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/types.h> 501da177e4SLinus Torvalds #include <linux/slab.h> 511da177e4SLinus Torvalds #include <linux/mm.h> 521da177e4SLinus Torvalds #include <linux/pagemap.h> 531da177e4SLinus Torvalds #include <linux/file.h> 541da177e4SLinus Torvalds #include <linux/mpage.h> 551da177e4SLinus Torvalds #include <linux/writeback.h> 561da177e4SLinus Torvalds 571da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h> 581da177e4SLinus Torvalds #include <linux/nfs_fs.h> 591da177e4SLinus Torvalds #include <linux/nfs_mount.h> 601da177e4SLinus Torvalds #include <linux/nfs_page.h> 611da177e4SLinus Torvalds #include <asm/uaccess.h> 621da177e4SLinus Torvalds #include <linux/smp_lock.h> 631da177e4SLinus Torvalds 641da177e4SLinus Torvalds #include "delegation.h" 6591d5b470SChuck Lever #include "iostat.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 int nfs_wait_on_write_congestion(struct address_space *, int); 801da177e4SLinus Torvalds static int nfs_wait_on_requests(struct inode *, unsigned long, unsigned int); 811da177e4SLinus Torvalds static int nfs_flush_inode(struct inode *inode, unsigned long idx_start, 821da177e4SLinus Torvalds unsigned int npages, int how); 83788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops; 84788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops; 85788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 861da177e4SLinus Torvalds 871da177e4SLinus Torvalds static kmem_cache_t *nfs_wdata_cachep; 883feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 891da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 901da177e4SLinus Torvalds 911da177e4SLinus Torvalds static DECLARE_WAIT_QUEUE_HEAD(nfs_write_congestion); 921da177e4SLinus Torvalds 93e9f7bee1STrond Myklebust struct nfs_write_data *nfs_commit_alloc(void) 941da177e4SLinus Torvalds { 951da177e4SLinus Torvalds struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, SLAB_NOFS); 9640859d7eSChuck Lever 971da177e4SLinus Torvalds if (p) { 981da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 991da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 1001da177e4SLinus Torvalds } 1011da177e4SLinus Torvalds return p; 1021da177e4SLinus Torvalds } 1031da177e4SLinus Torvalds 104e17b1fc4STrond Myklebust void nfs_commit_free(struct nfs_write_data *p) 1051da177e4SLinus Torvalds { 10640859d7eSChuck Lever if (p && (p->pagevec != &p->page_array[0])) 10740859d7eSChuck Lever kfree(p->pagevec); 1081da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 1091da177e4SLinus Torvalds } 1101da177e4SLinus Torvalds 111e9f7bee1STrond Myklebust struct nfs_write_data *nfs_writedata_alloc(size_t len) 1123feb2d49STrond Myklebust { 113e9f7bee1STrond Myklebust unsigned int pagecount = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1143feb2d49STrond Myklebust struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, SLAB_NOFS); 1153feb2d49STrond Myklebust 1163feb2d49STrond Myklebust if (p) { 1173feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 1183feb2d49STrond Myklebust INIT_LIST_HEAD(&p->pages); 119e9f7bee1STrond Myklebust p->npages = pagecount; 1200d0b5cb3SChuck Lever if (pagecount <= ARRAY_SIZE(p->page_array)) 1210d0b5cb3SChuck Lever p->pagevec = p->page_array; 1223feb2d49STrond Myklebust else { 1230d0b5cb3SChuck Lever p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); 1240d0b5cb3SChuck Lever if (!p->pagevec) { 1253feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 1263feb2d49STrond Myklebust p = NULL; 1273feb2d49STrond Myklebust } 1283feb2d49STrond Myklebust } 1293feb2d49STrond Myklebust } 1303feb2d49STrond Myklebust return p; 1313feb2d49STrond Myklebust } 1323feb2d49STrond Myklebust 133e4e20512SAdrian Bunk static void nfs_writedata_free(struct nfs_write_data *p) 1343feb2d49STrond Myklebust { 1353feb2d49STrond Myklebust if (p && (p->pagevec != &p->page_array[0])) 1363feb2d49STrond Myklebust kfree(p->pagevec); 1373feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 1383feb2d49STrond Myklebust } 1393feb2d49STrond Myklebust 140963d8fe5STrond Myklebust void nfs_writedata_release(void *wdata) 1411da177e4SLinus Torvalds { 1421da177e4SLinus Torvalds nfs_writedata_free(wdata); 1431da177e4SLinus Torvalds } 1441da177e4SLinus Torvalds 1451da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1461da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1471da177e4SLinus Torvalds { 1481da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 1491da177e4SLinus Torvalds loff_t end, i_size = i_size_read(inode); 1501da177e4SLinus Torvalds unsigned long end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 1511da177e4SLinus Torvalds 1521da177e4SLinus Torvalds if (i_size > 0 && page->index < end_index) 1531da177e4SLinus Torvalds return; 1541da177e4SLinus Torvalds end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); 1551da177e4SLinus Torvalds if (i_size >= end) 1561da177e4SLinus Torvalds return; 15791d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 1581da177e4SLinus Torvalds i_size_write(inode, end); 1591da177e4SLinus Torvalds } 1601da177e4SLinus Torvalds 1611da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 1621da177e4SLinus Torvalds * covers the full page. 1631da177e4SLinus Torvalds */ 1641da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 1651da177e4SLinus Torvalds { 1661da177e4SLinus Torvalds loff_t end_offs; 1671da177e4SLinus Torvalds 1681da177e4SLinus Torvalds if (PageUptodate(page)) 1691da177e4SLinus Torvalds return; 1701da177e4SLinus Torvalds if (base != 0) 1711da177e4SLinus Torvalds return; 1721da177e4SLinus Torvalds if (count == PAGE_CACHE_SIZE) { 1731da177e4SLinus Torvalds SetPageUptodate(page); 1741da177e4SLinus Torvalds return; 1751da177e4SLinus Torvalds } 1761da177e4SLinus Torvalds 1771da177e4SLinus Torvalds end_offs = i_size_read(page->mapping->host) - 1; 1781da177e4SLinus Torvalds if (end_offs < 0) 1791da177e4SLinus Torvalds return; 1801da177e4SLinus Torvalds /* Is this the last page? */ 1811da177e4SLinus Torvalds if (page->index != (unsigned long)(end_offs >> PAGE_CACHE_SHIFT)) 1821da177e4SLinus Torvalds return; 1831da177e4SLinus Torvalds /* This is the last page: set PG_uptodate if we cover the entire 1841da177e4SLinus Torvalds * extent of the data, then zero the rest of the page. 1851da177e4SLinus Torvalds */ 1861da177e4SLinus Torvalds if (count == (unsigned int)(end_offs & (PAGE_CACHE_SIZE - 1)) + 1) { 1871da177e4SLinus Torvalds memclear_highpage_flush(page, count, PAGE_CACHE_SIZE - count); 1881da177e4SLinus Torvalds SetPageUptodate(page); 1891da177e4SLinus Torvalds } 1901da177e4SLinus Torvalds } 1911da177e4SLinus Torvalds 1921da177e4SLinus Torvalds /* 1931da177e4SLinus Torvalds * Write a page synchronously. 1941da177e4SLinus Torvalds * Offset is the data offset within the page. 1951da177e4SLinus Torvalds */ 1961da177e4SLinus Torvalds static int nfs_writepage_sync(struct nfs_open_context *ctx, struct inode *inode, 1971da177e4SLinus Torvalds struct page *page, unsigned int offset, unsigned int count, 1981da177e4SLinus Torvalds int how) 1991da177e4SLinus Torvalds { 2001da177e4SLinus Torvalds unsigned int wsize = NFS_SERVER(inode)->wsize; 2011da177e4SLinus Torvalds int result, written = 0; 2021da177e4SLinus Torvalds struct nfs_write_data *wdata; 2031da177e4SLinus Torvalds 204e9f7bee1STrond Myklebust wdata = nfs_writedata_alloc(wsize); 2051da177e4SLinus Torvalds if (!wdata) 2061da177e4SLinus Torvalds return -ENOMEM; 2071da177e4SLinus Torvalds 2081da177e4SLinus Torvalds wdata->flags = how; 2091da177e4SLinus Torvalds wdata->cred = ctx->cred; 2101da177e4SLinus Torvalds wdata->inode = inode; 2111da177e4SLinus Torvalds wdata->args.fh = NFS_FH(inode); 2121da177e4SLinus Torvalds wdata->args.context = ctx; 2131da177e4SLinus Torvalds wdata->args.pages = &page; 2141da177e4SLinus Torvalds wdata->args.stable = NFS_FILE_SYNC; 2151da177e4SLinus Torvalds wdata->args.pgbase = offset; 2161da177e4SLinus Torvalds wdata->args.count = wsize; 2171da177e4SLinus Torvalds wdata->res.fattr = &wdata->fattr; 2181da177e4SLinus Torvalds wdata->res.verf = &wdata->verf; 2191da177e4SLinus Torvalds 2201da177e4SLinus Torvalds dprintk("NFS: nfs_writepage_sync(%s/%Ld %d@%Ld)\n", 2211da177e4SLinus Torvalds inode->i_sb->s_id, 2221da177e4SLinus Torvalds (long long)NFS_FILEID(inode), 2231da177e4SLinus Torvalds count, (long long)(page_offset(page) + offset)); 2241da177e4SLinus Torvalds 225bb713d6dSTrond Myklebust set_page_writeback(page); 2261da177e4SLinus Torvalds nfs_begin_data_update(inode); 2271da177e4SLinus Torvalds do { 2281da177e4SLinus Torvalds if (count < wsize) 2291da177e4SLinus Torvalds wdata->args.count = count; 2301da177e4SLinus Torvalds wdata->args.offset = page_offset(page) + wdata->args.pgbase; 2311da177e4SLinus Torvalds 2321da177e4SLinus Torvalds result = NFS_PROTO(inode)->write(wdata); 2331da177e4SLinus Torvalds 2341da177e4SLinus Torvalds if (result < 0) { 2351da177e4SLinus Torvalds /* Must mark the page invalid after I/O error */ 2361da177e4SLinus Torvalds ClearPageUptodate(page); 2371da177e4SLinus Torvalds goto io_error; 2381da177e4SLinus Torvalds } 2391da177e4SLinus Torvalds if (result < wdata->args.count) 2401da177e4SLinus Torvalds printk(KERN_WARNING "NFS: short write, count=%u, result=%d\n", 2411da177e4SLinus Torvalds wdata->args.count, result); 2421da177e4SLinus Torvalds 2431da177e4SLinus Torvalds wdata->args.offset += result; 2441da177e4SLinus Torvalds wdata->args.pgbase += result; 2451da177e4SLinus Torvalds written += result; 2461da177e4SLinus Torvalds count -= result; 