[PATCH] NFS: large non-page-aligned direct I/O clobbers memory

The logic in nfs_direct_read_schedule and nfs_direct_write_schedule can
allow data->npages to be one larger than rpages.  This causes a page
pointer to be written beyond the end of the pagevec in nfs_read_data (or
nfs_write_data).

Fix this by making nfs_(read|write)_alloc() calculate the size of the
pagevec array, and initialise data->npages.

Also get rid of the redundant argument to nfs_commit_alloc().

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Cc: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

3 files changed, 42 insertions, 69 deletions

diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c
index fecd3b095deb..76ca1cbc38f9 100644
--- a/fs/nfs/direct.c
+++ b/fs/nfs/direct.c
@@ -100,25 +100,6 @@ static inline int put_dreq(struct nfs_direct_req *dreq)
         return atomic_dec_and_test(&dreq->io_count);
 }
 
-/*
- * "size" is never larger than rsize or wsize.
- */
-static inline int nfs_direct_count_pages(unsigned long user_addr, size_t size)
-{
-        int page_count;
-
-        page_count = (user_addr + size + PAGE_SIZE - 1) >> PAGE_SHIFT;
-        page_count -= user_addr >> PAGE_SHIFT;
-        BUG_ON(page_count < 0);
-
-        return page_count;
-}
-
-static inline unsigned int nfs_max_pages(unsigned int size)
-{
-        return (size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
-}
-
 /**
  * nfs_direct_IO - NFS address space operation for direct I/O
  * @rw: direction (read or write)
@@ -276,28 +257,24 @@ static ssize_t nfs_direct_read_schedule(struct nfs_direct_req *dreq, unsigned lo
         struct nfs_open_context *ctx = dreq->ctx;
         struct inode *inode = ctx->dentry->d_inode;
         size_t rsize = NFS_SERVER(inode)->rsize;
-        unsigned int rpages = nfs_max_pages(rsize);
         unsigned int pgbase;
         int result;
         ssize_t started = 0;
 
         get_dreq(dreq);
 
-        pgbase = user_addr & ~PAGE_MASK;
         do {
                 struct nfs_read_data *data;
                 size_t bytes;
 
+                pgbase = user_addr & ~PAGE_MASK;
+                bytes = min(rsize,count);
+
                 result = -ENOMEM;
-                data = nfs_readdata_alloc(rpages);
+                data = nfs_readdata_alloc(pgbase + bytes);
                 if (unlikely(!data))
                         break;
 
-                bytes = rsize;
-                if (count < rsize)
-                        bytes = count;
-
-                data->npages = nfs_direct_count_pages(user_addr, bytes);
                 down_read(&current->mm->mmap_sem);
                 result = get_user_pages(current, current->mm, user_addr,
                                         data->npages, 1, 0, data->pagevec, NULL);
@@ -344,8 +321,10 @@ static ssize_t nfs_direct_read_schedule(struct nfs_direct_req *dreq, unsigned lo
                 started += bytes;
                 user_addr += bytes;
                 pos += bytes;
+                /* FIXME: Remove this unnecessary math from final patch */
                 pgbase += bytes;
                 pgbase &= ~PAGE_MASK;
+                BUG_ON(pgbase != (user_addr & ~PAGE_MASK));
 
                 count -= bytes;
         } while (count != 0);
@@ -524,7 +503,7 @@ static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode
 
 static void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
 {
-        dreq->commit_data = nfs_commit_alloc(0);
+        dreq->commit_data = nfs_commit_alloc();
         if (dreq->commit_data != NULL)
                 dreq->commit_data->req = (struct nfs_page *) dreq;
 }
@@ -605,28 +584,24 @@ static ssize_t nfs_direct_write_schedule(struct nfs_direct_req *dreq, unsigned l
         struct nfs_open_context *ctx = dreq->ctx;
         struct inode *inode = ctx->dentry->d_inode;
         size_t wsize = NFS_SERVER(inode)->wsize;
-        unsigned int wpages = nfs_max_pages(wsize);
         unsigned int pgbase;
         int result;
         ssize_t started = 0;
 
         get_dreq(dreq);
 
