11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/nfs/write.c 31da177e4SLinus Torvalds * 47c85d900STrond Myklebust * Write file data over NFS. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> 71da177e4SLinus Torvalds */ 81da177e4SLinus Torvalds 91da177e4SLinus Torvalds #include <linux/types.h> 101da177e4SLinus Torvalds #include <linux/slab.h> 111da177e4SLinus Torvalds #include <linux/mm.h> 121da177e4SLinus Torvalds #include <linux/pagemap.h> 131da177e4SLinus Torvalds #include <linux/file.h> 141da177e4SLinus Torvalds #include <linux/writeback.h> 1589a09141SPeter Zijlstra #include <linux/swap.h> 16074cc1deSTrond Myklebust #include <linux/migrate.h> 171da177e4SLinus Torvalds 181da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h> 191da177e4SLinus Torvalds #include <linux/nfs_fs.h> 201da177e4SLinus Torvalds #include <linux/nfs_mount.h> 211da177e4SLinus Torvalds #include <linux/nfs_page.h> 223fcfab16SAndrew Morton #include <linux/backing-dev.h> 23afeacc8cSPaul Gortmaker #include <linux/export.h> 243fcfab16SAndrew Morton 251da177e4SLinus Torvalds #include <asm/uaccess.h> 261da177e4SLinus Torvalds 271da177e4SLinus Torvalds #include "delegation.h" 2849a70f27STrond Myklebust #include "internal.h" 2991d5b470SChuck Lever #include "iostat.h" 30def6ed7eSAndy Adamson #include "nfs4_fs.h" 31074cc1deSTrond Myklebust #include "fscache.h" 3294ad1c80SFred Isaman #include "pnfs.h" 331da177e4SLinus Torvalds 34f4ce1299STrond Myklebust #include "nfstrace.h" 35f4ce1299STrond Myklebust 361da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 371da177e4SLinus Torvalds 381da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 391da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 401da177e4SLinus Torvalds 411da177e4SLinus Torvalds /* 421da177e4SLinus Torvalds * Local function declarations 431da177e4SLinus Torvalds */ 44f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 45788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 46061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; 47f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops; 484a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops; 491da177e4SLinus Torvalds 50e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 513feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 520b7c0153SFred Isaman static struct kmem_cache *nfs_cdata_cachep; 531da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 541da177e4SLinus Torvalds 550b7c0153SFred Isaman struct nfs_commit_data *nfs_commitdata_alloc(void) 561da177e4SLinus Torvalds { 57192e501bSMel Gorman struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); 5840859d7eSChuck Lever 591da177e4SLinus Torvalds if (p) { 601da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 611da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 621da177e4SLinus Torvalds } 631da177e4SLinus Torvalds return p; 641da177e4SLinus Torvalds } 65e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 661da177e4SLinus Torvalds 670b7c0153SFred Isaman void nfs_commit_free(struct nfs_commit_data *p) 681da177e4SLinus Torvalds { 691da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 701da177e4SLinus Torvalds } 71e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 721da177e4SLinus Torvalds 734a0de55cSAnna Schumaker static struct nfs_rw_header *nfs_writehdr_alloc(void) 743feb2d49STrond Myklebust { 75c0752cdfSAnna Schumaker struct nfs_rw_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); 763feb2d49STrond Myklebust 774a0de55cSAnna Schumaker if (p) 783feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 793feb2d49STrond Myklebust return p; 803feb2d49STrond Myklebust } 813feb2d49STrond Myklebust 824a0de55cSAnna Schumaker static void nfs_writehdr_free(struct nfs_rw_header *whdr) 833feb2d49STrond Myklebust { 84cd841605SFred Isaman mempool_free(whdr, nfs_wdata_mempool); 853feb2d49STrond Myklebust } 863feb2d49STrond Myklebust 877b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 887b159fc1STrond Myklebust { 897b159fc1STrond Myklebust ctx->error = error; 907b159fc1STrond Myklebust smp_wmb(); 917b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 927b159fc1STrond Myklebust } 937b159fc1STrond Myklebust 9429418aa4SMel Gorman static struct nfs_page * 9529418aa4SMel Gorman nfs_page_find_request_locked(struct nfs_inode *nfsi, struct page *page) 96277459d2STrond Myklebust { 97277459d2STrond Myklebust struct nfs_page *req = NULL; 98277459d2STrond Myklebust 9929418aa4SMel Gorman if (PagePrivate(page)) 100277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 10129418aa4SMel Gorman else if (unlikely(PageSwapCache(page))) { 10229418aa4SMel Gorman struct nfs_page *freq, *t; 10329418aa4SMel Gorman 10429418aa4SMel Gorman /* Linearly search the commit list for the correct req */ 10529418aa4SMel Gorman list_for_each_entry_safe(freq, t, &nfsi->commit_info.list, wb_list) { 10629418aa4SMel Gorman if (freq->wb_page == page) { 10729418aa4SMel Gorman req = freq; 10829418aa4SMel Gorman break; 109277459d2STrond Myklebust } 11029418aa4SMel Gorman } 11129418aa4SMel Gorman } 11229418aa4SMel Gorman 11329418aa4SMel Gorman if (req) 11429418aa4SMel Gorman kref_get(&req->wb_kref); 11529418aa4SMel Gorman 116277459d2STrond Myklebust return req; 117277459d2STrond Myklebust } 118277459d2STrond Myklebust 119277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request(struct page *page) 120277459d2STrond Myklebust { 121d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 122277459d2STrond Myklebust struct nfs_page *req = NULL; 123277459d2STrond Myklebust 124587142f8STrond Myklebust spin_lock(&inode->i_lock); 12529418aa4SMel Gorman req = nfs_page_find_request_locked(NFS_I(inode), page); 126587142f8STrond Myklebust spin_unlock(&inode->i_lock); 127277459d2STrond Myklebust return req; 128277459d2STrond Myklebust } 129277459d2STrond Myklebust 1301da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1311da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1321da177e4SLinus Torvalds { 133d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 134a3d01454STrond Myklebust loff_t end, i_size; 135a3d01454STrond Myklebust pgoff_t end_index; 1361da177e4SLinus Torvalds 137a3d01454STrond Myklebust spin_lock(&inode->i_lock); 138a3d01454STrond Myklebust i_size = i_size_read(inode); 139a3d01454STrond Myklebust end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 140d56b4ddfSMel Gorman if (i_size > 0 && page_file_index(page) < end_index) 141a3d01454STrond Myklebust goto out; 142d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 1431da177e4SLinus Torvalds if (i_size >= end) 144a3d01454STrond Myklebust goto out; 1451da177e4SLinus Torvalds i_size_write(inode, end); 146a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 147a3d01454STrond Myklebust out: 148a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 1491da177e4SLinus Torvalds } 1501da177e4SLinus Torvalds 151a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 152a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 153a301b777STrond Myklebust { 154d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 155a301b777STrond Myklebust } 156a301b777STrond Myklebust 1571da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 1581da177e4SLinus Torvalds * covers the full page. 1591da177e4SLinus Torvalds */ 1601da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 1611da177e4SLinus Torvalds { 1621da177e4SLinus Torvalds if (PageUptodate(page)) 1631da177e4SLinus Torvalds return; 1641da177e4SLinus Torvalds if (base != 0) 1651da177e4SLinus Torvalds return; 16649a70f27STrond Myklebust if (count != nfs_page_length(page)) 1671da177e4SLinus Torvalds return; 1681da177e4SLinus Torvalds SetPageUptodate(page); 1691da177e4SLinus Torvalds } 1701da177e4SLinus Torvalds 1711da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 1721da177e4SLinus Torvalds { 1731da177e4SLinus Torvalds if (wbc->for_reclaim) 174c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 175b17621feSWu Fengguang if (wbc->for_kupdate || wbc->for_background) 176b31268acSTrond Myklebust return FLUSH_LOWPRI | FLUSH_COND_STABLE; 177b31268acSTrond Myklebust return FLUSH_COND_STABLE; 1781da177e4SLinus Torvalds } 1791da177e4SLinus Torvalds 1801da177e4SLinus Torvalds /* 18189a09141SPeter Zijlstra * NFS congestion control 18289a09141SPeter Zijlstra */ 18389a09141SPeter Zijlstra 18489a09141SPeter Zijlstra int nfs_congestion_kb; 18589a09141SPeter Zijlstra 18689a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 18789a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 18889a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 18989a09141SPeter Zijlstra 190deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 19189a09141SPeter Zijlstra { 192deed85e7STrond Myklebust struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); 1935a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 1945a6d41b3STrond Myklebust 195deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 19689a09141SPeter Zijlstra 197277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 1988aa7e847SJens Axboe NFS_CONGESTION_ON_THRESH) { 1998aa7e847SJens Axboe set_bdi_congested(&nfss->backing_dev_info, 2008aa7e847SJens Axboe BLK_RW_ASYNC); 2018aa7e847SJens Axboe } 20289a09141SPeter Zijlstra } 20389a09141SPeter Zijlstra 20489a09141SPeter Zijlstra static void nfs_end_page_writeback(struct page *page) 20589a09141SPeter Zijlstra { 206d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 20789a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 20889a09141SPeter Zijlstra 20989a09141SPeter Zijlstra end_page_writeback(page); 210c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 2118aa7e847SJens Axboe clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 21289a09141SPeter Zijlstra } 21389a09141SPeter Zijlstra 214cfb506e1STrond Myklebust static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) 215e261f51fSTrond Myklebust { 216d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 217e261f51fSTrond Myklebust struct nfs_page *req; 218e261f51fSTrond Myklebust int ret; 219e261f51fSTrond Myklebust 220587142f8STrond Myklebust spin_lock(&inode->i_lock); 221e261f51fSTrond Myklebust for (;;) { 22229418aa4SMel Gorman req = nfs_page_find_request_locked(NFS_I(inode), page); 223074cc1deSTrond Myklebust if (req == NULL) 224074cc1deSTrond Myklebust break; 2257ad84aa9STrond Myklebust if (nfs_lock_request(req)) 226e261f51fSTrond Myklebust break; 227e261f51fSTrond Myklebust /* Note: If we hold the page lock, as is the case in nfs_writepage, 2287ad84aa9STrond Myklebust * then the call to nfs_lock_request() will always 229e261f51fSTrond Myklebust * succeed provided that someone hasn't already marked the 230e261f51fSTrond Myklebust * request as dirty (in which case we don't care). 