Ok, the fog is starting to clear and I am starting to get my hands around
how inodes work in ocfs. On the current tree, a 2.6 build will:
* never set a new inodes i_ino
* never bind the new inode
The reason inode->i_ino is never initialized is because iget5_locked (unlike
iget4) does not set the i_ino when it creates a new inode, but leaves that
job to the passed in set function.
I've made some considerable changes to fix this, including:
* Added a new ino entry to the find_inode_args type. This allows
the set function helper for iget5_locked to know how to initialize
the i_ino on the new inode
* changed ocfs_read_locked_inode() to take a find_inode_args pointer
instead of a ocfs_file_entry pointer. This allows ocfs_read_locked_inode
to bind new inodes by knowing when not to perform a bind in the case
of revalidating an inode.
* changed ocfs_init_locked_inode (the iget5_locked set function) to
set the new inode i_ino for more then root inodes.
* changed ocfs_iget() to pass in the args.ino instead of the bottom half
of the offset, and then also to just pass ocfs_read_locked_inode the
args and let ocfs_read_locked_inode read the file entry
* various places were calling ocfs_iget when they should be calling
ocfs_get_inode_from_offset() exactly like the 2.4 builds do
None of these changes should effect the 2.4 build, and this patch does
make it possible to mount ocfs2 on a 2.6 box, walk a directory of file
put on the file system from a 2.4 build, and cat the files.
There is still a fatal bug that will hang the system just about anytime
any kind of clean up happens, which makes the 2.6 build still unusable
(but closer to usable.)
--rusty
Index: src/inode.c
==================================================================---
src/inode.c (revision 754)
+++ src/inode.c (working copy)
@@ -317,14 +317,16 @@
* by this point, i_sb, i_dev, i_ino are filled in
*
*/
-void ocfs_read_locked_inode (struct inode *inode, ocfs_file_entry *entry)
+void ocfs_read_locked_inode (struct inode *inode, ocfs_find_inode_args *args)
{
struct super_block *sb;
ocfs_super *osb;
ocfs_inode *newoin;
umode_t mode;
+ __u64 voteoff;
+ ocfs_file_entry *fe = NULL;
- LOG_ENTRY_ARGS ("(0x%08x, 0x%08x)\n", inode, entry);
+ LOG_ENTRY_ARGS ("(%p, %p)\n", inode, args);
if (inode == NULL || inode->i_sb == NULL) {
LOG_ERROR_STR ("bad inode");
@@ -355,26 +357,26 @@
}
LOG_TRACE_ARGS("Populating non-root inode (i_ino = %lu)\n",
inode->i_ino);
-
- if (entry == NULL) {
+ if (args)
+ fe = (ocfs_file_entry *) OCFS_BH_GET_DATA_READ(args->fe_bh);
+ if (!fe) {
make_bad_inode (inode);
goto bail;
}
newoin = NULL;
- mode = entry->prot_bits;
-
- switch (entry->attribs) {
+ mode = fe->prot_bits;
+ switch (fe->attribs) {
case OCFS_ATTRIB_DIRECTORY:
mode |= S_IFDIR;
break;
case OCFS_ATTRIB_CHAR:
- inode->i_rdev = MKDEV (entry->dev_major, entry->dev_minor);
+ inode->i_rdev = MKDEV (fe->dev_major, fe->dev_minor);
mode |= S_IFCHR;
break;
case OCFS_ATTRIB_BLOCK:
- inode->i_rdev = MKDEV (entry->dev_major, entry->dev_minor);
+ inode->i_rdev = MKDEV (fe->dev_major, fe->dev_minor);
mode |= S_IFBLK;
break;
case OCFS_ATTRIB_FIFO:
@@ -391,9 +393,15 @@
mode |= S_IFREG;
break;
}
- ocfs_populate_inode (inode, entry, mode, newoin, false);
+ ocfs_populate_inode (inode, fe, mode, newoin, false);
+ voteoff = S_ISDIR (mode) ? fe->extents[0].disk_off : fe->this_sector;
+ if (!args->skip_bind)
+ ocfs_inode_hash_bind(osb, voteoff, inode);
bail:
+ if (fe)
+ OCFS_BH_PUT_DATA(args->fe_bh);
+
LOG_EXIT ();
return;
} /* ocfs_read_locked_inode */
@@ -578,17 +586,20 @@
* us to sleep.
