svclock.c

Linux中关于远程文件锁定的支持的源代码

/*
 * linux/fs/lockd/svclock.c
 *
 * Handling of server-side locks, mostly of the blocked variety.
 * This is the ugliest part of lockd because we tread on very thin ice.
 * GRANT and CANCEL calls may get stuck, meet in mid-flight, etc.
 * IMNSHO introducing the grant callback into the NLM protocol was one
 * of the worst ideas Sun ever had. Except maybe for the idea of doing
 * NFS file locking at all.
 *
 * I'm trying hard to avoid race conditions by protecting most accesses
 * to a file's list of blocked locks through a semaphore. The global
 * list of blocked locks is not protected in this fashion however.
 * Therefore, some functions (such as the RPC callback for the async grant
 * call) move blocked locks towards the head of the list *while some other
 * process might be traversing it*. This should not be a problem in
 * practice, because this will only cause functions traversing the list
 * to visit some blocks twice.
 *
 * Copyright (C) 1996, Olaf Kirch <[e-mail]>
 */

#include <linux/types.h>
#include <linux/errno.h>
#include <linux/kernel.h>
#include <linux/sched.h>
#include <linux/smp_lock.h>
#include <linux/sunrpc/clnt.h>
#include <linux/sunrpc/svc.h>
#include <linux/lockd/nlm.h>
#include <linux/lockd/lockd.h>
#include <linux/kthread.h>

#define NLMDBG_FACILITY		NLMDBG_SVCLOCK

#ifdef CONFIG_LOCKD_V4
#define nlm_deadlock	nlm4_deadlock
#else
#define nlm_deadlock	nlm_lck_denied
#endif

static void nlmsvc_release_block(struct nlm_block *block);
static void	nlmsvc_insert_block(struct nlm_block *block, unsigned long);
static void	nlmsvc_remove_block(struct nlm_block *block);

static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock);
static void nlmsvc_freegrantargs(struct nlm_rqst *call);
static const struct rpc_call_ops nlmsvc_grant_ops;

/*
 * The list of blocked locks to retry
 */
static LIST_HEAD(nlm_blocked);

/*
 * Insert a blocked lock into the global list
 */
static void
nlmsvc_insert_block(struct nlm_block *block, unsigned long when)
{
	struct nlm_block *b;
	struct list_head *pos;

	dprintk("lockd: nlmsvc_insert_block(%p, %ld)\n", block, when);
	if (list_empty(&block->b_list)) {
		kref_get(&block->b_count);
	} else {
		list_del_init(&block->b_list);
	}

	pos = &nlm_blocked;
	if (when != NLM_NEVER) {
		if ((when += jiffies) == NLM_NEVER)
			when ++;
		list_for_each(pos, &nlm_blocked) {
			b = list_entry(pos, struct nlm_block, b_list);
			if (time_after(b->b_when,when) || b->b_when == NLM_NEVER)
				break;
		}
		/* On normal exit from the loop, pos == &nlm_blocked,
		 * so we will be adding to the end of the list - good
		 */
	}

	list_add_tail(&block->b_list, pos);
	block->b_when = when;
}

/*
 * Remove a block from the global list
 */
static inline void
nlmsvc_remove_block(struct nlm_block *block)
{
	if (!list_empty(&block->b_list)) {
		list_del_init(&block->b_list);
		nlmsvc_release_block(block);
	}
}

/*
 * Find a block for a given lock
 */
static struct nlm_block *
nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock)
{
	struct nlm_block	*block;
	struct file_lock	*fl;

	dprintk("lockd: nlmsvc_lookup_block f=%p pd=%d %Ld-%Ld ty=%d\n",
				file, lock->fl.fl_pid,
				(long long)lock->fl.fl_start,
				(long long)lock->fl.fl_end, lock->fl.fl_type);
	list_for_each_entry(block, &nlm_blocked, b_list) {
		fl = &block->b_call->a_args.lock.fl;
		dprintk("lockd: check f=%p pd=%d %Ld-%Ld ty=%d cookie=%s\n",
				block->b_file, fl->fl_pid,
				(long long)fl->fl_start,
				(long long)fl->fl_end, fl->fl_type,
				nlmdbg_cookie2a(&block->b_call->a_args.cookie));
		if (block->b_file == file && nlm_compare_locks(fl, &lock->fl)) {
			kref_get(&block->b_count);
			return block;
		}
	}

	return NULL;
}

static inline int nlm_cookie_match(struct nlm_cookie *a, struct nlm_cookie *b)
{
	if (a->len != b->len)
		return 0;
	if (memcmp(a->data, b->data, a->len))
		return 0;
	return 1;
}

/*
 * Find a block with a given NLM cookie.
 */
static inline struct nlm_block *
nlmsvc_find_block(struct nlm_cookie *cookie)
{
	struct nlm_block *block;

	list_for_each_entry(block, &nlm_blocked, b_list) {
		if (nlm_cookie_match(&block->b_call->a_args.cookie,cookie))
			goto found;
	}

	return NULL;

found:
	dprintk("nlmsvc_find_block(%s): block=%p\n", nlmdbg_cookie2a(cookie), block);
	kref_get(&block->b_count);
	return block;
}

