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> 16074cc1deSTrond Myklebust #include <linux/migrate.h> 171da177e4SLinus Torvalds 181da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h> 191da177e4SLinus Torvalds #include <linux/nfs_fs.h> 201da177e4SLinus Torvalds #include <linux/nfs_mount.h> 211da177e4SLinus Torvalds #include <linux/nfs_page.h> 223fcfab16SAndrew Morton #include <linux/backing-dev.h> 23afeacc8cSPaul Gortmaker #include <linux/export.h> 243fcfab16SAndrew Morton 251da177e4SLinus Torvalds #include <asm/uaccess.h> 261da177e4SLinus Torvalds 271da177e4SLinus Torvalds #include "delegation.h" 2849a70f27STrond Myklebust #include "internal.h" 2991d5b470SChuck Lever #include "iostat.h" 30def6ed7eSAndy Adamson #include "nfs4_fs.h" 31074cc1deSTrond Myklebust #include "fscache.h" 3294ad1c80SFred Isaman #include "pnfs.h" 331da177e4SLinus Torvalds 341da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 351da177e4SLinus Torvalds 361da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 371da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 381da177e4SLinus Torvalds 391da177e4SLinus Torvalds /* 401da177e4SLinus Torvalds * Local function declarations 411da177e4SLinus Torvalds */ 42c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc, 43c63c7b05STrond Myklebust struct inode *inode, int ioflags); 44f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 45788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops; 46788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops; 47788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 481da177e4SLinus Torvalds 49e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 503feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 511da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 521da177e4SLinus Torvalds 53c9d8f89dSTrond Myklebust struct nfs_write_data *nfs_commitdata_alloc(void) 541da177e4SLinus Torvalds { 55e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS); 5640859d7eSChuck Lever 571da177e4SLinus Torvalds if (p) { 581da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 591da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 601da177e4SLinus Torvalds } 611da177e4SLinus Torvalds return p; 621da177e4SLinus Torvalds } 63e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 641da177e4SLinus Torvalds 655e4424afSTrond Myklebust void nfs_commit_free(struct nfs_write_data *p) 661da177e4SLinus Torvalds { 6740859d7eSChuck Lever if (p && (p->pagevec != &p->page_array[0])) 6840859d7eSChuck Lever kfree(p->pagevec); 691da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 701da177e4SLinus Torvalds } 71e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 721da177e4SLinus Torvalds 738d5658c9STrond Myklebust struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount) 743feb2d49STrond Myklebust { 75e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); 763feb2d49STrond Myklebust 773feb2d49STrond Myklebust if (p) { 783feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 793feb2d49STrond Myklebust INIT_LIST_HEAD(&p->pages); 80e9f7bee1STrond Myklebust p->npages = pagecount; 810d0b5cb3SChuck Lever if (pagecount <= ARRAY_SIZE(p->page_array)) 820d0b5cb3SChuck Lever p->pagevec = p->page_array; 833feb2d49STrond Myklebust else { 840d0b5cb3SChuck Lever p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); 850d0b5cb3SChuck Lever if (!p->pagevec) { 863feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 873feb2d49STrond Myklebust p = NULL; 883feb2d49STrond Myklebust } 893feb2d49STrond Myklebust } 903feb2d49STrond Myklebust } 913feb2d49STrond Myklebust return p; 923feb2d49STrond Myklebust } 933feb2d49STrond Myklebust 941ae88b2eSTrond Myklebust void nfs_writedata_free(struct nfs_write_data *p) 953feb2d49STrond Myklebust { 963feb2d49STrond Myklebust if (p && (p->pagevec != &p->page_array[0])) 973feb2d49STrond Myklebust kfree(p->pagevec); 983feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 993feb2d49STrond Myklebust } 1003feb2d49STrond Myklebust 101dce81290STrond Myklebust void nfs_writedata_release(struct nfs_write_data *wdata) 1021da177e4SLinus Torvalds { 1035053aa56SFred Isaman put_lseg(wdata->lseg); 104383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 1051da177e4SLinus Torvalds nfs_writedata_free(wdata); 1061da177e4SLinus Torvalds } 1071da177e4SLinus Torvalds 1087b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 1097b159fc1STrond Myklebust { 1107b159fc1STrond Myklebust ctx->error = error; 1117b159fc1STrond Myklebust smp_wmb(); 1127b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 1137b159fc1STrond Myklebust } 1147b159fc1STrond Myklebust 115277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request_locked(struct page *page) 116277459d2STrond Myklebust { 117277459d2STrond Myklebust struct nfs_page *req = NULL; 118277459d2STrond Myklebust 119277459d2STrond Myklebust if (PagePrivate(page)) { 120277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 121277459d2STrond Myklebust if (req != NULL) 122c03b4024STrond Myklebust kref_get(&req->wb_kref); 123277459d2STrond Myklebust } 124277459d2STrond Myklebust return req; 125277459d2STrond Myklebust } 126277459d2STrond Myklebust 127277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request(struct page *page) 128277459d2STrond Myklebust { 129587142f8STrond Myklebust struct inode *inode = page->mapping->host; 130277459d2STrond Myklebust struct nfs_page *req = NULL; 131277459d2STrond Myklebust 132587142f8STrond Myklebust spin_lock(&inode->i_lock); 133277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 134587142f8STrond Myklebust spin_unlock(&inode->i_lock); 135277459d2STrond Myklebust return req; 136277459d2STrond Myklebust } 137277459d2STrond Myklebust 1381da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1391da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1401da177e4SLinus Torvalds { 1411da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 142a3d01454STrond Myklebust loff_t end, i_size; 143a3d01454STrond Myklebust pgoff_t end_index; 1441da177e4SLinus Torvalds 145a3d01454STrond Myklebust spin_lock(&inode->i_lock); 146a3d01454STrond Myklebust i_size = i_size_read(inode); 147a3d01454STrond Myklebust end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 1481da177e4SLinus Torvalds if (i_size > 0 && page->index < end_index) 149a3d01454STrond Myklebust goto out; 1501da177e4SLinus Torvalds end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); 1511da177e4SLinus Torvalds if (i_size >= end) 152a3d01454STrond Myklebust goto out; 1531da177e4SLinus Torvalds i_size_write(inode, end); 154a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 155a3d01454STrond Myklebust out: 156a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 1571da177e4SLinus Torvalds } 1581da177e4SLinus Torvalds 159a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 160a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 161a301b777STrond Myklebust { 162a301b777STrond Myklebust SetPageError(page); 163a301b777STrond Myklebust nfs_zap_mapping(page->mapping->host, page->mapping); 164a301b777STrond Myklebust } 165a301b777STrond Myklebust 1661da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 1671da177e4SLinus Torvalds * covers the full page. 1681da177e4SLinus Torvalds */ 1691da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 1701da177e4SLinus Torvalds { 1711da177e4SLinus Torvalds if (PageUptodate(page)) 1721da177e4SLinus Torvalds return; 1731da177e4SLinus Torvalds if (base != 0) 1741da177e4SLinus Torvalds return; 17549a70f27STrond Myklebust if (count != nfs_page_length(page)) 1761da177e4SLinus Torvalds return; 1771da177e4SLinus Torvalds SetPageUptodate(page); 1781da177e4SLinus Torvalds } 1791da177e4SLinus Torvalds 1801da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 1811da177e4SLinus Torvalds { 1821da177e4SLinus Torvalds if (wbc->for_reclaim) 183c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 184b17621feSWu Fengguang if (wbc->for_kupdate || wbc->for_background) 185b31268acSTrond Myklebust return FLUSH_LOWPRI | FLUSH_COND_STABLE; 186b31268acSTrond Myklebust return FLUSH_COND_STABLE; 1871da177e4SLinus Torvalds } 1881da177e4SLinus Torvalds 1891da177e4SLinus Torvalds /* 19089a09141SPeter Zijlstra * NFS congestion control 19189a09141SPeter Zijlstra */ 19289a09141SPeter Zijlstra 19389a09141SPeter Zijlstra int nfs_congestion_kb; 19489a09141SPeter Zijlstra 19589a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 19689a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 19789a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 19889a09141SPeter Zijlstra 1995a6d41b3STrond Myklebust static int nfs_set_page_writeback(struct page *page) 20089a09141SPeter Zijlstra { 2015a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2025a6d41b3STrond Myklebust 2035a6d41b3STrond Myklebust if (!ret) { 20489a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 20589a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 20689a09141SPeter Zijlstra 207a6305ddbSTrond Myklebust page_cache_get(page); 208277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 2098aa7e847SJens Axboe NFS_CONGESTION_ON_THRESH) { 2108aa7e847SJens Axboe set_bdi_congested(&nfss->backing_dev_info, 2118aa7e847SJens Axboe BLK_RW_ASYNC); 2128aa7e847SJens Axboe } 21389a09141SPeter Zijlstra } 2145a6d41b3STrond Myklebust return ret; 21589a09141SPeter Zijlstra } 21689a09141SPeter Zijlstra 21789a09141SPeter Zijlstra static void nfs_end_page_writeback(struct page *page) 21889a09141SPeter Zijlstra { 21989a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 22089a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 22189a09141SPeter Zijlstra 22289a09141SPeter Zijlstra end_page_writeback(page); 223a6305ddbSTrond Myklebust page_cache_release(page); 224c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 2258aa7e847SJens Axboe clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 22689a09141SPeter Zijlstra } 22789a09141SPeter Zijlstra 228cfb506e1STrond Myklebust static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) 229e261f51fSTrond Myklebust { 230587142f8STrond Myklebust struct inode *inode = page->mapping->host; 231e261f51fSTrond Myklebust struct nfs_page *req; 232e261f51fSTrond Myklebust int ret; 233e261f51fSTrond Myklebust 234587142f8STrond Myklebust spin_lock(&inode->i_lock); 235e261f51fSTrond Myklebust for (;;) { 236e261f51fSTrond Myklebust req = nfs_page_find_request_locked(page); 237074cc1deSTrond Myklebust if (req == NULL) 238074cc1deSTrond Myklebust break; 2399994b62bSFred Isaman if (nfs_lock_request_dontget(req)) 240e261f51fSTrond Myklebust break; 241e261f51fSTrond Myklebust /* Note: If we hold the page lock, as is the case in nfs_writepage, 2429994b62bSFred Isaman * then the call to nfs_lock_request_dontget() will always 243e261f51fSTrond Myklebust * succeed provided that someone hasn't already marked the 244e261f51fSTrond Myklebust * request as dirty (in which case we don't care). 