Whamcloud - gitweb
706cc7488880b6ce3bd1c77ace90dda7dd52c951
[fs/lustre-release.git] / lustre / liblustre / dir.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2012, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/liblustre/dir.c
37  *
38  * Lustre Light directory handling
39  */
40
41 #define DEBUG_SUBSYSTEM S_LLITE
42
43 #include <errno.h>
44 #include <stddef.h>
45 #include <string.h>
46 #include <sys/stat.h>
47 #include <sys/types.h>
48 #include <libcfs/libcfs.h>
49 #include <lustre/lustre_idl.h>
50 #include <liblustre.h>
51 #include <lclient.h>
52 #include <lustre_dlm.h>
53 #include <lustre_lite.h>
54 #include <lustre_net.h>
55 #include <lustre_req_layout.h>
56 #include <obd.h>
57 #include <obd_class.h>
58 #include <obd_support.h>
59 #include "llite_lib.h"
60 #include <dirent.h>
61
62 /* (new) readdir implementation overview can be found in lustre/llite/dir.c */
63 static int llu_dir_do_readpage(struct inode *inode, struct page *page)
64 {
65         struct llu_inode_info *lli = llu_i2info(inode);
66         struct intnl_stat     *st = llu_i2stat(inode);
67         struct llu_sb_info    *sbi = llu_i2sbi(inode);
68         struct ptlrpc_request *request;
69         struct lustre_handle   lockh;
70         struct mdt_body       *body;
71         struct lookup_intent   it = { .it_op = IT_READDIR };
72         struct md_op_data      op_data = {{ 0 }};
73         ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_UPDATE } };
74         int rc = 0;
75         ENTRY;
76
77         llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, 0, LUSTRE_OPC_ANY);
78         rc = md_lock_match(sbi->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
79                            &lli->lli_fid, LDLM_IBITS, &policy, LCK_CR, &lockh);
80         if (!rc) {
81                 struct ldlm_enqueue_info einfo = {
82                         .ei_type        = LDLM_IBITS,
83                         .ei_mode        = LCK_CR,
84                         .ei_cb_bl       = llu_md_blocking_ast,
85                         .ei_cb_cp       = ldlm_completion_ast,
86                         .ei_cbdata      = inode,
87                 };
88
89                 rc = md_enqueue(sbi->ll_md_exp, &einfo, &it,
90                                 &op_data, &lockh, NULL, 0, NULL,
91                                 LDLM_FL_CANCEL_ON_BLOCK);
92                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
93                 if (request)
94                         ptlrpc_req_finished(request);
95                 if (rc < 0) {
96                         CERROR("lock enqueue: err: %d\n", rc);
97                         RETURN(rc);
98                 }
99         }
100         ldlm_lock_dump_handle(D_OTHER, &lockh);
101
102         op_data.op_hash_offset = hash_x_index(page->index, 0);
103         op_data.op_npages = 1;
104         rc = md_readpage(sbi->ll_md_exp, &op_data, &page, &request);
105         if (!rc) {
106                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
107                 LASSERT(body != NULL);         /* checked by md_readpage() */
108
109                 if (body->valid & OBD_MD_FLSIZE)
110                         st->st_size = body->size;
111         } else {
112                 CERROR("read_dir_page(%ld) error %d\n", page->index, rc);
113         }
114         ptlrpc_req_finished(request);
115         EXIT;
116
117         ldlm_lock_decref(&lockh, LCK_CR);
118         return rc;
119 }
120
121 static struct page *llu_dir_read_page(struct inode *ino, __u64 hash,
122                                      int exact, struct ll_dir_chain *chain)
123 {
124         struct page *page;
125         int rc;
126         ENTRY;
127
128         OBD_PAGE_ALLOC(page, 0);
129         if (!page)
130                 RETURN(ERR_PTR(-ENOMEM));
131         page->index = hash_x_index(hash, 0);
132
133         rc = llu_dir_do_readpage(ino, page);
134         if (rc) {
135                 OBD_PAGE_FREE(page);
136                 RETURN(ERR_PTR(rc));
137         }
138
139         return page;
140 }
141
142 static void *(*memmover)(void *, const void *, size_t) = memmove;
143
144 #define NAME_OFFSET(de) ((int) ((de)->d_name - (char *) (de)))
145 #define ROUND_UP64(x)   (((x)+sizeof(__u64)-1) & ~(sizeof(__u64)-1))
146 static int filldir(char *buf, int buflen, const char *name, int namelen,
147                    loff_t offset, ino_t ino, unsigned int d_type, int *filled)
148 {
149         struct intnl_dirent *dirent = (struct intnl_dirent *)(buf + *filled);
150         struct intnl_dirent holder;
151         int reclen = ROUND_UP64(NAME_OFFSET(dirent) + namelen + 1);
152
153         /*
154          * @buf is not guaranteed to be properly aligned. To work around,
155          * first fill stack-allocated @holder, then copy @holder into @buf by
156          * memmove().
