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 93*e9f7bee1STrond 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 111*e9f7bee1STrond Myklebust struct nfs_write_data *nfs_writedata_alloc(size_t len) 1123feb2d49STrond Myklebust { 113*e9f7bee1STrond 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); 119*e9f7bee1STrond 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 204*e9f7bee1STrond 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); 3991da177e4SLinus Torvalds return err; 4001da177e4SLinus Torvalds } 4011da177e4SLinus Torvalds 4021da177e4SLinus Torvalds /* 4031da177e4SLinus Torvalds * Insert a write request into an inode 4041da177e4SLinus Torvalds */ 4051da177e4SLinus Torvalds static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 4061da177e4SLinus Torvalds { 4071da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4081da177e4SLinus Torvalds int error; 4091da177e4SLinus Torvalds 4101da177e4SLinus Torvalds error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); 4111da177e4SLinus Torvalds BUG_ON(error == -EEXIST); 4121da177e4SLinus Torvalds if (error) 4131da177e4SLinus Torvalds return error; 4141da177e4SLinus Torvalds if (!nfsi->npages) { 4151da177e4SLinus Torvalds igrab(inode); 4161da177e4SLinus Torvalds nfs_begin_data_update(inode); 4171da177e4SLinus Torvalds if (nfs_have_delegation(inode, FMODE_WRITE)) 4181da177e4SLinus Torvalds nfsi->change_attr++; 4191da177e4SLinus Torvalds } 420deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 4211da177e4SLinus Torvalds nfsi->npages++; 4221da177e4SLinus Torvalds atomic_inc(&req->wb_count); 4231da177e4SLinus Torvalds return 0; 4241da177e4SLinus Torvalds } 4251da177e4SLinus Torvalds 4261da177e4SLinus Torvalds /* 4271da177e4SLinus Torvalds * Insert a write request into an inode 4281da177e4SLinus Torvalds */ 4291da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 4301da177e4SLinus Torvalds { 4311da177e4SLinus Torvalds struct inode *inode = req->wb_context->dentry->d_inode; 4321da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4331da177e4SLinus Torvalds 4341da177e4SLinus Torvalds BUG_ON (!NFS_WBACK_BUSY(req)); 4351da177e4SLinus Torvalds 4361da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 437deb7d638STrond Myklebust ClearPagePrivate(req->wb_page); 4381da177e4SLinus Torvalds radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); 4391da177e4SLinus Torvalds nfsi->npages--; 4401da177e4SLinus Torvalds if (!nfsi->npages) { 4411da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 442951a143bSTrond Myklebust nfs_end_data_update(inode); 4431da177e4SLinus Torvalds iput(inode); 4441da177e4SLinus Torvalds } else 4451da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 4461da177e4SLinus Torvalds nfs_clear_request(req); 4471da177e4SLinus Torvalds nfs_release_request(req); 4481da177e4SLinus Torvalds } 4491da177e4SLinus Torvalds 4501da177e4SLinus Torvalds /* 4511da177e4SLinus Torvalds * Find a request 4521da177e4SLinus Torvalds */ 4531da177e4SLinus Torvalds static inline struct nfs_page * 4541da177e4SLinus Torvalds _nfs_find_request(struct inode *inode, unsigned long index) 4551da177e4SLinus Torvalds { 4561da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4571da177e4SLinus Torvalds struct nfs_page *req; 4581da177e4SLinus Torvalds 4591da177e4SLinus Torvalds req = (struct nfs_page*)radix_tree_lookup(&nfsi->nfs_page_tree, index); 4601da177e4SLinus Torvalds if (req) 4611da177e4SLinus Torvalds atomic_inc(&req->wb_count); 4621da177e4SLinus Torvalds return req; 4631da177e4SLinus Torvalds } 4641da177e4SLinus Torvalds 4651da177e4SLinus Torvalds static struct nfs_page * 4661da177e4SLinus Torvalds nfs_find_request(struct inode *inode, unsigned long index) 4671da177e4SLinus Torvalds { 4681da177e4SLinus Torvalds struct nfs_page *req; 4691da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4701da177e4SLinus Torvalds 4711da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 4721da177e4SLinus Torvalds req = _nfs_find_request(inode, index); 4731da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 4741da177e4SLinus Torvalds return req; 4751da177e4SLinus Torvalds } 4761da177e4SLinus Torvalds 4771da177e4SLinus Torvalds /* 4781da177e4SLinus Torvalds * Add a request to the inode's dirty list. 4791da177e4SLinus Torvalds */ 4801da177e4SLinus Torvalds static void 4811da177e4SLinus Torvalds nfs_mark_request_dirty(struct nfs_page *req) 4821da177e4SLinus Torvalds { 4831da177e4SLinus Torvalds struct inode *inode = req->wb_context->dentry->d_inode; 4841da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4851da177e4SLinus Torvalds 4861da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 4873da28eb1STrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, 4883da28eb1STrond Myklebust req->wb_index, NFS_PAGE_TAG_DIRTY); 4891da177e4SLinus Torvalds nfs_list_add_request(req, &nfsi->dirty); 4901da177e4SLinus Torvalds nfsi->ndirty++; 4911da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 492b1e7a8fdSChristoph Lameter inc_zone_page_state(req->wb_page, NR_FILE_DIRTY); 4931da177e4SLinus Torvalds mark_inode_dirty(inode); 4941da177e4SLinus Torvalds } 4951da177e4SLinus Torvalds 4961da177e4SLinus Torvalds /* 4971da177e4SLinus Torvalds * Check if a request is dirty 4981da177e4SLinus Torvalds */ 4991da177e4SLinus Torvalds static inline int 5001da177e4SLinus Torvalds nfs_dirty_request(struct nfs_page *req) 5011da177e4SLinus Torvalds { 5021da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode); 5031da177e4SLinus Torvalds return !list_empty(&req->wb_list) && req->wb_list_head == &nfsi->dirty; 5041da177e4SLinus Torvalds } 5051da177e4SLinus Torvalds 5061da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 5071da177e4SLinus Torvalds /* 5081da177e4SLinus Torvalds * Add a request to the inode's commit list. 5091da177e4SLinus Torvalds */ 5101da177e4SLinus Torvalds static void 5111da177e4SLinus Torvalds nfs_mark_request_commit(struct nfs_page *req) 5121da177e4SLinus Torvalds { 5131da177e4SLinus Torvalds struct inode *inode = req->wb_context->dentry->d_inode; 5141da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5151da177e4SLinus Torvalds 5161da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 5171da177e4SLinus Torvalds nfs_list_add_request(req, &nfsi->commit); 5181da177e4SLinus Torvalds nfsi->ncommit++; 5191da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 520fd39fc85SChristoph Lameter inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 5211da177e4SLinus Torvalds mark_inode_dirty(inode); 5221da177e4SLinus Torvalds } 5231da177e4SLinus Torvalds #endif 5241da177e4SLinus Torvalds 5251da177e4SLinus Torvalds /* 5261da177e4SLinus Torvalds * Wait for a request to complete. 