- 24 Jul, 2008 40 commits
-
-
Ian Kent authored
Appologies, somehow I seem to have sent an out dated version of this patch. Here is an additional patch that brings the patch up to date. Signed-off-by:
Ian Kent <raven@themaw.net> Cc: Jeff Moyer <jmoyer@redhat.com> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
Ian Kent authored
For direct and offset type mounts that are covered by another mount we cannot check the AUTOFS_INF_EXPIRING flag during a path walk which leads to lookups walking into an expiring mount while it is being expired. For example, for the direct multi-mount map entry with a couple of offsets: /race/mm1 / <server1>:/<path1> /om1 <server2>:/<path2> /om2 <server1>:/<path3> an autofs trigger mount is mounted on /race/mm1 and when accessed it is over mounted and trigger mounts made for /race/mm1/om1 and /race/mm1/om2. So it isn't possible for path walks to see the expiring flag at all and they happily walk into the file system while it is expiring. When expiring these mounts follow_down() must stop at the autofs mount and all processes must block in the ->follow_link() method (except the daemon) until the expire is complete. This is done by decrementing the d_mounted field of the autofs trigger mount root dentry until the expire is completed. In ->follow_link() all processes wait on the expire and the mount following is completed for the daemon until the expire is complete. Signed-off-by: -
Ian Kent authored
The selection of a dentry for expiration and the setting of the AUTOFS_INF_EXPIRING flag isn't done atomically which can lead to lookups walking into an expiring mount. What happens is that an expire is initiated by the daemon and a dentry is selected for expire but, since there is no lock held between the selection and setting of the expiring flag, a process may find the flag clear and continue walking into the mount tree at the same time the daemon attempts the expire it. Signed-off-by:
Jeff Moyer <jmoyer@redhat.com> Signed-off-by:
-
Ian Kent authored
There are two cases for which a dentry that has a pending mount request does not wait for completion. One is via autofs4_revalidate() and the other via autofs4_follow_link(). In revalidate, after the mount point directory is created, but before the mount is done, the check in try_to_fill_dentry() can can fail to send the dentry to the wait queue since the dentry is positive and the lookup flags may contain only LOOKUP_FOLLOW. Although we don't trigger a mount for the LOOKUP_FOLLOW flag, if ther's one pending we might as well wait and use the mounted dentry for the lookup. In autofs4_follow_link() the dentry is not checked to see if it is pending so it may fail to call try_to_fill_dentry() and not wait for mount completion. A dentry that is pending must always be sent to the wait queue. Signed-off-by:
-
Ian Kent authored
The mount triggering functionality of readdir and related functions is no longer used (and is quite broken as well). The unused portions have been removed. Signed-off-by:
-
Ian Kent authored
We have been seeing mount requests comming to the automount daemon for keys of the form "<map key>/<non key directory>" which are lookups for invalid map keys. But we can check for this in the kernel module and return a fail immediately, without having to send a request to the daemon. It is possible to recognise these requests are invalid based on whether the request dentry is negative and its relation to the autofs file system root. For example, given the indirect multi-mount map entry: idm1 \ /mm1 <server>:/<path1> /mm2 <server>:/<path2> For a request to mount idm1, IS_ROOT((idm1)->d_parent) will be always be true and the dentry may be negative. But directories idm1/mm1 and idm1/mm2 will always be created as part of the mount request for idm1. So any mount request within idm1 itself must have a positive dentry otherwise the map key is invalid. In version 4 these multi-mount entries are all mounted and umounted as a single request and in version 5 the directories idm1/mm1 and idm1/mm2 are created and an autofs fs mounted on them to act as a mount trigger so the above is also true. This also holds true for the autofs version 4 pseudo direct mount feature. When this feature is used without the "--ghost" option automount(8) will create internal submounts as we go down the map key paths which are essentially normal indirect mounts for which the above holds. If the "--ghost" option is given the directories for map keys are created at daemon startup so valid map entries correspond to postive dentries in the autofs fs. autofs version 5 direct mount maps are similar except that the IS_ROOT check is not needed. This has been addressed in a previous patch tittled "autofs4 - detect invalid direct mount requests". For example, given the direct multi-mount map entry: /test/dm1 \ /mm1 <server>:/<path1> /mm2 <server>:/<path2> An autofs fs is mounted on /test/dm1 as a trigger mount and when a mount is triggered for /test/dm1, the multi-mount offset directories /test/dm1/mm1 and /test/dm1/mm2 are created and an autofs fs is mounted on them to act as mount triggers. So valid direct mount requests must always have a positive dentry if they correspond to a valid map entry. Signed-off-by: -
