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 /* This limit is arbitrary (32k clients on x86), but it is convenient to use
53 * 2^n * PAGE_SIZE * 8 for the number of bits that fit an order-n allocation. */
54 #define MDS_MAX_CLIENTS (PAGE_SIZE * 8)
56 #define LAST_RCVD "last_rcvd"
57 #define LOV_OBJID "lov_objid"
58 #define HEALTH_CHECK "health_check"
60 /* Add client data to the MDS. We use a bitmap to locate a free space
61 * in the last_rcvd file if cl_off is -1 (i.e. a new client).
62 * Otherwise, we have just read the data from the last_rcvd file and
65 * It should not be possible to fail adding an existing client - otherwise
66 * mds_init_server_data() callsite needs to be fixed.
68 int mds_client_add(struct obd_device *obd, struct mds_obd *mds,
69 struct mds_export_data *med, int cl_idx)
71 unsigned long *bitmap = mds->mds_client_bitmap;
72 int new_client = (cl_idx == -1);
75 LASSERT(bitmap != NULL);
76 LASSERTF(cl_idx > -2, "%d\n", cl_idx);
78 /* XXX if mcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
79 if (!strcmp(med->med_mcd->mcd_uuid, obd->obd_uuid.uuid))
82 /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
83 * there's no need for extra complication here
86 cl_idx = find_first_zero_bit(bitmap, MDS_MAX_CLIENTS);
88 if (cl_idx >= MDS_MAX_CLIENTS ||
89 OBD_FAIL_CHECK_ONCE(OBD_FAIL_MDS_CLIENT_ADD)) {
90 CERROR("no room for clients - fix MDS_MAX_CLIENTS\n");
93 if (test_and_set_bit(cl_idx, bitmap)) {
94 cl_idx = find_next_zero_bit(bitmap, MDS_MAX_CLIENTS,
99 if (test_and_set_bit(cl_idx, bitmap)) {
100 CERROR("MDS client %d: bit already set in bitmap!!\n",
106 CDEBUG(D_INFO, "client at idx %d with UUID '%s' added\n",
107 cl_idx, med->med_mcd->mcd_uuid);
109 med->med_lr_idx = cl_idx;
110 med->med_lr_off = le32_to_cpu(mds->mds_server_data->lsd_client_start) +
111 (cl_idx * le16_to_cpu(mds->mds_server_data->lsd_client_size));
112 LASSERTF(med->med_lr_off > 0, "med_lr_off = %llu\n", med->med_lr_off);
115 struct lvfs_run_ctxt saved;
116 loff_t off = med->med_lr_off;
117 struct file *file = mds->mds_rcvd_filp;
120 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
121 rc = fsfilt_write_record(obd, file, med->med_mcd,
122 sizeof(*med->med_mcd), &off, 1);
123 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
127 CDEBUG(D_INFO, "wrote client mcd at idx %u off %llu (len %u)\n",
128 med->med_lr_idx, med->med_lr_off,
129 (unsigned int)sizeof(*med->med_mcd));
134 int mds_client_free(struct obd_export *exp)
136 struct mds_export_data *med = &exp->exp_mds_data;
137 struct mds_obd *mds = &exp->exp_obd->u.mds;
138 struct obd_device *obd = exp->exp_obd;
139 struct mds_client_data zero_mcd;
140 struct lvfs_run_ctxt saved;
148 /* XXX if mcd_uuid were a real obd_uuid, I could use obd_uuid_equals */
149 if (!strcmp(med->med_mcd->mcd_uuid, obd->obd_uuid.uuid))
152 CDEBUG(D_INFO, "freeing client at idx %u, offset %lld with UUID '%s'\n",
153 med->med_lr_idx, med->med_lr_off, med->med_mcd->mcd_uuid);
155 LASSERT(mds->mds_client_bitmap != NULL);
157 off = med->med_lr_off;
159 /* Don't clear med_lr_idx here as it is likely also unset. At worst
