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_compat = cpu_to_le32(OBD_COMPAT_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 CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
258 obd->obd_uuid.uuid, lsd->lsd_uuid);
259 GOTO(err_msd, rc = -EINVAL);
261 mount_count = le64_to_cpu(lsd->lsd_mount_count);
263 if (lsd->lsd_feature_incompat & ~cpu_to_le32(MDT_INCOMPAT_SUPP)) {
264 CERROR("%s: unsupported incompat filesystem feature(s) %x\n",
265 obd->obd_name, le32_to_cpu(lsd->lsd_feature_incompat) &
267 GOTO(err_msd, rc = -EINVAL);
270 if (lsd->lsd_feature_rocompat & ~cpu_to_le32(MDT_ROCOMPAT_SUPP)) {
271 CERROR("%s: unsupported read-only filesystem feature(s) %x\n",
272 obd->obd_name, le32_to_cpu(lsd->lsd_feature_rocompat) &
274 /* Do something like remount filesystem read-only */
275 GOTO(err_msd, rc = -EINVAL);
278 if (!(lsd->lsd_feature_compat & cpu_to_le32(OBD_COMPAT_COMMON_LR))) {
279 CDEBUG(D_WARNING, "using old last_rcvd format\n");
280 lsd->lsd_mount_count = lsd->lsd_last_transno;
281 lsd->lsd_last_transno = lsd->lsd_unused;
282 /* If we update the last_rcvd, we can never go back to
283 an old install, so leave this in the old format for now.
284 lsd->lsd_feature_compat |= cpu_to_le32(LR_COMPAT_COMMON_LR);
288 mds->mds_last_transno = le64_to_cpu(lsd->lsd_last_transno);
290 CDEBUG(D_INODE, "%s: server last_transno: "LPU64"\n",
291 obd->obd_name, mds->mds_last_transno);
292 CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
293 obd->obd_name, mount_count + 1);
294 CDEBUG(D_INODE, "%s: server data size: %u\n",
295 obd->obd_name, le32_to_cpu(lsd->lsd_server_size));
296 CDEBUG(D_INODE, "%s: per-client data start: %u\n",
297 obd->obd_name, le32_to_cpu(lsd->lsd_client_start));
298 CDEBUG(D_INODE, "%s: per-client data size: %u\n",
299 obd->obd_name, le32_to_cpu(lsd->lsd_client_size));
300 CDEBUG(D_INODE, "%s: last_rcvd size: %lu\n",
301 obd->obd_name, last_rcvd_size);
302 CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
303 last_rcvd_size <= le32_to_cpu(lsd->lsd_client_start) ? 0 :
304 (last_rcvd_size - le32_to_cpu(lsd->lsd_client_start)) /
305 le16_to_cpu(lsd->lsd_client_size));
307 if (!lsd->lsd_server_size || !lsd->lsd_client_start ||
308 !lsd->lsd_client_size) {
309 CERROR("Bad last_rcvd contents!\n");
310 GOTO(err_msd, rc = -EINVAL);
313 /* When we do a clean MDS shutdown, we save the last_transno into
314 * the header. If we find clients with higher last_transno values
315 * then those clients may need recovery done. */
316 for (cl_idx = 0, off = le32_to_cpu(lsd->lsd_client_start);
317 off < last_rcvd_size; cl_idx++) {
319 struct obd_export *exp;
320 struct mds_export_data *med;
323 OBD_ALLOC_WAIT(mcd, sizeof(*mcd));
325 GOTO(err_client, rc = -ENOMEM);
328 /* Don't assume off is incremented properly by
329 * fsfilt_read_record(), in case sizeof(*mcd)
330 * isn't the same as lsd->lsd_client_size. */
331 off = le32_to_cpu(lsd->lsd_client_start) +
332 cl_idx * le16_to_cpu(lsd->lsd_client_size);
333 rc = fsfilt_read_record(obd, file, mcd, sizeof(*mcd), &off);
335 CERROR("error reading MDS %s idx %d, off %llu: rc %d\n",
336 LAST_RCVD, cl_idx, off, rc);
337 break; /* read error shouldn't cause startup to fail */
340 if (mcd->mcd_uuid[0] == '\0') {
341 CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
346 last_transno = le64_to_cpu(mcd->mcd_last_transno);
348 /* These exports are cleaned up by mds_disconnect(), so they
349 * need to be set up like real exports as mds_connect() does.