231e261f51fSTrond Myklebust */ 232587142f8STrond Myklebust spin_unlock(&inode->i_lock); 233cfb506e1STrond Myklebust if (!nonblock) 234e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 235cfb506e1STrond Myklebust else 236cfb506e1STrond Myklebust ret = -EAGAIN; 237e261f51fSTrond Myklebust nfs_release_request(req); 238e261f51fSTrond Myklebust if (ret != 0) 239074cc1deSTrond Myklebust return ERR_PTR(ret); 240587142f8STrond Myklebust spin_lock(&inode->i_lock); 241e261f51fSTrond Myklebust } 242587142f8STrond Myklebust spin_unlock(&inode->i_lock); 243074cc1deSTrond Myklebust return req; 244612c9384STrond Myklebust } 245074cc1deSTrond Myklebust 246074cc1deSTrond Myklebust /* 247074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 248074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 249074cc1deSTrond Myklebust */ 250074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 251cfb506e1STrond Myklebust struct page *page, bool nonblock) 252074cc1deSTrond Myklebust { 253074cc1deSTrond Myklebust struct nfs_page *req; 254074cc1deSTrond Myklebust int ret = 0; 255074cc1deSTrond Myklebust 256cfb506e1STrond Myklebust req = nfs_find_and_lock_request(page, nonblock); 257074cc1deSTrond Myklebust if (!req) 258074cc1deSTrond Myklebust goto out; 259074cc1deSTrond Myklebust ret = PTR_ERR(req); 260074cc1deSTrond Myklebust if (IS_ERR(req)) 261074cc1deSTrond Myklebust goto out; 262074cc1deSTrond Myklebust 263deed85e7STrond Myklebust nfs_set_page_writeback(page); 264deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 265074cc1deSTrond Myklebust 266deed85e7STrond Myklebust ret = 0; 267f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 268f8512ad0SFred Isaman nfs_redirty_request(req); 269074cc1deSTrond Myklebust ret = pgio->pg_error; 270f8512ad0SFred Isaman } 271074cc1deSTrond Myklebust out: 272074cc1deSTrond Myklebust return ret; 273e261f51fSTrond Myklebust } 274e261f51fSTrond Myklebust 275f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 276f758c885STrond Myklebust { 277d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 278cfb506e1STrond Myklebust int ret; 279f758c885STrond Myklebust 280f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 281f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 282f758c885STrond Myklebust 283d56b4ddfSMel Gorman nfs_pageio_cond_complete(pgio, page_file_index(page)); 2841b430beeSWu Fengguang ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); 285cfb506e1STrond Myklebust if (ret == -EAGAIN) { 286cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 287cfb506e1STrond Myklebust ret = 0; 288cfb506e1STrond Myklebust } 289cfb506e1STrond Myklebust return ret; 290f758c885STrond Myklebust } 291f758c885STrond Myklebust 292e261f51fSTrond Myklebust /* 2931da177e4SLinus Torvalds * Write an mmapped page to the server. 2941da177e4SLinus Torvalds */ 2954d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 2961da177e4SLinus Torvalds { 297f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 298e261f51fSTrond Myklebust int err; 2991da177e4SLinus Torvalds 300a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc), 301a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 302f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 303f758c885STrond Myklebust nfs_pageio_complete(&pgio); 304f758c885STrond Myklebust if (err < 0) 3054d770ccfSTrond Myklebust return err; 306f758c885STrond Myklebust if (pgio.pg_error < 0) 307f758c885STrond Myklebust return pgio.pg_error; 308f758c885STrond Myklebust return 0; 3094d770ccfSTrond Myklebust } 3104d770ccfSTrond Myklebust 3114d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 3124d770ccfSTrond Myklebust { 313f758c885STrond Myklebust int ret; 3144d770ccfSTrond Myklebust 315f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 3161da177e4SLinus Torvalds unlock_page(page); 317f758c885STrond Myklebust return ret; 318f758c885STrond Myklebust } 319f758c885STrond Myklebust 320f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 321f758c885STrond Myklebust { 322f758c885STrond Myklebust int ret; 323f758c885STrond Myklebust 324f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 325f758c885STrond Myklebust unlock_page(page); 326f758c885STrond Myklebust return ret; 3271da177e4SLinus Torvalds } 3281da177e4SLinus Torvalds 3291da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 3301da177e4SLinus Torvalds { 3311da177e4SLinus Torvalds struct inode *inode = mapping->host; 33272cb77f4STrond Myklebust unsigned long *bitlock = &NFS_I(inode)->flags; 333c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 3341da177e4SLinus Torvalds int err; 3351da177e4SLinus Torvalds 33672cb77f4STrond Myklebust /* Stop dirtying of new pages while we sync */ 33772cb77f4STrond Myklebust err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, 33872cb77f4STrond Myklebust nfs_wait_bit_killable, TASK_KILLABLE); 33972cb77f4STrond Myklebust if (err) 34072cb77f4STrond Myklebust goto out_err; 34172cb77f4STrond Myklebust 34291d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 34391d5b470SChuck Lever 344a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 345a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 346f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 347c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 34872cb77f4STrond Myklebust 34972cb77f4STrond Myklebust clear_bit_unlock(NFS_INO_FLUSHING, bitlock); 35072cb77f4STrond Myklebust smp_mb__after_clear_bit(); 35172cb77f4STrond Myklebust wake_up_bit(bitlock, NFS_INO_FLUSHING); 35272cb77f4STrond Myklebust 353f758c885STrond Myklebust if (err < 0) 35472cb77f4STrond Myklebust goto out_err; 35572cb77f4STrond Myklebust err = pgio.pg_error; 35672cb77f4STrond Myklebust if (err < 0) 35772cb77f4STrond Myklebust goto out_err; 358c63c7b05STrond Myklebust return 0; 35972cb77f4STrond Myklebust out_err: 36072cb77f4STrond Myklebust return err; 3611da177e4SLinus Torvalds } 3621da177e4SLinus Torvalds 3631da177e4SLinus Torvalds /* 3641da177e4SLinus Torvalds * Insert a write request into an inode 3651da177e4SLinus Torvalds */ 366d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 3671da177e4SLinus Torvalds { 3681da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 369e7d39069STrond Myklebust 370*2bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 371*2bfc6e56SWeston Andros Adamson 372e7d39069STrond Myklebust /* Lock the request! */ 3737ad84aa9STrond Myklebust nfs_lock_request(req); 374e7d39069STrond Myklebust 375e7d39069STrond Myklebust spin_lock(&inode->i_lock); 376011e2a7fSBryan Schumaker if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 377a9a4a87aSTrond Myklebust inode->i_version++; 37829418aa4SMel Gorman /* 37929418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 38029418aa4SMel Gorman * with invalidate/truncate. 38129418aa4SMel Gorman */ 38229418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 3832df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 384deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 385277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 38629418aa4SMel Gorman } 3871da177e4SLinus Torvalds nfsi->npages++; 388*2bfc6e56SWeston Andros Adamson set_bit(PG_INODE_REF, &req->wb_flags); 389c03b4024STrond Myklebust kref_get(&req->wb_kref); 390e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 3911da177e4SLinus Torvalds } 3921da177e4SLinus Torvalds 3931da177e4SLinus Torvalds /* 39489a09141SPeter Zijlstra * Remove a write request from an inode 3951da177e4SLinus Torvalds */ 3961da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 3971da177e4SLinus Torvalds { 3983d4ff43dSAl Viro struct inode *inode = req->wb_context->dentry->d_inode; 3991da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4001da177e4SLinus Torvalds 401587142f8STrond Myklebust spin_lock(&inode->i_lock); 40229418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 403277459d2STrond Myklebust set_page_private(req->wb_page, 0); 404deb7d638STrond Myklebust ClearPagePrivate(req->wb_page); 4052df485a7STrond Myklebust clear_bit(PG_MAPPED, &req->wb_flags); 40629418aa4SMel Gorman } 4071da177e4SLinus Torvalds nfsi->npages--; 408587142f8STrond Myklebust spin_unlock(&inode->i_lock); 4091da177e4SLinus Torvalds nfs_release_request(req); 4101da177e4SLinus Torvalds } 4111da177e4SLinus Torvalds 41261822ab5STrond Myklebust static void 4136d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 41461822ab5STrond Myklebust { 41561822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 41661822ab5STrond Myklebust } 41761822ab5STrond Myklebust 41889d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 4198dd37758STrond Myklebust /** 4208dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 4218dd37758STrond Myklebust * @req: pointer to a struct nfs_page 422ea2cf228SFred Isaman * @dst: commit list head 423ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 4248dd37758STrond Myklebust * 425ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 4268dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 4278dd37758STrond Myklebust * the MM page stats. 