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/writeback.h> 551da177e4SLinus Torvalds 561da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h> 571da177e4SLinus Torvalds #include <linux/nfs_fs.h> 581da177e4SLinus Torvalds #include <linux/nfs_mount.h> 591da177e4SLinus Torvalds #include <linux/nfs_page.h> 603fcfab16SAndrew Morton #include <linux/backing-dev.h> 613fcfab16SAndrew Morton 621da177e4SLinus Torvalds #include <asm/uaccess.h> 631da177e4SLinus Torvalds #include <linux/smp_lock.h> 641da177e4SLinus Torvalds 651da177e4SLinus Torvalds #include "delegation.h" 6649a70f27STrond Myklebust #include "internal.h" 6791d5b470SChuck Lever #include "iostat.h" 681da177e4SLinus Torvalds 691da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 701da177e4SLinus Torvalds 711da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 721da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 731da177e4SLinus Torvalds 741da177e4SLinus Torvalds /* 751da177e4SLinus Torvalds * Local function declarations 761da177e4SLinus Torvalds */ 771da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context*, 781da177e4SLinus Torvalds struct inode *, 791da177e4SLinus Torvalds struct page *, 801da177e4SLinus Torvalds unsigned int, unsigned int); 811da177e4SLinus Torvalds static int nfs_wait_on_write_congestion(struct address_space *, int); 821da177e4SLinus Torvalds static int nfs_wait_on_requests(struct inode *, unsigned long, unsigned int); 833f442547STrond Myklebust static long nfs_flush_mapping(struct address_space *mapping, struct writeback_control *wbc, int how); 841c75950bSTrond Myklebust static int nfs_wb_page_priority(struct inode *inode, struct page *page, int how); 85788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops; 86788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops; 87788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 881da177e4SLinus Torvalds 891da177e4SLinus Torvalds static kmem_cache_t *nfs_wdata_cachep; 903feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 911da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 921da177e4SLinus Torvalds 931da177e4SLinus Torvalds static DECLARE_WAIT_QUEUE_HEAD(nfs_write_congestion); 941da177e4SLinus Torvalds 95e9f7bee1STrond Myklebust struct nfs_write_data *nfs_commit_alloc(void) 961da177e4SLinus Torvalds { 971da177e4SLinus Torvalds struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, SLAB_NOFS); 9840859d7eSChuck Lever 991da177e4SLinus Torvalds if (p) { 1001da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 1011da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 1021da177e4SLinus Torvalds } 1031da177e4SLinus Torvalds return p; 1041da177e4SLinus Torvalds } 1051da177e4SLinus Torvalds 1068aca67f0STrond Myklebust void nfs_commit_rcu_free(struct rcu_head *head) 1071da177e4SLinus Torvalds { 1088aca67f0STrond Myklebust struct nfs_write_data *p = container_of(head, struct nfs_write_data, task.u.tk_rcu); 10940859d7eSChuck Lever if (p && (p->pagevec != &p->page_array[0])) 11040859d7eSChuck Lever kfree(p->pagevec); 1111da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 1121da177e4SLinus Torvalds } 1131da177e4SLinus Torvalds 1148aca67f0STrond Myklebust void nfs_commit_free(struct nfs_write_data *wdata) 1158aca67f0STrond Myklebust { 1168aca67f0STrond Myklebust call_rcu_bh(&wdata->task.u.tk_rcu, nfs_commit_rcu_free); 1178aca67f0STrond Myklebust } 1188aca67f0STrond Myklebust 119e9f7bee1STrond Myklebust struct nfs_write_data *nfs_writedata_alloc(size_t len) 1203feb2d49STrond Myklebust { 121e9f7bee1STrond Myklebust unsigned int pagecount = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1223feb2d49STrond Myklebust struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, SLAB_NOFS); 1233feb2d49STrond Myklebust 1243feb2d49STrond Myklebust if (p) { 1253feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 1263feb2d49STrond Myklebust INIT_LIST_HEAD(&p->pages); 127e9f7bee1STrond Myklebust p->npages = pagecount; 1280d0b5cb3SChuck Lever if (pagecount <= ARRAY_SIZE(p->page_array)) 1290d0b5cb3SChuck Lever p->pagevec = p->page_array; 1303feb2d49STrond Myklebust else { 1310d0b5cb3SChuck Lever p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); 1320d0b5cb3SChuck Lever if (!p->pagevec) { 1333feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 1343feb2d49STrond Myklebust p = NULL; 1353feb2d49STrond Myklebust } 1363feb2d49STrond Myklebust } 1373feb2d49STrond Myklebust } 1383feb2d49STrond Myklebust return p; 1393feb2d49STrond Myklebust } 1403feb2d49STrond Myklebust 1418aca67f0STrond Myklebust static void nfs_writedata_rcu_free(struct rcu_head *head) 1423feb2d49STrond Myklebust { 1438aca67f0STrond Myklebust struct nfs_write_data *p = container_of(head, struct nfs_write_data, task.u.tk_rcu); 1443feb2d49STrond Myklebust if (p && (p->pagevec != &p->page_array[0])) 1453feb2d49STrond Myklebust kfree(p->pagevec); 1463feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 1473feb2d49STrond Myklebust } 1483feb2d49STrond Myklebust 1498aca67f0STrond Myklebust static void nfs_writedata_free(struct nfs_write_data *wdata) 1508aca67f0STrond Myklebust { 1518aca67f0STrond Myklebust call_rcu_bh(&wdata->task.u.tk_rcu, nfs_writedata_rcu_free); 1528aca67f0STrond Myklebust } 1538aca67f0STrond Myklebust 154963d8fe5STrond Myklebust void nfs_writedata_release(void *wdata) 1551da177e4SLinus Torvalds { 1561da177e4SLinus Torvalds nfs_writedata_free(wdata); 1571da177e4SLinus Torvalds } 1581da177e4SLinus Torvalds 159277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request_locked(struct page *page) 160277459d2STrond Myklebust { 161277459d2STrond Myklebust struct nfs_page *req = NULL; 162277459d2STrond Myklebust 163277459d2STrond Myklebust if (PagePrivate(page)) { 164277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 165277459d2STrond Myklebust if (req != NULL) 166277459d2STrond Myklebust atomic_inc(&req->wb_count); 167277459d2STrond Myklebust } 168277459d2STrond Myklebust return req; 169277459d2STrond Myklebust } 170277459d2STrond Myklebust 171277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request(struct page *page) 172277459d2STrond Myklebust { 173277459d2STrond Myklebust struct nfs_page *req = NULL; 174277459d2STrond Myklebust spinlock_t *req_lock = &NFS_I(page->mapping->host)->req_lock; 175277459d2STrond Myklebust 176277459d2STrond Myklebust spin_lock(req_lock); 177277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 178277459d2STrond Myklebust spin_unlock(req_lock); 179277459d2STrond Myklebust return req; 180277459d2STrond Myklebust } 181277459d2STrond Myklebust 1821da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1831da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1841da177e4SLinus Torvalds { 1851da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 1861da177e4SLinus Torvalds loff_t end, i_size = i_size_read(inode); 1871da177e4SLinus Torvalds unsigned long end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 1881da177e4SLinus Torvalds 1891da177e4SLinus Torvalds if (i_size > 0 && page->index < end_index) 1901da177e4SLinus Torvalds return; 1911da177e4SLinus Torvalds end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); 1921da177e4SLinus Torvalds if (i_size >= end) 1931da177e4SLinus Torvalds return; 19491d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 1951da177e4SLinus Torvalds i_size_write(inode, end); 1961da177e4SLinus Torvalds } 1971da177e4SLinus Torvalds 1981da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 1991da177e4SLinus Torvalds * covers the full page. 2001da177e4SLinus Torvalds */ 2011da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 2021da177e4SLinus Torvalds { 2031da177e4SLinus Torvalds if (PageUptodate(page)) 2041da177e4SLinus Torvalds return; 2051da177e4SLinus Torvalds if (base != 0) 2061da177e4SLinus Torvalds return; 20749a70f27STrond Myklebust if (count != nfs_page_length(page)) 2081da177e4SLinus Torvalds return; 20949a70f27STrond Myklebust if (count != PAGE_CACHE_SIZE) 2101da177e4SLinus Torvalds memclear_highpage_flush(page, count, PAGE_CACHE_SIZE - count); 2111da177e4SLinus Torvalds SetPageUptodate(page); 2121da177e4SLinus Torvalds } 2131da177e4SLinus Torvalds 2141da177e4SLinus Torvalds /* 2151da177e4SLinus Torvalds * Write a page synchronously. 2161da177e4SLinus Torvalds * Offset is the data offset within the page. 2171da177e4SLinus Torvalds */ 2181da177e4SLinus Torvalds static int nfs_writepage_sync(struct nfs_open_context *ctx, struct inode *inode, 2191da177e4SLinus Torvalds struct page *page, unsigned int offset, unsigned int count, 2201da177e4SLinus Torvalds int how) 2211da177e4SLinus Torvalds { 2221da177e4SLinus Torvalds unsigned int wsize = NFS_SERVER(inode)->wsize; 2231da177e4SLinus Torvalds int result, written = 0; 2241da177e4SLinus Torvalds struct nfs_write_data *wdata; 2251da177e4SLinus Torvalds 226e9f7bee1STrond Myklebust wdata = nfs_writedata_alloc(wsize); 2271da177e4SLinus Torvalds if (!wdata) 2281da177e4SLinus Torvalds return -ENOMEM; 2291da177e4SLinus Torvalds 2301da177e4SLinus Torvalds wdata->flags = how; 2311da177e4SLinus Torvalds wdata->cred = ctx->cred; 2321da177e4SLinus Torvalds wdata->inode = inode; 2331da177e4SLinus Torvalds wdata->args.fh = NFS_FH(inode); 2341da177e4SLinus Torvalds wdata->args.context = ctx; 2351da177e4SLinus Torvalds wdata->args.pages = &page; 2361da177e4SLinus Torvalds wdata->args.stable = NFS_FILE_SYNC; 2371da177e4SLinus Torvalds wdata->args.pgbase = offset; 2381da177e4SLinus Torvalds wdata->args.count = wsize; 2391da177e4SLinus Torvalds wdata->res.fattr = &wdata->fattr; 2401da177e4SLinus Torvalds wdata->res.verf = &wdata->verf; 2411da177e4SLinus Torvalds 2421da177e4SLinus Torvalds dprintk("NFS: nfs_writepage_sync(%s/%Ld %d@%Ld)\n", 2431da177e4SLinus