nta.6

实现linux平台下零拷贝技术的软件包。

+	if (!ptr && !try) {
+		struct avl_node_entry *entry;
+
+		try = 1;
+
+		entry = avl_node_entry_alloc(gfp_mask, get_order(size));
+		if (entry) {
+			local_irq_save(flags);
+			avl_node_entry_commit(entry, smp_processor_id());
+			goto repeat;
+		}
+
+	}
+#endif
+	if (unlikely(!ptr && try))
+		if (net_ratelimit())
+			printk("%s: Failed to allocate %u bytes.\n", __func__, size);
+
+	return ptr;
+}
+
+/*
+ * Remove free chunk from the list.
+ */
+static inline struct avl_container *avl_search_container(void *ptr, unsigned int idx, int cpu)
+{
+	struct avl_container *c = ptr;
+
+	list_del(&c->centry);
+	c->ptr = ptr;
+
+	return c;
+}
+
+/*
+ * Combine neighbour free chunks into the one with bigger size
+ * and put new chunk into list of free chunks with appropriate size.
+ */
+static void avl_combine(struct avl_node *node, void *lp, unsigned int lbits, void *rp, unsigned int rbits,
+		void *cur_ptr, unsigned int cur_bits, int cpu)
+{
+	struct avl_container *lc, *rc, *c;
+	unsigned int idx;
+	void *ptr;
+
+	lc = rc = c = NULL;
+	idx = cur_bits - 1;
+	ptr = cur_ptr;
+
+	c = (struct avl_container *)cur_ptr;
+	c->ptr = cur_ptr;
+
+	if (rp) {
+		rc = avl_search_container(rp, rbits-1, cpu);
+		if (!rc) {
+			printk(KERN_ERR "%p.%p: Failed to find a container for right pointer %p, rbits: %u.\n",
+					node, cur_ptr, rp, rbits);
+			BUG();
+		}
+
+		c = rc;
+		idx += rbits;
+		ptr = c->ptr;
+	}
+
+	if (lp) {
+		lc = avl_search_container(lp, lbits-1, cpu);
+		if (!lc) {
+			printk(KERN_ERR "%p.%p: Failed to find a container for left pointer %p, lbits: %u.\n",
+					node, cur_ptr, lp, lbits);
+			BUG();
+		}
+
+		idx += lbits;
+		ptr = c->ptr;
+	}
+	avl_container_insert(c, idx, cpu);
+}
+
+/*
+ * Free memory region of given size.
+ * Must be called on the same CPU where allocation happend
+ * with disabled interrupts.
+ */
+static void __avl_free_local(void *ptr, unsigned int size)
+{
+	unsigned long val = avl_ptr_to_value(ptr);
+	unsigned int pos, idx, sbits = AVL_ALIGN(size)/AVL_MIN_SIZE;
+	unsigned int rbits, lbits, cpu = avl_get_cpu_ptr(val);
+	struct avl_node *node;
+	unsigned long p;
+	void *lp, *rp;
+
+	node = avl_get_node_ptr((unsigned long)ptr);
+
+	pos = avl_ptr_to_offset(ptr);
+	idx = pos/BITS_PER_LONG;
+
+	p = node->mask[idx] >> (pos%BITS_PER_LONG);
+
+	if ((p & 1)) {
+		if (net_ratelimit())
+			printk(KERN_ERR "%p.%p: Broken pointer: value: %lx, pos: %u, idx: %u, mask: %lx, p: %lx.\n",
+				node, ptr, val, pos, idx, node->mask[idx], p);
+		return;
+	}
+
+	avl_fill_bits(node->mask, ARRAY_SIZE(node->mask), pos, sbits, 1);
+
+	lp = rp = NULL;
+	rbits = lbits = 0;
+
+	idx = (pos+sbits)/BITS_PER_LONG;
+	p = (pos+sbits)%BITS_PER_LONG;
+
+	if ((node->mask[idx] >> p) & 1) {
+		lbits = avl_count_set_up(node->mask, ARRAY_SIZE(node->mask), pos+sbits);
+		if (lbits) {
+			lp = (void *)(val + (pos + sbits)*AVL_MIN_SIZE);
+		}
+	}
+
+	if (pos) {
+		idx = (pos-1)/BITS_PER_LONG;
+		p = (pos-1)%BITS_PER_LONG;
+		if ((node->mask[idx] >> p) & 1) {
+			rbits = avl_count_set_down(node->mask, pos-1);
+			if (rbits) {
+				rp = (void *)(val + (pos-rbits)*AVL_MIN_SIZE);
+			}
+		}
+	}
+
+	avl_combine(node, lp, lbits, rp, rbits, ptr, sbits, cpu);
+}
+
+/*
+ * Free memory region of given size.