160 * we leak a client slot that will be cleaned on the next recovery. */
162 CERROR("%s: client idx %d has offset %lld\n",
163 obd->obd_name, med->med_lr_idx, off);
164 GOTO(free, rc = -EINVAL);
167 /* Clear the bit _after_ zeroing out the client so we don't
168 race with mds_client_add and zero out new clients.*/
169 if (!test_bit(med->med_lr_idx, mds->mds_client_bitmap)) {
170 CERROR("MDS client %u: bit already clear in bitmap!!\n",
175 if (!(exp->exp_flags & OBD_OPT_FAILOVER)) {
176 memset(&zero_mcd, 0, sizeof zero_mcd);
177 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
178 rc = fsfilt_write_record(obd, mds->mds_rcvd_filp, &zero_mcd,
179 sizeof(zero_mcd), &off, 1);
180 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
182 CDEBUG(rc == 0 ? D_INFO : D_ERROR,
183 "zeroing out client %s idx %u in %s rc %d\n",
184 med->med_mcd->mcd_uuid, med->med_lr_idx, LAST_RCVD, rc);
187 if (!test_and_clear_bit(med->med_lr_idx, mds->mds_client_bitmap)) {
188 CERROR("MDS client %u: bit already clear in bitmap!!\n",
194 /* Make sure the server's last_transno is up to date. Do this
195 * after the client is freed so we know all the client's
196 * transactions have been committed. */
197 mds_update_server_data(exp->exp_obd, 0);
201 OBD_FREE(med->med_mcd, sizeof(*med->med_mcd));
207 static int mds_server_free_data(struct mds_obd *mds)
209 OBD_FREE(mds->mds_client_bitmap, MDS_MAX_CLIENTS / 8);
210 OBD_FREE(mds->mds_server_data, sizeof(*mds->mds_server_data));
211 mds->mds_server_data = NULL;
216 static int mds_init_server_data(struct obd_device *obd, struct file *file)
218 struct mds_obd *mds = &obd->u.mds;
219 struct lr_server_data *lsd;
220 struct mds_client_data *mcd = NULL;
222 unsigned long last_rcvd_size = file->f_dentry->d_inode->i_size;
227 /* ensure padding in the struct is the correct size */
228 LASSERT(offsetof(struct lr_server_data, lsd_padding) +
229 sizeof(lsd->lsd_padding) == LR_SERVER_SIZE);
230 LASSERT(offsetof(struct mds_client_data, mcd_padding) +
231 sizeof(mcd->mcd_padding) == MDS_LR_CLIENT_SIZE);
232 LASSERT(MDS_LR_CLIENT_SIZE == LR_CLIENT_SIZE);
233 LASSERT(MDS_LR_CLIENT_START == LR_CLIENT_START);
235 OBD_ALLOC_WAIT(lsd, sizeof(*lsd));
239 OBD_ALLOC_WAIT(mds->mds_client_bitmap, MDS_MAX_CLIENTS / 8);
240 if (!mds->mds_client_bitmap) {
241 OBD_FREE(lsd, sizeof(*lsd));
245 mds->mds_server_data = lsd;
247 if (last_rcvd_size == 0) {
248 CWARN("%s: initializing new %s\n", obd->obd_name, LAST_RCVD);
250 memcpy(lsd->lsd_uuid, obd->obd_uuid.uuid,sizeof(lsd->lsd_uuid));
251 lsd->lsd_last_transno = 0;
252 mount_count = lsd->lsd_mount_count = 0;
253 lsd->lsd_server_size = cpu_to_le32(MDS_LR_SERVER_SIZE);
254 lsd->lsd_client_start = cpu_to_le32(MDS_LR_CLIENT_START);
255 lsd->lsd_client_size = cpu_to_le16(MDS_LR_CLIENT_SIZE);
256 lsd->lsd_feature_rocompat = cpu_to_le32(MDS_ROCOMPAT_LOVOBJID);
258 rc = fsfilt_read_record(obd, file, lsd, sizeof(*lsd), &off);
260 CERROR("error reading MDS %s: rc = %d\n", LAST_RCVD, rc);
263 if (strcmp(lsd->lsd_uuid, obd->obd_uuid.uuid) != 0) {
264 CERROR("OBD UUID %s does not match last_rcvd UUID %s\n",
265 obd->obd_uuid.uuid, lsd->lsd_uuid);
