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 "llite_internal.h"
48 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
49 static int ll_test_inode(struct inode *inode, unsigned long ino, void *opaque)
51 static int ll_test_inode(struct inode *inode, void *opaque)
54 static int last_ino, last_gen, last_count;
55 struct lustre_md *md = opaque;
57 if (!(md->body->valid & (OBD_MD_FLGENER | OBD_MD_FLID))) {
58 CERROR("MDS body missing inum or generation\n");
62 if (last_ino == md->body->ino && last_gen == md->body->generation &&
67 CDEBUG(D_VFSTRACE, "compared %u/%u %u times\n",
68 last_ino, last_gen, last_count);
70 last_ino = md->body->ino;
71 last_gen = md->body->generation;
73 "comparing inode %p ino %lu/%u to body %u/%u\n",
74 inode, inode->i_ino, inode->i_generation,
75 md->body->ino, md->body->generation);
78 if (inode->i_generation != md->body->generation)
81 /* Apply the attributes in 'opaque' to this inode */
82 ll_update_inode(inode, md->body, md->lsm);
86 extern struct dentry_operations ll_d_ops;
88 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
92 ldlm_lock_decref(lockh, mode);
97 /* Get an inode by inode number (already instantiated by the intent lookup).
98 * Returns inode or NULL
100 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
101 int ll_set_inode(struct inode *inode, void *opaque)
103 ll_read_inode2(inode, opaque);
106 struct inode *ll_iget(struct super_block *sb, ino_t hash,
107 struct lustre_md *md)
112 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
115 if (inode->i_state & I_NEW)
116 unlock_new_inode(inode);
117 CDEBUG(D_VFSTRACE, "inode: %lu/%u(%p)\n", inode->i_ino,
118 inode->i_generation, inode);
124 struct inode *ll_iget(struct super_block *sb, ino_t hash,
125 struct lustre_md *md)
129 inode = iget4(sb, hash, ll_test_inode, md);
131 CDEBUG(D_VFSTRACE, "inode: %lu/%u(%p)\n", inode->i_ino,
132 inode->i_generation, inode);
137 int ll_mdc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
138 void *data, int flag)
141 struct lustre_handle lockh;
145 case LDLM_CB_BLOCKING:
146 ldlm_lock2handle(lock, &lockh);
147 rc = ldlm_cli_cancel(&lockh);
149 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
153 case LDLM_CB_CANCELING: {
154 struct inode *inode = ll_inode_from_lock(lock);
156 /* Invalidate all dentries associated with this inode */
160 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK,
161 &(ll_i2info(inode)->lli_flags));
163 if (lock->l_resource->lr_name.name[0] != inode->i_ino ||
164 lock->l_resource->lr_name.name[1] != inode->i_generation) {
165 LDLM_ERROR(lock, "data mismatch with ino %lu/%u",
166 inode->i_ino, inode->i_generation);
168 if (S_ISDIR(inode->i_mode)) {
169 CDEBUG(D_INODE, "invalidating inode %lu\n",
172 ll_invalidate_inode_pages(inode);
175 if (inode->i_sb->s_root &&
176 inode != inode->i_sb->s_root->d_inode)
177 ll_unhash_aliases(inode);
188 int ll_mdc_cancel_unused(struct lustre_handle *conn, struct inode *inode,
189 int flags, void *opaque)
191 struct ldlm_res_id res_id =
192 { .name = {inode->i_ino, inode->i_generation} };
193 struct obd_device *obddev = class_conn2obd(conn);
195 RETURN(ldlm_cli_cancel_unused(obddev->obd_namespace, &res_id, flags,
200 void ll_prepare_mdc_op_data(struct mdc_op_data *data, struct inode *i1,
201 struct inode *i2, const char *name, int namelen,
206 ll_i2uctxt(&data->ctxt, i1, i2);
207 ll_inode2fid(&data->fid1, i1);
210 ll_inode2fid(&data->fid2, i2);
213 data->namelen = namelen;
214 data->create_mode = mode;
215 data->mod_time = LTIME_S(CURRENT_TIME);
218 /* Search "inode"'s alias list for a dentry that has the same name and parent as
