11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/nfs/write.c 31da177e4SLinus Torvalds * 47c85d900STrond Myklebust * Write file data over NFS. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> 71da177e4SLinus Torvalds */ 81da177e4SLinus Torvalds 91da177e4SLinus Torvalds #include <linux/types.h> 101da177e4SLinus Torvalds #include <linux/slab.h> 111da177e4SLinus Torvalds #include <linux/mm.h> 121da177e4SLinus Torvalds #include <linux/pagemap.h> 131da177e4SLinus Torvalds #include <linux/file.h> 141da177e4SLinus Torvalds #include <linux/writeback.h> 1589a09141SPeter Zijlstra #include <linux/swap.h> 161da177e4SLinus Torvalds 171da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h> 181da177e4SLinus Torvalds #include <linux/nfs_fs.h> 191da177e4SLinus Torvalds #include <linux/nfs_mount.h> 201da177e4SLinus Torvalds #include <linux/nfs_page.h> 213fcfab16SAndrew Morton #include <linux/backing-dev.h> 223fcfab16SAndrew Morton 231da177e4SLinus Torvalds #include <asm/uaccess.h> 241da177e4SLinus Torvalds 251da177e4SLinus Torvalds #include "delegation.h" 2649a70f27STrond Myklebust #include "internal.h" 2791d5b470SChuck Lever #include "iostat.h" 281da177e4SLinus Torvalds 291da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 301da177e4SLinus Torvalds 311da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 321da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 331da177e4SLinus Torvalds 341da177e4SLinus Torvalds /* 351da177e4SLinus Torvalds * Local function declarations 361da177e4SLinus Torvalds */ 37c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc, 38c63c7b05STrond Myklebust struct inode *inode, int ioflags); 39f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 40788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops; 41788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops; 42788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 431da177e4SLinus Torvalds 44e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 453feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 461da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 471da177e4SLinus Torvalds 48c9d8f89dSTrond Myklebust struct nfs_write_data *nfs_commitdata_alloc(void) 491da177e4SLinus Torvalds { 50e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS); 5140859d7eSChuck Lever 521da177e4SLinus Torvalds if (p) { 531da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 541da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 551da177e4SLinus Torvalds } 561da177e4SLinus Torvalds return p; 571da177e4SLinus Torvalds } 581da177e4SLinus Torvalds 595e4424afSTrond Myklebust void nfs_commit_free(struct nfs_write_data *p) 601da177e4SLinus Torvalds { 6140859d7eSChuck Lever if (p && (p->pagevec != &p->page_array[0])) 6240859d7eSChuck Lever kfree(p->pagevec); 631da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 641da177e4SLinus Torvalds } 651da177e4SLinus Torvalds 668d5658c9STrond Myklebust struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount) 673feb2d49STrond Myklebust { 68e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); 693feb2d49STrond Myklebust 703feb2d49STrond Myklebust if (p) { 713feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 723feb2d49STrond Myklebust INIT_LIST_HEAD(&p->pages); 73e9f7bee1STrond Myklebust p->npages = pagecount; 740d0b5cb3SChuck Lever if (pagecount <= ARRAY_SIZE(p->page_array)) 750d0b5cb3SChuck Lever p->pagevec = p->page_array; 763feb2d49STrond Myklebust else { 770d0b5cb3SChuck Lever p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); 780d0b5cb3SChuck Lever if (!p->pagevec) { 793feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 803feb2d49STrond Myklebust p = NULL; 813feb2d49STrond Myklebust } 823feb2d49STrond Myklebust } 833feb2d49STrond Myklebust } 843feb2d49STrond Myklebust return p; 853feb2d49STrond Myklebust } 863feb2d49STrond Myklebust 875e4424afSTrond Myklebust static void nfs_writedata_free(struct nfs_write_data *p) 883feb2d49STrond Myklebust { 893feb2d49STrond Myklebust if (p && (p->pagevec != &p->page_array[0])) 903feb2d49STrond Myklebust kfree(p->pagevec); 913feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 923feb2d49STrond Myklebust } 933feb2d49STrond Myklebust 94383ba719STrond Myklebust void nfs_writedata_release(void *data) 951da177e4SLinus Torvalds { 96383ba719STrond Myklebust struct nfs_write_data *wdata = data; 97383ba719STrond Myklebust 98383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 991da177e4SLinus Torvalds nfs_writedata_free(wdata); 1001da177e4SLinus Torvalds } 1011da177e4SLinus Torvalds 1027b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 1037b159fc1STrond Myklebust { 1047b159fc1STrond Myklebust ctx->error = error; 1057b159fc1STrond Myklebust smp_wmb(); 1067b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 1077b159fc1STrond Myklebust } 1087b159fc1STrond Myklebust 109277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request_locked(struct page *page) 110277459d2STrond Myklebust { 111277459d2STrond Myklebust struct nfs_page *req = NULL; 112277459d2STrond Myklebust 113277459d2STrond Myklebust if (PagePrivate(page)) { 114277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 115277459d2STrond Myklebust if (req != NULL) 116c03b4024STrond Myklebust kref_get(&req->wb_kref); 117277459d2STrond Myklebust } 118277459d2STrond Myklebust return req; 119277459d2STrond Myklebust } 120277459d2STrond Myklebust 121277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request(struct page *page) 122277459d2STrond Myklebust { 123587142f8STrond Myklebust struct inode *inode = page->mapping->host; 124277459d2STrond Myklebust struct nfs_page *req = NULL; 125277459d2STrond Myklebust 126587142f8STrond Myklebust spin_lock(&inode->i_lock); 127277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 128587142f8STrond Myklebust spin_unlock(&inode->i_lock); 129277459d2STrond Myklebust return req; 130277459d2STrond Myklebust } 131277459d2STrond Myklebust 1321da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1331da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1341da177e4SLinus Torvalds { 1351da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 1361da177e4SLinus Torvalds loff_t end, i_size = i_size_read(inode); 137ca52fec1STrond Myklebust pgoff_t end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 1381da177e4SLinus Torvalds 1391da177e4SLinus Torvalds if (i_size > 0 && page->index < end_index) 1401da177e4SLinus Torvalds return; 1411da177e4SLinus Torvalds end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); 1421da177e4SLinus Torvalds if (i_size >= end) 1431da177e4SLinus Torvalds return; 14491d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 1451da177e4SLinus Torvalds i_size_write(inode, end); 1461da177e4SLinus Torvalds } 1471da177e4SLinus Torvalds 148a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 149a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 150a301b777STrond Myklebust { 151a301b777STrond Myklebust SetPageError(page); 152a301b777STrond Myklebust nfs_zap_mapping(page->mapping->host, page->mapping); 153a301b777STrond Myklebust } 154a301b777STrond Myklebust 1551da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 1561da177e4SLinus Torvalds * covers the full page. 1571da177e4SLinus Torvalds */ 1581da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 1591da177e4SLinus Torvalds { 1601da177e4SLinus Torvalds if (PageUptodate(page)) 1611da177e4SLinus Torvalds return; 1621da177e4SLinus Torvalds if (base != 0) 1631da177e4SLinus Torvalds return; 16449a70f27STrond Myklebust if (count != nfs_page_length(page)) 1651da177e4SLinus Torvalds return; 1661da177e4SLinus Torvalds SetPageUptodate(page); 1671da177e4SLinus Torvalds } 1681da177e4SLinus Torvalds 1691da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 1701da177e4SLinus Torvalds { 1711da177e4SLinus Torvalds if (wbc->for_reclaim) 172c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 1731da177e4SLinus Torvalds if (wbc->for_kupdate) 1741da177e4SLinus Torvalds return FLUSH_LOWPRI; 1751da177e4SLinus Torvalds return 0; 1761da177e4SLinus Torvalds } 1771da177e4SLinus Torvalds 1781da177e4SLinus Torvalds /* 17989a09141SPeter Zijlstra * NFS congestion control 18089a09141SPeter Zijlstra */ 18189a09141SPeter Zijlstra 18289a09141SPeter Zijlstra int nfs_congestion_kb; 18389a09141SPeter Zijlstra 18489a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 18589a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 18689a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 18789a09141SPeter Zijlstra 1885a6d41b3STrond Myklebust static int nfs_set_page_writeback(struct page *page) 18989a09141SPeter Zijlstra { 1905a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 1915a6d41b3STrond Myklebust 1925a6d41b3STrond Myklebust if (!ret) { 19389a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 19489a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 19589a09141SPeter Zijlstra 196277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 19789a09141SPeter Zijlstra NFS_CONGESTION_ON_THRESH) 19889a09141SPeter Zijlstra set_bdi_congested(&nfss->backing_dev_info, WRITE); 19989a09141SPeter Zijlstra } 2005a6d41b3STrond Myklebust return ret; 20189a09141SPeter Zijlstra } 20289a09141SPeter Zijlstra 20389a09141SPeter Zijlstra static void nfs_end_page_writeback(struct page *page) 20489a09141SPeter Zijlstra { 20589a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 20689a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 20789a09141SPeter Zijlstra 20889a09141SPeter Zijlstra end_page_writeback(page); 209c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 21089a09141SPeter Zijlstra clear_bdi_congested(&nfss->backing_dev_info, WRITE); 21189a09141SPeter Zijlstra } 21289a09141SPeter Zijlstra 21389a09141SPeter Zijlstra /* 214e261f51fSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 2159cccef95STrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 216e261f51fSTrond Myklebust */ 217c63c7b05STrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 218c63c7b05STrond Myklebust struct page *page) 219e261f51fSTrond Myklebust { 220587142f8STrond Myklebust struct inode *inode = page->mapping->host; 221e261f51fSTrond Myklebust struct nfs_page *req; 222e261f51fSTrond Myklebust int ret; 223e261f51fSTrond Myklebust 224587142f8STrond Myklebust spin_lock(&inode->i_lock); 225e261f51fSTrond Myklebust for(;;) { 226e261f51fSTrond Myklebust req = nfs_page_find_request_locked(page); 227e261f51fSTrond Myklebust if (req == NULL) { 228587142f8STrond Myklebust spin_unlock(&inode->i_lock); 2299cccef95STrond Myklebust return 0; 230e261f51fSTrond Myklebust } 231acee478aSTrond Myklebust if (nfs_set_page_tag_locked(req)) 232e261f51fSTrond Myklebust break; 233e261f51fSTrond Myklebust /* Note: If we hold the page lock, as is the case in nfs_writepage, 234acee478aSTrond Myklebust * then the call to nfs_set_page_tag_locked() will always 235e261f51fSTrond Myklebust * succeed provided that someone hasn't already marked the 236e261f51fSTrond Myklebust * request as dirty (in which case we don't care). 