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" 3194ad1c80SFred Isaman #include "pnfs.h" 321da177e4SLinus Torvalds 331da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 341da177e4SLinus Torvalds 351da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 361da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 371da177e4SLinus Torvalds 381da177e4SLinus Torvalds /* 391da177e4SLinus Torvalds * Local function declarations 401da177e4SLinus Torvalds */ 41c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc, 42c63c7b05STrond Myklebust struct inode *inode, int ioflags); 43f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 44788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops; 45788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops; 46788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 471da177e4SLinus Torvalds 48e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 493feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 501da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 511da177e4SLinus Torvalds 52c9d8f89dSTrond Myklebust struct nfs_write_data *nfs_commitdata_alloc(void) 531da177e4SLinus Torvalds { 54e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS); 5540859d7eSChuck Lever 561da177e4SLinus Torvalds if (p) { 571da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 581da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 591da177e4SLinus Torvalds } 601da177e4SLinus Torvalds return p; 611da177e4SLinus Torvalds } 62e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 631da177e4SLinus Torvalds 645e4424afSTrond Myklebust void nfs_commit_free(struct nfs_write_data *p) 651da177e4SLinus Torvalds { 6640859d7eSChuck Lever if (p && (p->pagevec != &p->page_array[0])) 6740859d7eSChuck Lever kfree(p->pagevec); 681da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 691da177e4SLinus Torvalds } 70e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 711da177e4SLinus Torvalds 728d5658c9STrond Myklebust struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount) 733feb2d49STrond Myklebust { 74e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); 753feb2d49STrond Myklebust 763feb2d49STrond Myklebust if (p) { 773feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 783feb2d49STrond Myklebust INIT_LIST_HEAD(&p->pages); 79e9f7bee1STrond Myklebust p->npages = pagecount; 800d0b5cb3SChuck Lever if (pagecount <= ARRAY_SIZE(p->page_array)) 810d0b5cb3SChuck Lever p->pagevec = p->page_array; 823feb2d49STrond Myklebust else { 830d0b5cb3SChuck Lever p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); 840d0b5cb3SChuck Lever if (!p->pagevec) { 853feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 863feb2d49STrond Myklebust p = NULL; 873feb2d49STrond Myklebust } 883feb2d49STrond Myklebust } 893feb2d49STrond Myklebust } 903feb2d49STrond Myklebust return p; 913feb2d49STrond Myklebust } 923feb2d49STrond Myklebust 931ae88b2eSTrond Myklebust void nfs_writedata_free(struct nfs_write_data *p) 943feb2d49STrond Myklebust { 953feb2d49STrond Myklebust if (p && (p->pagevec != &p->page_array[0])) 963feb2d49STrond Myklebust kfree(p->pagevec); 973feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 983feb2d49STrond Myklebust } 993feb2d49STrond Myklebust 100dce81290STrond Myklebust void nfs_writedata_release(struct nfs_write_data *wdata) 1011da177e4SLinus Torvalds { 1025053aa56SFred Isaman put_lseg(wdata->lseg); 103383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 1041da177e4SLinus Torvalds nfs_writedata_free(wdata); 1051da177e4SLinus Torvalds } 1061da177e4SLinus Torvalds 1077b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 1087b159fc1STrond Myklebust { 1097b159fc1STrond Myklebust ctx->error = error; 1107b159fc1STrond Myklebust smp_wmb(); 1117b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 1127b159fc1STrond Myklebust } 1137b159fc1STrond Myklebust 114277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request_locked(struct page *page) 115277459d2STrond Myklebust { 116277459d2STrond Myklebust struct nfs_page *req = NULL; 117277459d2STrond Myklebust 118277459d2STrond Myklebust if (PagePrivate(page)) { 119277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 120277459d2STrond Myklebust if (req != NULL) 121c03b4024STrond Myklebust kref_get(&req->wb_kref); 122277459d2STrond Myklebust } 123277459d2STrond Myklebust return req; 124277459d2STrond Myklebust } 125277459d2STrond Myklebust 126277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request(struct page *page) 127277459d2STrond Myklebust { 128587142f8STrond Myklebust struct inode *inode = page->mapping->host; 129277459d2STrond Myklebust struct nfs_page *req = NULL; 130277459d2STrond Myklebust 131587142f8STrond Myklebust spin_lock(&inode->i_lock); 132277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 133587142f8STrond Myklebust spin_unlock(&inode->i_lock); 134277459d2STrond Myklebust return req; 135277459d2STrond Myklebust } 136277459d2STrond Myklebust 1371da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1381da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1391da177e4SLinus Torvalds { 1401da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 141a3d01454STrond Myklebust loff_t end, i_size; 142a3d01454STrond Myklebust pgoff_t end_index; 1431da177e4SLinus Torvalds 144a3d01454STrond Myklebust spin_lock(&inode->i_lock); 145a3d01454STrond Myklebust i_size = i_size_read(inode); 146a3d01454STrond Myklebust end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 1471da177e4SLinus Torvalds if (i_size > 0 && page->index < end_index) 148a3d01454STrond Myklebust goto out; 1491da177e4SLinus Torvalds end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); 1501da177e4SLinus Torvalds if (i_size >= end) 151a3d01454STrond Myklebust goto out; 1521da177e4SLinus Torvalds i_size_write(inode, end); 153a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 154a3d01454STrond Myklebust out: 155a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 1561da177e4SLinus Torvalds } 1571da177e4SLinus Torvalds 158a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 159a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 160a301b777STrond Myklebust { 161a301b777STrond Myklebust SetPageError(page); 162a301b777STrond Myklebust nfs_zap_mapping(page->mapping->host, page->mapping); 163a301b777STrond Myklebust } 164a301b777STrond Myklebust 1651da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 1661da177e4SLinus Torvalds * covers the full page. 1671da177e4SLinus Torvalds */ 1681da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 1691da177e4SLinus Torvalds { 1701da177e4SLinus Torvalds if (PageUptodate(page)) 1711da177e4SLinus Torvalds return; 1721da177e4SLinus Torvalds if (base != 0) 1731da177e4SLinus Torvalds return; 17449a70f27STrond Myklebust if (count != nfs_page_length(page)) 1751da177e4SLinus Torvalds return; 1761da177e4SLinus Torvalds SetPageUptodate(page); 1771da177e4SLinus Torvalds } 1781da177e4SLinus Torvalds 1791da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 1801da177e4SLinus Torvalds { 1811da177e4SLinus Torvalds if (wbc->for_reclaim) 182c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 183b17621feSWu Fengguang if (wbc->for_kupdate || wbc->for_background) 184b31268acSTrond Myklebust return FLUSH_LOWPRI | FLUSH_COND_STABLE; 185b31268acSTrond Myklebust return FLUSH_COND_STABLE; 1861da177e4SLinus Torvalds } 1871da177e4SLinus Torvalds 1881da177e4SLinus Torvalds /* 18989a09141SPeter Zijlstra * NFS congestion control 19089a09141SPeter Zijlstra */ 19189a09141SPeter Zijlstra 19289a09141SPeter Zijlstra int nfs_congestion_kb; 19389a09141SPeter Zijlstra 19489a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 19589a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 19689a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 19789a09141SPeter Zijlstra 1985a6d41b3STrond Myklebust static int nfs_set_page_writeback(struct page *page) 19989a09141SPeter Zijlstra { 2005a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2015a6d41b3STrond Myklebust 2025a6d41b3STrond Myklebust if (!ret) { 20389a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 20489a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 20589a09141SPeter Zijlstra 206a6305ddbSTrond Myklebust page_cache_get(page); 207277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 2088aa7e847SJens Axboe NFS_CONGESTION_ON_THRESH) { 2098aa7e847SJens Axboe set_bdi_congested(&nfss->backing_dev_info, 2108aa7e847SJens Axboe BLK_RW_ASYNC); 2118aa7e847SJens Axboe } 21289a09141SPeter Zijlstra } 2135a6d41b3STrond Myklebust return ret; 21489a09141SPeter Zijlstra } 21589a09141SPeter Zijlstra 21689a09141SPeter Zijlstra static void nfs_end_page_writeback(struct page *page) 21789a09141SPeter Zijlstra { 21889a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 21989a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 22089a09141SPeter Zijlstra 22189a09141SPeter Zijlstra end_page_writeback(page); 222a6305ddbSTrond Myklebust page_cache_release(page); 223c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 2248aa7e847SJens Axboe clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 22589a09141SPeter Zijlstra } 22689a09141SPeter Zijlstra 227cfb506e1STrond Myklebust static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) 228e261f51fSTrond Myklebust { 229587142f8STrond Myklebust struct inode *inode = page->mapping->host; 230e261f51fSTrond Myklebust struct nfs_page *req; 231e261f51fSTrond Myklebust int ret; 232e261f51fSTrond Myklebust 233587142f8STrond Myklebust spin_lock(&inode->i_lock); 234e261f51fSTrond Myklebust for (;;) { 235e261f51fSTrond Myklebust req = nfs_page_find_request_locked(page); 236074cc1deSTrond Myklebust if (req == NULL) 237074cc1deSTrond Myklebust break; 238acee478aSTrond Myklebust if (nfs_set_page_tag_locked(req)) 239e261f51fSTrond Myklebust break; 240e261f51fSTrond Myklebust /* Note: If we hold the page lock, as is the case in nfs_writepage, 241acee478aSTrond Myklebust * then the call to nfs_set_page_tag_locked() will always 242e261f51fSTrond Myklebust * succeed provided that someone hasn't already marked the 243e261f51fSTrond Myklebust * request as dirty (in which case we don't care). 