1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2001-2003 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 #include <linux/sched.h>
24 #include <linux/smp_lock.h>
25 #include <linux/quotaops.h>
27 #define DEBUG_SUBSYSTEM S_LLITE
29 #include <linux/obd_support.h>
30 #include <linux/lustre_lite.h>
31 #include <linux/lustre_idl.h>
32 #include <linux/lustre_dlm.h>
33 #include <linux/lustre_version.h>
35 #include "llite_internal.h"
37 /* should NOT be called with the dcache lock, see fs/dcache.c */
38 static void ll_release(struct dentry *de)
40 struct ll_dentry_data *lld;
44 if (lld == NULL) { /* NFS copies the de->d_op methods (bug 4655) */
48 LASSERT(lld->lld_cwd_count == 0);
49 LASSERT(lld->lld_mnt_count == 0);
50 OBD_FREE(de->d_fsdata, sizeof(*lld));
55 /* Compare if two dentries are the same. Don't match if the existing dentry
56 * is marked DCACHE_LUSTRE_INVALID. Returns 1 if different, 0 if the same.
58 * This avoids a race where ll_lookup_it() instantiates a dentry, but we get
59 * an AST before calling d_revalidate_it(). The dentry still exists (marked
60 * INVALID) so d_lookup() matches it, but we have no lock on it (so
61 * lock_match() fails) and we spin around real_lookup(). */
62 static int ll_dcompare(struct dentry *parent, struct qstr *d_name,
65 struct dentry *dchild;
68 if (d_name->len != name->len)
71 if (memcmp(d_name->name, name->name, name->len))
74 dchild = container_of(d_name, struct dentry, d_name); /* ugh */
75 if (dchild->d_flags & DCACHE_LUSTRE_INVALID) {
76 CDEBUG(D_DENTRY,"INVALID dentry %p not matched, was bug 3784\n",
84 /* should NOT be called with the dcache lock, see fs/dcache.c */
85 static int ll_ddelete(struct dentry *de)
89 CDEBUG(D_DENTRY, "%s dentry %.*s (%p, parent %p, inode %p) %s%s\n",
90 (de->d_flags & DCACHE_LUSTRE_INVALID ? "deleting" : "keeping"),
91 de->d_name.len, de->d_name.name, de, de->d_parent, de->d_inode,
92 d_unhashed(de) ? "" : "hashed,",
93 list_empty(&de->d_subdirs) ? "" : "subdirs");
97 void ll_set_dd(struct dentry *de)
102 CDEBUG(D_DENTRY, "ldd on dentry %.*s (%p) parent %p inode %p refc %d\n",
103 de->d_name.len, de->d_name.name, de, de->d_parent, de->d_inode,
104 atomic_read(&de->d_count));
106 if (de->d_fsdata == NULL) {
107 OBD_ALLOC(de->d_fsdata, sizeof(struct ll_dentry_data));
114 void ll_intent_drop_lock(struct lookup_intent *it)
116 struct lustre_handle *handle;
118 if (it->it_op && it->d.lustre.it_lock_mode) {
119 handle = (struct lustre_handle *)&it->d.lustre.it_lock_handle;
120 CDEBUG(D_DLMTRACE, "releasing lock with cookie "LPX64
121 " from it %p\n", handle->cookie, it);
122 ldlm_lock_decref(handle, it->d.lustre.it_lock_mode);
124 /* bug 494: intent_release may be called multiple times, from
125 * this thread and we don't want to double-decref this lock */
126 it->d.lustre.it_lock_mode = 0;
130 void ll_intent_release(struct lookup_intent *it)
134 ll_intent_drop_lock(it);
136 it->it_op_release = 0;
137 it->d.lustre.it_disposition = 0;
138 it->d.lustre.it_data = NULL;
142 void ll_unhash_aliases(struct inode *inode)
144 struct list_head *tmp, *head;
145 struct ll_sb_info *sbi;
149 CERROR("unexpected NULL inode, tell phil\n");
153 CDEBUG(D_INODE, "marking dentries for ino %lu/%u(%p) invalid\n",
154 inode->i_ino, inode->i_generation, inode);
156 sbi = ll_i2sbi(inode);
157 head = &inode->i_dentry;
159 spin_lock(&dcache_lock);
161 while ((tmp = tmp->next) != head) {
