The Weird Way Linux Creates Processes

Comments · 753

• CO

  @CoreDumpped · 8 months ago · pinned

  👉Get Rust training from Let’s Get Rusty: <a href="https://letsgetrusty.com/start-with-jorge">https://letsgetrusty.com/start-with-jorge</a>

  13

• PI

  @pixwarrior · 8 months ago (edited)

  We need to create a child process!<br>Windows: Let&apos;s do that<br>Linux: Mitosis

  2.2K

• MT

  @mtxn · 8 months ago

  fork(), depending on the system configuration, does not copy whole memory when executing new process; it&apos;s using COW (copy on write) approach. When value is changed in the child process it&apos;s just stops pointing to the parent&apos;s one and allocates new one in it&apos;s own address space

  614

• XY

  @XyndraNerd · 8 months ago

  <a href="https://www.youtube.com/watch?v=SwIPOf2YAgI&amp;t=114">1:54</a> lol nice detail on memory size between chrome and firefox

  777

• KO

  @koop_ss · 8 months ago

  It&apos;s crazy how something can look complicated, but as soon as it&apos;s taught in a visual way, it becomes so much easier.<br>This channel is a gold mine, and I really appreciate the hard work you put into your presentations.

  344

• ZE

  @zerto111 · 8 months ago

  The game Factorio can utilize the fork on Linux to save the game without pausing it.<br>On Windows during an autosave the game pauses until the games state is written to disk.<br>On Linux the devs explained they can utilize fork which inherits the memory of the parent to save the game while the main process keeps running.<br>It increases memory usage for a short time. But you get a smoother experience.

  80

• NU

  @nullify. · 8 months ago

  Oh boy, we going to talk about PID 1 next.

  151

• NA

  @nasheethahmeda817 · 8 months ago

  at this point make a playlist for making our own OS by explaining all the implementation of existing OS&apos;s. Really looking forward to it.

  60

• AJ

  @AJMansfield1 · 8 months ago

  Also important to note, fork doesn&apos;t <i>actually</i> copy the underlying memory contents. It might seem like unix is doing redundant work there, especially if it&apos;s going to immediately overwrite the copied process with execve --- but the apparent &quot;copy&quot; is actually just a paper-thin illusion the kernel sets up via MMU shared pages.<br>For read-write memory areas, whichever fork first tries to write to its &quot;copy&quot; first might end up surprised to find that those read-write areas have been marked read-only --- except that the kernel catches that write-protection error before the process can even notice, and quickly makes an <i>actual</i> copy of just the modified page to switcheroo into the faulting process&apos;s page table before anyone notices. (Also, execve is smart enough to unmap and remap the memory pages instead of making the kernel do this switcheroo on every page.)<br>I&apos;ll also note that, to the keen eyed, it might then seem like it&apos;s <i>Windows</i> that ends up stuck making redundant extra copies in memory of the exact same worker process binary, but fortunately Windows has clever filesystem caching and page-mapping tricks of its own: processes running the same binary <i>still</i> end up sharing the same physical memory pages for their read-only program data. The specific runaround needed to make that happen isn&apos;t one I understand quite as well, and I suspect it&apos;s still a bit less efficient than fork; but if you <i>mean</i> to run multiple workers from the same shared program, you&apos;re meant to use CreateThread rather than CreateProcess.<br>(For sake of argument, I am of course ignoring systems with max-paranoia-level mitigations active for things like Spectre / Meltdown. In those configurations, you <i>do</i> just have hard copies.)<br>The real difference here is just that Unix prefers a small number of tightly-optimized abstractions that can handle multiple different use-cases, while Windows tends to provide multiple abstractions that are each more narrowly optimized, and leaves it to the programmer to pick the correct one.

  10

• PH

  @pheww81 · 8 months ago

  I&apos;m surprisingly by the first process being name P1 instead of P0

  179

• XY

  @Xylos144 · 8 months ago (edited)

  <a href="https://www.youtube.com/watch?v=SwIPOf2YAgI&amp;t=1000">16:40</a> in under 20 minutes we reached the genesis of process religion.

  8

• BE

  @beowulf4669 · 8 months ago

  Looking forward to learning about init systems!<br><br>It seems that `fork() + execv()` is more flexible than Windows&apos; approach, as fork can be used for much more than just subsequently spawing a subprocess. <br><br>If subprocess spawning is so frequent however, do you know why Unix systems didn&apos;t decide to provide an &apos;optimized&apos; syscall to skip the process duplication part?

  233