write.c

讲述linux的初始化过程

/*
 * linux/fs/nfs/write.c
 *
 * Writing file data over NFS.
 *
 * We do it like this: When a (user) process wishes to write data to an
 * NFS file, a write request is allocated that contains the RPC task data
 * plus some info on the page to be written, and added to the inode's
 * write chain. If the process writes past the end of the page, an async
 * RPC call to write the page is scheduled immediately; otherwise, the call
 * is delayed for a few seconds.
 *
 * Just like readahead, no async I/O is performed if wsize < PAGE_SIZE.
 *
 * Write requests are kept on the inode's writeback list. Each entry in
 * that list references the page (portion) to be written. When the
 * cache timeout has expired, the RPC task is woken up, and tries to
 * lock the page. As soon as it manages to do so, the request is moved
 * from the writeback list to the writelock list.
 *
 * Note: we must make sure never to confuse the inode passed in the
 * write_page request with the one in page->inode. As far as I understand
 * it, these are different when doing a swap-out.
 *
 * To understand everything that goes on here and in the NFS read code,
 * one should be aware that a page is locked in exactly one of the following
 * cases:
 *
 *  -	A write request is in progress.
 *  -	A user process is in generic_file_write/nfs_update_page
 *  -	A user process is in generic_file_read
 *
 * Also note that because of the way pages are invalidated in
 * nfs_revalidate_inode, the following assertions hold:
 *
 *  -	If a page is dirty, there will be no read requests (a page will
 *	not be re-read unless invalidated by nfs_revalidate_inode).
 *  -	If the page is not uptodate, there will be no pending write
 *	requests, and no process will be in nfs_update_page.
 *
 * FIXME: Interaction with the vmscan routines is not optimal yet.
 * Either vmscan must be made nfs-savvy, or we need a different page
 * reclaim concept that supports something like FS-independent
 * buffer_heads with a b_ops-> field.
 *
 * Copyright (C) 1996, 1997, Olaf Kirch <[e-mail]>
 */

#include <linux/config.h>
#include <linux/types.h>
#include <linux/malloc.h>
#include <linux/swap.h>
#include <linux/pagemap.h>
#include <linux/file.h>

#include <linux/sunrpc/clnt.h>
#include <linux/nfs_fs.h>
#include <linux/nfs_mount.h>
#include <linux/nfs_flushd.h>
#include <linux/nfs_page.h>
#include <asm/uaccess.h>
#include <linux/smp_lock.h>

#define NFS_PARANOIA 1
#define NFSDBG_FACILITY		NFSDBG_PAGECACHE

/*
 * Spinlock
 */
spinlock_t nfs_wreq_lock = SPIN_LOCK_UNLOCKED;
static atomic_t	nfs_nr_requests = ATOMIC_INIT(0);

/*
 * Local structures
 *
 * This is the struct where the WRITE/COMMIT arguments go.
 */
struct nfs_write_data {
	struct rpc_task		task;
	struct inode		*inode;
	struct rpc_cred		*cred;
	struct nfs_writeargs	args;		/* argument struct */
	struct nfs_writeres	res;		/* result struct */
	struct nfs_fattr	fattr;
	struct nfs_writeverf	verf;
	struct list_head	pages;		/* Coalesced requests we wish to flush */
};

/*
 * Local function declarations
 */
static struct nfs_page * nfs_update_request(struct file*, struct inode *,
					    struct page *,
					    unsigned int, unsigned int);
static void	nfs_strategy(struct inode *inode);
static void	nfs_writeback_done(struct rpc_task *);
#ifdef CONFIG_NFS_V3
static void	nfs_commit_done(struct rpc_task *);
#endif

