4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
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
27 * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/liblustre/dir.c
38 * Lustre Light directory handling
41 #define DEBUG_SUBSYSTEM S_LLITE
48 #include <sys/types.h>
51 #include <sys/queue.h>
53 #ifdef HAVE_LINUX_UNISTD_H
54 #include <linux/unistd.h>
55 #elif defined(HAVE_UNISTD_H)
60 #include "llite_lib.h"
63 /* (new) readdir implementation overview can be found in lustre/llite/dir.c */
65 static int llu_dir_do_readpage(struct inode *inode, struct page *page)
67 struct llu_inode_info *lli = llu_i2info(inode);
68 struct intnl_stat *st = llu_i2stat(inode);
69 struct llu_sb_info *sbi = llu_i2sbi(inode);
70 struct ptlrpc_request *request;
71 struct lustre_handle lockh;
72 struct mdt_body *body;
73 struct lookup_intent it = { .it_op = IT_READDIR };
74 struct md_op_data op_data = {{ 0 }};
75 ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_UPDATE } };
79 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, 0, LUSTRE_OPC_ANY);
80 rc = md_lock_match(sbi->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
81 &lli->lli_fid, LDLM_IBITS, &policy, LCK_CR, &lockh);
83 struct ldlm_enqueue_info einfo = {LDLM_IBITS, LCK_CR,
84 llu_md_blocking_ast, ldlm_completion_ast, NULL, NULL,
87 rc = md_enqueue(sbi->ll_md_exp, &einfo, &it,
88 &op_data, &lockh, NULL, 0, NULL,
89 LDLM_FL_CANCEL_ON_BLOCK);
90 request = (struct ptlrpc_request *)it.d.lustre.it_data;
92 ptlrpc_req_finished(request);
94 CERROR("lock enqueue: err: %d\n", rc);
98 ldlm_lock_dump_handle(D_OTHER, &lockh);
100 op_data.op_offset = (__u64)hash_x_index(page->index, 0);
101 op_data.op_npages = 1;
102 rc = md_readpage(sbi->ll_md_exp, &op_data, &page, &request);
104 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
105 LASSERT(body != NULL); /* checked by md_readpage() */
107 if (body->valid & OBD_MD_FLSIZE)
108 st->st_size = body->size;
110 CERROR("read_dir_page(%ld) error %d\n", page->index, rc);
112 ptlrpc_req_finished(request);
115 ldlm_lock_decref(&lockh, LCK_CR);
119 static cfs_page_t *llu_dir_read_page(struct inode *ino, __u64 hash,
120 int exact, struct ll_dir_chain *chain)
126 OBD_PAGE_ALLOC(page, 0);
128 RETURN(ERR_PTR(-ENOMEM));
129 page->index = hash_x_index(hash, 0);
131 rc = llu_dir_do_readpage(ino, page);
140 void *(*memmover)(void *, const void *, size_t) = memmove;
142 #define NAME_OFFSET(de) ((int) ((de)->d_name - (char *) (de)))
143 #define ROUND_UP64(x) (((x)+sizeof(__u64)-1) & ~(sizeof(__u64)-1))
144 static int filldir(char *buf, int buflen, const char *name, int namelen,
145 loff_t offset, ino_t ino, unsigned int d_type, int *filled)
147 struct dirent64 *dirent = (struct dirent64 *)(buf + *filled);
148 struct dirent64 holder;
149 int reclen = ROUND_UP64(NAME_OFFSET(dirent) + namelen + 1);
152 * @buf is not guaranteed to be properly aligned. To work around,
153 * first fill stack-allocated @holder, then copy @holder into @buf by
158 if ((*filled + reclen) > buflen)
162 #ifdef _DIRENT_HAVE_D_OFF
163 holder.d_off = offset;
165 holder.d_reclen = reclen;
166 #ifdef _DIRENT_HAVE_D_TYPE
167 holder.d_type = (unsigned short) d_type;
169 /* gcc unrolls memcpy() of structs into field-wise assignments,
170 * assuming proper alignment. Humor it. */
171 (*memmover)(dirent, &holder, NAME_OFFSET(dirent));
172 memcpy(dirent->d_name, name, namelen);
173 dirent->d_name[namelen] = 0;
181 * TODO: much of the code here is similar/identical to llite ll_readdir().
182 * These code can be factored out and shared in a common module.
185 ssize_t llu_iop_filldirentries(struct inode *dir, _SYSIO_OFF_T *basep,
186 char *buf, size_t nbytes)
188 struct llu_inode_info *lli = llu_i2info(dir);
189 struct intnl_stat *st = llu_i2stat(dir);
191 struct ll_dir_chain chain;
199 liblustre_wait_event(0);
201 if (st->st_size == 0) {
202 CWARN("dir size is 0?\n");
206 if (pos == MDS_DIR_END_OFF)
214 ll_dir_chain_init(&chain);
216 page = llu_dir_read_page(dir, pos, 0, &chain);
217 while (rc == 0 && !done) {
218 struct lu_dirpage *dp;
219 struct lu_dirent *ent;
223 * If page is empty (end of directoryis reached),
226 __u64 hash = MDS_DIR_END_OFF;
230 for (ent = lu_dirent_start(dp); ent != NULL && !done;
231 ent = lu_dirent_next(ent)) {
237 hash = le64_to_cpu(ent->lde_hash);
238 namelen = le16_to_cpu(ent->lde_namelen);
242 * Skip until we find target hash
254 name = ent->lde_name;
255 fid_le_to_cpu(&fid, &fid);
256 ino = cl_fid_build_ino(&fid, 0);
257 type = ll_dirent_type_get(ent);
258 done = filldir(buf, nbytes, name, namelen,
259 (loff_t)hash, ino, type,
262 next = le64_to_cpu(dp->ldp_hash_end);
266 if (pos == MDS_DIR_END_OFF)
268 * End of directory reached.
271 else if (1 /* chain is exhausted*/)
273 * Normal case: continue to the next
276 page = llu_dir_read_page(dir, pos, 1,
280 * go into overflow page.
286 GOTO(out, filled = -EINVAL);
290 CERROR("error reading dir "DFID" at %lu: rc %d\n",
291 PFID(&lli->lli_fid), (unsigned long)pos, rc);
294 lli->lli_dir_pos = (loff_t)pos;
295 *basep = lli->lli_dir_pos;
297 ll_dir_chain_fini(&chain);
298 liblustre_wait_event(0);