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> 233fcfab16SAndrew Morton 241da177e4SLinus Torvalds #include <asm/uaccess.h> 251da177e4SLinus Torvalds 261da177e4SLinus Torvalds #include "delegation.h" 2749a70f27STrond Myklebust #include "internal.h" 2891d5b470SChuck Lever #include "iostat.h" 29def6ed7eSAndy Adamson #include "nfs4_fs.h" 30074cc1deSTrond Myklebust #include "fscache.h" 311da177e4SLinus Torvalds 321da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 331da177e4SLinus Torvalds 341da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 351da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 361da177e4SLinus Torvalds 371da177e4SLinus Torvalds /* 381da177e4SLinus Torvalds * Local function declarations 391da177e4SLinus Torvalds */ 40c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc, 41c63c7b05STrond Myklebust struct inode *inode, int ioflags); 42f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 43788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops; 44788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops; 45788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 461da177e4SLinus Torvalds 47e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 483feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 491da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 501da177e4SLinus Torvalds 51c9d8f89dSTrond Myklebust struct nfs_write_data *nfs_commitdata_alloc(void) 521da177e4SLinus Torvalds { 53e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS); 5440859d7eSChuck Lever 551da177e4SLinus Torvalds if (p) { 561da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 571da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 585f7dbd5cSAndy Adamson p->res.seq_res.sr_slotid = NFS4_MAX_SLOT_TABLE; 591da177e4SLinus Torvalds } 601da177e4SLinus Torvalds return p; 611da177e4SLinus Torvalds } 621da177e4SLinus Torvalds 635e4424afSTrond Myklebust void nfs_commit_free(struct nfs_write_data *p) 641da177e4SLinus Torvalds { 6540859d7eSChuck Lever if (p && (p->pagevec != &p->page_array[0])) 6640859d7eSChuck Lever kfree(p->pagevec); 671da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 681da177e4SLinus Torvalds } 691da177e4SLinus Torvalds 708d5658c9STrond Myklebust struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount) 713feb2d49STrond Myklebust { 72e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); 733feb2d49STrond Myklebust 743feb2d49STrond Myklebust if (p) { 753feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 763feb2d49STrond Myklebust INIT_LIST_HEAD(&p->pages); 77e9f7bee1STrond Myklebust p->npages = pagecount; 785f7dbd5cSAndy Adamson p->res.seq_res.sr_slotid = NFS4_MAX_SLOT_TABLE; 790d0b5cb3SChuck Lever if (pagecount <= ARRAY_SIZE(p->page_array)) 800d0b5cb3SChuck Lever p->pagevec = p->page_array; 813feb2d49STrond Myklebust else { 820d0b5cb3SChuck Lever p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); 830d0b5cb3SChuck Lever if (!p->pagevec) { 843feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 853feb2d49STrond Myklebust p = NULL; 863feb2d49STrond Myklebust } 873feb2d49STrond Myklebust } 883feb2d49STrond Myklebust } 893feb2d49STrond Myklebust return p; 903feb2d49STrond Myklebust } 913feb2d49STrond Myklebust 921ae88b2eSTrond Myklebust void nfs_writedata_free(struct nfs_write_data *p) 933feb2d49STrond Myklebust { 943feb2d49STrond Myklebust if (p && (p->pagevec != &p->page_array[0])) 953feb2d49STrond Myklebust kfree(p->pagevec); 963feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 973feb2d49STrond Myklebust } 983feb2d49STrond Myklebust 991ae88b2eSTrond Myklebust static void nfs_writedata_release(struct nfs_write_data *wdata) 1001da177e4SLinus Torvalds { 101383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 1021da177e4SLinus Torvalds nfs_writedata_free(wdata); 1031da177e4SLinus Torvalds } 1041da177e4SLinus Torvalds 1057b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 1067b159fc1STrond Myklebust { 1077b159fc1STrond Myklebust ctx->error = error; 1087b159fc1STrond Myklebust smp_wmb(); 1097b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 1107b159fc1STrond Myklebust } 1117b159fc1STrond Myklebust 112277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request_locked(struct page *page) 113277459d2STrond Myklebust { 114277459d2STrond Myklebust struct nfs_page *req = NULL; 115277459d2STrond Myklebust 116277459d2STrond Myklebust if (PagePrivate(page)) { 117277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 118277459d2STrond Myklebust if (req != NULL) 119c03b4024STrond Myklebust kref_get(&req->wb_kref); 120277459d2STrond Myklebust } 121277459d2STrond Myklebust return req; 122277459d2STrond Myklebust } 123277459d2STrond Myklebust 124277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request(struct page *page) 125277459d2STrond Myklebust { 126587142f8STrond Myklebust struct inode *inode = page->mapping->host; 127277459d2STrond Myklebust struct nfs_page *req = NULL; 128277459d2STrond Myklebust 129587142f8STrond Myklebust spin_lock(&inode->i_lock); 130277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 131587142f8STrond Myklebust spin_unlock(&inode->i_lock); 132277459d2STrond Myklebust return req; 133277459d2STrond Myklebust } 134277459d2STrond Myklebust 1351da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1361da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1371da177e4SLinus Torvalds { 1381da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 139a3d01454STrond Myklebust loff_t end, i_size; 140a3d01454STrond Myklebust pgoff_t end_index; 1411da177e4SLinus Torvalds 142a3d01454STrond Myklebust spin_lock(&inode->i_lock); 143a3d01454STrond Myklebust i_size = i_size_read(inode); 144a3d01454STrond Myklebust end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 1451da177e4SLinus Torvalds if (i_size > 0 && page->index < end_index) 146a3d01454STrond Myklebust goto out; 1471da177e4SLinus Torvalds end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); 1481da177e4SLinus Torvalds if (i_size >= end) 149a3d01454STrond Myklebust goto out; 1501da177e4SLinus Torvalds i_size_write(inode, end); 151a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 152a3d01454STrond Myklebust out: 153a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 1541da177e4SLinus Torvalds } 1551da177e4SLinus Torvalds 156a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 157a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 158a301b777STrond Myklebust { 159a301b777STrond Myklebust SetPageError(page); 160a301b777STrond Myklebust nfs_zap_mapping(page->mapping->host, page->mapping); 161a301b777STrond Myklebust } 162a301b777STrond Myklebust 1631da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 1641da177e4SLinus Torvalds * covers the full page. 1651da177e4SLinus Torvalds */ 1661da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 1671da177e4SLinus Torvalds { 1681da177e4SLinus Torvalds if (PageUptodate(page)) 1691da177e4SLinus Torvalds return; 1701da177e4SLinus Torvalds if (base != 0) 1711da177e4SLinus Torvalds return; 17249a70f27STrond Myklebust if (count != nfs_page_length(page)) 1731da177e4SLinus Torvalds return; 1741da177e4SLinus Torvalds SetPageUptodate(page); 1751da177e4SLinus Torvalds } 1761da177e4SLinus Torvalds 1771da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 1781da177e4SLinus Torvalds { 1791da177e4SLinus Torvalds if (wbc->for_reclaim) 180c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 181b17621feSWu Fengguang if (wbc->for_kupdate || wbc->for_background) 1821da177e4SLinus Torvalds return FLUSH_LOWPRI; 1831da177e4SLinus Torvalds return 0; 1841da177e4SLinus Torvalds } 1851da177e4SLinus Torvalds 1861da177e4SLinus Torvalds /* 18789a09141SPeter Zijlstra * NFS congestion control 18889a09141SPeter Zijlstra */ 18989a09141SPeter Zijlstra 19089a09141SPeter Zijlstra int nfs_congestion_kb; 19189a09141SPeter Zijlstra 19289a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 19389a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 19489a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 19589a09141SPeter Zijlstra 1965a6d41b3STrond Myklebust static int nfs_set_page_writeback(struct page *page) 19789a09141SPeter Zijlstra { 1985a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 1995a6d41b3STrond Myklebust 2005a6d41b3STrond Myklebust if (!ret) { 20189a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 20289a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 20389a09141SPeter Zijlstra 204277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 2058aa7e847SJens Axboe NFS_CONGESTION_ON_THRESH) { 2068aa7e847SJens Axboe set_bdi_congested(&nfss->backing_dev_info, 2078aa7e847SJens Axboe BLK_RW_ASYNC); 2088aa7e847SJens Axboe } 20989a09141SPeter Zijlstra } 2105a6d41b3STrond Myklebust return ret; 21189a09141SPeter Zijlstra } 21289a09141SPeter Zijlstra 21389a09141SPeter Zijlstra static void nfs_end_page_writeback(struct page *page) 21489a09141SPeter Zijlstra { 21589a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 21689a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 21789a09141SPeter Zijlstra 21889a09141SPeter Zijlstra end_page_writeback(page); 219c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 2208aa7e847SJens Axboe clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 22189a09141SPeter Zijlstra } 22289a09141SPeter Zijlstra 223074cc1deSTrond Myklebust static struct nfs_page *nfs_find_and_lock_request(struct page *page) 224e261f51fSTrond Myklebust { 225587142f8STrond Myklebust struct inode *inode = page->mapping->host; 226e261f51fSTrond Myklebust struct nfs_page *req; 227e261f51fSTrond Myklebust int ret; 228e261f51fSTrond Myklebust 229587142f8STrond Myklebust spin_lock(&inode->i_lock); 230e261f51fSTrond Myklebust for (;;) { 231e261f51fSTrond Myklebust req = nfs_page_find_request_locked(page); 232074cc1deSTrond Myklebust if (req == NULL) 233074cc1deSTrond Myklebust break; 234acee478aSTrond Myklebust if (nfs_set_page_tag_locked(req)) 235e261f51fSTrond Myklebust break; 236e261f51fSTrond Myklebust /* Note: If we hold the page lock, as is the case in nfs_writepage, 237acee478aSTrond Myklebust * then the call to nfs_set_page_tag_locked() will always 238e261f51fSTrond Myklebust * succeed provided that someone hasn't already marked the 239e261f51fSTrond Myklebust * request as dirty (in which case we don't care). 