/*
 * Create a block and initialize it.
 *
 * Note: we explicitly set the cookie of the grant reply to that of
 * the blocked lock request. The spec explicitly mentions that the client
 * should _not_ rely on the callback containing the same cookie as the
 * request, but (as I found out later) that's because some implementations
 * do just this. Never mind the standards comittees, they support our
 * logging industries.
 *
 * 10 years later: I hope we can safely ignore these old and broken
 * clients by now. Let's fix this so we can uniquely identify an incoming
 * GRANTED_RES message by cookie, without having to rely on the client's IP
 * address. --okir
 */
static struct nlm_block *
nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_host *host,
		    struct nlm_file *file, struct nlm_lock *lock,
		    struct nlm_cookie *cookie)
{
	struct nlm_block	*block;
	struct nlm_rqst		*call = NULL;

	nlm_get_host(host);
	call = nlm_alloc_call(host);
	if (call == NULL)
		return NULL;

	/* Allocate memory for block, and initialize arguments */
	block = kzalloc(sizeof(*block), GFP_KERNEL);
	if (block == NULL)
		goto failed;
	kref_init(&block->b_count);
	INIT_LIST_HEAD(&block->b_list);
	INIT_LIST_HEAD(&block->b_flist);

	if (!nlmsvc_setgrantargs(call, lock))
		goto failed_free;

	/* Set notifier function for VFS, and init args */
	call->a_args.lock.fl.fl_flags |= FL_SLEEP;
	call->a_args.lock.fl.fl_lmops = &nlmsvc_lock_operations;
	nlmclnt_next_cookie(&call->a_args.cookie);

	dprintk("lockd: created block %p...\n", block);

	/* Create and initialize the block */
	block->b_daemon = rqstp->rq_server;
	block->b_host   = host;
	block->b_file   = file;
	block->b_fl = NULL;
	file->f_count++;

	/* Add to file's list of blocks */
	list_add(&block->b_flist, &file->f_blocks);

	/* Set up RPC arguments for callback */
	block->b_call = call;
	call->a_flags   = RPC_TASK_ASYNC;
	call->a_block = block;

	return block;

failed_free:
	kfree(block);
failed:
	nlm_release_call(call);
	return NULL;
}

/*
 * Delete a block.
 * It is the caller's responsibility to check whether the file
 * can be closed hereafter.
 */
static int nlmsvc_unlink_block(struct nlm_block *block)
{
	int status;
	dprintk("lockd: unlinking block %p...\n", block);

	/* Remove block from list */
	status = posix_unblock_lock(block->b_file->f_file, &block->b_call->a_args.lock.fl);
	nlmsvc_remove_block(block);
	return status;
}

static void nlmsvc_free_block(struct kref *kref)
{
	struct nlm_block *block = container_of(kref, struct nlm_block, b_count);
	struct nlm_file		*file = block->b_file;

	dprintk("lockd: freeing block %p...\n", block);

	/* Remove block from file's list of blocks */
	mutex_lock(&file->f_mutex);
	list_del_init(&block->b_flist);
	mutex_unlock(&file->f_mutex);

	nlmsvc_freegrantargs(block->b_call);
	nlm_release_call(block->b_call);
	nlm_release_file(block->b_file);
	kfree(block->b_fl);
	kfree(block);
}

static void nlmsvc_release_block(struct nlm_block *block)
{
	if (block != NULL)
		kref_put(&block->b_count, nlmsvc_free_block);
}

/*
 * Loop over all blocks and delete blocks held by
 * a matching host.
 */
void nlmsvc_traverse_blocks(struct nlm_host *host,
			struct nlm_file *file,
			nlm_host_match_fn_t match)
{
	struct nlm_block *block, *next;

restart:
	mutex_lock(&file->f_mutex);
	list_for_each_entry_safe(block, next, &file->f_blocks, b_flist) {
		if (!match(block->b_host, host))
			continue;
		/* Do not destroy blocks that are not on
		 * the global retry list - why? */
		if (list_empty(&block->b_list))
			continue;
		kref_get(&block->b_count);
		mutex_unlock(&file->f_mutex);
		nlmsvc_unlink_block(block);
		nlmsvc_release_block(block);
		goto restart;
	}
	mutex_unlock(&file->f_mutex);
}

/*
 * Initialize arguments for GRANTED call. The nlm_rqst structure
 * has been cleared already.
 */
static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock)
{
	locks_copy_lock(&call->a_args.lock.fl, &lock->fl);
	memcpy(&call->a_args.lock.fh, &lock->fh, sizeof(call->a_args.lock.fh));
	call->a_args.lock.caller = utsname()->nodename;
	call->a_args.lock.oh.len = lock->oh.len;