244e261f51fSTrond Myklebust */ 245587142f8STrond Myklebust spin_unlock(&inode->i_lock); 246cfb506e1STrond Myklebust if (!nonblock) 247e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 248cfb506e1STrond Myklebust else 249cfb506e1STrond Myklebust ret = -EAGAIN; 250e261f51fSTrond Myklebust nfs_release_request(req); 251e261f51fSTrond Myklebust if (ret != 0) 252074cc1deSTrond Myklebust return ERR_PTR(ret); 253587142f8STrond Myklebust spin_lock(&inode->i_lock); 254e261f51fSTrond Myklebust } 255587142f8STrond Myklebust spin_unlock(&inode->i_lock); 256074cc1deSTrond Myklebust return req; 257612c9384STrond Myklebust } 258074cc1deSTrond Myklebust 259074cc1deSTrond Myklebust /* 260074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 261074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 262074cc1deSTrond Myklebust */ 263074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 264cfb506e1STrond Myklebust struct page *page, bool nonblock) 265074cc1deSTrond Myklebust { 266074cc1deSTrond Myklebust struct nfs_page *req; 267074cc1deSTrond Myklebust int ret = 0; 268074cc1deSTrond Myklebust 269cfb506e1STrond Myklebust req = nfs_find_and_lock_request(page, nonblock); 270074cc1deSTrond Myklebust if (!req) 271074cc1deSTrond Myklebust goto out; 272074cc1deSTrond Myklebust ret = PTR_ERR(req); 273074cc1deSTrond Myklebust if (IS_ERR(req)) 274074cc1deSTrond Myklebust goto out; 275074cc1deSTrond Myklebust 276074cc1deSTrond Myklebust ret = nfs_set_page_writeback(page); 277074cc1deSTrond Myklebust BUG_ON(ret != 0); 278074cc1deSTrond Myklebust BUG_ON(test_bit(PG_CLEAN, &req->wb_flags)); 279074cc1deSTrond Myklebust 280f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 281f8512ad0SFred Isaman nfs_redirty_request(req); 282074cc1deSTrond Myklebust ret = pgio->pg_error; 283f8512ad0SFred Isaman } 284074cc1deSTrond Myklebust out: 285074cc1deSTrond Myklebust return ret; 286e261f51fSTrond Myklebust } 287e261f51fSTrond Myklebust 288f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 289f758c885STrond Myklebust { 290f758c885STrond Myklebust struct inode *inode = page->mapping->host; 291cfb506e1STrond Myklebust int ret; 292f758c885STrond Myklebust 293f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 294f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 295f758c885STrond Myklebust 296f758c885STrond Myklebust nfs_pageio_cond_complete(pgio, page->index); 2971b430beeSWu Fengguang ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); 298cfb506e1STrond Myklebust if (ret == -EAGAIN) { 299cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 300cfb506e1STrond Myklebust ret = 0; 301cfb506e1STrond Myklebust } 302cfb506e1STrond Myklebust return ret; 303f758c885STrond Myklebust } 304f758c885STrond Myklebust 305e261f51fSTrond Myklebust /* 3061da177e4SLinus Torvalds * Write an mmapped page to the server. 3071da177e4SLinus Torvalds */ 3084d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 3091da177e4SLinus Torvalds { 310f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 311e261f51fSTrond Myklebust int err; 3121da177e4SLinus Torvalds 313f758c885STrond Myklebust nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc)); 314f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 315f758c885STrond Myklebust nfs_pageio_complete(&pgio); 316f758c885STrond Myklebust if (err < 0) 3174d770ccfSTrond Myklebust return err; 318f758c885STrond Myklebust if (pgio.pg_error < 0) 319f758c885STrond Myklebust return pgio.pg_error; 320f758c885STrond Myklebust return 0; 3214d770ccfSTrond Myklebust } 3224d770ccfSTrond Myklebust 3234d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 3244d770ccfSTrond Myklebust { 325f758c885STrond Myklebust int ret; 3264d770ccfSTrond Myklebust 327f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 3281da177e4SLinus Torvalds unlock_page(page); 329f758c885STrond Myklebust return ret; 330f758c885STrond Myklebust } 331f758c885STrond Myklebust 332f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 333f758c885STrond Myklebust { 334f758c885STrond Myklebust int ret; 335f758c885STrond Myklebust 336f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 337f758c885STrond Myklebust unlock_page(page); 338f758c885STrond Myklebust return ret; 3391da177e4SLinus Torvalds } 3401da177e4SLinus Torvalds 3411da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 3421da177e4SLinus Torvalds { 3431da177e4SLinus Torvalds struct inode *inode = mapping->host; 34472cb77f4STrond Myklebust unsigned long *bitlock = &NFS_I(inode)->flags; 345c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 3461da177e4SLinus Torvalds int err; 3471da177e4SLinus Torvalds 34872cb77f4STrond Myklebust /* Stop dirtying of new pages while we sync */ 34972cb77f4STrond Myklebust err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, 35072cb77f4STrond Myklebust nfs_wait_bit_killable, TASK_KILLABLE); 35172cb77f4STrond Myklebust if (err) 35272cb77f4STrond Myklebust goto out_err; 35372cb77f4STrond Myklebust 35491d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 35591d5b470SChuck Lever 356c63c7b05STrond Myklebust nfs_pageio_init_write(&pgio, inode, wb_priority(wbc)); 357f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 358c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 35972cb77f4STrond Myklebust 36072cb77f4STrond Myklebust clear_bit_unlock(NFS_INO_FLUSHING, bitlock); 36172cb77f4STrond Myklebust smp_mb__after_clear_bit(); 36272cb77f4STrond Myklebust wake_up_bit(bitlock, NFS_INO_FLUSHING); 36372cb77f4STrond Myklebust 364f758c885STrond Myklebust if (err < 0) 36572cb77f4STrond Myklebust goto out_err; 36672cb77f4STrond Myklebust err = pgio.pg_error; 36772cb77f4STrond Myklebust if (err < 0) 36872cb77f4STrond Myklebust goto out_err; 369c63c7b05STrond Myklebust return 0; 37072cb77f4STrond Myklebust out_err: 37172cb77f4STrond Myklebust return err; 3721da177e4SLinus Torvalds } 3731da177e4SLinus Torvalds 3741da177e4SLinus Torvalds /* 3751da177e4SLinus Torvalds * Insert a write request into an inode 3761da177e4SLinus Torvalds */ 377e7d39069STrond Myklebust static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 3781da177e4SLinus Torvalds { 3791da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 3801da177e4SLinus Torvalds int error; 3811da177e4SLinus Torvalds 382e7d39069STrond Myklebust error = radix_tree_preload(GFP_NOFS); 383e7d39069STrond Myklebust if (error != 0) 384e7d39069STrond Myklebust goto out; 385e7d39069STrond Myklebust 386e7d39069STrond Myklebust /* Lock the request! */ 387e7d39069STrond Myklebust nfs_lock_request_dontget(req); 388e7d39069STrond Myklebust 389e7d39069STrond Myklebust spin_lock(&inode->i_lock); 3901da177e4SLinus Torvalds error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); 39127852596SNick Piggin BUG_ON(error); 3924d65c520STrond Myklebust if (!nfsi->npages && nfs_have_delegation(inode, FMODE_WRITE)) 3931da177e4SLinus Torvalds nfsi->change_attr++; 3942df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 395deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 396277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 3971da177e4SLinus Torvalds nfsi->npages++; 398c03b4024STrond Myklebust kref_get(&req->wb_kref); 39927852596SNick Piggin radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, 40027852596SNick Piggin NFS_PAGE_TAG_LOCKED); 401e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 402e7d39069STrond Myklebust radix_tree_preload_end(); 403e7d39069STrond Myklebust out: 404e7d39069STrond Myklebust return error; 4051da177e4SLinus Torvalds } 4061da177e4SLinus Torvalds 4071da177e4SLinus Torvalds /* 40889a09141SPeter Zijlstra * Remove a write request from an inode 4091da177e4SLinus Torvalds */ 4101da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 4111da177e4SLinus Torvalds { 4123d4ff43dSAl Viro struct inode *inode = req->wb_context->dentry->d_inode; 4131da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4141da177e4SLinus Torvalds 4151da177e4SLinus Torvalds BUG_ON (!NFS_WBACK_BUSY(req)); 4161da177e4SLinus Torvalds 417587142f8STrond Myklebust spin_lock(&inode->i_lock); 418277459d2STrond Myklebust set_page_private(req->wb_page, 0); 419deb7d638STrond Myklebust ClearPagePrivate(req->wb_page); 4202df485a7STrond Myklebust clear_bit(PG_MAPPED, &req->wb_flags); 4211da177e4SLinus Torvalds radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); 4221da177e4SLinus Torvalds nfsi->npages--; 423587142f8STrond Myklebust spin_unlock(&inode->i_lock); 4241da177e4SLinus Torvalds nfs_release_request(req); 4251da177e4SLinus Torvalds } 4261da177e4SLinus Torvalds 42761822ab5STrond Myklebust static void 4286d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 42961822ab5STrond Myklebust { 43061822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 43161822ab5STrond Myklebust } 43261822ab5STrond Myklebust 4331da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 4341da177e4SLinus Torvalds /* 4351da177e4SLinus Torvalds * Add a request to the inode's commit list. 4361da177e4SLinus Torvalds */ 4371da177e4SLinus Torvalds static void 438a861a1e1SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg) 4391da177e4SLinus Torvalds { 4403d4ff43dSAl Viro struct inode *inode = req->wb_context->dentry->d_inode; 4411da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4421da177e4SLinus Torvalds 443587142f8STrond Myklebust spin_lock(&inode->i_lock); 444e468bae9STrond Myklebust set_bit(PG_CLEAN, &(req)->wb_flags); 4455c369683STrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, 4465c369683STrond Myklebust req->wb_index, 4475c369683STrond Myklebust NFS_PAGE_TAG_COMMIT); 448ff778d02STrond Myklebust nfsi->ncommit++; 449587142f8STrond Myklebust spin_unlock(&inode->i_lock); 450a861a1e1SFred Isaman pnfs_mark_request_commit(req, lseg); 451fd39fc85SChristoph Lameter inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 452c9e51e41SPeter Zijlstra inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); 453a1803044STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 4541da177e4SLinus Torvalds } 4558e821cadSTrond Myklebust 456e468bae9STrond Myklebust static int 457e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 458e468bae9STrond Myklebust { 459e468bae9STrond Myklebust struct page *page = req->wb_page; 460e468bae9STrond Myklebust 461e468bae9STrond Myklebust if (test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) { 462e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 463e468bae9STrond Myklebust dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE); 464e468bae9STrond Myklebust return 1; 465e468bae9STrond Myklebust } 466e468bae9STrond Myklebust return 0; 467e468bae9STrond Myklebust } 468e468bae9STrond Myklebust 4698e821cadSTrond Myklebust static inline 4708e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4718e821cadSTrond Myklebust { 472465d5243SFred Isaman if (data->verf.committed == NFS_DATA_SYNC) 473465d5243SFred Isaman return data->lseg == NULL; 474465d5243SFred Isaman else 4758e821cadSTrond Myklebust return data->verf.committed != NFS_FILE_SYNC; 4768e821cadSTrond Myklebust } 4778e821cadSTrond Myklebust 4788e821cadSTrond Myklebust static inline 479a861a1e1SFred Isaman int nfs_reschedule_unstable_write(struct nfs_page *req, 480a861a1e1SFred Isaman struct nfs_write_data *data) 4818e821cadSTrond Myklebust { 482e468bae9STrond Myklebust if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) { 483a861a1e1SFred Isaman nfs_mark_request_commit(req, data->lseg); 4848e821cadSTrond Myklebust return 1; 4858e821cadSTrond Myklebust } 4868e821cadSTrond Myklebust if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) { 4876d884e8fSFred nfs_mark_request_dirty(req); 4888e821cadSTrond Myklebust return 1; 4898e821cadSTrond Myklebust } 4908e821cadSTrond Myklebust return 0; 4918e821cadSTrond Myklebust } 4928e821cadSTrond Myklebust #else 4938e821cadSTrond Myklebust static inline void 494a861a1e1SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg) 4958e821cadSTrond Myklebust { 4968e821cadSTrond Myklebust } 4978e821cadSTrond Myklebust 498e468bae9STrond Myklebust static inline int 499e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 500e468bae9STrond Myklebust { 501e468bae9STrond Myklebust return 0; 502e468bae9STrond Myklebust } 503e468bae9STrond Myklebust 5048e821cadSTrond Myklebust static inline 5058e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 5068e821cadSTrond Myklebust { 5078e821cadSTrond Myklebust return 0; 5088e821cadSTrond Myklebust } 5098e821cadSTrond Myklebust 5108e821cadSTrond Myklebust static inline 511a861a1e1SFred Isaman int nfs_reschedule_unstable_write(struct nfs_page *req, 512a861a1e1SFred Isaman struct nfs_write_data *data) 5138e821cadSTrond Myklebust { 5148e821cadSTrond Myklebust return 0; 5158e821cadSTrond Myklebust } 5161da177e4SLinus Torvalds #endif 5171da177e4SLinus Torvalds 51847c62564STrond Myklebust #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 519fb8a1f11STrond Myklebust static int 520fb8a1f11STrond Myklebust nfs_need_commit(struct nfs_inode *nfsi) 521fb8a1f11STrond Myklebust { 522fb8a1f11STrond Myklebust return radix_tree_tagged(&nfsi->nfs_page_tree, NFS_PAGE_TAG_COMMIT); 523fb8a1f11STrond Myklebust } 524fb8a1f11STrond Myklebust 5251da177e4SLinus Torvalds /* 5261da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 5271da177e4SLinus Torvalds * @inode: NFS inode to scan 5281da177e4SLinus Torvalds * @dst: destination list 5291da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 5301da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 5311da177e4SLinus Torvalds * 5321da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 5331da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 5341da177e4SLinus Torvalds */ 5351da177e4SLinus Torvalds static int 536ca52fec1STrond Myklebust nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 5371da177e4SLinus Torvalds { 5381da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 539ff778d02STrond Myklebust int ret; 5403da28eb1STrond Myklebust 541fb8a1f11STrond Myklebust if (!nfs_need_commit(nfsi)) 542fb8a1f11STrond Myklebust return 0; 543fb8a1f11STrond Myklebust 5440d88f6e8SDave Chinner spin_lock(&inode->i_lock); 545ff778d02STrond Myklebust ret = nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT); 546ff778d02STrond Myklebust if (ret > 0) 547ff778d02STrond Myklebust nfsi->ncommit -= ret; 5480d88f6e8SDave Chinner spin_unlock(&inode->i_lock); 5490d88f6e8SDave Chinner 5502928db1fSTrond Myklebust if (nfs_need_commit(NFS_I(inode))) 5512928db1fSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 5520d88f6e8SDave Chinner 553ff778d02STrond Myklebust return ret; 5541da177e4SLinus Torvalds } 555c42de9ddSTrond Myklebust #else 556fb8a1f11STrond Myklebust static inline int nfs_need_commit(struct nfs_inode *nfsi) 557fb8a1f11STrond Myklebust { 558fb8a1f11STrond Myklebust return 0; 559fb8a1f11STrond Myklebust } 560fb8a1f11STrond Myklebust 561ca52fec1STrond Myklebust static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 562c42de9ddSTrond Myklebust { 563c42de9ddSTrond Myklebust return 0; 564c42de9ddSTrond Myklebust } 5651da177e4SLinus Torvalds #endif 5661da177e4SLinus Torvalds 5671da177e4SLinus Torvalds /* 568e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 569e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 5701da177e4SLinus Torvalds * 571e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 572e7d39069STrond Myklebust * to disk. 5731da177e4SLinus Torvalds */ 574e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 575e7d39069STrond Myklebust struct page *page, 576e7d39069STrond Myklebust unsigned int offset, 577e7d39069STrond Myklebust unsigned int bytes) 5781da177e4SLinus Torvalds { 579e7d39069STrond Myklebust struct nfs_page *req; 580e7d39069STrond Myklebust unsigned int rqend; 581e7d39069STrond Myklebust unsigned int end; 5821da177e4SLinus Torvalds int error; 583277459d2STrond Myklebust 584e7d39069STrond Myklebust if (!PagePrivate(page)) 585e7d39069STrond Myklebust return NULL; 586e7d39069STrond Myklebust 587e7d39069STrond Myklebust end = offset + bytes; 588e7d39069STrond Myklebust spin_lock(&inode->i_lock); 589e7d39069STrond Myklebust 590e7d39069STrond Myklebust for (;;) { 591e7d39069STrond Myklebust req = nfs_page_find_request_locked(page); 592e7d39069STrond Myklebust if (req == NULL) 593e7d39069STrond Myklebust goto out_unlock; 594e7d39069STrond Myklebust 595e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 596e7d39069STrond Myklebust /* 597e7d39069STrond Myklebust * Tell the caller to flush out the request if 598e7d39069STrond Myklebust * the offsets are non-contiguous. 599e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 600e7d39069STrond Myklebust * have flushed out requests having wrong owners. 601e7d39069STrond Myklebust */ 602e468bae9STrond Myklebust if (offset > rqend 603e7d39069STrond Myklebust || end < req->wb_offset) 604e7d39069STrond Myklebust goto out_flushme; 605e7d39069STrond Myklebust 606e7d39069STrond Myklebust if (nfs_set_page_tag_locked(req)) 607e7d39069STrond Myklebust break; 608e7d39069STrond Myklebust 609e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 610587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6111da177e4SLinus Torvalds error = nfs_wait_on_request(req); 6121da177e4SLinus Torvalds nfs_release_request(req); 613e7d39069STrond Myklebust if (error != 0) 614e7d39069STrond Myklebust goto out_err; 615e7d39069STrond Myklebust spin_lock(&inode->i_lock); 6161da177e4SLinus Torvalds } 6171da177e4SLinus Torvalds 618ff778d02STrond Myklebust if (nfs_clear_request_commit(req) && 619e468bae9STrond Myklebust radix_tree_tag_clear(&NFS_I(inode)->nfs_page_tree, 620a861a1e1SFred Isaman req->wb_index, NFS_PAGE_TAG_COMMIT) != NULL) { 621ff778d02STrond Myklebust NFS_I(inode)->ncommit--; 622a861a1e1SFred Isaman pnfs_clear_request_commit(req); 623a861a1e1SFred Isaman } 624e468bae9STrond Myklebust 6251da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 6261da177e4SLinus Torvalds if (offset < req->wb_offset) { 6271da177e4SLinus Torvalds req->wb_offset = offset; 6281da177e4SLinus Torvalds req->wb_pgbase = offset; 6291da177e4SLinus Torvalds } 6301da177e4SLinus Torvalds if (end > rqend) 6311da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 632e7d39069STrond Myklebust else 633e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 634e7d39069STrond Myklebust out_unlock: 635e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 636e7d39069STrond Myklebust return req; 637e7d39069STrond Myklebust out_flushme: 638e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 639e7d39069STrond Myklebust nfs_release_request(req); 640e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 641e7d39069STrond Myklebust out_err: 642e7d39069STrond Myklebust return ERR_PTR(error); 643e7d39069STrond Myklebust } 6441da177e4SLinus Torvalds 645e7d39069STrond Myklebust /* 646e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 647e7d39069STrond Myklebust * 648e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 649e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 650e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 651e7d39069STrond Myklebust */ 652e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 653e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 654e7d39069STrond Myklebust { 655e7d39069STrond Myklebust struct inode *inode = page->mapping->host; 656e7d39069STrond Myklebust struct nfs_page *req; 657e7d39069STrond Myklebust int error; 658e7d39069STrond Myklebust 659e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 660e7d39069STrond Myklebust if (req != NULL) 661e7d39069STrond Myklebust goto out; 662e7d39069STrond Myklebust req = nfs_create_request(ctx, inode, page, offset, bytes); 663e7d39069STrond Myklebust if (IS_ERR(req)) 664e7d39069STrond Myklebust goto out; 665e7d39069STrond Myklebust error = nfs_inode_add_request(inode, req); 666e7d39069STrond Myklebust if (error != 0) { 667e7d39069STrond Myklebust nfs_release_request(req); 668e7d39069STrond Myklebust req = ERR_PTR(error); 669e7d39069STrond Myklebust } 670efc91ed0STrond Myklebust out: 67161e930a9STrond Myklebust return req; 6721da177e4SLinus Torvalds } 6731da177e4SLinus Torvalds 674e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 675e7d39069STrond Myklebust unsigned int offset, unsigned int count) 676e7d39069STrond Myklebust { 677e7d39069STrond Myklebust struct nfs_page *req; 678e7d39069STrond Myklebust 679e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 680e7d39069STrond Myklebust if (IS_ERR(req)) 681e7d39069STrond Myklebust return PTR_ERR(req); 682e7d39069STrond Myklebust /* Update file length */ 683e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 684e7d39069STrond Myklebust nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 685a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 686e7d39069STrond Myklebust nfs_clear_page_tag_locked(req); 687e7d39069STrond Myklebust return 0; 688e7d39069STrond Myklebust } 689e7d39069STrond Myklebust 6901da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 6911da177e4SLinus Torvalds { 692cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 6931da177e4SLinus Torvalds struct nfs_page *req; 6941a54533eSTrond Myklebust int do_flush, status; 6951da177e4SLinus Torvalds /* 6961da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 6971da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 6981da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 6991da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 7001da177e4SLinus Torvalds * Also do the same if we find a request from an existing 7011da177e4SLinus Torvalds * dropped page. 7021da177e4SLinus Torvalds */ 7031a54533eSTrond Myklebust do { 704277459d2STrond Myklebust req = nfs_page_find_request(page); 7051a54533eSTrond Myklebust if (req == NULL) 7061a54533eSTrond Myklebust return 0; 707f11ac8dbSTrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx || 708f11ac8dbSTrond Myklebust req->wb_lock_context->lockowner != current->files || 709f11ac8dbSTrond Myklebust req->wb_lock_context->pid != current->tgid; 7101da177e4SLinus Torvalds nfs_release_request(req); 7111a54533eSTrond Myklebust if (!do_flush) 7121a54533eSTrond Myklebust return 0; 713277459d2STrond Myklebust status = nfs_wb_page(page->mapping->host, page); 7141a54533eSTrond Myklebust } while (status == 0); 7151a54533eSTrond Myklebust return status; 7161da177e4SLinus Torvalds } 7171da177e4SLinus Torvalds 7181da177e4SLinus Torvalds /* 7195d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 7205d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 7215d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 7225d47a356STrond Myklebust */ 7235d47a356STrond Myklebust static int nfs_write_pageuptodate(struct page *page, struct inode *inode) 7245d47a356STrond Myklebust { 7255d47a356STrond Myklebust return PageUptodate(page) && 7265d47a356STrond Myklebust !