* return 0 on success, 1 on failure
*/
-int ocfs_init_locked_inode(struct inode * inode, void * opaque) {
+int ocfs_init_locked_inode(struct inode * inode, void * opaque)
+{
+ ocfs_find_inode_args *args = (ocfs_find_inode_args *) opaque;
int ret = 1;
- LOG_ENTRY_ARGS("inode = %x, opaque = %x\n", inode, opaque);
+ LOG_ENTRY_ARGS("inode = %p, opaque = %p\n", inode, opaque);
/* We might be called from ocfs_fill_super in which case, opaque
* will be null (but we'll be looking for the root inode
* anyway... */
- if (opaque == NULL) {
+ if (args == NULL) {
inode->i_ino = OCFS_ROOT_INODE_NUMBER;
+ } else {
+ inode->i_ino = args->ino;
}
-
ret = 0;
//bail:
LOG_EXIT_STATUS(ret);
@@ -605,8 +616,6 @@
struct inode * ocfs_iget(struct super_block *sb, ocfs_find_inode_args *args)
{
struct inode *inode = NULL;
- __u64 off;
- ocfs_file_entry *fe = NULL;
LOG_ENTRY();
@@ -616,13 +625,14 @@
/* if args == NULL, we want the root inode */
if (!args) {
LOG_TRACE_STR("looking for root inode");
- inode = iget5_locked(sb, OCFS_ROOT_INODE_NUMBER, ocfs_find_actor,
ocfs_init_locked_inode, NULL);
+ inode = iget5_locked(sb, OCFS_ROOT_INODE_NUMBER,
+ ocfs_find_actor, ocfs_init_locked_inode,
+ NULL);
} else {
LOG_TRACE_STR("looking for non-root inode");
- off = args->offset;
/* Need to verify that ocfs_find_inode does the right thing,
* and need to actually write ocfs_init_inode */
- inode = iget5_locked(sb, LO (off), ocfs_find_actor,
+ inode = iget5_locked(sb, args->ino, ocfs_find_actor,
ocfs_init_locked_inode, args);
}
@@ -632,11 +642,7 @@
/* inode was *not* in the inode cache */
if (inode->i_state & I_NEW) {
LOG_TRACE_STR("Inode was not in inode cache, reading it.");
- if (args)
- fe = OCFS_BH_GET_DATA_READ(args->fe_bh); /* read */
- ocfs_read_locked_inode(inode, fe);
- if (args)
- OCFS_BH_PUT_DATA(args->fe_bh);
+ ocfs_read_locked_inode(inode, args);
unlock_new_inode(inode);
}
Index: src/dcache.c
==================================================================---
src/dcache.c (revision 754)
+++ src/dcache.c (working copy)
@@ -50,10 +50,9 @@
struct qstr q;
struct buffer_head *fe_bh = NULL;
int needs_trunc;
+ ocfs_find_inode_args args;
#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
int flags = nd->flags;
-#else
- ocfs_find_inode_args args;
#endif
LOG_ENTRY_ARGS ("(0x%08x, %d, '%*s')\n", dentry,
flags,
@@ -134,14 +133,14 @@
LOG_TRACE_STR("found the file entry, but it has been
deleted or renamed!");
ret = 0; /* it is now officially stale :) */
} else {
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
- ocfs_read_locked_inode (inode, fe);
-#else
args.offset = fe->this_sector;
args.fe_bh = fe_bh;
args.skip_bind = 1;
OCFS_BH_PUT_DATA(fe_bh);
fe = NULL;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
+ ocfs_read_locked_inode (inode, &args);
+#else
ocfs_read_inode2 (inode, (void *) &args);
#endif
ret = 1;
Index: src/journal.c
==================================================================---
src/journal.c (revision 754)
+++ src/journal.c (working copy)
@@ -900,11 +900,7 @@
LO(fe->this_sector));
OCFS_BH_PUT_DATA(bh);
fe = NULL;
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
- inode = ocfs_iget (osb->sb, &args);
-#else
inode = ocfs_get_inode_from_offset(osb, args.offset, bh);
-#endif
if (inode == NULL) {
LOG_ERROR_STR("access error");
status = -EACCES;
@@ -1531,11 +1527,7 @@
args.skip_bind = 0;
OCFS_BH_PUT_DATA(bh);
fe = NULL;
