obd, 1);
408 GOTO(err_client, rc);
413 class_disconnect_exports(obd);
415 mds_server_free_data(mds);
419 int mds_fs_setup(struct obd_device *obd, struct vfsmount *mnt)
421 struct mds_obd *mds = &obd->u.mds;
422 struct lvfs_run_ctxt saved;
423 struct dentry *dentry;
428 rc = cleanup_group_info();
432 mds->mds_vfsmnt = mnt;
433 obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
435 fsfilt_setup(obd, obd->u.obt.obt_sb);
437 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
438 obd->obd_lvfs_ctxt.pwdmnt = mnt;
439 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
440 obd->obd_lvfs_ctxt.fs = get_ds();
441 obd->obd_lvfs_ctxt.cb_ops = mds_lvfs_ops;
443 /* setup the directory tree */
444 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
445 dentry = simple_mkdir(current->fs->pwd, "ROOT", 0755, 0);
446 if (IS_ERR(dentry)) {
447 rc = PTR_ERR(dentry);
448 CERROR("cannot create ROOT directory: rc = %d\n", rc);
452 mds->mds_rootfid.id = dentry->d_inode->i_ino;
453 mds->mds_rootfid.generation = dentry->d_inode->i_generation;
454 mds->mds_rootfid.f_type = S_IFDIR;
458 dentry = lookup_one_len("__iopen__", current->fs->pwd,
459 strlen("__iopen__"));
460 if (IS_ERR(dentry)) {
461 rc = PTR_ERR(dentry);
462 CERROR("cannot lookup __iopen__ directory: rc = %d\n", rc);
466 mds->mds_fid_de = dentry;
467 if (!dentry->d_inode || is_bad_inode(dentry->d_inode)) {
469 CERROR("__iopen__ directory has no inode? rc = %d\n", rc);
473 dentry = simple_mkdir(current->fs->pwd, "PENDING", 0777, 1);
474 if (IS_ERR(dentry)) {
475 rc = PTR_ERR(dentry);
476 CERROR("cannot create PENDING directory: rc = %d\n", rc);
479 mds->mds_pending_dir = dentry;
482 dentry = simple_mkdir(current->fs->pwd, MDT_LOGS_DIR, 0777, 1);
483 if (IS_ERR(dentry)) {
484 rc = PTR_ERR(dentry);
485 CERROR("cannot create %s directory: rc = %d\n",
487 GOTO(err_pending, rc);
489 mds->mds_logs_dir = dentry;
492 dentry = simple_mkdir(current->fs->pwd, "OBJECTS", 0777, 1);
493 if (IS_ERR(dentry)) {
494 rc = PTR_ERR(dentry);
495 CERROR("cannot create OBJECTS directory: rc = %d\n", rc);
498 mds->mds_objects_dir = dentry;
500 /* open and test the last rcvd file */
501 file = filp_open(LAST_RCVD, O_RDWR | O_CREAT, 0644);
504 CERROR("cannot open/create %s file: rc = %d\n", LAST_RCVD, rc);
505 GOTO(err_objects, rc = PTR_ERR(file));
507 mds->mds_rcvd_filp = file;
508 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
509 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
510 file->f_dentry->d_inode->i_mode);
511 GOTO(err_last_rcvd, rc = -ENOENT);
514 rc = mds_init_server_data(obd, file);
516 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
517 GOTO(err_last_rcvd, rc);
520 /* open and test the lov objd file */
521 file = filp_open(LOV_OBJID, O_RDWR | O_CREAT, 0644);
524 CERROR("cannot open/create %s file: rc = %d\n", LOV_OBJID, rc);
525 GOTO(err_client, rc = PTR_ERR(file));
527 mds->mds_lov_objid_filp = file;
528 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
529 CERROR("%s is not a regular file!: mode = %o\n", LOV_OBJID,
530 file->f_dentry->d_inode->i_mode);
531 GOTO(err_lov_objid, rc = -ENOENT);
534 /* open and test the check io file junk */