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); 7275d47a356STrond Myklebust } 7285d47a356STrond Myklebust 7295d47a356STrond Myklebust /* 7301da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 7311da177e4SLinus Torvalds * 7321da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 7331da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 7341da177e4SLinus Torvalds */ 7351da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 7361da177e4SLinus Torvalds unsigned int offset, unsigned int count) 7371da177e4SLinus Torvalds { 738cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 7391da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 7401da177e4SLinus Torvalds int status = 0; 7411da177e4SLinus Torvalds 74291d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 74391d5b470SChuck Lever 74448186c7dSChuck Lever dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", 74501cce933SJosef "Jeff" Sipek file->f_path.dentry->d_parent->d_name.name, 74601cce933SJosef "Jeff" Sipek file->f_path.dentry->d_name.name, count, 7470bbacc40SChuck Lever (long long)(page_offset(page) + offset)); 7481da177e4SLinus Torvalds 7491da177e4SLinus Torvalds /* If we're not using byte range locks, and we know the page 7505d47a356STrond Myklebust * is up to date, it may be more efficient to extend the write 7515d47a356STrond Myklebust * to cover the entire page in order to avoid fragmentation 7525d47a356STrond Myklebust * inefficiencies. 7531da177e4SLinus Torvalds */ 7545d47a356STrond Myklebust if (nfs_write_pageuptodate(page, inode) && 7555d47a356STrond Myklebust inode->i_flock == NULL && 7566b2f3d1fSChristoph Hellwig !(file->f_flags & O_DSYNC)) { 75749a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 7581da177e4SLinus Torvalds offset = 0; 7591da177e4SLinus Torvalds } 7601da177e4SLinus Torvalds 761e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 76203fa9e84STrond Myklebust if (status < 0) 76303fa9e84STrond Myklebust nfs_set_pageerror(page); 76459b7c05fSTrond Myklebust else 76559b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 7661da177e4SLinus Torvalds 76748186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 7681da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 7691da177e4SLinus Torvalds return status; 7701da177e4SLinus Torvalds } 7711da177e4SLinus Torvalds 772a861a1e1SFred Isaman static void nfs_writepage_release(struct nfs_page *req, 773a861a1e1SFred Isaman struct nfs_write_data *data) 7741da177e4SLinus Torvalds { 775a6305ddbSTrond Myklebust struct page *page = req->wb_page; 7768e821cadSTrond Myklebust 777a861a1e1SFred Isaman if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req, data)) 7781da177e4SLinus Torvalds nfs_inode_remove_request(req); 7799fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 780a6305ddbSTrond Myklebust nfs_end_page_writeback(page); 7811da177e4SLinus Torvalds } 7821da177e4SLinus Torvalds 7833ff7576dSTrond Myklebust static int flush_task_priority(int how) 7841da177e4SLinus Torvalds { 7851da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 7861da177e4SLinus Torvalds case FLUSH_HIGHPRI: 7871da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 7881da177e4SLinus Torvalds case FLUSH_LOWPRI: 7891da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 7901da177e4SLinus Torvalds } 7911da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 7921da177e4SLinus Torvalds } 7931da177e4SLinus Torvalds 794a69aef14SFred Isaman int nfs_initiate_write(struct nfs_write_data *data, 795d138d5d1SAndy Adamson struct rpc_clnt *clnt, 796788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 7971da177e4SLinus Torvalds int how) 7981da177e4SLinus Torvalds { 799d138d5d1SAndy Adamson struct inode *inode = data->inode; 8003ff7576dSTrond Myklebust int priority = flush_task_priority(how); 80107737691STrond Myklebust struct rpc_task *task; 802bdc7f021STrond Myklebust struct rpc_message msg = { 803bdc7f021STrond Myklebust .rpc_argp = &data->args, 804bdc7f021STrond Myklebust .rpc_resp = &data->res, 805d138d5d1SAndy Adamson .rpc_cred = data->cred, 806bdc7f021STrond Myklebust }; 80784115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 808d138d5d1SAndy Adamson .rpc_client = clnt, 80907737691STrond Myklebust .task = &data->task, 810bdc7f021STrond Myklebust .rpc_message = &msg, 81184115e1cSTrond Myklebust .callback_ops = call_ops, 81284115e1cSTrond Myklebust .callback_data = data, 813101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 8142c61be0aSTrond Myklebust .flags = RPC_TASK_ASYNC, 8153ff7576dSTrond Myklebust .priority = priority, 81684115e1cSTrond Myklebust }; 8172c61be0aSTrond Myklebust int ret = 0; 8181da177e4SLinus Torvalds 819d138d5d1SAndy Adamson /* Set up the initial task struct. */ 820d138d5d1SAndy Adamson NFS_PROTO(inode)->write_setup(data, &msg); 821d138d5d1SAndy Adamson 822d138d5d1SAndy Adamson dprintk("NFS: %5u initiated write call " 823d138d5d1SAndy Adamson "(req %s/%lld, %u bytes @ offset %llu)\n", 824d138d5d1SAndy Adamson data->task.tk_pid, 825d138d5d1SAndy Adamson inode->i_sb->s_id, 826d138d5d1SAndy Adamson (long long)NFS_FILEID(inode), 827d138d5d1SAndy Adamson data->args.count, 828d138d5d1SAndy Adamson (unsigned long long)data->args.offset); 829d138d5d1SAndy Adamson 830d138d5d1SAndy Adamson task = rpc_run_task(&task_setup_data); 831d138d5d1SAndy Adamson if (IS_ERR(task)) { 832d138d5d1SAndy Adamson ret = PTR_ERR(task); 833d138d5d1SAndy Adamson goto out; 834d138d5d1SAndy Adamson } 835d138d5d1SAndy Adamson if (how & FLUSH_SYNC) { 836d138d5d1SAndy Adamson ret = rpc_wait_for_completion_task(task); 837d138d5d1SAndy Adamson if (ret == 0) 838d138d5d1SAndy Adamson ret = task->tk_status; 839d138d5d1SAndy Adamson } 840d138d5d1SAndy Adamson rpc_put_task(task); 841d138d5d1SAndy Adamson out: 842d138d5d1SAndy Adamson return ret; 843d138d5d1SAndy Adamson } 844a69aef14SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_write); 845d138d5d1SAndy Adamson 846d138d5d1SAndy Adamson /* 847d138d5d1SAndy Adamson * Set up the argument/result storage required for the RPC call. 848d138d5d1SAndy Adamson */ 8496e4efd56STrond Myklebust static void nfs_write_rpcsetup(struct nfs_page *req, 850d138d5d1SAndy Adamson struct nfs_write_data *data, 851d138d5d1SAndy Adamson unsigned int count, unsigned int offset, 852d138d5d1SAndy Adamson int how) 853d138d5d1SAndy Adamson { 8543d4ff43dSAl Viro struct inode *inode = req->wb_context->dentry->d_inode; 855d138d5d1SAndy Adamson 8561da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 8571da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 8581da177e4SLinus Torvalds 8591da177e4SLinus Torvalds data->req = req; 8603d4ff43dSAl Viro data->inode = inode = req->wb_context->dentry->d_inode; 861d138d5d1SAndy Adamson data->cred = req->wb_context->cred; 8621da177e4SLinus Torvalds 8631da177e4SLinus Torvalds data->args.fh = NFS_FH(inode); 8641da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 8652bea038cSBoaz Harrosh /* pnfs_set_layoutcommit needs this */ 8662bea038cSBoaz Harrosh data->mds_offset = data->args.offset; 8671da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 8681da177e4SLinus Torvalds data->args.pages = data->pagevec; 8691da177e4SLinus Torvalds data->args.count = count; 870383ba719STrond Myklebust data->args.context = get_nfs_open_context(req->wb_context); 871f11ac8dbSTrond Myklebust data->args.lock_context = req->wb_lock_context; 872bdc7f021STrond Myklebust data->args.stable = NFS_UNSTABLE; 87387ed5eb4STrond Myklebust switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { 87487ed5eb4STrond Myklebust case 0: 87587ed5eb4STrond Myklebust break; 87687ed5eb4STrond Myklebust case FLUSH_COND_STABLE: 87787ed5eb4STrond Myklebust if (nfs_need_commit(NFS_I(inode))) 87887ed5eb4STrond Myklebust break; 87987ed5eb4STrond Myklebust default: 880bdc7f021STrond Myklebust data->args.stable = NFS_FILE_SYNC; 881bdc7f021STrond Myklebust } 8821da177e4SLinus Torvalds 8831da177e4SLinus Torvalds data->res.fattr = &data->fattr; 8841da177e4SLinus Torvalds data->res.count = count; 8851da177e4SLinus Torvalds data->res.verf = &data->verf; 8860e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 8876e4efd56STrond Myklebust } 8881da177e4SLinus Torvalds 8896e4efd56STrond Myklebust static int nfs_do_write(struct nfs_write_data *data, 8906e4efd56STrond Myklebust const struct rpc_call_ops *call_ops, 8916e4efd56STrond Myklebust int how) 8926e4efd56STrond Myklebust { 8935f00bcb3SStephen Rothwell struct inode *inode = data->args.context->dentry->d_inode; 8940382b744SAndy Adamson 895d138d5d1SAndy Adamson return nfs_initiate_write(data, NFS_CLIENT(inode), call_ops, how); 8961da177e4SLinus Torvalds } 8971da177e4SLinus Torvalds 898275acaafSTrond Myklebust static int nfs_do_multiple_writes(struct list_head *head, 899275acaafSTrond Myklebust const struct rpc_call_ops *call_ops, 900275acaafSTrond Myklebust int how) 901275acaafSTrond Myklebust { 902275acaafSTrond Myklebust struct nfs_write_data *data; 903275acaafSTrond Myklebust int ret = 0; 904275acaafSTrond Myklebust 905275acaafSTrond Myklebust while (!list_empty(head)) { 906275acaafSTrond Myklebust int ret2; 907275acaafSTrond Myklebust 908275acaafSTrond Myklebust data = list_entry(head->next, struct nfs_write_data, list); 909275acaafSTrond Myklebust list_del_init(&data->list); 910275acaafSTrond Myklebust 911dce81290STrond Myklebust ret2 = nfs_do_write(data, call_ops, how); 