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 <lustre_quota.h>
39 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
40 #include <linux/mount.h>
42 #include <lustre_mds.h>
43 #include <obd_class.h>
44 #include <obd_support.h>
45 #include <lustre_lib.h>
46 #include <lustre_fsfilt.h>
47 #include <lustre_disk.h>
48 #include <libcfs/list.h>
50 #include "mds_internal.h"
53 /* Add client data to the MDS. We use a bitmap to locate a free space
54 * in the last_rcvd file if cl_off is -1 (i.e. a new client).
55 * Otherwise, we have just read the data from the last_rcvd file and
58 * It should not be possible to fail adding an existing client - otherwise
59 * mds_init_server_data() callsite needs to be fixed.
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);
69 LASSERTF(cl_idx > -2, "%d\n", cl_idx);
71 /* XXX if mcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
72 if (!strcmp(med->med_mcd->mcd_uuid, obd->obd_uuid.uuid))
75 /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
76 * there's no need for extra complication here
79 cl_idx = find_first_zero_bit(bitmap, LR_MAX_CLIENTS);
81 if (cl_idx >= LR_MAX_CLIENTS ||
82 OBD_FAIL_CHECK_ONCE(OBD_FAIL_MDS_CLIENT_ADD)) {
83 CERROR("no room for clients - fix LR_MAX_CLIENTS\n");
86 if (test_and_set_bit(cl_idx, bitmap)) {
87 cl_idx = find_next_zero_bit(bitmap, LR_MAX_CLIENTS,
92 if (test_and_set_bit(cl_idx, bitmap)) {
93 CERROR("MDS client %d: bit already set in bitmap!!\n",
99 CDEBUG(D_INFO, "client at idx %d with UUID '%s' added\n",
100 cl_idx, med->med_mcd->mcd_uuid);
102 med->med_lr_idx = cl_idx;
103 med->med_lr_off = le32_to_cpu(mds->mds_server_data->lsd_client_start) +
104 (cl_idx * le16_to_cpu(mds->mds_server_data->lsd_client_size));
105 LASSERTF(med->med_lr_off > 0, "med_lr_off = %llu\n", med->med_lr_off);
108 struct lvfs_run_ctxt saved;
109 loff_t off = med->med_lr_off;
110 struct file *file = mds->mds_rcvd_filp;
113 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
114 rc = fsfilt_write_record(obd, file, med->med_mcd,
115 sizeof(*med->med_mcd), &off, 1);
116 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
120 CDEBUG(D_INFO, "wrote client mcd at idx %u off %llu (len %u)\n",
121 med->med_lr_idx, med->med_lr_off,
122 (unsigned int)sizeof(*med->med_mcd));
127 int mds_client_free(struct obd_export *exp)
129 struct mds_export_data *med = &exp->exp_mds_data;
130 struct mds_obd *mds = &exp->exp_obd->u.mds;
131 struct obd_device *obd = exp->exp_obd;
132 struct mds_client_data zero_mcd;
133 struct lvfs_run_ctxt saved;
141 /* XXX if mcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
142 if (!strcmp(med->med_mcd->mcd_uuid, obd->obd_uuid.uuid))
145 CDEBUG(D_INFO, "freeing client at idx %u, offset %lld with UUID '%s'\n",
146 med->med_lr_idx, med->med_lr_off, med->med_mcd->mcd_uuid);
148 LASSERT(mds->mds_client_bitmap != NULL);
150 off = med->med_lr_off;
152 /* Don't clear med_lr_idx here as it is likely also unset. At worst
153 * we leak a client slot that will be cleaned on the next recovery. */
155 CERROR("%s: client idx %d has offset %lld\n",
156 obd->obd_name, med->med_lr_idx, off);
157 GOTO(free, rc = -EINVAL);
160 /* Clear the bit _after_ zeroing out the client so we don't
161 race with mds_client_add and zero out new clients.*/
162 if (!test_bit(med->med_lr_idx, mds->mds_client_bitmap)) {
163 CERROR("MDS client %u: bit already clear in bitmap!!\n",