219 * de. If found, return it. If not found, return de. */
220 struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
222 struct list_head *tmp;
224 spin_lock(&dcache_lock);
225 list_for_each(tmp, &inode->i_dentry) {
226 struct dentry *dentry = list_entry(tmp, struct dentry, d_alias);
228 /* We are called here with 'de' already on the aliases list. */
234 if (dentry->d_parent != de->d_parent)
237 if (dentry->d_name.len != de->d_name.len)
240 if (memcmp(dentry->d_name.name, de->d_name.name,
241 de->d_name.len) != 0)
244 if (!list_empty(&dentry->d_lru))
245 list_del_init(&dentry->d_lru);
247 hlist_del_init(&dentry->d_hash);
248 __d_rehash(dentry, 0); /* avoid taking dcache_lock inside */
249 spin_unlock(&dcache_lock);
250 atomic_inc(&dentry->d_count);
252 dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
256 spin_unlock(&dcache_lock);
261 static int lookup_it_finish(struct ptlrpc_request *request, int offset,
262 struct lookup_intent *it, void *data)
264 struct it_cb_data *icbd = data;
265 struct dentry **de = icbd->icbd_childp;
266 struct inode *parent = icbd->icbd_parent;
267 struct ll_sb_info *sbi = ll_i2sbi(parent);
268 struct dentry *dentry = *de, *saved = *de;
269 struct inode *inode = NULL;
272 /* NB 1 request reference will be taken away by ll_intent_lock()
274 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
277 rc = ll_prep_inode(sbi->ll_osc_exp, &inode, request, offset,
282 /* If this is a stat, get the authoritative file size */
283 if (it->it_op == IT_GETATTR && S_ISREG(inode->i_mode) &&
284 ll_i2info(inode)->lli_smd != NULL) {
285 struct ldlm_extent extent = {0, OBD_OBJECT_EOF};
286 struct lustre_handle lockh = {0};
287 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
290 LASSERT(lsm->lsm_object_id != 0);
292 /* bug 2334: drop MDS lock before acquiring OST lock */
293 ll_intent_drop_lock(it);
295 rc = ll_extent_lock(NULL, inode, lsm, LCK_PR, &extent,
297 if (rc != ELDLM_OK) {
301 ll_extent_unlock(NULL, inode, lsm, LCK_PR, &lockh);
304 dentry = *de = ll_find_alias(inode, dentry);
309 dentry->d_op = &ll_d_ops;
313 d_add(dentry, inode);
319 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
320 struct lookup_intent *it, int flags)
322 struct dentry *save = dentry, *retval;
324 struct ll_uctxt ctxt;
325 struct it_cb_data icbd;
326 struct ptlrpc_request *req = NULL;
327 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
331 if (dentry->d_name.len > EXT3_NAME_LEN)
332 RETURN(ERR_PTR(-ENAMETOOLONG));
334 CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p),intent=%s\n",
335 dentry->d_name.name, parent->i_ino, parent->i_generation,
336 parent, LL_IT2STR(it));
338 if (d_mountpoint(dentry))
339 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
341 ll_frob_intent(&it, &lookup_it);
343 icbd.icbd_childp = &dentry;
344 icbd.icbd_parent = parent;
345 ll_inode2fid(&pfid, parent);
346 ll_i2uctxt(&ctxt, parent, NULL);
348 rc = mdc_intent_lock(ll_i2mdcexp(parent), &ctxt, &pfid,
349 dentry->d_name.name, dentry->d_name.len, NULL, 0,
350 NULL, it, flags, &req, ll_mdc_blocking_ast);
352 GOTO(out, retval = ERR_PTR(rc));
354 rc = lookup_it_finish(req, 1, it, &icbd);
356 ll_intent_release(it);
357 GOTO(out, retval = ERR_PTR(rc));
360 ll_lookup_finish_locks(it, dentry);
363 GOTO(out, retval = NULL);
365 GOTO(out, retval = dentry);
368 ptlrpc_req_finished(req);
372 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
373 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
374 struct nameidata *nd)
379 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
380 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
