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 1001ae88b2eSTrond Myklebust static 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 { 41288be9f99STrond Myklebust struct inode *inode = req->wb_context->path.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); 431b80c3cb6STrond Myklebust __mark_inode_dirty(req->wb_page->mapping->host, I_DIRTY_DATASYNC); 43261822ab5STrond Myklebust } 43361822ab5STrond Myklebust 4341da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 4351da177e4SLinus Torvalds /* 4361da177e4SLinus Torvalds * Add a request to the inode's commit list. 4371da177e4SLinus Torvalds */ 4381da177e4SLinus Torvalds static void 439a861a1e1SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg) 4401da177e4SLinus Torvalds { 44188be9f99STrond Myklebust struct inode *inode = req->wb_context->path.dentry->d_inode; 4421da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4431da177e4SLinus Torvalds 444587142f8STrond Myklebust spin_lock(&inode->i_lock); 445e468bae9STrond Myklebust set_bit(PG_CLEAN, &(req)->wb_flags); 4465c369683STrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, 4475c369683STrond Myklebust req->wb_index, 4485c369683STrond Myklebust NFS_PAGE_TAG_COMMIT); 449ff778d02STrond Myklebust nfsi->ncommit++; 450587142f8STrond Myklebust spin_unlock(&inode->i_lock); 451a861a1e1SFred Isaman pnfs_mark_request_commit(req, lseg); 452fd39fc85SChristoph Lameter inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 453c9e51e41SPeter Zijlstra inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE); 454a1803044STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 4551da177e4SLinus Torvalds } 4568e821cadSTrond Myklebust 457e468bae9STrond Myklebust static int 458e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 459e468bae9STrond Myklebust { 460e468bae9STrond Myklebust struct page *page = req->wb_page; 461e468bae9STrond Myklebust 462e468bae9STrond Myklebust if (test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) { 463e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 464e468bae9STrond Myklebust dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE); 465e468bae9STrond Myklebust return 1; 466e468bae9STrond Myklebust } 467e468bae9STrond Myklebust return 0; 468e468bae9STrond Myklebust } 469e468bae9STrond Myklebust 4708e821cadSTrond Myklebust static inline 4718e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4728e821cadSTrond Myklebust { 473465d5243SFred Isaman if (data->verf.committed == NFS_DATA_SYNC) 474465d5243SFred Isaman return data->lseg == NULL; 475465d5243SFred Isaman else 4768e821cadSTrond Myklebust return data->verf.committed != NFS_FILE_SYNC; 4778e821cadSTrond Myklebust } 4788e821cadSTrond Myklebust 4798e821cadSTrond Myklebust static inline 480a861a1e1SFred Isaman int nfs_reschedule_unstable_write(struct nfs_page *req, 481a861a1e1SFred Isaman struct nfs_write_data *data) 4828e821cadSTrond Myklebust { 483e468bae9STrond Myklebust if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) { 484a861a1e1SFred Isaman nfs_mark_request_commit(req, data->lseg); 4858e821cadSTrond Myklebust return 1; 4868e821cadSTrond Myklebust } 4878e821cadSTrond Myklebust if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) { 4886d884e8fSFred nfs_mark_request_dirty(req); 4898e821cadSTrond Myklebust return 1; 4908e821cadSTrond Myklebust } 4918e821cadSTrond Myklebust return 0; 4928e821cadSTrond Myklebust } 4938e821cadSTrond Myklebust #else 4948e821cadSTrond Myklebust static inline void 495a861a1e1SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg) 4968e821cadSTrond Myklebust { 4978e821cadSTrond Myklebust } 4988e821cadSTrond Myklebust 499e468bae9STrond Myklebust static inline int 500e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 501e468bae9STrond Myklebust { 502e468bae9STrond Myklebust return 0; 503e468bae9STrond Myklebust } 504e468bae9STrond Myklebust 5058e821cadSTrond Myklebust static inline 5068e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 5078e821cadSTrond Myklebust { 5088e821cadSTrond Myklebust return 0; 5098e821cadSTrond Myklebust } 5108e821cadSTrond Myklebust 5118e821cadSTrond Myklebust static inline 512a861a1e1SFred Isaman int nfs_reschedule_unstable_write(struct nfs_page *req, 513a861a1e1SFred Isaman struct nfs_write_data *data) 5148e821cadSTrond Myklebust { 5158e821cadSTrond Myklebust return 0; 5168e821cadSTrond Myklebust } 5171da177e4SLinus Torvalds #endif 5181da177e4SLinus Torvalds 51947c62564STrond Myklebust #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 520fb8a1f11STrond Myklebust static int 521fb8a1f11STrond Myklebust nfs_need_commit(struct nfs_inode *nfsi) 522fb8a1f11STrond Myklebust { 523fb8a1f11STrond Myklebust return radix_tree_tagged(&nfsi->nfs_page_tree, NFS_PAGE_TAG_COMMIT); 524fb8a1f11STrond Myklebust } 525fb8a1f11STrond Myklebust 5261da177e4SLinus Torvalds /* 5271da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 5281da177e4SLinus Torvalds * @inode: NFS inode to scan 5291da177e4SLinus Torvalds * @dst: destination list 5301da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 5311da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 5321da177e4SLinus Torvalds * 5331da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 5341da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 5351da177e4SLinus Torvalds */ 5361da177e4SLinus Torvalds static int 537ca52fec1STrond Myklebust nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 5381da177e4SLinus Torvalds { 5391da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 540ff778d02STrond Myklebust int ret; 5413da28eb1STrond Myklebust 542fb8a1f11STrond Myklebust if (!nfs_need_commit(nfsi)) 543fb8a1f11STrond Myklebust return 0; 544fb8a1f11STrond Myklebust 5450d88f6e8SDave Chinner spin_lock(&inode->i_lock); 546ff778d02STrond Myklebust ret = nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT); 547ff778d02STrond Myklebust if (ret > 0) 548ff778d02STrond Myklebust nfsi->ncommit -= ret; 5490d88f6e8SDave Chinner spin_unlock(&inode->i_lock); 5500d88f6e8SDave Chinner 5512928db1fSTrond Myklebust if (nfs_need_commit(NFS_I(inode))) 5522928db1fSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 5530d88f6e8SDave Chinner 554ff778d02STrond Myklebust return ret; 5551da177e4SLinus Torvalds } 556c42de9ddSTrond Myklebust #else 557fb8a1f11STrond Myklebust static inline int nfs_need_commit(struct nfs_inode *nfsi) 558fb8a1f11STrond Myklebust { 559fb8a1f11STrond Myklebust return 0; 560fb8a1f11STrond Myklebust } 561fb8a1f11STrond Myklebust 562ca52fec1STrond Myklebust static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages) 563c42de9ddSTrond Myklebust { 564c42de9ddSTrond Myklebust return 0; 565c42de9ddSTrond Myklebust } 5661da177e4SLinus Torvalds #endif 5671da177e4SLinus Torvalds 5681da177e4SLinus Torvalds /* 569e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 570e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 5711da177e4SLinus Torvalds * 572e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 573e7d39069STrond Myklebust * to disk. 5741da177e4SLinus Torvalds */ 575e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 576e7d39069STrond Myklebust struct page *page, 577e7d39069STrond Myklebust unsigned int offset, 578e7d39069STrond Myklebust unsigned int bytes) 5791da177e4SLinus Torvalds { 580e7d39069STrond Myklebust struct nfs_page *req; 581e7d39069STrond Myklebust unsigned int rqend; 582e7d39069STrond Myklebust unsigned int end; 5831da177e4SLinus Torvalds int error; 584277459d2STrond Myklebust 585e7d39069STrond Myklebust if (!PagePrivate(page)) 586e7d39069STrond Myklebust return NULL; 587e7d39069STrond Myklebust 588e7d39069STrond Myklebust end = offset + bytes; 589e7d39069STrond Myklebust spin_lock(&inode->i_lock); 590e7d39069STrond Myklebust 591e7d39069STrond Myklebust for (;;) { 592e7d39069STrond Myklebust req = nfs_page_find_request_locked(page); 593e7d39069STrond Myklebust if (req == NULL) 594e7d39069STrond Myklebust goto out_unlock; 595e7d39069STrond Myklebust 596e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 597e7d39069STrond Myklebust /* 598e7d39069STrond Myklebust * Tell the caller to flush out the request if 599e7d39069STrond Myklebust * the offsets are non-contiguous. 