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" 6691d5b470SChuck Lever #include "iostat.h" 671da177e4SLinus Torvalds 681da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 691da177e4SLinus Torvalds 701da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 711da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 721da177e4SLinus Torvalds 731da177e4SLinus Torvalds /* 741da177e4SLinus Torvalds * Local function declarations 751da177e4SLinus Torvalds */ 761da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context*, 771da177e4SLinus Torvalds struct inode *, 781da177e4SLinus Torvalds struct page *, 791da177e4SLinus Torvalds unsigned int, unsigned int); 801da177e4SLinus Torvalds static int nfs_wait_on_write_congestion(struct address_space *, int); 811da177e4SLinus Torvalds static int nfs_wait_on_requests(struct inode *, unsigned long, unsigned int); 821da177e4SLinus Torvalds static int nfs_flush_inode(struct inode *inode, unsigned long idx_start, 831da177e4SLinus Torvalds unsigned int npages, int how); 84788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops; 85788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops; 86788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 871da177e4SLinus Torvalds 881da177e4SLinus Torvalds static kmem_cache_t *nfs_wdata_cachep; 893feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 901da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 911da177e4SLinus Torvalds 921da177e4SLinus Torvalds static DECLARE_WAIT_QUEUE_HEAD(nfs_write_congestion); 931da177e4SLinus Torvalds 94e9f7bee1STrond Myklebust struct nfs_write_data *nfs_commit_alloc(void) 951da177e4SLinus Torvalds { 961da177e4SLinus Torvalds struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, SLAB_NOFS); 9740859d7eSChuck Lever 981da177e4SLinus Torvalds if (p) { 991da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 1001da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 1011da177e4SLinus Torvalds } 1021da177e4SLinus Torvalds return p; 1031da177e4SLinus Torvalds } 1041da177e4SLinus Torvalds 105*8aca67f0STrond Myklebust void nfs_commit_rcu_free(struct rcu_head *head) 1061da177e4SLinus Torvalds { 107*8aca67f0STrond Myklebust struct nfs_write_data *p = container_of(head, struct nfs_write_data, task.u.tk_rcu); 10840859d7eSChuck Lever if (p && (p->pagevec != &p->page_array[0])) 10940859d7eSChuck Lever kfree(p->pagevec); 1101da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 1111da177e4SLinus Torvalds } 1121da177e4SLinus Torvalds 113*8aca67f0STrond Myklebust void nfs_commit_free(struct nfs_write_data *wdata) 114*8aca67f0STrond Myklebust { 115*8aca67f0STrond Myklebust call_rcu_bh(&wdata->task.u.tk_rcu, nfs_commit_rcu_free); 116*8aca67f0STrond Myklebust } 117*8aca67f0STrond Myklebust 118e9f7bee1STrond Myklebust struct nfs_write_data *nfs_writedata_alloc(size_t len) 1193feb2d49STrond Myklebust { 120e9f7bee1STrond Myklebust unsigned int pagecount = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1213feb2d49STrond Myklebust struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, SLAB_NOFS); 1223feb2d49STrond Myklebust 1233feb2d49STrond Myklebust if (p) { 1243feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 1253feb2d49STrond Myklebust INIT_LIST_HEAD(&p->pages); 126e9f7bee1STrond Myklebust p->npages = pagecount; 1270d0b5cb3SChuck Lever if (pagecount <= ARRAY_SIZE(p->page_array)) 1280d0b5cb3SChuck Lever p->pagevec = p->page_array; 1293feb2d49STrond Myklebust else { 1300d0b5cb3SChuck Lever p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); 1310d0b5cb3SChuck Lever if (!p->pagevec) { 1323feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 1333feb2d49STrond Myklebust p = NULL; 1343feb2d49STrond Myklebust } 1353feb2d49STrond Myklebust } 1363feb2d49STrond Myklebust } 1373feb2d49STrond Myklebust return p; 1383feb2d49STrond Myklebust } 1393feb2d49STrond Myklebust 140*8aca67f0STrond Myklebust static void nfs_writedata_rcu_free(struct rcu_head *head) 1413feb2d49STrond Myklebust { 142*8aca67f0STrond Myklebust struct nfs_write_data *p = container_of(head, struct nfs_write_data, task.u.tk_rcu); 1433feb2d49STrond Myklebust if (p && (p->pagevec != &p->page_array[0])) 1443feb2d49STrond Myklebust kfree(p->pagevec); 1453feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 1463feb2d49STrond Myklebust } 1473feb2d49STrond Myklebust 148*8aca67f0STrond Myklebust static void nfs_writedata_free(struct nfs_write_data *wdata) 149*8aca67f0STrond Myklebust { 150*8aca67f0STrond Myklebust call_rcu_bh(&wdata->task.u.tk_rcu, nfs_writedata_rcu_free); 151*8aca67f0STrond Myklebust } 152*8aca67f0STrond Myklebust 153963d8fe5STrond Myklebust void nfs_writedata_release(void *wdata) 1541da177e4SLinus Torvalds { 1551da177e4SLinus Torvalds nfs_writedata_free(wdata); 1561da177e4SLinus Torvalds } 1571da177e4SLinus Torvalds 1581da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1591da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1601da177e4SLinus Torvalds { 1611da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 1621da177e4SLinus Torvalds loff_t end, i_size = i_size_read(inode); 1631da177e4SLinus Torvalds unsigned long end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 1641da177e4SLinus Torvalds 1651da177e4SLinus Torvalds if (i_size > 0 && page->index < end_index) 1661da177e4SLinus Torvalds return; 1671da177e4SLinus Torvalds end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); 1681da177e4SLinus Torvalds if (i_size >= end) 1691da177e4SLinus Torvalds return; 17091d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 1711da177e4SLinus Torvalds i_size_write(inode, end); 1721da177e4SLinus Torvalds } 1731da177e4SLinus Torvalds 1741da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 1751da177e4SLinus Torvalds * covers the full page. 1761da177e4SLinus Torvalds */ 1771da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 1781da177e4SLinus Torvalds { 1791da177e4SLinus Torvalds loff_t end_offs; 1801da177e4SLinus Torvalds 1811da177e4SLinus Torvalds if (PageUptodate(page)) 1821da177e4SLinus Torvalds return; 1831da177e4SLinus Torvalds if (base != 0) 1841da177e4SLinus Torvalds return; 1851da177e4SLinus Torvalds if (count == PAGE_CACHE_SIZE) { 1861da177e4SLinus Torvalds SetPageUptodate(page); 1871da177e4SLinus Torvalds return; 1881da177e4SLinus Torvalds } 1891da177e4SLinus Torvalds 1901da177e4SLinus Torvalds end_offs = i_size_read(page->mapping->host) - 1; 1911da177e4SLinus Torvalds if (end_offs < 0) 1921da177e4SLinus Torvalds return; 1931da177e4SLinus Torvalds /* Is this the last page? */ 1941da177e4SLinus Torvalds if (page->index != (unsigned long)(end_offs >> PAGE_CACHE_SHIFT)) 1951da177e4SLinus Torvalds return; 1961da177e4SLinus Torvalds /* This is the last page: set PG_uptodate if we cover the entire 1971da177e4SLinus Torvalds * extent of the data, then zero the rest of the page. 1981da177e4SLinus Torvalds */ 1991da177e4SLinus Torvalds if (count == (unsigned int)(end_offs & (PAGE_CACHE_SIZE - 1)) + 1) { 2001da177e4SLinus Torvalds memclear_highpage_flush(page, count, PAGE_CACHE_SIZE - count); 2011da177e4SLinus Torvalds SetPageUptodate(page); 2021da177e4SLinus Torvalds } 2031da177e4SLinus Torvalds } 2041da177e4SLinus Torvalds 2051da177e4SLinus Torvalds /* 2061da177e4SLinus Torvalds * Write a page synchronously. 2071da177e4SLinus Torvalds * Offset is the data offset within the page. 2081da177e4SLinus Torvalds */ 2091da177e4SLinus Torvalds static int nfs_writepage_sync(struct nfs_open_context *ctx, struct inode *inode, 2101da177e4SLinus Torvalds struct page *page, unsigned int offset, unsigned int count, 2111da177e4SLinus Torvalds int how) 2121da177e4SLinus Torvalds { 2131da177e4SLinus Torvalds unsigned int wsize = NFS_SERVER(inode)->wsize; 2141da177e4SLinus Torvalds int result, written = 0; 2151da177e4SLinus Torvalds struct nfs_write_data *wdata; 2161da177e4SLinus Torvalds 217e9f7bee1STrond Myklebust wdata = nfs_writedata_alloc(wsize); 2181da177e4SLinus Torvalds if (!wdata) 2191da177e4SLinus Torvalds return -ENOMEM; 2201da177e4SLinus Torvalds 2211da177e4SLinus Torvalds wdata->flags = how; 2221da177e4SLinus Torvalds wdata->cred = ctx->cred; 2231da177e4SLinus Torvalds wdata->inode = inode; 2241da177e4SLinus Torvalds wdata->args.fh = NFS_FH(inode); 2251da177e4SLinus Torvalds wdata->args.context = ctx; 2261da177e4SLinus Torvalds wdata->args.pages = &page; 2271da177e4SLinus Torvalds wdata->args.stable = NFS_FILE_SYNC; 2281da177e4SLinus Torvalds wdata->args.pgbase = offset; 2291da177e4SLinus Torvalds wdata->args.count = wsize; 2301da177e4SLinus Torvalds wdata->res.fattr = &wdata->fattr; 2311da177e4SLinus Torvalds wdata->res.verf = &wdata->verf; 2321da177e4SLinus Torvalds 2331da177e4SLinus Torvalds dprintk("NFS: nfs_writepage_sync(%s/%Ld %d@%Ld)\n", 2341da177e4SLinus Torvalds inode->i_sb->s_id, 2351da177e4SLinus Torvalds (long long)NFS_FILEID(inode), 2361da177e4SLinus Torvalds count, (long long)(page_offset(page) + offset)); 2371da177e4SLinus Torvalds 238bb713d6dSTrond Myklebust set_page_writeback(page); 2391da177e4SLinus Torvalds nfs_begin_data_update(inode); 2401da177e4SLinus Torvalds do { 2411da177e4SLinus Torvalds if (count < wsize) 2421da177e4SLinus Torvalds wdata->args.count = count; 2431da177e4SLinus Torvalds wdata->args.offset = page_offset(page) + wdata->args.pgbase; 2441da177e4SLinus Torvalds 2451da177e4SLinus Torvalds