NLM: Clean up nlmsvc_grant_reply locking

[linux-2.6] / fs / lockd / svclock.c

/*
 * 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 <okir@monad.swb.de>
 */

#include <linux/config.h>
#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>

#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_insert_block(struct nlm_block *block, unsigned long);
static int      nlmsvc_remove_block(struct nlm_block *block);

static const struct rpc_call_ops nlmsvc_grant_ops;

/*
 * The list of blocked locks to retry
 */
static struct nlm_block *       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 **bp, *b;

        dprintk("lockd: nlmsvc_insert_block(%p, %ld)\n", block, when);
        if (block->b_queued)
                nlmsvc_remove_block(block);
        bp = &nlm_blocked;
        if (when != NLM_NEVER) {
                if ((when += jiffies) == NLM_NEVER)
                        when ++;
                while ((b = *bp) && time_before_eq(b->b_when,when) && b->b_when != NLM_NEVER)
                        bp = &b->b_next;
        } else
                while ((b = *bp) != 0)
                        bp = &b->b_next;

        block->b_queued = 1;
        block->b_when = when;
        block->b_next = b;
        *bp = block;
}

/*
 * Remove a block from the global list
 */
static int
nlmsvc_remove_block(struct nlm_block *block)
{
        struct nlm_block **bp, *b;

        if (!block->b_queued)
                return 1;
        for (bp = &nlm_blocked; (b = *bp) != 0; bp = &b->b_next) {
                if (b == block) {
                        *bp = block->b_next;
                        block->b_queued = 0;
                        return 1;
                }
        }

        return 0;
}

/*
 * Find a block for a given lock and optionally remove it from
 * the list.
 */
static struct nlm_block *
nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock, int remove)
{
        struct nlm_block        **head, *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);
        for (head = &nlm_blocked; (block = *head) != 0; head = &block->b_next) {
                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)) {
                        if (remove) {
                                *head = block->b_next;
                                block->b_queued = 0;
                        }
                        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 sockaddr_in *sin)
{
        struct nlm_block *block;

        for (block = nlm_blocked; block; block = block->b_next) {
                dprintk("cookie: head of blocked queue %p, block %p\n", 
                        nlm_blocked, block);
                if (nlm_cookie_match(&block->b_call.a_args.cookie,cookie)
                                && nlm_cmp_addr(sin, &block->b_host->h_addr))
                        break;
        }

        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.
 */
static inline struct nlm_block *
nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_file *file,
                                struct nlm_lock *lock, struct nlm_cookie *cookie)
{
        struct nlm_block        *block;
        struct nlm_host         *host;
        struct nlm_rqst         *call;

        /* Create host handle for callback */
        host = nlmclnt_lookup_host(&rqstp->rq_addr,
                                rqstp->rq_prot, rqstp->rq_vers);
        if (host == NULL)
                return NULL;

        /* Allocate memory for block, and initialize arguments */
        if (!(block = (struct nlm_block *) kmalloc(sizeof(*block), GFP_KERNEL)))
                goto failed;
        memset(block, 0, sizeof(*block));
        locks_init_lock(&block->b_call.a_args.lock.fl);
        locks_init_lock(&block->b_call.a_res.lock.fl);

        if (!nlmclnt_setgrantargs(&block->b_call, lock))
                goto failed_free;

        /* Set notifier function for VFS, and init args */
        block->b_call.a_args.lock.fl.fl_lmops = &nlmsvc_lock_operations;
        block->b_call.a_args.cookie = *cookie;  /* see above */

        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;

        /* Add to file's list of blocks */
        block->b_fnext  = file->f_blocks;
        file->f_blocks  = block;

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

        return block;

failed_free:
        kfree(block);
failed:
        nlm_release_host(host);
        return NULL;
}

/*
 * Delete a block. If the lock was cancelled or the grant callback
 * failed, unlock is set to 1.
 * It is the caller's responsibility to check whether the file
 * can be closed hereafter.
 */
static void
nlmsvc_delete_block(struct nlm_block *block, int unlock)
{
        struct file_lock        *fl = &block->b_call.a_args.lock.fl;
        struct nlm_file         *file = block->b_file;
        struct nlm_block        **bp;