24791d5b470SChuck Lever nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, result); 2481da177e4SLinus Torvalds } while (count); 2491da177e4SLinus Torvalds /* Update file length */ 2501da177e4SLinus Torvalds nfs_grow_file(page, offset, written); 2511da177e4SLinus Torvalds /* Set the PG_uptodate flag? */ 2521da177e4SLinus Torvalds nfs_mark_uptodate(page, offset, written); 2531da177e4SLinus Torvalds 2541da177e4SLinus Torvalds if (PageError(page)) 2551da177e4SLinus Torvalds ClearPageError(page); 2561da177e4SLinus Torvalds 2571da177e4SLinus Torvalds io_error: 258951a143bSTrond Myklebust nfs_end_data_update(inode); 259bb713d6dSTrond Myklebust end_page_writeback(page); 2601da177e4SLinus Torvalds nfs_writedata_free(wdata); 2611da177e4SLinus Torvalds return written ? written : result; 2621da177e4SLinus Torvalds } 2631da177e4SLinus Torvalds 2641da177e4SLinus Torvalds static int nfs_writepage_async(struct nfs_open_context *ctx, 2651da177e4SLinus Torvalds struct inode *inode, struct page *page, 2661da177e4SLinus Torvalds unsigned int offset, unsigned int count) 2671da177e4SLinus Torvalds { 2681da177e4SLinus Torvalds struct nfs_page *req; 2691da177e4SLinus Torvalds 2701da177e4SLinus Torvalds req = nfs_update_request(ctx, inode, page, offset, count); 271abd3e641STrond Myklebust if (IS_ERR(req)) 272abd3e641STrond Myklebust return PTR_ERR(req); 2731da177e4SLinus Torvalds /* Update file length */ 2741da177e4SLinus Torvalds nfs_grow_file(page, offset, count); 2751da177e4SLinus Torvalds /* Set the PG_uptodate flag? */ 2761da177e4SLinus Torvalds nfs_mark_uptodate(page, offset, count); 2771da177e4SLinus Torvalds nfs_unlock_request(req); 278abd3e641STrond Myklebust return 0; 2791da177e4SLinus Torvalds } 2801da177e4SLinus Torvalds 2811da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 2821da177e4SLinus Torvalds { 2831da177e4SLinus Torvalds if (wbc->for_reclaim) 2841da177e4SLinus Torvalds return FLUSH_HIGHPRI; 2851da177e4SLinus Torvalds if (wbc->for_kupdate) 2861da177e4SLinus Torvalds return FLUSH_LOWPRI; 2871da177e4SLinus Torvalds return 0; 2881da177e4SLinus Torvalds } 2891da177e4SLinus Torvalds 2901da177e4SLinus Torvalds /* 2911da177e4SLinus Torvalds * Write an mmapped page to the server. 2921da177e4SLinus Torvalds */ 2931da177e4SLinus Torvalds int nfs_writepage(struct page *page, struct writeback_control *wbc) 2941da177e4SLinus Torvalds { 2951da177e4SLinus Torvalds struct nfs_open_context *ctx; 2961da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 2971da177e4SLinus Torvalds unsigned long end_index; 2981da177e4SLinus Torvalds unsigned offset = PAGE_CACHE_SIZE; 2991da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 3001da177e4SLinus Torvalds int inode_referenced = 0; 3011da177e4SLinus Torvalds int priority = wb_priority(wbc); 3021da177e4SLinus Torvalds int err; 3031da177e4SLinus Torvalds 30491d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 30591d5b470SChuck Lever nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 30691d5b470SChuck Lever 3071da177e4SLinus Torvalds /* 3081da177e4SLinus Torvalds * Note: We need to ensure that we have a reference to the inode 3091da177e4SLinus Torvalds * if we are to do asynchronous writes. If not, waiting 3101da177e4SLinus Torvalds * in nfs_wait_on_request() may deadlock with clear_inode(). 3111da177e4SLinus Torvalds * 3121da177e4SLinus Torvalds * If igrab() fails here, then it is in any case safe to 3131da177e4SLinus Torvalds * call nfs_wb_page(), since there will be no pending writes. 3141da177e4SLinus Torvalds */ 3151da177e4SLinus Torvalds if (igrab(inode) != 0) 3161da177e4SLinus Torvalds inode_referenced = 1; 3171da177e4SLinus Torvalds end_index = i_size >> PAGE_CACHE_SHIFT; 3181da177e4SLinus Torvalds 3191da177e4SLinus Torvalds /* Ensure we've flushed out any previous writes */ 3201da177e4SLinus Torvalds nfs_wb_page_priority(inode, page, priority); 3211da177e4SLinus Torvalds 3221da177e4SLinus Torvalds /* easy case */ 3231da177e4SLinus Torvalds if (page->index < end_index) 3241da177e4SLinus Torvalds goto do_it; 3251da177e4SLinus Torvalds /* things got complicated... */ 3261da177e4SLinus Torvalds offset = i_size & (PAGE_CACHE_SIZE-1); 3271da177e4SLinus Torvalds 3281da177e4SLinus Torvalds /* OK, are we completely out? */ 3291da177e4SLinus Torvalds err = 0; /* potential race with truncate - ignore */ 3301da177e4SLinus Torvalds if (page->index >= end_index+1 || !offset) 3311da177e4SLinus Torvalds goto out; 3321da177e4SLinus Torvalds do_it: 333d530838bSTrond Myklebust ctx = nfs_find_open_context(inode, NULL, FMODE_WRITE); 3341da177e4SLinus Torvalds if (ctx == NULL) { 3351da177e4SLinus Torvalds err = -EBADF; 3361da177e4SLinus Torvalds goto out; 3371da177e4SLinus Torvalds } 3381da177e4SLinus Torvalds lock_kernel(); 3391da177e4SLinus Torvalds if (!IS_SYNC(inode) && inode_referenced) { 3401da177e4SLinus Torvalds err = nfs_writepage_async(ctx, inode, page, 0, offset); 341abd3e641STrond Myklebust if (!wbc->for_writepages) 342abd3e641STrond Myklebust nfs_flush_inode(inode, 0, 0, wb_priority(wbc)); 3431da177e4SLinus Torvalds } else { 3441da177e4SLinus Torvalds err = nfs_writepage_sync(ctx, inode, page, 0, 3451da177e4SLinus Torvalds offset, priority); 3461da177e4SLinus Torvalds if (err >= 0) { 3471da177e4SLinus Torvalds if (err != offset) 3481da177e4SLinus Torvalds redirty_page_for_writepage(wbc, page); 3491da177e4SLinus Torvalds err = 0; 3501da177e4SLinus Torvalds } 3511da177e4SLinus Torvalds } 3521da177e4SLinus Torvalds unlock_kernel(); 3531da177e4SLinus Torvalds put_nfs_open_context(ctx); 3541da177e4SLinus Torvalds out: 3551da177e4SLinus Torvalds unlock_page(page); 3561da177e4SLinus Torvalds if (inode_referenced) 3571da177e4SLinus Torvalds iput(inode); 3581da177e4SLinus Torvalds return err; 3591da177e4SLinus Torvalds } 3601da177e4SLinus Torvalds 3611da177e4SLinus Torvalds /* 3621da177e4SLinus Torvalds * Note: causes nfs_update_request() to block on the assumption 3631da177e4SLinus Torvalds * that the writeback is generated due to memory pressure. 3641da177e4SLinus Torvalds */ 3651da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 3661da177e4SLinus Torvalds { 3671da177e4SLinus Torvalds struct backing_dev_info *bdi = mapping->backing_dev_info; 3681da177e4SLinus Torvalds struct inode *inode = mapping->host; 3691da177e4SLinus Torvalds int err; 3701da177e4SLinus Torvalds 37191d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 37291d5b470SChuck Lever 3731da177e4SLinus Torvalds err = generic_writepages(mapping, wbc); 3741da177e4SLinus Torvalds if (err) 3751da177e4SLinus Torvalds return err; 3761da177e4SLinus Torvalds while (test_and_set_bit(BDI_write_congested, &bdi->state) != 0) { 3771da177e4SLinus Torvalds if (wbc->nonblocking) 3781da177e4SLinus Torvalds return 0; 3791da177e4SLinus Torvalds nfs_wait_on_write_congestion(mapping, 0); 3801da177e4SLinus Torvalds } 3811da177e4SLinus Torvalds err = nfs_flush_inode(inode, 0, 0, wb_priority(wbc)); 3821da177e4SLinus Torvalds if (err < 0) 3831da177e4SLinus Torvalds goto out; 38491d5b470SChuck Lever nfs_add_stats(inode, NFSIOS_WRITEPAGES, err); 3851da177e4SLinus Torvalds wbc->nr_to_write -= err; 3861da177e4SLinus Torvalds if (!wbc->nonblocking && wbc->sync_mode == WB_SYNC_ALL) { 3871da177e4SLinus Torvalds err = nfs_wait_on_requests(inode, 0, 0); 3881da177e4SLinus Torvalds if (err < 0) 3891da177e4SLinus Torvalds goto out; 3901da177e4SLinus Torvalds } 3913da28eb1STrond Myklebust err = nfs_commit_inode(inode, wb_priority(wbc)); 3921da177e4SLinus Torvalds if (err > 0) { 3931da177e4SLinus Torvalds wbc->nr_to_write -= err; 3941da177e4SLinus Torvalds err = 0; 3951da177e4SLinus Torvalds } 3961da177e4SLinus Torvalds out: 3971da177e4SLinus Torvalds clear_bit(BDI_write_congested, &bdi->state); 3981da177e4SLinus Torvalds wake_up_all(&nfs_write_congestion); 399*275a082fSTrond Myklebust writeback_congestion_end(); 4001da177e4SLinus Torvalds return err; 4011da177e4SLinus Torvalds } 4021da177e4SLinus Torvalds 4031da177e4SLinus Torvalds /* 4041da177e4SLinus Torvalds * Insert a write request into an inode 4051da177e4SLinus Torvalds */ 4061da177e4SLinus Torvalds static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 4071da177e4SLinus Torvalds { 4081da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4091da177e4SLinus Torvalds int error; 4101da177e4SLinus Torvalds 4111da177e4SLinus Torvalds error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); 4121da177e4SLinus Torvalds BUG_ON(error == -EEXIST); 4131da177e4SLinus Torvalds if (error) 4141da177e4SLinus Torvalds return error; 4151da177e4SLinus Torvalds if (!nfsi->npages) { 4161da177e4SLinus Torvalds igrab(inode); 4171da177e4SLinus Torvalds nfs_begin_data_update(inode); 4181da177e4SLinus Torvalds if (nfs_have_delegation(inode, FMODE_WRITE)) 4191da177e4SLinus Torvalds nfsi->change_attr++; 4201da177e4SLinus Torvalds } 421deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 4221da177e4SLinus Torvalds nfsi->npages++; 4231da177e4SLinus Torvalds atomic_inc(&req->wb_count); 4241da177e4SLinus Torvalds return 0; 4251da177e4SLinus Torvalds } 4261da177e4SLinus Torvalds 4271da177e4SLinus Torvalds /* 4281da177e4SLinus Torvalds * Insert a write request into an inode 4291da177e4SLinus Torvalds */ 4301da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 4311da177e4SLinus Torvalds { 4321da177e4SLinus Torvalds struct inode *inode = req->wb_context->dentry->d_inode; 4331da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4341da177e4SLinus Torvalds 4351da177e4SLinus Torvalds BUG_ON (!NFS_WBACK_BUSY(req)); 