4288dd37758STrond Myklebust * 429ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 4308dd37758STrond Myklebust * holding the nfs_page lock. 4318dd37758STrond Myklebust */ 4328dd37758STrond Myklebust void 433ea2cf228SFred Isaman nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, 434ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 4358dd37758STrond Myklebust { 4368dd37758STrond Myklebust set_bit(PG_CLEAN, &(req)->wb_flags); 437ea2cf228SFred Isaman spin_lock(cinfo->lock); 438ea2cf228SFred Isaman nfs_list_add_request(req, dst); 439ea2cf228SFred Isaman cinfo->mds->ncommit++; 440ea2cf228SFred Isaman spin_unlock(cinfo->lock); 44156f9cd68SFred Isaman if (!cinfo->dreq) { 4428dd37758STrond Myklebust inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 443d56b4ddfSMel Gorman inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, 44456f9cd68SFred Isaman BDI_RECLAIMABLE); 44556f9cd68SFred Isaman __mark_inode_dirty(req->wb_context->dentry->d_inode, 44656f9cd68SFred Isaman I_DIRTY_DATASYNC); 44756f9cd68SFred Isaman } 4488dd37758STrond Myklebust } 4498dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 4508dd37758STrond Myklebust 4518dd37758STrond Myklebust /** 4528dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 4538dd37758STrond Myklebust * @req: pointer to a nfs_page 454ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 4558dd37758STrond Myklebust * 456ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 4578dd37758STrond Myklebust * number of outstanding requests requiring a commit 4588dd37758STrond Myklebust * It does not update the MM page stats. 4598dd37758STrond Myklebust * 460ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 4618dd37758STrond Myklebust */ 4628dd37758STrond Myklebust void 463ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 464ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 4658dd37758STrond Myklebust { 4668dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 4678dd37758STrond Myklebust return; 4688dd37758STrond Myklebust nfs_list_remove_request(req); 469ea2cf228SFred Isaman cinfo->mds->ncommit--; 4708dd37758STrond Myklebust } 4718dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 4728dd37758STrond Myklebust 473ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 474ea2cf228SFred Isaman struct inode *inode) 475ea2cf228SFred Isaman { 476ea2cf228SFred Isaman cinfo->lock = &inode->i_lock; 477ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 478ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 479b359f9d0SFred Isaman cinfo->dreq = NULL; 480f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 481ea2cf228SFred Isaman } 482ea2cf228SFred Isaman 483ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 484ea2cf228SFred Isaman struct inode *inode, 485ea2cf228SFred Isaman struct nfs_direct_req *dreq) 486ea2cf228SFred Isaman { 4871763da12SFred Isaman if (dreq) 4881763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 4891763da12SFred Isaman else 490ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 491ea2cf228SFred Isaman } 492ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 4938dd37758STrond Myklebust 4941da177e4SLinus Torvalds /* 4951da177e4SLinus Torvalds * Add a request to the inode's commit list. 4961da177e4SLinus Torvalds */ 4971763da12SFred Isaman void 498ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 499ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 5001da177e4SLinus Torvalds { 501ea2cf228SFred Isaman if (pnfs_mark_request_commit(req, lseg, cinfo)) 5028dd37758STrond Myklebust return; 503ea2cf228SFred Isaman nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); 5041da177e4SLinus Torvalds } 5058e821cadSTrond Myklebust 506d6d6dc7cSFred Isaman static void 507d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 508e468bae9STrond Myklebust { 509e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 510d56b4ddfSMel Gorman dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE); 511e468bae9STrond Myklebust } 512d6d6dc7cSFred Isaman 5138dd37758STrond Myklebust static void 514d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 515d6d6dc7cSFred Isaman { 5168dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 5178dd37758STrond Myklebust struct inode *inode = req->wb_context->dentry->d_inode; 518ea2cf228SFred Isaman struct nfs_commit_info cinfo; 519d6d6dc7cSFred Isaman 520ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 521ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 522ea2cf228SFred Isaman spin_lock(cinfo.lock); 523ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 524ea2cf228SFred Isaman spin_unlock(cinfo.lock); 525d6d6dc7cSFred Isaman } 5268dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 5278dd37758STrond Myklebust } 528e468bae9STrond Myklebust } 529e468bae9STrond Myklebust 5308e821cadSTrond Myklebust static inline 5319c7e1b3dSAnna Schumaker int nfs_write_need_commit(struct nfs_pgio_data *data) 5328e821cadSTrond Myklebust { 533465d5243SFred Isaman if (data->verf.committed == NFS_DATA_SYNC) 534cd841605SFred Isaman return data->header->lseg == NULL; 5358e821cadSTrond Myklebust return data->verf.committed != NFS_FILE_SYNC; 5368e821cadSTrond Myklebust } 5378e821cadSTrond Myklebust 5388e821cadSTrond Myklebust #else 53968cd6fa4SBryan Schumaker static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 54068cd6fa4SBryan Schumaker struct inode *inode) 54168cd6fa4SBryan Schumaker { 54268cd6fa4SBryan Schumaker } 54368cd6fa4SBryan Schumaker 54468cd6fa4SBryan Schumaker void nfs_init_cinfo(struct nfs_commit_info *cinfo, 54568cd6fa4SBryan Schumaker struct inode *inode, 54668cd6fa4SBryan Schumaker struct nfs_direct_req *dreq) 54768cd6fa4SBryan Schumaker { 54868cd6fa4SBryan Schumaker } 54968cd6fa4SBryan Schumaker 5501763da12SFred Isaman void 551ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 552ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 5538e821cadSTrond Myklebust { 5548e821cadSTrond Myklebust } 5558e821cadSTrond Myklebust 5568dd37758STrond Myklebust static void 557e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 558e468bae9STrond Myklebust { 559e468bae9STrond Myklebust } 560e468bae9STrond Myklebust 5618e821cadSTrond Myklebust static inline 5629c7e1b3dSAnna Schumaker int nfs_write_need_commit(struct nfs_pgio_data *data) 5638e821cadSTrond Myklebust { 5648e821cadSTrond Myklebust return 0; 5658e821cadSTrond Myklebust } 5668e821cadSTrond Myklebust 5671da177e4SLinus Torvalds #endif 5681da177e4SLinus Torvalds 569061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 5706c75dc0dSFred Isaman { 571ea2cf228SFred Isaman struct nfs_commit_info cinfo; 5726c75dc0dSFred Isaman unsigned long bytes = 0; 573*2bfc6e56SWeston Andros Adamson bool do_destroy; 5746c75dc0dSFred Isaman 5756c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 5766c75dc0dSFred Isaman goto out; 577ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 5786c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 5796c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 5806c75dc0dSFred Isaman 5816c75dc0dSFred Isaman bytes += req->wb_bytes; 5826c75dc0dSFred Isaman nfs_list_remove_request(req); 5836c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 5846c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 585d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 5866c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 5876c75dc0dSFred Isaman goto remove_req; 5886c75dc0dSFred Isaman } 5896c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) { 5906c75dc0dSFred Isaman nfs_mark_request_dirty(req); 5916c75dc0dSFred Isaman goto next; 5926c75dc0dSFred Isaman } 5936c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) { 594f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 595ea2cf228SFred Isaman nfs_mark_request_commit(req, hdr->lseg, &cinfo); 5966c75dc0dSFred Isaman goto next; 5976c75dc0dSFred Isaman } 5986c75dc0dSFred Isaman remove_req: 5996c75dc0dSFred Isaman nfs_inode_remove_request(req); 6006c75dc0dSFred Isaman next: 6011d1afcbcSTrond Myklebust nfs_unlock_request(req); 602d1182b33STrond Myklebust nfs_end_page_writeback(req->wb_page); 603*2bfc6e56SWeston Andros Adamson do_destroy = !test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags); 6043aff4ebbSTrond Myklebust nfs_release_request(req); 6056c75dc0dSFred Isaman } 6066c75dc0dSFred Isaman out: 6076c75dc0dSFred Isaman hdr->release(hdr); 6086c75dc0dSFred Isaman } 6096c75dc0dSFred Isaman 61089d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 611ce59515cSAnna Schumaker unsigned long 612ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 613fb8a1f11STrond Myklebust { 614ea2cf228SFred Isaman return cinfo->mds->ncommit; 615fb8a1f11STrond Myklebust } 616fb8a1f11STrond Myklebust 617ea2cf228SFred Isaman /* cinfo->lock held by caller */ 6181763da12SFred Isaman int 619ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 620ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 621d6d6dc7cSFred Isaman { 622d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 623d6d6dc7cSFred Isaman int ret = 0; 624d6d6dc7cSFred Isaman 625d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 6268dd37758STrond Myklebust if (!nfs_lock_request(req)) 6278dd37758STrond Myklebust continue; 6287ad84aa9STrond Myklebust kref_get(&req->wb_kref); 629ea2cf228SFred Isaman if (cond_resched_lock(cinfo->lock)) 6303b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 631ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 6328dd37758STrond Myklebust nfs_list_add_request(req, dst); 633d6d6dc7cSFred Isaman ret++; 6341763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 635d6d6dc7cSFred Isaman break; 636d6d6dc7cSFred