Re: [Gluster-devel] Invalid DIR * usage in quota xlator
On Tue, Oct 14, 2014 at 7:22 PM, Emmanuel Dreyfus m...@netbsd.org wrote: J. Bruce Fields bfie...@fieldses.org wrote: Is the result on non-Linux really to fail any readdir using an offset not returned from the current open? Yes, but thatnon-Linux behabvior is POSIX compliant. Linux just happens to do more than the standard here. I can't see how NFS READDIR will work on non-Linux platforms in that case. Differrent case: seekdir/telldir operate at libc level on data cached in userland. NFS READDIR operates on data in the kernel. Is there a way to get hold of the directory entry cookies used by NFS readdir from user-space? some sort of a NetBSD specific syscall (like getdents of Linux)? Thanks ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
Anand Avati av...@gluster.org wrote: Is there a way to get hold of the directory entry cookies used by NFS readdir from user-space? some sort of a NetBSD specific syscall (like getdents of Linux)? NetBSD has getdents(2) too, but it returns buffer of struct dirent that have no NFS cookies. On Linux, getdents(2) returns buffers of struct dirent or dirent64, but I see no NFS cookie there either. -- Emmanuel Dreyfus http://hcpnet.free.fr/pubz m...@netbsd.org ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
On Wed, Oct 15, 2014 at 01:26:19PM +0200, Emmanuel Dreyfus wrote: Anand Avati av...@gluster.org wrote: Is there a way to get hold of the directory entry cookies used by NFS readdir from user-space? some sort of a NetBSD specific syscall (like getdents of Linux)? NetBSD has getdents(2) too, but it returns buffer of struct dirent that have no NFS cookies. On Linux, getdents(2) returns buffers of struct dirent or dirent64, but I see no NFS cookie there either. On Linux, it's returned in the d_off field. Googling around the NetBSD manpage doesn't mention any d_off field, but it does have this: http://nixdoc.net/man-pages/netbsd/man2/getdents.2.html The current position pointer may be set and retrieved by lseek(2). The current position pointer should only be set to a value returned by lseek(2), or zero. So maybe that's what you want. --b. ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
On Mon, Oct 13, 2014 at 08:57:11AM +, Emmanuel Dreyfus wrote: On Mon, Oct 13, 2014 at 01:42:38PM +0530, Pranith Kumar Karampuri wrote: No bug here, just suboptimal behavior, both in glusterfs and NetBSD FUSE. oh!, but shouldn't it get op_ret = 0 instead of op_ret -1, op_errno EINVAL? It happens because of thei change I introduced: seekdir (dir, (long)off); #ifndef GF_LINUX_HOST_OS if (telldir(dir) != off) { gf_log (THIS-name, GF_LOG_ERROR, seekdir(%ld) failed on dir=%p: Invalid argument (offset reused from another DIR * structure?), (long)off, dir); errno = EINVAL; count = -1; goto out; } #endif /* GF_LINUX_HOST_OS */ The idea is to catch wrong opendir/closedir usage, and the errno = EINVAL is critical in such a situation to avoid an infinite loop. On the other hand it also fires for last entry. I did not notice it before because this EINVAL will not reach calling process, but it may cause trouble for internal glusterfs usage. I see an easy way yo fix it (patch below) - when posix xlator detects EOF, return an offset of -1 in last entry. - When called with an offset of -1, immediatly return op_ret = 0 with no entry You're assuming no filesystem uses -1 as a cookie? Might be true, but it's taking a small chance. --b. What do you think? diff --git a/xlators/storage/posix/src/posix.c b/xlators/storage/posix/src/posix index 6b12d39..3e539bb 100644 --- a/xlators/storage/posix/src/posix.c +++ b/xlators/storage/posix/src/posix.c @@ -79,6 +79,10 @@ extern char *marker_xattrs[]; #define SET_TO_OLD_FS_ID() #endif + +/* offset sizeof(struct dirent) cannot be valid */ +#define INVALID_DIRENT_OFFSET sizeof (void *) + int posix_forget (xlator_t *this, inode_t *inode) { @@ -4851,6 +4855,10 @@ posix_fill_readdir (fd_t *fd, DIR *dir, off_t off, size_t int len = 0; int ret = 0; +/* EOF was previously detected */ +if (off == INVALID_DIRENT_OFFSET) + goto out; + if (skip_dirs) { len = posix_handle_path (this, fd-inode-gfid, NULL, NULL, 0); hpath = alloca (len + 256); /* NAME_MAX */ @@ -4969,9 +4977,12 @@ posix_fill_readdir (fd_t *fd, DIR *dir, off_t off, size_t count ++; } -if ((!readdir (dir) (errno == 0))) +if ((!readdir (dir) (errno == 0))) { /* Indicate EOF */ errno = ENOENT; +/* Set invalid offset for later detection */ +this_entry-d_off = INVALID_DIRENT_OFFSET; +} out: return count; } -- Emmanuel Dreyfus m...@netbsd.org ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
J. Bruce Fields bfie...@fieldses.org wrote: You're assuming no filesystem uses -1 as a cookie? Might be true, but it's taking a small chance. Latest version does not work like this anymore: when EOF is reached, current offset is recorded in pfd-dir_eof, and we fire EINVAL if seekdir() failed to set the offset and the offset is not pfd-dir_eof: if (telldir(dir) != (long)off off != pfd-dir_eof) -- Emmanuel Dreyfus http://hcpnet.free.fr/pubz m...@netbsd.org ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
J. Bruce Fields bfie...@fieldses.org wrote: (It'd still be better to figure out how to get stable directory cookies, possibly using some other interface, but that's a preexisting problem not introduced by c65d4ea8a10a). getdents(2) seems the way to go. -- Emmanuel Dreyfus http://hcpnet.free.fr/pubz m...@netbsd.org ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
On Wed, Oct 15, 2014 at 07:42:25PM +0200, Emmanuel Dreyfus wrote: J. Bruce Fields bfie...@fieldses.org wrote: You're assuming no filesystem uses -1 as a cookie? Might be true, but it's taking a small chance. Latest version does not work like this anymore: when EOF is reached, current offset is recorded in pfd-dir_eof, and we fire EINVAL if seekdir() failed to set the offset and the offset is not pfd-dir_eof: if (telldir(dir) != (long)off off != pfd-dir_eof) But actually, I was wrong: telldir returns -1 for error or offset on success, so -1 is reserved. --b. ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
On Mon, Oct 13, 2014 at 10:04:33AM +, Emmanuel Dreyfus wrote: Here it is: http://review.gluster.org/8926 It passed regression, please review. Linux behavior is unaltered. -- Emmanuel Dreyfus m...@netbsd.org ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
On Sun, Oct 12, 2014 at 10:09:01AM +0200, Emmanuel Dreyfus wrote: When quota xlator is enabled, bricks get this: [2014-10-12 07:56:03.090516] E [posix.c:4874:posix_fill_readdir] 0-patchy-posix: seekdir(-1154801456) failed on dir=0xb99cb250: Invalid argument (offset reused from another DIR * structure?) This means something either seekdir on a DIR * that was not obtained by opendir(), or that was closedir() and then opendir() again. Can someone explain me what directory walks are introduced by quota, and where opendir/readdir/closedir happen for that? Apologies, I'm just catching up, but I'm confused by c65d4ea8a10a Fix invalid seekdir() usage Is the result on non-Linux really to fail any readdir using an offset not returned from the current open? I can't see how NFS READDIR will work on non-Linux platforms in that case. --b. ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
J. Bruce Fields bfie...@fieldses.org wrote: Is the result on non-Linux really to fail any readdir using an offset not returned from the current open? Yes, but thatnon-Linux behabvior is POSIX compliant. Linux just happens to do more than the standard here. I can't see how NFS READDIR will work on non-Linux platforms in that case. Differrent case: seekdir/telldir operate at libc level on data cached in userland. NFS READDIR operates on data in the kernel. -- Emmanuel Dreyfus http://hcpnet.free.fr/pubz m...@netbsd.org ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
