2002-07-17 17:56:50

by Robert Love

[permalink] [raw]
Subject: [PATCH] 2.5-rmap: VM strict overcommit

The following patch implements VM strict overcommit on top of
2.5.26-akpm-rmap. Strict overcommit couples address space accounting
with a strict commit rule to ensure all allocated memory is backed and
consequently we never OOM. All memory failures should be pushed into
the allocation routines - a page access should never result in a process
kill.

The new strict overcommit policies are implemented via sysctl.

This is based on Alan Cox's work in 2.4-ac with some cleanup and a new
overcommit mode for swapless machines. Hugh Dickins also contributed
some fixes for shmfs.

While this patch is diffed off the rmap VM, there is nothing inherently
in it that requires rmap. Additionally, it is relatively low-impact on
other code and does not change the behavior of the system in the default
overcommit policy.

Questions, comments, and plastic zebra figurines are welcome.

Robert Love

diff -urN linux-2.5.26-rmap/Documentation/sysctl/vm.txt linux/Documentation/sysctl/vm.txt
--- linux-2.5.26-rmap/Documentation/sysctl/vm.txt Wed Jul 17 10:44:09 2002
+++ linux/Documentation/sysctl/vm.txt Wed Jul 17 10:45:47 2002
@@ -78,7 +78,11 @@
programs that malloc() huge amounts of memory "just-in-case"
and don't use much of it.

-Look at: mm/mmap.c::vm_enough_memory() for more information.
+Flag values of 2 - 4 introduce a new "strict overcommit"
+policy that attempt to prevent any overcommit of memory.
+
+See Documentation/vm/overcommit-accounting and
+mm/mmap.c::vm_enough_memory() for more information.

==============================================================

diff -urN linux-2.5.26-rmap/Documentation/vm/overcommit-accounting linux/Documentation/vm/overcommit-accounting
--- linux-2.5.26-rmap/Documentation/vm/overcommit-accounting Wed Dec 31 16:00:00 1969
+++ linux/Documentation/vm/overcommit-accounting Wed Jul 17 10:45:47 2002
@@ -0,0 +1,77 @@
+The Linux kernel supports four overcommit handling modes
+
+0 - Heuristic overcommit handling. Obvious overcommits of
+ address space are refused. Used for a typical system. It
+ ensures a seriously wild allocation fails while allowing
+ overcommit to reduce swap usage. This is the default.
+
+1 - No overcommit handling. Appropriate for some scientific
+ applications.
+
+2 - (NEW) swapless strict overcommit. The total address space
+ commit for the system is not permitted to exceed 95% of
+ free memory. This mode utilizes the new stricter accounting
+ but does not impose a very strict rule. It is possible that
+ the system could kill a process accessing pages in certain
+ cases. If mode 3 is too strict when no swap is present
+ this is the best you can do.
+
+3 - (NEW) strict overcommit. The total address space commit
+ for the system is not permitted to exceed swap + half ram.
+ In almost all situations this means a process will not be
+ killed while accessing pages but only by malloc failures
+ that are reported back by the kernel mmap/brk code.
+
+4 - (NEW) paranoid overcommit. The total address space commit
+ for the system is not permitted to exceed swap. The machine
+ will never kill a process accessing pages it has mapped
+ except due to a bug (ie report it!).
+
+Gotchas
+-------
+
+The C language stack growth does an implicit mremap. If you want absolute
+guarantees and run close to the edge you MUST mmap your stack for the
+largest size you think you will need. For typical stack usage is does
+not matter much but its a corner case if you really really care
+
+In modes 2 and 3 the MAP_NORESERVE flag is ignored.
+
+
+How It Works
+------------
+
+The overcommit is based on the following rules
+
+For a file backed map
+ SHARED or READ only - 0 cost (the file is the map not swap)
+
+ WRITABLE SHARED - size of mapping per instance
+
+For a direct map
+ SHARED or READ only - size of mapping
+ PRIVATE WRITEABLE - size of mapping per instance
+
+Additional accounting
+ Pages made writable copies by mmap
+ shmfs memory drawn from the same pool
+
+Status
+------
+
+o We account mmap memory mappings
+o We account mprotect changes in commit
+o We account mremap changes in size
+o We account brk
+o We account munmap
+o We report the commit status in /proc
+o Account and check on fork
+o Review stack handling/building on exec
+o SHMfs accounting
+o Implement actual limit enforcement
+
+To Do
+-----
+o Account ptrace pages (this is hard)
+o Account for shared anonymous mappings properly
+ - right now we account them per instance
diff -urN linux-2.5.26-rmap/fs/exec.c linux/fs/exec.c
--- linux-2.5.26-rmap/fs/exec.c Wed Jul 17 10:44:19 2002
+++ linux/fs/exec.c Wed Jul 17 10:45:47 2002
@@ -313,8 +313,13 @@

mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
if (!mpnt)
- return -ENOMEM;
-
+ return -ENOMEM;
+
+ if (!vm_enough_memory((STACK_TOP - (PAGE_MASK & (unsigned long) bprm->p))>>PAGE_SHIFT)) {
+ kmem_cache_free(vm_area_cachep, mpnt);
+ return -ENOMEM;
+ }
+
down_write(&current->mm->mmap_sem);
{
mpnt->vm_mm = current->mm;
diff -urN linux-2.5.26-rmap/fs/proc/proc_misc.c linux/fs/proc/proc_misc.c
--- linux-2.5.26-rmap/fs/proc/proc_misc.c Wed Jul 17 10:44:18 2002
+++ linux/fs/proc/proc_misc.c Wed Jul 17 10:45:47 2002
@@ -126,11 +126,13 @@
return proc_calc_metrics(page, start, off, count, eof, len);
}

