Whamcloud - gitweb
a297c9be62b060a79ad7d86d47afe41fa8c9a358
[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  * Lustre Light directory handling
5  *
6  *  Copyright (c) 2002-2004 Cluster File Systems, Inc.
7  *
8  *   This file is part of Lustre, http://www.lustre.org.
9  *
10  *   Lustre is free software; you can redistribute it and/or
11  *   modify it under the terms of version 2 of the GNU General Public
12  *   License as published by the Free Software Foundation.
13  *
14  *   Lustre is distributed in the hope that it will be useful,
15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17  *   GNU General Public License for more details.
18  *
19  *   You should have received a copy of the GNU General Public License
20  *   along with Lustre; if not, write to the Free Software
21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22  */
23
24 #define DEBUG_SUBSYSTEM S_LLITE
25
26 #include <unistd.h>
27 #include <stdlib.h>
28 #include <string.h>
29 #include <assert.h>
30 #include <time.h>
31 #include <sys/types.h>
32 #include <sys/stat.h>
33 #include <fcntl.h>
34 #include <sys/queue.h>
35
36 #include <sysio.h>
37 #ifdef HAVE_XTIO_H
38 #include <xtio.h>
39 #endif
40 #include <fs.h>
41 #include <mount.h>
42 #include <inode.h>
43 #ifdef HAVE_FILE_H
44 #include <file.h>
45 #endif
46
47 #undef LIST_HEAD
48
49 #ifdef HAVE_ASM_TYPES_H
50 #include <asm/types.h>
51 #elif defined(HAVE_SYS_TYPES_H)
52 #include <sys/types.h>
53 #endif
54
55 #ifdef HAVE_LINUX_UNISTD_H
56 #include <linux/unistd.h>
57 #elif defined(HAVE_UNISTD_H)
58 #include <unistd.h>
59 #endif
60
61 #include <dirent.h>
62
63 #include "llite_lib.h"
64
65 /* (new) readdir implementation overview can be found in lustre/llite/dir.c */
66
67 static int llu_dir_do_readpage(struct inode *inode, struct page *page)
68 {
69         struct llu_inode_info *lli = llu_i2info(inode);
70         struct intnl_stat     *st = llu_i2stat(inode);
71         struct llu_sb_info    *sbi = llu_i2sbi(inode);
72         struct ptlrpc_request *request;
73         struct lustre_handle   lockh;
74         struct mdt_body       *body;
75         struct lookup_intent   it = { .it_op = IT_READDIR };
76         struct md_op_data      op_data = {{ 0 }};
77         ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_UPDATE } };
78         __u64 offset;
79         int rc = 0;
80         ENTRY;
81
82         rc = md_lock_match(sbi->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
83                            &lli->lli_fid, LDLM_IBITS, &policy, LCK_CR, &lockh);
84         if (!rc) {
85                 struct ldlm_enqueue_info einfo = {LDLM_IBITS, LCK_CR,
86                         llu_md_blocking_ast, ldlm_completion_ast, NULL, inode};
87
88                 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, 0,
89                                     LUSTRE_OPC_ANY);
90
91                 rc = md_enqueue(sbi->ll_md_exp, &einfo, &it,
92                                 &op_data, &lockh, NULL, 0,
93                                 LDLM_FL_CANCEL_ON_BLOCK);
94                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
95                 if (request)
96                         ptlrpc_req_finished(request);
97                 if (rc < 0) {
98                         CERROR("lock enqueue: err: %d\n", rc);
99                         RETURN(rc);
100                 }
101         }
102         ldlm_lock_dump_handle(D_OTHER, &lockh);
103
104         offset = (__u64)hash_x_index(page->index);
105         rc = md_readpage(sbi->ll_md_exp, &lli->lli_fid, NULL,
106                          offset, page, &request);
107         if (!rc) {
108                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
109                 LASSERT(body != NULL);         /* checked by md_readpage() */
110
111                 if (body->valid & OBD_MD_FLSIZE)
112                         st->st_size = body->size;
113         } else {
114                 CERROR("read_dir_page(%ld) error %d\n", page->index, rc);
115         }
116         ptlrpc_req_finished(request);
117         EXIT;
118
119         ldlm_lock_decref(&lockh, LCK_CR);
120         return rc;
121 }
122
123 static struct page *llu_dir_read_page(struct inode *ino, __u32 hash,
124                                       int exact, struct ll_dir_chain *chain)
125 {
126         struct page *page;
127         int rc;
128         ENTRY;
129
130         OBD_PAGE_ALLOC(page, 0);
131         if (!page)
132                 RETURN(ERR_PTR(-ENOMEM));
133         page->index = hash_x_index(hash);
134
135         rc = llu_dir_do_readpage(ino, page);
136         if (rc) {
137                 OBD_PAGE_FREE(page);
138                 RETURN(ERR_PTR(rc));
139         }
140
141         return page;
142 }
143
144 enum {
145         EXT2_FT_UNKNOWN,
146         EXT2_FT_REG_FILE,
147         EXT2_FT_DIR,
148         EXT2_FT_CHRDEV,
149         EXT2_FT_BLKDEV,
150         EXT2_FT_FIFO,
151         EXT2_FT_SOCK,
152         EXT2_FT_SYMLINK,
153         EXT2_FT_MAX
154 };
155
156 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
157         [EXT2_FT_UNKNOWN]       DT_UNKNOWN,
158         [EXT2_FT_REG_FILE]      DT_REG,
159         [EXT2_FT_DIR]           DT_DIR,
160         [EXT2_FT_CHRDEV]        DT_CHR,
161         [EXT2_FT_BLKDEV]        DT_BLK,
162         [EXT2_FT_FIFO]          DT_FIFO,
163         [EXT2_FT_SOCK]          DT_SOCK,
164         [EXT2_FT_SYMLINK]       DT_LNK,
165 };
166
167 #define NAME_OFFSET(de) ((int) ((de)->d_name - (char *) (de)))
168 #define ROUND_UP64(x)   (((x)+sizeof(__u64)-1) & ~(sizeof(__u64)-1))
169 static int filldir(char *buf, int buflen,
170                    const char *name, int namelen, loff_t offset,
171                    ino_t ino, unsigned int d_type, int *filled)
172 {