600e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 601e7d39069STrond Myklebust * have flushed out requests having wrong owners. 602e7d39069STrond Myklebust */ 603e468bae9STrond Myklebust if (offset > rqend 604e7d39069STrond Myklebust || end < req->wb_offset) 605e7d39069STrond Myklebust goto out_flushme; 606e7d39069STrond Myklebust 607e7d39069STrond Myklebust if (nfs_set_page_tag_locked(req)) 608e7d39069STrond Myklebust break; 609e7d39069STrond Myklebust 610e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 611587142f8STrond Myklebust spin_unlock(&inode->i_lock); 6121da177e4SLinus Torvalds error = nfs_wait_on_request(req); 6131da177e4SLinus Torvalds nfs_release_request(req); 614e7d39069STrond Myklebust if (error != 0) 615e7d39069STrond Myklebust goto out_err; 616e7d39069STrond Myklebust spin_lock(&inode->i_lock); 6171da177e4SLinus Torvalds } 6181da177e4SLinus Torvalds 619ff778d02STrond Myklebust if (nfs_clear_request_commit(req) && 620e468bae9STrond Myklebust radix_tree_tag_clear(&NFS_I(inode)->nfs_page_tree, 621a861a1e1SFred Isaman req->wb_index, NFS_PAGE_TAG_COMMIT) != NULL) { 622ff778d02STrond Myklebust NFS_I(inode)->ncommit--; 623a861a1e1SFred Isaman pnfs_clear_request_commit(req); 624a861a1e1SFred Isaman } 625e468bae9STrond Myklebust 6261da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 6271da177e4SLinus Torvalds if (offset < req->wb_offset) { 6281da177e4SLinus Torvalds req->wb_offset = offset; 6291da177e4SLinus Torvalds req->wb_pgbase = offset; 6301da177e4SLinus Torvalds } 6311da177e4SLinus Torvalds if (end > rqend) 6321da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 633e7d39069STrond Myklebust else 634e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 635e7d39069STrond Myklebust out_unlock: 636e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 637e7d39069STrond Myklebust return req; 638e7d39069STrond Myklebust out_flushme: 639e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 640e7d39069STrond Myklebust nfs_release_request(req); 641e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 642e7d39069STrond Myklebust out_err: 643e7d39069STrond Myklebust return ERR_PTR(error); 644e7d39069STrond Myklebust } 6451da177e4SLinus Torvalds 646e7d39069STrond Myklebust /* 647e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 648e7d39069STrond Myklebust * 649e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 650e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 651e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 652e7d39069STrond Myklebust */ 653e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 654e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 655e7d39069STrond Myklebust { 656e7d39069STrond Myklebust struct inode *inode = page->mapping->host; 657e7d39069STrond Myklebust struct nfs_page *req; 658e7d39069STrond Myklebust int error; 659e7d39069STrond Myklebust 660e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 661e7d39069STrond Myklebust if (req != NULL) 662e7d39069STrond Myklebust goto out; 663e7d39069STrond Myklebust req = nfs_create_request(ctx, inode, page, offset, bytes); 664e7d39069STrond Myklebust if (IS_ERR(req)) 665e7d39069STrond Myklebust goto out; 666e7d39069STrond Myklebust error = nfs_inode_add_request(inode, req); 667e7d39069STrond Myklebust if (error != 0) { 668e7d39069STrond Myklebust nfs_release_request(req); 669e7d39069STrond Myklebust req = ERR_PTR(error); 670e7d39069STrond Myklebust } 671efc91ed0STrond Myklebust out: 67261e930a9STrond Myklebust return req; 6731da177e4SLinus Torvalds } 6741da177e4SLinus Torvalds 675e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 676e7d39069STrond Myklebust unsigned int offset, unsigned int count) 677e7d39069STrond Myklebust { 678e7d39069STrond Myklebust struct nfs_page *req; 679e7d39069STrond Myklebust 680e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 681e7d39069STrond Myklebust if (IS_ERR(req)) 682e7d39069STrond Myklebust return PTR_ERR(req); 683e7d39069STrond Myklebust /* Update file length */ 684e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 685e7d39069STrond Myklebust nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 686a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 687e7d39069STrond Myklebust nfs_clear_page_tag_locked(req); 688e7d39069STrond Myklebust return 0; 689e7d39069STrond Myklebust } 690e7d39069STrond Myklebust 6911da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 6921da177e4SLinus Torvalds { 693cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 6941da177e4SLinus Torvalds struct nfs_page *req; 6951a54533eSTrond Myklebust int do_flush, status; 6961da177e4SLinus Torvalds /* 6971da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 6981da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 6991da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 7001da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 7011da177e4SLinus Torvalds * Also do the same if we find a request from an existing 7021da177e4SLinus Torvalds * dropped page. 7031da177e4SLinus Torvalds */ 7041a54533eSTrond Myklebust do { 705277459d2STrond Myklebust req = nfs_page_find_request(page); 7061a54533eSTrond Myklebust if (req == NULL) 7071a54533eSTrond Myklebust return 0; 708f11ac8dbSTrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx || 709f11ac8dbSTrond Myklebust req->wb_lock_context->lockowner != current->files || 710f11ac8dbSTrond Myklebust req->wb_lock_context->pid != current->tgid; 7111da177e4SLinus Torvalds nfs_release_request(req); 7121a54533eSTrond Myklebust if (!do_flush) 7131a54533eSTrond Myklebust return 0; 714277459d2STrond Myklebust status = nfs_wb_page(page->mapping->host, page); 7151a54533eSTrond Myklebust } while (status == 0); 7161a54533eSTrond Myklebust return status; 7171da177e4SLinus Torvalds } 7181da177e4SLinus Torvalds 7191da177e4SLinus Torvalds /* 7205d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 7215d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 7225d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 7235d47a356STrond Myklebust */ 7245d47a356STrond Myklebust static int nfs_write_pageuptodate(struct page *page, struct inode *inode) 7255d47a356STrond Myklebust { 7265d47a356STrond Myklebust return PageUptodate(page) && 7275d47a356STrond Myklebust !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA)); 7285d47a356STrond Myklebust } 7295d47a356STrond Myklebust 7305d47a356STrond Myklebust /* 7311da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 7321da177e4SLinus Torvalds * 7331da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 7341da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 7351da177e4SLinus Torvalds */ 7361da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 7371da177e4SLinus Torvalds unsigned int offset, unsigned int count) 7381da177e4SLinus Torvalds { 739cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 7401da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 7411da177e4SLinus Torvalds int status = 0; 7421da177e4SLinus Torvalds 74391d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 74491d5b470SChuck Lever 74548186c7dSChuck Lever dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", 74601cce933SJosef "Jeff" Sipek file->f_path.dentry->d_parent->d_name.name, 74701cce933SJosef "Jeff" Sipek file->f_path.dentry->d_name.name, count, 7480bbacc40SChuck Lever (long long)(page_offset(page) + offset)); 7491da177e4SLinus Torvalds 7501da177e4SLinus Torvalds /* If we're not using byte range locks, and we know the page 7515d47a356STrond Myklebust * is up to date, it may be more efficient to extend the write 7525d47a356STrond Myklebust * to cover the entire page in order to avoid fragmentation 7535d47a356STrond Myklebust * inefficiencies. 7541da177e4SLinus Torvalds */ 7555d47a356STrond Myklebust if (nfs_write_pageuptodate(page, inode) && 7565d47a356STrond Myklebust inode->i_flock == NULL && 7576b2f3d1fSChristoph Hellwig !