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> 161da177e4SLinus Torvalds 171da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h> 181da177e4SLinus Torvalds #include <linux/nfs_fs.h> 191da177e4SLinus Torvalds #include <linux/nfs_mount.h> 201da177e4SLinus Torvalds #include <linux/nfs_page.h> 213fcfab16SAndrew Morton #include <linux/backing-dev.h> 223fcfab16SAndrew Morton 231da177e4SLinus Torvalds #include <asm/uaccess.h> 241da177e4SLinus Torvalds #include <linux/smp_lock.h> 251da177e4SLinus Torvalds 261da177e4SLinus Torvalds #include "delegation.h" 2749a70f27STrond Myklebust #include "internal.h" 2891d5b470SChuck Lever #include "iostat.h" 291da177e4SLinus Torvalds 301da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 311da177e4SLinus Torvalds 321da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 331da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 341da177e4SLinus Torvalds 351da177e4SLinus Torvalds /* 361da177e4SLinus Torvalds * Local function declarations 371da177e4SLinus Torvalds */ 381da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context*, 391da177e4SLinus Torvalds struct page *, 401da177e4SLinus Torvalds unsigned int, unsigned int); 413f442547STrond Myklebust static long nfs_flush_mapping(struct address_space *mapping, struct writeback_control *wbc, int how); 42788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops; 43788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops; 44788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 451da177e4SLinus Torvalds 46e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 473feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 481da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 491da177e4SLinus Torvalds 50e9f7bee1STrond Myklebust struct nfs_write_data *nfs_commit_alloc(void) 511da177e4SLinus Torvalds { 52e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS); 5340859d7eSChuck Lever 541da177e4SLinus Torvalds if (p) { 551da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 561da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 571da177e4SLinus Torvalds } 581da177e4SLinus Torvalds return p; 591da177e4SLinus Torvalds } 601da177e4SLinus Torvalds 618aca67f0STrond Myklebust void nfs_commit_rcu_free(struct rcu_head *head) 621da177e4SLinus Torvalds { 638aca67f0STrond Myklebust struct nfs_write_data *p = container_of(head, struct nfs_write_data, task.u.tk_rcu); 6440859d7eSChuck Lever if (p && (p->pagevec != &p->page_array[0])) 6540859d7eSChuck Lever kfree(p->pagevec); 661da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 671da177e4SLinus Torvalds } 681da177e4SLinus Torvalds 698aca67f0STrond Myklebust void nfs_commit_free(struct nfs_write_data *wdata) 708aca67f0STrond Myklebust { 718aca67f0STrond Myklebust call_rcu_bh(&wdata->task.u.tk_rcu, nfs_commit_rcu_free); 728aca67f0STrond Myklebust } 738aca67f0STrond Myklebust 74e9f7bee1STrond Myklebust struct nfs_write_data *nfs_writedata_alloc(size_t len) 753feb2d49STrond Myklebust { 76e9f7bee1STrond Myklebust unsigned int pagecount = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 77e6b4f8daSChristoph Lameter struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS); 783feb2d49STrond Myklebust 793feb2d49STrond Myklebust if (p) { 803feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 813feb2d49STrond Myklebust INIT_LIST_HEAD(&p->pages); 82e9f7bee1STrond Myklebust p->npages = pagecount; 830d0b5cb3SChuck Lever if (pagecount <= ARRAY_SIZE(p->page_array)) 840d0b5cb3SChuck Lever p->pagevec = p->page_array; 853feb2d49STrond Myklebust else { 860d0b5cb3SChuck Lever p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS); 870d0b5cb3SChuck Lever if (!p->pagevec) { 883feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 893feb2d49STrond Myklebust p = NULL; 903feb2d49STrond Myklebust } 913feb2d49STrond Myklebust } 923feb2d49STrond Myklebust } 933feb2d49STrond Myklebust return p; 943feb2d49STrond