1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/liblustre/namei.c
38 * Lustre Light name resolution
41 #define DEBUG_SUBSYSTEM S_LLITE
47 #include <sys/types.h>
50 #include <sys/queue.h>
65 #include "llite_lib.h"
67 void ll_intent_drop_lock(struct lookup_intent *it)
69 struct lustre_handle *handle;
71 if (it->it_op && it->d.lustre.it_lock_mode) {
72 handle = (struct lustre_handle *)&it->d.lustre.it_lock_handle;
73 CDEBUG(D_DLMTRACE, "releasing lock with cookie "LPX64
74 " from it %p\n", handle->cookie, it);
75 ldlm_lock_decref(handle, it->d.lustre.it_lock_mode);
77 /* bug 494: intent_release may be called multiple times, from
78 * this thread and we don't want to double-decref this lock */
79 it->d.lustre.it_lock_mode = 0;
83 void ll_intent_release(struct lookup_intent *it)
87 ll_intent_drop_lock(it);
89 it->it_op_release = 0;
90 it->d.lustre.it_disposition = 0;
91 it->d.lustre.it_data = NULL;
95 void llu_lookup_finish_locks(struct lookup_intent *it, struct pnode *pnode)
101 inode = pnode->p_base->pb_ino;
102 if (it->d.lustre.it_lock_mode && inode != NULL) {
103 struct llu_sb_info *sbi;
105 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%llu/%lu)\n",
106 inode, (long long)llu_i2stat(inode)->st_ino,
107 llu_i2info(inode)->lli_st_generation);
109 sbi = llu_i2sbi(inode);
110 md_set_lock_data(sbi->ll_md_exp,
111 &it->d.lustre.it_lock_handle, inode);
114 /* drop lookup/getattr locks */
115 if (it->it_op & (IT_LOOKUP | IT_GETATTR))
116 ll_intent_release(it);
120 static inline void llu_invalidate_inode_pages(struct inode * inode)
125 int llu_md_blocking_ast(struct ldlm_lock *lock,
126 struct ldlm_lock_desc *desc,
127 void *data, int flag)
129 struct lustre_handle lockh;
135 case LDLM_CB_BLOCKING:
136 ldlm_lock2handle(lock, &lockh);
137 rc = ldlm_cli_cancel(&lockh);
139 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
143 case LDLM_CB_CANCELING: {
144 struct inode *inode = llu_inode_from_lock(lock);
145 struct llu_inode_info *lli;
146 struct intnl_stat *st;
147 __u64 bits = lock->l_policy_data.l_inodebits.bits;
150 /* Invalidate all dentries associated with this inode */
154 lli = llu_i2info(inode);
155 st = llu_i2stat(inode);
157 if (bits & MDS_INODELOCK_UPDATE)
158 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
161 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
162 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
163 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
164 LDLM_ERROR(lock,"data mismatch with ino %llu/%llu/%llu",
165 (long long)fid_seq(fid),
166 (long long)fid_oid(fid),
167 (long long)fid_ver(fid));
169 if (S_ISDIR(st->st_mode) &&
170 (bits & MDS_INODELOCK_UPDATE)) {
171 CDEBUG(D_INODE, "invalidating inode %llu\n",
172 (long long)st->st_ino);
174 llu_invalidate_inode_pages(inode);
178 if (inode->i_sb->s_root &&
179 inode != inode->i_sb->s_root->d_inode)
180 ll_unhash_aliases(inode);
192 static int pnode_revalidate_finish(struct ptlrpc_request *req,
193 struct lookup_intent *it,
196 struct inode *inode = pnode->p_base->pb_ino;
206 if (it_disposition(it, DISP_LOOKUP_NEG))
209 rc = md_get_lustre_md(llu_i2sbi(inode)->ll_md_exp, req,
210 llu_i2sbi(inode)->ll_dt_exp,
211 llu_i2sbi(inode)->ll_md_exp, &md);
215 llu_update_inode(inode, &md);
220 static int llu_pb_revalidate(struct pnode *pnode, int flags,
221 struct lookup_intent *it)
223 struct pnode_base *pb = pnode->p_base;
224 struct it_cb_data icbd;
225 struct md_op_data op_data = {{ 0 }};
226 struct ptlrpc_request *req = NULL;
227 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
228 struct obd_export *exp;
232 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,intent=%x\n",
233 (int)pb->pb_name.len, pb->pb_name.name, it ? it->it_op : 0);
235 /* We don't want to cache negative dentries, so return 0 immediately.
