1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
5 * Lustre Metadata Server (mds) handling of striped file data
7 * Copyright (C) 2001-2003 Cluster File Systems, Inc.
9 * This file is part of Lustre, http://www.lustre.org.
11 * Lustre is free software; you can redistribute it and/or
12 * modify it under the terms of version 2 of the GNU General Public
13 * License as published by the Free Software Foundation.
15 * Lustre is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
20 * You should have received a copy of the GNU General Public License
21 * along with Lustre; if not, write to the Free Software
22 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
26 # define EXPORT_SYMTAB
28 #define DEBUG_SUBSYSTEM S_MDS
30 #include <linux/module.h>
31 #include <linux/lustre_mds.h>
32 #include <linux/lustre_idl.h>
33 #include <linux/obd_class.h>
34 #include <linux/obd_lov.h>
35 #include <linux/lustre_lib.h>
36 #include <linux/lustre_fsfilt.h>
38 #include "mds_internal.h"
43 * - magic in mea struct
44 * - error handling is totally missed
47 int mds_lmv_connect(struct obd_device *obd, char * lmv_name)
49 struct mds_obd *mds = &obd->u.mds;
50 struct lustre_handle conn = {0};
51 int rc, valsize, value;
54 if (IS_ERR(mds->mds_lmv_obd))
55 RETURN(PTR_ERR(mds->mds_lmv_obd));
60 mds->mds_lmv_obd = class_name2obd(lmv_name);
61 if (!mds->mds_lmv_obd) {
62 CERROR("MDS cannot locate LMV %s\n",
64 mds->mds_lmv_obd = ERR_PTR(-ENOTCONN);
68 rc = obd_connect(&conn, mds->mds_lmv_obd, &obd->obd_uuid);
70 CERROR("MDS cannot connect to LMV %s (%d)\n",
72 mds->mds_lmv_obd = ERR_PTR(rc);
75 mds->mds_lmv_exp = class_conn2export(&conn);
76 if (mds->mds_lmv_exp == NULL)
77 CERROR("can't get export!\n");
79 rc = obd_register_observer(mds->mds_lmv_obd, obd);
81 CERROR("MDS cannot register as observer of LMV %s, "
82 "rc = %d\n", lmv_name, rc);
86 /* retrieve size of EA */
87 rc = obd_get_info(mds->mds_lmv_exp, strlen("mdsize"), "mdsize",
92 if (value > mds->mds_max_mdsize)
93 mds->mds_max_mdsize = value;
95 /* find our number in LMV cluster */
96 rc = obd_get_info(mds->mds_lmv_exp, strlen("mdsnum"), "mdsnum",
101 mds->mds_num = value;
103 rc = obd_set_info(mds->mds_lmv_exp, strlen("inter_mds"),
104 "inter_mds", 0, NULL);
111 obd_register_observer(mds->mds_lmv_obd, NULL);
113 obd_disconnect(mds->mds_lmv_exp, 0);
114 mds->mds_lmv_exp = NULL;
115 mds->mds_lmv_obd = ERR_PTR(rc);
119 int mds_lmv_postsetup(struct obd_device *obd)
121 struct mds_obd *mds = &obd->u.mds;
125 if (mds->mds_lmv_exp)
126 rc = obd_init_ea_size(mds->mds_lmv_exp, mds->mds_max_mdsize,
127 mds->mds_max_cookiesize);
132 int mds_lmv_disconnect(struct obd_device *obd, int flags)
134 struct mds_obd *mds = &obd->u.mds;
138 if (!IS_ERR(mds->mds_lmv_obd) && mds->mds_lmv_exp != NULL) {
139 obd_register_observer(mds->mds_lmv_obd, NULL);
141 /* if obd_disconnect fails (probably because the export was