240e261f51fSTrond Myklebust */ 241587142f8STrond Myklebust spin_unlock(&inode->i_lock); 242e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 243e261f51fSTrond Myklebust nfs_release_request(req); 244e261f51fSTrond Myklebust if (ret != 0) 245074cc1deSTrond Myklebust return ERR_PTR(ret); 246587142f8STrond Myklebust spin_lock(&inode->i_lock); 247e261f51fSTrond Myklebust } 248587142f8STrond Myklebust spin_unlock(&inode->i_lock); 249074cc1deSTrond Myklebust return req; 250612c9384STrond Myklebust } 251074cc1deSTrond Myklebust 252074cc1deSTrond Myklebust /* 253074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 254074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 255074cc1deSTrond Myklebust */ 256074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 257074cc1deSTrond Myklebust struct page *page) 258074cc1deSTrond Myklebust { 259074cc1deSTrond Myklebust struct nfs_page *req; 260074cc1deSTrond Myklebust int ret = 0; 261074cc1deSTrond Myklebust 262074cc1deSTrond Myklebust req = nfs_find_and_lock_request(page); 263074cc1deSTrond Myklebust if (!req) 264074cc1deSTrond Myklebust goto out; 265074cc1deSTrond Myklebust ret = PTR_ERR(req); 266074cc1deSTrond Myklebust if (IS_ERR(req)) 267074cc1deSTrond Myklebust goto out; 268074cc1deSTrond Myklebust 269074cc1deSTrond Myklebust ret = nfs_set_page_writeback(page); 270074cc1deSTrond Myklebust BUG_ON(ret != 0); 271074cc1deSTrond Myklebust BUG_ON(test_bit(PG_CLEAN, &req->wb_flags)); 272074cc1deSTrond Myklebust 273f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 274f8512ad0SFred Isaman nfs_redirty_request(req); 275074cc1deSTrond Myklebust ret = pgio->pg_error; 276f8512ad0SFred Isaman } 277074cc1deSTrond Myklebust out: 278074cc1deSTrond Myklebust return ret; 279e261f51fSTrond Myklebust } 280e261f51fSTrond Myklebust 281f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 282f758c885STrond Myklebust { 283f758c885STrond Myklebust struct inode *inode = page->mapping->host; 284f758c885STrond Myklebust 285f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 286f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 287f758c885STrond Myklebust 288f758c885STrond Myklebust nfs_pageio_cond_complete(pgio, page->index); 289f758c885STrond Myklebust return nfs_page_async_flush(pgio, page); 290f758c885STrond Myklebust } 291f758c885STrond Myklebust 292e261f51fSTrond Myklebust /* 2931da177e4SLinus Torvalds * Write an mmapped page to the server. 2941da177e4SLinus Torvalds */ 2954d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 2961da177e4SLinus Torvalds { 297f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 298e261f51fSTrond Myklebust int err; 2991da177e4SLinus Torvalds 300f758c885STrond Myklebust nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc)); 301f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 302f758c885STrond Myklebust nfs_pageio_complete(&pgio); 303f758c885STrond Myklebust if (err < 0) 3044d770ccfSTrond Myklebust return err; 305f758c885STrond Myklebust if (pgio.pg_error < 0) 306f758c885STrond Myklebust return pgio.pg_error; 307f758c885STrond Myklebust return 0; 3084d770ccfSTrond Myklebust } 3094d770ccfSTrond Myklebust 3104d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 3114d770ccfSTrond Myklebust { 312f758c885STrond Myklebust int ret; 3134d770ccfSTrond Myklebust 314f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 3151da177e4SLinus Torvalds unlock_page(page); 316f758c885STrond Myklebust return ret; 317f758c885STrond Myklebust } 318f758c885STrond Myklebust 319f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 320f758c885STrond Myklebust { 321f758c885STrond Myklebust int ret; 322f758c885STrond Myklebust 323f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 324f758c885STrond Myklebust unlock_page(page); 325f758c885STrond Myklebust return ret; 3261da177e4SLinus Torvalds } 3271da177e4SLinus Torvalds 3281da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 3291da177e4SLinus Torvalds { 3301da177e4SLinus Torvalds struct inode *inode = mapping->host; 33172cb77f4STrond Myklebust unsigned long *bitlock = &NFS_I(inode)->flags; 332c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 3331da177e4SLinus Torvalds int err; 3341da177e4SLinus Torvalds 33572cb77f4STrond Myklebust /* Stop dirtying of new pages while we sync */ 33672cb77f4STrond Myklebust err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, 33772cb77f4STrond Myklebust nfs_wait_bit_killable, TASK_KILLABLE); 33872cb77f4STrond Myklebust if (err) 33972cb77f4STrond Myklebust goto out_err; 34072cb77f4STrond Myklebust 34191d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 34291d5b470SChuck Lever 343c63c7b05STrond Myklebust nfs_pageio_init_write(&pgio, inode, wb_priority(wbc)); 344f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 345c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 34672cb77f4STrond Myklebust 34772cb77f4STrond Myklebust clear_bit_unlock(NFS_INO_FLUSHING, bitlock); 34872cb77f4STrond Myklebust smp_mb__after_clear_bit(); 34972cb77f4STrond Myklebust wake_up_bit(bitlock, NFS_INO_FLUSHING); 35072cb77f4STrond Myklebust 351f758c885STrond Myklebust if (err < 0) 35272cb77f4STrond Myklebust goto out_err; 35372cb77f4STrond Myklebust err = pgio.pg_error; 35472cb77f4STrond Myklebust if (err < 0) 35572cb77f4STrond Myklebust goto out_err; 356c63c7b05STrond Myklebust return 0; 35772cb77f4STrond Myklebust out_err: 35872cb77f4STrond Myklebust return err; 3591da177e4SLinus Torvalds } 3601da177e4SLinus Torvalds 3611da177e4SLinus Torvalds /* 3621da177e4SLinus Torvalds * Insert a write request into an inode 3631da177e4SLinus Torvalds */ 364e7d39069STrond Myklebust static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 3651da177e4SLinus Torvalds { 3661da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 3671da177e4SLinus Torvalds int error; 3681da177e4SLinus Torvalds 369e7d39069STrond Myklebust error = radix_tree_preload(GFP_NOFS); 370e7d39069STrond Myklebust if (error != 0) 371e7d39069STrond Myklebust goto out; 372e7d39069STrond Myklebust 373e7d39069STrond Myklebust /* Lock the request! */ 374e7d39069STrond Myklebust nfs_lock_request_dontget(req); 375e7d39069STrond Myklebust 376e7d39069STrond Myklebust spin_lock(&inode->i_lock); 3771da177e4SLinus Torvalds error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); 37827852596SNick Piggin BUG_ON(error); 3791da177e4SLinus Torvalds if (!nfsi->npages) { 3801da177e4SLinus Torvalds igrab(inode); 3811da177e4SLinus Torvalds if (nfs_have_delegation(inode, FMODE_WRITE)) 3821da177e4SLinus Torvalds nfsi->change_attr++; 3831da177e4SLinus Torvalds } 384deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 385277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 3861da177e4SLinus Torvalds nfsi->npages++; 387c03b4024STrond Myklebust kref_get(&req->wb_kref); 38827852596SNick Piggin radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, 38927852596SNick Piggin NFS_PAGE_TAG_LOCKED); 390e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 391e7d39069STrond Myklebust radix_tree_preload_end(); 392e7d39069STrond Myklebust out: 393e7d39069STrond Myklebust return error; 3941da177e4SLinus Torvalds } 3951da177e4SLinus Torvalds 3961da177e4SLinus Torvalds /* 39789a09141SPeter Zijlstra * Remove a write request from an inode 3981da177e4SLinus Torvalds */ 3991da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 4001da177e4SLinus Torvalds { 40188be9f99STrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 4021da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4031da177e4SLinus Torvalds 4041da177e4SLinus Torvalds BUG_ON (!NFS_WBACK_BUSY(req)); 4051da177e4SLinus Torvalds 406587142f8STrond Myklebust spin_lock(&inode->i_lock); 407277459d2STrond Myklebust set_page_private(req->wb_page, 0); 408deb7d638STrond Myklebust ClearPagePrivate(req->wb_page); 4091da177e4SLinus Torvalds radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); 4101da177e4SLinus Torvalds nfsi->npages--; 4111da177e4SLinus Torvalds if (!nfsi->npages) { 412587142f8STrond Myklebust spin_unlock(&inode->i_lock); 4131da177e4SLinus Torvalds iput(inode); 4141da177e4SLinus Torvalds } else 415587142f8STrond Myklebust spin_unlock(&inode->i_lock); 4161da177e4SLinus Torvalds nfs_clear_request(req); 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 4311da177e4SLinus Torvalds nfs_mark_request_commit(struct nfs_page *req) 4321da177e4SLinus Torvalds { 43388be9f99STrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 4341da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4351da177e4SLinus Torvalds 436587142f8STrond Myklebust spin_lock(&inode->i_lock); 437e468bae9STrond Myklebust set_bit(PG_CLEAN, &(req)->wb_flags); 4385c369683STrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, 4395c369683STrond Myklebust req->wb_index, 4405c369683STrond Myklebust NFS_PAGE_TAG_COMMIT); 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 448e468bae9STrond Myklebust static int 449e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 450e468bae9STrond Myklebust { 451e468bae9STrond Myklebust struct page *page = req->wb_page; 452e468bae9STrond Myklebust 453e468bae9STrond Myklebust if (test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) { 454e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 455e468bae9STrond Myklebust dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE); 456e468bae9STrond Myklebust return 1; 457e468bae9STrond Myklebust } 458e468bae9STrond Myklebust return 0; 459e468bae9STrond Myklebust } 460e468bae9STrond Myklebust 4618e821cadSTrond