245e261f51fSTrond Myklebust */ 246587142f8STrond Myklebust spin_unlock(&inode->i_lock); 247cfb506e1STrond Myklebust if (!nonblock) 248e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 249cfb506e1STrond Myklebust else 250cfb506e1STrond Myklebust ret = -EAGAIN; 251e261f51fSTrond Myklebust nfs_release_request(req); 252e261f51fSTrond Myklebust if (ret != 0) 253074cc1deSTrond Myklebust return ERR_PTR(ret); 254587142f8STrond Myklebust spin_lock(&inode->i_lock); 255e261f51fSTrond Myklebust } 256587142f8STrond Myklebust spin_unlock(&inode->i_lock); 257074cc1deSTrond Myklebust return req; 258612c9384STrond Myklebust } 259074cc1deSTrond Myklebust 260074cc1deSTrond Myklebust /* 261074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 262074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 263074cc1deSTrond Myklebust */ 264074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 265cfb506e1STrond Myklebust struct page *page, bool nonblock) 266074cc1deSTrond Myklebust { 267074cc1deSTrond Myklebust struct nfs_page *req; 268074cc1deSTrond Myklebust int ret = 0; 269074cc1deSTrond Myklebust 270cfb506e1STrond Myklebust req = nfs_find_and_lock_request(page, nonblock); 271074cc1deSTrond Myklebust if (!req) 272074cc1deSTrond Myklebust goto out; 273074cc1deSTrond Myklebust ret = PTR_ERR(req); 274074cc1deSTrond Myklebust if (IS_ERR(req)) 275074cc1deSTrond Myklebust goto out; 276074cc1deSTrond Myklebust 277074cc1deSTrond Myklebust ret = nfs_set_page_writeback(page); 278074cc1deSTrond Myklebust BUG_ON(ret != 0); 279074cc1deSTrond Myklebust BUG_ON(test_bit(PG_CLEAN, &req->wb_flags)); 280074cc1deSTrond Myklebust 281f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 282f8512ad0SFred Isaman nfs_redirty_request(req); 283074cc1deSTrond Myklebust ret = pgio->pg_error; 284f8512ad0SFred Isaman } 285074cc1deSTrond Myklebust out: 286074cc1deSTrond Myklebust return ret; 287e261f51fSTrond Myklebust } 288e261f51fSTrond Myklebust 289f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 290f758c885STrond Myklebust { 291f758c885STrond Myklebust struct inode *inode = page->mapping->host; 292cfb506e1STrond Myklebust int ret; 293f758c885STrond Myklebust 294f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 295f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 296f758c885STrond Myklebust 297f758c885STrond Myklebust nfs_pageio_cond_complete(pgio, page->index); 2981b430beeSWu Fengguang ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); 299cfb506e1STrond Myklebust if (ret == -EAGAIN) { 300cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 301cfb506e1STrond Myklebust ret = 0; 302cfb506e1STrond Myklebust } 303cfb506e1STrond Myklebust return ret; 304f758c885STrond Myklebust } 305f758c885STrond Myklebust 306e261f51fSTrond Myklebust /* 3071da177e4SLinus Torvalds * Write an mmapped page to the server. 3081da177e4SLinus Torvalds */ 3094d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 3101da177e4SLinus Torvalds { 311f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 312e261f51fSTrond Myklebust int err; 3131da177e4SLinus Torvalds 314f758c885STrond Myklebust nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc)); 315f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 316f758c885STrond Myklebust nfs_pageio_complete(&pgio); 317f758c885STrond Myklebust if (err < 0) 3184d770ccfSTrond Myklebust return err; 319f758c885STrond Myklebust if (pgio.pg_error < 0) 320f758c885STrond Myklebust return pgio.pg_error; 321f758c885STrond Myklebust return 0; 3224d770ccfSTrond Myklebust } 3234d770ccfSTrond Myklebust 3244d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 3254d770ccfSTrond Myklebust { 326f758c885STrond Myklebust int ret; 3274d770ccfSTrond Myklebust 328f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 3291da177e4SLinus Torvalds unlock_page(page); 330f758c885STrond Myklebust return ret; 331f758c885STrond Myklebust } 332f758c885STrond Myklebust 333f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 334f758c885STrond Myklebust { 335f758c885STrond Myklebust int ret; 336f758c885STrond Myklebust 337f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 338f758c885STrond Myklebust unlock_page(page); 339f758c885STrond Myklebust return ret; 3401da177e4SLinus Torvalds } 3411da177e4SLinus Torvalds 3421da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 3431da177e4SLinus Torvalds { 3441da177e4SLinus Torvalds struct inode *inode = mapping->host; 34572cb77f4STrond Myklebust unsigned long *bitlock = &NFS_I(inode)->flags; 346c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 3471da177e4SLinus Torvalds int err; 3481da177e4SLinus Torvalds 34972cb77f4STrond Myklebust /* Stop dirtying of new pages while we sync */ 35072cb77f4STrond Myklebust err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, 35172cb77f4STrond Myklebust nfs_wait_bit_killable, TASK_KILLABLE); 35272cb77f4STrond Myklebust if (err) 35372cb77f4STrond Myklebust goto out_err; 35472cb77f4STrond Myklebust 35591d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 35691d5b470SChuck Lever 357c63c7b05STrond Myklebust nfs_pageio_init_write(&pgio, inode, wb_priority(wbc)); 358f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 359c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 36072cb77f4STrond Myklebust 36172cb77f4STrond Myklebust clear_bit_unlock(NFS_INO_FLUSHING, bitlock); 36272cb77f4STrond Myklebust smp_mb__after_clear_bit(); 36372cb77f4STrond Myklebust wake_up_bit(bitlock, NFS_INO_FLUSHING); 36472cb77f4STrond Myklebust 365f758c885STrond Myklebust if (err < 0) 36672cb77f4STrond Myklebust goto out_err; 36772cb77f4STrond Myklebust err = pgio.pg_error; 36872cb77f4STrond Myklebust if (err < 0) 36972cb77f4STrond Myklebust goto out_err; 370c63c7b05STrond Myklebust return 0; 37172cb77f4STrond Myklebust out_err: 37272cb77f4STrond Myklebust return err; 3731da177e4SLinus Torvalds } 3741da177e4SLinus Torvalds 3751da177e4SLinus Torvalds /* 3761da177e4SLinus Torvalds * Insert a write request into an inode 3771da177e4SLinus Torvalds */ 378*d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 3791da177e4SLinus Torvalds { 3801da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 381e7d39069STrond Myklebust 382e7d39069STrond Myklebust /* Lock the request! */ 383e7d39069STrond Myklebust nfs_lock_request_dontget(req); 384e7d39069STrond Myklebust 385e7d39069STrond Myklebust spin_lock(&inode->i_lock); 3864d65c520STrond Myklebust if (!nfsi->npages && nfs_have_delegation(inode, FMODE_WRITE)) 387a9a4a87aSTrond Myklebust inode->i_version++; 3882df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 389deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 390277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 3911da177e4SLinus Torvalds nfsi->npages++; 392c03b4024STrond Myklebust kref_get(&req->wb_kref); 393e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 3941da177e4SLinus Torvalds } 3951da177e4SLinus Torvalds 396*d6d6dc7cSFred Isaman static struct pnfs_layout_segment *nfs_clear_request_commit(struct nfs_page *req); 397*d6d6dc7cSFred Isaman 3981da177e4SLinus Torvalds /* 39989a09141SPeter Zijlstra * Remove a write request from an inode 4001da177e4SLinus Torvalds */ 4011da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 4021da177e4SLinus Torvalds { 4033d4ff43dSAl Viro struct inode *inode = req->wb_context->dentry->d_inode; 4041da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 405*d6d6dc7cSFred Isaman struct pnfs_layout_segment *lseg; 4061da177e4SLinus Torvalds 4071da177e4SLinus Torvalds BUG_ON (!NFS_WBACK_BUSY(req)); 4081da177e4SLinus Torvalds 409587142f8STrond Myklebust spin_lock(&inode->i_lock); 410*d6d6dc7cSFred Isaman lseg = nfs_clear_request_commit(req); 411277459d2STrond Myklebust set_page_private(req->wb_page, 0); 412deb7d638STrond Myklebust ClearPagePrivate(req->wb_page); 4132df485a7STrond Myklebust clear_bit(PG_MAPPED, &req->wb_flags); 4141da177e4SLinus Torvalds nfsi->npages--; 415587142f8STrond Myklebust spin_unlock(&inode->i_lock); 416*d6d6dc7cSFred Isaman put_lseg(lseg); 4171da177e4SLinus Torvalds nfs_release_request(req); 4181da177e4SLinus Torvalds } 4191da177e4SLinus Torvalds 42061822ab5STrond Myklebust static void 4216d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 42261822ab5STrond Myklebust { 42361822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 42461822ab5STrond Myklebust } 42561822ab5STrond Myklebust 4261da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 4271da177e4SLinus Torvalds /* 4281da177e4SLinus Torvalds * Add a request to the inode's commit list. 4291da177e4SLinus Torvalds */ 4301da177e4SLinus Torvalds static void 431a861a1e1SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg) 4321da177e4SLinus Torvalds { 4333d4ff43dSAl Viro struct inode *inode = req->wb_context->dentry->d_inode; 4341da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 435*d6d6dc7cSFred Isaman struct list_head *clist; 4361da177e4SLinus Torvalds 437*d6d6dc7cSFred Isaman clist = pnfs_choose_commit_list(req, lseg); 438587142f8STrond Myklebust spin_lock(&inode->i_lock); 439e468bae9STrond Myklebust set_bit(PG_CLEAN, &(req)->wb_flags); 440*d6d6dc7cSFred Isaman nfs_list_add_request(req, clist); 441ff778d02STrond Myklebust nfsi->ncommit++; 442587142f8STrond Myklebust spin_unlock(&inode->i_lock); 443fd39fc85SChristoph Lameter inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 444c9e51e41SPeter Zijlstra inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); 445a1803044STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 4461da177e4SLinus Torvalds } 4478e821cadSTrond Myklebust 448*d6d6dc7cSFred Isaman static void 449*d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 450e468bae9STrond Myklebust { 451e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 452e468bae9STrond Myklebust dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE); 453e468bae9STrond Myklebust } 454*d6d6dc7cSFred Isaman 455*d6d6dc7cSFred Isaman static struct pnfs_layout_segment * 456*d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 457*d6d6dc7cSFred Isaman { 458*d6d6dc7cSFred Isaman struct pnfs_layout_segment *lseg = NULL; 459*d6d6dc7cSFred Isaman 460*d6d6dc7cSFred Isaman if (test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) { 461*d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 462*d6d6dc7cSFred Isaman lseg = pnfs_clear_request_commit(req); 463*d6d6dc7cSFred Isaman NFS_I(req->wb_context->dentry->d_inode)->ncommit--; 