Whamcloud - gitweb
Updating obdfs/ext2obd methods for new obdo parameters.
[fs/lustre-release.git] / lustre / obdfs / rw.c
index ff06852..aa18c72 100644 (file)
 #include <linux/pagemap.h>
 #include <linux/smp_lock.h>
 
-#include <../obd/linux/obd_support.h>
-#include <../obd/linux/obd_sim.h>
-#include <obdfs.h>
+#include <linux/obd_support.h>
+#include <linux/obd_ext2.h>
+#include <linux/obdfs.h>
+
+int console_loglevel;
 
 /* VFS super_block ops */
 
+#if 0
+int obdfs_brw(struct inode *dir, int rw, struct page *page, int create)
+{
+       return IOPS(dir, brw)(rw, IID(dir), dir, page, create);
+}
+#endif
+
 /* returns the page unlocked, but with a reference */
-int obdfs_readpage(struct file *file, struct page *page)
+int obdfs_readpage(struct dentry *dentry, struct page *page)
 {
-        struct obdfs_sb_info *sbi;
-       struct super_block *sb = file->f_dentry->d_inode->i_sb;
-       int rc;
+       struct inode *inode = dentry->d_inode;
+       struct obdo *oa;
+       int err;
 
         ENTRY;
-
-       /* XXX flush stuff */
-       sbi = sb->u.generic_sbp;
        PDEBUG(page, "READ");
-       rc =  sbi->osi_ops->o_brw(READ, sbi->osi_conn_info.conn_id, 
-                     file->f_dentry->d_inode->i_ino, page, 0);
-       if (rc == PAGE_SIZE ) {
+       oa = obdo_alloc();
+       if (!oa) {
+               printk("obdfs_readpage: obdo_alloc failure\n");
+               EXIT;
+               return -ENOMEM;
+       }
+
+       oa->o_id = inode->i_ino;
+       err = IOPS(inode, brw)(READ, IID(inode), oa, (char *)page_address(page),
+                              PAGE_SIZE, (page->index) << PAGE_SHIFT, 0);
+       obdo_to_inode(inode, oa); /* copy o_blocks to i_blocks */
+       obdo_free(oa);
+
+       if (err == PAGE_SIZE ) {
                SetPageUptodate(page);
                UnlockPage(page);
        } 
        PDEBUG(page, "READ");
-       if ( rc == PAGE_SIZE ) 
-               rc = 0;
-       return rc;
+       if ( err == PAGE_SIZE ) 
+               err = 0;
+       EXIT;
+       return err;
+}
+
+static kmem_cache_t *obdfs_wreq_cachep;
+
+int obdfs_init_wreqcache(void)
+{
+       /* XXX need to free this somewhere? */
+       ENTRY;
+       obdfs_wreq_cachep = kmem_cache_create("obdfs_wreq",
+                                             sizeof(struct obdfs_wreq),
+                                             0, SLAB_HWCACHE_ALIGN,
+                                             NULL, NULL);
+       if (obdfs_wreq_cachep == NULL) {
+               EXIT;
+               return -ENOMEM;
+       }
+       EXIT;
+       return 0;
+}
+
+/*
+ * Find a specific page in the page cache.  If it is found, we return
+ * the write request struct associated with it, if not found return NULL.
+ */
+static struct obdfs_wreq *
+obdfs_find_in_page_cache(struct inode *inode, struct page *page)
+{
+       struct list_head *list_head = &OBD_LIST(inode);
+       struct obdfs_wreq *head, *wreq;
+
+       ENTRY;
+       CDEBUG(D_INODE, "looking for inode %ld page %p\n", inode->i_ino, page);
+       if (list_empty(list_head)) {
+               CDEBUG(D_INODE, "empty list\n");
+               EXIT;
+               return NULL;
+       }
+       wreq = head = WREQ(list_head->next);
+       do {
+               CDEBUG(D_INODE, "checking page %p\n", wreq->wb_page);
+               if (wreq->wb_page == page) {
+                       CDEBUG(D_INODE, "found page %p in list\n", page);
+                       EXIT;
+                       return wreq;
+               }
+       } while ((wreq = WB_NEXT(wreq)) != head);
+
+       EXIT;
+       return NULL;
+}
+
+
+/*
+ * Remove a writeback request from a list
+ */
+static inline int
+obdfs_remove_from_page_cache(struct obdfs_wreq *wreq)
+{
+       struct inode *inode = wreq->wb_inode;
+       struct page *page = wreq->wb_page;
+       struct obdo *oa;
+       int err;
+
+       ENTRY;
+       CDEBUG(D_INODE, "removing inode %ld page %p, wreq: %p\n",
+              inode->i_ino, page, wreq);
+       oa = obdo_alloc();
+       if (!oa) {
+               printk("obdfs_remove_from_page_cache: obdo_alloc failure\n");
+               EXIT;
+               return -ENOMEM;
+       }
+       oa->o_id = inode->i_ino;
+       err = IOPS(inode, brw)(WRITE, IID(inode), oa,(char *)page_address(page),
+                              PAGE_SIZE, (page->index) << PAGE_SHIFT, 1);
+       obdo_to_inode(inode, oa); /* copy o_blocks to i_blocks */
+       /* XXX probably should handle error here somehow.  I think that
+        *     ext2 also does the same thing - discard write even if error?
+        */
+       put_page(page);
+        list_del(&wreq->wb_list);
+       kmem_cache_free(obdfs_wreq_cachep, wreq);
+
+       EXIT;
+       return err;
+}
+
+/*
+ * Add a page to the write request cache list for later writing
+ */
+static int
+obdfs_add_to_page_cache(struct inode *inode, struct page *page)
+{
+       struct obdfs_wreq *wreq;
+
+       ENTRY;
+       wreq = kmem_cache_alloc(obdfs_wreq_cachep, SLAB_KERNEL);
+       CDEBUG(D_INODE, "adding inode %ld page %p, wreq: %p\n",
+              inode->i_ino, page, wreq);
+       if (!wreq) {
+               EXIT;
+               return -ENOMEM;
+       }
+       memset(wreq, 0, sizeof(*wreq)); 
+
+       wreq->wb_page = page;
+       wreq->wb_inode = inode;
+
+       get_page(wreq->wb_page);
+       list_add(&wreq->wb_list, &OBD_LIST(inode));
 