Myklebust static inline 4628e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4638e821cadSTrond Myklebust { 4648e821cadSTrond Myklebust return data->verf.committed != NFS_FILE_SYNC; 4658e821cadSTrond Myklebust } 4668e821cadSTrond Myklebust 4678e821cadSTrond Myklebust static inline 4688e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req) 4698e821cadSTrond Myklebust { 470e468bae9STrond Myklebust if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) { 4718e821cadSTrond Myklebust nfs_mark_request_commit(req); 4728e821cadSTrond Myklebust return 1; 4738e821cadSTrond Myklebust } 4748e821cadSTrond Myklebust if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) { 4756d884e8fSFred nfs_mark_request_dirty(req); 4768e821cadSTrond Myklebust return 1; 4778e821cadSTrond Myklebust } 4788e821cadSTrond Myklebust return 0; 4798e821cadSTrond Myklebust } 4808e821cadSTrond Myklebust #else 4818e821cadSTrond Myklebust static inline void 4828e821cadSTrond Myklebust nfs_mark_request_commit(struct nfs_page *req) 4838e821cadSTrond Myklebust { 4848e821cadSTrond Myklebust } 4858e821cadSTrond Myklebust 486e468bae9STrond Myklebust static inline int 487e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 488e468bae9STrond Myklebust { 489e468bae9STrond Myklebust return 0; 490e468bae9STrond Myklebust } 491e468bae9STrond Myklebust 4928e821cadSTrond Myklebust static inline 4938e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4948e821cadSTrond Myklebust { 4958e821cadSTrond Myklebust return 0; 4968e821cadSTrond Myklebust } 4978e821cadSTrond Myklebust 4988e821cadSTrond Myklebust static inline 4998e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req) 5008e821cadSTrond Myklebust { 5018e821cadSTrond Myklebust return 0; 5028e821cadSTrond Myklebust } 5031da177e4SLinus Torvalds #endif 5041da177e4SLinus Torvalds 5051da177e4SLinus Torvalds /* 5061da177e4SLinus Torvalds * Wait for a request to complete. 5071da177e4SLinus Torvalds * 508150030b7SMatthew Wilcox * Interruptible by fatal signals only. 5091da177e4SLinus Torvalds */ 510ca52fec1STrond Myklebust static int nfs_wait_on_requests_locked(struct inode *inode, pgoff_t idx_start, unsigned int npages) 5111da177e4SLinus Torvalds { 5121da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5131da177e4SLinus Torvalds struct nfs_page *req; 514ca52fec1STrond Myklebust pgoff_t idx_end, next; 5151da177e4SLinus Torvalds unsigned int res = 0; 5161da177e4SLinus Torvalds int error; 5171da177e4SLinus Torvalds 5181da177e4SLinus Torvalds if (npages == 0) 5191da177e4SLinus Torvalds idx_end = ~0; 5201da177e4SLinus Torvalds else 5211da177e4SLinus Torvalds idx_end = idx_start + npages - 1; 5221da177e4SLinus Torvalds 5231da177e4SLinus Torvalds next = idx_start; 5249fd367f0STrond Myklebust while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_LOCKED)) { 5251da177e4SLinus Torvalds if (req->wb_index > idx_end) 5261da177e4SLinus Torvalds break; 5271da177e4SLinus Torvalds 5281da177e4SLinus Torvalds next = req->wb_index + 1; 529c6a556b8STrond Myklebust BUG_ON(!NFS_WBACK_BUSY(req)); 5301da177e4SLinus Torvalds 531c03b4024STrond Myklebust kref_get(&req->wb_kref); 532587142f8STrond Myklebust spin_unlock(&inode->i_lock); 5331da177e4SLinus Torvalds error = nfs_wait_on_request(req); 5341da177e4SLinus Torvalds nfs_release_request(req); 535587142f8STrond Myklebust spin_lock(&inode->i_lock); 5361da177e4SLinus Torvalds if (error < 0) 5371da177e4SLinus Torvalds return error; 5381da177e4SLinus Torvalds res++; 5391da177e4SLinus Torvalds } 5401da177e4SLinus Torvalds return res; 5411da177e4SLinus Torvalds } 5421da177e4SLinus Torvalds 54347c62564STrond Myklebust #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 544fb8a1f11STrond Myklebust static int 545fb8a1f11STrond Myklebust nfs_need_commit(struct nfs_inode *nfsi) 546fb8a1f11STrond Myklebust { 547fb8a1f11STrond Myklebust return radix_tree_tagged(&nfsi->nfs_page_tree, NFS_PAGE_TAG_COMMIT); 548fb8a1f11STrond Myklebust } 549fb8a1f11STrond Myklebust 5501da177e4SLinus Torvalds /* 5511da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 5521da177e4SLinus Torvalds * @inode: NFS inode to scan 5531da177e4SLinus Torvalds * @dst: destination list 5541da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 5551da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 5561da177e4SLinus Torvalds * 5571da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 5581da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 5591da177e4SLinus Torvalds */ 5601da177e4SLinus Torvalds static int 561ca52fec1STrond Myklebust nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 5621da177e4SLinus Torvalds { 5631da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 564ff778d02STrond Myklebust int ret; 5653da28eb1STrond Myklebust 566fb8a1f11STrond Myklebust if (!nfs_need_commit(nfsi)) 567fb8a1f11STrond Myklebust return 0; 568fb8a1f11STrond Myklebust 569ff778d02STrond Myklebust ret = nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT); 570ff778d02STrond Myklebust if (ret > 0) 571ff778d02STrond Myklebust nfsi->ncommit -= ret; 5722928db1fSTrond Myklebust if (nfs_need_commit(NFS_I(inode))) 5732928db1fSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 574ff778d02STrond Myklebust return ret; 5751da177e4SLinus Torvalds } 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 582ca52fec1STrond Myklebust static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 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; 604277459d2STrond Myklebust 605e7d39069STrond Myklebust if (!PagePrivate(page)) 606e7d39069STrond Myklebust return NULL; 607e7d39069STrond Myklebust 608e7d39069STrond Myklebust end = offset + bytes; 609e7d39069STrond Myklebust spin_lock(&inode->i_lock); 610e7d39069STrond Myklebust 611e7d39069STrond Myklebust for (;;) { 612e7d39069STrond Myklebust req = nfs_page_find_request_locked(page); 613e7d39069STrond Myklebust if (req == NULL) 614e7d39069STrond Myklebust goto out_unlock; 615e7d39069STrond Myklebust 616e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 617e7d39069STrond Myklebust /* 618e7d39069STrond Myklebust * Tell the caller to flush out the request if 619e7d39069STrond Myklebust * the offsets are non-contiguous. 620e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 621e7d39069STrond Myklebust * have flushed out requests having wrong owners. 622e7d39069STrond Myklebust */ 623e468bae9STrond Myklebust if (offset > rqend 624e7d39069STrond Myklebust || end < req->wb_offset) 625e7d39069STrond Myklebust goto out_flushme; 626e7d39069STrond Myklebust 627e7d39069STrond Myklebust if (nfs_set_page_tag_locked(req)) 628e7d39069STrond Myklebust break; 629e7d39069STrond Myklebust 630e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 631587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6321da177e4SLinus Torvalds error = nfs_wait_on_request(req); 6331da177e4SLinus Torvalds nfs_release_request(req); 634e7d39069STrond Myklebust if (error != 0) 635e7d39069STrond Myklebust goto out_err; 636e7d39069STrond Myklebust spin_lock(&inode->i_lock); 6371da177e4SLinus Torvalds } 6381da177e4SLinus Torvalds 639ff778d02STrond Myklebust if (nfs_clear_request_commit(req) && 640e468bae9STrond Myklebust radix_tree_tag_clear(&NFS_I(inode)->nfs_page_tree, 641ff778d02STrond Myklebust req->wb_index, NFS_PAGE_TAG_COMMIT) != NULL) 642ff778d02STrond Myklebust NFS_I(inode)->ncommit--; 643e468bae9STrond Myklebust 6441da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 6451da177e4SLinus Torvalds if (offset < req->wb_offset) { 6461da177e4SLinus Torvalds req->wb_offset = offset; 6471da177e4SLinus Torvalds req->wb_pgbase = offset; 6481da177e4SLinus Torvalds } 6491da177e4SLinus Torvalds if (end > rqend) 6501da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 651e7d39069STrond Myklebust else 652e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 653e7d39069STrond Myklebust out_unlock: 654e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 655e7d39069STrond Myklebust return req; 656e7d39069STrond Myklebust out_flushme: 657e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 658e7d39069STrond Myklebust nfs_release_request(req); 659e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 660e7d39069STrond Myklebust out_err: 661e7d39069STrond Myklebust return ERR_PTR(error); 662e7d39069STrond Myklebust } 6631da177e4SLinus Torvalds 664e7d39069STrond Myklebust /* 665e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 666e7d39069STrond Myklebust * 667e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 668e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 669e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 670e7d39069STrond Myklebust */ 671e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 672e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 673e7d39069STrond Myklebust { 674e7d39069STrond Myklebust struct inode *inode = page->mapping->host; 675e7d39069STrond Myklebust struct nfs_page *req; 676e7d39069STrond Myklebust int error; 677e7d39069STrond Myklebust 678e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 679e7d39069STrond Myklebust if (req != NULL) 680e7d39069STrond Myklebust goto out; 681e7d39069STrond Myklebust req = nfs_create_request(ctx, inode, page, offset, bytes); 682e7d39069STrond Myklebust if (IS_ERR(req)) 683e7d39069STrond Myklebust goto out; 684e7d39069STrond Myklebust error = nfs_inode_add_request(inode, req); 685e7d39069STrond Myklebust if (error != 0) { 686e7d39069STrond Myklebust nfs_release_request(req); 687e7d39069STrond Myklebust req = ERR_PTR(error); 688e7d39069STrond Myklebust } 689efc91ed0STrond Myklebust out: 69061e930a9STrond Myklebust return req; 6911da177e4SLinus Torvalds } 6921da177e4SLinus Torvalds 693e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 694e7d39069STrond Myklebust unsigned int offset, unsigned int count) 695e7d39069STrond Myklebust { 696e7d39069STrond Myklebust struct nfs_page *req; 697e7d39069STrond Myklebust 