5271da177e4SLinus Torvalds * 5281da177e4SLinus Torvalds * Interruptible by signals only if mounted with intr flag. 5291da177e4SLinus Torvalds */ 530c42de9ddSTrond Myklebust static int nfs_wait_on_requests_locked(struct inode *inode, unsigned long idx_start, unsigned int npages) 5311da177e4SLinus Torvalds { 5321da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5331da177e4SLinus Torvalds struct nfs_page *req; 5341da177e4SLinus Torvalds unsigned long idx_end, next; 5351da177e4SLinus Torvalds unsigned int res = 0; 5361da177e4SLinus Torvalds int error; 5371da177e4SLinus Torvalds 5381da177e4SLinus Torvalds if (npages == 0) 5391da177e4SLinus Torvalds idx_end = ~0; 5401da177e4SLinus Torvalds else 5411da177e4SLinus Torvalds idx_end = idx_start + npages - 1; 5421da177e4SLinus Torvalds 5431da177e4SLinus Torvalds next = idx_start; 544c6a556b8STrond Myklebust while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_WRITEBACK)) { 5451da177e4SLinus Torvalds if (req->wb_index > idx_end) 5461da177e4SLinus Torvalds break; 5471da177e4SLinus Torvalds 5481da177e4SLinus Torvalds next = req->wb_index + 1; 549c6a556b8STrond Myklebust BUG_ON(!NFS_WBACK_BUSY(req)); 5501da177e4SLinus Torvalds 5511da177e4SLinus Torvalds atomic_inc(&req->wb_count); 5521da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 5531da177e4SLinus Torvalds error = nfs_wait_on_request(req); 5541da177e4SLinus Torvalds nfs_release_request(req); 555c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 5561da177e4SLinus Torvalds if (error < 0) 5571da177e4SLinus Torvalds return error; 5581da177e4SLinus Torvalds res++; 5591da177e4SLinus Torvalds } 5601da177e4SLinus Torvalds return res; 5611da177e4SLinus Torvalds } 5621da177e4SLinus Torvalds 563c42de9ddSTrond Myklebust static int nfs_wait_on_requests(struct inode *inode, unsigned long idx_start, unsigned int npages) 564c42de9ddSTrond Myklebust { 565c42de9ddSTrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 566c42de9ddSTrond Myklebust int ret; 567c42de9ddSTrond Myklebust 568c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 569c42de9ddSTrond Myklebust ret = nfs_wait_on_requests_locked(inode, idx_start, npages); 570c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 571c42de9ddSTrond Myklebust return ret; 572c42de9ddSTrond Myklebust } 573c42de9ddSTrond Myklebust 57483715ad5STrond Myklebust static void nfs_cancel_dirty_list(struct list_head *head) 575d2ccddf0STrond Myklebust { 576d2ccddf0STrond Myklebust struct nfs_page *req; 577d2ccddf0STrond Myklebust while(!list_empty(head)) { 578d2ccddf0STrond Myklebust req = nfs_list_entry(head->next); 579d2ccddf0STrond Myklebust nfs_list_remove_request(req); 580d2ccddf0STrond Myklebust nfs_inode_remove_request(req); 581d2ccddf0STrond Myklebust nfs_clear_page_writeback(req); 582d2ccddf0STrond Myklebust } 583d2ccddf0STrond Myklebust } 584d2ccddf0STrond Myklebust 58583715ad5STrond Myklebust static void nfs_cancel_commit_list(struct list_head *head) 58683715ad5STrond Myklebust { 58783715ad5STrond Myklebust struct nfs_page *req; 58883715ad5STrond Myklebust 58983715ad5STrond Myklebust while(!list_empty(head)) { 59083715ad5STrond Myklebust req = nfs_list_entry(head->next); 59183715ad5STrond Myklebust nfs_list_remove_request(req); 59283715ad5STrond Myklebust nfs_inode_remove_request(req); 59383715ad5STrond Myklebust nfs_clear_page_writeback(req); 59483715ad5STrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 59583715ad5STrond Myklebust } 59683715ad5STrond Myklebust } 59783715ad5STrond Myklebust 5981da177e4SLinus Torvalds /* 5991da177e4SLinus Torvalds * nfs_scan_dirty - Scan an inode for dirty requests 6001da177e4SLinus Torvalds * @inode: NFS inode to scan 6011da177e4SLinus Torvalds * @dst: destination list 6021da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 6031da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 6041da177e4SLinus Torvalds * 6051da177e4SLinus Torvalds * Moves requests from the inode's dirty page list. 6061da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 6071da177e4SLinus Torvalds */ 6081da177e4SLinus Torvalds static int 6091da177e4SLinus Torvalds nfs_scan_dirty(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) 6101da177e4SLinus Torvalds { 6111da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 6123da28eb1STrond Myklebust int res = 0; 6133da28eb1STrond Myklebust 6143da28eb1STrond Myklebust if (nfsi->ndirty != 0) { 6153da28eb1STrond Myklebust res = nfs_scan_lock_dirty(nfsi, dst, idx_start, npages); 6161da177e4SLinus Torvalds nfsi->ndirty -= res; 6171da177e4SLinus Torvalds if ((nfsi->ndirty == 0) != list_empty(&nfsi->dirty)) 6181da177e4SLinus Torvalds printk(KERN_ERR "NFS: desynchronized value of nfs_i.ndirty.\n"); 6193da28eb1STrond Myklebust } 6201da177e4SLinus Torvalds return res; 6211da177e4SLinus Torvalds } 6221da177e4SLinus Torvalds 6231da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 6241da177e4SLinus Torvalds /* 6251da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 6261da177e4SLinus Torvalds * @inode: NFS inode to scan 6271da177e4SLinus Torvalds * @dst: destination list 6281da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 6291da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 6301da177e4SLinus Torvalds * 6311da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 6321da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 6331da177e4SLinus Torvalds */ 6341da177e4SLinus Torvalds static int 6351da177e4SLinus Torvalds nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) 6361da177e4SLinus Torvalds { 6371da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 6383da28eb1STrond Myklebust int res = 0; 6393da28eb1STrond Myklebust 6403da28eb1STrond Myklebust if (nfsi->ncommit != 0) { 641d2ccddf0STrond Myklebust res = nfs_scan_list(nfsi, &nfsi->commit, dst, idx_start, npages); 6421da177e4SLinus Torvalds nfsi->ncommit -= res; 6431da177e4SLinus Torvalds if ((nfsi->ncommit == 0) != list_empty(&nfsi->commit)) 6441da177e4SLinus Torvalds printk(KERN_ERR "NFS: desynchronized value of nfs_i.ncommit.\n"); 6453da28eb1STrond Myklebust } 6461da177e4SLinus Torvalds return res; 6471da177e4SLinus Torvalds } 648c42de9ddSTrond Myklebust #else 649c42de9ddSTrond Myklebust static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) 650c42de9ddSTrond Myklebust { 651c42de9ddSTrond Myklebust return 0; 652c42de9ddSTrond Myklebust } 6531da177e4SLinus Torvalds #endif 6541da177e4SLinus Torvalds 6551da177e4SLinus Torvalds static int nfs_wait_on_write_congestion(struct address_space *mapping, int intr) 6561da177e4SLinus Torvalds { 6571da177e4SLinus Torvalds struct backing_dev_info *bdi = mapping->backing_dev_info; 6581da177e4SLinus Torvalds DEFINE_WAIT(wait); 6591da177e4SLinus Torvalds int ret = 0; 6601da177e4SLinus Torvalds 6611da177e4SLinus Torvalds might_sleep(); 6621da177e4SLinus Torvalds 6631da177e4SLinus Torvalds if (!bdi_write_congested(bdi)) 6641da177e4SLinus Torvalds return 0; 66591d5b470SChuck Lever 66691d5b470SChuck Lever nfs_inc_stats(mapping->host, NFSIOS_CONGESTIONWAIT); 66791d5b470SChuck Lever 6681da177e4SLinus Torvalds if (intr) { 6691da177e4SLinus Torvalds struct rpc_clnt *clnt = NFS_CLIENT(mapping->host); 6701da177e4SLinus Torvalds sigset_t oldset; 6711da177e4SLinus Torvalds 6721da177e4SLinus Torvalds rpc_clnt_sigmask(clnt, &oldset); 6731da177e4SLinus Torvalds prepare_to_wait(&nfs_write_congestion, &wait, TASK_INTERRUPTIBLE); 6741da177e4SLinus Torvalds if (bdi_write_congested(bdi)) { 6751da177e4SLinus Torvalds if (signalled()) 6761da177e4SLinus Torvalds ret = -ERESTARTSYS; 6771da177e4SLinus Torvalds else 6781da177e4SLinus Torvalds schedule(); 6791da177e4SLinus Torvalds } 6801da177e4SLinus Torvalds rpc_clnt_sigunmask(clnt, &oldset); 6811da177e4SLinus Torvalds } else { 6821da177e4SLinus Torvalds prepare_to_wait(&nfs_write_congestion, &wait, TASK_UNINTERRUPTIBLE); 6831da177e4SLinus Torvalds if (bdi_write_congested(bdi)) 6841da177e4SLinus Torvalds schedule(); 6851da177e4SLinus Torvalds } 6861da177e4SLinus Torvalds finish_wait(&nfs_write_congestion, &wait); 6871da177e4SLinus Torvalds return ret; 6881da177e4SLinus Torvalds } 6891da177e4SLinus Torvalds 6901da177e4SLinus Torvalds 6911da177e4SLinus Torvalds /* 6921da177e4SLinus Torvalds * Try to update any existing write request, or create one if there is none. 6931da177e4SLinus Torvalds * In order to match, the request's credentials must match those of 6941da177e4SLinus Torvalds * the calling process. 