Isaman } 637d6d6dc7cSFred Isaman return ret; 638d6d6dc7cSFred Isaman } 639d6d6dc7cSFred Isaman 6401da177e4SLinus Torvalds /* 6411da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 6421da177e4SLinus Torvalds * @inode: NFS inode to scan 643ea2cf228SFred Isaman * @dst: mds destination list 644ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 6451da177e4SLinus Torvalds * 6461da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 6471da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 6481da177e4SLinus Torvalds */ 6491763da12SFred Isaman int 650ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 651ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 6521da177e4SLinus Torvalds { 653d6d6dc7cSFred Isaman int ret = 0; 654fb8a1f11STrond Myklebust 655ea2cf228SFred Isaman spin_lock(cinfo->lock); 656ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 6578dd37758STrond Myklebust const int max = INT_MAX; 658d6d6dc7cSFred Isaman 659ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 660ea2cf228SFred Isaman cinfo, max); 661ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 662d6d6dc7cSFred Isaman } 663ea2cf228SFred Isaman spin_unlock(cinfo->lock); 664ff778d02STrond Myklebust return ret; 6651da177e4SLinus Torvalds } 666d6d6dc7cSFred Isaman 667c42de9ddSTrond Myklebust #else 668ce59515cSAnna Schumaker unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 669fb8a1f11STrond Myklebust { 670fb8a1f11STrond Myklebust return 0; 671fb8a1f11STrond Myklebust } 672fb8a1f11STrond Myklebust 6731763da12SFred Isaman int nfs_scan_commit(struct inode *inode, struct list_head *dst, 674ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 675c42de9ddSTrond Myklebust { 676c42de9ddSTrond Myklebust return 0; 677c42de9ddSTrond Myklebust } 6781da177e4SLinus Torvalds #endif 6791da177e4SLinus Torvalds 6801da177e4SLinus Torvalds /* 681e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 682e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 6831da177e4SLinus Torvalds * 684e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 685e7d39069STrond Myklebust * to disk. 6861da177e4SLinus Torvalds */ 687e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 688e7d39069STrond Myklebust struct page *page, 689e7d39069STrond Myklebust unsigned int offset, 690e7d39069STrond Myklebust unsigned int bytes) 6911da177e4SLinus Torvalds { 692e7d39069STrond Myklebust struct nfs_page *req; 693e7d39069STrond Myklebust unsigned int rqend; 694e7d39069STrond Myklebust unsigned int end; 6951da177e4SLinus Torvalds int error; 696277459d2STrond Myklebust 697e7d39069STrond Myklebust if (!PagePrivate(page)) 698e7d39069STrond Myklebust return NULL; 699e7d39069STrond Myklebust 700e7d39069STrond Myklebust end = offset + bytes; 701e7d39069STrond Myklebust spin_lock(&inode->i_lock); 702e7d39069STrond Myklebust 703e7d39069STrond Myklebust for (;;) { 70429418aa4SMel Gorman req = nfs_page_find_request_locked(NFS_I(inode), page); 705e7d39069STrond Myklebust if (req == NULL) 706e7d39069STrond Myklebust goto out_unlock; 707e7d39069STrond Myklebust 708*2bfc6e56SWeston Andros Adamson /* should be handled by nfs_flush_incompatible */ 709*2bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 710*2bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 711*2bfc6e56SWeston Andros Adamson 712e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 713e7d39069STrond Myklebust /* 714e7d39069STrond Myklebust * Tell the caller to flush out the request if 715e7d39069STrond Myklebust * the offsets are non-contiguous. 716e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 717e7d39069STrond Myklebust * have flushed out requests having wrong owners. 718e7d39069STrond Myklebust */ 719e468bae9STrond Myklebust if (offset > rqend 720e7d39069STrond Myklebust || end < req->wb_offset) 721e7d39069STrond Myklebust goto out_flushme; 722e7d39069STrond Myklebust 7237ad84aa9STrond Myklebust if (nfs_lock_request(req)) 724e7d39069STrond Myklebust break; 725e7d39069STrond Myklebust 726e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 727587142f8STrond Myklebust spin_unlock(&inode->i_lock); 7281da177e4SLinus Torvalds error = nfs_wait_on_request(req); 7291da177e4SLinus Torvalds nfs_release_request(req); 730e7d39069STrond Myklebust if (error != 0) 731e7d39069STrond Myklebust goto out_err; 732e7d39069STrond Myklebust spin_lock(&inode->i_lock); 7331da177e4SLinus Torvalds } 7341da177e4SLinus Torvalds 7351da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 7361da177e4SLinus Torvalds if (offset < req->wb_offset) { 7371da177e4SLinus Torvalds req->wb_offset = offset; 7381da177e4SLinus Torvalds req->wb_pgbase = offset; 7391da177e4SLinus Torvalds } 7401da177e4SLinus Torvalds if (end > rqend) 7411da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 742e7d39069STrond Myklebust else 743e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 744e7d39069STrond Myklebust out_unlock: 745e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 746ca138f36SFred Isaman if (req) 7478dd37758STrond Myklebust nfs_clear_request_commit(req); 748e7d39069STrond Myklebust return req; 749e7d39069STrond Myklebust out_flushme: 750e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 751e7d39069STrond Myklebust nfs_release_request(req); 752e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 753e7d39069STrond Myklebust out_err: 754e7d39069STrond Myklebust return ERR_PTR(error); 755e7d39069STrond Myklebust } 7561da177e4SLinus Torvalds 757e7d39069STrond Myklebust /* 758e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 759e7d39069STrond Myklebust * 760e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 761e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 762e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 763e7d39069STrond Myklebust */ 764e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 765e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 766e7d39069STrond Myklebust { 767d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 768e7d39069STrond Myklebust struct nfs_page *req; 769e7d39069STrond Myklebust 770e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 771e7d39069STrond Myklebust if (req != NULL) 772e7d39069STrond Myklebust goto out; 773*2bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 774e7d39069STrond Myklebust if (IS_ERR(req)) 775e7d39069STrond Myklebust goto out; 776d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 777efc91ed0STrond Myklebust out: 77861e930a9STrond Myklebust return req; 7791da177e4SLinus Torvalds } 7801da177e4SLinus Torvalds 781e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 782e7d39069STrond Myklebust unsigned int offset, unsigned int count) 783e7d39069STrond Myklebust { 784e7d39069STrond Myklebust struct nfs_page *req; 785e7d39069STrond Myklebust 786e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 787e7d39069STrond Myklebust if (IS_ERR(req)) 788e7d39069STrond Myklebust return PTR_ERR(req); 789e7d39069STrond Myklebust /* Update file length */ 790e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 791e7d39069STrond Myklebust nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); 792a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 7931d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 794e7d39069STrond Myklebust return 0; 795e7d39069STrond Myklebust } 796e7d39069STrond Myklebust 7971da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 7981da177e4SLinus Torvalds { 799cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 8002a369153STrond Myklebust struct nfs_lock_context *l_ctx; 8011da177e4SLinus Torvalds struct nfs_page *req; 8021a54533eSTrond Myklebust int do_flush, status; 8031da177e4SLinus Torvalds /* 8041da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 8051da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 8061da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 8071da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 8081da177e4SLinus Torvalds * Also do the same if we find a request from an existing 8091da177e4SLinus Torvalds * dropped page. 8101da177e4SLinus Torvalds */ 8111a54533eSTrond Myklebust do { 812277459d2STrond Myklebust req = nfs_page_find_request(page); 8131a54533eSTrond Myklebust if (req == NULL) 8141a54533eSTrond Myklebust return 0; 8152a369153STrond Myklebust l_ctx = req->wb_lock_context; 8162a369153STrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx; 817*2bfc6e56SWeston Andros Adamson /* for now, flush if more than 1 request in page_group */ 818*2bfc6e56SWeston Andros Adamson do_flush |= req->wb_this_page != req; 8190f1d2605STrond Myklebust if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) { 8202a369153STrond Myklebust do_flush |= l_ctx->lockowner.l_owner != current->files 8212a369153STrond Myklebust || l_ctx->lockowner.l_pid != current->tgid; 8222a369153STrond Myklebust } 8231da177e4SLinus Torvalds nfs_release_request(req); 8241a54533eSTrond Myklebust if (!do_flush) 8251a54533eSTrond Myklebust return 0; 826d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 8271a54533eSTrond Myklebust } while (status == 0); 8281a54533eSTrond Myklebust return status; 8291da177e4SLinus Torvalds } 8301da177e4SLinus Torvalds 8311da177e4SLinus Torvalds /* 832dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 833dc24826bSAndy Adamson * expire soon. 834dc24826bSAndy Adamson * 835dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 836dc24826bSAndy Adamson * 837dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 838dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 839dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 840dc24826bSAndy Adamson */ 841dc24826bSAndy Adamson int 842dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 843dc24826bSAndy Adamson { 844dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 845dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 