+       /* For testing purposes, we write out the page here.
+        * In the future, a flush daemon will write out the page.
+        */
+       wreq = obdfs_find_in_page_cache(inode, page);
+       if (!wreq) {
+               CDEBUG(D_INODE, "XXXX Can't find page after adding it!!!\n");
+               return -EINVAL;
+       } else
+               return obdfs_remove_from_page_cache(wreq);
+
+       return 0;
+}
+
+
+/* returns the page unlocked, but with a reference */
+int obdfs_writepage(struct dentry *dentry, struct page *page)
+{
+        struct inode *inode = dentry->d_inode;
+       int err;
+
+        ENTRY;
+       PDEBUG(page, "WRITEPAGE");
+       /* XXX flush stuff */
+       err = obdfs_add_to_page_cache(inode, page);
+
+       if (!err)
+               SetPageUptodate(page);
+       PDEBUG(page,"WRITEPAGE");
+       return err;
 }
 
 /*
@@ -68,16 +225,28 @@ int obdfs_readpage(struct file *file, struct page *page)
  * If the writer ends up delaying the write, the writer needs to
  * increment the page use counts until he is done with the page.
  */
-int obdfs_write_one_page(struct file *file, struct page *page, unsigned long offset, unsigned long bytes, const char * buf)
+int obdfs_write_one_page(struct file *file, struct page *page,
+                        unsigned long offset, unsigned long bytes,
+                        const char * buf)
 {
        long status;
-        struct obdfs_sb_info *sbi = file->f_dentry->d_inode->i_sb->u.generic_sbp;
+        struct inode *inode = file->f_dentry->d_inode;
 