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

        /* Remove block from list */
        nlmsvc_remove_block(block);
        if (fl->fl_next)
                posix_unblock_lock(file->f_file, fl);
        if (unlock) {
                fl->fl_type = F_UNLCK;
                posix_lock_file(file->f_file, fl);
                block->b_granted = 0;
        }

        /* If the block is in the middle of a GRANT callback,
         * don't kill it yet. */
        if (block->b_incall) {
                nlmsvc_insert_block(block, NLM_NEVER);
                block->b_done = 1;
                return;
        }

        /* Remove block from file's list of blocks */
        for (bp = &file->f_blocks; *bp; bp = &(*bp)->b_fnext) {
                if (*bp == block) {
                        *bp = block->b_fnext;
                        break;
                }
        }

        if (block->b_host)
                nlm_release_host(block->b_host);
        nlmclnt_freegrantargs(&block->b_call);
        kfree(block);
}

/*
 * Loop over all blocks and perform the action specified.
 * (NLM_ACT_CHECK handled by nlmsvc_inspect_file).
 */
int
nlmsvc_traverse_blocks(struct nlm_host *host, struct nlm_file *file, int action)
{
        struct nlm_block        *block, *next;

        down(&file->f_sema);
        for (block = file->f_blocks; block; block = next) {
                next = block->b_fnext;
                if (action == NLM_ACT_MARK)
                        block->b_host->h_inuse = 1;
                else if (action == NLM_ACT_UNLOCK) {
                        if (host == NULL || host == block->b_host)
                                nlmsvc_delete_block(block, 1);
                }
        }
        up(&file->f_sema);
        return 0;
}

/*
 * Attempt to establish a lock, and if it can't be granted, block it
 * if required.
 */
u32
nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file,
                        struct nlm_lock *lock, int wait, struct nlm_cookie *cookie)
{
        struct file_lock        *conflock;
        struct nlm_block        *block;
        int                     error;

        dprintk("lockd: nlmsvc_lock(%s/%ld, ty=%d, pi=%d, %Ld-%Ld, bl=%d)\n",
                                file->f_file->f_dentry->d_inode->i_sb->s_id,
                                file->f_file->f_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);


        /* Get existing block (in case client is busy-waiting) */
        block = nlmsvc_lookup_block(file, lock, 0);

        lock->fl.fl_flags |= FL_LOCKD;

again:
        /* Lock file against concurrent access */
        down(&file->f_sema);

        if (!(conflock = posix_test_lock(file->f_file, &lock->fl))) {
                error = posix_lock_file(file->f_file, &lock->fl);

                if (block)
                        nlmsvc_delete_block(block, 0);
                up(&file->f_sema);

                dprintk("lockd: posix_lock_file returned %d\n", -error);
                switch(-error) {
                case 0:
                        return nlm_granted;
                case EDEADLK:
                        return nlm_deadlock;
                case EAGAIN:
                        return nlm_lck_denied;
                default:                        /* includes ENOLCK */
                        return nlm_lck_denied_nolocks;
                }
        }

        if (!wait) {
                up(&file->f_sema);
                return nlm_lck_denied;
        }

        if (posix_locks_deadlock(&lock->fl, conflock)) {
                up(&file->f_sema);
                return nlm_deadlock;
        }

        /* If we don't have a block, create and initialize it. Then
         * retry because we may have slept in kmalloc. */
        /* We have to release f_sema as nlmsvc_create_block may try to
         * to claim it while doing host garbage collection */
        if (block == NULL) {
                up(&file->f_sema);
                dprintk("lockd: blocking on this lock (allocating).\n");
                if (!(block = nlmsvc_create_block(rqstp, file, lock, cookie)))
                        return nlm_lck_denied_nolocks;
                goto again;
        }

