- 04 Jan, 2005 40 commits
-
-
Andi Kleen authored
This patch corrects a problem that was originally added with the nanosecond timestamps in stat patch. The problem is that some file systems don't have enough space in their on disk inode to save nanosecond timestamps, so they truncate the c/a/mtime to seconds when flushing an dirty node. In core the inode would have full jiffies granuality. This can be observed by programs as a timestamp that jumps backwards under specific loads when an inode is flushed and then reloaded from disk. The problem was already known when the original patch went in, but it wasn't deemed important enough at that time. So far there has been only one report of it causing problems. Now Tridge is worried that it will break running Excel over samba4 because Excel seems to do very anal timestamp checking and samba4 will supply 100ns timestamps over the network. This patch solves it by putting the time resolution into the superblock of a fs and always rounding the in core timestamps to that granuality. This also supercedes some previous ext2/3 hacks to flush the inode less often when only the subsecond timestamp changes. I tried to keep the overhead low, in particular it tries to keep divisions out of fast paths as far as possible. The patch is quite big but 99% of it is just relatively straight forward search'n'replace in a lot of fs. Unconverted filesystems will default to a 1ns granuality, but may still show the problem if they continue to use CURRENT_TIME. I converted all in tree fs. One possible future extension of this would be to have two time granualities per superblock - one that specifies the visible resolution, and the other to specify how often timestamps should be flushed to disk, which could be tuned with a mount option per fs (e.g. often m/atimes don't need to be flushed every second). Would be easy to do as an addon if someone is interested. Signed-off-by:
Andi Kleen <ak@suse.de> Signed-off-by:
Andrew Morton <akpm@osdl.org> Signed-off-by:
Linus Torvalds <torvalds@osdl.org>
-
Martin Schwidefsky authored
I realized that the best way to get the sys_time/sys_stime problem fixed is to make sys_time 64 bit safe by using "time_t *" instead of "int *" and to introduce two proper compat functions compat_sys_time and compat_sys_stime. The prototype change of sys_time is transparent for 32 bit architectures because both "int" and "time_t" are 32 bit. For 64 bit the type change would be wrong but luckily no 64 bit architecture uses sys_time/sys_stime in 64 bit mode. The patch makes the following change: ia64 : Remove sys32_time, use compat_sys_time and add (!!) compat_sys_stime to compat syscall table. mips : Use compat_sys_time/compat_sys_stime in 32 bit syscall table. Add #ifdef magic to compile sys_time/sys_stime and compat_sys_time/compat_sys_stime only if needed. parisc : Remove sys32_time, use compat_sys_time and compat_sys_stime. ppc64 : remove sys32_time, ppc64_sys32_stime and ppc64_sys_stime. Use common compat_sys_time, compat_sys_stime and sys_stime. s390 : Use compat_sys_stime. Add #ifdef magic to compile sys_time/sys_stime and compat_sys_time/compat_sys_stime only if needed. sparc64 : Use compat_sys_time/compat_Sys_stime in 32 bit syscall table. um : Remove um_time and um_stime. Use common functions sys_time and sys_stime. This adds a CAP_SYS_TIME check to UMs stime call. x86_64 : Remove sys32_time. Use compat_sys_time and compat_sys_stime in 32 bit syscall table. The original stime bug is fixed for mips, parisc, s390, sparc64 and x86_64. Can the arch-maintainers please take a look at this? From: Martin Schwidefsky <schwidefsky@de.ibm.com> Convert compat_time_t to time_t in 32 bit emulation for sys_stime and consolidate all the different implementation of sys_time, sys_stime and their 32-bit emulation parts. Signed-off-by:Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by:
-
Adrian Bunk authored
For the kernel, it would be logical to use -ffreestanding. The kernel is not a hosted environment with a standard C library. The gcc option -ffreestanding is supported by both gcc 2.95 and 3.4, which covers the whole range of currently supported compilers. Regarding changes caused by this patch: Andi Kleen reported: Newer gcc rewrites sprintf(buf,"%s",str) to strcpy(buf,str) transparently. This is only true with unit-at-a-time (disabled on i386 but enabled on x86_64). The Linux kernel doesn't offer a standard C library, and such transparent replacements of kernel functions with builtins are quite fragile. Even with -ffreestanding, it's still possilble to explicitely use a gcc builtin if desired. Signed-off-by:
Adrian Bunk <bunk@stusta.de> Signed-off-by:
-
Alexander Nyberg authored
