11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/nfs/write.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Writing file data over NFS. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * We do it like this: When a (user) process wishes to write data to an 71da177e4SLinus Torvalds * NFS file, a write request is allocated that contains the RPC task data 81da177e4SLinus Torvalds * plus some info on the page to be written, and added to the inode's 91da177e4SLinus Torvalds * write chain. If the process writes past the end of the page, an async 101da177e4SLinus Torvalds * RPC call to write the page is scheduled immediately; otherwise, the call 111da177e4SLinus Torvalds * is delayed for a few seconds. 121da177e4SLinus Torvalds * 131da177e4SLinus Torvalds * Just like readahead, no async I/O is performed if wsize < PAGE_SIZE. 141da177e4SLinus Torvalds * 151da177e4SLinus Torvalds * Write requests are kept on the inode's writeback list. Each entry in 161da177e4SLinus Torvalds * that list references the page (portion) to be written. When the 171da177e4SLinus Torvalds * cache timeout has expired, the RPC task is woken up, and tries to 181da177e4SLinus Torvalds * lock the page. As soon as it manages to do so, the request is moved 191da177e4SLinus Torvalds * from the writeback list to the writelock list. 201da177e4SLinus Torvalds * 211da177e4SLinus Torvalds * Note: we must make sure never to confuse the inode passed in the 221da177e4SLinus Torvalds * write_page request with the one in page->inode. As far as I understand 231da177e4SLinus Torvalds * it, these are different when doing a swap-out. 241da177e4SLinus Torvalds * 251da177e4SLinus Torvalds * To understand everything that goes on here and in the NFS read code, 261da177e4SLinus Torvalds * one should be aware that a page is locked in exactly one of the following 271da177e4SLinus Torvalds * cases: 281da177e4SLinus Torvalds * 291da177e4SLinus Torvalds * - A write request is in progress. 301da177e4SLinus Torvalds * - A user process is in generic_file_write/nfs_update_page 311da177e4SLinus Torvalds * - A user process is in generic_file_read 321da177e4SLinus Torvalds * 331da177e4SLinus Torvalds * Also note that because of the way pages are invalidated in 341da177e4SLinus Torvalds * nfs_revalidate_inode, the following assertions hold: 351da177e4SLinus Torvalds * 361da177e4SLinus Torvalds * - If a page is dirty, there will be no read requests (a page will 371da177e4SLinus Torvalds * not be re-read unless invalidated by nfs_revalidate_inode). 381da177e4SLinus Torvalds * - If the page is not uptodate, there will be no pending write 391da177e4SLinus Torvalds * requests, and no process will be in nfs_update_page. 401da177e4SLinus Torvalds * 411da177e4SLinus Torvalds * FIXME: Interaction with the vmscan routines is not optimal yet. 421da177e4SLinus Torvalds * Either vmscan must be made nfs-savvy, or we need a different page 431da177e4SLinus Torvalds * reclaim concept that supports something like FS-independent 441da177e4SLinus Torvalds * buffer_heads with a b_ops-> field. 451da177e4SLinus Torvalds * 461da177e4SLinus Torvalds * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> 471da177e4SLinus Torvalds */ 481da177e4SLinus Torvalds 491da177e4SLinus Torvalds #include <linux/config.h> 501da177e4SLinus Torvalds #include <linux/types.h> 511da177e4SLinus Torvalds #include <linux/slab.h> 521da177e4SLinus Torvalds #include <linux/mm.h> 531da177e4SLinus Torvalds #include <linux/pagemap.h> 541da177e4SLinus Torvalds #include <linux/file.h> 551da177e4SLinus Torvalds #include <linux/mpage.h> 561da177e4SLinus Torvalds #include <linux/writeback.h> 571da177e4SLinus Torvalds 581da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h> 591da177e4SLinus Torvalds #include <linux/nfs_fs.h> 601da177e4SLinus Torvalds #include <linux/nfs_mount.h> 611da177e4SLinus Torvalds #include <linux/nfs_page.h> 621da177e4SLinus Torvalds #include <asm/uaccess.h> 631da177e4SLinus Torvalds #include <linux/smp_lock.h> 641da177e4SLinus Torvalds 651da177e4SLinus Torvalds #include "delegation.h" 661da177e4SLinus Torvalds 671da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 681da177e4SLinus Torvalds 691da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 701da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 711da177e4SLinus Torvalds 721da177e4SLinus Torvalds /* 731da177e4SLinus Torvalds * Local function declarations 741da177e4SLinus Torvalds */ 751da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context*, 761da177e4SLinus Torvalds struct inode *, 771da177e4SLinus Torvalds struct page *, 781da177e4SLinus Torvalds unsigned int, unsigned int); 791da177e4SLinus Torvalds static void nfs_writeback_done_partial(struct nfs_write_data *, int); 801da177e4SLinus Torvalds static void nfs_writeback_done_full(struct nfs_write_data *, int); 811da177e4SLinus Torvalds static int nfs_wait_on_write_congestion(struct address_space *, int); 821da177e4SLinus Torvalds static int nfs_wait_on_requests(struct inode *, unsigned long, unsigned int); 831da177e4SLinus Torvalds static int nfs_flush_inode(struct inode *inode, unsigned long idx_start, 841da177e4SLinus Torvalds unsigned int npages, int how); 851da177e4SLinus Torvalds 861da177e4SLinus Torvalds static kmem_cache_t *nfs_wdata_cachep; 871da177e4SLinus Torvalds mempool_t *nfs_wdata_mempool; 881da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 891da177e4SLinus Torvalds 901da177e4SLinus Torvalds static DECLARE_WAIT_QUEUE_HEAD(nfs_write_congestion); 911da177e4SLinus Torvalds 9240859d7eSChuck Lever static inline struct nfs_write_data *nfs_commit_alloc(unsigned int pagecount) 931da177e4SLinus Torvalds { 941da177e4SLinus Torvalds struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, SLAB_NOFS); 9540859d7eSChuck Lever 961da177e4SLinus Torvalds if (p) { 971da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 981da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 9940859d7eSChuck Lever if (pagecount < NFS_PAGEVEC_SIZE) 10040859d7eSChuck Lever p->pagevec = &p->page_array[0]; 10140859d7eSChuck Lever else { 10240859d7eSChuck Lever size_t size = ++pagecount * sizeof(struct page *); 10340859d7eSChuck Lever p->pagevec = kmalloc(size, GFP_NOFS); 10440859d7eSChuck Lever if (p->pagevec) { 10540859d7eSChuck Lever memset(p->pagevec, 0, size); 10640859d7eSChuck Lever } else { 10740859d7eSChuck Lever mempool_free(p, nfs_commit_mempool); 10840859d7eSChuck Lever p = NULL; 10940859d7eSChuck Lever } 11040859d7eSChuck Lever } 1111da177e4SLinus Torvalds } 1121da177e4SLinus Torvalds return p; 1131da177e4SLinus Torvalds } 1141da177e4SLinus Torvalds 1151da177e4SLinus Torvalds static inline void nfs_commit_free(struct nfs_write_data *p) 1161da177e4SLinus Torvalds { 11740859d7eSChuck Lever if (p && (p->pagevec != &p->page_array[0])) 11840859d7eSChuck Lever kfree(p->pagevec); 1191da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 1201da177e4SLinus Torvalds } 1211da177e4SLinus Torvalds 122963d8fe5STrond Myklebust void nfs_writedata_release(void *wdata) 1231da177e4SLinus Torvalds { 1241da177e4SLinus Torvalds nfs_writedata_free(wdata); 1251da177e4SLinus Torvalds } 1261da177e4SLinus Torvalds 1271da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1281da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1291da177e4SLinus Torvalds { 1301da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 1311da177e4SLinus Torvalds loff_t end, i_size = i_size_read(inode); 1321da177e4SLinus Torvalds unsigned long end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 1331da177e4SLinus Torvalds 1341da177e4SLinus Torvalds if (i_size > 0 && page->index < end_index) 1351da177e4SLinus Torvalds return; 1361da177e4SLinus Torvalds end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); 1371da177e4SLinus Torvalds if (i_size >= end) 1381da177e4SLinus Torvalds return; 1391da177e4SLinus Torvalds i_size_write(inode, end); 1401da177e4SLinus Torvalds } 1411da177e4SLinus Torvalds 1421da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 1431da177e4SLinus Torvalds * covers the full page. 1441da177e4SLinus Torvalds */ 1451da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 1461da177e4SLinus Torvalds { 1471da177e4SLinus Torvalds loff_t end_offs; 1481da177e4SLinus Torvalds 1491da177e4SLinus Torvalds if (PageUptodate(page)) 1501da177e4SLinus Torvalds return; 1511da177e4SLinus Torvalds if (base != 0) 1521da177e4SLinus Torvalds return; 1531da177e4SLinus Torvalds if (count == PAGE_CACHE_SIZE) { 1541da177e4SLinus Torvalds SetPageUptodate(page); 1551da177e4SLinus Torvalds return; 1561da177e4SLinus Torvalds } 1571da177e4SLinus Torvalds 1581da177e4SLinus Torvalds end_offs = i_size_read(page->mapping->host) - 1; 1591da177e4SLinus Torvalds if (end_offs < 0) 1601da177e4SLinus Torvalds return; 1611da177e4SLinus Torvalds /* Is this the last page? */ 1621da177e4SLinus Torvalds if (page->index != (unsigned long)(end_offs >> PAGE_CACHE_SHIFT)) 1631da177e4SLinus Torvalds return; 1641da177e4SLinus Torvalds /* This is the last page: set PG_uptodate if we cover the entire 1651da177e4SLinus Torvalds * extent of the data, then zero the rest of the page. 1661da177e4SLinus Torvalds */ 1671da177e4SLinus Torvalds if (count == (unsigned int)(end_offs & (PAGE_CACHE_SIZE - 1)) + 1) { 1681da177e4SLinus Torvalds memclear_highpage_flush(page, count, PAGE_CACHE_SIZE - count); 1691da177e4SLinus Torvalds SetPageUptodate(page); 1701da177e4SLinus Torvalds } 1711da177e4SLinus Torvalds } 1721da177e4SLinus Torvalds 1731da177e4SLinus Torvalds /* 1741da177e4SLinus Torvalds * Write a page synchronously. 