+extern atomic_t vm_committed_space;
+
static int meminfo_read_proc(char *page, char **start, off_t off,
int count, int *eof, void *data)
{
struct sysinfo i;
- int len;
+ int len, committed;
struct page_state ps;

get_page_state(&ps);
@@ -140,6 +142,7 @@
#define K(x) ((x) << (PAGE_SHIFT - 10))
si_meminfo(&i);
si_swapinfo(&i);
+ committed = atomic_read(&vm_committed_space);

/*
* Tagged format, for easy grepping and expansion.
@@ -159,7 +162,8 @@
"SwapTotal: %8lu kB\n"
"SwapFree: %8lu kB\n"
"Dirty: %8lu kB\n"
- "Writeback: %8lu kB\n",
+ "Writeback: %8lu kB\n"
+ "Committed_AS: %8u kB\n",
K(i.totalram),
K(i.freeram),
K(i.sharedram),
@@ -174,7 +178,8 @@
K(i.totalswap),
K(i.freeswap),
K(ps.nr_dirty),
- K(ps.nr_writeback)
+ K(ps.nr_writeback),
+ K(committed)
);

return proc_calc_metrics(page, start, off, count, eof, len);
diff -urN linux-2.5.26-rmap/include/linux/mm.h linux/include/linux/mm.h
--- linux-2.5.26-rmap/include/linux/mm.h Wed Jul 17 10:43:54 2002
+++ linux/include/linux/mm.h Wed Jul 17 10:45:47 2002
@@ -103,8 +103,9 @@
#define VM_DONTCOPY 0x00020000 /* Do not copy this vma on fork */
#define VM_DONTEXPAND 0x00040000 /* Cannot expand with mremap() */
#define VM_RESERVED 0x00080000 /* Don't unmap it from swap_out */
+#define VM_ACCOUNT 0x00100000 /* Is a VM accounted object */

-#define VM_STACK_FLAGS (0x00000100 | VM_DATA_DEFAULT_FLAGS)
+#define VM_STACK_FLAGS (0x00000100 | VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT)

#define VM_READHINTMASK (VM_SEQ_READ | VM_RAND_READ)
#define VM_ClearReadHint(v) (v)->vm_flags &= ~VM_READHINTMASK
@@ -427,7 +428,7 @@
return ret;
}

-extern int do_munmap(struct mm_struct *, unsigned long, size_t);
+extern int do_munmap(struct mm_struct *, unsigned long, size_t, int);

extern unsigned long do_brk(unsigned long, unsigned long);

@@ -470,31 +471,8 @@
void page_cache_readaround(struct file *file, unsigned long offset);
void handle_ra_thrashing(struct file *file);

-/* vma is the first one with address < vma->vm_end,
- * and even address < vma->vm_start. Have to extend vma. */
-static inline int expand_stack(struct vm_area_struct * vma, unsigned long address)
-{
- unsigned long grow;
-
- /*
- * vma->vm_start/vm_end cannot change under us because the caller is required
- * to hold the mmap_sem in write mode. We need to get the spinlock only
- * before relocating the vma range ourself.
- */
- address &= PAGE_MASK;
- grow = (vma->vm_start - address) >> PAGE_SHIFT;
- if (vma->vm_end - address > current->rlim[RLIMIT_STACK].rlim_cur ||
- ((vma->vm_mm->total_vm + grow) << PAGE_SHIFT) > current->rlim[RLIMIT_AS].rlim_cur)
- return -ENOMEM;
- spin_lock(&vma->vm_mm->page_table_lock);
- vma->vm_start = address;
- vma->vm_pgoff -= grow;
- vma->vm_mm->total_vm += grow;
- if (vma->vm_flags & VM_LOCKED)
- vma->vm_mm->locked_vm += grow;
- spin_unlock(&vma->vm_mm->page_table_lock);
- return 0;
-}
+/* Do stack extension */
+extern int expand_stack(struct vm_area_struct * vma, unsigned long address);

/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
extern struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr);
diff -urN linux-2.5.26-rmap/include/linux/mman.h linux/include/linux/mman.h
--- linux-2.5.26-rmap/include/linux/mman.h Wed Jul 17 10:43:55 2002
+++ linux/include/linux/mman.h Wed Jul 17 10:45:47 2002
@@ -6,4 +6,7 @@
#define MREMAP_MAYMOVE 1
#define MREMAP_FIXED 2

+extern int vm_enough_memory(long pages);
+extern void vm_unacct_memory(long pages);
+
#endif /* _LINUX_MMAN_H */
diff -urN linux-2.5.26-rmap/ipc/shm.c linux/ipc/shm.c
--- linux-2.5.26-rmap/ipc/shm.c Wed Jul 17 10:44:19 2002
+++ linux/ipc/shm.c Wed Jul 17 10:45:47 2002
@@ -681,7 +681,7 @@
shmdnext = shmd->vm_next;
if (shmd->vm_ops == &shm_vm_ops
&& shmd->vm_start - (shmd->vm_pgoff << PAGE_SHIFT) == (ulong) shmaddr) {
- do_munmap(mm, shmd->vm_start, shmd->vm_end - shmd->vm_start);
+ do_munmap(mm, shmd->vm_start, shmd->vm_end - shmd->vm_start, 1);
retval = 0;
}
}
diff -urN linux-2.5.26-rmap/kernel/fork.c linux/kernel/fork.c
--- linux-2.5.26-rmap/kernel/fork.c Wed Jul 17 10:44:19 2002
+++ linux/kernel/fork.c Wed Jul 17 10:45:47 2002
@@ -25,6 +25,7 @@
#include <linux/binfmts.h>
#include <linux/fs.h>
#include <linux/mm.h>
+#include <linux/mman.h>