This fixes a theoretical bug indicated in: http://bugme.osdl.org/show_bug.cgi?id=240 It prevents overflow in case the required buffer is larger than the passed buffer. This I found to be the minimally intrusive change. Signed-off-by:
Alexander Nyberg <alexn@dsv.su.se> Signed-off-by:
-
Alex Tomas authored
1) intent of the patch is to get possibility to store EAs in the body of large inode. it saves space and improves performance in some cases 2) the patch is quite simple: it works the same way original xattr does, but using other storage (inode body). body has priority over separate block. original routines (ext3_xattr_get, ext3_xattr_list, ext3_xattr_set) are renamed to ext3_xattr_block_*. new routines that handle inode storate are added (ext3_xattr_ibody_get, ext3_xattr_ibody_list, ext3_xattr_ibody_set). routines ext3_xattr_get, ext3_xattr_list and ext3_xattr_set allow user to accesss both the storages transparently 3) the change makes sense on filesystem with inode size >= 256 bytes only. 2.4 kernels don't support such a filesystems, AFAIK. 2.6 kernels do support and ignore EAs stored in a body w/o the patch 4) debugfs and e2fsck need to be patched to deal with EAs in inode the patch will be sent later 5) testing results: a) Andrew Samba Master (tridge) has done successful tests b) we've been using ea-in-inode feature in Lustre for many months Signed-off-by:
Andreas Dilger <adilger@clusterfs.com> Signed-off-by:
Alex Tomas <alex@clusterfs.com> Signed-off-by:
-
Andrew Morton authored
We hold i_sem during the various sync() operations to prevent livelocks: if another thread is dirtying the file, a sync() may never return. Or at least, that used to be true when we were using the per-address_space page lists. Since writeback has used radix tree traversal it is not possible to livelock the sync() operations, because they only visit each page a single time. sync_page_range() (used by O_SYNC writes) has not been holding i_sem for quite some time, for the above reasons. The patch converts fsync(), fdatasync() and msync() to also not hold i_sem during the radix-tree-based writeback. Now, we _do_ still need to hold i_sem across the file->f_op->fsync() call, because that is still based on a list_head walk, and is still livelockable. But in the case of msync() I deliberately left i_sem untaken. This is because we're currently deadlockable in msync, because mmap_sem is already held, and mmap_sem nexts inside i_sem, due to direct-io.c. And yes, the ranking of down_read() veruss down() does matter: Task A Task B Task C down_read(rwsem) down(sem) down_write(rwsem) down(sem) down_read(rwsem) C's down_write() will cause B's down_read to block. B holds `sem', so A will never release `rwsem'. So the patch fixes a hard-to-hit triple-task deadlock, but adds a possible livelock in msync(). It is possible to fix sys_msync() so that it takes i_sem outside i_mmap_sem. Later. Signed-off-by:
-
Andrew Morton authored
We can call might_sleep() functions on the oops handling path (under do_exit). There seem little point in emitting spurious might_sleep() warnings into the logs after the kernel has oopsed. Signed-off-by:
-
Prasanna Meda authored
Bring the total_forks under tasklist_lock. When most of the fork code icluding nr_threads is moved to copy_process() from do_fork() code in 2.6, this is left out. Althought accuracy of total_forks is not important, it would be nice to add this. It does not involve additional cost, and the code will be cleaner if it is grouped with nr_threads. The difference is, total_forks will increase on fork, but nr_threads will increase on fork and decrease on the exit. I also moved extern decleration to sched.h from proc_misc.c. Signed-off-by:
Prasanna Meda <pmeda@akamai.com> Signed-off-by:
-
Li Shaohua authored
After resume from S3, 'date' shows time run too fast. Signed-off-by:
Li Shaohua <shaohua.li@intel.com> Signed-off-by:
-
Matthew Dobson authored
In the course of another patch I've been working on, I stumbled across some weirdness with some of the SD_*_INIT sched_domains initializers. A day or so of digging narrowed it down to the CPU_MASK_NONE initializer nested inside the sched_domain initializers. The errors I got were: kernel/sched.c:4812: error: initializer element is not constant kernel/sched.c:4812: error: (near initialization for `sched_domain_dummy') kernel/sched.c:4812: error: initializer element is not constant which was this line: static struct sched_domain sched_domain_dummy = SD_CPU_INIT; Janis Johnson, a GCC hacker, told me the following:
-
Stephen C. Tweedie authored