Torvalds inode->i_sb->s_id, 2441da177e4SLinus Torvalds (long long)NFS_FILEID(inode), 2451da177e4SLinus Torvalds count, (long long)(page_offset(page) + offset)); 2461da177e4SLinus Torvalds 247bb713d6dSTrond Myklebust set_page_writeback(page); 2481da177e4SLinus Torvalds nfs_begin_data_update(inode); 2491da177e4SLinus Torvalds do { 2501da177e4SLinus Torvalds if (count < wsize) 2511da177e4SLinus Torvalds wdata->args.count = count; 2521da177e4SLinus Torvalds wdata->args.offset = page_offset(page) + wdata->args.pgbase; 2531da177e4SLinus Torvalds 2541da177e4SLinus Torvalds result = NFS_PROTO(inode)->write(wdata); 2551da177e4SLinus Torvalds 2561da177e4SLinus Torvalds if (result < 0) { 2571da177e4SLinus Torvalds /* Must mark the page invalid after I/O error */ 2581da177e4SLinus Torvalds ClearPageUptodate(page); 2591da177e4SLinus Torvalds goto io_error; 2601da177e4SLinus Torvalds } 2611da177e4SLinus Torvalds if (result < wdata->args.count) 2621da177e4SLinus Torvalds printk(KERN_WARNING "NFS: short write, count=%u, result=%d\n", 2631da177e4SLinus Torvalds wdata->args.count, result); 2641da177e4SLinus Torvalds 2651da177e4SLinus Torvalds wdata->args.offset += result; 2661da177e4SLinus Torvalds wdata->args.pgbase += result; 2671da177e4SLinus Torvalds written += result; 2681da177e4SLinus Torvalds count -= result; 26991d5b470SChuck Lever nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, result); 2701da177e4SLinus Torvalds } while (count); 2711da177e4SLinus Torvalds /* Update file length */ 2721da177e4SLinus Torvalds nfs_grow_file(page, offset, written); 2731da177e4SLinus Torvalds /* Set the PG_uptodate flag? */ 2741da177e4SLinus Torvalds nfs_mark_uptodate(page, offset, written); 2751da177e4SLinus Torvalds 2761da177e4SLinus Torvalds if (PageError(page)) 2771da177e4SLinus Torvalds ClearPageError(page); 2781da177e4SLinus Torvalds 2791da177e4SLinus Torvalds io_error: 280951a143bSTrond Myklebust nfs_end_data_update(inode); 281bb713d6dSTrond Myklebust end_page_writeback(page); 2828aca67f0STrond Myklebust nfs_writedata_release(wdata); 2831da177e4SLinus Torvalds return written ? written : result; 2841da177e4SLinus Torvalds } 2851da177e4SLinus Torvalds 2861da177e4SLinus Torvalds static int nfs_writepage_async(struct nfs_open_context *ctx, 2871da177e4SLinus Torvalds struct inode *inode, struct page *page, 2881da177e4SLinus Torvalds unsigned int offset, unsigned int count) 2891da177e4SLinus Torvalds { 2901da177e4SLinus Torvalds struct nfs_page *req; 2911da177e4SLinus Torvalds 2921da177e4SLinus Torvalds req = nfs_update_request(ctx, inode, page, offset, count); 293abd3e641STrond Myklebust if (IS_ERR(req)) 294abd3e641STrond Myklebust return PTR_ERR(req); 2951da177e4SLinus Torvalds /* Update file length */ 2961da177e4SLinus Torvalds nfs_grow_file(page, offset, count); 2971da177e4SLinus Torvalds /* Set the PG_uptodate flag? */ 2981da177e4SLinus Torvalds nfs_mark_uptodate(page, offset, count); 2991da177e4SLinus Torvalds nfs_unlock_request(req); 300abd3e641STrond Myklebust return 0; 3011da177e4SLinus Torvalds } 3021da177e4SLinus Torvalds 3031da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 3041da177e4SLinus Torvalds { 3051da177e4SLinus Torvalds if (wbc->for_reclaim) 3061da177e4SLinus Torvalds return FLUSH_HIGHPRI; 3071da177e4SLinus Torvalds if (wbc->for_kupdate) 3081da177e4SLinus Torvalds return FLUSH_LOWPRI; 3091da177e4SLinus Torvalds return 0; 3101da177e4SLinus Torvalds } 3111da177e4SLinus Torvalds 3121da177e4SLinus Torvalds /* 3131da177e4SLinus Torvalds * Write an mmapped page to the server. 3141da177e4SLinus Torvalds */ 3151da177e4SLinus Torvalds int nfs_writepage(struct page *page, struct writeback_control *wbc) 3161da177e4SLinus Torvalds { 3171da177e4SLinus Torvalds struct nfs_open_context *ctx; 3181da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 31949a70f27STrond Myklebust unsigned offset; 3201da177e4SLinus Torvalds int inode_referenced = 0; 3211da177e4SLinus Torvalds int priority = wb_priority(wbc); 3221da177e4SLinus Torvalds int err; 3231da177e4SLinus Torvalds 32491d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 32591d5b470SChuck Lever nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 32691d5b470SChuck Lever 3271da177e4SLinus Torvalds /* 3281da177e4SLinus Torvalds * Note: We need to ensure that we have a reference to the inode 3291da177e4SLinus Torvalds * if we are to do asynchronous writes. If not, waiting 3301da177e4SLinus Torvalds * in nfs_wait_on_request() may deadlock with clear_inode(). 3311da177e4SLinus Torvalds * 3321da177e4SLinus Torvalds * If igrab() fails here, then it is in any case safe to 3331da177e4SLinus Torvalds * call nfs_wb_page(), since there will be no pending writes. 3341da177e4SLinus Torvalds */ 3351da177e4SLinus Torvalds if (igrab(inode) != 0) 3361da177e4SLinus Torvalds inode_referenced = 1; 3371da177e4SLinus Torvalds 3381da177e4SLinus Torvalds /* Ensure we've flushed out any previous writes */ 3391da177e4SLinus Torvalds nfs_wb_page_priority(inode, page, priority); 3401da177e4SLinus Torvalds 34149a70f27STrond Myklebust err = 0; 34249a70f27STrond Myklebust offset = nfs_page_length(page); 34349a70f27STrond Myklebust if (!offset) 3441da177e4SLinus Torvalds goto out; 34549a70f27STrond Myklebust 346d530838bSTrond Myklebust ctx = nfs_find_open_context(inode, NULL, FMODE_WRITE); 3471da177e4SLinus Torvalds if (ctx == NULL) { 3481da177e4SLinus Torvalds err = -EBADF; 3491da177e4SLinus Torvalds goto out; 3501da177e4SLinus Torvalds } 3511da177e4SLinus Torvalds lock_kernel(); 3521da177e4SLinus Torvalds if (!IS_SYNC(inode) && inode_referenced) { 3531da177e4SLinus Torvalds err = nfs_writepage_async(ctx, inode, page, 0, offset); 354abd3e641STrond Myklebust if (!wbc->for_writepages) 35528c6925fSTrond Myklebust nfs_flush_mapping(page->mapping, wbc, wb_priority(wbc)); 3561da177e4SLinus Torvalds } else { 3571da177e4SLinus Torvalds err = nfs_writepage_sync(ctx, inode, page, 0, 3581da177e4SLinus Torvalds offset, priority); 3591da177e4SLinus Torvalds if (err >= 0) { 3601da177e4SLinus Torvalds if (err != offset) 3611da177e4SLinus Torvalds redirty_page_for_writepage(wbc, page); 3621da177e4SLinus Torvalds err = 0; 3631da177e4SLinus Torvalds } 3641da177e4SLinus Torvalds } 3651da177e4SLinus Torvalds unlock_kernel(); 3661da177e4SLinus Torvalds put_nfs_open_context(ctx); 3671da177e4SLinus Torvalds out: 3681da177e4SLinus Torvalds unlock_page(page); 3691da177e4SLinus Torvalds if (inode_referenced) 3701da177e4SLinus Torvalds iput(inode); 3711da177e4SLinus Torvalds return err; 3721da177e4SLinus Torvalds } 3731da177e4SLinus Torvalds 3741da177e4SLinus Torvalds /* 3751da177e4SLinus Torvalds * Note: causes nfs_update_request() to block on the assumption 3761da177e4SLinus Torvalds * that the writeback is generated due to memory pressure. 3771da177e4SLinus Torvalds */ 3781da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 3791da177e4SLinus Torvalds { 3801da177e4SLinus Torvalds struct backing_dev_info *bdi = mapping->backing_dev_info; 3811da177e4SLinus Torvalds struct inode *inode = mapping->host; 3821da177e4SLinus Torvalds int err; 3831da177e4SLinus Torvalds 38491d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 38591d5b470SChuck Lever 3861da177e4SLinus Torvalds err = generic_writepages(mapping, wbc); 3871da177e4SLinus Torvalds if (err) 3881da177e4SLinus Torvalds return err; 3891da177e4SLinus Torvalds while (test_and_set_bit(BDI_write_congested, &bdi->state) != 0) { 3901da177e4SLinus Torvalds if (wbc->nonblocking) 3911da177e4SLinus Torvalds return 0; 3921da177e4SLinus Torvalds nfs_wait_on_write_congestion(mapping, 0); 3931da177e4SLinus Torvalds } 39428c6925fSTrond Myklebust err = nfs_flush_mapping(mapping, wbc, wb_priority(wbc)); 3951da177e4SLinus Torvalds if (err < 0) 3961da177e4SLinus Torvalds goto out; 39791d5b470SChuck Lever nfs_add_stats(inode, NFSIOS_WRITEPAGES, err); 3981da177e4SLinus Torvalds if (!wbc->nonblocking && wbc->sync_mode == WB_SYNC_ALL) { 3991da177e4SLinus Torvalds err = nfs_wait_on_requests(inode, 0, 0); 4001da177e4SLinus Torvalds if (err < 0) 4011da177e4SLinus Torvalds goto out; 4021da177e4SLinus Torvalds } 4033da28eb1STrond Myklebust err = nfs_commit_inode(inode, wb_priority(wbc)); 4043f442547STrond Myklebust if (err > 0) 4051da177e4SLinus Torvalds err = 0; 4061da177e4SLinus Torvalds out: 4071da177e4SLinus Torvalds clear_bit(BDI_write_congested, &bdi->state); 4081da177e4SLinus Torvalds wake_up_all(&nfs_write_congestion); 4093fcfab16SAndrew Morton congestion_end(WRITE); 4101da177e4SLinus Torvalds return err; 4111da177e4SLinus Torvalds } 4121da177e4SLinus Torvalds 4131da177e4SLinus Torvalds /* 4141da177e4SLinus Torvalds * Insert a write request into an inode 4151da177e4SLinus Torvalds */ 4161da177e4SLinus Torvalds static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 4171da177e4SLinus Torvalds { 4181da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4191da177e4SLinus Torvalds int error; 4201da177e4SLinus Torvalds 4211da177e4SLinus Torvalds error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); 4221da177e4SLinus Torvalds BUG_ON(error == -EEXIST); 4231da177e4SLinus Torvalds if (error) 4241da177e4SLinus Torvalds return error; 4251da177e4SLinus Torvalds if (!nfsi->npages) { 4261da177e4SLinus Torvalds igrab(inode); 4271da177e4SLinus Torvalds nfs_begin_data_update(inode); 4281da177e4SLinus Torvalds if (nfs_have_delegation(inode, FMODE_WRITE)) 4291da177e4SLinus Torvalds nfsi->change_attr++; 4301da177e4SLinus Torvalds } 431deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 432277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 4331da177e4SLinus Torvalds nfsi->npages++; 4341da177e4SLinus Torvalds atomic_inc(&req->wb_count); 4351da177e4SLinus Torvalds return 0; 4361da177e4SLinus Torvalds } 4371da177e4SLinus Torvalds 4381da177e4SLinus Torvalds /* 4391da177e4SLinus Torvalds * Insert a write request into an inode 4401da177e4SLinus Torvalds */ 4411da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 4421da177e4SLinus Torvalds { 4431da177e4SLinus Torvalds struct inode *inode = req->wb_context->dentry->d_inode; 4441da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4451da177e4SLinus Torvalds 4461da177e4SLinus Torvalds BUG_ON (!NFS_WBACK_BUSY(req)); 4471da177e4SLinus Torvalds 4481da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 449277459d2STrond Myklebust set_page_private(req->wb_page, 0); 450deb7d638STrond Myklebust ClearPagePrivate(req->wb_page); 4511da177e4SLinus Torvalds radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); 4521da177e4SLinus Torvalds nfsi->npages--; 4531da177e4SLinus Torvalds if (!nfsi->npages) { 4541da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 455951a143bSTrond Myklebust nfs_end_data_update(inode); 4561da177e4SLinus Torvalds iput(inode); 4571da177e4SLinus Torvalds } else 4581da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 4591da177e4SLinus Torvalds nfs_clear_request(req); 4601da177e4SLinus Torvalds nfs_release_request(req); 4611da177e4SLinus Torvalds } 4621da177e4SLinus Torvalds 4631da177e4SLinus Torvalds /* 4641da177e4SLinus Torvalds * Add a request to the inode's dirty list. 4651da177e4SLinus Torvalds */ 4661da177e4SLinus Torvalds static void 4671da177e4SLinus Torvalds nfs_mark_request_dirty(struct nfs_page *req) 4681da177e4SLinus Torvalds { 4691da177e4SLinus Torvalds struct inode *inode = req->wb_context->dentry->d_inode; 4701da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4711da177e4SLinus Torvalds 4721da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 4733da28eb1STrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, 4743da28eb1STrond Myklebust req->wb_index, NFS_PAGE_TAG_DIRTY); 4751da177e4SLinus Torvalds nfs_list_add_request(req, &nfsi->dirty); 4761da177e4SLinus Torvalds nfsi->ndirty++; 4771da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 478b1e7a8fdSChristoph Lameter inc_zone_page_state(req->wb_page, NR_FILE_DIRTY); 4791da177e4SLinus Torvalds mark_inode_dirty(inode); 4801da177e4SLinus Torvalds } 4811da177e4SLinus Torvalds 4821da177e4SLinus Torvalds /* 4831da177e4SLinus Torvalds * Check if a request is dirty 4841da177e4SLinus Torvalds */ 4851da177e4SLinus Torvalds static inline int 4861da177e4SLinus Torvalds nfs_dirty_request(struct nfs_page *req) 4871da177e4SLinus Torvalds { 4881da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode); 4891da177e4SLinus Torvalds return !list_empty(&req->wb_list) && req->wb_list_head == &nfsi->dirty; 4901da177e4SLinus Torvalds } 4911da177e4SLinus Torvalds 4921da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 4931da177e4SLinus Torvalds /* 4941da177e4SLinus Torvalds * Add a request to the inode's commit list. 4951da177e4SLinus Torvalds */ 4961da177e4SLinus Torvalds static void 4971da177e4SLinus Torvalds nfs_mark_request_commit(struct nfs_page *req) 4981da177e4SLinus Torvalds { 4991da177e4SLinus Torvalds struct inode *inode = req->wb_context->dentry->d_inode; 5001da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5011da177e4SLinus Torvalds 5021da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 5031da177e4SLinus Torvalds nfs_list_add_request(req, &nfsi->commit); 5041da177e4SLinus Torvalds nfsi->ncommit++; 5051da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 506fd39fc85SChristoph Lameter inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 5071da177e4SLinus Torvalds mark_inode_dirty(inode); 5081da177e4SLinus Torvalds } 5091da177e4SLinus Torvalds #endif 5101da177e4SLinus Torvalds 5111da177e4SLinus Torvalds /* 5121da177e4SLinus Torvalds * Wait for a request to complete. 5131da177e4SLinus Torvalds * 5141da177e4SLinus Torvalds * Interruptible by signals only if mounted with intr flag. 5151da177e4SLinus Torvalds */ 516c42de9ddSTrond Myklebust static int nfs_wait_on_requests_locked(struct inode *inode, unsigned long idx_start, unsigned int npages) 5171da177e4SLinus Torvalds { 5181da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5191da177e4SLinus Torvalds struct nfs_page *req; 5201da177e4SLinus Torvalds unsigned long idx_end, next; 5211da177e4SLinus Torvalds unsigned int res = 0; 5221da177e4SLinus Torvalds int error; 5231da177e4SLinus Torvalds 5241da177e4SLinus Torvalds if (npages == 0) 5251da177e4SLinus Torvalds idx_end = ~0; 5261da177e4SLinus Torvalds else 5271da177e4SLinus Torvalds idx_end = idx_start + npages - 1; 5281da177e4SLinus Torvalds 5291da177e4SLinus Torvalds next = idx_start; 530c6a556b8STrond Myklebust while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_WRITEBACK)) { 5311da177e4SLinus Torvalds if (req->wb_index > idx_end) 5321da177e4SLinus Torvalds break; 5331da177e4SLinus Torvalds 5341da177e4SLinus Torvalds next = req->wb_index + 1; 535c6a556b8STrond Myklebust BUG_ON(!NFS_WBACK_BUSY(req)); 5361da177e4SLinus Torvalds 5371da177e4SLinus Torvalds atomic_inc(&req->wb_count); 5381da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 5391da177e4SLinus Torvalds error = nfs_wait_on_request(req); 5401da177e4SLinus Torvalds nfs_release_request(req); 541c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 5421da177e4SLinus Torvalds if (error < 0) 5431da177e4SLinus Torvalds return error; 5441da177e4SLinus Torvalds res++; 5451da177e4SLinus Torvalds } 5461da177e4SLinus Torvalds return res; 5471da177e4SLinus Torvalds } 5481da177e4SLinus Torvalds 549c42de9ddSTrond Myklebust static int nfs_wait_on_requests(struct inode *inode, unsigned long idx_start, unsigned int npages) 550c42de9ddSTrond Myklebust { 551c42de9ddSTrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 552c42de9ddSTrond Myklebust int ret; 553c42de9ddSTrond Myklebust 554c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 555c42de9ddSTrond Myklebust ret = nfs_wait_on_requests_locked(inode, idx_start, npages); 556c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 557c42de9ddSTrond Myklebust return ret; 558c42de9ddSTrond Myklebust } 559c42de9ddSTrond Myklebust 56083715ad5STrond Myklebust static void nfs_cancel_dirty_list(struct list_head *head) 561d2ccddf0STrond Myklebust { 562d2ccddf0STrond Myklebust struct nfs_page *req; 563d2ccddf0STrond Myklebust while(!list_empty(head)) { 564d2ccddf0STrond Myklebust req = nfs_list_entry(head->next); 565d2ccddf0STrond Myklebust nfs_list_remove_request(req); 566d2ccddf0STrond Myklebust nfs_inode_remove_request(req); 567d2ccddf0STrond Myklebust nfs_clear_page_writeback(req); 568d2ccddf0STrond Myklebust } 569d2ccddf0STrond Myklebust } 570d2ccddf0STrond Myklebust 57183715ad5STrond Myklebust static void nfs_cancel_commit_list(struct list_head *head) 57283715ad5STrond Myklebust { 57383715ad5STrond Myklebust struct nfs_page *req; 57483715ad5STrond Myklebust 57583715ad5STrond Myklebust while(!list_empty(head)) { 57683715ad5STrond Myklebust req = nfs_list_entry(head->next); 577b6dff26aSTrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 57883715ad5STrond Myklebust nfs_list_remove_request(req); 57983715ad5STrond Myklebust nfs_inode_remove_request(req); 580b6dff26aSTrond Myklebust nfs_unlock_request(req); 58183715ad5STrond Myklebust } 58283715ad5STrond Myklebust } 58383715ad5STrond Myklebust 5841da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 5851da177e4SLinus Torvalds /* 5861da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 5871da177e4SLinus Torvalds * @inode: NFS inode to scan 5881da177e4SLinus Torvalds * @dst: destination list 5891da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 5901da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 5911da177e4SLinus Torvalds * 5921da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 5931da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 5941da177e4SLinus Torvalds */ 5951da177e4SLinus Torvalds static int 5961da177e4SLinus Torvalds nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) 5971da177e4SLinus Torvalds { 5981da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5993da28eb1STrond Myklebust int res = 0; 6003da28eb1STrond Myklebust 6013da28eb1STrond Myklebust if (nfsi->ncommit != 0) { 602d2ccddf0STrond Myklebust res = nfs_scan_list(nfsi, &nfsi->commit, dst, idx_start, npages); 6031da177e4SLinus Torvalds nfsi->ncommit -= res; 6041da177e4SLinus Torvalds if ((nfsi->ncommit == 0) != list_empty(&nfsi->commit)) 6051da177e4SLinus Torvalds printk(KERN_ERR "NFS: desynchronized value of nfs_i.ncommit.\n"); 6063da28eb1STrond Myklebust } 6071da177e4SLinus Torvalds return res; 6081da177e4SLinus Torvalds } 609c42de9ddSTrond Myklebust #else 610c42de9ddSTrond Myklebust static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) 611c42de9ddSTrond Myklebust { 612c42de9ddSTrond Myklebust return 0; 613c42de9ddSTrond Myklebust } 6141da177e4SLinus Torvalds #endif 6151da177e4SLinus Torvalds 6161da177e4SLinus Torvalds static int nfs_wait_on_write_congestion(struct address_space *mapping, int intr) 6171da177e4SLinus Torvalds { 6181da177e4SLinus Torvalds struct backing_dev_info *bdi = mapping->backing_dev_info; 6191da177e4SLinus Torvalds DEFINE_WAIT(wait); 6201da177e4SLinus Torvalds int ret = 0; 6211da177e4SLinus Torvalds 6221da177e4SLinus Torvalds might_sleep(); 6231da177e4SLinus Torvalds 6241da177e4SLinus Torvalds if (!bdi_write_congested(bdi)) 6251da177e4SLinus Torvalds return 0; 62691d5b470SChuck Lever 62791d5b470SChuck Lever nfs_inc_stats(mapping->host, NFSIOS_CONGESTIONWAIT); 62891d5b470SChuck Lever 6291da177e4SLinus Torvalds if (intr) { 6301da177e4SLinus Torvalds struct rpc_clnt *clnt = NFS_CLIENT(mapping->host); 6311da177e4SLinus Torvalds sigset_t oldset; 6321da177e4SLinus Torvalds 6331da177e4SLinus Torvalds rpc_clnt_sigmask(clnt, &oldset); 6341da177e4SLinus Torvalds prepare_to_wait(&nfs_write_congestion, &wait, TASK_INTERRUPTIBLE); 6351da177e4SLinus Torvalds if (bdi_write_congested(bdi)) { 6361da177e4SLinus Torvalds if (signalled()) 6371da177e4SLinus Torvalds ret = -ERESTARTSYS; 6381da177e4SLinus Torvalds else 6391da177e4SLinus Torvalds schedule(); 6401da177e4SLinus Torvalds } 6411da177e4SLinus Torvalds rpc_clnt_sigunmask(clnt, &oldset); 6421da177e4SLinus Torvalds } else { 6431da177e4SLinus Torvalds prepare_to_wait(&nfs_write_congestion, &wait, TASK_UNINTERRUPTIBLE); 6441da177e4SLinus Torvalds if (bdi_write_congested(bdi)) 6451da177e4SLinus Torvalds schedule(); 6461da177e4SLinus Torvalds } 6471da177e4SLinus Torvalds finish_wait(&nfs_write_congestion, &wait); 6481da177e4SLinus Torvalds return ret; 6491da177e4SLinus Torvalds } 6501da177e4SLinus Torvalds 6511da177e4SLinus Torvalds 6521da177e4SLinus Torvalds /* 6531da177e4SLinus Torvalds * Try to update any existing write request, or create one if there is none. 6541da177e4SLinus Torvalds * In order to match, the request's credentials must match those of 6551da177e4SLinus Torvalds * the calling process. 