#include <asm/pgtable.h>
#include <asm/pgalloc.h>
@@ -181,6 +182,7 @@
{
struct vm_area_struct * mpnt, *tmp, **pprev;
int retval;
+ unsigned long charge = 0;

flush_cache_mm(current->mm);
mm->locked_vm = 0;
@@ -208,6 +210,17 @@
retval = -ENOMEM;
if(mpnt->vm_flags & VM_DONTCOPY)
continue;
+
+ /*
+ * FIXME: shared writable map accounting should be one off
+ */
+ if (mpnt->vm_flags & VM_ACCOUNT) {
+ unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
+ if (!vm_enough_memory(len))
+ goto fail_nomem;
+ charge += len;
+ }
+
tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
if (!tmp)
goto fail_nomem;
@@ -248,9 +261,12 @@
retval = 0;
build_mmap_rb(mm);

-fail_nomem:
+out:
flush_tlb_mm(current->mm);
return retval;
+fail_nomem:
+ vm_unacct_memory(charge);
+ goto out;
}

spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
diff -urN linux-2.5.26-rmap/mm/mmap.c linux/mm/mmap.c
--- linux-2.5.26-rmap/mm/mmap.c Wed Jul 17 10:44:19 2002
+++ linux/mm/mmap.c Wed Jul 17 10:45:47 2002
@@ -1,8 +1,11 @@
/*
- * linux/mm/mmap.c
+ * mm/mmap.c
*
* Written by obz.
+ *
+ * Address space accounting code <[email protected]>
*/
+
#include <linux/slab.h>
#include <linux/shm.h>
#include <linux/mman.h>
@@ -49,51 +52,94 @@
};

int sysctl_overcommit_memory;
+atomic_t vm_committed_space = ATOMIC_INIT(0);

-/* Check that a process has enough memory to allocate a
- * new virtual mapping.
+/*
+ * Check that a process has enough memory to allocate a new virtual
+ * mapping. 1 means there is enough memory for the allocation to
+ * succeed and 0 implies there is not.
+ *
+ * We currently support four overcommit policies, which are set via the
+ * overcommit_memory sysctl. See Documentation/vm/overcommit-acounting
+ *
+ * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
*/
int vm_enough_memory(long pages)
{
- /* Stupid algorithm to decide if we have enough memory: while
- * simple, it hopefully works in most obvious cases.. Easy to
- * fool it, but this should catch most mistakes.
- */
- /* 23/11/98 NJC: Somewhat less stupid version of algorithm,
- * which tries to do "TheRightThing". Instead of using half of
- * (buffers+cache), use the minimum values. Allow an extra 2%
- * of num_physpages for safety margin.
- */
+ unsigned long free, allowed;

- unsigned long free;
-
- /* Sometimes we want to use more memory than we have. */
- if (sysctl_overcommit_memory)
- return 1;
-
- /* The page cache contains buffer pages these days.. */
- free = get_page_cache_size();
- free += nr_free_pages();
- free += nr_swap_pages;
+ atomic_add(pages, &vm_committed_space);

- /*
- * This double-counts: the nrpages are both in the page-cache
- * and in the swapper space. At the same time, this compensates
- * for the swap-space over-allocation (ie "nr_swap_pages" being
- * too small.
+ /*
+ * Sometimes we want to use more memory than we have
*/
- free += swapper_space.nrpages;
+ if (sysctl_overcommit_memory == 1)
+ return 1;

- /*
- * The code below doesn't account for free space in the inode
- * and dentry slab cache, slab cache fragmentation, inodes and
- * dentries which will become freeable under VM load, etc.
- * Lets just hope all these (complex) factors balance out...
- */
- free += (dentry_stat.nr_unused * sizeof(struct dentry)) >> PAGE_SHIFT;
- free += (inodes_stat.nr_unused * sizeof(struct inode)) >> PAGE_SHIFT;
+ if (sysctl_overcommit_memory == 0) {
+ free = get_page_cache_size();
+ free += nr_free_pages();
+ free += nr_swap_pages;
+
+ /*
+ * This double-counts: the nrpages are both in the
+ * page-cache and in the swapper space. At the same time,
+ * this compensates for the swap-space over-allocation
+ * (ie "nr_swap_pages" being too small).
+ */
+ free += swapper_space.nrpages;
+
+ /*
+ * The code below doesn't account for free space in the
+ * inode and dentry slab cache, slab cache fragmentation,
+ * inodes and dentries which will become freeable under
+ * VM load, etc. Lets just hope all these (complex)
+ * factors balance out...
+ */
+ free += (dentry_stat.nr_unused * sizeof(struct dentry)) >>
+ PAGE_SHIFT;
+ free += (inodes_stat.nr_unused * sizeof(struct inode)) >>
+ PAGE_SHIFT;
+
+ if (free > pages)
+ return 1;
+ atomic_sub(pages, &vm_committed_space);
+ return 0;
+ }
+
+ if (sysctl_overcommit_memory == 2) {
+ /*
+ * FIXME: need to add arch hooks to get the bits we need
+ * without the higher overhead crap
+ */
+ struct sysinfo i;
+ si_meminfo(&i);
+ allowed = i.totalram - (i.totalram / 20);
+ } else if (sysctl_overcommit_memory == 3) {
+ struct sysinfo i;
+ si_meminfo(&i);
+ allowed = total_swap_pages + (i.totalram >> 1);
+ } else /* sysctl_overcommit_memory == 4 */
+ allowed = total_swap_pages;
+
+ if (atomic_read(&vm_committed_space) < allowed)
+ return 1;
+
+ atomic_sub(pages, &vm_committed_space);
+
+ return 0;
+}