464*d6d6dc7cSFred Isaman list_del(&req->wb_list); 465*d6d6dc7cSFred Isaman } 466*d6d6dc7cSFred Isaman return lseg; 467e468bae9STrond Myklebust } 468e468bae9STrond Myklebust 4698e821cadSTrond Myklebust static inline 4708e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4718e821cadSTrond Myklebust { 472465d5243SFred Isaman if (data->verf.committed == NFS_DATA_SYNC) 473465d5243SFred Isaman return data->lseg == NULL; 474465d5243SFred Isaman else 4758e821cadSTrond Myklebust return data->verf.committed != NFS_FILE_SYNC; 4768e821cadSTrond Myklebust } 4778e821cadSTrond Myklebust 4788e821cadSTrond Myklebust static inline 479a861a1e1SFred Isaman int nfs_reschedule_unstable_write(struct nfs_page *req, 480a861a1e1SFred Isaman struct nfs_write_data *data) 4818e821cadSTrond Myklebust { 482e468bae9STrond Myklebust if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) { 483a861a1e1SFred Isaman nfs_mark_request_commit(req, data->lseg); 4848e821cadSTrond Myklebust return 1; 4858e821cadSTrond Myklebust } 4868e821cadSTrond Myklebust if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) { 4876d884e8fSFred nfs_mark_request_dirty(req); 4888e821cadSTrond Myklebust return 1; 4898e821cadSTrond Myklebust } 4908e821cadSTrond Myklebust return 0; 4918e821cadSTrond Myklebust } 4928e821cadSTrond Myklebust #else 4938e821cadSTrond Myklebust static inline void 494a861a1e1SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg) 4958e821cadSTrond Myklebust { 4968e821cadSTrond Myklebust } 4978e821cadSTrond Myklebust 498*d6d6dc7cSFred Isaman static inline struct pnfs_layout_segment * 499e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 500e468bae9STrond Myklebust { 501*d6d6dc7cSFred Isaman return NULL; 502e468bae9STrond Myklebust } 503e468bae9STrond Myklebust 5048e821cadSTrond Myklebust static inline 5058e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 5068e821cadSTrond Myklebust { 5078e821cadSTrond Myklebust return 0; 5088e821cadSTrond Myklebust } 5098e821cadSTrond Myklebust 5108e821cadSTrond Myklebust static inline 511a861a1e1SFred Isaman int nfs_reschedule_unstable_write(struct nfs_page *req, 512a861a1e1SFred Isaman struct nfs_write_data *data) 5138e821cadSTrond Myklebust { 5148e821cadSTrond Myklebust return 0; 5158e821cadSTrond Myklebust } 5161da177e4SLinus Torvalds #endif 5171da177e4SLinus Torvalds 51847c62564STrond Myklebust #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 519fb8a1f11STrond Myklebust static int 520fb8a1f11STrond Myklebust nfs_need_commit(struct nfs_inode *nfsi) 521fb8a1f11STrond Myklebust { 522*d6d6dc7cSFred Isaman return nfsi->ncommit > 0; 523fb8a1f11STrond Myklebust } 524fb8a1f11STrond Myklebust 525*d6d6dc7cSFred Isaman /* i_lock held by caller */ 526*d6d6dc7cSFred Isaman int 527*d6d6dc7cSFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, int max) 528*d6d6dc7cSFred Isaman { 529*d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 530*d6d6dc7cSFred Isaman int ret = 0; 531*d6d6dc7cSFred Isaman 532*d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 533*d6d6dc7cSFred Isaman if (nfs_lock_request_dontget(req)) { 534*d6d6dc7cSFred Isaman kref_get(&req->wb_kref); 535*d6d6dc7cSFred Isaman list_move_tail(&req->wb_list, dst); 536*d6d6dc7cSFred Isaman clear_bit(PG_CLEAN, &(req)->wb_flags); 537*d6d6dc7cSFred Isaman ret++; 538*d6d6dc7cSFred Isaman if (ret == max) 539*d6d6dc7cSFred Isaman break; 540*d6d6dc7cSFred Isaman } 541*d6d6dc7cSFred Isaman } 542*d6d6dc7cSFred Isaman return ret; 543*d6d6dc7cSFred Isaman } 544*d6d6dc7cSFred Isaman EXPORT_SYMBOL_GPL(nfs_scan_commit_list); 545*d6d6dc7cSFred Isaman 5461da177e4SLinus Torvalds /* 5471da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 5481da177e4SLinus Torvalds * @inode: NFS inode to scan 5491da177e4SLinus Torvalds * @dst: destination list 5501da177e4SLinus Torvalds * 5511da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 5521da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 5531da177e4SLinus Torvalds */ 5541da177e4SLinus Torvalds static int 555*d6d6dc7cSFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst) 5561da177e4SLinus Torvalds { 5571da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 558*d6d6dc7cSFred Isaman int ret = 0; 559fb8a1f11STrond Myklebust 5600d88f6e8SDave Chinner spin_lock(&inode->i_lock); 561*d6d6dc7cSFred Isaman if (nfsi->ncommit > 0) { 562*d6d6dc7cSFred Isaman int pnfs_ret; 563*d6d6dc7cSFred Isaman 564*d6d6dc7cSFred Isaman ret = nfs_scan_commit_list(&nfsi->commit_list, dst, INT_MAX); 565*d6d6dc7cSFred Isaman pnfs_ret = pnfs_scan_commit_lists(inode, INT_MAX - ret); 566*d6d6dc7cSFred Isaman if (pnfs_ret) { 567*d6d6dc7cSFred Isaman ret += pnfs_ret; 568*d6d6dc7cSFred Isaman set_bit(NFS_INO_PNFS_COMMIT, &nfsi->flags); 569*d6d6dc7cSFred Isaman } 570ff778d02STrond Myklebust nfsi->ncommit -= ret; 571*d6d6dc7cSFred Isaman } 5720d88f6e8SDave Chinner spin_unlock(&inode->i_lock); 573ff778d02STrond Myklebust return ret; 5741da177e4SLinus Torvalds } 575*d6d6dc7cSFred Isaman 576c42de9ddSTrond Myklebust #else 577fb8a1f11STrond Myklebust static inline int nfs_need_commit(struct nfs_inode *nfsi) 578fb8a1f11STrond Myklebust { 579fb8a1f11STrond Myklebust return 0; 580fb8a1f11STrond Myklebust } 581fb8a1f11STrond Myklebust 582*d6d6dc7cSFred Isaman static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst) 583c42de9ddSTrond Myklebust { 584c42de9ddSTrond Myklebust return 0; 585c42de9ddSTrond Myklebust } 5861da177e4SLinus Torvalds #endif 5871da177e4SLinus Torvalds 5881da177e4SLinus Torvalds /* 589e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 590e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 5911da177e4SLinus Torvalds * 592e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 593e7d39069STrond Myklebust * to disk. 5941da177e4SLinus Torvalds */ 595e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 596e7d39069STrond Myklebust struct page *page, 597e7d39069STrond Myklebust unsigned int offset, 598e7d39069STrond Myklebust unsigned int bytes) 5991da177e4SLinus Torvalds { 600e7d39069STrond Myklebust struct nfs_page *req; 601e7d39069STrond Myklebust unsigned int rqend; 602e7d39069STrond Myklebust unsigned int end; 6031da177e4SLinus Torvalds int error; 604*d6d6dc7cSFred Isaman struct pnfs_layout_segment *lseg = NULL; 605277459d2STrond Myklebust 606e7d39069STrond Myklebust if (!PagePrivate(page)) 607e7d39069STrond Myklebust return NULL; 608e7d39069STrond Myklebust 609e7d39069STrond Myklebust end = offset + bytes; 610e7d39069STrond Myklebust spin_lock(&inode->i_lock); 611e7d39069STrond Myklebust 612e7d39069STrond Myklebust for (;;) { 613e7d39069STrond Myklebust req = nfs_page_find_request_locked(page); 614e7d39069STrond Myklebust if (req == NULL) 615e7d39069STrond Myklebust goto out_unlock; 616e7d39069STrond Myklebust 617e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 618e7d39069STrond Myklebust /* 619e7d39069STrond Myklebust * Tell the caller to flush out the request if 620e7d39069STrond Myklebust * the offsets are non-contiguous. 621e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 622e7d39069STrond Myklebust * have flushed out requests having wrong owners. 623e7d39069STrond Myklebust */ 624e468bae9STrond Myklebust if (offset > rqend 625e7d39069STrond Myklebust || end < req->wb_offset) 626e7d39069STrond Myklebust goto out_flushme; 627e7d39069STrond Myklebust 6289994b62bSFred Isaman if (nfs_lock_request_dontget(req)) 629e7d39069STrond Myklebust break; 630e7d39069STrond Myklebust 631e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 632587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6331da177e4SLinus Torvalds error = nfs_wait_on_request(req); 6341da177e4SLinus Torvalds nfs_release_request(req); 635e7d39069STrond Myklebust if (error != 0) 636e7d39069STrond Myklebust goto out_err; 637e7d39069STrond Myklebust spin_lock(&inode->i_lock); 6381da177e4SLinus Torvalds } 6391da177e4SLinus Torvalds 640*d6d6dc7cSFred Isaman lseg = nfs_clear_request_commit(req); 641e468bae9STrond Myklebust 6421da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 6431da177e4SLinus Torvalds if (offset < req->wb_offset) { 6441da177e4SLinus Torvalds req->wb_offset = offset; 6451da177e4SLinus Torvalds req->wb_pgbase = offset; 6461da177e4SLinus Torvalds } 6471da177e4SLinus Torvalds if (end > rqend) 6481da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 649e7d39069STrond Myklebust else 650e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 651e7d39069STrond Myklebust out_unlock: 652e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 653*d6d6dc7cSFred Isaman put_lseg(lseg); 654e7d39069STrond Myklebust return req; 655e7d39069STrond Myklebust out_flushme: 656e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 657e7d39069STrond Myklebust nfs_release_request(req); 658e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 659e7d39069STrond Myklebust out_err: 660e7d39069STrond Myklebust return ERR_PTR(error); 661e7d39069STrond Myklebust } 6621da177e4SLinus Torvalds 663e7d39069STrond Myklebust /* 664e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 665e7d39069STrond Myklebust * 666e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 667e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 668e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 669e7d39069STrond Myklebust */ 670e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 671e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 672e7d39069STrond Myklebust { 673e7d39069STrond Myklebust struct inode *inode = page->mapping->host; 674e7d39069STrond Myklebust struct nfs_page *req; 675e7d39069STrond Myklebust 676e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 677e7d39069STrond Myklebust if (req != NULL) 678e7d39069STrond Myklebust goto out; 679e7d39069STrond Myklebust req = nfs_create_request(ctx, inode, page, offset, bytes); 680e7d39069STrond Myklebust if (IS_ERR(req)) 681e7d39069STrond Myklebust goto out; 682*d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 683efc91ed0STrond Myklebust out: 68461e930a9STrond Myklebust return req; 6851da177e4SLinus Torvalds } 6861da177e4SLinus Torvalds 687e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 688e7d39069STrond Myklebust unsigned int offset, unsigned int count) 689e7d39069STrond Myklebust { 690e7d39069STrond Myklebust struct nfs_page *req; 691e7d39069STrond Myklebust 692e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 693e7d39069STrond Myklebust if (IS_ERR(req)) 694e7d39069STrond Myklebust return PTR_ERR(req); 695e7d39069STrond Myklebust /* Update file length */ 696e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 697e7d39069STrond Myklebust nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 