142 * disconnected by class_disconnect_exports) then we just need
143 * to drop our ref. */
144 rc = obd_disconnect(mds->mds_lmv_exp, flags);
146 class_export_put(mds->mds_lmv_exp);
148 mds->mds_lmv_exp = NULL;
149 mds->mds_lmv_obd = NULL;
155 int mds_get_lmv_attr(struct obd_device *obd, struct inode *inode,
156 struct mea **mea, int *mea_size)
158 struct mds_obd *mds = &obd->u.mds;
162 if (!mds->mds_lmv_obd)
164 if (!S_ISDIR(inode->i_mode))
167 /* first calculate mea size */
168 *mea_size = obd_alloc_diskmd(mds->mds_lmv_exp,
169 (struct lov_mds_md **)mea);
170 if (*mea_size < 0 || *mea == NULL)
171 return *mea_size < 0 ? *mea_size : -EINVAL;
173 rc = mds_get_md(obd, inode, *mea, mea_size, 1);
176 OBD_FREE(*mea, *mea_size);
193 #define DIR_ROUND (DIR_PAD - 1)
194 #define DIR_REC_LEN(name_len) (((name_len) + 12 + DIR_ROUND) & ~DIR_ROUND)
196 /* this struct holds dir entries for particular MDS to be flushed */
198 struct list_head list;
203 struct brw_page brwc;
206 struct dirsplit_control {
207 struct obd_device *obd;
209 struct dentry *dentry;
211 struct dir_cache *cache;
214 static int dc_new_page_to_cache(struct dir_cache * dirc)
218 if (!list_empty(&dirc->list) && dirc->free > sizeof(__u16)) {
219 /* current page became full, mark the end */
220 struct dir_entry *de = dirc->cur;
224 page = alloc_page(GFP_KERNEL);
227 list_add_tail(&page->list, &dirc->list);
228 dirc->cur = page_address(page);
229 dirc->free = PAGE_SIZE;
233 static int retrieve_generation_numbers(struct dirsplit_control *dc, void *buf)
235 struct mds_obd *mds = &dc->obd->u.mds;
236 struct dir_entry *de;
237 struct dentry *dentry;
240 end = buf + PAGE_SIZE;
241 de = (struct dir_entry *) buf;
242 while ((char *) de < end && de->namelen) {
243 /* lookup an inode */
244 LASSERT(de->namelen <= 255);
245 dentry = ll_lookup_one_len(de->name, dc->dentry, de->namelen);
246 if (IS_ERR(dentry)) {
247 CERROR("can't lookup %*s: %d\n", de->namelen,
248 de->name, (int) PTR_ERR(dentry));
251 if (dentry->d_inode != NULL) {
252 de->mds = mds->mds_num;
253 de->ino = dentry->d_inode->i_ino;
254 de->generation = dentry->d_inode->i_generation;
255 } else if (dentry->d_flags & DCACHE_CROSS_REF) {
256 de->mds = dentry->d_mdsnum;
257 de->ino = dentry->d_inum;
258 de->generation = dentry->d_generation;
260 CERROR("can't lookup %*s\n", de->namelen, de->name);
266 de = (struct dir_entry *)
267 ((char *) de + DIR_REC_LEN(de->namelen));
272 static int flush_buffer_onto_mds(struct dirsplit_control *dc, int mdsnum)
274 struct mds_obd *mds = &dc->obd->u.mds;
275 struct list_head *cur, *tmp;
276 struct dir_cache *ca;
279 ca = dc->cache + mdsnum;
281 if (ca->free > sizeof(__u16)) {
282 /* current page became full, mark the end */
283 struct dir_entry *de = ca->cur;
287 list_for_each_safe(cur, tmp, &ca->list) {
290 page = list_entry(cur, struct page, list);
291 LASSERT(page != NULL);
293 retrieve_generation_numbers(dc, page_address(page));