237e261f51fSTrond Myklebust */ 238587142f8STrond Myklebust spin_unlock(&inode->i_lock); 239e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 240e261f51fSTrond Myklebust nfs_release_request(req); 241e261f51fSTrond Myklebust if (ret != 0) 242e261f51fSTrond Myklebust return ret; 243587142f8STrond Myklebust spin_lock(&inode->i_lock); 244e261f51fSTrond Myklebust } 245612c9384STrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 246612c9384STrond Myklebust /* This request is marked for commit */ 247587142f8STrond Myklebust spin_unlock(&inode->i_lock); 248acee478aSTrond Myklebust nfs_clear_page_tag_locked(req); 249c63c7b05STrond Myklebust nfs_pageio_complete(pgio); 2509cccef95STrond Myklebust return 0; 251612c9384STrond Myklebust } 252c63c7b05STrond Myklebust if (nfs_set_page_writeback(page) != 0) { 253587142f8STrond Myklebust spin_unlock(&inode->i_lock); 254c63c7b05STrond Myklebust BUG(); 255c63c7b05STrond Myklebust } 256587142f8STrond Myklebust spin_unlock(&inode->i_lock); 257f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 258f8512ad0SFred Isaman nfs_redirty_request(req); 259f8512ad0SFred Isaman return pgio->pg_error; 260f8512ad0SFred Isaman } 2619cccef95STrond Myklebust return 0; 262e261f51fSTrond Myklebust } 263e261f51fSTrond Myklebust 264f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 265f758c885STrond Myklebust { 266f758c885STrond Myklebust struct inode *inode = page->mapping->host; 267f758c885STrond Myklebust 268f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 269f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 270f758c885STrond Myklebust 271f758c885STrond Myklebust nfs_pageio_cond_complete(pgio, page->index); 272f758c885STrond Myklebust return nfs_page_async_flush(pgio, page); 273f758c885STrond Myklebust } 274f758c885STrond Myklebust 275e261f51fSTrond Myklebust /* 2761da177e4SLinus Torvalds * Write an mmapped page to the server. 2771da177e4SLinus Torvalds */ 2784d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 2791da177e4SLinus Torvalds { 280f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 281e261f51fSTrond Myklebust int err; 2821da177e4SLinus Torvalds 283f758c885STrond Myklebust nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc)); 284f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 285f758c885STrond Myklebust nfs_pageio_complete(&pgio); 286f758c885STrond Myklebust if (err < 0) 2874d770ccfSTrond Myklebust return err; 288f758c885STrond Myklebust if (pgio.pg_error < 0) 289f758c885STrond Myklebust return pgio.pg_error; 290f758c885STrond Myklebust return 0; 2914d770ccfSTrond Myklebust } 2924d770ccfSTrond Myklebust 2934d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 2944d770ccfSTrond Myklebust { 295f758c885STrond Myklebust int ret; 2964d770ccfSTrond Myklebust 297f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 2981da177e4SLinus Torvalds unlock_page(page); 299f758c885STrond Myklebust return ret; 300f758c885STrond Myklebust } 301f758c885STrond Myklebust 302f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 303f758c885STrond Myklebust { 304f758c885STrond Myklebust int ret; 305f758c885STrond Myklebust 306f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 307f758c885STrond Myklebust unlock_page(page); 308f758c885STrond Myklebust return ret; 3091da177e4SLinus Torvalds } 3101da177e4SLinus Torvalds 3111da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 3121da177e4SLinus Torvalds { 3131da177e4SLinus Torvalds struct inode *inode = mapping->host; 314c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 3151da177e4SLinus Torvalds int err; 3161da177e4SLinus Torvalds 31791d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 31891d5b470SChuck Lever 319c63c7b05STrond Myklebust nfs_pageio_init_write(&pgio, inode, wb_priority(wbc)); 320f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 321c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 322f758c885STrond Myklebust if (err < 0) 3231da177e4SLinus Torvalds return err; 324f758c885STrond Myklebust if (pgio.pg_error < 0) 325c63c7b05STrond Myklebust return pgio.pg_error; 326c63c7b05STrond Myklebust return 0; 3271da177e4SLinus Torvalds } 3281da177e4SLinus Torvalds 3291da177e4SLinus Torvalds /* 3301da177e4SLinus Torvalds * Insert a write request into an inode 3311da177e4SLinus Torvalds */ 332e7d39069STrond Myklebust static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 3331da177e4SLinus Torvalds { 3341da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 3351da177e4SLinus Torvalds int error; 3361da177e4SLinus Torvalds 337e7d39069STrond Myklebust error = radix_tree_preload(GFP_NOFS); 338e7d39069STrond Myklebust if (error != 0) 339e7d39069STrond Myklebust goto out; 340e7d39069STrond Myklebust 341e7d39069STrond Myklebust /* Lock the request! */ 342e7d39069STrond Myklebust nfs_lock_request_dontget(req); 343e7d39069STrond Myklebust 344e7d39069STrond Myklebust spin_lock(&inode->i_lock); 3451da177e4SLinus Torvalds error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); 34627852596SNick Piggin BUG_ON(error); 3471da177e4SLinus Torvalds if (!nfsi->npages) { 3481da177e4SLinus Torvalds igrab(inode); 3491da177e4SLinus Torvalds if (nfs_have_delegation(inode, FMODE_WRITE)) 3501da177e4SLinus Torvalds nfsi->change_attr++; 3511da177e4SLinus Torvalds } 352deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 353277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 3541da177e4SLinus Torvalds nfsi->npages++; 355c03b4024STrond Myklebust kref_get(&req->wb_kref); 35627852596SNick Piggin radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, 35727852596SNick Piggin NFS_PAGE_TAG_LOCKED); 358e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 359e7d39069STrond Myklebust radix_tree_preload_end(); 360e7d39069STrond Myklebust out: 361e7d39069STrond Myklebust return error; 3621da177e4SLinus Torvalds } 3631da177e4SLinus Torvalds 3641da177e4SLinus Torvalds /* 36589a09141SPeter Zijlstra * Remove a write request from an inode 3661da177e4SLinus Torvalds */ 3671da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 3681da177e4SLinus Torvalds { 36988be9f99STrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 3701da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 3711da177e4SLinus Torvalds 3721da177e4SLinus Torvalds BUG_ON (!NFS_WBACK_BUSY(req)); 3731da177e4SLinus Torvalds 374587142f8STrond Myklebust spin_lock(&inode->i_lock); 375277459d2STrond Myklebust set_page_private(req->wb_page, 0); 376deb7d638STrond Myklebust ClearPagePrivate(req->wb_page); 3771da177e4SLinus Torvalds radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); 3781da177e4SLinus Torvalds nfsi->npages--; 3791da177e4SLinus Torvalds if (!nfsi->npages) { 380587142f8STrond Myklebust spin_unlock(&inode->i_lock); 3811da177e4SLinus Torvalds iput(inode); 3821da177e4SLinus Torvalds } else 383587142f8STrond Myklebust spin_unlock(&inode->i_lock); 3841da177e4SLinus Torvalds nfs_clear_request(req); 3851da177e4SLinus Torvalds nfs_release_request(req); 3861da177e4SLinus Torvalds } 3871da177e4SLinus Torvalds 38861822ab5STrond Myklebust static void 3896d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 39061822ab5STrond Myklebust { 39161822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 39261822ab5STrond Myklebust } 39361822ab5STrond Myklebust 3941da177e4SLinus Torvalds /* 3951da177e4SLinus Torvalds * Check if a request is dirty 3961da177e4SLinus Torvalds */ 3971da177e4SLinus Torvalds static inline int 3981da177e4SLinus Torvalds nfs_dirty_request(struct nfs_page *req) 3991da177e4SLinus Torvalds { 4005a6d41b3STrond Myklebust struct page *page = req->wb_page; 4015a6d41b3STrond Myklebust 402612c9384STrond Myklebust if (page == NULL || test_bit(PG_NEED_COMMIT, &req->wb_flags)) 4035a6d41b3STrond Myklebust return 0; 40438def50fSFred Isaman return !PageWriteback(page); 4051da177e4SLinus Torvalds } 4061da177e4SLinus Torvalds 4071da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 4081da177e4SLinus Torvalds /* 4091da177e4SLinus Torvalds * Add a request to the inode's commit list. 4101da177e4SLinus Torvalds */ 4111da177e4SLinus Torvalds static void 4121da177e4SLinus Torvalds nfs_mark_request_commit(struct nfs_page *req) 4131da177e4SLinus Torvalds { 