1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 * derived in small part from linux/fs/ext2/namei.c
23 * Copyright (C) 1991, 1992 Linus Torvalds
25 * Big-endian to little-endian byte-swapping/bitmaps by
26 * David S. Miller (davem@caip.rutgers.edu), 1995
27 * Directory entry file type support and forward compatibility hooks
28 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
32 #include <linux/sched.h>
34 #include <linux/smp_lock.h>
35 #include <linux/quotaops.h>
36 #include <linux/highmem.h>
37 #include <linux/pagemap.h>
39 #define DEBUG_SUBSYSTEM S_LLITE
41 #include <linux/obd_support.h>
42 #include <linux/lustre_lite.h>
43 #include <linux/lustre_dlm.h>
44 #include <linux/lustre_version.h>
45 #include "llite_internal.h"
49 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
50 static int ll_test_inode(struct inode *inode, unsigned long ino, void *opaque)
52 static int ll_test_inode(struct inode *inode, void *opaque)
55 static int last_ino, last_gen, last_count;
56 struct lustre_md *md = opaque;
58 if (!(md->body->valid & (OBD_MD_FLGENER | OBD_MD_FLID))) {
59 CERROR("MDS body missing inum or generation\n");
63 if (last_ino == id_ino(&md->body->id1) &&
64 last_gen == id_gen(&md->body->id1) &&
69 CDEBUG(D_VFSTRACE, "compared %u/%u %u times\n",
70 last_ino, last_gen, last_count);
72 last_ino = id_ino(&md->body->id1);
73 last_gen = id_gen(&md->body->id1);
75 "comparing inode %p ino "DLID4" to body "DLID4"\n",
76 inode, OLID4(&ll_i2info(inode)->lli_id),
77 OLID4(&md->body->id1));
80 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
81 if (inode->i_ino != id_ino(&md->body->id1))
84 if (inode->i_generation != id_gen(&md->body->id1))
87 if (id_group(&ll_i2info(inode)->lli_id) != id_group(&md->body->id1))
90 /* apply the attributes in 'opaque' to this inode. */
91 ll_update_inode(inode, md);
95 extern struct dentry_operations ll_d_ops;
97 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
101 ldlm_lock_decref(lockh, mode);
107 * get an inode by inode number (already instantiated by the intent lookup).
108 * Returns inode or NULL.
110 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
111 int ll_set_inode(struct inode *inode, void *opaque)
113 ll_read_inode2(inode, opaque);
117 struct inode *ll_iget(struct super_block *sb, ino_t hash,
118 struct lustre_md *md)
123 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
126 if (inode->i_state & I_NEW)
127 unlock_new_inode(inode);
128 CDEBUG(D_VFSTRACE, "inode: %lu/%u(%p)\n", inode->i_ino,
129 inode->i_generation, inode);
135 struct inode *ll_iget(struct super_block *sb, ino_t hash,
136 struct lustre_md *md)
140 inode = iget4(sb, hash, ll_test_inode, md);
142 CDEBUG(D_VFSTRACE, "inode: %lu/%u(%p)\n", inode->i_ino,
143 inode->i_generation, inode);
148 int ll_mdc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
149 void *data, int flag)
152 struct lustre_handle lockh;
156 case LDLM_CB_BLOCKING:
157 ldlm_lock2handle(lock, &lockh);
158 rc = ldlm_cli_cancel(&lockh);
160 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
164 case LDLM_CB_CANCELING: {
165 struct inode *inode = ll_inode_from_lock(lock);
