1 files changed, 386 insertions, 89 deletions

diff --git a/fs/nfsd/nfscache.c b/fs/nfsd/nfscache.c
index 2cbac34a55da..e76244edd748 100644
--- a/fs/nfsd/nfscache.c
+++ b/fs/nfsd/nfscache.c
@@ -9,34 +9,63 @@
  */
 
 #include <linux/slab.h>
+#include <linux/sunrpc/addr.h>
+#include <linux/highmem.h>
+#include <linux/log2.h>
+#include <linux/hash.h>
+#include <net/checksum.h>
 
 #include "nfsd.h"
 #include "cache.h"
 
-/* Size of reply cache. Common values are:
- * 4.3BSD:      128
- * 4.4BSD:      256
- * Solaris2:    1024
- * DEC Unix:    512-4096
+#define NFSDDBG_FACILITY        NFSDDBG_REPCACHE
+
+/*
+ * We use this value to determine the number of hash buckets from the max
+ * cache size, the idea being that when the cache is at its maximum number
+ * of entries, then this should be the average number of entries per bucket.
  */
-#define CACHESIZE               1024
-#define HASHSIZE                64
+#define TARGET_BUCKET_SIZE      64
 
 static struct hlist_head *      cache_hash;
 static struct list_head         lru_head;
-static int                      cache_disabled = 1;
+static struct kmem_cache        *drc_slab;
+
+/* max number of entries allowed in the cache */
+static unsigned int             max_drc_entries;
+
+/* number of significant bits in the hash value */
+static unsigned int             maskbits;
 
 /*
- * Calculate the hash index from an XID.
+ * Stats and other tracking of on the duplicate reply cache. All of these and
+ * the "rc" fields in nfsdstats are protected by the cache_lock
  */
-static inline u32 request_hash(u32 xid)
-{
-        u32 h = xid;
-        h ^= (xid >> 24);
-        return h & (HASHSIZE-1);
-}
+
+/* total number of entries */
+static unsigned int             num_drc_entries;
+
+/* cache misses due only to checksum comparison failures */
+static unsigned int             payload_misses;
+
+/* amount of memory (in bytes) currently consumed by the DRC */
+static unsigned int             drc_mem_usage;
+
+/* longest hash chain seen */
+static unsigned int             longest_chain;
+
+/* size of cache when we saw the longest hash chain */
+static unsigned int             longest_chain_cachesize;
 
 static int      nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *vec);
+static void     cache_cleaner_func(struct work_struct *unused);
+static int      nfsd_reply_cache_shrink(struct shrinker *shrink,
+                                        struct shrink_control *sc);
+
+static struct shrinker nfsd_reply_cache_shrinker = {
+        .shrink = nfsd_reply_cache_shrink,
+        .seeks  = 1,
+};
 
 /*
  * locking for the reply cache:
@@ -44,30 +73,104 @@ static int	nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *vec);
  * Otherwise, it when accessing _prev or _next, the lock must be held.
  */
 static DEFINE_SPINLOCK(cache_lock);
+static DECLARE_DELAYED_WORK(cache_cleaner, cache_cleaner_func);
 
-int nfsd_reply_cache_init(void)
+/*
+ * Put a cap on the size of the DRC based on the amount of available
+ * low memory in the machine.
+ *
+ *  64MB:    8192
+ * 128MB:   11585
+ * 256MB:   16384
+ * 512MB:   23170
+ *   1GB:   32768
+ *   2GB:   46340
+ *   4GB:   65536
+ *   8GB:   92681
+ *  16GB:  131072
+ *
+ * ...with a hard cap of 256k entries. In the worst case, each entry will be
+ * ~1k, so the above numbers should give a rough max of the amount of memory
+ * used in k.
+ */
+static unsigned int
+nfsd_cache_size_limit(void)
+{
+        unsigned int limit;
+        unsigned long low_pages = totalram_pages - totalhigh_pages;
+
+        limit = (16 * int_sqrt(low_pages)) << (PAGE_SHIFT-10);
+        return min_t(unsigned int, limit, 256*1024);
+}
+
+/*
+ * Compute the number of hash buckets we need. Divide the max cachesize by
+ * the "target" max bucket size, and round up to next power of two.
+ */
+static unsigned int
+nfsd_hashsize(unsigned int limit)
+{
+        return roundup_pow_of_two(limit / TARGET_BUCKET_SIZE);
+}
+
+static struct svc_cacherep *
+nfsd_reply_cache_alloc(void)
 {
         struct svc_cacherep     *rp;
-        int                     i;
 
