Syslinux - Oct 2009 - [fsc]Adding a generic path_lookup function in VFS

Hi hpa,

I just added the generic path_lookup function for EXTLINUX. hpa, It changed
a lot, so I want hear something from you.

First , here is the main skeleton of the generic path_lookup function:
> if (*name == '/') {
>     inode = this_fs->fs_ops->iget_root();     // The way to get the
root
> inode is different from different fs
>     while(*name == '/')
>         name++;
>     } else {
>         inode = this_inode;                          // pwd
>     }
>     parent = inode;
>
>     while (*name) {
>         p = part;
>         while(*name && *name != '/')
>         *p++ = *name++;
>         *p = '\0';
>         inode = this_fs->fs_ops->iget(part, parent);  // get the
inode
>         if (!inode)
>             goto err;
>         if (inode->mode == I_SYMLINK) {               // handle the
SYMLINK
> file here
>             if (!this_fs->fs_ops->follow_symlink ||
>                 --symlink_count == 0               ||      /* limit check
> */
>                 inode->size >= (uint32_t)inode->blksize)
>                 goto err;
>             name = this_fs->fs_ops->follow_symlink(inode, name);
>             free_inode(inode);
>             continue;
>         }
>
>         if (parent != this_inode)                        // we can't
free
> 'pwd'
>             free_inode(parent);
>         parent = inode;
>         if (! *name)                                           // done
>             break;
>         while(*name == '/')
>             name++;
>     }
>
For I just applied it to EXTLINUX, and I don't want to break the other fs, I
added two union structure in file structure:
> struct file {
>     struct fs_info *fs;
>     union {
>         struct inode *inode;
>         struct open_file_t *open_file;
>     } u1;
>     union {
>         uint32_t offset;
>         uint32_t file_len;
>     } u2;
> };
>
the inode--offset pair is used in the generic path_lookup method, while the
open_file--file_len pair is for the old method. And with the inode-offset
pair, we don't need a more open_file_t structure in all fs more, and I did
remove the open_file_t structure in ext2.c. And as I expected, it did not
break the sys/iso/pxelinux.

So for those haven't applied the generic path_lookup method, I added a
detection in the beginning of searhdir function to make the older one work:

    /* for now, we just applied the universal path_lookup to EXTLINUX
*/
>     if (strcmp(this_fs->fs_ops->fs_name, "ext2") != 0) {
>         file->fs->fs_ops->searchdir(name, file);
>
>         if (file->u1.open_file) {
>             regs->esi.w[0]  = file_to_handle(file);
>             regs->eax.l     = file->u2.file_len;
>             regs->eflags.l &= ~EFLAGS_ZF;
>             return;
>         }
>
>         goto err;
>     }
>
then, followed by the generic path_lookup part.

And here is the inode struct:
> /*
>  * The inode structure, including the detail file information
>  */
> struct inode {
>     int          mode;   /* FILE , DIR or SYMLINK */
>     uint32_t     size;
>     uint32_t     ino;    /* Inode number */
>     uint32_t     atime;  /* Access time */
>     uint32_t     mtime;  /* Modify time */
>     uint32_t     ctime;  /* Create time */
>     uint32_t     dtime;  /* Delete time */
>     int          blocks; /* How many blocks the file take */
>     uint32_t *   data;   /* The block address array where the file stored
> */
>     uint32_t     flags;
>     int          blkbits;
>     int          blksize;
>     uint32_t     file_acl;
> };
>
maybe it's a little complicated, since we didn't need that much filed,
say
the time fields. But I think it would be useful when we want to implement a
fs tool in Syslinux, like the dir or ls command, or maybe even in hdt;)

And the next, I added a pair of *local* malloc-free functions in core
directory. The 'local' means temp, means it would be easily removed when
we
have a powerful memory manage system--- removing the malloc.c in core is OK.

hpa, be honest, I haven't checked the malloc function in com32/lib
carefully, since found it a little more complicated to be a *temp* method.
So, I added a really independent and temp-enough one.


Just as I expected, I removed a lots of 'local variables , structures and
functions' in ext2.c. The open_file_t is gone, and the ext2_searchdir is
gone, too. So, there is no definition of 'global variables' in the
beginning
of ext2.c now.

Anyway, with my limit test, it works.And if you find anything wrong, please
tell me. Thanks.

Last, you can check the commit
here<http://git.zytor.com/?p=users/liu/gsoc09_liu.git;a=commit;h=84b1443afbfe24eb6e338e8636e8e3759b9093d5>
:



-- 
regards
Liu Aleaxander