166 struct ll_inode_info *li = ll_i2info(inode);
167 __u64 bits = lock->l_policy_data.l_inodebits.bits;
169 /* For lookup locks: Invalidate all dentries associated with
170 this inode, for UPDATE locks - invalidate directory pages */
174 if (bits & MDS_INODELOCK_UPDATE)
175 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK,
176 &(ll_i2info(inode)->lli_flags));
179 if (lock->l_resource->lr_name.name[0] != id_fid(&li->lli_id) ||
180 lock->l_resource->lr_name.name[1] != id_group(&li->lli_id)) {
181 LDLM_ERROR(lock, "data mismatch with object %lu/%lu",
182 (unsigned long)id_fid(&li->lli_id),
183 (unsigned long)id_group(&li->lli_id));
186 if (bits & MDS_INODELOCK_OPEN) {
188 switch (lock->l_req_mode) {
199 CERROR("Unexpected lock mode for OPEN lock "
200 "%d, inode %ld\n", lock->l_req_mode,
203 ll_md_real_close(ll_i2mdexp(inode), inode, flags);
206 if (bits & MDS_INODELOCK_UPDATE)
207 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK,
208 &(ll_i2info(inode)->lli_flags));
211 /* If lookup lock is cancelled, we just drop the dentry and
212 this will cause us to reget data from MDS when we'd want to
213 access this dentry/inode again. If this is lock on
214 other parts of inode that is cancelled, we do not need to do
215 much (but need to discard data from readdir, if any), since
216 abscence of lock will cause ll_revalidate_it (called from
217 stat() and similar functions) to renew the data anyway */
218 if (S_ISDIR(inode->i_mode) &&
219 (bits & MDS_INODELOCK_UPDATE)) {
220 CDEBUG(D_INODE, "invalidating inode %lu/%u(%p)\n",
221 inode->i_ino, inode->i_generation, inode);
222 truncate_inode_pages(inode->i_mapping, 0);
225 if (inode->i_sb->s_root &&
226 inode != inode->i_sb->s_root->d_inode &&
227 (bits & MDS_INODELOCK_LOOKUP))
228 ll_unhash_aliases(inode);
239 int ll_mdc_cancel_unused(struct lustre_handle *conn, struct inode *inode,
240 int flags, void *opaque)
242 struct ll_inode_info *li = ll_i2info(inode);
243 struct ldlm_res_id res_id =
244 { .name = {id_fid(&li->lli_id), id_group(&li->lli_id)} };
245 struct obd_device *obddev = class_conn2obd(conn);
248 RETURN(ldlm_cli_cancel_unused(obddev->obd_namespace, &res_id, flags,
252 /* Search "inode"'s alias list for a dentry that has the same name and parent as
253 * de. If found, return it. If not found, return de. */
254 struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
256 struct list_head *tmp;
258 spin_lock(&dcache_lock);
259 list_for_each(tmp, &inode->i_dentry) {
260 struct dentry *dentry = list_entry(tmp, struct dentry, d_alias);
262 /* We are called here with 'de' already on the aliases list. */
268 if (dentry->d_parent != de->d_parent)
271 if (dentry->d_name.len != de->d_name.len)
274 if (memcmp(dentry->d_name.name, de->d_name.name,
275 de->d_name.len) != 0)
278 if (!list_empty(&dentry->d_lru))
279 list_del_init(&dentry->d_lru);
281 hlist_del_init(&dentry->d_hash);
282 __d_rehash(dentry); /* avoid taking dcache_lock inside */
283 spin_unlock(&dcache_lock);
284 atomic_inc(&dentry->d_count);
286 dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
287 CDEBUG(D_DENTRY, "alias dentry %*s (%p) parent %p inode %p "