168 if (!(exp->exp_flags & OBD_OPT_FAILOVER)) {
169 memset(&zero_mcd, 0, sizeof zero_mcd);
170 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
171 rc = fsfilt_write_record(obd, mds->mds_rcvd_filp, &zero_mcd,
172 sizeof(zero_mcd), &off, 1);
173 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
175 CDEBUG_EX(rc == 0 ? D_INFO : D_ERROR,
176 "zeroing out client %s idx %u in %s rc %d\n",
177 med->med_mcd->mcd_uuid, med->med_lr_idx, LAST_RCVD, rc);
180 if (!test_and_clear_bit(med->med_lr_idx, mds->mds_client_bitmap)) {
181 CERROR("MDS client %u: bit already clear in bitmap!!\n",
187 /* Make sure the server's last_transno is up to date. Do this
188 * after the client is freed so we know all the client's
189 * transactions have been committed. */
190 mds_update_server_data(exp->exp_obd, 0);
194 OBD_FREE(med->med_mcd, sizeof(*med->med_mcd));
200 static int mds_server_free_data(struct mds_obd *mds)
202 OBD_FREE(mds->mds_client_bitmap, LR_MAX_CLIENTS / 8);
203 OBD_FREE(mds->mds_server_data, sizeof(*mds->mds_server_data));
204 mds->mds_server_data = NULL;
209 static int mds_init_server_data(struct obd_device *obd, struct file *file)
211 struct mds_obd *mds = &obd->u.mds;
212 struct lr_server_data *lsd;
213 struct mds_client_data *mcd = NULL;
215 unsigned long last_rcvd_size = file->f_dentry->d_inode->i_size;
220 /* ensure padding in the struct is the correct size */
221 LASSERT(offsetof(struct lr_server_data, lsd_padding) +
222 sizeof(lsd->lsd_padding) == LR_SERVER_SIZE);
223 LASSERT(offsetof(struct mds_client_data, mcd_padding) +
224 sizeof(mcd->mcd_padding) == LR_CLIENT_SIZE);
226 OBD_ALLOC_WAIT(lsd, sizeof(*lsd));
230 OBD_ALLOC_WAIT(mds->mds_client_bitmap, LR_MAX_CLIENTS / 8);
231 if (!mds->mds_client_bitmap) {
232 OBD_FREE(lsd, sizeof(*lsd));
236 mds->mds_server_data = lsd;
238 if (last_rcvd_size == 0) {
239 LCONSOLE_WARN("%s: new disk, initializing\n", obd->obd_name);
241 memcpy(lsd->lsd_uuid, obd->obd_uuid.uuid,sizeof(lsd->lsd_uuid));
242 lsd->lsd_last_transno = 0;
243 mount_count = lsd->lsd_mount_count = 0;
244 lsd->lsd_server_size = cpu_to_le32(LR_SERVER_SIZE);
245 lsd->lsd_client_start = cpu_to_le32(LR_CLIENT_START);
246 lsd->lsd_client_size = cpu_to_le16(LR_CLIENT_SIZE);
247 lsd->lsd_feature_rocompat = cpu_to_le32(OBD_ROCOMPAT_LOVOBJID);
248 lsd->lsd_feature_incompat = cpu_to_le32(OBD_INCOMPAT_MDT |
249 OBD_INCOMPAT_COMMON_LR);
251 rc = fsfilt_read_record(obd, file, lsd, sizeof(*lsd), &off);
253 CERROR("error reading MDS %s: rc %d\n", LAST_RCVD, rc);
256 if (strcmp(lsd->lsd_uuid, obd->obd_uuid.uuid) != 0) {
257 LCONSOLE_ERROR("Trying to start OBD %s using the wrong"
258 " disk %s. Were the /dev/ assignments "
260 obd->obd_uuid.uuid, lsd->lsd_uuid);
261 GOTO(err_msd, rc = -EINVAL);
263 mount_count = le64_to_cpu(lsd->lsd_mount_count);
266 if (lsd->lsd_feature_incompat & ~cpu_to_le32(MDT_INCOMPAT_SUPP)) {
267 CERROR("%s: unsupported incompat filesystem feature(s) %x\n",
268 obd->obd_name, le32_to_cpu(lsd->lsd_feature_incompat) &
270 GOTO(err_msd, rc = -EINVAL);
272 if (lsd->lsd_feature_rocompat & ~cpu_to_le32(MDT_ROCOMPAT_SUPP)) {
273 CERROR("%s: unsupported read-only filesystem feature(s) %x\n",
274 obd->obd_name, le32_to_cpu(lsd->lsd_feature_rocompat) &
276 /* Do something like remount filesystem read-only */
277 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);
288 lsd->lsd_feature_compat = cpu_to_le32(OBD_COMPAT_MDT);
290 mds->mds_last_transno = le64_to_cpu(lsd->lsd_last_transno);
