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
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  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 static int llu_dir_do_readpage(struct inode *inode, struct page *page)
  83 {
  84         struct llu_inode_info *lli = llu_i2info(inode);
  85         struct intnl_stat *st = llu_i2stat(inode);
  86         struct llu_sb_info *sbi = llu_i2sbi(inode);
  87         struct ll_fid mdc_fid;
  88         __u64 offset;
  89         int rc = 0;
  90         struct ptlrpc_request *request;
  91         struct lustre_handle lockh;
  92         struct mds_body *body;
  93         struct lookup_intent it = { .it_op = IT_READDIR };
  94         struct mdc_op_data data;
  95         struct obd_device *obddev = class_exp2obd(sbi->ll_mdc_exp);
  96         struct ldlm_res_id res_id =
  97                 { .name = {st->st_ino, (__u64)lli->lli_st_generation} };
  98         ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_UPDATE } };
  99         ENTRY;
 100
 101         rc = ldlm_lock_match(obddev->obd_namespace, LDLM_FL_BLOCK_GRANTED,
 102                              &res_id, LDLM_IBITS, &policy, LCK_CR, &lockh);
 103         if (!rc) {
 104                 struct ldlm_enqueue_info einfo = {LDLM_IBITS, LCK_CR,
 105                         llu_mdc_blocking_ast, ldlm_completion_ast, NULL, inode};
 106
 107                 llu_prepare_mdc_op_data(&data, inode, NULL, NULL, 0, 0);
 108
 109                 rc = mdc_enqueue(sbi->ll_mdc_exp, &einfo, &it,
 110                                  &data, &lockh, NULL, 0,
 111                                  LDLM_FL_CANCEL_ON_BLOCK);
 112                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 113                 if (request)
 114                         ptlrpc_req_finished(request);
 115                 if (rc < 0) {
 116                         CERROR("lock enqueue: err: %d\n", rc);
 117                         RETURN(rc);
 118                 }
 119         }
 120         ldlm_lock_dump_handle(D_OTHER, &lockh);
 121
 122         ll_pack_fid(&mdc_fid, st->st_ino, lli->lli_st_generation, S_IFDIR);
 123
 124         offset = (__u64)page->index << CFS_PAGE_SHIFT;
 125         rc = mdc_readpage(sbi->ll_mdc_exp, &mdc_fid,
 126                           offset, page, &request);
 127         if (!rc) {
 128                 body = lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF,
 129                                       sizeof(*body));
 130                 LASSERT(body != NULL);         /* checked by mdc_readpage() */
 131                 /* swabbed by mdc_readpage() */
 132                 LASSERT(lustre_rep_swabbed(request, REPLY_REC_OFF));
 133
 134                 st->st_size = body->size;
 135         } else {
 136                 CERROR("read_dir_page(%ld) error %d\n", page->index, rc);
 137         }
 138         ptlrpc_req_finished(request);
 139         EXIT;
 140
 141         ldlm_lock_decref(&lockh, LCK_CR);
 142         return rc;
 143 }
 144
 145 static struct page *llu_dir_read_page(struct inode *ino, unsigned long pgidx)
 146 {
 147         struct page *page;
 148         int rc;
 149         ENTRY;
 150
 151         OBD_PAGE_ALLOC(page, 0);
 152         if (!page)
 153                 RETURN(ERR_PTR(-ENOMEM));
 154         page->index = pgidx;
 155
 156         rc = llu_dir_do_readpage(ino, page);
 157         if (rc) {
 158                 OBD_PAGE_FREE(page);
 159                 RETURN(ERR_PTR(rc));
 160         }
 161
 162         return page;
 163 }
 164
 165 enum {
 166         EXT2_FT_UNKNOWN,
 167         EXT2_FT_REG_FILE,
 168         EXT2_FT_DIR,
 169         EXT2_FT_CHRDEV,
 170         EXT2_FT_BLKDEV,
 171         EXT2_FT_FIFO,
 172         EXT2_FT_SOCK,
 173         EXT2_FT_SYMLINK,
 174         EXT2_FT_MAX
 175 };
 176
 177 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
 178         [EXT2_FT_UNKNOWN]       DT_UNKNOWN,
 179         [EXT2_FT_REG_FILE]      DT_REG,
 180         [EXT2_FT_DIR]           DT_DIR,
 181         [EXT2_FT_CHRDEV]        DT_CHR,
 182         [EXT2_FT_BLKDEV]        DT_BLK,
 183         [EXT2_FT_FIFO]          DT_FIFO,
 184         [EXT2_FT_SOCK]          DT_SOCK,
 185         [EXT2_FT_SYMLINK]       DT_LNK,
 186 };
 187
 188 #define NAME_OFFSET(de) ((int) ((de)->d_name - (char *) (de)))