41488be9f99STrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 4151da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4161da177e4SLinus Torvalds 417587142f8STrond Myklebust spin_lock(&inode->i_lock); 4181da177e4SLinus Torvalds nfsi->ncommit++; 419612c9384STrond Myklebust set_bit(PG_NEED_COMMIT, &(req)->wb_flags); 4205c369683STrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, 4215c369683STrond Myklebust req->wb_index, 4225c369683STrond Myklebust NFS_PAGE_TAG_COMMIT); 423587142f8STrond Myklebust spin_unlock(&inode->i_lock); 424fd39fc85SChristoph Lameter inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 425c9e51e41SPeter Zijlstra inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); 426a1803044STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 4271da177e4SLinus Torvalds } 4288e821cadSTrond Myklebust 4298e821cadSTrond Myklebust static inline 4308e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4318e821cadSTrond Myklebust { 4328e821cadSTrond Myklebust return data->verf.committed != NFS_FILE_SYNC; 4338e821cadSTrond Myklebust } 4348e821cadSTrond Myklebust 4358e821cadSTrond Myklebust static inline 4368e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req) 4378e821cadSTrond Myklebust { 438612c9384STrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 4398e821cadSTrond Myklebust nfs_mark_request_commit(req); 4408e821cadSTrond Myklebust return 1; 4418e821cadSTrond Myklebust } 4428e821cadSTrond Myklebust if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) { 4436d884e8fSFred nfs_mark_request_dirty(req); 4448e821cadSTrond Myklebust return 1; 4458e821cadSTrond Myklebust } 4468e821cadSTrond Myklebust return 0; 4478e821cadSTrond Myklebust } 4488e821cadSTrond Myklebust #else 4498e821cadSTrond Myklebust static inline void 4508e821cadSTrond Myklebust nfs_mark_request_commit(struct nfs_page *req) 4518e821cadSTrond Myklebust { 4528e821cadSTrond Myklebust } 4538e821cadSTrond Myklebust 4548e821cadSTrond Myklebust static inline 4558e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4568e821cadSTrond Myklebust { 4578e821cadSTrond Myklebust return 0; 4588e821cadSTrond Myklebust } 4598e821cadSTrond Myklebust 4608e821cadSTrond Myklebust static inline 4618e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req) 4628e821cadSTrond Myklebust { 4638e821cadSTrond Myklebust return 0; 4648e821cadSTrond Myklebust } 4651da177e4SLinus Torvalds #endif 4661da177e4SLinus Torvalds 4671da177e4SLinus Torvalds /* 4681da177e4SLinus Torvalds * Wait for a request to complete. 4691da177e4SLinus Torvalds * 470150030b7SMatthew Wilcox * Interruptible by fatal signals only. 4711da177e4SLinus Torvalds */ 472ca52fec1STrond Myklebust static int nfs_wait_on_requests_locked(struct inode *inode, pgoff_t idx_start, unsigned int npages) 4731da177e4SLinus Torvalds { 4741da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4751da177e4SLinus Torvalds struct nfs_page *req; 476ca52fec1STrond Myklebust pgoff_t idx_end, next; 4771da177e4SLinus Torvalds unsigned int res = 0; 4781da177e4SLinus Torvalds int error; 4791da177e4SLinus Torvalds 4801da177e4SLinus Torvalds if (npages == 0) 4811da177e4SLinus Torvalds idx_end = ~0; 4821da177e4SLinus Torvalds else 4831da177e4SLinus Torvalds idx_end = idx_start + npages - 1; 4841da177e4SLinus Torvalds 4851da177e4SLinus Torvalds next = idx_start; 4869fd367f0STrond Myklebust while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_LOCKED)) { 4871da177e4SLinus Torvalds if (req->wb_index > idx_end) 4881da177e4SLinus Torvalds break; 4891da177e4SLinus Torvalds 4901da177e4SLinus Torvalds next = req->wb_index + 1; 491c6a556b8STrond Myklebust BUG_ON(!NFS_WBACK_BUSY(req)); 4921da177e4SLinus Torvalds 493c03b4024STrond Myklebust kref_get(&req->wb_kref); 494587142f8STrond Myklebust spin_unlock(&inode->i_lock); 4951da177e4SLinus Torvalds error = nfs_wait_on_request(req); 4961da177e4SLinus Torvalds nfs_release_request(req); 497587142f8STrond Myklebust spin_lock(&inode->i_lock); 4981da177e4SLinus Torvalds if (error < 0) 4991da177e4SLinus Torvalds return error; 5001da177e4SLinus Torvalds res++; 5011da177e4SLinus Torvalds } 5021da177e4SLinus Torvalds return res; 5031da177e4SLinus Torvalds } 5041da177e4SLinus Torvalds 50583715ad5STrond Myklebust static void nfs_cancel_commit_list(struct list_head *head) 50683715ad5STrond Myklebust { 50783715ad5STrond Myklebust struct nfs_page *req; 50883715ad5STrond Myklebust 50983715ad5STrond Myklebust while(!list_empty(head)) { 51083715ad5STrond Myklebust req = nfs_list_entry(head->next); 511b6dff26aSTrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 512c9e51e41SPeter Zijlstra dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 513c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 51483715ad5STrond Myklebust nfs_list_remove_request(req); 515612c9384STrond Myklebust clear_bit(PG_NEED_COMMIT, &(req)->wb_flags); 51683715ad5STrond Myklebust nfs_inode_remove_request(req); 517b6dff26aSTrond Myklebust nfs_unlock_request(req); 51883715ad5STrond Myklebust } 51983715ad5STrond Myklebust } 52083715ad5STrond Myklebust 5211da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 5221da177e4SLinus Torvalds /* 5231da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 5241da177e4SLinus Torvalds * @inode: NFS inode to scan 5251da177e4SLinus Torvalds * @dst: destination list 5261da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 5271da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 5281da177e4SLinus Torvalds * 5291da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 5301da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 5311da177e4SLinus Torvalds */ 5321da177e4SLinus Torvalds static int 533ca52fec1STrond Myklebust nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 5341da177e4SLinus Torvalds { 5351da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5363da28eb1STrond Myklebust int res = 0; 5373da28eb1STrond Myklebust 5383da28eb1STrond Myklebust if (nfsi->ncommit != 0) { 5395c369683STrond Myklebust res = nfs_scan_list(nfsi, dst, idx_start, npages, 5405c369683STrond Myklebust NFS_PAGE_TAG_COMMIT); 5411da177e4SLinus Torvalds nfsi->ncommit -= res; 5423da28eb1STrond Myklebust } 5431da177e4SLinus Torvalds return res; 5441da177e4SLinus Torvalds } 545c42de9ddSTrond Myklebust #else 546ca52fec1STrond Myklebust static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 547c42de9ddSTrond Myklebust { 548c42de9ddSTrond Myklebust return 0; 549c42de9ddSTrond Myklebust } 5501da177e4SLinus Torvalds #endif 5511da177e4SLinus Torvalds 5521da177e4SLinus Torvalds /* 553e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 554e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 5551da177e4SLinus Torvalds * 556e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 557e7d39069STrond Myklebust * to disk. 5581da177e4SLinus Torvalds */ 559e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 560e7d39069STrond Myklebust struct page *page, 561e7d39069STrond Myklebust unsigned int offset, 562e7d39069STrond Myklebust unsigned int bytes) 5631da177e4SLinus Torvalds { 564e7d39069STrond Myklebust struct nfs_page *req; 565e7d39069STrond Myklebust unsigned int rqend; 566e7d39069STrond Myklebust unsigned int end; 5671da177e4SLinus Torvalds int error; 568277459d2STrond Myklebust 569e7d39069STrond Myklebust if (!PagePrivate(page)) 570e7d39069STrond Myklebust return NULL; 571e7d39069STrond Myklebust 572e7d39069STrond Myklebust end = offset + bytes; 573e7d39069STrond Myklebust spin_lock(&inode->i_lock); 574e7d39069STrond Myklebust 575e7d39069STrond Myklebust for (;;) { 576e7d39069STrond Myklebust req = nfs_page_find_request_locked(page); 577e7d39069STrond Myklebust if (req == NULL) 578e7d39069STrond Myklebust goto out_unlock; 579e7d39069STrond Myklebust 580e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 581e7d39069STrond Myklebust /* 582e7d39069STrond Myklebust * Tell the caller to flush out the request if 583e7d39069STrond Myklebust * the offsets are non-contiguous. 584e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 585e7d39069STrond Myklebust * have flushed out requests having wrong owners. 586e7d39069STrond Myklebust */ 587e7d39069STrond Myklebust if (!nfs_dirty_request(req) 588e7d39069STrond Myklebust || offset > rqend 589e7d39069STrond Myklebust || end < req->wb_offset) 590e7d39069STrond Myklebust goto out_flushme; 591e7d39069STrond Myklebust 592e7d39069STrond Myklebust if (nfs_set_page_tag_locked(req)) 593e7d39069STrond Myklebust break; 594e7d39069STrond Myklebust 595e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 596587142f8STrond Myklebust spin_unlock(&inode->i_lock); 5971da177e4SLinus Torvalds error = nfs_wait_on_request(req); 5981da177e4SLinus Torvalds nfs_release_request(req); 599e7d39069STrond Myklebust if (error != 0) 600e7d39069STrond Myklebust goto out_err; 601e7d39069STrond Myklebust spin_lock(&inode->i_lock); 6021da177e4SLinus Torvalds } 6031da177e4SLinus Torvalds 6041da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 6051da177e4SLinus Torvalds if (offset < req->wb_offset) { 6061da177e4SLinus Torvalds req->wb_offset = offset; 6071da177e4SLinus Torvalds req->wb_pgbase = offset; 6081da177e4SLinus Torvalds } 6091da177e4SLinus Torvalds if (end > rqend) 6101da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 611e7d39069STrond Myklebust else 612e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 613e7d39069STrond Myklebust out_unlock: 614e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 615e7d39069STrond Myklebust return req; 616e7d39069STrond Myklebust out_flushme: 617e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 618e7d39069STrond Myklebust nfs_release_request(req); 