-        pgbase = user_addr & ~PAGE_MASK;
         do {
                 struct nfs_write_data *data;
                 size_t bytes;
 
+                pgbase = user_addr & ~PAGE_MASK;
+                bytes = min(wsize,count);
+
                 result = -ENOMEM;
-                data = nfs_writedata_alloc(wpages);
+                data = nfs_writedata_alloc(pgbase + bytes);
                 if (unlikely(!data))
                         break;
 
-                bytes = wsize;
-                if (count < wsize)
-                        bytes = count;
-
-                data->npages = nfs_direct_count_pages(user_addr, bytes);
                 down_read(&current->mm->mmap_sem);
                 result = get_user_pages(current, current->mm, user_addr,
                                         data->npages, 0, 0, data->pagevec, NULL);
@@ -676,8 +651,11 @@ static ssize_t nfs_direct_write_schedule(struct nfs_direct_req *dreq, unsigned l
                 started += bytes;
                 user_addr += bytes;
                 pos += bytes;
+
+                /* FIXME: Remove this useless math from the final patch */
                 pgbase += bytes;
                 pgbase &= ~PAGE_MASK;
+                BUG_ON(pgbase != (user_addr & ~PAGE_MASK));
 
                 count -= bytes;
         } while (count != 0);
diff --git a/fs/nfs/read.c b/fs/nfs/read.c
index da9cf11c326f..7a9ee00e0c61 100644
--- a/fs/nfs/read.c
+++ b/fs/nfs/read.c
@@ -43,13 +43,15 @@ static mempool_t *nfs_rdata_mempool;
 
 #define MIN_POOL_READ   (32)
 
-struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount)
+struct nfs_read_data *nfs_readdata_alloc(size_t len)
 {
+        unsigned int pagecount = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
         struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, SLAB_NOFS);
 
         if (p) {
                 memset(p, 0, sizeof(*p));
                 INIT_LIST_HEAD(&p->pages);
+                p->npages = pagecount;
                 if (pagecount <= ARRAY_SIZE(p->page_array))
                         p->pagevec = p->page_array;
                 else {
@@ -140,7 +142,7 @@ static int nfs_readpage_sync(struct nfs_open_context *ctx, struct inode *inode,
         int             result;
         struct nfs_read_data *rdata;
 
-        rdata = nfs_readdata_alloc(1);
+        rdata = nfs_readdata_alloc(count);
         if (!rdata)
                 return -ENOMEM;
 
@@ -336,25 +338,25 @@ static int nfs_pagein_multi(struct list_head *head, struct inode *inode)
         struct nfs_page *req = nfs_list_entry(head->next);
         struct page *page = req->wb_page;
         struct nfs_read_data *data;
-        unsigned int rsize = NFS_SERVER(inode)->rsize;
-        unsigned int nbytes, offset;
+        size_t rsize = NFS_SERVER(inode)->rsize, nbytes;
+        unsigned int offset;
         int requests = 0;
         LIST_HEAD(list);
 
         nfs_list_remove_request(req);
 
         nbytes = req->wb_bytes;
-        for(;;) {
-                data = nfs_readdata_alloc(1);
+        do {
+                size_t len = min(nbytes,rsize);
+
+                data = nfs_readdata_alloc(len);
                 if (!data)
                         goto out_bad;
                 INIT_LIST_HEAD(&data->pages);
                 list_add(&data->pages, &list);
                 requests++;
-                if (nbytes <= rsize)
-                        break;
-                nbytes -= rsize;
-        }
+                nbytes -= len;
+        } while(nbytes != 0);
         atomic_set(&req->wb_complete, requests);
 
         ClearPageError(page);
@@ -402,7 +404,7 @@ static int nfs_pagein_one(struct list_head *head, struct inode *inode)
         if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE)
                 return nfs_pagein_multi(head, inode);
 
