Whamcloud - gitweb
b=20101 lfs getstripe -d test for sanity 27w
[fs/lustre-release.git] / lustre / liblustre / dir.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  * GPL HEADER START
5  *
6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
7  *
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.
11  *
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).
17  *
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
21  *
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
24  * have any questions.
25  *
26  * GPL HEADER END
27  */
28 /*
29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
30  * Use is subject to license terms.
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 <unistd.h>
44 #include <stdlib.h>
45 #include <string.h>
46 #include <assert.h>
47 #include <time.h>
48 #include <sys/types.h>
49 #include <sys/stat.h>
50 #include <fcntl.h>
51 #include <sys/queue.h>
52
53 #include <sysio.h>
54 #ifdef HAVE_XTIO_H
55 #include <xtio.h>
56 #endif
57 #include <fs.h>
58 #include <mount.h>
59 #include <inode.h>
60 #ifdef HAVE_FILE_H
61 #include <file.h>
62 #endif
63
64 #ifdef HAVE_LINUX_UNISTD_H
65 #include <linux/unistd.h>
66 #elif defined(HAVE_UNISTD_H)
67 #include <unistd.h>
68 #endif
69
70 #include <dirent.h>
71
72 #include "llite_lib.h"
73
74 /* (new) readdir implementation overview can be found in lustre/llite/dir.c */
75
76 static int llu_dir_do_readpage(struct inode *inode, struct page *page)
77 {
78         struct llu_inode_info *lli = llu_i2info(inode);
79         struct intnl_stat     *st = llu_i2stat(inode);
80         struct llu_sb_info    *sbi = llu_i2sbi(inode);
81         struct ptlrpc_request *request;
82         struct lustre_handle   lockh;
83         struct mdt_body       *body;
84         struct lookup_intent   it = { .it_op = IT_READDIR };
85         struct md_op_data      op_data = {{ 0 }};
86         ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_UPDATE } };
87         __u64 offset;
88         int rc = 0;
89         ENTRY;
90
91         rc = md_lock_match(sbi->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
92                            &lli->lli_fid, LDLM_IBITS, &policy, LCK_CR, &lockh);
93         if (!rc) {
94                 struct ldlm_enqueue_info einfo = {LDLM_IBITS, LCK_CR,
95                         llu_md_blocking_ast, ldlm_completion_ast, NULL, NULL,
96                         inode};
97
98                 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, 0,
99                                     LUSTRE_OPC_ANY);
100
101                 rc = md_enqueue(sbi->ll_md_exp, &einfo, &it,
102                                 &op_data, &lockh, NULL, 0, NULL,
103                                 LDLM_FL_CANCEL_ON_BLOCK);
104                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
105                 if (request)
106                         ptlrpc_req_finished(request);
107                 if (rc < 0) {
108                         CERROR("lock enqueue: err: %d\n", rc);
109                         RETURN(rc);
110                 }
111         }
112         ldlm_lock_dump_handle(D_OTHER, &lockh);
113
114         offset = (__u64)hash_x_index(page->index);
115         rc = md_readpage(sbi->ll_md_exp, &lli->lli_fid, NULL,
116                          offset, page, &request);
117         if (!rc) {
118                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
119                 LASSERT(body != NULL);         /* checked by md_readpage() */
120
121                 if (body->valid & OBD_MD_FLSIZE)
122                         st->st_size = body->size;
123         } else {
124                 CERROR("read_dir_page(%ld) error %d\n", page->index, rc);
125         }
126         ptlrpc_req_finished(request);
127         EXIT;
128
129         ldlm_lock_decref(&lockh, LCK_CR);
130         return rc;
131 }
132
133 static cfs_page_t *llu_dir_read_page(struct inode *ino, __u64 hash,
134                                      int exact, struct ll_dir_chain *chain)
135 {
136         cfs_page_t *page;
137         int rc;
138         ENTRY;
139
140         OBD_PAGE_ALLOC(page, 0);
141         if (!page)
142                 RETURN(ERR_PTR(-ENOMEM));
143         page->index = hash_x_index(hash);
144
145         rc = llu_dir_do_readpage(ino, page);
146         if (rc) {
147                 OBD_PAGE_FREE(page);
148                 RETURN(ERR_PTR(rc));
149         }
150
151         return page;
152 }
153
154 void *(*memmover)(void *, const void *, size_t) = memmove;
155
156 #define NAME_OFFSET(de) ((int) ((de)->d_name - (char *) (de)))
157 #define ROUND_UP64(x)   (((x)+sizeof(__u64)-1) & ~(sizeof(__u64)-1))
158 static int filldir(char *buf, int buflen,
159                    const char *name, int namelen, loff_t offset,
160                    ino_t ino, unsigned int d_type, int *filled)
161 {
162         cfs_dirent_t *dirent = (cfs_dirent_t *) (buf + *filled);
163         cfs_dirent_t  holder;
164         int reclen = ROUND_UP64(NAME_OFFSET(dirent) + namelen + 1);
165
166         /*
167          * @buf is not guaranteed to be properly aligned. To work around,
168          * first fill stack-allocated @holder, then copy @holder into @buf by
169          * memmove().