6561da177e4SLinus Torvalds * 6571da177e4SLinus Torvalds * Note: Should always be called with the Page Lock held! 6581da177e4SLinus Torvalds */ 6591da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx, 6601da177e4SLinus Torvalds struct inode *inode, struct page *page, 6611da177e4SLinus Torvalds unsigned int offset, unsigned int bytes) 6621da177e4SLinus Torvalds { 6631da177e4SLinus Torvalds struct nfs_server *server = NFS_SERVER(inode); 6641da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 6651da177e4SLinus Torvalds struct nfs_page *req, *new = NULL; 6661da177e4SLinus Torvalds unsigned long rqend, end; 6671da177e4SLinus Torvalds 6681da177e4SLinus Torvalds end = offset + bytes; 6691da177e4SLinus Torvalds 6701da177e4SLinus Torvalds if (nfs_wait_on_write_congestion(page->mapping, server->flags & NFS_MOUNT_INTR)) 6711da177e4SLinus Torvalds return ERR_PTR(-ERESTARTSYS); 6721da177e4SLinus Torvalds for (;;) { 6731da177e4SLinus Torvalds /* Loop over all inode entries and see if we find 6741da177e4SLinus Torvalds * A request for the page we wish to update 6751da177e4SLinus Torvalds */ 6761da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 677277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 6781da177e4SLinus Torvalds if (req) { 6791da177e4SLinus Torvalds if (!nfs_lock_request_dontget(req)) { 6801da177e4SLinus Torvalds int error; 681277459d2STrond Myklebust 6821da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 6831da177e4SLinus Torvalds error = nfs_wait_on_request(req); 6841da177e4SLinus Torvalds nfs_release_request(req); 6851dd594b2SNeil Brown if (error < 0) { 6861dd594b2SNeil Brown if (new) 6871dd594b2SNeil Brown nfs_release_request(new); 6881da177e4SLinus Torvalds return ERR_PTR(error); 6891dd594b2SNeil Brown } 6901da177e4SLinus Torvalds continue; 6911da177e4SLinus Torvalds } 6921da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 6931da177e4SLinus Torvalds if (new) 6941da177e4SLinus Torvalds nfs_release_request(new); 6951da177e4SLinus Torvalds break; 6961da177e4SLinus Torvalds } 6971da177e4SLinus Torvalds 6981da177e4SLinus Torvalds if (new) { 6991da177e4SLinus Torvalds int error; 7001da177e4SLinus Torvalds nfs_lock_request_dontget(new); 7011da177e4SLinus Torvalds error = nfs_inode_add_request(inode, new); 7021da177e4SLinus Torvalds if (error) { 7031da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 7041da177e4SLinus Torvalds nfs_unlock_request(new); 7051da177e4SLinus Torvalds return ERR_PTR(error); 7061da177e4SLinus Torvalds } 7071da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 7081da177e4SLinus Torvalds nfs_mark_request_dirty(new); 7091da177e4SLinus Torvalds return new; 7101da177e4SLinus Torvalds } 7111da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 7121da177e4SLinus Torvalds 7131da177e4SLinus Torvalds new = nfs_create_request(ctx, inode, page, offset, bytes); 7141da177e4SLinus Torvalds if (IS_ERR(new)) 7151da177e4SLinus Torvalds return new; 7161da177e4SLinus Torvalds } 7171da177e4SLinus Torvalds 7181da177e4SLinus Torvalds /* We have a request for our page. 7191da177e4SLinus Torvalds * If the creds don't match, or the 7201da177e4SLinus Torvalds * page addresses don't match, 7211da177e4SLinus Torvalds * tell the caller to wait on the conflicting 7221da177e4SLinus Torvalds * request. 7231da177e4SLinus Torvalds */ 7241da177e4SLinus Torvalds rqend = req->wb_offset + req->wb_bytes; 7251da177e4SLinus Torvalds if (req->wb_context != ctx 7261da177e4SLinus Torvalds || req->wb_page != page 7271da177e4SLinus Torvalds || !nfs_dirty_request(req) 7281da177e4SLinus Torvalds || offset > rqend || end < req->wb_offset) { 7291da177e4SLinus Torvalds nfs_unlock_request(req); 7301da177e4SLinus Torvalds return ERR_PTR(-EBUSY); 7311da177e4SLinus Torvalds } 7321da177e4SLinus Torvalds 7331da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 7341da177e4SLinus Torvalds if (offset < req->wb_offset) { 7351da177e4SLinus Torvalds req->wb_offset = offset; 7361da177e4SLinus Torvalds req->wb_pgbase = offset; 7371da177e4SLinus Torvalds req->wb_bytes = rqend - req->wb_offset; 7381da177e4SLinus Torvalds } 7391da177e4SLinus Torvalds 7401da177e4SLinus Torvalds if (end > rqend) 7411da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 7421da177e4SLinus Torvalds 7431da177e4SLinus Torvalds return req; 7441da177e4SLinus Torvalds } 7451da177e4SLinus Torvalds 7461da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 7471da177e4SLinus Torvalds { 7481da177e4SLinus Torvalds struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; 7491da177e4SLinus Torvalds struct nfs_page *req; 7501da177e4SLinus Torvalds int status = 0; 7511da177e4SLinus Torvalds /* 7521da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 7531da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 7541da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 7551da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 7561da177e4SLinus Torvalds * Also do the same if we find a request from an existing 7571da177e4SLinus Torvalds * dropped page. 7581da177e4SLinus Torvalds */ 759277459d2STrond Myklebust req = nfs_page_find_request(page); 760277459d2STrond Myklebust if (req != NULL) { 761277459d2STrond Myklebust int do_flush = req->wb_page != page || req->wb_context != ctx; 762277459d2STrond Myklebust 7631da177e4SLinus Torvalds nfs_release_request(req); 764277459d2STrond Myklebust if (do_flush) 765277459d2STrond Myklebust status = nfs_wb_page(page->mapping->host, page); 7661da177e4SLinus Torvalds } 7671da177e4SLinus Torvalds return (status < 0) ? status : 0; 7681da177e4SLinus Torvalds } 7691da177e4SLinus Torvalds 7701da177e4SLinus Torvalds /* 7711da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 7721da177e4SLinus Torvalds * 7731da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 7741da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 7751da177e4SLinus Torvalds */ 7761da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 7771da177e4SLinus Torvalds unsigned int offset, unsigned int count) 7781da177e4SLinus Torvalds { 7791da177e4SLinus Torvalds struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; 7801da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 7811da177e4SLinus Torvalds struct nfs_page *req; 7821da177e4SLinus Torvalds int status = 0; 7831da177e4SLinus Torvalds 78491d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 78591d5b470SChuck Lever 7861da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage(%s/%s %d@%Ld)\n", 7870bbacc40SChuck Lever file->f_dentry->d_parent->d_name.name, 7880bbacc40SChuck Lever file->f_dentry->d_name.name, count, 7890bbacc40SChuck Lever (long long)(page_offset(page) +offset)); 7901da177e4SLinus Torvalds 7911da177e4SLinus Torvalds if (IS_SYNC(inode)) { 7921da177e4SLinus Torvalds status = nfs_writepage_sync(ctx, inode, page, offset, count, 0); 7931da177e4SLinus Torvalds if (status > 0) { 7941da177e4SLinus Torvalds if (offset == 0 && status == PAGE_CACHE_SIZE) 7951da177e4SLinus Torvalds SetPageUptodate(page); 7961da177e4SLinus Torvalds return 0; 7971da177e4SLinus Torvalds } 7981da177e4SLinus Torvalds return status; 7991da177e4SLinus Torvalds } 8001da177e4SLinus Torvalds 8011da177e4SLinus Torvalds /* If we're not using byte range locks, and we know the page 8021da177e4SLinus Torvalds * is entirely in cache, it may be more efficient to avoid 8031da177e4SLinus Torvalds * fragmenting write requests. 8041da177e4SLinus Torvalds */ 805ab0a3dbeSTrond Myklebust if (PageUptodate(page) && inode->i_flock == NULL && !