297 ca->brwc.count = PAGE_SIZE;
299 ca->oa.o_mds = mdsnum;
300 rc = obd_brw(OBD_BRW_WRITE, mds->mds_lmv_exp, &ca->oa,
301 (struct lov_stripe_md *) dc->mea,
310 static int remove_entries_from_orig_dir(struct dirsplit_control *dc, int mdsnum)
312 struct list_head *cur, *tmp;
313 struct dentry *dentry;
314 struct dir_cache *ca;
315 struct dir_entry *de;
321 ca = dc->cache + mdsnum;
322 list_for_each_safe(cur, tmp, &ca->list) {
323 page = list_entry(cur, struct page, list);
324 buf = page_address(page);
325 end = buf + PAGE_SIZE;
327 de = (struct dir_entry *) buf;
328 while ((char *) de < end && de->namelen) {
329 /* lookup an inode */
330 LASSERT(de->namelen <= 255);
332 dentry = ll_lookup_one_len(de->name, dc->dentry,
334 if (IS_ERR(dentry)) {
335 CERROR("can't lookup %*s: %d\n", de->namelen,
336 de->name, (int) PTR_ERR(dentry));
339 rc = fsfilt_del_dir_entry(dc->obd, dentry);
342 de = (struct dir_entry *)
343 ((char *) de + DIR_REC_LEN(de->namelen));
349 static int filldir(void * __buf, const char * name, int namlen, loff_t offset,
350 ino_t ino, unsigned int d_type)
352 struct dirsplit_control *dc = __buf;
353 struct mds_obd *mds = &dc->obd->u.mds;
354 struct dir_cache *ca;
355 struct dir_entry *de;
360 if (name[0] == '.' && (namlen == 1 ||
361 (namlen == 2 && name[1] == '.'))) {
362 /* skip special entries */
367 newmds = mea_name2idx(dc->mea, (char *) name, namlen);
369 if (newmds == mds->mds_num) {
370 /* this entry remains on the current MDS, skip moving */
374 OBD_ALLOC(n, namlen + 1);
375 memcpy(n, name, namlen);
376 n[namlen] = (char) 0;
378 OBD_FREE(n, namlen + 1);
380 /* check for space in buffer for new entry */
381 ca = dc->cache + newmds;
382 if (DIR_REC_LEN(namlen) > ca->free) {
383 int err = dc_new_page_to_cache(ca);
387 /* insert found entry into buffer to be flushed later */
388 /* NOTE: we'll fill generations number later, because we
389 * it's stored in inode, thus we need to lookup an entry,
390 * but directory is locked for readdir(), so we delay this */
394 de->namelen = namlen;
395 memcpy(de->name, name, namlen);
396 ca->cur += DIR_REC_LEN(namlen);
397 ca->free -= DIR_REC_LEN(namlen);
403 int scan_and_distribute(struct obd_device *obd, struct dentry *dentry,
406 struct inode *dir = dentry->d_inode;
407 struct dirsplit_control dc;
412 nlen = strlen("__iopen__/") + 10 + 1;
413 OBD_ALLOC(file_name, nlen);
416 i = sprintf(file_name, "__iopen__/0x%lx", dentry->d_inode->i_ino);
418 file = filp_open(file_name, O_RDONLY, 0);
420 CERROR("can't open directory %s: %d\n",
421 file_name, (int) PTR_ERR(file));
422 OBD_FREE(file_name, nlen);
423 RETURN(PTR_ERR(file));
426 memset(&dc, 0, sizeof(dc));
431 OBD_ALLOC(dc.cache, sizeof(struct dir_cache) * mea->mea_count);
432 LASSERT(dc.cache != NULL);
433 for (i = 0; i < mea->mea_count; i++) {
434 INIT_LIST_HEAD(&dc.cache[i].list);
435 dc.cache[i].free = 0;
436 dc.cache[i].cached = 0;