1751da177e4SLinus Torvalds * Offset is the data offset within the page. 1761da177e4SLinus Torvalds */ 1771da177e4SLinus Torvalds static int nfs_writepage_sync(struct nfs_open_context *ctx, struct inode *inode, 1781da177e4SLinus Torvalds struct page *page, unsigned int offset, unsigned int count, 1791da177e4SLinus Torvalds int how) 1801da177e4SLinus Torvalds { 1811da177e4SLinus Torvalds unsigned int wsize = NFS_SERVER(inode)->wsize; 1821da177e4SLinus Torvalds int result, written = 0; 1831da177e4SLinus Torvalds struct nfs_write_data *wdata; 1841da177e4SLinus Torvalds 18540859d7eSChuck Lever wdata = nfs_writedata_alloc(1); 1861da177e4SLinus Torvalds if (!wdata) 1871da177e4SLinus Torvalds return -ENOMEM; 1881da177e4SLinus Torvalds 1891da177e4SLinus Torvalds wdata->flags = how; 1901da177e4SLinus Torvalds wdata->cred = ctx->cred; 1911da177e4SLinus Torvalds wdata->inode = inode; 1921da177e4SLinus Torvalds wdata->args.fh = NFS_FH(inode); 1931da177e4SLinus Torvalds wdata->args.context = ctx; 1941da177e4SLinus Torvalds wdata->args.pages = &page; 1951da177e4SLinus Torvalds wdata->args.stable = NFS_FILE_SYNC; 1961da177e4SLinus Torvalds wdata->args.pgbase = offset; 1971da177e4SLinus Torvalds wdata->args.count = wsize; 1981da177e4SLinus Torvalds wdata->res.fattr = &wdata->fattr; 1991da177e4SLinus Torvalds wdata->res.verf = &wdata->verf; 2001da177e4SLinus Torvalds 2011da177e4SLinus Torvalds dprintk("NFS: nfs_writepage_sync(%s/%Ld %d@%Ld)\n", 2021da177e4SLinus Torvalds inode->i_sb->s_id, 2031da177e4SLinus Torvalds (long long)NFS_FILEID(inode), 2041da177e4SLinus Torvalds count, (long long)(page_offset(page) + offset)); 2051da177e4SLinus Torvalds 206bb713d6dSTrond Myklebust set_page_writeback(page); 2071da177e4SLinus Torvalds nfs_begin_data_update(inode); 2081da177e4SLinus Torvalds do { 2091da177e4SLinus Torvalds if (count < wsize) 2101da177e4SLinus Torvalds wdata->args.count = count; 2111da177e4SLinus Torvalds wdata->args.offset = page_offset(page) + wdata->args.pgbase; 2121da177e4SLinus Torvalds 2131da177e4SLinus Torvalds result = NFS_PROTO(inode)->write(wdata); 2141da177e4SLinus Torvalds 2151da177e4SLinus Torvalds if (result < 0) { 2161da177e4SLinus Torvalds /* Must mark the page invalid after I/O error */ 2171da177e4SLinus Torvalds ClearPageUptodate(page); 2181da177e4SLinus Torvalds goto io_error; 2191da177e4SLinus Torvalds } 2201da177e4SLinus Torvalds if (result < wdata->args.count) 2211da177e4SLinus Torvalds printk(KERN_WARNING "NFS: short write, count=%u, result=%d\n", 2221da177e4SLinus Torvalds wdata->args.count, result); 2231da177e4SLinus Torvalds 2241da177e4SLinus Torvalds wdata->args.offset += result; 2251da177e4SLinus Torvalds wdata->args.pgbase += result; 2261da177e4SLinus Torvalds written += result; 2271da177e4SLinus Torvalds count -= result; 2281da177e4SLinus Torvalds } while (count); 2291da177e4SLinus Torvalds /* Update file length */ 2301da177e4SLinus Torvalds nfs_grow_file(page, offset, written); 2311da177e4SLinus Torvalds /* Set the PG_uptodate flag? */ 2321da177e4SLinus Torvalds nfs_mark_uptodate(page, offset, written); 2331da177e4SLinus Torvalds 2341da177e4SLinus Torvalds if (PageError(page)) 2351da177e4SLinus Torvalds ClearPageError(page); 2361da177e4SLinus Torvalds 2371da177e4SLinus Torvalds io_error: 238951a143bSTrond Myklebust nfs_end_data_update(inode); 239bb713d6dSTrond Myklebust end_page_writeback(page); 2401da177e4SLinus Torvalds nfs_writedata_free(wdata); 2411da177e4SLinus Torvalds return written ? written : result; 2421da177e4SLinus Torvalds } 2431da177e4SLinus Torvalds 2441da177e4SLinus Torvalds static int nfs_writepage_async(struct nfs_open_context *ctx, 2451da177e4SLinus Torvalds struct inode *inode, struct page *page, 2461da177e4SLinus Torvalds unsigned int offset, unsigned int count) 2471da177e4SLinus Torvalds { 2481da177e4SLinus Torvalds struct nfs_page *req; 2491da177e4SLinus Torvalds 2501da177e4SLinus Torvalds req = nfs_update_request(ctx, inode, page, offset, count); 251abd3e641STrond Myklebust if (IS_ERR(req)) 252abd3e641STrond Myklebust return PTR_ERR(req); 2531da177e4SLinus Torvalds /* Update file length */ 2541da177e4SLinus Torvalds nfs_grow_file(page, offset, count); 2551da177e4SLinus Torvalds /* Set the PG_uptodate flag? */ 2561da177e4SLinus Torvalds nfs_mark_uptodate(page, offset, count); 2571da177e4SLinus Torvalds nfs_unlock_request(req); 258abd3e641STrond Myklebust return 0; 2591da177e4SLinus Torvalds } 2601da177e4SLinus Torvalds 2611da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 2621da177e4SLinus Torvalds { 2631da177e4SLinus Torvalds if (wbc->for_reclaim) 2641da177e4SLinus Torvalds return FLUSH_HIGHPRI; 2651da177e4SLinus Torvalds if (wbc->for_kupdate) 2661da177e4SLinus Torvalds return FLUSH_LOWPRI; 2671da177e4SLinus Torvalds return 0; 2681da177e4SLinus Torvalds } 2691da177e4SLinus Torvalds 2701da177e4SLinus Torvalds /* 2711da177e4SLinus Torvalds * Write an mmapped page to the server. 2721da177e4SLinus Torvalds */ 2731da177e4SLinus Torvalds int nfs_writepage(struct page *page, struct writeback_control *wbc) 2741da177e4SLinus Torvalds { 2751da177e4SLinus Torvalds struct nfs_open_context *ctx; 2761da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 2771da177e4SLinus Torvalds unsigned long end_index; 2781da177e4SLinus Torvalds unsigned offset = PAGE_CACHE_SIZE; 2791da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 2801da177e4SLinus Torvalds int inode_referenced = 0; 2811da177e4SLinus Torvalds int priority = wb_priority(wbc); 2821da177e4SLinus Torvalds int err; 2831da177e4SLinus Torvalds 2841da177e4SLinus Torvalds /* 2851da177e4SLinus Torvalds * Note: We need to ensure that we have a reference to the inode 2861da177e4SLinus Torvalds * if we are to do asynchronous writes. If not, waiting 2871da177e4SLinus Torvalds * in nfs_wait_on_request() may deadlock with clear_inode(). 2881da177e4SLinus Torvalds * 2891da177e4SLinus Torvalds * If igrab() fails here, then it is in any case safe to 2901da177e4SLinus Torvalds * call nfs_wb_page(), since there will be no pending writes. 2911da177e4SLinus Torvalds */ 2921da177e4SLinus Torvalds if (igrab(inode) != 0) 2931da177e4SLinus Torvalds inode_referenced = 1; 2941da177e4SLinus Torvalds end_index = i_size >> PAGE_CACHE_SHIFT; 2951da177e4SLinus Torvalds 2961da177e4SLinus Torvalds /* Ensure we've flushed out any previous writes */ 2971da177e4SLinus Torvalds nfs_wb_page_priority(inode, page, priority); 2981da177e4SLinus Torvalds 2991da177e4SLinus Torvalds /* easy case */ 3001da177e4SLinus Torvalds if (page->index < end_index) 3011da177e4SLinus Torvalds goto do_it; 3021da177e4SLinus Torvalds /* things got complicated... */ 3031da177e4SLinus Torvalds offset = i_size & (PAGE_CACHE_SIZE-1); 3041da177e4SLinus Torvalds 3051da177e4SLinus Torvalds /* OK, are we completely out? */ 3061da177e4SLinus Torvalds err = 0; /* potential race with truncate - ignore */ 3071da177e4SLinus Torvalds if (page->index >= end_index+1 || !offset) 3081da177e4SLinus Torvalds goto out; 3091da177e4SLinus Torvalds do_it: 310d530838bSTrond Myklebust ctx = nfs_find_open_context(inode, NULL, FMODE_WRITE); 3111da177e4SLinus Torvalds if (ctx == NULL) { 3121da177e4SLinus Torvalds err = -EBADF; 3131da177e4SLinus Torvalds goto out; 3141da177e4SLinus Torvalds } 3151da177e4SLinus Torvalds lock_kernel(); 3161da177e4SLinus Torvalds if (!IS_SYNC(inode) && inode_referenced) { 3171da177e4SLinus Torvalds err = nfs_writepage_async(ctx, inode, page, 0, offset); 318abd3e641STrond Myklebust if (!wbc->for_writepages) 319abd3e641STrond Myklebust nfs_flush_inode(inode, 0, 0, wb_priority(wbc)); 3201da177e4SLinus Torvalds } else { 3211da177e4SLinus Torvalds err = nfs_writepage_sync(ctx, inode, page, 0, 3221da177e4SLinus Torvalds offset, priority); 3231da177e4SLinus Torvalds if (err >= 0) { 3241da177e4SLinus Torvalds if (err != offset) 3251da177e4SLinus Torvalds redirty_page_for_writepage(wbc, page); 3261da177e4SLinus Torvalds err = 0; 3271da177e4SLinus Torvalds } 3281da177e4SLinus Torvalds } 3291da177e4SLinus Torvalds unlock_kernel(); 3301da177e4SLinus Torvalds put_nfs_open_context(ctx); 3311da177e4SLinus Torvalds out: 3321da177e4SLinus Torvalds unlock_page(page); 3331da177e4SLinus Torvalds if (inode_referenced) 3341da177e4SLinus Torvalds iput(inode); 3351da177e4SLinus Torvalds return err; 3361da177e4SLinus Torvalds } 3371da177e4SLinus Torvalds 3381da177e4SLinus Torvalds /* 3391da177e4SLinus Torvalds * Note: causes nfs_update_request() to block on the assumption 3401da177e4SLinus Torvalds * that the writeback is generated due to memory pressure. 