- return free > pages;
+void inline vm_unacct_memory(long pages)
+{
+ atomic_sub(pages, &vm_committed_space);
+}
+
+void vm_unacct_vma(struct vm_area_struct *vma)
+{
+ int len = vma->vm_end - vma->vm_start;
+ if (vma->vm_flags & VM_ACCOUNT)
+ vm_unacct_memory(len >> PAGE_SHIFT);
}

/* Remove one vm structure from the inode's i_mapping address space. */
@@ -162,7 +208,7 @@

/* Always allow shrinking brk. */
if (brk <= mm->brk) {
- if (!do_munmap(mm, newbrk, oldbrk-newbrk))
+ if (!do_munmap(mm, newbrk, oldbrk-newbrk, 1))
goto set_brk;
goto out;
}
@@ -176,10 +222,6 @@
if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
goto out;

- /* Check if we have enough memory.. */
- if (!vm_enough_memory((newbrk-oldbrk) >> PAGE_SHIFT))
- goto out;
-
/* Ok, looks good - let it rip. */
if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
goto out;
@@ -385,8 +427,9 @@
return 0;
}

-unsigned long do_mmap_pgoff(struct file * file, unsigned long addr, unsigned long len,
- unsigned long prot, unsigned long flags, unsigned long pgoff)
+unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
+ unsigned long len, unsigned long prot,
+ unsigned long flags, unsigned long pgoff)
{
struct mm_struct * mm = current->mm;
struct vm_area_struct * vma, * prev;
@@ -394,6 +437,7 @@
int correct_wcount = 0;
int error;
rb_node_t ** rb_link, * rb_parent;
+ unsigned long charged = 0;

if (file && (!file->f_op || !file->f_op->mmap))
return -ENODEV;
@@ -480,7 +524,7 @@
munmap_back:
vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
if (vma && vma->vm_start < addr + len) {
- if (do_munmap(mm, addr, len))
+ if (do_munmap(mm, addr, len, 1))
return -ENOMEM;
goto munmap_back;
}
@@ -490,11 +534,17 @@
> current->rlim[RLIMIT_AS].rlim_cur)
return -ENOMEM;

+ if (sysctl_overcommit_memory > 1)
+ vm_flags &= ~MAP_NORESERVE;
+
/* Private writable mapping? Check memory availability.. */
- if ((vm_flags & (VM_SHARED | VM_WRITE)) == VM_WRITE &&
- !(flags & MAP_NORESERVE) &&
- !vm_enough_memory(len >> PAGE_SHIFT))
- return -ENOMEM;
+ if ((((vm_flags & (VM_SHARED | VM_WRITE)) == VM_WRITE) ||
+ (file == NULL)) && !(flags & MAP_NORESERVE)) {
+ charged = len >> PAGE_SHIFT;
+ if (!vm_enough_memory(charged))
+ return -ENOMEM;
+ vm_flags |= VM_ACCOUNT;
+ }

/* Can we just expand an old anonymous mapping? */
if (!file && !(vm_flags & VM_SHARED) && rb_parent)
@@ -506,8 +556,9 @@
* not unmapped, but the maps are removed from the list.
*/
vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
+ error = -ENOMEM;
if (!vma)
- return -ENOMEM;
+ goto unacct_error;

vma->vm_mm = mm;
vma->vm_start = addr;
@@ -570,6 +621,9 @@
zap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start);
free_vma:
kmem_cache_free(vm_area_cachep, vma);
+unacct_error:
+ if (charged)
+ vm_unacct_memory(charged);
return error;
}

@@ -699,6 +753,45 @@
return prev ? prev->vm_next : vma;
}

+/*
+ * vma is the first one with address < vma->vm_end,
+ * and even address < vma->vm_start. Have to extend vma.
+ */
+int expand_stack(struct vm_area_struct * vma, unsigned long address)
+{
+ unsigned long grow;
+
+ /*
+ * vma->vm_start/vm_end cannot change under us because the caller
+ * is required to hold the mmap_sem in write mode. We need to get
+ * the spinlock only before relocating the vma range ourself.
+ */
+ address &= PAGE_MASK;
+ spin_lock(&vma->vm_mm->page_table_lock);
+ grow = (vma->vm_start - address) >> PAGE_SHIFT;
+
+ /* Overcommit.. */
+ if(!vm_enough_memory(grow)) {
+ spin_unlock(&vma->vm_mm->page_table_lock);
+ return -ENOMEM;
+ }
+
+ if (vma->vm_end - address > current->rlim[RLIMIT_STACK].rlim_cur ||
+ ((vma->vm_mm->total_vm + grow) << PAGE_SHIFT) >
+ current->rlim[RLIMIT_AS].rlim_cur) {
+ spin_unlock(&vma->vm_mm->page_table_lock);
+ vm_unacct_memory(grow);
+ return -ENOMEM;
+ }
+ vma->vm_start = address;
+ vma->vm_pgoff -= grow;
+ vma->vm_mm->total_vm += grow;
+ if (vma->vm_flags & VM_LOCKED)
+ vma->vm_mm->locked_vm += grow;
+ spin_unlock(&vma->vm_mm->page_table_lock);
+ return 0;
+}
+
#ifdef ARCH_STACK_GROWSUP
struct vm_area_struct * find_extend_vma(struct mm_struct * mm, unsigned long addr)
{
@@ -824,7 +917,6 @@
kmem_cache_free(vm_area_cachep, area);
}