351 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
352 " srv lr: "LPU64" lx: "LPU64"\n", mcd->mcd_uuid, cl_idx,
353 last_transno, le64_to_cpu(lsd->lsd_last_transno),
354 le64_to_cpu(mcd->mcd_last_xid));
356 exp = class_new_export(obd);
358 GOTO(err_client, rc = -ENOMEM);
360 memcpy(&exp->exp_client_uuid.uuid, mcd->mcd_uuid,
361 sizeof exp->exp_client_uuid.uuid);
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 */
367 /* create helper if export init gets more complex */
368 INIT_LIST_HEAD(&med->med_open_head);
369 spin_lock_init(&med->med_open_lock);
372 exp->exp_replay_needed = 1;
373 obd->obd_recoverable_clients++;
374 obd->obd_max_recoverable_clients++;
375 class_export_put(exp);
377 CDEBUG(D_OTHER, "client at idx %d has last_transno = "LPU64"\n",
378 cl_idx, last_transno);
380 if (last_transno > mds->mds_last_transno)
381 mds->mds_last_transno = last_transno;
385 OBD_FREE(mcd, sizeof(*mcd));
387 obd->obd_last_committed = mds->mds_last_transno;
389 if (obd->obd_recoverable_clients) {
390 CWARN("RECOVERY: service %s, %d recoverable clients, "
391 "last_transno "LPU64"\n", obd->obd_name,
392 obd->obd_recoverable_clients, mds->mds_last_transno);
393 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
394 obd->obd_recovering = 1;
395 obd->obd_recovery_start = CURRENT_SECONDS;
396 /* Only used for lprocfs_status */
397 obd->obd_recovery_end = obd->obd_recovery_start +
398 OBD_RECOVERY_TIMEOUT / HZ;
401 mds->mds_mount_count = mount_count + 1;
402 lsd->lsd_mount_count = cpu_to_le64(mds->mds_mount_count);
404 /* save it, so mount count and last_transno is current */
405 rc = mds_update_server_data(obd, 1);
407 GOTO(err_client, rc);
412 class_disconnect_exports(obd);
414 mds_server_free_data(mds);
418 int mds_fs_setup(struct obd_device *obd, struct vfsmount *mnt)
420 struct mds_obd *mds = &obd->u.mds;
421 struct lvfs_run_ctxt saved;
422 struct dentry *dentry;
427 rc = cleanup_group_info();
431 mds->mds_vfsmnt = mnt;
432 obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
434 fsfilt_setup(obd, obd->u.obt.obt_sb);
436 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
437 obd->obd_lvfs_ctxt.pwdmnt = mnt;
438 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
439 obd->obd_lvfs_ctxt.fs = get_ds();
440 obd->obd_lvfs_ctxt.cb_ops = mds_lvfs_ops;
442 /* setup the directory tree */
443 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
444 dentry = simple_mkdir(current->fs->pwd, "ROOT", 0755, 0);
445 if (IS_ERR(dentry)) {
446 rc = PTR_ERR(dentry);
447 CERROR("cannot create ROOT directory: rc = %d\n", rc);
451 mds->mds_rootfid.id = dentry->d_inode->i_ino;
452 mds->mds_rootfid.generation = dentry->d_inode->i_generation;
453 mds->mds_rootfid.f_type = S_IFDIR;
457 dentry = lookup_one_len("__iopen__", current->fs->pwd,
458 strlen("__iopen__"));
459 if (IS_ERR(dentry)) {
460 rc = PTR_ERR(dentry);
461 CERROR("cannot lookup __iopen__ directory: rc = %d\n", rc);
465 mds->mds_fid_de = dentry;
466 if (!dentry->d_inode || is_bad_inode(dentry->d_inode)) {
468 CERROR("__iopen__ directory has no inode? rc = %d\n", rc);
472 dentry = simple_mkdir(current->fs->pwd, "PENDING", 0777, 1);
473 if (IS_ERR(dentry)) {
474 rc = PTR_ERR(dentry);
475 CERROR("cannot create PENDING directory: rc = %d\n", rc);
478 mds->mds_pending_dir = dentry;
481 dentry = simple_mkdir(current->fs->pwd, MDT_LOGS_DIR, 0777, 1);
482 if (IS_ERR(dentry)) {
483 rc = PTR_ERR(dentry);
484 CERROR("cannot create %s directory: rc = %d\n",
486 GOTO(err_pending, rc);
488 mds->mds_logs_dir = dentry;
491 dentry = simple_mkdir(current->fs->pwd, "OBJECTS", 0777, 1);
492 if (IS_ERR(dentry)) {
493 rc = PTR_ERR(dentry);
494 CERROR("cannot create OBJECTS directory: rc = %d\n", rc);
497 mds->mds_objects_dir = dentry;