288 "refc %d\n", de->d_name.len, de->d_name.name, de,
289 de->d_parent, de->d_inode, atomic_read(&de->d_count));
293 spin_unlock(&dcache_lock);
298 static int lookup_it_finish(struct ptlrpc_request *request, int offset,
299 struct lookup_intent *it, void *data)
301 struct it_cb_data *icbd = data;
302 struct dentry **de = icbd->icbd_childp;
303 struct inode *parent = icbd->icbd_parent;
304 struct ll_sb_info *sbi = ll_i2sbi(parent);
305 struct dentry *dentry = *de, *saved = *de;
306 struct inode *inode = NULL;
309 /* NB 1 request reference will be taken away by ll_intent_lock()
311 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
314 rc = ll_prep_inode(sbi->ll_dt_exp, sbi->ll_md_exp,
315 &inode, request, offset, dentry->d_sb);
319 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
320 inode, inode->i_ino, inode->i_generation);
322 mdc_set_lock_data(NULL, &LUSTRE_IT(it)->it_lock_handle, inode);
324 /* If this is a stat, get the authoritative file size */
325 if (it->it_op == IT_GETATTR && S_ISREG(inode->i_mode) &&
326 ll_i2info(inode)->lli_smd != NULL) {
327 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
330 LASSERT(lsm->lsm_object_id != 0);
332 /* bug 2334: drop MDS lock before acquiring OST lock */
333 ll_intent_drop_lock(it);
335 rc = ll_glimpse_size(inode);
342 dentry = *de = ll_find_alias(inode, dentry);
347 dentry->d_op = &ll_d_ops;
351 d_add(dentry, inode);
356 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
357 struct nameidata *nd, int flags)
359 struct dentry *save = dentry, *retval;
360 struct lookup_intent *it = flags ? &nd->intent.open : NULL;
361 struct lustre_id pid;
362 struct it_cb_data icbd;
363 struct ptlrpc_request *req = NULL;
364 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
368 if (dentry->d_name.len > EXT3_NAME_LEN)
369 RETURN(ERR_PTR(-ENAMETOOLONG));
371 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
372 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
373 parent->i_generation, parent, LL_IT2STR(it));
375 if (d_mountpoint(dentry))
376 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
379 nd->mnt->mnt_last_used = jiffies;
381 orig_it = it ? it->it_op : IT_OPEN;
382 ll_frob_intent(&it, &lookup_it);
384 icbd.icbd_childp = &dentry;
385 icbd.icbd_parent = parent;
386 ll_inode2id(&pid, parent);
388 rc = md_intent_lock(ll_i2mdexp(parent), &pid,
389 dentry->d_name.name, dentry->d_name.len, NULL, 0,
390 NULL, it, flags, &req, ll_mdc_blocking_ast);
392 GOTO(out, retval = ERR_PTR(rc));
394 rc = lookup_it_finish(req, 1, it, &icbd);
396 ll_intent_release(it);
397 GOTO(out, retval = ERR_PTR(rc));
400 ll_lookup_finish_locks(it, dentry);
403 dentry->d_inode != NULL && dentry->d_inode->i_mode & S_ISUID &&
404 S_ISDIR(dentry->d_inode->i_mode) &&
405 ((flags & LOOKUP_CONTINUE) || (orig_it & (IT_CHDIR | IT_OPEN))))
407 rc = ll_gns_mount_object(dentry, nd->mnt);
410 * making system to restart syscall as currently GNS is
411 * in mounting progress.
413 GOTO(out, retval = ERR_PTR(-ERESTARTSYS));
418 * just reporting about GNS failures, lookup() is
419 * successful, do not stop it.