439 err = vfs_readdir(file, filldir, &dc);
444 for (i = 0; i < mea->mea_count; i++) {
445 if (!dc.cache[i].cached)
447 err = flush_buffer_onto_mds(&dc, i);
452 for (i = 0; i < mea->mea_count; i++) {
453 if (!dc.cache[i].cached)
455 err = remove_entries_from_orig_dir(&dc, i);
461 for (i = 0; i < mea->mea_count; i++) {
462 struct list_head *cur, *tmp;
463 if (!dc.cache[i].cached)
465 list_for_each_safe(cur, tmp, &dc.cache[i].list) {
467 page = list_entry(cur, struct page, list);
468 list_del(&page->list);
472 OBD_FREE(dc.cache, sizeof(struct dir_cache) * mea->mea_count);
473 OBD_FREE(file_name, nlen);
478 #define MAX_DIR_SIZE (64 * 1024)
480 #define I_NON_SPLITTABLE 256
482 int mds_splitting_expected(struct obd_device *obd, struct dentry *dentry)
484 struct mds_obd *mds = &obd->u.mds;
485 struct mea *mea = NULL;
489 if (!mds->mds_lmv_obd)
490 RETURN(MDS_NO_SPLITTABLE);
493 if (dentry->d_inode == NULL)
494 return MDS_NO_SPLITTABLE;
496 /* a dir can be splitted only */
497 if (!S_ISDIR(dentry->d_inode->i_mode))
498 return MDS_NO_SPLITTABLE;
500 /* already splittied or slave directory (part of splitted dir) */
501 if (dentry->d_inode->i_flags & I_NON_SPLITTABLE)
502 return MDS_NO_SPLITTABLE;
504 /* don't split root directory */
505 if (dentry->d_inode->i_ino == mds->mds_rootfid.id)
506 return MDS_NO_SPLITTABLE;
508 /* large enough to be splitted? */
509 if (dentry->d_inode->i_size < MAX_DIR_SIZE)
510 return MDS_NO_SPLIT_EXPECTED;
512 mds_get_lmv_attr(obd, dentry->d_inode, &mea, &size);
514 /* already splitted or slave object: shouldn't be splitted */
515 rc = MDS_NO_SPLITTABLE;
516 /* mark to skip subsequent checks */
517 dentry->d_inode->i_flags |= I_NON_SPLITTABLE;
519 /* may be splitted */
520 rc = MDS_EXPECT_SPLIT;
529 * must not be called on already splitted directories.
531 int mds_try_to_split_dir(struct obd_device *obd, struct dentry *dentry,
532 struct mea **mea, int nstripes, int update_mode)
534 struct inode *dir = dentry->d_inode;
535 struct mds_obd *mds = &obd->u.mds;
536 struct mea *tmea = NULL;
537 struct obdo *oa = NULL;
538 int rc, mea_size = 0;
542 if (update_mode != LCK_EX)
544 /* TODO: optimization possible - we already may have mea here */
545 rc = mds_splitting_expected(obd, dentry);
546 if (rc == MDS_NO_SPLITTABLE)
548 if (rc == MDS_NO_SPLIT_EXPECTED && nstripes == 0)
550 if (nstripes && nstripes == 1)
553 LASSERT(mea == NULL || *mea == NULL);
555 CDEBUG(D_OTHER, "%s: split directory %u/%lu/%lu\n",
556 obd->obd_name, mds->mds_num, dir->i_ino,
557 (unsigned long) dir->i_generation);
561 mea_size = obd_size_diskmd(mds->mds_lmv_exp, NULL);
563 /* FIXME: Actually we may only want to allocate enough space for
564 * necessary amount of stripes, but on the other hand with this
565 * approach of allocating maximal possible amount of MDS slots,
566 * it would be easier to split the dir over more MDSes */
567 rc = obd_alloc_diskmd(mds->mds_lmv_exp, (void *)mea);