6951da177e4SLinus Torvalds * 6961da177e4SLinus Torvalds * Note: Should always be called with the Page Lock held! 6971da177e4SLinus Torvalds */ 6981da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx, 6991da177e4SLinus Torvalds struct inode *inode, struct page *page, 7001da177e4SLinus Torvalds unsigned int offset, unsigned int bytes) 7011da177e4SLinus Torvalds { 7021da177e4SLinus Torvalds struct nfs_server *server = NFS_SERVER(inode); 7031da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 7041da177e4SLinus Torvalds struct nfs_page *req, *new = NULL; 7051da177e4SLinus Torvalds unsigned long rqend, end; 7061da177e4SLinus Torvalds 7071da177e4SLinus Torvalds end = offset + bytes; 7081da177e4SLinus Torvalds 7091da177e4SLinus Torvalds if (nfs_wait_on_write_congestion(page->mapping, server->flags & NFS_MOUNT_INTR)) 7101da177e4SLinus Torvalds return ERR_PTR(-ERESTARTSYS); 7111da177e4SLinus Torvalds for (;;) { 7121da177e4SLinus Torvalds /* Loop over all inode entries and see if we find 7131da177e4SLinus Torvalds * A request for the page we wish to update 7141da177e4SLinus Torvalds */ 7151da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 7161da177e4SLinus Torvalds req = _nfs_find_request(inode, page->index); 7171da177e4SLinus Torvalds if (req) { 7181da177e4SLinus Torvalds if (!nfs_lock_request_dontget(req)) { 7191da177e4SLinus Torvalds int error; 7201da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 7211da177e4SLinus Torvalds error = nfs_wait_on_request(req); 7221da177e4SLinus Torvalds nfs_release_request(req); 7231dd594b2SNeil Brown if (error < 0) { 7241dd594b2SNeil Brown if (new) 7251dd594b2SNeil Brown nfs_release_request(new); 7261da177e4SLinus Torvalds return ERR_PTR(error); 7271dd594b2SNeil Brown } 7281da177e4SLinus Torvalds continue; 7291da177e4SLinus Torvalds } 7301da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 7311da177e4SLinus Torvalds if (new) 7321da177e4SLinus Torvalds nfs_release_request(new); 7331da177e4SLinus Torvalds break; 7341da177e4SLinus Torvalds } 7351da177e4SLinus Torvalds 7361da177e4SLinus Torvalds if (new) { 7371da177e4SLinus Torvalds int error; 7381da177e4SLinus Torvalds nfs_lock_request_dontget(new); 7391da177e4SLinus Torvalds error = nfs_inode_add_request(inode, new); 7401da177e4SLinus Torvalds if (error) { 7411da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 7421da177e4SLinus Torvalds nfs_unlock_request(new); 7431da177e4SLinus Torvalds return ERR_PTR(error); 7441da177e4SLinus Torvalds } 7451da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 7461da177e4SLinus Torvalds nfs_mark_request_dirty(new); 7471da177e4SLinus Torvalds return new; 7481da177e4SLinus Torvalds } 7491da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 7501da177e4SLinus Torvalds 7511da177e4SLinus Torvalds new = nfs_create_request(ctx, inode, page, offset, bytes); 7521da177e4SLinus Torvalds if (IS_ERR(new)) 7531da177e4SLinus Torvalds return new; 7541da177e4SLinus Torvalds } 7551da177e4SLinus Torvalds 7561da177e4SLinus Torvalds /* We have a request for our page. 7571da177e4SLinus Torvalds * If the creds don't match, or the 7581da177e4SLinus Torvalds * page addresses don't match, 7591da177e4SLinus Torvalds * tell the caller to wait on the conflicting 7601da177e4SLinus Torvalds * request. 7611da177e4SLinus Torvalds */ 7621da177e4SLinus Torvalds rqend = req->wb_offset + req->wb_bytes; 7631da177e4SLinus Torvalds if (req->wb_context != ctx 7641da177e4SLinus Torvalds || req->wb_page != page 7651da177e4SLinus Torvalds || !nfs_dirty_request(req) 7661da177e4SLinus Torvalds || offset > rqend || end < req->wb_offset) { 7671da177e4SLinus Torvalds nfs_unlock_request(req); 7681da177e4SLinus Torvalds return ERR_PTR(-EBUSY); 7691da177e4SLinus Torvalds } 7701da177e4SLinus Torvalds 7711da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 7721da177e4SLinus Torvalds if (offset < req->wb_offset) { 7731da177e4SLinus Torvalds req->wb_offset = offset; 7741da177e4SLinus Torvalds req->wb_pgbase = offset; 7751da177e4SLinus Torvalds req->wb_bytes = rqend - req->wb_offset; 7761da177e4SLinus Torvalds } 7771da177e4SLinus Torvalds 7781da177e4SLinus Torvalds if (end > rqend) 7791da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 7801da177e4SLinus Torvalds 7811da177e4SLinus Torvalds return req; 7821da177e4SLinus Torvalds } 7831da177e4SLinus Torvalds 7841da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 7851da177e4SLinus Torvalds { 7861da177e4SLinus Torvalds struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; 7871da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 7881da177e4SLinus Torvalds struct nfs_page *req; 7891da177e4SLinus Torvalds int status = 0; 7901da177e4SLinus Torvalds /* 7911da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 7921da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 7931da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 7941da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 7951da177e4SLinus Torvalds * Also do the same if we find a request from an existing 7961da177e4SLinus Torvalds * dropped page. 7971da177e4SLinus Torvalds */ 7981da177e4SLinus Torvalds req = nfs_find_request(inode, page->index); 7991da177e4SLinus Torvalds if (req) { 8001da177e4SLinus Torvalds if (req->wb_page != page || ctx != req->wb_context) 8011da177e4SLinus Torvalds status = nfs_wb_page(inode, page); 8021da177e4SLinus Torvalds nfs_release_request(req); 8031da177e4SLinus Torvalds } 8041da177e4SLinus Torvalds return (status < 0) ? status : 0; 8051da177e4SLinus Torvalds } 8061da177e4SLinus Torvalds 8071da177e4SLinus Torvalds /* 8081da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 8091da177e4SLinus Torvalds * 8101da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 8111da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 8121da177e4SLinus Torvalds */ 8131da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 8141da177e4SLinus Torvalds unsigned int offset, unsigned int count) 8151da177e4SLinus Torvalds { 8161da177e4SLinus Torvalds struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; 8171da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 8181da177e4SLinus Torvalds struct nfs_page *req; 8191da177e4SLinus Torvalds int status = 0; 8201da177e4SLinus Torvalds 82191d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 82291d5b470SChuck Lever 8231da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage(%s/%s %d@%Ld)\n", 8240bbacc40SChuck Lever file->f_dentry->d_parent->d_name.name, 8250bbacc40SChuck Lever file->f_dentry->d_name.name, count, 8260bbacc40SChuck Lever (long long)(page_offset(page) +offset)); 8271da177e4SLinus Torvalds 8281da177e4SLinus Torvalds if (IS_SYNC(inode)) { 8291da177e4SLinus Torvalds status = nfs_writepage_sync(ctx, inode, page, offset, count, 0); 8301da177e4SLinus Torvalds if (status > 0) { 8311da177e4SLinus Torvalds if (offset == 0 && status == PAGE_CACHE_SIZE) 8321da177e4SLinus Torvalds SetPageUptodate(page); 8331da177e4SLinus Torvalds return 0; 8341da177e4SLinus Torvalds } 8351da177e4SLinus Torvalds return status; 8361da177e4SLinus Torvalds } 8371da177e4SLinus Torvalds 8381da177e4SLinus Torvalds /* If we're not using byte range locks, and we know the page 8391da177e4SLinus Torvalds * is entirely in cache, it may be more efficient to avoid 8401da177e4SLinus Torvalds * fragmenting write requests. 