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         struct obd_device *obddev = class_exp2obd(sbi->ll_md_exp);
  78         struct ldlm_res_id res_id =
  79                 { .name = {fid_seq(&lli->lli_fid),
  80                            fid_oid(&lli->lli_fid),
  81                            fid_ver(&lli->lli_fid)} };
  82         ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_UPDATE } };
  83         ENTRY;
  84
  85         rc = ldlm_lock_match(obddev->obd_namespace, LDLM_FL_BLOCK_GRANTED,
  86                              &res_id, LDLM_IBITS, &policy, LCK_CR, &lockh);
  87         if (!rc) {
  88                 struct ldlm_enqueue_info einfo = {LDLM_IBITS, LCK_CR,
  89                         llu_md_blocking_ast, ldlm_completion_ast, NULL, inode};
  90
  91                 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, 0,
  92                                     LUSTRE_OPC_ANY);
  93
  94                 rc = md_enqueue(sbi->ll_md_exp, &einfo, &it,
  95                                 &op_data, &lockh, NULL, 0,
  96                                 LDLM_FL_CANCEL_ON_BLOCK);
  97                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
  98                 if (request)
  99                         ptlrpc_req_finished(request);
 100                 if (rc < 0) {
 101                         CERROR("lock enqueue: err: %d\n", rc);
 102                         RETURN(rc);
 103                 }
 104         }
 105         ldlm_lock_dump_handle(D_OTHER, &lockh);
 106
 107         offset = (__u64)page->index << CFS_PAGE_SHIFT;
 108         rc = md_readpage(sbi->ll_md_exp, &lli->lli_fid, NULL,
 109                          offset, page, &request);
 110         if (!rc) {
 111                 body = lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF,
 112                                       sizeof(*body));
 113                 LASSERT(body != NULL);         /* checked by md_readpage() */
 114                 /* swabbed by md_readpage() */
 115                 LASSERT(lustre_rep_swabbed(request, REPLY_REC_OFF));
 116
 117                 st->st_size = body->size;
 118         } else {
 119                 CERROR("read_dir_page(%ld) error %d\n", page->index, rc);
 120         }
 121         ptlrpc_req_finished(request);
 122         EXIT;
 123
 124         ldlm_lock_decref(&lockh, LCK_CR);
 125         return rc;
 126 }
 127
 128 static struct page *llu_dir_read_page(struct inode *ino, unsigned long pgidx)
 129 {
 130         struct page *page;
 131         int rc;
 132         ENTRY;
 133
 134         page = cfs_alloc_page(0);
 135         if (!page) {
 136                 CERROR("alloc page failed\n");
 137                 RETURN(ERR_PTR(-ENOMEM));
 138         }
 139         page->index = pgidx;
 140
 141         rc = llu_dir_do_readpage(ino, page);
 142         if (rc) {
 143                 free_page(page);
 144                 RETURN(ERR_PTR(rc));
 145         }
 146
 147         return page;
 148 }
 149
 150 enum {
 151         EXT2_FT_UNKNOWN,
 152         EXT2_FT_REG_FILE,
 153         EXT2_FT_DIR,
 154         EXT2_FT_CHRDEV,
 155         EXT2_FT_BLKDEV,
 156         EXT2_FT_FIFO,
 157         EXT2_FT_SOCK,
 158         EXT2_FT_SYMLINK,
 159         EXT2_FT_MAX
 160 };
 161
 162 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
 163         [EXT2_FT_UNKNOWN]       DT_UNKNOWN,
 164         [EXT2_FT_REG_FILE]      DT_REG,
 165         [EXT2_FT_DIR]           DT_DIR,
 166         [EXT2_FT_CHRDEV]        DT_CHR,
 167         [EXT2_FT_BLKDEV]        DT_BLK,
 168         [EXT2_FT_FIFO]          DT_FIFO,
 169         [EXT2_FT_SOCK]          DT_SOCK,
 170         [EXT2_FT_SYMLINK]       DT_LNK,
 171 };
 172
 173 #define NAME_OFFSET(de) ((int) ((de)->d_name - (char *) (de)))
 174 #define ROUND_UP64(x)   (((x)+sizeof(__u64)-1) & ~(sizeof(__u64)-1))
 175
 176 static int filldir(char *buf, int buflen,
 177                    const char *name, int namelen, loff_t offset,
 178                    ino_t ino, unsigned int d_type, int *filled)