421 * GNS failure may be that found object is found in SUID
422 * bit marked dir but it is not regular file and we
423 * should lookup further until we find correct mount
424 * object. This will allow to perform GNS mount is the
425 * following case for instance:
427 * /mnt/lustre/gns_mount/.mntinfo/.mntinfo/..../.mntinfo
428 * where all ".mntinfo" are dirs and only last one is
431 CDEBUG(D_INODE, "failed to mount %*s, err %d\n",
432 (int)dentry->d_name.len, dentry->d_name.name);
437 GOTO(out, retval = NULL);
439 GOTO(out, retval = dentry);
442 ptlrpc_req_finished(req);
443 if (it == &lookup_it)
444 ll_intent_release(it);
446 CDEBUG(D_INODE, "lookup 0x%p in %lu/%lu: %*s -> %lu/%lu\n",
448 (unsigned long) parent->i_ino,
449 (unsigned long) parent->i_generation,
450 dentry->d_name.len, dentry->d_name.name,
451 (unsigned long) dentry->d_inode->i_ino,
452 (unsigned long) dentry->d_inode->i_generation);
454 CDEBUG(D_INODE, "lookup 0x%p in %lu/%lu: %*s -> ??\n",
456 (unsigned long) parent->i_ino,
457 (unsigned long) parent->i_generation,
458 dentry->d_name.len, dentry->d_name.name);
462 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
463 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
464 struct nameidata *nd)
469 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
470 de = ll_lookup_it(parent, dentry, nd, nd->flags);
472 de = ll_lookup_it(parent, dentry, nd, 0);
478 /* We depend on "mode" being set with the proper file type/umask by now */
479 static struct inode *ll_create_node(struct inode *dir, const char *name,
480 int namelen, const void *data, int datalen,
481 int mode, __u64 extra,
482 struct lookup_intent *it)
484 struct inode *inode = NULL;
485 struct ptlrpc_request *request = NULL;
486 struct ll_sb_info *sbi = ll_i2sbi(dir);
491 LASSERT(it && LUSTRE_IT(it)->it_disposition);
493 request = LUSTRE_IT(it)->it_data;
494 rc = ll_prep_inode(sbi->ll_dt_exp, sbi->ll_md_exp,
495 &inode, request, 1, dir->i_sb);
497 GOTO(out, inode = ERR_PTR(rc));
499 LASSERT(list_empty(&inode->i_dentry));
501 /* We asked for a lock on the directory, but were granted a
502 * lock on the inode. Since we finally have an inode pointer,
503 * stuff it in the lock. */
504 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
505 inode, inode->i_ino, inode->i_generation);
506 mdc_set_lock_data(NULL, &LUSTRE_IT(it)->it_lock_handle, inode);
509 ptlrpc_req_finished(request);
514 * By the time this is called, we already have created the directory cache
515 * entry for the new file, but it is so far negative - it has no inode.
517 * We defer creating the OBD object(s) until open, to keep the intent and
518 * non-intent code paths similar, and also because we do not have the MDS
519 * inode number before calling ll_create_node() (which is needed for LOV),
520 * so we would need to do yet another RPC to the MDS to store the LOV EA
521 * data on the MDS. If needed, we would pass the PACKED lmm as data and
522 * lmm_size in datalen (the MDS still has code which will handle that).
524 * If the create succeeds, we fill in the inode information
525 * with d_instantiate().
527 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
528 struct lookup_intent *it)
531 struct ptlrpc_request *request = LUSTRE_IT(it)->it_data;
532 struct obd_export *md_exp = ll_i2mdexp(dir);
536 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
537 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
538 dir->i_generation, dir, LL_IT2STR(it));
540 rc = it_open_error(DISP_OPEN_CREATE, it);