4361da177e4SLinus Torvalds 4371da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 438deb7d638STrond Myklebust ClearPagePrivate(req->wb_page); 4391da177e4SLinus Torvalds radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); 4401da177e4SLinus Torvalds nfsi->npages--; 4411da177e4SLinus Torvalds if (!nfsi->npages) { 4421da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 443951a143bSTrond Myklebust nfs_end_data_update(inode); 4441da177e4SLinus Torvalds iput(inode); 4451da177e4SLinus Torvalds } else 4461da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 4471da177e4SLinus Torvalds nfs_clear_request(req); 4481da177e4SLinus Torvalds nfs_release_request(req); 4491da177e4SLinus Torvalds } 4501da177e4SLinus Torvalds 4511da177e4SLinus Torvalds /* 4521da177e4SLinus Torvalds * Find a request 4531da177e4SLinus Torvalds */ 4541da177e4SLinus Torvalds static inline struct nfs_page * 4551da177e4SLinus Torvalds _nfs_find_request(struct inode *inode, unsigned long index) 4561da177e4SLinus Torvalds { 4571da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4581da177e4SLinus Torvalds struct nfs_page *req; 4591da177e4SLinus Torvalds 4601da177e4SLinus Torvalds req = (struct nfs_page*)radix_tree_lookup(&nfsi->nfs_page_tree, index); 4611da177e4SLinus Torvalds if (req) 4621da177e4SLinus Torvalds atomic_inc(&req->wb_count); 4631da177e4SLinus Torvalds return req; 4641da177e4SLinus Torvalds } 4651da177e4SLinus Torvalds 4661da177e4SLinus Torvalds static struct nfs_page * 4671da177e4SLinus Torvalds nfs_find_request(struct inode *inode, unsigned long index) 4681da177e4SLinus Torvalds { 4691da177e4SLinus Torvalds struct nfs_page *req; 4701da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4711da177e4SLinus Torvalds 4721da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 4731da177e4SLinus Torvalds req = _nfs_find_request(inode, index); 4741da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 4751da177e4SLinus Torvalds return req; 4761da177e4SLinus Torvalds } 4771da177e4SLinus Torvalds 4781da177e4SLinus Torvalds /* 4791da177e4SLinus Torvalds * Add a request to the inode's dirty list. 4801da177e4SLinus Torvalds */ 4811da177e4SLinus Torvalds static void 4821da177e4SLinus Torvalds nfs_mark_request_dirty(struct nfs_page *req) 4831da177e4SLinus Torvalds { 4841da177e4SLinus Torvalds struct inode *inode = req->wb_context->dentry->d_inode; 4851da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4861da177e4SLinus Torvalds 4871da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 4883da28eb1STrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, 4893da28eb1STrond Myklebust req->wb_index, NFS_PAGE_TAG_DIRTY); 4901da177e4SLinus Torvalds nfs_list_add_request(req, &nfsi->dirty); 4911da177e4SLinus Torvalds nfsi->ndirty++; 4921da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 493b1e7a8fdSChristoph Lameter inc_zone_page_state(req->wb_page, NR_FILE_DIRTY); 4941da177e4SLinus Torvalds mark_inode_dirty(inode); 4951da177e4SLinus Torvalds } 4961da177e4SLinus Torvalds 4971da177e4SLinus Torvalds /* 4981da177e4SLinus Torvalds * Check if a request is dirty 4991da177e4SLinus Torvalds */ 5001da177e4SLinus Torvalds static inline int 5011da177e4SLinus Torvalds nfs_dirty_request(struct nfs_page *req) 5021da177e4SLinus Torvalds { 5031da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode); 5041da177e4SLinus Torvalds return !list_empty(&req->wb_list) && req->wb_list_head == &nfsi->dirty; 5051da177e4SLinus Torvalds } 5061da177e4SLinus Torvalds 5071da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 5081da177e4SLinus Torvalds /* 5091da177e4SLinus Torvalds * Add a request to the inode's commit list. 5101da177e4SLinus Torvalds */ 5111da177e4SLinus Torvalds static void 5121da177e4SLinus Torvalds nfs_mark_request_commit(struct nfs_page *req) 5131da177e4SLinus Torvalds { 5141da177e4SLinus Torvalds struct inode *inode = req->wb_context->dentry->d_inode; 5151da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5161da177e4SLinus Torvalds 5171da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 5181da177e4SLinus Torvalds nfs_list_add_request(req, &nfsi->commit); 5191da177e4SLinus Torvalds nfsi->ncommit++; 5201da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 521fd39fc85SChristoph Lameter inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 5221da177e4SLinus Torvalds mark_inode_dirty(inode); 5231da177e4SLinus Torvalds } 5241da177e4SLinus Torvalds #endif 5251da177e4SLinus Torvalds 5261da177e4SLinus Torvalds /* 5271da177e4SLinus Torvalds * Wait for a request to complete. 5281da177e4SLinus Torvalds * 5291da177e4SLinus Torvalds * Interruptible by signals only if mounted with intr flag. 5301da177e4SLinus Torvalds */ 531c42de9ddSTrond Myklebust static int nfs_wait_on_requests_locked(struct inode *inode, unsigned long idx_start, unsigned int npages) 5321da177e4SLinus Torvalds { 5331da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5341da177e4SLinus Torvalds struct nfs_page *req; 5351da177e4SLinus Torvalds unsigned long idx_end, next; 5361da177e4SLinus Torvalds unsigned int res = 0; 5371da177e4SLinus Torvalds int error; 5381da177e4SLinus Torvalds 5391da177e4SLinus Torvalds if (npages == 0) 5401da177e4SLinus Torvalds idx_end = ~0; 5411da177e4SLinus Torvalds else 5421da177e4SLinus Torvalds idx_end = idx_start + npages - 1; 5431da177e4SLinus Torvalds 5441da177e4SLinus Torvalds next = idx_start; 545c6a556b8STrond Myklebust while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_WRITEBACK)) { 5461da177e4SLinus Torvalds if (req->wb_index > idx_end) 5471da177e4SLinus Torvalds break; 5481da177e4SLinus Torvalds 5491da177e4SLinus Torvalds next = req->wb_index + 1; 550c6a556b8STrond Myklebust BUG_ON(!NFS_WBACK_BUSY(req)); 5511da177e4SLinus Torvalds 5521da177e4SLinus Torvalds atomic_inc(&req->wb_count); 5531da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 5541da177e4SLinus Torvalds error = nfs_wait_on_request(req); 5551da177e4SLinus Torvalds nfs_release_request(req); 556c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 5571da177e4SLinus Torvalds if (error < 0) 5581da177e4SLinus Torvalds return error; 5591da177e4SLinus Torvalds res++; 5601da177e4SLinus Torvalds } 5611da177e4SLinus Torvalds return res; 5621da177e4SLinus Torvalds } 5631da177e4SLinus Torvalds 564c42de9ddSTrond Myklebust static int nfs_wait_on_requests(struct inode *inode, unsigned long idx_start, unsigned int npages) 565c42de9ddSTrond Myklebust { 566c42de9ddSTrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 567c42de9ddSTrond Myklebust int ret; 568c42de9ddSTrond Myklebust 569c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 570c42de9ddSTrond Myklebust ret = nfs_wait_on_requests_locked(inode, idx_start, npages); 571c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 572c42de9ddSTrond Myklebust return ret; 573c42de9ddSTrond Myklebust } 574c42de9ddSTrond Myklebust 57583715ad5STrond Myklebust static void nfs_cancel_dirty_list(struct list_head *head) 576d2ccddf0STrond Myklebust { 577d2ccddf0STrond Myklebust struct nfs_page *req; 578d2ccddf0STrond Myklebust while(!list_empty(head)) { 579d2ccddf0STrond Myklebust req = nfs_list_entry(head->next); 580d2ccddf0STrond Myklebust nfs_list_remove_request(req); 581d2ccddf0STrond Myklebust nfs_inode_remove_request(req); 582d2ccddf0STrond Myklebust nfs_clear_page_writeback(req); 583d2ccddf0STrond Myklebust } 584d2ccddf0STrond Myklebust } 585d2ccddf0STrond Myklebust 58683715ad5STrond Myklebust static void nfs_cancel_commit_list(struct list_head *head) 58783715ad5STrond Myklebust { 58883715ad5STrond Myklebust struct nfs_page *req; 58983715ad5STrond Myklebust 59083715ad5STrond Myklebust while(!list_empty(head)) { 59183715ad5STrond Myklebust req = nfs_list_entry(head->next); 59283715ad5STrond Myklebust nfs_list_remove_request(req); 59383715ad5STrond Myklebust nfs_inode_remove_request(req); 59483715ad5STrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 5955c2d97cbSTrond Myklebust nfs_clear_page_writeback(req); 59683715ad5STrond Myklebust } 59783715ad5STrond Myklebust } 59883715ad5STrond Myklebust 5991da177e4SLinus Torvalds /* 6001da177e4SLinus Torvalds * nfs_scan_dirty - Scan an inode for dirty requests 6011da177e4SLinus Torvalds * @inode: NFS inode to scan 6021da177e4SLinus Torvalds * @dst: destination list 6031da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 6041da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 6051da177e4SLinus Torvalds * 6061da177e4SLinus Torvalds * Moves requests from the inode's dirty page list. 6071da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 6081da177e4SLinus Torvalds */ 6091da177e4SLinus Torvalds static int 6101da177e4SLinus Torvalds nfs_scan_dirty(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) 6111da177e4SLinus Torvalds { 6121da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 6133da28eb1STrond Myklebust int res = 0; 6143da28eb1STrond Myklebust 6153da28eb1STrond Myklebust if (nfsi->ndirty != 0) { 6163da28eb1STrond Myklebust res = nfs_scan_lock_dirty(nfsi, dst, idx_start, npages); 6171da177e4SLinus Torvalds nfsi->ndirty -= res; 6181da177e4SLinus Torvalds if ((nfsi->ndirty == 0) != list_empty(&nfsi->dirty)) 6191da177e4SLinus Torvalds printk(KERN_ERR "NFS: desynchronized value of nfs_i.ndirty.