266 GOTO(err_msd, rc = -EINVAL);
268 mount_count = le64_to_cpu(lsd->lsd_mount_count);
270 if (lsd->lsd_feature_incompat & ~cpu_to_le32(MDS_INCOMPAT_SUPP)) {
271 CERROR("unsupported incompat feature %x\n",
272 le32_to_cpu(lsd->lsd_feature_incompat) &
274 GOTO(err_msd, rc = -EINVAL);
277 if (lsd->lsd_feature_rocompat & ~cpu_to_le32(MDS_ROCOMPAT_SUPP)) {
278 CERROR("unsupported read-only feature %x\n",
279 le32_to_cpu(lsd->lsd_feature_rocompat) &
281 /* Do something like remount filesystem read-only */
282 GOTO(err_msd, rc = -EINVAL);
285 if (lsd->lsd_feature_compat & ~cpu_to_le32(LR_COMPAT_COMMON_LR)) {
286 CERROR("old last_rcvd format, updating\n");
287 lsd->lsd_mount_count = lsd->lsd_last_transno; //msd->msd_mount_count
288 lsd->lsd_last_transno = lsd->lsd_unused; //msd->msd_last_transno;
289 lsd->lsd_feature_compat |= cpu_to_le32(LR_COMPAT_COMMON_LR);
290 GOTO(err_msd, rc = -EINVAL);
293 mds->mds_last_transno = le64_to_cpu(lsd->lsd_last_transno);
295 CDEBUG(D_INODE, "%s: server last_transno: "LPU64"\n",
296 obd->obd_name, mds->mds_last_transno);
297 CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
298 obd->obd_name, mount_count + 1);
299 CDEBUG(D_INODE, "%s: server data size: %u\n",
300 obd->obd_name, le32_to_cpu(lsd->lsd_server_size));
301 CDEBUG(D_INODE, "%s: per-client data start: %u\n",
302 obd->obd_name, le32_to_cpu(lsd->lsd_client_start));
303 CDEBUG(D_INODE, "%s: per-client data size: %u\n",
304 obd->obd_name, le32_to_cpu(lsd->lsd_client_size));
305 CDEBUG(D_INODE, "%s: last_rcvd size: %lu\n",
306 obd->obd_name, last_rcvd_size);
307 CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
308 last_rcvd_size <= le32_to_cpu(lsd->lsd_client_start) ? 0 :
309 (last_rcvd_size - le32_to_cpu(lsd->lsd_client_start)) /
310 le16_to_cpu(lsd->lsd_client_size));
312 if (!lsd->lsd_server_size || !lsd->lsd_client_start ||
313 !lsd->lsd_client_size) {
314 CERROR("Bad last_rcvd contents!\n");
315 GOTO(err_msd, rc = -EINVAL);
318 /* When we do a clean MDS shutdown, we save the last_transno into
319 * the header. If we find clients with higher last_transno values
320 * then those clients may need recovery done. */
321 for (cl_idx = 0, off = le32_to_cpu(lsd->lsd_client_start);
322 off < last_rcvd_size; cl_idx++) {
324 struct obd_export *exp;
325 struct mds_export_data *med;
328 OBD_ALLOC_WAIT(mcd, sizeof(*mcd));
330 GOTO(err_client, rc = -ENOMEM);
333 /* Don't assume off is incremented properly by
334 * fsfilt_read_record(), in case sizeof(*mcd)
335 * isn't the same as lsd->lsd_client_size. */
336 off = le32_to_cpu(lsd->lsd_client_start) +
337 cl_idx * le16_to_cpu(lsd->lsd_client_size);
338 rc = fsfilt_read_record(obd, file, mcd, sizeof(*mcd), &off);
340 CERROR("error reading MDS %s idx %d, off %llu: rc %d\n",
341 LAST_RCVD, cl_idx, off, rc);
342 break; /* read error shouldn't cause startup to fail */
345 if (mcd->mcd_uuid[0] == '\0') {
346 CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
351 last_transno = le64_to_cpu(mcd->mcd_last_transno);
353 /* These exports are cleaned up by mds_disconnect(), so they
354 * need to be set up like real exports as mds_connect() does.