(file->f_mode & O_SYNC)) { 80649a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 8071da177e4SLinus Torvalds offset = 0; 8081da177e4SLinus Torvalds } 8091da177e4SLinus Torvalds 8101da177e4SLinus Torvalds /* 8111da177e4SLinus Torvalds * Try to find an NFS request corresponding to this page 8121da177e4SLinus Torvalds * and update it. 8131da177e4SLinus Torvalds * If the existing request cannot be updated, we must flush 8141da177e4SLinus Torvalds * it out now. 8151da177e4SLinus Torvalds */ 8161da177e4SLinus Torvalds do { 8171da177e4SLinus Torvalds req = nfs_update_request(ctx, inode, page, offset, count); 8181da177e4SLinus Torvalds status = (IS_ERR(req)) ? PTR_ERR(req) : 0; 8191da177e4SLinus Torvalds if (status != -EBUSY) 8201da177e4SLinus Torvalds break; 8211da177e4SLinus Torvalds /* Request could not be updated. Flush it out and try again */ 8221da177e4SLinus Torvalds status = nfs_wb_page(inode, page); 8231da177e4SLinus Torvalds } while (status >= 0); 8241da177e4SLinus Torvalds if (status < 0) 8251da177e4SLinus Torvalds goto done; 8261da177e4SLinus Torvalds 8271da177e4SLinus Torvalds status = 0; 8281da177e4SLinus Torvalds 8291da177e4SLinus Torvalds /* Update file length */ 8301da177e4SLinus Torvalds nfs_grow_file(page, offset, count); 8311da177e4SLinus Torvalds /* Set the PG_uptodate flag? */ 8321da177e4SLinus Torvalds nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 8331da177e4SLinus Torvalds nfs_unlock_request(req); 8341da177e4SLinus Torvalds done: 8351da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage returns %d (isize %Ld)\n", 8361da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 8371da177e4SLinus Torvalds if (status < 0) 8381da177e4SLinus Torvalds ClearPageUptodate(page); 8391da177e4SLinus Torvalds return status; 8401da177e4SLinus Torvalds } 8411da177e4SLinus Torvalds 8421da177e4SLinus Torvalds static void nfs_writepage_release(struct nfs_page *req) 8431da177e4SLinus Torvalds { 8441da177e4SLinus Torvalds end_page_writeback(req->wb_page); 8451da177e4SLinus Torvalds 8461da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 8471da177e4SLinus Torvalds if (!PageError(req->wb_page)) { 8481da177e4SLinus Torvalds if (NFS_NEED_RESCHED(req)) { 8491da177e4SLinus Torvalds nfs_mark_request_dirty(req); 8501da177e4SLinus Torvalds goto out; 8511da177e4SLinus Torvalds } else if (NFS_NEED_COMMIT(req)) { 8521da177e4SLinus Torvalds nfs_mark_request_commit(req); 8531da177e4SLinus Torvalds goto out; 8541da177e4SLinus Torvalds } 8551da177e4SLinus Torvalds } 8561da177e4SLinus Torvalds nfs_inode_remove_request(req); 8571da177e4SLinus Torvalds 8581da177e4SLinus Torvalds out: 8591da177e4SLinus Torvalds nfs_clear_commit(req); 8601da177e4SLinus Torvalds nfs_clear_reschedule(req); 8611da177e4SLinus Torvalds #else 8621da177e4SLinus Torvalds nfs_inode_remove_request(req); 8631da177e4SLinus Torvalds #endif 864c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 8651da177e4SLinus Torvalds } 8661da177e4SLinus Torvalds 8671da177e4SLinus Torvalds static inline int flush_task_priority(int how) 8681da177e4SLinus Torvalds { 8691da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 8701da177e4SLinus Torvalds case FLUSH_HIGHPRI: 8711da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 8721da177e4SLinus Torvalds case FLUSH_LOWPRI: 8731da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 8741da177e4SLinus Torvalds } 8751da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 8761da177e4SLinus Torvalds } 8771da177e4SLinus Torvalds 8781da177e4SLinus Torvalds /* 8791da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 8801da177e4SLinus Torvalds */ 8811da177e4SLinus Torvalds static void nfs_write_rpcsetup(struct nfs_page *req, 8821da177e4SLinus Torvalds struct nfs_write_data *data, 883788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 8841da177e4SLinus Torvalds unsigned int count, unsigned int offset, 8851da177e4SLinus Torvalds int how) 8861da177e4SLinus Torvalds { 8871da177e4SLinus Torvalds struct inode *inode; 888788e7a89STrond Myklebust int flags; 8891da177e4SLinus Torvalds 8901da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 8911da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 8921da177e4SLinus Torvalds 8931da177e4SLinus Torvalds data->req = req; 8941da177e4SLinus Torvalds data->inode = inode = req->wb_context->dentry->d_inode; 8951da177e4SLinus Torvalds data->cred = req->wb_context->cred; 8961da177e4SLinus Torvalds 8971da177e4SLinus Torvalds data->args.fh = NFS_FH(inode); 8981da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 8991da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 9001da177e4SLinus Torvalds data->args.pages = data->pagevec; 9011da177e4SLinus Torvalds data->args.count = count; 9021da177e4SLinus Torvalds data->args.context = req->wb_context; 9031da177e4SLinus Torvalds 9041da177e4SLinus Torvalds data->res.fattr = &data->fattr; 9051da177e4SLinus Torvalds data->res.count = count; 9061da177e4SLinus Torvalds data->res.verf = &data->verf; 9070e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 9081da177e4SLinus Torvalds 909788e7a89STrond Myklebust /* Set up the initial task struct. */ 910788e7a89STrond Myklebust flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 911788e7a89STrond Myklebust rpc_init_task(&data->task, NFS_CLIENT(inode), flags, call_ops, data); 9121da177e4SLinus Torvalds NFS_PROTO(inode)->write_setup(data, how); 9131da177e4SLinus Torvalds 9141da177e4SLinus Torvalds data->task.tk_priority = flush_task_priority(how); 9151da177e4SLinus Torvalds data->task.tk_cookie = (unsigned long)inode; 9161da177e4SLinus Torvalds 9171da177e4SLinus Torvalds dprintk("NFS: %4d initiated write call (req %s/%Ld, %u bytes @ offset %Lu)\n", 9180bbacc40SChuck Lever data->task.tk_pid, 9191da177e4SLinus Torvalds inode->i_sb->s_id, 9201da177e4SLinus Torvalds (long long)NFS_FILEID(inode), 9211da177e4SLinus Torvalds count, 9221da177e4SLinus Torvalds (unsigned long long)data->args.offset); 9231da177e4SLinus Torvalds } 9241da177e4SLinus Torvalds 9251da177e4SLinus Torvalds static void nfs_execute_write(struct nfs_write_data *data) 9261da177e4SLinus Torvalds { 9271da177e4SLinus Torvalds struct rpc_clnt *clnt = NFS_CLIENT(data->inode); 9281da177e4SLinus Torvalds sigset_t oldset; 9291da177e4SLinus Torvalds 9301da177e4SLinus Torvalds rpc_clnt_sigmask(clnt, &oldset); 9311da177e4SLinus Torvalds rpc_execute(&data->task); 9321da177e4SLinus Torvalds rpc_clnt_sigunmask(clnt, &oldset); 9331da177e4SLinus Torvalds } 9341da177e4SLinus Torvalds 9351da177e4SLinus Torvalds /* 9361da177e4SLinus Torvalds * Generate multiple small requests to write out a single 9371da177e4SLinus Torvalds * contiguous dirty area on one page. 9381da177e4SLinus Torvalds */ 9397d46a49fSTrond Myklebust static int nfs_flush_multi(struct inode *inode, struct list_head *head, int how) 9401da177e4SLinus Torvalds { 9411da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 9421da177e4SLinus Torvalds struct page *page = req->wb_page; 9431da177e4SLinus Torvalds struct nfs_write_data *data; 944e9f7bee1STrond Myklebust size_t wsize = NFS_SERVER(inode)->wsize, nbytes; 945e9f7bee1STrond Myklebust unsigned int offset; 9461da177e4SLinus Torvalds int requests = 0; 9471da177e4SLinus Torvalds LIST_HEAD(list); 9481da177e4SLinus Torvalds 9491da177e4SLinus Torvalds nfs_list_remove_request(req); 9501da177e4SLinus Torvalds 9511da177e4SLinus Torvalds nbytes = req->wb_bytes; 952e9f7bee1STrond Myklebust do { 953e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 954e9f7bee1STrond Myklebust 955e9f7bee1STrond Myklebust data = nfs_writedata_alloc(len); 9561da177e4SLinus Torvalds if (!data) 9571da177e4SLinus Torvalds goto out_bad; 9581da177e4SLinus Torvalds list_add(&data->pages, &list); 9591da177e4SLinus Torvalds requests++; 960e9f7bee1STrond Myklebust nbytes -= len; 961e9f7bee1STrond Myklebust } while (nbytes != 0); 9621da177e4SLinus Torvalds atomic_set(&req->wb_complete, requests); 9631da177e4SLinus Torvalds 9641da177e4SLinus Torvalds ClearPageError(page); 965bb713d6dSTrond Myklebust set_page_writeback(page); 9661da177e4SLinus Torvalds offset = 0; 9671da177e4SLinus Torvalds nbytes = req->wb_bytes; 9681da177e4SLinus Torvalds do { 9691da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 9701da177e4SLinus Torvalds list_del_init(&data->pages); 9711da177e4SLinus Torvalds 9721da177e4SLinus Torvalds data->pagevec[0] = page; 9731da177e4SLinus Torvalds 9741da177e4SLinus Torvalds if (nbytes > wsize) { 975788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 976788e7a89STrond Myklebust wsize, offset, how); 9771da177e4SLinus Torvalds offset += wsize; 9781da177e4SLinus Torvalds nbytes -= wsize; 9791da177e4SLinus Torvalds } else { 980788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 981788e7a89STrond Myklebust nbytes, offset, how); 9821da177e4SLinus Torvalds nbytes = 0; 9831da177e4SLinus Torvalds } 9841da177e4SLinus Torvalds nfs_execute_write(data); 9851da177e4SLinus Torvalds } while (nbytes != 0); 9861da177e4SLinus Torvalds 9871da177e4SLinus Torvalds return 0; 9881da177e4SLinus Torvalds 9891da177e4SLinus Torvalds out_bad: 9901da177e4SLinus Torvalds while (!list_empty(&list)) { 9911da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 9921da177e4SLinus Torvalds list_del(&data->pages); 9938aca67f0STrond Myklebust nfs_writedata_release(data); 9941da177e4SLinus Torvalds } 9951da177e4SLinus Torvalds nfs_mark_request_dirty(req); 996c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 9971da177e4SLinus Torvalds return -ENOMEM; 9981da177e4SLinus Torvalds } 9991da177e4SLinus Torvalds 10001da177e4SLinus Torvalds /* 10011da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 10021da177e4SLinus Torvalds * The page must have been locked by the caller. 10031da177e4SLinus Torvalds * 10041da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 10051da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 10061da177e4SLinus Torvalds * that has been written but not committed. 