/*
* OBDFS Super operations
*
+ * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
* Copryright (C) 1999 Stelias Computing Inc,
* (author Peter J. Braam <braam@stelias.com>)
* Copryright (C) 1999 Seagate Technology Inc.
- */
+*/
#include <linux/config.h>
#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>
+
+
+/* SYNCHRONOUS I/O for an inode */
+static int obdfs_brw(int rw, struct inode *inode, struct page *page, int create)
+{
+ obd_count num_obdo = 1;
+ obd_count bufs_per_obdo = 1;
+ struct obdo *oa;
+ char *buf = (char *)page_address(page);
+ obd_size count = PAGE_SIZE;
+ obd_off offset = ((obd_off)page->index) << PAGE_SHIFT;
+ obd_flag flags = create ? OBD_BRW_CREATE : 0;
+ int err;
+
+ ENTRY;
+ oa = obdo_fromid(IID(inode), inode->i_ino, OBD_MD_FLNOTOBD);
+ if ( IS_ERR(oa) ) {
+ EXIT;
+ return PTR_ERR(oa);
+ }
+ obdfs_from_inode(oa, inode);
-int console_loglevel;
+ err = IOPS(inode, brw)(rw, IID(inode), num_obdo, &oa, &bufs_per_obdo,
+ &buf, &count, &offset, &flags);
-/* VFS super_block ops */
+ if ( !err )
+ obdfs_to_inode(inode, oa); /* copy o_blocks to i_blocks */
+
+ obdo_free(oa);
+
+ EXIT;
+ return err;
+} /* obdfs_brw */
/* 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;
+ struct inode *inode = dentry->d_inode;
int rc;
- ENTRY;
-
- /* XXX flush stuff */
- sbi = sb->u.generic_sbp;
+ ENTRY;
PDEBUG(page, "READ");
- rc = sbi->osi_ops->o_brw(READ, sbi->osi_conn_info.conn_id,
- file->f_dentry->d_inode, page, 0);
- if (rc == PAGE_SIZE ) {
+ rc = obdfs_brw(READ, inode, page, 0);
+ if ( !rc ) {
SetPageUptodate(page);
UnlockPage(page);
}
PDEBUG(page, "READ");
- if ( rc == PAGE_SIZE )
- rc = 0;
+ EXIT;
return rc;
+} /* obdfs_readpage */
+
+static kmem_cache_t *obdfs_pgrq_cachep = NULL;
+
+/* XXX should probably have one of these per superblock */
+static int obdfs_cache_count = 0;
+int obdfs_init_pgrqcache(void)
+{
+ ENTRY;
+ if (obdfs_pgrq_cachep == NULL) {
+ CDEBUG(D_INODE, "allocating obdfs_pgrq_cache\n");
+ obdfs_pgrq_cachep = kmem_cache_create("obdfs_pgrq",
+ sizeof(struct obdfs_pgrq),
+ 0, SLAB_HWCACHE_ALIGN,
+ NULL, NULL);
+ if (obdfs_pgrq_cachep == NULL) {
+ EXIT;
+ return -ENOMEM;
+ } else {
+ CDEBUG(D_INODE, "allocated cache at %p\n",
+ obdfs_pgrq_cachep);
+ }
+ } else {
+ CDEBUG(D_INODE, "using existing cache at %p\n",
+ obdfs_pgrq_cachep);
+ }
+ EXIT;
+ return 0;
+} /* obdfs_init_wreqcache */
+
+inline void obdfs_pgrq_del(struct obdfs_pgrq *pgrq)
+{
+ obdfs_cache_count--;
+ CDEBUG(D_INODE, "deleting page %p from list [count %d]\n",
+ pgrq->rq_page, obdfs_cache_count);
+ list_del(&pgrq->rq_plist);
+ kmem_cache_free(obdfs_pgrq_cachep, pgrq);
}
+void obdfs_cleanup_pgrqcache(void)
+{
+ ENTRY;
+ if (obdfs_pgrq_cachep != NULL) {
+ CDEBUG(D_INODE, "destroying obdfs_pgrqcache at %p, count %d\n",
+ obdfs_pgrq_cachep, obdfs_cache_count);
+ if (kmem_cache_destroy(obdfs_pgrq_cachep))
+ printk(KERN_INFO "obd_cleanup_pgrqcache: unable to free all of cache\n");
+ } else
+ printk(KERN_INFO "obd_cleanup_pgrqcache: called with NULL cache pointer\n");
+
+ EXIT;
+} /* obdfs_cleanup_wreqcache */
+
+
/*
- * This does the "real" work of the write. The generic routine has
- * allocated the page, locked it, done all the page alignment stuff
- * calculations etc. Now we should just copy the data from user
- * space and write it back to the real medium..
