cifs: convert cifs_readpages to use async reads
authorJeff Layton <jlayton@redhat.com>
Wed, 19 Oct 2011 19:30:16 +0000 (15:30 -0400)
committerJeff Layton <jlayton@redhat.com>
Wed, 19 Oct 2011 19:30:16 +0000 (15:30 -0400)
Now that we have code in place to do asynchronous reads, convert
cifs_readpages to use it. The new cifs_readpages walks the page_list
that gets passed in, locks and adds the pages to the pagecache and
sets up cifs_readdata to handle the reads.

The rest is handled by the cifs_async_readv infrastructure.

Reviewed-and-Tested-by: Pavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: Jeff Layton <jlayton@redhat.com>
fs/cifs/file.c

index 852d1f39adae579816e794378493df40b05bc407..8f6917816fec4faf796353709a2c77d0144a04bb 100644 (file)
@@ -32,6 +32,7 @@
 #include <linux/delay.h>
 #include <linux/mount.h>
 #include <linux/slab.h>
+#include <linux/swap.h>
 #include <asm/div64.h>
 #include "cifsfs.h"
 #include "cifspdu.h"
@@ -2000,82 +2001,24 @@ int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
        return rc;
 }
 
-
-static void cifs_copy_cache_pages(struct address_space *mapping,
-       struct list_head *pages, int bytes_read, char *data)
-{
-       struct page *page;
-       char *target;
-
-       while (bytes_read > 0) {
-               if (list_empty(pages))
-                       break;
-
-               page = list_entry(pages->prev, struct page, lru);
-               list_del(&page->lru);
-
-               if (add_to_page_cache_lru(page, mapping, page->index,
-                                     GFP_KERNEL)) {
-                       page_cache_release(page);
-                       cFYI(1, "Add page cache failed");
-                       data += PAGE_CACHE_SIZE;
-                       bytes_read -= PAGE_CACHE_SIZE;
-                       continue;
-               }
-               page_cache_release(page);
-
-               target = kmap_atomic(page, KM_USER0);
-
-               if (PAGE_CACHE_SIZE > bytes_read) {
-                       memcpy(target, data, bytes_read);
-                       /* zero the tail end of this partial page */
-                       memset(target + bytes_read, 0,
-                              PAGE_CACHE_SIZE - bytes_read);
-                       bytes_read = 0;
-               } else {
-                       memcpy(target, data, PAGE_CACHE_SIZE);
-                       bytes_read -= PAGE_CACHE_SIZE;
-               }
-               kunmap_atomic(target, KM_USER0);
-
-               flush_dcache_page(page);
-               SetPageUptodate(page);
-               unlock_page(page);
-               data += PAGE_CACHE_SIZE;
-
-               /* add page to FS-Cache */
-               cifs_readpage_to_fscache(mapping->host, page);
-       }
-       return;
-}
-
 static int cifs_readpages(struct file *file, struct address_space *mapping,
        struct list_head *page_list, unsigned num_pages)
 {
-       int rc = -EACCES;
-       int xid;
-       loff_t offset;
-       struct page *page;
-       struct cifs_sb_info *cifs_sb;
-       struct cifs_tcon *pTcon;
-       unsigned int bytes_read = 0;
-       unsigned int read_size, i;
-       char *smb_read_data = NULL;
-       struct smb_com_read_rsp *pSMBr;
-       struct cifsFileInfo *open_file;
-       struct cifs_io_parms io_parms;
-       int buf_type = CIFS_NO_BUFFER;
-       __u32 pid;
+       int rc;
+       struct list_head tmplist;
+       struct cifsFileInfo *open_file = file->private_data;
+       struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
+       unsigned int rsize = cifs_sb->rsize;
+       pid_t pid;
 
-       xid = GetXid();
-       if (file->private_data == NULL) {
-               rc = -EBADF;
-               FreeXid(xid);
-               return rc;
-       }
-       open_file = file->private_data;
-       cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
-       pTcon = tlink_tcon(open_file->tlink);
+       /*
+        * Give up immediately if rsize is too small to read an entire page.
+        * The VFS will fall back to readpage. We should never reach this
+        * point however since we set ra_pages to 0 when the rsize is smaller
+        * than a cache page.
+        */
+       if (unlikely(rsize < PAGE_CACHE_SIZE))
+               return 0;
 
