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_LINUX_UNISTD_H
  67 #include <linux/unistd.h>
  68 #elif defined(HAVE_UNISTD_H)
  69 #include <unistd.h>
  70 #endif
  71
  72 #include <dirent.h>
  73
  74 #include "llite_lib.h"
  75
  76 /* (new) readdir implementation overview can be found in lustre/llite/dir.c */
  77
  78 static int llu_dir_do_readpage(struct inode *inode, struct page *page)
  79 {
  80         struct llu_inode_info *lli = llu_i2info(inode);
  81         struct intnl_stat     *st = llu_i2stat(inode);
  82         struct llu_sb_info    *sbi = llu_i2sbi(inode);
  83         struct ptlrpc_request *request;
  84         struct lustre_handle   lockh;
  85         struct mdt_body       *body;
  86         struct lookup_intent   it = { .it_op = IT_READDIR };
  87         struct md_op_data      op_data = {{ 0 }};
  88         ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_UPDATE } };
  89         __u64 offset;
  90         int rc = 0;
  91         ENTRY;
  92
  93         rc = md_lock_match(sbi->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
  94                            &lli->lli_fid, LDLM_IBITS, &policy, LCK_CR, &lockh);
  95         if (!rc) {
  96                 struct ldlm_enqueue_info einfo = {LDLM_IBITS, LCK_CR,
  97                         llu_md_blocking_ast, ldlm_completion_ast, NULL, NULL,
  98                         inode};
  99
 100                 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, 0,
 101                                     LUSTRE_OPC_ANY);
 102
 103                 rc = md_enqueue(sbi->ll_md_exp, &einfo, &it,
 104                                 &op_data, &lockh, NULL, 0, NULL,
 105                                 LDLM_FL_CANCEL_ON_BLOCK);
 106                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 107                 if (request)
 108                         ptlrpc_req_finished(request);
 109                 if (rc < 0) {
 110                         CERROR("lock enqueue: err: %d\n", rc);
 111                         RETURN(rc);
 112                 }
 113         }
 114         ldlm_lock_dump_handle(D_OTHER, &lockh);
 115
 116         offset = (__u64)hash_x_index(page->index);
 117         rc = md_readpage(sbi->ll_md_exp, &lli->lli_fid, NULL,
 118                          offset, page, &request);
 119         if (!rc) {
 120                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 121                 LASSERT(body != NULL);         /* checked by md_readpage() */
 122
 123                 if (body->valid & OBD_MD_FLSIZE)
 124                         st->st_size = body->size;
 125         } else {
 126                 CERROR("read_dir_page(%ld) error %d\n", page->index, rc);
 127         }
 128         ptlrpc_req_finished(request);
 129         EXIT;
 130
 131         ldlm_lock_decref(&lockh, LCK_CR);
 132         return rc;
 133 }
 134
 135 static cfs_page_t *llu_dir_read_page(struct inode *ino, __u64 hash,
 136                                      int exact, struct ll_dir_chain *chain)
 137 {
 138         cfs_page_t *page;
 139         int rc;
 140         ENTRY;
 141
 142         OBD_PAGE_ALLOC(page, 0);
 143         if (!page)
 144                 RETURN(ERR_PTR(-ENOMEM));
 145         page->index = hash_x_index(hash);
 146
 147         rc = llu_dir_do_readpage(ino, page);
 148         if (rc) {
 149                 OBD_PAGE_FREE(page);
 150                 RETURN(ERR_PTR(rc));
 151         }
 152
 153         return page;
 154 }
 155
 156 void *(*memmover)(void *, const void *, size_t) = memmove;
 157
 158 #define NAME_OFFSET(de) ((int) ((de)->d_name - (char *) (de)))
 159 #define ROUND_UP64(x)   (((x)+sizeof(__u64)-1) & ~(sizeof(__u64)-1))
 160 static int filldir(char *buf, int buflen,
 161                    const char *name, int namelen, loff_t offset,
 162                    ino_t ino, unsigned int d_type, int *filled)
 163 {
 164         cfs_dirent_t *dirent = (cfs_dirent_t *) (buf + *filled);
 165         cfs_dirent_t  holder;
 166         int reclen = ROUND_UP64(NAME_OFFSET(dirent) + namelen + 1);
 167
 168         /*
 169          * @buf is not guaranteed to be properly aligned. To work around,
 170          * first fill stack-allocated @holder, then copy @holder into @buf by
 171          * memmove().