- *
- * If the writer ends up delaying the write, the writer needs to
- * increment the page use counts until he is done with the page.
+ * 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.
+ * Called with the list lock held.
*/
-int obdfs_write_one_page(struct file *file, struct page *page, unsigned long offset, unsigned long bytes, const char * buf)
+static struct obdfs_pgrq *
+obdfs_find_in_page_list(struct inode *inode, struct page *page)
{
- long status;
- struct obdfs_sb_info *sbi = file->f_dentry->d_inode->i_sb->u.generic_sbp;
+ struct list_head *page_list = obdfs_iplist(inode);
+ struct list_head *tmp;
ENTRY;
- if ( !Page_Uptodate(page) ) {
- status = sbi->osi_ops->o_brw(READ,
- sbi->osi_conn_info.conn_id,
- file->f_dentry->d_inode,
- page, 1);
- if (status == PAGE_SIZE ) {
- SetPageUptodate(page);
- } else {
- return status;
+
+ CDEBUG(D_INODE, "looking for inode %ld page %p\n", inode->i_ino, page);
+ OIDEBUG(inode);
+
+ if (list_empty(page_list)) {
+ CDEBUG(D_INODE, "empty list\n");
+ EXIT;
+ return NULL;
+ }
+ tmp = page_list;
+ while ( (tmp = tmp->next) != page_list ) {
+ struct obdfs_pgrq *pgrq;
+
+ pgrq = list_entry(tmp, struct obdfs_pgrq, rq_plist);
+ if (pgrq->rq_page == page) {
+ CDEBUG(D_INODE, "found page %p in list\n", page);
+ EXIT;
+ return pgrq;
}
+ }
+
+ EXIT;
+ return NULL;
+} /* obdfs_find_in_page_list */
+
+
+/* called with the list lock held */
+static struct page* obdfs_find_page_index(struct inode *inode,
+ unsigned long index)
+{
+ struct list_head *page_list = obdfs_iplist(inode);
+ struct list_head *tmp;
+ struct page *page;
+
+ ENTRY;
+
+ CDEBUG(D_INODE, "looking for inode %ld pageindex %ld\n",
+ inode->i_ino, index);
+ OIDEBUG(inode);
+
+ if (list_empty(page_list)) {
+ EXIT;
+ return NULL;
}
- bytes -= copy_from_user((u8*)page_address(page) + offset, buf, bytes);
- status = -EFAULT;
+ tmp = page_list;
+ while ( (tmp = tmp->next) != page_list ) {
+ struct obdfs_pgrq *pgrq;
+
+ pgrq = list_entry(tmp, struct obdfs_pgrq, rq_plist);
+ page = pgrq->rq_page;
+ if (index == page->index) {
+ CDEBUG(D_INODE,
+ "INDEX SEARCH found page %p, index %ld\n",
+ page, index);
+ EXIT;
+ return page;
+ }
+ }
+
+ EXIT;
+ return NULL;
+} /* obdfs_find_page_index */
- if (bytes) {
- lock_kernel();
- status = obdfs_writepage(file, page);
- unlock_kernel();
+/* call and free pages from Linux page cache: called with io lock on inodes */
+int obdfs_do_vec_wr(struct inode **inodes, obd_count num_io,
+ obd_count num_obdos, struct obdo **obdos,
+ obd_count *oa_bufs, struct page **pages, char **bufs,
+ obd_size *counts, obd_off *offsets, obd_flag *flags)
+{
+ struct super_block *sb = inodes[0]->i_sb;
+ struct obdfs_sb_info *sbi = (struct obdfs_sb_info *)&sb->u.generic_sbp;
+ int err;
+
+ ENTRY;
+ CDEBUG(D_INODE, "writing %d page(s), %d obdo(s) in vector\n",
+ num_io, num_obdos);
+ err = OPS(sb, brw)(WRITE, &sbi->osi_conn, num_obdos, obdos, oa_bufs,
+ bufs, counts, offsets, flags);
+
+ /* release the pages from the page cache */
+ while ( num_io > 0 ) {
+ num_io--;
+ CDEBUG(D_INODE, "calling put_page for %p, index %ld\n",
+ pages[num_io], pages[num_io]->index);
+ put_page(pages[num_io]);
+ }
+
+ while ( num_obdos > 0) {
+ num_obdos--;
+ CDEBUG(D_INODE, "copy/free obdo %ld\n",
+ (long)obdos[num_obdos]->o_id);
+ obdfs_to_inode(inodes[num_obdos], obdos[num_obdos]);
+ obdo_free(obdos[num_obdos]);
}
EXIT;
- if ( status != PAGE_SIZE )
- return status;
- else
- return bytes;
+ return err;
}
+/*
+ * Add a page to the write request cache list for later writing
+ * ASYNCHRONOUS write method.