This patch improves ext3's ability to deal with corruption on-disk. If we try to delete a metadata block twice, we confuse ext3's internal revoke error-checking, resulting in a BUG(). But this can occur in practice due to a corrupt indirect block, so we should attempt to fail gracefully. Downgrade the assert failure to a JH_EXPECT_BH failure, and return EIO when it occurs. This is easily reproduced with a sample ext3 fs image containing an inode which references the same indirect block more than once. Deleting that inode will BUG() an unfixed kernel with: Assertion failure in journal_revoke() at fs/jbd/revoke.c:379: "!buffer_revoked(bh)" With the fix, ext3 recovers gracefully. Signed-off-by:
Stephen Tweedie <sct@redhat.com> Signed-off-by:
-
Stephen C. Tweedie authored
This patch improves ext3's ability to deal with corruption on-disk. If we ever get a corrupt inode or indirect block, then an attempt to delete it can end up trying to remove any block on the fs, including bitmap blocks. This can cause ext3 to assert-fail as we end up trying to do an ext3_forget on a buffer with b_committed_data set. The fix is to downgrade this to an IO error and journal abort, so that we take the filesystem readonly but don't bring down the whole kernel. Make J_EXPECT_JH() return a value so it can be easily tested and yet still retained as an assert failure if we build ext3 with full internal debugging enabled. Make journal_forget() return an error code so that in this case the error can be passed up to the caller. This is easily reproduced with a sample ext3 fs image containing an inode whose direct and indirect blocks refer to a block bitmap block. Allocating new blocks and then deleting that inode will BUG() with: Assertion failure in journal_forget() at fs/jbd/transaction.c:1228: "!jh->b_committed_data" With the fix, ext3 recovers gracefully. Signed-off-by:
-
Stephen C. Tweedie authored
This patch improves ext3's error logging when we encounter an on-disk corruption. Previously, a transaction (such as a truncate) which encountered many corruptions (eg. a single highly-corrupt indirect block) would emit copious "aborting transaction" errors to the log. Even worse, encountering an aborted journal can count as such an error, leading to a flood of spurious "aborting transaction: Journal has aborted" errors. With the fix, only emit that message on the first error. The patch also restores a missing \n in that printk path. Signed-off-by:
-
Adrian Bunk authored
All blk.h users were converted in 2.5, and at the same time blk.h began giving a warning. The patch below removes this obsolete file. Signed-off-by:
-
Corey Minyard authored
This patch removes some unneeded cruft that Adrian found, and also turns off the shutdown of the timer when removing the module. Since the timer is shutdown when the driver is closed (unless no way out is specified) this is unnecessary and defeats the no way out option. - remove some completely unused code - make some needlessly global code static - removal of some EXPORT_SYMBOL'ed code with zero users. - Removal of the timer shutdown on module removal Signed-off-by:
Corey Minyard <minyard@acm.org> Signed-off-by:
-
Robin Holt authored
Testing revealed long pauses of the entire system while autofs initiated umounts as a result of timing out the mounts. It was noticed that during a umount, the BKL is held while scanning the inode_list and removing and inodes that are candidates. This patch moves locking until after the first pass had gone through the inode_list. Testing revelead that on an ia64 machine with a filesystem that had 8.4 Million inodes, there were no observable pauses during the umount. This was down from over 4 seconds without this patch. Signed-Off-By:
Robin Holt <holt@sgi.com> Signed-off-by:
-
Christoph Hellwig authored
The only common field in irq_cpustat is __softirq_pending, i386 and ppc have some of their own. Remove all unused obsolete fields from various architectures. Signed-off-by:
-
Christoph Hellwig authored
This code is the same for all architectures with the following invariants: - arm gurantees irqs are disabled when calling irq_exit so it can call __do_softirq directly instead of do_softirq - arm26 is totally broken for about half a year, I didn't care for it - some architectures use softirq_pending(smp_processor_id()) instead of local_softirq_pending, but they always evaluate to the same This patch moves the out of line irq_exit implementation from kernel/irq/handle.c which depends on CONFIG_GENERIC_HARDIRQS to kernel/softirq.c which is always compiled, tweaks it for the arm special case and moves the irq_enter/irq_exit/nmi_enter/nmi_exit bits from asm-*/hardirq.h to linux/hardirq.h Signed-off-by:
-
Randy Dunlap authored
Fix module parameter quote handling. Module parameter strings (with spaces) are quoted like so: "modprm=this test" and not like this: modprm="this test" Signed-off-by:
Randy Dunlap <rddunlap@osdl.org> Signed-off-by:
-
Kirill Korotaev authored