+ * If freeing CPU is not the same as allocation one, chunk will
+ * be placed into list of to-be-freed objects on allocation CPU,
+ * otherwise chunk will be freed and combined with neighbours.
+ * Must be called with disabled interrupts.
+ */
+static void __avl_free(void *ptr, unsigned int size)
+{
+	int cpu = avl_get_cpu_ptr((unsigned long)ptr);
+
+	if (cpu != smp_processor_id()) {
+		struct avl_free_list *l, *this = ptr;
+		struct avl_allocator_data *alloc = &avl_allocator[cpu];
+
+		this->cpu = smp_processor_id();
+		this->size = size;
+
+		spin_lock(&alloc->avl_free_lock);
+		l = alloc->avl_free_list_head;
+		alloc->avl_free_list_head = this;
+		this->next = l;
+		spin_unlock(&alloc->avl_free_lock);
+		return;
+	}
+
+	__avl_free_local(ptr, size);
+}
+
+/*
+ * Free memory region of given size without sniffer data update.
+ */
+void avl_free_no_zc(void *ptr, unsigned int size)
+{
+	unsigned long flags;
+	struct avl_free_list *l;
+	struct avl_allocator_data *alloc;
+	struct avl_chunk *ch = avl_ptr_to_chunk(ptr, size);
+
+	if (unlikely((ch->canary != AVL_CANARY) || ch->size != size)) {
+		printk("Freeing destroyed object: ptr: %p, size: %u, canary: %x, must be %x, refcnt: %d, saved size: %u.\n",
+				ptr, size, ch->canary, AVL_CANARY, atomic_read(&ch->refcnt), ch->size);
+		return;
+	}
+
+	if (atomic_dec_and_test(&ch->refcnt)) {
+		local_irq_save(flags);
+		__avl_free(ptr, size);
+
+		alloc = &avl_allocator[smp_processor_id()];
+
+		while (alloc->avl_free_list_head) {
+			spin_lock(&alloc->avl_free_lock);
+			l = alloc->avl_free_list_head;
+			alloc->avl_free_list_head = l->next;
+			spin_unlock(&alloc->avl_free_lock);
+			__avl_free_local(l, l->size);
+		}
+		local_irq_restore(flags);
+	}
+}
+
+/*
+ * Free memory region of given size.
+ */
+void avl_free(void *ptr, unsigned int size)
+{
+	struct avl_chunk *ch = avl_ptr_to_chunk(ptr, size);
+
+	if (unlikely((ch->canary != AVL_CANARY) || ch->size != size)) {
+		printk("Freeing destroyed object: ptr: %p, size: %u, canary: %x, must be %x, refcnt: %d, saved size: %u.\n",
+				ptr, size, ch->canary, AVL_CANARY, atomic_read(&ch->refcnt), ch->size);
+		return;
+	}
+	avl_update_zc(avl_get_node_ptr((unsigned long)ptr), ptr, size);
+	avl_free_no_zc(ptr, size);
+}
+
+/*
+ * Initialize per-cpu allocator data.