8411da177e4SLinus Torvalds */ 842ab0a3dbeSTrond Myklebust if (PageUptodate(page) && inode->i_flock == NULL && !(file->f_mode & O_SYNC)) { 8431da177e4SLinus Torvalds loff_t end_offs = i_size_read(inode) - 1; 8441da177e4SLinus Torvalds unsigned long end_index = end_offs >> PAGE_CACHE_SHIFT; 8451da177e4SLinus Torvalds 8461da177e4SLinus Torvalds count += offset; 8471da177e4SLinus Torvalds offset = 0; 8481da177e4SLinus Torvalds if (unlikely(end_offs < 0)) { 8491da177e4SLinus Torvalds /* Do nothing */ 8501da177e4SLinus Torvalds } else if (page->index == end_index) { 8511da177e4SLinus Torvalds unsigned int pglen; 8521da177e4SLinus Torvalds pglen = (unsigned int)(end_offs & (PAGE_CACHE_SIZE-1)) + 1; 8531da177e4SLinus Torvalds if (count < pglen) 8541da177e4SLinus Torvalds count = pglen; 8551da177e4SLinus Torvalds } else if (page->index < end_index) 8561da177e4SLinus Torvalds count = PAGE_CACHE_SIZE; 8571da177e4SLinus Torvalds } 8581da177e4SLinus Torvalds 8591da177e4SLinus Torvalds /* 8601da177e4SLinus Torvalds * Try to find an NFS request corresponding to this page 8611da177e4SLinus Torvalds * and update it. 8621da177e4SLinus Torvalds * If the existing request cannot be updated, we must flush 8631da177e4SLinus Torvalds * it out now. 8641da177e4SLinus Torvalds */ 8651da177e4SLinus Torvalds do { 8661da177e4SLinus Torvalds req = nfs_update_request(ctx, inode, page, offset, count); 8671da177e4SLinus Torvalds status = (IS_ERR(req)) ? PTR_ERR(req) : 0; 8681da177e4SLinus Torvalds if (status != -EBUSY) 8691da177e4SLinus Torvalds break; 8701da177e4SLinus Torvalds /* Request could not be updated. Flush it out and try again */ 8711da177e4SLinus Torvalds status = nfs_wb_page(inode, page); 8721da177e4SLinus Torvalds } while (status >= 0); 8731da177e4SLinus Torvalds if (status < 0) 8741da177e4SLinus Torvalds goto done; 8751da177e4SLinus Torvalds 8761da177e4SLinus Torvalds status = 0; 8771da177e4SLinus Torvalds 8781da177e4SLinus Torvalds /* Update file length */ 8791da177e4SLinus Torvalds nfs_grow_file(page, offset, count); 8801da177e4SLinus Torvalds /* Set the PG_uptodate flag? */ 8811da177e4SLinus Torvalds nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 8821da177e4SLinus Torvalds nfs_unlock_request(req); 8831da177e4SLinus Torvalds done: 8841da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage returns %d (isize %Ld)\n", 8851da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 8861da177e4SLinus Torvalds if (status < 0) 8871da177e4SLinus Torvalds ClearPageUptodate(page); 8881da177e4SLinus Torvalds return status; 8891da177e4SLinus Torvalds } 8901da177e4SLinus Torvalds 8911da177e4SLinus Torvalds static void nfs_writepage_release(struct nfs_page *req) 8921da177e4SLinus Torvalds { 8931da177e4SLinus Torvalds end_page_writeback(req->wb_page); 8941da177e4SLinus Torvalds 8951da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 8961da177e4SLinus Torvalds if (!PageError(req->wb_page)) { 8971da177e4SLinus Torvalds if (NFS_NEED_RESCHED(req)) { 8981da177e4SLinus Torvalds nfs_mark_request_dirty(req); 8991da177e4SLinus Torvalds goto out; 9001da177e4SLinus Torvalds } else if (NFS_NEED_COMMIT(req)) { 9011da177e4SLinus Torvalds nfs_mark_request_commit(req); 9021da177e4SLinus Torvalds goto out; 9031da177e4SLinus Torvalds } 9041da177e4SLinus Torvalds } 9051da177e4SLinus Torvalds nfs_inode_remove_request(req); 9061da177e4SLinus Torvalds 9071da177e4SLinus Torvalds out: 9081da177e4SLinus Torvalds nfs_clear_commit(req); 9091da177e4SLinus Torvalds nfs_clear_reschedule(req); 9101da177e4SLinus Torvalds #else 9111da177e4SLinus Torvalds nfs_inode_remove_request(req); 9121da177e4SLinus Torvalds #endif 913c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 9141da177e4SLinus Torvalds } 9151da177e4SLinus Torvalds 9161da177e4SLinus Torvalds static inline int flush_task_priority(int how) 9171da177e4SLinus Torvalds { 9181da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 9191da177e4SLinus Torvalds case FLUSH_HIGHPRI: 9201da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 9211da177e4SLinus Torvalds case FLUSH_LOWPRI: 9221da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 9231da177e4SLinus Torvalds } 9241da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 9251da177e4SLinus Torvalds } 9261da177e4SLinus Torvalds 9271da177e4SLinus Torvalds /* 9281da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 9291da177e4SLinus Torvalds */ 9301da177e4SLinus Torvalds static void nfs_write_rpcsetup(struct nfs_page *req, 9311da177e4SLinus Torvalds struct nfs_write_data *data, 932788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 9331da177e4SLinus Torvalds unsigned int count, unsigned int offset, 9341da177e4SLinus Torvalds int how) 9351da177e4SLinus Torvalds { 9361da177e4SLinus Torvalds struct inode *inode; 937788e7a89STrond Myklebust int flags; 9381da177e4SLinus Torvalds 9391da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 9401da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 9411da177e4SLinus Torvalds 9421da177e4SLinus Torvalds data->req = req; 9431da177e4SLinus Torvalds data->inode = inode = req->wb_context->dentry->d_inode; 9441da177e4SLinus Torvalds data->cred = req->wb_context->cred; 9451da177e4SLinus Torvalds 9461da177e4SLinus Torvalds data->args.fh = NFS_FH(inode); 9471da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 9481da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 9491da177e4SLinus Torvalds data->args.pages = data->pagevec; 9501da177e4SLinus Torvalds data->args.count = count; 9511da177e4SLinus Torvalds data->args.context = req->wb_context; 9521da177e4SLinus Torvalds 9531da177e4SLinus Torvalds data->res.fattr = &data->fattr; 9541da177e4SLinus Torvalds data->res.count = count; 9551da177e4SLinus Torvalds data->res.verf = &data->verf; 9560e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 9571da177e4SLinus Torvalds 958788e7a89STrond Myklebust /* Set up the initial task struct. */ 959788e7a89STrond Myklebust flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 960788e7a89STrond Myklebust rpc_init_task(&data->task, NFS_CLIENT(inode), flags, call_ops, data); 9611da177e4SLinus Torvalds NFS_PROTO(inode)->write_setup(data, how); 9621da177e4SLinus Torvalds 9631da177e4SLinus Torvalds data->task.tk_priority = flush_task_priority(how); 9641da177e4SLinus Torvalds data->task.tk_cookie = (unsigned long)inode; 9651da177e4SLinus Torvalds 9661da177e4SLinus Torvalds dprintk("NFS: %4d initiated write call (req %s/%Ld, %u bytes @ offset %Lu)\n", 9670bbacc40SChuck Lever data->task.tk_pid, 9681da177e4SLinus Torvalds inode->i_sb->s_id, 9691da177e4SLinus Torvalds (long long)NFS_FILEID(inode), 9701da177e4SLinus Torvalds count, 9711da177e4SLinus Torvalds (unsigned long long)data->args.offset); 9721da177e4SLinus Torvalds } 9731da177e4SLinus Torvalds 9741da177e4SLinus Torvalds static void nfs_execute_write(struct nfs_write_data *data) 9751da177e4SLinus Torvalds { 9761da177e4SLinus Torvalds struct rpc_clnt *clnt = NFS_CLIENT(data->inode); 9771da177e4SLinus Torvalds sigset_t oldset; 9781da177e4SLinus Torvalds 9791da177e4SLinus Torvalds rpc_clnt_sigmask(clnt, &oldset); 9801da177e4SLinus Torvalds lock_kernel(); 9811da177e4SLinus Torvalds