698a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 6999994b62bSFred Isaman nfs_unlock_request(req); 700e7d39069STrond Myklebust return 0; 701e7d39069STrond Myklebust } 702e7d39069STrond Myklebust 7031da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 7041da177e4SLinus Torvalds { 705cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 7061da177e4SLinus Torvalds struct nfs_page *req; 7071a54533eSTrond Myklebust int do_flush, status; 7081da177e4SLinus Torvalds /* 7091da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 7101da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 7111da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 7121da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 7131da177e4SLinus Torvalds * Also do the same if we find a request from an existing 7141da177e4SLinus Torvalds * dropped page. 7151da177e4SLinus Torvalds */ 7161a54533eSTrond Myklebust do { 717277459d2STrond Myklebust req = nfs_page_find_request(page); 7181a54533eSTrond Myklebust if (req == NULL) 7191a54533eSTrond Myklebust return 0; 720f11ac8dbSTrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx || 721f11ac8dbSTrond Myklebust req->wb_lock_context->lockowner != current->files || 722f11ac8dbSTrond Myklebust req->wb_lock_context->pid != current->tgid; 7231da177e4SLinus Torvalds nfs_release_request(req); 7241a54533eSTrond Myklebust if (!do_flush) 7251a54533eSTrond Myklebust return 0; 726277459d2STrond Myklebust status = nfs_wb_page(page->mapping->host, page); 7271a54533eSTrond Myklebust } while (status == 0); 7281a54533eSTrond Myklebust return status; 7291da177e4SLinus Torvalds } 7301da177e4SLinus Torvalds 7311da177e4SLinus Torvalds /* 7325d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 7335d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 7345d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 7355d47a356STrond Myklebust */ 7365d47a356STrond Myklebust static int nfs_write_pageuptodate(struct page *page, struct inode *inode) 7375d47a356STrond Myklebust { 7385d47a356STrond Myklebust return PageUptodate(page) && 7395d47a356STrond Myklebust !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); 7405d47a356STrond Myklebust } 7415d47a356STrond Myklebust 7425d47a356STrond Myklebust /* 7431da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 7441da177e4SLinus Torvalds * 7451da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 7461da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 7471da177e4SLinus Torvalds */ 7481da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 7491da177e4SLinus Torvalds unsigned int offset, unsigned int count) 7501da177e4SLinus Torvalds { 751cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 7521da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 7531da177e4SLinus Torvalds int status = 0; 7541da177e4SLinus Torvalds 75591d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 75691d5b470SChuck Lever 75748186c7dSChuck Lever dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", 75801cce933SJosef "Jeff" Sipek file->f_path.dentry->d_parent->d_name.name, 75901cce933SJosef "Jeff" Sipek file->f_path.dentry->d_name.name, count, 7600bbacc40SChuck Lever (long long)(page_offset(page) + offset)); 7611da177e4SLinus Torvalds 7621da177e4SLinus Torvalds /* If we're not using byte range locks, and we know the page 7635d47a356STrond Myklebust * is up to date, it may be more efficient to extend the write 7645d47a356STrond Myklebust * to cover the entire page in order to avoid fragmentation 7655d47a356STrond Myklebust * inefficiencies. 7661da177e4SLinus Torvalds */ 7675d47a356STrond Myklebust if (nfs_write_pageuptodate(page, inode) && 7685d47a356STrond Myklebust inode->i_flock == NULL && 7696b2f3d1fSChristoph Hellwig !(file->f_flags & O_DSYNC)) { 77049a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 7711da177e4SLinus Torvalds offset = 0; 7721da177e4SLinus Torvalds } 7731da177e4SLinus Torvalds 774e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 77503fa9e84STrond Myklebust if (status < 0) 77603fa9e84STrond Myklebust nfs_set_pageerror(page); 77759b7c05fSTrond Myklebust else 77859b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 7791da177e4SLinus Torvalds 78048186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 7811da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 7821da177e4SLinus Torvalds return status; 7831da177e4SLinus Torvalds } 7841da177e4SLinus Torvalds 785a861a1e1SFred Isaman static void nfs_writepage_release(struct nfs_page *req, 786a861a1e1SFred Isaman struct nfs_write_data *data) 7871da177e4SLinus Torvalds { 788a6305ddbSTrond Myklebust struct page *page = req->wb_page; 7898e821cadSTrond Myklebust 790a861a1e1SFred Isaman if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req, data)) 7911da177e4SLinus Torvalds nfs_inode_remove_request(req); 7929994b62bSFred Isaman nfs_unlock_request(req); 793a6305ddbSTrond Myklebust nfs_end_page_writeback(page); 7941da177e4SLinus Torvalds } 7951da177e4SLinus Torvalds 7963ff7576dSTrond Myklebust static int flush_task_priority(int how) 7971da177e4SLinus Torvalds { 7981da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 7991da177e4SLinus Torvalds case FLUSH_HIGHPRI: 8001da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 8011da177e4SLinus Torvalds case FLUSH_LOWPRI: 8021da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 8031da177e4SLinus Torvalds } 8041da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 8051da177e4SLinus Torvalds } 8061da177e4SLinus Torvalds 807a69aef14SFred Isaman int nfs_initiate_write(struct nfs_write_data *data, 808d138d5d1SAndy Adamson struct rpc_clnt *clnt, 809788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 8101da177e4SLinus Torvalds int how) 8111da177e4SLinus Torvalds { 812d138d5d1SAndy Adamson struct inode *inode = data->inode; 8133ff7576dSTrond Myklebust int priority = flush_task_priority(how); 81407737691STrond Myklebust struct rpc_task *task; 815bdc7f021STrond Myklebust struct rpc_message msg = { 816bdc7f021STrond Myklebust .rpc_argp = &data->args, 817bdc7f021STrond Myklebust .rpc_resp = &data->res, 818d138d5d1SAndy Adamson .rpc_cred = data->cred, 819bdc7f021STrond Myklebust }; 82084115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 821d138d5d1SAndy Adamson .rpc_client = clnt, 82207737691STrond Myklebust .task = &data->task, 823bdc7f021STrond Myklebust .rpc_message = &msg, 82484115e1cSTrond Myklebust .callback_ops = call_ops, 82584115e1cSTrond Myklebust .callback_data = data, 826101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 8272c61be0aSTrond Myklebust .flags = RPC_TASK_ASYNC, 8283ff7576dSTrond Myklebust .priority = priority, 82984115e1cSTrond Myklebust }; 8302c61be0aSTrond Myklebust int ret = 0; 8311da177e4SLinus Torvalds 832d138d5d1SAndy Adamson /* Set up the initial task struct. */ 833d138d5d1SAndy Adamson NFS_PROTO(inode)->write_setup(data, &msg); 834d138d5d1SAndy Adamson 835d138d5d1SAndy Adamson dprintk("NFS: %5u initiated write call " 836d138d5d1SAndy Adamson "(req %s/%lld, %u bytes @ offset %llu)\n", 837d138d5d1SAndy Adamson data->task.tk_pid, 838d138d5d1SAndy Adamson inode->i_sb->s_id, 839d138d5d1SAndy Adamson (long long)NFS_FILEID(inode), 840d138d5d1SAndy Adamson data->args.count, 841d138d5d1SAndy Adamson (unsigned long long)data->args.offset); 842d138d5d1SAndy Adamson 843d138d5d1SAndy Adamson task = rpc_run_task(&task_setup_data); 844d138d5d1SAndy Adamson if (IS_ERR(task)) { 845d138d5d1SAndy Adamson ret = PTR_ERR(task); 846d138d5d1SAndy Adamson goto out; 847d138d5d1SAndy Adamson } 848d138d5d1SAndy Adamson if (how & FLUSH_SYNC) { 849d138d5d1SAndy Adamson ret = rpc_wait_for_completion_task(task); 850d138d5d1SAndy Adamson if (ret == 0) 851d138d5d1SAndy Adamson ret = task->tk_status; 852d138d5d1SAndy Adamson } 853d138d5d1SAndy Adamson rpc_put_task(task); 854d138d5d1SAndy Adamson out: 855d138d5d1SAndy Adamson return ret; 856d138d5d1SAndy Adamson } 857a69aef14SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_write); 858d138d5d1SAndy Adamson 859d138d5d1SAndy Adamson /* 860d138d5d1SAndy Adamson * Set up the argument/result storage required for the RPC call. 861d138d5d1SAndy Adamson */ 8626e4efd56STrond Myklebust static void nfs_write_rpcsetup(struct nfs_page *req, 863d138d5d1SAndy Adamson struct nfs_write_data *data, 864d138d5d1SAndy Adamson unsigned int count, unsigned int offset, 865d138d5d1SAndy Adamson int how) 866d138d5d1SAndy Adamson { 8673d4ff43dSAl Viro struct inode *inode = req->wb_context->dentry->d_inode; 868d138d5d1SAndy Adamson 8691da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 8701da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 8711da177e4SLinus Torvalds 8721da177e4SLinus Torvalds data->req = req; 8733d4ff43dSAl Viro data->inode = inode = req->wb_context->dentry->d_inode; 874d138d5d1SAndy Adamson data->cred = req->wb_context->cred; 8751da177e4SLinus Torvalds 8761da177e4SLinus Torvalds data->args.fh = NFS_FH(inode); 8771da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 8782bea038cSBoaz Harrosh /* pnfs_set_layoutcommit needs this */ 8792bea038cSBoaz Harrosh data->mds_offset = data->args.offset; 8801da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 8811da177e4SLinus Torvalds data->args.pages = data->pagevec; 8821da177e4SLinus Torvalds data->args.count = count; 883383ba719STrond Myklebust data->args.context = get_nfs_open_context(req->wb_context); 884f11ac8dbSTrond Myklebust data->args.lock_context = req->wb_lock_context; 885bdc7f021STrond Myklebust data->args.stable = NFS_UNSTABLE; 88687ed5eb4STrond Myklebust switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { 88787ed5eb4STrond Myklebust case 0: 88887ed5eb4STrond Myklebust break; 88987ed5eb4STrond Myklebust case FLUSH_COND_STABLE: 89087ed5eb4STrond Myklebust if (nfs_need_commit(NFS_I(inode))) 89187ed5eb4STrond Myklebust break; 89287ed5eb4STrond Myklebust default: 893bdc7f021STrond Myklebust data->args.stable = NFS_FILE_SYNC; 894bdc7f021STrond Myklebust } 8951da177e4SLinus Torvalds 8961da177e4SLinus Torvalds data->res.fattr = &data->fattr; 8971da177e4SLinus Torvalds data->res.count = count; 8981da177e4SLinus Torvalds data->res.verf = &data->verf; 8990e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 9006e4efd56STrond Myklebust } 9011da177e4SLinus Torvalds 9026e4efd56STrond Myklebust static int nfs_do_write(struct nfs_write_data *data, 9036e4efd56STrond Myklebust const struct rpc_call_ops *call_ops, 9046e4efd56STrond Myklebust int how) 9056e4efd56STrond Myklebust { 9065f00bcb3SStephen Rothwell struct inode *inode = data->args.context->dentry->d_inode; 9070382b744SAndy Adamson 908d138d5d1SAndy Adamson return