paoloambrosio / data-14 Goto Github PK

Block file opens in write mode if data has not being backed up yet.


Look at truncate and other operations that don't need an open file.

kmalloc can alloc at max 128KB (that means a bitmap for a 4GB file), we
should use vmalloc for bigger memory areas (or if it fails?).

(kmalloc size limit is KMALLOC_MAX_SIZE defined in linux/slab.h )

Truncate does not work very well.

If ftruncate(fd, length) is used, the message is queued immediately but
there could be writes until the file is closed. Now the file truncate works
only with truncate(path, length).

If a file is truncated and than expanded again, we need to fill the bitmap
or write range. If not, old data will be kept instead of zero-fill. 

Check for vfs function return value on post callbacks. Pass starting point
to write postcall from the precall. 

Move the bitmap to inode private data. The event will be an immediate
event, with less complexity. 

Context should call Event, not Message directly. Then Event will call Data. 

Log levels other than "debug" should go to the system log (prehaps a switch
to chose at compile time where they should go). In Linux there is no need
since syslog is more versatile. 

Log inode creation/modification times after each fs event (i_ctime on
create, i_mtime otherwise) 

Check for errors in filter and assure that every possibile error is handles
correctly. 

Change the range for write messages so that it reports the modified block
(not byte) range. This is needed to map the memory region with file
modifications to file blocks. 

Use bitmap functions defined in linux/bitmap.h