544 mdc_store_inode_generation(md_exp, request, MDS_REQ_INTENT_REC_OFF, 1);
545 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
546 NULL, 0, mode, 0, it);
548 RETURN(PTR_ERR(inode));
550 d_instantiate(dentry, inode);
554 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
555 static int ll_create_nd(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
557 return ll_create_it(dir, dentry, mode, &nd->intent.open);
561 static void ll_update_times(struct ptlrpc_request *request, int offset,
564 struct mds_body *body = lustre_msg_buf(request->rq_repmsg, offset,
568 if (body->valid & OBD_MD_FLMTIME &&
569 body->mtime > LTIME_S(inode->i_mtime)) {
570 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
571 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
572 LTIME_S(inode->i_mtime) = body->mtime;
574 if (body->valid & OBD_MD_FLCTIME &&
575 body->ctime > LTIME_S(inode->i_ctime))
576 LTIME_S(inode->i_ctime) = body->ctime;
579 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
581 struct ptlrpc_request *request = NULL;
582 struct inode *dir = nd->dentry->d_inode;
583 struct ll_sb_info *sbi = ll_i2sbi(dir);
584 struct mdc_op_data *op_data;
588 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
589 nd->last.len, nd->last.name, dir->i_ino, dir->i_generation, dir);
591 mode &= ~current->fs->umask;
593 switch (mode & S_IFMT) {
596 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
601 OBD_ALLOC(op_data, sizeof(*op_data));
604 ll_prepare_mdc_data(op_data, dir, NULL, nd->last.name,
606 err = md_create(sbi->ll_md_exp, op_data, NULL, 0, mode,
607 current->fsuid, current->fsgid, rdev,
609 OBD_FREE(op_data, sizeof(*op_data));
611 ll_update_times(request, 0, dir);
612 ptlrpc_req_finished(request);
623 static int ll_mknod(struct inode *dir, struct dentry *dchild,
624 int mode, ll_dev_t rdev)
626 struct ptlrpc_request *request = NULL;
627 struct inode *inode = NULL;
628 struct ll_sb_info *sbi = ll_i2sbi(dir);
629 struct mdc_op_data *op_data;
633 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
634 dchild->d_name.len, dchild->d_name.name,
635 dir->i_ino, dir->i_generation, dir);
637 mode &= ~current->fs->umask;
639 switch (mode & S_IFMT) {
642 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
647 OBD_ALLOC(op_data, sizeof(*op_data));
650 ll_prepare_mdc_data(op_data, dir, NULL, dchild->d_name.name,
651 dchild->d_name.len, 0);
652 err = md_create(sbi->ll_md_exp, op_data, NULL, 0, mode,
653 current->fsuid, current->fsgid, rdev,
655 OBD_FREE(op_data, sizeof(*op_data));
659 ll_update_times(request, 0, dir);
660 err = ll_prep_inode(sbi->ll_dt_exp, sbi->ll_md_exp,
661 &inode, request, 0, dchild->d_sb);
672 d_instantiate(dchild, inode);
675 ptlrpc_req_finished(request);
679 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
681 struct inode *dir = nd->dentry->d_inode;
682 struct ptlrpc_request *request = NULL;
683 struct ll_sb_info *sbi = ll_i2sbi(dir);
684 const char *name = nd->last.name;
685 struct mdc_op_data *op_data;
686 int len = nd->last.len;
690 CDEBUG(D_VFSTRACE, "VFS Op:name=%*s,dir=%lu/%u(%p),target=%s\n",
691 nd->last.len, nd->last.name, dir->i_ino, dir->i_generation,
694 if (dir->i_nlink >= EXT3_LINK_MAX)
697 OBD_ALLOC(op_data, sizeof(*op_data));
700 ll_prepare_mdc_data(op_data, dir, NULL, name, len, 0);
702 err = md_create(sbi->ll_md_exp, op_data,
703 tgt, strlen(tgt) + 1, S_IFLNK | S_IRWXUGO,
704 current->fsuid, current->fsgid, 0, &request);
705 OBD_FREE(op_data, sizeof(*op_data));
707 ll_update_times(request, 0, dir);
709 ptlrpc_req_finished(request);