3411da177e4SLinus Torvalds */ 3421da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 3431da177e4SLinus Torvalds { 3441da177e4SLinus Torvalds struct backing_dev_info *bdi = mapping->backing_dev_info; 3451da177e4SLinus Torvalds struct inode *inode = mapping->host; 3461da177e4SLinus Torvalds int err; 3471da177e4SLinus Torvalds 3481da177e4SLinus Torvalds err = generic_writepages(mapping, wbc); 3491da177e4SLinus Torvalds if (err) 3501da177e4SLinus Torvalds return err; 3511da177e4SLinus Torvalds while (test_and_set_bit(BDI_write_congested, &bdi->state) != 0) { 3521da177e4SLinus Torvalds if (wbc->nonblocking) 3531da177e4SLinus Torvalds return 0; 3541da177e4SLinus Torvalds nfs_wait_on_write_congestion(mapping, 0); 3551da177e4SLinus Torvalds } 3561da177e4SLinus Torvalds err = nfs_flush_inode(inode, 0, 0, wb_priority(wbc)); 3571da177e4SLinus Torvalds if (err < 0) 3581da177e4SLinus Torvalds goto out; 3591da177e4SLinus Torvalds wbc->nr_to_write -= err; 3601da177e4SLinus Torvalds if (!wbc->nonblocking && wbc->sync_mode == WB_SYNC_ALL) { 3611da177e4SLinus Torvalds err = nfs_wait_on_requests(inode, 0, 0); 3621da177e4SLinus Torvalds if (err < 0) 3631da177e4SLinus Torvalds goto out; 3641da177e4SLinus Torvalds } 3653da28eb1STrond Myklebust err = nfs_commit_inode(inode, wb_priority(wbc)); 3661da177e4SLinus Torvalds if (err > 0) { 3671da177e4SLinus Torvalds wbc->nr_to_write -= err; 3681da177e4SLinus Torvalds err = 0; 3691da177e4SLinus Torvalds } 3701da177e4SLinus Torvalds out: 3711da177e4SLinus Torvalds clear_bit(BDI_write_congested, &bdi->state); 3721da177e4SLinus Torvalds wake_up_all(&nfs_write_congestion); 3731da177e4SLinus Torvalds return err; 3741da177e4SLinus Torvalds } 3751da177e4SLinus Torvalds 3761da177e4SLinus Torvalds /* 3771da177e4SLinus Torvalds * Insert a write request into an inode 3781da177e4SLinus Torvalds */ 3791da177e4SLinus Torvalds static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 3801da177e4SLinus Torvalds { 3811da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 3821da177e4SLinus Torvalds int error; 3831da177e4SLinus Torvalds 3841da177e4SLinus Torvalds error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); 3851da177e4SLinus Torvalds BUG_ON(error == -EEXIST); 3861da177e4SLinus Torvalds if (error) 3871da177e4SLinus Torvalds return error; 3881da177e4SLinus Torvalds if (!nfsi->npages) { 3891da177e4SLinus Torvalds igrab(inode); 3901da177e4SLinus Torvalds nfs_begin_data_update(inode); 3911da177e4SLinus Torvalds if (nfs_have_delegation(inode, FMODE_WRITE)) 3921da177e4SLinus Torvalds nfsi->change_attr++; 3931da177e4SLinus Torvalds } 3941da177e4SLinus Torvalds nfsi->npages++; 3951da177e4SLinus Torvalds atomic_inc(&req->wb_count); 3961da177e4SLinus Torvalds return 0; 3971da177e4SLinus Torvalds } 3981da177e4SLinus Torvalds 3991da177e4SLinus Torvalds /* 4001da177e4SLinus Torvalds * Insert a write request into an inode 4011da177e4SLinus Torvalds */ 4021da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 4031da177e4SLinus Torvalds { 4041da177e4SLinus Torvalds struct inode *inode = req->wb_context->dentry->d_inode; 4051da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4061da177e4SLinus Torvalds 4071da177e4SLinus Torvalds BUG_ON (!NFS_WBACK_BUSY(req)); 4081da177e4SLinus Torvalds 4091da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 4101da177e4SLinus Torvalds radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); 4111da177e4SLinus Torvalds nfsi->npages--; 4121da177e4SLinus Torvalds if (!nfsi->npages) { 4131da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 414951a143bSTrond Myklebust nfs_end_data_update(inode); 4151da177e4SLinus Torvalds iput(inode); 4161da177e4SLinus Torvalds } else 4171da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 4181da177e4SLinus Torvalds nfs_clear_request(req); 4191da177e4SLinus Torvalds nfs_release_request(req); 4201da177e4SLinus Torvalds } 4211da177e4SLinus Torvalds 4221da177e4SLinus Torvalds /* 4231da177e4SLinus Torvalds * Find a request 4241da177e4SLinus Torvalds */ 4251da177e4SLinus Torvalds static inline struct nfs_page * 4261da177e4SLinus Torvalds _nfs_find_request(struct inode *inode, unsigned long index) 4271da177e4SLinus Torvalds { 4281da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4291da177e4SLinus Torvalds struct nfs_page *req; 4301da177e4SLinus Torvalds 4311da177e4SLinus Torvalds req = (struct nfs_page*)radix_tree_lookup(&nfsi->nfs_page_tree, index); 4321da177e4SLinus Torvalds if (req) 4331da177e4SLinus Torvalds atomic_inc(&req->wb_count); 4341da177e4SLinus Torvalds return req; 4351da177e4SLinus Torvalds } 4361da177e4SLinus Torvalds 4371da177e4SLinus Torvalds static struct nfs_page * 4381da177e4SLinus Torvalds nfs_find_request(struct inode *inode, unsigned long index) 4391da177e4SLinus Torvalds { 4401da177e4SLinus Torvalds struct nfs_page *req; 4411da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4421da177e4SLinus Torvalds 4431da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 4441da177e4SLinus Torvalds req = _nfs_find_request(inode, index); 4451da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 4461da177e4SLinus Torvalds return req; 4471da177e4SLinus Torvalds } 4481da177e4SLinus Torvalds 4491da177e4SLinus Torvalds /* 4501da177e4SLinus Torvalds * Add a request to the inode's dirty list. 4511da177e4SLinus Torvalds */ 4521da177e4SLinus Torvalds static void 4531da177e4SLinus Torvalds nfs_mark_request_dirty(struct nfs_page *req) 4541da177e4SLinus Torvalds { 4551da177e4SLinus Torvalds struct inode *inode = req->wb_context->dentry->d_inode; 4561da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4571da177e4SLinus Torvalds 4581da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 4593da28eb1STrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, 4603da28eb1STrond Myklebust req->wb_index, NFS_PAGE_TAG_DIRTY); 4611da177e4SLinus Torvalds nfs_list_add_request(req, &nfsi->dirty); 4621da177e4SLinus Torvalds nfsi->ndirty++; 4631da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 4641da177e4SLinus Torvalds inc_page_state(nr_dirty); 4651da177e4SLinus Torvalds mark_inode_dirty(inode); 4661da177e4SLinus Torvalds } 4671da177e4SLinus Torvalds 4681da177e4SLinus Torvalds /* 4691da177e4SLinus Torvalds * Check if a request is dirty 4701da177e4SLinus Torvalds */ 4711da177e4SLinus Torvalds static inline int 4721da177e4SLinus Torvalds nfs_dirty_request(struct nfs_page *req) 4731da177e4SLinus Torvalds { 4741da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode); 4751da177e4SLinus Torvalds return !list_empty(&req->wb_list) && req->wb_list_head == &nfsi->dirty; 4761da177e4SLinus Torvalds } 4771da177e4SLinus Torvalds 4781da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 4791da177e4SLinus Torvalds /* 4801da177e4SLinus Torvalds * Add a request to the inode's commit list. 4811da177e4SLinus Torvalds */ 4821da177e4SLinus Torvalds static void 4831da177e4SLinus Torvalds nfs_mark_request_commit(struct nfs_page *req) 4841da177e4SLinus Torvalds { 4851da177e4SLinus Torvalds struct inode *inode = req->wb_context->dentry->d_inode; 4861da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4871da177e4SLinus Torvalds 4881da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 4891da177e4SLinus Torvalds nfs_list_add_request(req, &nfsi->commit); 4901da177e4SLinus Torvalds nfsi->ncommit++; 4911da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 4921da177e4SLinus Torvalds inc_page_state(nr_unstable); 4931da177e4SLinus Torvalds mark_inode_dirty(inode); 4941da177e4SLinus Torvalds } 4951da177e4SLinus Torvalds #endif 4961da177e4SLinus Torvalds 4971da177e4SLinus Torvalds /* 4981da177e4SLinus Torvalds * Wait for a request to complete. 4991da177e4SLinus Torvalds * 5001da177e4SLinus Torvalds * Interruptible by signals only if mounted with intr flag. 5011da177e4SLinus Torvalds */ 5021da177e4SLinus Torvalds static int 5031da177e4SLinus Torvalds nfs_wait_on_requests(struct inode *inode, unsigned long idx_start, unsigned int npages) 5041da177e4SLinus Torvalds { 5051da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5061da177e4SLinus Torvalds struct nfs_page *req; 5071da177e4SLinus Torvalds unsigned long idx_end, next; 5081da177e4SLinus Torvalds unsigned int res = 0; 5091da177e4SLinus Torvalds int error; 5101da177e4SLinus Torvalds 5111da177e4SLinus Torvalds if (npages == 0) 5121da177e4SLinus Torvalds idx_end = ~0; 5131da177e4SLinus Torvalds else 5141da177e4SLinus Torvalds idx_end = idx_start + npages - 1; 5151da177e4SLinus Torvalds 5161da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 5171da177e4SLinus Torvalds next = idx_start; 518c6a556b8STrond Myklebust while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_WRITEBACK)) { 5191da177e4SLinus Torvalds if (req->wb_index > idx_end) 5201da177e4SLinus Torvalds break; 5211da177e4SLinus Torvalds 5221da177e4SLinus Torvalds next = req->wb_index + 1; 523c6a556b8STrond Myklebust BUG_ON(!NFS_WBACK_BUSY(req)); 5241da177e4SLinus Torvalds 5251da177e4SLinus Torvalds atomic_inc(&req->wb_count); 5261da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 5271da177e4SLinus Torvalds error = nfs_wait_on_request(req); 5281da177e4SLinus Torvalds nfs_release_request(req); 5291da177e4SLinus Torvalds if (error < 0) 5301da177e4SLinus