292 CDEBUG(D_INODE, "%s: server last_transno: "LPU64"\n",
293 obd->obd_name, mds->mds_last_transno);
294 CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
295 obd->obd_name, mount_count + 1);
296 CDEBUG(D_INODE, "%s: server data size: %u\n",
297 obd->obd_name, le32_to_cpu(lsd->lsd_server_size));
298 CDEBUG(D_INODE, "%s: per-client data start: %u\n",
299 obd->obd_name, le32_to_cpu(lsd->lsd_client_start));
300 CDEBUG(D_INODE, "%s: per-client data size: %u\n",
301 obd->obd_name, le32_to_cpu(lsd->lsd_client_size));
302 CDEBUG(D_INODE, "%s: last_rcvd size: %lu\n",
303 obd->obd_name, last_rcvd_size);
304 CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
305 last_rcvd_size <= le32_to_cpu(lsd->lsd_client_start) ? 0 :
306 (last_rcvd_size - le32_to_cpu(lsd->lsd_client_start)) /
307 le16_to_cpu(lsd->lsd_client_size));
309 if (!lsd->lsd_server_size || !lsd->lsd_client_start ||
310 !lsd->lsd_client_size) {
311 CERROR("Bad last_rcvd contents!\n");
312 GOTO(err_msd, rc = -EINVAL);
315 /* When we do a clean MDS shutdown, we save the last_transno into
316 * the header. If we find clients with higher last_transno values
317 * then those clients may need recovery done. */
318 for (cl_idx = 0, off = le32_to_cpu(lsd->lsd_client_start);
319 off < last_rcvd_size; cl_idx++) {
321 struct obd_export *exp;
322 struct mds_export_data *med;
325 OBD_ALLOC_WAIT(mcd, sizeof(*mcd));
327 GOTO(err_client, rc = -ENOMEM);
330 /* Don't assume off is incremented properly by
331 * fsfilt_read_record(), in case sizeof(*mcd)
332 * isn't the same as lsd->lsd_client_size. */
333 off = le32_to_cpu(lsd->lsd_client_start) +
334 cl_idx * le16_to_cpu(lsd->lsd_client_size);
335 rc = fsfilt_read_record(obd, file, mcd, sizeof(*mcd), &off);
337 CERROR("error reading MDS %s idx %d, off %llu: rc %d\n",
338 LAST_RCVD, cl_idx, off, rc);
339 break; /* read error shouldn't cause startup to fail */
342 if (mcd->mcd_uuid[0] == '\0') {
343 CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
348 last_transno = le64_to_cpu(mcd->mcd_last_transno);
350 /* These exports are cleaned up by mds_disconnect(), so they
351 * need to be set up like real exports as mds_connect() does.
353 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
354 " srv lr: "LPU64" lx: "LPU64"\n", mcd->mcd_uuid, cl_idx,
355 last_transno, le64_to_cpu(lsd->lsd_last_transno),
356 le64_to_cpu(mcd->mcd_last_xid));
358 exp = class_new_export(obd, (struct obd_uuid *)mcd->mcd_uuid);
360 GOTO(err_client, rc = PTR_ERR(exp));
362 med = &exp->exp_mds_data;
364 rc = mds_client_add(obd, mds, med, cl_idx);
365 LASSERTF(rc == 0, "rc = %d\n", rc); /* can't fail existing */
369 exp->exp_replay_needed = 1;
370 exp->exp_connecting = 0;
371 obd->obd_recoverable_clients++;
372 obd->obd_max_recoverable_clients++;
373 class_export_put(exp);
375 CDEBUG(D_OTHER, "client at idx %d has last_transno = "LPU64"\n",
376 cl_idx, last_transno);
378 if (last_transno > mds->mds_last_transno)
379 mds->mds_last_transno = last_transno;
383 OBD_FREE(mcd, sizeof(*mcd));
385 obd->obd_last_committed = mds->mds_last_transno;
387 if (obd->obd_recoverable_clients) {
388 CWARN("RECOVERY: service %s, %d recoverable clients, "
389 "last_transno "LPU64"\n", obd->obd_name,
390 obd->obd_recoverable_clients, mds->mds_last_transno);
391 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
392 obd->obd_recovering = 1;
393 obd->obd_recovery_start = CURRENT_SECONDS;
