https://github.com/torvalds/linux
Revision 9230a0b65b47fe6856c4468ec0175c4987e5bede authored by Dave Chinner on 20 November 2018, 06:50:08 UTC, committed by Darrick J. Wong on 21 November 2018, 18:10:53 UTC
Long saga. There have been days spent following this through dead end
after dead end in multi-GB event traces. This morning, after writing
a trace-cmd wrapper that enabled me to be more selective about XFS
trace points, I discovered that I could get just enough essential
tracepoints enabled that there was a 50:50 chance the fsx config
would fail at ~115k ops. If it didn't fail at op 115547, I stopped
fsx at op 115548 anyway.

That gave me two traces - one where the problem manifested, and one
where it didn't. After refining the traces to have the necessary
information, I found that in the failing case there was a real
extent in the COW fork compared to an unwritten extent in the
working case.

Walking back through the two traces to the point where the CWO fork
extents actually diverged, I found that the bad case had an extra
unwritten extent in it. This is likely because the bug it led me to
had triggered multiple times in those 115k ops, leaving stray
COW extents around. What I saw was a COW delalloc conversion to an
unwritten extent (as they should always be through
xfs_iomap_write_allocate()) resulted in a /written extent/:

xfs_writepage:        dev 259:0 ino 0x83 pgoff 0x17000 size 0x79a00 offset 0 length 0
xfs_iext_remove:      dev 259:0 ino 0x83 state RC|LF|RF|COW cur 0xffff888247b899c0/2 offset 32 block 152 count 20 flag 1 caller xfs_bmap_add_extent_delay_real
xfs_bmap_pre_update:  dev 259:0 ino 0x83 state RC|LF|RF|COW cur 0xffff888247b899c0/1 offset 1 block 4503599627239429 count 31 flag 0 caller xfs_bmap_add_extent_delay_real
xfs_bmap_post_update: dev 259:0 ino 0x83 state RC|LF|RF|COW cur 0xffff888247b899c0/1 offset 1 block 121 count 51 flag 0 caller xfs_bmap_add_ex

Basically, Cow fork before:

	0 1            32          52
	+H+DDDDDDDDDDDD+UUUUUUUUUUU+
	   PREV		RIGHT

COW delalloc conversion allocates:

	  1	       32
	  +uuuuuuuuuuuu+
	  NEW

And the result according to the xfs_bmap_post_update trace was:

	0 1            32          52
	+H+wwwwwwwwwwwwwwwwwwwwwwww+
	   PREV

Which is clearly wrong - it should be a merged unwritten extent,
not an unwritten extent.

That lead me to look at the LEFT_FILLING|RIGHT_FILLING|RIGHT_CONTIG
case in xfs_bmap_add_extent_delay_real(), and sure enough, there's
the bug.

It takes the old delalloc extent (PREV) and adds the length of the
RIGHT extent to it, takes the start block from NEW, removes the
RIGHT extent and then updates PREV with the new extent.

What it fails to do is update PREV.br_state. For delalloc, this is
always XFS_EXT_NORM, while in this case we are converting the
delayed allocation to unwritten, so it needs to be updated to
XFS_EXT_UNWRITTEN. This LF|RF|RC case does not do this, and so
the resultant extent is always written.

And that's the bug I've been chasing for a week - a bmap btree bug,
not a reflink/dedupe/copy_file_range bug, but a BMBT bug introduced
with the recent in core extent tree scalability enhancements.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
1 parent 2c30717
Raw File
Tip revision: 9230a0b65b47fe6856c4468ec0175c4987e5bede authored by Dave Chinner on 20 November 2018, 06:50:08 UTC
xfs: delalloc -> unwritten COW fork allocation can go wrong
Tip revision: 9230a0b
stackdepot.c
/*
 * Generic stack depot for storing stack traces.
 *
 * Some debugging tools need to save stack traces of certain events which can
 * be later presented to the user. For example, KASAN needs to safe alloc and
 * free stacks for each object, but storing two stack traces per object
 * requires too much memory (e.g. SLUB_DEBUG needs 256 bytes per object for
 * that).
 *
 * Instead, stack depot maintains a hashtable of unique stacktraces. Since alloc
 * and free stacks repeat a lot, we save about 100x space.
 * Stacks are never removed from depot, so we store them contiguously one after
 * another in a contiguos memory allocation.
 *
 * Author: Alexander Potapenko <glider@google.com>
 * Copyright (C) 2016 Google, Inc.
 *
 * Based on code by Dmitry Chernenkov.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * version 2 as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 */

#include <linux/gfp.h>
#include <linux/jhash.h>
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/percpu.h>
#include <linux/printk.h>
#include <linux/slab.h>
#include <linux/stacktrace.h>
#include <linux/stackdepot.h>
#include <linux/string.h>
#include <linux/types.h>