Myklebust } 953feb2d49STrond Myklebust 968aca67f0STrond Myklebust static void nfs_writedata_rcu_free(struct rcu_head *head) 973feb2d49STrond Myklebust { 988aca67f0STrond Myklebust struct nfs_write_data *p = container_of(head, struct nfs_write_data, task.u.tk_rcu); 993feb2d49STrond Myklebust if (p && (p->pagevec != &p->page_array[0])) 1003feb2d49STrond Myklebust kfree(p->pagevec); 1013feb2d49STrond Myklebust mempool_free(p, nfs_wdata_mempool); 1023feb2d49STrond Myklebust } 1033feb2d49STrond Myklebust 1048aca67f0STrond Myklebust static void nfs_writedata_free(struct nfs_write_data *wdata) 1058aca67f0STrond Myklebust { 1068aca67f0STrond Myklebust call_rcu_bh(&wdata->task.u.tk_rcu, nfs_writedata_rcu_free); 1078aca67f0STrond Myklebust } 1088aca67f0STrond Myklebust 109963d8fe5STrond Myklebust void nfs_writedata_release(void *wdata) 1101da177e4SLinus Torvalds { 1111da177e4SLinus Torvalds nfs_writedata_free(wdata); 1121da177e4SLinus Torvalds } 1131da177e4SLinus Torvalds 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) 121277459d2STrond Myklebust atomic_inc(&req->wb_count); 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 { 128277459d2STrond Myklebust struct nfs_page *req = NULL; 129277459d2STrond Myklebust spinlock_t *req_lock = &NFS_I(page->mapping->host)->req_lock; 130277459d2STrond Myklebust 131277459d2STrond Myklebust spin_lock(req_lock); 132277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 133277459d2STrond Myklebust spin_unlock(req_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; 1411da177e4SLinus Torvalds loff_t end, i_size = i_size_read(inode); 1421da177e4SLinus Torvalds unsigned long end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 1431da177e4SLinus Torvalds 1441da177e4SLinus Torvalds if (i_size > 0 && page->index < end_index) 1451da177e4SLinus Torvalds return; 1461da177e4SLinus Torvalds end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count); 1471da177e4SLinus Torvalds if (i_size >= end) 1481da177e4SLinus Torvalds return; 14991d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 1501da177e4SLinus Torvalds i_size_write(inode, end); 1511da177e4SLinus Torvalds } 1521da177e4SLinus Torvalds 153a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 154a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 155a301b777STrond Myklebust { 156a301b777STrond Myklebust SetPageError(page); 157a301b777STrond Myklebust nfs_zap_mapping(page->mapping->host, page->mapping); 158a301b777STrond Myklebust } 159a301b777STrond Myklebust 1601da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 1611da177e4SLinus Torvalds * covers the full page. 1621da177e4SLinus Torvalds */ 1631da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) 1641da177e4SLinus Torvalds { 1651da177e4SLinus Torvalds if (PageUptodate(page)) 1661da177e4SLinus Torvalds return; 1671da177e4SLinus Torvalds if (base != 0) 1681da177e4SLinus Torvalds return; 16949a70f27STrond Myklebust if (count != nfs_page_length(page)) 1701da177e4SLinus Torvalds return; 17149a70f27STrond Myklebust if (count != PAGE_CACHE_SIZE) 1721da177e4SLinus Torvalds memclear_highpage_flush(page, count, PAGE_CACHE_SIZE - count); 1731da177e4SLinus Torvalds SetPageUptodate(page); 1741da177e4SLinus Torvalds } 1751da177e4SLinus Torvalds 176e21195a7STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1771da177e4SLinus Torvalds unsigned int offset, unsigned int count) 1781da177e4SLinus Torvalds { 1791da177e4SLinus Torvalds struct nfs_page *req; 180e21195a7STrond Myklebust int ret; 1811da177e4SLinus Torvalds 182e21195a7STrond Myklebust for (;;) { 183e21195a7STrond Myklebust req = nfs_update_request(ctx, page, offset, count); 184e21195a7STrond Myklebust if (!IS_ERR(req)) 185e21195a7STrond Myklebust break; 186e21195a7STrond Myklebust ret = PTR_ERR(req); 187e21195a7STrond Myklebust if (ret != -EBUSY) 188e21195a7STrond Myklebust return ret; 189e21195a7STrond Myklebust ret = nfs_wb_page(page->mapping->host, page); 