inode.c

ocfs1.4.1 oracle分布式文件系统

/* -*- mode: c; c-basic-offset: 8; -*-
 * vim: noexpandtab sw=8 ts=8 sts=0:
 *
 * inode.c
 *
 * vfs' aops, fops, dops and iops
 *
 * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License as published by the Free Software Foundation; either
 * version 2 of the License, or (at your option) any later version.
 *
 * 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 for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 */

#include <linux/fs.h>
#include <linux/types.h>
#include <linux/slab.h>
#include <linux/highmem.h>
#include <linux/pagemap.h>

#include <asm/byteorder.h>

#define MLOG_MASK_PREFIX ML_INODE
#include <cluster/masklog.h>

#include "ocfs2.h"

#include "alloc.h"
#include "dlmglue.h"
#include "extent_map.h"
#include "file.h"
#include "heartbeat.h"
#include "inode.h"
#include "journal.h"
#include "namei.h"
#include "suballoc.h"
#include "super.h"
#include "symlink.h"
#include "sysfile.h"
#include "uptodate.h"

#include "buffer_head_io.h"

struct ocfs2_find_inode_args
{
	u64		fi_blkno;
	unsigned long	fi_ino;
	unsigned int	fi_flags;
	unsigned int	fi_sysfile_type;
};

#ifndef NO_LOCKDEP
static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
#endif

static int ocfs2_read_locked_inode(struct inode *inode,
				   struct ocfs2_find_inode_args *args);
static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
static int ocfs2_find_actor(struct inode *inode, void *opaque);
static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
				    struct inode *inode,
				    struct buffer_head *fe_bh);

void ocfs2_set_inode_flags(struct inode *inode)
{
	unsigned int flags = OCFS2_I(inode)->ip_attr;

	inode->i_flags &= ~(S_IMMUTABLE |
		S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);

	if (flags & OCFS2_IMMUTABLE_FL)
		inode->i_flags |= S_IMMUTABLE;

	if (flags & OCFS2_SYNC_FL)
		inode->i_flags |= S_SYNC;
	if (flags & OCFS2_APPEND_FL)
		inode->i_flags |= S_APPEND;
	if (flags & OCFS2_NOATIME_FL)
		inode->i_flags |= S_NOATIME;
	if (flags & OCFS2_DIRSYNC_FL)
		inode->i_flags |= S_DIRSYNC;
}

/* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
{
	unsigned int flags = oi->vfs_inode.i_flags;

	oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
			OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
	if (flags & S_SYNC)
		oi->ip_attr |= OCFS2_SYNC_FL;
	if (flags & S_APPEND)
		oi->ip_attr |= OCFS2_APPEND_FL;
	if (flags & S_IMMUTABLE)
		oi->ip_attr |= OCFS2_IMMUTABLE_FL;
	if (flags & S_NOATIME)
		oi->ip_attr |= OCFS2_NOATIME_FL;
	if (flags & S_DIRSYNC)
		oi->ip_attr |= OCFS2_DIRSYNC_FL;
}

struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
			 int sysfile_type)
{
	struct inode *inode = NULL;
	struct super_block *sb = osb->sb;
	struct ocfs2_find_inode_args args;

	mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);

	/* Ok. By now we've either got the offsets passed to us by the
	 * caller, or we just pulled them off the bh. Lets do some
	 * sanity checks to make sure they're OK. */
	if (blkno == 0) {
		inode = ERR_PTR(-EINVAL);
		mlog_errno(PTR_ERR(inode));
		goto bail;
	}

	args.fi_blkno = blkno;
	args.fi_flags = flags;
	args.fi_ino = ino_from_blkno(sb, blkno);
	args.fi_sysfile_type = sysfile_type;

	inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
			     ocfs2_init_locked_inode, &args);
	/* inode was *not* in the inode cache. 2.6.x requires
	 * us to do our own read_inode call and unlock it
	 * afterwards. */
	if (inode && inode->i_state & I_NEW) {
		mlog(0, "Inode was not in inode cache, reading it.\n");
		ocfs2_read_locked_inode(inode, &args);
		unlock_new_inode(inode);
	}
	if (inode == NULL) {
		inode = ERR_PTR(-ENOMEM);
		mlog_errno(PTR_ERR(inode));
		goto bail;
	}
	if (is_bad_inode(inode)) {
		iput(inode);
		inode = ERR_PTR(-ESTALE);
		goto bail;
	}

bail:
	if (!IS_ERR(inode)) {
		mlog(0, "returning inode with number %llu\n",
		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
		mlog_exit_ptr(inode);
	}

	return inode;
}


/*
 * here's how inodes get read from disk:
 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
 * found? : return the in-memory inode
 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
 */

static int ocfs2_find_actor(struct inode *inode, void *opaque)
{
	struct ocfs2_find_inode_args *args = NULL;
	struct ocfs2_inode_info *oi = OCFS2_I(inode);
	int ret = 0;

	mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);

	args = opaque;

	mlog_bug_on_msg(!inode, "No inode in find actor!\n");

	if (oi->ip_blkno != args->fi_blkno)
		goto bail;

	ret = 1;
bail:
	mlog_exit(ret);
	return ret;
}

/*
 * initialize the new inode, but don't do anything that would cause
 * us to sleep.
 * return 0 on success, 1 on failure
 */
static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
{
	struct ocfs2_find_inode_args *args = opaque;

	mlog_entry("inode = %p, opaque = %p\n", inode, opaque);

	inode->i_ino = args->fi_ino;
	OCFS2_I(inode)->ip_blkno = args->fi_blkno;
#ifndef NO_LOCKDEP
	if (args->fi_sysfile_type != 0)
		lockdep_set_class(&inode->i_mutex,
			&ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
#endif

	mlog_exit(0);
	return 0;
}

int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
		     	 int create_ino)
{
	struct super_block *sb;
	struct ocfs2_super *osb;
	int status = -EINVAL;

	mlog_entry("(0x%p, size:%llu)\n", inode,
		   (unsigned long long)le64_to_cpu(fe->i_size));

	sb = inode->i_sb;
	osb = OCFS2_SB(sb);

	/* this means that read_inode cannot create a superblock inode
	 * today.  change if needed. */
	if (!OCFS2_IS_VALID_DINODE(fe) ||
	    !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
		mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
		     "signature = %.*s, flags = 0x%x\n",
		     inode->i_ino,
		     (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
		     fe->i_signature, le32_to_cpu(fe->i_flags));
		goto bail;
	}

	if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) {
		mlog(ML_ERROR, "file entry generation does not match "
		     "superblock! osb->fs_generation=%x, "
		     "fe->i_fs_generation=%x\n",
		     osb->fs_generation, le32_to_cpu(fe->i_fs_generation));
		goto bail;
	}

	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
	OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
	OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);

	inode->i_version = 1;
	inode->i_generation = le32_to_cpu(fe->i_generation);
	inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
	inode->i_mode = le16_to_cpu(fe->i_mode);
	inode->i_uid = le32_to_cpu(fe->i_uid);
	inode->i_gid = le32_to_cpu(fe->i_gid);

	/* Fast symlinks will have i_size but no allocated clusters. */
	if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
		inode->i_blocks = 0;
	else
		inode->i_blocks = ocfs2_inode_sector_count(inode);
	inode->i_mapping->a_ops = &ocfs2_aops;
	inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
	inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
	inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
	inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
	inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
	inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);

	if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
		mlog(ML_ERROR,
		     "ip_blkno %llu != i_blkno %llu!\n",
		     (unsigned long long)OCFS2_I(inode)->ip_blkno,
		     (unsigned long long)le64_to_cpu(fe->i_blkno));

	inode->i_nlink = le16_to_cpu(fe->i_links_count);

	if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;

	if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
		mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
	} else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
	} else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
		mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
		/* we can't actually hit this as read_inode can't
		 * handle superblocks today ;-) */
		BUG();
	}