535 file = filp_open(HEALTH_CHECK, O_RDWR | O_CREAT, 0644);
538 CERROR("cannot open/create %s file: rc = %d\n", HEALTH_CHECK, rc);
539 GOTO(err_lov_objid, rc = PTR_ERR(file));
541 mds->mds_health_check_filp = file;
542 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
543 CERROR("%s is not a regular file!: mode = %o\n", HEALTH_CHECK,
544 file->f_dentry->d_inode->i_mode);
545 GOTO(err_health_check, rc = -ENOENT);
547 rc = lvfs_check_io_health(obd, file);
549 GOTO(err_health_check, rc);
551 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
556 if (mds->mds_health_check_filp &&
557 filp_close(mds->mds_health_check_filp, 0))
558 CERROR("can't close %s after error\n", HEALTH_CHECK);
560 if (mds->mds_lov_objid_filp && filp_close(mds->mds_lov_objid_filp, 0))
561 CERROR("can't close %s after error\n", LOV_OBJID);
563 class_disconnect_exports(obd);
565 if (mds->mds_rcvd_filp && filp_close(mds->mds_rcvd_filp, 0))
566 CERROR("can't close %s after error\n", LAST_RCVD);
568 dput(mds->mds_objects_dir);
570 dput(mds->mds_logs_dir);
572 dput(mds->mds_pending_dir);
574 dput(mds->mds_fid_de);
579 int mds_fs_cleanup(struct obd_device *obd)
581 struct mds_obd *mds = &obd->u.mds;
582 struct lvfs_run_ctxt saved;
586 CWARN("%s: shutting down for failover; client state will "
587 "be preserved.\n", obd->obd_name);
589 class_disconnect_exports(obd); /* cleans up client info too */
590 mds_server_free_data(mds);
592 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
593 if (mds->mds_rcvd_filp) {
594 rc = filp_close(mds->mds_rcvd_filp, 0);
595 mds->mds_rcvd_filp = NULL;
597 CERROR("%s file won't close, rc=%d\n", LAST_RCVD, rc);
599 if (mds->mds_lov_objid_filp) {
600 rc = filp_close(mds->mds_lov_objid_filp, 0);
601 mds->mds_lov_objid_filp = NULL;
603 CERROR("%s file won't close, rc=%d\n", LOV_OBJID, rc);
605 if (mds->mds_health_check_filp) {
606 rc = filp_close(mds->mds_health_check_filp, 0);
607 mds->mds_health_check_filp = NULL;
609 CERROR("%s file won't close, rc=%d\n", HEALTH_CHECK, rc);
611 if (mds->mds_objects_dir != NULL) {
612 l_dput(mds->mds_objects_dir);
613 mds->mds_objects_dir = NULL;
615 if (mds->mds_logs_dir) {
616 l_dput(mds->mds_logs_dir);
617 mds->mds_logs_dir = NULL;
619 if (mds->mds_pending_dir) {
620 l_dput(mds->mds_pending_dir);
621 mds->mds_pending_dir = NULL;
624 lquota_fs_cleanup(quota_interface, obd);
626 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
627 shrink_dcache_parent(mds->mds_fid_de);
628 dput(mds->mds_fid_de);
629 LL_DQUOT_OFF(obd->u.obt.obt_sb);
634 /* Creates an object with the same name as its fid. Because this is not at all
635 * performance sensitive, it is accomplished by creating a file, checking the
636 * fid, and renaming it. */
637 int mds_obd_create(struct obd_export *exp, struct obdo *oa,
638 struct lov_stripe_md **ea, struct obd_trans_info *oti)
640 struct mds_obd *mds = &exp->exp_obd->u.mds;
641 struct inode *parent_inode = mds->mds_objects_dir->d_inode;
642 unsigned int tmpname = ll_rand();
644 struct dentry *new_child;
645 struct lvfs_run_ctxt saved;
646 char fidname[LL_FID_NAMELEN];
648 struct lvfs_ucred ucred = { 0 };
649 int rc = 0, err, namelen;
652 /* the owner of object file should always be root */
