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); 441587142f8STrond Myklebust spin_unlock(&inode->i_lock); 442fd39fc85SChristoph Lameter inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 443c9e51e41SPeter Zijlstra inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); 444a1803044STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 4451da177e4SLinus Torvalds } 4468e821cadSTrond Myklebust 447e468bae9STrond Myklebust static int 448e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 449e468bae9STrond Myklebust { 450e468bae9STrond Myklebust struct page *page = req->wb_page; 451e468bae9STrond Myklebust 452e468bae9STrond Myklebust if (test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) { 453e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 454e468bae9STrond Myklebust dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE); 455e468bae9STrond Myklebust return 1; 456e468bae9STrond Myklebust } 457e468bae9STrond Myklebust return 0; 458e468bae9STrond Myklebust } 459e468bae9STrond Myklebust 4608e821cadSTrond Myklebust static inline 4618e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4628e821cadSTrond Myklebust { 4638e821cadSTrond Myklebust return data->verf.committed != NFS_FILE_SYNC; 4648e821cadSTrond Myklebust } 4658e821cadSTrond Myklebust 4668e821cadSTrond Myklebust static inline 4678e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req) 4688e821cadSTrond Myklebust { 469e468bae9STrond Myklebust if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) { 4708e821cadSTrond Myklebust nfs_mark_request_commit(req); 4718e821cadSTrond Myklebust return 1; 4728e821cadSTrond Myklebust } 4738e821cadSTrond Myklebust if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) { 4746d884e8fSFred nfs_mark_request_dirty(req); 4758e821cadSTrond Myklebust return 1; 4768e821cadSTrond Myklebust } 4778e821cadSTrond Myklebust return 0; 4788e821cadSTrond Myklebust } 4798e821cadSTrond Myklebust #else 4808e821cadSTrond Myklebust static inline void 4818e821cadSTrond Myklebust nfs_mark_request_commit(struct nfs_page *req) 4828e821cadSTrond Myklebust { 4838e821cadSTrond Myklebust } 4848e821cadSTrond Myklebust 485e468bae9STrond Myklebust static inline int 486e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 487e468bae9STrond Myklebust { 488e468bae9STrond Myklebust return 0; 489e468bae9STrond Myklebust } 490e468bae9STrond Myklebust 4918e821cadSTrond Myklebust static inline 4928e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4938e821cadSTrond Myklebust { 4948e821cadSTrond Myklebust return 0; 4958e821cadSTrond Myklebust } 4968e821cadSTrond Myklebust 4978e821cadSTrond Myklebust static inline 4988e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req) 4998e821cadSTrond Myklebust { 5008e821cadSTrond Myklebust return 0; 5018e821cadSTrond Myklebust } 5021da177e4SLinus Torvalds #endif 5031da177e4SLinus Torvalds 5041da177e4SLinus Torvalds /* 5051da177e4SLinus Torvalds * Wait for a request to complete. 5061da177e4SLinus Torvalds * 507150030b7SMatthew Wilcox * Interruptible by fatal signals only. 5081da177e4SLinus Torvalds */ 509ca52fec1STrond Myklebust static int nfs_wait_on_requests_locked(struct inode *inode, pgoff_t idx_start, unsigned int npages) 5101da177e4SLinus Torvalds { 5111da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5121da177e4SLinus Torvalds struct nfs_page *req; 513ca52fec1STrond Myklebust pgoff_t idx_end, next; 5141da177e4SLinus Torvalds unsigned int res = 0; 5151da177e4SLinus Torvalds int error; 5161da177e4SLinus Torvalds 5171da177e4SLinus Torvalds if (npages == 0) 5181da177e4SLinus Torvalds idx_end = ~0; 5191da177e4SLinus Torvalds else 5201da177e4SLinus Torvalds idx_end = idx_start + npages - 1; 5211da177e4SLinus Torvalds 5221da177e4SLinus Torvalds next = idx_start; 5239fd367f0STrond Myklebust while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_LOCKED)) { 5241da177e4SLinus Torvalds if (req->wb_index > idx_end) 5251da177e4SLinus Torvalds break; 5261da177e4SLinus Torvalds 5271da177e4SLinus Torvalds next = req->wb_index + 1; 528c6a556b8STrond Myklebust BUG_ON(!NFS_WBACK_BUSY(req)); 5291da177e4SLinus Torvalds 530c03b4024STrond Myklebust kref_get(&req->wb_kref); 531587142f8STrond Myklebust spin_unlock(&inode->i_lock); 5321da177e4SLinus Torvalds error = nfs_wait_on_request(req); 5331da177e4SLinus Torvalds nfs_release_request(req); 534587142f8STrond Myklebust spin_lock(&inode->i_lock); 5351da177e4SLinus Torvalds if (error < 0) 5361da177e4SLinus Torvalds return error; 5371da177e4SLinus Torvalds res++; 5381da177e4SLinus Torvalds } 5391da177e4SLinus Torvalds return res; 5401da177e4SLinus Torvalds } 5411da177e4SLinus Torvalds 54283715ad5STrond Myklebust static void nfs_cancel_commit_list(struct list_head *head) 54383715ad5STrond Myklebust { 54483715ad5STrond Myklebust struct nfs_page *req; 54583715ad5STrond Myklebust 54683715ad5STrond Myklebust while(!list_empty(head)) { 54783715ad5STrond Myklebust req = nfs_list_entry(head->next); 54883715ad5STrond Myklebust nfs_list_remove_request(req); 549e468bae9STrond Myklebust nfs_clear_request_commit(req); 55083715ad5STrond Myklebust nfs_inode_remove_request(req); 551b6dff26aSTrond Myklebust nfs_unlock_request(req); 55283715ad5STrond Myklebust } 55383715ad5STrond Myklebust } 55483715ad5STrond Myklebust 55547c62564STrond Myklebust #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 556fb8a1f11STrond Myklebust static int 557fb8a1f11STrond Myklebust nfs_need_commit(struct nfs_inode *nfsi) 558fb8a1f11STrond Myklebust { 559fb8a1f11STrond Myklebust return radix_tree_tagged(&nfsi->nfs_page_tree, NFS_PAGE_TAG_COMMIT); 560fb8a1f11STrond Myklebust } 561fb8a1f11STrond Myklebust 5621da177e4SLinus Torvalds /* 5631da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 5641da177e4SLinus Torvalds * @inode: NFS inode to scan 5651da177e4SLinus Torvalds * @dst: destination list 5661da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 5671da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 5681da177e4SLinus Torvalds * 5691da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 5701da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 5711da177e4SLinus Torvalds */ 5721da177e4SLinus Torvalds static int 573ca52fec1STrond Myklebust nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 5741da177e4SLinus Torvalds { 5751da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5763da28eb1STrond Myklebust 577fb8a1f11STrond Myklebust if (!nfs_need_commit(nfsi)) 578fb8a1f11STrond Myklebust return 0; 579fb8a1f11STrond Myklebust 580fb8a1f11STrond Myklebust return nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT); 5811da177e4SLinus Torvalds } 582c42de9ddSTrond Myklebust #else 583fb8a1f11STrond Myklebust static inline int nfs_need_commit(struct nfs_inode *nfsi) 584fb8a1f11STrond Myklebust { 585fb8a1f11STrond Myklebust return 0; 586fb8a1f11STrond Myklebust } 587fb8a1f11STrond Myklebust 588ca52fec1STrond Myklebust static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 589c42de9ddSTrond Myklebust { 590c42de9ddSTrond Myklebust return 0; 591c42de9ddSTrond Myklebust } 5921da177e4SLinus Torvalds #endif 5931da177e4SLinus Torvalds 5941da177e4SLinus Torvalds /* 595e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 596e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 5971da177e4SLinus Torvalds * 598e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 599e7d39069STrond Myklebust * to disk. 6001da177e4SLinus Torvalds */ 601e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 602e7d39069STrond Myklebust struct page *page, 603e7d39069STrond Myklebust unsigned int offset, 604e7d39069STrond Myklebust unsigned int bytes) 6051da177e4SLinus Torvalds { 606e7d39069STrond Myklebust struct nfs_page *req; 607e7d39069STrond Myklebust unsigned int rqend; 608e7d39069STrond Myklebust unsigned int end; 6091da177e4SLinus Torvalds int error; 610277459d2STrond Myklebust 611e7d39069STrond Myklebust if (!PagePrivate(page)) 612e7d39069STrond Myklebust return NULL; 613e7d39069STrond Myklebust 614e7d39069STrond Myklebust end = offset + bytes; 615e7d39069STrond Myklebust spin_lock(&inode->i_lock); 616e7d39069STrond Myklebust 617e7d39069STrond Myklebust for (;;) { 618e7d39069STrond Myklebust req = nfs_page_find_request_locked(page); 619e7d39069STrond Myklebust if (req == NULL) 620e7d39069STrond Myklebust goto out_unlock; 621e7d39069STrond Myklebust 622e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 623e7d39069STrond Myklebust /* 624e7d39069STrond Myklebust * Tell the caller to flush out the request if 625e7d39069STrond Myklebust * the offsets are non-contiguous. 