619e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 620e7d39069STrond Myklebust out_err: 621e7d39069STrond Myklebust return ERR_PTR(error); 622e7d39069STrond Myklebust } 6231da177e4SLinus Torvalds 624e7d39069STrond Myklebust /* 625e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 626e7d39069STrond Myklebust * 627e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 628e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 629e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 630e7d39069STrond Myklebust */ 631e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 632e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 633e7d39069STrond Myklebust { 634e7d39069STrond Myklebust struct inode *inode = page->mapping->host; 635e7d39069STrond Myklebust struct nfs_page *req; 636e7d39069STrond Myklebust int error; 637e7d39069STrond Myklebust 638e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 639e7d39069STrond Myklebust if (req != NULL) 640e7d39069STrond Myklebust goto out; 641e7d39069STrond Myklebust req = nfs_create_request(ctx, inode, page, offset, bytes); 642e7d39069STrond Myklebust if (IS_ERR(req)) 643e7d39069STrond Myklebust goto out; 644e7d39069STrond Myklebust error = nfs_inode_add_request(inode, req); 645e7d39069STrond Myklebust if (error != 0) { 646e7d39069STrond Myklebust nfs_release_request(req); 647e7d39069STrond Myklebust req = ERR_PTR(error); 648e7d39069STrond Myklebust } 649efc91ed0STrond Myklebust out: 65061e930a9STrond Myklebust return req; 6511da177e4SLinus Torvalds } 6521da177e4SLinus Torvalds 653e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 654e7d39069STrond Myklebust unsigned int offset, unsigned int count) 655e7d39069STrond Myklebust { 656e7d39069STrond Myklebust struct nfs_page *req; 657e7d39069STrond Myklebust 658e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 659e7d39069STrond Myklebust if (IS_ERR(req)) 660e7d39069STrond Myklebust return PTR_ERR(req); 661e7d39069STrond Myklebust /* Update file length */ 662e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 663e7d39069STrond Myklebust nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 664e7d39069STrond Myklebust nfs_clear_page_tag_locked(req); 665e7d39069STrond Myklebust return 0; 666e7d39069STrond Myklebust } 667e7d39069STrond Myklebust 6681da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 6691da177e4SLinus Torvalds { 670cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 6711da177e4SLinus Torvalds struct nfs_page *req; 6721a54533eSTrond Myklebust int do_flush, status; 6731da177e4SLinus Torvalds /* 6741da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 6751da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 6761da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 6771da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 6781da177e4SLinus Torvalds * Also do the same if we find a request from an existing 6791da177e4SLinus Torvalds * dropped page. 6801da177e4SLinus Torvalds */ 6811a54533eSTrond Myklebust do { 682277459d2STrond Myklebust req = nfs_page_find_request(page); 6831a54533eSTrond Myklebust if (req == NULL) 6841a54533eSTrond Myklebust return 0; 6851a54533eSTrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx 686e261f51fSTrond Myklebust || !nfs_dirty_request(req); 6871da177e4SLinus Torvalds nfs_release_request(req); 6881a54533eSTrond Myklebust if (!do_flush) 6891a54533eSTrond Myklebust return 0; 690277459d2STrond Myklebust status = nfs_wb_page(page->mapping->host, page); 6911a54533eSTrond Myklebust } while (status == 0); 6921a54533eSTrond Myklebust return status; 6931da177e4SLinus Torvalds } 6941da177e4SLinus Torvalds 6951da177e4SLinus Torvalds /* 6965d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 6975d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 6985d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 6995d47a356STrond Myklebust */ 7005d47a356STrond Myklebust static int nfs_write_pageuptodate(struct page *page, struct inode *inode) 7015d47a356STrond Myklebust { 7025d47a356STrond Myklebust return PageUptodate(page) && 7035d47a356STrond Myklebust !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); 7045d47a356STrond Myklebust } 7055d47a356STrond Myklebust 7065d47a356STrond Myklebust /* 7071da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 7081da177e4SLinus Torvalds * 7091da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 7101da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 7111da177e4SLinus Torvalds */ 7121da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 7131da177e4SLinus Torvalds unsigned int offset, unsigned int count) 7141da177e4SLinus Torvalds { 715cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 7161da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 7171da177e4SLinus Torvalds int status = 0; 7181da177e4SLinus Torvalds 71991d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 72091d5b470SChuck Lever 72148186c7dSChuck Lever dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", 72201cce933SJosef "Jeff" Sipek file->f_path.dentry->d_parent->d_name.name, 72301cce933SJosef "Jeff" Sipek file->f_path.dentry->d_name.name, count, 7240bbacc40SChuck Lever (long long)(page_offset(page) + offset)); 7251da177e4SLinus Torvalds 7261da177e4SLinus Torvalds /* If we're not using byte range locks, and we know the page 7275d47a356STrond Myklebust * is up to date, it may be more efficient to extend the write 7285d47a356STrond Myklebust * to cover the entire page in order to avoid fragmentation 7295d47a356STrond Myklebust * inefficiencies. 7301da177e4SLinus Torvalds */ 7315d47a356STrond Myklebust if (nfs_write_pageuptodate(page, inode) && 7325d47a356STrond Myklebust inode->i_flock == NULL && 7334b5621f6STrond Myklebust !(file->f_flags & O_SYNC)) { 73449a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 7351da177e4SLinus Torvalds offset = 0; 7361da177e4SLinus Torvalds } 7371da177e4SLinus Torvalds 738e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 73903fa9e84STrond Myklebust if (status < 0) 74003fa9e84STrond Myklebust nfs_set_pageerror(page); 74103fa9e84STrond Myklebust else 742e261f51fSTrond Myklebust __set_page_dirty_nobuffers(page); 7431da177e4SLinus Torvalds 74448186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 7451da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 7461da177e4SLinus Torvalds return status; 7471da177e4SLinus Torvalds } 7481da177e4SLinus Torvalds 7491da177e4SLinus Torvalds static void nfs_writepage_release(struct nfs_page *req) 7501da177e4SLinus Torvalds { 7518e821cadSTrond Myklebust 7527e5f6146STrond Myklebust if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req)) { 75389a09141SPeter Zijlstra nfs_end_page_writeback(req->wb_page); 7541da177e4SLinus Torvalds nfs_inode_remove_request(req); 7558e821cadSTrond Myklebust } else 7568e821cadSTrond Myklebust nfs_end_page_writeback(req->wb_page); 7579fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 7581da177e4SLinus Torvalds } 7591da177e4SLinus Torvalds 7603ff7576dSTrond Myklebust static int flush_task_priority(int how) 7611da177e4SLinus Torvalds { 7621da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 7631da177e4SLinus Torvalds case FLUSH_HIGHPRI: 7641da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 7651da177e4SLinus Torvalds case FLUSH_LOWPRI: 7661da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 7671da177e4SLinus Torvalds } 7681da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 7691da177e4SLinus Torvalds } 7701da177e4SLinus Torvalds 7711da177e4SLinus Torvalds /* 7721da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 7731da177e4SLinus Torvalds */ 774dbae4c73STrond Myklebust static int nfs_write_rpcsetup(struct nfs_page *req, 7751da177e4SLinus Torvalds struct nfs_write_data *data, 776788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 7771da177e4SLinus Torvalds unsigned int count, unsigned int offset, 7781da177e4SLinus Torvalds int how) 7791da177e4SLinus Torvalds { 78084115e1cSTrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 78184115e1cSTrond Myklebust int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 7823ff7576dSTrond Myklebust int priority = flush_task_priority(how); 78307737691STrond Myklebust struct rpc_task *task; 784bdc7f021STrond Myklebust struct rpc_message msg = { 785bdc7f021STrond Myklebust .rpc_argp = &data->args, 786bdc7f021STrond Myklebust .rpc_resp = &data->res, 787bdc7f021STrond Myklebust .rpc_cred = req->wb_context->cred, 788bdc7f021STrond Myklebust }; 78984115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 79084115e1cSTrond Myklebust .rpc_client = NFS_CLIENT(inode), 79107737691STrond Myklebust .task = &data->task, 792bdc7f021STrond Myklebust .rpc_message = &msg, 79384115e1cSTrond Myklebust .callback_ops = call_ops, 79484115e1cSTrond Myklebust .callback_data = data, 795101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 79684115e1cSTrond Myklebust .flags = flags, 7973ff7576dSTrond Myklebust .priority = priority, 79884115e1cSTrond Myklebust }; 7991da177e4SLinus Torvalds 8001da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 8011da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 8021da177e4SLinus Torvalds 8031da177e4SLinus Torvalds data->req = req; 80488be9f99STrond Myklebust data->inode = inode = req->wb_context->path.dentry->d_inode; 805bdc7f021STrond Myklebust data->cred = msg.rpc_cred; 8061da177e4SLinus Torvalds 8071da177e4SLinus Torvalds data->args.fh = NFS_FH(inode); 8081da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 8091da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 8101da177e4SLinus Torvalds data->args.pages = data->pagevec; 8111da177e4SLinus