Torvalds return error; 5311da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 5321da177e4SLinus Torvalds res++; 5331da177e4SLinus Torvalds } 5341da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 5351da177e4SLinus Torvalds return res; 5361da177e4SLinus Torvalds } 5371da177e4SLinus Torvalds 5381da177e4SLinus Torvalds /* 5391da177e4SLinus Torvalds * nfs_scan_dirty - Scan an inode for dirty requests 5401da177e4SLinus Torvalds * @inode: NFS inode to scan 5411da177e4SLinus Torvalds * @dst: destination list 5421da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 5431da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 5441da177e4SLinus Torvalds * 5451da177e4SLinus Torvalds * Moves requests from the inode's dirty page list. 5461da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 5471da177e4SLinus Torvalds */ 5481da177e4SLinus Torvalds static int 5491da177e4SLinus Torvalds nfs_scan_dirty(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) 5501da177e4SLinus Torvalds { 5511da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5523da28eb1STrond Myklebust int res = 0; 5533da28eb1STrond Myklebust 5543da28eb1STrond Myklebust if (nfsi->ndirty != 0) { 5553da28eb1STrond Myklebust res = nfs_scan_lock_dirty(nfsi, dst, idx_start, npages); 5561da177e4SLinus Torvalds nfsi->ndirty -= res; 5571da177e4SLinus Torvalds sub_page_state(nr_dirty,res); 5581da177e4SLinus Torvalds if ((nfsi->ndirty == 0) != list_empty(&nfsi->dirty)) 5591da177e4SLinus Torvalds printk(KERN_ERR "NFS: desynchronized value of nfs_i.ndirty.\n"); 5603da28eb1STrond Myklebust } 5611da177e4SLinus Torvalds return res; 5621da177e4SLinus Torvalds } 5631da177e4SLinus Torvalds 5641da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 5651da177e4SLinus Torvalds /* 5661da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 5671da177e4SLinus Torvalds * @inode: NFS inode to scan 5681da177e4SLinus Torvalds * @dst: destination list 5691da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 5701da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 5711da177e4SLinus Torvalds * 5721da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 5731da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 5741da177e4SLinus Torvalds */ 5751da177e4SLinus Torvalds static int 5761da177e4SLinus Torvalds nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) 5771da177e4SLinus Torvalds { 5781da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5793da28eb1STrond Myklebust int res = 0; 5803da28eb1STrond Myklebust 5813da28eb1STrond Myklebust if (nfsi->ncommit != 0) { 5821da177e4SLinus Torvalds res = nfs_scan_list(&nfsi->commit, dst, idx_start, npages); 5831da177e4SLinus Torvalds nfsi->ncommit -= res; 5841da177e4SLinus Torvalds if ((nfsi->ncommit == 0) != list_empty(&nfsi->commit)) 5851da177e4SLinus Torvalds printk(KERN_ERR "NFS: desynchronized value of nfs_i.ncommit.\n"); 5863da28eb1STrond Myklebust } 5871da177e4SLinus Torvalds return res; 5881da177e4SLinus Torvalds } 5891da177e4SLinus Torvalds #endif 5901da177e4SLinus Torvalds 5911da177e4SLinus Torvalds static int nfs_wait_on_write_congestion(struct address_space *mapping, int intr) 5921da177e4SLinus Torvalds { 5931da177e4SLinus Torvalds struct backing_dev_info *bdi = mapping->backing_dev_info; 5941da177e4SLinus Torvalds DEFINE_WAIT(wait); 5951da177e4SLinus Torvalds int ret = 0; 5961da177e4SLinus Torvalds 5971da177e4SLinus Torvalds might_sleep(); 5981da177e4SLinus Torvalds 5991da177e4SLinus Torvalds if (!bdi_write_congested(bdi)) 6001da177e4SLinus Torvalds return 0; 6011da177e4SLinus Torvalds if (intr) { 6021da177e4SLinus Torvalds struct rpc_clnt *clnt = NFS_CLIENT(mapping->host); 6031da177e4SLinus Torvalds sigset_t oldset; 6041da177e4SLinus Torvalds 6051da177e4SLinus Torvalds rpc_clnt_sigmask(clnt, &oldset); 6061da177e4SLinus Torvalds prepare_to_wait(&nfs_write_congestion, &wait, TASK_INTERRUPTIBLE); 6071da177e4SLinus Torvalds if (bdi_write_congested(bdi)) { 6081da177e4SLinus Torvalds if (signalled()) 6091da177e4SLinus Torvalds ret = -ERESTARTSYS; 6101da177e4SLinus Torvalds else 6111da177e4SLinus Torvalds schedule(); 6121da177e4SLinus Torvalds } 6131da177e4SLinus Torvalds rpc_clnt_sigunmask(clnt, &oldset); 6141da177e4SLinus Torvalds } else { 6151da177e4SLinus Torvalds prepare_to_wait(&nfs_write_congestion, &wait, TASK_UNINTERRUPTIBLE); 6161da177e4SLinus Torvalds if (bdi_write_congested(bdi)) 6171da177e4SLinus Torvalds schedule(); 6181da177e4SLinus Torvalds } 6191da177e4SLinus Torvalds finish_wait(&nfs_write_congestion, &wait); 6201da177e4SLinus Torvalds return ret; 6211da177e4SLinus Torvalds } 6221da177e4SLinus Torvalds 6231da177e4SLinus Torvalds 6241da177e4SLinus Torvalds /* 6251da177e4SLinus Torvalds * Try to update any existing write request, or create one if there is none. 6261da177e4SLinus Torvalds * In order to match, the request's credentials must match those of 6271da177e4SLinus Torvalds * the calling process. 6281da177e4SLinus Torvalds * 6291da177e4SLinus Torvalds * Note: Should always be called with the Page Lock held! 6301da177e4SLinus Torvalds */ 6311da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx, 6321da177e4SLinus Torvalds struct inode *inode, struct page *page, 6331da177e4SLinus Torvalds unsigned int offset, unsigned int bytes) 6341da177e4SLinus Torvalds { 6351da177e4SLinus Torvalds struct nfs_server *server = NFS_SERVER(inode); 6361da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 6371da177e4SLinus Torvalds struct nfs_page *req, *new = NULL; 6381da177e4SLinus Torvalds unsigned long rqend, end; 6391da177e4SLinus Torvalds 6401da177e4SLinus Torvalds end = offset + bytes; 6411da177e4SLinus Torvalds 6421da177e4SLinus Torvalds if (nfs_wait_on_write_congestion(page->mapping, server->flags & NFS_MOUNT_INTR)) 6431da177e4SLinus Torvalds return ERR_PTR(-ERESTARTSYS); 6441da177e4SLinus Torvalds for (;;) { 6451da177e4SLinus Torvalds /* Loop over all inode entries and see if we find 6461da177e4SLinus Torvalds * A request for the page we wish to update 6471da177e4SLinus Torvalds */ 6481da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 6491da177e4SLinus Torvalds req = _nfs_find_request(inode, page->index); 6501da177e4SLinus Torvalds if (req) { 6511da177e4SLinus Torvalds if (!nfs_lock_request_dontget(req)) { 6521da177e4SLinus Torvalds int error; 6531da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 6541da177e4SLinus Torvalds error = nfs_wait_on_request(req); 6551da177e4SLinus Torvalds nfs_release_request(req); 656*1dd594b2SNeil Brown if (error < 0) { 657*1dd594b2SNeil Brown if (new) 658*1dd594b2SNeil Brown nfs_release_request(new); 6591da177e4SLinus Torvalds return ERR_PTR(error); 660*1dd594b2SNeil Brown } 6611da177e4SLinus Torvalds continue; 6621da177e4SLinus Torvalds } 6631da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 6641da177e4SLinus Torvalds if (new) 6651da177e4SLinus Torvalds nfs_release_request(new); 6661da177e4SLinus Torvalds break; 6671da177e4SLinus Torvalds } 6681da177e4SLinus Torvalds 6691da177e4SLinus Torvalds if (new) { 6701da177e4SLinus Torvalds int error; 6711da177e4SLinus Torvalds nfs_lock_request_dontget(new); 6721da177e4SLinus Torvalds error = nfs_inode_add_request(inode, new); 6731da177e4SLinus Torvalds if (error) { 6741da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 6751da177e4SLinus Torvalds nfs_unlock_request(new); 6761da177e4SLinus Torvalds return ERR_PTR(error); 6771da177e4SLinus Torvalds } 6781da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 6791da177e4SLinus Torvalds nfs_mark_request_dirty(new); 6801da177e4SLinus Torvalds return new; 6811da177e4SLinus Torvalds } 6821da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 6831da177e4SLinus Torvalds 6841da177e4SLinus Torvalds new = nfs_create_request(ctx, inode, page, offset, bytes); 6851da177e4SLinus Torvalds if (IS_ERR(new)) 6861da177e4SLinus Torvalds return new; 6871da177e4SLinus Torvalds } 6881da177e4SLinus Torvalds 6891da177e4SLinus Torvalds /* We have a request for our page. 6901da177e4SLinus Torvalds * If the creds don't match, or the 6911da177e4SLinus Torvalds * page addresses don't match, 6921da177e4SLinus Torvalds * tell the caller to wait on the conflicting 6931da177e4SLinus Torvalds * request. 