356 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
357 " srv lr: "LPU64" lx: "LPU64"\n", mcd->mcd_uuid, cl_idx,
358 last_transno, le64_to_cpu(lsd->lsd_last_transno),
359 le64_to_cpu(mcd->mcd_last_xid));
361 exp = class_new_export(obd);
363 GOTO(err_client, rc = -ENOMEM);
365 memcpy(&exp->exp_client_uuid.uuid, mcd->mcd_uuid,
366 sizeof exp->exp_client_uuid.uuid);
367 med = &exp->exp_mds_data;
369 rc = mds_client_add(obd, mds, med, cl_idx);
370 LASSERTF(rc == 0, "rc = %d\n", rc); /* can't fail existing */
372 /* create helper if export init gets more complex */
373 INIT_LIST_HEAD(&med->med_open_head);
374 spin_lock_init(&med->med_open_lock);
377 exp->exp_replay_needed = 1;
378 obd->obd_recoverable_clients++;
379 obd->obd_max_recoverable_clients++;
380 class_export_put(exp);
382 CDEBUG(D_OTHER, "client at idx %d has last_transno = "LPU64"\n",
383 cl_idx, last_transno);
385 if (last_transno > mds->mds_last_transno)
386 mds->mds_last_transno = last_transno;
390 OBD_FREE(mcd, sizeof(*mcd));
392 obd->obd_last_committed = mds->mds_last_transno;
394 if (obd->obd_recoverable_clients) {
395 CWARN("RECOVERY: service %s, %d recoverable clients, "
396 "last_transno "LPU64"\n", obd->obd_name,
397 obd->obd_recoverable_clients, mds->mds_last_transno);
398 obd->obd_next_recovery_transno = obd->obd_last_committed + 1;
399 obd->obd_recovering = 1;
400 obd->obd_recovery_start = CURRENT_SECONDS;
401 /* Only used for lprocfs_status */
402 obd->obd_recovery_end = obd->obd_recovery_start +
403 OBD_RECOVERY_TIMEOUT / HZ;
406 mds->mds_mount_count = mount_count + 1;
407 lsd->lsd_mount_count = cpu_to_le64(mds->mds_mount_count);
409 /* save it, so mount count and last_transno is current */
410 rc = mds_update_server_data(obd, 1);
412 GOTO(err_client, rc);
417 class_disconnect_exports(obd);
419 mds_server_free_data(mds);
423 int mds_fs_setup(struct obd_device *obd, struct vfsmount *mnt)
425 struct mds_obd *mds = &obd->u.mds;
426 struct lvfs_run_ctxt saved;
427 struct dentry *dentry;
432 rc = cleanup_group_info();
436 mds->mds_vfsmnt = mnt;
437 mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
439 fsfilt_setup(obd, mds->mds_sb);
441 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
442 obd->obd_lvfs_ctxt.pwdmnt = mnt;
443 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
444 obd->obd_lvfs_ctxt.fs = get_ds();
445 obd->obd_lvfs_ctxt.cb_ops = mds_lvfs_ops;
447 /* setup the directory tree */
448 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
449 dentry = simple_mkdir(current->fs->pwd, "ROOT", 0755, 0);
450 if (IS_ERR(dentry)) {
451 rc = PTR_ERR(dentry);
452 CERROR("cannot create ROOT directory: rc = %d\n", rc);
456 mds->mds_rootfid.id = dentry->d_inode->i_ino;
457 mds->mds_rootfid.generation = dentry->d_inode->i_generation;
458 mds->mds_rootfid.f_type = S_IFDIR;
462 dentry = lookup_one_len("__iopen__", current->fs->pwd,
463 strlen("__iopen__"));
464 if (IS_ERR(dentry)) {
465 rc = PTR_ERR(dentry);
466 CERROR("cannot lookup __iopen__ directory: rc = %d\n", rc);
470 mds->mds_fid_de = dentry;
471 if (!dentry->d_inode || is_bad_inode(dentry->d_inode)) {
473 CERROR("__iopen__ directory has no inode? rc = %d\n", rc);
477 dentry = simple_mkdir(current->fs->pwd, "PENDING", 0777, 1);
478 if (IS_ERR(dentry)) {
479 rc = PTR_ERR(dentry);
480 CERROR("cannot create PENDING directory: rc = %d\n", rc);
483 mds->mds_pending_dir = dentry;
485 dentry = simple_mkdir(current->fs->pwd, MOUNT_CONFIGS_DIR, 0777, 1);
486 if (IS_ERR(dentry)) {
487 rc = PTR_ERR(dentry);
488 CERROR("cannot create %s directory: rc = %d\n",
489 MOUNT_CONFIGS_DIR, rc);
490 GOTO(err_pending, rc);
492 mds->mds_logs_dir = dentry;
494 dentry = simple_mkdir(current->fs->pwd, "OBJECTS", 0777, 1);
495 if (IS_ERR(dentry)) {
496 rc = PTR_ERR(dentry);