698e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 699e7d39069STrond Myklebust if (IS_ERR(req)) 700e7d39069STrond Myklebust return PTR_ERR(req); 701e7d39069STrond Myklebust /* Update file length */ 702e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 703e7d39069STrond Myklebust nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 704e7d39069STrond Myklebust nfs_clear_page_tag_locked(req); 705e7d39069STrond Myklebust return 0; 706e7d39069STrond Myklebust } 707e7d39069STrond Myklebust 7081da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 7091da177e4SLinus Torvalds { 710cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 7111da177e4SLinus Torvalds struct nfs_page *req; 7121a54533eSTrond Myklebust int do_flush, status; 7131da177e4SLinus Torvalds /* 7141da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 7151da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 7161da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 7171da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 7181da177e4SLinus Torvalds * Also do the same if we find a request from an existing 7191da177e4SLinus Torvalds * dropped page. 7201da177e4SLinus Torvalds */ 7211a54533eSTrond Myklebust do { 722277459d2STrond Myklebust req = nfs_page_find_request(page); 7231a54533eSTrond Myklebust if (req == NULL) 7241a54533eSTrond Myklebust return 0; 725e468bae9STrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx; 7261da177e4SLinus Torvalds nfs_release_request(req); 7271a54533eSTrond Myklebust if (!do_flush) 7281a54533eSTrond Myklebust return 0; 729277459d2STrond Myklebust status = nfs_wb_page(page->mapping->host, page); 7301a54533eSTrond Myklebust } while (status == 0); 7311a54533eSTrond Myklebust return status; 7321da177e4SLinus Torvalds } 7331da177e4SLinus Torvalds 7341da177e4SLinus Torvalds /* 7355d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 7365d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 7375d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 7385d47a356STrond Myklebust */ 7395d47a356STrond Myklebust static int nfs_write_pageuptodate(struct page *page, struct inode *inode) 7405d47a356STrond Myklebust { 7415d47a356STrond Myklebust return PageUptodate(page) && 7425d47a356STrond Myklebust !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); 7435d47a356STrond Myklebust } 7445d47a356STrond Myklebust 7455d47a356STrond Myklebust /* 7461da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 7471da177e4SLinus Torvalds * 7481da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 7491da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 7501da177e4SLinus Torvalds */ 7511da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 7521da177e4SLinus Torvalds unsigned int offset, unsigned int count) 7531da177e4SLinus Torvalds { 754cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 7551da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 7561da177e4SLinus Torvalds int status = 0; 7571da177e4SLinus Torvalds 75891d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 75991d5b470SChuck Lever 76048186c7dSChuck Lever dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", 76101cce933SJosef "Jeff" Sipek file->f_path.dentry->d_parent->d_name.name, 76201cce933SJosef "Jeff" Sipek file->f_path.dentry->d_name.name, count, 7630bbacc40SChuck Lever (long long)(page_offset(page) + offset)); 7641da177e4SLinus Torvalds 7651da177e4SLinus Torvalds /* If we're not using byte range locks, and we know the page 7665d47a356STrond Myklebust * is up to date, it may be more efficient to extend the write 7675d47a356STrond Myklebust * to cover the entire page in order to avoid fragmentation 7685d47a356STrond Myklebust * inefficiencies. 7691da177e4SLinus Torvalds */ 7705d47a356STrond Myklebust if (nfs_write_pageuptodate(page, inode) && 7715d47a356STrond Myklebust inode->i_flock == NULL && 7726b2f3d1fSChristoph Hellwig !(file->f_flags & O_DSYNC)) { 77349a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 7741da177e4SLinus Torvalds offset = 0; 7751da177e4SLinus Torvalds } 7761da177e4SLinus Torvalds 777e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 77803fa9e84STrond Myklebust if (status < 0) 77903fa9e84STrond Myklebust nfs_set_pageerror(page); 78003fa9e84STrond Myklebust else 781e261f51fSTrond Myklebust __set_page_dirty_nobuffers(page); 7821da177e4SLinus Torvalds 78348186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 7841da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 7851da177e4SLinus Torvalds return status; 7861da177e4SLinus Torvalds } 7871da177e4SLinus Torvalds 7881da177e4SLinus Torvalds static void nfs_writepage_release(struct nfs_page *req) 7891da177e4SLinus Torvalds { 7908e821cadSTrond Myklebust 7917e5f6146STrond Myklebust if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req)) { 79289a09141SPeter Zijlstra nfs_end_page_writeback(req->wb_page); 7931da177e4SLinus Torvalds nfs_inode_remove_request(req); 7948e821cadSTrond Myklebust } else 7958e821cadSTrond Myklebust nfs_end_page_writeback(req->wb_page); 7969fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 7971da177e4SLinus Torvalds } 7981da177e4SLinus Torvalds 7993ff7576dSTrond Myklebust static int flush_task_priority(int how) 8001da177e4SLinus Torvalds { 8011da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 8021da177e4SLinus Torvalds case FLUSH_HIGHPRI: 8031da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 8041da177e4SLinus Torvalds case FLUSH_LOWPRI: 8051da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 8061da177e4SLinus Torvalds } 8071da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 8081da177e4SLinus Torvalds } 8091da177e4SLinus Torvalds 8101da177e4SLinus Torvalds /* 8111da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 8121da177e4SLinus Torvalds */ 813dbae4c73STrond Myklebust static int nfs_write_rpcsetup(struct nfs_page *req, 8141da177e4SLinus Torvalds struct nfs_write_data *data, 815788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 8161da177e4SLinus Torvalds unsigned int count, unsigned int offset, 8171da177e4SLinus Torvalds int how) 8181da177e4SLinus Torvalds { 81984115e1cSTrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 82084115e1cSTrond Myklebust int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 8213ff7576dSTrond Myklebust int priority = flush_task_priority(how); 82207737691STrond Myklebust struct rpc_task *task; 823bdc7f021STrond Myklebust struct rpc_message msg = { 824bdc7f021STrond Myklebust .rpc_argp = &data->args, 825bdc7f021STrond Myklebust .rpc_resp = &data->res, 826bdc7f021STrond Myklebust .rpc_cred = req->wb_context->cred, 827bdc7f021STrond Myklebust }; 82884115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 82984115e1cSTrond Myklebust .rpc_client = NFS_CLIENT(inode), 83007737691STrond Myklebust .task = &data->task, 831bdc7f021STrond Myklebust .rpc_message = &msg, 83284115e1cSTrond Myklebust .callback_ops = call_ops, 83384115e1cSTrond Myklebust .callback_data = data, 834101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 83584115e1cSTrond Myklebust .flags = flags, 8363ff7576dSTrond Myklebust .priority = priority, 83784115e1cSTrond Myklebust }; 8381da177e4SLinus Torvalds 8391da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 8401da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 8411da177e4SLinus Torvalds 8421da177e4SLinus Torvalds data->req = req; 84388be9f99STrond Myklebust data->inode = inode = req->wb_context->path.dentry->d_inode; 844bdc7f021STrond Myklebust data->cred = msg.rpc_cred; 8451da177e4SLinus Torvalds 8461da177e4SLinus Torvalds data->args.fh = NFS_FH(inode); 8471da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 8481da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 8491da177e4SLinus Torvalds data->args.pages = data->pagevec; 8501da177e4SLinus Torvalds data->args.count = count; 851383ba719STrond Myklebust data->args.context = get_nfs_open_context(req->wb_context); 852bdc7f021STrond Myklebust data->args.stable = NFS_UNSTABLE; 853bdc7f021STrond Myklebust if (how & FLUSH_STABLE) { 854bdc7f021STrond Myklebust data->args.stable = NFS_DATA_SYNC; 855fb8a1f11STrond Myklebust if (!nfs_need_commit(NFS_I(inode))) 856bdc7f021STrond Myklebust data->args.stable = NFS_FILE_SYNC; 857bdc7f021STrond Myklebust } 8581da177e4SLinus Torvalds 8591da177e4SLinus Torvalds data->res.fattr = &data->fattr; 8601da177e4SLinus Torvalds data->res.count = count; 8611da177e4SLinus Torvalds data->res.verf = &data->verf; 8620e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 8631da177e4SLinus Torvalds 864788e7a89STrond Myklebust /* Set up the initial task struct. */ 865bdc7f021STrond Myklebust NFS_PROTO(inode)->write_setup(data, &msg); 8661da177e4SLinus Torvalds 867a3f565b1SChuck Lever dprintk("NFS: %5u initiated write call " 86848186c7dSChuck Lever "(req %s/%lld, %u bytes @ offset %llu)\n", 8690bbacc40SChuck Lever data->task.tk_pid, 8701da177e4SLinus Torvalds inode->i_sb->s_id, 8711da177e4SLinus Torvalds (long long)NFS_FILEID(inode), 8721da177e4SLinus Torvalds count, 8731da177e4SLinus Torvalds (unsigned long long)data->args.offset); 8741da177e4SLinus Torvalds 87507737691STrond Myklebust task = rpc_run_task(&task_setup_data); 876dbae4c73STrond Myklebust if (IS_ERR(task)) 877dbae4c73STrond Myklebust return PTR_ERR(task); 87807737691STrond Myklebust rpc_put_task(task); 879dbae4c73STrond Myklebust return 0; 8801da177e4SLinus Torvalds } 8811da177e4SLinus Torvalds 8826d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 8836d884e8fSFred * call this on each, which will prepare them to be retried on next 8846d884e8fSFred * writeback using standard nfs. 8856d884e8fSFred */ 8866d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 8876d884e8fSFred { 8886d884e8fSFred nfs_mark_request_dirty(req); 8896d884e8fSFred nfs_end_page_writeback(req->wb_page); 8906d884e8fSFred nfs_clear_page_tag_locked(req); 8916d884e8fSFred } 8926d884e8fSFred 8931da177e4SLinus Torvalds /* 