result = NFS_PROTO(inode)->write(wdata); 2461da177e4SLinus Torvalds 2471da177e4SLinus Torvalds if (result < 0) { 2481da177e4SLinus Torvalds /* Must mark the page invalid after I/O error */ 2491da177e4SLinus Torvalds ClearPageUptodate(page); 2501da177e4SLinus Torvalds goto io_error; 2511da177e4SLinus Torvalds } 2521da177e4SLinus Torvalds if (result < wdata->args.count) 2531da177e4SLinus Torvalds printk(KERN_WARNING "NFS: short write, count=%u, result=%d\n", 2541da177e4SLinus Torvalds wdata->args.count, result); 2551da177e4SLinus Torvalds 2561da177e4SLinus Torvalds wdata->args.offset += result; 2571da177e4SLinus Torvalds wdata->args.pgbase += result; 2581da177e4SLinus Torvalds written += result; 2591da177e4SLinus Torvalds count -= result; 26091d5b470SChuck Lever nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, result); 2611da177e4SLinus Torvalds } while (count); 2621da177e4SLinus Torvalds /* Update file length */ 2631da177e4SLinus Torvalds nfs_grow_file(page, offset, written); 2641da177e4SLinus Torvalds /* Set the PG_uptodate flag? */ 2651da177e4SLinus Torvalds nfs_mark_uptodate(page, offset, written); 2661da177e4SLinus Torvalds 2671da177e4SLinus Torvalds if (PageError(page)) 2681da177e4SLinus Torvalds ClearPageError(page); 2691da177e4SLinus Torvalds 2701da177e4SLinus Torvalds io_error: 271951a143bSTrond Myklebust nfs_end_data_update(inode); 272bb713d6dSTrond Myklebust end_page_writeback(page); 273*8aca67f0STrond Myklebust nfs_writedata_release(wdata); 2741da177e4SLinus Torvalds return written ? written : result; 2751da177e4SLinus Torvalds } 2761da177e4SLinus Torvalds 2771da177e4SLinus Torvalds static int nfs_writepage_async(struct nfs_open_context *ctx, 2781da177e4SLinus Torvalds struct inode *inode, struct page *page, 2791da177e4SLinus Torvalds unsigned int offset, unsigned int count) 2801da177e4SLinus Torvalds { 2811da177e4SLinus Torvalds struct nfs_page *req; 2821da177e4SLinus Torvalds 2831da177e4SLinus Torvalds req = nfs_update_request(ctx, inode, page, offset, count); 284abd3e641STrond Myklebust if (IS_ERR(req)) 285abd3e641STrond Myklebust return PTR_ERR(req); 2861da177e4SLinus Torvalds /* Update file length */ 2871da177e4SLinus Torvalds nfs_grow_file(page, offset, count); 2881da177e4SLinus Torvalds /* Set the PG_uptodate flag? */ 2891da177e4SLinus Torvalds nfs_mark_uptodate(page, offset, count); 2901da177e4SLinus Torvalds nfs_unlock_request(req); 291abd3e641STrond Myklebust return 0; 2921da177e4SLinus Torvalds } 2931da177e4SLinus Torvalds 2941da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 2951da177e4SLinus Torvalds { 2961da177e4SLinus Torvalds if (wbc->for_reclaim) 2971da177e4SLinus Torvalds return FLUSH_HIGHPRI; 2981da177e4SLinus Torvalds if (wbc->for_kupdate) 2991da177e4SLinus Torvalds return FLUSH_LOWPRI; 3001da177e4SLinus Torvalds return 0; 3011da177e4SLinus Torvalds } 3021da177e4SLinus Torvalds 3031da177e4SLinus Torvalds /* 3041da177e4SLinus Torvalds * Write an mmapped page to the server. 3051da177e4SLinus Torvalds */ 3061da177e4SLinus Torvalds int nfs_writepage(struct page *page, struct writeback_control *wbc) 3071da177e4SLinus Torvalds { 3081da177e4SLinus Torvalds struct nfs_open_context *ctx; 3091da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 3101da177e4SLinus Torvalds unsigned long end_index; 3111da177e4SLinus Torvalds unsigned offset = PAGE_CACHE_SIZE; 3121da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 3131da177e4SLinus Torvalds int inode_referenced = 0; 3141da177e4SLinus Torvalds int priority = wb_priority(wbc); 3151da177e4SLinus Torvalds int err; 3161da177e4SLinus Torvalds 31791d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 31891d5b470SChuck Lever nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 31991d5b470SChuck Lever 3201da177e4SLinus Torvalds /* 3211da177e4SLinus Torvalds * Note: We need to ensure that we have a reference to the inode 3221da177e4SLinus Torvalds * if we are to do asynchronous writes. If not, waiting 3231da177e4SLinus Torvalds * in nfs_wait_on_request() may deadlock with clear_inode(). 3241da177e4SLinus Torvalds * 3251da177e4SLinus Torvalds * If igrab() fails here, then it is in any case safe to 3261da177e4SLinus Torvalds * call nfs_wb_page(), since there will be no pending writes. 3271da177e4SLinus Torvalds */ 3281da177e4SLinus Torvalds if (igrab(inode) != 0) 3291da177e4SLinus Torvalds inode_referenced = 1; 3301da177e4SLinus Torvalds end_index = i_size >> PAGE_CACHE_SHIFT; 3311da177e4SLinus Torvalds 3321da177e4SLinus Torvalds /* Ensure we've flushed out any previous writes */ 3331da177e4SLinus Torvalds nfs_wb_page_priority(inode, page, priority); 3341da177e4SLinus Torvalds 3351da177e4SLinus Torvalds /* easy case */ 3361da177e4SLinus Torvalds if (page->index < end_index) 3371da177e4SLinus Torvalds goto do_it; 3381da177e4SLinus Torvalds /* things got complicated... */ 3391da177e4SLinus Torvalds offset = i_size & (PAGE_CACHE_SIZE-1); 3401da177e4SLinus Torvalds 3411da177e4SLinus Torvalds /* OK, are we completely out? */ 3421da177e4SLinus Torvalds err = 0; /* potential race with truncate - ignore */ 3431da177e4SLinus Torvalds if (page->index >= end_index+1 || !offset) 3441da177e4SLinus Torvalds goto out; 3451da177e4SLinus Torvalds do_it: 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) 355abd3e641STrond Myklebust nfs_flush_inode(inode, 0, 0, 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 } 3941da177e4SLinus Torvalds err = nfs_flush_inode(inode, 0, 0, wb_priority(wbc)); 3951da177e4SLinus Torvalds if (err < 0) 3961da177e4SLinus Torvalds goto out; 39791d5b470SChuck Lever nfs_add_stats(inode, NFSIOS_WRITEPAGES, err); 3981da177e4SLinus Torvalds wbc->nr_to_write -= err; 3991da177e4SLinus Torvalds if (!wbc->nonblocking && wbc->sync_mode == WB_SYNC_ALL) { 4001da177e4SLinus Torvalds err = nfs_wait_on_requests(inode, 0, 0); 4011da177e4SLinus Torvalds if (err < 0) 4021da177e4SLinus Torvalds goto out; 4031da177e4SLinus Torvalds } 4043da28eb1STrond Myklebust err = nfs_commit_inode(inode, wb_priority(wbc)); 4051da177e4SLinus Torvalds if (err > 0) { 4061da177e4SLinus Torvalds wbc->nr_to_write -= err; 4071da177e4SLinus Torvalds err = 0; 4081da177e4SLinus Torvalds } 4091da177e4SLinus Torvalds out: 4101da177e4SLinus Torvalds clear_bit(BDI_write_congested, &bdi->state); 4111da177e4SLinus Torvalds wake_up_all(&nfs_write_congestion); 4123fcfab16SAndrew Morton congestion_end(WRITE); 4131da177e4SLinus Torvalds return err; 4141da177e4SLinus Torvalds } 4151da177e4SLinus Torvalds 4161da177e4SLinus Torvalds /* 4171da177e4SLinus Torvalds * Insert a write request into an inode 4181da177e4SLinus Torvalds */ 4191da177e4SLinus Torvalds static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 4201da177e4SLinus Torvalds { 4211da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4221da177e4SLinus Torvalds int error; 4231da177e4SLinus Torvalds 4241da177e4SLinus Torvalds error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); 4251da177e4SLinus Torvalds BUG_ON(error == -EEXIST); 4261da177e4SLinus Torvalds if (error) 4271da177e4SLinus Torvalds return error; 4281da177e4SLinus Torvalds if (!nfsi->npages) { 4291da177e4SLinus Torvalds igrab(inode); 4301da177e4SLinus Torvalds nfs_begin_data_update(inode); 4311da177e4SLinus Torvalds if (nfs_have_delegation(inode, FMODE_WRITE)) 4321da177e4SLinus Torvalds nfsi->change_attr++; 4331da177e4SLinus Torvalds } 434deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 4351da177e4SLinus Torvalds nfsi->npages++; 4361da177e4SLinus Torvalds atomic_inc(&req->wb_count); 4371da177e4SLinus Torvalds return 0; 4381da177e4SLinus Torvalds } 4391da177e4SLinus Torvalds 4401da177e4SLinus Torvalds /* 4411da177e4SLinus Torvalds * Insert a write request into an inode 4421da177e4SLinus Torvalds */ 4431da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 4441da177e4SLinus Torvalds { 4451da177e4SLinus Torvalds struct inode *inode = req->wb_context->dentry->d_inode; 4461da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4471da177e4SLinus Torvalds 4481da177e4SLinus Torvalds BUG_ON (!NFS_WBACK_BUSY(req)); 4491da177e4SLinus Torvalds 4501da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 451deb7d638STrond Myklebust ClearPagePrivate(req->wb_page); 4521da177e4SLinus Torvalds radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); 4531da177e4SLinus Torvalds nfsi->npages--; 4541da177e4SLinus Torvalds if (!nfsi->npages) { 4551da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 456951a143bSTrond Myklebust nfs_end_data_update(inode); 4571da177e4SLinus Torvalds iput(inode); 4581da177e4SLinus Torvalds } else 4591da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 4601da177e4SLinus Torvalds nfs_clear_request(req); 4611da177e4SLinus Torvalds nfs_release_request(req); 4621da177e4SLinus Torvalds } 4631da177e4SLinus Torvalds 4641da177e4SLinus Torvalds /* 4651da177e4SLinus Torvalds * Find a request 4661da177e4SLinus Torvalds */ 4671da177e4SLinus Torvalds static inline struct nfs_page * 4681da177e4SLinus Torvalds _nfs_find_request(struct inode *inode, unsigned long index) 4691da177e4SLinus Torvalds { 4701da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4711da177e4SLinus Torvalds struct nfs_page *req; 4721da177e4SLinus Torvalds 4731da177e4SLinus Torvalds req = (struct nfs_page*)radix_tree_lookup(&nfsi->nfs_page_tree, index); 4741da177e4SLinus Torvalds