+       ENTRY;
        if ( !Page_Uptodate(page) ) {
-               status =  sbi->osi_ops->o_brw(READ, 
-                                             sbi->osi_conn_info.conn_id, 
-                                             file->f_dentry->d_inode->i_ino, 
-                                             page, 1);
+               struct obdo *oa;
+               oa = obdo_alloc();
+               if (!oa) {
+                       printk("obdfs_write_one_page: obdo_alloc failure\n");
+                       EXIT;
+                       return -ENOMEM;
+               }
+               oa->o_id = inode->i_ino;
+               status = IOPS(inode, brw)(READ, IID(inode), oa,
+                                         (char *)page_address(page), PAGE_SIZE,                                          (page->index) << PAGE_SHIFT, 1);
+               obdo_to_inode(inode, oa); /* copy o_blocks to i_blocks */
+               obdo_free(oa);
+
                if (status == PAGE_SIZE ) {
                        SetPageUptodate(page);
                } else { 
@@ -89,9 +258,10 @@ int obdfs_write_one_page(struct file *file, struct page *page, unsigned long off
 
        if (bytes) {
                lock_kernel();
-               status = obdfs_writepage(file, page);
+               status = obdfs_writepage(file->f_dentry, page);
                unlock_kernel();
        }
+       EXIT;
        if ( status != PAGE_SIZE ) 
                return status;
        else
@@ -101,119 +271,86 @@ int obdfs_write_one_page(struct file *file, struct page *page, unsigned long off
 
 
 
-/* returns the page unlocked, but with a reference */
-int obdfs_writepage(struct file *file, struct page *page)
-{
-        struct obdfs_sb_info *sbi = file->f_dentry->d_inode->i_sb->u.generic_sbp;
-       int rc;
-
-        ENTRY;
-       PDEBUG(page, "WRITEPAGE");
-       /* XXX flush stuff */
 
-       rc = sbi->osi_ops->o_brw(WRITE, sbi->osi_conn_info.conn_id, 
-                     file->f_dentry->d_inode->i_ino, page, 1);
-       SetPageUptodate(page);
-       PDEBUG(page,"WRITEPAGE");
-       return rc;
+void report_inode(struct page * page) {
+       struct inode *inode = (struct inode *)0;
+       int offset = (int)&inode->i_data;
+       inode = (struct inode *)( (char *)page->mapping - offset);
+       if ( inode->i_sb->s_magic == 0x4711 )
+               printk("----> ino %ld , dev %d\n", inode->i_ino, inode->i_dev);
 }
 
-
 /* 
-   page is returned unlocked, with the up to date flag set, 
-   and held, i.e. caller must do a page_put
+   return an up to date page:
+    - if locked is true then is returned locked
+    - if create is true the corresponding disk blocks are created 
+    - page is held, i.e. caller must release the page
+
+   modeled on NFS code.
 */
-struct page *obdfs_getpage(struct inode *inode, unsigned long offset, int create, int locked)
+struct page *obdfs_getpage(struct inode *inode, unsigned long offset,
+                          int create, int locked)
 {
-       unsigned long new_page;
+       struct page *page_cache;
        struct page ** hash;
-       struct page * page; 
-       struct obdfs_sb_info *sbi;
-       struct super_block *sb = inode->i_sb;
+       struct page * page;
+       struct obdo *oa;
+       int err;
 
         ENTRY;
 
-       sbi = sb->u.generic_sbp;
+       offset = offset & PAGE_CACHE_MASK;
+       CDEBUG(D_INODE, "\n");
        
