CINXE.COM

<?xml version="1.0" encoding="UTF-8" standalone="yes"?> <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"><html xmlns="http://www.w3.org/1999/xhtml"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8" /><title>LKML: Jan Hudec: Re: filesystem transactions API</title><link href="/css/message.css" rel="stylesheet" type="text/css" /><link href="/css/wrap.css" rel="alternate stylesheet" type="text/css" title="wrap" /><link href="/css/nowrap.css" rel="stylesheet" type="text/css" title="nowrap" /><link href="/favicon.ico" rel="shortcut icon" /><script src="/js/simple-calendar.js" type="text/javascript"></script><script src="/js/styleswitcher.js" type="text/javascript"></script><link rel="alternate" type="application/rss+xml" title="lkml.org : last 100 messages" href="/rss.php" /><link rel="alternate" type="application/rss+xml" title="lkml.org : last messages by Jan Hudec" href="/groupie.php?aid=2647" /><script> var _paq = window._paq = window._paq || []; /* tracker methods like "setCustomDimension" should be called before "trackPageView" */ _paq.push(["setDoNotTrack", true]); _paq.push(["disableCookies"]); _paq.push(['trackPageView']); _paq.push(['enableLinkTracking']); (function() { var u="//m.lkml.org/"; _paq.push(['setTrackerUrl', u+'matomo.php']); _paq.push(['setSiteId', '1']); var d=document, g=d.createElement('script'), s=d.getElementsByTagName('script')[0]; g.async=true; g.src=u+'matomo.js'; s.parentNode.insertBefore(g,s); })(); </script></head><body onload="es.jasper.simpleCalendar.init();" itemscope="itemscope" itemtype="http://schema.org/BlogPosting"><table border="0" cellpadding="0" cellspacing="0"><tr><td width="180" align="center"><a href="/"><img style="border:0;width:135px;height:32px" src="/images/toprowlk.gif" alt="lkml.org" /></a></td><td width="32">聽</td><td class="nb"><div><a class="nb" href="/lkml"> [lkml]</a> 聽 <a class="nb" href="/lkml/2005"> [2005]</a> 聽 <a class="nb" href="/lkml/2005/4"> [Apr]</a> 聽 <a class="nb" href="/lkml/2005/4/27"> [27]</a> 聽 <a class="nb" href="/lkml/last100"> [last100]</a> 聽 <a href="/rss.php"><img src="/images/rss-or.gif" border="0" alt="RSS Feed" /></a></div><div>Views: <a href="#" class="nowrap" onclick="setActiveStyleSheet('wrap');return false;">[wrap]</a><a href="#" class="wrap" onclick="setActiveStyleSheet('nowrap');return false;">[no wrap]</a> 聽 <a class="nb" href="/lkml/mheaders/2005/4/27/46" onclick="this.href='/lkml/headers'+'/2005/4/27/46';">[headers]</a>聽 <a href="/lkml/bounce/2005/4/27/46">[forward]</a>聽 </div></td><td width="32">聽</td></tr><tr><td valign="top"><div class="es-jasper-simpleCalendar" baseurl="/lkml/"></div><div class="threadlist">Messages in this thread</div><ul class="threadlist"><li class="root"><a href="/lkml/2005/4/24/76">First message in thread</a></li><li><a href="/lkml/2005/4/26/172">"John Stoffel"</a><ul><li><a href="/lkml/2005/4/26/176">Lars Marowsky-Bree</a></li><li><a href="/lkml/2005/4/26/179">Trond Myklebust</a><ul><li><a href="/lkml/2005/4/26/192">Jamie Lokier</a><ul><li><a href="/lkml/2005/4/26/194">"Artem B. Bityuckiy"</a><ul><li><a href="/lkml/2005/4/26/197">Jamie Lokier</a></li></ul></li></ul></li></ul></li><li><a href="/lkml/2005/4/26/181">Jamie Lokier</a><ul><li><a href="/lkml/2005/4/26/229">Diego Calleja</a><ul><li><a href="/lkml/2005/4/26/240">Jamie Lokier</a></li></ul></li><li class="origin"><a href="/lkml/2005/4/27/94">Jan Hudec</a><ul><li><a href="/lkml/2005/4/27/94">Ville Herva</a><ul><li><a href="/lkml/2005/4/27/126">Jamie Lokier</a></li></ul></li></ul></li></ul></li></ul></li></ul></td><td width="32" rowspan="2" class="c" valign="top"><img src="/images/icornerl.gif" width="32" height="32" alt="/" /></td><td class="c" rowspan="2" valign="top" style="padding-top: 1em"><table><tr><td><table><tr><td class="lp">Date</td><td class="rp" itemprop="datePublished">Wed, 27 Apr 2005 11:34:12 +0200</td></tr><tr><td class="lp">From</td><td class="rp" itemprop="author">Jan Hudec <></td></tr><tr><td class="lp">Subject</td><td class="rp" itemprop="name">Re: filesystem transactions API</td></tr></table></td><td></td></tr></table><pre itemprop="articleBody">On Tue, Apr 26, 2005 at 16:24:34 +0100, Jamie Lokier wrote:<br />> John Stoffel wrote:<br />> > >>>>> "Jamie" == Jamie Lokier <jamie@shareable.org> writes:<br />> > <br />> > Jamie> No. A transaction means that _all_ processes will see the<br />> > Jamie> whole transaction or not.