497 CERROR("cannot create OBJECTS directory: rc = %d\n", rc);
500 mds->mds_objects_dir = dentry;
502 /* open and test the last rcvd file */
503 file = filp_open(LAST_RCVD, O_RDWR | O_CREAT, 0644);
506 CERROR("cannot open/create %s file: rc = %d\n", LAST_RCVD, rc);
507 GOTO(err_objects, rc = PTR_ERR(file));
509 mds->mds_rcvd_filp = file;
510 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
511 CERROR("%s is not a regular file!: mode = %o\n", LAST_RCVD,
512 file->f_dentry->d_inode->i_mode);
513 GOTO(err_last_rcvd, rc = -ENOENT);
516 rc = mds_init_server_data(obd, file);
518 CERROR("cannot read %s: rc = %d\n", LAST_RCVD, rc);
519 GOTO(err_last_rcvd, rc);
522 /* open and test the lov objd file */
523 file = filp_open(LOV_OBJID, O_RDWR | O_CREAT, 0644);
526 CERROR("cannot open/create %s file: rc = %d\n", LOV_OBJID, rc);
527 GOTO(err_client, rc = PTR_ERR(file));
529 mds->mds_lov_objid_filp = file;
530 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
531 CERROR("%s is not a regular file!: mode = %o\n", LOV_OBJID,
532 file->f_dentry->d_inode->i_mode);
533 GOTO(err_lov_objid, rc = -ENOENT);
536 /* open and test the check io file junk */
537 file = filp_open(HEALTH_CHECK, O_RDWR | O_CREAT, 0644);
540 CERROR("cannot open/create %s file: rc = %d\n", HEALTH_CHECK, rc);
541 GOTO(err_lov_objid, rc = PTR_ERR(file));
543 mds->mds_health_check_filp = file;
544 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
545 CERROR("%s is not a regular file!: mode = %o\n", HEALTH_CHECK,
546 file->f_dentry->d_inode->i_mode);
547 GOTO(err_health_check, rc = -ENOENT);
549 rc = lvfs_check_io_health(obd, file);
551 GOTO(err_health_check, rc);
553 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
558 if (mds->mds_health_check_filp &&
559 filp_close(mds->mds_health_check_filp, 0))
560 CERROR("can't close %s after error\n", HEALTH_CHECK);
562 if (mds->mds_lov_objid_filp && filp_close(mds->mds_lov_objid_filp, 0))
563 CERROR("can't close %s after error\n", LOV_OBJID);
565 class_disconnect_exports(obd);
567 if (mds->mds_rcvd_filp && filp_close(mds->mds_rcvd_filp, 0))
568 CERROR("can't close %s after error\n", LAST_RCVD);
570 dput(mds->mds_objects_dir);
572 dput(mds->mds_logs_dir);
574 dput(mds->mds_pending_dir);
576 dput(mds->mds_fid_de);
581 int mds_fs_cleanup(struct obd_device *obd)
583 struct mds_obd *mds = &obd->u.mds;
584 struct lvfs_run_ctxt saved;
588 CERROR("%s: shutting down for failover; client state will"
589 " be preserved.\n", obd->obd_name);
591 class_disconnect_exports(obd); /* cleans up client info too */
592 mds_server_free_data(mds);
594 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
595 if (mds->mds_rcvd_filp) {
596 rc = filp_close(mds->mds_rcvd_filp, 0);
597 mds->mds_rcvd_filp = NULL;
599 CERROR("%s file won't close, rc=%d\n", LAST_RCVD, rc);
601 if (mds->mds_lov_objid_filp) {
602 rc = filp_close(mds->mds_lov_objid_filp, 0);
603 mds->mds_lov_objid_filp = NULL;
605 CERROR("%s file won't close, rc=%d\n", LOV_OBJID, rc);
607 if (mds->mds_health_check_filp) {
608 rc = filp_close(mds->mds_health_check_filp, 0);
609 mds->mds_health_check_filp = NULL;
611 CERROR("%s file won't close, rc=%d\n", HEALTH_CHECK, rc);
613 if (mds->mds_objects_dir != NULL) {
614 l_dput(mds->mds_objects_dir);
615 mds->mds_objects_dir = NULL;
617 if (mds->mds_logs_dir) {
618 l_dput(mds->mds_logs_dir);
619 mds->mds_logs_dir = NULL;
621 if (mds->mds_pending_dir) {
622 l_dput(mds->mds_pending_dir);
623 mds->mds_pending_dir = NULL;
626 mds_fs_quota_cleanup(mds);
628 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
629 shrink_dcache_parent(mds->mds_fid_de);
630 dput(mds->mds_fid_de);
631 LL_DQUOT_OFF(mds->mds_sb);
636 /* Creates an object with the same name as its fid. Because this is not at all