-
/*
* Update the VMA and inode share lists.
*
@@ -851,19 +943,25 @@
struct vm_area_struct *mpnt,
struct vm_area_struct *prev,
unsigned long start,
- unsigned long end)
+ unsigned long end,
+ int acct)
{
mmu_gather_t *tlb;

tlb = tlb_gather_mmu(mm, 0);

do {
- unsigned long from, to;
+ unsigned long from, to, len;

from = start < mpnt->vm_start ? mpnt->vm_start : start;
to = end > mpnt->vm_end ? mpnt->vm_end : end;

unmap_page_range(tlb, mpnt, from, to);
+
+ if (acct && (mpnt->vm_flags & VM_ACCOUNT)) {
+ len = to - from;
+ vm_unacct_memory(len >> PAGE_SHIFT);
+ }
} while ((mpnt = mpnt->vm_next) != NULL);

free_pgtables(tlb, prev, start, end);
@@ -941,7 +1039,7 @@
* work. This now handles partial unmappings.
* Jeremy Fitzhardine <[email protected]>
*/
-int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
+int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, int acct)
{
unsigned long end;
struct vm_area_struct *mpnt, *prev, *last;
@@ -985,7 +1083,7 @@
*/
spin_lock(&mm->page_table_lock);
mpnt = touched_by_munmap(mm, mpnt, prev, end);
- unmap_region(mm, mpnt, prev, start, end);
+ unmap_region(mm, mpnt, prev, start, end, acct);
spin_unlock(&mm->page_table_lock);

/* Fix up all other VM information */
@@ -1000,7 +1098,7 @@
struct mm_struct *mm = current->mm;

down_write(&mm->mmap_sem);
- ret = do_munmap(mm, addr, len);
+ ret = do_munmap(mm, addr, len, 1);
up_write(&mm->mmap_sem);
return ret;
}
@@ -1037,7 +1135,7 @@
munmap_back:
vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
if (vma && vma->vm_start < addr + len) {
- if (do_munmap(mm, addr, len))
+ if (do_munmap(mm, addr, len, 1))
return -ENOMEM;
goto munmap_back;
}
@@ -1053,7 +1151,7 @@
if (!vm_enough_memory(len >> PAGE_SHIFT))
return -ENOMEM;

- flags = VM_DATA_DEFAULT_FLAGS | mm->def_flags;
+ flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;

/* Can we just expand an old anonymous mapping? */
if (rb_parent && vma_merge(mm, prev, rb_parent, addr, addr + len, flags))
@@ -1063,8 +1161,10 @@
* create a vma struct for an anonymous mapping
*/
vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
- if (!vma)
+ if (!vma) {
+ vm_unacct_memory(len >> PAGE_SHIFT);
return -ENOMEM;
+ }

vma->vm_mm = mm;
vma->vm_start = addr;
@@ -1120,6 +1220,13 @@
unsigned long start = mpnt->vm_start;
unsigned long end = mpnt->vm_end;

+ /*
+ * If the VMA has been charged for, account for its
+ * removal
+ */
+ if (mpnt->vm_flags & VM_ACCOUNT)
+ vm_unacct_vma(mpnt);
+
mm->map_count--;
unmap_page_range(tlb, mpnt, start, end);
mpnt = mpnt->vm_next;
diff -urN linux-2.5.26-rmap/mm/mprotect.c linux/mm/mprotect.c
--- linux-2.5.26-rmap/mm/mprotect.c Wed Jul 17 10:44:19 2002
+++ linux/mm/mprotect.c Wed Jul 17 10:45:47 2002
@@ -1,7 +1,10 @@
/*
- * linux/mm/mprotect.c
+ * mm/mprotect.c
*
* (C) Copyright 1994 Linus Torvalds
+ *
+ * Address space accounting code <[email protected]>
+ * (C) Copyright 2002 Red Hat Inc, All Rights Reserved
*/
#include <linux/mm.h>
#include <linux/slab.h>
@@ -247,6 +250,7 @@
{
pgprot_t newprot;
int error;
+ unsigned long charged = 0;

if (newflags == vma->vm_flags) {
*pprev = vma;
@@ -263,9 +267,18 @@
else
error = mprotect_fixup_middle(vma, pprev, start, end, newflags, newprot);

- if (error)
+ if (error) {
+ if (newflags & PROT_WRITE)
+ vm_unacct_memory(charged);
return error;
+ }

+ /*
+ * Delayed accounting for reduction of memory use - done last to
+ * avoid allocation races
+ */
+ if (charged && !(newflags & PROT_WRITE))
+ vm_unacct_memory(charged);
change_protection(vma, start, end, newprot);
return 0;
}
diff -urN linux-2.5.26-rmap/mm/mremap.c linux/mm/mremap.c
--- linux-2.5.26-rmap/mm/mremap.c Wed Jul 17 10:44:19 2002
+++ linux/mm/mremap.c Wed Jul 17 10:45:47 2002
@@ -1,7 +1,10 @@
/*
- * linux/mm/remap.c
+ * mm/remap.c
*
* (C) Copyright 1996 Linus Torvalds
+ *
+ * Address space accounting code <[email protected]>
+ * (C) Copyright 2002 Red Hat Inc, All Rights Reserved
*/