912275acaafSTrond Myklebust if (ret == 0) 913275acaafSTrond Myklebust ret = ret2; 914275acaafSTrond Myklebust } 915275acaafSTrond Myklebust return ret; 916275acaafSTrond Myklebust } 917275acaafSTrond Myklebust 9186d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 9196d884e8fSFred * call this on each, which will prepare them to be retried on next 9206d884e8fSFred * writeback using standard nfs. 9216d884e8fSFred */ 9226d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 9236d884e8fSFred { 924a6305ddbSTrond Myklebust struct page *page = req->wb_page; 925a6305ddbSTrond Myklebust 9266d884e8fSFred nfs_mark_request_dirty(req); 9276d884e8fSFred nfs_clear_page_tag_locked(req); 928a6305ddbSTrond Myklebust nfs_end_page_writeback(page); 9296d884e8fSFred } 9306d884e8fSFred 9311da177e4SLinus Torvalds /* 9321da177e4SLinus Torvalds * Generate multiple small requests to write out a single 9331da177e4SLinus Torvalds * contiguous dirty area on one page. 9341da177e4SLinus Torvalds */ 935275acaafSTrond Myklebust static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, struct list_head *res) 9361da177e4SLinus Torvalds { 937c76069bdSFred Isaman struct nfs_page *req = nfs_list_entry(desc->pg_list.next); 9381da177e4SLinus Torvalds struct page *page = req->wb_page; 9391da177e4SLinus Torvalds struct nfs_write_data *data; 940d097971dSTrond Myklebust size_t wsize = desc->pg_bsize, nbytes; 941e9f7bee1STrond Myklebust unsigned int offset; 9421da177e4SLinus Torvalds int requests = 0; 943dbae4c73STrond Myklebust int ret = 0; 9441da177e4SLinus Torvalds 9451da177e4SLinus Torvalds nfs_list_remove_request(req); 9461da177e4SLinus Torvalds 947b31268acSTrond Myklebust if ((desc->pg_ioflags & FLUSH_COND_STABLE) && 948b31268acSTrond Myklebust (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit || 949b31268acSTrond Myklebust desc->pg_count > wsize)) 950b31268acSTrond Myklebust desc->pg_ioflags &= ~FLUSH_COND_STABLE; 951b31268acSTrond Myklebust 952b31268acSTrond Myklebust 953275acaafSTrond Myklebust offset = 0; 954c76069bdSFred Isaman nbytes = desc->pg_count; 955e9f7bee1STrond Myklebust do { 956e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 957e9f7bee1STrond Myklebust 9588d5658c9STrond Myklebust data = nfs_writedata_alloc(1); 9591da177e4SLinus Torvalds if (!data) 9601da177e4SLinus Torvalds goto out_bad; 961275acaafSTrond Myklebust data->pagevec[0] = page; 962f13c3620STrond Myklebust nfs_write_rpcsetup(req, data, len, offset, desc->pg_ioflags); 963275acaafSTrond Myklebust list_add(&data->list, res); 9641da177e4SLinus Torvalds requests++; 965e9f7bee1STrond Myklebust nbytes -= len; 966275acaafSTrond Myklebust offset += len; 967e9f7bee1STrond Myklebust } while (nbytes != 0); 9681da177e4SLinus Torvalds atomic_set(&req->wb_complete, requests); 96950828d7eSTrond Myklebust desc->pg_rpc_callops = &nfs_write_partial_ops; 970dbae4c73STrond Myklebust return ret; 9711da177e4SLinus Torvalds 9721da177e4SLinus Torvalds out_bad: 973275acaafSTrond Myklebust while (!list_empty(res)) { 974275acaafSTrond Myklebust data = list_entry(res->next, struct nfs_write_data, list); 9756e4efd56STrond Myklebust list_del(&data->list); 9760da2a4acSFred Isaman nfs_writedata_free(data); 9771da177e4SLinus Torvalds } 97861822ab5STrond Myklebust nfs_redirty_request(req); 9791da177e4SLinus Torvalds return -ENOMEM; 9801da177e4SLinus Torvalds } 9811da177e4SLinus Torvalds 9821da177e4SLinus Torvalds /* 9831da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 9841da177e4SLinus Torvalds * The page must have been locked by the caller. 9851da177e4SLinus Torvalds * 9861da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 9871da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 9881da177e4SLinus Torvalds * that has been written but not committed. 9891da177e4SLinus Torvalds */ 990275acaafSTrond Myklebust static int nfs_flush_one(struct nfs_pageio_descriptor *desc, struct list_head *res) 9911da177e4SLinus Torvalds { 9921da177e4SLinus Torvalds struct nfs_page *req; 9931da177e4SLinus Torvalds struct page **pages; 9941da177e4SLinus Torvalds struct nfs_write_data *data; 995c76069bdSFred Isaman struct list_head *head = &desc->pg_list; 9963b609184SPeng Tao int ret = 0; 9971da177e4SLinus Torvalds 998c76069bdSFred Isaman data = nfs_writedata_alloc(nfs_page_array_len(desc->pg_base, 999c76069bdSFred Isaman desc->pg_count)); 100044b83799SFred Isaman if (!data) { 100144b83799SFred Isaman while (!list_empty(head)) { 100244b83799SFred Isaman req = nfs_list_entry(head->next); 100344b83799SFred Isaman nfs_list_remove_request(req); 100444b83799SFred Isaman nfs_redirty_request(req); 100544b83799SFred Isaman } 100644b83799SFred Isaman ret = -ENOMEM; 100744b83799SFred Isaman goto out; 100844b83799SFred Isaman } 10091da177e4SLinus Torvalds pages = data->pagevec; 10101da177e4SLinus Torvalds while (!list_empty(head)) { 10111da177e4SLinus Torvalds req = nfs_list_entry(head->next); 10121da177e4SLinus Torvalds nfs_list_remove_request(req); 10131da177e4SLinus Torvalds nfs_list_add_request(req, &data->pages); 10141da177e4SLinus Torvalds ClearPageError(req->wb_page); 10151da177e4SLinus Torvalds *pages++ = req->wb_page; 10161da177e4SLinus Torvalds } 10171da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 10181da177e4SLinus Torvalds 1019b31268acSTrond Myklebust if ((desc->pg_ioflags & FLUSH_COND_STABLE) && 1020b31268acSTrond Myklebust (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit)) 1021b31268acSTrond Myklebust desc->pg_ioflags &= ~FLUSH_COND_STABLE; 1022b31268acSTrond Myklebust 10231da177e4SLinus Torvalds /* Set up the argument struct */ 10246e4efd56STrond Myklebust nfs_write_rpcsetup(req, data, desc->pg_count, 0, desc->pg_ioflags); 1025275acaafSTrond Myklebust list_add(&data->list, res); 102650828d7eSTrond Myklebust desc->pg_rpc_callops = &nfs_write_full_ops; 102744b83799SFred Isaman out: 102844b83799SFred Isaman return ret; 10291da177e4SLinus Torvalds } 10301da177e4SLinus Torvalds 1031dce81290STrond Myklebust int nfs_generic_flush(struct nfs_pageio_descriptor *desc, struct list_head *head) 1032dce81290STrond Myklebust { 1033dce81290STrond Myklebust if (desc->pg_bsize < PAGE_CACHE_SIZE) 1034dce81290STrond Myklebust return nfs_flush_multi(desc, head); 1035dce81290STrond Myklebust return nfs_flush_one(desc, head); 1036dce81290STrond Myklebust } 1037dce81290STrond Myklebust 1038dce81290STrond Myklebust static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) 10391751c363STrond Myklebust { 1040275acaafSTrond Myklebust LIST_HEAD(head); 1041275acaafSTrond Myklebust int ret; 1042275acaafSTrond Myklebust 1043dce81290STrond Myklebust ret = nfs_generic_flush(desc, &head); 1044275acaafSTrond Myklebust if (ret == 0) 104550828d7eSTrond Myklebust ret = nfs_do_multiple_writes(&head, desc->pg_rpc_callops, 1046dce81290STrond Myklebust desc->pg_ioflags); 1047275acaafSTrond Myklebust return ret; 10481751c363STrond Myklebust } 10491751c363STrond Myklebust 10501751c363STrond Myklebust static const struct nfs_pageio_ops nfs_pageio_write_ops = { 10511751c363STrond Myklebust .pg_test = nfs_generic_pg_test, 10521751c363STrond Myklebust .pg_doio = nfs_generic_pg_writepages, 10531751c363STrond Myklebust }; 10541751c363STrond Myklebust 10551f945357STrond Myklebust static void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio, 10561751c363STrond Myklebust struct inode *inode, int ioflags) 10571751c363STrond Myklebust { 10581751c363STrond Myklebust nfs_pageio_init(pgio, inode, &nfs_pageio_write_ops, 10591751c363STrond Myklebust NFS_SERVER(inode)->wsize, ioflags); 10601751c363STrond Myklebust } 10611751c363STrond Myklebust 1062dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1063dce81290STrond Myklebust { 1064dce81290STrond Myklebust pgio->pg_ops = &nfs_pageio_write_ops; 1065dce81290STrond Myklebust pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1066dce81290STrond Myklebust } 10671f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1068dce81290STrond Myklebust 1069c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1070c63c7b05STrond Myklebust struct inode *inode, int ioflags) 10711da177e4SLinus Torvalds { 10721751c363STrond Myklebust if (!pnfs_pageio_init_write(pgio, inode, ioflags)) 10731751c363STrond Myklebust nfs_pageio_init_write_mds(pgio, inode, ioflags); 10741da177e4SLinus Torvalds } 10751da177e4SLinus Torvalds 10761da177e4SLinus Torvalds /* 10771da177e4SLinus Torvalds * Handle a write reply that flushed part of a page. 10781da177e4SLinus Torvalds */ 1079788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) 10801da177e4SLinus Torvalds { 1081788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 10821da177e4SLinus Torvalds 108348186c7dSChuck Lever dprintk("NFS: %5u write(%s/%lld %d@%lld)", 108448186c7dSChuck Lever task->tk_pid, 10853d4ff43dSAl Viro data->req->wb_context->dentry->d_inode->i_sb->s_id, 108648186c7dSChuck Lever (long long) 10873d4ff43dSAl Viro NFS_FILEID(data->req->wb_context->dentry->d_inode), 108848186c7dSChuck Lever data->req->wb_bytes, (long long)req_offset(data->req)); 10891da177e4SLinus Torvalds 1090c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 1091c9d8f89dSTrond Myklebust } 1092788e7a89STrond Myklebust 1093c9d8f89dSTrond Myklebust static void nfs_writeback_release_partial(void *calldata) 1094c9d8f89dSTrond Myklebust { 1095c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1096c9d8f89dSTrond Myklebust struct nfs_page *req = data->req; 1097c9d8f89dSTrond Myklebust struct page *page = req->wb_page; 1098c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1099c9d8f89dSTrond Myklebust 1100c9d8f89dSTrond Myklebust if (status < 0) { 1101a301b777STrond Myklebust nfs_set_pageerror(page); 1102c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 1103c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 11048e821cadSTrond Myklebust goto out; 11058e821cadSTrond Myklebust } 11068e821cadSTrond Myklebust 11078e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 1108587142f8STrond Myklebust struct inode *inode = page->mapping->host; 11098e821cadSTrond Myklebust 1110587142f8STrond Myklebust spin_lock(&inode->i_lock); 11118e821cadSTrond Myklebust if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { 11128e821cadSTrond Myklebust /* Do nothing we need to resend the writes */ 11138e821cadSTrond Myklebust } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { 11141da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 11151da177e4SLinus Torvalds dprintk(" defer commit\n"); 11161da177e4SLinus Torvalds } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { 11178e821cadSTrond Myklebust