637 * performance sensitive, it is accomplished by creating a file, checking the
638 * fid, and renaming it. */
639 int mds_obd_create(struct obd_export *exp, struct obdo *oa,
640 struct lov_stripe_md **ea, struct obd_trans_info *oti)
642 struct mds_obd *mds = &exp->exp_obd->u.mds;
643 struct inode *parent_inode = mds->mds_objects_dir->d_inode;
644 unsigned int tmpname = ll_insecure_random_int();
646 struct dentry *new_child;
647 struct lvfs_run_ctxt saved;
648 char fidname[LL_FID_NAMELEN];
650 struct lvfs_ucred ucred = { 0 };
651 int rc = 0, err, namelen;
654 /* the owner of object file should always be root */
655 ucred.luc_cap = current->cap_effective | CAP_SYS_RESOURCE;
657 push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, &ucred);
659 sprintf(fidname, "OBJECTS/%u.%u", tmpname, current->pid);
660 filp = filp_open(fidname, O_CREAT | O_EXCL, 0666);
664 CERROR("impossible object name collision %u\n",
668 CERROR("error creating tmp object %u: rc %d\n", tmpname, rc);
672 LASSERT(mds->mds_objects_dir == filp->f_dentry->d_parent);
674 oa->o_id = filp->f_dentry->d_inode->i_ino;
675 oa->o_generation = filp->f_dentry->d_inode->i_generation;
676 namelen = ll_fid2str(fidname, oa->o_id, oa->o_generation);
678 down(&parent_inode->i_sem);
679 new_child = lookup_one_len(fidname, mds->mds_objects_dir, namelen);
681 if (IS_ERR(new_child)) {
682 CERROR("getting neg dentry for obj rename: %d\n", rc);
683 GOTO(out_close, rc = PTR_ERR(new_child));
685 if (new_child->d_inode != NULL) {
686 CERROR("impossible non-negative obj dentry " LPU64":%u!\n",
687 oa->o_id, oa->o_generation);
691 handle = fsfilt_start(exp->exp_obd, mds->mds_objects_dir->d_inode,
692 FSFILT_OP_RENAME, NULL);
694 GOTO(out_dput, rc = PTR_ERR(handle));
697 rc = vfs_rename(mds->mds_objects_dir->d_inode, filp->f_dentry,
698 mds->mds_objects_dir->d_inode, new_child);
701 CERROR("error renaming new object "LPU64":%u: rc %d\n",
702 oa->o_id, oa->o_generation, rc);
704 err = fsfilt_commit(exp->exp_obd, mds->mds_objects_dir->d_inode,
707 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGENER;
713 up(&parent_inode->i_sem);
714 err = filp_close(filp, 0);
716 CERROR("closing tmpfile %u: rc %d\n", tmpname, rc);
721 pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, &ucred);
725 int mds_obd_destroy(struct obd_export *exp, struct obdo *oa,
726 struct lov_stripe_md *ea, struct obd_trans_info *oti)
728 struct mds_obd *mds = &exp->exp_obd->u.mds;
729 struct inode *parent_inode = mds->mds_objects_dir->d_inode;
730 struct obd_device *obd = exp->exp_obd;
731 struct lvfs_run_ctxt saved;
732 struct lvfs_ucred ucred = { 0 };
733 char fidname[LL_FID_NAMELEN];
736 int err, namelen, rc = 0;
739 ucred.luc_cap = current->cap_effective | CAP_SYS_RESOURCE;
740 push_ctxt(&saved, &obd->obd_lvfs_ctxt, &ucred);
742 namelen = ll_fid2str(fidname, oa->o_id, oa->o_generation);
744 down(&parent_inode->i_sem);
745 de = lookup_one_len(fidname, mds->mds_objects_dir, namelen);
749 CERROR("error looking up object "LPU64" %s: rc %d\n",
750 oa->o_id, fidname, rc);
753 if (de->d_inode == NULL) {
754 CERROR("destroying non-existent object "LPU64" %s: rc %d\n",
755 oa->o_id, fidname, rc);
756 GOTO(out_dput, rc = -ENOENT);
759 /* Stripe count is 1 here since this is some MDS specific stuff
760 that is unlinked, not spanned across multiple OSTs */
761 handle = fsfilt_start_log(obd, mds->mds_objects_dir->d_inode,
762 FSFILT_OP_UNLINK, oti, 1);
765 GOTO(out_dput, rc = PTR_ERR(handle));
767 rc = vfs_unlink(mds->mds_objects_dir->d_inode, de);
769 CERROR("error destroying object "LPU64":%u: rc %d\n",
770 oa->o_id, oa->o_generation, rc);
772 err = fsfilt_commit(obd, mds->mds_objects_dir->d_inode, handle, 0);
778 up(&parent_inode->i_sem);
780 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &ucred);