8941da177e4SLinus Torvalds * Generate multiple small requests to write out a single 8951da177e4SLinus Torvalds * contiguous dirty area on one page. 8961da177e4SLinus Torvalds */ 8978d5658c9STrond Myklebust static int nfs_flush_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) 8981da177e4SLinus Torvalds { 8991da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 9001da177e4SLinus Torvalds struct page *page = req->wb_page; 9011da177e4SLinus Torvalds struct nfs_write_data *data; 902e9f7bee1STrond Myklebust size_t wsize = NFS_SERVER(inode)->wsize, nbytes; 903e9f7bee1STrond Myklebust unsigned int offset; 9041da177e4SLinus Torvalds int requests = 0; 905dbae4c73STrond Myklebust int ret = 0; 9061da177e4SLinus Torvalds LIST_HEAD(list); 9071da177e4SLinus Torvalds 9081da177e4SLinus Torvalds nfs_list_remove_request(req); 9091da177e4SLinus Torvalds 910bcb71bbaSTrond Myklebust nbytes = count; 911e9f7bee1STrond Myklebust do { 912e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 913e9f7bee1STrond Myklebust 9148d5658c9STrond Myklebust data = nfs_writedata_alloc(1); 9151da177e4SLinus Torvalds if (!data) 9161da177e4SLinus Torvalds goto out_bad; 9171da177e4SLinus Torvalds list_add(&data->pages, &list); 9181da177e4SLinus Torvalds requests++; 919e9f7bee1STrond Myklebust nbytes -= len; 920e9f7bee1STrond Myklebust } while (nbytes != 0); 9211da177e4SLinus Torvalds atomic_set(&req->wb_complete, requests); 9221da177e4SLinus Torvalds 9231da177e4SLinus Torvalds ClearPageError(page); 9241da177e4SLinus Torvalds offset = 0; 925bcb71bbaSTrond Myklebust nbytes = count; 9261da177e4SLinus Torvalds do { 927dbae4c73STrond Myklebust int ret2; 928dbae4c73STrond Myklebust 9291da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 9301da177e4SLinus Torvalds list_del_init(&data->pages); 9311da177e4SLinus Torvalds 9321da177e4SLinus Torvalds data->pagevec[0] = page; 9331da177e4SLinus Torvalds 934bcb71bbaSTrond Myklebust if (nbytes < wsize) 935bcb71bbaSTrond Myklebust wsize = nbytes; 936dbae4c73STrond Myklebust ret2 = nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 937788e7a89STrond Myklebust wsize, offset, how); 938dbae4c73STrond Myklebust if (ret == 0) 939dbae4c73STrond Myklebust ret = ret2; 9401da177e4SLinus Torvalds offset += wsize; 9411da177e4SLinus Torvalds nbytes -= wsize; 9421da177e4SLinus Torvalds } while (nbytes != 0); 9431da177e4SLinus Torvalds 944dbae4c73STrond Myklebust return ret; 9451da177e4SLinus Torvalds 9461da177e4SLinus Torvalds out_bad: 9471da177e4SLinus Torvalds while (!list_empty(&list)) { 9481da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 9491da177e4SLinus Torvalds list_del(&data->pages); 9508aca67f0STrond Myklebust nfs_writedata_release(data); 9511da177e4SLinus Torvalds } 95261822ab5STrond Myklebust nfs_redirty_request(req); 9531da177e4SLinus Torvalds return -ENOMEM; 9541da177e4SLinus Torvalds } 9551da177e4SLinus Torvalds 9561da177e4SLinus Torvalds /* 9571da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 9581da177e4SLinus Torvalds * The page must have been locked by the caller. 9591da177e4SLinus Torvalds * 9601da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 9611da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 9621da177e4SLinus Torvalds * that has been written but not committed. 9631da177e4SLinus Torvalds */ 9648d5658c9STrond Myklebust static int nfs_flush_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) 9651da177e4SLinus Torvalds { 9661da177e4SLinus Torvalds struct nfs_page *req; 9671da177e4SLinus Torvalds struct page **pages; 9681da177e4SLinus Torvalds struct nfs_write_data *data; 9691da177e4SLinus Torvalds 9708d5658c9STrond Myklebust data = nfs_writedata_alloc(npages); 9711da177e4SLinus Torvalds if (!data) 9721da177e4SLinus Torvalds goto out_bad; 9731da177e4SLinus Torvalds 9741da177e4SLinus Torvalds pages = data->pagevec; 9751da177e4SLinus Torvalds while (!list_empty(head)) { 9761da177e4SLinus Torvalds req = nfs_list_entry(head->next); 9771da177e4SLinus Torvalds nfs_list_remove_request(req); 9781da177e4SLinus Torvalds nfs_list_add_request(req, &data->pages); 9791da177e4SLinus Torvalds ClearPageError(req->wb_page); 9801da177e4SLinus Torvalds *pages++ = req->wb_page; 9811da177e4SLinus Torvalds } 9821da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 9831da177e4SLinus Torvalds 9841da177e4SLinus Torvalds /* Set up the argument struct */ 985dbae4c73STrond Myklebust return nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how); 9861da177e4SLinus Torvalds out_bad: 9871da177e4SLinus Torvalds while (!list_empty(head)) { 98810afec90STrond Myklebust req = nfs_list_entry(head->next); 9891da177e4SLinus Torvalds nfs_list_remove_request(req); 99061822ab5STrond Myklebust nfs_redirty_request(req); 9911da177e4SLinus Torvalds } 9921da177e4SLinus Torvalds return -ENOMEM; 9931da177e4SLinus Torvalds } 9941da177e4SLinus Torvalds 995c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 996c63c7b05STrond Myklebust struct inode *inode, int ioflags) 9971da177e4SLinus Torvalds { 998bf4285e7SChuck Lever size_t wsize = NFS_SERVER(inode)->wsize; 9991da177e4SLinus Torvalds 1000bcb71bbaSTrond Myklebust if (wsize < PAGE_CACHE_SIZE) 1001c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags); 1002bcb71bbaSTrond Myklebust else 1003c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags); 10041da177e4SLinus Torvalds } 10051da177e4SLinus Torvalds 10061da177e4SLinus Torvalds /* 10071da177e4SLinus Torvalds * Handle a write reply that flushed part of a page. 10081da177e4SLinus Torvalds */ 1009788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) 10101da177e4SLinus Torvalds { 1011788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 10121da177e4SLinus Torvalds 101348186c7dSChuck Lever dprintk("NFS: %5u write(%s/%lld %d@%lld)", 101448186c7dSChuck Lever task->tk_pid, 101548186c7dSChuck Lever data->req->wb_context->path.dentry->d_inode->i_sb->s_id, 101648186c7dSChuck Lever (long long) 101748186c7dSChuck Lever NFS_FILEID(data->req->wb_context->path.dentry->d_inode), 101848186c7dSChuck Lever data->req->wb_bytes, (long long)req_offset(data->req)); 10191da177e4SLinus Torvalds 1020c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 1021c9d8f89dSTrond Myklebust } 1022788e7a89STrond Myklebust 1023c9d8f89dSTrond Myklebust static void nfs_writeback_release_partial(void *calldata) 1024c9d8f89dSTrond Myklebust { 1025c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1026c9d8f89dSTrond Myklebust struct nfs_page *req = data->req; 1027c9d8f89dSTrond Myklebust struct page *page = req->wb_page; 1028c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1029c9d8f89dSTrond Myklebust 1030c9d8f89dSTrond Myklebust if (status < 0) { 1031a301b777STrond Myklebust nfs_set_pageerror(page); 1032c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 1033c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 10348e821cadSTrond Myklebust goto out; 10358e821cadSTrond Myklebust } 10368e821cadSTrond Myklebust 10378e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 1038587142f8STrond Myklebust struct inode *inode = page->mapping->host; 10398e821cadSTrond Myklebust 1040587142f8STrond Myklebust spin_lock(&inode->i_lock); 10418e821cadSTrond Myklebust if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { 10428e821cadSTrond Myklebust /* Do nothing we need to resend the writes */ 10438e821cadSTrond Myklebust } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { 10441da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 10451da177e4SLinus Torvalds dprintk(" defer commit\n"); 10461da177e4SLinus Torvalds } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { 10478e821cadSTrond Myklebust set_bit(PG_NEED_RESCHED, &req->wb_flags); 10488e821cadSTrond Myklebust clear_bit(PG_NEED_COMMIT, &req->wb_flags); 10491da177e4SLinus Torvalds dprintk(" server reboot detected\n"); 10501da177e4SLinus Torvalds } 1051587142f8STrond Myklebust spin_unlock(&inode->i_lock); 10521da177e4SLinus Torvalds } else 10531da177e4SLinus Torvalds dprintk(" OK\n"); 10541da177e4SLinus Torvalds 10558e821cadSTrond Myklebust out: 10561da177e4SLinus Torvalds if (atomic_dec_and_test(&req->wb_complete)) 10571da177e4SLinus Torvalds nfs_writepage_release(req); 1058c9d8f89dSTrond Myklebust nfs_writedata_release(calldata); 10591da177e4SLinus Torvalds } 10601da177e4SLinus Torvalds 1061def6ed7eSAndy Adamson #if defined(CONFIG_NFS_V4_1) 1062def6ed7eSAndy Adamson void nfs_write_prepare(struct rpc_task *task, void *calldata) 1063def6ed7eSAndy Adamson { 1064def6ed7eSAndy Adamson struct nfs_write_data *data = calldata; 1065def6ed7eSAndy Adamson struct nfs_client *clp = (NFS_SERVER(data->inode))->nfs_client; 1066def6ed7eSAndy Adamson 1067def6ed7eSAndy Adamson if (nfs4_setup_sequence(clp, &data->args.seq_args, 1068def6ed7eSAndy Adamson &data->res.seq_res, 1, task)) 1069def6ed7eSAndy Adamson return; 1070def6ed7eSAndy Adamson rpc_call_start(task); 1071def6ed7eSAndy Adamson } 1072def6ed7eSAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1073def6ed7eSAndy Adamson 1074788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops = { 1075def6ed7eSAndy Adamson #if defined(CONFIG_NFS_V4_1) 1076def6ed7eSAndy Adamson .rpc_call_prepare = nfs_write_prepare, 1077def6ed7eSAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1078788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_partial, 1079c9d8f89dSTrond Myklebust .rpc_release = nfs_writeback_release_partial, 1080788e7a89STrond Myklebust }; 1081788e7a89STrond Myklebust 10821da177e4SLinus Torvalds /* 10831da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 10841da177e4SLinus Torvalds * 10851da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 10861da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 10871da177e4SLinus Torvalds * as the page has a write request pending. 