+ */
+static int avl_init_cpu(int cpu)
+{
+	unsigned int i;
+	struct avl_allocator_data *alloc = &avl_allocator[cpu];
+	struct avl_node_entry *entry;
+
+	spin_lock_init(&alloc->avl_free_lock);
+	spin_lock_init(&alloc->avl_node_lock);
+	INIT_LIST_HEAD(&alloc->avl_node_list);
+
+	alloc->avl_container_array = kzalloc(sizeof(struct list_head) * AVL_CONTAINER_ARRAY_SIZE, GFP_KERNEL);
+	if (!alloc->avl_container_array)
+		goto err_out_exit;
+
+	for (i=0; i<AVL_CONTAINER_ARRAY_SIZE; ++i)
+		INIT_LIST_HEAD(&alloc->avl_container_array[i]);
+
+	entry = avl_node_entry_alloc(GFP_KERNEL, AVL_ORDER);
+	if (!entry)
+		goto err_out_free_container;
+
+	avl_node_entry_commit(entry, cpu);
+
+	return 0;
+
+err_out_free_container:
+	kfree(alloc->avl_container_array);
+err_out_exit:
+	return -ENOMEM;
+}
+
+/*
+ * Initialize network allocator.
+ */
+int avl_init(void)
+{
+	int err, cpu;
+
+	for_each_possible_cpu(cpu) {
+		err = avl_init_cpu(cpu);
+		if (err)
+			goto err_out;
+	}
+
+	err = avl_init_zc();
+
+	printk(KERN_INFO "Network tree allocator has been initialized.\n");
+	return 0;
+
+err_out:
+	panic("Failed to initialize network allocator.\n");
+
+	return -ENOMEM;
+}
diff --git a/net/core/alloc/zc.c b/net/core/alloc/zc.c
new file mode 100644
index 0000000..8be4d7d
--- /dev/null
+++ b/net/core/alloc/zc.c
@@ -0,0 +1,487 @@
+/*
+ * 	zc.c
+ *
+ * 2006 Copyright (c) Evgeniy Polyakov <[e-mail]>
+ * All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * 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.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/string.h>
+#include <linux/errno.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/percpu.h>
+#include <linux/list.h>
+#include <linux/mm.h>
+#include <linux/fs.h>
+#include <linux/poll.h>
+#include <linux/ioctl.h>
+#include <linux/skbuff.h>
+#include <linux/netfilter.h>
+#include <linux/netfilter_ipv4.h>
+#include <linux/ip.h>
+#include <net/flow.h>
+#include <net/dst.h>
+#include <net/route.h>
+#include <asm/uaccess.h>
+#include <linux/avl.h>
+
+struct zc_private
+{
+	struct zc_data	*zcb;
+	struct mutex	lock;
+	int		cpu;
+};
+
+static char zc_name[] = "zc";
+static int zc_major;
+struct zc_control zc_sniffer;
+
+static int zc_release(struct inode *inode, struct file *file)
+{
+	struct zc_private *priv = file->private_data;
+
+	kfree(priv);
+	return 0;
+}
+
+static int zc_open(struct inode *inode, struct file *file)
+{
+	struct zc_private *priv;
+	struct zc_control *ctl = &zc_sniffer;
+
+	priv = kzalloc(sizeof(struct zc_private) + ctl->zc_num * sizeof(struct zc_data), GFP_KERNEL);
+	if (!priv)
+		return -ENOMEM;
+	priv->zcb = (struct zc_data *)(priv+1);
+	priv->cpu = 0; /* Use CPU0 by default */
+	mutex_init(&priv->lock);
+
+	file->private_data = priv;
+
+	return 0;
+}
+
+static int zc_mmap(struct file *file, struct vm_area_struct *vma)
+{
+	struct zc_private *priv = file->private_data;
+	struct avl_allocator_data *alloc = &avl_allocator[priv->cpu];
+	struct avl_node_entry *e;
+	unsigned long flags, start = vma->vm_start;
+	int err = 0, idx, off;
+	unsigned int i, j, st, num, total_num;
+
+	st = vma->vm_pgoff;
+	total_num = (vma->vm_end - vma->vm_start)/PAGE_SIZE;