Ian Kent authored
autofs v5 direct and offset mounts within an autofs filesystem are triggered by existing autofs triger mounts so the mount point dentry must be positive. If the mount point dentry is negative then the trigger doesn't exist so we can return fail immediately. Signed-off-by:
-
Ian Kent authored
If an autofs mount becomes catatonic before autofs4_wait_release() is called the wait queue counter will not be decremented down to zero and the entry will never be freed. There are also races decrementing the wait counter in the wait release function. To deal with this the counter needs to be updated while holding the wait queue mutex and waiters need to be woken up unconditionally when the wait is removed from the queue to ensure we eventually free the wait. Signed-off-by:
-
Ian Kent authored
It is possible for an autofs mount to become catatonic (and for the daemon communication pipe to become NULL) after a wait has been initiallized but before the request has been sent to the daemon. We need to check for this before sending the request packet. Signed-off-by:
-
Ian Kent authored
It see that the patch tittled "autofs4 - fix pending mount race" is missing a change that I had recently made. It's missing a kfree for the case mutex_lock_interruptible() fails to aquire the wait queue mutex. Signed-off-by:
-
Ian Kent authored
Close a race between a pending mount that is about to finish and a new lookup for the same directory. Process P1 triggers a mount of directory foo. It sets DCACHE_AUTOFS_PENDING in the ->lookup routine, creates a waitq entry for 'foo', and calls out to the daemon to perform the mount. The autofs daemon will then create the directory 'foo', using a new dentry that will be hashed in the dcache. Before the mount completes, another process, P2, tries to walk into the 'foo' directory. The vfs path walking code finds an entry for 'foo' and calls the revalidate method. Revalidate finds that the entry is not PENDING (because PENDING was never set on the dentry created by the mkdir), but it does find the directory is empty. Revalidate calls try_to_fill_dentry, which sets the PENDING flag and then calls into the autofs4 wait code to trigger or wait for a mount of 'foo'. The wait code finds the entry for 'foo' and goes to sleep waiting for the completion of the mount. Yet another process, P3, tries to walk into the 'foo' directory. This process again finds a dentry in the dcache for 'foo', and calls into the autofs revalidate code. The revalidate code finds that the PENDING flag is set, and so calls try_to_fill_dentry. a) try_to_fill_dentry sets the PENDING flag redundantly for this dentry, then calls into the autofs4 wait code. b) the autofs4 wait code takes the waitq mutex and searches for an entry for 'foo' Between a and b, P1 is woken up because the mount completed. P1 takes the wait queue mutex, clears the PENDING flag from the dentry, and removes the waitqueue entry for 'foo' from the list. When it releases the waitq mutex, P3 (eventually) acquires it. At this time, it looks for an existing waitq for 'foo', finds none, and so creates a new one and calls out to the daemon to mount the 'foo' directory. Now, the reason that three processes are required to trigger this race is that, because the PENDING flag is not set on the dentry created by mkdir, the window for the race would be way to slim for it to ever occur. Basically, between the testing of d_mountpoint(dentry) and the taking of the waitq mutex, the mount would have to complete and the daemon would have to be woken up, and that in turn would have to wake up P1. This is simply impossible. Add the third process, though, and it becomes slightly more likely. Signed-off-by:
-
Ian Kent authored
The autofs4_catatonic_mode() function accesses the wait queue without any locking but can be called at any time. This could lead to a possible double free of the name field of the wait and a double fput of the daemon communication pipe or an fput of a NULL file pointer. Signed-off-by:
-
Jeff Moyer authored
The autofs_wait_queue already contains all of the fields of the struct qstr, so change it into a qstr. This patch, from Jeff Moyer, has been modified a liitle by myself. Signed-off-by:
-
Ian Kent authored
