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 (c) 2003, 2010, Oracle and/or its affiliates. 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>
52 #include "llite_lib.h"
54 void ll_intent_drop_lock(struct lookup_intent *it)
56 struct lustre_handle *handle;
58 if (it->it_op && it->d.lustre.it_lock_mode) {
59 handle = (struct lustre_handle *)&it->d.lustre.it_lock_handle;
60 CDEBUG(D_DLMTRACE, "releasing lock with cookie "LPX64
61 " from it %p\n", handle->cookie, it);
62 ldlm_lock_decref(handle, it->d.lustre.it_lock_mode);
64 /* bug 494: intent_release may be called multiple times, from
65 * this thread and we don't want to double-decref this lock */
66 it->d.lustre.it_lock_mode = 0;
70 void ll_intent_release(struct lookup_intent *it)
74 ll_intent_drop_lock(it);
76 it->it_op_release = 0;
77 it->d.lustre.it_disposition = 0;
78 it->d.lustre.it_data = NULL;
82 void llu_lookup_finish_locks(struct lookup_intent *it, struct pnode *pnode)
88 inode = pnode->p_base->pb_ino;
89 if (it->d.lustre.it_lock_mode && inode != NULL) {
90 struct llu_sb_info *sbi;
92 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%llu/%lu)\n",
93 inode, (long long)llu_i2stat(inode)->st_ino,
94 llu_i2info(inode)->lli_st_generation);
96 sbi = llu_i2sbi(inode);
97 md_set_lock_data(sbi->ll_md_exp,
98 &it->d.lustre.it_lock_handle, inode, NULL);
101 /* drop lookup/getattr locks */
102 if (it->it_op & (IT_LOOKUP | IT_GETATTR))
103 ll_intent_release(it);
107 static inline void llu_invalidate_inode_pages(struct inode * inode)
112 int llu_md_blocking_ast(struct ldlm_lock *lock,
113 struct ldlm_lock_desc *desc,
114 void *data, int flag)
116 struct lustre_handle lockh;
122 case LDLM_CB_BLOCKING:
123 ldlm_lock2handle(lock, &lockh);
124 rc = ldlm_cli_cancel(&lockh);
126 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
130 case LDLM_CB_CANCELING: {
131 struct inode *inode = llu_inode_from_lock(lock);
132 struct llu_inode_info *lli;
133 struct intnl_stat *st;
134 __u64 bits = lock->l_policy_data.l_inodebits.bits;
137 /* Invalidate all dentries associated with this inode */
141 lli = llu_i2info(inode);
142 st = llu_i2stat(inode);
144 if (bits & MDS_INODELOCK_UPDATE)
145 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
148 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
149 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
150 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
151 LDLM_ERROR(lock,"data mismatch with ino %llu/%llu/%llu",
152 (long long)fid_seq(fid),
153 (long long)fid_oid(fid),
154 (long long)fid_ver(fid));
156 if (S_ISDIR(st->st_mode) &&
157 (bits & MDS_INODELOCK_UPDATE)) {
158 CDEBUG(D_INODE, "invalidating inode %llu\n",
159 (long long)st->st_ino);
161 llu_invalidate_inode_pages(inode);
165 if (inode->i_sb->s_root &&
166 inode != inode->i_sb->s_root->d_inode)
167 ll_unhash_aliases(inode);
179 static int pnode_revalidate_finish(struct ptlrpc_request *req,
180 struct lookup_intent *it,
183 struct inode *inode = pnode->p_base->pb_ino;
193 if (it_disposition(it, DISP_LOOKUP_NEG))
196 rc = md_get_lustre_md(llu_i2sbi(inode)->ll_md_exp, req,
197 llu_i2sbi(inode)->ll_dt_exp,
198 llu_i2sbi(inode)->ll_md_exp, &md);
202 llu_update_inode(inode, &md);
207 static int llu_pb_revalidate(struct pnode *pnode, int flags,
208 struct lookup_intent *it)
210 struct pnode_base *pb = pnode->p_base;
211 struct it_cb_data icbd;
212 struct md_op_data op_data = {{ 0 }};
213 struct ptlrpc_request *req = NULL;
214 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
215 struct obd_export *exp;
219 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,intent=%x\n",
220 (int)pb->pb_name.len, pb->pb_name.name, it ? it->it_op : 0);
222 /* We don't want to cache negative dentries, so return 0 immediately.