Torvalds data->args.count = count; 812383ba719STrond Myklebust data->args.context = get_nfs_open_context(req->wb_context); 813bdc7f021STrond Myklebust data->args.stable = NFS_UNSTABLE; 814bdc7f021STrond Myklebust if (how & FLUSH_STABLE) { 815bdc7f021STrond Myklebust data->args.stable = NFS_DATA_SYNC; 816bdc7f021STrond Myklebust if (!NFS_I(inode)->ncommit) 817bdc7f021STrond Myklebust data->args.stable = NFS_FILE_SYNC; 818bdc7f021STrond Myklebust } 8191da177e4SLinus Torvalds 8201da177e4SLinus Torvalds data->res.fattr = &data->fattr; 8211da177e4SLinus Torvalds data->res.count = count; 8221da177e4SLinus Torvalds data->res.verf = &data->verf; 8230e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 8241da177e4SLinus Torvalds 825788e7a89STrond Myklebust /* Set up the initial task struct. */ 826bdc7f021STrond Myklebust NFS_PROTO(inode)->write_setup(data, &msg); 8271da177e4SLinus Torvalds 828a3f565b1SChuck Lever dprintk("NFS: %5u initiated write call " 82948186c7dSChuck Lever "(req %s/%lld, %u bytes @ offset %llu)\n", 8300bbacc40SChuck Lever data->task.tk_pid, 8311da177e4SLinus Torvalds inode->i_sb->s_id, 8321da177e4SLinus Torvalds (long long)NFS_FILEID(inode), 8331da177e4SLinus Torvalds count, 8341da177e4SLinus Torvalds (unsigned long long)data->args.offset); 8351da177e4SLinus Torvalds 83607737691STrond Myklebust task = rpc_run_task(&task_setup_data); 837dbae4c73STrond Myklebust if (IS_ERR(task)) 838dbae4c73STrond Myklebust return PTR_ERR(task); 83907737691STrond Myklebust rpc_put_task(task); 840dbae4c73STrond Myklebust return 0; 8411da177e4SLinus Torvalds } 8421da177e4SLinus Torvalds 8436d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 8446d884e8fSFred * call this on each, which will prepare them to be retried on next 8456d884e8fSFred * writeback using standard nfs. 8466d884e8fSFred */ 8476d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 8486d884e8fSFred { 8496d884e8fSFred nfs_mark_request_dirty(req); 8506d884e8fSFred nfs_end_page_writeback(req->wb_page); 8516d884e8fSFred nfs_clear_page_tag_locked(req); 8526d884e8fSFred } 8536d884e8fSFred 8541da177e4SLinus Torvalds /* 8551da177e4SLinus Torvalds * Generate multiple small requests to write out a single 8561da177e4SLinus Torvalds * contiguous dirty area on one page. 8571da177e4SLinus Torvalds */ 8588d5658c9STrond Myklebust static int nfs_flush_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) 8591da177e4SLinus Torvalds { 8601da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 8611da177e4SLinus Torvalds struct page *page = req->wb_page; 8621da177e4SLinus Torvalds struct nfs_write_data *data; 863e9f7bee1STrond Myklebust size_t wsize = NFS_SERVER(inode)->wsize, nbytes; 864e9f7bee1STrond Myklebust unsigned int offset; 8651da177e4SLinus Torvalds int requests = 0; 866dbae4c73STrond Myklebust int ret = 0; 8671da177e4SLinus Torvalds LIST_HEAD(list); 8681da177e4SLinus Torvalds 8691da177e4SLinus Torvalds nfs_list_remove_request(req); 8701da177e4SLinus Torvalds 871bcb71bbaSTrond Myklebust nbytes = count; 872e9f7bee1STrond Myklebust do { 873e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 874e9f7bee1STrond Myklebust 8758d5658c9STrond Myklebust data = nfs_writedata_alloc(1); 8761da177e4SLinus Torvalds if (!data) 8771da177e4SLinus Torvalds goto out_bad; 8781da177e4SLinus Torvalds list_add(&data->pages, &list); 8791da177e4SLinus Torvalds requests++; 880e9f7bee1STrond Myklebust nbytes -= len; 881e9f7bee1STrond Myklebust } while (nbytes != 0); 8821da177e4SLinus Torvalds atomic_set(&req->wb_complete, requests); 8831da177e4SLinus Torvalds 8841da177e4SLinus Torvalds ClearPageError(page); 8851da177e4SLinus Torvalds offset = 0; 886bcb71bbaSTrond Myklebust nbytes = count; 8871da177e4SLinus Torvalds do { 888dbae4c73STrond Myklebust int ret2; 889dbae4c73STrond Myklebust 8901da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 8911da177e4SLinus Torvalds list_del_init(&data->pages); 8921da177e4SLinus Torvalds 8931da177e4SLinus Torvalds data->pagevec[0] = page; 8941da177e4SLinus Torvalds 895bcb71bbaSTrond Myklebust if (nbytes < wsize) 896bcb71bbaSTrond Myklebust wsize = nbytes; 897dbae4c73STrond Myklebust ret2 = nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 898788e7a89STrond Myklebust wsize, offset, how); 899dbae4c73STrond Myklebust if (ret == 0) 900dbae4c73STrond Myklebust ret = ret2; 9011da177e4SLinus Torvalds offset += wsize; 9021da177e4SLinus Torvalds nbytes -= wsize; 9031da177e4SLinus Torvalds } while (nbytes != 0); 9041da177e4SLinus Torvalds 905dbae4c73STrond Myklebust return ret; 9061da177e4SLinus Torvalds 9071da177e4SLinus Torvalds out_bad: 9081da177e4SLinus Torvalds while (!list_empty(&list)) { 9091da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 9101da177e4SLinus Torvalds list_del(&data->pages); 9118aca67f0STrond Myklebust nfs_writedata_release(data); 9121da177e4SLinus Torvalds } 91361822ab5STrond Myklebust nfs_redirty_request(req); 9141da177e4SLinus Torvalds return -ENOMEM; 9151da177e4SLinus Torvalds } 9161da177e4SLinus Torvalds 9171da177e4SLinus Torvalds /* 9181da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 9191da177e4SLinus Torvalds * The page must have been locked by the caller. 9201da177e4SLinus Torvalds * 9211da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 9221da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 9231da177e4SLinus Torvalds * that has been written but not committed. 9241da177e4SLinus Torvalds */ 9258d5658c9STrond Myklebust static int nfs_flush_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) 9261da177e4SLinus Torvalds { 9271da177e4SLinus Torvalds struct nfs_page *req; 9281da177e4SLinus Torvalds struct page **pages; 9291da177e4SLinus Torvalds struct nfs_write_data *data; 9301da177e4SLinus Torvalds 9318d5658c9STrond Myklebust data = nfs_writedata_alloc(npages); 9321da177e4SLinus Torvalds if (!data) 9331da177e4SLinus Torvalds goto out_bad; 9341da177e4SLinus Torvalds 9351da177e4SLinus Torvalds pages = data->pagevec; 9361da177e4SLinus Torvalds while (!list_empty(head)) { 9371da177e4SLinus Torvalds req = nfs_list_entry(head->next); 9381da177e4SLinus Torvalds nfs_list_remove_request(req); 9391da177e4SLinus Torvalds nfs_list_add_request(req, &data->pages); 9401da177e4SLinus Torvalds ClearPageError(req->wb_page); 9411da177e4SLinus Torvalds *pages++ = req->wb_page; 9421da177e4SLinus Torvalds } 9431da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 9441da177e4SLinus Torvalds 9451da177e4SLinus Torvalds /* Set up the argument struct */ 946dbae4c73STrond Myklebust return nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how); 9471da177e4SLinus Torvalds out_bad: 9481da177e4SLinus Torvalds while (!list_empty(head)) { 94910afec90STrond Myklebust req = nfs_list_entry(head->next); 9501da177e4SLinus Torvalds nfs_list_remove_request(req); 95161822ab5STrond Myklebust nfs_redirty_request(req); 9521da177e4SLinus Torvalds } 9531da177e4SLinus Torvalds return -ENOMEM; 9541da177e4SLinus Torvalds } 9551da177e4SLinus Torvalds 956c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 957c63c7b05STrond Myklebust struct inode *inode, int ioflags) 9581da177e4SLinus Torvalds { 959bf4285e7SChuck Lever size_t wsize = NFS_SERVER(inode)->wsize; 9601da177e4SLinus Torvalds 961bcb71bbaSTrond Myklebust if (wsize < PAGE_CACHE_SIZE) 962c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags); 963bcb71bbaSTrond Myklebust else 964c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags); 9651da177e4SLinus Torvalds } 9661da177e4SLinus Torvalds 9671da177e4SLinus Torvalds /* 9681da177e4SLinus Torvalds * Handle a write reply that flushed part of a page. 9691da177e4SLinus Torvalds */ 970788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) 9711da177e4SLinus Torvalds { 972788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 9731da177e4SLinus Torvalds 97448186c7dSChuck Lever dprintk("NFS: %5u write(%s/%lld %d@%lld)", 97548186c7dSChuck Lever task->tk_pid, 97648186c7dSChuck Lever data->req->wb_context->path.dentry->d_inode->i_sb->s_id, 97748186c7dSChuck Lever (long long) 97848186c7dSChuck Lever NFS_FILEID(data->req->wb_context->path.dentry->d_inode), 97948186c7dSChuck Lever data->req->wb_bytes, (long long)req_offset(data->req)); 9801da177e4SLinus Torvalds 981c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 982c9d8f89dSTrond Myklebust } 983788e7a89STrond Myklebust 984c9d8f89dSTrond Myklebust static void nfs_writeback_release_partial(void *calldata) 985c9d8f89dSTrond Myklebust { 986c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 987c9d8f89dSTrond Myklebust struct nfs_page *req = data->req; 988c9d8f89dSTrond Myklebust struct page *page = req->wb_page; 989c9d8f89dSTrond Myklebust int status = data->task.tk_status; 990c9d8f89dSTrond Myklebust 991c9d8f89dSTrond Myklebust if (status < 0) { 992a301b777STrond Myklebust nfs_set_pageerror(page); 993c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 994c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 9958e821cadSTrond Myklebust goto out; 9968e821cadSTrond Myklebust } 9978e821cadSTrond Myklebust 9988e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 999587142f8STrond Myklebust struct inode *inode = page->mapping->host; 10008e821cadSTrond Myklebust 1001587142f8STrond Myklebust spin_lock(&inode->i_lock); 10028e821cadSTrond Myklebust if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { 10038e821cadSTrond Myklebust /* Do nothing we need to resend the writes */ 10048e821cadSTrond Myklebust } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { 10051da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 10061da177e4SLinus Torvalds dprintk(" defer commit\n"); 10071da177e4SLinus Torvalds } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { 10088e821cadSTrond Myklebust set_bit(PG_NEED_RESCHED, &req->wb_flags); 10098e821cadSTrond Myklebust clear_bit(PG_NEED_COMMIT, &req->wb_flags); 10101da177e4SLinus Torvalds dprintk(" server reboot detected\n"); 10111da177e4SLinus Torvalds } 1012587142f8STrond Myklebust spin_unlock(&inode->i_lock); 10131da177e4SLinus Torvalds } else 10141da177e4SLinus Torvalds dprintk(" OK\n"); 10151da177e4SLinus Torvalds 10168e821cadSTrond Myklebust out: 10171da177e4SLinus Torvalds if (atomic_dec_and_test(&req->wb_complete)) 10181da177e4SLinus Torvalds nfs_writepage_release(req); 1019c9d8f89dSTrond Myklebust nfs_writedata_release(calldata); 10201da177e4SLinus Torvalds } 10211da177e4SLinus Torvalds 1022788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops = { 1023788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_partial, 1024c9d8f89dSTrond Myklebust .rpc_release = nfs_writeback_release_partial, 1025788e7a89STrond Myklebust }; 1026788e7a89STrond Myklebust 10271da177e4SLinus Torvalds /* 10281da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 10291da177e4SLinus Torvalds * 10301da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 10311da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 10321da177e4SLinus Torvalds * as the page has a write request pending. 