Whamcloud - gitweb
LU-3321 clio: revert LU-2622 for removing global env list
[fs/lustre-release.git] / lustre / liblustre / dir.c
index a297c9b..c7560e8 100644 (file)
@@ -1,24 +1,41 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
+/*
+ * GPL HEADER START
  *
- * Lustre Light directory handling
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
  *
- *  Copyright (c) 2002-2004 Cluster File Systems, Inc.
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  *
- *   This file is part of Lustre, http://www.lustre.org.
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Use is subject to license terms.
  *
- *   Lustre is free software; you can redistribute it and/or
- *   modify it under the terms of version 2 of the GNU General Public
- *   License as published by the Free Software Foundation.
+ * Copyright (c) 2011, 2012, Intel Corporation.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
  *
- *   Lustre is distributed in the hope that it will be useful,
- *   but WITHOUT ANY WARRANTY; without even the implied warranty of
- *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- *   GNU General Public License for more details.
+ * lustre/liblustre/dir.c
  *
- *   You should have received a copy of the GNU General Public License
- *   along with Lustre; if not, write to the Free Software
- *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * Lustre Light directory handling
  */
 
 #define DEBUG_SUBSYSTEM S_LLITE
 #include <fcntl.h>
 #include <sys/queue.h>
 
-#include <sysio.h>
-#ifdef HAVE_XTIO_H
-#include <xtio.h>
-#endif
-#include <fs.h>
-#include <mount.h>
-#include <inode.h>
-#ifdef HAVE_FILE_H
-#include <file.h>
-#endif
-
-#undef LIST_HEAD
-
-#ifdef HAVE_ASM_TYPES_H
-#include <asm/types.h>
-#elif defined(HAVE_SYS_TYPES_H)
-#include <sys/types.h>
-#endif
-
 #ifdef HAVE_LINUX_UNISTD_H
 #include <linux/unistd.h>
 #elif defined(HAVE_UNISTD_H)
 #include <unistd.h>
 #endif
 
-#include <dirent.h>
 
 #include "llite_lib.h"
+#include <dirent.h>
 
 /* (new) readdir implementation overview can be found in lustre/llite/dir.c */
 
@@ -75,21 +73,23 @@ static int llu_dir_do_readpage(struct inode *inode, struct page *page)
         struct lookup_intent   it = { .it_op = IT_READDIR };
         struct md_op_data      op_data = {{ 0 }};
         ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_UPDATE } };
-        __u64 offset;
         int rc = 0;
         ENTRY;
 
+        llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, 0, LUSTRE_OPC_ANY);
         rc = md_lock_match(sbi->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
                            &lli->lli_fid, LDLM_IBITS, &policy, LCK_CR, &lockh);
         if (!rc) {
-                struct ldlm_enqueue_info einfo = {LDLM_IBITS, LCK_CR,
-                        llu_md_blocking_ast, ldlm_completion_ast, NULL, inode};
-
-                llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, 0,
-                                    LUSTRE_OPC_ANY);
+               struct ldlm_enqueue_info einfo = {
+                       .ei_type        = LDLM_IBITS,
+                       .ei_mode        = LCK_CR,
+                       .ei_cb_bl       = llu_md_blocking_ast,
+                       .ei_cb_cp       = ldlm_completion_ast,
+                       .ei_cbdata      = inode,
+               };
 
                 rc = md_enqueue(sbi->ll_md_exp, &einfo, &it,
-                                &op_data, &lockh, NULL, 0,
+                                &op_data, &lockh, NULL, 0, NULL,
                                 LDLM_FL_CANCEL_ON_BLOCK);
                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
                 if (request)
@@ -101,9 +101,9 @@ static int llu_dir_do_readpage(struct inode *inode, struct page *page)
         }
         ldlm_lock_dump_handle(D_OTHER, &lockh);
 
-        offset = (__u64)hash_x_index(page->index);
-        rc = md_readpage(sbi->ll_md_exp, &lli->lli_fid, NULL,
-                         offset, page, &request);
+        op_data.op_offset = (__u64)hash_x_index(page->index, 0);
+        op_data.op_npages = 1;
+        rc = md_readpage(sbi->ll_md_exp, &op_data, &page, &request);
         if (!rc) {
                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
                 LASSERT(body != NULL);         /* checked by md_readpage() */
@@ -120,17 +120,17 @@ static int llu_dir_do_readpage(struct inode *inode, struct page *page)
         return rc;
 }
 