/* Hack for future NFS swap support */
#ifndef IS_SWAPFILE
# define IS_SWAPFILE(inode)	(0)
#endif

static kmem_cache_t *nfs_page_cachep;
static kmem_cache_t *nfs_wdata_cachep;

static __inline__ struct nfs_page *nfs_page_alloc(void)
{
	struct nfs_page	*p;
	p = kmem_cache_alloc(nfs_page_cachep, SLAB_KERNEL);
	if (p) {
		memset(p, 0, sizeof(*p));
		INIT_LIST_HEAD(&p->wb_hash);
		INIT_LIST_HEAD(&p->wb_list);
		init_waitqueue_head(&p->wb_wait);
	}
	return p;
}

static __inline__ void nfs_page_free(struct nfs_page *p)
{
	kmem_cache_free(nfs_page_cachep, p);
}

static __inline__ struct nfs_write_data *nfs_writedata_alloc(void)
{
	struct nfs_write_data	*p;
	p = kmem_cache_alloc(nfs_wdata_cachep, SLAB_NFS);
	if (p) {
		memset(p, 0, sizeof(*p));
		INIT_LIST_HEAD(&p->pages);
	}
	return p;
}

static __inline__ void nfs_writedata_free(struct nfs_write_data *p)
{
	kmem_cache_free(nfs_wdata_cachep, p);
}

static void nfs_writedata_release(struct rpc_task *task)
{
	struct nfs_write_data	*wdata = (struct nfs_write_data *)task->tk_calldata;
	nfs_writedata_free(wdata);
}

/*
 * This function will be used to simulate weak cache consistency
 * under NFSv2 when the NFSv3 attribute patch is included.
 * For the moment, we just call nfs_refresh_inode().
 */
static __inline__ int
nfs_write_attributes(struct inode *inode, struct nfs_fattr *fattr)
{
	if ((fattr->valid & NFS_ATTR_FATTR) && !(fattr->valid & NFS_ATTR_WCC)) {
		fattr->pre_size  = NFS_CACHE_ISIZE(inode);
		fattr->pre_mtime = NFS_CACHE_MTIME(inode);
		fattr->pre_ctime = NFS_CACHE_CTIME(inode);
		fattr->valid |= NFS_ATTR_WCC;
	}
	return nfs_refresh_inode(inode, fattr);
}

/*
 * Write a page synchronously.
 * Offset is the data offset within the page.
 */
static int
nfs_writepage_sync(struct file *file, struct inode *inode, struct page *page,
		   unsigned int offset, unsigned int count)
{
	struct rpc_cred	*cred = NULL;
	loff_t		base;
	unsigned int	wsize = NFS_SERVER(inode)->wsize;
	int		result, refresh = 0, written = 0, flags;
	u8		*buffer;
	struct nfs_fattr fattr;
	struct nfs_writeverf verf;


	if (file)
		cred = nfs_file_cred(file);

	lock_kernel();
	dprintk("NFS:      nfs_writepage_sync(%x/%Ld %d@%Ld)\n",
		inode->i_dev, (long long)NFS_FILEID(inode),
		count, (long long)(page_offset(page) + offset));

	buffer = kmap(page) + offset;
	base = page_offset(page) + offset;

	flags = ((IS_SWAPFILE(inode)) ? NFS_RW_SWAP : 0) | NFS_RW_SYNC;

	do {
		if (count < wsize && !IS_SWAPFILE(inode))
			wsize = count;

		result = NFS_PROTO(inode)->write(inode, cred, &fattr, flags,
						 base, wsize, buffer, &verf);
		nfs_write_attributes(inode, &fattr);

		if (result < 0) {
			/* Must mark the page invalid after I/O error */
			ClearPageUptodate(page);
			goto io_error;
		}
		if (result != wsize)
			printk("NFS: short write, wsize=%u, result=%d\n",
			wsize, result);
		refresh = 1;
		buffer  += wsize;
	        base    += wsize;
		written += wsize;
		count   -= wsize;
		/*
		 * If we've extended the file, update the inode
		 * now so we don't invalidate the cache.
		 */
		if (base > inode->i_size)
			inode->i_size = base;
	} while (count);

	if (PageError(page))
		ClearPageError(page);

io_error:
	kunmap(page);

	unlock_kernel();
	return written? written : result;
}

static int
nfs_writepage_async(struct file *file, struct inode *inode, struct page *page,
		    unsigned int offset, unsigned int count)
{
	struct nfs_page	*req;
	int		status;

	req = nfs_update_request(file, inode, page, offset, count);
	status = (IS_ERR(req)) ? PTR_ERR(req) : 0;
	if (status < 0)
		goto out;
	nfs_release_request(req);
	nfs_strategy(inode);
 out:
	return status;
}