10331da177e4SLinus Torvalds */ 1034788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) 10351da177e4SLinus Torvalds { 1036788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 10371da177e4SLinus Torvalds 1038c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 1039c9d8f89dSTrond Myklebust } 1040c9d8f89dSTrond Myklebust 1041c9d8f89dSTrond Myklebust static void nfs_writeback_release_full(void *calldata) 1042c9d8f89dSTrond Myklebust { 1043c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1044c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1045788e7a89STrond Myklebust 10461da177e4SLinus Torvalds /* Update attributes as result of writeback. */ 10471da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 1048c9d8f89dSTrond Myklebust struct nfs_page *req = nfs_list_entry(data->pages.next); 1049c9d8f89dSTrond Myklebust struct page *page = req->wb_page; 1050c9d8f89dSTrond Myklebust 10511da177e4SLinus Torvalds nfs_list_remove_request(req); 10521da177e4SLinus Torvalds 105348186c7dSChuck Lever dprintk("NFS: %5u write (%s/%lld %d@%lld)", 105448186c7dSChuck Lever data->task.tk_pid, 105588be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 105688be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 10571da177e4SLinus Torvalds req->wb_bytes, 10581da177e4SLinus Torvalds (long long)req_offset(req)); 10591da177e4SLinus Torvalds 1060c9d8f89dSTrond Myklebust if (status < 0) { 1061a301b777STrond Myklebust nfs_set_pageerror(page); 1062c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 1063c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 10648e821cadSTrond Myklebust goto remove_request; 10651da177e4SLinus Torvalds } 10661da177e4SLinus Torvalds 10678e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 10681da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 10691da177e4SLinus Torvalds nfs_mark_request_commit(req); 10708e821cadSTrond Myklebust nfs_end_page_writeback(page); 10711da177e4SLinus Torvalds dprintk(" marked for commit\n"); 10728e821cadSTrond Myklebust goto next; 10738e821cadSTrond Myklebust } 10748e821cadSTrond Myklebust dprintk(" OK\n"); 10758e821cadSTrond Myklebust remove_request: 10768e821cadSTrond Myklebust nfs_end_page_writeback(page); 10771da177e4SLinus Torvalds nfs_inode_remove_request(req); 10781da177e4SLinus Torvalds next: 10799fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 10801da177e4SLinus Torvalds } 1081c9d8f89dSTrond Myklebust nfs_writedata_release(calldata); 10821da177e4SLinus Torvalds } 10831da177e4SLinus Torvalds 1084788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = { 1085788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_full, 1086c9d8f89dSTrond Myklebust .rpc_release = nfs_writeback_release_full, 1087788e7a89STrond Myklebust }; 1088788e7a89STrond Myklebust 1089788e7a89STrond Myklebust 10901da177e4SLinus Torvalds /* 10911da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 10921da177e4SLinus Torvalds */ 1093462d5b32SChuck Lever int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 10941da177e4SLinus Torvalds { 10951da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 10961da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1097788e7a89STrond Myklebust int status; 10981da177e4SLinus Torvalds 1099a3f565b1SChuck Lever dprintk("NFS: %5u nfs_writeback_done (status %d)\n", 11001da177e4SLinus Torvalds task->tk_pid, task->tk_status); 11011da177e4SLinus Torvalds 1102f551e44fSChuck Lever /* 1103f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1104f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1105f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1106f551e44fSChuck Lever * another writer had changed the file, but some applications 1107f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1108f551e44fSChuck Lever */ 1109788e7a89STrond Myklebust status = NFS_PROTO(data->inode)->write_done(task, data); 1110788e7a89STrond Myklebust if (status != 0) 1111788e7a89STrond Myklebust return status; 111291d5b470SChuck Lever nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 111391d5b470SChuck Lever 11141da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 11151da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 11161da177e4SLinus Torvalds /* We tried a write call, but the server did not 11171da177e4SLinus Torvalds * commit data to stable storage even though we 11181da177e4SLinus Torvalds * requested it. 11191da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 11201da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 11211da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 11221da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 11231da177e4SLinus Torvalds */ 11241da177e4SLinus Torvalds static unsigned long complain; 11251da177e4SLinus Torvalds 11261da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 11271da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 11281da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 112954ceac45SDavid Howells NFS_SERVER(data->inode)->nfs_client->cl_hostname, 11301da177e4SLinus Torvalds resp->verf->committed, argp->stable); 11311da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11321da177e4SLinus Torvalds } 11331da177e4SLinus Torvalds } 11341da177e4SLinus Torvalds #endif 11351da177e4SLinus Torvalds /* Is this a short write? */ 11361da177e4SLinus Torvalds if (task->tk_status >= 0 && resp->count < argp->count) { 11371da177e4SLinus Torvalds static unsigned long complain; 11381da177e4SLinus Torvalds 113991d5b470SChuck Lever nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); 114091d5b470SChuck Lever 11411da177e4SLinus Torvalds /* Has the server at least made some progress? */ 11421da177e4SLinus Torvalds if (resp->count != 0) { 11431da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 11441da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 11451da177e4SLinus Torvalds /* Resend from where the server left off */ 11461da177e4SLinus Torvalds argp->offset += resp->count; 11471da177e4SLinus Torvalds argp->pgbase += resp->count; 11481da177e4SLinus Torvalds argp->count -= resp->count; 11491da177e4SLinus Torvalds } else { 11501da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 11511da177e4SLinus Torvalds * headaches in the case of a server crash. 11521da177e4SLinus Torvalds */ 11531da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 11541da177e4SLinus Torvalds } 11551da177e4SLinus Torvalds rpc_restart_call(task); 1156788e7a89STrond Myklebust return -EAGAIN; 11571da177e4SLinus Torvalds } 11581da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 11591da177e4SLinus Torvalds printk(KERN_WARNING 11601da177e4SLinus Torvalds "NFS: Server wrote zero bytes, expected %u.\n", 11611da177e4SLinus Torvalds argp->count); 11621da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11631da177e4SLinus Torvalds } 11641da177e4SLinus Torvalds /* Can't do anything about it except throw an error. */ 11651da177e4SLinus Torvalds task->tk_status = -EIO; 11661da177e4SLinus Torvalds } 1167788e7a89STrond Myklebust return 0; 11681da177e4SLinus Torvalds } 11691da177e4SLinus Torvalds 11701da177e4SLinus Torvalds 11711da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 1172c9d8f89dSTrond Myklebust void nfs_commitdata_release(void *data) 11731da177e4SLinus Torvalds { 1174383ba719STrond Myklebust struct nfs_write_data *wdata = data; 1175383ba719STrond Myklebust 1176383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 11771da177e4SLinus Torvalds nfs_commit_free(wdata); 11781da177e4SLinus Torvalds } 11791da177e4SLinus Torvalds 11801da177e4SLinus Torvalds /* 11811da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 11821da177e4SLinus Torvalds */ 1183dbae4c73STrond Myklebust static int nfs_commit_rpcsetup(struct list_head *head, 1184788e7a89STrond Myklebust struct nfs_write_data *data, 1185788e7a89STrond Myklebust int how) 11861da177e4SLinus Torvalds { 118784115e1cSTrond Myklebust struct nfs_page *first = nfs_list_entry(head->next); 118884115e1cSTrond Myklebust struct inode *inode = first->wb_context->path.dentry->d_inode; 118984115e1cSTrond Myklebust int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 11903ff7576dSTrond Myklebust int priority = flush_task_priority(how); 119107737691STrond Myklebust struct rpc_task *task; 1192bdc7f021STrond Myklebust struct rpc_message