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 [sun.com URL with a
  20  * copy of GPLv2].
  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 #undef LIST_HEAD
  65
  66 #ifdef HAVE_ASM_TYPES_H
  67 #include <asm/types.h>
  68 #elif defined(HAVE_SYS_TYPES_H)
  69 #include <sys/types.h>
  70 #endif
  71
  72 #ifdef HAVE_LINUX_UNISTD_H
  73 #include <linux/unistd.h>
  74 #elif defined(HAVE_UNISTD_H)
  75 #include <unistd.h>
  76 #endif
  77
  78 #include <dirent.h>
  79
  80 #include "llite_lib.h"
  81
  82 /* (new) readdir implementation overview can be found in lustre/llite/dir.c */
  83
  84 static int llu_dir_do_readpage(struct inode *inode, struct page *page)
  85 {
  86         struct llu_inode_info *lli = llu_i2info(inode);
  87         struct intnl_stat     *st = llu_i2stat(inode);
  88         struct llu_sb_info    *sbi = llu_i2sbi(inode);
  89         struct ptlrpc_request *request;
  90         struct lustre_handle   lockh;
  91         struct mdt_body       *body;
  92         struct lookup_intent   it = { .it_op = IT_READDIR };
  93         struct md_op_data      op_data = {{ 0 }};
  94         ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_UPDATE } };
  95         __u64 offset;
  96         int rc = 0;
  97         ENTRY;
  98
  99         rc = md_lock_match(sbi->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 100                            &lli->lli_fid, LDLM_IBITS, &policy, LCK_CR, &lockh);
 101         if (!rc) {
 102                 struct ldlm_enqueue_info einfo = {LDLM_IBITS, LCK_CR,
 103                         llu_md_blocking_ast, ldlm_completion_ast, NULL, inode};
 104
 105                 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, 0,
 106                                     LUSTRE_OPC_ANY);
 107
 108                 rc = md_enqueue(sbi->ll_md_exp, &einfo, &it,
 109                                 &op_data, &lockh, NULL, 0,
 110                                 LDLM_FL_CANCEL_ON_BLOCK);
 111                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 112                 if (request)
 113                         ptlrpc_req_finished(request);
 114                 if (rc < 0) {
 115                         CERROR("lock enqueue: err: %d\n", rc);
 116                         RETURN(rc);
 117                 }
 118         }
 119         ldlm_lock_dump_handle(D_OTHER, &lockh);
 120
 121         offset = (__u64)hash_x_index(page->index);
 122         rc = md_readpage(sbi->ll_md_exp, &lli->lli_fid, NULL,
 123                          offset, page, &request);
 124         if (!rc) {
 125                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 126                 LASSERT(body != NULL);         /* checked by md_readpage() */
 127
 128                 if (body->valid & OBD_MD_FLSIZE)
 129                         st->st_size = body->size;
 130         } else {
 131                 CERROR("read_dir_page(%ld) error %d\n", page->index, rc);
 132         }
 133         ptlrpc_req_finished(request);
 134         EXIT;
 135
 136         ldlm_lock_decref(&lockh, LCK_CR);
 137         return rc;
 138 }
 139
 140 static struct page *llu_dir_read_page(struct inode *ino, __u32 hash,
 141                                       int exact, struct ll_dir_chain *chain)
 142 {
 143         struct page *page;
 144         int rc;
 145         ENTRY;
 146
 147         OBD_PAGE_ALLOC(page, 0);
 148         if (!page)
 149                 RETURN(ERR_PTR(-ENOMEM));
 150         page->index = hash_x_index(hash);
 151
 152         rc = llu_dir_do_readpage(ino, page);
 153         if (rc) {
 154                 OBD_PAGE_FREE(page);
 155                 RETURN(ERR_PTR(rc));
 156         }
 157
 158         return page;
 159 }
 160
 161 void *(*memmover)(void *, const void *, size_t) = memmove;
 162
 163 #define NAME_OFFSET(de) ((int) ((de)->d_name - (char *) (de)))
 164 #define ROUND_UP64(x)   (((x)+sizeof(__u64)-1) & ~(sizeof(__u64)-1))
 165 static int filldir(char *buf, int buflen,
 166                    const char *name, int namelen, loff_t offset,
 167                    ino_t ino, unsigned int d_type, int *filled)
 168 {
 169         cfs_dirent_t *dirent = (cfs_dirent_t *) (buf + *filled);
 170         cfs_dirent_t  holder;
 171         int reclen = ROUND_UP64(NAME_OFFSET(dirent) + namelen + 1);
 172
 173         /*
 174          * @buf is not guaranteed to be properly aligned. To work around,
 175          * first fill stack-allocated @holder, then copy @holder into @buf by
 176          * memmove().