nfs_initiate_write(data, NFS_CLIENT(inode), call_ops, how); 9091da177e4SLinus Torvalds } 9101da177e4SLinus Torvalds 911275acaafSTrond Myklebust static int nfs_do_multiple_writes(struct list_head *head, 912275acaafSTrond Myklebust const struct rpc_call_ops *call_ops, 913275acaafSTrond Myklebust int how) 914275acaafSTrond Myklebust { 915275acaafSTrond Myklebust struct nfs_write_data *data; 916275acaafSTrond Myklebust int ret = 0; 917275acaafSTrond Myklebust 918275acaafSTrond Myklebust while (!list_empty(head)) { 919275acaafSTrond Myklebust int ret2; 920275acaafSTrond Myklebust 921275acaafSTrond Myklebust data = list_entry(head->next, struct nfs_write_data, list); 922275acaafSTrond Myklebust list_del_init(&data->list); 923275acaafSTrond Myklebust 924dce81290STrond Myklebust ret2 = nfs_do_write(data, call_ops, how); 925275acaafSTrond Myklebust if (ret == 0) 926275acaafSTrond Myklebust ret = ret2; 927275acaafSTrond Myklebust } 928275acaafSTrond Myklebust return ret; 929275acaafSTrond Myklebust } 930275acaafSTrond Myklebust 9316d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 9326d884e8fSFred * call this on each, which will prepare them to be retried on next 9336d884e8fSFred * writeback using standard nfs. 9346d884e8fSFred */ 9356d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 9366d884e8fSFred { 937a6305ddbSTrond Myklebust struct page *page = req->wb_page; 938a6305ddbSTrond Myklebust 9396d884e8fSFred nfs_mark_request_dirty(req); 9409994b62bSFred Isaman nfs_unlock_request(req); 941a6305ddbSTrond Myklebust nfs_end_page_writeback(page); 9426d884e8fSFred } 9436d884e8fSFred 9441da177e4SLinus Torvalds /* 9451da177e4SLinus Torvalds * Generate multiple small requests to write out a single 9461da177e4SLinus Torvalds * contiguous dirty area on one page. 9471da177e4SLinus Torvalds */ 948275acaafSTrond Myklebust static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, struct list_head *res) 9491da177e4SLinus Torvalds { 950c76069bdSFred Isaman struct nfs_page *req = nfs_list_entry(desc->pg_list.next); 9511da177e4SLinus Torvalds struct page *page = req->wb_page; 9521da177e4SLinus Torvalds struct nfs_write_data *data; 953d097971dSTrond Myklebust size_t wsize = desc->pg_bsize, nbytes; 954e9f7bee1STrond Myklebust unsigned int offset; 9551da177e4SLinus Torvalds int requests = 0; 956dbae4c73STrond Myklebust int ret = 0; 9571da177e4SLinus Torvalds 9581da177e4SLinus Torvalds nfs_list_remove_request(req); 9591da177e4SLinus Torvalds 960b31268acSTrond Myklebust if ((desc->pg_ioflags & FLUSH_COND_STABLE) && 961b31268acSTrond Myklebust (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit || 962b31268acSTrond Myklebust desc->pg_count > wsize)) 963b31268acSTrond Myklebust desc->pg_ioflags &= ~FLUSH_COND_STABLE; 964b31268acSTrond Myklebust 965b31268acSTrond Myklebust 966275acaafSTrond Myklebust offset = 0; 967c76069bdSFred Isaman nbytes = desc->pg_count; 968e9f7bee1STrond Myklebust do { 969e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 970e9f7bee1STrond Myklebust 9718d5658c9STrond Myklebust data = nfs_writedata_alloc(1); 9721da177e4SLinus Torvalds if (!data) 9731da177e4SLinus Torvalds goto out_bad; 974275acaafSTrond Myklebust data->pagevec[0] = page; 975f13c3620STrond Myklebust nfs_write_rpcsetup(req, data, len, offset, desc->pg_ioflags); 976275acaafSTrond Myklebust list_add(&data->list, res); 9771da177e4SLinus Torvalds requests++; 978e9f7bee1STrond Myklebust nbytes -= len; 979275acaafSTrond Myklebust offset += len; 980e9f7bee1STrond Myklebust } while (nbytes != 0); 9811da177e4SLinus Torvalds atomic_set(&req->wb_complete, requests); 98250828d7eSTrond Myklebust desc->pg_rpc_callops = &nfs_write_partial_ops; 983dbae4c73STrond Myklebust return ret; 9841da177e4SLinus Torvalds 9851da177e4SLinus Torvalds out_bad: 986275acaafSTrond Myklebust while (!list_empty(res)) { 987275acaafSTrond Myklebust data = list_entry(res->next, struct nfs_write_data, list); 9886e4efd56STrond Myklebust list_del(&data->list); 9890da2a4acSFred Isaman nfs_writedata_free(data); 9901da177e4SLinus Torvalds } 99161822ab5STrond Myklebust nfs_redirty_request(req); 9921da177e4SLinus Torvalds return -ENOMEM; 9931da177e4SLinus Torvalds } 9941da177e4SLinus Torvalds 9951da177e4SLinus Torvalds /* 9961da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 9971da177e4SLinus Torvalds * The page must have been locked by the caller. 9981da177e4SLinus Torvalds * 9991da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 10001da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 10011da177e4SLinus Torvalds * that has been written but not committed. 10021da177e4SLinus Torvalds */ 1003275acaafSTrond Myklebust static int nfs_flush_one(struct nfs_pageio_descriptor *desc, struct list_head *res) 10041da177e4SLinus Torvalds { 10051da177e4SLinus Torvalds struct nfs_page *req; 10061da177e4SLinus Torvalds struct page **pages; 10071da177e4SLinus Torvalds struct nfs_write_data *data; 1008c76069bdSFred Isaman struct list_head *head = &desc->pg_list; 10093b609184SPeng Tao int ret = 0; 10101da177e4SLinus Torvalds 1011c76069bdSFred Isaman data = nfs_writedata_alloc(nfs_page_array_len(desc->pg_base, 1012c76069bdSFred Isaman desc->pg_count)); 101344b83799SFred Isaman if (!data) { 101444b83799SFred Isaman while (!list_empty(head)) { 101544b83799SFred Isaman req = nfs_list_entry(head->next); 101644b83799SFred Isaman nfs_list_remove_request(req); 101744b83799SFred Isaman nfs_redirty_request(req); 101844b83799SFred Isaman } 101944b83799SFred Isaman ret = -ENOMEM; 102044b83799SFred Isaman goto out; 102144b83799SFred Isaman } 10221da177e4SLinus Torvalds pages = data->pagevec; 10231da177e4SLinus Torvalds while (!list_empty(head)) { 10241da177e4SLinus Torvalds req = nfs_list_entry(head->next); 10251da177e4SLinus Torvalds nfs_list_remove_request(req); 10261da177e4SLinus Torvalds nfs_list_add_request(req, &data->pages); 10271da177e4SLinus Torvalds *pages++ = req->wb_page; 10281da177e4SLinus Torvalds } 10291da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 10301da177e4SLinus Torvalds 1031b31268acSTrond Myklebust if ((desc->pg_ioflags & FLUSH_COND_STABLE) && 1032b31268acSTrond Myklebust (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit)) 1033b31268acSTrond Myklebust desc->pg_ioflags &= ~FLUSH_COND_STABLE; 1034b31268acSTrond Myklebust 10351da177e4SLinus Torvalds /* Set up the argument struct */ 10366e4efd56STrond Myklebust nfs_write_rpcsetup(req, data, desc->pg_count, 0, desc->pg_ioflags); 1037275acaafSTrond Myklebust list_add(&data->list, res); 103850828d7eSTrond Myklebust desc->pg_rpc_callops = &nfs_write_full_ops; 103944b83799SFred Isaman out: 104044b83799SFred Isaman return ret; 10411da177e4SLinus Torvalds } 10421da177e4SLinus Torvalds 1043dce81290STrond Myklebust int nfs_generic_flush(struct nfs_pageio_descriptor *desc, struct list_head *head) 1044dce81290STrond Myklebust { 1045dce81290STrond Myklebust if (desc->pg_bsize < PAGE_CACHE_SIZE) 1046dce81290STrond Myklebust return nfs_flush_multi(desc, head); 1047dce81290STrond Myklebust return nfs_flush_one(desc, head); 1048dce81290STrond Myklebust } 1049dce81290STrond Myklebust 1050dce81290STrond Myklebust static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) 10511751c363STrond Myklebust { 1052275acaafSTrond Myklebust LIST_HEAD(head); 1053275acaafSTrond Myklebust int ret; 1054275acaafSTrond Myklebust 1055dce81290STrond Myklebust ret = nfs_generic_flush(desc, &head); 1056275acaafSTrond Myklebust if (ret == 0) 105750828d7eSTrond Myklebust ret = nfs_do_multiple_writes(&head, desc->pg_rpc_callops, 1058dce81290STrond Myklebust desc->pg_ioflags); 1059275acaafSTrond Myklebust return ret; 10601751c363STrond Myklebust } 10611751c363STrond Myklebust 10621751c363STrond Myklebust static const struct nfs_pageio_ops nfs_pageio_write_ops = { 10631751c363STrond Myklebust .pg_test = nfs_generic_pg_test, 10641751c363STrond Myklebust .pg_doio = nfs_generic_pg_writepages, 10651751c363STrond Myklebust }; 10661751c363STrond Myklebust 1067e2fecb21STrond Myklebust void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio, 10681751c363STrond Myklebust struct inode *inode, int ioflags) 10691751c363STrond Myklebust { 10701751c363STrond Myklebust nfs_pageio_init(pgio, inode, &nfs_pageio_write_ops, 10711751c363STrond Myklebust NFS_SERVER(inode)->wsize, ioflags); 10721751c363STrond Myklebust } 10731751c363STrond Myklebust 1074dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1075dce81290STrond Myklebust { 1076dce81290STrond Myklebust pgio->pg_ops = &nfs_pageio_write_ops; 1077dce81290STrond Myklebust pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1078dce81290STrond Myklebust } 10791f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1080dce81290STrond Myklebust 1081c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1082c63c7b05STrond Myklebust struct inode *inode, int ioflags) 10831da177e4SLinus Torvalds { 10841751c363STrond Myklebust if (!pnfs_pageio_init_write(pgio, inode, ioflags)) 10851751c363STrond Myklebust nfs_pageio_init_write_mds(pgio, inode, ioflags); 10861da177e4SLinus Torvalds } 10871da177e4SLinus Torvalds 10881da177e4SLinus Torvalds /* 10891da177e4SLinus Torvalds * Handle a write reply that flushed part of a page. 10901da177e4SLinus Torvalds */ 1091788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) 10921da177e4SLinus Torvalds { 1093788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 10941da177e4SLinus Torvalds 109548186c7dSChuck Lever dprintk("NFS: %5u write(%s/%lld %d@%lld)", 109648186c7dSChuck Lever task->tk_pid, 10973d4ff43dSAl Viro data->req->wb_context->dentry->d_inode->i_sb->s_id, 109848186c7dSChuck Lever (long long) 10993d4ff43dSAl Viro NFS_FILEID(data->req->wb_context->dentry->d_inode), 110048186c7dSChuck Lever data->req->wb_bytes, (long long)req_offset(data->req)); 11011da177e4SLinus Torvalds 1102c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 1103c9d8f89dSTrond Myklebust } 1104788e7a89STrond Myklebust 1105c9d8f89dSTrond Myklebust static void nfs_writeback_release_partial(void *calldata) 1106c9d8f89dSTrond Myklebust { 1107c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1108c9d8f89dSTrond Myklebust struct nfs_page *req = data->req; 1109c9d8f89dSTrond Myklebust struct page *page = req->wb_page; 1110c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1111c9d8f89dSTrond Myklebust 1112c9d8f89dSTrond Myklebust if (status < 0) { 1113a301b777STrond Myklebust nfs_set_pageerror(page); 1114c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 1115c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 11168e821cadSTrond Myklebust goto out; 11178e821cadSTrond Myklebust } 11188e821cadSTrond Myklebust 11198e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 