#define DEPOT_STACK_BITS (sizeof(depot_stack_handle_t) * 8)

#define STACK_ALLOC_NULL_PROTECTION_BITS 1
#define STACK_ALLOC_ORDER 2 /* 'Slab' size order for stack depot, 4 pages */
#define STACK_ALLOC_SIZE (1LL << (PAGE_SHIFT + STACK_ALLOC_ORDER))
#define STACK_ALLOC_ALIGN 4
#define STACK_ALLOC_OFFSET_BITS (STACK_ALLOC_ORDER + PAGE_SHIFT - \
					STACK_ALLOC_ALIGN)
#define STACK_ALLOC_INDEX_BITS (DEPOT_STACK_BITS - \
		STACK_ALLOC_NULL_PROTECTION_BITS - STACK_ALLOC_OFFSET_BITS)
#define STACK_ALLOC_SLABS_CAP 8192
#define STACK_ALLOC_MAX_SLABS \
	(((1LL << (STACK_ALLOC_INDEX_BITS)) < STACK_ALLOC_SLABS_CAP) ? \
	 (1LL << (STACK_ALLOC_INDEX_BITS)) : STACK_ALLOC_SLABS_CAP)

/* The compact structure to store the reference to stacks. */
union handle_parts {
	depot_stack_handle_t handle;
	struct {
		u32 slabindex : STACK_ALLOC_INDEX_BITS;
		u32 offset : STACK_ALLOC_OFFSET_BITS;
		u32 valid : STACK_ALLOC_NULL_PROTECTION_BITS;
	};
};

struct stack_record {
	struct stack_record *next;	/* Link in the hashtable */
	u32 hash;			/* Hash in the hastable */
	u32 size;			/* Number of frames in the stack */
	union handle_parts handle;
	unsigned long entries[1];	/* Variable-sized array of entries. */
};

static void *stack_slabs[STACK_ALLOC_MAX_SLABS];

static int depot_index;
static int next_slab_inited;
static size_t depot_offset;
static DEFINE_SPINLOCK(depot_lock);

static bool init_stack_slab(void **prealloc)
{
	if (!*prealloc)
		return false;
	/*
	 * This smp_load_acquire() pairs with smp_store_release() to
	 * |next_slab_inited| below and in depot_alloc_stack().
	 */
	if (smp_load_acquire(&next_slab_inited))
		return true;
	if (stack_slabs[depot_index] == NULL) {
		stack_slabs[depot_index] = *prealloc;
	} else {
		stack_slabs[depot_index + 1] = *prealloc;
		/*
		 * This smp_store_release pairs with smp_load_acquire() from
		 * |next_slab_inited| above and in depot_save_stack().
		 */
		smp_store_release(&next_slab_inited, 1);
	}
	*prealloc = NULL;
	return true;
}

/* Allocation of a new stack in raw storage */
static struct stack_record *depot_alloc_stack(unsigned long *entries, int size,
		u32 hash, void **prealloc, gfp_t alloc_flags)
{
	int required_size = offsetof(struct stack_record, entries) +
		sizeof(unsigned long) * size;
	struct stack_record *stack;

	required_size = ALIGN(required_size, 1 << STACK_ALLOC_ALIGN);

	if (unlikely(depot_offset + required_size > STACK_ALLOC_SIZE)) {
		if (unlikely(depot_index + 1 >= STACK_ALLOC_MAX_SLABS)) {
			WARN_ONCE(1, "Stack depot reached limit capacity");
			return NULL;
		}
		depot_index++;
		depot_offset = 0;
		/*
		 * smp_store_release() here pairs with smp_load_acquire() from
		 * |next_slab_inited| in depot_save_stack() and
		 * init_stack_slab().
		 */
		if (depot_index + 1 < STACK_ALLOC_MAX_SLABS)
			smp_store_release(&next_slab_inited, 0);
	}
	init_stack_slab(prealloc);
	if (stack_slabs[depot_index] == NULL)
		return NULL;

	stack = stack_slabs[depot_index] + depot_offset;

	stack->hash = hash;
	stack->size = size;
	stack->handle.slabindex = depot_index;
	stack->handle.offset = depot_offset >> STACK_ALLOC_ALIGN;
	stack->handle.valid = 1;
	memcpy(stack->entries, entries, size * sizeof(unsigned long));
	depot_offset += required_size;

	return stack;
}

#define STACK_HASH_ORDER 20
#define STACK_HASH_SIZE (1L << STACK_HASH_ORDER)
#define STACK_HASH_MASK (STACK_HASH_SIZE - 1)
#define STACK_HASH_SEED 0x9747b28c

static struct stack_record *stack_table[STACK_HASH_SIZE] = {
	[0 ...	STACK_HASH_SIZE - 1] = NULL
};