626e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 627e7d39069STrond Myklebust * have flushed out requests having wrong owners. 628e7d39069STrond Myklebust */ 629e468bae9STrond Myklebust if (offset > rqend 630e7d39069STrond Myklebust || end < req->wb_offset) 631e7d39069STrond Myklebust goto out_flushme; 632e7d39069STrond Myklebust 633e7d39069STrond Myklebust if (nfs_set_page_tag_locked(req)) 634e7d39069STrond Myklebust break; 635e7d39069STrond Myklebust 636e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 637587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6381da177e4SLinus Torvalds error = nfs_wait_on_request(req); 6391da177e4SLinus Torvalds nfs_release_request(req); 640e7d39069STrond Myklebust if (error != 0) 641e7d39069STrond Myklebust goto out_err; 642e7d39069STrond Myklebust spin_lock(&inode->i_lock); 6431da177e4SLinus Torvalds } 6441da177e4SLinus Torvalds 645e468bae9STrond Myklebust if (nfs_clear_request_commit(req)) 646e468bae9STrond Myklebust radix_tree_tag_clear(&NFS_I(inode)->nfs_page_tree, 647e468bae9STrond Myklebust req->wb_index, NFS_PAGE_TAG_COMMIT); 648e468bae9STrond Myklebust 6491da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 6501da177e4SLinus Torvalds if (offset < req->wb_offset) { 6511da177e4SLinus Torvalds req->wb_offset = offset; 6521da177e4SLinus Torvalds req->wb_pgbase = offset; 6531da177e4SLinus Torvalds } 6541da177e4SLinus Torvalds if (end > rqend) 6551da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 656e7d39069STrond Myklebust else 657e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 658e7d39069STrond Myklebust out_unlock: 659e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 660e7d39069STrond Myklebust return req; 661e7d39069STrond Myklebust out_flushme: 662e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 663e7d39069STrond Myklebust nfs_release_request(req); 664e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 665e7d39069STrond Myklebust out_err: 666e7d39069STrond Myklebust return ERR_PTR(error); 667e7d39069STrond Myklebust } 6681da177e4SLinus Torvalds 669e7d39069STrond Myklebust /* 670e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 671e7d39069STrond Myklebust * 672e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 673e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 674e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 675e7d39069STrond Myklebust */ 676e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 677e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 678e7d39069STrond Myklebust { 679e7d39069STrond Myklebust struct inode *inode = page->mapping->host; 680e7d39069STrond Myklebust struct nfs_page *req; 681e7d39069STrond Myklebust int error; 682e7d39069STrond Myklebust 683e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 684e7d39069STrond Myklebust if (req != NULL) 685e7d39069STrond Myklebust goto out; 686e7d39069STrond Myklebust req = nfs_create_request(ctx, inode, page, offset, bytes); 687e7d39069STrond Myklebust if (IS_ERR(req)) 688e7d39069STrond Myklebust goto out; 689e7d39069STrond Myklebust error = nfs_inode_add_request(inode, req); 690e7d39069STrond Myklebust if (error != 0) { 691e7d39069STrond Myklebust nfs_release_request(req); 692e7d39069STrond Myklebust req = ERR_PTR(error); 693e7d39069STrond Myklebust } 694efc91ed0STrond Myklebust out: 69561e930a9STrond Myklebust return req; 6961da177e4SLinus Torvalds } 6971da177e4SLinus Torvalds 698e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 699e7d39069STrond Myklebust unsigned int offset, unsigned int count) 700e7d39069STrond Myklebust { 701e7d39069STrond Myklebust struct nfs_page *req; 702e7d39069STrond Myklebust 703e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 704e7d39069STrond Myklebust if (IS_ERR(req)) 705e7d39069STrond Myklebust return PTR_ERR(req); 706e7d39069STrond Myklebust /* Update file length */ 707e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 708e7d39069STrond Myklebust nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 709e7d39069STrond Myklebust nfs_clear_page_tag_locked(req); 710e7d39069STrond Myklebust return 0; 711e7d39069STrond Myklebust } 712e7d39069STrond Myklebust 7131da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 7141da177e4SLinus Torvalds { 715cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 7161da177e4SLinus Torvalds struct nfs_page *req; 7171a54533eSTrond Myklebust int do_flush, status; 7181da177e4SLinus Torvalds /* 7191da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 7201da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 7211da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 7221da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 7231da177e4SLinus Torvalds * Also do the same if we find a request from an existing 7241da177e4SLinus Torvalds * dropped page. 7251da177e4SLinus Torvalds */ 7261a54533eSTrond Myklebust do { 727277459d2STrond Myklebust req = nfs_page_find_request(page); 7281a54533eSTrond Myklebust if (req == NULL) 7291a54533eSTrond Myklebust return 0; 730e468bae9STrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx; 7311da177e4SLinus Torvalds nfs_release_request(req); 7321a54533eSTrond Myklebust if (!do_flush) 7331a54533eSTrond Myklebust return 0; 734277459d2STrond Myklebust status = nfs_wb_page(page->mapping->host, page); 7351a54533eSTrond Myklebust } while (status == 0); 7361a54533eSTrond Myklebust return status; 7371da177e4SLinus Torvalds } 7381da177e4SLinus Torvalds 7391da177e4SLinus Torvalds /* 7405d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 7415d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 7425d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 7435d47a356STrond Myklebust */ 7445d47a356STrond Myklebust static int nfs_write_pageuptodate(struct page *page, struct inode *inode) 7455d47a356STrond Myklebust { 7465d47a356STrond Myklebust return PageUptodate(page) && 7475d47a356STrond Myklebust !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); 7485d47a356STrond Myklebust } 7495d47a356STrond Myklebust 7505d47a356STrond Myklebust /* 7511da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 7521da177e4SLinus Torvalds * 7531da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 7541da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 7551da177e4SLinus Torvalds */ 7561da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 7571da177e4SLinus Torvalds unsigned int offset, unsigned int count) 7581da177e4SLinus Torvalds { 759cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 7601da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 7611da177e4SLinus Torvalds int status = 0; 7621da177e4SLinus Torvalds 76391d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 76491d5b470SChuck Lever 76548186c7dSChuck Lever dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", 76601cce933SJosef "Jeff" Sipek file->f_path.dentry->d_parent->d_name.name, 76701cce933SJosef "Jeff" Sipek file->f_path.dentry->d_name.name, count, 7680bbacc40SChuck Lever (long long)(page_offset(page) + offset)); 7691da177e4SLinus Torvalds 7701da177e4SLinus Torvalds /* If we're not using byte range locks, and we know the page 7715d47a356STrond Myklebust * is up to date, it may be more efficient to extend the write 7725d47a356STrond Myklebust * to cover the entire page in order to avoid fragmentation 7735d47a356STrond Myklebust * inefficiencies. 