846dc24826bSAndy Adamson 847dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 848dc24826bSAndy Adamson } 849dc24826bSAndy Adamson 850dc24826bSAndy Adamson /* 851dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 852dc24826bSAndy Adamson */ 853dc24826bSAndy Adamson bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) 854dc24826bSAndy Adamson { 855dc24826bSAndy Adamson return rpcauth_cred_key_to_expire(ctx->cred); 856dc24826bSAndy Adamson } 857dc24826bSAndy Adamson 858dc24826bSAndy Adamson /* 8595d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 8605d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 8615d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 8625d47a356STrond Myklebust */ 8638d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 8645d47a356STrond Myklebust { 865d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 866d529ef83SJeff Layton 8678d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 8688d197a56STrond Myklebust goto out; 869d529ef83SJeff Layton if (nfsi->cache_validity & (NFS_INO_INVALID_DATA|NFS_INO_REVAL_PAGECACHE)) 870d529ef83SJeff Layton return false; 8714db72b40SJeff Layton smp_rmb(); 872d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 8738d197a56STrond Myklebust return false; 8748d197a56STrond Myklebust out: 8758d197a56STrond Myklebust return PageUptodate(page) != 0; 8765d47a356STrond Myklebust } 8775d47a356STrond Myklebust 878c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 879c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 880c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 881c7559663SScott Mayhew * inefficiencies. 882c7559663SScott Mayhew * 883263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 884263b4509SScott Mayhew * of the checks. 885c7559663SScott Mayhew */ 886c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 887c7559663SScott Mayhew { 888c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 889c7559663SScott Mayhew return 0; 890263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 891263b4509SScott Mayhew return 0; 892c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 893c7559663SScott Mayhew return 1; 894263b4509SScott Mayhew if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 && 895c7559663SScott Mayhew inode->i_flock->fl_end == OFFSET_MAX && 896263b4509SScott Mayhew inode->i_flock->fl_type != F_RDLCK)) 897c7559663SScott Mayhew return 1; 898c7559663SScott Mayhew return 0; 899c7559663SScott Mayhew } 900c7559663SScott Mayhew 9015d47a356STrond Myklebust /* 9021da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 9031da177e4SLinus Torvalds * 9041da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 9051da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 9061da177e4SLinus Torvalds */ 9071da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 9081da177e4SLinus Torvalds unsigned int offset, unsigned int count) 9091da177e4SLinus Torvalds { 910cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 911d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 9121da177e4SLinus Torvalds int status = 0; 9131da177e4SLinus Torvalds 91491d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 91591d5b470SChuck Lever 9166de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 9176de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 9181da177e4SLinus Torvalds 919c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 92049a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 9211da177e4SLinus Torvalds offset = 0; 9221da177e4SLinus Torvalds } 9231da177e4SLinus Torvalds 924e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 92503fa9e84STrond Myklebust if (status < 0) 92603fa9e84STrond Myklebust nfs_set_pageerror(page); 92759b7c05fSTrond Myklebust else 92859b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 9291da177e4SLinus Torvalds 93048186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 9311da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 9321da177e4SLinus Torvalds return status; 9331da177e4SLinus Torvalds } 9341da177e4SLinus Torvalds 9353ff7576dSTrond Myklebust static int flush_task_priority(int how) 9361da177e4SLinus Torvalds { 9371da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 9381da177e4SLinus Torvalds case FLUSH_HIGHPRI: 9391da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 9401da177e4SLinus Torvalds case FLUSH_LOWPRI: 9411da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 9421da177e4SLinus Torvalds } 9431da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 9441da177e4SLinus Torvalds } 9451da177e4SLinus Torvalds 9461ed26f33SAnna Schumaker static void nfs_initiate_write(struct nfs_pgio_data *data, struct rpc_message *msg, 9471ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 9481da177e4SLinus Torvalds { 949cd841605SFred Isaman struct inode *inode = data->header->inode; 9503ff7576dSTrond Myklebust int priority = flush_task_priority(how); 9511da177e4SLinus Torvalds 9521ed26f33SAnna Schumaker task_setup_data->priority = priority; 9531ed26f33SAnna Schumaker NFS_PROTO(inode)->write_setup(data, msg); 954d138d5d1SAndy Adamson 9558c21c62cSWeston Andros Adamson nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client, 9561ed26f33SAnna Schumaker &task_setup_data->rpc_client, msg, data); 957d138d5d1SAndy Adamson } 958d138d5d1SAndy Adamson 9596d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 9606d884e8fSFred * call this on each, which will prepare them to be retried on next 9616d884e8fSFred * writeback using standard nfs. 9626d884e8fSFred */ 9636d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 9646d884e8fSFred { 9656d884e8fSFred nfs_mark_request_dirty(req); 9661d1afcbcSTrond Myklebust nfs_unlock_request(req); 967d1182b33STrond Myklebust nfs_end_page_writeback(req->wb_page); 9683aff4ebbSTrond Myklebust nfs_release_request(req); 9696d884e8fSFred } 9706d884e8fSFred 971061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 9726c75dc0dSFred Isaman { 9736c75dc0dSFred Isaman struct nfs_page *req; 9746c75dc0dSFred Isaman 9756c75dc0dSFred Isaman while (!list_empty(head)) { 9766c75dc0dSFred Isaman req = nfs_list_entry(head->next); 9776c75dc0dSFred Isaman nfs_list_remove_request(req); 9786c75dc0dSFred Isaman nfs_redirty_request(req); 9796c75dc0dSFred Isaman } 9806c75dc0dSFred Isaman } 9816c75dc0dSFred Isaman 982061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 983061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 984061ae2edSFred Isaman .completion = nfs_write_completion, 985061ae2edSFred Isaman }; 986061ae2edSFred Isaman 98757208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 988a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 989061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 9901751c363STrond Myklebust { 991a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 99241d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 993a20c93e3SChristoph Hellwig 994a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 995a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 996a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 997a20c93e3SChristoph Hellwig #endif 9984a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 9994a0de55cSAnna Schumaker server->wsize, ioflags); 10001751c363STrond Myklebust } 1001ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 10021751c363STrond Myklebust 1003dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1004dce81290STrond Myklebust { 100541d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1006dce81290STrond Myklebust pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1007dce81290STrond Myklebust } 10081f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1009dce81290STrond Myklebust 10101da177e4SLinus Torvalds 10110b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 10120b7c0153SFred Isaman { 10130b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 10140b7c0153SFred Isaman 10150b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 10160b7c0153SFred Isaman } 10170b7c0153SFred Isaman 1018a4cdda59SAnna Schumaker static void nfs_writeback_release_common(struct nfs_pgio_data *data) 1019c9d8f89dSTrond Myklebust { 1020cd841605SFred Isaman struct nfs_pgio_header *hdr = data->header; 1021e2fecb21STrond Myklebust int status = data->task.tk_status; 1022788e7a89STrond Myklebust 10236c75dc0dSFred Isaman if ((status >= 0) && nfs_write_need_commit(data)) { 10246c75dc0dSFred Isaman spin_lock(&hdr->lock); 10256c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) 10266c75dc0dSFred Isaman ; /* Do nothing */ 10276c75dc0dSFred Isaman else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) 1028f79d06f5SAnna Schumaker memcpy(&hdr->verf, &data->verf, sizeof(hdr->verf)); 1029f79d06f5SAnna Schumaker else if (memcmp(&hdr->verf, &data->verf, sizeof(hdr->verf))) 10306c75dc0dSFred Isaman set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags); 10316c75dc0dSFred Isaman spin_unlock(&hdr->lock); 10321da177e4SLinus Torvalds } 10331da177e4SLinus Torvalds } 10341da177e4SLinus Torvalds 10351f2edbe3STrond Myklebust /* 10361f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 10371f2edbe3STrond Myklebust */ 10381f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 10391f2edbe3STrond Myklebust { 10401f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 10411f2edbe3STrond Myklebust int kill = 0; 10421f2edbe3STrond Myklebust 10431f2edbe3STrond Myklebust /* suid always must be killed */ 10441f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 10451f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 10461f2edbe3STrond Myklebust 10471f2edbe3STrond Myklebust /* 10481f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 10491f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 10501f2edbe3STrond Myklebust */ 10511f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 