-        INIT_LIST_HEAD(&lru_head);
-        i = CACHESIZE;
-        while (i) {
-                rp = kmalloc(sizeof(*rp), GFP_KERNEL);
-                if (!rp)
-                        goto out_nomem;
-                list_add(&rp->c_lru, &lru_head);
+        rp = kmem_cache_alloc(drc_slab, GFP_KERNEL);
+        if (rp) {
                 rp->c_state = RC_UNUSED;
                 rp->c_type = RC_NOCACHE;
+                INIT_LIST_HEAD(&rp->c_lru);
                 INIT_HLIST_NODE(&rp->c_hash);
-                i--;
         }
+        return rp;
+}
+
+static void
+nfsd_reply_cache_free_locked(struct svc_cacherep *rp)
+{
+        if (rp->c_type == RC_REPLBUFF && rp->c_replvec.iov_base) {
+                drc_mem_usage -= rp->c_replvec.iov_len;
+                kfree(rp->c_replvec.iov_base);
+        }
+        if (!hlist_unhashed(&rp->c_hash))
+                hlist_del(&rp->c_hash);
+        list_del(&rp->c_lru);
+        --num_drc_entries;
+        drc_mem_usage -= sizeof(*rp);
+        kmem_cache_free(drc_slab, rp);
+}
 
-        cache_hash = kcalloc (HASHSIZE, sizeof(struct hlist_head), GFP_KERNEL);
+static void
+nfsd_reply_cache_free(struct svc_cacherep *rp)
+{
+        spin_lock(&cache_lock);
+        nfsd_reply_cache_free_locked(rp);
+        spin_unlock(&cache_lock);
+}
+
+int nfsd_reply_cache_init(void)
+{
+        unsigned int hashsize;
+
+        INIT_LIST_HEAD(&lru_head);
+        max_drc_entries = nfsd_cache_size_limit();
+        num_drc_entries = 0;
+        hashsize = nfsd_hashsize(max_drc_entries);
+        maskbits = ilog2(hashsize);
+
+        register_shrinker(&nfsd_reply_cache_shrinker);
+        drc_slab = kmem_cache_create("nfsd_drc", sizeof(struct svc_cacherep),
+                                        0, 0, NULL);
+        if (!drc_slab)
+                goto out_nomem;
+
+        cache_hash = kcalloc(hashsize, sizeof(struct hlist_head), GFP_KERNEL);
         if (!cache_hash)
                 goto out_nomem;
 
-        cache_disabled = 0;
         return 0;
 out_nomem:
         printk(KERN_ERR "nfsd: failed to allocate reply cache\n");
@@ -79,27 +182,33 @@ void nfsd_reply_cache_shutdown(void)
 {
         struct svc_cacherep     *rp;
 
+        unregister_shrinker(&nfsd_reply_cache_shrinker);
+        cancel_delayed_work_sync(&cache_cleaner);
+
         while (!list_empty(&lru_head)) {
                 rp = list_entry(lru_head.next, struct svc_cacherep, c_lru);
-                if (rp->c_state == RC_DONE && rp->c_type == RC_REPLBUFF)
-                        kfree(rp->c_replvec.iov_base);
-                list_del(&rp->c_lru);
-                kfree(rp);
+                nfsd_reply_cache_free_locked(rp);
         }
 
-        cache_disabled = 1;
-
         kfree (cache_hash);
         cache_hash = NULL;
+
+        if (drc_slab) {
+                kmem_cache_destroy(drc_slab);
+                drc_slab = NULL;
+        }
 }
 