set_bit(PG_NEED_RESCHED, &req->wb_flags); 11188e821cadSTrond Myklebust clear_bit(PG_NEED_COMMIT, &req->wb_flags); 11191da177e4SLinus Torvalds dprintk(" server reboot detected\n"); 11201da177e4SLinus Torvalds } 1121587142f8STrond Myklebust spin_unlock(&inode->i_lock); 11221da177e4SLinus Torvalds } else 11231da177e4SLinus Torvalds dprintk(" OK\n"); 11241da177e4SLinus Torvalds 11258e821cadSTrond Myklebust out: 11261da177e4SLinus Torvalds if (atomic_dec_and_test(&req->wb_complete)) 1127a861a1e1SFred Isaman nfs_writepage_release(req, data); 1128c9d8f89dSTrond Myklebust nfs_writedata_release(calldata); 11291da177e4SLinus Torvalds } 11301da177e4SLinus Torvalds 1131def6ed7eSAndy Adamson #if defined(CONFIG_NFS_V4_1) 1132def6ed7eSAndy Adamson void nfs_write_prepare(struct rpc_task *task, void *calldata) 1133def6ed7eSAndy Adamson { 1134def6ed7eSAndy Adamson struct nfs_write_data *data = calldata; 1135def6ed7eSAndy Adamson 1136035168abSTrond Myklebust if (nfs4_setup_sequence(NFS_SERVER(data->inode), 1137035168abSTrond Myklebust &data->args.seq_args, 1138def6ed7eSAndy Adamson &data->res.seq_res, 1, task)) 1139def6ed7eSAndy Adamson return; 1140def6ed7eSAndy Adamson rpc_call_start(task); 1141def6ed7eSAndy Adamson } 1142def6ed7eSAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1143def6ed7eSAndy Adamson 1144788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_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_partial, 1149c9d8f89dSTrond Myklebust .rpc_release = nfs_writeback_release_partial, 1150788e7a89STrond Myklebust }; 1151788e7a89STrond Myklebust 11521da177e4SLinus Torvalds /* 11531da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 11541da177e4SLinus Torvalds * 11551da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 11561da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 11571da177e4SLinus Torvalds * as the page has a write request pending. 11581da177e4SLinus Torvalds */ 1159788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) 11601da177e4SLinus Torvalds { 1161788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 11621da177e4SLinus Torvalds 1163c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 1164c9d8f89dSTrond Myklebust } 1165c9d8f89dSTrond Myklebust 1166c9d8f89dSTrond Myklebust static void nfs_writeback_release_full(void *calldata) 1167c9d8f89dSTrond Myklebust { 1168c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1169c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1170788e7a89STrond Myklebust 11711da177e4SLinus Torvalds /* Update attributes as result of writeback. */ 11721da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 1173c9d8f89dSTrond Myklebust struct nfs_page *req = nfs_list_entry(data->pages.next); 1174c9d8f89dSTrond Myklebust struct page *page = req->wb_page; 1175c9d8f89dSTrond Myklebust 11761da177e4SLinus Torvalds nfs_list_remove_request(req); 11771da177e4SLinus Torvalds 117848186c7dSChuck Lever dprintk("NFS: %5u write (%s/%lld %d@%lld)", 117948186c7dSChuck Lever data->task.tk_pid, 11803d4ff43dSAl Viro req->wb_context->dentry->d_inode->i_sb->s_id, 11813d4ff43dSAl Viro (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 11821da177e4SLinus Torvalds req->wb_bytes, 11831da177e4SLinus Torvalds (long long)req_offset(req)); 11841da177e4SLinus Torvalds 1185c9d8f89dSTrond Myklebust if (status < 0) { 1186a301b777STrond Myklebust nfs_set_pageerror(page); 1187c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 1188c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 11898e821cadSTrond Myklebust goto remove_request; 11901da177e4SLinus Torvalds } 11911da177e4SLinus Torvalds 11928e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 11931da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 1194a861a1e1SFred Isaman nfs_mark_request_commit(req, data->lseg); 11951da177e4SLinus Torvalds dprintk(" marked for commit\n"); 11968e821cadSTrond Myklebust goto next; 11978e821cadSTrond Myklebust } 11988e821cadSTrond Myklebust dprintk(" OK\n"); 11998e821cadSTrond Myklebust remove_request: 12001da177e4SLinus Torvalds nfs_inode_remove_request(req); 12011da177e4SLinus Torvalds next: 12029fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 1203a6305ddbSTrond Myklebust nfs_end_page_writeback(page); 12041da177e4SLinus Torvalds } 1205c9d8f89dSTrond Myklebust nfs_writedata_release(calldata); 12061da177e4SLinus Torvalds } 12071da177e4SLinus Torvalds 1208788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = { 1209def6ed7eSAndy Adamson #if defined(CONFIG_NFS_V4_1) 1210def6ed7eSAndy Adamson .rpc_call_prepare = nfs_write_prepare, 1211def6ed7eSAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1212788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_full, 1213c9d8f89dSTrond Myklebust .rpc_release = nfs_writeback_release_full, 1214788e7a89STrond Myklebust }; 1215788e7a89STrond Myklebust 1216788e7a89STrond Myklebust 12171da177e4SLinus Torvalds /* 12181da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 12191da177e4SLinus Torvalds */ 122013602896SFred Isaman void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 12211da177e4SLinus Torvalds { 12221da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 12231da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1224eedc020eSAndy Adamson struct nfs_server *server = NFS_SERVER(data->inode); 1225788e7a89STrond Myklebust int status; 12261da177e4SLinus Torvalds 1227a3f565b1SChuck Lever dprintk("NFS: %5u nfs_writeback_done (status %d)\n", 12281da177e4SLinus Torvalds task->tk_pid, task->tk_status); 12291da177e4SLinus Torvalds 1230f551e44fSChuck Lever /* 1231f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1232f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1233f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1234f551e44fSChuck Lever * another writer had changed the file, but some applications 1235f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1236f551e44fSChuck Lever */ 1237788e7a89STrond Myklebust status = NFS_PROTO(data->inode)->write_done(task, data); 1238788e7a89STrond Myklebust if (status != 0) 123913602896SFred Isaman return; 124091d5b470SChuck Lever nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 124191d5b470SChuck Lever 12421da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 12431da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 12441da177e4SLinus Torvalds /* We tried a write call, but the server did not 12451da177e4SLinus Torvalds * commit data to stable storage even though we 12461da177e4SLinus Torvalds * requested it. 12471da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 12481da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 12491da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 12501da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 12511da177e4SLinus Torvalds */ 12521da177e4SLinus Torvalds static unsigned long complain; 12531da177e4SLinus Torvalds 1254a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 12551da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 12561da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 12571da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1258eedc020eSAndy Adamson server->nfs_client->cl_hostname, 12591da177e4SLinus Torvalds resp->verf->committed, argp->stable); 12601da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 12611da177e4SLinus Torvalds } 12621da177e4SLinus Torvalds } 12631da177e4SLinus Torvalds #endif 12641da177e4SLinus Torvalds /* Is this a short write? */ 12651da177e4SLinus Torvalds if (task->tk_status >= 0 && resp->count < argp->count) { 12661da177e4SLinus Torvalds static unsigned long complain; 12671da177e4SLinus Torvalds 126891d5b470SChuck Lever nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); 126991d5b470SChuck Lever 12701da177e4SLinus Torvalds /* Has the server at least made some progress? */ 12711da177e4SLinus Torvalds if (resp->count != 0) { 12721da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 12731da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 12741da177e4SLinus Torvalds /* Resend from where the server left off */ 1275a69aef14SFred Isaman data->mds_offset += resp->count; 12761da177e4SLinus Torvalds argp->offset += resp->count; 12771da177e4SLinus Torvalds argp->pgbase += resp->count; 12781da177e4SLinus Torvalds argp->count -= resp->count; 12791da177e4SLinus Torvalds } else { 12801da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 12811da177e4SLinus Torvalds * headaches in the case of a server crash. 12821da177e4SLinus Torvalds */ 12831da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 12841da177e4SLinus Torvalds } 12850110ee15STrond Myklebust nfs_restart_rpc(task, server->nfs_client); 128613602896SFred Isaman return; 12871da177e4SLinus Torvalds } 12881da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 12891da177e4SLinus Torvalds printk(KERN_WARNING 12901da177e4SLinus Torvalds "NFS: Server wrote zero bytes, expected %u.\n", 12911da177e4SLinus Torvalds argp->count); 12921da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 12931da177e4SLinus Torvalds } 12941da177e4SLinus Torvalds /* Can't do anything about it except throw an error. */ 12951da177e4SLinus Torvalds task->tk_status = -EIO; 12961da177e4SLinus Torvalds } 129713602896SFred Isaman return; 12981da177e4SLinus Torvalds } 12991da177e4SLinus Torvalds 13001da177e4SLinus Torvalds 13011da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 130271d0a611STrond Myklebust static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) 130371d0a611STrond Myklebust { 1304b8413f98STrond Myklebust int ret; 1305b8413f98STrond Myklebust 130671d0a611STrond Myklebust if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) 130771d0a611STrond Myklebust return 1; 1308b8413f98STrond Myklebust if (!may_wait) 130971d0a611STrond Myklebust return 0; 1310b8413f98STrond Myklebust ret = out_of_line_wait_on_bit_lock(&nfsi->flags, 1311b8413f98STrond Myklebust NFS_INO_COMMIT, 1312b8413f98STrond Myklebust nfs_wait_bit_killable, 1313b8413f98STrond Myklebust TASK_KILLABLE); 1314b8413f98STrond Myklebust return (ret < 0) ? ret : 1; 131571d0a611STrond Myklebust } 131671d0a611STrond Myklebust 1317e0c2b380SFred Isaman void nfs_commit_clear_lock(struct nfs_inode *nfsi) 131871d0a611STrond Myklebust { 131971d0a611STrond Myklebust clear_bit(NFS_INO_COMMIT, &nfsi->flags); 132071d0a611STrond Myklebust smp_mb__after_clear_bit(); 132171d0a611STrond Myklebust wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); 132271d0a611STrond Myklebust } 1323e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_clear_lock); 132471d0a611STrond Myklebust 1325e0c2b380SFred Isaman void nfs_commitdata_release(void *data) 13261da177e4SLinus Torvalds { 1327383ba719STrond Myklebust struct nfs_write_data *wdata = data; 1328383ba719STrond Myklebust 1329988b6dceSFred Isaman put_lseg(wdata->lseg); 1330383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 13311da177e4SLinus Torvalds nfs_commit_free(wdata); 13321da177e4SLinus Torvalds } 1333e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 13341da177e4SLinus Torvalds 1335e0c2b380SFred Isaman int nfs_initiate_commit(struct nfs_write_data *data, struct rpc_clnt *clnt, 13369ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 1337788e7a89STrond Myklebust int how) 13381da177e4SLinus Torvalds { 133907737691STrond Myklebust struct rpc_task *task; 13409ace33cdSFred Isaman int priority = flush_task_priority(how); 1341bdc7f021STrond Myklebust struct rpc_message msg = { 1342bdc7f021STrond Myklebust .rpc_argp = &data->args, 1343bdc7f021STrond Myklebust .rpc_resp = &data->res, 13449ace33cdSFred Isaman .rpc_cred = data->cred, 1345bdc7f021STrond Myklebust }; 134684115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 134707737691STrond Myklebust .task = &data->task, 13489ace33cdSFred Isaman .rpc_client = clnt, 1349bdc7f021STrond Myklebust .rpc_message = &msg, 13509ace33cdSFred Isaman .callback_ops = call_ops, 135184115e1cSTrond Myklebust .callback_data = data, 1352101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 13532c61be0aSTrond Myklebust .flags = RPC_TASK_ASYNC, 13543ff7576dSTrond Myklebust .priority = priority, 135584115e1cSTrond Myklebust }; 1356788e7a89STrond Myklebust /* Set up the initial task struct. */ 13579ace33cdSFred Isaman NFS_PROTO(data->inode)->commit_setup(data, &msg); 13581da177e4SLinus Torvalds 1359a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1360bdc7f021STrond Myklebust 136107737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1362dbae4c73STrond Myklebust if (IS_ERR(task)) 1363dbae4c73STrond Myklebust return PTR_ERR(task); 1364d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1365d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 136607737691STrond Myklebust rpc_put_task(task); 1367dbae4c73STrond Myklebust return 0; 13681da177e4SLinus Torvalds } 1369e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 13701da177e4SLinus Torvalds 13711da177e4SLinus Torvalds /* 13729ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 13739ace33cdSFred Isaman */ 1374e0c2b380SFred Isaman void nfs_init_commit(struct nfs_write_data *data, 1375988b6dceSFred Isaman struct list_head *head, 1376988b6dceSFred Isaman struct pnfs_layout_segment *lseg) 13779ace33cdSFred Isaman { 13789ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 13793d4ff43dSAl Viro struct inode *inode = first->wb_context->dentry->d_inode; 13809ace33cdSFred Isaman 13819ace33cdSFred Isaman /* Set up the RPC argument and reply structs 13829ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 13839ace33cdSFred Isaman 13849ace33cdSFred Isaman list_splice_init(head, &data->pages); 13859ace33cdSFred Isaman 13869ace33cdSFred Isaman data->inode = inode; 13879ace33cdSFred Isaman data->cred = first->wb_context->cred; 1388988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 13899ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 13909ace33cdSFred Isaman 13919ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 13929ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 13939ace33cdSFred Isaman data->args.offset = 0; 13949ace33cdSFred Isaman data->args.count = 0; 13959ace33cdSFred Isaman data->args.context = get_nfs_open_context(first->wb_context); 13969ace33cdSFred Isaman data->res.count = 0; 13979ace33cdSFred Isaman data->res.fattr = &data->fattr; 13989ace33cdSFred Isaman data->res.verf = &data->verf; 13999ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 14009ace33cdSFred Isaman } 1401e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 14029ace33cdSFred Isaman 1403e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1404a861a1e1SFred Isaman struct pnfs_layout_segment *lseg) 140564bfeb49SFred Isaman { 140664bfeb49SFred Isaman struct nfs_page *req; 140764bfeb49SFred Isaman 140864bfeb49SFred Isaman while (!list_empty(page_list)) { 140964bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 141064bfeb49SFred Isaman nfs_list_remove_request(req); 1411a861a1e1SFred Isaman nfs_mark_request_commit(req, lseg); 141264bfeb49SFred Isaman dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 141364bfeb49SFred Isaman dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 141464bfeb49SFred Isaman BDI_RECLAIMABLE); 141564bfeb49SFred Isaman nfs_clear_page_tag_locked(req); 141664bfeb49SFred Isaman } 141764bfeb49SFred Isaman } 1418e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 141964bfeb49SFred Isaman 14209ace33cdSFred Isaman /* 14211da177e4SLinus Torvalds * Commit dirty pages 14221da177e4SLinus Torvalds */ 14231da177e4SLinus Torvalds static int 142440859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how) 14251da177e4SLinus Torvalds { 14261da177e4SLinus Torvalds struct nfs_write_data *data; 14271da177e4SLinus Torvalds 1428c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 14291da177e4SLinus Torvalds 14301da177e4SLinus Torvalds if (!data) 14311da177e4SLinus Torvalds goto out_bad; 14321da177e4SLinus Torvalds 14331da177e4SLinus Torvalds /* Set up the argument struct */ 1434988b6dceSFred Isaman nfs_init_commit(data, head, NULL); 14359ace33cdSFred Isaman return nfs_initiate_commit(data, NFS_CLIENT(inode), data->mds_ops, how); 14361da177e4SLinus Torvalds out_bad: 1437a861a1e1SFred Isaman nfs_retry_commit(head, NULL); 143871d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(inode)); 14391da177e4SLinus Torvalds return -ENOMEM; 14401da177e4SLinus Torvalds } 14411da177e4SLinus Torvalds 14421da177e4SLinus Torvalds /* 14431da177e4SLinus Torvalds * COMMIT call returned 14441da177e4SLinus Torvalds */ 1445788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 14461da177e4SLinus Torvalds { 1447963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 14481da177e4SLinus Torvalds 1449a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 14501da177e4SLinus Torvalds task->tk_pid, task->tk_status); 14511da177e4SLinus Torvalds 1452788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1453c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1454c9d8f89dSTrond Myklebust } 1455c9d8f89dSTrond Myklebust 1456e0c2b380SFred Isaman void nfs_commit_release_pages(struct nfs_write_data *data) 1457c9d8f89dSTrond Myklebust { 1458c9d8f89dSTrond Myklebust struct nfs_page *req; 1459c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1460788e7a89STrond Myklebust 14611da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 14621da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 14631da177e4SLinus Torvalds nfs_list_remove_request(req); 1464e468bae9STrond Myklebust nfs_clear_request_commit(req); 14651da177e4SLinus Torvalds 146648186c7dSChuck Lever dprintk("NFS: commit (%s/%lld %d@%lld)", 14673d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 14683d4ff43dSAl Viro (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 14691da177e4SLinus Torvalds req->wb_bytes, 14701da177e4SLinus Torvalds (long long)req_offset(req)); 1471c9d8f89dSTrond Myklebust if (status < 0) { 1472c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 14731da177e4SLinus Torvalds nfs_inode_remove_request(req); 1474c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 14751da177e4SLinus Torvalds goto next; 14761da177e4SLinus Torvalds } 14771da177e4SLinus Torvalds 14781da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 14791da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 14801da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 14811da177e4SLinus Torvalds /* We have a match */ 14821da177e4SLinus Torvalds nfs_inode_remove_request(req); 14831da177e4SLinus Torvalds dprintk(" OK\n"); 14841da177e4SLinus Torvalds goto next; 14851da177e4SLinus Torvalds } 14861da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 14871da177e4SLinus Torvalds dprintk(" mismatch\n"); 14886d884e8fSFred nfs_mark_request_dirty(req); 14891da177e4SLinus Torvalds next: 14909fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 14911da177e4SLinus Torvalds } 14925917ce84SFred Isaman } 1493e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_release_pages); 14945917ce84SFred Isaman 14955917ce84SFred Isaman static void nfs_commit_release(void *calldata) 14965917ce84SFred Isaman { 14975917ce84SFred Isaman struct nfs_write_data *data = calldata; 14985917ce84SFred Isaman 14995917ce84SFred Isaman nfs_commit_release_pages(data); 150071d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(data->inode)); 1501c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 15021da177e4SLinus Torvalds } 1503788e7a89STrond Myklebust 1504788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 150521d9a851SAndy Adamson #if defined(CONFIG_NFS_V4_1) 150621d9a851SAndy Adamson .rpc_call_prepare = nfs_write_prepare, 150721d9a851SAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1508788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1509788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1510788e7a89STrond Myklebust }; 15111da177e4SLinus Torvalds 1512b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 15131da177e4SLinus Torvalds { 15141da177e4SLinus Torvalds LIST_HEAD(head); 151571d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1516b8413f98STrond Myklebust int res; 15171da177e4SLinus Torvalds 1518b8413f98STrond Myklebust res = nfs_commit_set_lock(NFS_I(inode), may_wait); 1519b8413f98STrond Myklebust if (res <= 0) 1520c5efa5fcSTrond Myklebust goto out_mark_dirty; 15213da28eb1STrond Myklebust res = nfs_scan_commit(inode, &head, 0, 0); 15221da177e4SLinus Torvalds if (res) { 1523a861a1e1SFred Isaman int error; 1524a861a1e1SFred Isaman 1525a861a1e1SFred Isaman error = pnfs_commit_list(inode, &head, how); 1526a861a1e1SFred Isaman if (error == PNFS_NOT_ATTEMPTED) 1527a861a1e1SFred Isaman error = nfs_commit_list(inode, &head, how); 15281da177e4SLinus Torvalds if (error < 0) 15291da177e4SLinus Torvalds return error; 1530b8413f98STrond Myklebust if (!may_wait) 1531b8413f98STrond Myklebust goto out_mark_dirty; 1532b8413f98STrond Myklebust error = wait_on_bit(&NFS_I(inode)->flags, 1533b8413f98STrond Myklebust NFS_INO_COMMIT, 153471d0a611STrond Myklebust nfs_wait_bit_killable, 153571d0a611STrond Myklebust TASK_KILLABLE); 1536b8413f98STrond Myklebust if (error < 0) 1537b8413f98STrond Myklebust return error; 153871d0a611STrond Myklebust } else 153971d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(inode)); 