(file->f_flags & O_DSYNC)) { 75849a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 7591da177e4SLinus Torvalds offset = 0; 7601da177e4SLinus Torvalds } 7611da177e4SLinus Torvalds 762e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 76303fa9e84STrond Myklebust if (status < 0) 76403fa9e84STrond Myklebust nfs_set_pageerror(page); 7651da177e4SLinus Torvalds 76648186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 7671da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 7681da177e4SLinus Torvalds return status; 7691da177e4SLinus Torvalds } 7701da177e4SLinus Torvalds 771a861a1e1SFred Isaman static void nfs_writepage_release(struct nfs_page *req, 772a861a1e1SFred Isaman struct nfs_write_data *data) 7731da177e4SLinus Torvalds { 774a6305ddbSTrond Myklebust struct page *page = req->wb_page; 7758e821cadSTrond Myklebust 776a861a1e1SFred Isaman if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req, data)) 7771da177e4SLinus Torvalds nfs_inode_remove_request(req); 7789fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 779a6305ddbSTrond Myklebust nfs_end_page_writeback(page); 7801da177e4SLinus Torvalds } 7811da177e4SLinus Torvalds 7823ff7576dSTrond Myklebust static int flush_task_priority(int how) 7831da177e4SLinus Torvalds { 7841da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 7851da177e4SLinus Torvalds case FLUSH_HIGHPRI: 7861da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 7871da177e4SLinus Torvalds case FLUSH_LOWPRI: 7881da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 7891da177e4SLinus Torvalds } 7901da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 7911da177e4SLinus Torvalds } 7921da177e4SLinus Torvalds 793a69aef14SFred Isaman int nfs_initiate_write(struct nfs_write_data *data, 794d138d5d1SAndy Adamson struct rpc_clnt *clnt, 795788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 7961da177e4SLinus Torvalds int how) 7971da177e4SLinus Torvalds { 798d138d5d1SAndy Adamson struct inode *inode = data->inode; 7993ff7576dSTrond Myklebust int priority = flush_task_priority(how); 80007737691STrond Myklebust struct rpc_task *task; 801bdc7f021STrond Myklebust struct rpc_message msg = { 802bdc7f021STrond Myklebust .rpc_argp = &data->args, 803bdc7f021STrond Myklebust .rpc_resp = &data->res, 804d138d5d1SAndy Adamson .rpc_cred = data->cred, 805bdc7f021STrond Myklebust }; 80684115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 807d138d5d1SAndy Adamson .rpc_client = clnt, 80807737691STrond Myklebust .task = &data->task, 809bdc7f021STrond Myklebust .rpc_message = &msg, 81084115e1cSTrond Myklebust .callback_ops = call_ops, 81184115e1cSTrond Myklebust .callback_data = data, 812101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 8132c61be0aSTrond Myklebust .flags = RPC_TASK_ASYNC, 8143ff7576dSTrond Myklebust .priority = priority, 81584115e1cSTrond Myklebust }; 8162c61be0aSTrond Myklebust int ret = 0; 8171da177e4SLinus Torvalds 818d138d5d1SAndy Adamson /* Set up the initial task struct. */ 819d138d5d1SAndy Adamson NFS_PROTO(inode)->write_setup(data, &msg); 820d138d5d1SAndy Adamson 821d138d5d1SAndy Adamson dprintk("NFS: %5u initiated write call " 822d138d5d1SAndy Adamson "(req %s/%lld, %u bytes @ offset %llu)\n", 823d138d5d1SAndy Adamson data->task.tk_pid, 824d138d5d1SAndy Adamson inode->i_sb->s_id, 825d138d5d1SAndy Adamson (long long)NFS_FILEID(inode), 826d138d5d1SAndy Adamson data->args.count, 827d138d5d1SAndy Adamson (unsigned long long)data->args.offset); 828d138d5d1SAndy Adamson 829d138d5d1SAndy Adamson task = rpc_run_task(&task_setup_data); 830d138d5d1SAndy Adamson if (IS_ERR(task)) { 831d138d5d1SAndy Adamson ret = PTR_ERR(task); 832d138d5d1SAndy Adamson goto out; 833d138d5d1SAndy Adamson } 834d138d5d1SAndy Adamson if (how & FLUSH_SYNC) { 835d138d5d1SAndy Adamson ret = rpc_wait_for_completion_task(task); 836d138d5d1SAndy Adamson if (ret == 0) 837d138d5d1SAndy Adamson ret = task->tk_status; 838d138d5d1SAndy Adamson } 839d138d5d1SAndy Adamson rpc_put_task(task); 840d138d5d1SAndy Adamson out: 841d138d5d1SAndy Adamson return ret; 842d138d5d1SAndy Adamson } 843a69aef14SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_write); 844d138d5d1SAndy Adamson 845d138d5d1SAndy Adamson /* 846d138d5d1SAndy Adamson * Set up the argument/result storage required for the RPC call. 847d138d5d1SAndy Adamson */ 848d138d5d1SAndy Adamson static int nfs_write_rpcsetup(struct nfs_page *req, 849d138d5d1SAndy Adamson struct nfs_write_data *data, 850d138d5d1SAndy Adamson const struct rpc_call_ops *call_ops, 851d138d5d1SAndy Adamson unsigned int count, unsigned int offset, 8525053aa56SFred Isaman struct pnfs_layout_segment *lseg, 853d138d5d1SAndy Adamson int how) 854d138d5d1SAndy Adamson { 855d138d5d1SAndy Adamson struct inode *inode = req->wb_context->path.dentry->d_inode; 856d138d5d1SAndy Adamson 8571da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 8581da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 8591da177e4SLinus Torvalds 8601da177e4SLinus Torvalds data->req = req; 86188be9f99STrond Myklebust data->inode = inode = req->wb_context->path.dentry->d_inode; 862d138d5d1SAndy Adamson data->cred = req->wb_context->cred; 8635053aa56SFred Isaman data->lseg = get_lseg(lseg); 8641da177e4SLinus Torvalds 8651da177e4SLinus Torvalds data->args.fh = NFS_FH(inode); 8661da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 867*2bea038cSBoaz Harrosh /* pnfs_set_layoutcommit needs this */ 868*2bea038cSBoaz Harrosh data->mds_offset = data->args.offset; 8691da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 8701da177e4SLinus Torvalds data->args.pages = data->pagevec; 8711da177e4SLinus Torvalds data->args.count = count; 872383ba719STrond Myklebust data->args.context = get_nfs_open_context(req->wb_context); 873f11ac8dbSTrond Myklebust data->args.lock_context = req->wb_lock_context; 874bdc7f021STrond Myklebust data->args.stable = NFS_UNSTABLE; 875b31268acSTrond Myklebust if (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { 876bdc7f021STrond Myklebust data->args.stable = NFS_DATA_SYNC; 877fb8a1f11STrond Myklebust if (!nfs_need_commit(NFS_I(inode))) 878bdc7f021STrond Myklebust data->args.stable = NFS_FILE_SYNC; 879bdc7f021STrond Myklebust } 8801da177e4SLinus Torvalds 8811da177e4SLinus Torvalds data->res.fattr = &data->fattr; 8821da177e4SLinus Torvalds data->res.count = count; 8831da177e4SLinus Torvalds data->res.verf = &data->verf; 8840e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 8851da177e4SLinus Torvalds 8860382b744SAndy Adamson if (data->lseg && 8870382b744SAndy Adamson (pnfs_try_to_write_data(data, call_ops, how) == PNFS_ATTEMPTED)) 8880382b744SAndy Adamson return 0; 8890382b744SAndy Adamson 890d138d5d1SAndy Adamson return nfs_initiate_write(data, NFS_CLIENT(inode), call_ops, how); 8911da177e4SLinus Torvalds } 8921da177e4SLinus Torvalds 8936d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 8946d884e8fSFred * call this on each, which will prepare them to be retried on next 8956d884e8fSFred * writeback using standard nfs. 8966d884e8fSFred */ 8976d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 8986d884e8fSFred { 899a6305ddbSTrond Myklebust struct page *page = req->wb_page; 900a6305ddbSTrond Myklebust 9016d884e8fSFred nfs_mark_request_dirty(req); 9026d884e8fSFred nfs_clear_page_tag_locked(req); 903a6305ddbSTrond Myklebust nfs_end_page_writeback(page); 9046d884e8fSFred } 9056d884e8fSFred 9061da177e4SLinus Torvalds /* 9071da177e4SLinus Torvalds * Generate multiple small requests to write out a single 9081da177e4SLinus Torvalds * contiguous dirty area on one page. 9091da177e4SLinus Torvalds */ 910c76069bdSFred Isaman static int nfs_flush_multi(struct nfs_pageio_descriptor *desc) 9111da177e4SLinus Torvalds { 912c76069bdSFred Isaman struct nfs_page *req = nfs_list_entry(desc->pg_list.next); 9131da177e4SLinus Torvalds struct page *page = req->wb_page; 9141da177e4SLinus Torvalds struct nfs_write_data *data; 915c76069bdSFred Isaman size_t wsize = NFS_SERVER(desc->pg_inode)->wsize, nbytes; 916e9f7bee1STrond Myklebust unsigned int offset; 9171da177e4SLinus Torvalds int requests = 0; 918dbae4c73STrond Myklebust int ret = 0; 919c76069bdSFred Isaman struct pnfs_layout_segment *lseg; 9201da177e4SLinus Torvalds LIST_HEAD(list); 9211da177e4SLinus Torvalds 9221da177e4SLinus Torvalds nfs_list_remove_request(req); 9231da177e4SLinus Torvalds 924b31268acSTrond Myklebust if ((desc->pg_ioflags & FLUSH_COND_STABLE) && 925b31268acSTrond Myklebust (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit || 926b31268acSTrond Myklebust desc->pg_count > wsize)) 927b31268acSTrond Myklebust desc->pg_ioflags &= ~FLUSH_COND_STABLE; 928b31268acSTrond Myklebust 929b31268acSTrond Myklebust 930c76069bdSFred Isaman nbytes = desc->pg_count; 931e9f7bee1STrond Myklebust do { 932e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 933e9f7bee1STrond Myklebust 9348d5658c9STrond Myklebust data = nfs_writedata_alloc(1); 9351da177e4SLinus Torvalds if (!data) 9361da177e4SLinus Torvalds goto out_bad; 9371da177e4SLinus Torvalds list_add(&data->pages, &list); 9381da177e4SLinus Torvalds requests++; 939e9f7bee1STrond Myklebust nbytes -= len; 940e9f7bee1STrond Myklebust } while (nbytes != 0); 9411da177e4SLinus Torvalds atomic_set(&req->wb_complete, requests); 9421da177e4SLinus Torvalds 943c76069bdSFred Isaman BUG_ON(desc->pg_lseg); 944fb3296ebSBenny Halevy lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, 945fb3296ebSBenny Halevy req_offset(req), desc->pg_count, 946fb3296ebSBenny Halevy IOMODE_RW, GFP_NOFS); 9471da177e4SLinus Torvalds ClearPageError(page); 9481da177e4SLinus Torvalds offset = 0; 949c76069bdSFred Isaman nbytes = desc->pg_count; 9501da177e4SLinus Torvalds do { 951dbae4c73STrond Myklebust int ret2; 952dbae4c73STrond Myklebust 9531da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 9541da177e4SLinus Torvalds list_del_init(&data->pages); 9551da177e4SLinus Torvalds 9561da177e4SLinus Torvalds data->pagevec[0] = page; 9571da177e4SLinus Torvalds 958bcb71bbaSTrond Myklebust if (nbytes < wsize) 959bcb71bbaSTrond Myklebust wsize = nbytes; 960dbae4c73STrond Myklebust ret2 = nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 961c76069bdSFred Isaman wsize, offset, lseg, desc->pg_ioflags); 962dbae4c73STrond Myklebust if (ret == 0) 963dbae4c73STrond Myklebust ret = ret2; 9641da177e4SLinus Torvalds offset += wsize; 9651da177e4SLinus Torvalds nbytes -= wsize; 9661da177e4SLinus Torvalds } while (nbytes != 0); 9671da177e4SLinus Torvalds 96844b83799SFred Isaman put_lseg(lseg); 96936fe432dSFred Isaman desc->pg_lseg = NULL; 970dbae4c73STrond Myklebust return ret; 9711da177e4SLinus Torvalds 9721da177e4SLinus Torvalds out_bad: 9731da177e4SLinus Torvalds while (!list_empty(&list)) { 9741da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 9751da177e4SLinus Torvalds list_del(&data->pages); 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 */ 990c76069bdSFred Isaman static int nfs_flush_one(struct nfs_pageio_descriptor *desc) 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; 996c76069bdSFred Isaman struct pnfs_layout_segment *lseg = desc->pg_lseg; 99744b83799SFred Isaman int ret; 9981da177e4SLinus Torvalds 999c76069bdSFred Isaman data = nfs_writedata_alloc(nfs_page_array_len(desc->pg_base, 1000c76069bdSFred Isaman desc->pg_count)); 100144b83799SFred Isaman if (!data) { 100244b83799SFred Isaman while (!list_empty(head)) { 100344b83799SFred Isaman req = nfs_list_entry(head->next); 100444b83799SFred Isaman nfs_list_remove_request(req); 100544b83799SFred Isaman nfs_redirty_request(req); 100644b83799SFred Isaman } 100744b83799SFred Isaman ret = -ENOMEM; 100844b83799SFred Isaman goto out; 100944b83799SFred Isaman } 10101da177e4SLinus Torvalds pages = data->pagevec; 10111da177e4SLinus Torvalds while (!list_empty(head)) { 10121da177e4SLinus Torvalds req = nfs_list_entry(head->next); 10131da177e4SLinus Torvalds nfs_list_remove_request(req); 10141da177e4SLinus Torvalds nfs_list_add_request(req, &data->pages); 10151da177e4SLinus Torvalds ClearPageError(req->wb_page); 10161da177e4SLinus Torvalds *pages++ = req->wb_page; 10171da177e4SLinus Torvalds } 10181da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 101944b83799SFred Isaman if ((!lseg) && list_is_singular(&data->pages)) 1020fb3296ebSBenny Halevy lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, 1021fb3296ebSBenny Halevy req_offset(req), desc->pg_count, 1022fb3296ebSBenny Halevy IOMODE_RW, GFP_NOFS); 10231da177e4SLinus Torvalds 1024b31268acSTrond Myklebust if ((desc->pg_ioflags & FLUSH_COND_STABLE) && 1025b31268acSTrond Myklebust (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit)) 1026b31268acSTrond Myklebust desc->pg_ioflags &= ~FLUSH_COND_STABLE; 1027b31268acSTrond Myklebust 10281da177e4SLinus Torvalds /* Set up the argument struct */ 1029c76069bdSFred Isaman ret = nfs_write_rpcsetup(req, data, &nfs_write_full_ops, desc->pg_count, 0, lseg, desc->pg_ioflags); 103044b83799SFred Isaman out: 103144b83799SFred Isaman put_lseg(lseg); /* Cleans any gotten in ->pg_test */ 103236fe432dSFred Isaman desc->pg_lseg = NULL; 103344b83799SFred Isaman return ret; 10341da177e4SLinus Torvalds } 10351da177e4SLinus Torvalds 1036c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1037c63c7b05STrond Myklebust struct inode *inode, int ioflags) 10381da177e4SLinus Torvalds { 1039bf4285e7SChuck Lever size_t wsize = NFS_SERVER(inode)->wsize; 10401da177e4SLinus Torvalds 1041bcb71bbaSTrond Myklebust if (wsize < PAGE_CACHE_SIZE) 1042c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags); 1043bcb71bbaSTrond Myklebust else 1044c63c7b05STrond Myklebust nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags); 10451da177e4SLinus Torvalds } 10461da177e4SLinus Torvalds 10471da177e4SLinus Torvalds /* 10481da177e4SLinus Torvalds * Handle a write reply that flushed part of a page. 10491da177e4SLinus Torvalds */ 1050788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) 10511da177e4SLinus Torvalds { 1052788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 10531da177e4SLinus Torvalds 105448186c7dSChuck Lever dprintk("NFS: %5u write(%s/%lld %d@%lld)", 105548186c7dSChuck Lever task->tk_pid, 105648186c7dSChuck Lever data->req->wb_context->path.dentry->d_inode->i_sb->s_id, 105748186c7dSChuck Lever (long long) 105848186c7dSChuck Lever NFS_FILEID(data->req->wb_context->path.dentry->d_inode), 105948186c7dSChuck Lever data->req->wb_bytes, (long long)req_offset(data->req)); 10601da177e4SLinus Torvalds 1061c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 1062c9d8f89dSTrond Myklebust } 1063788e7a89STrond Myklebust 1064c9d8f89dSTrond Myklebust static void nfs_writeback_release_partial(void *calldata) 1065c9d8f89dSTrond Myklebust { 1066c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1067c9d8f89dSTrond Myklebust struct nfs_page *req = data->req; 1068c9d8f89dSTrond Myklebust struct page *page = req->wb_page; 1069c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1070c9d8f89dSTrond Myklebust 1071c9d8f89dSTrond Myklebust if (status < 0) { 1072a301b777STrond Myklebust nfs_set_pageerror(page); 1073c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 1074c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 10758e821cadSTrond Myklebust goto out; 10768e821cadSTrond Myklebust } 10778e821cadSTrond Myklebust 10788e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 1079587142f8STrond Myklebust struct inode *inode = page->mapping->host; 10808e821cadSTrond Myklebust 1081587142f8STrond Myklebust spin_lock(&inode->i_lock); 10828e821cadSTrond Myklebust if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { 10838e821cadSTrond Myklebust /* Do nothing we need to resend the writes */ 10848e821cadSTrond Myklebust } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { 10851da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 10861da177e4SLinus Torvalds dprintk(" defer commit\n"); 10871da177e4SLinus Torvalds } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { 10888e821cadSTrond Myklebust