 /*
- * Move cache entry to end of LRU list
+ * Move cache entry to end of LRU list, and queue the cleaner to run if it's
+ * not already scheduled.
  */
 static void
 lru_put_end(struct svc_cacherep *rp)
 {
+        rp->c_timestamp = jiffies;
         list_move_tail(&rp->c_lru, &lru_head);
+        schedule_delayed_work(&cache_cleaner, RC_EXPIRE);
 }
 
 /*
@@ -109,89 +218,247 @@ static void
 hash_refile(struct svc_cacherep *rp)
 {
         hlist_del_init(&rp->c_hash);
-        hlist_add_head(&rp->c_hash, cache_hash + request_hash(rp->c_xid));
+        hlist_add_head(&rp->c_hash, cache_hash + hash_32(rp->c_xid, maskbits));
+}
+
+static inline bool
+nfsd_cache_entry_expired(struct svc_cacherep *rp)
+{
+        return rp->c_state != RC_INPROG &&
+               time_after(jiffies, rp->c_timestamp + RC_EXPIRE);
+}
+
+/*
+ * Walk the LRU list and prune off entries that are older than RC_EXPIRE.
+ * Also prune the oldest ones when the total exceeds the max number of entries.
+ */
+static void
+prune_cache_entries(void)
+{
+        struct svc_cacherep *rp, *tmp;
+
+        list_for_each_entry_safe(rp, tmp, &lru_head, c_lru) {
+                if (!nfsd_cache_entry_expired(rp) &&
+                    num_drc_entries <= max_drc_entries)
+                        break;
+                nfsd_reply_cache_free_locked(rp);
+        }
+
+        /*
+         * Conditionally rearm the job. If we cleaned out the list, then
+         * cancel any pending run (since there won't be any work to do).
+         * Otherwise, we rearm the job or modify the existing one to run in
+         * RC_EXPIRE since we just ran the pruner.
+         */
+        if (list_empty(&lru_head))
+                cancel_delayed_work(&cache_cleaner);
+        else
+                mod_delayed_work(system_wq, &cache_cleaner, RC_EXPIRE);
+}
+
+static void
+cache_cleaner_func(struct work_struct *unused)
+{
+        spin_lock(&cache_lock);
+        prune_cache_entries();
+        spin_unlock(&cache_lock);
+}
+
+static int
+nfsd_reply_cache_shrink(struct shrinker *shrink, struct shrink_control *sc)
+{
+        unsigned int num;
+
+        spin_lock(&cache_lock);
+        if (sc->nr_to_scan)
+                prune_cache_entries();
+        num = num_drc_entries;
+        spin_unlock(&cache_lock);
+
+        return num;
+}
+
+/*
+ * Walk an xdr_buf and get a CRC for at most the first RC_CSUMLEN bytes
+ */
+static __wsum
+nfsd_cache_csum(struct svc_rqst *rqstp)
+{
+        int idx;
+        unsigned int base;
+        __wsum csum;
+        struct xdr_buf *buf = &rqstp->rq_arg;
+        const unsigned char *p = buf->head[0].iov_base;
+        size_t csum_len = min_t(size_t, buf->head[0].iov_len + buf->page_len,
+                                RC_CSUMLEN);
+        size_t len = min(buf->head[0].iov_len, csum_len);
+
+        /* rq_arg.head first */
+        csum = csum_partial(p, len, 0);
+        csum_len -= len;
+
+        /* Continue into page array */
+        idx = buf->page_base / PAGE_SIZE;
+        base = buf->page_base & ~PAGE_MASK;
+        while (csum_len) {
+                p = page_address(buf->pages[idx]) + base;
+                len = min_t(size_t, PAGE_SIZE - base, csum_len);
+                csum = csum_partial(p, len, csum);
+                csum_len -= len;
+                base = 0;
+                ++idx;
+        }
+        return csum;
+}
+
+static bool
+nfsd_cache_match(struct svc_rqst *rqstp, __wsum csum, struct svc_cacherep *rp)
+{
+        /* Check RPC header info first */
+        if (rqstp->rq_xid != rp->c_xid || rqstp->rq_proc != rp->c_proc ||
+            rqstp->rq_prot != rp->c_prot || rqstp->rq_vers != rp->c_vers ||
+            rqstp->rq_arg.len != rp->c_len ||
+            !rpc_cmp_addr(svc_addr(rqstp), (struct sockaddr *)&rp->c_addr) ||
+            rpc_get_port(svc_addr(rqstp)) != rpc_get_port((struct sockaddr *)&rp->c_addr))
+                return false;
+
+        /* compare checksum of NFS data */
+        if (csum != rp->c_csum) {
+                ++payload_misses;
+                return false;
+        }
+
+        return true;
+}
+
+/*
+ * Search the request hash for an entry that matches the given rqstp.
+ * Must be called with cache_lock held. Returns the found entry or
+ * NULL on failure.
+ */
+static struct svc_cacherep *
+nfsd_cache_search(struct svc_rqst *rqstp, __wsum csum)
+{
+        struct svc_cacherep     *rp, *ret = NULL;
+        struct hlist_head       *rh;
+        unsigned int            entries = 0;
+
+        rh = &cache_hash[hash_32(rqstp->rq_xid, maskbits)];
+        hlist_for_each_entry(rp, rh, c_hash) {
+                ++entries;
+                if (nfsd_cache_match(rqstp, csum, rp)) {
+                        ret = rp;
+                        break;
+                }
+        }
+
+        /* tally hash chain length stats */
+        if (entries > longest_chain) {
+                longest_chain = entries;
+                longest_chain_cachesize = num_drc_entries;
+        } else if (entries == longest_chain) {
+                /* prefer to keep the smallest cachesize possible here */
+                longest_chain_cachesize = min(longest_chain_cachesize,
+                                                num_drc_entries);
+        }
+
+        return ret;
 }
 