On Mon, Oct 13, 2014 at 12:39:33PM +0530, Pranith Kumar Karampuri wrote: Op_errno is valid only if 'op_ret 0'. so that doesn't say much. After the last readdir call with op_ret 0, there will be one more readdir call for which op_ret will come as '0'. That means reading of the directory is complete. Right, I have 3 calls for end of directory: op_ret 0, op_errno = ENOENT, but glusterfs does an extra call and gets: op_ret = -1, op_errno = EINVAL, but NetBSD FUSE does an extra call ang gets: op_ret = -1, op_errno = EINVAL No bug here, just suboptimal behavior, both in glusterfs and NetBSD FUSE. -- Emmanuel Dreyfus m...@netbsd.org ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
On 10/13/2014 01:14 PM, Emmanuel Dreyfus wrote: On Mon, Oct 13, 2014 at 12:39:33PM +0530, Pranith Kumar Karampuri wrote: Op_errno is valid only if 'op_ret 0'. so that doesn't say much. After the last readdir call with op_ret 0, there will be one more readdir call for which op_ret will come as '0'. That means reading of the directory is complete. Right, I have 3 calls for end of directory: op_ret 0, op_errno = ENOENT, but glusterfs does an extra call and gets: op_ret = -1, op_errno = EINVAL, but NetBSD FUSE does an extra call ang gets: op_ret = -1, op_errno = EINVAL No bug here, just suboptimal behavior, both in glusterfs and NetBSD FUSE. oh!, but shouldn't it get op_ret = 0 instead of op_ret -1, op_errno EINVAL? Pranith ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
On Mon, Oct 13, 2014 at 01:42:38PM +0530, Pranith Kumar Karampuri wrote: No bug here, just suboptimal behavior, both in glusterfs and NetBSD FUSE. oh!, but shouldn't it get op_ret = 0 instead of op_ret -1, op_errno EINVAL? It happens because of thei change I introduced: seekdir (dir, (long)off); #ifndef GF_LINUX_HOST_OS if (telldir(dir) != off) { gf_log (THIS-name, GF_LOG_ERROR, seekdir(%ld) failed on dir=%p: Invalid argument (offset reused from another DIR * structure?), (long)off, dir); errno = EINVAL; count = -1; goto out; } #endif /* GF_LINUX_HOST_OS */ The idea is to catch wrong opendir/closedir usage, and the errno = EINVAL is critical in such a situation to avoid an infinite loop. On the other hand it also fires for last entry. I did not notice it before because this EINVAL will not reach calling process, but it may cause trouble for internal glusterfs usage. I see an easy way yo fix it (patch below) - when posix xlator detects EOF, return an offset of -1 in last entry. - When called with an offset of -1, immediatly return op_ret = 0 with no entry What do you think? diff --git a/xlators/storage/posix/src/posix.c b/xlators/storage/posix/src/posix index 6b12d39..3e539bb 100644 --- a/xlators/storage/posix/src/posix.c +++ b/xlators/storage/posix/src/posix.c @@ -79,6 +79,10 @@ extern char *marker_xattrs[]; #define SET_TO_OLD_FS_ID() #endif + +/* offset sizeof(struct dirent) cannot be valid */ +#define INVALID_DIRENT_OFFSET sizeof (void *) + int posix_forget (xlator_t *this, inode_t *inode) { @@ -4851,6 +4855,10 @@ posix_fill_readdir (fd_t *fd, DIR *dir, off_t off, size_t int len = 0; int ret = 0; +/* EOF was previously detected */ +if (off == INVALID_DIRENT_OFFSET) + goto out; + if (skip_dirs) { len = posix_handle_path (this, fd-inode-gfid, NULL, NULL, 0); hpath = alloca (len + 256); /* NAME_MAX */ @@ -4969,9 +4977,12 @@ posix_fill_readdir (fd_t *fd, DIR *dir, off_t off, size_t count ++; } -if ((!readdir (dir) (errno == 0))) +if ((!readdir (dir) (errno == 0))) { /* Indicate EOF */ errno = ENOENT; +/* Set invalid offset for later detection */ +this_entry-d_off = INVALID_DIRENT_OFFSET; +} out: return count; } -- Emmanuel Dreyfus m...@netbsd.org ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