10881da177e4SLinus Torvalds */ 1089788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) 10901da177e4SLinus Torvalds { 1091788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 10921da177e4SLinus Torvalds 1093c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 1094c9d8f89dSTrond Myklebust } 1095c9d8f89dSTrond Myklebust 1096c9d8f89dSTrond Myklebust static void nfs_writeback_release_full(void *calldata) 1097c9d8f89dSTrond Myklebust { 1098c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1099c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1100788e7a89STrond Myklebust 11011da177e4SLinus Torvalds /* Update attributes as result of writeback. */ 11021da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 1103c9d8f89dSTrond Myklebust struct nfs_page *req = nfs_list_entry(data->pages.next); 1104c9d8f89dSTrond Myklebust struct page *page = req->wb_page; 1105c9d8f89dSTrond Myklebust 11061da177e4SLinus Torvalds nfs_list_remove_request(req); 11071da177e4SLinus Torvalds 110848186c7dSChuck Lever dprintk("NFS: %5u write (%s/%lld %d@%lld)", 110948186c7dSChuck Lever data->task.tk_pid, 111088be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 111188be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 11121da177e4SLinus Torvalds req->wb_bytes, 11131da177e4SLinus Torvalds (long long)req_offset(req)); 11141da177e4SLinus Torvalds 1115c9d8f89dSTrond Myklebust if (status < 0) { 1116a301b777STrond Myklebust nfs_set_pageerror(page); 1117c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 1118c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 11198e821cadSTrond Myklebust goto remove_request; 11201da177e4SLinus Torvalds } 11211da177e4SLinus Torvalds 11228e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 11231da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 11241da177e4SLinus Torvalds nfs_mark_request_commit(req); 11258e821cadSTrond Myklebust nfs_end_page_writeback(page); 11261da177e4SLinus Torvalds dprintk(" marked for commit\n"); 11278e821cadSTrond Myklebust goto next; 11288e821cadSTrond Myklebust } 11298e821cadSTrond Myklebust dprintk(" OK\n"); 11308e821cadSTrond Myklebust remove_request: 11318e821cadSTrond Myklebust nfs_end_page_writeback(page); 11321da177e4SLinus Torvalds nfs_inode_remove_request(req); 11331da177e4SLinus Torvalds next: 11349fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 11351da177e4SLinus Torvalds } 1136c9d8f89dSTrond Myklebust nfs_writedata_release(calldata); 11371da177e4SLinus Torvalds } 11381da177e4SLinus Torvalds 1139788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = { 1140def6ed7eSAndy Adamson #if defined(CONFIG_NFS_V4_1) 1141def6ed7eSAndy Adamson .rpc_call_prepare = nfs_write_prepare, 1142def6ed7eSAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1143788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_full, 1144c9d8f89dSTrond Myklebust .rpc_release = nfs_writeback_release_full, 1145788e7a89STrond Myklebust }; 1146788e7a89STrond Myklebust 1147788e7a89STrond Myklebust 11481da177e4SLinus Torvalds /* 11491da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 11501da177e4SLinus Torvalds */ 1151462d5b32SChuck Lever int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 11521da177e4SLinus Torvalds { 11531da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 11541da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1155eedc020eSAndy Adamson struct nfs_server *server = NFS_SERVER(data->inode); 1156788e7a89STrond Myklebust int status; 11571da177e4SLinus Torvalds 1158a3f565b1SChuck Lever dprintk("NFS: %5u nfs_writeback_done (status %d)\n", 11591da177e4SLinus Torvalds task->tk_pid, task->tk_status); 11601da177e4SLinus Torvalds 1161f551e44fSChuck Lever /* 1162f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1163f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1164f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1165f551e44fSChuck Lever * another writer had changed the file, but some applications 1166f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1167f551e44fSChuck Lever */ 1168788e7a89STrond Myklebust status = NFS_PROTO(data->inode)->write_done(task, data); 1169788e7a89STrond Myklebust if (status != 0) 1170788e7a89STrond Myklebust return status; 117191d5b470SChuck Lever nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 117291d5b470SChuck Lever 11731da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 11741da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 11751da177e4SLinus Torvalds /* We tried a write call, but the server did not 11761da177e4SLinus Torvalds * commit data to stable storage even though we 11771da177e4SLinus Torvalds * requested it. 11781da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 11791da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 11801da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 11811da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 11821da177e4SLinus Torvalds */ 11831da177e4SLinus Torvalds static unsigned long complain; 11841da177e4SLinus Torvalds 11851da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 11861da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 11871da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1188eedc020eSAndy Adamson server->nfs_client->cl_hostname, 11891da177e4SLinus Torvalds resp->verf->committed, argp->stable); 11901da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11911da177e4SLinus Torvalds } 11921da177e4SLinus Torvalds } 11931da177e4SLinus Torvalds #endif 11941da177e4SLinus Torvalds /* Is this a short write? */ 11951da177e4SLinus Torvalds if (task->tk_status >= 0 && resp->count < argp->count) { 11961da177e4SLinus Torvalds static unsigned long complain; 11971da177e4SLinus Torvalds 119891d5b470SChuck Lever nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); 119991d5b470SChuck Lever 12001da177e4SLinus Torvalds /* Has the server at least made some progress? */ 12011da177e4SLinus Torvalds if (resp->count != 0) { 12021da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 12031da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 12041da177e4SLinus Torvalds /* Resend from where the server left off */ 12051da177e4SLinus Torvalds argp->offset += resp->count; 12061da177e4SLinus Torvalds argp->pgbase += resp->count; 12071da177e4SLinus Torvalds argp->count -= resp->count; 12081da177e4SLinus Torvalds } else { 12091da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 12101da177e4SLinus Torvalds * headaches in the case of a server crash. 12111da177e4SLinus Torvalds */ 12121da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 12131da177e4SLinus Torvalds } 12140110ee15STrond Myklebust nfs_restart_rpc(task, server->nfs_client); 1215788e7a89STrond Myklebust return -EAGAIN; 12161da177e4SLinus Torvalds } 12171da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 12181da177e4SLinus Torvalds printk(KERN_WARNING 12191da177e4SLinus Torvalds "NFS: Server wrote zero bytes, expected %u.\n", 12201da177e4SLinus Torvalds argp->count); 12211da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 12221da177e4SLinus Torvalds } 12231da177e4SLinus Torvalds /* Can't do anything about it except throw an error. */ 12241da177e4SLinus Torvalds task->tk_status = -EIO; 12251da177e4SLinus Torvalds } 1226788e7a89STrond Myklebust return 0; 12271da177e4SLinus Torvalds } 12281da177e4SLinus Torvalds 12291da177e4SLinus Torvalds 12301da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 12310aa05887SH Hartley Sweeten static void nfs_commitdata_release(void *data) 12321da177e4SLinus Torvalds { 1233383ba719STrond Myklebust struct nfs_write_data *wdata = data; 1234383ba719STrond Myklebust 1235383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 12361da177e4SLinus Torvalds nfs_commit_free(wdata); 12371da177e4SLinus Torvalds } 12381da177e4SLinus Torvalds 12391da177e4SLinus Torvalds /* 12401da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 12411da177e4SLinus Torvalds */ 1242dbae4c73STrond Myklebust static int nfs_commit_rpcsetup(struct list_head *head, 1243788e7a89STrond Myklebust struct nfs_write_data *data, 1244788e7a89STrond Myklebust int how) 12451da177e4SLinus Torvalds { 124684115e1cSTrond Myklebust struct nfs_page *first = nfs_list_entry(head->next); 124784115e1cSTrond Myklebust struct inode *inode = first->wb_context->path.dentry->d_inode; 124884115e1cSTrond Myklebust int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 12493ff7576dSTrond Myklebust int priority = flush_task_priority(how); 125007737691STrond Myklebust struct rpc_task *task; 1251bdc7f021STrond Myklebust struct rpc_message msg = { 1252bdc7f021STrond Myklebust .rpc_argp = &data->args, 1253bdc7f021STrond Myklebust .rpc_resp = &data->res, 1254bdc7f021STrond Myklebust .rpc_cred = first->wb_context->cred, 1255bdc7f021STrond Myklebust }; 125684115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 125707737691STrond Myklebust .task = &data->task, 125884115e1cSTrond Myklebust .rpc_client = NFS_CLIENT(inode), 1259bdc7f021STrond Myklebust .rpc_message = &msg, 126084115e1cSTrond Myklebust .callback_ops = &nfs_commit_ops, 126184115e1cSTrond Myklebust .callback_data = data, 1262101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 126384115e1cSTrond Myklebust .flags = flags, 12643ff7576dSTrond Myklebust .priority = priority, 126584115e1cSTrond Myklebust }; 12661da177e4SLinus Torvalds 12671da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 12681da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 12691da177e4SLinus Torvalds 12701da177e4SLinus Torvalds list_splice_init(head, &data->pages); 12711da177e4SLinus Torvalds 12721da177e4SLinus Torvalds data->inode = inode; 1273bdc7f021STrond Myklebust data->cred = msg.rpc_cred; 12741da177e4SLinus Torvalds 12751da177e4SLinus Torvalds data->args.fh = NFS_FH(data->inode); 12763da28eb1STrond Myklebust /* Note: we always request a commit of the entire inode */ 12773da28eb1STrond Myklebust data->args.offset = 0; 12783da28eb1STrond Myklebust data->args.count = 0; 1279383ba719STrond Myklebust data->args.context = get_nfs_open_context(first->wb_context); 12803da28eb1STrond