\n"); 6203da28eb1STrond Myklebust } 6211da177e4SLinus Torvalds return res; 6221da177e4SLinus Torvalds } 6231da177e4SLinus Torvalds 6241da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 6251da177e4SLinus Torvalds /* 6261da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 6271da177e4SLinus Torvalds * @inode: NFS inode to scan 6281da177e4SLinus Torvalds * @dst: destination list 6291da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 6301da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 6311da177e4SLinus Torvalds * 6321da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 6331da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 6341da177e4SLinus Torvalds */ 6351da177e4SLinus Torvalds static int 6361da177e4SLinus Torvalds nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) 6371da177e4SLinus Torvalds { 6381da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 6393da28eb1STrond Myklebust int res = 0; 6403da28eb1STrond Myklebust 6413da28eb1STrond Myklebust if (nfsi->ncommit != 0) { 642d2ccddf0STrond Myklebust res = nfs_scan_list(nfsi, &nfsi->commit, dst, idx_start, npages); 6431da177e4SLinus Torvalds nfsi->ncommit -= res; 6441da177e4SLinus Torvalds if ((nfsi->ncommit == 0) != list_empty(&nfsi->commit)) 6451da177e4SLinus Torvalds printk(KERN_ERR "NFS: desynchronized value of nfs_i.ncommit.\n"); 6463da28eb1STrond Myklebust } 6471da177e4SLinus Torvalds return res; 6481da177e4SLinus Torvalds } 649c42de9ddSTrond Myklebust #else 650c42de9ddSTrond Myklebust static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) 651c42de9ddSTrond Myklebust { 652c42de9ddSTrond Myklebust return 0; 653c42de9ddSTrond Myklebust } 6541da177e4SLinus Torvalds #endif 6551da177e4SLinus Torvalds 6561da177e4SLinus Torvalds static int nfs_wait_on_write_congestion(struct address_space *mapping, int intr) 6571da177e4SLinus Torvalds { 6581da177e4SLinus Torvalds struct backing_dev_info *bdi = mapping->backing_dev_info; 6591da177e4SLinus Torvalds DEFINE_WAIT(wait); 6601da177e4SLinus Torvalds int ret = 0; 6611da177e4SLinus Torvalds 6621da177e4SLinus Torvalds might_sleep(); 6631da177e4SLinus Torvalds 6641da177e4SLinus Torvalds if (!bdi_write_congested(bdi)) 6651da177e4SLinus Torvalds return 0; 66691d5b470SChuck Lever 66791d5b470SChuck Lever nfs_inc_stats(mapping->host, NFSIOS_CONGESTIONWAIT); 66891d5b470SChuck Lever 6691da177e4SLinus Torvalds if (intr) { 6701da177e4SLinus Torvalds struct rpc_clnt *clnt = NFS_CLIENT(mapping->host); 6711da177e4SLinus Torvalds sigset_t oldset; 6721da177e4SLinus Torvalds 6731da177e4SLinus Torvalds rpc_clnt_sigmask(clnt, &oldset); 6741da177e4SLinus Torvalds prepare_to_wait(&nfs_write_congestion, &wait, TASK_INTERRUPTIBLE); 6751da177e4SLinus Torvalds if (bdi_write_congested(bdi)) { 6761da177e4SLinus Torvalds if (signalled()) 6771da177e4SLinus Torvalds ret = -ERESTARTSYS; 6781da177e4SLinus Torvalds else 6791da177e4SLinus Torvalds schedule(); 6801da177e4SLinus Torvalds } 6811da177e4SLinus Torvalds rpc_clnt_sigunmask(clnt, &oldset); 6821da177e4SLinus Torvalds } else { 6831da177e4SLinus Torvalds prepare_to_wait(&nfs_write_congestion, &wait, TASK_UNINTERRUPTIBLE); 6841da177e4SLinus Torvalds if (bdi_write_congested(bdi)) 6851da177e4SLinus Torvalds schedule(); 6861da177e4SLinus Torvalds } 6871da177e4SLinus Torvalds finish_wait(&nfs_write_congestion, &wait); 6881da177e4SLinus Torvalds return ret; 6891da177e4SLinus Torvalds } 6901da177e4SLinus Torvalds 6911da177e4SLinus Torvalds 6921da177e4SLinus Torvalds /* 6931da177e4SLinus Torvalds * Try to update any existing write request, or create one if there is none. 6941da177e4SLinus Torvalds * In order to match, the request's credentials must match those of 6951da177e4SLinus Torvalds * the calling process. 6961da177e4SLinus Torvalds * 6971da177e4SLinus Torvalds * Note: Should always be called with the Page Lock held! 6981da177e4SLinus Torvalds */ 6991da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx, 7001da177e4SLinus Torvalds struct inode *inode, struct page *page, 7011da177e4SLinus Torvalds unsigned int offset, unsigned int bytes) 7021da177e4SLinus Torvalds { 7031da177e4SLinus Torvalds struct nfs_server *server = NFS_SERVER(inode); 7041da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 7051da177e4SLinus Torvalds struct nfs_page *req, *new = NULL; 7061da177e4SLinus Torvalds unsigned long rqend, end; 7071da177e4SLinus Torvalds 7081da177e4SLinus Torvalds end = offset + bytes; 7091da177e4SLinus Torvalds 7101da177e4SLinus Torvalds if (nfs_wait_on_write_congestion(page->mapping, server->flags & NFS_MOUNT_INTR)) 7111da177e4SLinus Torvalds return ERR_PTR(-ERESTARTSYS); 7121da177e4SLinus Torvalds for (;;) { 7131da177e4SLinus Torvalds /* Loop over all inode entries and see if we find 7141da177e4SLinus Torvalds * A request for the page we wish to update 7151da177e4SLinus Torvalds */ 7161da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 7171da177e4SLinus Torvalds req = _nfs_find_request(inode, page->index); 7181da177e4SLinus Torvalds if (req) { 7191da177e4SLinus Torvalds if (!nfs_lock_request_dontget(req)) { 7201da177e4SLinus Torvalds int error; 7211da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 7221da177e4SLinus Torvalds error = nfs_wait_on_request(req); 7231da177e4SLinus Torvalds nfs_release_request(req); 7241dd594b2SNeil Brown if (error < 0) { 7251dd594b2SNeil Brown if (new) 7261dd594b2SNeil Brown nfs_release_request(new); 7271da177e4SLinus Torvalds return ERR_PTR(error); 7281dd594b2SNeil Brown } 7291da177e4SLinus Torvalds continue; 7301da177e4SLinus Torvalds } 7311da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 7321da177e4SLinus Torvalds if (new) 7331da177e4SLinus Torvalds nfs_release_request(new); 7341da177e4SLinus Torvalds break; 7351da177e4SLinus Torvalds } 7361da177e4SLinus Torvalds 7371da177e4SLinus Torvalds if (new) { 7381da177e4SLinus Torvalds int error; 7391da177e4SLinus Torvalds nfs_lock_request_dontget(new); 7401da177e4SLinus Torvalds error = nfs_inode_add_request(inode, new); 7411da177e4SLinus Torvalds if (error) { 7421da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 7431da177e4SLinus Torvalds nfs_unlock_request(new); 7441da177e4SLinus Torvalds return ERR_PTR(error); 7451da177e4SLinus Torvalds } 7461da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 7471da177e4SLinus Torvalds nfs_mark_request_dirty(new); 7481da177e4SLinus Torvalds return new; 7491da177e4SLinus Torvalds } 7501da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 7511da177e4SLinus Torvalds 7521da177e4SLinus Torvalds new = nfs_create_request(ctx, inode, page, offset, bytes); 7531da177e4SLinus Torvalds if (IS_ERR(new)) 7541da177e4SLinus Torvalds return new; 7551da177e4SLinus Torvalds } 7561da177e4SLinus Torvalds 7571da177e4SLinus Torvalds /* We have a request for our page. 7581da177e4SLinus Torvalds * If the creds don't match, or the 7591da177e4SLinus Torvalds * page addresses don't match, 7601da177e4SLinus Torvalds * tell the caller to wait on the conflicting 7611da177e4SLinus Torvalds * request. 7621da177e4SLinus Torvalds */ 7631da177e4SLinus Torvalds rqend = req->wb_offset + req->wb_bytes; 7641da177e4SLinus Torvalds if (req->wb_context != ctx 7651da177e4SLinus Torvalds || req->wb_page != page 7661da177e4SLinus Torvalds || !nfs_dirty_request(req) 7671da177e4SLinus Torvalds || offset > rqend || end < req->wb_offset) { 7681da177e4SLinus Torvalds nfs_unlock_request(req); 7691da177e4SLinus Torvalds return ERR_PTR(-EBUSY); 7701da177e4SLinus Torvalds } 7711da177e4SLinus Torvalds 7721da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 7731da177e4SLinus Torvalds if (offset < req->wb_offset) { 7741da177e4SLinus Torvalds req->wb_offset = offset; 7751da177e4SLinus Torvalds req->wb_pgbase = offset; 7761da177e4SLinus Torvalds req->wb_bytes = rqend - req->wb_offset; 7771da177e4SLinus Torvalds } 7781da177e4SLinus Torvalds 7791da177e4SLinus Torvalds if (end > rqend) 7801da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 7811da177e4SLinus Torvalds 7821da177e4SLinus Torvalds return req; 7831da177e4SLinus Torvalds } 7841da177e4SLinus Torvalds 7851da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 7861da177e4SLinus Torvalds { 7871da177e4SLinus Torvalds struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; 7881da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 7891da177e4SLinus Torvalds struct nfs_page *req; 7901da177e4SLinus Torvalds int status = 0; 7911da177e4SLinus Torvalds /* 7921da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 7931da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 7941da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 7951da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 7961da177e4SLinus Torvalds * Also do the same if we find a request from an existing 7971da177e4SLinus Torvalds * dropped page. 7981da177e4SLinus Torvalds */ 7991da177e4SLinus Torvalds req = nfs_find_request(inode, page->index); 8001da177e4SLinus Torvalds if (req) { 8011da177e4SLinus Torvalds if (req->wb_page != page || ctx != req->wb_context) 8021da177e4SLinus Torvalds status = nfs_wb_page(inode, page); 8031da177e4SLinus Torvalds nfs_release_request(req); 8041da177e4SLinus Torvalds } 8051da177e4SLinus Torvalds return (status < 0) ? status : 0; 8061da177e4SLinus Torvalds } 8071da177e4SLinus Torvalds 8081da177e4SLinus Torvalds /* 8091da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 8101da177e4SLinus Torvalds * 8111da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 8121da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 8131da177e4SLinus Torvalds */ 8141da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 8151da177e4SLinus Torvalds unsigned int offset, unsigned int count) 8161da177e4SLinus Torvalds { 8171da177e4SLinus Torvalds struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; 8181da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 8191da177e4SLinus Torvalds struct nfs_page *req; 8201da177e4SLinus Torvalds int status = 0; 8211da177e4SLinus Torvalds 82291d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 82391d5b470SChuck Lever 8241da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage(%s/%s %d@%Ld)\n", 8250bbacc40SChuck Lever file->f_dentry->d_parent->d_name.name, 8260bbacc40SChuck Lever file->f_dentry->d_name.name, count, 8270bbacc40SChuck Lever (long long)(page_offset(page) +offset)); 8281da177e4SLinus Torvalds 8291da177e4SLinus Torvalds if (IS_SYNC(inode)) { 8301da177e4SLinus Torvalds status = nfs_writepage_sync(ctx, inode, page, offset, count, 0); 8311da177e4SLinus Torvalds if (status > 0) { 8321da177e4SLinus Torvalds if (offset == 0 && status == PAGE_CACHE_SIZE) 8331da177e4SLinus Torvalds SetPageUptodate(page); 8341da177e4SLinus Torvalds return 0; 8351da177e4SLinus Torvalds } 8361da177e4SLinus Torvalds return status; 8371da177e4SLinus Torvalds } 8381da177e4SLinus Torvalds 8391da177e4SLinus Torvalds /* If we're not using byte range locks, and we know the page 8401da177e4SLinus Torvalds * is entirely in cache, it may be more efficient to avoid 8411da177e4SLinus Torvalds * fragmenting write requests. 