394 /* Only used for lprocfs_status */
395 obd->obd_recovery_end = obd->obd_recovery_start +
396 OBD_RECOVERY_TIMEOUT;
399 mds->mds_mount_count = mount_count + 1;
400 lsd->lsd_mount_count = cpu_to_le64(mds->mds_mount_count);
402 /* save it, so mount count and last_transno is current */
403 rc = mds_update_server_data(obd, 1);
405 GOTO(err_client, rc);
410 class_disconnect_exports(obd);
412 mds_server_free_data(mds);
416 void mds_init_ctxt(struct obd_device *obd, struct vfsmount *mnt)
418 struct mds_obd *mds = &obd->u.mds;
420 mds->mds_vfsmnt = mnt;
421 /* why not mnt->mnt_sb instead of mnt->mnt_root->d_inode->i_sb? */
422 obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
424 fsfilt_setup(obd, obd->u.obt.obt_sb);
426 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
427 obd->obd_lvfs_ctxt.pwdmnt = mnt;
428 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
429 obd->obd_lvfs_ctxt.fs = get_ds();
430 obd->obd_lvfs_ctxt.cb_ops = mds_lvfs_ops;
434 int mds_fs_setup(struct obd_device *obd, struct vfsmount *mnt)
436 struct mds_obd *mds = &obd->u.mds;
437 struct lvfs_run_ctxt saved;
438 struct dentry *dentry;
443 rc = cleanup_group_info();
447 mds_init_ctxt(obd, mnt);
449 /* setup the directory tree */
450 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
451 dentry = simple_mkdir(current->fs->pwd, "ROOT", 0755, 0);
452 if (IS_ERR(dentry)) {
453 rc = PTR_ERR(dentry);
454 CERROR("cannot create ROOT directory: rc = %d\n", rc);
458 mds->mds_rootfid.id = dentry->d_inode->i_ino;
459 mds->mds_rootfid.generation = dentry->d_inode->i_generation;
460 mds->mds_rootfid.f_type = S_IFDIR;
464 dentry = lookup_one_len("__iopen__", current->fs->pwd,
465 strlen("__iopen__"));
466 if (IS_ERR(dentry)) {
467 rc = PTR_ERR(dentry);
468 CERROR("cannot lookup __iopen__ directory: rc = %d\n", rc);
472 mds->mds_fid_de = dentry;
473 if (!dentry->d_inode || is_bad_inode(dentry->d_inode)) {
475 CERROR("__iopen__ directory has no inode? rc = %d\n", rc);
479 dentry = simple_mkdir(current->fs->pwd, "PENDING", 0777, 1);
480 if (IS_ERR(dentry)) {
481 rc = PTR_ERR(dentry);
482 CERROR("cannot create PENDING directory: rc = %d\n", rc);
485 mds->mds_pending_dir = dentry;
488 dentry = simple_mkdir(current->fs->pwd, MDT_LOGS_DIR, 0777, 1);
489 if (IS_ERR(dentry)) {
490 rc = PTR_ERR(dentry);
491 CERROR("cannot create %s directory: rc = %d\n",
493 GOTO(err_pending, rc);
495 mds->mds_logs_dir = dentry;
498 dentry = simple_mkdir(current->fs->pwd, "OBJECTS", 0777, 1);
499 if (IS_ERR(dentry)) {
500 rc = PTR_ERR(dentry);
501 CERROR("cannot create OBJECTS directory: rc = %d\n", rc);
504 mds->mds_objects_dir = dentry;
506 /* open and test the last rcvd file */
507 file = filp_open(LAST_RCVD, O_RDWR | O_CREAT, 0644);
510 CERROR("cannot open/create %s file: rc = %d\n", LAST_RCVD, rc);
511 GOTO(err_objects, rc = PTR_ERR(file));
513 mds->mds_rcvd_filp = file;
514 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
515 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
516 file->f_dentry->d_inode->i_mode);
517 GOTO(err_last_rcvd, rc = -ENOENT);
520 rc = mds_init_server_data(obd, file);
522 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
523 GOTO(err_last_rcvd, rc);
526 /* open and test the lov objd file */
527 file = filp_open(LOV_OBJID, O_RDWR | O_CREAT, 0644);
530 CERROR("cannot open/create %s file: rc = %d\n", LOV_OBJID, rc);
531 GOTO(err_client, rc = PTR_ERR(file));