+
+	printk("%s: start: %lx, end: %lx, total_num: %u, st: %u.\n", __func__, start, vma->vm_end, total_num, st);
+
+	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
+	vma->vm_flags |= VM_RESERVED;
+	vma->vm_file = file;
+
+	spin_lock_irqsave(&alloc->avl_node_lock, flags);
+	list_for_each_entry(e, &alloc->avl_node_list, node_entry) {
+		if (st >= e->avl_node_num*(1U<<e->avl_node_order)) {
+#if 0
+			printk("%s: continue on cpu: %d, e: %p, total_num: %u, node_num: %u, node_order: %u, pages_in_node: %u, st: %u.\n",
+					__func__, priv->cpu, e, total_num, e->avl_node_num, e->avl_node_order,
+					e->avl_node_num*(1U<<e->avl_node_order), st);
+#endif
+			st -= e->avl_node_num*(1U<<e->avl_node_order);
+			continue;
+		}
+		num = min_t(unsigned int, total_num, e->avl_node_num*(1<<e->avl_node_order));
+
+		printk("%s: cpu: %d, e: %p, total_num: %u, node_num: %u, node_order: %u, st: %u, num: %u.\n",
+				__func__, priv->cpu, e, total_num, e->avl_node_num, e->avl_node_order, st, num);
+
+		idx = 0;
+		off = st;
+		for (i=st; i<num;) {
+			struct avl_node *node = &e->avl_node_array[idx][off];
+
+			if (++off >= AVL_NODES_ON_PAGE) {
+				idx++;
+				off = 0;
+			}
+
+			for (j=0; (j<(1<<e->avl_node_order)) && (i<num); ++j, ++i) {
+				unsigned long virt = node->value + (j<<PAGE_SHIFT);
+				err = vm_insert_page(vma, start, virt_to_page(virt));
+				if (err) {
+					printk("\n%s: Failed to insert page for addr %lx into %lx, err: %d.\n",
+							__func__, virt, start, err);
+					break;
+				}
+				start += PAGE_SIZE;
+			}
+		}
+		if (err)
+			break;
+		total_num -= num;
+
+		if (total_num == 0)
+			break;
+	}
+	spin_unlock_irqrestore(&alloc->avl_node_lock, flags);
+
+	return err;
+}
+
+static ssize_t zc_write(struct file *file, const char __user *buf, size_t size, loff_t *off)
+{
+	ssize_t sz = 0;
+	struct zc_private *priv = file->private_data;
+	unsigned long flags;
+	unsigned int req_num = size/sizeof(struct zc_data), cnum, csize, i;
+	struct zc_control *ctl = &zc_sniffer;
+
+	while (size) {
+		cnum = min_t(unsigned int, req_num, ctl->zc_num);
+		csize = cnum*sizeof(struct zc_data);
+
+		if (copy_from_user(priv->zcb, buf, csize)) {
+			printk("%s: copy_from_user() failed.\n", __func__);
+			break;
+		}
+
+		spin_lock_irqsave(&ctl->zc_lock, flags);
+		for (i=0; i<cnum; ++i)
+			avl_free_no_zc(priv->zcb[i].data.ptr, priv->zcb[i].size);
+		ctl->zc_used -= cnum;
+		spin_unlock_irqrestore(&ctl->zc_lock, flags);
+
+		sz += csize;
+		size -= csize;
+		buf += csize;
+	}
+
+	return sz;
+}
+
+static ssize_t zc_read(struct file *file, char __user *buf, size_t size, loff_t *off)
+{
+	ssize_t sz = 0;
+	struct zc_private *priv = file->private_data;
+	unsigned long flags;
+	unsigned int pos, req_num = size/sizeof(struct zc_data), cnum, csize;
+	struct zc_control *ctl = &zc_sniffer;
+
+	wait_event_interruptible(ctl->zc_wait, ctl->zc_used > 0);
+
+	spin_lock_irqsave(&ctl->zc_lock, flags);
+	cnum = min_t(unsigned int, req_num, ctl->zc_used);
+	csize = cnum*sizeof(struct zc_data);
+	if (ctl->zc_used) {
+		if (ctl->zc_pos >= ctl->zc_used) {
+			pos = ctl->zc_pos - ctl->zc_used;
+			memcpy(priv->zcb, &ctl->zcb[pos], csize);