On 10/13/2014 02:27 PM, Emmanuel Dreyfus wrote: On Mon, Oct 13, 2014 at 01:42:38PM +0530, Pranith Kumar Karampuri wrote: No bug here, just suboptimal behavior, both in glusterfs and NetBSD FUSE. oh!, but shouldn't it get op_ret = 0 instead of op_ret -1, op_errno EINVAL? It happens because of thei change I introduced: seekdir (dir, (long)off); #ifndef GF_LINUX_HOST_OS if (telldir(dir) != off) { gf_log (THIS-name, GF_LOG_ERROR, seekdir(%ld) failed on dir=%p: Invalid argument (offset reused from another DIR * structure?), (long)off, dir); errno = EINVAL; count = -1; goto out; } #endif /* GF_LINUX_HOST_OS */ The idea is to catch wrong opendir/closedir usage, and the errno = EINVAL is critical in such a situation to avoid an infinite loop. On the other hand it also fires for last entry. I did not notice it before because this EINVAL will not reach calling process, but it may cause trouble for internal glusterfs usage. I see an easy way yo fix it (patch below) - when posix xlator detects EOF, return an offset of -1 in last entry. - When called with an offset of -1, immediatly return op_ret = 0 with no entry What do you think? I am not aware of backend filesystems that much, may be someone with that knowledge can comment here, what happens when new entries are created in the directory after this readdir is responded with '-1'? diff --git a/xlators/storage/posix/src/posix.c b/xlators/storage/posix/src/posix index 6b12d39..3e539bb 100644 --- a/xlators/storage/posix/src/posix.c +++ b/xlators/storage/posix/src/posix.c @@ -79,6 +79,10 @@ extern char *marker_xattrs[]; #define SET_TO_OLD_FS_ID() #endif + +/* offset sizeof(struct dirent) cannot be valid */ +#define INVALID_DIRENT_OFFSET sizeof (void *) + int posix_forget (xlator_t *this, inode_t *inode) { @@ -4851,6 +4855,10 @@ posix_fill_readdir (fd_t *fd, DIR *dir, off_t off, size_t int len = 0; int ret = 0; +/* EOF was previously detected */ +if (off == INVALID_DIRENT_OFFSET) + goto out; + if (skip_dirs) { len = posix_handle_path (this, fd-inode-gfid, NULL, NULL, 0); hpath = alloca (len + 256); /* NAME_MAX */ @@ -4969,9 +4977,12 @@ posix_fill_readdir (fd_t *fd, DIR *dir, off_t off, size_t count ++; } -if ((!readdir (dir) (errno == 0))) +if ((!readdir (dir) (errno == 0))) { /* Indicate EOF */ errno = ENOENT; Now I understand why you kept saying ENOENT ;-). This may not be a good idea IMO. Code all through gluster doesn't care what errno it unwinds upwards. +/* Set invalid offset for later detection */ +this_entry-d_off = INVALID_DIRENT_OFFSET; +} out: return count; } ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
On 10/13/2014 02:37 PM, Pranith Kumar Karampuri wrote: On 10/13/2014 02:27 PM, Emmanuel Dreyfus wrote: On Mon, Oct 13, 2014 at 01:42:38PM +0530, Pranith Kumar Karampuri wrote: No bug here, just suboptimal behavior, both in glusterfs and NetBSD FUSE. oh!, but shouldn't it get op_ret = 0 instead of op_ret -1, op_errno EINVAL? It happens because of thei change I introduced: seekdir (dir, (long)off); #ifndef GF_LINUX_HOST_OS if (telldir(dir) != off) { gf_log (THIS-name, GF_LOG_ERROR, seekdir(%ld) failed on dir=%p: Invalid argument (offset reused from another DIR * structure?), (long)off, dir); errno = EINVAL; count = -1; goto out; } #endif /* GF_LINUX_HOST_OS */ The idea is to catch wrong opendir/closedir usage, and the errno = EINVAL is critical in such a situation to avoid an infinite loop. On the other hand it also fires for last entry. I did not notice it before because this EINVAL will not reach calling process, but it may cause trouble for internal glusterfs usage. I see an easy way yo fix it (patch below) - when posix