 179 {
 180         struct dirent64 *dirent = (struct dirent64 *) (buf + *filled);
 181         int reclen = ROUND_UP64(NAME_OFFSET(dirent) + namelen + 1);
 182
 183         /* check overflow */
 184         if ((*filled + reclen) > buflen)
 185                 return 1;
 186
 187         dirent->d_ino = ino;
 188         dirent->d_off = offset;
 189         dirent->d_reclen = reclen;
 190 #ifndef _AIX
 191         dirent->d_type = (unsigned short) d_type;
 192 #endif
 193         memcpy(dirent->d_name, name, namelen);
 194         dirent->d_name[namelen] = 0;
 195
 196         *filled += reclen;
 197
 198         return 0;
 199 }
 200
 201 ssize_t llu_iop_filldirentries(struct inode *ino, _SYSIO_OFF_T *basep,
 202                                char *buf, size_t nbytes)
 203 {
 204         struct llu_inode_info *lli = llu_i2info(ino);
 205         struct intnl_stat *st = llu_i2stat(ino);
 206         loff_t pos = *basep, offset;
 207         unsigned long maxpages, pgidx;
 208         int filled = 0;
 209         ENTRY;
 210
 211         liblustre_wait_event(0);
 212
 213         if (st->st_size == 0) {
 214                 CWARN("dir size is 0?\n");
 215                 RETURN(0);
 216         }
 217
 218         if (pos == -1)
 219                 pos = lli->lli_dir_pos;
 220
 221         maxpages = (st->st_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
 222         pgidx = pos >> CFS_PAGE_SHIFT;
 223         offset = pos & ~CFS_PAGE_MASK;
 224
 225         for ( ; pgidx < maxpages ; pgidx++, offset = 0) {
 226                 struct page *page;
 227                 struct ext2_dirent *de;
 228                 char *addr, *limit;
 229
 230                 page = llu_dir_read_page(ino, pgidx);
 231                 if (IS_ERR(page))
 232                         continue;
 233
 234                 /* size might have been updated by md_readpage */
 235                 maxpages = (st->st_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
 236
 237                 /* fill in buffer */
 238                 addr = page->addr;
 239                 limit = addr + CFS_PAGE_SIZE - EXT2_DIR_REC_LEN(1);
 240                 de = (struct ext2_dirent *) (addr + offset);
 241
 242                 for ( ; (char*) de <= limit; de = ext2_next_entry(de)) {
 243                         if (de->inode) {
 244                                 int over;
 245                                 unsigned char d_type = DT_UNKNOWN;
 246
 247                                 if (de->file_type < EXT2_FT_MAX)
 248                                         d_type = ext2_filetype_table[de->file_type];
 249
 250                                 offset = (char*) de - addr;
 251                                 over =  filldir(buf, nbytes, de->name, de->name_len,
 252                                                 (((__u64)pgidx << PAGE_SHIFT) | offset) +
 253                                                 le16_to_cpu(de->rec_len),
 254                                                 le32_to_cpu(de->inode), d_type, &filled);
 255                                 if (over) {
 256                                         free_page(page);
 257                                         /*
 258                                          * if buffer overflow with no data
 259                                          * returned yet, then report error
 260                                          * instead of eof
 261                                          */
 262                                         if (filled == 0)
 263                                                 RETURN(-EINVAL);
 264                                         GOTO(done, 0);
 265                                 }
 266                         }
 267                 }
 268
 269                 free_page(page);
 270         }
 271 done:
 272         lli->lli_dir_pos = pgidx << CFS_PAGE_SHIFT | offset;
 273         *basep = lli->lli_dir_pos;
 274         liblustre_wait_event(0);
 275         RETURN(filled);
 276 }