1120587142f8STrond Myklebust struct inode *inode = page->mapping->host; 11218e821cadSTrond Myklebust 1122587142f8STrond Myklebust spin_lock(&inode->i_lock); 11238e821cadSTrond Myklebust if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { 11248e821cadSTrond Myklebust /* Do nothing we need to resend the writes */ 11258e821cadSTrond Myklebust } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { 11261da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 11271da177e4SLinus Torvalds dprintk(" defer commit\n"); 11281da177e4SLinus Torvalds } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { 11298e821cadSTrond Myklebust set_bit(PG_NEED_RESCHED, &req->wb_flags); 11308e821cadSTrond Myklebust clear_bit(PG_NEED_COMMIT, &req->wb_flags); 11311da177e4SLinus Torvalds dprintk(" server reboot detected\n"); 11321da177e4SLinus Torvalds } 1133587142f8STrond Myklebust spin_unlock(&inode->i_lock); 11341da177e4SLinus Torvalds } else 11351da177e4SLinus Torvalds dprintk(" OK\n"); 11361da177e4SLinus Torvalds 11378e821cadSTrond Myklebust out: 11381da177e4SLinus Torvalds if (atomic_dec_and_test(&req->wb_complete)) 1139a861a1e1SFred Isaman nfs_writepage_release(req, data); 1140c9d8f89dSTrond Myklebust nfs_writedata_release(calldata); 11411da177e4SLinus Torvalds } 11421da177e4SLinus Torvalds 1143def6ed7eSAndy Adamson #if defined(CONFIG_NFS_V4_1) 1144def6ed7eSAndy Adamson void nfs_write_prepare(struct rpc_task *task, void *calldata) 1145def6ed7eSAndy Adamson { 1146def6ed7eSAndy Adamson struct nfs_write_data *data = calldata; 1147def6ed7eSAndy Adamson 1148035168abSTrond Myklebust if (nfs4_setup_sequence(NFS_SERVER(data->inode), 1149035168abSTrond Myklebust &data->args.seq_args, 11509d12b216STrond Myklebust &data->res.seq_res, task)) 1151def6ed7eSAndy Adamson return; 1152def6ed7eSAndy Adamson rpc_call_start(task); 1153def6ed7eSAndy Adamson } 1154def6ed7eSAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1155def6ed7eSAndy Adamson 1156788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops = { 1157def6ed7eSAndy Adamson #if defined(CONFIG_NFS_V4_1) 1158def6ed7eSAndy Adamson .rpc_call_prepare = nfs_write_prepare, 1159def6ed7eSAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1160788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_partial, 1161c9d8f89dSTrond Myklebust .rpc_release = nfs_writeback_release_partial, 1162788e7a89STrond Myklebust }; 1163788e7a89STrond Myklebust 11641da177e4SLinus Torvalds /* 11651da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 11661da177e4SLinus Torvalds * 11671da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 11681da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 11691da177e4SLinus Torvalds * as the page has a write request pending. 11701da177e4SLinus Torvalds */ 1171788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) 11721da177e4SLinus Torvalds { 1173788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 11741da177e4SLinus Torvalds 1175c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 1176c9d8f89dSTrond Myklebust } 1177c9d8f89dSTrond Myklebust 1178c9d8f89dSTrond Myklebust static void nfs_writeback_release_full(void *calldata) 1179c9d8f89dSTrond Myklebust { 1180c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1181e2fecb21STrond Myklebust int status = data->task.tk_status; 1182788e7a89STrond Myklebust 11831da177e4SLinus Torvalds /* Update attributes as result of writeback. */ 11841da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 1185c9d8f89dSTrond Myklebust struct nfs_page *req = nfs_list_entry(data->pages.next); 1186c9d8f89dSTrond Myklebust struct page *page = req->wb_page; 1187c9d8f89dSTrond Myklebust 11881da177e4SLinus Torvalds nfs_list_remove_request(req); 11891da177e4SLinus Torvalds 119048186c7dSChuck Lever dprintk("NFS: %5u write (%s/%lld %d@%lld)", 119148186c7dSChuck Lever data->task.tk_pid, 11923d4ff43dSAl Viro req->wb_context->dentry->d_inode->i_sb->s_id, 11933d4ff43dSAl Viro (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 11941da177e4SLinus Torvalds req->wb_bytes, 11951da177e4SLinus Torvalds (long long)req_offset(req)); 11961da177e4SLinus Torvalds 1197c9d8f89dSTrond Myklebust if (status < 0) { 1198a301b777STrond Myklebust nfs_set_pageerror(page); 1199c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 1200c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 12018e821cadSTrond Myklebust goto remove_request; 12021da177e4SLinus Torvalds } 12031da177e4SLinus Torvalds 12048e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 12051da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 1206a861a1e1SFred Isaman nfs_mark_request_commit(req, data->lseg); 12071da177e4SLinus Torvalds dprintk(" marked for commit\n"); 12088e821cadSTrond Myklebust goto next; 12098e821cadSTrond Myklebust } 12108e821cadSTrond Myklebust dprintk(" OK\n"); 12118e821cadSTrond Myklebust remove_request: 12121da177e4SLinus Torvalds nfs_inode_remove_request(req); 12131da177e4SLinus Torvalds next: 12149994b62bSFred Isaman nfs_unlock_request(req); 1215a6305ddbSTrond Myklebust nfs_end_page_writeback(page); 12161da177e4SLinus Torvalds } 1217c9d8f89dSTrond Myklebust nfs_writedata_release(calldata); 12181da177e4SLinus Torvalds } 12191da177e4SLinus Torvalds 1220788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = { 1221def6ed7eSAndy Adamson #if defined(CONFIG_NFS_V4_1) 1222def6ed7eSAndy Adamson .rpc_call_prepare = nfs_write_prepare, 1223def6ed7eSAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1224788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_full, 1225c9d8f89dSTrond Myklebust .rpc_release = nfs_writeback_release_full, 1226788e7a89STrond Myklebust }; 1227788e7a89STrond Myklebust 1228788e7a89STrond Myklebust 12291da177e4SLinus Torvalds /* 12301da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 12311da177e4SLinus Torvalds */ 123213602896SFred Isaman void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 12331da177e4SLinus Torvalds { 12341da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 12351da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1236788e7a89STrond Myklebust int status; 12371da177e4SLinus Torvalds 1238a3f565b1SChuck Lever dprintk("NFS: %5u nfs_writeback_done (status %d)\n", 12391da177e4SLinus Torvalds task->tk_pid, task->tk_status); 12401da177e4SLinus Torvalds 1241f551e44fSChuck Lever /* 1242f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1243f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1244f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1245f551e44fSChuck Lever * another writer had changed the file, but some applications 1246f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1247f551e44fSChuck Lever */ 1248788e7a89STrond Myklebust status = NFS_PROTO(data->inode)->write_done(task, data); 1249788e7a89STrond Myklebust if (status != 0) 125013602896SFred Isaman return; 125191d5b470SChuck Lever nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 125291d5b470SChuck Lever 12531da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 12541da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 12551da177e4SLinus Torvalds /* We tried a write call, but the server did not 12561da177e4SLinus Torvalds * commit data to stable storage even though we 12571da177e4SLinus Torvalds * requested it. 12581da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 12591da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 12601da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 12611da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 12621da177e4SLinus Torvalds */ 12631da177e4SLinus Torvalds static unsigned long complain; 12641da177e4SLinus Torvalds 1265a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 12661da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 12671da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 12681da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 12692b72c9ccSRakib Mullick NFS_SERVER(data->inode)->nfs_client->cl_hostname, 12701da177e4SLinus Torvalds resp->verf->committed, argp->stable); 12711da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 12721da177e4SLinus Torvalds } 12731da177e4SLinus Torvalds } 12741da177e4SLinus Torvalds #endif 12751da177e4SLinus Torvalds /* Is this a short write? */ 12761da177e4SLinus Torvalds if (task->tk_status >= 0 && resp->count < argp->count) { 12771da177e4SLinus Torvalds static unsigned long complain; 12781da177e4SLinus Torvalds 127991d5b470SChuck Lever nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); 128091d5b470SChuck Lever 12811da177e4SLinus Torvalds /* Has the server at least made some progress? */ 12821da177e4SLinus Torvalds if (resp->count != 0) { 12831da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 12841da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 12851da177e4SLinus Torvalds /* Resend from where the server left off */ 1286a69aef14SFred Isaman data->mds_offset += resp->count; 12871da177e4SLinus Torvalds argp->offset += resp->count; 12881da177e4SLinus Torvalds argp->pgbase += resp->count; 12891da177e4SLinus Torvalds argp->count -= resp->count; 12901da177e4SLinus Torvalds } else { 12911da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 12921da177e4SLinus Torvalds * headaches in the case of a server crash. 12931da177e4SLinus Torvalds */ 12941da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 12951da177e4SLinus Torvalds } 1296d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 129713602896SFred Isaman return; 12981da177e4SLinus Torvalds } 12991da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 13001da177e4SLinus Torvalds printk(KERN_WARNING 13011da177e4SLinus Torvalds "NFS: Server wrote zero bytes, expected %u.