 177          */
 178
 179         /* check overflow */
 180         if ((*filled + reclen) > buflen)
 181                 return 1;
 182
 183         holder.d_ino = ino;
 184 #ifdef _DIRENT_HAVE_D_OFF
 185         holder.d_off = offset;
 186 #endif
 187         holder.d_reclen = reclen;
 188 #ifdef _DIRENT_HAVE_D_TYPE
 189         holder.d_type = (unsigned short) d_type;
 190 #endif
 191         /* gcc unrolls memcpy() of structs into field-wise assignments,
 192          * assuming proper alignment. Humor it. */
 193         (*memmover)(dirent, &holder, NAME_OFFSET(dirent));
 194         memcpy(dirent->d_name, name, namelen);
 195         dirent->d_name[namelen] = 0;
 196
 197         *filled += reclen;
 198
 199         return 0;
 200 }
 201
 202 /*
 203  * TODO: much of the code here is similar/identical to llite ll_readdir().
 204  * These code can be factored out and shared in a common module.
 205  */
 206
 207 ssize_t llu_iop_filldirentries(struct inode *dir, _SYSIO_OFF_T *basep,
 208                                char *buf, size_t nbytes)
 209 {
 210         struct llu_inode_info *lli = llu_i2info(dir);
 211         struct intnl_stat     *st = llu_i2stat(dir);
 212         loff_t                 pos = *basep;
 213         struct ll_dir_chain    chain;
 214         struct page           *page;
 215         int filled = 0;
 216         int rc;
 217         int done;
 218         int shift;
 219         ENTRY;
 220
 221         liblustre_wait_event(0);
 222
 223         if (st->st_size == 0) {
 224                 CWARN("dir size is 0?\n");
 225                 RETURN(0);
 226         }
 227
 228         if (pos == DIR_END_OFF)
 229                 /*
 230                  * end-of-file.
 231                  */
 232                 RETURN(0);
 233
 234         rc    = 0;
 235         done  = 0;
 236         shift = 0;
 237         ll_dir_chain_init(&chain);
 238
 239         page = llu_dir_read_page(dir, pos, 0, &chain);
 240         while (rc == 0 && !done) {
 241                 struct lu_dirpage *dp;
 242                 struct lu_dirent  *ent;
 243
 244                 if (!IS_ERR(page)) {
 245                         /*
 246                          * If page is empty (end of directoryis reached),
 247                          * use this value.
 248                          */
 249                         __u64 hash = DIR_END_OFF;
 250                         __u64 next;
 251
 252                         dp = page->addr;
 253                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 254                              ent = lu_dirent_next(ent)) {
 255                                 char          *name;
 256                                 int            namelen;
 257                                 struct lu_fid  fid;
 258                                 ino_t          ino;
 259
 260                                 hash    = le64_to_cpu(ent->lde_hash);
 261                                 namelen = le16_to_cpu(ent->lde_namelen);
 262
 263                                 if (hash < pos)
 264                                         /*
 265                                          * Skip until we find target hash
 266                                          * value.
 267                                          */
 268                                         continue;
 269
 270                                 if (namelen == 0)
 271                                         /*
 272                                          * Skip dummy record.
 273                                          */
 274                                         continue;
 275
 276                                 fid  = ent->lde_fid;
 277                                 name = ent->lde_name;
 278                                 fid_le_to_cpu(&fid, &fid);
 279                                 ino  = llu_fid_build_ino(llu_i2sbi(dir), &fid);
 280
 281                                 done = filldir(buf, nbytes, name, namelen,
 282                                                (loff_t)hash, ino, DT_UNKNOWN,
 283                                                &filled);
 284                         }
 285                         next = le64_to_cpu(dp->ldp_hash_end);
 286                         OBD_PAGE_FREE(page);
 287                         if (!done) {
 288                                 pos = next;
 289                                 if (pos == DIR_END_OFF)
 290                                         /*
 291                                          * End of directory reached.
 292                                          */
 293                                         done = 1;
 294                                 else if (1 /* chain is exhausted*/)
 295                                         /*
 296                                          * Normal case: continue to the next
 297                                          * page.
 298                                          */
 299                                         page = llu_dir_read_page(dir, pos, 1,
 300                                                                &chain);
 301                                 else {
 302                                         /*
 303                                          * go into overflow page.
 304                                          */
 305                                 }
 306                         } else {
 307                                 pos = hash;
 308                                 if (filled == 0)
 309                                         GOTO(out, filled = -EINVAL);
 310                         }
 311                 } else {
 312                         rc = PTR_ERR(page);
 313                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 314                                PFID(&lli->lli_fid), (unsigned long)pos, rc);
 315                 }
 316         }
 317         lli->lli_dir_pos = (loff_t)(__s32)pos;
 318         *basep = lli->lli_dir_pos;
 319 out:
 320         ll_dir_chain_fini(&chain);
 321         liblustre_wait_event(0);
 322         RETURN(filled);
 323 }