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] != lli->lli_st_ino ||
157 lock->l_resource->lr_name.name[1] != lli->lli_st_generation) {
158 LDLM_ERROR(lock, "data mismatch with ino %lu/%lu",
159 lli->lli_st_ino, lli->lli_st_generation);
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);
170 if (inode->i_sb->s_root &&
171 inode != inode->i_sb->s_root->d_inode)
172 ll_unhash_aliases(inode);
184 static int pnode_revalidate_finish(struct ptlrpc_request *req,
186 struct lookup_intent *it,
189 struct inode *inode = pnode->p_base->pb_ino;
199 if (it_disposition(it, DISP_LOOKUP_NEG))
202 rc = mdc_req2lustre_md(llu_i2sbi(inode)->ll_mdc_exp, req, offset,
203 llu_i2sbi(inode)->ll_osc_exp, &md);
207 llu_update_inode(inode, md.body, md.lsm);
212 int llu_pb_revalidate(struct pnode *pnode, int flags, struct lookup_intent *it)
214 struct pnode_base *pb = pnode->p_base;
215 struct ll_fid pfid, cfid;
216 struct it_cb_data icbd;
217 struct ll_uctxt ctxt;
218 struct ptlrpc_request *req = NULL;
219 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
220 struct obd_export *exp;
224 CDEBUG(D_VFSTRACE, "VFS Op:name=%s,intent=%x\n",
225 pb->pb_name.name, it ? it->it_op : 0);
227 /* We don't want to cache negative dentries, so return 0 immediately.
228 * We believe that this is safe, that negative dentries cannot be
229 * pinned by someone else */
230 if (pb->pb_ino == NULL) {
231 CDEBUG(D_INODE, "negative pb\n");
235 /* This is due to bad interaction with libsysio. remove this when we
236 * switched to libbsdio XXX
239 struct llu_inode_info *lli = llu_i2info(pb->pb_ino);
241 CDEBUG(D_INODE, "inode %lu still have intent "
242 "%p(opc 0x%x), release it\n",
243 lli->lli_st_ino, lli->lli_it,
245 ll_intent_release(lli->lli_it);
246 OBD_FREE(lli->lli_it, sizeof(*lli->lli_it));
251 exp = llu_i2mdcexp(pb->pb_ino);
252 ll_inode2fid(&pfid, pnode->p_parent->p_base->pb_ino);
253 ll_inode2fid(&cfid, pb->pb_ino);
254 icbd.icbd_parent = pnode->p_parent->p_base->pb_ino;
255 icbd.icbd_child = pnode;
259 it->it_op_release = ll_intent_release;
262 ll_i2uctxt(&ctxt, pnode->p_parent->p_base->pb_ino, pb->pb_ino);
264 rc = mdc_intent_lock(exp, &ctxt, &pfid,
265 pb->pb_name.name, pb->pb_name.len,
266 NULL, 0, &cfid, it, flags, &req,
267 llu_mdc_blocking_ast);
268 /* If req is NULL, then mdc_intent_lock only tried to do a lock match;
269 * if all was well, it will return 1 if it found locks, 0 otherwise. */
270 if (req == NULL && rc >= 0)
276 rc = pnode_revalidate_finish(req, 1, it, pnode);
278 ll_intent_release(it);
283 /* Note: ll_intent_lock may cause a callback, check this! */
285 if (it->it_op & IT_OPEN)
286 LL_SAVE_INTENT(pb->pb_ino, it);
290 ptlrpc_req_finished(req);
296 llu_lookup_finish_locks(it, pnode);
301 static int lookup_it_finish(struct ptlrpc_request *request, int offset,
302 struct lookup_intent *it, void *data)
304 struct it_cb_data *icbd = data;
305 struct pnode *child = icbd->icbd_child;
306 struct inode *parent = icbd->icbd_parent;
307 struct llu_sb_info *sbi = llu_i2sbi(parent);
308 struct inode *inode = NULL;
311 /* libsysio require us generate inode right away if success.