#include <linux/mm.h>
@@ -18,8 +21,6 @@
#include <asm/cacheflush.h>
#include <asm/tlbflush.h>

-extern int vm_enough_memory(long pages);
-
static inline pte_t *get_one_pte_map_nested(struct mm_struct *mm, unsigned long addr)
{
pgd_t * pgd;
@@ -209,7 +210,11 @@
new_vma->vm_ops->open(new_vma);
insert_vm_struct(current->mm, new_vma);
}
- do_munmap(current->mm, addr, old_len);
+ /*
+ * The old VMA has been accounted for,
+ * don't double account
+ */
+ do_munmap(current->mm, addr, old_len, 0);
current->mm->total_vm += new_len >> PAGE_SHIFT;
if (new_vma->vm_flags & VM_LOCKED) {
current->mm->locked_vm += new_len >> PAGE_SHIFT;
@@ -224,6 +229,8 @@
return -ENOMEM;
}

+extern int sysctl_overcommit_memory; /* FIXME!! */
+
/*
* Expand (or shrink) an existing mapping, potentially moving it at the
* same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
@@ -237,6 +244,7 @@
{
struct vm_area_struct *vma;
unsigned long ret = -EINVAL;
+ unsigned long charged = 0;

if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))
goto out;
@@ -266,16 +274,17 @@
if ((addr <= new_addr) && (addr+old_len) > new_addr)
goto out;

- do_munmap(current->mm, new_addr, new_len);
+ do_munmap(current->mm, new_addr, new_len, 1);
}

/*
* Always allow a shrinking remap: that just unmaps
* the unnecessary pages..
+ * do_munmap does all the needed commit accounting
*/
ret = addr;
if (old_len >= new_len) {
- do_munmap(current->mm, addr+new_len, old_len - new_len);
+ do_munmap(current->mm, addr+new_len, old_len - new_len, 1);
if (!(flags & MREMAP_FIXED) || (new_addr == addr))
goto out;
}
@@ -305,11 +314,14 @@
if ((current->mm->total_vm << PAGE_SHIFT) + (new_len - old_len)
> current->rlim[RLIMIT_AS].rlim_cur)
goto out;
- /* Private writable mapping? Check memory availability.. */
- if ((vma->vm_flags & (VM_SHARED | VM_WRITE)) == VM_WRITE &&
- !(flags & MAP_NORESERVE) &&
- !vm_enough_memory((new_len - old_len) >> PAGE_SHIFT))
- goto out;
+
+ if (sysctl_overcommit_memory > 1)
+ flags &= ~MAP_NORESERVE;
+ if (vma->vm_flags & VM_ACCOUNT) {
+ charged = (new_len - old_len) >> PAGE_SHIFT;
+ if (!vm_enough_memory(charged))
+ goto out_nc;
+ }

/* old_len exactly to the end of the area..
* And we're not relocating the area.
@@ -356,6 +368,9 @@
ret = move_vma(vma, addr, old_len, new_len, new_addr);
}
out:
+ if (ret & ~PAGE_MASK)
+ vm_unacct_memory(charged);
+out_nc:
return ret;
}

diff -urN linux-2.5.26-rmap/mm/shmem.c linux/mm/shmem.c
--- linux-2.5.26-rmap/mm/shmem.c Wed Jul 17 10:44:19 2002
+++ linux/mm/shmem.c Wed Jul 17 10:45:47 2002
@@ -5,7 +5,8 @@
* 2000 Transmeta Corp.
* 2000-2001 Christoph Rohland
* 2000-2001 SAP AG
- *
+ * 2002 Red Hat Inc.
+ *
* This file is released under the GPL.
*/

@@ -21,6 +22,7 @@
#include <linux/devfs_fs_kernel.h>
#include <linux/fs.h>
#include <linux/mm.h>
+#include <linux/mman.h>
#include <linux/file.h>
#include <linux/swap.h>
#include <linux/pagemap.h>
@@ -35,6 +37,13 @@
#define TMPFS_MAGIC 0x01021994

#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
+#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
+
+#define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + ENTRIES_PER_PAGE * \
+ (ENTRIES_PER_PAGE/2) * (ENTRIES_PER_PAGE+1))
+#define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << \
+ PAGE_CACHE_SHIFT)
+#define VM_ACCT(size) (((size) + PAGE_CACHE_SIZE - 1) >> PAGE_SHIFT)

static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
{
@@ -54,8 +63,6 @@

static struct page *shmem_getpage_locked(struct shmem_inode_info *, struct inode *, unsigned long);

-#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
-
/*
* shmem_recalc_inode - recalculate the size of an inode
*
@@ -132,8 +139,6 @@
* +-> 52-55
*/

-#define SHMEM_MAX_BLOCKS (SHMEM_NR_DIRECT + ENTRIES_PER_PAGE * ENTRIES_PER_PAGE/2*(ENTRIES_PER_PAGE+1))
-
static swp_entry_t * shmem_swp_entry (struct shmem_inode_info *info, unsigned long index, unsigned long page)
{
unsigned long offset;
@@ -186,7 +191,7 @@
unsigned long page = 0;
swp_entry_t * res;