        /* Append to list of blocked */
        nlmsvc_insert_block(block, NLM_NEVER);

        if (list_empty(&block->b_call.a_args.lock.fl.fl_block)) {
                /* Now add block to block list of the conflicting lock
                   if we haven't done so. */
                dprintk("lockd: blocking on this lock.\n");
                posix_block_lock(conflock, &block->b_call.a_args.lock.fl);
        }

        up(&file->f_sema);
        return nlm_lck_blocked;
}

/*
 * Test for presence of a conflicting lock.
 */
u32
nlmsvc_testlock(struct nlm_file *file, struct nlm_lock *lock,
                                       struct nlm_lock *conflock)
{
        struct file_lock        *fl;

        dprintk("lockd: nlmsvc_testlock(%s/%ld, ty=%d, %Ld-%Ld)\n",
                                file->f_file->f_dentry->d_inode->i_sb->s_id,
                                file->f_file->f_dentry->d_inode->i_ino,
                                lock->fl.fl_type,
                                (long long)lock->fl.fl_start,
                                (long long)lock->fl.fl_end);

        if ((fl = posix_test_lock(file->f_file, &lock->fl)) != NULL) {
                dprintk("lockd: conflicting lock(ty=%d, %Ld-%Ld)\n",
                                fl->fl_type, (long long)fl->fl_start,
                                (long long)fl->fl_end);
                conflock->caller = "somehost";  /* FIXME */
                conflock->oh.len = 0;           /* don't return OH info */
                conflock->fl = *fl;
                return nlm_lck_denied;
        }

        return nlm_granted;
}

/*
 * Remove a lock.
 * This implies a CANCEL call: We send a GRANT_MSG, the client replies
 * with a GRANT_RES call which gets lost, and calls UNLOCK immediately
 * afterwards. In this case the block will still be there, and hence
 * must be removed.
 */
u32
nlmsvc_unlock(struct nlm_file *file, struct nlm_lock *lock)
{
        int     error;

        dprintk("lockd: nlmsvc_unlock(%s/%ld, pi=%d, %Ld-%Ld)\n",
                                file->f_file->f_dentry->d_inode->i_sb->s_id,
                                file->f_file->f_dentry->d_inode->i_ino,
                                lock->fl.fl_pid,
                                (long long)lock->fl.fl_start,
                                (long long)lock->fl.fl_end);

        /* First, cancel any lock that might be there */
        nlmsvc_cancel_blocked(file, lock);

        lock->fl.fl_type = F_UNLCK;
        error = posix_lock_file(file->f_file, &lock->fl);

        return (error < 0)? nlm_lck_denied_nolocks : nlm_granted;
}

/*
 * Cancel a previously blocked request.
 *
 * A cancel request always overrides any grant that may currently
 * be in progress.
 * The calling procedure must check whether the file can be closed.
 */
u32
nlmsvc_cancel_blocked(struct nlm_file *file, struct nlm_lock *lock)
{
        struct nlm_block        *block;

        dprintk("lockd: nlmsvc_cancel(%s/%ld, pi=%d, %Ld-%Ld)\n",
                                file->f_file->f_dentry->d_inode->i_sb->s_id,
                                file->f_file->f_dentry->d_inode->i_ino,
                                lock->fl.fl_pid,
                                (long long)lock->fl.fl_start,
                                (long long)lock->fl.fl_end);

        down(&file->f_sema);
        if ((block = nlmsvc_lookup_block(file, lock, 1)) != NULL)
                nlmsvc_delete_block(block, 1);
        up(&file->f_sema);
        return nlm_granted;
}