190e21195a7STrond Myklebust if (ret != 0) 191e21195a7STrond Myklebust return ret; 192e21195a7STrond Myklebust } 1931da177e4SLinus Torvalds /* Update file length */ 1941da177e4SLinus Torvalds nfs_grow_file(page, offset, count); 1951da177e4SLinus Torvalds /* Set the PG_uptodate flag? */ 1961da177e4SLinus Torvalds nfs_mark_uptodate(page, offset, count); 1971da177e4SLinus Torvalds nfs_unlock_request(req); 198abd3e641STrond Myklebust return 0; 1991da177e4SLinus Torvalds } 2001da177e4SLinus Torvalds 2011da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 2021da177e4SLinus Torvalds { 2031da177e4SLinus Torvalds if (wbc->for_reclaim) 2041da177e4SLinus Torvalds return FLUSH_HIGHPRI; 2051da177e4SLinus Torvalds if (wbc->for_kupdate) 2061da177e4SLinus Torvalds return FLUSH_LOWPRI; 2071da177e4SLinus Torvalds return 0; 2081da177e4SLinus Torvalds } 2091da177e4SLinus Torvalds 2101da177e4SLinus Torvalds /* 21189a09141SPeter Zijlstra * NFS congestion control 21289a09141SPeter Zijlstra */ 21389a09141SPeter Zijlstra 21489a09141SPeter Zijlstra int nfs_congestion_kb; 21589a09141SPeter Zijlstra 21689a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 21789a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 21889a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 21989a09141SPeter Zijlstra 2205a6d41b3STrond Myklebust static int nfs_set_page_writeback(struct page *page) 22189a09141SPeter Zijlstra { 2225a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2235a6d41b3STrond Myklebust 2245a6d41b3STrond Myklebust if (!ret) { 22589a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 22689a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 22789a09141SPeter Zijlstra 22889a09141SPeter Zijlstra if (atomic_inc_return(&nfss->writeback) > 22989a09141SPeter Zijlstra NFS_CONGESTION_ON_THRESH) 23089a09141SPeter Zijlstra set_bdi_congested(&nfss->backing_dev_info, WRITE); 23189a09141SPeter Zijlstra } 2325a6d41b3STrond Myklebust return ret; 23389a09141SPeter Zijlstra } 23489a09141SPeter Zijlstra 23589a09141SPeter Zijlstra static void nfs_end_page_writeback(struct page *page) 23689a09141SPeter Zijlstra { 23789a09141SPeter Zijlstra struct inode *inode = page->mapping->host; 23889a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 23989a09141SPeter Zijlstra 24089a09141SPeter Zijlstra end_page_writeback(page); 24189a09141SPeter Zijlstra if (atomic_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) { 24289a09141SPeter Zijlstra clear_bdi_congested(&nfss->backing_dev_info, WRITE); 24389a09141SPeter Zijlstra congestion_end(WRITE); 24489a09141SPeter Zijlstra } 24589a09141SPeter Zijlstra } 24689a09141SPeter Zijlstra 24789a09141SPeter Zijlstra /* 248e261f51fSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 249e261f51fSTrond Myklebust * Returns 1 if there was no write request, or if the request was 250e261f51fSTrond Myklebust * already tagged by nfs_set_page_dirty.Returns 0 if the request 251e261f51fSTrond Myklebust * was not tagged. 252e261f51fSTrond Myklebust * May also return an error if the user signalled nfs_wait_on_request(). 253e261f51fSTrond Myklebust */ 254e261f51fSTrond Myklebust static int nfs_page_mark_flush(struct page *page) 255e261f51fSTrond Myklebust { 256e261f51fSTrond Myklebust struct nfs_page *req; 257612c9384STrond Myklebust struct nfs_inode *nfsi = NFS_I(page->mapping->host); 258612c9384STrond Myklebust spinlock_t *req_lock = &nfsi->req_lock; 259e261f51fSTrond Myklebust int ret; 260e261f51fSTrond Myklebust 261e261f51fSTrond Myklebust spin_lock(req_lock); 262e261f51fSTrond Myklebust for(;;) { 263e261f51fSTrond Myklebust req = nfs_page_find_request_locked(page); 264e261f51fSTrond Myklebust if (req == NULL) { 265e261f51fSTrond Myklebust spin_unlock(req_lock); 266e261f51fSTrond Myklebust return 1; 267e261f51fSTrond Myklebust } 268e261f51fSTrond Myklebust if (nfs_lock_request_dontget(req)) 269e261f51fSTrond Myklebust break; 270e261f51fSTrond Myklebust /* Note: If we hold the page lock, as is the case in nfs_writepage, 271e261f51fSTrond Myklebust * then the call to nfs_lock_request_dontget() will always 272e261f51fSTrond Myklebust * succeed provided that someone hasn't already marked the 273e261f51fSTrond Myklebust * request as dirty (in which case we don't care). 