 /*
  * Try to find an entry matching the current call in the cache. When none
- * is found, we grab the oldest unlocked entry off the LRU list.
- * Note that no operation within the loop may sleep.
+ * is found, we try to grab the oldest expired entry off the LRU list. If
+ * a suitable one isn't there, then drop the cache_lock and allocate a
+ * new one, then search again in case one got inserted while this thread
+ * didn't hold the lock.
  */
 int
 nfsd_cache_lookup(struct svc_rqst *rqstp)
 {
-        struct hlist_node       *hn;
-        struct hlist_head       *rh;
-        struct svc_cacherep     *rp;
+        struct svc_cacherep     *rp, *found;
         __be32                  xid = rqstp->rq_xid;
         u32                     proto =  rqstp->rq_prot,
                                 vers = rqstp->rq_vers,
                                 proc = rqstp->rq_proc;
+        __wsum                  csum;
         unsigned long           age;
         int type = rqstp->rq_cachetype;
-        int rtn;
+        int rtn = RC_DOIT;
 
         rqstp->rq_cacherep = NULL;
-        if (cache_disabled || type == RC_NOCACHE) {
+        if (type == RC_NOCACHE) {
                 nfsdstats.rcnocache++;
-                return RC_DOIT;
+                return rtn;
         }
 
-        spin_lock(&cache_lock);
-        rtn = RC_DOIT;
+        csum = nfsd_cache_csum(rqstp);
 
-        rh = &cache_hash[request_hash(xid)];
-        hlist_for_each_entry(rp, hn, rh, c_hash) {
-                if (rp->c_state != RC_UNUSED &&
-                    xid == rp->c_xid && proc == rp->c_proc &&
-                    proto == rp->c_prot && vers == rp->c_vers &&
-                    time_before(jiffies, rp->c_timestamp + 120*HZ) &&
-                    memcmp((char*)&rqstp->rq_addr, (char*)&rp->c_addr, sizeof(rp->c_addr))==0) {
-                        nfsdstats.rchits++;
-                        goto found_entry;
+        /*
+         * Since the common case is a cache miss followed by an insert,
+         * preallocate an entry. First, try to reuse the first entry on the LRU
+         * if it works, then go ahead and prune the LRU list.
+         */
+        spin_lock(&cache_lock);
+        if (!list_empty(&lru_head)) {
+                rp = list_first_entry(&lru_head, struct svc_cacherep, c_lru);
+                if (nfsd_cache_entry_expired(rp) ||
+                    num_drc_entries >= max_drc_entries) {
+                        lru_put_end(rp);
+                        prune_cache_entries();
+                        goto search_cache;
                 }
         }
-        nfsdstats.rcmisses++;
 
