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) 2002-2004 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;
75 * remove the stale inode from pnode
77 void unhook_stale_inode(struct pnode *pno)
79 struct inode *inode = pno->p_base->pb_ino;
83 LASSERT(llu_i2info(inode)->lli_stale_flag);
85 pno->p_base->pb_ino = NULL;
88 if (!llu_i2info(inode)->lli_open_count) {
89 CDEBUG(D_INODE, "unhook inode %p (ino %lu) from pno %p\n",
90 inode, llu_i2info(inode)->lli_st_ino, pno);
100 void llu_lookup_finish_locks(struct lookup_intent *it, struct pnode *pnode)
105 if (it && pnode->p_base->pb_ino != NULL) {
106 struct inode *inode = pnode->p_base->pb_ino;
107 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%lu)\n",
108 inode, llu_i2info(inode)->lli_st_ino,
109 llu_i2info(inode)->lli_st_generation);
110 mdc_set_lock_data(NULL, &it->d.lustre.it_lock_handle, inode);
113 /* drop lookup/getattr locks */
114 if (it->it_op == IT_LOOKUP || it->it_op == IT_GETATTR)
115 ll_intent_release(it);
119 static inline void llu_invalidate_inode_pages(struct inode * inode)
124 int llu_mdc_blocking_ast(struct ldlm_lock *lock,
125 struct ldlm_lock_desc *desc,
126 void *data, int flag)
129 struct lustre_handle lockh;
134 case LDLM_CB_BLOCKING:
135 ldlm_lock2handle(lock, &lockh);
136 rc = ldlm_cli_cancel(&lockh);
138 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
142 case LDLM_CB_CANCELING: {
143 struct inode *inode = llu_inode_from_lock(lock);
144 struct llu_inode_info *lli;
145 __u64 bits = lock->l_policy_data.l_inodebits.bits;
147 /* Invalidate all dentries associated with this inode */
151 lli = llu_i2info(inode);
153 if (bits & MDS_INODELOCK_UPDATE)
154 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
156 if (lock->l_resource->lr_name.name[0] != id_fid(&lli->lli_id) ||
157 lock->l_resource->lr_name.name[1] != id_group(&lli->lli_id)) {
158 LDLM_ERROR(lock, "data mismatch with object "DLID4,
159 OLID4(&lli->lli_id));
161 if (S_ISDIR(lli->lli_st_mode) &&
162 (bits & MDS_INODELOCK_UPDATE)) {
163 CDEBUG(D_INODE, "invalidating inode %lu\n",
166 llu_invalidate_inode_pages(inode);
179 static int pnode_revalidate_finish(struct ptlrpc_request *req,
181 struct lookup_intent *it,
184 struct inode *inode = pnode->p_base->pb_ino;
194 if (it_disposition(it, DISP_LOOKUP_NEG))
197 rc = mdc_req2lustre_md(llu_i2sbi(inode)->ll_lmv_exp, req, offset,
198 llu_i2sbi(inode)->ll_lov_exp, &md);
202 llu_update_inode(inode, md.body, md.lsm);
207 int llu_pb_revalidate(struct pnode *pnode, int flags, struct lookup_intent *it)
209 struct pnode_base *pb = pnode->p_base;
210 struct lustre_id pid, cid;
211 struct it_cb_data icbd;
212 struct ptlrpc_request *req = NULL;
213 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
214 struct obd_export *exp;
218 CDEBUG(D_VFSTRACE, "VFS Op:name=%s,intent=%x\n",
219 pb->pb_name.name, it ? it->it_op : 0);
221 /* We don't want to cache negative dentries, so return 0 immediately.