if (req) 4751da177e4SLinus Torvalds atomic_inc(&req->wb_count); 4761da177e4SLinus Torvalds return req; 4771da177e4SLinus Torvalds } 4781da177e4SLinus Torvalds 4791da177e4SLinus Torvalds static struct nfs_page * 4801da177e4SLinus Torvalds nfs_find_request(struct inode *inode, unsigned long index) 4811da177e4SLinus Torvalds { 4821da177e4SLinus Torvalds struct nfs_page *req; 4831da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4841da177e4SLinus Torvalds 4851da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 4861da177e4SLinus Torvalds req = _nfs_find_request(inode, index); 4871da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 4881da177e4SLinus Torvalds return req; 4891da177e4SLinus Torvalds } 4901da177e4SLinus Torvalds 4911da177e4SLinus Torvalds /* 4921da177e4SLinus Torvalds * Add a request to the inode's dirty list. 4931da177e4SLinus Torvalds */ 4941da177e4SLinus Torvalds static void 4951da177e4SLinus Torvalds nfs_mark_request_dirty(struct nfs_page *req) 4961da177e4SLinus Torvalds { 4971da177e4SLinus Torvalds struct inode *inode = req->wb_context->dentry->d_inode; 4981da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4991da177e4SLinus Torvalds 5001da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 5013da28eb1STrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, 5023da28eb1STrond Myklebust req->wb_index, NFS_PAGE_TAG_DIRTY); 5031da177e4SLinus Torvalds nfs_list_add_request(req, &nfsi->dirty); 5041da177e4SLinus Torvalds nfsi->ndirty++; 5051da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 506b1e7a8fdSChristoph Lameter inc_zone_page_state(req->wb_page, NR_FILE_DIRTY); 5071da177e4SLinus Torvalds mark_inode_dirty(inode); 5081da177e4SLinus Torvalds } 5091da177e4SLinus Torvalds 5101da177e4SLinus Torvalds /* 5111da177e4SLinus Torvalds * Check if a request is dirty 5121da177e4SLinus Torvalds */ 5131da177e4SLinus Torvalds static inline int 5141da177e4SLinus Torvalds nfs_dirty_request(struct nfs_page *req) 5151da177e4SLinus Torvalds { 5161da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode); 5171da177e4SLinus Torvalds return !list_empty(&req->wb_list) && req->wb_list_head == &nfsi->dirty; 5181da177e4SLinus Torvalds } 5191da177e4SLinus Torvalds 5201da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 5211da177e4SLinus Torvalds /* 5221da177e4SLinus Torvalds * Add a request to the inode's commit list. 5231da177e4SLinus Torvalds */ 5241da177e4SLinus Torvalds static void 5251da177e4SLinus Torvalds nfs_mark_request_commit(struct nfs_page *req) 5261da177e4SLinus Torvalds { 5271da177e4SLinus Torvalds struct inode *inode = req->wb_context->dentry->d_inode; 5281da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5291da177e4SLinus Torvalds 5301da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 5311da177e4SLinus Torvalds nfs_list_add_request(req, &nfsi->commit); 5321da177e4SLinus Torvalds nfsi->ncommit++; 5331da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 534fd39fc85SChristoph Lameter inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 5351da177e4SLinus Torvalds mark_inode_dirty(inode); 5361da177e4SLinus Torvalds } 5371da177e4SLinus Torvalds #endif 5381da177e4SLinus Torvalds 5391da177e4SLinus Torvalds /* 5401da177e4SLinus Torvalds * Wait for a request to complete. 5411da177e4SLinus Torvalds * 5421da177e4SLinus Torvalds * Interruptible by signals only if mounted with intr flag. 5431da177e4SLinus Torvalds */ 544c42de9ddSTrond Myklebust static int nfs_wait_on_requests_locked(struct inode *inode, unsigned long idx_start, unsigned int npages) 5451da177e4SLinus Torvalds { 5461da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5471da177e4SLinus Torvalds struct nfs_page *req; 5481da177e4SLinus Torvalds unsigned long idx_end, next; 5491da177e4SLinus Torvalds unsigned int res = 0; 5501da177e4SLinus Torvalds int error; 5511da177e4SLinus Torvalds 5521da177e4SLinus Torvalds if (npages == 0) 5531da177e4SLinus Torvalds idx_end = ~0; 5541da177e4SLinus Torvalds else 5551da177e4SLinus Torvalds idx_end = idx_start + npages - 1; 5561da177e4SLinus Torvalds 5571da177e4SLinus Torvalds next = idx_start; 558c6a556b8STrond Myklebust while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_WRITEBACK)) { 5591da177e4SLinus Torvalds if (req->wb_index > idx_end) 5601da177e4SLinus Torvalds break; 5611da177e4SLinus Torvalds 5621da177e4SLinus Torvalds next = req->wb_index + 1; 563c6a556b8STrond Myklebust BUG_ON(!NFS_WBACK_BUSY(req)); 5641da177e4SLinus Torvalds 5651da177e4SLinus Torvalds atomic_inc(&req->wb_count); 5661da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 5671da177e4SLinus Torvalds error = nfs_wait_on_request(req); 5681da177e4SLinus Torvalds nfs_release_request(req); 569c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 5701da177e4SLinus Torvalds if (error < 0) 5711da177e4SLinus Torvalds return error; 5721da177e4SLinus Torvalds res++; 5731da177e4SLinus Torvalds } 5741da177e4SLinus Torvalds return res; 5751da177e4SLinus Torvalds } 5761da177e4SLinus Torvalds 577c42de9ddSTrond Myklebust static int nfs_wait_on_requests(struct inode *inode, unsigned long idx_start, unsigned int npages) 578c42de9ddSTrond Myklebust { 579c42de9ddSTrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 580c42de9ddSTrond Myklebust int ret; 581c42de9ddSTrond Myklebust 582c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 583c42de9ddSTrond Myklebust ret = nfs_wait_on_requests_locked(inode, idx_start, npages); 584c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 585c42de9ddSTrond Myklebust return ret; 586c42de9ddSTrond Myklebust } 587c42de9ddSTrond Myklebust 58883715ad5STrond Myklebust static void nfs_cancel_dirty_list(struct list_head *head) 589d2ccddf0STrond Myklebust { 590d2ccddf0STrond Myklebust struct nfs_page *req; 591d2ccddf0STrond Myklebust while(!list_empty(head)) { 592d2ccddf0STrond Myklebust req = nfs_list_entry(head->next); 593d2ccddf0STrond Myklebust nfs_list_remove_request(req); 594d2ccddf0STrond Myklebust nfs_inode_remove_request(req); 595d2ccddf0STrond Myklebust nfs_clear_page_writeback(req); 596d2ccddf0STrond Myklebust } 597d2ccddf0STrond Myklebust } 598d2ccddf0STrond Myklebust 59983715ad5STrond Myklebust static void nfs_cancel_commit_list(struct list_head *head) 60083715ad5STrond Myklebust { 60183715ad5STrond Myklebust struct nfs_page *req; 60283715ad5STrond Myklebust 60383715ad5STrond Myklebust while(!list_empty(head)) { 60483715ad5STrond Myklebust req = nfs_list_entry(head->next); 605b6dff26aSTrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 60683715ad5STrond Myklebust nfs_list_remove_request(req); 60783715ad5STrond Myklebust nfs_inode_remove_request(req); 608b6dff26aSTrond Myklebust nfs_unlock_request(req); 60983715ad5STrond Myklebust } 61083715ad5STrond Myklebust } 61183715ad5STrond Myklebust 6121da177e4SLinus Torvalds /* 6131da177e4SLinus Torvalds * nfs_scan_dirty - Scan an inode for dirty requests 6141da177e4SLinus Torvalds * @inode: NFS inode to scan 6151da177e4SLinus Torvalds * @dst: destination list 6161da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 6171da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 6181da177e4SLinus Torvalds * 6191da177e4SLinus Torvalds * Moves requests from the inode's dirty page list. 6201da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 6211da177e4SLinus Torvalds */ 6221da177e4SLinus Torvalds static int 6231da177e4SLinus Torvalds nfs_scan_dirty(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) 6241da177e4SLinus Torvalds { 6251da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 6263da28eb1STrond Myklebust int res = 0; 6273da28eb1STrond Myklebust 6283da28eb1STrond Myklebust if (nfsi->ndirty != 0) { 6293da28eb1STrond Myklebust res = nfs_scan_lock_dirty(nfsi, dst, idx_start, npages); 6301da177e4SLinus Torvalds nfsi->ndirty -= res; 6311da177e4SLinus Torvalds if ((nfsi->ndirty == 0) != list_empty(&nfsi->dirty)) 6321da177e4SLinus Torvalds printk(KERN_ERR "NFS: desynchronized value of nfs_i.ndirty.\n"); 6333da28eb1STrond Myklebust } 6341da177e4SLinus Torvalds return res; 6351da177e4SLinus Torvalds } 6361da177e4SLinus Torvalds 6371da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 6381da177e4SLinus Torvalds /* 6391da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 6401da177e4SLinus Torvalds * @inode: NFS inode to scan 6411da177e4SLinus Torvalds * @dst: destination list 6421da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 6431da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 6441da177e4SLinus Torvalds * 6451da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 6461da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 6471da177e4SLinus Torvalds */ 6481da177e4SLinus Torvalds static int 6491da177e4SLinus Torvalds nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) 6501da177e4SLinus Torvalds { 6511da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 6523da28eb1STrond Myklebust int res = 0; 6533da28eb1STrond Myklebust 6543da28eb1STrond Myklebust if (nfsi->ncommit != 0) { 655d2ccddf0STrond Myklebust res = nfs_scan_list(nfsi, &nfsi->commit, dst, idx_start, npages); 6561da177e4SLinus Torvalds nfsi->ncommit -= res; 6571da177e4SLinus Torvalds if ((nfsi->ncommit == 0) != list_empty(&nfsi->commit)) 6581da177e4SLinus Torvalds printk(KERN_ERR "NFS: desynchronized value of nfs_i.ncommit.