6941da177e4SLinus Torvalds */ 6951da177e4SLinus Torvalds rqend = req->wb_offset + req->wb_bytes; 6961da177e4SLinus Torvalds if (req->wb_context != ctx 6971da177e4SLinus Torvalds || req->wb_page != page 6981da177e4SLinus Torvalds || !nfs_dirty_request(req) 6991da177e4SLinus Torvalds || offset > rqend || end < req->wb_offset) { 7001da177e4SLinus Torvalds nfs_unlock_request(req); 7011da177e4SLinus Torvalds return ERR_PTR(-EBUSY); 7021da177e4SLinus Torvalds } 7031da177e4SLinus Torvalds 7041da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 7051da177e4SLinus Torvalds if (offset < req->wb_offset) { 7061da177e4SLinus Torvalds req->wb_offset = offset; 7071da177e4SLinus Torvalds req->wb_pgbase = offset; 7081da177e4SLinus Torvalds req->wb_bytes = rqend - req->wb_offset; 7091da177e4SLinus Torvalds } 7101da177e4SLinus Torvalds 7111da177e4SLinus Torvalds if (end > rqend) 7121da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 7131da177e4SLinus Torvalds 7141da177e4SLinus Torvalds return req; 7151da177e4SLinus Torvalds } 7161da177e4SLinus Torvalds 7171da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 7181da177e4SLinus Torvalds { 7191da177e4SLinus Torvalds struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; 7201da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 7211da177e4SLinus Torvalds struct nfs_page *req; 7221da177e4SLinus Torvalds int status = 0; 7231da177e4SLinus Torvalds /* 7241da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 7251da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 7261da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 7271da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 7281da177e4SLinus Torvalds * Also do the same if we find a request from an existing 7291da177e4SLinus Torvalds * dropped page. 7301da177e4SLinus Torvalds */ 7311da177e4SLinus Torvalds req = nfs_find_request(inode, page->index); 7321da177e4SLinus Torvalds if (req) { 7331da177e4SLinus Torvalds if (req->wb_page != page || ctx != req->wb_context) 7341da177e4SLinus Torvalds status = nfs_wb_page(inode, page); 7351da177e4SLinus Torvalds nfs_release_request(req); 7361da177e4SLinus Torvalds } 7371da177e4SLinus Torvalds return (status < 0) ? status : 0; 7381da177e4SLinus Torvalds } 7391da177e4SLinus Torvalds 7401da177e4SLinus Torvalds /* 7411da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 7421da177e4SLinus Torvalds * 7431da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 7441da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 7451da177e4SLinus Torvalds */ 7461da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 7471da177e4SLinus Torvalds unsigned int offset, unsigned int count) 7481da177e4SLinus Torvalds { 7491da177e4SLinus Torvalds struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; 7501da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 7511da177e4SLinus Torvalds struct nfs_page *req; 7521da177e4SLinus Torvalds int status = 0; 7531da177e4SLinus Torvalds 7541da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage(%s/%s %d@%Ld)\n", 7550bbacc40SChuck Lever file->f_dentry->d_parent->d_name.name, 7560bbacc40SChuck Lever file->f_dentry->d_name.name, count, 7570bbacc40SChuck Lever (long long)(page_offset(page) +offset)); 7581da177e4SLinus Torvalds 7591da177e4SLinus Torvalds if (IS_SYNC(inode)) { 7601da177e4SLinus Torvalds status = nfs_writepage_sync(ctx, inode, page, offset, count, 0); 7611da177e4SLinus Torvalds if (status > 0) { 7621da177e4SLinus Torvalds if (offset == 0 && status == PAGE_CACHE_SIZE) 7631da177e4SLinus Torvalds SetPageUptodate(page); 7641da177e4SLinus Torvalds return 0; 7651da177e4SLinus Torvalds } 7661da177e4SLinus Torvalds return status; 7671da177e4SLinus Torvalds } 7681da177e4SLinus Torvalds 7691da177e4SLinus Torvalds /* If we're not using byte range locks, and we know the page 7701da177e4SLinus Torvalds * is entirely in cache, it may be more efficient to avoid 7711da177e4SLinus Torvalds * fragmenting write requests. 7721da177e4SLinus Torvalds */ 773ab0a3dbeSTrond Myklebust if (PageUptodate(page) && inode->i_flock == NULL && !(file->f_mode & O_SYNC)) { 7741da177e4SLinus Torvalds loff_t end_offs = i_size_read(inode) - 1; 7751da177e4SLinus Torvalds unsigned long end_index = end_offs >> PAGE_CACHE_SHIFT; 7761da177e4SLinus Torvalds 7771da177e4SLinus Torvalds count += offset; 7781da177e4SLinus Torvalds offset = 0; 7791da177e4SLinus Torvalds if (unlikely(end_offs < 0)) { 7801da177e4SLinus Torvalds /* Do nothing */ 7811da177e4SLinus Torvalds } else if (page->index == end_index) { 7821da177e4SLinus Torvalds unsigned int pglen; 7831da177e4SLinus Torvalds pglen = (unsigned int)(end_offs & (PAGE_CACHE_SIZE-1)) + 1; 7841da177e4SLinus Torvalds if (count < pglen) 7851da177e4SLinus Torvalds count = pglen; 7861da177e4SLinus Torvalds } else if (page->index < end_index) 7871da177e4SLinus Torvalds count = PAGE_CACHE_SIZE; 7881da177e4SLinus Torvalds } 7891da177e4SLinus Torvalds 7901da177e4SLinus Torvalds /* 7911da177e4SLinus Torvalds * Try to find an NFS request corresponding to this page 7921da177e4SLinus Torvalds * and update it. 7931da177e4SLinus Torvalds * If the existing request cannot be updated, we must flush 7941da177e4SLinus Torvalds * it out now. 7951da177e4SLinus Torvalds */ 7961da177e4SLinus Torvalds do { 7971da177e4SLinus Torvalds req = nfs_update_request(ctx, inode, page, offset, count); 7981da177e4SLinus Torvalds status = (IS_ERR(req)) ? PTR_ERR(req) : 0; 7991da177e4SLinus Torvalds if (status != -EBUSY) 8001da177e4SLinus Torvalds break; 8011da177e4SLinus Torvalds /* Request could not be updated. Flush it out and try again */ 8021da177e4SLinus Torvalds status = nfs_wb_page(inode, page); 8031da177e4SLinus Torvalds } while (status >= 0); 8041da177e4SLinus Torvalds if (status < 0) 8051da177e4SLinus Torvalds goto done; 8061da177e4SLinus Torvalds 8071da177e4SLinus Torvalds status = 0; 8081da177e4SLinus Torvalds 8091da177e4SLinus Torvalds /* Update file length */ 8101da177e4SLinus Torvalds nfs_grow_file(page, offset, count); 8111da177e4SLinus Torvalds /* Set the PG_uptodate flag? */ 8121da177e4SLinus Torvalds nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 8131da177e4SLinus Torvalds nfs_unlock_request(req); 8141da177e4SLinus Torvalds done: 8151da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage returns %d (isize %Ld)\n", 8161da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 8171da177e4SLinus Torvalds if (status < 0) 8181da177e4SLinus Torvalds ClearPageUptodate(page); 8191da177e4SLinus Torvalds return status; 8201da177e4SLinus Torvalds } 8211da177e4SLinus Torvalds 8221da177e4SLinus Torvalds static void nfs_writepage_release(struct nfs_page *req) 8231da177e4SLinus Torvalds { 8241da177e4SLinus Torvalds end_page_writeback(req->wb_page); 8251da177e4SLinus Torvalds 8261da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 8271da177e4SLinus Torvalds if (!PageError(req->wb_page)) { 8281da177e4SLinus Torvalds if (NFS_NEED_RESCHED(req)) { 8291da177e4SLinus Torvalds nfs_mark_request_dirty(req); 8301da177e4SLinus Torvalds goto out; 8311da177e4SLinus Torvalds } else if (NFS_NEED_COMMIT(req)) { 8321da177e4SLinus Torvalds nfs_mark_request_commit(req); 8331da177e4SLinus Torvalds goto out; 8341da177e4SLinus Torvalds } 8351da177e4SLinus Torvalds } 8361da177e4SLinus Torvalds nfs_inode_remove_request(req); 8371da177e4SLinus Torvalds 8381da177e4SLinus Torvalds out: 8391da177e4SLinus Torvalds nfs_clear_commit(req); 8401da177e4SLinus Torvalds nfs_clear_reschedule(req); 8411da177e4SLinus Torvalds #else 8421da177e4SLinus Torvalds nfs_inode_remove_request(req); 8431da177e4SLinus Torvalds #endif 844c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 8451da177e4SLinus Torvalds } 8461da177e4SLinus Torvalds 8471da177e4SLinus Torvalds static inline int flush_task_priority(int how) 8481da177e4SLinus Torvalds { 8491da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 8501da177e4SLinus Torvalds case FLUSH_HIGHPRI: 8511da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 8521da177e4SLinus Torvalds case FLUSH_LOWPRI: 8531da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 8541da177e4SLinus Torvalds } 8551da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 8561da177e4SLinus Torvalds } 8571da177e4SLinus Torvalds 8581da177e4SLinus Torvalds /* 8591da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 8601da177e4SLinus Torvalds */ 8611da177e4SLinus Torvalds static void nfs_write_rpcsetup(struct nfs_page *req, 8621da177e4SLinus Torvalds struct nfs_write_data *data, 8631da177e4SLinus Torvalds unsigned int count, unsigned int offset, 8641da177e4SLinus Torvalds int how) 8651da177e4SLinus Torvalds { 8661da177e4SLinus Torvalds struct inode *inode; 8671da177e4SLinus Torvalds 8681da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 8691da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 8701da177e4SLinus Torvalds 8711da177e4SLinus Torvalds data->req = req; 8721da177e4SLinus Torvalds data->inode = inode = req->wb_context->dentry->d_inode; 8731da177e4SLinus Torvalds data->cred = req->wb_context->cred; 8741da177e4SLinus Torvalds 8751da177e4SLinus Torvalds data->args.fh = NFS_FH(inode); 8761da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 8771da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 8781da177e4SLinus Torvalds data->args.pages = data->pagevec; 8791da177e4SLinus Torvalds data->args.count = count; 8801da177e4SLinus Torvalds data->args.context = req->wb_context; 8811da177e4SLinus Torvalds 8821da177e4SLinus Torvalds data->res.fattr = &data->fattr; 8831da177e4SLinus Torvalds data->res.count = count; 8841da177e4SLinus Torvalds data->res.verf = &data->verf; 8850e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 8861da177e4SLinus Torvalds 8871da177e4SLinus Torvalds NFS_PROTO(inode)->write_setup(data, how); 8881da177e4SLinus Torvalds 8891da177e4SLinus Torvalds data->task.tk_priority = flush_task_priority(how); 8901da177e4SLinus Torvalds data->task.tk_cookie = (unsigned long)inode; 8911da177e4SLinus Torvalds 8921da177e4SLinus Torvalds dprintk("NFS: %4d