10521f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 10531f2edbe3STrond Myklebust 10541f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 10551f2edbe3STrond Myklebust return kill; 10561f2edbe3STrond Myklebust 10571f2edbe3STrond Myklebust return 0; 10581f2edbe3STrond Myklebust } 1059788e7a89STrond Myklebust 10601da177e4SLinus Torvalds /* 10611da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 10621da177e4SLinus Torvalds */ 10630eecb214SAnna Schumaker static int nfs_writeback_done(struct rpc_task *task, struct nfs_pgio_data *data, 10640eecb214SAnna Schumaker struct inode *inode) 10651da177e4SLinus Torvalds { 1066788e7a89STrond Myklebust int status; 10671da177e4SLinus Torvalds 1068f551e44fSChuck Lever /* 1069f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1070f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1071f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1072f551e44fSChuck Lever * another writer had changed the file, but some applications 1073f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1074f551e44fSChuck Lever */ 1075cd841605SFred Isaman status = NFS_PROTO(inode)->write_done(task, data); 1076788e7a89STrond Myklebust if (status != 0) 10770eecb214SAnna Schumaker return status; 10780eecb214SAnna Schumaker nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, data->res.count); 107991d5b470SChuck Lever 108089d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 10810eecb214SAnna Schumaker if (data->res.verf->committed < data->args.stable && task->tk_status >= 0) { 10821da177e4SLinus Torvalds /* We tried a write call, but the server did not 10831da177e4SLinus Torvalds * commit data to stable storage even though we 10841da177e4SLinus Torvalds * requested it. 10851da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 10861da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 10871da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 10881da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 10891da177e4SLinus Torvalds */ 10901da177e4SLinus Torvalds static unsigned long complain; 10911da177e4SLinus Torvalds 1092a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 10931da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 10941da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 10951da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1096cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 10970eecb214SAnna Schumaker data->res.verf->committed, data->args.stable); 10981da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 10991da177e4SLinus Torvalds } 11001da177e4SLinus Torvalds } 11011da177e4SLinus Torvalds #endif 11021f2edbe3STrond Myklebust 11031f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 11041f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 11051f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 11060eecb214SAnna Schumaker return 0; 11070eecb214SAnna Schumaker } 11080eecb214SAnna Schumaker 11090eecb214SAnna Schumaker /* 11100eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 11110eecb214SAnna Schumaker */ 11120eecb214SAnna Schumaker static void nfs_writeback_result(struct rpc_task *task, struct nfs_pgio_data *data) 11130eecb214SAnna Schumaker { 11140eecb214SAnna Schumaker struct nfs_pgio_args *argp = &data->args; 11150eecb214SAnna Schumaker struct nfs_pgio_res *resp = &data->res; 11161f2edbe3STrond Myklebust 11171f2edbe3STrond Myklebust if (resp->count < argp->count) { 11181da177e4SLinus Torvalds static unsigned long complain; 11191da177e4SLinus Torvalds 11206c75dc0dSFred Isaman /* This a short write! */ 11210eecb214SAnna Schumaker nfs_inc_stats(data->header->inode, NFSIOS_SHORTWRITE); 112291d5b470SChuck Lever 11231da177e4SLinus Torvalds /* Has the server at least made some progress? */ 11246c75dc0dSFred Isaman if (resp->count == 0) { 11256c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 11266c75dc0dSFred Isaman printk(KERN_WARNING 11276c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 11286c75dc0dSFred Isaman argp->count); 11296c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 11306c75dc0dSFred Isaman } 11316c75dc0dSFred Isaman nfs_set_pgio_error(data->header, -EIO, argp->offset); 11326c75dc0dSFred Isaman task->tk_status = -EIO; 11336c75dc0dSFred Isaman return; 11346c75dc0dSFred Isaman } 11351da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 11361da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 11371da177e4SLinus Torvalds /* Resend from where the server left off */ 1138a69aef14SFred Isaman data->mds_offset += resp->count; 11391da177e4SLinus Torvalds argp->offset += resp->count; 11401da177e4SLinus Torvalds argp->pgbase += resp->count; 11411da177e4SLinus Torvalds argp->count -= resp->count; 11421da177e4SLinus Torvalds } else { 11431da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 11441da177e4SLinus Torvalds * headaches in the case of a server crash. 11451da177e4SLinus Torvalds */ 11461da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 11471da177e4SLinus Torvalds } 1148d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 11491da177e4SLinus Torvalds } 11501da177e4SLinus Torvalds } 11511da177e4SLinus Torvalds 11521da177e4SLinus Torvalds 115389d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 115471d0a611STrond Myklebust static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) 115571d0a611STrond Myklebust { 1156b8413f98STrond Myklebust int ret; 1157b8413f98STrond Myklebust 115871d0a611STrond Myklebust if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) 115971d0a611STrond Myklebust return 1; 1160b8413f98STrond Myklebust if (!may_wait) 116171d0a611STrond Myklebust return 0; 1162b8413f98STrond Myklebust ret = out_of_line_wait_on_bit_lock(&nfsi->flags, 1163b8413f98STrond Myklebust NFS_INO_COMMIT, 1164b8413f98STrond Myklebust nfs_wait_bit_killable, 1165b8413f98STrond Myklebust TASK_KILLABLE); 1166b8413f98STrond Myklebust return (ret < 0) ? ret : 1; 116771d0a611STrond Myklebust } 116871d0a611STrond Myklebust 1169f453a54aSFred Isaman static void nfs_commit_clear_lock(struct nfs_inode *nfsi) 117071d0a611STrond Myklebust { 117171d0a611STrond Myklebust clear_bit(NFS_INO_COMMIT, &nfsi->flags); 117271d0a611STrond Myklebust smp_mb__after_clear_bit(); 117371d0a611STrond Myklebust wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); 117471d0a611STrond Myklebust } 117571d0a611STrond Myklebust 11760b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 11771da177e4SLinus Torvalds { 11780b7c0153SFred Isaman put_nfs_open_context(data->context); 11790b7c0153SFred Isaman nfs_commit_free(data); 11801da177e4SLinus Torvalds } 1181e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 11821da177e4SLinus Torvalds 11830b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 11849ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 11859f0ec176SAndy Adamson int how, int flags) 11861da177e4SLinus Torvalds { 118707737691STrond Myklebust struct rpc_task *task; 11889ace33cdSFred Isaman int priority = flush_task_priority(how); 1189bdc7f021STrond Myklebust struct rpc_message msg = { 1190bdc7f021STrond Myklebust .rpc_argp = &data->args, 1191bdc7f021STrond Myklebust .rpc_resp = &data->res, 11929ace33cdSFred Isaman .rpc_cred = data->cred, 1193bdc7f021STrond Myklebust }; 119484115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 119507737691STrond Myklebust .task = &data->task, 11969ace33cdSFred Isaman .rpc_client = clnt, 1197bdc7f021STrond Myklebust .rpc_message = &msg, 11989ace33cdSFred Isaman .callback_ops = call_ops, 119984115e1cSTrond Myklebust .callback_data = data, 1200101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 12019f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 12023ff7576dSTrond Myklebust .priority = priority, 120384115e1cSTrond Myklebust }; 1204788e7a89STrond Myklebust /* Set up the initial task struct. */ 12059ace33cdSFred Isaman NFS_PROTO(data->inode)->commit_setup(data, &msg); 12061da177e4SLinus Torvalds 1207a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1208bdc7f021STrond Myklebust 12098c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 12108c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 12118c21c62cSWeston Andros Adamson 121207737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1213dbae4c73STrond Myklebust if (IS_ERR(task)) 1214dbae4c73STrond Myklebust return PTR_ERR(task); 1215d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1216d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 121707737691STrond Myklebust rpc_put_task(task); 1218dbae4c73STrond Myklebust return 0; 12191da177e4SLinus Torvalds } 1220e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 12211da177e4SLinus Torvalds 12221da177e4SLinus Torvalds /* 12239ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 12249ace33cdSFred Isaman */ 12250b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1226988b6dceSFred Isaman struct list_head *head, 1227f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1228f453a54aSFred Isaman struct nfs_commit_info *cinfo) 12299ace33cdSFred Isaman { 12309ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 12313d4ff43dSAl Viro struct inode *inode = first->wb_context->dentry->d_inode; 12329ace33cdSFred Isaman 12339ace33cdSFred Isaman /* Set up the RPC argument and reply structs 12349ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 12359ace33cdSFred Isaman 12369ace33cdSFred Isaman list_splice_init(head, &data->pages); 12379ace33cdSFred Isaman 12389ace33cdSFred Isaman data->inode = inode; 12399ace33cdSFred Isaman data->cred = first->wb_context->cred; 1240988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 12419ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1242f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1243b359f9d0SFred Isaman data->dreq = cinfo->dreq; 12449ace33cdSFred Isaman 