569 CERROR("obd_alloc_diskmd() failed, error %d.\n", rc);
575 (*mea)->mea_count = nstripes;
577 /* 1) create directory objects on slave MDS'es */
578 /* FIXME: should this be OBD method? */
583 oa->o_id = dir->i_ino;
584 oa->o_generation = dir->i_generation;
586 obdo_from_inode(oa, dir, OBD_MD_FLTYPE | OBD_MD_FLATIME |
587 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
588 OBD_MD_FLUID | OBD_MD_FLGID);
590 oa->o_gr = FILTER_GROUP_FIRST_MDS + mds->mds_num;
591 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
592 oa->o_mode = dir->i_mode;
594 CDEBUG(D_OTHER, "%s: create subdirs with mode %o, uid %u, gid %u\n",
595 obd->obd_name, dir->i_mode, dir->i_uid, dir->i_gid);
597 rc = obd_create(mds->mds_lmv_exp, oa,
598 (struct lov_stripe_md **)mea, NULL);
602 CDEBUG(D_OTHER, "%d dirobjects created\n", (int)(*mea)->mea_count);
604 /* 2) update dir attribute */
607 handle = fsfilt_start(obd, dir, FSFILT_OP_SETATTR, NULL);
608 if (IS_ERR(handle)) {
610 CERROR("fsfilt_start() failed: %d\n", (int) PTR_ERR(handle));
611 GOTO(err_oa, rc = PTR_ERR(handle));
614 rc = fsfilt_set_md(obd, dir, handle, *mea, mea_size);
617 CERROR("fsfilt_set_md() failed, error %d.\n", rc);
621 rc = fsfilt_commit(obd, mds->mds_sb, dir, handle, 0);
624 CERROR("fsfilt_commit() failed, error %d.\n", rc);
631 /* 3) read through the dir and distribute it over objects */
632 rc = scan_and_distribute(obd, dentry, *mea);
634 obd_free_diskmd(mds->mds_lmv_exp, (struct lov_mds_md **)mea);
636 CERROR("scan_and_distribute() failed, error %d.\n", rc);
647 static int filter_start_page_write(struct inode *inode,
648 struct niobuf_local *lnb)
650 struct page *page = alloc_pages(GFP_HIGHUSER, 0);
652 CERROR("no memory for a temp page\n");
653 RETURN(lnb->rc = -ENOMEM);
655 POISON_PAGE(page, 0xf1);
656 page->index = lnb->offset >> PAGE_SHIFT;
662 struct dentry *filter_fid2dentry(struct obd_device *obd,
663 struct dentry *dir_dentry,
664 obd_gr group, obd_id id);
666 int mds_preprw(int cmd, struct obd_export *exp, struct obdo *oa,
667 int objcount, struct obd_ioobj *obj,
668 int niocount, struct niobuf_remote *nb,
669 struct niobuf_local *res,
670 struct obd_trans_info *oti)
672 struct mds_obd *mds = &exp->exp_obd->u.mds;
673 struct niobuf_remote *rnb;
674 struct niobuf_local *lnb = NULL;
675 int rc = 0, i, tot_bytes = 0;
676 unsigned long now = jiffies;
677 struct dentry *dentry;
680 LASSERT(objcount == 1);
681 LASSERT(obj->ioo_bufcnt > 0);
683 memset(res, 0, niocount * sizeof(*res));
685 fid.id = obj->ioo_id;
686 fid.generation = obj->ioo_gr;
687 dentry = mds_fid2dentry(mds, &fid, NULL);
688 if (IS_ERR(dentry)) {
689 CERROR("can't get dentry for "LPU64"/%u: %d\n",
690 fid.id, fid.generation, (int) PTR_ERR(dentry));
691 GOTO(cleanup, rc = (int) PTR_ERR(dentry));
694 if (dentry->d_inode == NULL) {
695 CERROR("trying to BRW to non-existent file "LPU64"\n",
698 GOTO(cleanup, rc = -ENOENT);