	/* set default data area */
	call->a_args.lock.oh.data = call->a_owner;
	call->a_args.lock.svid = lock->fl.fl_pid;

	if (lock->oh.len > NLMCLNT_OHSIZE) {
		void *data = kmalloc(lock->oh.len, GFP_KERNEL);
		if (!data)
			return 0;
		call->a_args.lock.oh.data = (u8 *) data;
	}

	memcpy(call->a_args.lock.oh.data, lock->oh.data, lock->oh.len);
	return 1;
}

static void nlmsvc_freegrantargs(struct nlm_rqst *call)
{
	if (call->a_args.lock.oh.data != call->a_owner)
		kfree(call->a_args.lock.oh.data);
}

/*
 * Deferred lock request handling for non-blocking lock
 */
static __be32
nlmsvc_defer_lock_rqst(struct svc_rqst *rqstp, struct nlm_block *block)
{
	__be32 status = nlm_lck_denied_nolocks;

	block->b_flags |= B_QUEUED;

	nlmsvc_insert_block(block, NLM_TIMEOUT);

	block->b_cache_req = &rqstp->rq_chandle;
	if (rqstp->rq_chandle.defer) {
		block->b_deferred_req =
			rqstp->rq_chandle.defer(block->b_cache_req);
		if (block->b_deferred_req != NULL)
			status = nlm_drop_reply;
	}
	dprintk("lockd: nlmsvc_defer_lock_rqst block %p flags %d status %d\n",
		block, block->b_flags, ntohl(status));

	return status;
}

/*
 * Attempt to establish a lock, and if it can't be granted, block it
 * if required.
 */
__be32
nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file,
	    struct nlm_host *host, struct nlm_lock *lock, int wait,
	    struct nlm_cookie *cookie, int reclaim)
{
	struct nlm_block	*block = NULL;
	int			error;
	__be32			ret;

	dprintk("lockd: nlmsvc_lock(%s/%ld, ty=%d, pi=%d, %Ld-%Ld, bl=%d)\n",
				file->f_file->f_path.dentry->d_inode->i_sb->s_id,
				file->f_file->f_path.dentry->d_inode->i_ino,
				lock->fl.fl_type, lock->fl.fl_pid,
				(long long)lock->fl.fl_start,
				(long long)lock->fl.fl_end,
				wait);

	/* Lock file against concurrent access */
	mutex_lock(&file->f_mutex);
	/* Get existing block (in case client is busy-waiting)
	 * or create new block
	 */
	block = nlmsvc_lookup_block(file, lock);
	if (block == NULL) {
		block = nlmsvc_create_block(rqstp, host, file, lock, cookie);
		ret = nlm_lck_denied_nolocks;
		if (block == NULL)
			goto out;
		lock = &block->b_call->a_args.lock;
	} else
		lock->fl.fl_flags &= ~FL_SLEEP;

	if (block->b_flags & B_QUEUED) {
		dprintk("lockd: nlmsvc_lock deferred block %p flags %d\n",
							block, block->b_flags);
		if (block->b_granted) {
			nlmsvc_unlink_block(block);
			ret = nlm_granted;
			goto out;
		}
		if (block->b_flags & B_TIMED_OUT) {
			nlmsvc_unlink_block(block);
			ret = nlm_lck_denied;
			goto out;
		}
		ret = nlm_drop_reply;
		goto out;
	}

	if (locks_in_grace() && !reclaim) {
		ret = nlm_lck_denied_grace_period;
		goto out;
	}
	if (reclaim && !locks_in_grace()) {
		ret = nlm_lck_denied_grace_period;
		goto out;
	}

	if (!wait)
		lock->fl.fl_flags &= ~FL_SLEEP;
	error = vfs_lock_file(file->f_file, F_SETLK, &lock->fl, NULL);
	lock->fl.fl_flags &= ~FL_SLEEP;

	dprintk("lockd: vfs_lock_file returned %d\n", error);
	switch (error) {
		case 0:
			ret = nlm_granted;
			goto out;
		case -EAGAIN:
			ret = nlm_lck_denied;
			break;
		case FILE_LOCK_DEFERRED:
			if (wait)
				break;
			/* Filesystem lock operation is in progress
			   Add it to the queue waiting for callback */
			ret = nlmsvc_defer_lock_rqst(rqstp, block);
			goto out;
		case -EDEADLK:
			ret = nlm_deadlock;
			goto out;
		default:			/* includes ENOLCK */
			ret = nlm_lck_denied_nolocks;
			goto out;
	}

	ret = nlm_lck_denied;
	if (!wait)
		goto out;

	ret = nlm_lck_blocked;

	/* Append to list of blocked */
	nlmsvc_insert_block(block, NLM_NEVER);
out:
	mutex_unlock(&file->f_mutex);
	nlmsvc_release_block(block);
	dprintk("lockd: nlmsvc_lock returned %u\n", ret);
	return ret;
}