When an open(2) call is made on an autofs mount point directory that already exists and the O_DIRECTORY flag is not used the needed mount callback to the daemon is not done. This leads to the path walk continuing resulting in a callback to the daemon with an incorrect key. open(2) is called without O_DIRECTORY by the "find" utility but this should be handled properly anyway. This happens because autofs needs to use the lookup flags to decide when to callback to the daemon to perform a mount to prevent mount storms. For example, an autofs indirect mount map that has the "browse" option will have the mount point directories are pre-created and the stat(2) call made by a color ls against each directory will cause all these directories to be mounted. It is unfortunate we need to resort to this but mount maps can be quite large. Additionally, if a user manually umounts an autofs indirect mount the directory isn't removed which also leads to this situation. To resolve this autofs needs to use the lookup intent flags to enable it to make this decision. This patch adds this check and triggers a call back if any of the lookup intent flags are set as all these calls warrant a mount attempt be requested. I know that external VFS code which uses the lookup flags is something that the VFS would like to eliminate but I have no choice as I can't see any other way to do this. A VFS dentry or inode operation callback which returns the lookup "type" (requires a definition) would be sufficient. But this change is needed now and I'm not aware of the form that coming VFS changes will take so I'm not willing to propose anything along these lines. If anyone can provide an alternate method I would be happy to use it. [akpm@linux-foundation.org: fix build for concurrent VFS changes] Signed-off-by:
-
Ian Kent authored
Since we now delay hashing of dentrys until the ->mkdir() call, droping and re-taking the directory mutex within the ->lookup() function when we are being called by user space is not needed. This can lead to a race when other processes are attempting to access the same directory during mount point directory creation. In this case we need to hang onto the mutex to ensure we don't get user processes trying to create a mount request for a newly created dentry after the mount point entry has already been created. This ensures that when we need to check a dentry passed to autofs4_wait(), if it is hashed, it is always the mount point dentry and not a new dentry created by another lookup during directory creation. Signed-off-by:
-
Ian Kent authored
The length of the symlink name has been moved but it needs to be set before allocating space for it in the dentry info struct. This corrects a mistake in a recent patch. Signed-off-by:
-
Ian Kent authored
A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by:
-
Ian Kent authored
This patch series enables the use of a single dentry for lookups prior to the dentry being hashed and so we no longer need to redo the lookup. This patch reverts the patch of commit 03379044. Signed-off-by:
-
Ian Kent authored
Correct the error of making a positive dentry negative after it has been instantiated. The code that makes this error attempts to re-use the dentry from a concurrent expire and mount to resolve a race and the dentry used for the lookup must be negative for mounts to trigger in the required cases. The fact is that the dentry doesn't need to be re-used because all that is needed is to preserve the flag that indicates an expire is still incomplete at the time of the mount request. This change uses the the dentry to check the flag and wait for the expire to complete then discards it instead of attempting to re-use it. Signed-off-by:
-
Michael Halcrow authored
There is no good reason to immediately open the lower file, and that can cause problems with files that the user does not intend to immediately open, such as device nodes. This patch removes the persistent file open from the interpose step and pushes that to the locations where eCryptfs really does need the lower persistent file, such as just before reading or writing the metadata stored in the lower file header. Two functions are jumping to out_dput when they should just be jumping to out on error paths. This patch also fixes these. Signed-off-by:
Michael Halcrow <mhalcrow@us.ibm.com> Signed-off-by:
-
Michael Halcrow authored
When creating device nodes, eCryptfs needs to delay actually opening the lower persistent file until an application tries to open. Device handles may not be backed by anything when they first come into existence. [Valdis.Kletnieks@vt.edu: build fix] Signed-off-by:
-
Harvey Harrison authored
