- 21 Jan, 2020 40 commits
-
-
Russell King authored
Avoid using the inode number as the indirect disc address, even though these currently have the same value. Signed-off-by:
Russell King <rmk+kernel@armlinux.org.uk> Signed-off-by:
Al Viro <viro@zeniv.linux.org.uk>
-
Russell King authored
Add support for ADFS E and E+ floppy image formats, which, unlike their hard disk variants, do not have a filesystem boot block - they have a single map zone, with the map fragment stored at sector 0. Signed-off-by:
-
Russell King authored
Separate the filesystem block probing from the superblock filling so we can support other ADFS filesystem formats, such as the single-zone E and E+ floppy image formats which do not have a boot block. Signed-off-by:
-
Russell King authored
Remove the noisy debug in adfs_dir_update(). Signed-off-by:
-
Russell King authored
When we have write support enabled, we must not drop inodes before they have been written back, otherwise we lose updates to the filesystem on umount. Keep the inodes around unless we are built in read-only mode, or we are mounted read-only. Signed-off-by:
-
Russell King authored
Implement big directory entry update support in the same way that we do for new directories. Signed-off-by:
-
Russell King authored
When reading a big directory, calculate the validate the directory checkbyte to ensure that the directory contents are valid. Signed-off-by:
-
Russell King authored
Strengthen the directory validation by ensuring that the header fields contain sensible values that fit inside the directory, and limit the directory size to 4MB as per RISC OS requirements. Signed-off-by:
-
Russell King authored
Extract the directory validation from the directory reading function as we will want to re-use this code. Signed-off-by:
-
Russell King authored
Factor out the directory entry byte offset calculation. Signed-off-by:
-
Russell King authored
After changing a directory, we need to update the sequence numbers and calculate the new check byte before the directory is scheduled to be written back to the media. Since this needs to happen for any change to the directory, move this into a separate method. Signed-off-by:
-
Russell King authored
__adfs_dir_put() and adfs_dir_find_entry() are only called from adfs_f_update(), so move them into this function, removing some unnecessary entry copying by doing so. Signed-off-by:
-
Russell King authored
adfs_dir_read() is only called from adfs_f_read(), so merge it into that function. As new directories are always 2048 bytes in size, (which we rely on elsewhere) we can consolidate some of the code. Signed-off-by:
-
Russell King authored
Check that the lastmask and reserved fields are all zero, as per the documentation. Signed-off-by:
-
Russell King authored
We have two locations where we validate the new directory format, so factor this out to a helper. Signed-off-by:
-
Russell King authored
Add and use pointers in the adfs_dir structure to access the directory head and tail structures, which will always be contiguous in a buffer. This allows us to avoid memcpy()ing the data in the new directory code, making it slightly more efficient. Signed-off-by:
-
Russell King authored
Rather than using setpos + getnext to iterate through the directory entries, pass iterate() down to the dir format code to populate the dirents. Signed-off-by:
-
Russell King authored
There is nothing in our readdir (aka iterate) method that relies on the directory inode being exclusively locked, so switch to using the iterate_shared() hook rather than iterate(). Signed-off-by:
-
Russell King authored
Get rid of the ifdef, using IS_ENABLED() instead to detect whether the code should be callable. This allows the compiler to always parse the following code, reducing the chances of errors being missed. Signed-off-by:
-
Russell King authored
When we update a directory, a number of errors may happen. If we failed to find the entry to update, we can just release the directory buffers as normal. However, if we have some other error, we may have partially updated the buffers, resulting in an invalid directory. In this case, we need to discard the buffers to avoid writing the contents back to the media, and later re-read the directory from the media. Signed-off-by:
-
Russell King authored
Use __u8 and pack the structures for on-disk directories. Signed-off-by:
-
Russell King authored
Update directory locking such that it covers the validation of the directory, which could fail if another thread is concurrently writing to the same directory. Since we may sleep, we need to use a rwsem rather than a rw spinlock. Signed-off-by:
-
Russell King authored
Provide a helper for marking directory buffers dirty so they get written back to disk. Signed-off-by:
-
Russell King authored
Add a helper to read a directory using the inode, which we do in two places. Signed-off-by:
-
Russell King authored
Both directory formats code the mechanics of fetching the directory buffers using their own implementations. Consolidate these into one implementation. Signed-off-by:
-
Russell King authored
Directories can span multiple buffers, and we currently open-code memcpy access to these buffers, including dealing with entries that are split across multiple buffers. Such code exists in both directory format implementations. Provide common functions to allow data to be copied from/to the directory buffers as if they were a contiguous set of buffers, and use them when accessing directories. Signed-off-by:
-
Russell King authored
adfs_fplus_sync() can be used for both directory formats since we now have a common way to access the buffer heads, so move it into dir.c and appropriately rename it. Remove the directory-format specific implementations. Signed-off-by:
-
Russell King authored
With the bhs pointer in place, we have no need for separate per-format free() methods, since a generic version will do. Provide a generic implementation, remove the format specific implementations and the method function pointer. Signed-off-by:
-
Russell King authored
Initialise the dir object before we pass it down to the directory format specific read handler. This allows us to get rid of the initialisation inside those handlers. Signed-off-by:
-
Russell King authored
Rename bh_fplus to bhs in preparation to make some of the directory handling code sharable between implementations. Signed-off-by:
-
Russell King authored
When scanning the map for a fragment id, we need to keep track of the free space links, so we don't inadvertently believe that the freespace link is a valid fragment id. Signed-off-by:
-
Russell King authored
Move map specific superblock initialisation to map.c, rather than having it spread into super.c. Signed-off-by:
-
Russell King authored
Use find_next_bit_le() to find the end of a fragment in the map rather than open-coding this functionality. Signed-off-by:
-
Russell King authored
lookup_zone() and scan_free_map() cope in different ways with the location of the map data within a zone: 1. lookup_zone() adds a four byte offset to the map data pointer to skip over the check and free link bytes. 2. scan_free_map() needs to use the free link pointer, which is an offset from itself, so we end up adding a 32-bit offset to the end pointer (aka mapsize) which is really confusing. Rename mapsize to endbit as this is really what it is, and incorporate the 32-bit offset into the map layout. This means that both dm_startbit and dm_endbit are now bit offsets from the start of the buffer, rather than four bytes in to the buffer. Signed-off-by:
-
Russell King authored
We have several places which deal with releasing the map buffers and freeing the map array. Provide a helper for this. Signed-off-by:
-
Russell King authored
Split up adfs_read_map() into separate helpers to layout the map, read the map, and release the map buffers. Signed-off-by:
-
Russell King authored
adfs_map_free() is not obvious whether it is freeing the map or returning the number of free blocks on the filesystem. Rename it to the more generic statfs() to make it clear that it's a statistic function. Signed-off-by:
-
Russell King authored
Keep all the map code together in map.c, rather than having some in super.c Signed-off-by:
-
Russell King authored
Fix adfs_mode2atts() to actually update the file permissions on the media rather than using the current inode mode. Note also that directories do not have read/write permissions stored on the media. Signed-off-by:
-
Russell King authored
Despite ADFS timestamps having centi-second granularity, and Linux gaining fine-grained timestamp support in v2.5.48, fs/adfs was never updated. Update fs/adfs to centi-second support, and ensure that the inode ctime always reflects what is written in underlying media. Signed-off-by:
-
