linux - Are Unix reads and writes to a single file atomically serialized? -

i'd know if writes upon single file done atomically such write("bla bla") , subsequent write("herp derp") same file never results in interleaving, e.g. "bla herp bla derp". assuming these writes happen in different processes or threads, governs gets done first?

also, read() return data reflecting file in state of previous writes completed (whether data has been written disk or not)? example, after write("herp derp"), subsequent reads reflect full data written file, or subsequent read reflect "herp" not "derp" (or reflect none of data @ all)? if reads , writes occur in different processes/threads?

i'm not interested in concurrent file access strategies. want know read , write exactly.

separate write() calls processed separately, not single atomic write transaction, , interleaving entirely possible when multiple processes/threads writing same file. order of actual writes determined schedulers (both kernel process scheduler, , "green" threads thread library's scheduler).

unless specify otherwise (o_direct open flag or similar, if supported), read() , write() operate on kernel buffers , read() use loaded buffer in preference reading disk again.

note may complicated local file buffering; example, stdio , iostreams read file data blocks buffer in process independent of kernel buffers, write() elsewhere data buffered in stdio won't seen. likewise, output buffering there won't actual kernel-level output until output buffer flushed, either automatically because has filled or manually due fflush() or c++'s endl (which implicitly flushes output buffer).