-        /* This loop shouldn't take more than a few iterations normally */
-        {
-        int     safe = 0;
-        list_for_each_entry(rp, &lru_head, c_lru) {
-                if (rp->c_state != RC_INPROG)
-                        break;
-                if (safe++ > CACHESIZE) {
-                        printk("nfsd: loop in repcache LRU list\n");
-                        cache_disabled = 1;
-                        goto out;
-                }
-        }
+        /* No expired ones available, allocate a new one. */
+        spin_unlock(&cache_lock);
+        rp = nfsd_reply_cache_alloc();
+        spin_lock(&cache_lock);
+        if (likely(rp)) {
+                ++num_drc_entries;
+                drc_mem_usage += sizeof(*rp);
         }
 
-        /* All entries on the LRU are in-progress. This should not happen */
-        if (&rp->c_lru == &lru_head) {
-                static int      complaints;
+search_cache:
+        found = nfsd_cache_search(rqstp, csum);
+        if (found) {
+                if (likely(rp))
+                        nfsd_reply_cache_free_locked(rp);
+                rp = found;
+                goto found_entry;
+        }
 
-                printk(KERN_WARNING "nfsd: all repcache entries locked!\n");
-                if (++complaints > 5) {
-                        printk(KERN_WARNING "nfsd: disabling repcache.\n");
-                        cache_disabled = 1;
-                }
+        if (!rp) {
+                dprintk("nfsd: unable to allocate DRC entry!\n");
                 goto out;
         }
 
+        /*
+         * We're keeping the one we just allocated. Are we now over the
+         * limit? Prune one off the tip of the LRU in trade for the one we
+         * just allocated if so.
+         */
+        if (num_drc_entries >= max_drc_entries)
+                nfsd_reply_cache_free_locked(list_first_entry(&lru_head,
+                                                struct svc_cacherep, c_lru));
+
+        nfsdstats.rcmisses++;
         rqstp->rq_cacherep = rp;
         rp->c_state = RC_INPROG;
         rp->c_xid = xid;
         rp->c_proc = proc;
-        memcpy(&rp->c_addr, svc_addr_in(rqstp), sizeof(rp->c_addr));
+        rpc_copy_addr((struct sockaddr *)&rp->c_addr, svc_addr(rqstp));
+        rpc_set_port((struct sockaddr *)&rp->c_addr, rpc_get_port(svc_addr(rqstp)));
         rp->c_prot = proto;
         rp->c_vers = vers;
-        rp->c_timestamp = jiffies;
+        rp->c_len = rqstp->rq_arg.len;
+        rp->c_csum = csum;
 
         hash_refile(rp);
+        lru_put_end(rp);
 
         /* release any buffer */
         if (rp->c_type == RC_REPLBUFF) {
+                drc_mem_usage -= rp->c_replvec.iov_len;
                 kfree(rp->c_replvec.iov_base);
                 rp->c_replvec.iov_base = NULL;
         }
@@ -201,9 +468,9 @@ nfsd_cache_lookup(struct svc_rqst *rqstp)
         return rtn;
 
 found_entry:
+        nfsdstats.rchits++;
         /* We found a matching entry which is either in progress or done. */
         age = jiffies - rp->c_timestamp;
-        rp->c_timestamp = jiffies;
         lru_put_end(rp);
 