533 mds->mds_lov_objid_filp = file;
534 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
535 CERROR("%s is not a regular file!: mode = %o\n", LOV_OBJID,
536 file->f_dentry->d_inode->i_mode);
537 GOTO(err_lov_objid, rc = -ENOENT);
540 /* open and test the check io file junk */
541 file = filp_open(HEALTH_CHECK, O_RDWR | O_CREAT, 0644);
544 CERROR("cannot open/create %s file: rc = %d\n", HEALTH_CHECK, rc);
545 GOTO(err_lov_objid, rc = PTR_ERR(file));
547 mds->mds_health_check_filp = file;
548 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
549 CERROR("%s is not a regular file!: mode = %o\n", HEALTH_CHECK,
550 file->f_dentry->d_inode->i_mode);
551 GOTO(err_health_check, rc = -ENOENT);
553 rc = lvfs_check_io_health(obd, file);
555 GOTO(err_health_check, rc);
557 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
562 if (mds->mds_health_check_filp &&
563 filp_close(mds->mds_health_check_filp, 0))
564 CERROR("can't close %s after error\n", HEALTH_CHECK);
566 if (mds->mds_lov_objid_filp &&
567 filp_close((struct file *)mds->mds_lov_objid_filp, 0))
568 CERROR("can't close %s after error\n", LOV_OBJID);
570 class_disconnect_exports(obd);
572 if (mds->mds_rcvd_filp && filp_close(mds->mds_rcvd_filp, 0))
573 CERROR("can't close %s after error\n", LAST_RCVD);
575 dput(mds->mds_objects_dir);
577 dput(mds->mds_logs_dir);
579 dput(mds->mds_pending_dir);
581 dput(mds->mds_fid_de);
586 int mds_fs_cleanup(struct obd_device *obd)
588 struct mds_obd *mds = &obd->u.mds;
589 struct lvfs_run_ctxt saved;
593 LCONSOLE_WARN("%s: shutting down for failover; client state "
594 "will be preserved.\n", obd->obd_name);
596 class_disconnect_exports(obd); /* cleans up client info too */
597 mds_server_free_data(mds);
599 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
600 if (mds->mds_rcvd_filp) {
601 rc = filp_close(mds->mds_rcvd_filp, 0);
602 mds->mds_rcvd_filp = NULL;
604 CERROR("%s file won't close, rc=%d\n", LAST_RCVD, rc);
606 if (mds->mds_lov_objid_filp) {
607 rc = filp_close((struct file *)mds->mds_lov_objid_filp, 0);
608 mds->mds_lov_objid_filp = NULL;
610 CERROR("%s file won't close, rc=%d\n", LOV_OBJID, rc);
612 if (mds->mds_health_check_filp) {
613 rc = filp_close(mds->mds_health_check_filp, 0);
614 mds->mds_health_check_filp = NULL;
616 CERROR("%s file won't close, rc=%d\n", HEALTH_CHECK, rc);
618 if (mds->mds_objects_dir != NULL) {
619 l_dput(mds->mds_objects_dir);
620 mds->mds_objects_dir = NULL;
622 if (mds->mds_logs_dir) {
623 l_dput(mds->mds_logs_dir);
624 mds->mds_logs_dir = NULL;
626 if (mds->mds_pending_dir) {
627 l_dput(mds->mds_pending_dir);
628 mds->mds_pending_dir = NULL;
631 lquota_fs_cleanup(quota_interface, obd);
633 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
634 shrink_dcache_parent(mds->mds_fid_de);
635 dput(mds->mds_fid_de);
636 LL_DQUOT_OFF(obd->u.obt.obt_sb);
641 /* Creates an object with the same name as its fid. Because this is not at all
642 * performance sensitive, it is accomplished by creating a file, checking the
643 * fid, and renaming it. */
644 int mds_obd_create(struct obd_export *exp, struct obdo *oa,
645 struct lov_stripe_md **ea, struct obd_trans_info *oti)
647 struct mds_obd *mds = &exp->exp_obd->u.mds;
648 struct inode *parent_inode = mds->mds_objects_dir->d_inode;
649 unsigned int tmpname = ll_rand();
651 struct dentry *new_child;
652 struct lvfs_run_ctxt saved;
653 char fidname[LL_FID_NAMELEN];