	switch (inode->i_mode & S_IFMT) {
	    case S_IFREG:
		    inode->i_fop = &ocfs2_fops;
		    inode->i_op = &ocfs2_file_iops;
		    i_size_write(inode, le64_to_cpu(fe->i_size));
		    break;
	    case S_IFDIR:
		    inode->i_op = &ocfs2_dir_iops;
		    inode->i_fop = &ocfs2_dops;
		    i_size_write(inode, le64_to_cpu(fe->i_size));
		    break;
	    case S_IFLNK:
		    if (ocfs2_inode_is_fast_symlink(inode))
			inode->i_op = &ocfs2_fast_symlink_inode_operations;
		    else
			inode->i_op = &ocfs2_symlink_inode_operations;
		    i_size_write(inode, le64_to_cpu(fe->i_size));
		    break;
	    default:
		    inode->i_op = &ocfs2_special_file_iops;
		    init_special_inode(inode, inode->i_mode,
				       inode->i_rdev);
		    break;
	}

	if (create_ino) {
		inode->i_ino = ino_from_blkno(inode->i_sb,
			       le64_to_cpu(fe->i_blkno));

		/*
		 * If we ever want to create system files from kernel,
		 * the generation argument to
		 * ocfs2_inode_lock_res_init() will have to change.
		 */
		BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);

		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
					  OCFS2_LOCK_TYPE_META, 0, inode);

		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
					  OCFS2_LOCK_TYPE_OPEN, 0, inode);
	}

	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
				  OCFS2_LOCK_TYPE_RW, inode->i_generation,
				  inode);

	ocfs2_set_inode_flags(inode);

	status = 0;
bail:
	mlog_exit(status);
	return status;
}

static int ocfs2_read_locked_inode(struct inode *inode,
				   struct ocfs2_find_inode_args *args)
{
	struct super_block *sb;
	struct ocfs2_super *osb;
	struct ocfs2_dinode *fe;
	struct buffer_head *bh = NULL;
	int status, can_lock;
	u32 generation = 0;

	mlog_entry("(0x%p, 0x%p)\n", inode, args);

	status = -EINVAL;
	if (inode == NULL || inode->i_sb == NULL) {
		mlog(ML_ERROR, "bad inode\n");
		return status;
	}
	sb = inode->i_sb;
	osb = OCFS2_SB(sb);

	if (!args) {
		mlog(ML_ERROR, "bad inode args\n");
		make_bad_inode(inode);
		return status;
	}

	/*
	 * To improve performance of cold-cache inode stats, we take
	 * the cluster lock here if possible.
	 *
	 * Generally, OCFS2 never trusts the contents of an inode
	 * unless it's holding a cluster lock, so taking it here isn't
	 * a correctness issue as much as it is a performance
	 * improvement.
	 *
	 * There are three times when taking the lock is not a good idea:
	 *
	 * 1) During startup, before we have initialized the DLM.
	 *
	 * 2) If we are reading certain system files which never get
	 *    cluster locks (local alloc, truncate log).
	 *
	 * 3) If the process doing the iget() is responsible for
	 *    orphan dir recovery. We're holding the orphan dir lock and
	 *    can get into a deadlock with another process on another
	 *    node in ->delete_inode().
	 *
	 * #1 and #2 can be simply solved by never taking the lock
	 * here for system files (which are the only type we read
	 * during mount). It's a heavier approach, but our main
	 * concern is user-accesible files anyway.
	 *
	 * #3 works itself out because we'll eventually take the
	 * cluster lock before trusting anything anyway.
	 */
	can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
		&& !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
		&& !ocfs2_mount_local(osb);

	/*
	 * To maintain backwards compatibility with older versions of
	 * ocfs2-tools, we still store the generation value for system
	 * files. The only ones that actually matter to userspace are
	 * the journals, but it's easier and inexpensive to just flag
	 * all system files similarly.
	 */
	if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
		generation = osb->fs_generation;

	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
				  OCFS2_LOCK_TYPE_META,
				  generation, inode);

	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
				  OCFS2_LOCK_TYPE_OPEN,
				  0, inode);