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
- inode = ocfs_iget (osb->sb, &args);
-#else
inode = ocfs_get_inode_from_offset(osb, args.offset, bh);
-#endif
if (inode == NULL) {
LOG_ERROR_STR("access error");
status = -EACCES;
Index: src/hash.c
==================================================================--- src/hash.c
(revision 754)
+++ src/hash.c (working copy)
@@ -1714,7 +1714,6 @@
{
struct inode *inode = NULL;
struct super_block *sb = osb->sb;
- unsigned long ino = 0;
__u64 fe_off;
ocfs_file_entry *fe;
ocfs_find_inode_args args;
@@ -1780,12 +1779,12 @@
/* alright, allocate a new inode number for this guy
* and insert it into the hash. It's not bound yet --
* read_inode2 binds the actual inode to it. */
- ino = ocfs_inode_hash_insert(osb, offset, fe_off);
+ args.ino = ocfs_inode_hash_insert(osb, offset, fe_off);
#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
inode = ocfs_iget (sb, &args);
#else
inode - iget4 (sb, ino,
+ iget4 (sb, args.ino,
(find_inode_t) ocfs_find_inode,
(void *) (&args));
#endif
Index: src/inc/ocfs.h
==================================================================---
src/inc/ocfs.h (revision 754)
+++ src/inc/ocfs.h (working copy)
@@ -27,10 +27,6 @@
#ifndef OCFS_H
#define OCFS_H
-/* XXX Hack to avoid warning */
-struct mem_dqinfo;
-extern inline void mark_info_dirty(struct mem_dqinfo *info);
-
/*
** System header files
*/
@@ -49,6 +45,10 @@
#endif
extern inline int generic_fls(int x);
extern inline int get_bitmask_order(unsigned int count);
+
+/* XXX Hack to avoid warning */
+struct mem_dqinfo;
+extern inline void mark_info_dirty(struct mem_dqinfo *info);
#endif /* 2.4 kernel */
#include <linux/module.h>
@@ -2589,6 +2589,7 @@
__u64 offset;
struct buffer_head *fe_bh;
int skip_bind;
+ unsigned long ino;
}
ocfs_find_inode_args;
/* timeout structure taken from Ben's aio.c */
Index: src/inc/proto.h
==================================================================---
src/inc/proto.h (revision 754)
+++ src/inc/proto.h (working copy)
@@ -86,7 +86,7 @@
int ocfs_find_inode (struct inode *inode, unsigned long ino, void *opaque);
void ocfs_populate_inode (struct inode *inode, ocfs_file_entry *fe, umode_t
mode, void *genptr, bool create_ino);
-void ocfs_read_locked_inode (struct inode *inode, ocfs_file_entry *entry);
+void ocfs_read_locked_inode (struct inode *inode, ocfs_find_inode_args *args);
void ocfs_read_inode2 (struct inode *inode, void *opaque);
void ocfs_read_inode (struct inode *inode);
struct inode * ocfs_iget(struct super_block *sb, ocfs_find_inode_args *args);
Index: src/namei.c
==================================================================---
src/namei.c (revision 754)
+++ src/namei.c (working copy)
@@ -74,7 +74,6 @@
struct super_block *sb = dir->i_sb;
struct dentry *ret;
ocfs_super *osb = (ocfs_super *) OCFS_GENERIC_SB_P(sb);
- unsigned long ino;
__u64 inode_off;
LOG_ENTRY_ARGS ("(0x%08x, 0x%08x, '%*s')\n", dir, dentry,
@@ -125,12 +124,12 @@
/* alright, allocate a new inode number for this guy
* and insert it into the hash. It's not bound yet --
* read_inode2 binds the actual inode to it. */
- ino = ocfs_inode_hash_insert(osb, inode_off, fe_off);
+ args.ino = ocfs_inode_hash_insert(osb, inode_off, fe_off);
#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
inode = ocfs_iget (sb, &args);
#else
inode - iget4 (sb, ino,
+ iget4 (sb, args.ino,
(find_inode_t) ocfs_find_inode,
(void *) (&args));
#endif