8421da177e4SLinus Torvalds */ 843ab0a3dbeSTrond Myklebust if (PageUptodate(page) && inode->i_flock == NULL && !(file->f_mode & O_SYNC)) { 8441da177e4SLinus Torvalds loff_t end_offs = i_size_read(inode) - 1; 8451da177e4SLinus Torvalds unsigned long end_index = end_offs >> PAGE_CACHE_SHIFT; 8461da177e4SLinus Torvalds 8471da177e4SLinus Torvalds count += offset; 8481da177e4SLinus Torvalds offset = 0; 8491da177e4SLinus Torvalds if (unlikely(end_offs < 0)) { 8501da177e4SLinus Torvalds /* Do nothing */ 8511da177e4SLinus Torvalds } else if (page->index == end_index) { 8521da177e4SLinus Torvalds unsigned int pglen; 8531da177e4SLinus Torvalds pglen = (unsigned int)(end_offs & (PAGE_CACHE_SIZE-1)) + 1; 8541da177e4SLinus Torvalds if (count < pglen) 8551da177e4SLinus Torvalds count = pglen; 8561da177e4SLinus Torvalds } else if (page->index < end_index) 8571da177e4SLinus Torvalds count = PAGE_CACHE_SIZE; 8581da177e4SLinus Torvalds } 8591da177e4SLinus Torvalds 8601da177e4SLinus Torvalds /* 8611da177e4SLinus Torvalds * Try to find an NFS request corresponding to this page 8621da177e4SLinus Torvalds * and update it. 8631da177e4SLinus Torvalds * If the existing request cannot be updated, we must flush 8641da177e4SLinus Torvalds * it out now. 8651da177e4SLinus Torvalds */ 8661da177e4SLinus Torvalds do { 8671da177e4SLinus Torvalds req = nfs_update_request(ctx, inode, page, offset, count); 8681da177e4SLinus Torvalds status = (IS_ERR(req)) ? PTR_ERR(req) : 0; 8691da177e4SLinus Torvalds if (status != -EBUSY) 8701da177e4SLinus Torvalds break; 8711da177e4SLinus Torvalds /* Request could not be updated. Flush it out and try again */ 8721da177e4SLinus Torvalds status = nfs_wb_page(inode, page); 8731da177e4SLinus Torvalds } while (status >= 0); 8741da177e4SLinus Torvalds if (status < 0) 8751da177e4SLinus Torvalds goto done; 8761da177e4SLinus Torvalds 8771da177e4SLinus Torvalds status = 0; 8781da177e4SLinus Torvalds 8791da177e4SLinus Torvalds /* Update file length */ 8801da177e4SLinus Torvalds nfs_grow_file(page, offset, count); 8811da177e4SLinus Torvalds /* Set the PG_uptodate flag? */ 8821da177e4SLinus Torvalds nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 8831da177e4SLinus Torvalds nfs_unlock_request(req); 8841da177e4SLinus Torvalds done: 8851da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage returns %d (isize %Ld)\n", 8861da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 8871da177e4SLinus Torvalds if (status < 0) 8881da177e4SLinus Torvalds ClearPageUptodate(page); 8891da177e4SLinus Torvalds return status; 8901da177e4SLinus Torvalds } 8911da177e4SLinus Torvalds 8921da177e4SLinus Torvalds static void nfs_writepage_release(struct nfs_page *req) 8931da177e4SLinus Torvalds { 8941da177e4SLinus Torvalds end_page_writeback(req->wb_page); 8951da177e4SLinus Torvalds 8961da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 8971da177e4SLinus Torvalds if (!PageError(req->wb_page)) { 8981da177e4SLinus Torvalds if (NFS_NEED_RESCHED(req)) { 8991da177e4SLinus Torvalds nfs_mark_request_dirty(req); 9001da177e4SLinus Torvalds goto out; 9011da177e4SLinus Torvalds } else if (NFS_NEED_COMMIT(req)) { 9021da177e4SLinus Torvalds nfs_mark_request_commit(req); 9031da177e4SLinus Torvalds goto out; 9041da177e4SLinus Torvalds } 9051da177e4SLinus Torvalds } 9061da177e4SLinus Torvalds nfs_inode_remove_request(req); 9071da177e4SLinus Torvalds 9081da177e4SLinus Torvalds out: 9091da177e4SLinus Torvalds nfs_clear_commit(req); 9101da177e4SLinus Torvalds nfs_clear_reschedule(req); 9111da177e4SLinus Torvalds #else 9121da177e4SLinus Torvalds nfs_inode_remove_request(req); 9131da177e4SLinus Torvalds #endif 914c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 9151da177e4SLinus Torvalds } 9161da177e4SLinus Torvalds 9171da177e4SLinus Torvalds static inline int flush_task_priority(int how) 9181da177e4SLinus Torvalds { 9191da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 9201da177e4SLinus Torvalds case FLUSH_HIGHPRI: 9211da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 9221da177e4SLinus Torvalds case FLUSH_LOWPRI: 9231da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 9241da177e4SLinus Torvalds } 9251da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 9261da177e4SLinus Torvalds } 9271da177e4SLinus Torvalds 9281da177e4SLinus Torvalds /* 9291da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 9301da177e4SLinus Torvalds */ 9311da177e4SLinus Torvalds static void nfs_write_rpcsetup(struct nfs_page *req, 9321da177e4SLinus Torvalds struct nfs_write_data *data, 933788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 9341da177e4SLinus Torvalds unsigned int count, unsigned int offset, 9351da177e4SLinus Torvalds int how) 9361da177e4SLinus Torvalds { 9371da177e4SLinus Torvalds struct inode *inode; 938788e7a89STrond Myklebust int flags; 9391da177e4SLinus Torvalds 9401da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 9411da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 9421da177e4SLinus Torvalds 9431da177e4SLinus Torvalds data->req = req; 9441da177e4SLinus Torvalds data->inode = inode = req->wb_context->dentry->d_inode; 9451da177e4SLinus Torvalds data->cred = req->wb_context->cred; 9461da177e4SLinus Torvalds 9471da177e4SLinus Torvalds data->args.fh = NFS_FH(inode); 9481da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 9491da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 9501da177e4SLinus Torvalds data->args.pages = data->pagevec; 9511da177e4SLinus Torvalds data->args.count = count; 9521da177e4SLinus Torvalds data->args.context = req->wb_context; 9531da177e4SLinus Torvalds 9541da177e4SLinus Torvalds data->res.fattr = &data->fattr; 9551da177e4SLinus Torvalds data->res.count = count; 9561da177e4SLinus Torvalds data->res.verf = &data->verf; 9570e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 9581da177e4SLinus Torvalds 959788e7a89STrond Myklebust /* Set up the initial task struct. */ 960788e7a89STrond Myklebust flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 961788e7a89STrond Myklebust rpc_init_task(&data->task, NFS_CLIENT(inode), flags, call_ops, data); 9621da177e4SLinus Torvalds NFS_PROTO(inode)->write_setup(data, how); 9631da177e4SLinus Torvalds 9641da177e4SLinus Torvalds data->task.tk_priority = flush_task_priority(how); 9651da177e4SLinus Torvalds data->task.tk_cookie = (unsigned long)inode; 9661da177e4SLinus Torvalds 9671da177e4SLinus Torvalds dprintk("NFS: %4d initiated write call (req %s/%Ld, %u bytes @ offset %Lu)\n", 9680bbacc40SChuck Lever data->task.tk_pid, 9691da177e4SLinus Torvalds inode->i_sb->s_id, 9701da177e4SLinus Torvalds (long long)NFS_FILEID(inode), 9711da177e4SLinus Torvalds count, 9721da177e4SLinus Torvalds (unsigned long long)data->args.offset); 9731da177e4SLinus Torvalds } 9741da177e4SLinus Torvalds 9751da177e4SLinus Torvalds static void nfs_execute_write(struct nfs_write_data *data) 9761da177e4SLinus Torvalds { 9771da177e4SLinus Torvalds struct rpc_clnt *clnt = NFS_CLIENT(data->inode); 9781da177e4SLinus Torvalds sigset_t oldset; 9791da177e4SLinus Torvalds 9801da177e4SLinus Torvalds rpc_clnt_sigmask(clnt, &oldset); 9811da177e4SLinus Torvalds lock_kernel(); 9821da177e4SLinus Torvalds rpc_execute(&data->task); 9831da177e4SLinus Torvalds unlock_kernel(); 9841da177e4SLinus Torvalds rpc_clnt_sigunmask(clnt, &oldset); 9851da177e4SLinus Torvalds } 9861da177e4SLinus Torvalds 9871da177e4SLinus Torvalds /* 9881da177e4SLinus Torvalds * Generate multiple small requests to write out a single 9891da177e4SLinus Torvalds * contiguous dirty area on one page. 9901da177e4SLinus Torvalds */ 9917d46a49fSTrond Myklebust static int nfs_flush_multi(struct inode *inode, struct list_head *head, int how) 9921da177e4SLinus Torvalds { 9931da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 9941da177e4SLinus Torvalds struct page *page = req->wb_page; 9951da177e4SLinus Torvalds struct nfs_write_data *data; 996e9f7bee1STrond Myklebust size_t wsize = NFS_SERVER(inode)->wsize, nbytes; 997e9f7bee1STrond Myklebust unsigned int offset; 9981da177e4SLinus Torvalds int requests = 0; 9991da177e4SLinus Torvalds LIST_HEAD(list); 10001da177e4SLinus Torvalds 10011da177e4SLinus Torvalds nfs_list_remove_request(req); 10021da177e4SLinus Torvalds 10031da177e4SLinus Torvalds nbytes = req->wb_bytes; 1004e9f7bee1STrond Myklebust do { 1005e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 1006e9f7bee1STrond Myklebust 1007e9f7bee1STrond Myklebust data = nfs_writedata_alloc(len); 10081da177e4SLinus Torvalds if (!data) 10091da177e4SLinus Torvalds goto out_bad; 10101da177e4SLinus Torvalds list_add(&data->pages, &list); 10111da177e4SLinus Torvalds requests++; 1012e9f7bee1STrond Myklebust nbytes -= len; 1013e9f7bee1STrond Myklebust } while (nbytes != 0); 10141da177e4SLinus Torvalds atomic_set(&req->wb_complete, requests); 10151da177e4SLinus Torvalds 10161da177e4SLinus Torvalds ClearPageError(page); 1017bb713d6dSTrond Myklebust set_page_writeback(page); 10181da177e4SLinus Torvalds offset = 0; 10191da177e4SLinus Torvalds nbytes = req->wb_bytes; 10201da177e4SLinus Torvalds do { 10211da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 10221da177e4SLinus Torvalds list_del_init(&data->pages); 10231da177e4SLinus Torvalds 10241da177e4SLinus Torvalds data->pagevec[0] = page; 10251da177e4SLinus Torvalds 10261da177e4SLinus Torvalds if (nbytes > wsize) { 1027788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 1028788e7a89STrond Myklebust wsize, offset, how); 10291da177e4SLinus Torvalds offset += wsize; 10301da177e4SLinus Torvalds nbytes -= wsize; 10311da177e4SLinus Torvalds } else { 1032788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 1033788e7a89STrond Myklebust nbytes, offset, how); 10341da177e4SLinus Torvalds nbytes = 0; 10351da177e4SLinus Torvalds } 10361da177e4SLinus Torvalds nfs_execute_write(data); 10371da177e4SLinus Torvalds } while (nbytes != 0); 10381da177e4SLinus Torvalds 10391da177e4SLinus Torvalds return 0; 10401da177e4SLinus Torvalds 10411da177e4SLinus Torvalds out_bad: 10421da177e4SLinus Torvalds while (!list_empty(&list)) { 10431da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 10441da177e4SLinus Torvalds list_del(&data->pages); 10451da177e4SLinus Torvalds nfs_writedata_free(data); 10461da177e4SLinus Torvalds } 10471da177e4SLinus Torvalds nfs_mark_request_dirty(req); 1048c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 10491da177e4SLinus Torvalds return -ENOMEM; 10501da177e4SLinus Torvalds } 10511da177e4SLinus Torvalds 10521da177e4SLinus Torvalds /* 10531da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 10541da177e4SLinus Torvalds * The page must have been locked by the caller. 10551da177e4SLinus Torvalds * 10561da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 10571da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 10581da177e4SLinus Torvalds * that has been written but not committed. 10591da177e4SLinus Torvalds */ 10607d46a49fSTrond Myklebust static int nfs_flush_one(struct inode *inode, struct list_head *head, int how) 10611da177e4SLinus Torvalds { 10621da177e4SLinus Torvalds struct nfs_page *req; 10631da177e4SLinus Torvalds struct page **pages; 10641da177e4SLinus Torvalds struct nfs_write_data *data; 10651da177e4SLinus Torvalds unsigned int count; 10661da177e4SLinus Torvalds 1067e9f7bee1STrond Myklebust data = nfs_writedata_alloc(NFS_SERVER(inode)->wsize); 10681da177e4SLinus Torvalds if (!data) 10691da177e4SLinus Torvalds goto out_bad; 10701da177e4SLinus Torvalds 10711da177e4SLinus Torvalds pages = data->pagevec; 10721da177e4SLinus Torvalds count = 0; 10731da177e4SLinus Torvalds while (!list_empty(head)) { 10741da177e4SLinus Torvalds req = nfs_list_entry(head->next); 10751da177e4SLinus Torvalds nfs_list_remove_request(req); 10761da177e4SLinus Torvalds nfs_list_add_request(req, &data->pages); 10771da177e4SLinus Torvalds ClearPageError(req->wb_page); 1078bb713d6dSTrond Myklebust set_page_writeback(req->wb_page); 10791da177e4SLinus Torvalds *pages++ = req->wb_page; 10801da177e4SLinus Torvalds count += req->wb_bytes; 10811da177e4SLinus Torvalds } 10821da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 10831da177e4SLinus Torvalds 10841da177e4SLinus Torvalds /* Set up the argument struct */ 1085788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how); 10861da177e4SLinus Torvalds 10871da177e4SLinus Torvalds nfs_execute_write(data); 10881da177e4SLinus Torvalds return 0; 10891da177e4SLinus Torvalds out_bad: 10901da177e4SLinus Torvalds while (!list_empty(head)) { 10911da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 10921da177e4SLinus Torvalds nfs_list_remove_request(req); 10931da177e4SLinus Torvalds nfs_mark_request_dirty(req); 1094c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 10951da177e4SLinus Torvalds } 10961da177e4SLinus Torvalds return -ENOMEM; 10971da177e4SLinus Torvalds } 10981da177e4SLinus Torvalds 10997d46a49fSTrond Myklebust static int nfs_flush_list(struct inode *inode, struct list_head *head, int npages, int how) 11001da177e4SLinus Torvalds { 11011da177e4SLinus Torvalds LIST_HEAD(one_request); 11027d46a49fSTrond Myklebust int (*flush_one)(struct inode *, struct list_head *, int); 11031da177e4SLinus Torvalds struct nfs_page *req; 11047d46a49fSTrond Myklebust int wpages = NFS_SERVER(inode)->wpages; 11057d46a49fSTrond Myklebust int wsize = NFS_SERVER(inode)->wsize; 11067d46a49fSTrond Myklebust int error; 11071da177e4SLinus Torvalds 11087d46a49fSTrond Myklebust flush_one = nfs_flush_one; 11097d46a49fSTrond Myklebust if (wsize < PAGE_CACHE_SIZE) 11107d46a49fSTrond Myklebust flush_one = nfs_flush_multi; 11117d46a49fSTrond Myklebust /* For single writes, FLUSH_STABLE is more efficient */ 11127d46a49fSTrond Myklebust if (npages <= wpages && npages == NFS_I(inode)->npages 11137d46a49fSTrond Myklebust && nfs_list_entry(head->next)->wb_bytes <= wsize) 11147d46a49fSTrond Myklebust how |= FLUSH_STABLE; 11157d46a49fSTrond Myklebust 11167d46a49fSTrond Myklebust do { 11177d46a49fSTrond Myklebust nfs_coalesce_requests(head, &one_request, wpages); 11181da177e4SLinus Torvalds req = nfs_list_entry(one_request.next); 11197d46a49fSTrond Myklebust error = flush_one(inode, &one_request, how); 11201da177e4SLinus Torvalds if (error < 0) 11217d46a49fSTrond Myklebust goto out_err; 11227d46a49fSTrond Myklebust } while (!list_empty(head)); 11237d46a49fSTrond Myklebust return 0; 11247d46a49fSTrond Myklebust out_err: 11251da177e4SLinus Torvalds while (!list_empty(head)) { 11261da177e4SLinus Torvalds req = nfs_list_entry(head->next); 11271da177e4SLinus Torvalds nfs_list_remove_request(req); 11281da177e4SLinus Torvalds nfs_mark_request_dirty(req); 1129c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 11301da177e4SLinus Torvalds } 11311da177e4SLinus Torvalds return error; 11321da177e4SLinus Torvalds } 11331da177e4SLinus Torvalds 11341da177e4SLinus Torvalds /* 11351da177e4SLinus Torvalds * Handle a write reply that flushed part of a page. 11361da177e4SLinus Torvalds */ 1137788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) 11381da177e4SLinus Torvalds { 1139788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 11401da177e4SLinus Torvalds struct nfs_page *req = data->req; 11411da177e4SLinus Torvalds struct page *page = req->wb_page; 11421da177e4SLinus Torvalds 11431da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 11441da177e4SLinus Torvalds req->wb_context->dentry->d_inode->i_sb->s_id, 11451da177e4SLinus Torvalds (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 11461da177e4SLinus Torvalds req->wb_bytes, 11471da177e4SLinus Torvalds (long long)req_offset(req)); 11481da177e4SLinus Torvalds 1149788e7a89STrond Myklebust if (nfs_writeback_done(task, data) != 0) 1150788e7a89STrond Myklebust return; 1151788e7a89STrond Myklebust 1152788e7a89STrond Myklebust if (task->tk_status < 0) { 11531da177e4SLinus Torvalds ClearPageUptodate(page); 11541da177e4SLinus Torvalds SetPageError(page); 1155788e7a89STrond Myklebust req->wb_context->error = task->tk_status; 1156788e7a89STrond Myklebust dprintk(", error = %d\n", task->tk_status); 11571da177e4SLinus Torvalds } else { 11581da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 11591da177e4SLinus Torvalds if (data->verf.committed < NFS_FILE_SYNC) { 11601da177e4SLinus Torvalds if (!NFS_NEED_COMMIT(req)) { 11611da177e4SLinus Torvalds nfs_defer_commit(req); 11621da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 11631da177e4SLinus Torvalds dprintk(" defer commit\n"); 11641da177e4SLinus Torvalds } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { 11651da177e4SLinus Torvalds nfs_defer_reschedule(req); 11661da177e4SLinus Torvalds dprintk(" server reboot detected\n"); 11671da177e4SLinus Torvalds } 11681da177e4SLinus Torvalds } else 11691da177e4SLinus Torvalds #endif 11701da177e4SLinus Torvalds dprintk(" OK\n"); 11711da177e4SLinus Torvalds } 11721da177e4SLinus Torvalds 11731da177e4SLinus Torvalds if (atomic_dec_and_test(&req->wb_complete)) 11741da177e4SLinus Torvalds nfs_writepage_release(req); 11751da177e4SLinus Torvalds } 11761da177e4SLinus Torvalds 1177788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops = { 1178788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_partial, 1179788e7a89STrond Myklebust .rpc_release = nfs_writedata_release, 1180788e7a89STrond Myklebust }; 1181788e7a89STrond Myklebust 11821da177e4SLinus Torvalds /* 11831da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 11841da177e4SLinus Torvalds * 11851da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 11861da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 11871da177e4SLinus Torvalds * as the page has a write request pending. 