222 * We believe that this is safe, that negative dentries cannot be
223 * pinned by someone else */
224 if (pb->pb_ino == NULL) {
225 CDEBUG(D_INODE, "negative pb\n");
229 /* This is due to bad interaction with libsysio. remove this when we
230 * switched to libbsdio XXX
233 struct llu_inode_info *lli = llu_i2info(pb->pb_ino);
235 CDEBUG(D_INODE, "inode %lu still have intent "
236 "%p(opc 0x%x), release it\n",
237 lli->lli_st_ino, lli->lli_it,
239 ll_intent_release(lli->lli_it);
240 OBD_FREE(lli->lli_it, sizeof(*lli->lli_it));
245 exp = llu_i2mdcexp(pb->pb_ino);
246 ll_inode2id(&pid, pnode->p_parent->p_base->pb_ino);
247 ll_inode2id(&cid, pb->pb_ino);
248 icbd.icbd_parent = pnode->p_parent->p_base->pb_ino;
249 icbd.icbd_child = pnode;
253 it->it_op_release = ll_intent_release;
256 rc = mdc_intent_lock(exp, &pid, pb->pb_name.name, pb->pb_name.len,
257 NULL, 0, &cid, it, flags, &req, llu_mdc_blocking_ast);
258 /* If req is NULL, then mdc_intent_lock only tried to do a lock match;
259 * if all was well, it will return 1 if it found locks, 0 otherwise. */
260 if (req == NULL && rc >= 0)
266 rc = pnode_revalidate_finish(req, 1, it, pnode);
268 ll_intent_release(it);
273 /* Note: ll_intent_lock may cause a callback, check this! */
275 if (it->it_op & IT_OPEN)
276 LL_SAVE_INTENT(pb->pb_ino, it);
280 ptlrpc_req_finished(req);
286 llu_lookup_finish_locks(it, pnode);
291 static int lookup_it_finish(struct ptlrpc_request *request, int offset,
292 struct lookup_intent *it, void *data)
294 struct it_cb_data *icbd = data;
295 struct pnode *child = icbd->icbd_child;
296 struct inode *parent = icbd->icbd_parent;
297 struct llu_sb_info *sbi = llu_i2sbi(parent);
298 struct inode *inode = NULL;
301 /* libsysio require us generate inode right away if success.
302 * so if mds created new inode for us we need make sure it
303 * succeeded. thus for any error we can't delay to the
304 * llu_file_open() time. */
305 if (it_disposition(it, DISP_OPEN_CREATE) &&
306 it_open_error(DISP_OPEN_CREATE, it)) {
307 CDEBUG(D_INODE, "detect mds create error\n");
308 return it_open_error(DISP_OPEN_CREATE, it);
310 if (it_disposition(it, DISP_OPEN_OPEN) &&
311 it_open_error(DISP_OPEN_OPEN, it)) {
312 CDEBUG(D_INODE, "detect mds open error\n");
313 /* undo which did by mdc_intent_lock */
314 if (it_disposition(it, DISP_OPEN_CREATE) &&
315 !it_open_error(DISP_OPEN_CREATE, it)) {
317 LASSERT(atomic_read(&request->rq_refcount) > 1);
318 CDEBUG(D_INODE, "dec a ref of req %p\n", request);
319 ptlrpc_req_finished(request);
321 return it_open_error(DISP_OPEN_OPEN, it);
324 /* NB 1 request reference will be taken away by ll_intent_lock()
327 if (!it_disposition(it, DISP_LOOKUP_NEG) ||
328 (it->it_op & IT_CREAT)) {
330 struct llu_inode_info *lli;
333 rc = mdc_req2lustre_md(sbi->ll_lmv_exp, request, offset,
334 sbi->ll_lov_exp, &md);
338 inode = llu_iget(parent->i_fs, &md);
339 if (!inode || IS_ERR(inode)) {
340 /* free the lsm if we allocated one above */
342 obd_free_memmd(sbi->ll_lov_exp, &md.lsm);
343 RETURN(inode ? PTR_ERR(inode) : -ENOMEM);
344 } else if (md.lsm != NULL &&
345 llu_i2info(inode)->lli_smd != md.lsm) {
346 obd_free_memmd(sbi->ll_lov_exp, &md.lsm);
349 lli = llu_i2info(inode);
351 /* If this is a stat, get the authoritative file size */
352 if (it->it_op == IT_GETATTR && S_ISREG(lli->lli_st_mode) &&
353 lli->lli_smd != NULL) {
354 struct lov_stripe_md *lsm = lli->lli_smd;
357 LASSERT(lsm->lsm_object_id != 0);
359 /* bug 2334: drop MDS lock before acquiring OST lock */
360 ll_intent_drop_lock(it);
362 rc = llu_glimpse_size(inode);