Fixe sparse warnings: fs/ecryptfs/inode.c:368:15: warning: cast to restricted __be64 fs/ecryptfs/mmap.c:385:12: warning: incorrect type in assignment (different base types) fs/ecryptfs/mmap.c:385:12: expected unsigned long long [unsigned] [assigned] [usertype] file_size fs/ecryptfs/mmap.c:385:12: got restricted __be64 [usertype] <noident> fs/ecryptfs/mmap.c:428:12: warning: incorrect type in assignment (different base types) fs/ecryptfs/mmap.c:428:12: expected unsigned long long [unsigned] [assigned] [usertype] file_size fs/ecryptfs/mmap.c:428:12: got restricted __be64 [usertype] <noident> Signed-off-by:
Harvey Harrison <harvey.harrison@gmail.com> Cc: Michael Halcrow <mhalcrow@us.ibm.com> Signed-off-by:
-
Harvey Harrison authored
Fixes the following sparse warnings: fs/ecryptfs/crypto.c:1036:8: warning: cast to restricted __be32 fs/ecryptfs/crypto.c:1038:8: warning: cast to restricted __be32 fs/ecryptfs/crypto.c:1077:10: warning: cast to restricted __be32 fs/ecryptfs/crypto.c:1103:6: warning: incorrect type in assignment (different base types) fs/ecryptfs/crypto.c:1105:6: warning: incorrect type in assignment (different base types) fs/ecryptfs/crypto.c:1124:8: warning: incorrect type in assignment (different base types) fs/ecryptfs/crypto.c:1241:21: warning: incorrect type in assignment (different base types) fs/ecryptfs/crypto.c:1244:30: warning: incorrect type in assignment (different base types) fs/ecryptfs/crypto.c:1414:23: warning: cast to restricted __be32 fs/ecryptfs/crypto.c:1417:32: warning: cast to restricted __be16 Signed-off-by:
-
Miklos Szeredi authored
Clean up overcomplicated string copy, which also gets rid of this bogus warning: fs/ecryptfs/main.c: In function 'ecryptfs_parse_options': include/asm/arch/string_32.h:75: warning: array subscript is above array bounds Signed-off-by:
Miklos Szeredi <mszeredi@suse.cz> Cc: Michael Halcrow <mhalcrow@us.ibm.com> Signed-off-by:
-
Eric Sandeen authored
Mounting with invalid key signatures should probably fail, if they were specifically requested but not available. Also fix case checks in process_request_key_err() for the right sign of the errnos, as spotted by Jan Tluka. Signed-off-by:
Eric Sandeen <sandeen@redhat.com> Reviewed-by:
Jan Tluka <jtluka@redhat.com> Acked-by:
-
Tyler Hicks authored
The userspace eCryptfs daemon sends HELO and QUIT messages to the kernel for per-user daemon (un)registration. These messages are required when netlink is used as the transport, but (un)registration is handled by opening and closing the device file when miscdev is the transport. These messages should be discarded in the miscdev transport so that a daemon isn't registered twice. Signed-off-by:
Tyler Hicks <tyhicks@linux.vnet.ibm.com> Cc: Michael Halcrow <mhalcrow@us.ibm.com> Signed-off-by:
-
Michael Halcrow authored
eCryptfs would really like to have read-write access to all files in the lower filesystem. Right now, the persistent lower file may be opened read-only if the attempt to open it read-write fails. One way to keep from having to do that is to have a privileged kthread that can open the lower persistent file on behalf of the user opening the eCryptfs file; this patch implements this functionality. This patch will properly allow a less-privileged user to open the eCryptfs file, followed by a more-privileged user opening the eCryptfs file, with the first user only being able to read and the second user being able to both read and write. eCryptfs currently does this wrong; it will wind up calling vfs_write() on a file that was opened read-only. This is fixed in this patch. Signed-off-by:
-
Wang Chen authored
Compile warning: ignoring return value of `sysfs_create_link', declared with attribute warn_unused_result. If sysfs_create_link failed, take care of the return value and do some error handle after the failure. Since sysfs_remove_link() will check whether a link exists, when removing the link in error path, we don't need to care whether a link was created. Signed-off-by:
Wang Chen <wangchen@cn.fujitsu.com> Cc: Markus Lidel <Markus.Lidel@shadowconnect.com> Cc: Jens Axboe <jens.axboe@oracle.com> Signed-off-by:
-
Stefano Stabellini authored
vt.c DO_UPDATE macro checks if the console is visible but doesn't check if the console is blanked. In fact updating fbcon while the console is blanked is not only unnecessary but can even cause screen corruption. Therefore I am adding a simple check on console_blanked in DO_UPDATE. Signed-off-by:
Stefano Stabellini <stefano.stabellini@eu.citrix.com> Cc: Krzysztof Helt <krzysztof.h1@poczta.fm> Cc: "Antonino A. Daplas" <adaplas@pol.net> Signed-off-by:
-
Jiri Slaby authored
Hold console sem while creating/destroying sysfs files. Serialisation is so far done by BKL held in tty release_dev and chrdev_open, but no other locks are held in open path. Signed-off-by:
Jiri Slaby <jirislaby@gmail.com> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Cc: Aristeu Rozanski <aris@ruivo.org> Signed-off-by:
-
Jan Nikitenko authored
