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, Whamcloud, Inc.
  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 #ifdef HAVE_LINUX_UNISTD_H
  54 #include <linux/unistd.h>
  55 #elif defined(HAVE_UNISTD_H)
  56 #include <unistd.h>
  57 #endif
  58
  59
  60 #include "llite_lib.h"
  61 #include <dirent.h>
  62
  63 /* (new) readdir implementation overview can be found in lustre/llite/dir.c */
  64
  65 static int llu_dir_do_readpage(struct inode *inode, struct page *page)
  66 {
  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 } };
  76         int rc = 0;
  77         ENTRY;
  78
  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);
  82         if (!rc) {
  83                 struct ldlm_enqueue_info einfo = {LDLM_IBITS, LCK_CR,
  84                         llu_md_blocking_ast, ldlm_completion_ast, NULL, NULL,
  85                         inode};
  86
  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;
  91                 if (request)
  92                         ptlrpc_req_finished(request);
  93                 if (rc < 0) {
  94                         CERROR("lock enqueue: err: %d\n", rc);
  95                         RETURN(rc);
  96                 }
  97         }
  98         ldlm_lock_dump_handle(D_OTHER, &lockh);
  99
 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);
 103         if (!rc) {
 104                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 105                 LASSERT(body != NULL);         /* checked by md_readpage() */
 106
 107                 if (body->valid & OBD_MD_FLSIZE)
 108                         st->st_size = body->size;
 109         } else {
 110                 CERROR("read_dir_page(%ld) error %d\n", page->index, rc);
 111         }
 112         ptlrpc_req_finished(request);
 113         EXIT;
 114
 115         ldlm_lock_decref(&lockh, LCK_CR);
 116         return rc;
 117 }
 118
 119 static cfs_page_t *llu_dir_read_page(struct inode *ino, __u64 hash,
 120                                      int exact, struct ll_dir_chain *chain)
 121 {
 122         cfs_page_t *page;
 123         int rc;
 124         ENTRY;
 125
 126         OBD_PAGE_ALLOC(page, 0);
 127         if (!page)
 128                 RETURN(ERR_PTR(-ENOMEM));
 129         page->index = hash_x_index(hash, 0);
 130
 131         rc = llu_dir_do_readpage(ino, page);
 132         if (rc) {
 133                 OBD_PAGE_FREE(page);
 134                 RETURN(ERR_PTR(rc));
 135         }
 136
 137         return page;
 138 }
 139
 140 void *(*memmover)(void *, const void *, size_t) = memmove;
 141
 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,
 145                    const char *name, int namelen, loff_t offset,
 146                    ino_t ino, unsigned int d_type, int *filled)
 147 {
 148         cfs_dirent_t *dirent = (cfs_dirent_t *) (buf + *filled);
 149         cfs_dirent_t  holder;
 150         int reclen = ROUND_UP64(NAME_OFFSET(dirent) + namelen + 1);
 151
 152         /*
 153          * @buf is not guaranteed to be properly aligned. To work around,
 154          * first fill stack-allocated @holder, then copy @holder into @buf by
 155          * memmove().
 156          */
 157
 158         /* check overflow */
 159         if ((*filled + reclen) > buflen)
 160                 return 1;
 161
 162         holder.d_ino = ino;
 163 #ifdef _DIRENT_HAVE_D_OFF
 164         holder.d_off = offset;
 165 #endif
 166         holder.d_reclen = reclen;
 167 #ifdef _DIRENT_HAVE_D_TYPE
 168         holder.d_type = (unsigned short) d_type;
 169 #endif
 170         /* gcc unrolls memcpy() of structs into field-wise assignments,
 171          * assuming proper alignment. Humor it. */
 172         (*memmover)(dirent, &holder, NAME_OFFSET(dirent));
 173         memcpy(dirent->d_name, name, namelen);
 174         dirent->d_name[namelen] = 0;
 175
 176         *filled += reclen;
 177
 178         return 0;
 179 }
 180
 181 /*
 182  * TODO: much of the code here is similar/identical to llite ll_readdir().