set_bit(PG_NEED_RESCHED, &req->wb_flags); 10898e821cadSTrond Myklebust clear_bit(PG_NEED_COMMIT, &req->wb_flags); 10901da177e4SLinus Torvalds dprintk(" server reboot detected\n"); 10911da177e4SLinus Torvalds } 1092587142f8STrond Myklebust spin_unlock(&inode->i_lock); 10931da177e4SLinus Torvalds } else 10941da177e4SLinus Torvalds dprintk(" OK\n"); 10951da177e4SLinus Torvalds 10968e821cadSTrond Myklebust out: 10971da177e4SLinus Torvalds if (atomic_dec_and_test(&req->wb_complete)) 1098a861a1e1SFred Isaman nfs_writepage_release(req, data); 1099c9d8f89dSTrond Myklebust nfs_writedata_release(calldata); 11001da177e4SLinus Torvalds } 11011da177e4SLinus Torvalds 1102def6ed7eSAndy Adamson #if defined(CONFIG_NFS_V4_1) 1103def6ed7eSAndy Adamson void nfs_write_prepare(struct rpc_task *task, void *calldata) 1104def6ed7eSAndy Adamson { 1105def6ed7eSAndy Adamson struct nfs_write_data *data = calldata; 1106def6ed7eSAndy Adamson 1107035168abSTrond Myklebust if (nfs4_setup_sequence(NFS_SERVER(data->inode), 1108035168abSTrond Myklebust &data->args.seq_args, 1109def6ed7eSAndy Adamson &data->res.seq_res, 1, task)) 1110def6ed7eSAndy Adamson return; 1111def6ed7eSAndy Adamson rpc_call_start(task); 1112def6ed7eSAndy Adamson } 1113def6ed7eSAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1114def6ed7eSAndy Adamson 1115788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops = { 1116def6ed7eSAndy Adamson #if defined(CONFIG_NFS_V4_1) 1117def6ed7eSAndy Adamson .rpc_call_prepare = nfs_write_prepare, 1118def6ed7eSAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1119788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_partial, 1120c9d8f89dSTrond Myklebust .rpc_release = nfs_writeback_release_partial, 1121788e7a89STrond Myklebust }; 1122788e7a89STrond Myklebust 11231da177e4SLinus Torvalds /* 11241da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 11251da177e4SLinus Torvalds * 11261da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 11271da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 11281da177e4SLinus Torvalds * as the page has a write request pending. 11291da177e4SLinus Torvalds */ 1130788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) 11311da177e4SLinus Torvalds { 1132788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 11331da177e4SLinus Torvalds 1134c9d8f89dSTrond Myklebust nfs_writeback_done(task, data); 1135c9d8f89dSTrond Myklebust } 1136c9d8f89dSTrond Myklebust 1137c9d8f89dSTrond Myklebust static void nfs_writeback_release_full(void *calldata) 1138c9d8f89dSTrond Myklebust { 1139c9d8f89dSTrond Myklebust struct nfs_write_data *data = calldata; 1140c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1141788e7a89STrond Myklebust 11421da177e4SLinus Torvalds /* Update attributes as result of writeback. */ 11431da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 1144c9d8f89dSTrond Myklebust struct nfs_page *req = nfs_list_entry(data->pages.next); 1145c9d8f89dSTrond Myklebust struct page *page = req->wb_page; 1146c9d8f89dSTrond Myklebust 11471da177e4SLinus Torvalds nfs_list_remove_request(req); 11481da177e4SLinus Torvalds 114948186c7dSChuck Lever dprintk("NFS: %5u write (%s/%lld %d@%lld)", 115048186c7dSChuck Lever data->task.tk_pid, 115188be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 115288be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 11531da177e4SLinus Torvalds req->wb_bytes, 11541da177e4SLinus Torvalds (long long)req_offset(req)); 11551da177e4SLinus Torvalds 1156c9d8f89dSTrond Myklebust if (status < 0) { 1157a301b777STrond Myklebust nfs_set_pageerror(page); 1158c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 1159c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 11608e821cadSTrond Myklebust goto remove_request; 11611da177e4SLinus Torvalds } 11621da177e4SLinus Torvalds 11638e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 11641da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 1165a861a1e1SFred Isaman nfs_mark_request_commit(req, data->lseg); 11661da177e4SLinus Torvalds dprintk(" marked for commit\n"); 11678e821cadSTrond Myklebust goto next; 11688e821cadSTrond Myklebust } 11698e821cadSTrond Myklebust dprintk(" OK\n"); 11708e821cadSTrond Myklebust remove_request: 11711da177e4SLinus Torvalds nfs_inode_remove_request(req); 11721da177e4SLinus Torvalds next: 11739fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 1174a6305ddbSTrond Myklebust nfs_end_page_writeback(page); 11751da177e4SLinus Torvalds } 1176c9d8f89dSTrond Myklebust nfs_writedata_release(calldata); 11771da177e4SLinus Torvalds } 11781da177e4SLinus Torvalds 1179788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = { 1180def6ed7eSAndy Adamson #if defined(CONFIG_NFS_V4_1) 1181def6ed7eSAndy Adamson .rpc_call_prepare = nfs_write_prepare, 1182def6ed7eSAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1183788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_full, 1184c9d8f89dSTrond Myklebust .rpc_release = nfs_writeback_release_full, 1185788e7a89STrond Myklebust }; 1186788e7a89STrond Myklebust 1187788e7a89STrond Myklebust 11881da177e4SLinus Torvalds /* 11891da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 11901da177e4SLinus Torvalds */ 119113602896SFred Isaman void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 11921da177e4SLinus Torvalds { 11931da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 11941da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1195eedc020eSAndy Adamson struct nfs_server *server = NFS_SERVER(data->inode); 1196788e7a89STrond Myklebust int status; 11971da177e4SLinus Torvalds 1198a3f565b1SChuck Lever dprintk("NFS: %5u nfs_writeback_done (status %d)\n", 11991da177e4SLinus Torvalds task->tk_pid, task->tk_status); 12001da177e4SLinus Torvalds 1201f551e44fSChuck Lever /* 1202f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1203f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1204f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1205f551e44fSChuck Lever * another writer had changed the file, but some applications 1206f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1207f551e44fSChuck Lever */ 1208788e7a89STrond Myklebust status = NFS_PROTO(data->inode)->write_done(task, data); 1209788e7a89STrond Myklebust if (status != 0) 121013602896SFred Isaman return; 121191d5b470SChuck Lever nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 121291d5b470SChuck Lever 12131da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 12141da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 12151da177e4SLinus Torvalds /* We tried a write call, but the server did not 12161da177e4SLinus Torvalds * commit data to stable storage even though we 12171da177e4SLinus Torvalds * requested it. 12181da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 12191da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 12201da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 12211da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 12221da177e4SLinus Torvalds */ 12231da177e4SLinus Torvalds static unsigned long complain; 12241da177e4SLinus Torvalds 1225a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 12261da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 12271da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 12281da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1229eedc020eSAndy Adamson server->nfs_client->cl_hostname, 12301da177e4SLinus Torvalds resp->verf->committed, argp->stable); 12311da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 12321da177e4SLinus Torvalds } 12331da177e4SLinus Torvalds } 12341da177e4SLinus Torvalds #endif 12351da177e4SLinus Torvalds /* Is this a short write? */ 12361da177e4SLinus Torvalds if (task->tk_status >= 0 && resp->count < argp->count) { 12371da177e4SLinus Torvalds static unsigned long complain; 12381da177e4SLinus Torvalds 123991d5b470SChuck Lever nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); 124091d5b470SChuck Lever 12411da177e4SLinus Torvalds /* Has the server at least made some progress? */ 12421da177e4SLinus Torvalds if (resp->count != 0) { 12431da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 12441da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 12451da177e4SLinus Torvalds /* Resend from where the server left off */ 1246a69aef14SFred Isaman data->mds_offset += resp->count; 12471da177e4SLinus Torvalds argp->offset += resp->count; 12481da177e4SLinus Torvalds argp->pgbase += resp->count; 12491da177e4SLinus Torvalds argp->count -= resp->count; 12501da177e4SLinus Torvalds } else { 12511da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 12521da177e4SLinus Torvalds * headaches in the case of a server crash. 