655 struct lvfs_ucred ucred = { 0 };
656 int rc = 0, err, namelen;
659 /* the owner of object file should always be root */
660 ucred.luc_cap = current->cap_effective | CAP_SYS_RESOURCE;
662 if (strcmp(exp->exp_obd->obd_name, MDD_OBD_NAME)) {
666 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, &ucred);
668 sprintf(fidname, "OBJECTS/%u.%u", tmpname, current->pid);
669 filp = filp_open(fidname, O_CREAT | O_EXCL, 0666);
673 CERROR("impossible object name collision %u\n",
677 CERROR("error creating tmp object %u: rc %d\n", tmpname, rc);
681 LASSERT(mds->mds_objects_dir == filp->f_dentry->d_parent);
683 oa->o_id = filp->f_dentry->d_inode->i_ino;
684 oa->o_generation = filp->f_dentry->d_inode->i_generation;
685 namelen = mds_fid2str(fidname, oa->o_id, oa->o_generation);
687 LOCK_INODE_MUTEX(parent_inode);
688 new_child = lookup_one_len(fidname, mds->mds_objects_dir, namelen);
690 if (IS_ERR(new_child)) {
691 CERROR("getting neg dentry for obj rename: %d\n", rc);
692 GOTO(out_close, rc = PTR_ERR(new_child));
694 if (new_child->d_inode != NULL) {
695 CERROR("impossible non-negative obj dentry " LPU64":%u!\n",
696 oa->o_id, oa->o_generation);
700 handle = fsfilt_start(exp->exp_obd, mds->mds_objects_dir->d_inode,
701 FSFILT_OP_RENAME, NULL);
703 GOTO(out_dput, rc = PTR_ERR(handle));
706 rc = vfs_rename(mds->mds_objects_dir->d_inode, filp->f_dentry,
707 mds->mds_objects_dir->d_inode, new_child);
710 CERROR("error renaming new object "LPU64":%u: rc %d\n",
711 oa->o_id, oa->o_generation, rc);
713 err = fsfilt_commit(exp->exp_obd, mds->mds_objects_dir->d_inode,
716 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGENER;
722 UNLOCK_INODE_MUTEX(parent_inode);
723 err = filp_close(filp, 0);
725 CERROR("closing tmpfile %u: rc %d\n", tmpname, rc);
730 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, &ucred);
734 int mds_obd_destroy(struct obd_export *exp, struct obdo *oa,
735 struct lov_stripe_md *ea, struct obd_trans_info *oti,
736 struct obd_export *md_exp)
738 struct mds_obd *mds = &exp->exp_obd->u.mds;
739 struct inode *parent_inode = mds->mds_objects_dir->d_inode;
740 struct obd_device *obd = exp->exp_obd;
741 struct lvfs_run_ctxt saved;
742 struct lvfs_ucred ucred = { 0 };
743 char fidname[LL_FID_NAMELEN];
746 int err, namelen, rc = 0;
749 ucred.luc_cap = current->cap_effective | CAP_SYS_RESOURCE;
750 push_ctxt(&saved, &obd->obd_lvfs_ctxt, &ucred);
752 namelen = mds_fid2str(fidname, oa->o_id, oa->o_generation);
754 LOCK_INODE_MUTEX(parent_inode);
755 de = lookup_one_len(fidname, mds->mds_objects_dir, namelen);
759 CERROR("error looking up object "LPU64" %s: rc %d\n",
760 oa->o_id, fidname, rc);
763 if (de->d_inode == NULL) {
764 CERROR("destroying non-existent object "LPU64" %s: rc %d\n",
765 oa->o_id, fidname, rc);
766 GOTO(out_dput, rc = -ENOENT);
769 /* Stripe count is 1 here since this is some MDS specific stuff
770 that is unlinked, not spanned across multiple OSTs */
771 handle = fsfilt_start_log(obd, mds->mds_objects_dir->d_inode,
772 FSFILT_OP_UNLINK, oti, 1);
775 GOTO(out_dput, rc = PTR_ERR(handle));
777 rc = vfs_unlink(mds->mds_objects_dir->d_inode, de);
779 CERROR("error destroying object "LPU64":%u: rc %d\n",
780 oa->o_id, oa->o_generation, rc);
782 err = fsfilt_commit(obd, mds->mds_objects_dir->d_inode, handle, 0);
788 UNLOCK_INODE_MUTEX(parent_inode);
790 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &ucred);