rpc_execute(&data->task); 9821da177e4SLinus Torvalds unlock_kernel(); 9831da177e4SLinus Torvalds rpc_clnt_sigunmask(clnt, &oldset); 9841da177e4SLinus Torvalds } 9851da177e4SLinus Torvalds 9861da177e4SLinus Torvalds /* 9871da177e4SLinus Torvalds * Generate multiple small requests to write out a single 9881da177e4SLinus Torvalds * contiguous dirty area on one page. 9891da177e4SLinus Torvalds */ 9907d46a49fSTrond Myklebust static int nfs_flush_multi(struct inode *inode, struct list_head *head, int how) 9911da177e4SLinus Torvalds { 9921da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 9931da177e4SLinus Torvalds struct page *page = req->wb_page; 9941da177e4SLinus Torvalds struct nfs_write_data *data; 995*e9f7bee1STrond Myklebust size_t wsize = NFS_SERVER(inode)->wsize, nbytes; 996*e9f7bee1STrond Myklebust unsigned int offset; 9971da177e4SLinus Torvalds int requests = 0; 9981da177e4SLinus Torvalds LIST_HEAD(list); 9991da177e4SLinus Torvalds 10001da177e4SLinus Torvalds nfs_list_remove_request(req); 10011da177e4SLinus Torvalds 10021da177e4SLinus Torvalds nbytes = req->wb_bytes; 1003*e9f7bee1STrond Myklebust do { 1004*e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 1005*e9f7bee1STrond Myklebust 1006*e9f7bee1STrond Myklebust data = nfs_writedata_alloc(len); 10071da177e4SLinus Torvalds if (!data) 10081da177e4SLinus Torvalds goto out_bad; 10091da177e4SLinus Torvalds list_add(&data->pages, &list); 10101da177e4SLinus Torvalds requests++; 1011*e9f7bee1STrond Myklebust nbytes -= len; 1012*e9f7bee1STrond Myklebust } while (nbytes != 0); 10131da177e4SLinus Torvalds atomic_set(&req->wb_complete, requests); 10141da177e4SLinus Torvalds 10151da177e4SLinus Torvalds ClearPageError(page); 1016bb713d6dSTrond Myklebust set_page_writeback(page); 10171da177e4SLinus Torvalds offset = 0; 10181da177e4SLinus Torvalds nbytes = req->wb_bytes; 10191da177e4SLinus Torvalds do { 10201da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 10211da177e4SLinus Torvalds list_del_init(&data->pages); 10221da177e4SLinus Torvalds 10231da177e4SLinus Torvalds data->pagevec[0] = page; 10241da177e4SLinus Torvalds 10251da177e4SLinus Torvalds if (nbytes > wsize) { 1026788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 1027788e7a89STrond Myklebust wsize, offset, how); 10281da177e4SLinus Torvalds offset += wsize; 10291da177e4SLinus Torvalds nbytes -= wsize; 10301da177e4SLinus Torvalds } else { 1031788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 1032788e7a89STrond Myklebust nbytes, offset, how); 10331da177e4SLinus Torvalds nbytes = 0; 10341da177e4SLinus Torvalds } 10351da177e4SLinus Torvalds nfs_execute_write(data); 10361da177e4SLinus Torvalds } while (nbytes != 0); 10371da177e4SLinus Torvalds 10381da177e4SLinus Torvalds return 0; 10391da177e4SLinus Torvalds 10401da177e4SLinus Torvalds out_bad: 10411da177e4SLinus Torvalds while (!list_empty(&list)) { 10421da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 10431da177e4SLinus Torvalds list_del(&data->pages); 10441da177e4SLinus Torvalds nfs_writedata_free(data); 10451da177e4SLinus Torvalds } 10461da177e4SLinus Torvalds nfs_mark_request_dirty(req); 1047c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 10481da177e4SLinus Torvalds return -ENOMEM; 10491da177e4SLinus Torvalds } 10501da177e4SLinus Torvalds 10511da177e4SLinus Torvalds /* 10521da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 10531da177e4SLinus Torvalds * The page must have been locked by the caller. 10541da177e4SLinus Torvalds * 10551da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 10561da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 10571da177e4SLinus Torvalds * that has been written but not committed. 10581da177e4SLinus Torvalds */ 10597d46a49fSTrond Myklebust static int nfs_flush_one(struct inode *inode, struct list_head *head, int how) 10601da177e4SLinus Torvalds { 10611da177e4SLinus Torvalds struct nfs_page *req; 10621da177e4SLinus Torvalds struct page **pages; 10631da177e4SLinus Torvalds struct nfs_write_data *data; 10641da177e4SLinus Torvalds unsigned int count; 10651da177e4SLinus Torvalds 1066*e9f7bee1STrond Myklebust data = nfs_writedata_alloc(NFS_SERVER(inode)->wsize); 10671da177e4SLinus Torvalds if (!data) 10681da177e4SLinus Torvalds goto out_bad; 10691da177e4SLinus Torvalds 10701da177e4SLinus Torvalds pages = data->pagevec; 10711da177e4SLinus Torvalds count = 0; 10721da177e4SLinus Torvalds while (!list_empty(head)) { 10731da177e4SLinus Torvalds req = nfs_list_entry(head->next); 10741da177e4SLinus Torvalds nfs_list_remove_request(req); 10751da177e4SLinus Torvalds nfs_list_add_request(req, &data->pages); 10761da177e4SLinus Torvalds ClearPageError(req->wb_page); 1077bb713d6dSTrond Myklebust set_page_writeback(req->wb_page); 10781da177e4SLinus Torvalds *pages++ = req->wb_page; 10791da177e4SLinus Torvalds count += req->wb_bytes; 10801da177e4SLinus Torvalds } 10811da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 10821da177e4SLinus Torvalds 10831da177e4SLinus Torvalds /* Set up the argument struct */ 1084788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how); 10851da177e4SLinus Torvalds 10861da177e4SLinus Torvalds nfs_execute_write(data); 10871da177e4SLinus Torvalds return 0; 10881da177e4SLinus Torvalds out_bad: 10891da177e4SLinus Torvalds while (!list_empty(head)) { 10901da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 10911da177e4SLinus Torvalds nfs_list_remove_request(req); 10921da177e4SLinus Torvalds nfs_mark_request_dirty(req); 1093c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 10941da177e4SLinus Torvalds } 10951da177e4SLinus Torvalds return -ENOMEM; 10961da177e4SLinus Torvalds } 10971da177e4SLinus Torvalds 10987d46a49fSTrond Myklebust static int nfs_flush_list(struct inode *inode, struct list_head *head, int npages, int how) 10991da177e4SLinus Torvalds { 11001da177e4SLinus Torvalds LIST_HEAD(one_request); 11017d46a49fSTrond Myklebust int (*flush_one)(struct inode *, struct list_head *, int); 11021da177e4SLinus Torvalds struct nfs_page *req; 11037d46a49fSTrond Myklebust int wpages = NFS_SERVER(inode)->wpages; 11047d46a49fSTrond Myklebust int wsize = NFS_SERVER(inode)->wsize; 11057d46a49fSTrond Myklebust int error; 11061da177e4SLinus Torvalds 11077d46a49fSTrond Myklebust flush_one = nfs_flush_one; 11087d46a49fSTrond Myklebust if (wsize < PAGE_CACHE_SIZE) 11097d46a49fSTrond Myklebust flush_one = nfs_flush_multi; 11107d46a49fSTrond Myklebust /* For single writes, FLUSH_STABLE is more efficient */ 11117d46a49fSTrond Myklebust if (npages <= wpages && npages == NFS_I(inode)->npages 11127d46a49fSTrond Myklebust && nfs_list_entry(head->next)->wb_bytes <= wsize) 11137d46a49fSTrond Myklebust how |= FLUSH_STABLE; 11147d46a49fSTrond Myklebust 11157d46a49fSTrond Myklebust do { 11167d46a49fSTrond Myklebust nfs_coalesce_requests(head, &one_request, wpages); 11171da177e4SLinus Torvalds req = nfs_list_entry(one_request.next); 11187d46a49fSTrond Myklebust error = flush_one(inode, &one_request, how); 11191da177e4SLinus Torvalds if (error < 0) 11207d46a49fSTrond Myklebust goto out_err; 11217d46a49fSTrond Myklebust } while (!list_empty(head)); 11227d46a49fSTrond Myklebust return 0; 11237d46a49fSTrond Myklebust out_err: 11241da177e4SLinus Torvalds while (!list_empty(head)) { 11251da177e4SLinus Torvalds req = nfs_list_entry(head->next); 11261da177e4SLinus Torvalds nfs_list_remove_request(req); 11271da177e4SLinus Torvalds