701 if (time_after(jiffies, now + 15 * HZ))
702 CERROR("slow preprw_write setup %lus\n", (jiffies - now) / HZ);
704 CDEBUG(D_INFO, "preprw_write setup: %lu jiffies\n",
707 for (i = 0, rnb = nb, lnb = res; i < obj->ioo_bufcnt;
709 lnb->dentry = dentry;
710 lnb->offset = rnb->offset;
712 lnb->flags = rnb->flags;
714 rc = filter_start_page_write(dentry->d_inode, lnb);
716 CDEBUG(rc == -ENOSPC ? D_INODE : D_ERROR, "page err %u@"
717 LPU64" %u/%u %p: rc %d\n", lnb->len, lnb->offset,
718 i, obj->ioo_bufcnt, dentry, rc);
720 __free_pages(lnb->page, 0);
724 tot_bytes += lnb->len;
727 if (time_after(jiffies, now + 15 * HZ))
728 CERROR("slow start_page_write %lus\n", (jiffies - now) / HZ);
730 CDEBUG(D_INFO, "start_page_write: %lu jiffies\n",
738 int mds_commitrw(int cmd, struct obd_export *exp, struct obdo *oa,
739 int objcount, struct obd_ioobj *obj, int niocount,
740 struct niobuf_local *res, struct obd_trans_info *oti,
743 struct obd_device *obd = exp->exp_obd;
744 struct niobuf_local *lnb;
745 struct inode *inode = NULL;
746 int rc = 0, i, cleanup_phase = 0, err, entries = 0;
749 LASSERT(objcount == 1);
750 LASSERT(current->journal_info == NULL);
753 inode = res->dentry->d_inode;
755 for (i = 0, lnb = res; i < obj->ioo_bufcnt; i++, lnb++) {
757 struct dir_entry *de;
759 buf = kmap(lnb->page);
760 LASSERT(buf != NULL);
761 end = buf + lnb->len;
762 de = (struct dir_entry *) buf;
763 while ((char *) de < end && de->namelen) {
764 err = fsfilt_add_dir_entry(obd, res->dentry, de->name,
765 de->namelen, de->ino,
766 de->generation, de->mds);
768 CERROR("can't add dir entry %*s->%u/%u/%u"
770 de->namelen, de->name,
771 de->mds, (unsigned) de->ino,
772 (unsigned) de->generation,
773 res->dentry->d_inode->i_ino,
774 res->dentry->d_inode->i_generation,
780 de = (struct dir_entry *)
781 ((char *) de + DIR_REC_LEN(de->namelen));
787 for (i = 0, lnb = res; i < obj->ioo_bufcnt; i++, lnb++)
788 __free_page(lnb->page);
794 int mds_choose_mdsnum(struct obd_device *obd, const char *name, int len, int flags)
797 struct mds_obd *mds = &obd->u.mds;
798 int i = mds->mds_num;
800 if (flags & REC_REINT_CREATE) {
802 } else if (mds->mds_lmv_exp) {
803 lmv = &mds->mds_lmv_exp->exp_obd->u.lmv;
804 i = raw_name2idx(MEA_MAGIC_LAST_CHAR, lmv->desc.ld_tgt_count, name, len);
809 int mds_lock_slave_objs(struct obd_device *obd, struct dentry *dentry,
810 struct lustre_handle **rlockh)
812 struct mds_obd *mds = &obd->u.mds;
813 struct mdc_op_data op_data;
814 struct lookup_intent it;
815 struct mea *mea = NULL;
818 LASSERT(rlockh != NULL);
819 LASSERT(dentry != NULL);
820 LASSERT(dentry->d_inode != NULL);
823 if (!mds->mds_lmv_obd)
826 /* a dir can be splitted only */
827 if (!S_ISDIR(dentry->d_inode->i_mode))
830 rc = mds_get_lmv_attr(obd, dentry->d_inode, &mea, &mea_size);
836 if (mea->mea_count == 0) {
837 /* this is slave object */
838 GOTO(cleanup, rc = 0);