12531da177e4SLinus Torvalds */ 12541da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 12551da177e4SLinus Torvalds } 12560110ee15STrond Myklebust nfs_restart_rpc(task, server->nfs_client); 125713602896SFred Isaman return; 12581da177e4SLinus Torvalds } 12591da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 12601da177e4SLinus Torvalds printk(KERN_WARNING 12611da177e4SLinus Torvalds "NFS: Server wrote zero bytes, expected %u.\n", 12621da177e4SLinus Torvalds argp->count); 12631da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 12641da177e4SLinus Torvalds } 12651da177e4SLinus Torvalds /* Can't do anything about it except throw an error. */ 12661da177e4SLinus Torvalds task->tk_status = -EIO; 12671da177e4SLinus Torvalds } 126813602896SFred Isaman return; 12691da177e4SLinus Torvalds } 12701da177e4SLinus Torvalds 12711da177e4SLinus Torvalds 12721da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 127371d0a611STrond Myklebust static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) 127471d0a611STrond Myklebust { 1275b8413f98STrond Myklebust int ret; 1276b8413f98STrond Myklebust 127771d0a611STrond Myklebust if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) 127871d0a611STrond Myklebust return 1; 1279b8413f98STrond Myklebust if (!may_wait) 128071d0a611STrond Myklebust return 0; 1281b8413f98STrond Myklebust ret = out_of_line_wait_on_bit_lock(&nfsi->flags, 1282b8413f98STrond Myklebust NFS_INO_COMMIT, 1283b8413f98STrond Myklebust nfs_wait_bit_killable, 1284b8413f98STrond Myklebust TASK_KILLABLE); 1285b8413f98STrond Myklebust return (ret < 0) ? ret : 1; 128671d0a611STrond Myklebust } 128771d0a611STrond Myklebust 1288e0c2b380SFred Isaman void nfs_commit_clear_lock(struct nfs_inode *nfsi) 128971d0a611STrond Myklebust { 129071d0a611STrond Myklebust clear_bit(NFS_INO_COMMIT, &nfsi->flags); 129171d0a611STrond Myklebust smp_mb__after_clear_bit(); 129271d0a611STrond Myklebust wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); 129371d0a611STrond Myklebust } 1294e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_clear_lock); 129571d0a611STrond Myklebust 1296e0c2b380SFred Isaman void nfs_commitdata_release(void *data) 12971da177e4SLinus Torvalds { 1298383ba719STrond Myklebust struct nfs_write_data *wdata = data; 1299383ba719STrond Myklebust 1300988b6dceSFred Isaman put_lseg(wdata->lseg); 1301383ba719STrond Myklebust put_nfs_open_context(wdata->args.context); 13021da177e4SLinus Torvalds nfs_commit_free(wdata); 13031da177e4SLinus Torvalds } 1304e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 13051da177e4SLinus Torvalds 1306e0c2b380SFred Isaman int nfs_initiate_commit(struct nfs_write_data *data, struct rpc_clnt *clnt, 13079ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 1308788e7a89STrond Myklebust int how) 13091da177e4SLinus Torvalds { 131007737691STrond Myklebust struct rpc_task *task; 13119ace33cdSFred Isaman int priority = flush_task_priority(how); 1312bdc7f021STrond Myklebust struct rpc_message msg = { 1313bdc7f021STrond Myklebust .rpc_argp = &data->args, 1314bdc7f021STrond Myklebust .rpc_resp = &data->res, 13159ace33cdSFred Isaman .rpc_cred = data->cred, 1316bdc7f021STrond Myklebust }; 131784115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 131807737691STrond Myklebust .task = &data->task, 13199ace33cdSFred Isaman .rpc_client = clnt, 1320bdc7f021STrond Myklebust .rpc_message = &msg, 13219ace33cdSFred Isaman .callback_ops = call_ops, 132284115e1cSTrond Myklebust .callback_data = data, 1323101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 13242c61be0aSTrond Myklebust .flags = RPC_TASK_ASYNC, 13253ff7576dSTrond Myklebust .priority = priority, 132684115e1cSTrond Myklebust }; 1327788e7a89STrond Myklebust /* Set up the initial task struct. */ 13289ace33cdSFred Isaman NFS_PROTO(data->inode)->commit_setup(data, &msg); 13291da177e4SLinus Torvalds 1330a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1331bdc7f021STrond Myklebust 133207737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1333dbae4c73STrond Myklebust if (IS_ERR(task)) 1334dbae4c73STrond Myklebust return PTR_ERR(task); 1335d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1336d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 133707737691STrond Myklebust rpc_put_task(task); 1338dbae4c73STrond Myklebust return 0; 13391da177e4SLinus Torvalds } 1340e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 13411da177e4SLinus Torvalds 13421da177e4SLinus Torvalds /* 13439ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 13449ace33cdSFred Isaman */ 1345e0c2b380SFred Isaman void nfs_init_commit(struct nfs_write_data *data, 1346988b6dceSFred Isaman struct list_head *head, 1347988b6dceSFred Isaman struct pnfs_layout_segment *lseg) 13489ace33cdSFred Isaman { 13499ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 13509ace33cdSFred Isaman struct inode *inode = first->wb_context->path.dentry->d_inode; 13519ace33cdSFred Isaman 13529ace33cdSFred Isaman /* Set up the RPC argument and reply structs 13539ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 13549ace33cdSFred Isaman 13559ace33cdSFred Isaman list_splice_init(head, &data->pages); 13569ace33cdSFred Isaman 13579ace33cdSFred Isaman data->inode = inode; 13589ace33cdSFred Isaman data->cred = first->wb_context->cred; 1359988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 13609ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 13619ace33cdSFred Isaman 13629ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 13639ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 13649ace33cdSFred Isaman data->args.offset = 0; 13659ace33cdSFred Isaman data->args.count = 0; 13669ace33cdSFred Isaman data->args.context = get_nfs_open_context(first->wb_context); 13679ace33cdSFred Isaman data->res.count = 0; 13689ace33cdSFred Isaman data->res.fattr = &data->fattr; 13699ace33cdSFred Isaman data->res.verf = &data->verf; 13709ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 13719ace33cdSFred Isaman } 1372e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 13739ace33cdSFred Isaman 1374e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1375a861a1e1SFred Isaman struct pnfs_layout_segment *lseg) 137664bfeb49SFred Isaman { 137764bfeb49SFred Isaman struct nfs_page *req; 137864bfeb49SFred Isaman 137964bfeb49SFred Isaman while (!list_empty(page_list)) { 138064bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 138164bfeb49SFred Isaman nfs_list_remove_request(req); 1382a861a1e1SFred Isaman nfs_mark_request_commit(req, lseg); 138364bfeb49SFred Isaman dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 138464bfeb49SFred Isaman dec_bdi_stat(req->wb_page->mapping->backing_dev_info, 138564bfeb49SFred Isaman BDI_RECLAIMABLE); 138664bfeb49SFred Isaman nfs_clear_page_tag_locked(req); 138764bfeb49SFred Isaman } 138864bfeb49SFred Isaman } 1389e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 139064bfeb49SFred Isaman 13919ace33cdSFred Isaman /* 13921da177e4SLinus Torvalds * Commit dirty pages 13931da177e4SLinus Torvalds */ 13941da177e4SLinus Torvalds static int 139540859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how) 13961da177e4SLinus Torvalds { 13971da177e4SLinus Torvalds struct nfs_write_data *data; 13981da177e4SLinus Torvalds 1399c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 14001da177e4SLinus Torvalds 14011da177e4SLinus Torvalds if (!data) 14021da177e4SLinus Torvalds goto out_bad; 