7741da177e4SLinus Torvalds */ 7755d47a356STrond Myklebust if (nfs_write_pageuptodate(page, inode) && 7765d47a356STrond Myklebust inode->i_flock == NULL && 777*6b2f3d1fSChristoph Hellwig !(file->f_flags & O_DSYNC)) { 77849a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 7791da177e4SLinus Torvalds offset = 0; 7801da177e4SLinus Torvalds } 7811da177e4SLinus Torvalds 782e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 78303fa9e84STrond Myklebust if (status < 0) 78403fa9e84STrond Myklebust nfs_set_pageerror(page); 78503fa9e84STrond Myklebust else 786e261f51fSTrond Myklebust __set_page_dirty_nobuffers(page); 7871da177e4SLinus Torvalds 78848186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 7891da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 7901da177e4SLinus Torvalds return status; 7911da177e4SLinus Torvalds } 7921da177e4SLinus Torvalds 7931da177e4SLinus Torvalds static void nfs_writepage_release(struct nfs_page *req) 7941da177e4SLinus Torvalds { 7958e821cadSTrond Myklebust 7967e5f6146STrond Myklebust if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req)) { 79789a09141SPeter Zijlstra nfs_end_page_writeback(req->wb_page); 7981da177e4SLinus Torvalds nfs_inode_remove_request(req); 7998e821cadSTrond Myklebust } else 8008e821cadSTrond Myklebust nfs_end_page_writeback(req->wb_page); 8019fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 8021da177e4SLinus Torvalds } 8031da177e4SLinus Torvalds 8043ff7576dSTrond Myklebust static int flush_task_priority(int how) 8051da177e4SLinus Torvalds { 8061da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 8071da177e4SLinus Torvalds case FLUSH_HIGHPRI: 8081da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 8091da177e4SLinus Torvalds case FLUSH_LOWPRI: 8101da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 8111da177e4SLinus Torvalds } 8121da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 8131da177e4SLinus Torvalds } 8141da177e4SLinus Torvalds 8151da177e4SLinus Torvalds /* 8161da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 8171da177e4SLinus Torvalds */ 818dbae4c73STrond Myklebust static int nfs_write_rpcsetup(struct nfs_page *req, 8191da177e4SLinus Torvalds struct nfs_write_data *data, 820788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 8211da177e4SLinus Torvalds unsigned int count, unsigned int offset, 8221da177e4SLinus Torvalds int how) 8231da177e4SLinus Torvalds { 82484115e1cSTrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 82584115e1cSTrond Myklebust int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 8263ff7576dSTrond Myklebust int priority = flush_task_priority(how); 82707737691STrond Myklebust struct rpc_task *task; 828bdc7f021STrond Myklebust struct rpc_message msg = { 829bdc7f021STrond Myklebust .rpc_argp = &data->args, 830bdc7f021STrond Myklebust .rpc_resp = &data->res, 831bdc7f021STrond Myklebust .rpc_cred = req->wb_context->cred, 832bdc7f021STrond Myklebust }; 83384115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 83484115e1cSTrond Myklebust .rpc_client = NFS_CLIENT(inode), 83507737691STrond Myklebust .task = &data->task, 836bdc7f021STrond Myklebust .rpc_message = &msg, 83784115e1cSTrond Myklebust .callback_ops = call_ops, 83884115e1cSTrond Myklebust .callback_data = data, 839101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 84084115e1cSTrond Myklebust .flags = flags, 8413ff7576dSTrond Myklebust .priority = priority, 84284115e1cSTrond Myklebust }; 8431da177e4SLinus Torvalds 8441da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 8451da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 8461da177e4SLinus Torvalds 8471da177e4SLinus Torvalds data->req = req; 84888be9f99STrond Myklebust data->inode = inode = req->wb_context->path.dentry->d_inode; 849bdc7f021STrond Myklebust data->cred = msg.rpc_cred; 8501da177e4SLinus Torvalds 8511da177e4SLinus Torvalds data->args.fh = NFS_FH(inode); 8521da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 8531da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 8541da177e4SLinus Torvalds data->args.pages = data->pagevec; 8551da177e4SLinus Torvalds data->args.count = count; 856383ba719STrond Myklebust data->args.context = get_nfs_open_context(req->wb_context); 857bdc7f021STrond Myklebust data->args.stable = NFS_UNSTABLE; 858bdc7f021STrond Myklebust if (how & FLUSH_STABLE) { 859bdc7f021STrond Myklebust data->args.stable = NFS_DATA_SYNC; 860fb8a1f11STrond Myklebust if (!nfs_need_commit(NFS_I(inode))) 861bdc7f021STrond Myklebust data->args.stable = NFS_FILE_SYNC; 862bdc7f021STrond Myklebust } 8631da177e4SLinus Torvalds 8641da177e4SLinus Torvalds data->res.fattr = &data->fattr; 8651da177e4SLinus Torvalds data->res.count = count; 8661da177e4SLinus Torvalds data->res.verf = &data->verf; 8670e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 8681da177e4SLinus Torvalds 869788e7a89STrond Myklebust /* Set up the initial task struct. */ 870bdc7f021STrond Myklebust NFS_PROTO(inode)->write_setup(data, &msg); 8711da177e4SLinus Torvalds 872a3f565b1SChuck Lever dprintk("NFS: %5u initiated write call " 87348186c7dSChuck Lever "(req %s/%lld, %u bytes @ offset %llu)\n", 8740bbacc40SChuck Lever data->task.tk_pid, 8751da177e4SLinus Torvalds inode->i_sb->s_id, 8761da177e4SLinus Torvalds (long long)NFS_FILEID(inode), 8771da177e4SLinus Torvalds count, 8781da177e4SLinus Torvalds (unsigned long long)data->args.offset); 8791da177e4SLinus Torvalds 88007737691STrond Myklebust task = rpc_run_task(&task_setup_data); 881dbae4c73STrond Myklebust if (IS_ERR(task)) 882dbae4c73STrond Myklebust return PTR_ERR(task); 88307737691STrond Myklebust rpc_put_task(task); 884dbae4c73STrond Myklebust return 0; 8851da177e4SLinus Torvalds } 8861da177e4SLinus Torvalds 8876d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 8886d884e8fSFred * call this on each, which will prepare them to be retried on next 8896d884e8fSFred * writeback using standard nfs. 8906d884e8fSFred */ 8916d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 8926d884e8fSFred { 8936d884e8fSFred nfs_mark_request_dirty(req); 8946d884e8fSFred nfs_end_page_writeback(req->wb_page); 8956d884e8fSFred nfs_clear_page_tag_locked(req); 8966d884e8fSFred } 8976d884e8fSFred 8981da177e4SLinus Torvalds /* 8991da177e4SLinus Torvalds * Generate multiple small requests to write out a single 9001da177e4SLinus Torvalds * contiguous dirty area on one page. 9011da177e4SLinus Torvalds */ 9028d5658c9STrond Myklebust static int nfs_flush_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) 9031da177e4SLinus Torvalds { 9041da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 9051da177e4SLinus Torvalds struct page *page = req->wb_page; 9061da177e4SLinus Torvalds struct nfs_write_data *data; 907e9f7bee1STrond Myklebust size_t wsize = NFS_SERVER(inode)->wsize, nbytes; 908e9f7bee1STrond Myklebust unsigned int offset; 9091da177e4SLinus Torvalds int requests = 0; 910dbae4c73STrond Myklebust int ret = 0; 9111da177e4SLinus Torvalds LIST_HEAD(list); 9121da177e4SLinus Torvalds 9131da177e4SLinus