841 CDEBUG(D_OTHER, "%s: lock slaves for %lu/%lu\n", obd->obd_name,
842 (unsigned long) dentry->d_inode->i_ino,
843 (unsigned long) dentry->d_inode->i_generation);
845 OBD_ALLOC(*rlockh, sizeof(struct lustre_handle) * mea->mea_count);
847 GOTO(cleanup, rc = -ENOMEM);
848 memset(*rlockh, 0, sizeof(struct lustre_handle) * mea->mea_count);
850 memset(&op_data, 0, sizeof(op_data));
852 it.it_op = IT_UNLINK;
853 rc = md_enqueue(mds->mds_lmv_exp, LDLM_IBITS, &it, LCK_EX, &op_data,
854 *rlockh, NULL, 0, ldlm_completion_ast, mds_blocking_ast,
857 OBD_FREE(mea, mea_size);
861 void mds_unlock_slave_objs(struct obd_device *obd, struct dentry *dentry,
862 struct lustre_handle *lockh)
864 struct mds_obd *mds = &obd->u.mds;
865 struct mea *mea = NULL;
871 LASSERT(mds->mds_lmv_obd != NULL);
872 LASSERT(S_ISDIR(dentry->d_inode->i_mode));
874 rc = mds_get_lmv_attr(obd, dentry->d_inode, &mea, &mea_size);
876 CERROR("locks are leaked\n");
879 LASSERT(mea_size != 0);
880 LASSERT(mea != NULL);
881 LASSERT(mea->mea_count != 0);
883 CDEBUG(D_OTHER, "%s: unlock slaves for %lu/%lu\n", obd->obd_name,
884 (unsigned long) dentry->d_inode->i_ino,
885 (unsigned long) dentry->d_inode->i_generation);
887 for (i = 0; i < mea->mea_count; i++) {
888 if (lockh[i].cookie != 0)
889 ldlm_lock_decref(lockh + i, LCK_EX);
892 OBD_FREE(lockh, sizeof(struct lustre_handle) * mea->mea_count);
893 OBD_FREE(mea, mea_size);
897 int mds_unlink_slave_objs(struct obd_device *obd, struct dentry *dentry)
899 struct mds_obd *mds = &obd->u.mds;
900 struct ptlrpc_request *req = NULL;
901 struct mdc_op_data op_data;
902 struct mea *mea = NULL;
906 if (!mds->mds_lmv_obd)
909 /* a dir can be splitted only */
910 if (!S_ISDIR(dentry->d_inode->i_mode))
913 rc = mds_get_lmv_attr(obd, dentry->d_inode, &mea, &mea_size);
919 if (mea->mea_count == 0)
920 GOTO(cleanup, rc = 0);
922 CDEBUG(D_OTHER, "%s: unlink slaves for %lu/%lu\n", obd->obd_name,
923 (unsigned long) dentry->d_inode->i_ino,
924 (unsigned long) dentry->d_inode->i_generation);
926 memset(&op_data, 0, sizeof(op_data));
928 rc = md_unlink(mds->mds_lmv_exp, &op_data, &req);
929 LASSERT(req == NULL);
931 OBD_FREE(mea, mea_size);
935 struct ide_tracking {
940 int mds_ide_filldir(void *__buf, const char *name, int namelen,
941 loff_t offset, ino_t ino, unsigned int d_type)
943 struct ide_tracking *it = __buf;
953 if (name[0] == '.' && namelen == 1)
955 if (name[0] == '.' && name[1] == '.' && namelen == 2)
962 int mds_is_dir_empty(struct obd_device *obd, struct dentry *dentry)
964 struct ide_tracking it;
972 nlen = strlen("__iopen__/") + 10 + 1;
973 OBD_ALLOC(file_name, nlen);
976 i = sprintf(file_name, "__iopen__/0x%lx", dentry->d_inode->i_ino);
978 file = filp_open(file_name, O_RDONLY, 0);
980 CERROR("can't open directory %s: %d\n",
981 file_name, (int) PTR_ERR(file));
982 GOTO(cleanup, rc = PTR_ERR(file));
985 rc = vfs_readdir(file, mds_ide_filldir, &it);
988 if (it.empty && rc == 0)