653 ucred.luc_cap = current->cap_effective | CAP_SYS_RESOURCE;
655 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, &ucred);
657 sprintf(fidname, "OBJECTS/%u.%u", tmpname, current->pid);
658 filp = filp_open(fidname, O_CREAT | O_EXCL, 0666);
662 CERROR("impossible object name collision %u\n",
666 CERROR("error creating tmp object %u: rc %d\n", tmpname, rc);
670 LASSERT(mds->mds_objects_dir == filp->f_dentry->d_parent);
672 oa->o_id = filp->f_dentry->d_inode->i_ino;
673 oa->o_generation = filp->f_dentry->d_inode->i_generation;
674 namelen = ll_fid2str(fidname, oa->o_id, oa->o_generation);
676 down(&parent_inode->i_sem);
677 new_child = lookup_one_len(fidname, mds->mds_objects_dir, namelen);
679 if (IS_ERR(new_child)) {
680 CERROR("getting neg dentry for obj rename: %d\n", rc);
681 GOTO(out_close, rc = PTR_ERR(new_child));
683 if (new_child->d_inode != NULL) {
684 CERROR("impossible non-negative obj dentry " LPU64":%u!\n",
685 oa->o_id, oa->o_generation);
689 handle = fsfilt_start(exp->exp_obd, mds->mds_objects_dir->d_inode,
690 FSFILT_OP_RENAME, NULL);
692 GOTO(out_dput, rc = PTR_ERR(handle));
695 rc = vfs_rename(mds->mds_objects_dir->d_inode, filp->f_dentry,
696 mds->mds_objects_dir->d_inode, new_child);
699 CERROR("error renaming new object "LPU64":%u: rc %d\n",
700 oa->o_id, oa->o_generation, rc);
702 err = fsfilt_commit(exp->exp_obd, mds->mds_objects_dir->d_inode,
705 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGENER;
711 up(&parent_inode->i_sem);
712 err = filp_close(filp, 0);
714 CERROR("closing tmpfile %u: rc %d\n", tmpname, rc);
719 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, &ucred);
723 int mds_obd_destroy(struct obd_export *exp, struct obdo *oa,
724 struct lov_stripe_md *ea, struct obd_trans_info *oti,
725 struct obd_export *md_exp)
727 struct mds_obd *mds = &exp->exp_obd->u.mds;
728 struct inode *parent_inode = mds->mds_objects_dir->d_inode;
729 struct obd_device *obd = exp->exp_obd;
730 struct lvfs_run_ctxt saved;
731 struct lvfs_ucred ucred = { 0 };
732 char fidname[LL_FID_NAMELEN];
735 int err, namelen, rc = 0;
738 ucred.luc_cap = current->cap_effective | CAP_SYS_RESOURCE;
739 push_ctxt(&saved, &obd->obd_lvfs_ctxt, &ucred);
741 namelen = ll_fid2str(fidname, oa->o_id, oa->o_generation);
743 down(&parent_inode->i_sem);
744 de = lookup_one_len(fidname, mds->mds_objects_dir, namelen);
748 CERROR("error looking up object "LPU64" %s: rc %d\n",
749 oa->o_id, fidname, rc);
752 if (de->d_inode == NULL) {
753 CERROR("destroying non-existent object "LPU64" %s: rc %d\n",
754 oa->o_id, fidname, rc);
755 GOTO(out_dput, rc = -ENOENT);
758 /* Stripe count is 1 here since this is some MDS specific stuff
759 that is unlinked, not spanned across multiple OSTs */
760 handle = fsfilt_start_log(obd, mds->mds_objects_dir->d_inode,
761 FSFILT_OP_UNLINK, oti, 1);
764 GOTO(out_dput, rc = PTR_ERR(handle));
766 rc = vfs_unlink(mds->mds_objects_dir->d_inode, de);
768 CERROR("error destroying object "LPU64":%u: rc %d\n",
769 oa->o_id, oa->o_generation, rc);
771 err = fsfilt_commit(obd, mds->mds_objects_dir->d_inode, handle, 0);
777 up(&parent_inode->i_sem);
779 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &ucred);