1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Lustre Light name resolution
6 * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
8 * This file is part of Lustre, http://www.lustre.org.
10 * Lustre is free software; you can redistribute it and/or
11 * modify it under the terms of version 2 of the GNU General Public
12 * License as published by the Free Software Foundation.
14 * Lustre is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with Lustre; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 #define DEBUG_SUBSYSTEM S_LLITE
30 #include <sys/types.h>
32 #include <sys/fcntl.h>
33 #include <sys/queue.h>
43 #include "llite_lib.h"
45 static void ll_intent_drop_lock(struct lookup_intent *it)
47 struct lustre_handle *handle;
49 if (it->it_op && it->d.lustre.it_lock_mode) {
50 handle = (struct lustre_handle *)&it->d.lustre.it_lock_handle;
51 CDEBUG(D_DLMTRACE, "releasing lock with cookie "LPX64
52 " from it %p\n", handle->cookie, it);
53 ldlm_lock_decref(handle, it->d.lustre.it_lock_mode);
55 /* bug 494: intent_release may be called multiple times, from
56 * this thread and we don't want to double-decref this lock */
57 it->d.lustre.it_lock_mode = 0;
61 static void ll_intent_release(struct lookup_intent *it)
65 ll_intent_drop_lock(it);
67 it->it_op_release = 0;
68 it->d.lustre.it_disposition = 0;
69 it->d.lustre.it_data = NULL;
74 * remove the stale inode from pnode
76 void unhook_stale_inode(struct pnode *pno)
78 struct inode *inode = pno->p_base->pb_ino;
82 LASSERT(llu_i2info(inode)->lli_stale_flag);
84 pno->p_base->pb_ino = NULL;
87 if (!llu_i2info(inode)->lli_open_count) {
88 CDEBUG(D_INODE, "unhook inode %p (ino %lu) from pno %p\n",
89 inode, llu_i2info(inode)->lli_st_ino, pno);
98 void llu_lookup_finish_locks(struct lookup_intent *it, struct pnode *pnode)
103 if (it && pnode->p_base->pb_ino != NULL) {
104 struct inode *inode = pnode->p_base->pb_ino;
105 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%lu)\n",
106 inode, llu_i2info(inode)->lli_st_ino,
107 llu_i2info(inode)->lli_st_generation);
108 mdc_set_lock_data(NULL, &it->d.lustre.it_lock_handle, inode);
111 /* drop lookup/getattr locks */
112 if (it->it_op == IT_LOOKUP || it->it_op == IT_GETATTR)
113 ll_intent_release(it);
117 static inline void llu_invalidate_inode_pages(struct inode * inode)
122 int llu_mdc_blocking_ast(struct ldlm_lock *lock,
123 struct ldlm_lock_desc *desc,
124 void *data, int flag)
127 struct lustre_handle lockh;
132 case LDLM_CB_BLOCKING:
133 ldlm_lock2handle(lock, &lockh);
134 rc = ldlm_cli_cancel(&lockh);
136 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
140 case LDLM_CB_CANCELING: {
141 struct inode *inode = llu_inode_from_lock(lock);
142 struct llu_inode_info *lli;
143 __u64 bits = lock->l_policy_data.l_inodebits.bits;
145 /* Invalidate all dentries associated with this inode */
149 lli = llu_i2info(inode);
151 if (bits & MDS_INODELOCK_UPDATE)
152 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
154 if (lock->l_resource->lr_name.name[0] != lli->lli_st_ino ||
155 lock->l_resource->lr_name.name[1] != lli->lli_st_generation) {
156 LDLM_ERROR(lock, "data mismatch with ino %lu/%lu",
157 lli->lli_st_ino, lli->lli_st_generation);
159 if (S_ISDIR(lli->lli_st_mode) &&
160 (bits & MDS_INODELOCK_UPDATE)) {
161 CDEBUG(D_INODE, "invalidating inode %lu\n",
164 llu_invalidate_inode_pages(inode);
168 if (inode->i_sb->s_root &&
169 inode != inode->i_sb->s_root->d_inode)
170 ll_unhash_aliases(inode);
182 static int pnode_revalidate_finish(struct ptlrpc_request *req,
184 struct lookup_intent *it,
187 struct inode *inode = pnode->p_base->pb_ino;
197 if (it_disposition(it, DISP_LOOKUP_NEG))
200 rc = mdc_req2lustre_md(llu_i2sbi(inode)->ll_mdc_exp, req, offset,
201 llu_i2sbi(inode)->ll_osc_exp, &md);
205 llu_update_inode(inode, md.body, md.lsm);
210 int llu_pb_revalidate(struct pnode *pnode, int flags, struct lookup_intent *it)
212 struct pnode_base *pb = pnode->p_base;
213 struct ll_fid pfid, cfid;
214 struct it_cb_data icbd;
215 struct ll_uctxt ctxt;
216 struct ptlrpc_request *req = NULL;
217 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
218 struct obd_export *exp;
222 CDEBUG(D_VFSTRACE, "VFS Op:name=%s,intent=%x\n",
223 pb->pb_name.name, it ? it->it_op : 0);
225 /* We don't want to cache negative dentries, so return 0 immediately.