10071da177e4SLinus Torvalds */ 10087d46a49fSTrond Myklebust static int nfs_flush_one(struct inode *inode, struct list_head *head, int how) 10091da177e4SLinus Torvalds { 10101da177e4SLinus Torvalds struct nfs_page *req; 10111da177e4SLinus Torvalds struct page **pages; 10121da177e4SLinus Torvalds struct nfs_write_data *data; 10131da177e4SLinus Torvalds unsigned int count; 10141da177e4SLinus Torvalds 1015e9f7bee1STrond Myklebust data = nfs_writedata_alloc(NFS_SERVER(inode)->wsize); 10161da177e4SLinus Torvalds if (!data) 10171da177e4SLinus Torvalds goto out_bad; 10181da177e4SLinus Torvalds 10191da177e4SLinus Torvalds pages = data->pagevec; 10201da177e4SLinus Torvalds count = 0; 10211da177e4SLinus Torvalds while (!list_empty(head)) { 10221da177e4SLinus Torvalds req = nfs_list_entry(head->next); 10231da177e4SLinus Torvalds nfs_list_remove_request(req); 10241da177e4SLinus Torvalds nfs_list_add_request(req, &data->pages); 10251da177e4SLinus Torvalds ClearPageError(req->wb_page); 1026bb713d6dSTrond Myklebust set_page_writeback(req->wb_page); 10271da177e4SLinus Torvalds *pages++ = req->wb_page; 10281da177e4SLinus Torvalds count += req->wb_bytes; 10291da177e4SLinus Torvalds } 10301da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 10311da177e4SLinus Torvalds 10321da177e4SLinus Torvalds /* Set up the argument struct */ 1033788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how); 10341da177e4SLinus Torvalds 10351da177e4SLinus Torvalds nfs_execute_write(data); 10361da177e4SLinus Torvalds return 0; 10371da177e4SLinus Torvalds out_bad: 10381da177e4SLinus Torvalds while (!list_empty(head)) { 10391da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 10401da177e4SLinus Torvalds nfs_list_remove_request(req); 10411da177e4SLinus Torvalds nfs_mark_request_dirty(req); 1042c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 10431da177e4SLinus Torvalds } 10441da177e4SLinus Torvalds return -ENOMEM; 10451da177e4SLinus Torvalds } 10461da177e4SLinus Torvalds 10477d46a49fSTrond Myklebust static int nfs_flush_list(struct inode *inode, struct list_head *head, int npages, int how) 10481da177e4SLinus Torvalds { 10491da177e4SLinus Torvalds LIST_HEAD(one_request); 10507d46a49fSTrond Myklebust int (*flush_one)(struct inode *, struct list_head *, int); 10511da177e4SLinus Torvalds struct nfs_page *req; 10527d46a49fSTrond Myklebust int wpages = NFS_SERVER(inode)->wpages; 10537d46a49fSTrond Myklebust int wsize = NFS_SERVER(inode)->wsize; 10547d46a49fSTrond Myklebust int error; 10551da177e4SLinus Torvalds 10567d46a49fSTrond Myklebust flush_one = nfs_flush_one; 10577d46a49fSTrond Myklebust if (wsize < PAGE_CACHE_SIZE) 10587d46a49fSTrond Myklebust flush_one = nfs_flush_multi; 10597d46a49fSTrond Myklebust /* For single writes, FLUSH_STABLE is more efficient */ 10607d46a49fSTrond Myklebust if (npages <= wpages && npages == NFS_I(inode)->npages 10617d46a49fSTrond Myklebust && nfs_list_entry(head->next)->wb_bytes <= wsize) 10627d46a49fSTrond Myklebust how |= FLUSH_STABLE; 10637d46a49fSTrond Myklebust 10647d46a49fSTrond Myklebust do { 10657d46a49fSTrond Myklebust nfs_coalesce_requests(head, &one_request, wpages); 10661da177e4SLinus Torvalds req = nfs_list_entry(one_request.next); 10677d46a49fSTrond Myklebust error = flush_one(inode, &one_request, how); 10681da177e4SLinus Torvalds if (error < 0) 10697d46a49fSTrond Myklebust goto out_err; 10707d46a49fSTrond Myklebust } while (!list_empty(head)); 10717d46a49fSTrond Myklebust return 0; 10727d46a49fSTrond Myklebust out_err: 10731da177e4SLinus Torvalds while (!list_empty(head)) { 10741da177e4SLinus Torvalds req = nfs_list_entry(head->next); 10751da177e4SLinus Torvalds nfs_list_remove_request(req); 10761da177e4SLinus Torvalds nfs_mark_request_dirty(req); 1077c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 10781da177e4SLinus Torvalds } 10791da177e4SLinus Torvalds return error; 10801da177e4SLinus Torvalds } 10811da177e4SLinus Torvalds 10821da177e4SLinus Torvalds /* 10831da177e4SLinus Torvalds * Handle a write reply that flushed part of a page. 10841da177e4SLinus Torvalds */ 1085788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) 10861da177e4SLinus Torvalds { 1087788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 10881da177e4SLinus Torvalds struct nfs_page *req = data->req; 10891da177e4SLinus Torvalds struct page *page = req->wb_page; 10901da177e4SLinus Torvalds 10911da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 10921da177e4SLinus Torvalds req->wb_context->dentry->d_inode->i_sb->s_id, 10931da177e4SLinus Torvalds (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 10941da177e4SLinus Torvalds req->wb_bytes, 10951da177e4SLinus Torvalds (long long)req_offset(req)); 10961da177e4SLinus Torvalds 1097788e7a89STrond Myklebust if (nfs_writeback_done(task, data) != 0) 1098788e7a89STrond Myklebust return; 1099788e7a89STrond Myklebust 1100788e7a89STrond Myklebust if (task->tk_status < 0) { 11011da177e4SLinus Torvalds ClearPageUptodate(page); 11021da177e4SLinus Torvalds SetPageError(page); 1103788e7a89STrond Myklebust req->wb_context->error = task->tk_status; 1104788e7a89STrond Myklebust dprintk(", error = %d\n", task->tk_status); 11051da177e4SLinus Torvalds } else { 11061da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 11071da177e4SLinus Torvalds if (data->verf.committed < NFS_FILE_SYNC) { 11081da177e4SLinus Torvalds if (!NFS_NEED_COMMIT(req)) { 11091da177e4SLinus Torvalds nfs_defer_commit(req); 11101da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 11111da177e4SLinus Torvalds dprintk(" defer commit\n"); 11121da177e4SLinus Torvalds } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { 11131da177e4SLinus Torvalds nfs_defer_reschedule(req); 11141da177e4SLinus Torvalds dprintk(" server reboot detected\n"); 11151da177e4SLinus Torvalds } 11161da177e4SLinus Torvalds } else 11171da177e4SLinus Torvalds #endif 11181da177e4SLinus Torvalds dprintk(" OK\n"); 11191da177e4SLinus Torvalds } 11201da177e4SLinus Torvalds 11211da177e4SLinus Torvalds if (atomic_dec_and_test(&req->wb_complete)) 11221da177e4SLinus Torvalds nfs_writepage_release(req); 11231da177e4SLinus Torvalds } 11241da177e4SLinus Torvalds 1125788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops = { 1126788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_partial, 1127788e7a89STrond Myklebust .rpc_release = nfs_writedata_release, 1128788e7a89STrond Myklebust }; 1129788e7a89STrond Myklebust 11301da177e4SLinus Torvalds /* 11311da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 11321da177e4SLinus Torvalds * 11331da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 11341da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 11351da177e4SLinus Torvalds * as the page has a write request pending. 11361da177e4SLinus Torvalds */ 1137788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) 11381da177e4SLinus Torvalds { 1139788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 11401da177e4SLinus Torvalds struct nfs_page *req; 11411da177e4SLinus Torvalds struct page *page; 11421da177e4SLinus Torvalds 1143788e7a89STrond Myklebust if (nfs_writeback_done(task, data) != 0) 1144788e7a89STrond Myklebust return; 1145788e7a89STrond Myklebust 11461da177e4SLinus Torvalds /* Update attributes as result of writeback. */ 11471da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 11481da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 11491da177e4SLinus Torvalds nfs_list_remove_request(req); 11501da177e4SLinus Torvalds page = req->wb_page; 11511da177e4SLinus Torvalds 11521da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 11531da177e4SLinus Torvalds req->wb_context->dentry->d_inode->i_sb->s_id, 11541da177e4SLinus Torvalds (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 11551da177e4SLinus Torvalds req->wb_bytes, 11561da177e4SLinus Torvalds (long long)req_offset(req)); 11571da177e4SLinus Torvalds 1158788e7a89STrond Myklebust if (task->tk_status < 0) { 11591da177e4SLinus Torvalds ClearPageUptodate(page); 11601da177e4SLinus Torvalds SetPageError(page); 1161788e7a89STrond Myklebust req->wb_context->error = task->tk_status; 11621da177e4SLinus Torvalds end_page_writeback(page); 11631da177e4SLinus Torvalds nfs_inode_remove_request(req); 1164788e7a89STrond Myklebust dprintk(", error = %d\n", task->tk_status); 11651da177e4SLinus Torvalds goto next; 11661da177e4SLinus Torvalds } 11671da177e4SLinus Torvalds end_page_writeback(page); 11681da177e4SLinus Torvalds 11691da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 11701da177e4SLinus Torvalds if (data->args.stable != NFS_UNSTABLE || data->verf.committed == NFS_FILE_SYNC) { 11711da177e4SLinus Torvalds nfs_inode_remove_request(req); 11721da177e4SLinus Torvalds dprintk(" OK\n"); 11731da177e4SLinus Torvalds goto next; 11741da177e4SLinus Torvalds } 11751da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 11761da177e4SLinus Torvalds nfs_mark_request_commit(req); 11771da177e4SLinus Torvalds dprintk(" marked for commit\n"); 11781da177e4SLinus Torvalds #else 11791da177e4SLinus Torvalds nfs_inode_remove_request(req); 11801da177e4SLinus Torvalds #endif 11811da177e4SLinus Torvalds next: 1182c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 11831da177e4SLinus Torvalds } 11841da177e4SLinus Torvalds } 11851da177e4SLinus Torvalds 1186788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = { 1187788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_full, 1188788e7a89STrond Myklebust .rpc_release = nfs_writedata_release, 1189788e7a89STrond Myklebust }; 1190788e7a89STrond Myklebust 1191788e7a89STrond Myklebust 11921da177e4SLinus Torvalds /* 11931da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 11941da177e4SLinus Torvalds */ 1195462d5b32SChuck Lever int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 11961da177e4SLinus Torvalds { 11971da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 11981da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1199788e7a89STrond Myklebust int status; 12001da177e4SLinus Torvalds 12011da177e4SLinus Torvalds dprintk("NFS: %4d nfs_writeback_done (status %d)\n", 12021da177e4SLinus Torvalds task->tk_pid, task->tk_status); 12031da177e4SLinus Torvalds 1204f551e44fSChuck Lever /* 1205f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1206f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1207f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1208f551e44fSChuck Lever * another writer had changed the file, but some applications 1209f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1210f551e44fSChuck Lever */ 1211788e7a89STrond Myklebust status = NFS_PROTO(data->inode)->write_done(task, data); 1212788e7a89STrond Myklebust if (status != 0) 1213788e7a89STrond Myklebust return status; 121491d5b470SChuck Lever nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 121591d5b470SChuck Lever 12161da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 12171da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 12181da177e4SLinus Torvalds /* We tried a write call, but the server did not 12191da177e4SLinus Torvalds * commit data to stable storage even though we 12201da177e4SLinus Torvalds * requested it. 12211da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 12221da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 12231da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 12241da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 12251da177e4SLinus Torvalds */ 12261da177e4SLinus Torvalds static unsigned long complain; 12271da177e4SLinus Torvalds 12281da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 12291da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 12301da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 123154ceac45SDavid Howells NFS_SERVER(data->inode)->nfs_client->cl_hostname, 12321da177e4SLinus Torvalds resp->verf->committed, argp->stable); 12331da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 12341da177e4SLinus Torvalds } 12351da177e4SLinus Torvalds } 12361da177e4SLinus Torvalds #endif 12371da177e4SLinus Torvalds /* Is this a short write? */ 12381da177e4SLinus Torvalds if (task->tk_status >= 0 && resp->count < argp->count) { 12391da177e4SLinus Torvalds static unsigned long complain; 12401da177e4SLinus Torvalds 124191d5b470SChuck Lever nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); 124291d5b470SChuck Lever 12431da177e4SLinus Torvalds /* Has the server at least made some progress? */ 12441da177e4SLinus Torvalds if (resp->count != 0) { 12451da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 12461da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 12471da177e4SLinus Torvalds /* Resend from where the server left off */ 12481da177e4SLinus Torvalds argp->offset += resp->count; 12491da177e4SLinus Torvalds argp->pgbase += resp->count; 12501da177e4SLinus Torvalds argp->count -= resp->count; 12511da177e4SLinus Torvalds } else { 12521da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 12531da177e4SLinus Torvalds * headaches in the case of a server crash. 12541da177e4SLinus Torvalds */ 12551da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 12561da177e4SLinus Torvalds } 12571da177e4SLinus Torvalds rpc_restart_call(task); 1258788e7a89STrond Myklebust return -EAGAIN; 12591da177e4SLinus Torvalds } 12601da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 12611da177e4SLinus Torvalds printk(KERN_WARNING 12621da177e4SLinus Torvalds "NFS: Server wrote zero bytes, expected %u.\n", 12631da177e4SLinus Torvalds argp->count); 12641da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 12651da177e4SLinus Torvalds } 12661da177e4SLinus Torvalds /* Can't do anything about it except throw an error. */ 12671da177e4SLinus Torvalds task->tk_status = -EIO; 12681da177e4SLinus Torvalds } 1269788e7a89STrond Myklebust return 0; 12701da177e4SLinus Torvalds } 12711da177e4SLinus Torvalds 12721da177e4SLinus Torvalds 12731da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 1274963d8fe5STrond Myklebust void nfs_commit_release(void *wdata) 12751da177e4SLinus Torvalds { 12761da177e4SLinus Torvalds nfs_commit_free(wdata); 12771da177e4SLinus Torvalds } 12781da177e4SLinus Torvalds 12791da177e4SLinus Torvalds /* 12801da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 12811da177e4SLinus Torvalds */ 12821da177e4SLinus Torvalds static void nfs_commit_rpcsetup(struct list_head *head, 1283788e7a89STrond Myklebust struct nfs_write_data *data, 1284788e7a89STrond Myklebust int how) 12851da177e4SLinus Torvalds { 12863da28eb1STrond Myklebust struct nfs_page *first; 12871da177e4SLinus Torvalds struct inode *inode; 1288788e7a89STrond Myklebust int flags; 12891da177e4SLinus Torvalds 12901da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 12911da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 12921da177e4SLinus Torvalds 12931da177e4SLinus Torvalds list_splice_init(head, &data->pages); 12941da177e4SLinus Torvalds first = nfs_list_entry(data->pages.next); 12951da177e4SLinus Torvalds inode = first->wb_context->dentry->d_inode; 12961da177e4SLinus Torvalds 12971da177e4SLinus Torvalds data->inode = inode; 12981da177e4SLinus Torvalds data->cred = first->wb_context->cred; 12991da177e4SLinus Torvalds 13001da177e4SLinus Torvalds data->args.fh = NFS_FH(data->inode); 13013da28eb1STrond Myklebust /* Note: we always request a commit of the entire inode */ 13023da28eb1STrond Myklebust data->args.offset = 0; 13033da28eb1STrond Myklebust data->args.count = 0; 13043da28eb1STrond Myklebust data->res.count = 0; 13051da177e4SLinus Torvalds data->res.fattr = &data->fattr; 13061da177e4SLinus Torvalds data->res.verf = &data->verf; 13070e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 13081da177e4SLinus Torvalds 1309788e7a89STrond Myklebust /* Set up the initial task struct. */ 1310788e7a89STrond Myklebust flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 1311788e7a89STrond Myklebust rpc_init_task(&data->task, NFS_CLIENT(inode), flags, &nfs_commit_ops, data); 13121da177e4SLinus Torvalds NFS_PROTO(inode)->commit_setup(data, how); 13131da177e4SLinus Torvalds 13141da177e4SLinus Torvalds data->task.tk_priority = flush_task_priority(how); 13151da177e4SLinus Torvalds data->task.tk_cookie = (unsigned long)inode; 13161da177e4SLinus Torvalds 13170bbacc40SChuck Lever dprintk("NFS: %4d initiated commit call\n", data->task.tk_pid); 13181da177e4SLinus Torvalds } 13191da177e4SLinus Torvalds 13201da177e4SLinus Torvalds /* 13211da177e4SLinus Torvalds * Commit dirty pages 13221da177e4SLinus Torvalds */ 13231da177e4SLinus Torvalds static int 132440859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how) 13251da177e4SLinus Torvalds { 13261da177e4SLinus Torvalds struct nfs_write_data *data; 13271da177e4SLinus Torvalds struct nfs_page *req; 13281da177e4SLinus Torvalds 1329e9f7bee1STrond Myklebust data = nfs_commit_alloc(); 13301da177e4SLinus Torvalds 13311da177e4SLinus Torvalds if (!data) 13321da177e4SLinus Torvalds goto out_bad; 13331da177e4SLinus Torvalds 13341da177e4SLinus Torvalds /* Set up the argument struct */ 13351da177e4SLinus Torvalds nfs_commit_rpcsetup(head, data, how); 13361da177e4SLinus Torvalds 13371da177e4SLinus Torvalds nfs_execute_write(data); 13381da177e4SLinus Torvalds return 0; 13391da177e4SLinus Torvalds out_bad: 13401da177e4SLinus Torvalds while (!list_empty(head)) { 13411da177e4SLinus Torvalds req = nfs_list_entry(head->next); 13421da177e4SLinus Torvalds nfs_list_remove_request(req); 13431da177e4SLinus Torvalds nfs_mark_request_commit(req); 134483715ad5STrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 13455c2d97cbSTrond Myklebust nfs_clear_page_writeback(req); 13461da177e4SLinus Torvalds } 13471da177e4SLinus Torvalds return -ENOMEM; 13481da177e4SLinus Torvalds } 13491da177e4SLinus Torvalds 13501da177e4SLinus Torvalds /* 13511da177e4SLinus Torvalds * COMMIT call returned 13521da177e4SLinus Torvalds */ 1353788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 13541da177e4SLinus Torvalds { 1355963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 13561da177e4SLinus Torvalds struct nfs_page *req; 13571da177e4SLinus Torvalds 13581da177e4SLinus Torvalds dprintk("NFS: %4d nfs_commit_done (status %d)\n", 13591da177e4SLinus Torvalds task->tk_pid, task->tk_status); 13601da177e4SLinus Torvalds 1361788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1362788e7a89STrond Myklebust if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) 1363788e7a89STrond Myklebust return; 1364788e7a89STrond Myklebust 13651da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 13661da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 13671da177e4SLinus Torvalds nfs_list_remove_request(req); 1368fd39fc85SChristoph Lameter dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 13691da177e4SLinus Torvalds 13701da177e4SLinus Torvalds dprintk("NFS: commit (%s/%Ld %d@%Ld)", 13711da177e4SLinus Torvalds req->wb_context->dentry->d_inode->i_sb->s_id, 13721da177e4SLinus Torvalds (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 13731da177e4SLinus Torvalds req->wb_bytes, 13741da177e4SLinus Torvalds (long long)req_offset(req)); 13751da177e4SLinus Torvalds if (task->tk_status < 0) { 13761da177e4SLinus Torvalds req->wb_context->error = task->tk_status; 13771da177e4SLinus Torvalds nfs_inode_remove_request(req); 13781da177e4SLinus Torvalds dprintk(", error = %d\n", task->tk_status); 13791da177e4SLinus Torvalds goto next; 13801da177e4SLinus Torvalds } 13811da177e4SLinus Torvalds 13821da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 13831da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 13841da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 13851da177e4SLinus Torvalds /* We have a match */ 13861da177e4SLinus Torvalds nfs_inode_remove_request(req); 13871da177e4SLinus Torvalds dprintk(" OK\n"); 13881da177e4SLinus Torvalds goto next; 13891da177e4SLinus Torvalds } 13901da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 13911da177e4SLinus Torvalds dprintk(" mismatch\n"); 13921da177e4SLinus Torvalds nfs_mark_request_dirty(req); 13931da177e4SLinus Torvalds next: 1394c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 13951da177e4SLinus Torvalds } 13961da177e4SLinus Torvalds } 1397788e7a89STrond Myklebust 1398788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 1399788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1400788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1401788e7a89STrond