 183  * These code can be factored out and shared in a common module.
 184  */
 185
 186 ssize_t llu_iop_filldirentries(struct inode *dir, _SYSIO_OFF_T *basep,
 187                                char *buf, size_t nbytes)
 188 {
 189         struct llu_inode_info *lli = llu_i2info(dir);
 190         struct intnl_stat     *st = llu_i2stat(dir);
 191         loff_t                 pos = *basep;
 192         struct ll_dir_chain    chain;
 193         cfs_page_t            *page;
 194         int filled = 0;
 195         int rc;
 196         int done;
 197         __u16 type;
 198         ENTRY;
 199
 200         liblustre_wait_event(0);
 201
 202         if (st->st_size == 0) {
 203                 CWARN("dir size is 0?\n");
 204                 RETURN(0);
 205         }
 206
 207         if (pos == MDS_DIR_END_OFF)
 208                 /*
 209                  * end-of-file.
 210                  */
 211                 RETURN(0);
 212
 213         rc    = 0;
 214         done  = 0;
 215         ll_dir_chain_init(&chain);
 216
 217         page = llu_dir_read_page(dir, pos, 0, &chain);
 218         while (rc == 0 && !done) {
 219                 struct lu_dirpage *dp;
 220                 struct lu_dirent  *ent;
 221
 222                 if (!IS_ERR(page)) {
 223                         /*
 224                          * If page is empty (end of directoryis reached),
 225                          * use this value.
 226                          */
 227                         __u64 hash = MDS_DIR_END_OFF;
 228                         __u64 next;
 229
 230                         dp = page->addr;
 231                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 232                              ent = lu_dirent_next(ent)) {
 233                                 char          *name;
 234                                 int            namelen;
 235                                 struct lu_fid  fid;
 236                                 __u64          ino;
 237
 238                                 hash    = le64_to_cpu(ent->lde_hash);
 239                                 namelen = le16_to_cpu(ent->lde_namelen);
 240
 241                                 if (hash < pos)
 242                                         /*
 243                                          * Skip until we find target hash
 244                                          * value.
 245                                          */
 246                                         continue;
 247
 248                                 if (namelen == 0)
 249                                         /*
 250                                          * Skip dummy record.
 251                                          */
 252                                         continue;
 253
 254                                 fid  = ent->lde_fid;
 255                                 name = ent->lde_name;
 256                                 fid_le_to_cpu(&fid, &fid);
 257                                 ino  = cl_fid_build_ino(&fid, 0);
 258                                 type = ll_dirent_type_get(ent);
 259                                 done = filldir(buf, nbytes, name, namelen,
 260                                                (loff_t)hash, ino, type,
 261                                                &filled);
 262                         }
 263                         next = le64_to_cpu(dp->ldp_hash_end);
 264                         OBD_PAGE_FREE(page);
 265                         if (!done) {
 266                                 pos = next;
 267                                 if (pos == MDS_DIR_END_OFF)
 268                                         /*
 269                                          * End of directory reached.
 270                                          */
 271                                         done = 1;
 272                                 else if (1 /* chain is exhausted*/)
 273                                         /*
 274                                          * Normal case: continue to the next
 275                                          * page.
 276                                          */
 277                                         page = llu_dir_read_page(dir, pos, 1,
 278                                                                &chain);
 279                                 else {
 280                                         /*
 281                                          * go into overflow page.
 282                                          */
 283                                 }
 284                         } else {
 285                                 pos = hash;
 286                                 if (filled == 0)
 287                                         GOTO(out, filled = -EINVAL);
 288                         }
 289                 } else {
 290                         rc = PTR_ERR(page);
 291                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 292                                PFID(&lli->lli_fid), (unsigned long)pos, rc);
 293                 }
 294         }
 295         lli->lli_dir_pos = (loff_t)pos;
 296         *basep = lli->lli_dir_pos;
 297 out:
 298         ll_dir_chain_fini(&chain);
 299         liblustre_wait_event(0);
 300         RETURN(filled);
 301 }