226 * We believe that this is safe, that negative dentries cannot be
227 * pinned by someone else */
228 if (pb->pb_ino == NULL) {
229 CDEBUG(D_INODE, "negative pb\n");
233 /* check stale inode */
234 if (llu_i2info(pb->pb_ino)->lli_stale_flag)
235 unhook_stale_inode(pnode);
237 /* check again because unhook_stale_inode() might generate
239 if (pb->pb_ino == NULL) {
240 CDEBUG(D_INODE, "negative pb\n");
244 /* This is due to bad interaction with libsysio. remove this when we
245 * switched to libbsdio XXX
248 struct llu_inode_info *lli = llu_i2info(pb->pb_ino);
250 CDEBUG(D_INODE, "inode %lu still have intent "
251 "%p(opc 0x%x), release it\n",
252 lli->lli_st_ino, lli->lli_it,
254 ll_intent_release(lli->lli_it);
255 OBD_FREE(lli->lli_it, sizeof(*lli->lli_it));
260 exp = llu_i2mdcexp(pb->pb_ino);
261 ll_inode2fid(&pfid, pnode->p_parent->p_base->pb_ino);
262 ll_inode2fid(&cfid, pb->pb_ino);
263 icbd.icbd_parent = pnode->p_parent->p_base->pb_ino;
264 icbd.icbd_child = pnode;
268 it->it_op_release = ll_intent_release;
271 ll_i2uctxt(&ctxt, pnode->p_parent->p_base->pb_ino, pb->pb_ino);
273 rc = mdc_intent_lock(exp, &ctxt, &pfid,
274 pb->pb_name.name, pb->pb_name.len,
275 NULL, 0, &cfid, it, flags, &req,
276 llu_mdc_blocking_ast);
277 /* If req is NULL, then mdc_intent_lock only tried to do a lock match;
278 * if all was well, it will return 1 if it found locks, 0 otherwise. */
279 if (req == NULL && rc >= 0)
285 rc = pnode_revalidate_finish(req, 1, it, pnode);
287 ll_intent_release(it);
292 /* Note: ll_intent_lock may cause a callback, check this! */
294 if (it->it_op & IT_OPEN)
295 LL_SAVE_INTENT(pb->pb_ino, it);
299 ptlrpc_req_finished(req);
305 llu_lookup_finish_locks(it, pnode);
306 llu_i2info(pb->pb_ino)->lli_stale_flag = 0;
311 static int lookup_it_finish(struct ptlrpc_request *request, int offset,
312 struct lookup_intent *it, void *data)
314 struct it_cb_data *icbd = data;
315 struct pnode *child = icbd->icbd_child;
316 struct inode *parent = icbd->icbd_parent;
317 struct llu_sb_info *sbi = llu_i2sbi(parent);
318 struct inode *inode = NULL;
321 /* NB 1 request reference will be taken away by ll_intent_lock()
324 /* FIXME: for CREAT, libsysio require the inode must be generated here
325 * currently here we don't know the whether the create is successful
326 * or failed on mds. thus blinded return -EPERM in llu_iget(). need
329 if ((it->it_op & IT_CREAT) || !it_disposition(it, DISP_LOOKUP_NEG)) {
331 struct llu_inode_info *lli;
334 rc = mdc_req2lustre_md(sbi->ll_mdc_exp, request, offset,
335 sbi->ll_osc_exp, &md);
339 inode = llu_iget(parent->i_fs, &md);
340 if (!inode || IS_ERR(inode)) {
341 /* free the lsm if we allocated one above */
343 obd_free_memmd(sbi->ll_osc_exp, &md.lsm);
344 RETURN(inode ? PTR_ERR(inode) : -ENOMEM);
345 } else if (md.lsm != NULL &&
346 llu_i2info(inode)->lli_smd != md.lsm) {
347 obd_free_memmd(sbi->ll_osc_exp, &md.lsm);
350 lli = llu_i2info(inode);
352 /* If this is a stat, get the authoritative file size */
353 if (it->it_op == IT_GETATTR && S_ISREG(lli->lli_st_mode) &&
354 lli->lli_smd != NULL) {
355 struct lov_stripe_md *lsm = lli->lli_smd;
359 LASSERT(lsm->lsm_object_id != 0);
361 /* bug 2334: drop MDS lock before acquiring OST lock */
362 ll_intent_drop_lock(it);
364 rc = llu_glimpse_size(inode, &lvb);
369 lli->lli_st_size = lvb.lvb_size;
375 /* intent will be further used in cases of open()/getattr() */
376 if (inode && (it->it_op & IT_OPEN))
377 LL_SAVE_INTENT(inode, it);