Torvalds nfs_list_remove_request(req); 9141da177e4SLinus Torvalds 915bcb71bbaSTrond Myklebust nbytes = count; 916e9f7bee1STrond Myklebust do { 917e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 918e9f7bee1STrond Myklebust 9198d5658c9STrond Myklebust data = nfs_writedata_alloc(1); 9201da177e4SLinus Torvalds if (!data) 9211da177e4SLinus Torvalds goto out_bad; 9221da177e4SLinus Torvalds list_add(&data->pages, &list); 9231da177e4SLinus Torvalds requests++; 924e9f7bee1STrond Myklebust nbytes -= len; 925e9f7bee1STrond Myklebust } while (nbytes != 0); 9261da177e4SLinus Torvalds atomic_set(&req->wb_complete, requests); 9271da177e4SLinus Torvalds 9281da177e4SLinus Torvalds ClearPageError(page); 9291da177e4SLinus Torvalds offset = 0; 930bcb71bbaSTrond Myklebust nbytes = count; 9311da177e4SLinus Torvalds do { 932dbae4c73STrond Myklebust int ret2; 933dbae4c73STrond Myklebust 9341da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 9351da177e4SLinus Torvalds list_del_init(&data->pages); 9361da177e4SLinus Torvalds 9371da177e4SLinus Torvalds data->pagevec[0] = page; 9381da177e4SLinus Torvalds 939bcb71bbaSTrond Myklebust if (nbytes < wsize) 940bcb71bbaSTrond Myklebust wsize = nbytes; 941dbae4c73STrond Myklebust ret2 = nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 942788e7a89STrond Myklebust wsize, offset, how); 943dbae4c73STrond Myklebust if (ret == 0) 944dbae4c73STrond Myklebust ret = ret2; 9451da177e4SLinus Torvalds offset += wsize; 9461da177e4SLinus Torvalds nbytes -= wsize; 9471da177e4SLinus Torvalds } while (nbytes != 0); 9481da177e4SLinus Torvalds 949dbae4c73STrond Myklebust return ret; 9501da177e4SLinus Torvalds 9511da177e4SLinus Torvalds out_bad: 9521da177e4SLinus Torvalds while (!list_empty(&list)) { 9531da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 9541da177e4SLinus Torvalds list_del(&data->pages); 9558aca67f0STrond Myklebust nfs_writedata_release(data); 9561da177e4SLinus Torvalds } 95761822ab5STrond Myklebust nfs_redirty_request(req); 9581da177e4SLinus Torvalds return -ENOMEM; 9591da177e4SLinus Torvalds } 9601da177e4SLinus Torvalds 9611da177e4SLinus Torvalds /* 9621da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 9631da177e4SLinus Torvalds * The page must have been locked by the caller. 9641da177e4SLinus Torvalds * 9651da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 9661da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 9671da177e4SLinus Torvalds * that has been written but not committed. 9681da177e4SLinus Torvalds */ 9698d5658c9STrond Myklebust static int nfs_flush_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how) 9701da177e4SLinus Torvalds { 9711da177e4SLinus Torvalds struct nfs_page *req; 9721da177e4SLinus Torvalds struct page **pages; 9731da177e4SLinus Torvalds struct nfs_write_data *data; 9741da177e4SLinus Torvalds 9758d5658c9STrond Myklebust data = nfs_writedata_alloc(npages); 9761da177e4SLinus Torvalds if (!data) 9771da177e4SLinus Torvalds goto out_bad; 9781da177e4SLinus Torvalds 9791da177e4SLinus Torvalds pages = data->pagevec; 9801da177e4SLinus Torvalds while (!list_empty(head)) { 9811da177e4SLinus Torvalds req = nfs_list_entry(head->next); 9821da177e4SLinus Torvalds nfs_list_remove_request(req); 9831da177e4SLinus Torvalds nfs_list_add_request(req, &data->pages); 9841da177e4SLinus Torvalds ClearPageError(req->wb_page); 9851da177e4SLinus Torvalds *pages++ = req->wb_page; 9861da177e4SLinus Torvalds } 9871da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 9881da177e4SLinus Torvalds 9891da177e4SLinus Torvalds /* Set up the argument struct */ 990dbae4c73STrond Myklebust return nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how); 9911da177e4SLinus Torvalds out_bad: 9921da177e4SLinus Torvalds while (!list_empty(head)) { 99310afec90STrond Myklebust req = nfs_list_entry(head->next); 9941da177e4SLinus Torvalds nfs_list_remove_request(req); 99561822ab5STrond Myklebust nfs_redirty_request(req); 9961da177e4SLinus Torvalds } 9971da177e4SLinus Torvalds return -ENOMEM; 9981da177e4SLinus Torvalds } 9991da177e4SLinus Torvalds 1000c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1001c63c7b05STrond Myklebust struct inode *inode, int ioflags) 10021da177e4SLinus Torvalds { 1003bf4285e7SChuck Lever size_t wsize = NFS_SERVER(inode)->wsize; 10041da177e4SLinus Torvalds 1005bcb71bbaSTrond Myklebust if (wsize < PAGE_CACHE_SIZE) 1006c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags); 1007bcb71bbaSTrond Myklebust else 1008c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags); 10091da177e4SLinus Torvalds } 10101da177e4SLinus Torvalds 10111da177e4SLinus Torvalds /* 10121da177e4SLinus Torvalds * Handle a write reply that flushed part of a page. 10131da177e4SLinus Torvalds */ 1014788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) 10151da177e4SLinus Torvalds { 1016788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 10171da177e4SLinus Torvalds 101848186c7dSChuck Lever dprintk("NFS: %5u write(%s/%lld %d@%lld)", 101948186c7dSChuck Lever task->tk_pid, 102048186c7dSChuck Lever data->req->wb_context->path.dentry->d_inode->i_sb->s_id, 102148186c7dSChuck Lever (long long) 102248186c7dSChuck Lever NFS_FILEID(data->req->wb_context->path.dentry->d_inode), 102348186c7dSChuck Lever data->req->wb_bytes, (long long)req_offset(data->req)); 10241da177e4SLinus Torvalds 1025c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 1026c9d8f89dSTrond Myklebust } 1027788e7a89STrond Myklebust 1028c9d8f89dSTrond Myklebust static void nfs_writeback_release_partial(void *calldata) 1029c9d8f89dSTrond Myklebust { 1030c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1031c9d8f89dSTrond Myklebust struct nfs_page *req = data->req; 1032c9d8f89dSTrond Myklebust struct page *page = req->wb_page; 1033c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1034c9d8f89dSTrond Myklebust 1035c9d8f89dSTrond Myklebust if (status < 0) { 1036a301b777STrond Myklebust nfs_set_pageerror(page); 1037c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 1038c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 10398e821cadSTrond Myklebust goto out; 10408e821cadSTrond Myklebust } 10418e821cadSTrond Myklebust 10428e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 1043587142f8STrond Myklebust struct inode *inode = page->mapping->host; 10448e821cadSTrond Myklebust 1045587142f8STrond Myklebust spin_lock(&inode->i_lock); 10468e821cadSTrond Myklebust if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { 10478e821cadSTrond Myklebust /* Do nothing we need to resend the writes */ 10488e821cadSTrond Myklebust } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { 10491da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 10501da177e4SLinus Torvalds dprintk(" defer commit\n"); 10511da177e4SLinus Torvalds } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { 10528e821cadSTrond Myklebust set_bit(PG_NEED_RESCHED, &req->wb_flags); 10538e821cadSTrond Myklebust clear_bit(PG_NEED_COMMIT, &req->wb_flags); 10541da177e4SLinus Torvalds dprintk(" server reboot detected\n"); 10551da177e4SLinus Torvalds } 1056587142f8STrond Myklebust spin_unlock(&inode->i_lock); 10571da177e4SLinus Torvalds } else 10581da177e4SLinus Torvalds dprintk(" OK\n"); 10591da177e4SLinus Torvalds 10608e821cadSTrond Myklebust out: 10611da177e4SLinus Torvalds if (atomic_dec_and_test(&req->wb_complete)) 10621da177e4SLinus Torvalds nfs_writepage_release(req); 1063c9d8f89dSTrond Myklebust nfs_writedata_release(calldata); 10641da177e4SLinus Torvalds } 10651da177e4SLinus Torvalds 1066def6ed7eSAndy Adamson #if defined(CONFIG_NFS_V4_1) 1067def6ed7eSAndy Adamson void nfs_write_prepare(struct rpc_task *task, void *calldata) 1068def6ed7eSAndy Adamson { 1069def6ed7eSAndy Adamson struct nfs_write_data *data = calldata; 1070def6ed7eSAndy Adamson struct nfs_client *clp = (NFS_SERVER(data->inode))->nfs_client; 1071def6ed7eSAndy Adamson 1072def6ed7eSAndy Adamson if (nfs4_setup_sequence(clp, &data->args.seq_args, 1073def6ed7eSAndy Adamson &data->res.seq_res, 1, task)) 1074def6ed7eSAndy Adamson return; 1075def6ed7eSAndy Adamson rpc_call_start(task); 1076def6ed7eSAndy Adamson } 1077def6ed7eSAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1078def6ed7eSAndy Adamson 1079788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops = { 1080def6ed7eSAndy Adamson #if defined(CONFIG_NFS_V4_1) 1081def6ed7eSAndy Adamson .rpc_call_prepare = nfs_write_prepare, 1082def6ed7eSAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1083788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_partial, 1084c9d8f89dSTrond Myklebust .rpc_release = nfs_writeback_release_partial, 1085788e7a89STrond Myklebust }; 1086788e7a89STrond Myklebust 10871da177e4SLinus Torvalds /* 10881da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 10891da177e4SLinus Torvalds * 10901da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 10911da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 10921da177e4SLinus Torvalds * as the page has a write request pending. 10931da177e4SLinus Torvalds */ 1094788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) 10951da177e4SLinus Torvalds { 1096788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 10971da177e4SLinus Torvalds 1098c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 1099c9d8f89dSTrond Myklebust } 1100c9d8f89dSTrond Myklebust 1101c9d8f89dSTrond Myklebust static void nfs_writeback_release_full(void *calldata) 1102c9d8f89dSTrond Myklebust { 1103c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1104c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1105788e7a89STrond Myklebust 11061da177e4SLinus Torvalds /* Update attributes as result of writeback. */ 11071da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 1108c9d8f89dSTrond Myklebust struct nfs_page *req = nfs_list_entry(data->pages.next); 1109c9d8f89dSTrond Myklebust struct page *page = req->wb_page; 1110c9d8f89dSTrond Myklebust 11111da177e4SLinus Torvalds nfs_list_remove_request(req); 11121da177e4SLinus Torvalds 111348186c7dSChuck Lever dprintk("NFS: %5u write (%s/%lld %d@%lld)", 111448186c7dSChuck Lever data->task.tk_pid, 111588be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 111688be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 11171da177e4SLinus Torvalds req->wb_bytes, 11181da177e4SLinus Torvalds (long long)req_offset(req)); 11191da177e4SLinus Torvalds 1120c9d8f89dSTrond Myklebust if (status < 0) { 1121a301b777STrond Myklebust nfs_set_pageerror(page); 1122c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 1123c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 11248e821cadSTrond Myklebust goto remove_request; 11251da177e4SLinus Torvalds } 11261da177e4SLinus Torvalds 11278e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 11281da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 11291da177e4SLinus Torvalds nfs_mark_request_commit(req); 11308e821cadSTrond Myklebust nfs_end_page_writeback(page); 11311da177e4SLinus Torvalds dprintk(" marked for commit\n"); 11328e821cadSTrond Myklebust goto next; 11338e821cadSTrond Myklebust } 11348e821cadSTrond Myklebust dprintk(" OK\n"); 11358e821cadSTrond Myklebust remove_request: 11368e821cadSTrond Myklebust nfs_end_page_writeback(page); 11371da177e4SLinus Torvalds nfs_inode_remove_request(req); 11381da177e4SLinus Torvalds next: 11399fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 11401da177e4SLinus Torvalds } 1141c9d8f89dSTrond Myklebust nfs_writedata_release(calldata); 11421da177e4SLinus Torvalds } 11431da177e4SLinus Torvalds 1144788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = { 1145def6ed7eSAndy Adamson #if defined(CONFIG_NFS_V4_1) 1146def6ed7eSAndy Adamson .rpc_call_prepare = nfs_write_prepare, 1147def6ed7eSAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1148788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_full, 1149c9d8f89dSTrond Myklebust .rpc_release = nfs_writeback_release_full, 1150788e7a89STrond Myklebust }; 1151788e7a89STrond Myklebust 1152788e7a89STrond Myklebust 11531da177e4SLinus Torvalds /* 11541da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 11551da177e4SLinus Torvalds */ 1156462d5b32SChuck Lever int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 11571da177e4SLinus Torvalds { 11581da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 11591da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1160eedc020eSAndy Adamson struct nfs_server *server = NFS_SERVER(data->inode); 1161788e7a89STrond Myklebust int status; 11621da177e4SLinus Torvalds 1163a3f565b1SChuck Lever dprintk("NFS: %5u nfs_writeback_done (status %d)\n", 11641da177e4SLinus Torvalds task->tk_pid, task->tk_status); 11651da177e4SLinus Torvalds 1166f551e44fSChuck Lever /* 1167f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1168f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1169f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1170f551e44fSChuck Lever * another writer had changed the file, but some applications 1171f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1172f551e44fSChuck Lever */ 1173788e7a89STrond Myklebust status = NFS_PROTO(data->inode)->write_done(task, data); 1174788e7a89STrond Myklebust if (status != 0) 1175788e7a89STrond Myklebust return status; 117691d5b470SChuck Lever nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 117791d5b470SChuck Lever 11781da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 11791da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 11801da177e4SLinus Torvalds /* We tried a write call, but the server did not 11811da177e4SLinus Torvalds * commit data to stable storage even though we 11821da177e4SLinus Torvalds * requested it. 11831da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 11841da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 11851da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 11861da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 11871da177e4SLinus Torvalds */ 11881da177e4SLinus Torvalds static unsigned long complain; 11891da177e4SLinus Torvalds 11901da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 11911da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 11921da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1193eedc020eSAndy Adamson server->nfs_client->cl_hostname, 11941da177e4SLinus Torvalds resp->verf->committed, argp->stable); 11951da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11961da177e4SLinus Torvalds } 11971da177e4SLinus Torvalds } 11981da177e4SLinus Torvalds #endif 11991da177e4SLinus Torvalds /* Is this a short write? */ 12001da177e4SLinus Torvalds if (task->tk_status >= 0 && resp->count < argp->count) { 12011da177e4SLinus Torvalds static unsigned long complain; 12021da177e4SLinus Torvalds 120391d5b470SChuck Lever nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); 120491d5b470SChuck Lever 12051da177e4SLinus Torvalds /* Has the server at least made some progress? */ 12061da177e4SLinus Torvalds if (resp->count != 0) { 12071da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 12081da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 12091da177e4SLinus Torvalds /* Resend from where the server left off */ 12101da177e4SLinus Torvalds argp->offset += resp->count; 12111da177e4SLinus Torvalds argp->pgbase += resp->count; 12121da177e4SLinus Torvalds argp->count -= resp->count; 12131da177e4SLinus Torvalds } else { 12141da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 12151da177e4SLinus Torvalds * headaches in the case of a server crash. 