11881da177e4SLinus Torvalds */ 1189788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) 11901da177e4SLinus Torvalds { 1191788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 11921da177e4SLinus Torvalds struct nfs_page *req; 11931da177e4SLinus Torvalds struct page *page; 11941da177e4SLinus Torvalds 1195788e7a89STrond Myklebust if (nfs_writeback_done(task, data) != 0) 1196788e7a89STrond Myklebust return; 1197788e7a89STrond Myklebust 11981da177e4SLinus Torvalds /* Update attributes as result of writeback. */ 11991da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 12001da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 12011da177e4SLinus Torvalds nfs_list_remove_request(req); 12021da177e4SLinus Torvalds page = req->wb_page; 12031da177e4SLinus Torvalds 12041da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 12051da177e4SLinus Torvalds req->wb_context->dentry->d_inode->i_sb->s_id, 12061da177e4SLinus Torvalds (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 12071da177e4SLinus Torvalds req->wb_bytes, 12081da177e4SLinus Torvalds (long long)req_offset(req)); 12091da177e4SLinus Torvalds 1210788e7a89STrond Myklebust if (task->tk_status < 0) { 12111da177e4SLinus Torvalds ClearPageUptodate(page); 12121da177e4SLinus Torvalds SetPageError(page); 1213788e7a89STrond Myklebust req->wb_context->error = task->tk_status; 12141da177e4SLinus Torvalds end_page_writeback(page); 12151da177e4SLinus Torvalds nfs_inode_remove_request(req); 1216788e7a89STrond Myklebust dprintk(", error = %d\n", task->tk_status); 12171da177e4SLinus Torvalds goto next; 12181da177e4SLinus Torvalds } 12191da177e4SLinus Torvalds end_page_writeback(page); 12201da177e4SLinus Torvalds 12211da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 12221da177e4SLinus Torvalds if (data->args.stable != NFS_UNSTABLE || data->verf.committed == NFS_FILE_SYNC) { 12231da177e4SLinus Torvalds nfs_inode_remove_request(req); 12241da177e4SLinus Torvalds dprintk(" OK\n"); 12251da177e4SLinus Torvalds goto next; 12261da177e4SLinus Torvalds } 12271da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 12281da177e4SLinus Torvalds nfs_mark_request_commit(req); 12291da177e4SLinus Torvalds dprintk(" marked for commit\n"); 12301da177e4SLinus Torvalds #else 12311da177e4SLinus Torvalds nfs_inode_remove_request(req); 12321da177e4SLinus Torvalds #endif 12331da177e4SLinus Torvalds next: 1234c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 12351da177e4SLinus Torvalds } 12361da177e4SLinus Torvalds } 12371da177e4SLinus Torvalds 1238788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = { 1239788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_full, 1240788e7a89STrond Myklebust .rpc_release = nfs_writedata_release, 1241788e7a89STrond Myklebust }; 1242788e7a89STrond Myklebust 1243788e7a89STrond Myklebust 12441da177e4SLinus Torvalds /* 12451da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 12461da177e4SLinus Torvalds */ 1247462d5b32SChuck Lever int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 12481da177e4SLinus Torvalds { 12491da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 12501da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1251788e7a89STrond Myklebust int status; 12521da177e4SLinus Torvalds 12531da177e4SLinus Torvalds dprintk("NFS: %4d nfs_writeback_done (status %d)\n", 12541da177e4SLinus Torvalds task->tk_pid, task->tk_status); 12551da177e4SLinus Torvalds 1256788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1257788e7a89STrond Myklebust status = NFS_PROTO(data->inode)->write_done(task, data); 1258788e7a89STrond Myklebust if (status != 0) 1259788e7a89STrond Myklebust return status; 126091d5b470SChuck Lever nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 126191d5b470SChuck Lever 12621da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 12631da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 12641da177e4SLinus Torvalds /* We tried a write call, but the server did not 12651da177e4SLinus Torvalds * commit data to stable storage even though we 12661da177e4SLinus Torvalds * requested it. 12671da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 12681da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 12691da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 12701da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 12711da177e4SLinus Torvalds */ 12721da177e4SLinus Torvalds static unsigned long complain; 12731da177e4SLinus Torvalds 12741da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 12751da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 12761da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 127754ceac45SDavid Howells NFS_SERVER(data->inode)->nfs_client->cl_hostname, 12781da177e4SLinus Torvalds resp->verf->committed, argp->stable); 12791da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 12801da177e4SLinus Torvalds } 12811da177e4SLinus Torvalds } 12821da177e4SLinus Torvalds #endif 12831da177e4SLinus Torvalds /* Is this a short write? */ 12841da177e4SLinus Torvalds if (task->tk_status >= 0 && resp->count < argp->count) { 12851da177e4SLinus Torvalds static unsigned long complain; 12861da177e4SLinus Torvalds 128791d5b470SChuck Lever nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); 128891d5b470SChuck Lever 12891da177e4SLinus Torvalds /* Has the server at least made some progress? */ 12901da177e4SLinus Torvalds if (resp->count != 0) { 12911da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 12921da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 12931da177e4SLinus Torvalds /* Resend from where the server left off */ 12941da177e4SLinus Torvalds argp->offset += resp->count; 12951da177e4SLinus Torvalds argp->pgbase += resp->count; 12961da177e4SLinus Torvalds argp->count -= resp->count; 12971da177e4SLinus Torvalds } else { 12981da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 12991da177e4SLinus Torvalds * headaches in the case of a server crash. 13001da177e4SLinus Torvalds */ 13011da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 13021da177e4SLinus Torvalds } 13031da177e4SLinus Torvalds rpc_restart_call(task); 1304788e7a89STrond Myklebust return -EAGAIN; 13051da177e4SLinus Torvalds } 13061da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 13071da177e4SLinus Torvalds printk(KERN_WARNING 13081da177e4SLinus Torvalds "NFS: Server wrote zero bytes, expected %u.\n", 13091da177e4SLinus Torvalds argp->count); 13101da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 13111da177e4SLinus Torvalds } 13121da177e4SLinus Torvalds /* Can't do anything about it except throw an error. */ 13131da177e4SLinus Torvalds task->tk_status = -EIO; 13141da177e4SLinus Torvalds } 1315788e7a89STrond Myklebust return 0; 13161da177e4SLinus Torvalds } 13171da177e4SLinus Torvalds 13181da177e4SLinus Torvalds 13191da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 1320963d8fe5STrond Myklebust void nfs_commit_release(void *wdata) 13211da177e4SLinus Torvalds { 13221da177e4SLinus Torvalds nfs_commit_free(wdata); 13231da177e4SLinus Torvalds } 13241da177e4SLinus Torvalds 13251da177e4SLinus Torvalds /* 13261da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 13271da177e4SLinus Torvalds */ 13281da177e4SLinus Torvalds static void nfs_commit_rpcsetup(struct list_head *head, 1329788e7a89STrond Myklebust struct nfs_write_data *data, 1330788e7a89STrond Myklebust int how) 13311da177e4SLinus Torvalds { 13323da28eb1STrond Myklebust struct nfs_page *first; 13331da177e4SLinus Torvalds struct inode *inode; 1334788e7a89STrond Myklebust int flags; 13351da177e4SLinus Torvalds 13361da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 13371da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 13381da177e4SLinus Torvalds 13391da177e4SLinus Torvalds list_splice_init(head, &data->pages); 13401da177e4SLinus Torvalds first = nfs_list_entry(data->pages.next); 13411da177e4SLinus Torvalds inode = first->wb_context->dentry->d_inode; 13421da177e4SLinus Torvalds 13431da177e4SLinus Torvalds data->inode = inode; 13441da177e4SLinus Torvalds data->cred = first->wb_context->cred; 13451da177e4SLinus Torvalds 13461da177e4SLinus Torvalds data->args.fh = NFS_FH(data->inode); 13473da28eb1STrond Myklebust /* Note: we always request a commit of the entire inode */ 13483da28eb1STrond Myklebust data->args.offset = 0; 13493da28eb1STrond Myklebust data->args.count = 0; 13503da28eb1STrond Myklebust data->res.count = 0; 13511da177e4SLinus Torvalds data->res.fattr = &data->fattr; 13521da177e4SLinus Torvalds data->res.verf = &data->verf; 13530e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 13541da177e4SLinus Torvalds 1355788e7a89STrond Myklebust /* Set up the initial task struct. */ 1356788e7a89STrond Myklebust flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 1357788e7a89STrond Myklebust rpc_init_task(&data->task, NFS_CLIENT(inode), flags, &nfs_commit_ops, data); 13581da177e4SLinus Torvalds NFS_PROTO(inode)->commit_setup(data, how); 13591da177e4SLinus Torvalds 13601da177e4SLinus Torvalds data->task.tk_priority = flush_task_priority(how); 13611da177e4SLinus Torvalds data->task.tk_cookie = (unsigned long)inode; 13621da177e4SLinus Torvalds 13630bbacc40SChuck Lever dprintk("NFS: %4d initiated commit call\n", data->task.tk_pid); 13641da177e4SLinus Torvalds } 13651da177e4SLinus Torvalds 13661da177e4SLinus Torvalds /* 13671da177e4SLinus Torvalds * Commit dirty pages 13681da177e4SLinus Torvalds */ 13691da177e4SLinus Torvalds static int 137040859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how) 13711da177e4SLinus Torvalds { 13721da177e4SLinus Torvalds struct nfs_write_data *data; 13731da177e4SLinus Torvalds struct nfs_page *req; 13741da177e4SLinus Torvalds 1375e9f7bee1STrond