/*
 * Write an mmapped page to the server.
 */
int
nfs_writepage(struct page *page)
{
	struct inode *inode;
	unsigned long end_index;
	unsigned offset = PAGE_CACHE_SIZE;
	int err;
	struct address_space *mapping = page->mapping;

	if (!mapping)
		BUG();
	inode = mapping->host;
	if (!inode)
		BUG();
	end_index = inode->i_size >> PAGE_CACHE_SHIFT;

	/* Ensure we've flushed out any previous writes */
	nfs_wb_page(inode,page);

	/* easy case */
	if (page->index < end_index)
		goto do_it;
	/* things got complicated... */
	offset = inode->i_size & (PAGE_CACHE_SIZE-1);

	/* OK, are we completely out? */
	err = -EIO;
	if (page->index >= end_index+1 || !offset)
		goto out;
do_it:
	if (!PageError(page) && NFS_SERVER(inode)->rsize >= PAGE_CACHE_SIZE) {
		err = nfs_writepage_async(NULL, inode, page, 0, offset);
		if (err >= 0)
			goto out_ok;
	}
	err = nfs_writepage_sync(NULL, inode, page, 0, offset); 
	if ( err == offset) {
out_ok:
		err = 0;
	}
out:
	UnlockPage(page);
	return err; 
}

/*
 * Check whether the file range we want to write to is locked by
 * us.
 */
static int
region_locked(struct inode *inode, struct nfs_page *req)
{
	struct file_lock	*fl;
	loff_t			rqstart, rqend;

	/* Don't optimize writes if we don't use NLM */
	if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
		return 0;

	rqstart = page_offset(req->wb_page) + req->wb_offset;
	rqend = rqstart + req->wb_bytes;
	for (fl = inode->i_flock; fl; fl = fl->fl_next) {
		if (fl->fl_owner == current->files && (fl->fl_flags & FL_POSIX)
		    && fl->fl_type == F_WRLCK
		    && fl->fl_start <= rqstart && rqend <= fl->fl_end) {
			return 1;
		}
	}

	return 0;
}

/*
 * Insert a write request into an inode
 */
static inline void
nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
{
	if (!list_empty(&req->wb_hash))
		return;
	if (!NFS_WBACK_BUSY(req))
		printk(KERN_ERR "NFS: unlocked request attempted hashed!\n");
	if (list_empty(&inode->u.nfs_i.writeback))
		atomic_inc(&inode->i_count);
	inode->u.nfs_i.npages++;
	list_add(&req->wb_hash, &inode->u.nfs_i.writeback);
	req->wb_count++;
}

/*
 * Insert a write request into an inode
 */
static inline void
nfs_inode_remove_request(struct nfs_page *req)
{
	struct inode *inode;
	spin_lock(&nfs_wreq_lock);
	if (list_empty(&req->wb_hash)) {
		spin_unlock(&nfs_wreq_lock);
		return;
	}
	if (!NFS_WBACK_BUSY(req))
		printk(KERN_ERR "NFS: unlocked request attempted unhashed!\n");
	inode = req->wb_inode;
	list_del(&req->wb_hash);
	INIT_LIST_HEAD(&req->wb_hash);
	inode->u.nfs_i.npages--;
	if ((inode->u.nfs_i.npages == 0) != list_empty(&inode->u.nfs_i.writeback))
		printk(KERN_ERR "NFS: desynchronized value of nfs_i.npages.\n");
	if (list_empty(&inode->u.nfs_i.writeback))
		iput(inode);
	if (!nfs_have_writebacks(inode) && !nfs_have_read(inode))
		inode_remove_flushd(inode);
	spin_unlock(&nfs_wreq_lock);
	nfs_release_request(req);
}

/*
 * Find a request
 */
static inline struct nfs_page *
_nfs_find_request(struct inode *inode, struct page *page)
{
	struct list_head	*head, *next;

	head = &inode->u.nfs_i.writeback;
	next = head->next;
	while (next != head) {
		struct nfs_page *req = nfs_inode_wb_entry(next);
		next = next->next;
		if (page_index(req->wb_page) != page_index(page))
			continue;
		req->wb_count++;
		return req;
	}
	return NULL;
}

static struct nfs_page *
nfs_find_request(struct inode *inode, struct page *page)
{
	struct nfs_page		*req;

	spin_lock(&nfs_wreq_lock);
	req = _nfs_find_request(inode, page);
	spin_unlock(&nfs_wreq_lock);
	return req;
}