initiated write call (req %s/%Ld, %u bytes @ offset %Lu)\n", 8930bbacc40SChuck Lever data->task.tk_pid, 8941da177e4SLinus Torvalds inode->i_sb->s_id, 8951da177e4SLinus Torvalds (long long)NFS_FILEID(inode), 8961da177e4SLinus Torvalds count, 8971da177e4SLinus Torvalds (unsigned long long)data->args.offset); 8981da177e4SLinus Torvalds } 8991da177e4SLinus Torvalds 9001da177e4SLinus Torvalds static void nfs_execute_write(struct nfs_write_data *data) 9011da177e4SLinus Torvalds { 9021da177e4SLinus Torvalds struct rpc_clnt *clnt = NFS_CLIENT(data->inode); 9031da177e4SLinus Torvalds sigset_t oldset; 9041da177e4SLinus Torvalds 9051da177e4SLinus Torvalds rpc_clnt_sigmask(clnt, &oldset); 9061da177e4SLinus Torvalds lock_kernel(); 9071da177e4SLinus Torvalds rpc_execute(&data->task); 9081da177e4SLinus Torvalds unlock_kernel(); 9091da177e4SLinus Torvalds rpc_clnt_sigunmask(clnt, &oldset); 9101da177e4SLinus Torvalds } 9111da177e4SLinus Torvalds 9121da177e4SLinus Torvalds /* 9131da177e4SLinus Torvalds * Generate multiple small requests to write out a single 9141da177e4SLinus Torvalds * contiguous dirty area on one page. 9151da177e4SLinus Torvalds */ 9161da177e4SLinus Torvalds static int nfs_flush_multi(struct list_head *head, struct inode *inode, int how) 9171da177e4SLinus Torvalds { 9181da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 9191da177e4SLinus Torvalds struct page *page = req->wb_page; 9201da177e4SLinus Torvalds struct nfs_write_data *data; 9211da177e4SLinus Torvalds unsigned int wsize = NFS_SERVER(inode)->wsize; 9221da177e4SLinus Torvalds unsigned int nbytes, offset; 9231da177e4SLinus Torvalds int requests = 0; 9241da177e4SLinus Torvalds LIST_HEAD(list); 9251da177e4SLinus Torvalds 9261da177e4SLinus Torvalds nfs_list_remove_request(req); 9271da177e4SLinus Torvalds 9281da177e4SLinus Torvalds nbytes = req->wb_bytes; 9291da177e4SLinus Torvalds for (;;) { 93040859d7eSChuck Lever data = nfs_writedata_alloc(1); 9311da177e4SLinus Torvalds if (!data) 9321da177e4SLinus Torvalds goto out_bad; 9331da177e4SLinus Torvalds list_add(&data->pages, &list); 9341da177e4SLinus Torvalds requests++; 9351da177e4SLinus Torvalds if (nbytes <= wsize) 9361da177e4SLinus Torvalds break; 9371da177e4SLinus Torvalds nbytes -= wsize; 9381da177e4SLinus Torvalds } 9391da177e4SLinus Torvalds atomic_set(&req->wb_complete, requests); 9401da177e4SLinus Torvalds 9411da177e4SLinus Torvalds ClearPageError(page); 942bb713d6dSTrond Myklebust set_page_writeback(page); 9431da177e4SLinus Torvalds offset = 0; 9441da177e4SLinus Torvalds nbytes = req->wb_bytes; 9451da177e4SLinus Torvalds do { 9461da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 9471da177e4SLinus Torvalds list_del_init(&data->pages); 9481da177e4SLinus Torvalds 9491da177e4SLinus Torvalds data->pagevec[0] = page; 9501da177e4SLinus Torvalds data->complete = nfs_writeback_done_partial; 9511da177e4SLinus Torvalds 9521da177e4SLinus Torvalds if (nbytes > wsize) { 9531da177e4SLinus Torvalds nfs_write_rpcsetup(req, data, wsize, offset, how); 9541da177e4SLinus Torvalds offset += wsize; 9551da177e4SLinus Torvalds nbytes -= wsize; 9561da177e4SLinus Torvalds } else { 9571da177e4SLinus Torvalds nfs_write_rpcsetup(req, data, nbytes, offset, how); 9581da177e4SLinus Torvalds nbytes = 0; 9591da177e4SLinus Torvalds } 9601da177e4SLinus Torvalds nfs_execute_write(data); 9611da177e4SLinus Torvalds } while (nbytes != 0); 9621da177e4SLinus Torvalds 9631da177e4SLinus Torvalds return 0; 9641da177e4SLinus Torvalds 9651da177e4SLinus Torvalds out_bad: 9661da177e4SLinus Torvalds while (!list_empty(&list)) { 9671da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 9681da177e4SLinus Torvalds list_del(&data->pages); 9691da177e4SLinus Torvalds nfs_writedata_free(data); 9701da177e4SLinus Torvalds } 9711da177e4SLinus Torvalds nfs_mark_request_dirty(req); 972c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 9731da177e4SLinus Torvalds return -ENOMEM; 9741da177e4SLinus Torvalds } 9751da177e4SLinus Torvalds 9761da177e4SLinus Torvalds /* 9771da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 9781da177e4SLinus Torvalds * The page must have been locked by the caller. 9791da177e4SLinus Torvalds * 9801da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 9811da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 9821da177e4SLinus Torvalds * that has been written but not committed. 9831da177e4SLinus Torvalds */ 9841da177e4SLinus Torvalds static int nfs_flush_one(struct list_head *head, struct inode *inode, int how) 9851da177e4SLinus Torvalds { 9861da177e4SLinus Torvalds struct nfs_page *req; 9871da177e4SLinus Torvalds struct page **pages; 9881da177e4SLinus Torvalds struct nfs_write_data *data; 9891da177e4SLinus Torvalds unsigned int count; 9901da177e4SLinus Torvalds 9911da177e4SLinus Torvalds if (NFS_SERVER(inode)->wsize < PAGE_CACHE_SIZE) 9921da177e4SLinus Torvalds return nfs_flush_multi(head, inode, how); 9931da177e4SLinus Torvalds 99440859d7eSChuck Lever data = nfs_writedata_alloc(NFS_SERVER(inode)->wpages); 9951da177e4SLinus Torvalds if (!data) 9961da177e4SLinus Torvalds goto out_bad; 9971da177e4SLinus Torvalds 9981da177e4SLinus Torvalds pages = data->pagevec; 9991da177e4SLinus Torvalds count = 0; 10001da177e4SLinus Torvalds while (!list_empty(head)) { 10011da177e4SLinus Torvalds req = nfs_list_entry(head->next); 10021da177e4SLinus Torvalds nfs_list_remove_request(req); 10031da177e4SLinus Torvalds nfs_list_add_request(req, &data->pages); 10041da177e4SLinus Torvalds ClearPageError(req->wb_page); 1005bb713d6dSTrond Myklebust set_page_writeback(req->wb_page); 10061da177e4SLinus Torvalds *pages++ = req->wb_page; 10071da177e4SLinus Torvalds count += req->wb_bytes; 10081da177e4SLinus Torvalds } 10091da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 10101da177e4SLinus Torvalds 10111da177e4SLinus Torvalds data->complete = nfs_writeback_done_full; 10121da177e4SLinus Torvalds /* Set up the argument struct */ 10131da177e4SLinus Torvalds nfs_write_rpcsetup(req, data, count, 0, how); 10141da177e4SLinus Torvalds 10151da177e4SLinus Torvalds nfs_execute_write(data); 10161da177e4SLinus Torvalds return 0; 10171da177e4SLinus Torvalds out_bad: 10181da177e4SLinus Torvalds while (!list_empty(head)) { 10191da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 10201da177e4SLinus Torvalds nfs_list_remove_request(req); 10211da177e4SLinus Torvalds nfs_mark_request_dirty(req); 1022c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 10231da177e4SLinus Torvalds } 10241da177e4SLinus Torvalds return -ENOMEM; 10251da177e4SLinus Torvalds } 10261da177e4SLinus Torvalds 10271da177e4SLinus Torvalds static int 10281da177e4SLinus Torvalds nfs_flush_list(struct list_head *head, int wpages, int how) 10291da177e4SLinus Torvalds { 10301da177e4SLinus Torvalds LIST_HEAD(one_request); 10311da177e4SLinus Torvalds struct nfs_page *req; 10321da177e4SLinus Torvalds int error = 0; 10331da177e4SLinus Torvalds unsigned int pages = 0; 10341da177e4SLinus Torvalds 10351da177e4SLinus Torvalds while (!list_empty(head)) { 10361da177e4SLinus Torvalds pages += nfs_coalesce_requests(head, &one_request, wpages); 10371da177e4SLinus Torvalds req = nfs_list_entry(one_request.next); 10381da177e4SLinus Torvalds error = nfs_flush_one(&one_request, req->wb_context->dentry->d_inode, how); 10391da177e4SLinus Torvalds if (error < 0) 10401da177e4SLinus Torvalds break; 10411da177e4SLinus Torvalds } 10421da177e4SLinus Torvalds if (error >= 0) 10431da177e4SLinus Torvalds return pages; 10441da177e4SLinus Torvalds 10451da177e4SLinus Torvalds while (!list_empty(head)) { 10461da177e4SLinus Torvalds req = nfs_list_entry(head->next); 10471da177e4SLinus Torvalds nfs_list_remove_request(req); 10481da177e4SLinus Torvalds nfs_mark_request_dirty(req); 1049c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 10501da177e4SLinus Torvalds } 10511da177e4SLinus Torvalds return error; 10521da177e4SLinus Torvalds } 10531da177e4SLinus Torvalds 10541da177e4SLinus Torvalds /* 10551da177e4SLinus Torvalds * Handle a write reply that flushed part of a page. 10561da177e4SLinus Torvalds */ 10571da177e4SLinus Torvalds static void nfs_writeback_done_partial(struct nfs_write_data *data, int status) 10581da177e4SLinus Torvalds { 10591da177e4SLinus Torvalds struct nfs_page *req = data->req; 10601da177e4SLinus Torvalds struct page *page = req->wb_page; 10611da177e4SLinus Torvalds 10621da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 10631da177e4SLinus Torvalds req->wb_context->dentry->d_inode->i_sb->s_id, 10641da177e4SLinus Torvalds (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 10651da177e4SLinus Torvalds req->wb_bytes, 10661da177e4SLinus Torvalds (long long)req_offset(req)); 10671da177e4SLinus Torvalds 10681da177e4SLinus Torvalds if (status < 0) { 10691da177e4SLinus Torvalds ClearPageUptodate(page); 10701da177e4SLinus Torvalds SetPageError(page); 10711da177e4SLinus Torvalds req->wb_context->error = status; 10721da177e4SLinus Torvalds dprintk(", error = %d\n", status); 10731da177e4SLinus Torvalds } else { 10741da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 10751da177e4SLinus Torvalds if (data->verf.committed < NFS_FILE_SYNC) { 10761da177e4SLinus Torvalds if (!NFS_NEED_COMMIT(req)) { 10771da177e4SLinus Torvalds nfs_defer_commit(req); 10781da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 10791da177e4SLinus Torvalds dprintk(" defer commit\n"); 10801da177e4SLinus Torvalds } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { 10811da177e4SLinus Torvalds nfs_defer_reschedule(req); 10821da177e4SLinus Torvalds dprintk(" server reboot detected\n"); 10831da177e4SLinus Torvalds } 10841da177e4SLinus Torvalds } else 10851da177e4SLinus Torvalds #endif 10861da177e4SLinus Torvalds dprintk(" OK\n"); 10871da177e4SLinus Torvalds } 10881da177e4SLinus Torvalds 10891da177e4SLinus Torvalds if (atomic_dec_and_test(&req->wb_complete)) 10901da177e4SLinus Torvalds nfs_writepage_release(req); 10911da177e4SLinus Torvalds } 10921da177e4SLinus Torvalds 10931da177e4SLinus Torvalds /* 10941da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 10951da177e4SLinus Torvalds * 10961da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 10971da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 10981da177e4SLinus Torvalds * as the page has a write request pending. 