xlator detects EOF, return an offset of -1 in last entry. - When called with an offset of -1, immediatly return op_ret = 0 with no entry What do you think? I am not aware of backend filesystems that much, may be someone with that knowledge can comment here, what happens when new entries are created in the directory after this readdir is responded with '-1'? diff --git a/xlators/storage/posix/src/posix.c b/xlators/storage/posix/src/posix index 6b12d39..3e539bb 100644 --- a/xlators/storage/posix/src/posix.c +++ b/xlators/storage/posix/src/posix.c @@ -79,6 +79,10 @@ extern char *marker_xattrs[]; #define SET_TO_OLD_FS_ID() #endif + +/* offset sizeof(struct dirent) cannot be valid */ +#define INVALID_DIRENT_OFFSET sizeof (void *) + int posix_forget (xlator_t *this, inode_t *inode) { @@ -4851,6 +4855,10 @@ posix_fill_readdir (fd_t *fd, DIR *dir, off_t off, size_t int len = 0; int ret = 0; +/* EOF was previously detected */ +if (off == INVALID_DIRENT_OFFSET) + goto out; + if (skip_dirs) { len = posix_handle_path (this, fd-inode-gfid, NULL, NULL, 0); hpath = alloca (len + 256); /* NAME_MAX */ @@ -4969,9 +4977,12 @@ posix_fill_readdir (fd_t *fd, DIR *dir, off_t off, size_t count ++; } -if ((!readdir (dir) (errno == 0))) +if ((!readdir (dir) (errno == 0))) { /* Indicate EOF */ errno = ENOENT; Now I understand why you kept saying ENOENT ;-). This may not be a good idea IMO. Code all through gluster doesn't care what errno it unwinds upwards. In case op_ret is zero I mean :-) Pranith +/* Set invalid offset for later detection */ +this_entry-d_off = INVALID_DIRENT_OFFSET; +} out: return count; } ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
On Mon, Oct 13, 2014 at 02:37:12PM +0530, Pranith Kumar Karampuri wrote: I am not aware of backend filesystems that much, may be someone with that knowledge can comment here, what happens when new entries are created in the directory after this readdir is responded with '-1'? In the meantime, I changed -1 to sizeof (void *) to address alignment issues. I am not sure we have cases where we loop awaiting for new entries to be added at the end of the directory. That is doomed to fail anyway because getdents(2) cn return deleted entries, and a new entry may be added before the end of the buffer. -- Emmanuel Dreyfus m...@netbsd.org ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
On 10/13/2014 02:45 PM, Emmanuel Dreyfus wrote: On Mon, Oct 13, 2014 at 02:37:12PM +0530, Pranith Kumar Karampuri wrote: I am not aware of backend filesystems that much, may be someone with that knowledge can comment here, what happens when new entries are created in the directory after this readdir is responded with '-1'? In the meantime, I changed -1 to sizeof (void *) to address alignment issues. I am not sure we have cases where we loop awaiting for new entries to be added at the end of the directory. That is doomed to fail anyway because getdents(2) cn return deleted entries, and a new entry may be added before the end of the buffer. I do not know :-(. Send the patch out for review, may be someone with this knowledge can do the review... Pranith ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
On Mon, Oct 13, 2014 at 02:58:12PM +0530, Pranith Kumar Karampuri wrote: I do not know :-(. Send the patch out for review, may be someone with this knowledge can do the review... Here it is: http://review.gluster.org/8926 -- Emmanuel Dreyfus m...@netbsd.org ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
Emmanuel Dreyfus m...@netbsd.org wrote: This means something either seekdir on a DIR * that was not obtained by opendir(), or that was closedir() and then opendir() again. Can someone explain me what directory walks are introduced by quota, and where opendir/readdir/closedir happen for that? Erratum: it happens because it attempts to seekdir to the offset obtained for last record. But I still have to find what code is sending the request. -- Emmanuel Dreyfus http://hcpnet.free.fr/pubz m...@netbsd.org ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