Myklebust }; 1402c42de9ddSTrond Myklebust #else 1403c42de9ddSTrond Myklebust static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how) 1404c42de9ddSTrond Myklebust { 1405c42de9ddSTrond Myklebust return 0; 1406c42de9ddSTrond Myklebust } 14071da177e4SLinus Torvalds #endif 14081da177e4SLinus Torvalds 14093f442547STrond Myklebust static long nfs_flush_mapping(struct address_space *mapping, struct writeback_control *wbc, int how) 14101da177e4SLinus Torvalds { 141128c6925fSTrond Myklebust struct nfs_inode *nfsi = NFS_I(mapping->host); 14121da177e4SLinus Torvalds LIST_HEAD(head); 14133f442547STrond Myklebust long res; 14141da177e4SLinus Torvalds 14151da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 14163f442547STrond Myklebust res = nfs_scan_dirty(mapping, wbc, &head); 14171da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 1418ab0a3dbeSTrond Myklebust if (res) { 141928c6925fSTrond Myklebust int error = nfs_flush_list(mapping->host, &head, res, how); 14201da177e4SLinus Torvalds if (error < 0) 14211da177e4SLinus Torvalds return error; 14227d46a49fSTrond Myklebust } 14231da177e4SLinus Torvalds return res; 14241da177e4SLinus Torvalds } 14251da177e4SLinus Torvalds 14261da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 14273da28eb1STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 14281da177e4SLinus Torvalds { 14291da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 14301da177e4SLinus Torvalds LIST_HEAD(head); 14317d46a49fSTrond Myklebust int res; 14321da177e4SLinus Torvalds 14331da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 14343da28eb1STrond Myklebust res = nfs_scan_commit(inode, &head, 0, 0); 14353da28eb1STrond Myklebust spin_unlock(&nfsi->req_lock); 14361da177e4SLinus Torvalds if (res) { 14377d46a49fSTrond Myklebust int error = nfs_commit_list(inode, &head, how); 14381da177e4SLinus Torvalds if (error < 0) 14391da177e4SLinus Torvalds return error; 14403da28eb1STrond Myklebust } 14411da177e4SLinus Torvalds return res; 14421da177e4SLinus Torvalds } 14431da177e4SLinus Torvalds #endif 14441da177e4SLinus Torvalds 14451c75950bSTrond Myklebust long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how) 14461da177e4SLinus Torvalds { 14471c75950bSTrond Myklebust struct inode *inode = mapping->host; 1448c42de9ddSTrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 14491c75950bSTrond Myklebust unsigned long idx_start, idx_end; 14501c75950bSTrond Myklebust unsigned int npages = 0; 1451c42de9ddSTrond Myklebust LIST_HEAD(head); 145270b9ecbdSTrond Myklebust int nocommit = how & FLUSH_NOCOMMIT; 14533f442547STrond Myklebust long pages, ret; 14541da177e4SLinus Torvalds 14551c75950bSTrond Myklebust /* FIXME */ 14561c75950bSTrond Myklebust if (wbc->range_cyclic) 14571c75950bSTrond Myklebust idx_start = 0; 14581c75950bSTrond Myklebust else { 14591c75950bSTrond Myklebust idx_start = wbc->range_start >> PAGE_CACHE_SHIFT; 14601c75950bSTrond Myklebust idx_end = wbc->range_end >> PAGE_CACHE_SHIFT; 14611c75950bSTrond Myklebust if (idx_end > idx_start) { 14621c75950bSTrond Myklebust unsigned long l_npages = 1 + idx_end - idx_start; 14631c75950bSTrond Myklebust npages = l_npages; 14641c75950bSTrond Myklebust if (sizeof(npages) != sizeof(l_npages) && 14651c75950bSTrond Myklebust (unsigned long)npages != l_npages) 14661c75950bSTrond Myklebust npages = 0; 14671c75950bSTrond Myklebust } 14681c75950bSTrond Myklebust } 1469c42de9ddSTrond Myklebust how &= ~FLUSH_NOCOMMIT; 1470c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 14711da177e4SLinus Torvalds do { 14721c75950bSTrond Myklebust wbc->pages_skipped = 0; 1473c42de9ddSTrond Myklebust ret = nfs_wait_on_requests_locked(inode, idx_start, npages); 1474c42de9ddSTrond Myklebust if (ret != 0) 1475c42de9ddSTrond Myklebust continue; 14761c75950bSTrond Myklebust pages = nfs_scan_dirty(mapping, wbc, &head); 1477c42de9ddSTrond Myklebust if (pages != 0) { 1478c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 1479e8e058e8STrond Myklebust if (how & FLUSH_INVALIDATE) { 148083715ad5STrond Myklebust nfs_cancel_dirty_list(&head); 1481e8e058e8STrond Myklebust ret = pages; 1482e8e058e8STrond Myklebust } else 1483c42de9ddSTrond Myklebust ret = nfs_flush_list(inode, &head, pages, how); 1484c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 148570b9ecbdSTrond Myklebust continue; 148670b9ecbdSTrond Myklebust } 14871c75950bSTrond Myklebust if (wbc->pages_skipped != 0) 14881c75950bSTrond Myklebust continue; 1489c42de9ddSTrond Myklebust if (nocommit) 1490c42de9ddSTrond Myklebust break; 1491d2ccddf0STrond Myklebust pages = nfs_scan_commit(inode, &head, idx_start, npages); 14921c75950bSTrond Myklebust if (pages == 0) { 14931c75950bSTrond Myklebust if (wbc->pages_skipped != 0) 14941c75950bSTrond Myklebust continue; 1495c42de9ddSTrond Myklebust break; 14961c75950bSTrond Myklebust } 1497d2ccddf0STrond Myklebust if (how & FLUSH_INVALIDATE) { 1498d2ccddf0STrond Myklebust spin_unlock(&nfsi->req_lock); 149983715ad5STrond Myklebust nfs_cancel_commit_list(&head); 1500e8e058e8STrond Myklebust ret = pages; 1501d2ccddf0STrond Myklebust spin_lock(&nfsi->req_lock); 1502d2ccddf0STrond Myklebust continue; 1503d2ccddf0STrond Myklebust } 1504d2ccddf0STrond Myklebust pages += nfs_scan_commit(inode, &head, 0, 0); 1505c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 1506c42de9ddSTrond Myklebust ret = nfs_commit_list(inode, &head, how); 1507c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 1508c42de9ddSTrond Myklebust } while (ret >= 0); 1509c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 1510c42de9ddSTrond Myklebust return ret; 15111da177e4SLinus Torvalds } 15121da177e4SLinus Torvalds 15131c75950bSTrond Myklebust /* 15141c75950bSTrond Myklebust * flush the inode to disk. 15151c75950bSTrond Myklebust */ 15161c75950bSTrond Myklebust int nfs_wb_all(struct inode *inode) 15171c75950bSTrond Myklebust { 15181c75950bSTrond Myklebust struct address_space *mapping = inode->i_mapping; 15191c75950bSTrond Myklebust struct writeback_control wbc = { 15201c75950bSTrond Myklebust .bdi = mapping->backing_dev_info, 15211c75950bSTrond Myklebust .sync_mode = WB_SYNC_ALL, 15221c75950bSTrond Myklebust .nr_to_write = LONG_MAX, 15231c75950bSTrond Myklebust .range_cyclic = 1, 15241c75950bSTrond Myklebust }; 15251c75950bSTrond Myklebust int ret; 15261c75950bSTrond Myklebust 15271c75950bSTrond Myklebust ret = nfs_sync_mapping_wait(mapping, &wbc, 0); 15281c75950bSTrond Myklebust if (ret >= 0) 15291c75950bSTrond Myklebust return 0; 15301c75950bSTrond Myklebust return ret; 15311c75950bSTrond Myklebust } 15321c75950bSTrond Myklebust 15331c75950bSTrond Myklebust int nfs_sync_mapping_range(struct address_space *mapping, loff_t range_start, loff_t range_end, int how) 15341c75950bSTrond Myklebust { 15351c75950bSTrond Myklebust struct writeback_control wbc = { 15361c75950bSTrond Myklebust .bdi = mapping->backing_dev_info, 15371c75950bSTrond Myklebust .sync_mode = WB_SYNC_ALL, 15381c75950bSTrond Myklebust .nr_to_write = LONG_MAX, 15391c75950bSTrond Myklebust .range_start = range_start, 15401c75950bSTrond Myklebust .range_end = range_end, 15411c75950bSTrond Myklebust }; 15421c75950bSTrond Myklebust int ret; 15431c75950bSTrond Myklebust 15441c75950bSTrond Myklebust ret = nfs_sync_mapping_wait(mapping, &wbc, how); 15451c75950bSTrond Myklebust if (ret >= 0) 15461c75950bSTrond Myklebust return 0; 15471c75950bSTrond Myklebust return ret; 15481c75950bSTrond Myklebust } 15491c75950bSTrond Myklebust 15501c75950bSTrond Myklebust static int nfs_wb_page_priority(struct inode *inode, struct page *page, int how) 15511c75950bSTrond Myklebust { 15521c75950bSTrond Myklebust loff_t range_start = page_offset(page); 15531c75950bSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 15541c75950bSTrond Myklebust 15551c75950bSTrond Myklebust return nfs_sync_mapping_range(inode->i_mapping, range_start, range_end, how | FLUSH_STABLE); 15561c75950bSTrond Myklebust } 15571c75950bSTrond Myklebust 15581c75950bSTrond Myklebust /* 15591c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 15601c75950bSTrond Myklebust */ 15611c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page* page) 15621c75950bSTrond Myklebust { 15631c75950bSTrond Myklebust return nfs_wb_page_priority(inode, page, 0); 15641c75950bSTrond Myklebust } 15651c75950bSTrond Myklebust 15661c75950bSTrond Myklebust 1567f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 15681da177e4SLinus Torvalds { 15691da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 15701da177e4SLinus Torvalds sizeof(struct nfs_write_data), 15711da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 15721da177e4SLinus Torvalds NULL, NULL); 15731da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 15741da177e4SLinus Torvalds return -ENOMEM; 15751da177e4SLinus Torvalds 157693d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 15771da177e4SLinus Torvalds nfs_wdata_cachep); 15781da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 15791da177e4SLinus Torvalds return -ENOMEM; 15801da177e4SLinus Torvalds 158193d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 15821da177e4SLinus Torvalds nfs_wdata_cachep); 15831da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 15841da177e4SLinus Torvalds return -ENOMEM; 15851da177e4SLinus Torvalds 15861da177e4SLinus Torvalds return 0; 15871da177e4SLinus Torvalds } 15881da177e4SLinus Torvalds 1589266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 15901da177e4SLinus Torvalds { 15911da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 15921da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 15931a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 15941da177e4SLinus Torvalds } 15951da177e4SLinus Torvalds 1596