<br />> > <br />> > This is really hard. How do you handle the case where process X<br />> > starts a transaction modifies files a, b & c, but process Y has file b<br />> > open for writing, and never lets it go? Or the file gets unlinked? <br />> <br />> Then it starts to depend on what kind of transactions you want to<br />> implement.<br />> <br />> You can say that a transaction isn't allowed when a process has one of<br />> the files opened for writing. Or you can say a transaction is<br />> equivalent to calling all of the I/O system calls at once. You can<br />> also decide if you want the reads and directory lookups performed in<br />> the transactions to become prerequisites for the transaction<br />> completing (so it's aborted if another process writes to those file<br />> regions or changes the directory structure in a way which breaks a<br />> prerequisite), or if you want those to lock the things which are read<br />> for the duration of the transaction, or even just ignore reads for<br />> transaction purposes. Or, you can say that transactions are limited<br />> to just directory structure, and not file contents (that's good enough<br />> for package management), or you can say they're limited to just file<br />> contents (that's good enough for databases and text file edits).<br />> <br />> Etc, etc, quite a lot of semantic choices.<br /><br />How do we specify which calls belong to a transaction? By some kind of<br />extra file handle?<br /><br />I'd think having global per-process transaction is not the best way.<br />So I think we should have some kind of transaction handle (probably in<br />the file handle space) and a way to say that a syscall is done within<br />a transaction. To avoid duplicating all syscalls, we could have<br />set_active_transaction() operation.<br /><br />Now I think the criteria for semantics should be serializability. That<br />would mean, that lookup paths would have to be locked IFF the lookup was<br />done within the transaction -- but you would be free to open a file<br />without transaction, then set_active_transaction and write that file.<br />That way the write would become atomic, but someone else could freely<br />rename the file from under you.<br /><br />Note: Editors currently write to a temporary file and rename over the<br />original (if they have permissions to do it), which is as good<br />transaction as they need.<br /><br />> > What about programs that are already open and running? <br />> > <br />> > It might be doable in some sense, but I can see that details are<br />> > really hard to get right. Esp without breaking existing Unix<br />> > semantics. <br />> <br />> It's even harder without kernel support! :)<br /><br />If every syscall (touching filesystem) was turned into a transaction of<br />it's own, it wouldn't break any semantics.<br /><br />-------------------------------------------------------------------------------<br /> Jan 'Bulb' Hudec <bulb@ucw.cz><br />[unhandled content-type:application/pgp-signature]</pre></td><td width="32" rowspan="2" class="c" valign="top"><img src="/images/icornerr.gif" width="32" height="32" alt="\" /></td></tr><tr><td align="right" valign="bottom"> 聽 </td></tr><tr><td align="right" valign="bottom">聽</td><td class="c" valign="bottom" style="padding-bottom: 0px"><img src="/images/bcornerl.gif" width="32" height="32" alt="\" /></td><td class="c">聽</td><td class="c" valign="bottom" style="padding-bottom: 0px"><img src="/images/bcornerr.gif" width="32" height="32" alt="/" /></td></tr><tr><td align="right" valign="top" colspan="2"> 聽 </td><td class="lm">Last update: 2005-04-27 11:38 聽聽 [from the cache]<br />漏2003-2020 <a href="http://blog.jasper.es/"><span itemprop="editor">Jasper Spaans</span></a>|hosted at <a href="https://www.digitalocean.com/?refcode=9a8e99d24cf9">Digital Ocean</a> and my Meterkast|<a href="http://blog.jasper.es/categories.html#lkml-ref">Read the blog</a></td><td>聽</td></tr></table><script language="javascript" src="/js/styleswitcher.js" type="text/javascript"></script></body></html>