\n"); 6593da28eb1STrond Myklebust } 6601da177e4SLinus Torvalds return res; 6611da177e4SLinus Torvalds } 662c42de9ddSTrond Myklebust #else 663c42de9ddSTrond Myklebust static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) 664c42de9ddSTrond Myklebust { 665c42de9ddSTrond Myklebust return 0; 666c42de9ddSTrond Myklebust } 6671da177e4SLinus Torvalds #endif 6681da177e4SLinus Torvalds 6691da177e4SLinus Torvalds static int nfs_wait_on_write_congestion(struct address_space *mapping, int intr) 6701da177e4SLinus Torvalds { 6711da177e4SLinus Torvalds struct backing_dev_info *bdi = mapping->backing_dev_info; 6721da177e4SLinus Torvalds DEFINE_WAIT(wait); 6731da177e4SLinus Torvalds int ret = 0; 6741da177e4SLinus Torvalds 6751da177e4SLinus Torvalds might_sleep(); 6761da177e4SLinus Torvalds 6771da177e4SLinus Torvalds if (!bdi_write_congested(bdi)) 6781da177e4SLinus Torvalds return 0; 67991d5b470SChuck Lever 68091d5b470SChuck Lever nfs_inc_stats(mapping->host, NFSIOS_CONGESTIONWAIT); 68191d5b470SChuck Lever 6821da177e4SLinus Torvalds if (intr) { 6831da177e4SLinus Torvalds struct rpc_clnt *clnt = NFS_CLIENT(mapping->host); 6841da177e4SLinus Torvalds sigset_t oldset; 6851da177e4SLinus Torvalds 6861da177e4SLinus Torvalds rpc_clnt_sigmask(clnt, &oldset); 6871da177e4SLinus Torvalds prepare_to_wait(&nfs_write_congestion, &wait, TASK_INTERRUPTIBLE); 6881da177e4SLinus Torvalds if (bdi_write_congested(bdi)) { 6891da177e4SLinus Torvalds if (signalled()) 6901da177e4SLinus Torvalds ret = -ERESTARTSYS; 6911da177e4SLinus Torvalds else 6921da177e4SLinus Torvalds schedule(); 6931da177e4SLinus Torvalds } 6941da177e4SLinus Torvalds rpc_clnt_sigunmask(clnt, &oldset); 6951da177e4SLinus Torvalds } else { 6961da177e4SLinus Torvalds prepare_to_wait(&nfs_write_congestion, &wait, TASK_UNINTERRUPTIBLE); 6971da177e4SLinus Torvalds if (bdi_write_congested(bdi)) 6981da177e4SLinus Torvalds schedule(); 6991da177e4SLinus Torvalds } 7001da177e4SLinus Torvalds finish_wait(&nfs_write_congestion, &wait); 7011da177e4SLinus Torvalds return ret; 7021da177e4SLinus Torvalds } 7031da177e4SLinus Torvalds 7041da177e4SLinus Torvalds 7051da177e4SLinus Torvalds /* 7061da177e4SLinus Torvalds * Try to update any existing write request, or create one if there is none. 7071da177e4SLinus Torvalds * In order to match, the request's credentials must match those of 7081da177e4SLinus Torvalds * the calling process. 7091da177e4SLinus Torvalds * 7101da177e4SLinus Torvalds * Note: Should always be called with the Page Lock held! 7111da177e4SLinus Torvalds */ 7121da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx, 7131da177e4SLinus Torvalds struct inode *inode, struct page *page, 7141da177e4SLinus Torvalds unsigned int offset, unsigned int bytes) 7151da177e4SLinus Torvalds { 7161da177e4SLinus Torvalds struct nfs_server *server = NFS_SERVER(inode); 7171da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 7181da177e4SLinus Torvalds struct nfs_page *req, *new = NULL; 7191da177e4SLinus Torvalds unsigned long rqend, end; 7201da177e4SLinus Torvalds 7211da177e4SLinus Torvalds end = offset + bytes; 7221da177e4SLinus Torvalds 7231da177e4SLinus Torvalds if (nfs_wait_on_write_congestion(page->mapping, server->flags & NFS_MOUNT_INTR)) 7241da177e4SLinus Torvalds return ERR_PTR(-ERESTARTSYS); 7251da177e4SLinus Torvalds for (;;) { 7261da177e4SLinus Torvalds /* Loop over all inode entries and see if we find 7271da177e4SLinus Torvalds * A request for the page we wish to update 7281da177e4SLinus Torvalds */ 7291da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 7301da177e4SLinus Torvalds req = _nfs_find_request(inode, page->index); 7311da177e4SLinus Torvalds if (req) { 7321da177e4SLinus Torvalds if (!nfs_lock_request_dontget(req)) { 7331da177e4SLinus Torvalds int error; 7341da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 7351da177e4SLinus Torvalds error = nfs_wait_on_request(req); 7361da177e4SLinus Torvalds nfs_release_request(req); 7371dd594b2SNeil Brown if (error < 0) { 7381dd594b2SNeil Brown if (new) 7391dd594b2SNeil Brown nfs_release_request(new); 7401da177e4SLinus Torvalds return ERR_PTR(error); 7411dd594b2SNeil Brown } 7421da177e4SLinus Torvalds continue; 7431da177e4SLinus Torvalds } 7441da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 7451da177e4SLinus Torvalds if (new) 7461da177e4SLinus Torvalds nfs_release_request(new); 7471da177e4SLinus Torvalds break; 7481da177e4SLinus Torvalds } 7491da177e4SLinus Torvalds 7501da177e4SLinus Torvalds if (new) { 7511da177e4SLinus Torvalds int error; 7521da177e4SLinus Torvalds nfs_lock_request_dontget(new); 7531da177e4SLinus Torvalds error = nfs_inode_add_request(inode, new); 7541da177e4SLinus Torvalds if (error) { 7551da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 7561da177e4SLinus Torvalds nfs_unlock_request(new); 7571da177e4SLinus Torvalds return ERR_PTR(error); 7581da177e4SLinus Torvalds } 7591da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 7601da177e4SLinus Torvalds nfs_mark_request_dirty(new); 7611da177e4SLinus Torvalds return new; 7621da177e4SLinus Torvalds } 7631da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 7641da177e4SLinus Torvalds 7651da177e4SLinus Torvalds new = nfs_create_request(ctx, inode, page, offset, bytes); 7661da177e4SLinus Torvalds if (IS_ERR(new)) 7671da177e4SLinus Torvalds return new; 7681da177e4SLinus Torvalds } 7691da177e4SLinus Torvalds 7701da177e4SLinus Torvalds /* We have a request for our page. 7711da177e4SLinus Torvalds * If the creds don't match, or the 7721da177e4SLinus Torvalds * page addresses don't match, 7731da177e4SLinus Torvalds * tell the caller to wait on the conflicting 7741da177e4SLinus Torvalds * request. 7751da177e4SLinus Torvalds */ 7761da177e4SLinus Torvalds rqend = req->wb_offset + req->wb_bytes; 7771da177e4SLinus Torvalds if (req->wb_context != ctx 7781da177e4SLinus Torvalds || req->wb_page != page 7791da177e4SLinus Torvalds || !nfs_dirty_request(req) 7801da177e4SLinus Torvalds || offset > rqend || end < req->wb_offset) { 7811da177e4SLinus Torvalds nfs_unlock_request(req); 7821da177e4SLinus Torvalds return ERR_PTR(-EBUSY); 7831da177e4SLinus Torvalds } 7841da177e4SLinus Torvalds 7851da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 7861da177e4SLinus Torvalds if (offset < req->wb_offset) { 7871da177e4SLinus Torvalds req->wb_offset = offset; 7881da177e4SLinus Torvalds req->wb_pgbase = offset; 7891da177e4SLinus Torvalds req->wb_bytes = rqend - req->wb_offset; 7901da177e4SLinus Torvalds } 7911da177e4SLinus Torvalds 7921da177e4SLinus Torvalds if (end > rqend) 7931da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 7941da177e4SLinus Torvalds 7951da177e4SLinus Torvalds return req; 7961da177e4SLinus Torvalds } 7971da177e4SLinus Torvalds 7981da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 7991da177e4SLinus Torvalds { 8001da177e4SLinus Torvalds struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; 8011da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 8021da177e4SLinus Torvalds struct nfs_page *req; 8031da177e4SLinus Torvalds int status = 0; 8041da177e4SLinus Torvalds /* 8051da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 8061da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 8071da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 8081da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 8091da177e4SLinus Torvalds * Also do the same if we find a request from an existing 8101da177e4SLinus Torvalds * dropped page. 8111da177e4SLinus Torvalds */ 8121da177e4SLinus Torvalds req = nfs_find_request(inode, page->index); 8131da177e4SLinus Torvalds if (req) { 8141da177e4SLinus Torvalds if (req->wb_page != page || ctx != req->wb_context) 8151da177e4SLinus Torvalds status = nfs_wb_page(inode, page); 8161da177e4SLinus Torvalds nfs_release_request(req); 8171da177e4SLinus Torvalds } 8181da177e4SLinus Torvalds return (status < 0) ? status : 0; 8191da177e4SLinus Torvalds } 8201da177e4SLinus Torvalds 8211da177e4SLinus Torvalds /* 8221da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 8231da177e4SLinus Torvalds * 8241da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 8251da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 8261da177e4SLinus Torvalds */ 8271da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 8281da177e4SLinus Torvalds unsigned int offset, unsigned int count) 8291da177e4SLinus Torvalds { 8301da177e4SLinus Torvalds struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; 8311da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 8321da177e4SLinus Torvalds struct nfs_page *req; 8331da177e4SLinus Torvalds int status = 0; 8341da177e4SLinus Torvalds 83591d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 83691d5b470SChuck Lever 8371da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage(%s/%s %d@%Ld)\n", 8380bbacc40SChuck Lever file->f_dentry->d_parent->d_name.name, 8390bbacc40SChuck Lever file->f_dentry->d_name.name, count, 8400bbacc40SChuck Lever (long long)(page_offset(page) +offset)); 8411da177e4SLinus Torvalds 8421da177e4SLinus Torvalds if (IS_SYNC(inode)) { 8431da177e4SLinus Torvalds status = nfs_writepage_sync(ctx, inode, page, offset, count, 0); 8441da177e4SLinus Torvalds if (status > 0) { 8451da177e4SLinus Torvalds if (offset == 0 && status == PAGE_CACHE_SIZE) 8461da177e4SLinus Torvalds SetPageUptodate(page); 8471da177e4SLinus Torvalds return 0; 8481da177e4SLinus Torvalds } 8491da177e4SLinus Torvalds return status; 8501da177e4SLinus Torvalds } 8511da177e4SLinus Torvalds 8521da177e4SLinus Torvalds /* If we're not using byte range locks, and we know the page 8531da177e4SLinus Torvalds * is entirely in cache, it may be more efficient to avoid 8541da177e4SLinus Torvalds * fragmenting write requests. 