msg = { 1193bdc7f021STrond Myklebust .rpc_argp = &data->args, 1194bdc7f021STrond Myklebust .rpc_resp = &data->res, 1195bdc7f021STrond Myklebust .rpc_cred = first->wb_context->cred, 1196bdc7f021STrond Myklebust }; 119784115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 119807737691STrond Myklebust .task = &data->task, 119984115e1cSTrond Myklebust .rpc_client = NFS_CLIENT(inode), 1200bdc7f021STrond Myklebust .rpc_message = &msg, 120184115e1cSTrond Myklebust .callback_ops = &nfs_commit_ops, 120284115e1cSTrond Myklebust .callback_data = data, 1203101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 120484115e1cSTrond Myklebust .flags = flags, 12053ff7576dSTrond Myklebust .priority = priority, 120684115e1cSTrond Myklebust }; 12071da177e4SLinus Torvalds 12081da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 12091da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 12101da177e4SLinus Torvalds 12111da177e4SLinus Torvalds list_splice_init(head, &data->pages); 12121da177e4SLinus Torvalds 12131da177e4SLinus Torvalds data->inode = inode; 1214bdc7f021STrond Myklebust data->cred = msg.rpc_cred; 12151da177e4SLinus Torvalds 12161da177e4SLinus Torvalds data->args.fh = NFS_FH(data->inode); 12173da28eb1STrond Myklebust /* Note: we always request a commit of the entire inode */ 12183da28eb1STrond Myklebust data->args.offset = 0; 12193da28eb1STrond Myklebust data->args.count = 0; 1220383ba719STrond Myklebust data->args.context = get_nfs_open_context(first->wb_context); 12213da28eb1STrond Myklebust data->res.count = 0; 12221da177e4SLinus Torvalds data->res.fattr = &data->fattr; 12231da177e4SLinus Torvalds data->res.verf = &data->verf; 12240e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 12251da177e4SLinus Torvalds 1226788e7a89STrond Myklebust /* Set up the initial task struct. */ 1227bdc7f021STrond Myklebust NFS_PROTO(inode)->commit_setup(data, &msg); 12281da177e4SLinus Torvalds 1229a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1230bdc7f021STrond Myklebust 123107737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1232dbae4c73STrond Myklebust if (IS_ERR(task)) 1233dbae4c73STrond Myklebust return PTR_ERR(task); 123407737691STrond Myklebust rpc_put_task(task); 1235dbae4c73STrond Myklebust return 0; 12361da177e4SLinus Torvalds } 12371da177e4SLinus Torvalds 12381da177e4SLinus Torvalds /* 12391da177e4SLinus Torvalds * Commit dirty pages 12401da177e4SLinus Torvalds */ 12411da177e4SLinus Torvalds static int 124240859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how) 12431da177e4SLinus Torvalds { 12441da177e4SLinus Torvalds struct nfs_write_data *data; 12451da177e4SLinus Torvalds struct nfs_page *req; 12461da177e4SLinus Torvalds 1247c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 12481da177e4SLinus Torvalds 12491da177e4SLinus Torvalds if (!data) 12501da177e4SLinus Torvalds goto out_bad; 12511da177e4SLinus Torvalds 12521da177e4SLinus Torvalds /* Set up the argument struct */ 1253dbae4c73STrond Myklebust return nfs_commit_rpcsetup(head, data, how); 12541da177e4SLinus Torvalds out_bad: 12551da177e4SLinus Torvalds while (!list_empty(head)) { 12561da177e4SLinus Torvalds req = nfs_list_entry(head->next); 12571da177e4SLinus Torvalds nfs_list_remove_request(req); 12581da177e4SLinus Torvalds nfs_mark_request_commit(req); 125983715ad5STrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1260c9e51e41SPeter Zijlstra dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 1261c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 12629fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 12631da177e4SLinus Torvalds } 12641da177e4SLinus Torvalds return -ENOMEM; 12651da177e4SLinus Torvalds } 12661da177e4SLinus Torvalds 12671da177e4SLinus Torvalds /* 12681da177e4SLinus Torvalds * COMMIT call returned 12691da177e4SLinus Torvalds */ 1270788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 12711da177e4SLinus Torvalds { 1272963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 12731da177e4SLinus Torvalds 1274a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 12751da177e4SLinus Torvalds task->tk_pid, task->tk_status); 12761da177e4SLinus Torvalds 1277788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1278788e7a89STrond Myklebust if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) 1279788e7a89STrond Myklebust return; 1280c9d8f89dSTrond Myklebust } 1281c9d8f89dSTrond Myklebust 1282c9d8f89dSTrond Myklebust static void nfs_commit_release(void *calldata) 1283c9d8f89dSTrond Myklebust { 1284c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1285c9d8f89dSTrond Myklebust struct nfs_page *req; 1286c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1287788e7a89STrond Myklebust 12881da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 12891da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 12901da177e4SLinus Torvalds nfs_list_remove_request(req); 1291612c9384STrond Myklebust clear_bit(PG_NEED_COMMIT, &(req)->wb_flags); 1292fd39fc85SChristoph Lameter dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1293c9e51e41SPeter Zijlstra dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 1294c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 12951da177e4SLinus Torvalds 129648186c7dSChuck Lever dprintk("NFS: commit (%s/%lld %d@%lld)", 129788be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 129888be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 12991da177e4SLinus Torvalds req->wb_bytes, 13001da177e4SLinus Torvalds (long long)req_offset(req)); 1301c9d8f89dSTrond Myklebust if (status < 0) { 1302c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 13031da177e4SLinus Torvalds nfs_inode_remove_request(req); 1304c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 13051da177e4SLinus Torvalds goto next; 13061da177e4SLinus Torvalds } 13071da177e4SLinus Torvalds 13081da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 13091da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 13101da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 13111da177e4SLinus Torvalds /* We have a match */ 13121da177e4SLinus Torvalds nfs_inode_remove_request(req); 13131da177e4SLinus Torvalds dprintk(" OK\n"); 13141da177e4SLinus Torvalds goto next; 13151da177e4SLinus Torvalds } 13161da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 13171da177e4SLinus Torvalds dprintk(" mismatch\n"); 13186d884e8fSFred nfs_mark_request_dirty(req); 13191da177e4SLinus Torvalds next: 13209fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 13211da177e4SLinus Torvalds } 1322c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 13231da177e4SLinus Torvalds } 1324788e7a89STrond Myklebust 1325788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 1326788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1327788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1328788e7a89STrond Myklebust }; 13291da177e4SLinus Torvalds 13303da28eb1STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 13311da177e4SLinus Torvalds { 13321da177e4SLinus Torvalds LIST_HEAD(head); 13337d46a49fSTrond Myklebust int res; 13341da177e4SLinus Torvalds 1335587142f8STrond Myklebust spin_lock(&inode->i_lock); 13363da28eb1STrond Myklebust res = nfs_scan_commit(inode, &head, 0, 0); 1337587142f8STrond Myklebust spin_unlock(&inode->i_lock); 13381da177e4SLinus Torvalds if (res) { 13397d46a49fSTrond Myklebust int error = nfs_commit_list(inode, &head, how); 13401da177e4SLinus Torvalds if (error < 0) 13411da177e4SLinus Torvalds return error; 13423da28eb1STrond Myklebust } 13431da177e4SLinus Torvalds return res; 13441da177e4SLinus Torvalds } 1345c63c7b05STrond Myklebust #else 1346c63c7b05STrond Myklebust static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how) 1347c63c7b05STrond Myklebust { 1348c63c7b05STrond Myklebust return 0; 1349c63c7b05STrond Myklebust } 13501da177e4SLinus Torvalds #endif 13511da177e4SLinus Torvalds 13521c75950bSTrond Myklebust long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how) 13531da177e4SLinus Torvalds { 13541c75950bSTrond Myklebust struct inode *inode = mapping->host; 1355ca52fec1STrond Myklebust pgoff_t idx_start, idx_end; 13561c75950bSTrond Myklebust unsigned int npages = 0; 1357c42de9ddSTrond Myklebust LIST_HEAD(head); 135870b9ecbdSTrond Myklebust int nocommit = how & FLUSH_NOCOMMIT; 13593f442547STrond Myklebust long pages, ret; 13601da177e4SLinus Torvalds 13611c75950bSTrond Myklebust /* FIXME */ 13621c75950bSTrond Myklebust if (wbc->range_cyclic) 13631c75950bSTrond Myklebust idx_start = 0; 13641c75950bSTrond Myklebust else { 13651c75950bSTrond Myklebust idx_start = wbc->range_start >> PAGE_CACHE_SHIFT; 13661c75950bSTrond Myklebust idx_end = wbc->range_end >> PAGE_CACHE_SHIFT; 13671c75950bSTrond Myklebust if (idx_end > idx_start) { 1368ca52fec1STrond Myklebust pgoff_t l_npages = 1 + idx_end - idx_start; 13691c75950bSTrond Myklebust npages = l_npages; 13701c75950bSTrond Myklebust if (sizeof(npages) != sizeof(l_npages) && 1371ca52fec1STrond Myklebust (pgoff_t)npages != l_npages) 13721c75950bSTrond Myklebust npages = 0; 13731c75950bSTrond Myklebust } 13741c75950bSTrond Myklebust } 1375c42de9ddSTrond Myklebust how &= ~FLUSH_NOCOMMIT; 1376587142f8STrond Myklebust spin_lock(&inode->i_lock); 13771da177e4SLinus Torvalds do { 1378c42de9ddSTrond Myklebust ret = nfs_wait_on_requests_locked(inode, idx_start, npages); 1379c42de9ddSTrond Myklebust if (ret != 0) 1380c42de9ddSTrond Myklebust continue; 1381c42de9ddSTrond Myklebust if (nocommit) 1382c42de9ddSTrond Myklebust break; 1383d2ccddf0STrond Myklebust pages = nfs_scan_commit(inode, &head, idx_start, npages); 1384724c439cSTrond Myklebust if (pages == 0) 1385c42de9ddSTrond