994 OBD_FREE(file_name, nlen);
998 int mds_lock_and_check_slave(int offset, struct ptlrpc_request *req,
999 struct lustre_handle *lockh)
1001 struct obd_device *obd = req->rq_export->exp_obd;
1002 struct dentry *dentry = NULL;
1003 struct lvfs_run_ctxt saved;
1004 int cleanup_phase = 0;
1005 struct mds_body *body;
1006 struct lvfs_ucred uc;
1007 int rc, update_mode;
1010 body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1011 lustre_swab_mds_body);
1013 CERROR("Can't swab mds_body\n");
1014 GOTO(cleanup, rc = -EFAULT);
1016 CDEBUG(D_OTHER, "%s: check slave %lu/%lu\n", obd->obd_name,
1017 (unsigned long) body->fid1.id,
1018 (unsigned long) body->fid1.generation);
1019 dentry = mds_fid2locked_dentry(obd, &body->fid1, NULL, LCK_EX, lockh,
1020 &update_mode, NULL, 0,
1021 MDS_INODELOCK_UPDATE);
1022 if (IS_ERR(dentry)) {
1023 CERROR("can't find inode: %d\n", (int) PTR_ERR(dentry));
1024 GOTO(cleanup, rc = PTR_ERR(dentry));
1028 LASSERT(S_ISDIR(dentry->d_inode->i_mode));
1030 uc.luc_fsuid = body->fsuid;
1031 uc.luc_fsgid = body->fsgid;
1032 uc.luc_cap = body->capability;
1033 uc.luc_suppgid1 = body->suppgid;
1034 uc.luc_suppgid2 = -1;
1035 push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1038 if (!mds_is_dir_empty(obd, dentry))
1042 switch(cleanup_phase) {
1045 ldlm_lock_decref(lockh, LCK_EX);
1047 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1054 int mds_convert_mea_ea(struct obd_device *obd, struct inode *inode,
1055 struct lov_mds_md *lmm, int lmmsize)
1057 int i, rc, err, size;
1058 struct mea_old *old;
1064 mea = (struct mea *) lmm;
1065 if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
1066 mea->mea_magic == MEA_MAGIC_ALL_CHARS)
1069 old = (struct mea_old *) lmm;
1070 rc = sizeof(struct ll_fid) * old->mea_count + sizeof(struct mea_old);
1071 if (old->mea_count > 256 || old->mea_master > 256 || lmmsize < rc
1072 || old->mea_master > old->mea_count) {
1073 CWARN("unknown MEA format, dont convert it\n");
1074 CWARN(" count %u, master %u, size %u\n",
1075 old->mea_count, old->mea_master, rc);
1079 CWARN("converting MEA EA on %lu/%u from V0 to V1 (%u/%u)\n",
1080 inode->i_ino, inode->i_generation, old->mea_count, old->mea_master);
1082 size = sizeof(struct ll_fid) * old->mea_count + sizeof(struct mea);
1083 OBD_ALLOC(new, size);
1087 new->mea_magic = MEA_MAGIC_LAST_CHAR;
1088 new->mea_count = old->mea_count;
1089 new->mea_master = old->mea_master;
1090 for (i = 0; i < new->mea_count; i++)
1091 new->mea_fids[i] = old->mea_fids[i];
1093 handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
1094 if (IS_ERR(handle)) {
1095 rc = PTR_ERR(handle);
1096 GOTO(conv_free, rc);
1099 rc = fsfilt_set_md(obd, inode, handle, (struct lov_mds_md *) new, size);
1102 memcpy(lmm, new, size);
1104 err = fsfilt_commit(obd, obd->u.mds.mds_sb, inode, handle, 0);
1106 rc = err ? err : size;
1107 GOTO(conv_free, rc);
1109 OBD_FREE(new, size);