+ */
+static int obdfs_add_page_to_cache(struct inode *inode, struct page *page)
+{
+ int res = 0;
+
+ ENTRY;
+
+ /* If this page isn't already in the inode page list, add it */
+ obd_down(&obdfs_i2sbi(inode)->osi_list_mutex);
+ if ( !obdfs_find_in_page_list(inode, page) ) {
+ struct obdfs_pgrq *pgrq;
+ pgrq = kmem_cache_alloc(obdfs_pgrq_cachep, SLAB_KERNEL);
+ CDEBUG(D_INODE, "adding inode %ld page %p, pgrq: %p, cache count [%d]\n",
+ inode->i_ino, page, pgrq, obdfs_cache_count);
+ if (!pgrq) {
+ EXIT;
+ obd_up(&obdfs_i2sbi(inode)->osi_list_mutex);
+ return -ENOMEM;
+ }
+ memset(pgrq, 0, sizeof(*pgrq));
+
+ pgrq->rq_page = page;
+ get_page(pgrq->rq_page);
+ list_add(&pgrq->rq_plist, obdfs_iplist(inode));
+ obdfs_cache_count++;
+ }
+
+ /* If inode isn't already on the superblock inodes list, add it,
+ * and increase ref count on inode so it doesn't disappear on us.
+ */
+ if ( list_empty(obdfs_islist(inode)) ) {
+ iget(inode->i_sb, inode->i_ino);
+ CDEBUG(D_INODE, "adding inode %ld to superblock list %p\n",
+ inode->i_ino, obdfs_slist(inode));
+ list_add(obdfs_islist(inode), obdfs_slist(inode));
+ }
+
+ /* XXX For testing purposes, we write out the page here.
+ * In the future, a flush daemon will write out the page.
+ res = obdfs_flush_reqs(obdfs_slist(inode), 0);
+ obdfs_flush_dirty_pages(1);
+ */
+ obd_up(&obdfs_i2sbi(inode)->osi_list_mutex);
+
+ EXIT;
+ return res;
+} /* obdfs_add_page_to_cache */
+
+
+/* select between SYNC and ASYNC I/O methods */
+int obdfs_do_writepage(struct inode *inode, struct page *page, int sync)
+{
+ int err;
+ ENTRY;
+ /* PDEBUG(page, "WRITEPAGE"); */
+ if ( sync )
+ err = obdfs_brw(WRITE, inode, page, 1);
+ else {
+ err = obdfs_add_page_to_cache(inode, page);
+ CDEBUG(D_IOCTL, "DO_WR ino: %ld, page %p, err %d, uptodata %d\n", inode->i_ino, page, err, Page_Uptodate(page));
+ }
+
+ if ( !err )
+ SetPageUptodate(page);
+ /* PDEBUG(page,"WRITEPAGE"); */
+ EXIT;
+ return err;
+} /* obdfs_do_writepage */
/* returns the page unlocked, but with a reference */
-int obdfs_writepage(struct file *file, struct page *page)
+int obdfs_writepage(struct dentry *dentry, struct page *page)
{
- struct obdfs_sb_info *sbi = file->f_dentry->d_inode->i_sb->u.generic_sbp;
- int rc;
+ return obdfs_do_writepage(dentry->d_inode, page, 0);
+}
- ENTRY;
- PDEBUG(page, "WRITEPAGE");
- /* XXX flush stuff */
+/*
+ * This does the "real" work of the write. The generic routine has
+ * allocated the page, locked it, done all the page alignment stuff
+ * calculations etc. Now we should just copy the data from user
+ * space and write it back to the real medium..