Myklebust break; 1386d2ccddf0STrond Myklebust if (how & FLUSH_INVALIDATE) { 1387587142f8STrond Myklebust spin_unlock(&inode->i_lock); 138883715ad5STrond Myklebust nfs_cancel_commit_list(&head); 1389e8e058e8STrond Myklebust ret = pages; 1390587142f8STrond Myklebust spin_lock(&inode->i_lock); 1391d2ccddf0STrond Myklebust continue; 1392d2ccddf0STrond Myklebust } 1393d2ccddf0STrond Myklebust pages += nfs_scan_commit(inode, &head, 0, 0); 1394587142f8STrond Myklebust spin_unlock(&inode->i_lock); 1395c42de9ddSTrond Myklebust ret = nfs_commit_list(inode, &head, how); 1396587142f8STrond Myklebust spin_lock(&inode->i_lock); 1397587142f8STrond Myklebust 1398c42de9ddSTrond Myklebust } while (ret >= 0); 1399587142f8STrond Myklebust spin_unlock(&inode->i_lock); 1400c42de9ddSTrond Myklebust return ret; 14011da177e4SLinus Torvalds } 14021da177e4SLinus Torvalds 140334901f70STrond Myklebust static int __nfs_write_mapping(struct address_space *mapping, struct writeback_control *wbc, int how) 14041c75950bSTrond Myklebust { 14051c75950bSTrond Myklebust int ret; 14061c75950bSTrond Myklebust 140734901f70STrond Myklebust ret = nfs_writepages(mapping, wbc); 140861822ab5STrond Myklebust if (ret < 0) 140961822ab5STrond Myklebust goto out; 141034901f70STrond Myklebust ret = nfs_sync_mapping_wait(mapping, wbc, how); 1411ed90ef51STrond Myklebust if (ret < 0) 1412ed90ef51STrond Myklebust goto out; 14131c75950bSTrond Myklebust return 0; 141461822ab5STrond Myklebust out: 1415e507d9ebSTrond Myklebust __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 14161c75950bSTrond Myklebust return ret; 14171c75950bSTrond Myklebust } 14181c75950bSTrond Myklebust 141934901f70STrond Myklebust /* Two pass sync: first using WB_SYNC_NONE, then WB_SYNC_ALL */ 142034901f70STrond Myklebust static int nfs_write_mapping(struct address_space *mapping, int how) 142134901f70STrond Myklebust { 142234901f70STrond Myklebust struct writeback_control wbc = { 142334901f70STrond Myklebust .bdi = mapping->backing_dev_info, 142434901f70STrond Myklebust .sync_mode = WB_SYNC_NONE, 142534901f70STrond Myklebust .nr_to_write = LONG_MAX, 142634901f70STrond Myklebust .for_writepages = 1, 142734901f70STrond Myklebust .range_cyclic = 1, 142834901f70STrond Myklebust }; 142934901f70STrond Myklebust int ret; 143034901f70STrond Myklebust 143134901f70STrond Myklebust ret = __nfs_write_mapping(mapping, &wbc, how); 143234901f70STrond Myklebust if (ret < 0) 143334901f70STrond Myklebust return ret; 143434901f70STrond Myklebust wbc.sync_mode = WB_SYNC_ALL; 143534901f70STrond Myklebust return __nfs_write_mapping(mapping, &wbc, how); 143634901f70STrond Myklebust } 143734901f70STrond Myklebust 1438ed90ef51STrond Myklebust /* 1439ed90ef51STrond Myklebust * flush the inode to disk. 1440ed90ef51STrond Myklebust */ 1441ed90ef51STrond Myklebust int nfs_wb_all(struct inode *inode) 14421c75950bSTrond Myklebust { 1443ed90ef51STrond Myklebust return nfs_write_mapping(inode->i_mapping, 0); 1444ed90ef51STrond Myklebust } 14451c75950bSTrond Myklebust 1446ed90ef51STrond Myklebust int nfs_wb_nocommit(struct inode *inode) 1447ed90ef51STrond Myklebust { 1448ed90ef51STrond Myklebust return nfs_write_mapping(inode->i_mapping, FLUSH_NOCOMMIT); 14491c75950bSTrond Myklebust } 14501c75950bSTrond Myklebust 14511b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 14521b3b4a1aSTrond Myklebust { 14531b3b4a1aSTrond Myklebust struct nfs_page *req; 14541b3b4a1aSTrond Myklebust loff_t range_start = page_offset(page); 14551b3b4a1aSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 14561b3b4a1aSTrond Myklebust struct writeback_control wbc = { 14571b3b4a1aSTrond Myklebust .bdi = page->mapping->backing_dev_info, 14581b3b4a1aSTrond Myklebust .sync_mode = WB_SYNC_ALL, 14591b3b4a1aSTrond Myklebust .nr_to_write = LONG_MAX, 14601b3b4a1aSTrond Myklebust .range_start = range_start, 14611b3b4a1aSTrond Myklebust .range_end = range_end, 14621b3b4a1aSTrond Myklebust }; 14631b3b4a1aSTrond Myklebust int ret = 0; 14641b3b4a1aSTrond Myklebust 14651b3b4a1aSTrond Myklebust BUG_ON(!PageLocked(page)); 14661b3b4a1aSTrond Myklebust for (;;) { 14671b3b4a1aSTrond Myklebust req = nfs_page_find_request(page); 14681b3b4a1aSTrond Myklebust if (req == NULL) 14691b3b4a1aSTrond Myklebust goto out; 14701b3b4a1aSTrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 14711b3b4a1aSTrond Myklebust nfs_release_request(req); 14721b3b4a1aSTrond Myklebust break; 14731b3b4a1aSTrond Myklebust } 14741b3b4a1aSTrond Myklebust if (nfs_lock_request_dontget(req)) { 14751b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 14761b3b4a1aSTrond Myklebust /* 14771b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 14781b3b4a1aSTrond Myklebust * page as being dirty 14791b3b4a1aSTrond Myklebust */ 14801b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 14811b3b4a1aSTrond Myklebust nfs_unlock_request(req); 14821b3b4a1aSTrond Myklebust break; 14831b3b4a1aSTrond Myklebust } 14841b3b4a1aSTrond Myklebust ret = nfs_wait_on_request(req); 14851b3b4a1aSTrond Myklebust if (ret < 0) 14861b3b4a1aSTrond Myklebust goto out; 14871b3b4a1aSTrond Myklebust } 14881b3b4a1aSTrond Myklebust if (!PagePrivate(page)) 14891b3b4a1aSTrond Myklebust return 0; 14901b3b4a1aSTrond Myklebust ret = nfs_sync_mapping_wait(page->mapping, &wbc, FLUSH_INVALIDATE); 14911b3b4a1aSTrond Myklebust out: 14921b3b4a1aSTrond Myklebust return ret; 14931b3b4a1aSTrond Myklebust } 14941b3b4a1aSTrond Myklebust 14955334eb13SAdrian Bunk static int nfs_wb_page_priority(struct inode *inode, struct page *page, 14965334eb13SAdrian Bunk int how) 14971c75950bSTrond Myklebust { 14981c75950bSTrond Myklebust loff_t range_start = page_offset(page); 14991c75950bSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 15004d770ccfSTrond Myklebust struct writeback_control wbc = { 15014d770ccfSTrond Myklebust .bdi = page->mapping->backing_dev_info, 15024d770ccfSTrond Myklebust .sync_mode = WB_SYNC_ALL, 15034d770ccfSTrond Myklebust .nr_to_write = LONG_MAX, 15044d770ccfSTrond Myklebust .range_start = range_start, 15054d770ccfSTrond Myklebust .range_end = range_end, 15064d770ccfSTrond Myklebust }; 15074d770ccfSTrond Myklebust int ret; 15081c75950bSTrond Myklebust 150973e3302fSTrond Myklebust do { 1510c63c7b05STrond Myklebust if (clear_page_dirty_for_io(page)) { 15114d770ccfSTrond Myklebust ret = nfs_writepage_locked(page, &wbc); 15124d770ccfSTrond Myklebust if (ret < 0) 151373e3302fSTrond Myklebust goto out_error; 151473e3302fSTrond Myklebust } else if (!PagePrivate(page)) 151573e3302fSTrond Myklebust break; 15164d770ccfSTrond Myklebust ret = nfs_sync_mapping_wait(page->mapping, &wbc, how); 151773e3302fSTrond Myklebust if (ret < 0) 151873e3302fSTrond Myklebust goto out_error; 151973e3302fSTrond Myklebust } while (PagePrivate(page)); 15204d770ccfSTrond Myklebust return 0; 152173e3302fSTrond Myklebust out_error: 1522e507d9ebSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_PAGES); 15234d770ccfSTrond Myklebust return ret; 15241c75950bSTrond Myklebust } 15251c75950bSTrond Myklebust 15261c75950bSTrond Myklebust /* 15271c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 15281c75950bSTrond Myklebust */ 15291c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page* page) 15301c75950bSTrond Myklebust { 15314d770ccfSTrond Myklebust return nfs_wb_page_priority(inode, page, FLUSH_STABLE); 15321c75950bSTrond Myklebust } 15331c75950bSTrond Myklebust 1534f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 15351da177e4SLinus Torvalds { 15361da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 15371da177e4SLinus Torvalds sizeof(struct nfs_write_data), 15381da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 153920c2df83SPaul Mundt NULL); 15401da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 15411da177e4SLinus Torvalds return -ENOMEM; 15421da177e4SLinus Torvalds 154393d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 15441da177e4SLinus Torvalds nfs_wdata_cachep); 15451da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 15461da177e4SLinus Torvalds return -ENOMEM; 15471da177e4SLinus Torvalds 154893d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 15491da177e4SLinus Torvalds nfs_wdata_cachep); 15501da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 15511da177e4SLinus Torvalds return -ENOMEM; 15521da177e4SLinus Torvalds 155389a09141SPeter Zijlstra /* 155489a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 155589a09141SPeter Zijlstra * 155689a09141SPeter Zijlstra * 64MB: 8192k 155789a09141SPeter Zijlstra * 128MB: 11585k 155889a09141SPeter Zijlstra * 256MB: 16384k 155989a09141SPeter Zijlstra * 512MB: 23170k 156089a09141SPeter Zijlstra * 1GB: 32768k 156189a09141SPeter Zijlstra * 2GB: 46340k 156289a09141SPeter Zijlstra * 4GB: 65536k 156389a09141SPeter Zijlstra * 8GB: 92681k 156489a09141SPeter Zijlstra * 16GB: 131072k 156589a09141SPeter Zijlstra * 156689a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 156789a09141SPeter Zijlstra * Limit the default to 256M 156889a09141SPeter Zijlstra */ 156989a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 157089a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 157189a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 157289a09141SPeter Zijlstra 15731da177e4SLinus Torvalds return 0; 15741da177e4SLinus Torvalds } 15751da177e4SLinus Torvalds 1576266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 15771da177e4SLinus Torvalds { 15781da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 15791da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 15801a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 15811da177e4SLinus Torvalds } 15821da177e4SLinus Torvalds 1583