382 de = ll_lookup_it(parent, dentry, NULL, 0);
388 /* We depend on "mode" being set with the proper file type/umask by now */
389 static struct inode *ll_create_node(struct inode *dir, const char *name,
390 int namelen, const void *data, int datalen,
391 int mode, __u64 extra,
392 struct lookup_intent *it)
394 struct inode *inode = NULL;
395 struct ptlrpc_request *request = NULL;
396 struct ll_sb_info *sbi = ll_i2sbi(dir);
400 LASSERT(it && it->d.lustre.it_disposition);
402 request = it->d.lustre.it_data;
403 rc = ll_prep_inode(sbi->ll_osc_exp, &inode, request, 1, dir->i_sb);
405 GOTO(out, inode = ERR_PTR(rc));
407 LASSERT(list_empty(&inode->i_dentry));
409 /* We asked for a lock on the directory, but were granted a
410 * lock on the inode. Since we finally have an inode pointer,
411 * stuff it in the lock. */
412 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
413 inode, inode->i_ino, inode->i_generation);
414 mdc_set_lock_data(&it->d.lustre.it_lock_handle, inode);
417 ptlrpc_req_finished(request);
422 * By the time this is called, we already have created the directory cache
423 * entry for the new file, but it is so far negative - it has no inode.
425 * We defer creating the OBD object(s) until open, to keep the intent and
426 * non-intent code paths similar, and also because we do not have the MDS
427 * inode number before calling ll_create_node() (which is needed for LOV),
428 * so we would need to do yet another RPC to the MDS to store the LOV EA
429 * data on the MDS. If needed, we would pass the PACKED lmm as data and
430 * lmm_size in datalen (the MDS still has code which will handle that).
432 * If the create succeeds, we fill in the inode information
433 * with d_instantiate().
435 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
436 struct lookup_intent *it)
439 struct ptlrpc_request *request = it->d.lustre.it_data;
443 CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p),intent=%s\n",
444 dentry->d_name.name, dir->i_ino, dir->i_generation, dir,
447 rc = it_open_error(DISP_OPEN_CREATE, it);
451 mdc_store_inode_generation(request, 2, 1);
452 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
453 NULL, 0, mode, 0, it);
455 RETURN(PTR_ERR(inode));
458 d_instantiate(dentry, inode);
462 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
463 static int ll_create_nd(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
465 return ll_create_it(dir, dentry, mode, &nd->intent);
469 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
471 struct ptlrpc_request *request = NULL;
472 struct inode *dir = nd->dentry->d_inode;
473 const char *name = nd->last.name;
474 int len = nd->last.len;
475 struct ll_sb_info *sbi = ll_i2sbi(dir);
476 struct mdc_op_data op_data;
480 CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p)\n",
481 name, dir->i_ino, dir->i_generation, dir);
483 if (dir->i_nlink >= EXT3_LINK_MAX)
486 mode &= ~current->fs->umask;
488 switch (mode & S_IFMT) {
491 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
496 ll_prepare_mdc_op_data(&op_data, dir, NULL, name, len, 0);
497 err = mdc_create(sbi->ll_mdc_exp, &op_data, NULL, 0, mode,
498 current->fsuid, current->fsgid,
500 ptlrpc_req_finished(request);
511 static int ll_mknod(struct inode *dir, struct dentry *child, int mode,
514 struct ptlrpc_request *request = NULL;
515 struct inode *inode = NULL;
516 const char *name = child->d_name.name;
517 int len = child->d_name.len;
518 struct ll_sb_info *sbi = ll_i2sbi(dir);
519 struct mdc_op_data op_data;
523 CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p)\n",
524 name, dir->i_ino, dir->i_generation, dir);
526 if (dir->i_nlink >= EXT3_LINK_MAX)