nfs_mark_request_dirty(req); 1128c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 11291da177e4SLinus Torvalds } 11301da177e4SLinus Torvalds return error; 11311da177e4SLinus Torvalds } 11321da177e4SLinus Torvalds 11331da177e4SLinus Torvalds /* 11341da177e4SLinus Torvalds * Handle a write reply that flushed part of a page. 11351da177e4SLinus Torvalds */ 1136788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) 11371da177e4SLinus Torvalds { 1138788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 11391da177e4SLinus Torvalds struct nfs_page *req = data->req; 11401da177e4SLinus Torvalds struct page *page = req->wb_page; 11411da177e4SLinus Torvalds 11421da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 11431da177e4SLinus Torvalds req->wb_context->dentry->d_inode->i_sb->s_id, 11441da177e4SLinus Torvalds (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 11451da177e4SLinus Torvalds req->wb_bytes, 11461da177e4SLinus Torvalds (long long)req_offset(req)); 11471da177e4SLinus Torvalds 1148788e7a89STrond Myklebust if (nfs_writeback_done(task, data) != 0) 1149788e7a89STrond Myklebust return; 1150788e7a89STrond Myklebust 1151788e7a89STrond Myklebust if (task->tk_status < 0) { 11521da177e4SLinus Torvalds ClearPageUptodate(page); 11531da177e4SLinus Torvalds SetPageError(page); 1154788e7a89STrond Myklebust req->wb_context->error = task->tk_status; 1155788e7a89STrond Myklebust dprintk(", error = %d\n", task->tk_status); 11561da177e4SLinus Torvalds } else { 11571da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 11581da177e4SLinus Torvalds if (data->verf.committed < NFS_FILE_SYNC) { 11591da177e4SLinus Torvalds if (!NFS_NEED_COMMIT(req)) { 11601da177e4SLinus Torvalds nfs_defer_commit(req); 11611da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 11621da177e4SLinus Torvalds dprintk(" defer commit\n"); 11631da177e4SLinus Torvalds } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { 11641da177e4SLinus Torvalds nfs_defer_reschedule(req); 11651da177e4SLinus Torvalds dprintk(" server reboot detected\n"); 11661da177e4SLinus Torvalds } 11671da177e4SLinus Torvalds } else 11681da177e4SLinus Torvalds #endif 11691da177e4SLinus Torvalds dprintk(" OK\n"); 11701da177e4SLinus Torvalds } 11711da177e4SLinus Torvalds 11721da177e4SLinus Torvalds if (atomic_dec_and_test(&req->wb_complete)) 11731da177e4SLinus Torvalds nfs_writepage_release(req); 11741da177e4SLinus Torvalds } 11751da177e4SLinus Torvalds 1176788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops = { 1177788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_partial, 1178788e7a89STrond Myklebust .rpc_release = nfs_writedata_release, 1179788e7a89STrond Myklebust }; 1180788e7a89STrond Myklebust 11811da177e4SLinus Torvalds /* 11821da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 11831da177e4SLinus Torvalds * 11841da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 11851da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 11861da177e4SLinus Torvalds * as the page has a write request pending. 11871da177e4SLinus Torvalds */ 1188788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) 11891da177e4SLinus Torvalds { 1190788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 11911da177e4SLinus Torvalds struct nfs_page *req; 11921da177e4SLinus Torvalds struct page *page; 11931da177e4SLinus Torvalds 1194788e7a89STrond Myklebust if (nfs_writeback_done(task, data) != 0) 1195788e7a89STrond Myklebust return; 1196788e7a89STrond Myklebust 11971da177e4SLinus Torvalds /* Update attributes as result of writeback. */ 11981da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 11991da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 12001da177e4SLinus Torvalds nfs_list_remove_request(req); 12011da177e4SLinus Torvalds page = req->wb_page; 12021da177e4SLinus Torvalds 12031da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 12041da177e4SLinus Torvalds req->wb_context->dentry->d_inode->i_sb->s_id, 12051da177e4SLinus Torvalds (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 12061da177e4SLinus Torvalds req->wb_bytes, 12071da177e4SLinus Torvalds (long long)req_offset(req)); 12081da177e4SLinus Torvalds 1209788e7a89STrond Myklebust if (task->tk_status < 0) { 12101da177e4SLinus Torvalds ClearPageUptodate(page); 12111da177e4SLinus Torvalds SetPageError(page); 1212788e7a89STrond Myklebust req->wb_context->error = task->tk_status; 12131da177e4SLinus Torvalds end_page_writeback(page); 12141da177e4SLinus Torvalds nfs_inode_remove_request(req); 1215788e7a89STrond Myklebust dprintk(", error = %d\n", task->tk_status); 12161da177e4SLinus Torvalds goto next; 12171da177e4SLinus Torvalds } 12181da177e4SLinus Torvalds end_page_writeback(page); 12191da177e4SLinus Torvalds 12201da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 12211da177e4SLinus Torvalds if (data->args.stable != NFS_UNSTABLE || data->verf.committed == NFS_FILE_SYNC) { 12221da177e4SLinus Torvalds nfs_inode_remove_request(req); 12231da177e4SLinus Torvalds dprintk(" OK\n"); 12241da177e4SLinus Torvalds goto next; 12251da177e4SLinus Torvalds } 12261da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 12271da177e4SLinus Torvalds nfs_mark_request_commit(req); 12281da177e4SLinus Torvalds dprintk(" marked for commit\n"); 12291da177e4SLinus Torvalds #else 12301da177e4SLinus Torvalds nfs_inode_remove_request(req); 12311da177e4SLinus Torvalds #endif 12321da177e4SLinus Torvalds next: 1233c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 12341da177e4SLinus Torvalds } 12351da177e4SLinus Torvalds } 12361da177e4SLinus Torvalds 1237788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = { 1238788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_full, 1239788e7a89STrond Myklebust .rpc_release = nfs_writedata_release, 1240788e7a89STrond Myklebust }; 1241788e7a89STrond Myklebust 1242788e7a89STrond Myklebust 12431da177e4SLinus Torvalds /* 12441da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 12451da177e4SLinus Torvalds */ 1246462d5b32SChuck Lever int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 12471da177e4SLinus Torvalds { 12481da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 12491da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1250788e7a89STrond Myklebust int status; 12511da177e4SLinus Torvalds 12521da177e4SLinus Torvalds dprintk("NFS: %4d nfs_writeback_done (status %d)\n", 12531da177e4SLinus Torvalds task->tk_pid, task->tk_status); 12541da177e4SLinus Torvalds 1255788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1256788e7a89STrond Myklebust status = NFS_PROTO(data->inode)->write_done(task, data); 1257788e7a89STrond Myklebust if (status != 0) 1258788e7a89STrond Myklebust return status; 125991d5b470SChuck Lever nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 126091d5b470SChuck Lever 12611da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 12621da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 12631da177e4SLinus Torvalds /* We tried a write call, but the server did not 12641da177e4SLinus Torvalds * commit data to stable storage even though we 12651da177e4SLinus Torvalds * requested it. 12661da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 12671da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 12681da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 12691da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 12701da177e4SLinus Torvalds */ 12711da177e4SLinus Torvalds