274e261f51fSTrond Myklebust */ 275e261f51fSTrond Myklebust spin_unlock(req_lock); 276e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 277e261f51fSTrond Myklebust nfs_release_request(req); 278e261f51fSTrond Myklebust if (ret != 0) 279e261f51fSTrond Myklebust return ret; 280e261f51fSTrond Myklebust spin_lock(req_lock); 281e261f51fSTrond Myklebust } 282612c9384STrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 283612c9384STrond Myklebust /* This request is marked for commit */ 284e261f51fSTrond Myklebust spin_unlock(req_lock); 285612c9384STrond Myklebust nfs_unlock_request(req); 286612c9384STrond Myklebust return 1; 287612c9384STrond Myklebust } 288eb4cac10STrond Myklebust if (nfs_set_page_writeback(page) == 0) { 289eb4cac10STrond Myklebust nfs_list_remove_request(req); 290612c9384STrond Myklebust /* add the request to the inode's dirty list. */ 291612c9384STrond Myklebust radix_tree_tag_set(&nfsi->nfs_page_tree, 292612c9384STrond Myklebust req->wb_index, NFS_PAGE_TAG_DIRTY); 293612c9384STrond Myklebust nfs_list_add_request(req, &nfsi->dirty); 294612c9384STrond Myklebust nfsi->ndirty++; 295612c9384STrond Myklebust spin_unlock(req_lock); 296612c9384STrond Myklebust __mark_inode_dirty(page->mapping->host, I_DIRTY_PAGES); 297612c9384STrond Myklebust } else 298612c9384STrond Myklebust spin_unlock(req_lock); 299e261f51fSTrond Myklebust ret = test_bit(PG_NEED_FLUSH, &req->wb_flags); 300e261f51fSTrond Myklebust nfs_unlock_request(req); 301e261f51fSTrond Myklebust return ret; 302e261f51fSTrond Myklebust } 303e261f51fSTrond Myklebust 304e261f51fSTrond Myklebust /* 3051da177e4SLinus Torvalds * Write an mmapped page to the server. 3061da177e4SLinus Torvalds */ 3074d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 3081da177e4SLinus Torvalds { 3091da177e4SLinus Torvalds struct nfs_open_context *ctx; 3101da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 31149a70f27STrond Myklebust unsigned offset; 312e261f51fSTrond Myklebust int err; 3131da177e4SLinus Torvalds 31491d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 31591d5b470SChuck Lever nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 31691d5b470SChuck Lever 317e261f51fSTrond Myklebust err = nfs_page_mark_flush(page); 318e261f51fSTrond Myklebust if (err <= 0) 3191a54533eSTrond Myklebust goto out; 320e261f51fSTrond Myklebust err = 0; 32149a70f27STrond Myklebust offset = nfs_page_length(page); 32249a70f27STrond Myklebust if (!offset) 3231da177e4SLinus Torvalds goto out; 32449a70f27STrond Myklebust 325d530838bSTrond Myklebust ctx = nfs_find_open_context(inode, NULL, FMODE_WRITE); 3261da177e4SLinus Torvalds if (ctx == NULL) { 3271da177e4SLinus Torvalds err = -EBADF; 3281da177e4SLinus Torvalds goto out; 3291da177e4SLinus Torvalds } 330e21195a7STrond Myklebust err = nfs_writepage_setup(ctx, page, 0, offset); 331200baa21STrond Myklebust put_nfs_open_context(ctx); 332e261f51fSTrond Myklebust if (err != 0) 333e261f51fSTrond Myklebust goto out; 334e261f51fSTrond Myklebust err = nfs_page_mark_flush(page); 335e261f51fSTrond Myklebust if (err > 0) 336e261f51fSTrond Myklebust err = 0; 337200baa21STrond Myklebust out: 338abd3e641STrond Myklebust if (!wbc->for_writepages) 33902241bc4STrond Myklebust nfs_flush_mapping(page->mapping, wbc, FLUSH_STABLE|wb_priority(wbc)); 3404d770ccfSTrond Myklebust return err; 3414d770ccfSTrond Myklebust } 3424d770ccfSTrond Myklebust 3434d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 3444d770ccfSTrond Myklebust { 3454d770ccfSTrond Myklebust int err; 3464d770ccfSTrond Myklebust 3474d770ccfSTrond Myklebust err = nfs_writepage_locked(page, wbc); 