713 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
715 struct inode *src = srcnd->dentry->d_inode;
716 struct inode *dir = tgtnd->dentry->d_inode;
717 struct ptlrpc_request *request = NULL;
718 struct mdc_op_data *op_data;
720 struct ll_sb_info *sbi = ll_i2sbi(dir);
724 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
725 src->i_ino, src->i_generation, src, dir->i_ino,
726 dir->i_generation, dir, tgtnd->last.len, tgtnd->last.name);
728 OBD_ALLOC(op_data, sizeof(*op_data));
731 ll_prepare_mdc_data(op_data, src, dir, tgtnd->last.name,
733 err = md_link(sbi->ll_md_exp, op_data, &request);
734 OBD_FREE(op_data, sizeof(*op_data));
736 ll_update_times(request, 0, dir);
737 ptlrpc_req_finished(request);
742 static int ll_mkdir_raw(struct nameidata *nd, int mode)
744 struct inode *dir = nd->dentry->d_inode;
745 struct ptlrpc_request *request = NULL;
746 struct ll_sb_info *sbi = ll_i2sbi(dir);
747 struct mdc_op_data *op_data;
750 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
751 nd->last.len, nd->last.name, dir->i_ino, dir->i_generation, dir);
753 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
754 OBD_ALLOC(op_data, sizeof(*op_data));
757 ll_prepare_mdc_data(op_data, dir, NULL, nd->last.name,
759 err = md_create(sbi->ll_md_exp, op_data, NULL, 0, mode,
760 current->fsuid, current->fsgid, 0, &request);
761 OBD_FREE(op_data, sizeof(*op_data));
763 ll_update_times(request, 0, dir);
764 ptlrpc_req_finished(request);
768 static int ll_rmdir_raw(struct nameidata *nd)
770 struct inode *dir = nd->dentry->d_inode;
771 struct ptlrpc_request *request = NULL;
772 struct mdc_op_data *op_data;
776 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
777 nd->last.len, nd->last.name, dir->i_ino, dir->i_generation, dir);
779 OBD_ALLOC(op_data, sizeof(*op_data));
782 ll_prepare_mdc_data(op_data, dir, NULL, nd->last.name,
783 nd->last.len, S_IFDIR);
784 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
785 OBD_FREE(op_data, sizeof(*op_data));
787 ll_update_times(request, 0, dir);
788 ptlrpc_req_finished(request);
792 int ll_objects_destroy(struct ptlrpc_request *request,
793 struct inode *dir, int offset)
795 struct mds_body *body;
796 struct lov_mds_md *eadata;
797 struct lov_stripe_md *lsm = NULL;
798 struct obd_trans_info oti = { 0 };
803 /* req is swabbed so this is safe */
804 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
806 if (!(body->valid & OBD_MD_FLEASIZE))
809 if (body->eadatasize == 0) {
810 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
811 GOTO(out, rc = -EPROTO);
815 * the MDS sent back the EA because we unlinked the last reference to
816 * this file. Use this EA to unlink the objects on the OST. It's opaque
817 * so we don't swab here; we leave it to obd_unpackmd() to check it is
818 * complete and sensible.
820 eadata = lustre_swab_repbuf(request, 1, body->eadatasize, NULL);
821 LASSERT(eadata != NULL);
822 if (eadata == NULL) {
823 CERROR("Can't unpack MDS EA data\n");
824 GOTO(out, rc = -EPROTO);
827 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
829 CERROR("obd_unpackmd: %d\n", rc);
832 LASSERT(rc >= sizeof(*lsm));
836 GOTO(out_free_memmd, rc = -ENOMEM);
838 oa->o_id = lsm->lsm_object_id;
839 oa->o_gr = lsm->lsm_object_gr;
840 oa->o_mode = body->mode & S_IFMT;
841 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
843 if (body->valid & OBD_MD_FLCOOKIE) {
844 int length = sizeof(struct llog_cookie) *
845 lsm->lsm_stripe_count;
846 oa->o_valid |= OBD_MD_FLCOOKIE;
848 lustre_msg_buf(request->rq_repmsg, 2, length);