static unsigned long complain; 12721da177e4SLinus Torvalds 12731da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 12741da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 12751da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 12761da177e4SLinus Torvalds NFS_SERVER(data->inode)->hostname, 12771da177e4SLinus Torvalds resp->verf->committed, argp->stable); 12781da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 12791da177e4SLinus Torvalds } 12801da177e4SLinus Torvalds } 12811da177e4SLinus Torvalds #endif 12821da177e4SLinus Torvalds /* Is this a short write? */ 12831da177e4SLinus Torvalds if (task->tk_status >= 0 && resp->count < argp->count) { 12841da177e4SLinus Torvalds static unsigned long complain; 12851da177e4SLinus Torvalds 128691d5b470SChuck Lever nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); 128791d5b470SChuck Lever 12881da177e4SLinus Torvalds /* Has the server at least made some progress? */ 12891da177e4SLinus Torvalds if (resp->count != 0) { 12901da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 12911da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 12921da177e4SLinus Torvalds /* Resend from where the server left off */ 12931da177e4SLinus Torvalds argp->offset += resp->count; 12941da177e4SLinus Torvalds argp->pgbase += resp->count; 12951da177e4SLinus Torvalds argp->count -= resp->count; 12961da177e4SLinus Torvalds } else { 12971da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 12981da177e4SLinus Torvalds * headaches in the case of a server crash. 12991da177e4SLinus Torvalds */ 13001da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 13011da177e4SLinus Torvalds } 13021da177e4SLinus Torvalds rpc_restart_call(task); 1303788e7a89STrond Myklebust return -EAGAIN; 13041da177e4SLinus Torvalds } 13051da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 13061da177e4SLinus Torvalds printk(KERN_WARNING 13071da177e4SLinus Torvalds "NFS: Server wrote zero bytes, expected %u.\n", 13081da177e4SLinus Torvalds argp->count); 13091da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 13101da177e4SLinus Torvalds } 13111da177e4SLinus Torvalds /* Can't do anything about it except throw an error. */ 13121da177e4SLinus Torvalds task->tk_status = -EIO; 13131da177e4SLinus Torvalds } 1314788e7a89STrond Myklebust return 0; 13151da177e4SLinus Torvalds } 13161da177e4SLinus Torvalds 13171da177e4SLinus Torvalds 13181da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 1319963d8fe5STrond Myklebust void nfs_commit_release(void *wdata) 13201da177e4SLinus Torvalds { 13211da177e4SLinus Torvalds nfs_commit_free(wdata); 13221da177e4SLinus Torvalds } 13231da177e4SLinus Torvalds 13241da177e4SLinus Torvalds /* 13251da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 13261da177e4SLinus Torvalds */ 13271da177e4SLinus Torvalds static void nfs_commit_rpcsetup(struct list_head *head, 1328788e7a89STrond Myklebust struct nfs_write_data *data, 1329788e7a89STrond Myklebust int how) 13301da177e4SLinus Torvalds { 13313da28eb1STrond Myklebust struct nfs_page *first; 13321da177e4SLinus Torvalds struct inode *inode; 1333788e7a89STrond Myklebust int flags; 13341da177e4SLinus Torvalds 13351da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 13361da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 13371da177e4SLinus Torvalds 13381da177e4SLinus Torvalds list_splice_init(head, &data->pages); 13391da177e4SLinus Torvalds first = nfs_list_entry(data->pages.next); 13401da177e4SLinus Torvalds inode = first->wb_context->dentry->d_inode; 13411da177e4SLinus Torvalds 13421da177e4SLinus Torvalds data->inode = inode; 13431da177e4SLinus Torvalds data->cred = first->wb_context->cred; 13441da177e4SLinus Torvalds 13451da177e4SLinus Torvalds data->args.fh = NFS_FH(data->inode); 13463da28eb1STrond Myklebust /* Note: we always request a commit of the entire inode */ 13473da28eb1STrond Myklebust data->args.offset = 0; 13483da28eb1STrond Myklebust data->args.count = 0; 13493da28eb1STrond Myklebust data->res.count = 0; 13501da177e4SLinus Torvalds data->res.fattr = &data->fattr; 13511da177e4SLinus Torvalds data->res.verf = &data->verf; 13520e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 13531da177e4SLinus Torvalds 1354788e7a89STrond Myklebust /* Set up the initial task struct. */ 1355788e7a89STrond Myklebust flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 1356788e7a89STrond Myklebust rpc_init_task(&data->task, NFS_CLIENT(inode), flags, &nfs_commit_ops, data); 13571da177e4SLinus Torvalds NFS_PROTO(inode)->commit_setup(data, how); 13581da177e4SLinus Torvalds 13591da177e4SLinus Torvalds data->task.tk_priority = flush_task_priority(how); 13601da177e4SLinus Torvalds data->task.tk_cookie = (unsigned long)inode; 13611da177e4SLinus Torvalds 13620bbacc40SChuck Lever dprintk("NFS: %4d initiated commit call\n", data->task.tk_pid); 13631da177e4SLinus Torvalds } 13641da177e4SLinus Torvalds 13651da177e4SLinus Torvalds /* 13661da177e4SLinus Torvalds * Commit dirty pages 13671da177e4SLinus Torvalds */ 13681da177e4SLinus Torvalds static int 136940859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how) 13701da177e4SLinus Torvalds { 13711da177e4SLinus Torvalds struct nfs_write_data *data; 13721da177e4SLinus Torvalds struct nfs_page *req; 13731da177e4SLinus Torvalds 1374*e9f7bee1STrond Myklebust data = nfs_commit_alloc(); 13751da177e4SLinus Torvalds 13761da177e4SLinus Torvalds if (!data) 13771da177e4SLinus Torvalds goto out_bad; 13781da177e4SLinus Torvalds 13791da177e4SLinus Torvalds /* Set up the argument struct */ 13801da177e4SLinus Torvalds nfs_commit_rpcsetup(head, data, how); 13811da177e4SLinus Torvalds 13821da177e4SLinus Torvalds nfs_execute_write(data); 13831da177e4SLinus Torvalds return 0; 13841da177e4SLinus Torvalds out_bad: 13851da177e4SLinus Torvalds while (!list_empty(head)) { 13861da177e4SLinus Torvalds req = nfs_list_entry(head->next); 13871da177e4SLinus Torvalds nfs_list_remove_request(req); 13881da177e4SLinus Torvalds nfs_mark_request_commit(req); 1389c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 139083715ad5STrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 13911da177e4SLinus Torvalds } 13921da177e4SLinus Torvalds return -ENOMEM; 13931da177e4SLinus Torvalds } 13941da177e4SLinus Torvalds 13951da177e4SLinus Torvalds /* 13961da177e4SLinus Torvalds * COMMIT call returned 13971da177e4SLinus Torvalds */ 1398788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 13991da177e4SLinus Torvalds { 1400963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 14011da177e4SLinus Torvalds struct nfs_page *req; 14021da177e4SLinus Torvalds 14031da177e4SLinus Torvalds dprintk("NFS: %4d nfs_commit_done (status %d)\n", 14041da177e4SLinus Torvalds task->tk_pid, task->tk_status); 14051da177e4SLinus Torvalds 1406788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1407788e7a89STrond Myklebust if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) 1408788e7a89STrond Myklebust return; 1409788e7a89STrond Myklebust 14101da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 14111da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 14121da177e4SLinus Torvalds nfs_list_remove_request(req); 1413fd39fc85SChristoph Lameter dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 14141da177e4SLinus Torvalds 14151da177e4SLinus Torvalds dprintk("NFS: commit (%s/%Ld %d@%Ld)", 14161da177e4SLinus Torvalds req->wb_context->dentry->d_inode->i_sb->s_id, 