-static struct page *llu_dir_read_page(struct inode *ino, __u32 hash,
-                                      int exact, struct ll_dir_chain *chain)
+static struct page *llu_dir_read_page(struct inode *ino, __u64 hash,
+                                     int exact, struct ll_dir_chain *chain)
 {
-        struct page *page;
+       struct page *page;
         int rc;
         ENTRY;
 
         OBD_PAGE_ALLOC(page, 0);
         if (!page)
                 RETURN(ERR_PTR(-ENOMEM));
-        page->index = hash_x_index(hash);
+        page->index = hash_x_index(hash, 0);
 
         rc = llu_dir_do_readpage(ino, page);
         if (rc) {
@@ -141,48 +141,38 @@ static struct page *llu_dir_read_page(struct inode *ino, __u32 hash,
         return page;
 }
 
-enum {
-        EXT2_FT_UNKNOWN,
-        EXT2_FT_REG_FILE,
-        EXT2_FT_DIR,
-        EXT2_FT_CHRDEV,
-        EXT2_FT_BLKDEV,
-        EXT2_FT_FIFO,
-        EXT2_FT_SOCK,
-        EXT2_FT_SYMLINK,
-        EXT2_FT_MAX
-};
-
-static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
-        [EXT2_FT_UNKNOWN]       DT_UNKNOWN,
-        [EXT2_FT_REG_FILE]      DT_REG,
-        [EXT2_FT_DIR]           DT_DIR,
-        [EXT2_FT_CHRDEV]        DT_CHR,
-        [EXT2_FT_BLKDEV]        DT_BLK,
-        [EXT2_FT_FIFO]          DT_FIFO,
-        [EXT2_FT_SOCK]          DT_SOCK,
-        [EXT2_FT_SYMLINK]       DT_LNK,
-};
+void *(*memmover)(void *, const void *, size_t) = memmove;
 
 #define NAME_OFFSET(de) ((int) ((de)->d_name - (char *) (de)))
 #define ROUND_UP64(x)   (((x)+sizeof(__u64)-1) & ~(sizeof(__u64)-1))
-static int filldir(char *buf, int buflen,
-                   const char *name, int namelen, loff_t offset,
-                   ino_t ino, unsigned int d_type, int *filled)
+static int filldir(char *buf, int buflen, const char *name, int namelen,
+                  loff_t offset, ino_t ino, unsigned int d_type, int *filled)
 {
-        struct dirent64 *dirent = (struct dirent64 *) (buf + *filled);
-        int reclen = ROUND_UP64(NAME_OFFSET(dirent) + namelen + 1);
+       struct dirent64 *dirent = (struct dirent64 *)(buf + *filled);
+       struct dirent64  holder;
+       int reclen = ROUND_UP64(NAME_OFFSET(dirent) + namelen + 1);
+
+        /*
+         * @buf is not guaranteed to be properly aligned. To work around,
+         * first fill stack-allocated @holder, then copy @holder into @buf by
+         * memmove().
+         */
 
         /* check overflow */
         if ((*filled + reclen) > buflen)
                 return 1;
 
-        dirent->d_ino = ino;
-        dirent->d_off = offset;
-        dirent->d_reclen = reclen;
-#ifndef _AIX
-        dirent->d_type = (unsigned short) d_type;
+        holder.d_ino = ino;
+#ifdef _DIRENT_HAVE_D_OFF
+        holder.d_off = offset;
 #endif
+        holder.d_reclen = reclen;
+#ifdef _DIRENT_HAVE_D_TYPE
+        holder.d_type = (unsigned short) d_type;
+#endif
+        /* gcc unrolls memcpy() of structs into field-wise assignments,
+         * assuming proper alignment. Humor it. */
+        (*memmover)(dirent, &holder, NAME_OFFSET(dirent));
         memcpy(dirent->d_name, name, namelen);
         dirent->d_name[namelen] = 0;
 
@@ -191,23 +181,23 @@ static int filldir(char *buf, int buflen,
         return 0;
 }
 