Myklebust data->res.count = 0; 12811da177e4SLinus Torvalds data->res.fattr = &data->fattr; 12821da177e4SLinus Torvalds data->res.verf = &data->verf; 12830e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 12841da177e4SLinus Torvalds 1285788e7a89STrond Myklebust /* Set up the initial task struct. */ 1286bdc7f021STrond Myklebust NFS_PROTO(inode)->commit_setup(data, &msg); 12871da177e4SLinus Torvalds 1288a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1289bdc7f021STrond Myklebust 129007737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1291dbae4c73STrond Myklebust if (IS_ERR(task)) 1292dbae4c73STrond Myklebust return PTR_ERR(task); 129307737691STrond Myklebust rpc_put_task(task); 1294dbae4c73STrond Myklebust return 0; 12951da177e4SLinus Torvalds } 12961da177e4SLinus Torvalds 12971da177e4SLinus Torvalds /* 12981da177e4SLinus Torvalds * Commit dirty pages 12991da177e4SLinus Torvalds */ 13001da177e4SLinus Torvalds static int 130140859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how) 13021da177e4SLinus Torvalds { 13031da177e4SLinus Torvalds struct nfs_write_data *data; 13041da177e4SLinus Torvalds struct nfs_page *req; 13051da177e4SLinus Torvalds 1306c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 13071da177e4SLinus Torvalds 13081da177e4SLinus Torvalds if (!data) 13091da177e4SLinus Torvalds goto out_bad; 13101da177e4SLinus Torvalds 13111da177e4SLinus Torvalds /* Set up the argument struct */ 1312dbae4c73STrond Myklebust return nfs_commit_rpcsetup(head, data, how); 13131da177e4SLinus Torvalds out_bad: 13141da177e4SLinus Torvalds while (!list_empty(head)) { 13151da177e4SLinus Torvalds req = nfs_list_entry(head->next); 13161da177e4SLinus Torvalds nfs_list_remove_request(req); 13171da177e4SLinus Torvalds nfs_mark_request_commit(req); 131883715ad5STrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1319c9e51e41SPeter Zijlstra dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 1320c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 13219fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 13221da177e4SLinus Torvalds } 13231da177e4SLinus Torvalds return -ENOMEM; 13241da177e4SLinus Torvalds } 13251da177e4SLinus Torvalds 13261da177e4SLinus Torvalds /* 13271da177e4SLinus Torvalds * COMMIT call returned 13281da177e4SLinus Torvalds */ 1329788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 13301da177e4SLinus Torvalds { 1331963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 13321da177e4SLinus Torvalds 1333a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 13341da177e4SLinus Torvalds task->tk_pid, task->tk_status); 13351da177e4SLinus Torvalds 1336788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1337788e7a89STrond Myklebust if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) 1338788e7a89STrond Myklebust return; 1339c9d8f89dSTrond Myklebust } 1340c9d8f89dSTrond Myklebust 1341c9d8f89dSTrond Myklebust static void nfs_commit_release(void *calldata) 1342c9d8f89dSTrond Myklebust { 1343c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1344c9d8f89dSTrond Myklebust struct nfs_page *req; 1345c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1346788e7a89STrond Myklebust 13471da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 13481da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 13491da177e4SLinus Torvalds nfs_list_remove_request(req); 1350e468bae9STrond Myklebust nfs_clear_request_commit(req); 13511da177e4SLinus Torvalds 135248186c7dSChuck Lever dprintk("NFS: commit (%s/%lld %d@%lld)", 135388be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 135488be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 13551da177e4SLinus Torvalds req->wb_bytes, 13561da177e4SLinus Torvalds (long long)req_offset(req)); 1357c9d8f89dSTrond Myklebust if (status < 0) { 1358c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 13591da177e4SLinus Torvalds nfs_inode_remove_request(req); 1360c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 13611da177e4SLinus Torvalds goto next; 13621da177e4SLinus Torvalds } 13631da177e4SLinus Torvalds 13641da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 13651da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 13661da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 13671da177e4SLinus Torvalds /* We have a match */ 13681da177e4SLinus Torvalds nfs_inode_remove_request(req); 13691da177e4SLinus Torvalds dprintk(" OK\n"); 13701da177e4SLinus Torvalds goto next; 13711da177e4SLinus Torvalds } 13721da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 13731da177e4SLinus Torvalds dprintk(" mismatch\n"); 13746d884e8fSFred nfs_mark_request_dirty(req); 13751da177e4SLinus Torvalds next: 13769fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 13771da177e4SLinus Torvalds } 1378c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 13791da177e4SLinus Torvalds } 1380788e7a89STrond Myklebust 1381788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 138221d9a851SAndy Adamson #if defined(CONFIG_NFS_V4_1) 138321d9a851SAndy Adamson .rpc_call_prepare = nfs_write_prepare, 138421d9a851SAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1385788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1386788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1387788e7a89STrond Myklebust }; 13881da177e4SLinus Torvalds 13898fc795f7STrond Myklebust static int nfs_commit_inode(struct inode *inode, int how) 13901da177e4SLinus Torvalds { 13911da177e4SLinus Torvalds LIST_HEAD(head); 13927d46a49fSTrond Myklebust int res; 13931da177e4SLinus Torvalds 1394587142f8STrond Myklebust spin_lock(&inode->i_lock); 13953da28eb1STrond Myklebust res = nfs_scan_commit(inode, &head, 0, 0); 1396587142f8STrond Myklebust spin_unlock(&inode->i_lock); 13971da177e4SLinus Torvalds if (res) { 13987d46a49fSTrond Myklebust int error = nfs_commit_list(inode, &head, how); 13991da177e4SLinus Torvalds if (error < 0) 14001da177e4SLinus Torvalds return error; 14013da28eb1STrond Myklebust } 14021da177e4SLinus Torvalds return res; 14031da177e4SLinus Torvalds } 14048fc795f7STrond Myklebust 14058fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 14068fc795f7STrond Myklebust { 1407420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1408420e3646STrond Myklebust int flags = FLUSH_SYNC; 1409420e3646STrond Myklebust int ret = 0; 14108fc795f7STrond Myklebust 1411420e3646STrond Myklebust /* Don't commit yet if this is a non-blocking flush and there are 1412420e3646STrond Myklebust * lots of outstanding writes for this mapping. 1413420e3646STrond Myklebust */ 1414420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE && 1415420e3646STrond Myklebust nfsi->ncommit <= (nfsi->npages >> 1)) 1416420e3646STrond Myklebust goto out_mark_dirty; 1417420e3646STrond Myklebust 14185bad5abeSTrond Myklebust if (wbc->nonblocking || wbc->for_background) 1419420e3646STrond Myklebust flags = 0; 1420420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1421420e3646STrond Myklebust if (ret >= 0) { 1422420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1423420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1424420e3646STrond Myklebust wbc->nr_to_write -= ret; 1425420e3646STrond Myklebust else 1426420e3646STrond Myklebust wbc->nr_to_write = 0; 1427420e3646STrond Myklebust } 14288fc795f7STrond Myklebust return 0; 1429420e3646STrond Myklebust } 1430420e3646STrond Myklebust out_mark_dirty: 14318fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 14328fc795f7STrond Myklebust return ret; 14338fc795f7STrond Myklebust } 1434c63c7b05STrond Myklebust #else 1435c63c7b05STrond Myklebust static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how) 1436c63c7b05STrond Myklebust { 1437c63c7b05STrond Myklebust return 0; 1438c63c7b05STrond Myklebust } 14398fc795f7STrond Myklebust 14408fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 14418fc795f7STrond Myklebust { 14428fc795f7STrond Myklebust return 0; 14438fc795f7STrond Myklebust } 14441da177e4SLinus Torvalds #endif 14451da177e4SLinus Torvalds 14468fc795f7STrond Myklebust int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 14478fc795f7STrond Myklebust { 14488fc795f7STrond Myklebust return nfs_commit_unstable_pages(inode, wbc); 14498fc795f7STrond Myklebust } 14508fc795f7STrond Myklebust 14511c75950bSTrond Myklebust long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how) 14521da177e4SLinus Torvalds { 14531c75950bSTrond Myklebust struct inode *inode = mapping->host; 1454ca52fec1STrond Myklebust pgoff_t idx_start, idx_end; 14551c75950bSTrond Myklebust unsigned int npages = 0; 1456c42de9ddSTrond Myklebust LIST_HEAD(head); 145770b9ecbdSTrond Myklebust int nocommit = how & FLUSH_NOCOMMIT; 14583f442547STrond Myklebust long pages, ret; 14591da177e4SLinus Torvalds 14601c75950bSTrond Myklebust /* FIXME */ 14611c75950bSTrond Myklebust if (wbc->range_cyclic) 14621c75950bSTrond Myklebust idx_start = 0; 14631c75950bSTrond Myklebust else { 14641c75950bSTrond Myklebust idx_start = wbc->range_start >> PAGE_CACHE_SHIFT; 14651c75950bSTrond Myklebust idx_end = wbc->range_end >> PAGE_CACHE_SHIFT; 14661c75950bSTrond Myklebust if (idx_end > idx_start) { 1467ca52fec1STrond Myklebust pgoff_t l_npages = 1 + idx_end - idx_start; 14681c75950bSTrond Myklebust npages = l_npages; 14691c75950bSTrond Myklebust if (sizeof(npages) != sizeof(l_npages) && 1470ca52fec1STrond Myklebust (pgoff_t)npages != l_npages) 14711c75950bSTrond Myklebust npages = 0; 14721c75950bSTrond Myklebust } 14731c75950bSTrond Myklebust } 1474c42de9ddSTrond Myklebust how &= ~FLUSH_NOCOMMIT; 1475587142f8STrond Myklebust spin_lock(&inode->i_lock); 14761da177e4SLinus Torvalds do { 1477c42de9ddSTrond Myklebust ret = nfs_wait_on_requests_locked(inode, idx_start, npages); 1478c42de9ddSTrond Myklebust if (ret != 0) 1479c42de9ddSTrond Myklebust continue; 1480c42de9ddSTrond Myklebust if (nocommit) 1481c42de9ddSTrond Myklebust break; 1482d2ccddf0STrond Myklebust pages = nfs_scan_commit(inode, &head, idx_start, npages); 