157          */
158
159         /* check overflow */
160         if ((*filled + reclen) > buflen)
161                 return 1;
162
163         holder.d_ino = ino;
164 #ifdef _DIRENT_HAVE_D_OFF
165         holder.d_off = offset;
166 #endif
167         holder.d_reclen = reclen;
168 #ifdef _DIRENT_HAVE_D_TYPE
169         holder.d_type = (unsigned short) d_type;
170 #endif
171         /* gcc unrolls memcpy() of structs into field-wise assignments,
172          * assuming proper alignment. Humor it. */
173         (*memmover)(dirent, &holder, NAME_OFFSET(dirent));
174         memcpy(dirent->d_name, name, namelen);
175         dirent->d_name[namelen] = 0;
176
177         *filled += reclen;
178
179         return 0;
180 }
181
182 /*
183  * TODO: much of the code here is similar/identical to llite ll_readdir().
184  * These code can be factored out and shared in a common module.
185  */
186
187 ssize_t llu_iop_filldirentries(struct inode *dir, _SYSIO_OFF_T *basep,
188                                char *buf, size_t nbytes)
189 {
190         struct llu_inode_info *lli = llu_i2info(dir);
191         struct intnl_stat     *st = llu_i2stat(dir);
192         loff_t                 pos = *basep;
193         struct ll_dir_chain    chain;
194         struct page            *page;
195         int filled = 0;
196         int rc;
197         int done;
198         __u16 type;
199         ENTRY;
200
201         liblustre_wait_event(0);
202
203         if (st->st_size == 0) {
204                 CWARN("dir size is 0?\n");
205                 RETURN(0);
206         }
207
208         if (pos == MDS_DIR_END_OFF)
209                 /*
210                  * end-of-file.
211                  */
212                 RETURN(0);
213
214         rc    = 0;
215         done  = 0;
216         ll_dir_chain_init(&chain);
217
218         page = llu_dir_read_page(dir, pos, 0, &chain);
219         while (rc == 0 && !done) {
220                 struct lu_dirpage *dp;
221                 struct lu_dirent  *ent;
222
223                 if (!IS_ERR(page)) {
224                         /*
225                          * If page is empty (end of directoryis reached),
226                          * use this value.
227                          */
228                         __u64 hash = MDS_DIR_END_OFF;
229                         __u64 next;
230
231                         dp = page->addr;
232                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
233                              ent = lu_dirent_next(ent)) {
234                                 char          *name;
235                                 int            namelen;
236                                 struct lu_fid  fid;
237                                 __u64          ino;
238
239                                 hash    = le64_to_cpu(ent->lde_hash);
240                                 namelen = le16_to_cpu(ent->lde_namelen);
241
242                                 if (hash < pos)
243                                         /*
244                                          * Skip until we find target hash
245                                          * value.
246                                          */
247                                         continue;
248
249                                 if (namelen == 0)
250                                         /*
251                                          * Skip dummy record.
252                                          */
253                                         continue;
254
255                                 fid  = ent->lde_fid;
256                                 name = ent->lde_name;
257                                 fid_le_to_cpu(&fid, &fid);
258                                 ino  = cl_fid_build_ino(&fid, 0);
259                                 type = ll_dirent_type_get(ent);
260                                 done = filldir(buf, nbytes, name, namelen,
261                                                (loff_t)hash, ino, type,
262                                                &filled);
263                         }
264                         next = le64_to_cpu(dp->ldp_hash_end);
265                         OBD_PAGE_FREE(page);
266                         if (!done) {
267                                 pos = next;
268                                 if (pos == MDS_DIR_END_OFF)
269                                         /*
270                                          * End of directory reached.
271                                          */
272                                         done = 1;
273                                 else if (1 /* chain is exhausted*/)
274                                         /*
275                                          * Normal case: continue to the next
276                                          * page.
277                                          */
278                                         page = llu_dir_read_page(dir, pos, 1,
279                                                                &chain);
280                                 else {
281                                         /*
282                                          * go into overflow page.
283                                          */
284                                 }
285                         } else {
286                                 pos = hash;
287                                 if (filled == 0)
288                                         GOTO(out, filled = -EINVAL);
289                         }
290                 } else {
291                         rc = PTR_ERR(page);
292                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
293                                PFID(&lli->lli_fid), (unsigned long)pos, rc);
294                 }
295         }
296         lli->lli_dir_pos = (loff_t)pos;
297         *basep = lli->lli_dir_pos;
298 out:
299         ll_dir_chain_fini(&chain);
300         liblustre_wait_event(0);
301         RETURN(filled);
302 }