312 * so if mds created new inode for us we need make sure it
313 * succeeded. thus for any error we can't delay to the
314 * llu_file_open() time. */
315 if (it_disposition(it, DISP_OPEN_CREATE) &&
316 it_open_error(DISP_OPEN_CREATE, it)) {
317 CDEBUG(D_INODE, "detect mds create error\n");
318 return it_open_error(DISP_OPEN_CREATE, it);
320 if (it_disposition(it, DISP_OPEN_OPEN) &&
321 it_open_error(DISP_OPEN_OPEN, it)) {
322 CDEBUG(D_INODE, "detect mds open error\n");
323 /* undo which did by mdc_intent_lock */
324 if (it_disposition(it, DISP_OPEN_CREATE) &&
325 !it_open_error(DISP_OPEN_CREATE, it)) {
327 LASSERT(atomic_read(&request->rq_refcount) > 1);
328 CDEBUG(D_INODE, "dec a ref of req %p\n", request);
329 ptlrpc_req_finished(request);
331 return it_open_error(DISP_OPEN_OPEN, it);
334 /* NB 1 request reference will be taken away by ll_intent_lock()
337 if (!it_disposition(it, DISP_LOOKUP_NEG) ||
338 (it->it_op & IT_CREAT)) {
340 struct llu_inode_info *lli;
343 rc = mdc_req2lustre_md(sbi->ll_mdc_exp, request, offset,
344 sbi->ll_osc_exp, &md);
348 inode = llu_iget(parent->i_fs, &md);
349 if (!inode || IS_ERR(inode)) {
350 /* free the lsm if we allocated one above */
352 obd_free_memmd(sbi->ll_osc_exp, &md.lsm);
353 RETURN(inode ? PTR_ERR(inode) : -ENOMEM);
354 } else if (md.lsm != NULL &&
355 llu_i2info(inode)->lli_smd != md.lsm) {
356 obd_free_memmd(sbi->ll_osc_exp, &md.lsm);
359 lli = llu_i2info(inode);
361 /* If this is a stat, get the authoritative file size */
362 if (it->it_op == IT_GETATTR && S_ISREG(lli->lli_st_mode) &&
363 lli->lli_smd != NULL) {
364 struct lov_stripe_md *lsm = lli->lli_smd;
367 LASSERT(lsm->lsm_object_id != 0);
369 /* bug 2334: drop MDS lock before acquiring OST lock */
370 ll_intent_drop_lock(it);
372 rc = llu_glimpse_size(inode);
382 /* intent will be further used in cases of open()/getattr() */
383 if (inode && (it->it_op & IT_OPEN))
384 LL_SAVE_INTENT(inode, it);
386 child->p_base->pb_ino = inode;
391 struct inode *llu_inode_from_lock(struct ldlm_lock *lock)
394 l_lock(&lock->l_resource->lr_namespace->ns_lock);
396 if (lock->l_ast_data) {
397 inode = (struct inode *)lock->l_ast_data;
402 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
406 static int llu_lookup_it(struct inode *parent, struct pnode *pnode,
407 struct lookup_intent *it, int flags)
410 struct ll_uctxt ctxt;
411 struct it_cb_data icbd;
412 struct ptlrpc_request *req = NULL;
413 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
417 if (pnode->p_base->pb_name.len > EXT2_NAME_LEN)
418 RETURN(-ENAMETOOLONG);
422 it->it_op_release = ll_intent_release;
425 icbd.icbd_child = pnode;
426 icbd.icbd_parent = parent;
427 icbd.icbd_child = pnode;
428 ll_inode2fid(&pfid, parent);
429 ll_i2uctxt(&ctxt, parent, NULL);
431 rc = mdc_intent_lock(llu_i2mdcexp(parent), &ctxt, &pfid,
432 pnode->p_base->pb_name.name,
433 pnode->p_base->pb_name.len,
434 NULL, 0, NULL, it, flags, &req,
435 llu_mdc_blocking_ast);
439 rc = lookup_it_finish(req, 1, it, &icbd);
441 ll_intent_release(it);
445 llu_lookup_finish_locks(it, pnode);
449 ptlrpc_req_finished(req);
453 static struct lookup_intent*
454 translate_lookup_intent(struct intent *intent, const char *path)
456 struct lookup_intent *it;
460 if (!intent || path) {
461 CDEBUG(D_VFSTRACE, "not intent needed\n");
465 OBD_ALLOC(it, sizeof(*it));
468 memset(it, 0, sizeof(*it));
470 /* libsysio will assign intent like following:
471 * NOTE: INT_CREAT has include INT_UPDPARENT
473 * open: INT_OPEN [| INT_CREAT]
476 * unlink: INT_UPDPARENT
477 * rmdir: INT_UPDPARENT
482 * following logic is adjusted for libsysio
485 it->it_flags = intent->int_arg2 ? *((int*)intent->int_arg2) : 0;
487 if (intent->int_opmask & INT_OPEN) {
488 it->it_op |= IT_OPEN;
490 /* convert access mode from O_ to FMODE_ */
491 if (it->it_flags & O_WRONLY)
493 else if (it->it_flags & O_RDWR)
494 fmode = FMODE_READ | FMODE_WRITE;
497 it->it_flags &= ~O_ACCMODE;
498 it->it_flags |= fmode;
501 /* XXX libsysio has strange code on intent handling,
502 * more check later */
503 if (it->it_flags & O_CREAT) {
504 it->it_op |= IT_CREAT;
505 it->it_create_mode = *((int*)intent->int_arg1);
508 if (intent->int_opmask & INT_GETATTR)
509 it->it_op |= IT_GETATTR;
511 LASSERT(!(intent->int_opmask & INT_SETATTR));
513 /* libsysio is different to linux vfs when doing unlink/rmdir,
514 * INT_UPDPARENT was passed down during name resolution. Here
515 * we treat it as normal lookup, later unlink()/rmdir() will
516 * do the actual work */
518 /* conform to kernel code, if only IT_LOOKUP was set, don't
520 if (!it->it_op || it->it_op == IT_LOOKUP) {
521 OBD_FREE(it, sizeof(*it));
525 it->it_op_release = ll_intent_release;
527 CDEBUG(D_VFSTRACE, "final intent 0x%x\n", it ? it->it_op : 0);
531 int llu_iop_lookup(struct pnode *pnode,
533 struct intent *intnt,
536 struct lookup_intent *it;
540 liblustre_wait_event(0);
544 /* the mount root inode have no name, so don't call
545 * remote in this case. but probably we need revalidate
547 if (pnode->p_mount->mnt_root == pnode) {
548 struct inode *i = pnode->p_base->pb_ino;
553 if (!pnode->p_base->pb_name.len)
556 it = translate_lookup_intent(intnt, path);
558 /* param flags is not used, let it be 0 */
559 if (llu_pb_revalidate(pnode, 0, it)) {
560 LASSERT(pnode->p_base->pb_ino);
561 *inop = pnode->p_base->pb_ino;
565 rc = llu_lookup_it(pnode->p_parent->p_base->pb_ino, pnode, it, 0);
567 if (!pnode->p_base->pb_ino)
570 *inop = pnode->p_base->pb_ino;