-        data = nfs_readdata_alloc(NFS_SERVER(inode)->rpages);
+        data = nfs_readdata_alloc(NFS_SERVER(inode)->rsize);
         if (!data)
                 goto out_bad;
 
diff --git a/fs/nfs/write.c b/fs/nfs/write.c
index 50774991f8d5..8ab3cf10d792 100644
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -90,22 +90,13 @@ static mempool_t *nfs_commit_mempool;
 
 static DECLARE_WAIT_QUEUE_HEAD(nfs_write_congestion);
 
-struct nfs_write_data *nfs_commit_alloc(unsigned int pagecount)
+struct nfs_write_data *nfs_commit_alloc(void)
 {
         struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, SLAB_NOFS);
 
         if (p) {
                 memset(p, 0, sizeof(*p));
                 INIT_LIST_HEAD(&p->pages);
-                if (pagecount <= ARRAY_SIZE(p->page_array))
-                        p->pagevec = p->page_array;
-                else {
-                        p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
-                        if (!p->pagevec) {
-                                mempool_free(p, nfs_commit_mempool);
-                                p = NULL;
-                        }
-                }
         }
         return p;
 }
@@ -117,13 +108,15 @@ void nfs_commit_free(struct nfs_write_data *p)
         mempool_free(p, nfs_commit_mempool);
 }
 
-struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount)
+struct nfs_write_data *nfs_writedata_alloc(size_t len)
 {
+        unsigned int pagecount = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
         struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, SLAB_NOFS);
 
         if (p) {
                 memset(p, 0, sizeof(*p));
                 INIT_LIST_HEAD(&p->pages);
+                p->npages = pagecount;
                 if (pagecount <= ARRAY_SIZE(p->page_array))
                         p->pagevec = p->page_array;
                 else {
@@ -208,7 +201,7 @@ static int nfs_writepage_sync(struct nfs_open_context *ctx, struct inode *inode,
         int             result, written = 0;
         struct nfs_write_data *wdata;
 
-        wdata = nfs_writedata_alloc(1);
+        wdata = nfs_writedata_alloc(wsize);
         if (!wdata)
                 return -ENOMEM;
 
@@ -999,24 +992,24 @@ static int nfs_flush_multi(struct inode *inode, struct list_head *head, int how)
         struct nfs_page *req = nfs_list_entry(head->next);
         struct page *page = req->wb_page;
         struct nfs_write_data *data;
-        unsigned int wsize = NFS_SERVER(inode)->wsize;
-        unsigned int nbytes, offset;
+        size_t wsize = NFS_SERVER(inode)->wsize, nbytes;
+        unsigned int offset;
         int requests = 0;
         LIST_HEAD(list);
 
         nfs_list_remove_request(req);
 
         nbytes = req->wb_bytes;
-        for (;;) {
-                data = nfs_writedata_alloc(1);
+        do {
+                size_t len = min(nbytes, wsize);
+
+                data = nfs_writedata_alloc(len);
                 if (!data)
                         goto out_bad;
                 list_add(&data->pages, &list);
                 requests++;
-                if (nbytes <= wsize)
-                        break;
-                nbytes -= wsize;
-        }
+                nbytes -= len;
+        } while (nbytes != 0);
         atomic_set(&req->wb_complete, requests);
 
         ClearPageError(page);
@@ -1070,7 +1063,7 @@ static int nfs_flush_one(struct inode *inode, struct list_head *head, int how)
         struct nfs_write_data   *data;
         unsigned int            count;
 
-        data = nfs_writedata_alloc(NFS_SERVER(inode)->wpages);
+        data = nfs_writedata_alloc(NFS_SERVER(inode)->wsize);
         if (!data)
                 goto out_bad;
 
@@ -1378,7 +1371,7 @@ nfs_commit_list(struct inode *inode, struct list_head *head, int how)
         struct nfs_write_data   *data;
         struct nfs_page         *req;
 
-        data = nfs_commit_alloc(NFS_SERVER(inode)->wpages);
+        data = nfs_commit_alloc();
 
         if (!data)
                 goto out_bad;