223 * We believe that this is safe, that negative dentries cannot be
224 * pinned by someone else */
225 if (pb->pb_ino == NULL) {
226 CDEBUG(D_INODE, "negative pb\n");
230 /* This is due to bad interaction with libsysio. remove this when we
231 * switched to libbsdio XXX
234 struct llu_inode_info *lli = llu_i2info(pb->pb_ino);
235 struct intnl_stat *st = llu_i2stat(pb->pb_ino);
237 CDEBUG(D_INODE, "inode %llu still have intent "
238 "%p(opc 0x%x), release it\n",
239 (long long) st->st_ino, lli->lli_it,
241 ll_intent_release(lli->lli_it);
242 OBD_FREE(lli->lli_it, sizeof(*lli->lli_it));
247 exp = llu_i2mdexp(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 llu_prep_md_op_data(&op_data, pnode->p_parent->p_base->pb_ino,
257 pb->pb_ino, pb->pb_name.name, pb->pb_name.len,
260 rc = md_intent_lock(exp, &op_data, NULL, 0, it, flags,
261 &req, llu_md_blocking_ast,
262 LDLM_FL_CANCEL_ON_BLOCK);
263 /* If req is NULL, then md_intent_lock only tried to do a lock match;
264 * if all was well, it will return 1 if it found locks, 0 otherwise. */
265 if (req == NULL && rc >= 0)
271 rc = pnode_revalidate_finish(req, it, pnode);
273 ll_intent_release(it);
278 /* Note: ll_intent_lock may cause a callback, check this! */
280 if (it->it_op & IT_OPEN)
281 LL_SAVE_INTENT(pb->pb_ino, it);
285 ptlrpc_req_finished(req);
291 llu_lookup_finish_locks(it, pnode);
296 static int lookup_it_finish(struct ptlrpc_request *request, int offset,
297 struct lookup_intent *it, void *data)
299 struct it_cb_data *icbd = data;
300 struct pnode *child = icbd->icbd_child;
301 struct inode *parent = icbd->icbd_parent;
302 struct llu_sb_info *sbi = llu_i2sbi(parent);
303 struct inode *inode = NULL;
306 /* libsysio require us generate inode right away if success.
307 * so if mds created new inode for us we need make sure it
308 * succeeded. thus for any error we can't delay to the
309 * llu_file_open() time. */
310 if (it_disposition(it, DISP_OPEN_CREATE) &&
311 it_open_error(DISP_OPEN_CREATE, it)) {
312 CDEBUG(D_INODE, "detect mds create error\n");
313 return it_open_error(DISP_OPEN_CREATE, it);
315 if (it_disposition(it, DISP_OPEN_OPEN) &&
316 it_open_error(DISP_OPEN_OPEN, it)) {
317 CDEBUG(D_INODE, "detect mds open error\n");
318 /* undo which did by md_intent_lock */
319 if (it_disposition(it, DISP_OPEN_CREATE) &&
320 !it_open_error(DISP_OPEN_CREATE, it)) {
322 LASSERT(cfs_atomic_read(&request->rq_refcount) > 1);
323 CDEBUG(D_INODE, "dec a ref of req %p\n", request);
324 ptlrpc_req_finished(request);
326 return it_open_error(DISP_OPEN_OPEN, it);
329 /* NB 1 request reference will be taken away by ll_intent_lock()
332 if (!it_disposition(it, DISP_LOOKUP_NEG) || (it->it_op & IT_CREAT)) {
334 struct llu_inode_info *lli;
335 struct intnl_stat *st;
338 if (it_disposition(it, DISP_OPEN_CREATE))
339 ptlrpc_req_finished(request);
341 rc = md_get_lustre_md(sbi->ll_md_exp, request,
342 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
346 inode = llu_iget(parent->i_fs, &md);
347 if (!inode || IS_ERR(inode)) {
348 /* free the lsm if we allocated one above */
350 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
351 RETURN(inode ? PTR_ERR(inode) : -ENOMEM);
352 } else if (md.lsm != NULL &&
353 llu_i2info(inode)->lli_smd != md.lsm) {
354 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
357 lli = llu_i2info(inode);
358 st = llu_i2stat(inode);
360 /* If this is a stat, get the authoritative file size */
361 if (it->it_op == IT_GETATTR && S_ISREG(st->st_mode) &&
362 lli->lli_smd != NULL) {
363 struct lov_stripe_md *lsm = lli->lli_smd;
366 LASSERT(lsm->lsm_object_id != 0);
368 /* bug 2334: drop MDS lock before acquiring OST lock */
369 ll_intent_drop_lock(it);
371 rc = cl_glimpse_size(inode);
381 /* intent will be further used in cases of open()/getattr() */
382 if (inode && (it->it_op & IT_OPEN))
383 LL_SAVE_INTENT(inode, it);
385 child->p_base->pb_ino = inode;
390 struct inode *llu_inode_from_lock(struct ldlm_lock *lock)