10991da177e4SLinus Torvalds */ 11001da177e4SLinus Torvalds static void nfs_writeback_done_full(struct nfs_write_data *data, int status) 11011da177e4SLinus Torvalds { 11021da177e4SLinus Torvalds struct nfs_page *req; 11031da177e4SLinus Torvalds struct page *page; 11041da177e4SLinus Torvalds 11051da177e4SLinus Torvalds /* Update attributes as result of writeback. */ 11061da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 11071da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 11081da177e4SLinus Torvalds nfs_list_remove_request(req); 11091da177e4SLinus Torvalds page = req->wb_page; 11101da177e4SLinus Torvalds 11111da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 11121da177e4SLinus Torvalds req->wb_context->dentry->d_inode->i_sb->s_id, 11131da177e4SLinus Torvalds (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 11141da177e4SLinus Torvalds req->wb_bytes, 11151da177e4SLinus Torvalds (long long)req_offset(req)); 11161da177e4SLinus Torvalds 11171da177e4SLinus Torvalds if (status < 0) { 11181da177e4SLinus Torvalds ClearPageUptodate(page); 11191da177e4SLinus Torvalds SetPageError(page); 11201da177e4SLinus Torvalds req->wb_context->error = status; 11211da177e4SLinus Torvalds end_page_writeback(page); 11221da177e4SLinus Torvalds nfs_inode_remove_request(req); 11231da177e4SLinus Torvalds dprintk(", error = %d\n", status); 11241da177e4SLinus Torvalds goto next; 11251da177e4SLinus Torvalds } 11261da177e4SLinus Torvalds end_page_writeback(page); 11271da177e4SLinus Torvalds 11281da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 11291da177e4SLinus Torvalds if (data->args.stable != NFS_UNSTABLE || data->verf.committed == NFS_FILE_SYNC) { 11301da177e4SLinus Torvalds nfs_inode_remove_request(req); 11311da177e4SLinus Torvalds dprintk(" OK\n"); 11321da177e4SLinus Torvalds goto next; 11331da177e4SLinus Torvalds } 11341da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 11351da177e4SLinus Torvalds nfs_mark_request_commit(req); 11361da177e4SLinus Torvalds dprintk(" marked for commit\n"); 11371da177e4SLinus Torvalds #else 11381da177e4SLinus Torvalds nfs_inode_remove_request(req); 11391da177e4SLinus Torvalds #endif 11401da177e4SLinus Torvalds next: 1141c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 11421da177e4SLinus Torvalds } 11431da177e4SLinus Torvalds } 11441da177e4SLinus Torvalds 11451da177e4SLinus Torvalds /* 11461da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 11471da177e4SLinus Torvalds */ 1148963d8fe5STrond Myklebust void nfs_writeback_done(struct rpc_task *task, void *calldata) 11491da177e4SLinus Torvalds { 1150963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 11511da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 11521da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 11531da177e4SLinus Torvalds 11541da177e4SLinus Torvalds dprintk("NFS: %4d nfs_writeback_done (status %d)\n", 11551da177e4SLinus Torvalds task->tk_pid, task->tk_status); 11561da177e4SLinus Torvalds 11571da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 11581da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 11591da177e4SLinus Torvalds /* We tried a write call, but the server did not 11601da177e4SLinus Torvalds * commit data to stable storage even though we 11611da177e4SLinus Torvalds * requested it. 11621da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 11631da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 11641da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 11651da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 11661da177e4SLinus Torvalds */ 11671da177e4SLinus Torvalds static unsigned long complain; 11681da177e4SLinus Torvalds 11691da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 11701da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 11711da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 11721da177e4SLinus Torvalds NFS_SERVER(data->inode)->hostname, 11731da177e4SLinus Torvalds resp->verf->committed, argp->stable); 11741da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11751da177e4SLinus Torvalds } 11761da177e4SLinus Torvalds } 11771da177e4SLinus Torvalds #endif 11781da177e4SLinus Torvalds /* Is this a short write? */ 11791da177e4SLinus Torvalds if (task->tk_status >= 0 && resp->count < argp->count) { 11801da177e4SLinus Torvalds static unsigned long complain; 11811da177e4SLinus Torvalds 11821da177e4SLinus Torvalds /* Has the server at least made some progress? */ 11831da177e4SLinus Torvalds if (resp->count != 0) { 11841da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 11851da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 11861da177e4SLinus Torvalds /* Resend from where the server left off */ 11871da177e4SLinus Torvalds argp->offset += resp->count; 11881da177e4SLinus Torvalds argp->pgbase += resp->count; 11891da177e4SLinus Torvalds argp->count -= resp->count; 11901da177e4SLinus Torvalds } else { 11911da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 11921da177e4SLinus Torvalds * headaches in the case of a server crash. 11931da177e4SLinus Torvalds */ 11941da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 11951da177e4SLinus Torvalds } 11961da177e4SLinus Torvalds rpc_restart_call(task); 11971da177e4SLinus Torvalds return; 11981da177e4SLinus Torvalds } 11991da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 12001da177e4SLinus Torvalds printk(KERN_WARNING 12011da177e4SLinus Torvalds "NFS: Server wrote zero bytes, expected %u.\n", 12021da177e4SLinus Torvalds argp->count); 12031da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 12041da177e4SLinus Torvalds } 12051da177e4SLinus Torvalds /* Can't do anything about it except throw an error. */ 12061da177e4SLinus Torvalds task->tk_status = -EIO; 12071da177e4SLinus Torvalds } 12081da177e4SLinus Torvalds 12091da177e4SLinus Torvalds /* 12101da177e4SLinus Torvalds * Process the nfs_page list 12111da177e4SLinus Torvalds */ 12121da177e4SLinus Torvalds data->complete(data, task->tk_status); 12131da177e4SLinus Torvalds } 12141da177e4SLinus Torvalds 12151da177e4SLinus Torvalds 12161da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 1217963d8fe5STrond Myklebust void nfs_commit_release(void *wdata) 12181da177e4SLinus Torvalds { 12191da177e4SLinus Torvalds nfs_commit_free(wdata); 12201da177e4SLinus Torvalds } 12211da177e4SLinus Torvalds 12221da177e4SLinus Torvalds /* 12231da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 12241da177e4SLinus Torvalds */ 12251da177e4SLinus Torvalds static void nfs_commit_rpcsetup(struct list_head *head, 12261da177e4SLinus Torvalds struct nfs_write_data *data, int how) 12271da177e4SLinus Torvalds { 12283da28eb1STrond Myklebust struct nfs_page *first; 12291da177e4SLinus Torvalds struct inode *inode; 12301da177e4SLinus Torvalds 12311da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 12321da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 12331da177e4SLinus Torvalds 12341da177e4SLinus Torvalds list_splice_init(head, &data->pages); 12351da177e4SLinus Torvalds first = nfs_list_entry(data->pages.next); 12361da177e4SLinus Torvalds inode = first->wb_context->dentry->d_inode; 12371da177e4SLinus Torvalds 12381da177e4SLinus Torvalds data->inode = inode; 12391da177e4SLinus Torvalds data->cred = first->wb_context->cred; 12401da177e4SLinus Torvalds 12411da177e4SLinus Torvalds data->args.fh = NFS_FH(data->inode); 12423da28eb1STrond Myklebust /* Note: we always request a commit of the entire inode */ 12433da28eb1STrond Myklebust data->args.offset = 0; 12443da28eb1STrond Myklebust data->args.count = 0; 12453da28eb1STrond Myklebust data->res.count = 0; 12461da177e4SLinus Torvalds data->res.fattr = &data->fattr; 12471da177e4SLinus Torvalds data->res.verf = &data->verf; 12480e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 12491da177e4SLinus Torvalds 12501da177e4SLinus Torvalds NFS_PROTO(inode)->commit_setup(data, how); 12511da177e4SLinus Torvalds 12521da177e4SLinus Torvalds data->task.tk_priority = flush_task_priority(how); 