236 * We believe that this is safe, that negative dentries cannot be
237 * pinned by someone else */
238 if (pb->pb_ino == NULL) {
239 CDEBUG(D_INODE, "negative pb\n");
243 /* This is due to bad interaction with libsysio. remove this when we
244 * switched to libbsdio XXX
247 struct llu_inode_info *lli = llu_i2info(pb->pb_ino);
248 struct intnl_stat *st = llu_i2stat(pb->pb_ino);
250 CDEBUG(D_INODE, "inode %llu still have intent "
251 "%p(opc 0x%x), release it\n",
252 (long long) st->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_i2mdexp(pb->pb_ino);
261 icbd.icbd_parent = pnode->p_parent->p_base->pb_ino;
262 icbd.icbd_child = pnode;
266 it->it_op_release = ll_intent_release;
269 llu_prep_md_op_data(&op_data, pnode->p_parent->p_base->pb_ino,
270 pb->pb_ino, pb->pb_name.name, pb->pb_name.len,
273 rc = md_intent_lock(exp, &op_data, NULL, 0, it, flags,
274 &req, llu_md_blocking_ast,
275 LDLM_FL_CANCEL_ON_BLOCK);
276 /* If req is NULL, then md_intent_lock only tried to do a lock match;
277 * if all was well, it will return 1 if it found locks, 0 otherwise. */
278 if (req == NULL && rc >= 0)
284 rc = pnode_revalidate_finish(req, it, pnode);
286 ll_intent_release(it);
291 /* Note: ll_intent_lock may cause a callback, check this! */
293 if (it->it_op & IT_OPEN)
294 LL_SAVE_INTENT(pb->pb_ino, it);
298 ptlrpc_req_finished(req);
304 llu_lookup_finish_locks(it, pnode);
309 static int lookup_it_finish(struct ptlrpc_request *request, int offset,
310 struct lookup_intent *it, void *data)
312 struct it_cb_data *icbd = data;
313 struct pnode *child = icbd->icbd_child;
314 struct inode *parent = icbd->icbd_parent;
315 struct llu_sb_info *sbi = llu_i2sbi(parent);
316 struct inode *inode = NULL;
319 /* libsysio require us generate inode right away if success.
320 * so if mds created new inode for us we need make sure it
321 * succeeded. thus for any error we can't delay to the
322 * llu_file_open() time. */
323 if (it_disposition(it, DISP_OPEN_CREATE) &&
324 it_open_error(DISP_OPEN_CREATE, it)) {
325 CDEBUG(D_INODE, "detect mds create error\n");
326 return it_open_error(DISP_OPEN_CREATE, it);
328 if (it_disposition(it, DISP_OPEN_OPEN) &&
329 it_open_error(DISP_OPEN_OPEN, it)) {
330 CDEBUG(D_INODE, "detect mds open error\n");
331 /* undo which did by md_intent_lock */
332 if (it_disposition(it, DISP_OPEN_CREATE) &&
333 !it_open_error(DISP_OPEN_CREATE, it)) {
335 LASSERT(atomic_read(&request->rq_refcount) > 1);
336 CDEBUG(D_INODE, "dec a ref of req %p\n", request);
337 ptlrpc_req_finished(request);
339 return it_open_error(DISP_OPEN_OPEN, it);
342 /* NB 1 request reference will be taken away by ll_intent_lock()
345 if (!it_disposition(it, DISP_LOOKUP_NEG) || (it->it_op & IT_CREAT)) {
347 struct llu_inode_info *lli;
348 struct intnl_stat *st;
351 if (it_disposition(it, DISP_OPEN_CREATE))
352 ptlrpc_req_finished(request);
354 rc = md_get_lustre_md(sbi->ll_md_exp, request,
355 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
359 inode = llu_iget(parent->i_fs, &md);
360 if (!inode || IS_ERR(inode)) {
361 /* free the lsm if we allocated one above */
363 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
364 RETURN(inode ? PTR_ERR(inode) : -ENOMEM);
365 } else if (md.lsm != NULL &&
366 llu_i2info(inode)->lli_smd != md.lsm) {
367 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
370 lli = llu_i2info(inode);
371 st = llu_i2stat(inode);
373 /* If this is a stat, get the authoritative file size */
374 if (it->it_op == IT_GETATTR && S_ISREG(st->st_mode) &&
375 lli->lli_smd != NULL) {
376 struct lov_stripe_md *lsm = lli->lli_smd;
379 LASSERT(lsm->lsm_object_id != 0);
381 /* bug 2334: drop MDS lock before acquiring OST lock */
382 ll_intent_drop_lock(it);
384 rc = cl_glimpse_size(inode);
394 /* intent will be further used in cases of open()/getattr() */
395 if (inode && (it->it_op & IT_OPEN))
396 LL_SAVE_INTENT(inode, it);
398 child->p_base->pb_ino = inode;
403 struct inode *llu_inode_from_lock(struct ldlm_lock *lock)