529 mode &= ~current->fs->umask;
531 switch (mode & S_IFMT) {
534 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
539 ll_prepare_mdc_op_data(&op_data, dir, NULL, name, len, 0);
540 err = mdc_create(sbi->ll_mdc_exp, &op_data, NULL, 0, mode,
541 current->fsuid, current->fsgid,
543 err = ll_prep_inode(sbi->ll_osc_exp, &inode, request, 0,
555 d_instantiate(child, inode);
557 ptlrpc_req_finished(request);
561 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
563 struct inode *dir = nd->dentry->d_inode;
564 const char *name = nd->last.name;
565 int len = nd->last.len;
566 struct ptlrpc_request *request = NULL;
567 struct ll_sb_info *sbi = ll_i2sbi(dir);
568 struct mdc_op_data op_data;
572 CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p),target=%s\n",
573 name, dir->i_ino, dir->i_generation, dir, tgt);
575 if (dir->i_nlink >= EXT3_LINK_MAX)
578 ll_prepare_mdc_op_data(&op_data, dir, NULL, name, len, 0);
579 err = mdc_create(sbi->ll_mdc_exp, &op_data,
580 tgt, strlen(tgt) + 1, S_IFLNK | S_IRWXUGO,
581 current->fsuid, current->fsgid, 0, &request);
582 ptlrpc_req_finished(request);
586 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
588 struct inode *src = srcnd->dentry->d_inode;
589 struct inode *dir = tgtnd->dentry->d_inode;
590 const char *name = tgtnd->last.name;
591 int len = tgtnd->last.len;
592 struct ptlrpc_request *request = NULL;
593 struct mdc_op_data op_data;
595 struct ll_sb_info *sbi = ll_i2sbi(dir);
598 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),dir=%lu/%u(%p),target=%s\n",
599 src->i_ino, src->i_generation, src,
600 dir->i_ino, dir->i_generation, dir, name);
602 ll_prepare_mdc_op_data(&op_data, src, dir, name, len, 0);
603 err = mdc_link(sbi->ll_mdc_exp, &op_data, &request);
604 ptlrpc_req_finished(request);
610 static int ll_mkdir_raw(struct nameidata *nd, int mode)
612 struct inode *dir = nd->dentry->d_inode;
613 const char *name = nd->last.name;
614 int len = nd->last.len;
615 struct ptlrpc_request *request = NULL;
616 struct ll_sb_info *sbi = ll_i2sbi(dir);
617 struct mdc_op_data op_data;
620 CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p)\n",
621 name, dir->i_ino, dir->i_generation, dir);
623 if (dir->i_nlink >= EXT3_LINK_MAX)
626 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
627 ll_prepare_mdc_op_data(&op_data, dir, NULL, name, len, 0);
628 err = mdc_create(sbi->ll_mdc_exp, &op_data, NULL, 0, mode,
629 current->fsuid, current->fsgid, 0, &request);
630 ptlrpc_req_finished(request);
634 static int ll_rmdir_raw(struct nameidata *nd)
636 struct inode *dir = nd->dentry->d_inode;
637 const char *name = nd->last.name;
638 int len = nd->last.len;
639 struct ptlrpc_request *request = NULL;
640 struct mdc_op_data op_data;
643 CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p)\n",
644 name, dir->i_ino, dir->i_generation, dir);
646 ll_prepare_mdc_op_data(&op_data, dir, NULL, name, len, S_IFDIR);
647 rc = mdc_unlink(ll_i2sbi(dir)->ll_mdc_exp, &op_data, &request);
648 ptlrpc_req_finished(request);
652 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
654 struct mds_body *body;
655 struct lov_mds_md *eadata;
656 struct lov_stripe_md *lsm = NULL;
657 struct obd_trans_info oti = { 0 };
662 /* req is swabbed so this is safe */
663 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
665 if (!(body->valid & OBD_MD_FLEASIZE))
668 if (body->eadatasize == 0) {
669 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
670 GOTO(out, rc = -EPROTO);
673 /* The MDS sent back the EA because we unlinked the last reference
674 * to this file. Use this EA to unlink the objects on the OST.