393 lock_res_and_lock(lock);
395 if (lock->l_ast_data) {
396 inode = (struct inode *)lock->l_ast_data;
401 unlock_res_and_lock(lock);
405 static int llu_lookup_it(struct inode *parent, struct pnode *pnode,
406 struct lookup_intent *it, int flags)
408 struct md_op_data op_data = {{ 0 }};
409 struct it_cb_data icbd;
410 struct ptlrpc_request *req = NULL;
411 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
416 if (pnode->p_base->pb_name.len > EXT2_NAME_LEN)
417 RETURN(-ENAMETOOLONG);
421 it->it_op_release = ll_intent_release;
424 icbd.icbd_child = pnode;
425 icbd.icbd_parent = parent;
427 if (it->it_op & IT_CREAT ||
428 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT)) {
429 opc = LUSTRE_OPC_CREATE;
431 opc = LUSTRE_OPC_ANY;
434 llu_prep_md_op_data(&op_data, parent, NULL,
435 pnode->p_base->pb_name.name,
436 pnode->p_base->pb_name.len, flags, opc);
438 rc = md_intent_lock(llu_i2mdexp(parent), &op_data, NULL, 0, it,
439 flags, &req, llu_md_blocking_ast,
440 LDLM_FL_CANCEL_ON_BLOCK);
444 rc = lookup_it_finish(req, DLM_REPLY_REC_OFF, it, &icbd);
446 ll_intent_release(it);
450 llu_lookup_finish_locks(it, pnode);
454 ptlrpc_req_finished(req);
458 static struct lookup_intent*
459 translate_lookup_intent(struct intent *intent, const char *path)
461 struct lookup_intent *it;
465 if (!intent || path) {
466 CDEBUG(D_VFSTRACE, "not intent needed\n");
470 OBD_ALLOC(it, sizeof(*it));
473 memset(it, 0, sizeof(*it));
475 /* libsysio will assign intent like following:
476 * NOTE: INT_CREAT has include INT_UPDPARENT
478 * open: INT_OPEN [| INT_CREAT]
481 * unlink: INT_UPDPARENT
482 * rmdir: INT_UPDPARENT
487 * following logic is adjusted for libsysio
490 it->it_flags = intent->int_arg2 ? *((int*)intent->int_arg2) : 0;
492 if (intent->int_opmask & INT_OPEN) {
493 it->it_op |= IT_OPEN;
495 /* convert access mode from O_ to FMODE_ */
496 if (it->it_flags & O_WRONLY)
498 else if (it->it_flags & O_RDWR)
499 fmode = FMODE_READ | FMODE_WRITE;
502 it->it_flags &= ~O_ACCMODE;
503 it->it_flags |= fmode;
506 /* XXX libsysio has strange code on intent handling,
507 * more check later */
508 if (it->it_flags & O_CREAT) {
509 it->it_op |= IT_CREAT;
510 it->it_create_mode = *((int*)intent->int_arg1);
511 /* bug 7278: libsysio hack. For O_EXCL, libsysio depends on
512 this lookup to return negative result, but then there is no
513 way to find out original intent in ll_iop_open(). So we just
514 clear O_EXCL from libsysio flags here to avoid checking
515 for negative result. O_EXCL will be enforced by MDS. */
516 *((int*)intent->int_arg2) &= ~O_EXCL;
519 if (intent->int_opmask & INT_GETATTR)
520 it->it_op |= IT_GETATTR;
522 LASSERT(!(intent->int_opmask & INT_SETATTR));
524 /* libsysio is different to linux vfs when doing unlink/rmdir,
525 * INT_UPDPARENT was passed down during name resolution. Here
526 * we treat it as normal lookup, later unlink()/rmdir() will
527 * do the actual work */
529 /* conform to kernel code, if only IT_LOOKUP was set, don't
531 if (!it->it_op || it->it_op & IT_LOOKUP) {
532 OBD_FREE(it, sizeof(*it));
536 it->it_op_release = ll_intent_release;
538 CDEBUG(D_VFSTRACE, "final intent 0x%x\n", it ? it->it_op : 0);
542 int llu_iop_lookup(struct pnode *pnode,
544 struct intent *intnt,
547 struct lookup_intent *it;
551 liblustre_wait_event(0);
555 /* the mount root inode have no name, so don't call
556 * remote in this case. but probably we need revalidate
558 if (pnode->p_mount->mnt_root == pnode) {
559 struct inode *i = pnode->p_base->pb_ino;
564 if (!pnode->p_base->pb_name.len)
567 it = translate_lookup_intent(intnt, path);
569 /* param flags is not used, let it be 0 */
570 if (llu_pb_revalidate(pnode, 0, it)) {
571 LASSERT(pnode->p_base->pb_ino);
572 *inop = pnode->p_base->pb_ino;
576 rc = llu_lookup_it(pnode->p_parent->p_base->pb_ino, pnode, it, 0);
578 if (!pnode->p_base->pb_ino)
581 *inop = pnode->p_base->pb_ino;
586 OBD_FREE(it, sizeof(*it));
587 liblustre_wait_event(0);