        /*
         * Reads as many pages as possible from fscache. Returns -ENOBUFS
@@ -2084,125 +2027,127 @@ static int cifs_readpages(struct file *file, struct address_space *mapping,
        rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
                                         &num_pages);
        if (rc == 0)
-               goto read_complete;
+               return rc;
 
-       cFYI(DBG2, "rpages: num pages %d", num_pages);
        if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
                pid = open_file->pid;
        else
                pid = current->tgid;
 
-       for (i = 0; i < num_pages; ) {
-               unsigned contig_pages;
-               struct page *tmp_page;
-               unsigned long expected_index;
+       rc = 0;
+       INIT_LIST_HEAD(&tmplist);
 
-               if (list_empty(page_list))
-                       break;
+       cFYI(1, "%s: file=%p mapping=%p num_pages=%u", __func__, file,
+               mapping, num_pages);
+
+       /*
+        * Start with the page at end of list and move it to private
+        * list. Do the same with any following pages until we hit
+        * the rsize limit, hit an index discontinuity, or run out of
+        * pages. Issue the async read and then start the loop again
+        * until the list is empty.
+        *
+        * Note that list order is important. The page_list is in
+        * the order of declining indexes. When we put the pages in
+        * the rdata->pages, then we want them in increasing order.
+        */
+       while (!list_empty(page_list)) {
+               unsigned int bytes = PAGE_CACHE_SIZE;
+               unsigned int expected_index;
+               unsigned int nr_pages = 1;
+               loff_t offset;
+               struct page *page, *tpage;
+               struct cifs_readdata *rdata;
 
                page = list_entry(page_list->prev, struct page, lru);
+
+               /*
+                * Lock the page and put it in the cache. Since no one else
+                * should have access to this page, we're safe to simply set
+                * PG_locked without checking it first.
+                */
+               __set_page_locked(page);
+               rc = add_to_page_cache_locked(page, mapping,
+                                             page->index, GFP_KERNEL);
+
+               /* give up if we can't stick it in the cache */
+               if (rc) {
+                       __clear_page_locked(page);
+                       break;
+               }
+
+               /* move first page to the tmplist */
                offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
+               list_move_tail(&page->lru, &tmplist);
+
+               /* now try and add more pages onto the request */
+               expected_index = page->index + 1;
+               list_for_each_entry_safe_reverse(page, tpage, page_list, lru) {
+                       /* discontinuity ? */
+                       if (page->index != expected_index)
+                               break;
+
+                       /* would this page push the read over the rsize? */
+                       if (bytes + PAGE_CACHE_SIZE > rsize)
+                               break;
 
