Does file_write(...) incur I/O???

[nickchen-cpu]

Do we have another I/O when writing to cow file? I mean when file_write(...), why don't we get infinite loop ?

static int __cow_write_data(struct cow_manager *cm, void *buf){
        int ret;
        char *abs_path = NULL;
        int abs_path_len;
        uint64_t curr_size = cm->curr_pos * COW_BLOCK_SIZE;

        if(curr_size >= cm->file_max){
                ret = -EFBIG;

                file_get_absolute_pathname(cm->filp, &abs_path, &abs_path_len);
                if(!abs_path){
                        LOG_ERROR(ret, "cow file max size exceeded (%llu/%llu)", curr_size, cm->file_max);
                }else{
                        LOG_ERROR(ret, "cow file '%s' max size exceeded (%llu/%llu)", abs_path, curr_size, cm->file_max);
                        kfree(abs_path);
                }

                goto error;
        }

        ret = file_write(cm->filp, buf, curr_size, COW_BLOCK_SIZE);
        if(ret) goto error;

        cm->curr_pos++;

        return 0;

error:
        LOG_ERROR(ret, "error writing cow data");
        return ret;
}

[nixomose]

because somewhere in the cow manager (I forget where) it checks to see if the block being written is for the cow file, and if it is, it won't try and cow it.

[nixomose]

from your email:

Hi, I mean every writBig to cow file will incur other I/O on machine, does it make sense?

yes, while in snapshot/cowing mode, every write (well, every unique write within the snapshot) will add two additional I/O's, one to read the block to be cowed, and one to write that block to the cow file.