- if (index >= SHMEM_MAX_BLOCKS)
+ if (index >= SHMEM_MAX_INDEX)
return ERR_PTR(-EFBIG);

if (info->next_index <= index)
@@ -358,15 +363,46 @@
up(&info->sem);
}

+static int shmem_setattr(struct dentry * dentry, struct iattr *attr)
+{
+ struct inode *inode = dentry->d_inode;
+ long change = 0;
+ int error;
+
+ if ((attr->ia_valid & ATTR_SIZE) && (attr->ia_size <= SHMEM_MAX_BYTES)){
+ /*
+ * Account swap file usage based on new file size,
+ * but just let vmtruncate fail on out-of-range sizes.
+ */
+ long change = (attr->ia_size>>PAGE_SHIFT) - (inode->i_size >> PAGE_SHIFT);
+
+ change = VM_ACCT(attr->ia_size) - VM_ACCT(inode->i_size);
+ if (change > 0) {
+ if (!vm_enough_memory(change))
+ return -ENOMEM;
+ } else
+ vm_unacct_memory(-change);
+ }
+
+ error = inode_change_ok(inode, attr);
+ if (!error)
+ error = inode_setattr(inode, attr);
+ if (error && change)
+ vm_unacct_memory(change);
+ return error;
+}
+
+
static void shmem_delete_inode(struct inode * inode)
{
struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);

- inode->i_size = 0;
- if (inode->i_op->truncate == shmem_truncate){
+ if (inode->i_op->truncate == shmem_truncate) {
spin_lock (&shmem_ilock);
list_del (&SHMEM_I(inode)->list);
spin_unlock (&shmem_ilock);
+ vm_unacct_memory(VM_ACCT(inode->i_size));
+ inode->i_size = 0;
shmem_truncate (inode);
}
spin_lock (&sbinfo->stat_lock);
@@ -823,6 +859,7 @@
unsigned long written;
long status;
int err;
+ loff_t maxpos;

if ((ssize_t) count < 0)
return -EINVAL;
@@ -835,12 +872,12 @@
pos = *ppos;
err = -EINVAL;
if (pos < 0)
- goto out;
+ goto out_nc;

err = file->f_error;
if (err) {
file->f_error = 0;
- goto out;
+ goto out_nc;
}

written = 0;
@@ -848,6 +885,17 @@
if (file->f_flags & O_APPEND)
pos = inode->i_size;

+ maxpos = inode->i_size;
+ if (pos + count > inode->i_size) {
+ maxpos = pos + count;
+ if (maxpos > SHMEM_MAX_BYTES)
+ maxpos = SHMEM_MAX_BYTES;
+ if (!vm_enough_memory(VM_ACCT(maxpos) - VM_ACCT(inode->i_size))){
+ err = -ENOMEM;
+ goto out_nc;
+ }
+ }
+
/*
* Check whether we've reached the file size limit.
*/
@@ -937,6 +985,10 @@

err = written ? written : status;
out:
+ /* Short writes give back address space */
+ if (inode->i_size != maxpos)
+ vm_unacct_memory(VM_ACCT(maxpos) - VM_ACCT(inode->i_size));
+out_nc:
up(&inode->i_sem);
return err;
fail_write:
@@ -1386,7 +1438,7 @@
sbinfo->free_blocks = blocks;
sbinfo->max_inodes = inodes;
sbinfo->free_inodes = inodes;
- sb->s_maxbytes = (unsigned long long) SHMEM_MAX_BLOCKS << PAGE_CACHE_SHIFT;
+ sb->s_maxbytes = SHMEM_MAX_BYTES;
sb->s_blocksize = PAGE_CACHE_SIZE;
sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
sb->s_magic = TMPFS_MAGIC;
@@ -1480,6 +1532,7 @@

static struct inode_operations shmem_inode_operations = {
truncate: shmem_truncate,
+ setattr: shmem_setattr,
};

static struct inode_operations shmem_dir_inode_operations = {
@@ -1603,17 +1656,16 @@
*/
struct file *shmem_file_setup(char * name, loff_t size)
{
- int error;
+ int error = -ENOMEM;
struct file *file;
struct inode * inode;
struct dentry *dentry, *root;
struct qstr this;
- int vm_enough_memory(long pages);

- if (size > (unsigned long long) SHMEM_MAX_BLOCKS << PAGE_CACHE_SHIFT)
+ if (size > SHMEM_MAX_BYTES)
return ERR_PTR(-EINVAL);

- if (!vm_enough_memory((size) >> PAGE_CACHE_SHIFT))
+ if (!vm_enough_memory(VM_ACCT(size)))
return ERR_PTR(-ENOMEM);

this.name = name;
@@ -1622,7 +1674,7 @@
root = shm_mnt->mnt_root;
dentry = d_alloc(root, &this);
if (!dentry)
- return ERR_PTR(-ENOMEM);
+ goto put_memory;

error = -ENFILE;
file = get_empty_filp();
@@ -1648,8 +1700,11 @@
put_filp(file);
put_dentry:
dput (dentry);
+put_memory:
+ vm_unacct_memory(VM_ACCT(size));
return ERR_PTR(error);
}
+
/*
* shmem_zero_setup - setup a shared anonymous mapping
*




2002-07-26 10:28:29

by Pavel Machek

[permalink] [raw]
Subject: Re: [PATCH] 2.5-rmap: VM strict overcommit

Hi!