-               /* count adjacent pages that we will read into */
-               contig_pages = 0;
-               expected_index =
-                       list_entry(page_list->prev, struct page, lru)->index;
-               list_for_each_entry_reverse(tmp_page, page_list, lru) {
-                       if (tmp_page->index == expected_index) {
-                               contig_pages++;
-                               expected_index++;
-                       } else
+                       __set_page_locked(page);
+                       if (add_to_page_cache_locked(page, mapping,
+                                               page->index, GFP_KERNEL)) {
+                               __clear_page_locked(page);
                                break;
+                       }
+                       list_move_tail(&page->lru, &tmplist);
+                       bytes += PAGE_CACHE_SIZE;
+                       expected_index++;
+                       nr_pages++;
                }
-               if (contig_pages + i >  num_pages)
-                       contig_pages = num_pages - i;
-
-               /* for reads over a certain size could initiate async
-                  read ahead */
-
-               read_size = contig_pages * PAGE_CACHE_SIZE;
-               /* Read size needs to be in multiples of one page */
-               read_size = min_t(const unsigned int, read_size,
-                                 cifs_sb->rsize & PAGE_CACHE_MASK);
-               cFYI(DBG2, "rpages: read size 0x%x  contiguous pages %d",
-                               read_size, contig_pages);
-               rc = -EAGAIN;
-               while (rc == -EAGAIN) {
+
+               rdata = cifs_readdata_alloc(nr_pages);
+               if (!rdata) {
+                       /* best to give up if we're out of mem */
+                       list_for_each_entry_safe(page, tpage, &tmplist, lru) {
+                               list_del(&page->lru);
+                               lru_cache_add_file(page);
+                               unlock_page(page);
+                               page_cache_release(page);
+                       }
+                       rc = -ENOMEM;
+                       break;
+               }
+
+               spin_lock(&cifs_file_list_lock);
+               cifsFileInfo_get(open_file);
+               spin_unlock(&cifs_file_list_lock);
+               rdata->cfile = open_file;
+               rdata->mapping = mapping;
+               rdata->offset = offset;
+               rdata->bytes = bytes;
+               rdata->pid = pid;
+               list_splice_init(&tmplist, &rdata->pages);
+
+               do {
                        if (open_file->invalidHandle) {
                                rc = cifs_reopen_file(open_file, true);
                                if (rc != 0)
-                                       break;
+                                       continue;
                        }
-                       io_parms.netfid = open_file->netfid;
-                       io_parms.pid = pid;
-                       io_parms.tcon = pTcon;
-                       io_parms.offset = offset;
-                       io_parms.length = read_size;
-                       rc = CIFSSMBRead(xid, &io_parms, &bytes_read,
-                                        &smb_read_data, &buf_type);
-                       /* BB more RC checks ? */
-                       if (rc == -EAGAIN) {
-                               if (smb_read_data) {
-                                       if (buf_type == CIFS_SMALL_BUFFER)
-                                               cifs_small_buf_release(smb_read_data);
-                                       else if (buf_type == CIFS_LARGE_BUFFER)
-                                               cifs_buf_release(smb_read_data);
-                                       smb_read_data = NULL;
-                               }
-                       }
-               }
-               if ((rc < 0) || (smb_read_data == NULL)) {
-                       cFYI(1, "Read error in readpages: %d", rc);
-                       break;
-               } else if (bytes_read > 0) {
-                       task_io_account_read(bytes_read);
-                       pSMBr = (struct smb_com_read_rsp *)smb_read_data;
-                       cifs_copy_cache_pages(mapping, page_list, bytes_read,
-                               smb_read_data + 4 /* RFC1001 hdr */ +
-                               le16_to_cpu(pSMBr->DataOffset));
-
-                       i +=  bytes_read >> PAGE_CACHE_SHIFT;
-                       cifs_stats_bytes_read(pTcon, bytes_read);
-                       if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
-                               i++; /* account for partial page */
-
-                               /* server copy of file can have smaller size
-                                  than client */
-                               /* BB do we need to verify this common case ?
-                                  this case is ok - if we are at server EOF
-                                  we will hit it on next read */
+                       rc = cifs_async_readv(rdata);
+               } while (rc == -EAGAIN);
 
-                               /* break; */
+               if (rc != 0) {
+                       list_for_each_entry_safe(page, tpage, &rdata->pages,
+                                                lru) {
+                               list_del(&page->lru);
+                               lru_cache_add_file(page);
+                               unlock_page(page);
+                               page_cache_release(page);
                        }
-               } else {
-                       cFYI(1, "No bytes read (%d) at offset %lld . "
-                               "Cleaning remaining pages from readahead list",
-                               bytes_read, offset);
-                       /* BB turn off caching and do new lookup on
-                          file size at server? */
+                       cifs_readdata_free(rdata);
                        break;
                }
-               if (smb_read_data) {
-                       if (buf_type == CIFS_SMALL_BUFFER)
-                               cifs_small_buf_release(smb_read_data);
-                       else if (buf_type == CIFS_LARGE_BUFFER)
-                               cifs_buf_release(smb_read_data);
-                       smb_read_data = NULL;
-               }
-               bytes_read = 0;
-       }
-
-/* need to free smb_read_data buf before exit */
-       if (smb_read_data) {
-               if (buf_type == CIFS_SMALL_BUFFER)
-                       cifs_small_buf_release(smb_read_data);
-               else if (buf_type == CIFS_LARGE_BUFFER)
-                       cifs_buf_release(smb_read_data);
-               smb_read_data = NULL;
        }
 
-read_complete:
-       FreeXid(xid);
        return rc;
 }