173         struct dirent64 *dirent = (struct dirent64 *) (buf + *filled);
174         int reclen = ROUND_UP64(NAME_OFFSET(dirent) + namelen + 1);
175
176         /* check overflow */
177         if ((*filled + reclen) > buflen)
178                 return 1;
179
180         dirent->d_ino = ino;
181         dirent->d_off = offset;
182         dirent->d_reclen = reclen;
183 #ifndef _AIX
184         dirent->d_type = (unsigned short) d_type;
185 #endif
186         memcpy(dirent->d_name, name, namelen);
187         dirent->d_name[namelen] = 0;
188
189         *filled += reclen;
190
191         return 0;
192 }
193
194 /* 
195  * TODO: much of the code here is similar/identical to llite ll_readdir().
196  * These code can be factored out and shared in a common module.
197  */
198
199 ssize_t llu_iop_filldirentries(struct inode *dir, _SYSIO_OFF_T *basep, 
200                                char *buf, size_t nbytes)
201 {
202         struct llu_inode_info *lli = llu_i2info(dir);
203         struct intnl_stat     *st = llu_i2stat(dir);
204         loff_t                 pos = *basep;
205         struct ll_dir_chain    chain;
206         struct page           *page;
207         int filled = 0;
208         int rc;
209         int done;
210         int shift;
211         ENTRY;
212
213         liblustre_wait_event(0);
214
215         if (st->st_size == 0) {
216                 CWARN("dir size is 0?\n");
217                 RETURN(0);
218         }
219
220         if (pos == DIR_END_OFF)
221                 /*
222                  * end-of-file.
223                  */
224                 RETURN(0);
225
226         rc    = 0;
227         done  = 0;
228         shift = 0;
229         ll_dir_chain_init(&chain);
230
231         page = llu_dir_read_page(dir, pos, 0, &chain);
232         while (rc == 0 && !done) {
233                 struct lu_dirpage *dp;
234                 struct lu_dirent  *ent;
235
236                 if (!IS_ERR(page)) {
237                         /* 
238                          * If page is empty (end of directoryis reached),
239                          * use this value. 
240                          */
241                         __u32 hash = DIR_END_OFF;
242                         __u32 next;
243
244                         dp = page->addr;
245                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
246                              ent = lu_dirent_next(ent)) {
247                                 char          *name;
248                                 int            namelen;
249                                 struct lu_fid  fid;
250                                 ino_t          ino;
251
252                                 hash    = le32_to_cpu(ent->lde_hash);
253                                 namelen = le16_to_cpu(ent->lde_namelen);
254
255                                 if (hash < pos)
256                                         /*
257                                          * Skip until we find target hash
258                                          * value.
259                                          */
260                                         continue;
261
262                                 if (namelen == 0)
263                                         /*
264                                          * Skip dummy record.
265                                          */
266                                         continue;
267
268                                 fid  = ent->lde_fid;
269                                 name = ent->lde_name;
270                                 fid_le_to_cpu(&fid, &fid);
271                                 ino  = llu_fid_build_ino(llu_i2sbi(dir), &fid);
272
273                                 done = filldir(buf, nbytes, name, namelen,
274                                                (loff_t)hash, ino, DT_UNKNOWN,
275                                                &filled);
276                         }
277                         next = le32_to_cpu(dp->ldp_hash_end);
278                         OBD_PAGE_FREE(page);
279                         if (!done) {
280                                 pos = next;
281                                 if (pos == DIR_END_OFF)
282                                         /*
283                                          * End of directory reached.
284                                          */
285                                         done = 1;
286                                 else if (1 /* chain is exhausted*/)
287                                         /*
288                                          * Normal case: continue to the next
289                                          * page.
290                                          */
291                                         page = llu_dir_read_page(dir, pos, 1,
292                                                                &chain);
293                                 else {
294                                         /*
295                                          * go into overflow page.
296                                          */
297                                 }
298                         } else {
299                                 pos = hash;
300                                 if (filled == 0)
301                                         GOTO(out, filled = -EINVAL);
302                         }
303                 } else {
304                         rc = PTR_ERR(page);
305                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
306                                PFID(&lli->lli_fid), (unsigned long)pos, rc);
307                 }
308         }
309         lli->lli_dir_pos = (loff_t)(__s32)pos;
310         *basep = lli->lli_dir_pos;
311 out:
312         ll_dir_chain_fini(&chain);
313         liblustre_wait_event(0);
314         RETURN(filled);
315 }