/* Calculate hash for a stack */
static inline u32 hash_stack(unsigned long *entries, unsigned int size)
{
	return jhash2((u32 *)entries,
			       size * sizeof(unsigned long) / sizeof(u32),
			       STACK_HASH_SEED);
}

/* Use our own, non-instrumented version of memcmp().
 *
 * We actually don't care about the order, just the equality.
 */
static inline
int stackdepot_memcmp(const unsigned long *u1, const unsigned long *u2,
			unsigned int n)
{
	for ( ; n-- ; u1++, u2++) {
		if (*u1 != *u2)
			return 1;
	}
	return 0;
}

/* Find a stack that is equal to the one stored in entries in the hash */
static inline struct stack_record *find_stack(struct stack_record *bucket,
					     unsigned long *entries, int size,
					     u32 hash)
{
	struct stack_record *found;

	for (found = bucket; found; found = found->next) {
		if (found->hash == hash &&
		    found->size == size &&
		    !stackdepot_memcmp(entries, found->entries, size))
			return found;
	}
	return NULL;
}

void depot_fetch_stack(depot_stack_handle_t handle, struct stack_trace *trace)
{
	union handle_parts parts = { .handle = handle };
	void *slab = stack_slabs[parts.slabindex];
	size_t offset = parts.offset << STACK_ALLOC_ALIGN;
	struct stack_record *stack = slab + offset;

	trace->nr_entries = trace->max_entries = stack->size;
	trace->entries = stack->entries;
	trace->skip = 0;
}
EXPORT_SYMBOL_GPL(depot_fetch_stack);

/**
 * depot_save_stack - save stack in a stack depot.
 * @trace - the stacktrace to save.
 * @alloc_flags - flags for allocating additional memory if required.
 *
 * Returns the handle of the stack struct stored in depot.
 */
depot_stack_handle_t depot_save_stack(struct stack_trace *trace,
				    gfp_t alloc_flags)
{
	u32 hash;
	depot_stack_handle_t retval = 0;
	struct stack_record *found = NULL, **bucket;
	unsigned long flags;
	struct page *page = NULL;
	void *prealloc = NULL;

	if (unlikely(trace->nr_entries == 0))
		goto fast_exit;

	hash = hash_stack(trace->entries, trace->nr_entries);
	bucket = &stack_table[hash & STACK_HASH_MASK];

	/*
	 * Fast path: look the stack trace up without locking.
	 * The smp_load_acquire() here pairs with smp_store_release() to
	 * |bucket| below.
	 */
	found = find_stack(smp_load_acquire(bucket), trace->entries,
			   trace->nr_entries, hash);
	if (found)
		goto exit;

	/*
	 * Check if the current or the next stack slab need to be initialized.
	 * If so, allocate the memory - we won't be able to do that under the
	 * lock.
	 *
	 * The smp_load_acquire() here pairs with smp_store_release() to
	 * |next_slab_inited| in depot_alloc_stack() and init_stack_slab().
	 */
	if (unlikely(!smp_load_acquire(&next_slab_inited))) {
		/*
		 * Zero out zone modifiers, as we don't have specific zone
		 * requirements. Keep the flags related to allocation in atomic
		 * contexts and I/O.
		 */
		alloc_flags &= ~GFP_ZONEMASK;
		alloc_flags &= (GFP_ATOMIC | GFP_KERNEL);
		alloc_flags |= __GFP_NOWARN;
		page = alloc_pages(alloc_flags, STACK_ALLOC_ORDER);
		if (page)
			prealloc = page_address(page);
	}

	spin_lock_irqsave(&depot_lock, flags);

	found = find_stack(*bucket, trace->entries, trace->nr_entries, hash);
	if (!found) {
		struct stack_record *new =
			depot_alloc_stack(trace->entries, trace->nr_entries,
					  hash, &prealloc, alloc_flags);
		if (new) {
			new->next = *bucket;
			/*
			 * This smp_store_release() pairs with
			 * smp_load_acquire() from |bucket| above.
			 */
			smp_store_release(bucket, new);
			found = new;
		}
	} else if (prealloc) {
		/*
		 * We didn't need to store this stack trace, but let's keep
		 * the preallocated memory for the future.
		 */
		WARN_ON(!init_stack_slab(&prealloc));
	}

	spin_unlock_irqrestore(&depot_lock, flags);
exit:
	if (prealloc) {
		/* Nobody used this memory, ok to free it. */
		free_pages((unsigned long)prealloc, STACK_ALLOC_ORDER);
	}
	if (found)
		retval = found->handle.handle;
fast_exit:
	return retval;
}
EXPORT_SYMBOL_GPL(depot_save_stack);
back to top