14031da177e4SLinus Torvalds 14041da177e4SLinus Torvalds /* Set up the argument struct */ 1405988b6dceSFred Isaman nfs_init_commit(data, head, NULL); 14069ace33cdSFred Isaman return nfs_initiate_commit(data, NFS_CLIENT(inode), data->mds_ops, how); 14071da177e4SLinus Torvalds out_bad: 1408a861a1e1SFred Isaman nfs_retry_commit(head, NULL); 140971d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(inode)); 14101da177e4SLinus Torvalds return -ENOMEM; 14111da177e4SLinus Torvalds } 14121da177e4SLinus Torvalds 14131da177e4SLinus Torvalds /* 14141da177e4SLinus Torvalds * COMMIT call returned 14151da177e4SLinus Torvalds */ 1416788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 14171da177e4SLinus Torvalds { 1418963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 14191da177e4SLinus Torvalds 1420a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 14211da177e4SLinus Torvalds task->tk_pid, task->tk_status); 14221da177e4SLinus Torvalds 1423788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1424c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1425c9d8f89dSTrond Myklebust } 1426c9d8f89dSTrond Myklebust 1427e0c2b380SFred Isaman void nfs_commit_release_pages(struct nfs_write_data *data) 1428c9d8f89dSTrond Myklebust { 1429c9d8f89dSTrond Myklebust struct nfs_page *req; 1430c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1431788e7a89STrond Myklebust 14321da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 14331da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 14341da177e4SLinus Torvalds nfs_list_remove_request(req); 1435e468bae9STrond Myklebust nfs_clear_request_commit(req); 14361da177e4SLinus Torvalds 143748186c7dSChuck Lever dprintk("NFS: commit (%s/%lld %d@%lld)", 143888be9f99STrond Myklebust req->wb_context->path.dentry->d_inode->i_sb->s_id, 143988be9f99STrond Myklebust (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), 14401da177e4SLinus Torvalds req->wb_bytes, 14411da177e4SLinus Torvalds (long long)req_offset(req)); 1442c9d8f89dSTrond Myklebust if (status < 0) { 1443c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 14441da177e4SLinus Torvalds nfs_inode_remove_request(req); 1445c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 14461da177e4SLinus Torvalds goto next; 14471da177e4SLinus Torvalds } 14481da177e4SLinus Torvalds 14491da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 14501da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 14511da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 14521da177e4SLinus Torvalds /* We have a match */ 14531da177e4SLinus Torvalds nfs_inode_remove_request(req); 14541da177e4SLinus Torvalds dprintk(" OK\n"); 14551da177e4SLinus Torvalds goto next; 14561da177e4SLinus Torvalds } 14571da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 14581da177e4SLinus Torvalds dprintk(" mismatch\n"); 14596d884e8fSFred nfs_mark_request_dirty(req); 14601da177e4SLinus Torvalds next: 14619fd367f0STrond Myklebust nfs_clear_page_tag_locked(req); 14621da177e4SLinus Torvalds } 14635917ce84SFred Isaman } 1464e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_release_pages); 14655917ce84SFred Isaman 14665917ce84SFred Isaman static void nfs_commit_release(void *calldata) 14675917ce84SFred Isaman { 14685917ce84SFred Isaman struct nfs_write_data *data = calldata; 14695917ce84SFred Isaman 14705917ce84SFred Isaman nfs_commit_release_pages(data); 147171d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(data->inode)); 1472c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 14731da177e4SLinus Torvalds } 1474788e7a89STrond Myklebust 1475788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 147621d9a851SAndy Adamson #if defined(CONFIG_NFS_V4_1) 147721d9a851SAndy Adamson .rpc_call_prepare = nfs_write_prepare, 147821d9a851SAndy Adamson #endif /* CONFIG_NFS_V4_1 */ 1479788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1480788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1481788e7a89STrond Myklebust }; 14821da177e4SLinus Torvalds 1483b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 14841da177e4SLinus Torvalds { 14851da177e4SLinus Torvalds LIST_HEAD(head); 148671d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1487b8413f98STrond Myklebust int res; 14881da177e4SLinus Torvalds 1489b8413f98STrond Myklebust res = nfs_commit_set_lock(NFS_I(inode), may_wait); 1490b8413f98STrond Myklebust if (res <= 0) 1491c5efa5fcSTrond Myklebust goto out_mark_dirty; 14923da28eb1STrond Myklebust res = nfs_scan_commit(inode, &head, 0, 0); 14931da177e4SLinus Torvalds if (res) { 1494a861a1e1SFred Isaman int error; 1495a861a1e1SFred Isaman 1496a861a1e1SFred Isaman error = pnfs_commit_list(inode, &head, how); 1497a861a1e1SFred Isaman if (error == PNFS_NOT_ATTEMPTED) 1498a861a1e1SFred Isaman error = nfs_commit_list(inode, &head, how); 14991da177e4SLinus Torvalds if (error < 0) 15001da177e4SLinus Torvalds return error; 1501b8413f98STrond Myklebust if (!may_wait) 1502b8413f98STrond Myklebust goto out_mark_dirty; 1503b8413f98STrond Myklebust error = wait_on_bit(&NFS_I(inode)->flags, 1504b8413f98STrond Myklebust NFS_INO_COMMIT, 150571d0a611STrond Myklebust nfs_wait_bit_killable, 150671d0a611STrond Myklebust TASK_KILLABLE); 1507b8413f98STrond Myklebust if (error < 0) 1508b8413f98STrond Myklebust return error; 150971d0a611STrond Myklebust } else 151071d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(inode)); 1511c5efa5fcSTrond Myklebust return res; 1512c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1513c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1514c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1515c5efa5fcSTrond Myklebust * that the data is on the disk. 1516c5efa5fcSTrond Myklebust */ 1517c5efa5fcSTrond Myklebust out_mark_dirty: 1518c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 15191da177e4SLinus Torvalds return res; 15201da177e4SLinus Torvalds } 15218fc795f7STrond Myklebust 15228fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 15238fc795f7STrond Myklebust { 1524420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1525420e3646STrond Myklebust int flags = FLUSH_SYNC; 1526420e3646STrond Myklebust int ret = 0; 15278fc795f7STrond Myklebust 1528a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1529a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1530a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1531420e3646STrond Myklebust */ 1532a00dd6c0SJeff Layton if (nfsi->ncommit <= (nfsi->npages >> 1)) 1533420e3646STrond Myklebust goto out_mark_dirty; 1534420e3646STrond Myklebust 1535a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1536420e3646STrond Myklebust flags = 0; 1537a00dd6c0SJeff Layton } 1538a00dd6c0SJeff Layton 1539420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1540420e3646STrond Myklebust if (ret >= 0) { 1541420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1542420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1543420e3646STrond Myklebust wbc->nr_to_write -= ret; 1544420e3646STrond Myklebust else 1545420e3646STrond Myklebust wbc->nr_to_write = 0; 1546420e3646STrond Myklebust } 15478fc795f7STrond Myklebust return 0; 1548420e3646STrond Myklebust } 1549420e3646STrond Myklebust out_mark_dirty: 15508fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 15518fc795f7STrond Myklebust return ret; 15528fc795f7STrond Myklebust } 1553c63c7b05STrond Myklebust #else 15548fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 15558fc795f7STrond Myklebust { 15568fc795f7STrond Myklebust return 0; 15578fc795f7STrond Myklebust } 15581da177e4SLinus Torvalds #endif 15591da177e4SLinus Torvalds 15608fc795f7STrond Myklebust int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 15618fc795f7STrond Myklebust { 1562863a3c6cSAndy Adamson int ret; 1563863a3c6cSAndy Adamson 1564863a3c6cSAndy Adamson ret = nfs_commit_unstable_pages(inode, wbc); 1565863a3c6cSAndy Adamson if (ret >= 0 && test_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags)) { 1566ef311537SAndy Adamson int status; 1567ef311537SAndy Adamson bool sync = true; 