 189 #define ROUND_UP64(x)   (((x)+sizeof(__u64)-1) & ~(sizeof(__u64)-1))
 190
 191 static int filldir(char *buf, int buflen,
 192                    const char *name, int namelen, loff_t offset,
 193                    ino_t ino, unsigned int d_type, int *filled)
 194 {
 195         struct dirent64 *dirent = (struct dirent64 *) (buf + *filled);
 196         int reclen = ROUND_UP64(NAME_OFFSET(dirent) + namelen + 1);
 197
 198         /* check overflow */
 199         if ((*filled + reclen) > buflen)
 200                 return 1;
 201
 202         dirent->d_ino = ino;
 203         dirent->d_off = offset;
 204         dirent->d_reclen = reclen;
 205 #ifndef _AIX
 206         dirent->d_type = (unsigned short) d_type;
 207 #endif
 208         memcpy(dirent->d_name, name, namelen);
 209         dirent->d_name[namelen] = 0;
 210
 211         *filled += reclen;
 212
 213         return 0;
 214 }
 215
 216 ssize_t llu_iop_filldirentries(struct inode *ino, _SYSIO_OFF_T *basep,
 217                                char *buf, size_t nbytes)
 218 {
 219         struct llu_inode_info *lli = llu_i2info(ino);
 220         struct intnl_stat *st = llu_i2stat(ino);
 221         loff_t pos = *basep, offset;
 222         int filled = 0;
 223         unsigned long pgidx, maxpages;
 224         ENTRY;
 225
 226         liblustre_wait_event(0);
 227
 228         if (st->st_size == 0) {
 229                 CWARN("dir size is 0?\n");
 230                 RETURN(0);
 231         }
 232
 233         if (pos == -1)
 234                 pos = lli->lli_dir_pos;
 235
 236         maxpages = (st->st_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
 237         pgidx = pos >> CFS_PAGE_SHIFT;
 238         offset = pos & ~CFS_PAGE_MASK;
 239
 240         for ( ; pgidx < maxpages ; pgidx++, offset = 0) {
 241                 struct page *page;
 242                 struct ext2_dirent *de;
 243                 char *addr, *limit;
 244
 245                 page = llu_dir_read_page(ino, pgidx);
 246                 if (IS_ERR(page))
 247                         continue;
 248
 249                 /* size might have been updated by mdc_readpage */
 250                 maxpages = (st->st_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
 251
 252                 /* fill in buffer */
 253                 addr = page->addr;
 254                 limit = addr + CFS_PAGE_SIZE - EXT2_DIR_REC_LEN(1);
 255                 de = (struct ext2_dirent *) (addr + offset);
 256
 257                 for ( ; (char*) de <= limit; de = ext2_next_entry(de)) {
 258                         if (de->inode) {
 259                                 int over;
 260                                 unsigned char d_type = DT_UNKNOWN;
 261
 262                                 if (de->file_type < EXT2_FT_MAX)
 263                                         d_type = ext2_filetype_table[de->file_type];
 264
 265                                 offset = (char*) de - addr;
 266                                 over =  filldir(buf, nbytes, de->name, de->name_len,
 267                                                 (((__u64)pgidx << CFS_PAGE_SHIFT) | offset)
 268                                                 + le16_to_cpu(de->rec_len),
 269                                                 le32_to_cpu(de->inode), d_type, &filled);
 270                                 if (over) {
 271                                         OBD_PAGE_FREE(page);
 272                                         /*
 273                                          * if buffer overflow with no data
 274                                          * returned yet, then report error
 275                                          * instead of eof
 276                                          */
 277                                         if (filled == 0)
 278                                                 RETURN(-EINVAL);
 279
 280                                         GOTO(done, 0);
 281                                 }
 282                         }
 283                 }
 284
 285                 OBD_PAGE_FREE(page);
 286         }
 287 done:
 288         lli->lli_dir_pos = (__u64)pgidx << CFS_PAGE_SHIFT | offset;
 289         *basep = lli->lli_dir_pos;
 290         liblustre_wait_event(0);
 291         RETURN(filled);
 292 }