14171da177e4SLinus Torvalds (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 14181da177e4SLinus Torvalds req->wb_bytes, 14191da177e4SLinus Torvalds (long long)req_offset(req)); 14201da177e4SLinus Torvalds if (task->tk_status < 0) { 14211da177e4SLinus Torvalds req->wb_context->error = task->tk_status; 14221da177e4SLinus Torvalds nfs_inode_remove_request(req); 14231da177e4SLinus Torvalds dprintk(", error = %d\n", task->tk_status); 14241da177e4SLinus Torvalds goto next; 14251da177e4SLinus Torvalds } 14261da177e4SLinus Torvalds 14271da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 14281da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 14291da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 14301da177e4SLinus Torvalds /* We have a match */ 14311da177e4SLinus Torvalds nfs_inode_remove_request(req); 14321da177e4SLinus Torvalds dprintk(" OK\n"); 14331da177e4SLinus Torvalds goto next; 14341da177e4SLinus Torvalds } 14351da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 14361da177e4SLinus Torvalds dprintk(" mismatch\n"); 14371da177e4SLinus Torvalds nfs_mark_request_dirty(req); 14381da177e4SLinus Torvalds next: 1439c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 14401da177e4SLinus Torvalds } 14411da177e4SLinus Torvalds } 1442788e7a89STrond Myklebust 1443788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 1444788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1445788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1446788e7a89STrond Myklebust }; 1447c42de9ddSTrond Myklebust #else 1448c42de9ddSTrond Myklebust static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how) 1449c42de9ddSTrond Myklebust { 1450c42de9ddSTrond Myklebust return 0; 1451c42de9ddSTrond Myklebust } 14521da177e4SLinus Torvalds #endif 14531da177e4SLinus Torvalds 14541da177e4SLinus Torvalds static int nfs_flush_inode(struct inode *inode, unsigned long idx_start, 14551da177e4SLinus Torvalds unsigned int npages, int how) 14561da177e4SLinus Torvalds { 14571da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 14581da177e4SLinus Torvalds LIST_HEAD(head); 14597d46a49fSTrond Myklebust int res; 14601da177e4SLinus Torvalds 14611da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 14621da177e4SLinus Torvalds res = nfs_scan_dirty(inode, &head, idx_start, npages); 14631da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 1464ab0a3dbeSTrond Myklebust if (res) { 14657d46a49fSTrond Myklebust int error = nfs_flush_list(inode, &head, res, how); 14661da177e4SLinus Torvalds if (error < 0) 14671da177e4SLinus Torvalds return error; 14687d46a49fSTrond Myklebust } 14691da177e4SLinus Torvalds return res; 14701da177e4SLinus Torvalds } 14711da177e4SLinus Torvalds 14721da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 14733da28eb1STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 14741da177e4SLinus Torvalds { 14751da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 14761da177e4SLinus Torvalds LIST_HEAD(head); 14777d46a49fSTrond Myklebust int res; 14781da177e4SLinus Torvalds 14791da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 14803da28eb1STrond Myklebust res = nfs_scan_commit(inode, &head, 0, 0); 14813da28eb1STrond Myklebust spin_unlock(&nfsi->req_lock); 14821da177e4SLinus Torvalds if (res) { 14837d46a49fSTrond Myklebust int error = nfs_commit_list(inode, &head, how); 14841da177e4SLinus Torvalds if (error < 0) 14851da177e4SLinus Torvalds return error; 14863da28eb1STrond Myklebust } 14871da177e4SLinus Torvalds return res; 14881da177e4SLinus Torvalds } 14891da177e4SLinus Torvalds #endif 14901da177e4SLinus Torvalds 1491c42de9ddSTrond Myklebust int nfs_sync_inode_wait(struct inode *inode, unsigned long idx_start, 14921da177e4SLinus Torvalds unsigned int npages, int how) 14931da177e4SLinus Torvalds { 1494c42de9ddSTrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1495c42de9ddSTrond Myklebust LIST_HEAD(head); 149670b9ecbdSTrond Myklebust int nocommit = how & FLUSH_NOCOMMIT; 1497c42de9ddSTrond Myklebust int pages, ret; 14981da177e4SLinus Torvalds 1499c42de9ddSTrond Myklebust how &= ~FLUSH_NOCOMMIT; 1500c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 15011da177e4SLinus Torvalds do { 1502c42de9ddSTrond Myklebust ret = nfs_wait_on_requests_locked(inode, idx_start, npages); 1503c42de9ddSTrond Myklebust if (ret != 0) 1504c42de9ddSTrond Myklebust continue; 1505c42de9ddSTrond Myklebust pages = nfs_scan_dirty(inode, &head, idx_start, npages); 1506c42de9ddSTrond Myklebust if (pages != 0) { 1507c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 1508d2ccddf0STrond Myklebust if (how & FLUSH_INVALIDATE) 150983715ad5STrond Myklebust nfs_cancel_dirty_list(&head); 1510d2ccddf0STrond Myklebust else 1511c42de9ddSTrond Myklebust ret = nfs_flush_list(inode, &head, pages, how); 1512c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 151370b9ecbdSTrond Myklebust continue; 151470b9ecbdSTrond Myklebust } 1515c42de9ddSTrond Myklebust if (nocommit) 1516c42de9ddSTrond Myklebust break; 1517d2ccddf0STrond Myklebust pages = nfs_scan_commit(inode, &head, idx_start, npages); 1518c42de9ddSTrond Myklebust if (pages == 0) 1519c42de9ddSTrond Myklebust break; 1520d2ccddf0STrond Myklebust if (how & FLUSH_INVALIDATE) { 1521d2ccddf0STrond Myklebust spin_unlock(&nfsi->req_lock); 152283715ad5STrond Myklebust nfs_cancel_commit_list(&head); 1523d2ccddf0STrond Myklebust spin_lock(&nfsi->req_lock); 1524d2ccddf0STrond Myklebust continue; 1525d2ccddf0STrond Myklebust } 1526d2ccddf0STrond Myklebust pages += nfs_scan_commit(inode, &head, 0, 0); 1527c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 1528c42de9ddSTrond Myklebust ret = nfs_commit_list(inode, &head, how); 1529c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 1530c42de9ddSTrond Myklebust } while (ret >= 0); 1531c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 1532c42de9ddSTrond Myklebust return ret; 15331da177e4SLinus Torvalds } 15341da177e4SLinus Torvalds 1535f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 15361da177e4SLinus Torvalds { 15371da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 15381da177e4SLinus Torvalds sizeof(struct nfs_write_data), 15391da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 15401da177e4SLinus Torvalds NULL, NULL); 15411da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 15421da177e4SLinus Torvalds return -ENOMEM; 15431da177e4SLinus Torvalds 154493d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 15451da177e4SLinus Torvalds nfs_wdata_cachep); 15461da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 15471da177e4SLinus Torvalds return -ENOMEM; 15481da177e4SLinus Torvalds 154993d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 15501da177e4SLinus Torvalds nfs_wdata_cachep); 15511da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 15521da177e4SLinus Torvalds return -ENOMEM; 15531da177e4SLinus Torvalds 15541da177e4SLinus Torvalds return 0; 15551da177e4SLinus Torvalds } 15561da177e4SLinus Torvalds 1557266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 15581da177e4SLinus Torvalds { 15591da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 15601da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 15611da177e4SLinus Torvalds if (kmem_cache_destroy(nfs_wdata_cachep)) 15621da177e4SLinus Torvalds printk(KERN_INFO "nfs_write_data: not all structures were freed\n"); 15631da177e4SLinus Torvalds } 15641da177e4SLinus Torvalds 1565