8551da177e4SLinus Torvalds */ 856ab0a3dbeSTrond Myklebust if (PageUptodate(page) && inode->i_flock == NULL && !(file->f_mode & O_SYNC)) { 8571da177e4SLinus Torvalds loff_t end_offs = i_size_read(inode) - 1; 8581da177e4SLinus Torvalds unsigned long end_index = end_offs >> PAGE_CACHE_SHIFT; 8591da177e4SLinus Torvalds 8601da177e4SLinus Torvalds count += offset; 8611da177e4SLinus Torvalds offset = 0; 8621da177e4SLinus Torvalds if (unlikely(end_offs < 0)) { 8631da177e4SLinus Torvalds /* Do nothing */ 8641da177e4SLinus Torvalds } else if (page->index == end_index) { 8651da177e4SLinus Torvalds unsigned int pglen; 8661da177e4SLinus Torvalds pglen = (unsigned int)(end_offs & (PAGE_CACHE_SIZE-1)) + 1; 8671da177e4SLinus Torvalds if (count < pglen) 8681da177e4SLinus Torvalds count = pglen; 8691da177e4SLinus Torvalds } else if (page->index < end_index) 8701da177e4SLinus Torvalds count = PAGE_CACHE_SIZE; 8711da177e4SLinus Torvalds } 8721da177e4SLinus Torvalds 8731da177e4SLinus Torvalds /* 8741da177e4SLinus Torvalds * Try to find an NFS request corresponding to this page 8751da177e4SLinus Torvalds * and update it. 8761da177e4SLinus Torvalds * If the existing request cannot be updated, we must flush 8771da177e4SLinus Torvalds * it out now. 8781da177e4SLinus Torvalds */ 8791da177e4SLinus Torvalds do { 8801da177e4SLinus Torvalds req = nfs_update_request(ctx, inode, page, offset, count); 8811da177e4SLinus Torvalds status = (IS_ERR(req)) ? PTR_ERR(req) : 0; 8821da177e4SLinus Torvalds if (status != -EBUSY) 8831da177e4SLinus Torvalds break; 8841da177e4SLinus Torvalds /* Request could not be updated. Flush it out and try again */ 8851da177e4SLinus Torvalds status = nfs_wb_page(inode, page); 8861da177e4SLinus Torvalds } while (status >= 0); 8871da177e4SLinus Torvalds if (status < 0) 8881da177e4SLinus Torvalds goto done; 8891da177e4SLinus Torvalds 8901da177e4SLinus Torvalds status = 0; 8911da177e4SLinus Torvalds 8921da177e4SLinus Torvalds /* Update file length */ 8931da177e4SLinus Torvalds nfs_grow_file(page, offset, count); 8941da177e4SLinus Torvalds /* Set the PG_uptodate flag? */ 8951da177e4SLinus Torvalds nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 8961da177e4SLinus Torvalds nfs_unlock_request(req); 8971da177e4SLinus Torvalds done: 8981da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage returns %d (isize %Ld)\n", 8991da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 9001da177e4SLinus Torvalds if (status < 0) 9011da177e4SLinus Torvalds ClearPageUptodate(page); 9021da177e4SLinus Torvalds return status; 9031da177e4SLinus Torvalds } 9041da177e4SLinus Torvalds 9051da177e4SLinus Torvalds static void nfs_writepage_release(struct nfs_page *req) 9061da177e4SLinus Torvalds { 9071da177e4SLinus Torvalds end_page_writeback(req->wb_page); 9081da177e4SLinus Torvalds 9091da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 9101da177e4SLinus Torvalds if (!PageError(req->wb_page)) { 9111da177e4SLinus Torvalds if (NFS_NEED_RESCHED(req)) { 9121da177e4SLinus Torvalds nfs_mark_request_dirty(req); 9131da177e4SLinus Torvalds goto out; 9141da177e4SLinus Torvalds } else if (NFS_NEED_COMMIT(req)) { 9151da177e4SLinus Torvalds nfs_mark_request_commit(req); 9161da177e4SLinus Torvalds goto out; 9171da177e4SLinus Torvalds } 9181da177e4SLinus Torvalds } 9191da177e4SLinus Torvalds nfs_inode_remove_request(req); 9201da177e4SLinus Torvalds 9211da177e4SLinus Torvalds out: 9221da177e4SLinus Torvalds nfs_clear_commit(req); 9231da177e4SLinus Torvalds nfs_clear_reschedule(req); 9241da177e4SLinus Torvalds #else 9251da177e4SLinus Torvalds nfs_inode_remove_request(req); 9261da177e4SLinus Torvalds #endif 927c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 9281da177e4SLinus Torvalds } 9291da177e4SLinus Torvalds 9301da177e4SLinus Torvalds static inline int flush_task_priority(int how) 9311da177e4SLinus Torvalds { 9321da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 9331da177e4SLinus Torvalds case FLUSH_HIGHPRI: 9341da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 9351da177e4SLinus Torvalds case FLUSH_LOWPRI: 9361da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 9371da177e4SLinus Torvalds } 9381da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 9391da177e4SLinus Torvalds } 9401da177e4SLinus Torvalds 9411da177e4SLinus Torvalds /* 9421da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 9431da177e4SLinus Torvalds */ 9441da177e4SLinus Torvalds static void nfs_write_rpcsetup(struct nfs_page *req, 9451da177e4SLinus Torvalds struct nfs_write_data *data, 946788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 9471da177e4SLinus Torvalds unsigned int count, unsigned int offset, 9481da177e4SLinus Torvalds int how) 9491da177e4SLinus Torvalds { 9501da177e4SLinus Torvalds struct inode *inode; 951788e7a89STrond Myklebust int flags; 9521da177e4SLinus Torvalds 9531da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 9541da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 9551da177e4SLinus Torvalds 9561da177e4SLinus Torvalds data->req = req; 9571da177e4SLinus Torvalds data->inode = inode = req->wb_context->dentry->d_inode; 9581da177e4SLinus Torvalds data->cred = req->wb_context->cred; 9591da177e4SLinus Torvalds 9601da177e4SLinus Torvalds data->args.fh = NFS_FH(inode); 9611da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 9621da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 9631da177e4SLinus Torvalds data->args.pages = data->pagevec; 9641da177e4SLinus Torvalds data->args.count = count; 9651da177e4SLinus Torvalds data->args.context = req->wb_context; 9661da177e4SLinus Torvalds 9671da177e4SLinus Torvalds data->res.fattr = &data->fattr; 9681da177e4SLinus Torvalds data->res.count = count; 9691da177e4SLinus Torvalds data->res.verf = &data->verf; 9700e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 9711da177e4SLinus Torvalds 972788e7a89STrond Myklebust /* Set up the initial task struct. */ 973788e7a89STrond Myklebust flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 974788e7a89STrond Myklebust rpc_init_task(&data->task, NFS_CLIENT(inode), flags, call_ops, data); 9751da177e4SLinus Torvalds NFS_PROTO(inode)->write_setup(data, how); 9761da177e4SLinus Torvalds 9771da177e4SLinus Torvalds data->task.tk_priority = flush_task_priority(how); 9781da177e4SLinus Torvalds data->task.tk_cookie = (unsigned long)inode; 9791da177e4SLinus Torvalds 9801da177e4SLinus Torvalds dprintk("NFS: %4d initiated write call (req %s/%Ld, %u bytes @ offset %Lu)\n", 9810bbacc40SChuck Lever data->task.tk_pid, 9821da177e4SLinus Torvalds inode->i_sb->s_id, 9831da177e4SLinus Torvalds (long long)NFS_FILEID(inode), 9841da177e4SLinus Torvalds count, 9851da177e4SLinus Torvalds (unsigned long long)data->args.offset); 9861da177e4SLinus Torvalds } 9871da177e4SLinus Torvalds 9881da177e4SLinus Torvalds static void nfs_execute_write(struct nfs_write_data *data) 9891da177e4SLinus Torvalds { 9901da177e4SLinus Torvalds struct rpc_clnt *clnt = NFS_CLIENT(data->inode); 9911da177e4SLinus Torvalds sigset_t oldset; 9921da177e4SLinus Torvalds 9931da177e4SLinus Torvalds rpc_clnt_sigmask(clnt, &oldset); 9941da177e4SLinus Torvalds lock_kernel(); 9951da177e4SLinus Torvalds rpc_execute(&data->task); 9961da177e4SLinus Torvalds unlock_kernel(); 9971da177e4SLinus Torvalds rpc_clnt_sigunmask(clnt, &oldset); 9981da177e4SLinus Torvalds } 9991da177e4SLinus Torvalds 10001da177e4SLinus Torvalds /* 10011da177e4SLinus Torvalds * Generate multiple small requests to write out a single 10021da177e4SLinus Torvalds * contiguous dirty area on one page. 10031da177e4SLinus Torvalds */ 10047d46a49fSTrond Myklebust static int nfs_flush_multi(struct inode *inode, struct list_head *head, int how) 10051da177e4SLinus Torvalds { 10061da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 10071da177e4SLinus Torvalds struct page *page = req->wb_page; 10081da177e4SLinus Torvalds struct nfs_write_data *data; 1009e9f7bee1STrond Myklebust size_t wsize = NFS_SERVER(inode)->wsize, nbytes; 1010e9f7bee1STrond Myklebust unsigned int offset; 10111da177e4SLinus Torvalds int requests = 0; 10121da177e4SLinus Torvalds LIST_HEAD(list); 10131da177e4SLinus Torvalds 10141da177e4SLinus Torvalds nfs_list_remove_request(req); 10151da177e4SLinus Torvalds 10161da177e4SLinus Torvalds nbytes = req->wb_bytes; 1017e9f7bee1STrond Myklebust do { 1018e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 1019e9f7bee1STrond Myklebust 1020e9f7bee1STrond Myklebust data = nfs_writedata_alloc(len); 10211da177e4SLinus Torvalds if (!data) 10221da177e4SLinus Torvalds goto out_bad; 10231da177e4SLinus Torvalds list_add(&data->pages, &list); 10241da177e4SLinus Torvalds requests++; 1025e9f7bee1STrond Myklebust nbytes -= len; 1026e9f7bee1STrond Myklebust } while (nbytes != 0); 10271da177e4SLinus Torvalds atomic_set(&req->wb_complete, requests); 10281da177e4SLinus Torvalds 10291da177e4SLinus Torvalds ClearPageError(page); 1030bb713d6dSTrond Myklebust set_page_writeback(page); 10311da177e4SLinus Torvalds offset = 0; 10321da177e4SLinus