849 if (oti.oti_logcookies == NULL) {
850 oa->o_valid &= ~OBD_MD_FLCOOKIE;
851 body->valid &= ~OBD_MD_FLCOOKIE;
853 /* copy llog cookies to request to replay unlink
854 * so that the same llog file and records as those created
855 * during fail can be re-created while doing replay
858 memcpy(lustre_msg_buf(request->rq_reqmsg, offset, 0),
859 oti.oti_logcookies, length);
863 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti);
866 CERROR("obd destroy objid "LPX64" error %d\n",
867 lsm->lsm_object_id, rc);
870 obd_free_memmd(ll_i2dtexp(dir), &lsm);
875 static int ll_unlink_raw(struct nameidata *nd)
877 struct inode *dir = nd->dentry->d_inode;
878 struct ptlrpc_request *request = NULL;
879 struct mdc_op_data *op_data;
882 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
883 nd->last.len, nd->last.name, dir->i_ino, dir->i_generation, dir);
885 OBD_ALLOC(op_data, sizeof(*op_data));
888 ll_prepare_mdc_data(op_data, dir, NULL, nd->last.name, nd->last.len, 0);
889 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
890 OBD_FREE(op_data, sizeof(*op_data));
893 ll_update_times(request, 0, dir);
895 rc = ll_objects_destroy(request, dir, 2);
898 ptlrpc_req_finished(request);
902 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
904 struct inode *src = srcnd->dentry->d_inode;
905 struct inode *tgt = tgtnd->dentry->d_inode;
906 struct ptlrpc_request *request = NULL;
907 struct ll_sb_info *sbi = ll_i2sbi(src);
908 struct mdc_op_data *op_data;
912 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
913 "tgt_dir=%lu/%u(%p)\n", srcnd->last.len, srcnd->last.name,
914 src->i_ino, src->i_generation, src, tgtnd->last.len,
915 tgtnd->last.name, tgt->i_ino, tgt->i_generation, tgt);
917 OBD_ALLOC(op_data, sizeof(*op_data));
920 ll_prepare_mdc_data(op_data, src, tgt, NULL, 0, 0);
921 err = md_rename(sbi->ll_md_exp, op_data, srcnd->last.name,
922 srcnd->last.len, tgtnd->last.name, tgtnd->last.len,
924 OBD_FREE(op_data, sizeof(*op_data));
926 ll_update_times(request, 0, src);
927 ll_update_times(request, 0, tgt);
928 err = ll_objects_destroy(request, src, 3);
931 ptlrpc_req_finished(request);
935 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
936 #define LLITE_IT_RAWOPS (IT_MKNOD|IT_MKDIR|IT_SYMLINK|IT_LINK|IT_UNLINK|IT_RMDIR|IT_RENAME)
937 static int ll_rawop_from_intent(struct nameidata *nd)
941 if (!nd || !(nd->intent.open.op & LLITE_IT_RAWOPS))
944 switch (nd->intent.open.op) {
946 error = ll_mknod_raw(nd, nd->intent.open.create_mode,
947 nd->intent.open.create.dev);
950 error = ll_mkdir_raw(nd, nd->intent.open.create_mode);
953 error = ll_rmdir_raw(nd);
956 error = ll_unlink_raw(nd);
959 LASSERT(nd->intent.open.create.link);
960 error = ll_symlink_raw(nd, nd->intent.open.create.link);
963 error = ll_link_raw(nd->intent.open.create.source_nd, nd);
966 LASSERT(nd->intent.open.create.source_nd);
967 error = ll_rename_raw(nd->intent.open.create.source_nd, nd);
972 if (error != -EOPNOTSUPP)
973 nd->intent.open.flags |= IT_STATUS_RAW;
979 struct inode_operations ll_dir_inode_operations = {
981 .setattr = ll_setattr,
982 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
983 .create_it = ll_create_it,
984 .lookup_it = ll_lookup_it,
985 .revalidate_it = ll_inode_revalidate_it,
987 .lookup = ll_lookup_nd,
988 .create = ll_create_nd,
989 .getattr = ll_getattr,
990 .endparentlookup = ll_rawop_from_intent,
992 .setxattr = ll_setxattr,
993 .getxattr = ll_getxattr,
994 .listxattr = ll_listxattr,
995 .removexattr = ll_removexattr,
996 .permission = ll_inode_permission,