/*
 * Insert a write request into a sorted list
 */
void nfs_list_add_request(struct nfs_page *req, struct list_head *head)
{
	struct list_head *prev;

	if (!list_empty(&req->wb_list)) {
		printk(KERN_ERR "NFS: Add to list failed!\n");
		return;
	}
	if (!NFS_WBACK_BUSY(req))
		printk(KERN_ERR "NFS: unlocked request attempted added to list!\n");
	prev = head->prev;
	while (prev != head) {
		struct nfs_page	*p = nfs_list_entry(prev);
		if (page_index(p->wb_page) < page_index(req->wb_page))
			break;
		prev = prev->prev;
	}
	list_add(&req->wb_list, prev);
	req->wb_list_head = head;
}

/*
 * Insert a write request into an inode
 */
void nfs_list_remove_request(struct nfs_page *req)
{
	if (list_empty(&req->wb_list))
		return;
	if (!NFS_WBACK_BUSY(req))
		printk(KERN_ERR "NFS: unlocked request attempted removed from list!\n");
	list_del(&req->wb_list);
	INIT_LIST_HEAD(&req->wb_list);
	req->wb_list_head = NULL;
}

/*
 * Add a request to the inode's dirty list.
 */
static inline void
nfs_mark_request_dirty(struct nfs_page *req)
{
	struct inode *inode = req->wb_inode;

	spin_lock(&nfs_wreq_lock);
	if (list_empty(&req->wb_list)) {
		nfs_list_add_request(req, &inode->u.nfs_i.dirty);
		inode->u.nfs_i.ndirty++;
	}
	spin_unlock(&nfs_wreq_lock);
	/*
	 * NB: the call to inode_schedule_scan() must lie outside the
	 *     spinlock since it can run flushd().
	 */
	inode_schedule_scan(inode, req->wb_timeout);
}

/*
 * Check if a request is dirty
 */
static inline int
nfs_dirty_request(struct nfs_page *req)
{
	struct inode *inode = req->wb_inode;
	return !list_empty(&req->wb_list) && req->wb_list_head == &inode->u.nfs_i.dirty;
}

#ifdef CONFIG_NFS_V3
/*
 * Add a request to the inode's commit list.
 */
static inline void
nfs_mark_request_commit(struct nfs_page *req)
{
	struct inode *inode = req->wb_inode;

	spin_lock(&nfs_wreq_lock);
	if (list_empty(&req->wb_list)) {
		nfs_list_add_request(req, &inode->u.nfs_i.commit);
		inode->u.nfs_i.ncommit++;
	}
	spin_unlock(&nfs_wreq_lock);
	/*
	 * NB: the call to inode_schedule_scan() must lie outside the
	 *     spinlock since it can run flushd().
	 */
	inode_schedule_scan(inode, req->wb_timeout);
}
#endif

/*
 * Create a write request.
 * Page must be locked by the caller. This makes sure we never create
 * two different requests for the same page, and avoids possible deadlock
 * when we reach the hard limit on the number of dirty pages.
 * It should be safe to sleep here.
 */
struct nfs_page *nfs_create_request(struct file *file, struct inode *inode,
				    struct page *page,
				    unsigned int offset, unsigned int count)
{
	struct nfs_reqlist	*cache = NFS_REQUESTLIST(inode);
	struct nfs_page		*req = NULL;
	long			timeout;

	/* Deal with hard/soft limits.
	 */
	do {
		/* If we're over the global soft limit, wake up all requests */
		if (atomic_read(&nfs_nr_requests) >= MAX_REQUEST_SOFT) {
			dprintk("NFS:      hit soft limit (%d requests)\n",
				atomic_read(&nfs_nr_requests));
			if (!cache->task)
				nfs_reqlist_init(NFS_SERVER(inode));
			nfs_wake_flushd();
		}

		/* If we haven't reached the local hard limit yet,
		 * try to allocate the request struct */
		if (atomic_read(&cache->nr_requests) < MAX_REQUEST_HARD) {