170          */
171
172         /* check overflow */
173         if ((*filled + reclen) > buflen)
174                 return 1;
175
176         holder.d_ino = ino;
177 #ifdef _DIRENT_HAVE_D_OFF
178         holder.d_off = offset;
179 #endif
180         holder.d_reclen = reclen;
181 #ifdef _DIRENT_HAVE_D_TYPE
182         holder.d_type = (unsigned short) d_type;
183 #endif
184         /* gcc unrolls memcpy() of structs into field-wise assignments,
185          * assuming proper alignment. Humor it. */
186         (*memmover)(dirent, &holder, NAME_OFFSET(dirent));
187         memcpy(dirent->d_name, name, namelen);
188         dirent->d_name[namelen] = 0;
189
190         *filled += reclen;
191
192         return 0;
193 }
194
195 /*
196  * TODO: much of the code here is similar/identical to llite ll_readdir().
197  * These code can be factored out and shared in a common module.
198  */
199
200 ssize_t llu_iop_filldirentries(struct inode *dir, _SYSIO_OFF_T *basep,
201                                char *buf, size_t nbytes)
202 {
203         struct llu_inode_info *lli = llu_i2info(dir);
204         struct intnl_stat     *st = llu_i2stat(dir);
205         loff_t                 pos = *basep;
206         struct ll_dir_chain    chain;
207         cfs_page_t            *page;
208         int filled = 0;
209         int rc;
210         int done;
211         int shift;
212         __u16 type;
213         ENTRY;
214
215         liblustre_wait_event(0);
216
217         if (st->st_size == 0) {
218                 CWARN("dir size is 0?\n");
219                 RETURN(0);
220         }
221
222         if (pos == DIR_END_OFF)
223                 /*
224                  * end-of-file.
225                  */
226                 RETURN(0);
227
228         rc    = 0;
229         done  = 0;
230         shift = 0;
231         ll_dir_chain_init(&chain);
232
233         page = llu_dir_read_page(dir, pos, 0, &chain);
234         while (rc == 0 && !done) {
235                 struct lu_dirpage *dp;
236                 struct lu_dirent  *ent;
237
238                 if (!IS_ERR(page)) {
239                         /*
240                          * If page is empty (end of directoryis reached),
241                          * use this value.
242                          */
243                         __u64 hash = DIR_END_OFF;
244                         __u64 next;
245
246                         dp = page->addr;
247                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
248                              ent = lu_dirent_next(ent)) {
249                                 char          *name;
250                                 int            namelen;
251                                 struct lu_fid  fid;
252                                 ino_t          ino;
253
254                                 hash    = le64_to_cpu(ent->lde_hash);
255                                 namelen = le16_to_cpu(ent->lde_namelen);
256
257                                 if (hash < pos)
258                                         /*
259                                          * Skip until we find target hash
260                                          * value.
261                                          */
262                                         continue;
263
264                                 if (namelen == 0)
265                                         /*
266                                          * Skip dummy record.
267                                          */
268                                         continue;
269
270                                 fid  = ent->lde_fid;
271                                 name = ent->lde_name;
272                                 fid_le_to_cpu(&fid, &fid);
273                                 ino  = cl_fid_build_ino(&fid);
274                                 type = ll_dirent_type_get(ent);
275                                 done = filldir(buf, nbytes, name, namelen,
276                                                (loff_t)hash, ino, type,
277                                                &filled);
278                         }
279                         next = le64_to_cpu(dp->ldp_hash_end);
280                         OBD_PAGE_FREE(page);
281                         if (!done) {
282                                 pos = next;
283                                 if (pos == DIR_END_OFF)
284                                         /*
285                                          * End of directory reached.
286                                          */
287                                         done = 1;
288                                 else if (1 /* chain is exhausted*/)
289                                         /*
290                                          * Normal case: continue to the next
291                                          * page.
292                                          */
293                                         page = llu_dir_read_page(dir, pos, 1,
294                                                                &chain);
295                                 else {
296                                         /*
297                                          * go into overflow page.
298                                          */
299                                 }
300                         } else {
301                                 pos = hash;
302                                 if (filled == 0)
303                                         GOTO(out, filled = -EINVAL);
304                         }
305                 } else {
306                         rc = PTR_ERR(page);
307                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
308                                PFID(&lli->lli_fid), (unsigned long)pos, rc);
309                 }
310         }
311         lli->lli_dir_pos = (loff_t)pos;
312         *basep = lli->lli_dir_pos;
313 out:
314         ll_dir_chain_fini(&chain);
315         liblustre_wait_event(0);
316         RETURN(filled);
317 }