1568863a3c6cSAndy Adamson 1569ef311537SAndy Adamson if (wbc->sync_mode == WB_SYNC_NONE || wbc->nonblocking || 1570ef311537SAndy Adamson wbc->for_background) 1571ef311537SAndy Adamson sync = false; 1572863a3c6cSAndy Adamson 1573863a3c6cSAndy Adamson status = pnfs_layoutcommit_inode(inode, sync); 1574863a3c6cSAndy Adamson if (status < 0) 1575863a3c6cSAndy Adamson return status; 1576863a3c6cSAndy Adamson } 1577863a3c6cSAndy Adamson return ret; 15788fc795f7STrond Myklebust } 15798fc795f7STrond Myklebust 1580acdc53b2STrond Myklebust /* 1581acdc53b2STrond Myklebust * flush the inode to disk. 1582acdc53b2STrond Myklebust */ 1583acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 158434901f70STrond Myklebust { 158534901f70STrond Myklebust struct writeback_control wbc = { 158672cb77f4STrond Myklebust .sync_mode = WB_SYNC_ALL, 158734901f70STrond Myklebust .nr_to_write = LONG_MAX, 1588d7fb1207STrond Myklebust .range_start = 0, 1589d7fb1207STrond Myklebust .range_end = LLONG_MAX, 159034901f70STrond Myklebust }; 159134901f70STrond Myklebust 1592acdc53b2STrond Myklebust return sync_inode(inode, &wbc); 15931c75950bSTrond Myklebust } 15941c75950bSTrond Myklebust 15951b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 15961b3b4a1aSTrond Myklebust { 15971b3b4a1aSTrond Myklebust struct nfs_page *req; 15981b3b4a1aSTrond Myklebust int ret = 0; 15991b3b4a1aSTrond Myklebust 16001b3b4a1aSTrond Myklebust BUG_ON(!PageLocked(page)); 16011b3b4a1aSTrond Myklebust for (;;) { 1602ba8b06e6STrond Myklebust wait_on_page_writeback(page); 16031b3b4a1aSTrond Myklebust req = nfs_page_find_request(page); 16041b3b4a1aSTrond Myklebust if (req == NULL) 16051b3b4a1aSTrond Myklebust break; 16061b3b4a1aSTrond Myklebust if (nfs_lock_request_dontget(req)) { 16071b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 16081b3b4a1aSTrond Myklebust /* 16091b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 16101b3b4a1aSTrond Myklebust * page as being dirty 16111b3b4a1aSTrond Myklebust */ 16121b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 16131b3b4a1aSTrond Myklebust nfs_unlock_request(req); 16141b3b4a1aSTrond Myklebust break; 16151b3b4a1aSTrond Myklebust } 16161b3b4a1aSTrond Myklebust ret = nfs_wait_on_request(req); 1617c9edda71STrond Myklebust nfs_release_request(req); 16181b3b4a1aSTrond Myklebust if (ret < 0) 1619c988950eSTrond Myklebust break; 16201b3b4a1aSTrond Myklebust } 16211b3b4a1aSTrond Myklebust return ret; 16221b3b4a1aSTrond Myklebust } 16231b3b4a1aSTrond Myklebust 16241c75950bSTrond Myklebust /* 16251c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 16261c75950bSTrond Myklebust */ 16271c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 16281c75950bSTrond Myklebust { 16297f2f12d9STrond Myklebust loff_t range_start = page_offset(page); 16307f2f12d9STrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 16317f2f12d9STrond Myklebust struct writeback_control wbc = { 16327f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 16337f2f12d9STrond Myklebust .nr_to_write = 0, 16347f2f12d9STrond Myklebust .range_start = range_start, 16357f2f12d9STrond Myklebust .range_end = range_end, 16367f2f12d9STrond Myklebust }; 16377f2f12d9STrond Myklebust int ret; 16387f2f12d9STrond Myklebust 16390522f6adSTrond Myklebust for (;;) { 1640ba8b06e6STrond Myklebust wait_on_page_writeback(page); 16417f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 16427f2f12d9STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 16437f2f12d9STrond Myklebust if (ret < 0) 16447f2f12d9STrond Myklebust goto out_error; 16450522f6adSTrond Myklebust continue; 16467f2f12d9STrond Myklebust } 16470522f6adSTrond Myklebust if (!PagePrivate(page)) 16480522f6adSTrond Myklebust break; 16490522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 16507f2f12d9STrond Myklebust if (ret < 0) 16517f2f12d9STrond Myklebust goto out_error; 16527f2f12d9STrond Myklebust } 16537f2f12d9STrond Myklebust return 0; 16547f2f12d9STrond Myklebust out_error: 16557f2f12d9STrond Myklebust return ret; 16561c75950bSTrond Myklebust } 16571c75950bSTrond Myklebust 1658074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 1659074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 1660074cc1deSTrond Myklebust struct page *page) 1661074cc1deSTrond Myklebust { 1662074cc1deSTrond Myklebust struct nfs_page *req; 1663074cc1deSTrond Myklebust int ret; 1664074cc1deSTrond Myklebust 1665074cc1deSTrond Myklebust nfs_fscache_release_page(page, GFP_KERNEL); 1666074cc1deSTrond Myklebust 1667cfb506e1STrond Myklebust req = nfs_find_and_lock_request(page, false); 1668074cc1deSTrond Myklebust ret = PTR_ERR(req); 1669074cc1deSTrond Myklebust if (IS_ERR(req)) 1670074cc1deSTrond Myklebust goto out; 1671074cc1deSTrond Myklebust 1672074cc1deSTrond Myklebust ret = migrate_page(mapping, newpage, page); 1673074cc1deSTrond Myklebust if (!req) 1674074cc1deSTrond Myklebust goto out; 1675074cc1deSTrond Myklebust if (ret) 1676074cc1deSTrond Myklebust goto out_unlock; 1677074cc1deSTrond Myklebust page_cache_get(newpage); 1678190f38e5STrond Myklebust spin_lock(&mapping->host->i_lock); 1679074cc1deSTrond Myklebust req->wb_page = newpage; 1680074cc1deSTrond Myklebust SetPagePrivate(newpage); 1681190f38e5STrond Myklebust set_page_private(newpage, (unsigned long)req); 1682074cc1deSTrond Myklebust ClearPagePrivate(page); 1683074cc1deSTrond Myklebust set_page_private(page, 0); 1684190f38e5STrond Myklebust spin_unlock(&mapping->host->i_lock); 1685074cc1deSTrond Myklebust page_cache_release(page); 1686074cc1deSTrond Myklebust out_unlock: 1687074cc1deSTrond Myklebust nfs_clear_page_tag_locked(req); 1688074cc1deSTrond Myklebust out: 1689074cc1deSTrond Myklebust return ret; 1690074cc1deSTrond Myklebust } 1691074cc1deSTrond Myklebust #endif 1692074cc1deSTrond Myklebust 1693f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 16941da177e4SLinus Torvalds { 16951da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 16961da177e4SLinus Torvalds sizeof(struct nfs_write_data), 16971da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 169820c2df83SPaul Mundt NULL); 16991da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 17001da177e4SLinus Torvalds return -ENOMEM; 17011da177e4SLinus Torvalds 170293d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 17031da177e4SLinus Torvalds nfs_wdata_cachep); 17041da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 17051da177e4SLinus Torvalds return -ENOMEM; 17061da177e4SLinus Torvalds 170793d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 17081da177e4SLinus Torvalds nfs_wdata_cachep); 17091da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 17101da177e4SLinus Torvalds return -ENOMEM; 17111da177e4SLinus Torvalds 171289a09141SPeter Zijlstra /* 171389a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 171489a09141SPeter Zijlstra * 171589a09141SPeter Zijlstra * 64MB: 8192k 171689a09141SPeter Zijlstra * 128MB: 11585k 171789a09141SPeter Zijlstra * 256MB: 16384k 171889a09141SPeter Zijlstra * 512MB: 23170k 171989a09141SPeter Zijlstra * 1GB: 32768k 172089a09141SPeter Zijlstra * 2GB: 46340k 172189a09141SPeter Zijlstra * 4GB: 65536k 172289a09141SPeter Zijlstra * 8GB: 92681k 172389a09141SPeter Zijlstra * 16GB: 131072k 172489a09141SPeter Zijlstra * 172589a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 172689a09141SPeter Zijlstra * Limit the default to 256M 172789a09141SPeter Zijlstra */ 172889a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 172989a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 173089a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 173189a09141SPeter Zijlstra 17321da177e4SLinus Torvalds return 0; 17331da177e4SLinus Torvalds } 17341da177e4SLinus Torvalds 1735266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 17361da177e4SLinus Torvalds { 17371da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 17381da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 17391a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 17401da177e4SLinus Torvalds } 17411da177e4SLinus Torvalds 1742