/*
 * Unblock a blocked lock request. This is a callback invoked from the
 * VFS layer when a lock on which we blocked is removed.
 *
 * This function doesn't grant the blocked lock instantly, but rather moves
 * the block to the head of nlm_blocked where it can be picked up by lockd.
 */
static void
nlmsvc_notify_blocked(struct file_lock *fl)
{
        struct nlm_block        **bp, *block;

        dprintk("lockd: VFS unblock notification for block %p\n", fl);
        for (bp = &nlm_blocked; (block = *bp) != 0; bp = &block->b_next) {
                if (nlm_compare_locks(&block->b_call.a_args.lock.fl, fl)) {
                        nlmsvc_insert_block(block, 0);
                        svc_wake_up(block->b_daemon);
                        return;
                }
        }

        printk(KERN_WARNING "lockd: notification for unknown block!\n");
}

static int nlmsvc_same_owner(struct file_lock *fl1, struct file_lock *fl2)
{
        return fl1->fl_owner == fl2->fl_owner && fl1->fl_pid == fl2->fl_pid;
}

struct lock_manager_operations nlmsvc_lock_operations = {
        .fl_compare_owner = nlmsvc_same_owner,
        .fl_notify = nlmsvc_notify_blocked,
};

/*
 * Try to claim a lock that was previously blocked.
 *
 * Note that we use both the RPC_GRANTED_MSG call _and_ an async
 * RPC thread when notifying the client. This seems like overkill...
 * Here's why:
 *  -   we don't want to use a synchronous RPC thread, otherwise
 *      we might find ourselves hanging on a dead portmapper.
 *  -   Some lockd implementations (e.g. HP) don't react to
 *      RPC_GRANTED calls; they seem to insist on RPC_GRANTED_MSG calls.
 */
static void
nlmsvc_grant_blocked(struct nlm_block *block)
{
        struct nlm_file         *file = block->b_file;
        struct nlm_lock         *lock = &block->b_call.a_args.lock;
        struct file_lock        *conflock;
        int                     error;

        dprintk("lockd: grant blocked lock %p\n", block);

        /* First thing is lock the file */
        down(&file->f_sema);

        /* Unlink block request from list */
        nlmsvc_remove_block(block);

        /* If b_granted is true this means we've been here before.
         * Just retry the grant callback, possibly refreshing the RPC
         * binding */
        if (block->b_granted) {
                nlm_rebind_host(block->b_host);
                goto callback;
        }

        /* Try the lock operation again */
        if ((conflock = posix_test_lock(file->f_file, &lock->fl)) != NULL) {
                /* Bummer, we blocked again */
                dprintk("lockd: lock still blocked\n");
                nlmsvc_insert_block(block, NLM_NEVER);
                posix_block_lock(conflock, &lock->fl);
                up(&file->f_sema);
                return;
        }

        /* Alright, no conflicting lock. Now lock it for real. If the
         * following yields an error, this is most probably due to low
         * memory. Retry the lock in a few seconds.
         */
        if ((error = posix_lock_file(file->f_file, &lock->fl)) < 0) {
                printk(KERN_WARNING "lockd: unexpected error %d in %s!\n",
                                -error, __FUNCTION__);
                nlmsvc_insert_block(block, 10 * HZ);
                up(&file->f_sema);
                return;
        }

callback:
        /* Lock was granted by VFS. */
        dprintk("lockd: GRANTing blocked lock.\n");
        block->b_granted = 1;
        block->b_incall  = 1;

        /* Schedule next grant callback in 30 seconds */
        nlmsvc_insert_block(block, 30 * HZ);

        /* Call the client */
        nlm_get_host(block->b_call.a_host);
        if (nlmsvc_async_call(&block->b_call, NLMPROC_GRANTED_MSG,
                                                &nlmsvc_grant_ops) < 0)
                nlm_release_host(block->b_call.a_host);
        up(&file->f_sema);
}