372 /* intent will be further used in cases of open()/getattr() */
373 if (inode && (it->it_op & IT_OPEN))
374 LL_SAVE_INTENT(inode, it);
376 child->p_base->pb_ino = inode;
381 struct inode *llu_inode_from_lock(struct ldlm_lock *lock)
384 l_lock(&lock->l_resource->lr_namespace->ns_lock);
386 if (lock->l_ast_data) {
387 inode = (struct inode *)lock->l_ast_data;
392 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
396 static int llu_lookup_it(struct inode *parent, struct pnode *pnode,
397 struct lookup_intent *it, int flags)
399 struct lustre_id pid;
400 struct it_cb_data icbd;
401 struct ptlrpc_request *req = NULL;
402 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
406 if (pnode->p_base->pb_name.len > EXT2_NAME_LEN)
407 RETURN(-ENAMETOOLONG);
411 it->it_op_release = ll_intent_release;
414 icbd.icbd_child = pnode;
415 icbd.icbd_parent = parent;
416 icbd.icbd_child = pnode;
417 ll_inode2id(&pid, parent);
419 rc = mdc_intent_lock(llu_i2mdcexp(parent), &pid,
420 pnode->p_base->pb_name.name,
421 pnode->p_base->pb_name.len,
422 NULL, 0, NULL, it, flags, &req,
423 llu_mdc_blocking_ast);
427 rc = lookup_it_finish(req, 1, it, &icbd);
429 ll_intent_release(it);
433 llu_lookup_finish_locks(it, pnode);
437 ptlrpc_req_finished(req);
441 static struct lookup_intent*
442 translate_lookup_intent(struct intent *intent, const char *path)
444 struct lookup_intent *it;
448 if (!intent || path) {
449 CDEBUG(D_VFSTRACE, "not intent needed\n");
453 OBD_ALLOC(it, sizeof(*it));
456 memset(it, 0, sizeof(*it));
458 /* libsysio will assign intent like following:
459 * NOTE: INT_CREAT has include INT_UPDPARENT
461 * open: INT_OPEN [| INT_CREAT]
464 * unlink: INT_UPDPARENT
465 * rmdir: INT_UPDPARENT
470 * following logic is adjusted for libsysio
473 it->it_flags = intent->int_arg2 ? *((int*)intent->int_arg2) : 0;
475 if (intent->int_opmask & INT_OPEN) {
476 it->it_op |= IT_OPEN;
478 /* convert access mode from O_ to FMODE_ */
479 if (it->it_flags & O_WRONLY)
481 else if (it->it_flags & O_RDWR)
482 fmode = FMODE_READ | FMODE_WRITE;
485 it->it_flags &= ~O_ACCMODE;
486 it->it_flags |= fmode;
489 /* XXX libsysio has strange code on intent handling,
490 * more check later */
491 if (it->it_flags & O_CREAT) {
492 it->it_op |= IT_CREAT;
493 it->it_create_mode = *((int*)intent->int_arg1);
496 if (intent->int_opmask & INT_GETATTR)
497 it->it_op |= IT_GETATTR;
499 LASSERT(!(intent->int_opmask & INT_SETATTR));
501 /* libsysio is different to linux vfs when doing unlink/rmdir,
502 * INT_UPDPARENT was passed down during name resolution. Here
503 * we treat it as normal lookup, later unlink()/rmdir() will
504 * do the actual work */
506 /* conform to kernel code, if only IT_LOOKUP was set, don't
508 if (!it->it_op || it->it_op == IT_LOOKUP) {
509 OBD_FREE(it, sizeof(*it));
513 it->it_op_release = ll_intent_release;
515 CDEBUG(D_VFSTRACE, "final intent 0x%x\n", it ? it->it_op : 0);
519 int llu_iop_lookup(struct pnode *pnode,
521 struct intent *intnt,
524 struct lookup_intent *it;
528 liblustre_wait_event(0);
532 /* the mount root inode have no name, so don't call
533 * remote in this case. but probably we need revalidate
535 if (pnode->p_mount->mnt_root == pnode) {
536 struct inode *i = pnode->p_base->pb_ino;
541 if (!pnode->p_base->pb_name.len)
544 it = translate_lookup_intent(intnt, path);
546 /* param flags is not used, let it be 0 */
547 if (llu_pb_revalidate(pnode, 0, it)) {
548 LASSERT(pnode->p_base->pb_ino);
549 *inop = pnode->p_base->pb_ino;
553 rc = llu_lookup_it(pnode->p_parent->p_base->pb_ino, pnode, it, 0);
555 if (!pnode->p_base->pb_ino)
558 *inop = pnode->p_base->pb_ino;