499 /* open and test the last rcvd file */
500 file = filp_open(LAST_RCVD, O_RDWR | O_CREAT, 0644);
503 CERROR("cannot open/create %s file: rc = %d\n", LAST_RCVD, rc);
504 GOTO(err_objects, rc = PTR_ERR(file));
506 mds->mds_rcvd_filp = file;
507 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
508 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
509 file->f_dentry->d_inode->i_mode);
510 GOTO(err_last_rcvd, rc = -ENOENT);
513 rc = mds_init_server_data(obd, file);
515 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
516 GOTO(err_last_rcvd, rc);
519 /* open and test the lov objd file */
520 file = filp_open(LOV_OBJID, O_RDWR | O_CREAT, 0644);
523 CERROR("cannot open/create %s file: rc = %d\n", LOV_OBJID, rc);
524 GOTO(err_client, rc = PTR_ERR(file));
526 mds->mds_lov_objid_filp = file;
527 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
528 CERROR("%s is not a regular file!: mode = %o\n", LOV_OBJID,
529 file->f_dentry->d_inode->i_mode);
530 GOTO(err_lov_objid, rc = -ENOENT);
533 /* open and test the check io file junk */
534 file = filp_open(HEALTH_CHECK, O_RDWR | O_CREAT, 0644);
537 CERROR("cannot open/create %s file: rc = %d\n", HEALTH_CHECK, rc);
538 GOTO(err_lov_objid, rc = PTR_ERR(file));
540 mds->mds_health_check_filp = file;
541 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
542 CERROR("%s is not a regular file!: mode = %o\n", HEALTH_CHECK,
543 file->f_dentry->d_inode->i_mode);
544 GOTO(err_health_check, rc = -ENOENT);
546 rc = lvfs_check_io_health(obd, file);
548 GOTO(err_health_check, rc);
550 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
555 if (mds->mds_health_check_filp &&
556 filp_close(mds->mds_health_check_filp, 0))
557 CERROR("can't close %s after error\n", HEALTH_CHECK);
559 if (mds->mds_lov_objid_filp && filp_close(mds->mds_lov_objid_filp, 0))
560 CERROR("can't close %s after error\n", LOV_OBJID);
562 class_disconnect_exports(obd);
564 if (mds->mds_rcvd_filp && filp_close(mds->mds_rcvd_filp, 0))
565 CERROR("can't close %s after error\n", LAST_RCVD);
567 dput(mds->mds_objects_dir);
569 dput(mds->mds_logs_dir);
571 dput(mds->mds_pending_dir);
573 dput(mds->mds_fid_de);
578 int mds_fs_cleanup(struct obd_device *obd)
580 struct mds_obd *mds = &obd->u.mds;
581 struct lvfs_run_ctxt saved;
585 CWARN("%s: shutting down for failover; client state will "
586 "be preserved.\n", obd->obd_name);
588 class_disconnect_exports(obd); /* cleans up client info too */
589 mds_server_free_data(mds);
591 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
592 if (mds->mds_rcvd_filp) {
593 rc = filp_close(mds->mds_rcvd_filp, 0);
594 mds->mds_rcvd_filp = NULL;
596 CERROR("%s file won't close, rc=%d\n", LAST_RCVD, rc);
598 if (mds->mds_lov_objid_filp) {
599 rc = filp_close(mds->mds_lov_objid_filp, 0);
600 mds->mds_lov_objid_filp = NULL;
602 CERROR("%s file won't close, rc=%d\n", LOV_OBJID, rc);
604 if (mds->mds_health_check_filp) {
605 rc = filp_close(mds->mds_health_check_filp, 0);
606 mds->mds_health_check_filp = NULL;
608 CERROR("%s file won't close, rc=%d\n", HEALTH_CHECK, rc);
610 if (mds->mds_objects_dir != NULL) {
611 l_dput(mds->mds_objects_dir);
612 mds->mds_objects_dir = NULL;
614 if (mds->mds_logs_dir) {
615 l_dput(mds->mds_logs_dir);
616 mds->mds_logs_dir = NULL;
618 if (mds->mds_pending_dir) {
619 l_dput(mds->mds_pending_dir);
620 mds->mds_pending_dir = NULL;
623 lquota_fs_cleanup(quota_interface, obd);
625 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
626 shrink_dcache_parent(mds->mds_fid_de);
627 dput(mds->mds_fid_de);
628 LL_DQUOT_OFF(obd->u.obt.obt_sb);
633 /* Creates an object with the same name as its fid. Because this is not at all
634 * performance sensitive, it is accomplished by creating a file, checking the