12161da177e4SLinus Torvalds */ 12171da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 12181da177e4SLinus Torvalds } 1219eedc020eSAndy Adamson nfs4_restart_rpc(task, server->nfs_client); 1220788e7a89STrond Myklebust return -EAGAIN; 12211da177e4SLinus Torvalds } 12221da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 12231da177e4SLinus Torvalds printk(KERN_WARNING 12241da177e4SLinus Torvalds "NFS: Server wrote zero bytes, expected %u.\n", 12251da177e4SLinus Torvalds argp->count); 12261da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 12271da177e4SLinus Torvalds } 12281da177e4SLinus Torvalds /* Can't do anything about it except throw an error. */ 12291da177e4SLinus Torvalds task->tk_status = -EIO; 12301da177e4SLinus Torvalds } 1231eedc020eSAndy Adamson nfs4_sequence_free_slot(server->nfs_client, &data->res.seq_res); 1232788e7a89STrond Myklebust return 0; 12331da177e4SLinus Torvalds } 12341da177e4SLinus Torvalds 12351da177e4SLinus Torvalds 12361da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 1237c9d8f89dSTrond Myklebust void nfs_commitdata_release(void *data) 12381da177e4SLinus Torvalds { 1239383ba719STrond Myklebust struct nfs_write_data *wdata = data; 1240383ba719STrond Myklebust 1241383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 12421da177e4SLinus Torvalds nfs_commit_free(wdata); 12431da177e4SLinus Torvalds } 12441da177e4SLinus Torvalds 12451da177e4SLinus Torvalds /* 12461da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 12471da177e4SLinus Torvalds */ 1248dbae4c73STrond Myklebust static int nfs_commit_rpcsetup(struct list_head *head, 1249788e7a89STrond Myklebust struct nfs_write_data *data, 1250788e7a89STrond Myklebust int how) 12511da177e4SLinus Torvalds { 125284115e1cSTrond Myklebust struct nfs_page *first = nfs_list_entry(head->next); 125384115e1cSTrond Myklebust struct inode *inode = first->wb_context->path.dentry->d_inode; 125484115e1cSTrond Myklebust int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 12553ff7576dSTrond Myklebust int priority = flush_task_priority(how); 125607737691STrond Myklebust struct rpc_task *task; 1257bdc7f021STrond Myklebust struct rpc_message msg = { 1258bdc7f021STrond Myklebust .rpc_argp = &data->args, 1259bdc7f021STrond Myklebust .rpc_resp = &data->res, 1260bdc7f021STrond Myklebust .rpc_cred = first->wb_context->cred, 1261bdc7f021STrond Myklebust }; 126284115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 126307737691STrond Myklebust .task = &data->task, 126484115e1cSTrond Myklebust .rpc_client = NFS_CLIENT(inode), 1265bdc7f021STrond Myklebust .rpc_message = &msg, 126684115e1cSTrond Myklebust .callback_ops = &nfs_commit_ops, 126784115e1cSTrond Myklebust .callback_data = data, 1268101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 126984115e1cSTrond Myklebust .flags = flags, 12703ff7576dSTrond Myklebust .priority = priority, 127184115e1cSTrond Myklebust }; 12721da177e4SLinus Torvalds 12731da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 12741da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 12751da177e4SLinus Torvalds 12761da177e4SLinus Torvalds list_splice_init(head, &data->pages); 12771da177e4SLinus Torvalds 12781da177e4SLinus Torvalds data->inode = inode; 1279bdc7f021STrond Myklebust data->cred = msg.rpc_cred; 12801da177e4SLinus Torvalds 12811da177e4SLinus Torvalds data->args.fh = NFS_FH(data->inode); 12823da28eb1STrond Myklebust /* Note: we always request a commit of the entire inode */ 12833da28eb1STrond Myklebust data->args.offset = 0; 12843da28eb1STrond Myklebust data->args.count = 0; 1285383ba719STrond Myklebust data->args.context = get_nfs_open_context(first->wb_context); 12863da28eb1STrond Myklebust data->res.count = 0; 12871da177e4SLinus Torvalds data->res.fattr = &data->fattr; 12881da177e4SLinus Torvalds data->res.verf = &data->verf; 12890e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 12901da177e4SLinus Torvalds 1291788e7a89STrond Myklebust /* Set up the initial task struct. */ 1292bdc7f021STrond Myklebust NFS_PROTO(inode)->commit_setup(data, &msg); 12931da177e4SLinus Torvalds 1294a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1295bdc7f021STrond Myklebust 129607737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1297dbae4c73STrond Myklebust if (IS_ERR(task)) 1298dbae4c73STrond Myklebust return PTR_ERR(task); 129907737691STrond Myklebust rpc_put_task(task); 1300dbae4c73STrond Myklebust return 0; 13011da177e4SLinus Torvalds } 13021da177e4SLinus Torvalds 13031da177e4SLinus Torvalds /* 13041da177e4SLinus Torvalds * Commit dirty pages 13051da177e4SLinus Torvalds */ 13061da177e4SLinus Torvalds static int 130740859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how) 13081da177e4SLinus Torvalds { 13091da177e4SLinus Torvalds struct nfs_write_data *data; 13101da177e4SLinus Torvalds struct nfs_page *req; 13111da177e4SLinus Torvalds 1312c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 13131da177e4SLinus Torvalds 13141da177e4SLinus Torvalds if (!data) 13151da177e4SLinus Torvalds goto out_bad; 13161da177e4SLinus Torvalds 13171da177e4SLinus Torvalds /* Set up the argument struct */ 1318dbae4c73STrond Myklebust return nfs_commit_rpcsetup(head, data, how); 13191da177e4SLinus Torvalds out_bad: 13201da177e4SLinus Torvalds while (!list_empty(head)) { 13211da177e4SLinus Torvalds req = nfs_list_entry(head->next); 13221da177e4SLinus Torvalds nfs_list_remove_request(req); 13231da177e4SLinus Torvalds nfs_mark_request_commit(req); 132483715ad5STrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1325c9e51e41SPeter Zijlstra dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 1326c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 13279fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 13281da177e4SLinus Torvalds } 13291da177e4SLinus Torvalds return -ENOMEM; 13301da177e4SLinus Torvalds } 13311da177e4SLinus Torvalds 13321da177e4SLinus Torvalds /* 13331da177e4SLinus Torvalds * COMMIT call returned 13341da177e4SLinus Torvalds */ 1335788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 13361da177e4SLinus Torvalds { 1337963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 13381da177e4SLinus Torvalds 1339a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 13401da177e4SLinus Torvalds task->tk_pid, task->tk_status); 13411da177e4SLinus Torvalds 1342788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1343788e7a89STrond Myklebust if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) 1344788e7a89STrond Myklebust return; 1345c9d8f89dSTrond Myklebust } 1346c9d8f89dSTrond Myklebust 1347c9d8f89dSTrond Myklebust static void nfs_commit_release(void *calldata) 1348c9d8f89dSTrond Myklebust { 1349c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1350c9d8f89dSTrond Myklebust struct nfs_page *req; 1351c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1352788e7a89STrond Myklebust 13531da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 13541da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 13551da177e4SLinus Torvalds nfs_list_remove_request(req); 1356e468bae9STrond Myklebust nfs_clear_request_commit(req); 13571da177e4SLinus Torvalds 135848186c7dSChuck Lever dprintk("NFS: commit (%s/%lld %d@%lld)", 135988be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 136088be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 13611da177e4SLinus Torvalds req->wb_bytes, 13621da177e4SLinus Torvalds (long long)req_offset(req)); 1363c9d8f89dSTrond Myklebust if (status < 0) { 