Improve PIO transfer mode of au1550 spi controller by continuing of spi transfer, instead of aborting transfer when transmit underflow interrupt occurrs. Verified by oscilloscope that the spi clock pauses on trasmit underflow, so transfer continuation is perfectly valid even though au1550 datasheet says that on tx underflow zeroes will be transfered. Also make some error messages more specific. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by:
Jan Nikitenko <jan.nikitenko@gmail.com> Signed-off-by:
David Brownell <dbrownell@users.sourceforge.net> Signed-off-by:
-
Manuel Lauss authored
Remove the Au1550 resource table and instead extract MMIO/IRQ/DMA resources from platform resource information like any well-behaved platform driver. Signed-off-by:
Manuel Lauss <mano@roarinelk.homelinux.net> Signed-off-by:
-
Alan Cox authored
Another step to removing ->ioctl and to removing the BKL [dbrownell@users.sourceforge.net: take final step; BKL not needed] Signed-off-by:
Alan Cox <alan@redhat.com> Signed-off-by:
-
Grant Likely authored
Currently, 'modalias' in the spi_device structure is a 'const char *'. The spi_new_device() function fills in the modalias value from a passed in spi_board_info data block. Since it is a pointer copy, the new spi_device remains dependent on the spi_board_info structure after the new spi_device is registered (no other fields in spi_device directly depend on the spi_board_info structure; all of the other data is copied). This causes a problem when dynamically propulating the list of attached SPI devices. For example, in arch/powerpc, the list of SPI devices can be populated from data in the device tree. With the current code, the device tree adapter must kmalloc() a new spi_board_info structure for each new SPI device it finds in the device tree, and there is no simple mechanism in place for keeping track of these allocations. This patch changes modalias from a 'const char *' to a fixed char array. By copying the modalias string instead of referencing it, the dependency on the spi_board_info structure is eliminated and an outside caller does not need to maintain a separate spi_board_info allocation for each device. If searched through the code to the best of my ability for any references to modalias which may be affected by this change and haven't found anything. It has been tested with the lite5200b platform in arch/powerpc. [dbrownell@users.sourceforge.net: cope with linux-next changes: KOBJ_NAME_LEN obliterated, etc] Signed-off-by:
Grant Likely <grant.likely@secretlab.ca> Signed-off-by:
-
Robert P. J. Day authored
Use "if SPI_MASTER" to remove numerous dependencies. [dbrownell@users.sourceforge.net: remove a couple now-needless EXPERIMENTAL dependencies too] Signed-off-by:
Robert P. J. Day <rpjday@crashcourse.ca> Signed-off-by:
-
Roel Kluin authored
xilinx_spi->irq is unsigned, so the test fails Signed-off-by:
Roel Kluin <12o3l@tiscali.nl> Cc: David Brownell <dbrownell@users.sourceforge.net> Cc: Andrei Konovalov <akonovalov@ru.mvista.com> Cc: Yuri Frolov <yfrolov@ru.mvista.com> Signed-off-by:
-
Chen Gong authored
This updates the SPI clock rate calculations for the spi_mpc83xx driver. Some boundary conditions were wrong, and in several cases divide-by-16 wasn't always needed Signed-off-by:
Chen Gong <g.chen@freescale.com> Signed-off-by:
-
Andre Haupt authored
Signed-off-by:
Andre Haupt <andre@bitwigglers.org> Acked-by:
-
Nye Liu authored
Before setting STOP_TX, set _brkcr to 0 so the SMC does not send a break character. The driver appears to properly re-initialize _brkcr when the SMC is restarted. Do not interrupt RX/TX when the termios is being adjusted; it results in corrupted characters appearing on the line. Cc: Vitaly Bordug <vbordug@ru.mvista.com> Cc: Scott Wood <scottwood@freescale.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Kumar Gala <galak@kernel.crashing.org> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Signed-off-by:
-
Maciej W. Rozycki authored
Changes to the generic console support code that happened a while ago introduced a scenario where the initial console is used in parallel with the final console during a brief period when switching between the two is in progress. During that time a message about the switch-over is printed. With some combinations of chips, firmware and drivers, such as the DEC DZ11 clone used with the DECstation, a hang may happen because the firmware used for the initial console may not expect the state of the chip after it has been initialised by the driver. This is a workaround for the DZ11 which reuses the power-management callback to keep the transmitter of the line associated with the console enabled. It reflects the consensus reached in a discussion a while ago. Signed-off-by:
Maciej W. Rozycki <macro@linux-mips.org> Cc: Jiri Slaby <jirislaby@gmail.com> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Signed-off-by:
-