Torvalds nbytes = req->wb_bytes; 10331da177e4SLinus Torvalds do { 10341da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 10351da177e4SLinus Torvalds list_del_init(&data->pages); 10361da177e4SLinus Torvalds 10371da177e4SLinus Torvalds data->pagevec[0] = page; 10381da177e4SLinus Torvalds 10391da177e4SLinus Torvalds if (nbytes > wsize) { 1040788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 1041788e7a89STrond Myklebust wsize, offset, how); 10421da177e4SLinus Torvalds offset += wsize; 10431da177e4SLinus Torvalds nbytes -= wsize; 10441da177e4SLinus Torvalds } else { 1045788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 1046788e7a89STrond Myklebust nbytes, offset, how); 10471da177e4SLinus Torvalds nbytes = 0; 10481da177e4SLinus Torvalds } 10491da177e4SLinus Torvalds nfs_execute_write(data); 10501da177e4SLinus Torvalds } while (nbytes != 0); 10511da177e4SLinus Torvalds 10521da177e4SLinus Torvalds return 0; 10531da177e4SLinus Torvalds 10541da177e4SLinus Torvalds out_bad: 10551da177e4SLinus Torvalds while (!list_empty(&list)) { 10561da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 10571da177e4SLinus Torvalds list_del(&data->pages); 1058*8aca67f0STrond Myklebust nfs_writedata_release(data); 10591da177e4SLinus Torvalds } 10601da177e4SLinus Torvalds nfs_mark_request_dirty(req); 1061c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 10621da177e4SLinus Torvalds return -ENOMEM; 10631da177e4SLinus Torvalds } 10641da177e4SLinus Torvalds 10651da177e4SLinus Torvalds /* 10661da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 10671da177e4SLinus Torvalds * The page must have been locked by the caller. 10681da177e4SLinus Torvalds * 10691da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 10701da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 10711da177e4SLinus Torvalds * that has been written but not committed. 10721da177e4SLinus Torvalds */ 10737d46a49fSTrond Myklebust static int nfs_flush_one(struct inode *inode, struct list_head *head, int how) 10741da177e4SLinus Torvalds { 10751da177e4SLinus Torvalds struct nfs_page *req; 10761da177e4SLinus Torvalds struct page **pages; 10771da177e4SLinus Torvalds struct nfs_write_data *data; 10781da177e4SLinus Torvalds unsigned int count; 10791da177e4SLinus Torvalds 1080e9f7bee1STrond Myklebust data = nfs_writedata_alloc(NFS_SERVER(inode)->wsize); 10811da177e4SLinus Torvalds if (!data) 10821da177e4SLinus Torvalds goto out_bad; 10831da177e4SLinus Torvalds 10841da177e4SLinus Torvalds pages = data->pagevec; 10851da177e4SLinus Torvalds count = 0; 10861da177e4SLinus Torvalds while (!list_empty(head)) { 10871da177e4SLinus Torvalds req = nfs_list_entry(head->next); 10881da177e4SLinus Torvalds nfs_list_remove_request(req); 10891da177e4SLinus Torvalds nfs_list_add_request(req, &data->pages); 10901da177e4SLinus Torvalds ClearPageError(req->wb_page); 1091bb713d6dSTrond Myklebust set_page_writeback(req->wb_page); 10921da177e4SLinus Torvalds *pages++ = req->wb_page; 10931da177e4SLinus Torvalds count += req->wb_bytes; 10941da177e4SLinus Torvalds } 10951da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 10961da177e4SLinus Torvalds 10971da177e4SLinus Torvalds /* Set up the argument struct */ 1098788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how); 10991da177e4SLinus Torvalds 11001da177e4SLinus Torvalds nfs_execute_write(data); 11011da177e4SLinus Torvalds return 0; 11021da177e4SLinus Torvalds out_bad: 11031da177e4SLinus Torvalds while (!list_empty(head)) { 11041da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 11051da177e4SLinus Torvalds nfs_list_remove_request(req); 11061da177e4SLinus Torvalds nfs_mark_request_dirty(req); 1107c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 11081da177e4SLinus Torvalds } 11091da177e4SLinus Torvalds return -ENOMEM; 11101da177e4SLinus Torvalds } 11111da177e4SLinus Torvalds 11127d46a49fSTrond Myklebust static int nfs_flush_list(struct inode *inode, struct list_head *head, int npages, int how) 11131da177e4SLinus Torvalds { 11141da177e4SLinus Torvalds LIST_HEAD(one_request); 11157d46a49fSTrond Myklebust int (*flush_one)(struct inode *, struct list_head *, int); 11161da177e4SLinus Torvalds struct nfs_page *req; 11177d46a49fSTrond Myklebust int wpages = NFS_SERVER(inode)->wpages; 11187d46a49fSTrond Myklebust int wsize = NFS_SERVER(inode)->wsize; 11197d46a49fSTrond Myklebust int error; 11201da177e4SLinus Torvalds 11217d46a49fSTrond Myklebust flush_one = nfs_flush_one; 11227d46a49fSTrond Myklebust if (wsize < PAGE_CACHE_SIZE) 11237d46a49fSTrond Myklebust flush_one = nfs_flush_multi; 11247d46a49fSTrond Myklebust /* For single writes, FLUSH_STABLE is more efficient */ 11257d46a49fSTrond Myklebust if (npages <= wpages && npages == NFS_I(inode)->npages 11267d46a49fSTrond Myklebust && nfs_list_entry(head->next)->wb_bytes <= wsize) 11277d46a49fSTrond Myklebust how |= FLUSH_STABLE; 11287d46a49fSTrond Myklebust 11297d46a49fSTrond Myklebust do { 11307d46a49fSTrond Myklebust nfs_coalesce_requests(head, &one_request, wpages); 11311da177e4SLinus Torvalds req = nfs_list_entry(one_request.next); 11327d46a49fSTrond Myklebust error = flush_one(inode, &one_request, how); 11331da177e4SLinus Torvalds if (error < 0) 11347d46a49fSTrond Myklebust goto out_err; 11357d46a49fSTrond Myklebust } while (!list_empty(head)); 11367d46a49fSTrond Myklebust return 0; 11377d46a49fSTrond Myklebust out_err: 11381da177e4SLinus Torvalds while (!list_empty(head)) { 11391da177e4SLinus Torvalds req = nfs_list_entry(head->next); 11401da177e4SLinus Torvalds nfs_list_remove_request(req); 11411da177e4SLinus Torvalds nfs_mark_request_dirty(req); 1142c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 11431da177e4SLinus Torvalds } 11441da177e4SLinus Torvalds return error; 11451da177e4SLinus Torvalds } 11461da177e4SLinus Torvalds 11471da177e4SLinus Torvalds /* 11481da177e4SLinus Torvalds * Handle a write reply that flushed part of a page. 11491da177e4SLinus Torvalds */ 1150788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) 11511da177e4SLinus Torvalds { 1152788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 11531da177e4SLinus Torvalds struct nfs_page *req = data->req; 11541da177e4SLinus Torvalds struct page *page = req->wb_page; 11551da177e4SLinus Torvalds 11561da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 11571da177e4SLinus Torvalds req->wb_context->dentry->d_inode->i_sb->s_id, 11581da177e4SLinus Torvalds (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 11591da177e4SLinus Torvalds req->wb_bytes, 11601da177e4SLinus Torvalds (long long)req_offset(req)); 11611da177e4SLinus Torvalds 1162788e7a89STrond Myklebust if (nfs_writeback_done(task, data) != 0) 1163788e7a89STrond Myklebust return; 1164788e7a89STrond Myklebust 1165788e7a89STrond Myklebust if (task->tk_status < 0) { 11661da177e4SLinus Torvalds ClearPageUptodate(page); 11671da177e4SLinus Torvalds SetPageError(page); 1168788e7a89STrond Myklebust req->wb_context->error = task->tk_status; 1169788e7a89STrond Myklebust dprintk(", error = %d\n", task->tk_status); 11701da177e4SLinus Torvalds } else { 11711da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 11721da177e4SLinus Torvalds if (data->verf.committed < NFS_FILE_SYNC) { 11731da177e4SLinus Torvalds if (!NFS_NEED_COMMIT(req)) { 11741da177e4SLinus Torvalds nfs_defer_commit(req); 11751da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 11761da177e4SLinus Torvalds dprintk(" defer commit\n"); 11771da177e4SLinus Torvalds } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { 11781da177e4SLinus Torvalds nfs_defer_reschedule(req); 11791da177e4SLinus Torvalds dprintk(" server reboot detected\n"); 11801da177e4SLinus Torvalds } 11811da177e4SLinus Torvalds } else 11821da177e4SLinus Torvalds #endif 11831da177e4SLinus Torvalds dprintk(" OK\n"); 11841da177e4SLinus Torvalds } 11851da177e4SLinus Torvalds 11861da177e4SLinus Torvalds if (atomic_dec_and_test(&req->wb_complete)) 11871da177e4SLinus Torvalds nfs_writepage_release(req); 11881da177e4SLinus Torvalds } 11891da177e4SLinus Torvalds 1190788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops = { 1191788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_partial, 1192788e7a89STrond Myklebust .rpc_release = nfs_writedata_release, 1193788e7a89STrond Myklebust }; 1194788e7a89STrond Myklebust 11951da177e4SLinus Torvalds /* 11961da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 11971da177e4SLinus Torvalds * 11981da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 11991da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 12001da177e4SLinus Torvalds * as the page has a write request pending. 