> diff -urN linux-2.5.26-rmap/Documentation/vm/overcommit-accounting linux/Documentation/vm/overcommit-accounting
> --- linux-2.5.26-rmap/Documentation/vm/overcommit-accounting Wed Dec 31 16:00:00 1969
> +++ linux/Documentation/vm/overcommit-accounting Wed Jul 17 10:45:47 2002
> @@ -0,0 +1,77 @@
> +The Linux kernel supports four overcommit handling modes
> +
> +0 - Heuristic overcommit handling. Obvious overcommits of
> + address space are refused. Used for a typical system. It
> + ensures a seriously wild allocation fails while allowing
> + overcommit to reduce swap usage. This is the default.
> +
> +1 - No overcommit handling. Appropriate for some scientific
> + applications.
> +
> +2 - (NEW) swapless strict overcommit. The total address space
> + commit for the system is not permitted to exceed 95% of
> + free memory. This mode utilizes the new stricter accounting
> + but does not impose a very strict rule. It is possible that
> + the system could kill a process accessing pages in certain
> + cases. If mode 3 is too strict when no swap is present
> + this is the best you can do.
> +
> +3 - (NEW) strict overcommit. The total address space commit
> + for the system is not permitted to exceed swap + half ram.
> + In almost all situations this means a process will not be
> + killed while accessing pages but only by malloc failures
> + that are reported back by the kernel mmap/brk code.

In what scenario can "strict overcommit" kill?

> +4 - (NEW) paranoid overcommit. The total address space commit
> + for the system is not permitted to exceed swap. The machine
> + will never kill a process accessing pages it has mapped
> + except due to a bug (ie report it!).

...and why is that scenario impossible on "paranoid overcommit"?
Pavel
--
I'm [email protected]. "In my country we have almost anarchy and I don't care."
Panos Katsaloulis describing me w.r.t. patents at [email protected]

2002-07-26 13:29:23

by Alan

[permalink] [raw]
Subject: Re: [PATCH] 2.5-rmap: VM strict overcommit

On Fri, 2002-07-26 at 11:31, Pavel Machek wrote:
> In what scenario can "strict overcommit" kill?

When the kernel grabs over 50% of RAM. Remember that includes page
tables. I've seen the kernel taking 35% of RAM.

>
> > +4 - (NEW) paranoid overcommit. The total address space commit
> > + for the system is not permitted to exceed swap. The machine
> > + will never kill a process accessing pages it has mapped
> > + except due to a bug (ie report it!).
>
> ...and why is that scenario impossible on "paranoid overcommit"?

You are guaranteed user pages are either discardable or all fit in swap.
If you run out of memory the kernel itself runs out of allocation space
for any kernel owned page.

2002-07-30 17:44:37

by Pavel Machek

[permalink] [raw]
Subject: Re: [PATCH] 2.5-rmap: VM strict overcommit

Hi!

> > In what scenario can "strict overcommit" kill?
>
> When the kernel grabs over 50% of RAM. Remember that includes page
> tables. I've seen the kernel taking 35% of RAM.

But it could happen that kernel would attempt to allocate 101% of RAM
for page tables, right? At that even "paranoid overcommit" might be
OOM, right?
Pavel
--
Worst form of spam? Adding advertisment signatures ala sourceforge.net.
What goes next? Inserting advertisment *into* email?

2002-07-30 17:46:31

by Rik van Riel

[permalink] [raw]
Subject: Re: [PATCH] 2.5-rmap: VM strict overcommit

On Tue, 30 Jul 2002, Pavel Machek wrote:

> > > In what scenario can "strict overcommit" kill?
> >
> > When the kernel grabs over 50% of RAM. Remember that includes page
> > tables. I've seen the kernel taking 35% of RAM.
>
> But it could happen that kernel would attempt to allocate 101% of RAM
> for page tables, right? At that even "paranoid overcommit" might be OOM,
> right?

Indeed, there are a number of places where memory allocation
by the kernel is pretty much unbound.

IMHO we need to fix those.

regards,

Rik
--
http://www.linuxsymposium.org/2002/
"You're one of those condescending OLS attendants"
"Here's a nickle kid. Go buy yourself a real t-shirt"

http://www.surriel.com/ http://distro.conectiva.com/

2002-07-30 17:52:36

by William Lee Irwin III

[permalink] [raw]
Subject: Re: [PATCH] 2.5-rmap: VM strict overcommit

At some point in the past, someone said:
>> When the kernel grabs over 50% of RAM. Remember that includes page
>> tables. I've seen the kernel taking 35% of RAM.

On Tue, Jul 30, 2002 at 12:20:52AM +0200, Pavel Machek wrote:
> But it could happen that kernel would attempt to allocate 101% of RAM
> for page tables, right? At that even "paranoid overcommit" might be
> OOM, right?

There is already accounting for this, so it is no longer necessary to
let them fly below the kernel's radar.


Cheers,
Bill

2002-07-30 18:50:13

by Alan

[permalink] [raw]
Subject: Re: [PATCH] 2.5-rmap: VM strict overcommit

On Mon, 2002-07-29 at 23:20, Pavel Machek wrote:
> But it could happen that kernel would attempt to allocate 101% of RAM
> for page tables, right? At that even "paranoid overcommit" might be
> OOM, right?

Paranoid mode guarantees that the process wont get killed for OOM. It
doesn't guarantee that the machine itself won't die. For obvious reasons
thats a rather more complex guarantee, but one I'd like to see possible
in future