635 * fid, and renaming it. */
636 int mds_obd_create(struct obd_export *exp, struct obdo *oa,
637 struct lov_stripe_md **ea, struct obd_trans_info *oti)
639 struct mds_obd *mds = &exp->exp_obd->u.mds;
640 struct inode *parent_inode = mds->mds_objects_dir->d_inode;
641 unsigned int tmpname = ll_rand();
643 struct dentry *new_child;
644 struct lvfs_run_ctxt saved;
645 char fidname[LL_FID_NAMELEN];
647 struct lvfs_ucred ucred = { 0 };
648 int rc = 0, err, namelen;
651 /* the owner of object file should always be root */
652 ucred.luc_cap = current->cap_effective | CAP_SYS_RESOURCE;
654 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, &ucred);
656 sprintf(fidname, "OBJECTS/%u.%u", tmpname, current->pid);
657 filp = filp_open(fidname, O_CREAT | O_EXCL, 0666);
661 CERROR("impossible object name collision %u\n",
665 CERROR("error creating tmp object %u: rc %d\n", tmpname, rc);
669 LASSERT(mds->mds_objects_dir == filp->f_dentry->d_parent);
671 oa->o_id = filp->f_dentry->d_inode->i_ino;
672 oa->o_generation = filp->f_dentry->d_inode->i_generation;
673 namelen = ll_fid2str(fidname, oa->o_id, oa->o_generation);
675 down(&parent_inode->i_sem);
676 new_child = lookup_one_len(fidname, mds->mds_objects_dir, namelen);
678 if (IS_ERR(new_child)) {
679 CERROR("getting neg dentry for obj rename: %d\n", rc);
680 GOTO(out_close, rc = PTR_ERR(new_child));
682 if (new_child->d_inode != NULL) {
683 CERROR("impossible non-negative obj dentry " LPU64":%u!\n",
684 oa->o_id, oa->o_generation);
688 handle = fsfilt_start(exp->exp_obd, mds->mds_objects_dir->d_inode,
689 FSFILT_OP_RENAME, NULL);
691 GOTO(out_dput, rc = PTR_ERR(handle));
694 rc = vfs_rename(mds->mds_objects_dir->d_inode, filp->f_dentry,
695 mds->mds_objects_dir->d_inode, new_child);
698 CERROR("error renaming new object "LPU64":%u: rc %d\n",
699 oa->o_id, oa->o_generation, rc);
701 err = fsfilt_commit(exp->exp_obd, mds->mds_objects_dir->d_inode,
704 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGENER;
710 up(&parent_inode->i_sem);
711 err = filp_close(filp, 0);
713 CERROR("closing tmpfile %u: rc %d\n", tmpname, rc);
718 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, &ucred);
722 int mds_obd_destroy(struct obd_export *exp, struct obdo *oa,
723 struct lov_stripe_md *ea, struct obd_trans_info *oti,
724 struct obd_export *md_exp)
726 struct mds_obd *mds = &exp->exp_obd->u.mds;
727 struct inode *parent_inode = mds->mds_objects_dir->d_inode;
728 struct obd_device *obd = exp->exp_obd;
729 struct lvfs_run_ctxt saved;
730 struct lvfs_ucred ucred = { 0 };
731 char fidname[LL_FID_NAMELEN];
734 int err, namelen, rc = 0;
737 ucred.luc_cap = current->cap_effective | CAP_SYS_RESOURCE;
738 push_ctxt(&saved, &obd->obd_lvfs_ctxt, &ucred);
740 namelen = ll_fid2str(fidname, oa->o_id, oa->o_generation);
742 down(&parent_inode->i_sem);
743 de = lookup_one_len(fidname, mds->mds_objects_dir, namelen);
747 CERROR("error looking up object "LPU64" %s: rc %d\n",
748 oa->o_id, fidname, rc);
751 if (de->d_inode == NULL) {
752 CERROR("destroying non-existent object "LPU64" %s: rc %d\n",
753 oa->o_id, fidname, rc);
754 GOTO(out_dput, rc = -ENOENT);
757 /* Stripe count is 1 here since this is some MDS specific stuff
758 that is unlinked, not spanned across multiple OSTs */
759 handle = fsfilt_start_log(obd, mds->mds_objects_dir->d_inode,
760 FSFILT_OP_UNLINK, oti, 1);
763 GOTO(out_dput, rc = PTR_ERR(handle));
765 rc = vfs_unlink(mds->mds_objects_dir->d_inode, de);
767 CERROR("error destroying object "LPU64":%u: rc %d\n",
768 oa->o_id, oa->o_generation, rc);
770 err = fsfilt_commit(obd, mds->mds_objects_dir->d_inode, handle, 0);
776 up(&parent_inode->i_sem);
778 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &ucred);