12011da177e4SLinus Torvalds */ 1202788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) 12031da177e4SLinus Torvalds { 1204788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 12051da177e4SLinus Torvalds struct nfs_page *req; 12061da177e4SLinus Torvalds struct page *page; 12071da177e4SLinus Torvalds 1208788e7a89STrond Myklebust if (nfs_writeback_done(task, data) != 0) 1209788e7a89STrond Myklebust return; 1210788e7a89STrond Myklebust 12111da177e4SLinus Torvalds /* Update attributes as result of writeback. */ 12121da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 12131da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 12141da177e4SLinus Torvalds nfs_list_remove_request(req); 12151da177e4SLinus Torvalds page = req->wb_page; 12161da177e4SLinus Torvalds 12171da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 12181da177e4SLinus Torvalds req->wb_context->dentry->d_inode->i_sb->s_id, 12191da177e4SLinus Torvalds (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 12201da177e4SLinus Torvalds req->wb_bytes, 12211da177e4SLinus Torvalds (long long)req_offset(req)); 12221da177e4SLinus Torvalds 1223788e7a89STrond Myklebust if (task->tk_status < 0) { 12241da177e4SLinus Torvalds ClearPageUptodate(page); 12251da177e4SLinus Torvalds SetPageError(page); 1226788e7a89STrond Myklebust req->wb_context->error = task->tk_status; 12271da177e4SLinus Torvalds end_page_writeback(page); 12281da177e4SLinus Torvalds nfs_inode_remove_request(req); 1229788e7a89STrond Myklebust dprintk(", error = %d\n", task->tk_status); 12301da177e4SLinus Torvalds goto next; 12311da177e4SLinus Torvalds } 12321da177e4SLinus Torvalds end_page_writeback(page); 12331da177e4SLinus Torvalds 12341da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 12351da177e4SLinus Torvalds if (data->args.stable != NFS_UNSTABLE || data->verf.committed == NFS_FILE_SYNC) { 12361da177e4SLinus Torvalds nfs_inode_remove_request(req); 12371da177e4SLinus Torvalds dprintk(" OK\n"); 12381da177e4SLinus Torvalds goto next; 12391da177e4SLinus Torvalds } 12401da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 12411da177e4SLinus Torvalds nfs_mark_request_commit(req); 12421da177e4SLinus Torvalds dprintk(" marked for commit\n"); 12431da177e4SLinus Torvalds #else 12441da177e4SLinus Torvalds nfs_inode_remove_request(req); 12451da177e4SLinus Torvalds #endif 12461da177e4SLinus Torvalds next: 1247c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 12481da177e4SLinus Torvalds } 12491da177e4SLinus Torvalds } 12501da177e4SLinus Torvalds 1251788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = { 1252788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_full, 1253788e7a89STrond Myklebust .rpc_release = nfs_writedata_release, 1254788e7a89STrond Myklebust }; 1255788e7a89STrond Myklebust 1256788e7a89STrond Myklebust 12571da177e4SLinus Torvalds /* 12581da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 12591da177e4SLinus Torvalds */ 1260462d5b32SChuck Lever int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 12611da177e4SLinus Torvalds { 12621da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 12631da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1264788e7a89STrond Myklebust int status; 12651da177e4SLinus Torvalds 12661da177e4SLinus Torvalds dprintk("NFS: %4d nfs_writeback_done (status %d)\n", 12671da177e4SLinus Torvalds task->tk_pid, task->tk_status); 12681da177e4SLinus Torvalds 1269f551e44fSChuck Lever /* 1270f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1271f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1272f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1273f551e44fSChuck Lever * another writer had changed the file, but some applications 1274f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1275f551e44fSChuck Lever */ 1276788e7a89STrond Myklebust status = NFS_PROTO(data->inode)->write_done(task, data); 1277788e7a89STrond Myklebust if (status != 0) 1278788e7a89STrond Myklebust return status; 127991d5b470SChuck Lever nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 128091d5b470SChuck Lever 12811da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 12821da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 12831da177e4SLinus Torvalds /* We tried a write call, but the server did not 12841da177e4SLinus Torvalds * commit data to stable storage even though we 12851da177e4SLinus Torvalds * requested it. 12861da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 12871da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 12881da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 12891da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 12901da177e4SLinus Torvalds */ 12911da177e4SLinus Torvalds static unsigned long complain; 12921da177e4SLinus Torvalds 12931da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 12941da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 12951da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 129654ceac45SDavid Howells NFS_SERVER(data->inode)->nfs_client->cl_hostname, 12971da177e4SLinus Torvalds resp->verf->committed, argp->stable); 12981da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 12991da177e4SLinus Torvalds } 13001da177e4SLinus Torvalds } 13011da177e4SLinus Torvalds #endif 13021da177e4SLinus Torvalds /* Is this a short write? */ 13031da177e4SLinus Torvalds if (task->tk_status >= 0 && resp->count < argp->count) { 13041da177e4SLinus Torvalds static unsigned long complain; 13051da177e4SLinus Torvalds 130691d5b470SChuck Lever nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); 130791d5b470SChuck Lever 13081da177e4SLinus Torvalds /* Has the server at least made some progress? */ 13091da177e4SLinus Torvalds if (resp->count != 0) { 13101da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 13111da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 13121da177e4SLinus Torvalds /* Resend from where the server left off */ 13131da177e4SLinus Torvalds argp->offset += resp->count; 13141da177e4SLinus Torvalds argp->pgbase += resp->count; 13151da177e4SLinus Torvalds argp->count -= resp->count; 13161da177e4SLinus Torvalds } else { 13171da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 13181da177e4SLinus Torvalds * headaches in the case of a server crash. 13191da177e4SLinus Torvalds */ 13201da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 13211da177e4SLinus Torvalds } 13221da177e4SLinus Torvalds rpc_restart_call(task); 1323788e7a89STrond Myklebust return -EAGAIN; 13241da177e4SLinus Torvalds } 13251da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 13261da177e4SLinus Torvalds printk(KERN_WARNING 13271da177e4SLinus Torvalds "NFS: Server wrote zero bytes, expected %u.\n", 13281da177e4SLinus Torvalds argp->count); 13291da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 13301da177e4SLinus Torvalds } 13311da177e4SLinus Torvalds /* Can't do anything about it except throw an error. */ 13321da177e4SLinus Torvalds task->tk_status = -EIO; 13331da177e4SLinus Torvalds } 1334788e7a89STrond Myklebust return 0; 13351da177e4SLinus Torvalds } 13361da177e4SLinus Torvalds 13371da177e4SLinus Torvalds 13381da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 1339963d8fe5STrond Myklebust void nfs_commit_release(void *wdata) 13401da177e4SLinus Torvalds { 13411da177e4SLinus Torvalds nfs_commit_free(wdata); 13421da177e4SLinus Torvalds } 13431da177e4SLinus Torvalds 13441da177e4SLinus Torvalds /* 13451da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 13461da177e4SLinus Torvalds */ 13471da177e4SLinus Torvalds static void nfs_commit_rpcsetup(struct list_head *head, 1348788e7a89STrond Myklebust struct nfs_write_data *data, 1349788e7a89STrond Myklebust int how) 13501da177e4SLinus Torvalds { 13513da28eb1STrond Myklebust struct nfs_page *first; 13521da177e4SLinus Torvalds struct inode *inode; 1353788e7a89STrond Myklebust int flags; 13541da177e4SLinus Torvalds 13551da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 13561da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 13571da177e4SLinus Torvalds 13581da177e4SLinus Torvalds list_splice_init(head, &data->pages); 13591da177e4SLinus Torvalds first = nfs_list_entry(data->pages.next); 13601da177e4SLinus Torvalds inode = first->wb_context->dentry->d_inode; 13611da177e4SLinus Torvalds 13621da177e4SLinus Torvalds data->inode = inode; 13631da177e4SLinus Torvalds data->cred = first->wb_context->cred; 13641da177e4SLinus Torvalds 13651da177e4SLinus Torvalds data->args.fh = NFS_FH(data->inode); 13663da28eb1STrond Myklebust /* Note: we always request a commit of the entire inode */ 13673da28eb1STrond Myklebust data->args.offset = 0; 13683da28eb1STrond Myklebust data->args.count = 0; 13693da28eb1STrond Myklebust data->res.count = 0; 13701da177e4SLinus Torvalds data->res.fattr = &data->fattr; 13711da177e4SLinus Torvalds data->res.verf = &data->verf; 13720e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 13731da177e4SLinus Torvalds 1374788e7a89STrond Myklebust /* Set up the initial task struct. */ 1375788e7a89STrond Myklebust flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 1376788e7a89STrond Myklebust rpc_init_task(&data->task, NFS_CLIENT(inode), flags, &nfs_commit_ops, data); 13771da177e4SLinus Torvalds NFS_PROTO(inode)->commit_setup(data, how); 13781da177e4SLinus Torvalds 13791da177e4SLinus Torvalds data->task.tk_priority = flush_task_priority(how); 13801da177e4SLinus Torvalds data->task.tk_cookie = (unsigned long)inode; 13811da177e4SLinus Torvalds 13820bbacc40SChuck Lever dprintk("NFS: %4d initiated commit call\n", data->task.tk_pid); 13831da177e4SLinus Torvalds } 13841da177e4SLinus Torvalds 13851da177e4SLinus Torvalds /* 13861da177e4SLinus Torvalds * Commit dirty pages 13871da177e4SLinus Torvalds */ 13881da177e4SLinus Torvalds static int 138940859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how) 13901da177e4SLinus Torvalds { 13911da177e4SLinus Torvalds struct nfs_write_data *data; 13921da177e4SLinus Torvalds struct nfs_page *req; 13931da177e4SLinus Torvalds 1394e9f7bee1STrond Myklebust data = nfs_commit_alloc(); 13951da177e4SLinus Torvalds 