Myklebust data = nfs_commit_alloc(); 13761da177e4SLinus Torvalds 13771da177e4SLinus Torvalds if (!data) 13781da177e4SLinus Torvalds goto out_bad; 13791da177e4SLinus Torvalds 13801da177e4SLinus Torvalds /* Set up the argument struct */ 13811da177e4SLinus Torvalds nfs_commit_rpcsetup(head, data, how); 13821da177e4SLinus Torvalds 13831da177e4SLinus Torvalds nfs_execute_write(data); 13841da177e4SLinus Torvalds return 0; 13851da177e4SLinus Torvalds out_bad: 13861da177e4SLinus Torvalds while (!list_empty(head)) { 13871da177e4SLinus Torvalds req = nfs_list_entry(head->next); 13881da177e4SLinus Torvalds nfs_list_remove_request(req); 13891da177e4SLinus Torvalds nfs_mark_request_commit(req); 139083715ad5STrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 13915c2d97cbSTrond Myklebust nfs_clear_page_writeback(req); 13921da177e4SLinus Torvalds } 13931da177e4SLinus Torvalds return -ENOMEM; 13941da177e4SLinus Torvalds } 13951da177e4SLinus Torvalds 13961da177e4SLinus Torvalds /* 13971da177e4SLinus Torvalds * COMMIT call returned 13981da177e4SLinus Torvalds */ 1399788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 14001da177e4SLinus Torvalds { 1401963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 14021da177e4SLinus Torvalds struct nfs_page *req; 14031da177e4SLinus Torvalds 14041da177e4SLinus Torvalds dprintk("NFS: %4d nfs_commit_done (status %d)\n", 14051da177e4SLinus Torvalds task->tk_pid, task->tk_status); 14061da177e4SLinus Torvalds 1407788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1408788e7a89STrond Myklebust if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) 1409788e7a89STrond Myklebust return; 1410788e7a89STrond Myklebust 14111da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 14121da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 14131da177e4SLinus Torvalds nfs_list_remove_request(req); 1414fd39fc85SChristoph Lameter dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 14151da177e4SLinus Torvalds 14161da177e4SLinus Torvalds dprintk("NFS: commit (%s/%Ld %d@%Ld)", 14171da177e4SLinus Torvalds req->wb_context->dentry->d_inode->i_sb->s_id, 14181da177e4SLinus Torvalds (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 14191da177e4SLinus Torvalds req->wb_bytes, 14201da177e4SLinus Torvalds (long long)req_offset(req)); 14211da177e4SLinus Torvalds if (task->tk_status < 0) { 14221da177e4SLinus Torvalds req->wb_context->error = task->tk_status; 14231da177e4SLinus Torvalds nfs_inode_remove_request(req); 14241da177e4SLinus Torvalds dprintk(", error = %d\n", task->tk_status); 14251da177e4SLinus Torvalds goto next; 14261da177e4SLinus Torvalds } 14271da177e4SLinus Torvalds 14281da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 14291da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 14301da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 14311da177e4SLinus Torvalds /* We have a match */ 14321da177e4SLinus Torvalds nfs_inode_remove_request(req); 14331da177e4SLinus Torvalds dprintk(" OK\n"); 14341da177e4SLinus Torvalds goto next; 14351da177e4SLinus Torvalds } 14361da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 14371da177e4SLinus Torvalds dprintk(" mismatch\n"); 14381da177e4SLinus Torvalds nfs_mark_request_dirty(req); 14391da177e4SLinus Torvalds next: 1440c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 14411da177e4SLinus Torvalds } 14421da177e4SLinus Torvalds } 1443788e7a89STrond Myklebust 1444788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 1445788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1446788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1447788e7a89STrond Myklebust }; 1448c42de9ddSTrond Myklebust #else 1449c42de9ddSTrond Myklebust static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how) 1450c42de9ddSTrond Myklebust { 1451c42de9ddSTrond Myklebust return 0; 1452c42de9ddSTrond Myklebust } 14531da177e4SLinus Torvalds #endif 14541da177e4SLinus Torvalds 14551da177e4SLinus Torvalds static int nfs_flush_inode(struct inode *inode, unsigned long idx_start, 14561da177e4SLinus Torvalds unsigned int npages, int how) 14571da177e4SLinus Torvalds { 14581da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 14591da177e4SLinus Torvalds LIST_HEAD(head); 14607d46a49fSTrond Myklebust int res; 14611da177e4SLinus Torvalds 14621da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 14631da177e4SLinus Torvalds res = nfs_scan_dirty(inode, &head, idx_start, npages); 14641da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 1465ab0a3dbeSTrond Myklebust if (res) { 14667d46a49fSTrond Myklebust int error = nfs_flush_list(inode, &head, res, how); 14671da177e4SLinus Torvalds if (error < 0) 14681da177e4SLinus Torvalds return error; 14697d46a49fSTrond Myklebust } 14701da177e4SLinus Torvalds return res; 14711da177e4SLinus Torvalds } 14721da177e4SLinus Torvalds 14731da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 14743da28eb1STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 14751da177e4SLinus Torvalds { 14761da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 14771da177e4SLinus Torvalds LIST_HEAD(head); 14787d46a49fSTrond Myklebust int res; 14791da177e4SLinus Torvalds 14801da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 14813da28eb1STrond Myklebust res = nfs_scan_commit(inode, &head, 0, 0); 14823da28eb1STrond Myklebust spin_unlock(&nfsi->req_lock); 14831da177e4SLinus Torvalds if (res) { 14847d46a49fSTrond Myklebust int error = nfs_commit_list(inode, &head, how); 14851da177e4SLinus Torvalds if (error < 0) 14861da177e4SLinus Torvalds return error; 14873da28eb1STrond Myklebust } 14881da177e4SLinus Torvalds return res; 14891da177e4SLinus Torvalds } 14901da177e4SLinus Torvalds #endif 14911da177e4SLinus Torvalds 1492c42de9ddSTrond Myklebust int nfs_sync_inode_wait(struct inode *inode, unsigned long idx_start, 14931da177e4SLinus Torvalds unsigned int npages, int how) 14941da177e4SLinus Torvalds { 1495c42de9ddSTrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1496c42de9ddSTrond Myklebust LIST_HEAD(head); 149770b9ecbdSTrond Myklebust int nocommit = how & FLUSH_NOCOMMIT; 1498c42de9ddSTrond Myklebust int pages, ret; 14991da177e4SLinus Torvalds 1500c42de9ddSTrond Myklebust how &= ~FLUSH_NOCOMMIT; 1501c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 15021da177e4SLinus Torvalds do { 1503c42de9ddSTrond Myklebust ret = nfs_wait_on_requests_locked(inode, idx_start, npages); 1504c42de9ddSTrond Myklebust if (ret != 0) 1505c42de9ddSTrond Myklebust continue; 1506c42de9ddSTrond Myklebust pages = nfs_scan_dirty(inode, &head, idx_start, npages); 1507c42de9ddSTrond Myklebust if (pages != 0) { 1508c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 1509d2ccddf0STrond Myklebust if (how & FLUSH_INVALIDATE) 151083715ad5STrond Myklebust nfs_cancel_dirty_list(&head); 1511d2ccddf0STrond Myklebust else 1512c42de9ddSTrond Myklebust ret = nfs_flush_list(inode, &head, pages, how); 1513c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 151470b9ecbdSTrond Myklebust continue; 151570b9ecbdSTrond Myklebust } 1516c42de9ddSTrond Myklebust if (nocommit) 1517c42de9ddSTrond Myklebust break; 1518d2ccddf0STrond Myklebust pages = nfs_scan_commit(inode, &head, idx_start, npages); 1519c42de9ddSTrond Myklebust if (pages == 0) 1520c42de9ddSTrond Myklebust break; 1521d2ccddf0STrond Myklebust if (how & FLUSH_INVALIDATE) { 1522d2ccddf0STrond Myklebust spin_unlock(&nfsi->req_lock); 152383715ad5STrond Myklebust nfs_cancel_commit_list(&head); 1524d2ccddf0STrond Myklebust spin_lock(&nfsi->req_lock); 1525d2ccddf0STrond Myklebust continue; 1526d2ccddf0STrond Myklebust } 1527d2ccddf0STrond Myklebust pages += nfs_scan_commit(inode, &head, 0, 0); 1528c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 1529c42de9ddSTrond Myklebust ret = nfs_commit_list(inode, &head, how); 1530c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 1531c42de9ddSTrond Myklebust } while (ret >= 0); 1532c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 1533c42de9ddSTrond Myklebust return ret; 15341da177e4SLinus Torvalds } 15351da177e4SLinus Torvalds 1536f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 15371da177e4SLinus Torvalds { 15381da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 15391da177e4SLinus Torvalds sizeof(struct nfs_write_data), 15401da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 15411da177e4SLinus Torvalds NULL, NULL); 15421da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 15431da177e4SLinus Torvalds return -ENOMEM; 15441da177e4SLinus Torvalds 154593d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 15461da177e4SLinus Torvalds nfs_wdata_cachep); 15471da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 15481da177e4SLinus Torvalds return -ENOMEM; 15491da177e4SLinus Torvalds 155093d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 15511da177e4SLinus Torvalds nfs_wdata_cachep); 15521da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 15531da177e4SLinus Torvalds return -ENOMEM; 15541da177e4SLinus Torvalds 15551da177e4SLinus Torvalds return 0; 15561da177e4SLinus Torvalds } 15571da177e4SLinus Torvalds 1558266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 15591da177e4SLinus Torvalds { 15601da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 15611da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 15621da177e4SLinus Torvalds if (kmem_cache_destroy(nfs_wdata_cachep)) 15631da177e4SLinus Torvalds printk(KERN_INFO "nfs_write_data: not all structures were freed\n"); 15641da177e4SLinus Torvalds } 15651da177e4SLinus Torvalds 1566