12459ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 12469ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 12479ace33cdSFred Isaman data->args.offset = 0; 12489ace33cdSFred Isaman data->args.count = 0; 12490b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 12509ace33cdSFred Isaman data->res.fattr = &data->fattr; 12519ace33cdSFred Isaman data->res.verf = &data->verf; 12529ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 12539ace33cdSFred Isaman } 1254e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 12559ace33cdSFred Isaman 1256e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1257ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1258ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 125964bfeb49SFred Isaman { 126064bfeb49SFred Isaman struct nfs_page *req; 126164bfeb49SFred Isaman 126264bfeb49SFred Isaman while (!list_empty(page_list)) { 126364bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 126464bfeb49SFred Isaman nfs_list_remove_request(req); 1265ea2cf228SFred Isaman nfs_mark_request_commit(req, lseg, cinfo); 126656f9cd68SFred Isaman if (!cinfo->dreq) { 126764bfeb49SFred Isaman dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1268d56b4ddfSMel Gorman dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, 126964bfeb49SFred Isaman BDI_RECLAIMABLE); 127056f9cd68SFred Isaman } 12711d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 127264bfeb49SFred Isaman } 127364bfeb49SFred Isaman } 1274e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 127564bfeb49SFred Isaman 12769ace33cdSFred Isaman /* 12771da177e4SLinus Torvalds * Commit dirty pages 12781da177e4SLinus Torvalds */ 12791da177e4SLinus Torvalds static int 1280ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1281ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 12821da177e4SLinus Torvalds { 12830b7c0153SFred Isaman struct nfs_commit_data *data; 12841da177e4SLinus Torvalds 1285c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 12861da177e4SLinus Torvalds 12871da177e4SLinus Torvalds if (!data) 12881da177e4SLinus Torvalds goto out_bad; 12891da177e4SLinus Torvalds 12901da177e4SLinus Torvalds /* Set up the argument struct */ 1291f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1292f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 12939f0ec176SAndy Adamson return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, 12949f0ec176SAndy Adamson how, 0); 12951da177e4SLinus Torvalds out_bad: 1296ea2cf228SFred Isaman nfs_retry_commit(head, NULL, cinfo); 1297f453a54aSFred Isaman cinfo->completion_ops->error_cleanup(NFS_I(inode)); 12981da177e4SLinus Torvalds return -ENOMEM; 12991da177e4SLinus Torvalds } 13001da177e4SLinus Torvalds 13011da177e4SLinus Torvalds /* 13021da177e4SLinus Torvalds * COMMIT call returned 13031da177e4SLinus Torvalds */ 1304788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 13051da177e4SLinus Torvalds { 13060b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 13071da177e4SLinus Torvalds 1308a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 13091da177e4SLinus Torvalds task->tk_pid, task->tk_status); 13101da177e4SLinus Torvalds 1311788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1312c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1313c9d8f89dSTrond Myklebust } 1314c9d8f89dSTrond Myklebust 1315f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1316c9d8f89dSTrond Myklebust { 1317c9d8f89dSTrond Myklebust struct nfs_page *req; 1318c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1319f453a54aSFred Isaman struct nfs_commit_info cinfo; 1320788e7a89STrond Myklebust 13211da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 13221da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 13231da177e4SLinus Torvalds nfs_list_remove_request(req); 1324d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 13251da177e4SLinus Torvalds 13261e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 13273d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 13281e8968c5SNiels de Vos (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode), 13291da177e4SLinus Torvalds req->wb_bytes, 13301da177e4SLinus Torvalds (long long)req_offset(req)); 1331c9d8f89dSTrond Myklebust if (status < 0) { 1332c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 13331da177e4SLinus Torvalds nfs_inode_remove_request(req); 1334c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 13351da177e4SLinus Torvalds goto next; 13361da177e4SLinus Torvalds } 13371da177e4SLinus Torvalds 13381da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 13391da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 13402f2c63bcSTrond Myklebust if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { 13411da177e4SLinus Torvalds /* We have a match */ 13421da177e4SLinus Torvalds nfs_inode_remove_request(req); 13431da177e4SLinus Torvalds dprintk(" OK\n"); 13441da177e4SLinus Torvalds goto next; 13451da177e4SLinus Torvalds } 13461da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 13471da177e4SLinus Torvalds dprintk(" mismatch\n"); 13486d884e8fSFred nfs_mark_request_dirty(req); 134905990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 13501da177e4SLinus Torvalds next: 13511d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 13521da177e4SLinus Torvalds } 1353f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1354f453a54aSFred Isaman if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) 1355f453a54aSFred Isaman nfs_commit_clear_lock(NFS_I(data->inode)); 13565917ce84SFred Isaman } 13575917ce84SFred Isaman 13585917ce84SFred Isaman static void nfs_commit_release(void *calldata) 13595917ce84SFred Isaman { 13600b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 13615917ce84SFred Isaman 1362f453a54aSFred Isaman data->completion_ops->completion(data); 1363c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 13641da177e4SLinus Torvalds } 1365788e7a89STrond Myklebust 1366788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 13670b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1368788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1369788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1370788e7a89STrond Myklebust }; 13711da177e4SLinus Torvalds 1372f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1373f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1374f453a54aSFred Isaman .error_cleanup = nfs_commit_clear_lock, 1375f453a54aSFred Isaman }; 1376f453a54aSFred Isaman 13771763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1378ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 137984c53ab5SFred Isaman { 138084c53ab5SFred Isaman int status; 138184c53ab5SFred Isaman 1382ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 138384c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1384ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 138584c53ab5SFred Isaman return status; 138684c53ab5SFred Isaman } 138784c53ab5SFred Isaman 1388b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 13891da177e4SLinus Torvalds { 13901da177e4SLinus Torvalds LIST_HEAD(head); 1391ea2cf228SFred Isaman struct nfs_commit_info cinfo; 139271d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1393b8413f98STrond Myklebust int res; 13941da177e4SLinus Torvalds 1395b8413f98STrond Myklebust res = nfs_commit_set_lock(NFS_I(inode), may_wait); 1396b8413f98STrond Myklebust if (res <= 0) 1397c5efa5fcSTrond Myklebust goto out_mark_dirty; 1398ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1399ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 14001da177e4SLinus Torvalds if (res) { 1401a861a1e1SFred Isaman int error; 1402a861a1e1SFred Isaman 1403ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 14041da177e4SLinus Torvalds if (error < 0) 14051da177e4SLinus Torvalds return error; 1406b8413f98STrond Myklebust if (!may_wait) 1407b8413f98STrond Myklebust goto out_mark_dirty; 1408b8413f98STrond Myklebust error = wait_on_bit(&NFS_I(inode)->flags, 1409b8413f98STrond Myklebust NFS_INO_COMMIT, 141071d0a611STrond Myklebust nfs_wait_bit_killable, 141171d0a611STrond Myklebust TASK_KILLABLE); 1412b8413f98STrond Myklebust if (error < 0) 1413b8413f98STrond Myklebust return error; 141471d0a611STrond Myklebust } else 141571d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(inode)); 1416c5efa5fcSTrond Myklebust return res; 1417c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1418c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1419c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1420c5efa5fcSTrond Myklebust * that the data is on the disk. 1421c5efa5fcSTrond Myklebust */ 1422c5efa5fcSTrond Myklebust out_mark_dirty: 1423c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 14241da177e4SLinus Torvalds return res; 14251da177e4SLinus Torvalds } 14268fc795f7STrond Myklebust 14278fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 14288fc795f7STrond Myklebust { 1429420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1430420e3646STrond Myklebust int flags = FLUSH_SYNC; 1431420e3646STrond Myklebust int ret = 0; 14328fc795f7STrond Myklebust 14333236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1434ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 14353236c3e1SJeff Layton return ret; 14363236c3e1SJeff Layton 1437a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1438a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1439a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1440420e3646STrond Myklebust */ 1441ea2cf228SFred Isaman if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1)) 1442420e3646STrond Myklebust goto out_mark_dirty; 1443420e3646STrond Myklebust 1444a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1445420e3646STrond Myklebust flags = 0; 1446a00dd6c0SJeff Layton } 1447a00dd6c0SJeff Layton 1448420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1449420e3646STrond