379 child->p_base->pb_ino = inode;
384 struct inode *llu_inode_from_lock(struct ldlm_lock *lock)
387 l_lock(&lock->l_resource->lr_namespace->ns_lock);
389 if (lock->l_ast_data) {
390 inode = (struct inode *)lock->l_ast_data;
395 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
399 static int llu_lookup_it(struct inode *parent, struct pnode *pnode,
400 struct lookup_intent *it, int flags)
403 struct ll_uctxt ctxt;
404 struct it_cb_data icbd;
405 struct ptlrpc_request *req = NULL;
406 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
410 if (pnode->p_base->pb_name.len > EXT2_NAME_LEN)
411 RETURN(-ENAMETOOLONG);
415 it->it_op_release = ll_intent_release;
418 icbd.icbd_child = pnode;
419 icbd.icbd_parent = parent;
420 icbd.icbd_child = pnode;
421 ll_inode2fid(&pfid, parent);
422 ll_i2uctxt(&ctxt, parent, NULL);
424 rc = mdc_intent_lock(llu_i2mdcexp(parent), &ctxt, &pfid,
425 pnode->p_base->pb_name.name,
426 pnode->p_base->pb_name.len,
427 NULL, 0, NULL, it, flags, &req,
428 llu_mdc_blocking_ast);
432 rc = lookup_it_finish(req, 1, it, &icbd);
434 ll_intent_release(it);
438 llu_lookup_finish_locks(it, pnode);
442 ptlrpc_req_finished(req);
446 static struct lookup_intent*
447 translate_lookup_intent(struct intent *intent, const char *path)
449 struct lookup_intent *it;
453 if (!intent || path) {
454 CDEBUG(D_VFSTRACE, "not intent needed\n");
458 OBD_ALLOC(it, sizeof(*it));
461 memset(it, 0, sizeof(*it));
463 /* libsysio will assign intent like following:
464 * NOTE: INT_CREAT has include INT_UPDPARENT
466 * open: INT_OPEN [| INT_CREAT]
469 * unlink: INT_UPDPARENT
470 * rmdir: INT_UPDPARENT
475 * following logic is adjusted for libsysio
478 it->it_flags = intent->int_arg2 ? *((int*)intent->int_arg2) : 0;
480 if (intent->int_opmask & INT_OPEN) {
481 it->it_op |= IT_OPEN;
483 /* convert access mode from O_ to FMODE_ */
484 if (it->it_flags & O_WRONLY)
486 else if (it->it_flags & O_RDWR)
487 fmode = FMODE_READ | FMODE_WRITE;
490 it->it_flags &= ~O_ACCMODE;
491 it->it_flags |= fmode;
494 /* XXX libsysio has strange code on intent handling,
495 * more check later */
496 if (it->it_flags & O_CREAT) {
497 it->it_op |= IT_CREAT;
498 it->it_create_mode = *((int*)intent->int_arg1);
501 if (intent->int_opmask & INT_GETATTR)
502 it->it_op |= IT_GETATTR;
504 LASSERT(!(intent->int_opmask & INT_SETATTR));
506 /* libsysio is different to linux vfs when doing unlink/rmdir,
507 * INT_UPDPARENT was passed down during name resolution. Here
508 * we treat it as normal lookup, later unlink()/rmdir() will
509 * do the actual work */
511 /* conform to kernel code, if only IT_LOOKUP was set, don't
513 if (!it->it_op || it->it_op == IT_LOOKUP) {
514 OBD_FREE(it, sizeof(*it));
518 it->it_op_release = ll_intent_release;
520 CDEBUG(D_VFSTRACE, "final intent 0x%x\n", it ? it->it_op : 0);
524 int llu_iop_lookup(struct pnode *pnode,
526 struct intent *intnt,
529 struct lookup_intent *it;
535 /* the mount root inode have no name, so don't call
536 * remote in this case. but probably we need revalidate
538 if (pnode->p_mount->mnt_root == pnode) {
539 struct inode *i = pnode->p_base->pb_ino;
544 if (!pnode->p_base->pb_name.len)
547 it = translate_lookup_intent(intnt, path);
549 /* param flags is not used, let it be 0 */
550 if (llu_pb_revalidate(pnode, 0, it)) {
551 LASSERT(pnode->p_base->pb_ino);
552 *inop = pnode->p_base->pb_ino;
556 rc = llu_lookup_it(pnode->p_parent->p_base->pb_ino, pnode, it, 0);
558 if (!pnode->p_base->pb_ino)
561 *inop = pnode->p_base->pb_ino;