-       page = find_lock_page(inode, offset); 
-       if (page && Page_Uptodate(page)) { 
-               PDEBUG(page,"GETPAGE");
+       page = NULL;
+       page_cache = page_cache_alloc();
+       if ( ! page_cache ) 
+               return NULL;
+       CDEBUG(D_INODE, "page_cache %p\n", page_cache);
+
+       hash = page_hash(&inode->i_data, offset);
+       page = grab_cache_page(&inode->i_data, offset);
+
+       PDEBUG(page, "obdfs_getpage: got page - before reading\n");
+       /* now check if the data in the page is up to date */
+       if ( Page_Uptodate(page)) { 
                if (!locked)
                        UnlockPage(page);
+               EXIT;
                return page;
        } 
-               
-       if (page && !Page_Uptodate(page) ) {
-               CDEBUG(D_INODE, "Page found but not up to date\n");
+
+       /* it's not: read it */
+       if (! page) {
+               printk("obdfs_getpage: says no dice ...\n");
+               return NULL;
        }
 
-       /* page_cache_alloc returns address of page */
-       new_page = page_cache_alloc();
-       if (!new_page)
+       oa = obdo_alloc();
+       if (!oa) {
+               printk("obdfs_getpage: obdo_alloc failure\n");
+               EXIT;
                return NULL;
-       
-       /* corresponding struct page in the mmap */
-       hash = page_hash(inode, offset);
-       page = page_cache_entry(new_page);
-       PDEBUG(page, "GETPAGE");
-       if (!add_to_page_cache_unique(page, inode, offset, hash)) {
-               CDEBUG(D_INODE, "Page not found. Reading it.\n");
-               PDEBUG(page,"GETPAGE");
-               sbi->osi_ops->o_brw(READ, sbi->osi_conn_info.conn_id, 
-                                   inode->i_ino, page, create);
-               if ( !locked )
-                       UnlockPage(page);
-               SetPageUptodate(page);
-               PDEBUG(page,"GETPAGE");
+       }
+       oa->o_id = inode->i_ino;
+       err = IOPS(inode, brw)(READ, IID(inode), oa, (char *)page_address(page),
+                              PAGE_SIZE, (page->index) << PAGE_SHIFT, create);
+
+       obdo_to_inode(inode, oa); /* XXX to copy o_blocks */
+       obdo_free(oa);
+
+       if ( err != PAGE_SIZE ) {
+               SetPageError(page);
+               UnlockPage(page);
                return page;
        }
-       /*
-        * We arrive here in the unlikely event that someone 
-        * raced with us and added our page to the cache first.
-        */
-       CDEBUG(D_INODE, "Page not found. Someone raced us.\n");
-       PDEBUG(page,"GETPAGE");
+
+       if ( !locked )
+               UnlockPage(page);
+       SetPageUptodate(page);
+       PDEBUG(page,"GETPAGE - after reading");
+       EXIT;
        return page;
 }
 
 
-
-struct file_operations obdfs_file_ops = {
-       NULL,                   /* lseek - default */
-       generic_file_read,      /* read */
-       obdfs_file_write,     /* write - bad */
-        obdfs_readdir,         /* readdir */
-       NULL,                   /* poll - default */
-       NULL,                   /* ioctl */
-       NULL,                   /* mmap */
-       NULL,                   /* no special open code */
-       NULL,                   /* flush */
-       NULL,                   /* no special release code */
-       NULL,                   /* fsync */
-       NULL,                   /* fasync */
-       NULL,                   /* check_media_change */
-       NULL                    /* revalidate */
-};
-
-struct inode_operations obdfs_inode_ops = {
-       &obdfs_file_ops,        /* default directory file-ops */
-       obdfs_create,   /* create */
-       obdfs_lookup,   /* lookup */
-       obdfs_link,     /* link */
-       obdfs_unlink,   /* unlink */
-       obdfs_symlink,  /* symlink */
-       obdfs_mkdir,    /* mkdir */
-       obdfs_rmdir,    /* rmdir */
-       obdfs_mknod,    /* mknod */
-       obdfs_rename,   /* rename */
-       NULL,           /* readlink */
-       NULL,           /* follow_link */
-       NULL,           /* get_block */
-       obdfs_readpage, /* readpage */
-       obdfs_writepage, /* writepage */
-       NULL,           /* flushpage */
-       NULL,           /* truncate */
-       NULL,           /* permission */
-       NULL,           /* smap */
-       NULL            /* revalidate */
-};