 172          */
 173
 174         /* check overflow */
 175         if ((*filled + reclen) > buflen)
 176                 return 1;
 177
 178         holder.d_ino = ino;
 179 #ifdef _DIRENT_HAVE_D_OFF
 180         holder.d_off = offset;
 181 #endif
 182         holder.d_reclen = reclen;
 183 #ifdef _DIRENT_HAVE_D_TYPE
 184         holder.d_type = (unsigned short) d_type;
 185 #endif
 186         /* gcc unrolls memcpy() of structs into field-wise assignments,
 187          * assuming proper alignment. Humor it. */
 188         (*memmover)(dirent, &holder, NAME_OFFSET(dirent));
 189         memcpy(dirent->d_name, name, namelen);
 190         dirent->d_name[namelen] = 0;
 191
 192         *filled += reclen;
 193
 194         return 0;
 195 }
 196
 197 /*
 198  * TODO: much of the code here is similar/identical to llite ll_readdir().
 199  * These code can be factored out and shared in a common module.
 200  */
 201
 202 ssize_t llu_iop_filldirentries(struct inode *dir, _SYSIO_OFF_T *basep,
 203                                char *buf, size_t nbytes)
 204 {
 205         struct llu_inode_info *lli = llu_i2info(dir);
 206         struct intnl_stat     *st = llu_i2stat(dir);
 207         loff_t                 pos = *basep;
 208         struct ll_dir_chain    chain;
 209         cfs_page_t            *page;
 210         int filled = 0;
 211         int rc;
 212         int done;
 213         int shift;
 214         __u16 type;
 215         ENTRY;
 216
 217         liblustre_wait_event(0);
 218
 219         if (st->st_size == 0) {
 220                 CWARN("dir size is 0?\n");
 221                 RETURN(0);
 222         }
 223
 224         if (pos == DIR_END_OFF)
 225                 /*
 226                  * end-of-file.
 227                  */
 228                 RETURN(0);
 229
 230         rc    = 0;
 231         done  = 0;
 232         shift = 0;
 233         ll_dir_chain_init(&chain);
 234
 235         page = llu_dir_read_page(dir, pos, 0, &chain);
 236         while (rc == 0 && !done) {
 237                 struct lu_dirpage *dp;
 238                 struct lu_dirent  *ent;
 239
 240                 if (!IS_ERR(page)) {
 241                         /*
 242                          * If page is empty (end of directoryis reached),
 243                          * use this value.
 244                          */
 245                         __u64 hash = DIR_END_OFF;
 246                         __u64 next;
 247
 248                         dp = page->addr;
 249                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 250                              ent = lu_dirent_next(ent)) {
 251                                 char          *name;
 252                                 int            namelen;
 253                                 struct lu_fid  fid;
 254                                 ino_t          ino;
 255
 256                                 hash    = le64_to_cpu(ent->lde_hash);
 257                                 namelen = le16_to_cpu(ent->lde_namelen);
 258
 259                                 if (hash < pos)
 260                                         /*
 261                                          * Skip until we find target hash
 262                                          * value.
 263                                          */
 264                                         continue;
 265
 266                                 if (namelen == 0)
 267                                         /*
 268                                          * Skip dummy record.
 269                                          */
 270                                         continue;
 271
 272                                 fid  = ent->lde_fid;
 273                                 name = ent->lde_name;
 274                                 fid_le_to_cpu(&fid, &fid);
 275                                 ino  = cl_fid_build_ino(&fid);
 276                                 type = ll_dirent_type_get(ent);
 277                                 done = filldir(buf, nbytes, name, namelen,
 278                                                (loff_t)hash, ino, type,
 279                                                &filled);
 280                         }
 281                         next = le64_to_cpu(dp->ldp_hash_end);
 282                         OBD_PAGE_FREE(page);
 283                         if (!done) {
 284                                 pos = next;
 285                                 if (pos == DIR_END_OFF)
 286                                         /*
 287                                          * End of directory reached.
 288                                          */
 289                                         done = 1;
 290                                 else if (1 /* chain is exhausted*/)
 291                                         /*
 292                                          * Normal case: continue to the next
 293                                          * page.
 294                                          */
 295                                         page = llu_dir_read_page(dir, pos, 1,
 296                                                                &chain);
 297                                 else {
 298                                         /*
 299                                          * go into overflow page.
 300                                          */
 301                                 }
 302                         } else {
 303                                 pos = hash;
 304                                 if (filled == 0)
 305                                         GOTO(out, filled = -EINVAL);
 306                         }
 307                 } else {
 308                         rc = PTR_ERR(page);
 309                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 310                                PFID(&lli->lli_fid), (unsigned long)pos, rc);
 311                 }
 312         }
 313         lli->lli_dir_pos = (loff_t)pos;
 314         *basep = lli->lli_dir_pos;
 315 out:
 316         ll_dir_chain_fini(&chain);
 317         liblustre_wait_event(0);
 318         RETURN(filled);
 319 }