From owner-freebsd-hackers@FreeBSD.ORG Fri Nov 6 17:52:42 2009 Return-Path: Delivered-To: freebsd-hackers@freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:4f8:fff6::34]) by hub.freebsd.org (Postfix) with ESMTP id 9DE9D106566B; Fri, 6 Nov 2009 17:52:42 +0000 (UTC) (envelope-from a_best01@uni-muenster.de) Received: from zivm-exrelay1.uni-muenster.de (ZIVM-EXRELAY1.UNI-MUENSTER.DE [128.176.192.14]) by mx1.freebsd.org (Postfix) with ESMTP id A11BD8FC16; Fri, 6 Nov 2009 17:52:41 +0000 (UTC) X-IronPort-AV: E=Sophos;i="4.44,694,1249250400"; d="scan'208";a="287569039" Received: from zivmaildisp1.uni-muenster.de (HELO ZIVMAILUSER03.UNI-MUENSTER.DE) ([128.176.188.85]) by zivm-relay1.uni-muenster.de with ESMTP; 06 Nov 2009 18:52:40 +0100 Received: by ZIVMAILUSER03.UNI-MUENSTER.DE (Postfix, from userid 149459) id B2BBF1B0751; Fri, 6 Nov 2009 18:52:40 +0100 (CET) Date: Fri, 06 Nov 2009 18:52:40 +0100 (CET) From: Alexander Best Sender: Organization: Westfaelische Wilhelms-Universitaet Muenster To: Alexander Best , , Alexander Best Message-ID: In-Reply-To: MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Cc: freebsd-hackers@FreeBSD.org, Alex Dupre Subject: Re: rmdir(2) and mkdir(2) both return EISDIR for argument "/" X-BeenThere: freebsd-hackers@freebsd.org X-Mailman-Version: 2.1.5 Precedence: list List-Id: Technical Discussions relating to FreeBSD List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Fri, 06 Nov 2009 17:52:42 -0000 Alexander Best schrieb am 2009-11-06: > Gary Jennejohn schrieb am 2009-11-06: > > On Fri, 06 Nov 2009 17:43:06 +0100 (CET) > > Alexander Best wrote: > > > Gary Jennejohn schrieb am 2009-11-06: > > > > On Fri, 06 Nov 2009 16:32:22 +0100 (CET) > > > > Alexander Best wrote: > > > > > Alex Dupre schrieb am 2009-11-06: > > > > > > Alexander Best ha scritto: > > > > > > > i dug up this old pr > > > > > > > http://www.freebsd.org/cgi/query-pr.cgi?pr=kern/59739 > > > > > > I think the EISDIR error is coming from kern/vfs_lookup.c, > > > > > > lookup() > > > > > > function with cn_nameptr = "": > > > > > > /* > > > > > > * Check for degenerate name (e.g. / or "") > > > > > > * which is a way of talking about a directory, > > > > > > * e.g. like "/." or ".". > > > > > > */ > > > > > > if (cnp->cn_nameptr[0] == '\0') { > > > > > > ... > > > > > > if (cnp->cn_nameiop != LOOKUP) { > > > > > > error = EISDIR; > > > > > > goto bad; > > > > > > } > > > > > > ... > > > > > thanks a lot for finding the problem in the src. what do you > > > > > think > > > > > of the > > > > > patch attached to this message? after applying it the example > > > > > code > > > > > i posted in > > > > > my previous message returns the following output (instead of > > > > > EISDIR): > > > > > rmdir errno: 16 (which is EBUSY) > > > > > mkdir errno: 17 (which is EEXIST) > > > > > i don't know if these really are the correct return values, > > > > > but > > > > > it's what the > > > > > originator of the PR requested. > > > > What if cn_nameiop is != LOOKUP but also neither DELETE nor > > > > CREATE, > > > > assuming that case is possible? I'd leave the original > > > > if-clause > > > > at > > > > the end to catch that. > > > > --- > > > > Gary Jennejohn > > > how about this patch? > > > 1. i've added "if (cnp->cn_nameiop != LOOKUP)" although i don't > > > think it's > > > necessary since the first blocks should cover all the possible > > > cases. > > > 2. i've used rename() to test the case (cnp->cn_nameiop != > > > RENAME). > > > is this > > > correct or does rename() use a combo of DELETE and CREATE? > > > problem > > > is that the > > > rename(2) manual doesn't seem to cover the case that arg 1 is a > > > mountpoint. > > > right now EBUSY gets returned if cnp->cn_nameiop != RENAME. > > > however > > > BUSY needs > > > to be added to all manuals which use cnp->cn_nameiop != RENAME > > > (shouldn't be > > > too many). or are there any other suggestions what rename() > > > should > > > return if > > > arg 1 is a mountpoint? > > Hmm. In rename(2) there's > > [EINVAL] The from argument is a parent directory of to, > > or > > an > > attempt is made to rename `.' or `..'. > > and a few lines below your patch this case is handled for ISDOTDOT > > for both RENAME and DELETE. I don't see off hand where renaming or > > deleting "." is handled. > > According to the comment above your patch the case of "/." or "." > > is > > being checked, which would seem to correspond to the above part of > > rename(2), i.e. perhaps EINVAL should be returned for RENAME and > > DELETE. > > --- > > Gary Jennejohn > that would be an option. however in the case of rmdir(2) EINVAL and > EBUYS > would both fit. depends whether be forbid deletion of / because it is > a > mountpoint or because / is actually /. and paths ending with . are > forbidden > as arg in rmdir(2). > i guess we have to take a look at the POSIX specs before we can > decide how to > handle this. > also i've discovered that permission checks for / seem to be handled > differently than any other dir. on my machine /usr is a mountpoint. > doing > rmdir /usr returns EACCES as regular user and EBUSY as superuser. > doing rmdir > / as regular user however doesn't seem to check permission but > returns EBUSY > right away. but that's not a problem i guess. this is probably > happening > because the kern/vfs_lookup.c code is being executed before anything > else > (including permission checks). > i'll have a look what POSIX has to say about the return values. but i > agree > with you. returning EINVAL seems logical. > alex. ok. here it goes. POSIX says: rmdir() [http://www.opengroup.org/onlinepubs/9699919799/functions/rmdir.html#tag_16_618]: If the directory is the root directory or the current working directory of any process, it is unspecified whether the function succeeds, or whether it shall fail and set errno to [EBUSY]. mkdir() [http://www.opengroup.org/onlinepubs/9699919799/functions/mkdir.html#tag_16_372]: nothing regarding /. rename() [http://www.opengroup.org/onlinepubs/9699919799/functions/rename.html#tag_16_614]: If either pathname argument refers to a path whose final component is either dot or dot-dot, rename() shall fail. so at least rmdir() should return EBUSY. also POSIX defines EBUSY for return() which isn't documented in our return(2) manual. alex