1364c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 13651da177e4SLinus Torvalds nfs_inode_remove_request(req); 1366c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 13671da177e4SLinus Torvalds goto next; 13681da177e4SLinus Torvalds } 13691da177e4SLinus Torvalds 13701da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 13711da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 13721da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 13731da177e4SLinus Torvalds /* We have a match */ 13741da177e4SLinus Torvalds nfs_inode_remove_request(req); 13751da177e4SLinus Torvalds dprintk(" OK\n"); 13761da177e4SLinus Torvalds goto next; 13771da177e4SLinus Torvalds } 13781da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 13791da177e4SLinus Torvalds dprintk(" mismatch\n"); 13806d884e8fSFred nfs_mark_request_dirty(req); 13811da177e4SLinus Torvalds next: 13829fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 13831da177e4SLinus Torvalds } 1384c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 13851da177e4SLinus Torvalds } 1386788e7a89STrond Myklebust 1387788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 138821d9a851SAndy Adamson #if defined(CONFIG_NFS_V4_1) 138921d9a851SAndy Adamson .rpc_call_prepare = nfs_write_prepare, 139021d9a851SAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1391788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1392788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1393788e7a89STrond Myklebust }; 13941da177e4SLinus Torvalds 13953da28eb1STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 13961da177e4SLinus Torvalds { 13971da177e4SLinus Torvalds LIST_HEAD(head); 13987d46a49fSTrond Myklebust int res; 13991da177e4SLinus Torvalds 1400587142f8STrond Myklebust spin_lock(&inode->i_lock); 14013da28eb1STrond Myklebust res = nfs_scan_commit(inode, &head, 0, 0); 1402587142f8STrond Myklebust spin_unlock(&inode->i_lock); 14031da177e4SLinus Torvalds if (res) { 14047d46a49fSTrond Myklebust int error = nfs_commit_list(inode, &head, how); 14051da177e4SLinus Torvalds if (error < 0) 14061da177e4SLinus Torvalds return error; 14073da28eb1STrond Myklebust } 14081da177e4SLinus Torvalds return res; 14091da177e4SLinus Torvalds } 1410c63c7b05STrond Myklebust #else 1411c63c7b05STrond Myklebust static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how) 1412c63c7b05STrond Myklebust { 1413c63c7b05STrond Myklebust return 0; 1414c63c7b05STrond Myklebust } 14151da177e4SLinus Torvalds #endif 14161da177e4SLinus Torvalds 14171c75950bSTrond Myklebust long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how) 14181da177e4SLinus Torvalds { 14191c75950bSTrond Myklebust struct inode *inode = mapping->host; 1420ca52fec1STrond Myklebust pgoff_t idx_start, idx_end; 14211c75950bSTrond Myklebust unsigned int npages = 0; 1422c42de9ddSTrond Myklebust LIST_HEAD(head); 142370b9ecbdSTrond Myklebust int nocommit = how & FLUSH_NOCOMMIT; 14243f442547STrond Myklebust long pages, ret; 14251da177e4SLinus Torvalds 14261c75950bSTrond Myklebust /* FIXME */ 14271c75950bSTrond Myklebust if (wbc->range_cyclic) 14281c75950bSTrond Myklebust idx_start = 0; 14291c75950bSTrond Myklebust else { 14301c75950bSTrond Myklebust idx_start = wbc->range_start >> PAGE_CACHE_SHIFT; 14311c75950bSTrond Myklebust idx_end = wbc->range_end >> PAGE_CACHE_SHIFT; 14321c75950bSTrond Myklebust if (idx_end > idx_start) { 1433ca52fec1STrond Myklebust pgoff_t l_npages = 1 + idx_end - idx_start; 14341c75950bSTrond Myklebust npages = l_npages; 14351c75950bSTrond Myklebust if (sizeof(npages) != sizeof(l_npages) && 1436ca52fec1STrond Myklebust (pgoff_t)npages != l_npages) 14371c75950bSTrond Myklebust npages = 0; 14381c75950bSTrond Myklebust } 14391c75950bSTrond Myklebust } 1440c42de9ddSTrond Myklebust how &= ~FLUSH_NOCOMMIT; 1441587142f8STrond Myklebust spin_lock(&inode->i_lock); 14421da177e4SLinus Torvalds do { 1443c42de9ddSTrond Myklebust ret = nfs_wait_on_requests_locked(inode, idx_start, npages); 1444c42de9ddSTrond Myklebust if (ret != 0) 1445c42de9ddSTrond Myklebust continue; 1446c42de9ddSTrond Myklebust if (nocommit) 1447c42de9ddSTrond Myklebust break; 1448d2ccddf0STrond Myklebust pages = nfs_scan_commit(inode, &head, idx_start, npages); 1449724c439cSTrond Myklebust if (pages == 0) 1450c42de9ddSTrond Myklebust break; 1451d2ccddf0STrond Myklebust if (how & FLUSH_INVALIDATE) { 1452587142f8STrond Myklebust spin_unlock(&inode->i_lock); 145383715ad5STrond Myklebust nfs_cancel_commit_list(&head); 1454e8e058e8STrond Myklebust ret = pages; 1455587142f8STrond Myklebust spin_lock(&inode->i_lock); 1456d2ccddf0STrond Myklebust continue; 1457d2ccddf0STrond Myklebust } 1458d2ccddf0STrond Myklebust pages += nfs_scan_commit(inode, &head, 0, 0); 1459587142f8STrond Myklebust spin_unlock(&inode->i_lock); 1460c42de9ddSTrond Myklebust ret = nfs_commit_list(inode, &head, how); 1461587142f8STrond Myklebust spin_lock(&inode->i_lock); 1462587142f8STrond Myklebust 1463c42de9ddSTrond Myklebust } while (ret >= 0); 1464587142f8STrond Myklebust spin_unlock(&inode->i_lock); 1465c42de9ddSTrond Myklebust return ret; 14661da177e4SLinus Torvalds } 14671da177e4SLinus Torvalds 146834901f70STrond Myklebust static int __nfs_write_mapping(struct address_space *mapping, struct writeback_control *wbc, int how) 14691c75950bSTrond Myklebust { 14701c75950bSTrond Myklebust int ret; 14711c75950bSTrond Myklebust 147234901f70STrond Myklebust ret = nfs_writepages(mapping, wbc); 147361822ab5STrond Myklebust if (ret < 0) 147461822ab5STrond Myklebust goto out; 147534901f70STrond Myklebust ret = nfs_sync_mapping_wait(mapping, wbc, how); 1476ed90ef51STrond Myklebust if (ret < 0) 1477ed90ef51STrond Myklebust goto out; 14781c75950bSTrond Myklebust return 0; 147961822ab5STrond Myklebust out: 1480e507d9ebSTrond Myklebust __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 14811c75950bSTrond Myklebust return ret; 14821c75950bSTrond Myklebust } 14831c75950bSTrond Myklebust 148434901f70STrond Myklebust /* Two pass sync: first using WB_SYNC_NONE, then WB_SYNC_ALL */ 148534901f70STrond Myklebust static int nfs_write_mapping(struct address_space *mapping, int how) 148634901f70STrond Myklebust { 148734901f70STrond Myklebust struct writeback_control wbc = { 148834901f70STrond Myklebust .bdi = mapping->backing_dev_info, 148972cb77f4STrond Myklebust .sync_mode = WB_SYNC_ALL, 149034901f70STrond Myklebust .nr_to_write = LONG_MAX, 1491d7fb1207STrond Myklebust .range_start = 0, 1492d7fb1207STrond Myklebust .range_end = LLONG_MAX, 149334901f70STrond Myklebust }; 149434901f70STrond Myklebust 149534901f70STrond Myklebust return __nfs_write_mapping(mapping, &wbc, how); 149634901f70STrond Myklebust } 149734901f70STrond Myklebust 1498ed90ef51STrond Myklebust /* 1499ed90ef51STrond Myklebust * flush the inode to disk. 1500ed90ef51STrond Myklebust */ 1501ed90ef51STrond Myklebust int nfs_wb_all(struct inode *inode) 15021c75950bSTrond Myklebust { 1503ed90ef51STrond Myklebust return nfs_write_mapping(inode->i_mapping, 0); 1504ed90ef51STrond Myklebust } 15051c75950bSTrond Myklebust 1506ed90ef51STrond Myklebust int nfs_wb_nocommit(struct inode *inode) 1507ed90ef51STrond Myklebust { 1508ed90ef51STrond Myklebust return nfs_write_mapping(inode->i_mapping, FLUSH_NOCOMMIT); 15091c75950bSTrond Myklebust } 15101c75950bSTrond Myklebust 15111b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 15121b3b4a1aSTrond Myklebust { 15131b3b4a1aSTrond Myklebust struct nfs_page *req; 15141b3b4a1aSTrond Myklebust loff_t range_start = page_offset(page); 15151b3b4a1aSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 15161b3b4a1aSTrond Myklebust struct writeback_control wbc = { 15171b3b4a1aSTrond Myklebust .bdi = page->mapping->backing_dev_info, 15181b3b4a1aSTrond Myklebust .sync_mode = WB_SYNC_ALL, 15191b3b4a1aSTrond Myklebust .nr_to_write = LONG_MAX, 15201b3b4a1aSTrond Myklebust .range_start = range_start, 15211b3b4a1aSTrond Myklebust .range_end = range_end, 15221b3b4a1aSTrond Myklebust }; 15231b3b4a1aSTrond Myklebust int ret = 0; 15241b3b4a1aSTrond Myklebust 15251b3b4a1aSTrond Myklebust BUG_ON(!PageLocked(page)); 15261b3b4a1aSTrond Myklebust for (;;) { 15271b3b4a1aSTrond Myklebust req = nfs_page_find_request(page); 15281b3b4a1aSTrond Myklebust if (req == NULL) 15291b3b4a1aSTrond Myklebust goto out; 1530e468bae9STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 15311b3b4a1aSTrond Myklebust nfs_release_request(req); 15321b3b4a1aSTrond Myklebust break; 15331b3b4a1aSTrond Myklebust } 15341b3b4a1aSTrond Myklebust if (nfs_lock_request_dontget(req)) { 15351b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 15361b3b4a1aSTrond Myklebust /* 15371b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 15381b3b4a1aSTrond Myklebust * page as being dirty 15391b3b4a1aSTrond Myklebust */ 15401b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 15411b3b4a1aSTrond Myklebust nfs_unlock_request(req); 15421b3b4a1aSTrond Myklebust break; 15431b3b4a1aSTrond Myklebust } 15441b3b4a1aSTrond Myklebust ret = nfs_wait_on_request(req); 15451b3b4a1aSTrond Myklebust if (ret < 0) 15461b3b4a1aSTrond Myklebust goto out; 15471b3b4a1aSTrond Myklebust } 15481b3b4a1aSTrond Myklebust if (!PagePrivate(page)) 15491b3b4a1aSTrond Myklebust return 0; 15501b3b4a1aSTrond Myklebust ret = nfs_sync_mapping_wait(page->mapping, &wbc, FLUSH_INVALIDATE); 15511b3b4a1aSTrond Myklebust out: 15521b3b4a1aSTrond Myklebust return ret; 15531b3b4a1aSTrond Myklebust } 15541b3b4a1aSTrond Myklebust 15555334eb13SAdrian Bunk static int nfs_wb_page_priority(struct inode *inode, struct page *page, 15565334eb13SAdrian Bunk int how) 15571c75950bSTrond Myklebust { 15581c75950bSTrond Myklebust loff_t range_start = page_offset(page); 15591c75950bSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 15604d770ccfSTrond Myklebust struct writeback_control wbc = { 15614d770ccfSTrond Myklebust .bdi = page->mapping->backing_dev_info, 15624d770ccfSTrond Myklebust .sync_mode = WB_SYNC_ALL, 15634d770ccfSTrond Myklebust .nr_to_write = LONG_MAX, 15644d770ccfSTrond Myklebust .range_start = range_start, 15654d770ccfSTrond Myklebust .range_end = range_end, 15664d770ccfSTrond Myklebust }; 15674d770ccfSTrond Myklebust int ret; 15681c75950bSTrond Myklebust 156973e3302fSTrond Myklebust do { 1570c63c7b05STrond Myklebust if (clear_page_dirty_for_io(page)) { 15714d770ccfSTrond Myklebust ret = nfs_writepage_locked(page, &wbc); 15724d770ccfSTrond Myklebust if (ret < 0) 157373e3302fSTrond Myklebust goto out_error; 157473e3302fSTrond Myklebust } else if (!PagePrivate(page)) 157573e3302fSTrond Myklebust break; 15764d770ccfSTrond Myklebust ret = nfs_sync_mapping_wait(page->mapping, &wbc, how); 157773e3302fSTrond Myklebust if (ret < 0) 157873e3302fSTrond Myklebust goto out_error; 157973e3302fSTrond Myklebust } while (PagePrivate(page)); 15804d770ccfSTrond Myklebust return 0; 158173e3302fSTrond Myklebust out_error: 1582e507d9ebSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_PAGES); 15834d770ccfSTrond Myklebust return ret; 15841c75950bSTrond Myklebust } 15851c75950bSTrond Myklebust 15861c75950bSTrond Myklebust /* 15871c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 15881c75950bSTrond Myklebust */ 15891c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page* page) 15901c75950bSTrond Myklebust { 15914d770ccfSTrond Myklebust return nfs_wb_page_priority(inode, page, FLUSH_STABLE); 15921c75950bSTrond Myklebust } 15931c75950bSTrond Myklebust 1594074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 1595074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 1596074cc1deSTrond Myklebust struct page *page) 1597074cc1deSTrond Myklebust { 1598074cc1deSTrond Myklebust struct nfs_page *req; 1599074cc1deSTrond Myklebust int ret; 1600074cc1deSTrond Myklebust 1601074cc1deSTrond Myklebust if (PageFsCache(page)) 1602074cc1deSTrond Myklebust nfs_fscache_release_page(page, GFP_KERNEL); 1603074cc1deSTrond Myklebust 1604074cc1deSTrond Myklebust req = nfs_find_and_lock_request(page); 1605074cc1deSTrond Myklebust ret = PTR_ERR(req); 1606074cc1deSTrond Myklebust if (IS_ERR(req)) 1607074cc1deSTrond Myklebust goto out; 1608074cc1deSTrond Myklebust 1609074cc1deSTrond Myklebust ret = migrate_page(mapping, newpage, page); 1610074cc1deSTrond Myklebust if (!req) 1611074cc1deSTrond Myklebust goto out; 1612074cc1deSTrond Myklebust if (ret) 1613074cc1deSTrond Myklebust goto out_unlock; 1614074cc1deSTrond Myklebust page_cache_get(newpage); 1615074cc1deSTrond Myklebust req->wb_page = newpage; 1616074cc1deSTrond Myklebust SetPagePrivate(newpage); 1617074cc1deSTrond Myklebust set_page_private(newpage, page_private(page)); 1618074cc1deSTrond Myklebust ClearPagePrivate(page); 1619074cc1deSTrond Myklebust set_page_private(page, 0); 1620074cc1deSTrond Myklebust page_cache_release(page); 1621074cc1deSTrond Myklebust out_unlock: 1622074cc1deSTrond Myklebust nfs_clear_page_tag_locked(req); 1623074cc1deSTrond Myklebust nfs_release_request(req); 1624074cc1deSTrond Myklebust out: 1625074cc1deSTrond Myklebust return ret; 1626074cc1deSTrond Myklebust } 1627074cc1deSTrond Myklebust #endif 1628074cc1deSTrond Myklebust 1629f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 16301da177e4SLinus Torvalds { 16311da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 16321da177e4SLinus Torvalds sizeof(struct nfs_write_data), 16331da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 163420c2df83SPaul Mundt NULL); 16351da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 16361da177e4SLinus Torvalds return -ENOMEM; 16371da177e4SLinus Torvalds 163893d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 16391da177e4SLinus Torvalds nfs_wdata_cachep); 16401da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 16411da177e4SLinus Torvalds return -ENOMEM; 16421da177e4SLinus Torvalds 164393d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 16441da177e4SLinus Torvalds nfs_wdata_cachep); 16451da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 16461da177e4SLinus Torvalds return -ENOMEM; 16471da177e4SLinus Torvalds 164889a09141SPeter Zijlstra /* 164989a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 165089a09141SPeter Zijlstra * 165189a09141SPeter Zijlstra * 64MB: 8192k 165289a09141SPeter Zijlstra * 128MB: 11585k 165389a09141SPeter Zijlstra * 256MB: 16384k 165489a09141SPeter Zijlstra * 512MB: 23170k 165589a09141SPeter Zijlstra * 1GB: 32768k 165689a09141SPeter Zijlstra * 2GB: 46340k 165789a09141SPeter Zijlstra * 4GB: 65536k 165889a09141SPeter Zijlstra * 8GB: 92681k 165989a09141SPeter Zijlstra * 16GB: 131072k 166089a09141SPeter Zijlstra * 166189a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 166289a09141SPeter Zijlstra * Limit the default to 256M 166389a09141SPeter Zijlstra */ 166489a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 166589a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 166689a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 166789a09141SPeter Zijlstra 16681da177e4SLinus Torvalds return 0; 16691da177e4SLinus Torvalds } 16701da177e4SLinus Torvalds 1671266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 16721da177e4SLinus Torvalds { 16731da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 16741da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 16751a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 16761da177e4SLinus Torvalds } 16771da177e4SLinus Torvalds 1678