/*
 * This is the callback from the RPC layer when the NLM_GRANTED_MSG
 * RPC call has succeeded or timed out.
 * Like all RPC callbacks, it is invoked by the rpciod process, so it
 * better not sleep. Therefore, we put the blocked lock on the nlm_blocked
 * chain once more in order to have it removed by lockd itself (which can
 * then sleep on the file semaphore without disrupting e.g. the nfs client).
 */
static void nlmsvc_grant_callback(struct rpc_task *task, void *data)
{
        struct nlm_rqst         *call = data;
        struct nlm_block        *block;
        unsigned long           timeout;
        struct sockaddr_in      *peer_addr = RPC_PEERADDR(task->tk_client);

        dprintk("lockd: GRANT_MSG RPC callback\n");
        dprintk("callback: looking for cookie %s, host (%u.%u.%u.%u)\n",
                nlmdbg_cookie2a(&call->a_args.cookie),
                NIPQUAD(peer_addr->sin_addr.s_addr));
        if (!(block = nlmsvc_find_block(&call->a_args.cookie, peer_addr))) {
                dprintk("lockd: no block for cookie %s, host (%u.%u.%u.%u)\n",
                        nlmdbg_cookie2a(&call->a_args.cookie),
                        NIPQUAD(peer_addr->sin_addr.s_addr));
                return;
        }

        /* Technically, we should down the file semaphore here. Since we
         * move the block towards the head of the queue only, no harm
         * can be done, though. */
        if (task->tk_status < 0) {
                /* RPC error: Re-insert for retransmission */
                timeout = 10 * HZ;
        } else if (block->b_done) {
                /* Block already removed, kill it for real */
                timeout = 0;
        } else {
                /* Call was successful, now wait for client callback */
                timeout = 60 * HZ;
        }
        nlmsvc_insert_block(block, timeout);
        svc_wake_up(block->b_daemon);
        block->b_incall = 0;

        nlm_release_host(call->a_host);
}

static const struct rpc_call_ops nlmsvc_grant_ops = {
        .rpc_call_done = nlmsvc_grant_callback,
};

/*
 * We received a GRANT_RES callback. Try to find the corresponding
 * block.
 */
void
nlmsvc_grant_reply(struct svc_rqst *rqstp, struct nlm_cookie *cookie, u32 status)
{
        struct nlm_block        *block;
        struct nlm_file         *file;

        dprintk("grant_reply: looking for cookie %x, host (%08x), s=%d \n", 
                *(unsigned int *)(cookie->data), 
                ntohl(rqstp->rq_addr.sin_addr.s_addr), status);
        if (!(block = nlmsvc_find_block(cookie, &rqstp->rq_addr)))
                return;
        file = block->b_file;

        file->f_count++;
        down(&file->f_sema);
        block = nlmsvc_find_block(cookie, &rqstp->rq_addr);
        if (block) {
                if (status == NLM_LCK_DENIED_GRACE_PERIOD) {
                        /* Try again in a couple of seconds */
                        nlmsvc_insert_block(block, 10 * HZ);
                        up(&file->f_sema);
                } else {
                        /* Lock is now held by client, or has been rejected.
                         * In both cases, the block should be removed. */
                        up(&file->f_sema);
                        if (status == NLM_LCK_GRANTED)
                                nlmsvc_delete_block(block, 0);
                        else
                                nlmsvc_delete_block(block, 1);
                }
        }
        nlm_release_file(file);
}

/*
 * Retry all blocked locks that have been notified. This is where lockd
 * picks up locks that can be granted, or grant notifications that must
 * be retransmitted.
 */
unsigned long
nlmsvc_retry_blocked(void)
{
        struct nlm_block        *block;

        dprintk("nlmsvc_retry_blocked(%p, when=%ld)\n",
                        nlm_blocked,
                        nlm_blocked? nlm_blocked->b_when : 0);
        while ((block = nlm_blocked) != 0) {
                if (block->b_when == NLM_NEVER)
                        break;
                if (time_after(block->b_when,jiffies))
                        break;
                dprintk("nlmsvc_retry_blocked(%p, when=%ld, done=%d)\n",
                        block, block->b_when, block->b_done);
                if (block->b_done)
                        nlmsvc_delete_block(block, 0);
                else
                        nlmsvc_grant_blocked(block);
        }

        if ((block = nlm_blocked) && block->b_when != NLM_NEVER)
                return (block->b_when - jiffies);

        return MAX_SCHEDULE_TIMEOUT;
}