+ *
+ * If the writer ends up delaying the write, the writer needs to
+ * increment the page use counts until he is done with the page.
+ *
+ * Return value is the number of bytes written.
+ */
+int obdfs_write_one_page(struct file *file, struct page *page,
+ unsigned long offset, unsigned long bytes,
+ const char * buf)
+{
+ struct inode *inode = file->f_dentry->d_inode;
+ int err;
- rc = sbi->osi_ops->o_brw(WRITE, sbi->osi_conn_info.conn_id,
- file->f_dentry->d_inode, page, 1);
+ ENTRY;
+ if ( !Page_Uptodate(page) ) {
+ err = obdfs_brw(READ, inode, page, 1);
+ if ( !err )
+ SetPageUptodate(page);
+ else
+ return err;
+ }
- SetPageUptodate(page);
- PDEBUG(page,"WRITEPAGE");
- return rc;
-}
+ if (copy_from_user((u8*)page_address(page) + offset, buf, bytes))
+ return -EFAULT;
+ lock_kernel();
+ err = obdfs_writepage(file->f_dentry, page);
+ unlock_kernel();
-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);
-}
+ return (err < 0 ? err : bytes);
+} /* obdfs_write_one_page */
/*
- 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)
+ * 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 *page_cache;
+ int index;
struct page ** hash;
struct page * page;
- struct obdfs_sb_info *sbi;
- struct super_block *sb = inode->i_sb;
- int rc;
+ int err;
- ENTRY;
+ ENTRY;
offset = offset & PAGE_CACHE_MASK;
- sbi = sb->u.generic_sbp;
- CDEBUG(D_INODE, "\n");
-
+ CDEBUG(D_INODE, "ino: %ld, offset %ld, create %d, locked %d\n",
+ inode->i_ino, offset, create, locked);
+ index = offset >> PAGE_CACHE_SHIFT;
+
+
page = NULL;
page_cache = page_cache_alloc();
- if ( ! page_cache )
+ if ( ! page_cache ) {
+ EXIT;
return NULL;
+ }
CDEBUG(D_INODE, "page_cache %p\n", page_cache);
- hash = page_hash(&inode->i_data, offset);
- repeat:
- CDEBUG(D_INODE, "Finding page\n");
- IDEBUG(inode);
+ hash = page_hash(&inode->i_data, index);
+ page = grab_cache_page(&inode->i_data, index);
- page = __find_lock_page(&inode->i_data, offset, hash);
- if ( page ) {
- CDEBUG(D_INODE, "Page found freeing\n");
- page_cache_free(page_cache);
- } else {
- page = page_cache;
- if ( page->buffers ) {
- PDEBUG(page, "GETPAGE: buffers bug\n");
- UnlockPage(page);
- return NULL;
- }
- if (add_to_page_cache_unique(page, &inode->i_data, offset, hash)) {
- page_cache_release(page);
- CDEBUG(D_INODE, "Someone raced: try again\n");
- goto repeat;
- }
+ /* Yuck, no page */
+ if (! page) {
+ printk("grab_cache_page says no dice ...\n");
+ EXIT;
+ return 0;
}
PDEBUG(page, "GETPAGE: got page - before reading\n");
return page;
}
- /* it's not: read it */
- if (! page) {
- printk("get_page_map says no dice ...\n");
- return 0;
- }
+ if ( obdfs_find_page_index(inode, index) ) {
+ CDEBUG(D_INODE, "OVERWRITE: found dirty page %p, index %ld\n",
+ page, page->index);
+ }
+ err = obdfs_brw(READ, inode, page, create);
- rc = sbi->osi_ops->o_brw(READ, sbi->osi_conn_info.conn_id,
- inode, page, create);
- if ( rc != PAGE_SIZE ) {
+ if ( err ) {
SetPageError(page);
UnlockPage(page);
+ EXIT;
return page;
}
PDEBUG(page,"GETPAGE - after reading");
EXIT;
return page;
-}
+} /* obdfs_getpage */
-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 */
-};
git://git.whamcloud.com / fs / lustre-release.git / blobdiff