-/* 
+/*
  * TODO: much of the code here is similar/identical to llite ll_readdir().
  * These code can be factored out and shared in a common module.
  */
 
-ssize_t llu_iop_filldirentries(struct inode *dir, _SYSIO_OFF_T *basep, 
+ssize_t llu_iop_filldirentries(struct inode *dir, _SYSIO_OFF_T *basep,
                               char *buf, size_t nbytes)
 {
         struct llu_inode_info *lli = llu_i2info(dir);
         struct intnl_stat     *st = llu_i2stat(dir);
         loff_t                 pos = *basep;
         struct ll_dir_chain    chain;
-        struct page           *page;
+       struct page            *page;
         int filled = 0;
         int rc;
         int done;
-        int shift;
+        __u16 type;
         ENTRY;
 
         liblustre_wait_event(0);
@@ -217,7 +207,7 @@ ssize_t llu_iop_filldirentries(struct inode *dir, _SYSIO_OFF_T *basep,
                 RETURN(0);
         }
 
-        if (pos == DIR_END_OFF)
+        if (pos == MDS_DIR_END_OFF)
                 /*
                  * end-of-file.
                  */
@@ -225,7 +215,6 @@ ssize_t llu_iop_filldirentries(struct inode *dir, _SYSIO_OFF_T *basep,
 
         rc    = 0;
         done  = 0;
-        shift = 0;
         ll_dir_chain_init(&chain);
 
         page = llu_dir_read_page(dir, pos, 0, &chain);
@@ -234,12 +223,12 @@ ssize_t llu_iop_filldirentries(struct inode *dir, _SYSIO_OFF_T *basep,
                 struct lu_dirent  *ent;
 
                 if (!IS_ERR(page)) {
-                        /* 
+                        /*
                          * If page is empty (end of directoryis reached),
-                         * use this value. 
+                         * use this value.
                          */
-                        __u32 hash = DIR_END_OFF;
-                        __u32 next;
+                        __u64 hash = MDS_DIR_END_OFF;
+                        __u64 next;
 
                         dp = page->addr;
                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
@@ -247,9 +236,9 @@ ssize_t llu_iop_filldirentries(struct inode *dir, _SYSIO_OFF_T *basep,
                                 char          *name;
                                 int            namelen;
                                 struct lu_fid  fid;
-                                ino_t          ino;
+                                __u64          ino;
 
-                                hash    = le32_to_cpu(ent->lde_hash);
+                                hash    = le64_to_cpu(ent->lde_hash);
                                 namelen = le16_to_cpu(ent->lde_namelen);
 
                                 if (hash < pos)
@@ -268,17 +257,17 @@ ssize_t llu_iop_filldirentries(struct inode *dir, _SYSIO_OFF_T *basep,
                                 fid  = ent->lde_fid;
                                 name = ent->lde_name;
                                 fid_le_to_cpu(&fid, &fid);
-                                ino  = llu_fid_build_ino(llu_i2sbi(dir), &fid);
-
+                                ino  = cl_fid_build_ino(&fid, 0);
+                                type = ll_dirent_type_get(ent);
                                 done = filldir(buf, nbytes, name, namelen,
-                                               (loff_t)hash, ino, DT_UNKNOWN,
+                                               (loff_t)hash, ino, type,
                                                &filled);
                         }
-                        next = le32_to_cpu(dp->ldp_hash_end);
+                        next = le64_to_cpu(dp->ldp_hash_end);
                         OBD_PAGE_FREE(page);
                         if (!done) {
                                 pos = next;
-                                if (pos == DIR_END_OFF)
+                                if (pos == MDS_DIR_END_OFF)
                                         /*
                                          * End of directory reached.
                                          */
@@ -306,7 +295,7 @@ ssize_t llu_iop_filldirentries(struct inode *dir, _SYSIO_OFF_T *basep,
                                PFID(&lli->lli_fid), (unsigned long)pos, rc);
                 }
         }
-        lli->lli_dir_pos = (loff_t)(__s32)pos;
+        lli->lli_dir_pos = (loff_t)pos;
         *basep = lli->lli_dir_pos;
 out:
         ll_dir_chain_fini(&chain);