\n", 13021da177e4SLinus Torvalds argp->count); 13031da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 13041da177e4SLinus Torvalds } 13051da177e4SLinus Torvalds /* Can't do anything about it except throw an error. */ 13061da177e4SLinus Torvalds task->tk_status = -EIO; 13071da177e4SLinus Torvalds } 130813602896SFred Isaman return; 13091da177e4SLinus Torvalds } 13101da177e4SLinus Torvalds 13111da177e4SLinus Torvalds 13121da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 131371d0a611STrond Myklebust static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) 131471d0a611STrond Myklebust { 1315b8413f98STrond Myklebust int ret; 1316b8413f98STrond Myklebust 131771d0a611STrond Myklebust if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) 131871d0a611STrond Myklebust return 1; 1319b8413f98STrond Myklebust if (!may_wait) 132071d0a611STrond Myklebust return 0; 1321b8413f98STrond Myklebust ret = out_of_line_wait_on_bit_lock(&nfsi->flags, 1322b8413f98STrond Myklebust NFS_INO_COMMIT, 1323b8413f98STrond Myklebust nfs_wait_bit_killable, 1324b8413f98STrond Myklebust TASK_KILLABLE); 1325b8413f98STrond Myklebust return (ret < 0) ? ret : 1; 132671d0a611STrond Myklebust } 132771d0a611STrond Myklebust 1328e0c2b380SFred Isaman void nfs_commit_clear_lock(struct nfs_inode *nfsi) 132971d0a611STrond Myklebust { 133071d0a611STrond Myklebust clear_bit(NFS_INO_COMMIT, &nfsi->flags); 133171d0a611STrond Myklebust smp_mb__after_clear_bit(); 133271d0a611STrond Myklebust wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); 133371d0a611STrond Myklebust } 1334e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_clear_lock); 133571d0a611STrond Myklebust 1336e0c2b380SFred Isaman void nfs_commitdata_release(void *data) 13371da177e4SLinus Torvalds { 1338383ba719STrond Myklebust struct nfs_write_data *wdata = data; 1339383ba719STrond Myklebust 1340988b6dceSFred Isaman put_lseg(wdata->lseg); 1341383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 13421da177e4SLinus Torvalds nfs_commit_free(wdata); 13431da177e4SLinus Torvalds } 1344e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 13451da177e4SLinus Torvalds 1346e0c2b380SFred Isaman int nfs_initiate_commit(struct nfs_write_data *data, struct rpc_clnt *clnt, 13479ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 1348788e7a89STrond Myklebust int how) 13491da177e4SLinus Torvalds { 135007737691STrond Myklebust struct rpc_task *task; 13519ace33cdSFred Isaman int priority = flush_task_priority(how); 1352bdc7f021STrond Myklebust struct rpc_message msg = { 1353bdc7f021STrond Myklebust .rpc_argp = &data->args, 1354bdc7f021STrond Myklebust .rpc_resp = &data->res, 13559ace33cdSFred Isaman .rpc_cred = data->cred, 1356bdc7f021STrond Myklebust }; 135784115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 135807737691STrond Myklebust .task = &data->task, 13599ace33cdSFred Isaman .rpc_client = clnt, 1360bdc7f021STrond Myklebust .rpc_message = &msg, 13619ace33cdSFred Isaman .callback_ops = call_ops, 136284115e1cSTrond Myklebust .callback_data = data, 1363101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 13642c61be0aSTrond Myklebust .flags = RPC_TASK_ASYNC, 13653ff7576dSTrond Myklebust .priority = priority, 136684115e1cSTrond Myklebust }; 1367788e7a89STrond Myklebust /* Set up the initial task struct. */ 13689ace33cdSFred Isaman NFS_PROTO(data->inode)->commit_setup(data, &msg); 13691da177e4SLinus Torvalds 1370a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1371bdc7f021STrond Myklebust 137207737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1373dbae4c73STrond Myklebust if (IS_ERR(task)) 1374dbae4c73STrond Myklebust return PTR_ERR(task); 1375d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1376d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 137707737691STrond Myklebust rpc_put_task(task); 1378dbae4c73STrond Myklebust return 0; 13791da177e4SLinus Torvalds } 1380e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 13811da177e4SLinus Torvalds 13821da177e4SLinus Torvalds /* 13839ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 13849ace33cdSFred Isaman */ 1385e0c2b380SFred Isaman void nfs_init_commit(struct nfs_write_data *data, 1386988b6dceSFred Isaman struct list_head *head, 1387988b6dceSFred Isaman struct pnfs_layout_segment *lseg) 13889ace33cdSFred Isaman { 13899ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 13903d4ff43dSAl Viro struct inode *inode = first->wb_context->dentry->d_inode; 13919ace33cdSFred Isaman 13929ace33cdSFred Isaman /* Set up the RPC argument and reply structs 13939ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 13949ace33cdSFred Isaman 13959ace33cdSFred Isaman list_splice_init(head, &data->pages); 13969ace33cdSFred Isaman 13979ace33cdSFred Isaman data->inode = inode; 13989ace33cdSFred Isaman data->cred = first->wb_context->cred; 1399988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 14009ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 14019ace33cdSFred Isaman 14029ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 14039ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 14049ace33cdSFred Isaman data->args.offset = 0; 14059ace33cdSFred Isaman data->args.count = 0; 14069ace33cdSFred Isaman data->args.context = get_nfs_open_context(first->wb_context); 14079ace33cdSFred Isaman data->res.count = 0; 14089ace33cdSFred Isaman data->res.fattr = &data->fattr; 14099ace33cdSFred Isaman data->res.verf = &data->verf; 14109ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 14119ace33cdSFred Isaman } 1412e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 14139ace33cdSFred Isaman 1414e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1415a861a1e1SFred Isaman struct pnfs_layout_segment *lseg) 141664bfeb49SFred Isaman { 141764bfeb49SFred Isaman struct nfs_page *req; 141864bfeb49SFred Isaman 141964bfeb49SFred Isaman while (!list_empty(page_list)) { 142064bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 142164bfeb49SFred Isaman nfs_list_remove_request(req); 1422a861a1e1SFred Isaman nfs_mark_request_commit(req, lseg); 142364bfeb49SFred Isaman dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 142464bfeb49SFred Isaman dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 142564bfeb49SFred Isaman BDI_RECLAIMABLE); 14269994b62bSFred Isaman nfs_unlock_request(req); 142764bfeb49SFred Isaman } 142864bfeb49SFred Isaman } 1429e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 143064bfeb49SFred Isaman 14319ace33cdSFred Isaman /* 14321da177e4SLinus Torvalds * Commit dirty pages 14331da177e4SLinus Torvalds */ 14341da177e4SLinus Torvalds static int 143540859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how) 14361da177e4SLinus Torvalds { 14371da177e4SLinus Torvalds struct nfs_write_data *data; 14381da177e4SLinus Torvalds 1439c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 14401da177e4SLinus Torvalds 14411da177e4SLinus Torvalds if (!data) 14421da177e4SLinus Torvalds goto out_bad; 14431da177e4SLinus Torvalds 14441da177e4SLinus Torvalds /* Set up the argument struct */ 1445988b6dceSFred Isaman nfs_init_commit(data, head, NULL); 14469ace33cdSFred Isaman return nfs_initiate_commit(data, NFS_CLIENT(inode), data->mds_ops, how); 14471da177e4SLinus Torvalds out_bad: 1448a861a1e1SFred Isaman nfs_retry_commit(head, NULL); 144971d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(inode)); 14501da177e4SLinus Torvalds return -ENOMEM; 14511da177e4SLinus Torvalds } 14521da177e4SLinus Torvalds 14531da177e4SLinus Torvalds /* 14541da177e4SLinus Torvalds * COMMIT call returned 14551da177e4SLinus Torvalds */ 1456788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 14571da177e4SLinus Torvalds { 1458963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 14591da177e4SLinus Torvalds 1460a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 14611da177e4SLinus Torvalds task->tk_pid, task->tk_status); 14621da177e4SLinus Torvalds 1463788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1464c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1465c9d8f89dSTrond Myklebust } 1466c9d8f89dSTrond Myklebust 1467e0c2b380SFred Isaman void nfs_commit_release_pages(struct nfs_write_data *data) 1468c9d8f89dSTrond Myklebust { 1469c9d8f89dSTrond Myklebust struct nfs_page *req; 1470c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1471788e7a89STrond Myklebust 14721da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 14731da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 14741da177e4SLinus Torvalds nfs_list_remove_request(req); 1475*d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 14761da177e4SLinus Torvalds 147748186c7dSChuck Lever dprintk("NFS: commit (%s/%lld %d@%lld)", 14783d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 14793d4ff43dSAl Viro (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 14801da177e4SLinus Torvalds req->wb_bytes, 14811da177e4SLinus Torvalds (long long)req_offset(req)); 1482c9d8f89dSTrond Myklebust if (status < 0) { 1483c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 14841da177e4SLinus Torvalds nfs_inode_remove_request(req); 1485c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 14861da177e4SLinus Torvalds goto next; 14871da177e4SLinus Torvalds } 14881da177e4SLinus Torvalds 14891da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 14901da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 14911da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 14921da177e4SLinus Torvalds /* We have a match */ 14931da177e4SLinus Torvalds nfs_inode_remove_request(req); 14941da177e4SLinus Torvalds dprintk(" OK\n"); 14951da177e4SLinus Torvalds goto next; 14961da177e4SLinus Torvalds } 14971da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 14981da177e4SLinus Torvalds dprintk(" mismatch\n"); 14996d884e8fSFred nfs_mark_request_dirty(req); 15001da177e4SLinus Torvalds next: 15019994b62bSFred Isaman nfs_unlock_request(req); 15021da177e4SLinus Torvalds } 15035917ce84SFred Isaman } 1504e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_release_pages); 15055917ce84SFred Isaman 15065917ce84SFred Isaman static void nfs_commit_release(void *calldata) 15075917ce84SFred Isaman { 15085917ce84SFred Isaman struct nfs_write_data *data = calldata; 15095917ce84SFred Isaman 15105917ce84SFred Isaman nfs_commit_release_pages(data); 151171d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(data->inode)); 1512c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 15131da177e4SLinus Torvalds } 1514788e7a89STrond Myklebust 1515788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 151621d9a851SAndy Adamson #if defined(CONFIG_NFS_V4_1) 151721d9a851SAndy Adamson .rpc_call_prepare = nfs_write_prepare, 151821d9a851SAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1519788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1520788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1521788e7a89STrond Myklebust }; 15221da177e4SLinus Torvalds 1523b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 15241da177e4SLinus Torvalds { 15251da177e4SLinus Torvalds LIST_HEAD(head); 152671d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1527b8413f98STrond Myklebust int res; 15281da177e4SLinus Torvalds 1529b8413f98STrond Myklebust res = nfs_commit_set_lock(NFS_I(inode), may_wait); 1530b8413f98STrond Myklebust if (res <= 0) 1531c5efa5fcSTrond Myklebust goto out_mark_dirty; 1532*d6d6dc7cSFred Isaman res = nfs_scan_commit(inode, &head); 15331da177e4SLinus Torvalds if (res) { 1534a861a1e1SFred Isaman int error; 1535a861a1e1SFred Isaman 1536a861a1e1SFred Isaman error = pnfs_commit_list(inode, &head, how); 1537a861a1e1SFred Isaman if (error == PNFS_NOT_ATTEMPTED) 1538a861a1e1SFred Isaman error = nfs_commit_list(inode, &head, how); 15391da177e4SLinus Torvalds if (error < 0) 15401da177e4SLinus Torvalds return error; 1541b8413f98STrond Myklebust if (!may_wait) 1542b8413f98STrond Myklebust goto out_mark_dirty; 1543b8413f98STrond Myklebust error = wait_on_bit(&NFS_I(inode)->flags, 1544b8413f98STrond Myklebust NFS_INO_COMMIT, 154571d0a611STrond Myklebust nfs_wait_bit_killable, 154671d0a611STrond Myklebust TASK_KILLABLE); 1547b8413f98STrond Myklebust if (error < 0) 1548b8413f98STrond Myklebust return error; 154971d0a611STrond Myklebust } else 155071d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(inode)); 1551c5efa5fcSTrond Myklebust return res; 1552c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1553c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1554c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1555c5efa5fcSTrond Myklebust * that the data is on the disk. 1556c5efa5fcSTrond Myklebust */ 1557c5efa5fcSTrond Myklebust out_mark_dirty: 1558c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 15591da177e4SLinus Torvalds return res; 15601da177e4SLinus Torvalds } 15618fc795f7STrond Myklebust 15628fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 15638fc795f7STrond Myklebust { 1564420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1565420e3646STrond Myklebust int flags = FLUSH_SYNC; 1566420e3646STrond Myklebust int ret = 0; 15678fc795f7STrond Myklebust 15683236c3e1SJeff Layton /* no commits means nothing needs to be done */ 15693236c3e1SJeff Layton if (!nfsi->ncommit) 15703236c3e1SJeff Layton return ret; 15713236c3e1SJeff Layton 1572a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1573a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1574a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1575420e3646STrond Myklebust */ 1576a00dd6c0SJeff Layton if (nfsi->ncommit <= (nfsi->npages >> 1)) 1577420e3646STrond Myklebust goto out_mark_dirty; 1578420e3646STrond Myklebust 1579a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1580420e3646STrond Myklebust flags = 0; 1581a00dd6c0SJeff Layton } 1582a00dd6c0SJeff Layton 1583420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1584420e3646STrond Myklebust if (ret >= 0) { 1585420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1586420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1587420e3646STrond Myklebust wbc->nr_to_write -= ret; 1588420e3646STrond Myklebust else 1589420e3646STrond Myklebust wbc->nr_to_write = 0; 1590420e3646STrond Myklebust } 15918fc795f7STrond Myklebust return 0; 1592420e3646STrond Myklebust } 1593420e3646STrond Myklebust out_mark_dirty: 15948fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 15958fc795f7STrond Myklebust return ret; 15968fc795f7STrond Myklebust } 1597c63c7b05STrond Myklebust #else 15988fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 15998fc795f7STrond Myklebust { 16008fc795f7STrond Myklebust return 0; 16018fc795f7STrond Myklebust } 16021da177e4SLinus Torvalds #endif 16031da177e4SLinus Torvalds 16048fc795f7STrond Myklebust int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 16058fc795f7STrond Myklebust { 1606863a3c6cSAndy Adamson int ret; 1607863a3c6cSAndy Adamson 1608863a3c6cSAndy Adamson ret = nfs_commit_unstable_pages(inode, wbc); 1609863a3c6cSAndy Adamson if (ret >= 0 && test_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags)) { 1610ef311537SAndy Adamson int status; 1611ef311537SAndy Adamson bool sync = true; 1612863a3c6cSAndy Adamson 1613846d5a09SWu Fengguang if (wbc->sync_mode == WB_SYNC_NONE) 1614ef311537SAndy Adamson sync = false; 1615863a3c6cSAndy Adamson 1616863a3c6cSAndy Adamson status = pnfs_layoutcommit_inode(inode, sync); 1617863a3c6cSAndy Adamson if (status < 0) 1618863a3c6cSAndy Adamson return status; 1619863a3c6cSAndy Adamson } 1620863a3c6cSAndy Adamson return ret; 16218fc795f7STrond Myklebust } 16228fc795f7STrond Myklebust 1623acdc53b2STrond Myklebust /* 1624acdc53b2STrond Myklebust * flush the inode to disk. 1625acdc53b2STrond Myklebust */ 1626acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 162734901f70STrond Myklebust { 162834901f70STrond Myklebust struct writeback_control wbc = { 162972cb77f4STrond Myklebust .sync_mode = WB_SYNC_ALL, 163034901f70STrond Myklebust .nr_to_write = LONG_MAX, 1631d7fb1207STrond Myklebust .range_start = 0, 1632d7fb1207STrond Myklebust .range_end = LLONG_MAX, 163334901f70STrond Myklebust }; 163434901f70STrond Myklebust 1635acdc53b2STrond Myklebust return sync_inode(inode, &wbc); 16361c75950bSTrond Myklebust } 16371c75950bSTrond Myklebust 16381b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 16391b3b4a1aSTrond Myklebust { 16401b3b4a1aSTrond Myklebust struct nfs_page *req; 16411b3b4a1aSTrond Myklebust int ret = 0; 16421b3b4a1aSTrond Myklebust 16431b3b4a1aSTrond Myklebust BUG_ON(!PageLocked(page)); 16441b3b4a1aSTrond Myklebust for (;;) { 1645ba8b06e6STrond Myklebust wait_on_page_writeback(page); 16461b3b4a1aSTrond Myklebust req = nfs_page_find_request(page); 16471b3b4a1aSTrond Myklebust if (req == NULL) 16481b3b4a1aSTrond Myklebust break; 16491b3b4a1aSTrond Myklebust if (nfs_lock_request_dontget(req)) { 16501b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 16511b3b4a1aSTrond Myklebust /* 16521b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 16531b3b4a1aSTrond Myklebust * page as being dirty 16541b3b4a1aSTrond Myklebust */ 16551b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 16561b3b4a1aSTrond Myklebust nfs_unlock_request(req); 16571b3b4a1aSTrond Myklebust break; 16581b3b4a1aSTrond Myklebust } 16591b3b4a1aSTrond Myklebust ret = nfs_wait_on_request(req); 1660c9edda71STrond Myklebust nfs_release_request(req); 16611b3b4a1aSTrond Myklebust if (ret < 0) 1662c988950eSTrond Myklebust break; 16631b3b4a1aSTrond Myklebust } 16641b3b4a1aSTrond Myklebust return ret; 16651b3b4a1aSTrond Myklebust } 16661b3b4a1aSTrond Myklebust 16671c75950bSTrond Myklebust /* 16681c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 16691c75950bSTrond Myklebust */ 16701c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 16711c75950bSTrond Myklebust { 16727f2f12d9STrond Myklebust loff_t range_start = page_offset(page); 16737f2f12d9STrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 16747f2f12d9STrond Myklebust struct writeback_control wbc = { 16757f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 16767f2f12d9STrond Myklebust .nr_to_write = 0, 16777f2f12d9STrond Myklebust .range_start = range_start, 16787f2f12d9STrond Myklebust .range_end = range_end, 16797f2f12d9STrond Myklebust }; 16807f2f12d9STrond Myklebust int ret; 16817f2f12d9STrond Myklebust 16820522f6adSTrond Myklebust for (;;) { 1683ba8b06e6STrond Myklebust wait_on_page_writeback(page); 16847f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 16857f2f12d9STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 16867f2f12d9STrond Myklebust if (ret < 0) 16877f2f12d9STrond Myklebust goto out_error; 16880522f6adSTrond Myklebust continue; 16897f2f12d9STrond Myklebust } 16900522f6adSTrond Myklebust if (!PagePrivate(page)) 16910522f6adSTrond Myklebust break; 16920522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 16937f2f12d9STrond Myklebust if (ret < 0) 16947f2f12d9STrond Myklebust goto out_error; 16957f2f12d9STrond Myklebust } 16967f2f12d9STrond Myklebust return 0; 16977f2f12d9STrond Myklebust out_error: 16987f2f12d9STrond Myklebust return ret; 16991c75950bSTrond Myklebust } 17001c75950bSTrond Myklebust 1701074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 1702074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 1703a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 1704074cc1deSTrond Myklebust { 17052da95652SJeff Layton /* 17062da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 17072da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 17082da95652SJeff Layton * 17092da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 17102da95652SJeff Layton * that we can safely release the inode reference while holding 17112da95652SJeff Layton * the page lock. 17122da95652SJeff Layton */ 17132da95652SJeff Layton if (PagePrivate(page)) 17142da95652SJeff Layton return -EBUSY; 1715074cc1deSTrond Myklebust 1716074cc1deSTrond Myklebust nfs_fscache_release_page(page, GFP_KERNEL); 1717074cc1deSTrond Myklebust 1718a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 1719074cc1deSTrond Myklebust } 1720074cc1deSTrond Myklebust #endif 1721074cc1deSTrond Myklebust 1722f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 17231da177e4SLinus Torvalds { 17241da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 17251da177e4SLinus Torvalds sizeof(struct nfs_write_data), 17261da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 172720c2df83SPaul Mundt NULL); 17281da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 17291da177e4SLinus Torvalds return -ENOMEM; 17301da177e4SLinus Torvalds 173193d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 17321da177e4SLinus Torvalds nfs_wdata_cachep); 17331da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 17341da177e4SLinus Torvalds return -ENOMEM; 17351da177e4SLinus Torvalds 173693d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 17371da177e4SLinus Torvalds nfs_wdata_cachep); 17381da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 17391da177e4SLinus Torvalds return -ENOMEM; 17401da177e4SLinus Torvalds 174189a09141SPeter Zijlstra /* 174289a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 174389a09141SPeter Zijlstra * 174489a09141SPeter Zijlstra * 64MB: 8192k 174589a09141SPeter Zijlstra * 128MB: 11585k 174689a09141SPeter Zijlstra * 256MB: 16384k 174789a09141SPeter Zijlstra * 512MB: 23170k 174889a09141SPeter Zijlstra * 1GB: 32768k 174989a09141SPeter Zijlstra * 2GB: 46340k 175089a09141SPeter Zijlstra * 4GB: 65536k 175189a09141SPeter Zijlstra * 8GB: 92681k 175289a09141SPeter Zijlstra * 16GB: 131072k 175389a09141SPeter Zijlstra * 175489a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 175589a09141SPeter Zijlstra * Limit the default to 256M 175689a09141SPeter Zijlstra */ 175789a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 175889a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 175989a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 176089a09141SPeter Zijlstra 17611da177e4SLinus Torvalds return 0; 17621da177e4SLinus Torvalds } 17631da177e4SLinus Torvalds 1764266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 17651da177e4SLinus Torvalds { 17661da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 17671da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 17681a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 17691da177e4SLinus Torvalds } 17701da177e4SLinus Torvalds 1771