406 lock_res_and_lock(lock);
408 if (lock->l_ast_data) {
409 inode = (struct inode *)lock->l_ast_data;
414 unlock_res_and_lock(lock);
418 static int llu_lookup_it(struct inode *parent, struct pnode *pnode,
419 struct lookup_intent *it, int flags)
421 struct md_op_data op_data = {{ 0 }};
422 struct it_cb_data icbd;
423 struct ptlrpc_request *req = NULL;
424 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
429 if (pnode->p_base->pb_name.len > EXT2_NAME_LEN)
430 RETURN(-ENAMETOOLONG);
434 it->it_op_release = ll_intent_release;
437 icbd.icbd_child = pnode;
438 icbd.icbd_parent = parent;
440 if (it->it_op & IT_CREAT ||
441 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT)) {
442 opc = LUSTRE_OPC_CREATE;
444 opc = LUSTRE_OPC_ANY;
447 llu_prep_md_op_data(&op_data, parent, NULL,
448 pnode->p_base->pb_name.name,
449 pnode->p_base->pb_name.len, flags, opc);
451 rc = md_intent_lock(llu_i2mdexp(parent), &op_data, NULL, 0, it,
452 flags, &req, llu_md_blocking_ast,
453 LDLM_FL_CANCEL_ON_BLOCK);
457 rc = lookup_it_finish(req, DLM_REPLY_REC_OFF, it, &icbd);
459 ll_intent_release(it);
463 llu_lookup_finish_locks(it, pnode);
467 ptlrpc_req_finished(req);
471 static struct lookup_intent*
472 translate_lookup_intent(struct intent *intent, const char *path)
474 struct lookup_intent *it;
478 if (!intent || path) {
479 CDEBUG(D_VFSTRACE, "not intent needed\n");
483 OBD_ALLOC(it, sizeof(*it));
486 memset(it, 0, sizeof(*it));
488 /* libsysio will assign intent like following:
489 * NOTE: INT_CREAT has include INT_UPDPARENT
491 * open: INT_OPEN [| INT_CREAT]
494 * unlink: INT_UPDPARENT
495 * rmdir: INT_UPDPARENT
500 * following logic is adjusted for libsysio
503 it->it_flags = intent->int_arg2 ? *((int*)intent->int_arg2) : 0;
505 if (intent->int_opmask & INT_OPEN) {
506 it->it_op |= IT_OPEN;
508 /* convert access mode from O_ to FMODE_ */
509 if (it->it_flags & O_WRONLY)
511 else if (it->it_flags & O_RDWR)
512 fmode = FMODE_READ | FMODE_WRITE;
515 it->it_flags &= ~O_ACCMODE;
516 it->it_flags |= fmode;
519 /* XXX libsysio has strange code on intent handling,
520 * more check later */
521 if (it->it_flags & O_CREAT) {
522 it->it_op |= IT_CREAT;
523 it->it_create_mode = *((int*)intent->int_arg1);
524 /* bug 7278: libsysio hack. For O_EXCL, libsysio depends on
525 this lookup to return negative result, but then there is no
526 way to find out original intent in ll_iop_open(). So we just
527 clear O_EXCL from libsysio flags here to avoid checking
528 for negative result. O_EXCL will be enforced by MDS. */
529 *((int*)intent->int_arg2) &= ~O_EXCL;
532 if (intent->int_opmask & INT_GETATTR)
533 it->it_op |= IT_GETATTR;
535 LASSERT(!(intent->int_opmask & INT_SETATTR));
537 /* libsysio is different to linux vfs when doing unlink/rmdir,
538 * INT_UPDPARENT was passed down during name resolution. Here
539 * we treat it as normal lookup, later unlink()/rmdir() will
540 * do the actual work */
542 /* conform to kernel code, if only IT_LOOKUP was set, don't
544 if (!it->it_op || it->it_op & IT_LOOKUP) {
545 OBD_FREE(it, sizeof(*it));
549 it->it_op_release = ll_intent_release;
551 CDEBUG(D_VFSTRACE, "final intent 0x%x\n", it ? it->it_op : 0);
555 int llu_iop_lookup(struct pnode *pnode,
557 struct intent *intnt,
560 struct lookup_intent *it;
564 liblustre_wait_event(0);
568 /* the mount root inode have no name, so don't call
569 * remote in this case. but probably we need revalidate
571 if (pnode->p_mount->mnt_root == pnode) {
572 struct inode *i = pnode->p_base->pb_ino;
577 if (!pnode->p_base->pb_name.len)
580 it = translate_lookup_intent(intnt, path);
582 /* param flags is not used, let it be 0 */
583 if (llu_pb_revalidate(pnode, 0, it)) {
584 LASSERT(pnode->p_base->pb_ino);
585 *inop = pnode->p_base->pb_ino;
589 rc = llu_lookup_it(pnode->p_parent->p_base->pb_ino, pnode, it, 0);
591 if (!pnode->p_base->pb_ino)
594 *inop = pnode->p_base->pb_ino;
599 OBD_FREE(it, sizeof(*it));
600 liblustre_wait_event(0);