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 void *(*memmover)(void *, const void *, size_t) = memmove;
 145
 146 #define NAME_OFFSET(de) ((int) ((de)->d_name - (char *) (de)))
 147 #define ROUND_UP64(x)   (((x)+sizeof(__u64)-1) & ~(sizeof(__u64)-1))
 148 static int filldir(char *buf, int buflen,
 149                    const char *name, int namelen, loff_t offset,
 150                    ino_t ino, unsigned int d_type, int *filled)
 151 {
 152         cfs_dirent_t *dirent = (cfs_dirent_t *) (buf + *filled);
 153         cfs_dirent_t  holder;
 154         int reclen = ROUND_UP64(NAME_OFFSET(dirent) + namelen + 1);
 155
 156         /*
 157          * @buf is not guaranteed to be properly aligned. To work around,
 158          * first fill stack-allocated @holder, then copy @holder into @buf by
 159          * memmove().
 160          */
 161
 162         /* check overflow */
 163         if ((*filled + reclen) > buflen)
 164                 return 1;
 165
 166         holder.d_ino = ino;
 167 #ifdef _DIRENT_HAVE_D_OFF
 168         holder.d_off = offset;
 169 #endif
 170         holder.d_reclen = reclen;
 171 #ifdef _DIRENT_HAVE_D_TYPE
 172         holder.d_type = (unsigned short) d_type;
 173 #endif
 174         /* gcc unrolls memcpy() of structs into field-wise assignments,
 175          * assuming proper alignment. Humor it. */
 176         (*memmover)(dirent, &holder, NAME_OFFSET(dirent));
 177         memcpy(dirent->d_name, name, namelen);
 178         dirent->d_name[namelen] = 0;
 179
 180         *filled += reclen;
 181
 182         return 0;
 183 }
 184
 185 /*
 186  * TODO: much of the code here is similar/identical to llite ll_readdir().
 187  * These code can be factored out and shared in a common module.
 188  */
 189
 190 ssize_t llu_iop_filldirentries(struct inode *dir, _SYSIO_OFF_T *basep,
 191                                char *buf, size_t nbytes)
 192 {
 193         struct llu_inode_info *lli = llu_i2info(dir);
 194         struct intnl_stat     *st = llu_i2stat(dir);
 195         loff_t                 pos = *basep;
 196         struct ll_dir_chain    chain;
 197         struct page           *page;
 198         int filled = 0;
 199         int rc;
 200         int done;
 201         int shift;
 202         ENTRY;
 203
 204         liblustre_wait_event(0);
 205
 206         if (st->st_size == 0) {
 207                 CWARN("dir size is 0?\n");
 208                 RETURN(0);
 209         }
 210
 211         if (pos == DIR_END_OFF)
 212                 /*
 213                  * end-of-file.
 214                  */
 215                 RETURN(0);
 216
 217         rc    = 0;
 218         done  = 0;
 219         shift = 0;
 220         ll_dir_chain_init(&chain);
 221
 222         page = llu_dir_read_page(dir, pos, 0, &chain);
 223         while (rc == 0 && !done) {
 224                 struct lu_dirpage *dp;
 225                 struct lu_dirent  *ent;
 226
 227                 if (!IS_ERR(page)) {
 228                         /*
 229                          * If page is empty (end of directoryis reached),
 230                          * use this value.
 231                          */
 232                         __u64 hash = DIR_END_OFF;
 233                         __u64 next;
 234
 235                         dp = page->addr;
 236                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 237                              ent = lu_dirent_next(ent)) {
 238                                 char          *name;
 239                                 int            namelen;
 240                                 struct lu_fid  fid;
 241                                 ino_t          ino;
 242
 243                                 hash    = le64_to_cpu(ent->lde_hash);
 244                                 namelen = le16_to_cpu(ent->lde_namelen);
 245
 246                                 if (hash < pos)
 247                                         /*
 248                                          * Skip until we find target hash
 249                                          * value.
 250                                          */
 251                                         continue;
 252
 253                                 if (namelen == 0)
 254                                         /*
 255                                          * Skip dummy record.
 256                                          */
 257                                         continue;
 258
 259                                 fid  = ent->lde_fid;
 260                                 name = ent->lde_name;
 261                                 fid_le_to_cpu(&fid, &fid);
 262                                 ino  = llu_fid_build_ino(llu_i2sbi(dir), &fid);
 263
 264                                 done = filldir(buf, nbytes, name, namelen,
 265                                                (loff_t)hash, ino, DT_UNKNOWN,
 266                                                &filled);
 267                         }
 268                         next = le64_to_cpu(dp->ldp_hash_end);
 269                         OBD_PAGE_FREE(page);
 270                         if (!done) {
 271                                 pos = next;
 272                                 if (pos == DIR_END_OFF)
 273                                         /*
 274                                          * End of directory reached.
 275                                          */
 276                                         done = 1;
 277                                 else if (1 /* chain is exhausted*/)
 278                                         /*
 279                                          * Normal case: continue to the next
 280                                          * page.
 281                                          */
 282                                         page = llu_dir_read_page(dir, pos, 1,
 283                                                                &chain);
 284                                 else {
 285                                         /*
 286                                          * go into overflow page.
 287                                          */
 288                                 }
 289                         } else {
 290                                 pos = hash;
 291                                 if (filled == 0)
 292                                         GOTO(out, filled = -EINVAL);
 293                         }
 294                 } else {
 295                         rc = PTR_ERR(page);
 296                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 297                                PFID(&lli->lli_fid), (unsigned long)pos, rc);
 298                 }
 299         }
 300         lli->lli_dir_pos = (loff_t)(__s32)pos;
 301         *basep = lli->lli_dir_pos;
 302 out:
 303         ll_dir_chain_fini(&chain);
 304         liblustre_wait_event(0);
 305         RETURN(filled);
 306 }