12531da177e4SLinus Torvalds data->task.tk_cookie = (unsigned long)inode; 12541da177e4SLinus Torvalds 12550bbacc40SChuck Lever dprintk("NFS: %4d initiated commit call\n", data->task.tk_pid); 12561da177e4SLinus Torvalds } 12571da177e4SLinus Torvalds 12581da177e4SLinus Torvalds /* 12591da177e4SLinus Torvalds * Commit dirty pages 12601da177e4SLinus Torvalds */ 12611da177e4SLinus Torvalds static int 126240859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how) 12631da177e4SLinus Torvalds { 12641da177e4SLinus Torvalds struct nfs_write_data *data; 12651da177e4SLinus Torvalds struct nfs_page *req; 12661da177e4SLinus Torvalds 126740859d7eSChuck Lever data = nfs_commit_alloc(NFS_SERVER(inode)->wpages); 12681da177e4SLinus Torvalds 12691da177e4SLinus Torvalds if (!data) 12701da177e4SLinus Torvalds goto out_bad; 12711da177e4SLinus Torvalds 12721da177e4SLinus Torvalds /* Set up the argument struct */ 12731da177e4SLinus Torvalds nfs_commit_rpcsetup(head, data, how); 12741da177e4SLinus Torvalds 12751da177e4SLinus Torvalds nfs_execute_write(data); 12761da177e4SLinus Torvalds return 0; 12771da177e4SLinus Torvalds out_bad: 12781da177e4SLinus Torvalds while (!list_empty(head)) { 12791da177e4SLinus Torvalds req = nfs_list_entry(head->next); 12801da177e4SLinus Torvalds nfs_list_remove_request(req); 12811da177e4SLinus Torvalds nfs_mark_request_commit(req); 1282c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 12831da177e4SLinus Torvalds } 12841da177e4SLinus Torvalds return -ENOMEM; 12851da177e4SLinus Torvalds } 12861da177e4SLinus Torvalds 12871da177e4SLinus Torvalds /* 12881da177e4SLinus Torvalds * COMMIT call returned 12891da177e4SLinus Torvalds */ 1290963d8fe5STrond Myklebust void nfs_commit_done(struct rpc_task *task, void *calldata) 12911da177e4SLinus Torvalds { 1292963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 12931da177e4SLinus Torvalds struct nfs_page *req; 12941da177e4SLinus Torvalds int res = 0; 12951da177e4SLinus Torvalds 12961da177e4SLinus Torvalds dprintk("NFS: %4d nfs_commit_done (status %d)\n", 12971da177e4SLinus Torvalds task->tk_pid, task->tk_status); 12981da177e4SLinus Torvalds 12991da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 13001da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 13011da177e4SLinus Torvalds nfs_list_remove_request(req); 13021da177e4SLinus Torvalds 13031da177e4SLinus Torvalds dprintk("NFS: commit (%s/%Ld %d@%Ld)", 13041da177e4SLinus Torvalds req->wb_context->dentry->d_inode->i_sb->s_id, 13051da177e4SLinus Torvalds (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 13061da177e4SLinus Torvalds req->wb_bytes, 13071da177e4SLinus Torvalds (long long)req_offset(req)); 13081da177e4SLinus Torvalds if (task->tk_status < 0) { 13091da177e4SLinus Torvalds req->wb_context->error = task->tk_status; 13101da177e4SLinus Torvalds nfs_inode_remove_request(req); 13111da177e4SLinus Torvalds dprintk(", error = %d\n", task->tk_status); 13121da177e4SLinus Torvalds goto next; 13131da177e4SLinus Torvalds } 13141da177e4SLinus Torvalds 13151da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 13161da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 13171da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 13181da177e4SLinus Torvalds /* We have a match */ 13191da177e4SLinus Torvalds nfs_inode_remove_request(req); 13201da177e4SLinus Torvalds dprintk(" OK\n"); 13211da177e4SLinus Torvalds goto next; 13221da177e4SLinus Torvalds } 13231da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 13241da177e4SLinus Torvalds dprintk(" mismatch\n"); 13251da177e4SLinus Torvalds nfs_mark_request_dirty(req); 13261da177e4SLinus Torvalds next: 1327c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 13281da177e4SLinus Torvalds res++; 13291da177e4SLinus Torvalds } 13301da177e4SLinus Torvalds sub_page_state(nr_unstable,res); 13311da177e4SLinus Torvalds } 13321da177e4SLinus Torvalds #endif 13331da177e4SLinus Torvalds 13341da177e4SLinus Torvalds static int nfs_flush_inode(struct inode *inode, unsigned long idx_start, 13351da177e4SLinus Torvalds unsigned int npages, int how) 13361da177e4SLinus Torvalds { 13371da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 13381da177e4SLinus Torvalds LIST_HEAD(head); 13391da177e4SLinus Torvalds int res, 13401da177e4SLinus Torvalds error = 0; 13411da177e4SLinus Torvalds 13421da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 13431da177e4SLinus Torvalds res = nfs_scan_dirty(inode, &head, idx_start, npages); 13441da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 1345ab0a3dbeSTrond Myklebust if (res) { 1346ab0a3dbeSTrond Myklebust struct nfs_server *server = NFS_SERVER(inode); 1347ab0a3dbeSTrond Myklebust 1348ab0a3dbeSTrond Myklebust /* For single writes, FLUSH_STABLE is more efficient */ 1349ab0a3dbeSTrond Myklebust if (res == nfsi->npages && nfsi->npages <= server->wpages) { 1350ab0a3dbeSTrond Myklebust if (res > 1 || nfs_list_entry(head.next)->wb_bytes <= server->wsize) 1351ab0a3dbeSTrond Myklebust how |= FLUSH_STABLE; 1352ab0a3dbeSTrond Myklebust } 1353ab0a3dbeSTrond Myklebust error = nfs_flush_list(&head, server->wpages, how); 1354ab0a3dbeSTrond Myklebust } 13551da177e4SLinus Torvalds if (error < 0) 13561da177e4SLinus Torvalds return error; 13571da177e4SLinus Torvalds return res; 13581da177e4SLinus Torvalds } 13591da177e4SLinus Torvalds 13601da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 13613da28eb1STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 13621da177e4SLinus Torvalds { 13631da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 13641da177e4SLinus Torvalds LIST_HEAD(head); 13651da177e4SLinus Torvalds int res, 13661da177e4SLinus Torvalds error = 0; 13671da177e4SLinus Torvalds 13681da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 13693da28eb1STrond Myklebust res = nfs_scan_commit(inode, &head, 0, 0); 13703da28eb1STrond Myklebust spin_unlock(&nfsi->req_lock); 13711da177e4SLinus Torvalds if (res) { 137240859d7eSChuck Lever error = nfs_commit_list(inode, &head, how); 13731da177e4SLinus Torvalds if (error < 0) 13741da177e4SLinus Torvalds return error; 13753da28eb1STrond Myklebust } 13761da177e4SLinus Torvalds return res; 13771da177e4SLinus Torvalds } 13781da177e4SLinus Torvalds #endif 13791da177e4SLinus Torvalds 13801da177e4SLinus Torvalds int nfs_sync_inode(struct inode *inode, unsigned long idx_start, 13811da177e4SLinus Torvalds unsigned int npages, int how) 13821da177e4SLinus Torvalds { 138370b9ecbdSTrond Myklebust int nocommit = how & FLUSH_NOCOMMIT; 138470b9ecbdSTrond Myklebust int wait = how & FLUSH_WAIT; 138570b9ecbdSTrond Myklebust int error; 13861da177e4SLinus Torvalds 138770b9ecbdSTrond Myklebust how &= ~(FLUSH_WAIT|FLUSH_NOCOMMIT); 13881da177e4SLinus Torvalds 13891da177e4SLinus Torvalds do { 139070b9ecbdSTrond Myklebust if (wait) { 13911da177e4SLinus Torvalds error = nfs_wait_on_requests(inode, idx_start, npages); 139270b9ecbdSTrond Myklebust if (error != 0) 139370b9ecbdSTrond Myklebust continue; 139470b9ecbdSTrond Myklebust } 13951da177e4SLinus Torvalds error = nfs_flush_inode(inode, idx_start, npages, how); 139670b9ecbdSTrond Myklebust if (error != 0) 139770b9ecbdSTrond Myklebust continue; 139870b9ecbdSTrond Myklebust if (!nocommit) 13993da28eb1STrond Myklebust error = nfs_commit_inode(inode, how); 14001da177e4SLinus Torvalds } while (error > 0); 14011da177e4SLinus Torvalds return error; 14021da177e4SLinus Torvalds } 14031da177e4SLinus Torvalds 14041da177e4SLinus Torvalds int nfs_init_writepagecache(void) 14051da177e4SLinus Torvalds { 14061da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 14071da177e4SLinus Torvalds sizeof(struct nfs_write_data), 14081da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 14091da177e4SLinus Torvalds NULL, NULL); 14101da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 14111da177e4SLinus Torvalds return -ENOMEM; 14121da177e4SLinus Torvalds 14131da177e4SLinus Torvalds nfs_wdata_mempool = mempool_create(MIN_POOL_WRITE, 14141da177e4SLinus Torvalds mempool_alloc_slab, 14151da177e4SLinus Torvalds mempool_free_slab, 14161da177e4SLinus Torvalds nfs_wdata_cachep); 14171da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 14181da177e4SLinus Torvalds return -ENOMEM; 14191da177e4SLinus Torvalds 14201da177e4SLinus Torvalds nfs_commit_mempool = mempool_create(MIN_POOL_COMMIT, 14211da177e4SLinus Torvalds mempool_alloc_slab, 14221da177e4SLinus Torvalds mempool_free_slab, 14231da177e4SLinus Torvalds nfs_wdata_cachep); 14241da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 14251da177e4SLinus Torvalds return -ENOMEM; 14261da177e4SLinus Torvalds 14271da177e4SLinus Torvalds return 0; 14281da177e4SLinus Torvalds } 14291da177e4SLinus Torvalds 14301da177e4SLinus Torvalds void nfs_destroy_writepagecache(void) 14311da177e4SLinus Torvalds { 14321da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 14331da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 14341da177e4SLinus Torvalds if (kmem_cache_destroy(nfs_wdata_cachep)) 14351da177e4SLinus Torvalds printk(KERN_INFO "nfs_write_data: not all structures were freed\n"); 14361da177e4SLinus Torvalds } 14371da177e4SLinus Torvalds 1438