1483724c439cSTrond Myklebust if (pages == 0) 1484c42de9ddSTrond Myklebust break; 1485d2ccddf0STrond Myklebust pages += nfs_scan_commit(inode, &head, 0, 0); 1486587142f8STrond Myklebust spin_unlock(&inode->i_lock); 1487c42de9ddSTrond Myklebust ret = nfs_commit_list(inode, &head, how); 1488587142f8STrond Myklebust spin_lock(&inode->i_lock); 1489587142f8STrond Myklebust 1490c42de9ddSTrond Myklebust } while (ret >= 0); 1491587142f8STrond Myklebust spin_unlock(&inode->i_lock); 1492c42de9ddSTrond Myklebust return ret; 14931da177e4SLinus Torvalds } 14941da177e4SLinus Torvalds 149534901f70STrond Myklebust static int __nfs_write_mapping(struct address_space *mapping, struct writeback_control *wbc, int how) 14961c75950bSTrond Myklebust { 14971c75950bSTrond Myklebust int ret; 14981c75950bSTrond Myklebust 149934901f70STrond Myklebust ret = nfs_writepages(mapping, wbc); 150061822ab5STrond Myklebust if (ret < 0) 150161822ab5STrond Myklebust goto out; 150234901f70STrond Myklebust ret = nfs_sync_mapping_wait(mapping, wbc, how); 1503ed90ef51STrond Myklebust if (ret < 0) 1504ed90ef51STrond Myklebust goto out; 15051c75950bSTrond Myklebust return 0; 150661822ab5STrond Myklebust out: 1507e507d9ebSTrond Myklebust __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 15081c75950bSTrond Myklebust return ret; 15091c75950bSTrond Myklebust } 15101c75950bSTrond Myklebust 151134901f70STrond Myklebust /* Two pass sync: first using WB_SYNC_NONE, then WB_SYNC_ALL */ 151234901f70STrond Myklebust static int nfs_write_mapping(struct address_space *mapping, int how) 151334901f70STrond Myklebust { 151434901f70STrond Myklebust struct writeback_control wbc = { 151534901f70STrond Myklebust .bdi = mapping->backing_dev_info, 151672cb77f4STrond Myklebust .sync_mode = WB_SYNC_ALL, 151734901f70STrond Myklebust .nr_to_write = LONG_MAX, 1518d7fb1207STrond Myklebust .range_start = 0, 1519d7fb1207STrond Myklebust .range_end = LLONG_MAX, 152034901f70STrond Myklebust }; 152134901f70STrond Myklebust 152234901f70STrond Myklebust return __nfs_write_mapping(mapping, &wbc, how); 152334901f70STrond Myklebust } 152434901f70STrond Myklebust 1525ed90ef51STrond Myklebust /* 1526ed90ef51STrond Myklebust * flush the inode to disk. 1527ed90ef51STrond Myklebust */ 1528ed90ef51STrond Myklebust int nfs_wb_all(struct inode *inode) 15291c75950bSTrond Myklebust { 1530ed90ef51STrond Myklebust return nfs_write_mapping(inode->i_mapping, 0); 1531ed90ef51STrond Myklebust } 15321c75950bSTrond Myklebust 1533ed90ef51STrond Myklebust int nfs_wb_nocommit(struct inode *inode) 1534ed90ef51STrond Myklebust { 1535ed90ef51STrond Myklebust return nfs_write_mapping(inode->i_mapping, FLUSH_NOCOMMIT); 15361c75950bSTrond Myklebust } 15371c75950bSTrond Myklebust 15381b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 15391b3b4a1aSTrond Myklebust { 15401b3b4a1aSTrond Myklebust struct nfs_page *req; 15411b3b4a1aSTrond Myklebust int ret = 0; 15421b3b4a1aSTrond Myklebust 15431b3b4a1aSTrond Myklebust BUG_ON(!PageLocked(page)); 15441b3b4a1aSTrond Myklebust for (;;) { 15451b3b4a1aSTrond Myklebust req = nfs_page_find_request(page); 15461b3b4a1aSTrond Myklebust if (req == NULL) 15471b3b4a1aSTrond Myklebust break; 15481b3b4a1aSTrond Myklebust if (nfs_lock_request_dontget(req)) { 15491b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 15501b3b4a1aSTrond Myklebust /* 15511b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 15521b3b4a1aSTrond Myklebust * page as being dirty 15531b3b4a1aSTrond Myklebust */ 15541b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 15551b3b4a1aSTrond Myklebust nfs_unlock_request(req); 15561b3b4a1aSTrond Myklebust break; 15571b3b4a1aSTrond Myklebust } 15581b3b4a1aSTrond Myklebust ret = nfs_wait_on_request(req); 1559c9edda71STrond Myklebust nfs_release_request(req); 15601b3b4a1aSTrond Myklebust if (ret < 0) 1561*c988950eSTrond Myklebust break; 15621b3b4a1aSTrond Myklebust } 15631b3b4a1aSTrond Myklebust return ret; 15641b3b4a1aSTrond Myklebust } 15651b3b4a1aSTrond Myklebust 15665334eb13SAdrian Bunk static int nfs_wb_page_priority(struct inode *inode, struct page *page, 15675334eb13SAdrian Bunk int how) 15681c75950bSTrond Myklebust { 15691c75950bSTrond Myklebust loff_t range_start = page_offset(page); 15701c75950bSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 15714d770ccfSTrond Myklebust struct writeback_control wbc = { 15724d770ccfSTrond Myklebust .bdi = page->mapping->backing_dev_info, 15734d770ccfSTrond Myklebust .sync_mode = WB_SYNC_ALL, 15744d770ccfSTrond Myklebust .nr_to_write = LONG_MAX, 15754d770ccfSTrond Myklebust .range_start = range_start, 15764d770ccfSTrond Myklebust .range_end = range_end, 15774d770ccfSTrond Myklebust }; 15784d770ccfSTrond Myklebust int ret; 15791c75950bSTrond Myklebust 158073e3302fSTrond Myklebust do { 1581c63c7b05STrond Myklebust if (clear_page_dirty_for_io(page)) { 15824d770ccfSTrond Myklebust ret = nfs_writepage_locked(page, &wbc); 15834d770ccfSTrond Myklebust if (ret < 0) 158473e3302fSTrond Myklebust goto out_error; 158573e3302fSTrond Myklebust } else if (!PagePrivate(page)) 158673e3302fSTrond Myklebust break; 15874d770ccfSTrond Myklebust ret = nfs_sync_mapping_wait(page->mapping, &wbc, how); 158873e3302fSTrond Myklebust if (ret < 0) 158973e3302fSTrond Myklebust goto out_error; 159073e3302fSTrond Myklebust } while (PagePrivate(page)); 15914d770ccfSTrond Myklebust return 0; 159273e3302fSTrond Myklebust out_error: 1593e507d9ebSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_PAGES); 15944d770ccfSTrond Myklebust return ret; 15951c75950bSTrond Myklebust } 15961c75950bSTrond Myklebust 15971c75950bSTrond Myklebust /* 15981c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 15991c75950bSTrond Myklebust */ 16001c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page* page) 16011c75950bSTrond Myklebust { 16024d770ccfSTrond Myklebust return nfs_wb_page_priority(inode, page, FLUSH_STABLE); 16031c75950bSTrond Myklebust } 16041c75950bSTrond Myklebust 1605074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 1606074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 1607074cc1deSTrond Myklebust struct page *page) 1608074cc1deSTrond Myklebust { 1609074cc1deSTrond Myklebust struct nfs_page *req; 1610074cc1deSTrond Myklebust int ret; 1611074cc1deSTrond Myklebust 1612074cc1deSTrond Myklebust nfs_fscache_release_page(page, GFP_KERNEL); 1613074cc1deSTrond Myklebust 1614074cc1deSTrond Myklebust req = nfs_find_and_lock_request(page); 1615074cc1deSTrond Myklebust ret = PTR_ERR(req); 1616074cc1deSTrond Myklebust if (IS_ERR(req)) 1617074cc1deSTrond Myklebust goto out; 1618074cc1deSTrond Myklebust 1619074cc1deSTrond Myklebust ret = migrate_page(mapping, newpage, page); 1620074cc1deSTrond Myklebust if (!req) 1621074cc1deSTrond Myklebust goto out; 1622074cc1deSTrond Myklebust if (ret) 1623074cc1deSTrond Myklebust goto out_unlock; 1624074cc1deSTrond Myklebust page_cache_get(newpage); 1625190f38e5STrond Myklebust spin_lock(&mapping->host->i_lock); 1626074cc1deSTrond Myklebust req->wb_page = newpage; 1627074cc1deSTrond Myklebust SetPagePrivate(newpage); 1628190f38e5STrond Myklebust set_page_private(newpage, (unsigned long)req); 1629074cc1deSTrond Myklebust ClearPagePrivate(page); 1630074cc1deSTrond Myklebust set_page_private(page, 0); 1631190f38e5STrond Myklebust spin_unlock(&mapping->host->i_lock); 1632074cc1deSTrond Myklebust page_cache_release(page); 1633074cc1deSTrond Myklebust out_unlock: 1634074cc1deSTrond Myklebust nfs_clear_page_tag_locked(req); 1635074cc1deSTrond Myklebust out: 1636074cc1deSTrond Myklebust return ret; 1637074cc1deSTrond Myklebust } 1638074cc1deSTrond Myklebust #endif 1639074cc1deSTrond Myklebust 1640f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 16411da177e4SLinus Torvalds { 16421da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 16431da177e4SLinus Torvalds sizeof(struct nfs_write_data), 16441da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 164520c2df83SPaul Mundt NULL); 16461da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 16471da177e4SLinus Torvalds return -ENOMEM; 16481da177e4SLinus Torvalds 164993d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 16501da177e4SLinus Torvalds nfs_wdata_cachep); 16511da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 16521da177e4SLinus Torvalds return -ENOMEM; 16531da177e4SLinus Torvalds 165493d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 16551da177e4SLinus Torvalds nfs_wdata_cachep); 16561da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 16571da177e4SLinus Torvalds return -ENOMEM; 16581da177e4SLinus Torvalds 165989a09141SPeter Zijlstra /* 166089a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 166189a09141SPeter Zijlstra * 166289a09141SPeter Zijlstra * 64MB: 8192k 166389a09141SPeter Zijlstra * 128MB: 11585k 166489a09141SPeter Zijlstra * 256MB: 16384k 166589a09141SPeter Zijlstra * 512MB: 23170k 166689a09141SPeter Zijlstra * 1GB: 32768k 166789a09141SPeter Zijlstra * 2GB: 46340k 166889a09141SPeter Zijlstra * 4GB: 65536k 166989a09141SPeter Zijlstra * 8GB: 92681k 167089a09141SPeter Zijlstra * 16GB: 131072k 167189a09141SPeter Zijlstra * 167289a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 167389a09141SPeter Zijlstra * Limit the default to 256M 167489a09141SPeter Zijlstra */ 167589a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 167689a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 167789a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 167889a09141SPeter Zijlstra 16791da177e4SLinus Torvalds return 0; 16801da177e4SLinus Torvalds } 16811da177e4SLinus Torvalds 1682266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 16831da177e4SLinus Torvalds { 16841da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 16851da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 16861a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 16871da177e4SLinus Torvalds } 16881da177e4SLinus Torvalds 1689