162 struct dentry *dentry = list_entry(tmp, struct dentry, d_alias);
164 if (dentry->d_name.len == 1 && dentry->d_name.name[0] == '/') {
165 CERROR("called on root (?) dentry=%p, inode=%p "
166 "ino=%lu\n", dentry, inode, inode->i_ino);
167 lustre_dump_dentry(dentry, 1);
168 libcfs_debug_dumpstack(NULL);
169 } else if (d_mountpoint(dentry)) {
170 /* For mountpoints we skip removal of the dentry
171 which happens solely because we have a lock on it
172 obtained when this dentry was not a mountpoint yet */
173 CDEBUG(D_DENTRY, "Skippind mountpoint dentry removal "
174 "%.*s (%p) parent %p\n",
177 dentry, dentry->d_parent);
182 if (atomic_read(&dentry->d_count) == 0) {
183 CDEBUG(D_DENTRY, "deleting dentry %.*s (%p) parent %p "
184 "inode %p\n", dentry->d_name.len,
185 dentry->d_name.name, dentry, dentry->d_parent,
189 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
190 INIT_HLIST_NODE(&dentry->d_hash);
192 spin_unlock(&dcache_lock);
195 } else if (!(dentry->d_flags & DCACHE_LUSTRE_INVALID)) {
196 CDEBUG(D_DENTRY, "unhashing dentry %.*s (%p) parent %p "
197 "inode %p refc %d\n", dentry->d_name.len,
198 dentry->d_name.name, dentry, dentry->d_parent,
199 dentry->d_inode, atomic_read(&dentry->d_count));
200 hlist_del_init(&dentry->d_hash);
201 dentry->d_flags |= DCACHE_LUSTRE_INVALID;
202 hlist_add_head(&dentry->d_hash,
203 &sbi->ll_orphan_dentry_list);
206 spin_unlock(&dcache_lock);
210 static int revalidate_it_finish(struct ptlrpc_request *request, int offset,
211 struct lookup_intent *it,
214 struct ll_sb_info *sbi;
221 if (it_disposition(it, DISP_LOOKUP_NEG))
224 sbi = ll_i2sbi(de->d_inode);
225 rc = ll_prep_inode(sbi->ll_osc_exp, &de->d_inode, request, offset,NULL);
230 void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry)
233 LASSERT(dentry != NULL);
235 if (it->d.lustre.it_lock_mode && dentry->d_inode != NULL) {
236 struct inode *inode = dentry->d_inode;
237 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
238 inode, inode->i_ino, inode->i_generation);
239 mdc_set_lock_data(&it->d.lustre.it_lock_handle, inode);
242 /* drop lookup or getattr locks immediately */
243 if (it->it_op == IT_LOOKUP || it->it_op == IT_GETATTR) {
244 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
245 /* on 2.6 there are situation when several lookups and
246 * revalidations may be requested during single operation.
247 * therefore, we don't release intent here -bzzz */
248 ll_intent_drop_lock(it);
250 ll_intent_release(it);
255 void ll_frob_intent(struct lookup_intent **itp, struct lookup_intent *deft)
257 struct lookup_intent *it = *itp;
258 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
260 LASSERTF(it->it_magic == INTENT_MAGIC, "bad intent magic: %x\n",
265 if (!it || it->it_op == IT_GETXATTR)
268 it->it_op_release = ll_intent_release;
271 int ll_revalidate_it(struct dentry *de, int lookup_flags,
272 struct lookup_intent *it)
275 struct it_cb_data icbd;
276 struct mdc_op_data op_data;
277 struct ptlrpc_request *req = NULL;
278 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
279 struct obd_export *exp;
282 CDEBUG(D_VFSTRACE, "VFS Op:name=%s,intent=%s\n", de->d_name.name,
285 /* Cached negative dentries are unsafe for now - look them up again */
286 if (de->d_inode == NULL)
289 exp = ll_i2mdcexp(de->d_inode);
290 icbd.icbd_parent = de->d_parent->d_inode;
291 icbd.icbd_childp = &de;
293 /* Never execute intents for mount points.