         rtn = RC_DROPIT;
@@ -232,7 +499,7 @@ found_entry:
                 break;
         default:
                 printk(KERN_WARNING "nfsd: bad repcache type %d\n", rp->c_type);
-                rp->c_state = RC_UNUSED;
+                nfsd_reply_cache_free_locked(rp);
         }
 
         goto out;
@@ -257,11 +524,12 @@ found_entry:
 void
 nfsd_cache_update(struct svc_rqst *rqstp, int cachetype, __be32 *statp)
 {
-        struct svc_cacherep *rp;
+        struct svc_cacherep *rp = rqstp->rq_cacherep;
         struct kvec     *resv = &rqstp->rq_res.head[0], *cachv;
         int             len;
+        size_t          bufsize = 0;
 
-        if (!(rp = rqstp->rq_cacherep) || cache_disabled)
+        if (!rp)
                 return;
 
         len = resv->iov_len - ((char*)statp - (char*)resv->iov_base);
@@ -269,7 +537,7 @@ nfsd_cache_update(struct svc_rqst *rqstp, int cachetype, __be32 *statp)
 
         /* Don't cache excessive amounts of data and XDR failures */
         if (!statp || len > (256 >> 2)) {
-                rp->c_state = RC_UNUSED;
+                nfsd_reply_cache_free(rp);
                 return;
         }
 
@@ -281,23 +549,25 @@ nfsd_cache_update(struct svc_rqst *rqstp, int cachetype, __be32 *statp)
                 break;
         case RC_REPLBUFF:
                 cachv = &rp->c_replvec;
-                cachv->iov_base = kmalloc(len << 2, GFP_KERNEL);
+                bufsize = len << 2;
+                cachv->iov_base = kmalloc(bufsize, GFP_KERNEL);
                 if (!cachv->iov_base) {
-                        spin_lock(&cache_lock);
-                        rp->c_state = RC_UNUSED;
-                        spin_unlock(&cache_lock);
+                        nfsd_reply_cache_free(rp);
                         return;
                 }
-                cachv->iov_len = len << 2;
-                memcpy(cachv->iov_base, statp, len << 2);
+                cachv->iov_len = bufsize;
+                memcpy(cachv->iov_base, statp, bufsize);
                 break;
+        case RC_NOCACHE:
+                nfsd_reply_cache_free(rp);
+                return;
         }
         spin_lock(&cache_lock);
+        drc_mem_usage += bufsize;
         lru_put_end(rp);
         rp->c_secure = rqstp->rq_secure;
         rp->c_type = cachetype;
         rp->c_state = RC_DONE;
-        rp->c_timestamp = jiffies;
         spin_unlock(&cache_lock);
         return;
 }
@@ -321,3 +591,30 @@ nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *data)
         vec->iov_len += data->iov_len;
         return 1;
 }
+
+/*
+ * Note that fields may be added, removed or reordered in the future. Programs
+ * scraping this file for info should test the labels to ensure they're
+ * getting the correct field.
+ */
+static int nfsd_reply_cache_stats_show(struct seq_file *m, void *v)
+{
+        spin_lock(&cache_lock);
+        seq_printf(m, "max entries:           %u\n", max_drc_entries);
+        seq_printf(m, "num entries:           %u\n", num_drc_entries);
+        seq_printf(m, "hash buckets:          %u\n", 1 << maskbits);
+        seq_printf(m, "mem usage:             %u\n", drc_mem_usage);
+        seq_printf(m, "cache hits:            %u\n", nfsdstats.rchits);
+        seq_printf(m, "cache misses:          %u\n", nfsdstats.rcmisses);
+        seq_printf(m, "not cached:            %u\n", nfsdstats.rcnocache);
+        seq_printf(m, "payload misses:        %u\n", payload_misses);
+        seq_printf(m, "longest chain len:     %u\n", longest_chain);
+        seq_printf(m, "cachesize at longest:  %u\n", longest_chain_cachesize);
+        spin_unlock(&cache_lock);
+        return 0;
+}
+
+int nfsd_reply_cache_stats_open(struct inode *inode, struct file *file)
+{
+        return single_open(file, nfsd_reply_cache_stats_show, NULL);
+}