Emmanuel Dreyfus m...@netbsd.org wrote: Erratum: it happens because it attempts to seekdir to the offset obtained for last record. But I still have to find what code is sending the request. Here is the king of debug trace I have: [quota.c:3959:quota_readdirp] 0-XXXmanu: fd = 0xbb242818, offset = 0 [posix.c:4882:posix_fill_readdir] 0-XXXmanu: telldir(0xb99c7250) = -1154801520 [posix.c:4966:posix_fill_readdir] 0-XXXmanu: telldir(0xb99c7250) = -1154801504 [posix.c:4882:posix_fill_readdir] 0-XXXmanu: telldir(0xb99c7250) = -1154801504 [posix.c:4966:posix_fill_readdir] 0-XXXmanu: telldir(0xb99c7250) = -1154801488 [posix.c:4882:posix_fill_readdir] 0-XXXmanu: telldir(0xb99c7250) = -1154801488 [posix.c:4882:posix_fill_readdir] 0-XXXmanu: telldir(0xb99c7250) = -1154801472 [posix.c:4966:posix_fill_readdir] 0-XXXmanu: telldir(0xb99c7250) = -1154801456 [posix.c:4882:posix_fill_readdir] 0-XXXmanu: telldir(0xb99c7250) = -1154801456 [quota.c:3892:quota_readdirp_cbk] 0-XXXmanu: op_ret = 3, op_errno = ENOENT [quota.c:3959:quota_readdirp] 0-XXXmanu: fd = 0xbb242818, offset = -1154801456 [posix.c:4865:posix_fill_readdir] 0-XXXmanu: seekdir(0xb99c7250, -1154801456) [posix.c:4872:posix_fill_readdir] 0-patchy-posix: seekdir(-1154801456) failed on dir=0xb99c7250: Invalid argument (offset reused from another DIR * structure?) [quota.c:3892:quota_readdirp_cbk] 0-XXXmanu: op_ret = -1, op_errno = EINVAL [2014-10-12 14:51:35.176375] I [server-rpc-fops.c:1917:server_readdirp_cbk] 0-patchy-server: 19: READDIRP 0 (----0001) == (Invalid argument) [quota.c:3959:quota_readdirp] 0-XXXmanu: fd = 0xbb242818, offset = -1154801456 [posix.c:4865:posix_fill_readdir] 0-XXXmanu: seekdir(0xb99c7250, -1154801456) [posix.c:4872:posix_fill_readdir] 0-patchy-posix: seekdir(-1154801456) failed on dir=0xb99c7250: Invalid argument (offset reused from another DIR * structure?) -- Emmanuel Dreyfus http://hcpnet.free.fr/pubz m...@netbsd.org ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
Emmanuel Dreyfus m...@netbsd.org wrote: Erratum: it happens because it attempts to seekdir to the offset obtained for last record. But I still have to find what code is sending the request. This is a find(1) probably forked by quota-crawld glusterd. In fact running a find . in the fs is enough to trigger the phenomenon: once end of directory is reached (op_ret 0, op_errno = ENOENT), readdir is called 2 more times (op_ret = -1, op_errno = EINVAL), which suggests end of directory information does not reach the caller process correctly. -- Emmanuel Dreyfus http://hcpnet.free.fr/pubz m...@netbsd.org ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel
Re: [Gluster-devel] Invalid DIR * usage in quota xlator
On 10/13/2014 09:45 AM, Emmanuel Dreyfus wrote: Emmanuel Dreyfus m...@netbsd.org wrote: Erratum: it happens because it attempts to seekdir to the offset obtained for last record. But I still have to find what code is sending the request. This is a find(1) probably forked by quota-crawld glusterd. In fact running a find . in the fs is enough to trigger the phenomenon: once end of directory is reached (op_ret 0, op_errno = ENOENT), readdir is called 2 more times (op_ret = -1, op_errno = EINVAL), which suggests end of directory information does not reach the caller process correctly. Emmanuel, End of directory is determined by 'op_ret == 0'. On my system with xfs as backend, things are working fine. Which backend filesystem are you using? Pranith ___ Gluster-devel mailing list Gluster-devel@gluster.org http://supercolony.gluster.org/mailman/listinfo/gluster-devel