675 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
676 * check it is complete and sensible. */
677 eadata = lustre_swab_repbuf(request, 1, body->eadatasize, NULL);
678 LASSERT(eadata != NULL);
679 if (eadata == NULL) {
680 CERROR("Can't unpack MDS EA data\n");
681 GOTO(out, rc = -EPROTO);
684 rc = obd_unpackmd(ll_i2obdexp(dir), &lsm, eadata, body->eadatasize);
686 CERROR("obd_unpackmd: %d\n", rc);
689 LASSERT(rc >= sizeof(*lsm));
693 GOTO(out_free_memmd, rc = -ENOMEM);
695 oa->o_id = lsm->lsm_object_id;
696 oa->o_mode = body->mode & S_IFMT;
697 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
699 if (body->valid & OBD_MD_FLCOOKIE) {
700 oa->o_valid |= OBD_MD_FLCOOKIE;
702 lustre_msg_buf(request->rq_repmsg, 2,
703 sizeof(struct llog_cookie) *
704 lsm->lsm_stripe_count);
705 if (oti.oti_logcookies == NULL) {
706 oa->o_valid &= ~OBD_MD_FLCOOKIE;
707 body->valid &= ~OBD_MD_FLCOOKIE;
711 rc = obd_destroy(ll_i2obdexp(dir), oa, lsm, &oti);
714 CERROR("obd destroy objid "LPX64" error %d\n",
715 lsm->lsm_object_id, rc);
717 obd_free_memmd(ll_i2obdexp(dir), &lsm);
722 static int ll_unlink_raw(struct nameidata *nd)
724 struct inode *dir = nd->dentry->d_inode;
725 const char *name = nd->last.name;
726 int len = nd->last.len;
727 struct ptlrpc_request *request = NULL;
728 struct mdc_op_data op_data;
731 CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu/%u(%p)\n",
732 name, dir->i_ino, dir->i_generation, dir);
734 ll_prepare_mdc_op_data(&op_data, dir, NULL, name, len, 0);
735 rc = mdc_unlink(ll_i2sbi(dir)->ll_mdc_exp, &op_data, &request);
739 rc = ll_objects_destroy(request, dir);
741 ptlrpc_req_finished(request);
745 static int ll_rename_raw(struct nameidata *oldnd, struct nameidata *newnd)
747 struct inode *src = oldnd->dentry->d_inode;
748 struct inode *tgt = newnd->dentry->d_inode;
749 const char *oldname = oldnd->last.name;
750 int oldlen = oldnd->last.len;
751 const char *newname = newnd->last.name;
752 int newlen = newnd->last.len;
753 struct ptlrpc_request *request = NULL;
754 struct ll_sb_info *sbi = ll_i2sbi(src);
755 struct mdc_op_data op_data;
758 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%s,src_dir=%lu/%u(%p),newname=%s,"
759 "tgt_dir=%lu/%u(%p)\n", oldname, src->i_ino, src->i_generation,
760 src, newname, tgt->i_ino, tgt->i_generation, tgt);
762 ll_prepare_mdc_op_data(&op_data, src, tgt, NULL, 0, 0);
763 err = mdc_rename(sbi->ll_mdc_exp, &op_data,
764 oldname, oldlen, newname, newlen, &request);
766 err = ll_objects_destroy(request, src);
769 ptlrpc_req_finished(request);
774 struct inode_operations ll_dir_inode_operations = {
775 link_raw: ll_link_raw,
776 unlink_raw: ll_unlink_raw,
777 symlink_raw: ll_symlink_raw,
778 mkdir_raw: ll_mkdir_raw,
779 rmdir_raw: ll_rmdir_raw,
780 mknod_raw: ll_mknod_raw,
782 rename_raw: ll_rename_raw,
784 setattr_raw: ll_setattr_raw,
785 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
786 create_it: ll_create_it,
787 lookup_it: ll_lookup_it,
788 revalidate_it: ll_inode_revalidate_it,
790 lookup: ll_lookup_nd,
791 create: ll_create_nd,
792 getattr_it: ll_getattr,