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 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         __u64 offset;
  71         int rc = 0;
  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;
  77         ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_UPDATE } };
  78         ENTRY;
  79
  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, inode};
  85
  86                 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, 0,
  87                                     LUSTRE_OPC_ANY);
  88
  89                 rc = md_enqueue(sbi->ll_md_exp, &einfo, &it,
  90                                 &op_data, &lockh, NULL, 0,
  91                                 LDLM_FL_CANCEL_ON_BLOCK);
  92                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
  93                 if (request)
  94                         ptlrpc_req_finished(request);
  95                 if (rc < 0) {
  96                         CERROR("lock enqueue: err: %d\n", rc);
  97                         RETURN(rc);
  98                 }
  99         }
 100         ldlm_lock_dump_handle(D_OTHER, &lockh);
 101
 102         offset = (__u64)page->index << CFS_PAGE_SHIFT;
 103         rc = md_readpage(sbi->ll_md_exp, &lli->lli_fid, NULL,
 104                          offset, page, &request);
 105         if (!rc) {
 106                 body = lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF,
 107                                       sizeof(*body));
 108                 LASSERT(body != NULL);         /* checked by md_readpage() */
 109                 /* swabbed by md_readpage() */
 110                 LASSERT(lustre_rep_swabbed(request, REPLY_REC_OFF));
 111
 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, unsigned long pgidx)
 124 {
 125         struct page *page;
 126         int rc;
 127         ENTRY;
 128
 129         page = cfs_alloc_page(0);
 130         if (!page) {
 131                 CERROR("alloc page failed\n");
 132                 RETURN(ERR_PTR(-ENOMEM));
 133         }
 134         page->index = pgidx;
 135
 136         rc = llu_dir_do_readpage(ino, page);
 137         if (rc) {
 138                 free_page(page);
 139                 RETURN(ERR_PTR(rc));
 140         }
 141
 142         return page;
 143 }
 144
 145 enum {
 146         EXT2_FT_UNKNOWN,
 147         EXT2_FT_REG_FILE,
 148         EXT2_FT_DIR,
 149         EXT2_FT_CHRDEV,
 150         EXT2_FT_BLKDEV,
 151         EXT2_FT_FIFO,
 152         EXT2_FT_SOCK,
 153         EXT2_FT_SYMLINK,
 154         EXT2_FT_MAX
 155 };
 156
 157 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
 158         [EXT2_FT_UNKNOWN]       DT_UNKNOWN,
 159         [EXT2_FT_REG_FILE]      DT_REG,
 160         [EXT2_FT_DIR]           DT_DIR,
 161         [EXT2_FT_CHRDEV]        DT_CHR,
 162         [EXT2_FT_BLKDEV]        DT_BLK,
 163         [EXT2_FT_FIFO]          DT_FIFO,
 164         [EXT2_FT_SOCK]          DT_SOCK,
 165         [EXT2_FT_SYMLINK]       DT_LNK,
 166 };
 167
 168 #define NAME_OFFSET(de) ((int) ((de)->d_name - (char *) (de)))
 169 #define ROUND_UP64(x)   (((x)+sizeof(__u64)-1) & ~(sizeof(__u64)-1))
 170
 171 static int filldir(char *buf, int buflen,
 172                    const char *name, int namelen, loff_t offset,
 173                    ino_t ino, unsigned int d_type, int *filled)
 174 {
 175         struct dirent64 *dirent = (struct dirent64 *) (buf + *filled);
 176         int reclen = ROUND_UP64(NAME_OFFSET(dirent) + namelen + 1);
 177
 178         /* check overflow */
 179         if ((*filled + reclen) > buflen)
 180                 return 1;
 181
 182         dirent->d_ino = ino;
 183         dirent->d_off = offset;
 184         dirent->d_reclen = reclen;
 185 #ifndef _AIX
 186         dirent->d_type = (unsigned short) d_type;
 187 #endif
 188         memcpy(dirent->d_name, name, namelen);
 189         dirent->d_name[namelen] = 0;
 190
 191         *filled += reclen;
 192
 193         return 0;
 194 }
 195
 196 ssize_t llu_iop_filldirentries(struct inode *ino, _SYSIO_OFF_T *basep,
 197                                char *buf, size_t nbytes)
 198 {
 199         struct llu_inode_info *lli = llu_i2info(ino);
 200         struct intnl_stat *st = llu_i2stat(ino);
 201         loff_t pos = *basep, offset;
 202         unsigned long maxpages, pgidx;
 203         int filled = 0;
 204         ENTRY;
 205
 206         liblustre_wait_event(0);
 207
 208         if (st->st_size == 0) {
 209                 CWARN("dir size is 0?\n");
 210                 RETURN(0);
 211         }
 212
 213         if (pos == -1)
 214                 pos = lli->lli_dir_pos;
 215
 216         maxpages = (st->st_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
 217         pgidx = pos >> CFS_PAGE_SHIFT;
 218         offset = pos & ~CFS_PAGE_MASK;
 219
 220         for ( ; pgidx < maxpages ; pgidx++, offset = 0) {
 221                 struct page *page;
 222                 struct ext2_dirent *de;
 223                 char *addr, *limit;
 224
 225                 page = llu_dir_read_page(ino, pgidx);
 226                 if (IS_ERR(page))
 227                         continue;
 228
 229                 /* size might have been updated by md_readpage */
 230                 maxpages = (st->st_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
 231
 232                 /* fill in buffer */
 233                 addr = page->addr;
 234                 limit = addr + CFS_PAGE_SIZE - EXT2_DIR_REC_LEN(1);
 235                 de = (struct ext2_dirent *) (addr + offset);
 236
 237                 for ( ; (char*) de <= limit; de = ext2_next_entry(de)) {
 238                         if (de->inode) {
 239                                 int over;
 240                                 unsigned char d_type = DT_UNKNOWN;
 241
 242                                 if (de->file_type < EXT2_FT_MAX)
 243                                         d_type = ext2_filetype_table[de->file_type];
 244
 245                                 offset = (char*) de - addr;
 246                                 over =  filldir(buf, nbytes, de->name, de->name_len,
 247                                                 (((__u64)pgidx << PAGE_SHIFT) | offset) +
 248                                                 le16_to_cpu(de->rec_len),
 249                                                 le32_to_cpu(de->inode), d_type, &filled);
 250                                 if (over) {
 251                                         free_page(page);
 252                                         /*
 253                                          * if buffer overflow with no data
 254                                          * returned yet, then report error
 255                                          * instead of eof
 256                                          */
 257                                         if (filled == 0)
 258                                                 RETURN(-EINVAL);
 259                                         GOTO(done, 0);
 260                                 }
 261                         }
 262                 }
 263
 264                 free_page(page);
 265         }
 266 done:
 267         lli->lli_dir_pos = pgidx << CFS_PAGE_SHIFT | offset;
 268         *basep = lli->lli_dir_pos;
 269         liblustre_wait_event(0);
 270         RETURN(filled);
 271 }