This patch fixes incorrect address range check in do_getname(). Theoretically this can lead to do_getname() failure on kernel address space string on the TASK_SIZE boundary addresses when 4GB split is ON. (akpm: I don't see why this check exists at all, actually. afaict the only effect of removing it is that we'll then generate -EFAULT on a non-null-terminated pathname which ends exactly at TASK_SIZE). Signed-Off-By:
Kirill Korotaev <dev@sw.ru> Signed-off-by:
-
Jay Lan authored
This patch is to offer common accounting data collection method at memory usage for various accounting packages including BSD accounting, ELSA, CSA and any other acct packages that use a common layer of data collection. New struct fields are added to mm_struct to save high watermarks of rss usage as well as virtual memory usage. New struct fields are added to task_struct to collect accumulated rss usage and vm usages. These data are collected on per process basis. Signed-off-by:
Jay Lan <jlan@sgi.com> Signed-off-by:
-
Jay Lan authored
This patch is to offer common accounting data collection method at I/O for various accounting packages including BSD accounting, ELSA, CSA and any other acct packages that use a common layer of data collection. Patch is made to fs/read_write.c to collect per process data on character read/written in bytes and number of read/write syscalls made. New struct fields are added to task_struct to store the data. These data are collected on per process basis. Signed-off-by:
-
David Howells authored
The attached patch includes prio-tree support and adds cross-referencing of VMAs with address spaces back in, as is done under normal MMU Linux. Signed-Off-By:
David Howells <dhowells@redhat.com> Signed-off-by:
-
David Howells authored
The attached patch applies some further fixes and extensions to the nommu mmap implementation: (1) /proc/maps distinguishes shareable private mappings and real shared mappings by marking the former with 's' and the latter with 'S'. (2) Rearrange and optimise the checking portion of do_mmap_pgoff() to make it easier to follow. (3) Only set VM_SHARED on MAP_SHARED mappings. Its presence indicates that the backing memory is supplied by the underlying file or chardev. VM_MAYSHARE indicates that a VMA may be shared if it's a private VMA. The memory for a private VMA is allocated by do_mmap_pgoff() from a kmalloc slab and then the file contents are read into it before returning. (4) Permit MAP_SHARED + PROT_WRITE on memory-backed files[*] and chardevs to indicate a contiguous area of memory when its get_unmapped_area() is called if the backing fs/chardev is willing. [*] file->f_mapping->backing_dev_info->memory_backed == 1 (5) Require chardevs and files that support to provide a get_unmapped_area() file operation. (6) Made sure a private mapping of /dev/zero is possible. Shared mappings of /dev/zero are not currently supported because this'd need greater interaction of mmap with the chardev driver than is currently supported. (7) Add in some extra checks from mm/mmap.c: security, file having write access for a writable shared mapping, file not being in append mode. (8) Only account the mapping memory if it's allocated here; memory belonging to a shared chardev or file is not accounted. With this patch it should be possible to map contiguous flash files directly out of ROM simply by providing get_unmapped_area() for a read-only/shared mapping. I think that it might be worth splitting do_mmap_pgoff() up into smaller subfunctions: one to handle the checking, one to handle shared mappings and one to handle private mappings. Signed-Off-By: -
David Howells authored
The attached patch does the following things: (1) It uniquifies permitted overlapping VMAs (eg: MAP_SHARED on chardevs) in nommu_vma_tree. Identical entries break the assumptions on which rbtrees work. Since we don't need to share VMAs in this case, we uniquify such VMAs by using the pointer to the VMA. They're only kept in the tree for /proc/maps visibility. (2) Extracts VMA unlinking into its own function so that the source is adjacent to the VMA linking function. (3) No longer releases memory belonging to a shared chardev or file (the underlying driver is expected to provide mappable memory). (4) Frees the file attached to a VMA whether or not that VMA is shared or is a memory-mapped I/O mapping. Signed-Off-By: -
David Howells authored
The attached patch implements a nommu version of find_vma(). Signed-Off-By:
-
David Howells authored
The attached patch changes the PML4 bits of the FRV arch to the new PUD way. Signed-Off-By:
-
David Howells authored
The attached patch fixes the following issues with support for the FR55x CPUs: (1) The FR555 has a 64-byte cacheline size; everything else that we've come across has a 32-byte cacheline size. (2) Fix machine_restart() for FR55x. (3) Fix frv_cpu_suspend() for FR55x. Signed-Off-By:Mark Salter <msalter@redhat.com> Signed-off-by:
-
David Howells authored