Myklebust if (ret >= 0) { 1450420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1451420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1452420e3646STrond Myklebust wbc->nr_to_write -= ret; 1453420e3646STrond Myklebust else 1454420e3646STrond Myklebust wbc->nr_to_write = 0; 1455420e3646STrond Myklebust } 14568fc795f7STrond Myklebust return 0; 1457420e3646STrond Myklebust } 1458420e3646STrond Myklebust out_mark_dirty: 14598fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 14608fc795f7STrond Myklebust return ret; 14618fc795f7STrond Myklebust } 1462c63c7b05STrond Myklebust #else 14638fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 14648fc795f7STrond Myklebust { 14658fc795f7STrond Myklebust return 0; 14668fc795f7STrond Myklebust } 14671da177e4SLinus Torvalds #endif 14681da177e4SLinus Torvalds 14698fc795f7STrond Myklebust int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 14708fc795f7STrond Myklebust { 1471a8d8f02cSBryan Schumaker return nfs_commit_unstable_pages(inode, wbc); 1472a8d8f02cSBryan Schumaker } 147389d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1474863a3c6cSAndy Adamson 1475acdc53b2STrond Myklebust /* 1476acdc53b2STrond Myklebust * flush the inode to disk. 1477acdc53b2STrond Myklebust */ 1478acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 147934901f70STrond Myklebust { 148034901f70STrond Myklebust struct writeback_control wbc = { 148172cb77f4STrond Myklebust .sync_mode = WB_SYNC_ALL, 148234901f70STrond Myklebust .nr_to_write = LONG_MAX, 1483d7fb1207STrond Myklebust .range_start = 0, 1484d7fb1207STrond Myklebust .range_end = LLONG_MAX, 148534901f70STrond Myklebust }; 1486f4ce1299STrond Myklebust int ret; 148734901f70STrond Myklebust 1488f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1489f4ce1299STrond Myklebust 1490f4ce1299STrond Myklebust ret = sync_inode(inode, &wbc); 1491f4ce1299STrond Myklebust 1492f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1493f4ce1299STrond Myklebust return ret; 14941c75950bSTrond Myklebust } 1495ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 14961c75950bSTrond Myklebust 14971b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 14981b3b4a1aSTrond Myklebust { 14991b3b4a1aSTrond Myklebust struct nfs_page *req; 15001b3b4a1aSTrond Myklebust int ret = 0; 15011b3b4a1aSTrond Myklebust 15021b3b4a1aSTrond Myklebust for (;;) { 1503ba8b06e6STrond Myklebust wait_on_page_writeback(page); 15041b3b4a1aSTrond Myklebust req = nfs_page_find_request(page); 15051b3b4a1aSTrond Myklebust if (req == NULL) 15061b3b4a1aSTrond Myklebust break; 15077ad84aa9STrond Myklebust if (nfs_lock_request(req)) { 15088dd37758STrond Myklebust nfs_clear_request_commit(req); 15091b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 15101b3b4a1aSTrond Myklebust /* 15111b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 15121b3b4a1aSTrond Myklebust * page as being dirty 15131b3b4a1aSTrond Myklebust */ 15141b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 15151d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 15161b3b4a1aSTrond Myklebust break; 15171b3b4a1aSTrond Myklebust } 15181b3b4a1aSTrond Myklebust ret = nfs_wait_on_request(req); 1519c9edda71STrond Myklebust nfs_release_request(req); 15201b3b4a1aSTrond Myklebust if (ret < 0) 1521c988950eSTrond Myklebust break; 15221b3b4a1aSTrond Myklebust } 15231b3b4a1aSTrond Myklebust return ret; 15241b3b4a1aSTrond Myklebust } 15251b3b4a1aSTrond Myklebust 15261c75950bSTrond Myklebust /* 15271c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 15281c75950bSTrond Myklebust */ 15291c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 15301c75950bSTrond Myklebust { 153129418aa4SMel Gorman loff_t range_start = page_file_offset(page); 15327f2f12d9STrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 15337f2f12d9STrond Myklebust struct writeback_control wbc = { 15347f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 15357f2f12d9STrond Myklebust .nr_to_write = 0, 15367f2f12d9STrond Myklebust .range_start = range_start, 15377f2f12d9STrond Myklebust .range_end = range_end, 15387f2f12d9STrond Myklebust }; 15397f2f12d9STrond Myklebust int ret; 15407f2f12d9STrond Myklebust 1541f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 1542f4ce1299STrond Myklebust 15430522f6adSTrond Myklebust for (;;) { 1544ba8b06e6STrond Myklebust wait_on_page_writeback(page); 15457f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 15467f2f12d9STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 15477f2f12d9STrond Myklebust if (ret < 0) 15487f2f12d9STrond Myklebust goto out_error; 15490522f6adSTrond Myklebust continue; 15507f2f12d9STrond Myklebust } 1551f4ce1299STrond Myklebust ret = 0; 15520522f6adSTrond Myklebust if (!PagePrivate(page)) 15530522f6adSTrond Myklebust break; 15540522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 15557f2f12d9STrond Myklebust if (ret < 0) 15567f2f12d9STrond Myklebust goto out_error; 15577f2f12d9STrond Myklebust } 15587f2f12d9STrond Myklebust out_error: 1559f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 15607f2f12d9STrond Myklebust return ret; 15611c75950bSTrond Myklebust } 15621c75950bSTrond Myklebust 1563074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 1564074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 1565a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 1566074cc1deSTrond Myklebust { 15672da95652SJeff Layton /* 15682da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 15692da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 15702da95652SJeff Layton * 15712da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 15722da95652SJeff Layton * that we can safely release the inode reference while holding 15732da95652SJeff Layton * the page lock. 15742da95652SJeff Layton */ 15752da95652SJeff Layton if (PagePrivate(page)) 15762da95652SJeff Layton return -EBUSY; 1577074cc1deSTrond Myklebust 15788c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 15798c209ce7SDavid Howells return -EBUSY; 1580074cc1deSTrond Myklebust 1581a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 1582074cc1deSTrond Myklebust } 1583074cc1deSTrond Myklebust #endif 1584074cc1deSTrond Myklebust 1585f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 15861da177e4SLinus Torvalds { 15871da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 1588c0752cdfSAnna Schumaker sizeof(struct nfs_rw_header), 15891da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 159020c2df83SPaul Mundt NULL); 15911da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 15921da177e4SLinus Torvalds return -ENOMEM; 15931da177e4SLinus Torvalds 159493d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 15951da177e4SLinus Torvalds nfs_wdata_cachep); 15961da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 15973dd4765fSJeff Layton goto out_destroy_write_cache; 15981da177e4SLinus Torvalds 15990b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 16000b7c0153SFred Isaman sizeof(struct nfs_commit_data), 16010b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 16020b7c0153SFred Isaman NULL); 16030b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 16043dd4765fSJeff Layton goto out_destroy_write_mempool; 16050b7c0153SFred Isaman 160693d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 16074c100210SYanchuan Nian nfs_cdata_cachep); 16081da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 16093dd4765fSJeff Layton goto out_destroy_commit_cache; 16101da177e4SLinus Torvalds 161189a09141SPeter Zijlstra /* 161289a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 161389a09141SPeter Zijlstra * 161489a09141SPeter Zijlstra * 64MB: 8192k 161589a09141SPeter Zijlstra * 128MB: 11585k 161689a09141SPeter Zijlstra * 256MB: 16384k 161789a09141SPeter Zijlstra * 512MB: 23170k 161889a09141SPeter Zijlstra * 1GB: 32768k 161989a09141SPeter Zijlstra * 2GB: 46340k 162089a09141SPeter Zijlstra * 4GB: 65536k 162189a09141SPeter Zijlstra * 8GB: 92681k 162289a09141SPeter Zijlstra * 16GB: 131072k 162389a09141SPeter Zijlstra * 162489a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 162589a09141SPeter Zijlstra * Limit the default to 256M 162689a09141SPeter Zijlstra */ 162789a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 162889a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 162989a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 163089a09141SPeter Zijlstra 16311da177e4SLinus Torvalds return 0; 16323dd4765fSJeff Layton 16333dd4765fSJeff Layton out_destroy_commit_cache: 16343dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 16353dd4765fSJeff Layton out_destroy_write_mempool: 16363dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 16373dd4765fSJeff Layton out_destroy_write_cache: 16383dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 16393dd4765fSJeff Layton return -ENOMEM; 16401da177e4SLinus Torvalds } 16411da177e4SLinus Torvalds 1642266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 16431da177e4SLinus Torvalds { 16441da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 16453dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 16461da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 16471a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 16481da177e4SLinus Torvalds } 16491da177e4SLinus Torvalds 16504a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 1651a4cdda59SAnna Schumaker .rw_mode = FMODE_WRITE, 16524a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 16534a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 1654a4cdda59SAnna Schumaker .rw_release = nfs_writeback_release_common, 16550eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 16560eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 16571ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 16584a0de55cSAnna Schumaker }; 1659