294 * Attributes will be fixed up in ll_inode_revalidate_it */
295 if (d_mountpoint(de))
298 OBD_FAIL_TIMEOUT(OBD_FAIL_MDC_REVALIDATE_PAUSE, 5);
299 ll_frob_intent(&it, &lookup_it);
302 ll_prepare_mdc_op_data(&op_data, de->d_parent->d_inode, de->d_inode,
303 de->d_name.name, de->d_name.len, 0);
305 rc = mdc_intent_lock(exp, &op_data, NULL, 0, it, lookup_flags,
306 &req, ll_mdc_blocking_ast);
307 /* If req is NULL, then mdc_intent_lock only tried to do a lock match;
308 * if all was well, it will return 1 if it found locks, 0 otherwise. */
309 if (req == NULL && rc >= 0)
314 CDEBUG(D_INFO, "ll_intent_lock: rc %d : it->it_status "
315 "%d\n", rc, it->d.lustre.it_status);
320 rc = revalidate_it_finish(req, 1, it, de);
322 ll_intent_release(it);
327 /* unfortunately ll_intent_lock may cause a callback and revoke our
329 spin_lock(&dcache_lock);
330 hlist_del_init(&de->d_hash);
332 spin_unlock(&dcache_lock);
335 /* If we had succesful it lookup on mds, but it happened to be negative,
336 we do not free request as it will be reused during lookup (see
337 comment in mdc/mdc_locks.c::mdc_intent_lock(). But if
338 request was not completed, we need to free it. (bug 5154) */
339 if (req != NULL && (rc == 1 || !it_disposition(it, DISP_ENQ_COMPLETE)))
340 ptlrpc_req_finished(req);
342 ll_unhash_aliases(de->d_inode);
343 /* done in ll_unhash_aliases()
344 dentry->d_flags |= DCACHE_LUSTRE_INVALID; */
346 CDEBUG(D_DENTRY, "revalidated dentry %.*s (%p) parent %p "
347 "inode %p refc %d\n", de->d_name.len,
348 de->d_name.name, de, de->d_parent, de->d_inode,
349 atomic_read(&de->d_count));
350 ll_lookup_finish_locks(it, de);
351 de->d_flags &= ~DCACHE_LUSTRE_INVALID;
356 /*static*/ void ll_pin(struct dentry *de, struct vfsmount *mnt, int flag)
358 struct inode *inode= de->d_inode;
359 struct ll_sb_info *sbi = ll_i2sbi(inode);
360 struct ll_dentry_data *ldd = ll_d2d(de);
361 struct obd_client_handle *handle;
367 /* Strictly speaking this introduces an additional race: the
368 * increments should wait until the rpc has returned.
369 * However, given that at present the function is void, this
371 if (flag == 1 && (++ldd->lld_mnt_count) > 1) {
377 if (flag == 0 && (++ldd->lld_cwd_count) > 1) {
384 handle = (flag) ? &ldd->lld_mnt_och : &ldd->lld_cwd_och;
385 rc = obd_pin(sbi->ll_mdc_exp, inode->i_ino, inode->i_generation,
386 inode->i_mode & S_IFMT, handle, flag);
390 memset(handle, 0, sizeof(*handle));
392 ldd->lld_cwd_count--;
394 ldd->lld_mnt_count--;
402 /*static*/ void ll_unpin(struct dentry *de, struct vfsmount *mnt, int flag)
404 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
405 struct ll_dentry_data *ldd = ll_d2d(de);
406 struct obd_client_handle handle;
412 /* Strictly speaking this introduces an additional race: the
413 * increments should wait until the rpc has returned.
414 * However, given that at present the function is void, this
416 handle = (flag) ? ldd->lld_mnt_och : ldd->lld_cwd_och;
417 if (handle.och_magic != OBD_CLIENT_HANDLE_MAGIC) {
418 /* the "pin" failed */
425 count = --ldd->lld_mnt_count;
427 count = --ldd->lld_cwd_count;
435 rc = obd_unpin(sbi->ll_mdc_exp, &handle, flag);
440 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
441 static int ll_revalidate_nd(struct dentry *dentry, struct nameidata *nd)
446 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
447 rc = ll_revalidate_it(dentry, nd->flags, &nd->intent);
449 rc = ll_revalidate_it(dentry, 0, NULL);
455 struct dentry_operations ll_d_ops = {
456 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
457 .d_revalidate = ll_revalidate_nd,
459 .d_revalidate_it = ll_revalidate_it,
461 .d_release = ll_release,
462 .d_delete = ll_ddelete,
463 .d_compare = ll_dcompare,