13961da177e4SLinus Torvalds if (!data) 13971da177e4SLinus Torvalds goto out_bad; 13981da177e4SLinus Torvalds 13991da177e4SLinus Torvalds /* Set up the argument struct */ 14001da177e4SLinus Torvalds nfs_commit_rpcsetup(head, data, how); 14011da177e4SLinus Torvalds 14021da177e4SLinus Torvalds nfs_execute_write(data); 14031da177e4SLinus Torvalds return 0; 14041da177e4SLinus Torvalds out_bad: 14051da177e4SLinus Torvalds while (!list_empty(head)) { 14061da177e4SLinus Torvalds req = nfs_list_entry(head->next); 14071da177e4SLinus Torvalds nfs_list_remove_request(req); 14081da177e4SLinus Torvalds nfs_mark_request_commit(req); 140983715ad5STrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 14105c2d97cbSTrond Myklebust nfs_clear_page_writeback(req); 14111da177e4SLinus Torvalds } 14121da177e4SLinus Torvalds return -ENOMEM; 14131da177e4SLinus Torvalds } 14141da177e4SLinus Torvalds 14151da177e4SLinus Torvalds /* 14161da177e4SLinus Torvalds * COMMIT call returned 14171da177e4SLinus Torvalds */ 1418788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 14191da177e4SLinus Torvalds { 1420963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 14211da177e4SLinus Torvalds struct nfs_page *req; 14221da177e4SLinus Torvalds 14231da177e4SLinus Torvalds dprintk("NFS: %4d nfs_commit_done (status %d)\n", 14241da177e4SLinus Torvalds task->tk_pid, task->tk_status); 14251da177e4SLinus Torvalds 1426788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1427788e7a89STrond Myklebust if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) 1428788e7a89STrond Myklebust return; 1429788e7a89STrond Myklebust 14301da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 14311da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 14321da177e4SLinus Torvalds nfs_list_remove_request(req); 1433fd39fc85SChristoph Lameter dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 14341da177e4SLinus Torvalds 14351da177e4SLinus Torvalds dprintk("NFS: commit (%s/%Ld %d@%Ld)", 14361da177e4SLinus Torvalds req->wb_context->dentry->d_inode->i_sb->s_id, 14371da177e4SLinus Torvalds (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 14381da177e4SLinus Torvalds req->wb_bytes, 14391da177e4SLinus Torvalds (long long)req_offset(req)); 14401da177e4SLinus Torvalds if (task->tk_status < 0) { 14411da177e4SLinus Torvalds req->wb_context->error = task->tk_status; 14421da177e4SLinus Torvalds nfs_inode_remove_request(req); 14431da177e4SLinus Torvalds dprintk(", error = %d\n", task->tk_status); 14441da177e4SLinus Torvalds goto next; 14451da177e4SLinus Torvalds } 14461da177e4SLinus Torvalds 14471da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 14481da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 14491da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 14501da177e4SLinus Torvalds /* We have a match */ 14511da177e4SLinus Torvalds nfs_inode_remove_request(req); 14521da177e4SLinus Torvalds dprintk(" OK\n"); 14531da177e4SLinus Torvalds goto next; 14541da177e4SLinus Torvalds } 14551da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 14561da177e4SLinus Torvalds dprintk(" mismatch\n"); 14571da177e4SLinus Torvalds nfs_mark_request_dirty(req); 14581da177e4SLinus Torvalds next: 1459c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 14601da177e4SLinus Torvalds } 14611da177e4SLinus Torvalds } 1462788e7a89STrond Myklebust 1463788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 1464788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1465788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1466788e7a89STrond Myklebust }; 1467c42de9ddSTrond Myklebust #else 1468c42de9ddSTrond Myklebust static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how) 1469c42de9ddSTrond Myklebust { 1470c42de9ddSTrond Myklebust return 0; 1471c42de9ddSTrond Myklebust } 14721da177e4SLinus Torvalds #endif 14731da177e4SLinus Torvalds 14741da177e4SLinus Torvalds static int nfs_flush_inode(struct inode *inode, unsigned long idx_start, 14751da177e4SLinus Torvalds unsigned int npages, int how) 14761da177e4SLinus Torvalds { 14771da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 14781da177e4SLinus Torvalds LIST_HEAD(head); 14797d46a49fSTrond Myklebust int res; 14801da177e4SLinus Torvalds 14811da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 14821da177e4SLinus Torvalds res = nfs_scan_dirty(inode, &head, idx_start, npages); 14831da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 1484ab0a3dbeSTrond Myklebust if (res) { 14857d46a49fSTrond Myklebust int error = nfs_flush_list(inode, &head, res, how); 14861da177e4SLinus Torvalds if (error < 0) 14871da177e4SLinus Torvalds return error; 14887d46a49fSTrond Myklebust } 14891da177e4SLinus Torvalds return res; 14901da177e4SLinus Torvalds } 14911da177e4SLinus Torvalds 14921da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 14933da28eb1STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 14941da177e4SLinus Torvalds { 14951da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 14961da177e4SLinus Torvalds LIST_HEAD(head); 14977d46a49fSTrond Myklebust int res; 14981da177e4SLinus Torvalds 14991da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 15003da28eb1STrond Myklebust res = nfs_scan_commit(inode, &head, 0, 0); 15013da28eb1STrond Myklebust spin_unlock(&nfsi->req_lock); 15021da177e4SLinus Torvalds if (res) { 15037d46a49fSTrond Myklebust int error = nfs_commit_list(inode, &head, how); 15041da177e4SLinus Torvalds if (error < 0) 15051da177e4SLinus Torvalds return error; 15063da28eb1STrond Myklebust } 15071da177e4SLinus Torvalds return res; 15081da177e4SLinus Torvalds } 15091da177e4SLinus Torvalds #endif 15101da177e4SLinus Torvalds 1511c42de9ddSTrond Myklebust int nfs_sync_inode_wait(struct inode *inode, unsigned long idx_start, 15121da177e4SLinus Torvalds unsigned int npages, int how) 15131da177e4SLinus Torvalds { 1514c42de9ddSTrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1515c42de9ddSTrond Myklebust LIST_HEAD(head); 151670b9ecbdSTrond Myklebust int nocommit = how & FLUSH_NOCOMMIT; 1517c42de9ddSTrond Myklebust int pages, ret; 15181da177e4SLinus Torvalds 1519c42de9ddSTrond Myklebust how &= ~FLUSH_NOCOMMIT; 1520c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 15211da177e4SLinus Torvalds do { 1522c42de9ddSTrond Myklebust ret = nfs_wait_on_requests_locked(inode, idx_start, npages); 1523c42de9ddSTrond Myklebust if (ret != 0) 1524c42de9ddSTrond Myklebust continue; 1525c42de9ddSTrond Myklebust pages = nfs_scan_dirty(inode, &head, idx_start, npages); 1526c42de9ddSTrond Myklebust if (pages != 0) { 1527c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 1528d2ccddf0STrond Myklebust if (how & FLUSH_INVALIDATE) 152983715ad5STrond Myklebust nfs_cancel_dirty_list(&head); 1530d2ccddf0STrond Myklebust else 1531c42de9ddSTrond Myklebust ret = nfs_flush_list(inode, &head, pages, how); 1532c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 153370b9ecbdSTrond Myklebust continue; 153470b9ecbdSTrond Myklebust } 1535c42de9ddSTrond Myklebust if (nocommit) 1536c42de9ddSTrond Myklebust break; 1537d2ccddf0STrond Myklebust pages = nfs_scan_commit(inode, &head, idx_start, npages); 1538c42de9ddSTrond Myklebust if (pages == 0) 1539c42de9ddSTrond Myklebust break; 1540d2ccddf0STrond Myklebust if (how & FLUSH_INVALIDATE) { 1541d2ccddf0STrond Myklebust spin_unlock(&nfsi->req_lock); 154283715ad5STrond Myklebust nfs_cancel_commit_list(&head); 1543d2ccddf0STrond Myklebust spin_lock(&nfsi->req_lock); 1544d2ccddf0STrond Myklebust continue; 1545d2ccddf0STrond Myklebust } 1546d2ccddf0STrond Myklebust pages += nfs_scan_commit(inode, &head, 0, 0); 1547c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 1548c42de9ddSTrond Myklebust ret = nfs_commit_list(inode, &head, how); 1549c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 1550c42de9ddSTrond Myklebust } while (ret >= 0); 1551c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 1552c42de9ddSTrond Myklebust return ret; 15531da177e4SLinus Torvalds } 15541da177e4SLinus Torvalds 1555f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 15561da177e4SLinus Torvalds { 15571da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 15581da177e4SLinus Torvalds sizeof(struct nfs_write_data), 15591da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 15601da177e4SLinus Torvalds NULL, NULL); 15611da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 15621da177e4SLinus Torvalds return -ENOMEM; 15631da177e4SLinus Torvalds 156493d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 15651da177e4SLinus Torvalds nfs_wdata_cachep); 15661da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 15671da177e4SLinus Torvalds return -ENOMEM; 15681da177e4SLinus Torvalds 156993d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 15701da177e4SLinus Torvalds nfs_wdata_cachep); 15711da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 15721da177e4SLinus Torvalds return -ENOMEM; 15731da177e4SLinus Torvalds 15741da177e4SLinus Torvalds return 0; 15751da177e4SLinus Torvalds } 15761da177e4SLinus Torvalds 1577266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 15781da177e4SLinus Torvalds { 15791da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 15801da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 15811a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 15821da177e4SLinus Torvalds } 15831da177e4SLinus Torvalds 1584