1540c5efa5fcSTrond Myklebust return res; 1541c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1542c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1543c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1544c5efa5fcSTrond Myklebust * that the data is on the disk. 1545c5efa5fcSTrond Myklebust */ 1546c5efa5fcSTrond Myklebust out_mark_dirty: 1547c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 15481da177e4SLinus Torvalds return res; 15491da177e4SLinus Torvalds } 15508fc795f7STrond Myklebust 15518fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 15528fc795f7STrond Myklebust { 1553420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1554420e3646STrond Myklebust int flags = FLUSH_SYNC; 1555420e3646STrond Myklebust int ret = 0; 15568fc795f7STrond Myklebust 1557a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1558a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1559a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1560420e3646STrond Myklebust */ 1561a00dd6c0SJeff Layton if (nfsi->ncommit <= (nfsi->npages >> 1)) 1562420e3646STrond Myklebust goto out_mark_dirty; 1563420e3646STrond Myklebust 1564a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1565420e3646STrond Myklebust flags = 0; 1566a00dd6c0SJeff Layton } 1567a00dd6c0SJeff Layton 1568420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1569420e3646STrond Myklebust if (ret >= 0) { 1570420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1571420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1572420e3646STrond Myklebust wbc->nr_to_write -= ret; 1573420e3646STrond Myklebust else 1574420e3646STrond Myklebust wbc->nr_to_write = 0; 1575420e3646STrond Myklebust } 15768fc795f7STrond Myklebust return 0; 1577420e3646STrond Myklebust } 1578420e3646STrond Myklebust out_mark_dirty: 15798fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 15808fc795f7STrond Myklebust return ret; 15818fc795f7STrond Myklebust } 1582c63c7b05STrond Myklebust #else 15838fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 15848fc795f7STrond Myklebust { 15858fc795f7STrond Myklebust return 0; 15868fc795f7STrond Myklebust } 15871da177e4SLinus Torvalds #endif 15881da177e4SLinus Torvalds 15898fc795f7STrond Myklebust int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 15908fc795f7STrond Myklebust { 1591863a3c6cSAndy Adamson int ret; 1592863a3c6cSAndy Adamson 1593863a3c6cSAndy Adamson ret = nfs_commit_unstable_pages(inode, wbc); 1594863a3c6cSAndy Adamson if (ret >= 0 && test_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags)) { 1595ef311537SAndy Adamson int status; 1596ef311537SAndy Adamson bool sync = true; 1597863a3c6cSAndy Adamson 1598846d5a09SWu Fengguang if (wbc->sync_mode == WB_SYNC_NONE) 1599ef311537SAndy Adamson sync = false; 1600863a3c6cSAndy Adamson 1601863a3c6cSAndy Adamson status = pnfs_layoutcommit_inode(inode, sync); 1602863a3c6cSAndy Adamson if (status < 0) 1603863a3c6cSAndy Adamson return status; 1604863a3c6cSAndy Adamson } 1605863a3c6cSAndy Adamson return ret; 16068fc795f7STrond Myklebust } 16078fc795f7STrond Myklebust 1608acdc53b2STrond Myklebust /* 1609acdc53b2STrond Myklebust * flush the inode to disk. 1610acdc53b2STrond Myklebust */ 1611acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 161234901f70STrond Myklebust { 161334901f70STrond Myklebust struct writeback_control wbc = { 161472cb77f4STrond Myklebust .sync_mode = WB_SYNC_ALL, 161534901f70STrond Myklebust .nr_to_write = LONG_MAX, 1616d7fb1207STrond Myklebust .range_start = 0, 1617d7fb1207STrond Myklebust .range_end = LLONG_MAX, 161834901f70STrond Myklebust }; 161934901f70STrond Myklebust 1620acdc53b2STrond Myklebust return sync_inode(inode, &wbc); 16211c75950bSTrond Myklebust } 16221c75950bSTrond Myklebust 16231b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 16241b3b4a1aSTrond Myklebust { 16251b3b4a1aSTrond Myklebust struct nfs_page *req; 16261b3b4a1aSTrond Myklebust int ret = 0; 16271b3b4a1aSTrond Myklebust 16281b3b4a1aSTrond Myklebust BUG_ON(!PageLocked(page)); 16291b3b4a1aSTrond Myklebust for (;;) { 1630ba8b06e6STrond Myklebust wait_on_page_writeback(page); 16311b3b4a1aSTrond Myklebust req = nfs_page_find_request(page); 16321b3b4a1aSTrond Myklebust if (req == NULL) 16331b3b4a1aSTrond Myklebust break; 16341b3b4a1aSTrond Myklebust if (nfs_lock_request_dontget(req)) { 16351b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 16361b3b4a1aSTrond Myklebust /* 16371b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 16381b3b4a1aSTrond Myklebust * page as being dirty 16391b3b4a1aSTrond Myklebust */ 16401b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 16411b3b4a1aSTrond Myklebust nfs_unlock_request(req); 16421b3b4a1aSTrond Myklebust break; 16431b3b4a1aSTrond Myklebust } 16441b3b4a1aSTrond Myklebust ret = nfs_wait_on_request(req); 1645c9edda71STrond Myklebust nfs_release_request(req); 16461b3b4a1aSTrond Myklebust if (ret < 0) 1647c988950eSTrond Myklebust break; 16481b3b4a1aSTrond Myklebust } 16491b3b4a1aSTrond Myklebust return ret; 16501b3b4a1aSTrond Myklebust } 16511b3b4a1aSTrond Myklebust 16521c75950bSTrond Myklebust /* 16531c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 16541c75950bSTrond Myklebust */ 16551c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 16561c75950bSTrond Myklebust { 16577f2f12d9STrond Myklebust loff_t range_start = page_offset(page); 16587f2f12d9STrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 16597f2f12d9STrond Myklebust struct writeback_control wbc = { 16607f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 16617f2f12d9STrond Myklebust .nr_to_write = 0, 16627f2f12d9STrond Myklebust .range_start = range_start, 16637f2f12d9STrond Myklebust .range_end = range_end, 16647f2f12d9STrond Myklebust }; 16657f2f12d9STrond Myklebust int ret; 16667f2f12d9STrond Myklebust 16670522f6adSTrond Myklebust for (;;) { 1668ba8b06e6STrond Myklebust wait_on_page_writeback(page); 16697f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 16707f2f12d9STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 16717f2f12d9STrond Myklebust if (ret < 0) 16727f2f12d9STrond Myklebust goto out_error; 16730522f6adSTrond Myklebust continue; 16747f2f12d9STrond Myklebust } 16750522f6adSTrond Myklebust if (!PagePrivate(page)) 16760522f6adSTrond Myklebust break; 16770522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 16787f2f12d9STrond Myklebust if (ret < 0) 16797f2f12d9STrond Myklebust goto out_error; 16807f2f12d9STrond Myklebust } 16817f2f12d9STrond Myklebust return 0; 16827f2f12d9STrond Myklebust out_error: 16837f2f12d9STrond Myklebust return ret; 16841c75950bSTrond Myklebust } 16851c75950bSTrond Myklebust 1686074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 1687074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 1688074cc1deSTrond Myklebust struct page *page) 1689074cc1deSTrond Myklebust { 1690074cc1deSTrond Myklebust struct nfs_page *req; 1691074cc1deSTrond Myklebust int ret; 1692074cc1deSTrond Myklebust 1693074cc1deSTrond Myklebust nfs_fscache_release_page(page, GFP_KERNEL); 1694074cc1deSTrond Myklebust 1695cfb506e1STrond Myklebust req = nfs_find_and_lock_request(page, false); 1696074cc1deSTrond Myklebust ret = PTR_ERR(req); 1697074cc1deSTrond Myklebust if (IS_ERR(req)) 1698074cc1deSTrond Myklebust goto out; 1699074cc1deSTrond Myklebust 1700074cc1deSTrond Myklebust ret = migrate_page(mapping, newpage, page); 1701074cc1deSTrond Myklebust if (!req) 1702074cc1deSTrond Myklebust goto out; 1703074cc1deSTrond Myklebust if (ret) 1704074cc1deSTrond Myklebust goto out_unlock; 1705074cc1deSTrond Myklebust page_cache_get(newpage); 1706190f38e5STrond Myklebust spin_lock(&mapping->host->i_lock); 1707074cc1deSTrond Myklebust req->wb_page = newpage; 1708074cc1deSTrond Myklebust SetPagePrivate(newpage); 1709190f38e5STrond Myklebust set_page_private(newpage, (unsigned long)req); 1710074cc1deSTrond Myklebust ClearPagePrivate(page); 1711074cc1deSTrond Myklebust set_page_private(page, 0); 1712190f38e5STrond Myklebust spin_unlock(&mapping->host->i_lock); 1713074cc1deSTrond Myklebust page_cache_release(page); 1714074cc1deSTrond Myklebust out_unlock: 1715074cc1deSTrond Myklebust nfs_clear_page_tag_locked(req); 1716074cc1deSTrond Myklebust out: 1717074cc1deSTrond Myklebust return ret; 1718074cc1deSTrond Myklebust } 1719074cc1deSTrond Myklebust #endif 1720074cc1deSTrond Myklebust 1721f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 17221da177e4SLinus Torvalds { 17231da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 17241da177e4SLinus Torvalds sizeof(struct nfs_write_data), 17251da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 172620c2df83SPaul Mundt NULL); 17271da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 17281da177e4SLinus Torvalds return -ENOMEM; 17291da177e4SLinus Torvalds 173093d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 17311da177e4SLinus Torvalds nfs_wdata_cachep); 17321da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 17331da177e4SLinus Torvalds return -ENOMEM; 17341da177e4SLinus Torvalds 173593d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 17361da177e4SLinus Torvalds nfs_wdata_cachep); 17371da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 17381da177e4SLinus Torvalds return -ENOMEM; 17391da177e4SLinus Torvalds 174089a09141SPeter Zijlstra /* 174189a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 174289a09141SPeter Zijlstra * 174389a09141SPeter Zijlstra * 64MB: 8192k 174489a09141SPeter Zijlstra * 128MB: 11585k 174589a09141SPeter Zijlstra * 256MB: 16384k 174689a09141SPeter Zijlstra * 512MB: 23170k 174789a09141SPeter Zijlstra * 1GB: 32768k 174889a09141SPeter Zijlstra * 2GB: 46340k 174989a09141SPeter Zijlstra * 4GB: 65536k 175089a09141SPeter Zijlstra * 8GB: 92681k 175189a09141SPeter Zijlstra * 16GB: 131072k 175289a09141SPeter Zijlstra * 175389a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 175489a09141SPeter Zijlstra * Limit the default to 256M 175589a09141SPeter Zijlstra */ 175689a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 175789a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 175889a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 175989a09141SPeter Zijlstra 17601da177e4SLinus Torvalds return 0; 17611da177e4SLinus Torvalds } 17621da177e4SLinus Torvalds 1763266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 17641da177e4SLinus Torvalds { 17651da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 17661da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 17671a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 17681da177e4SLinus Torvalds } 17691da177e4SLinus Torvalds 1770