The attached patch makes the following fixes to the frv arch: (1) pte_offset() should no longer be around; the fault handler should use pte_offset_kernel() instead when fixing up vmalloc misses. (2) The PGEs/PMEs do not hold PTEs. They have greater address resolution and fewer control bits. (3) The data access error pattern in ESR15.EC should be 10000 not 10100. Signed-Off-By: -
David Howells authored
The attached patch stops the FRV kernel-instruction-TLB-miss handler from setting the write-protect bit on a mapping entry when punting an entry from the mapping fast cache registers (DAMR1/IAMR1) to the TLB. This patch derives the WP value from the DAMPR1 register (which actually has a WP bit) rather than the IAMPR1 register (which does not). Signed-Off-By:
-
David Howells authored
The attached patch updates the FRV trap tables comment to make it more appropriate. Signed-Off-By:
-
David Howells authored
The attached patch gets rid of the perfctr_info syscall from the FRV arch now that its implementation has gone and it has been removed from the i386 arch and the i386 syscalls have been renumbered. Signed-Off-By:
-
David Howells authored
The attached patch does two things: (1) Implements the ext2/ext3 bitops in terms of the main bitops functions. (2) Changes the Minix bitops to use the ext2 bitops (LE) rather than the main bitops (BE). Signed-Off-By: -
David Howells authored
The attached patch fixes three debugging problems in the frv arch: (1) Single-stepping in userspace steps through into the kernel-mode interrupt handler when a hardware interrupt happens, and sometimes it gets past where the debug-mode handler would normally catch it. This patch extends the range of detected PC values. (2) When setting up the kernel-mode exception frame from the debug-mode handler for a userspace debugging event, we weren't setting the LR register to generate a return to the exception handler epilogue. (3) sys_ptrace() now needs to "put" the inferior task_struct not "free" it as was done in 2.4. Signed-Off-By: -
David Howells authored
The attached patch makes more syscalls available for the FR-V arch. Signed-Off-By:
-
David Howells authored
The attached patch changes the nommu bits of the FRV arch to incorporate the name changes made to the nommu core stuff. Signed-Off-By:
-
David Howells authored
The attached patch changes the nommu procfs routines to match the nommu changes in patch 1/1. This is an exercise in structure renaming and handling the fact that the list of VMAs in the system is now held together by vma->vm_rb. Signed-Off-By:
-
David Howells authored
The attached patch further changes the nommu stuff previously changed. These new changes do the following: (0) Some additional variables have been defined to make nommu even compile. (1) Get rid of the alternate vm_area_struct. The nommu mmap now uses the normal one. There's a refcount field added to the normal one, contingent on !CONFIG_MMU. (2) vm_rb is now used to keep track of the VMAs in an rbtree rather than adding a separate list. (3) mm_tblock_struct is now vm_list_struct. (4) put_vma() now calls vma->vm_ops->close() if available on nommu. (5) A dummy generic_file_vm_ops has been provided. It does nothing, but permits tiny-shmem to compile. tiny-shmem and ramfs still need attention, such that files contained therein can be mmapped shared-writably to some extent on nommu. Signed-Off-By: -
David Howells authored
The attached patch fixes the following problems in the ELF-FDPIC binfmt driver: (1) elf_fdpic_map_file() should be passed an mm_struct pointer, not NULL. (2) do_mmap() should be called with the mmap_sem held. (3) mm_struct::end_brk doesn't exist in 2.6 (debugging only). (4) Avoid debugging warnings by casting certain values to unsigned long before printing them. Signed-Off-By: -
David Howells authored
The attached patch adds a new binary format driver that allows a special variety of ELF to be used that permits the dynamic sections that comprise an executable, its dynamic loader and its shared libaries and its stack and data to be located anywhere within the address space. This is used to provide shared libraries and shared executables (at least, as far as the read-only dynamic sections go) on uClinux. Not only that, but the same binaries can be run on MMU linux without a problem. This is achieved by: (1) Passing loadmaps to the dynamic loader (or to a statically linked executable) to indicate the whereabouts of the various dynamic sections. (2) Using a GOT inside the program. (3) Passing setup_arg_pages() the stack pointer to be. (4) Allowing the arch greated control over how an executable is laid out in memory in MMU Linux. (5) Rewriting mm/nommu.c to support MAP_PRIVATE on files, thus allowing _mmap_ to handle sharing of private-readonly mappings. Signed-Off-By:
-