3481da177e4SLinus Torvalds unlock_page(page); 3491da177e4SLinus Torvalds return err; 3501da177e4SLinus Torvalds } 3511da177e4SLinus Torvalds 3521da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 3531da177e4SLinus Torvalds { 3541da177e4SLinus Torvalds struct inode *inode = mapping->host; 3551da177e4SLinus Torvalds int err; 3561da177e4SLinus Torvalds 35791d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 35891d5b470SChuck Lever 3591da177e4SLinus Torvalds err = generic_writepages(mapping, wbc); 3601da177e4SLinus Torvalds if (err) 3611da177e4SLinus Torvalds return err; 36228c6925fSTrond Myklebust err = nfs_flush_mapping(mapping, wbc, wb_priority(wbc)); 3631da177e4SLinus Torvalds if (err < 0) 3641da177e4SLinus Torvalds goto out; 36591d5b470SChuck Lever nfs_add_stats(inode, NFSIOS_WRITEPAGES, err); 3661da177e4SLinus Torvalds err = 0; 3671da177e4SLinus Torvalds out: 3681da177e4SLinus Torvalds return err; 3691da177e4SLinus Torvalds } 3701da177e4SLinus Torvalds 3711da177e4SLinus Torvalds /* 3721da177e4SLinus Torvalds * Insert a write request into an inode 3731da177e4SLinus Torvalds */ 3741da177e4SLinus Torvalds static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 3751da177e4SLinus Torvalds { 3761da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 3771da177e4SLinus Torvalds int error; 3781da177e4SLinus Torvalds 3791da177e4SLinus Torvalds error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req); 3801da177e4SLinus Torvalds BUG_ON(error == -EEXIST); 3811da177e4SLinus Torvalds if (error) 3821da177e4SLinus Torvalds return error; 3831da177e4SLinus Torvalds if (!nfsi->npages) { 3841da177e4SLinus Torvalds igrab(inode); 3851da177e4SLinus Torvalds nfs_begin_data_update(inode); 3861da177e4SLinus Torvalds if (nfs_have_delegation(inode, FMODE_WRITE)) 3871da177e4SLinus Torvalds nfsi->change_attr++; 3881da177e4SLinus Torvalds } 389deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 390277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 391*2b82f190STrond Myklebust if (PageDirty(req->wb_page)) 392*2b82f190STrond Myklebust set_bit(PG_NEED_FLUSH, &req->wb_flags); 3931da177e4SLinus Torvalds nfsi->npages++; 3941da177e4SLinus Torvalds atomic_inc(&req->wb_count); 3951da177e4SLinus Torvalds return 0; 3961da177e4SLinus Torvalds } 3971da177e4SLinus Torvalds 3981da177e4SLinus Torvalds /* 39989a09141SPeter Zijlstra * Remove a write request from an inode 4001da177e4SLinus Torvalds */ 4011da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 4021da177e4SLinus Torvalds { 4031da177e4SLinus Torvalds struct inode *inode = req->wb_context->dentry->d_inode; 4041da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4051da177e4SLinus Torvalds 4061da177e4SLinus Torvalds BUG_ON (!NFS_WBACK_BUSY(req)); 4071da177e4SLinus Torvalds 4081da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 409277459d2STrond Myklebust set_page_private(req->wb_page, 0); 410deb7d638STrond Myklebust ClearPagePrivate(req->wb_page); 4111da177e4SLinus Torvalds radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index); 412*2b82f190STrond Myklebust if (test_and_clear_bit(PG_NEED_FLUSH, &req->wb_flags)) 413*2b82f190STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 4141da177e4SLinus Torvalds nfsi->npages--; 4151da177e4SLinus Torvalds if (!nfsi->npages) { 4161da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 417951a143bSTrond Myklebust nfs_end_data_update(inode); 4181da177e4SLinus Torvalds iput(inode); 4191da177e4SLinus Torvalds } else 4201da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 4211da177e4SLinus Torvalds nfs_clear_request(req); 4221da177e4SLinus Torvalds nfs_release_request(req); 4231da177e4SLinus Torvalds } 4241da177e4SLinus Torvalds 42561822ab5STrond Myklebust static void 42661822ab5STrond Myklebust nfs_redirty_request(struct nfs_page *req) 42761822ab5STrond Myklebust { 42861822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 42961822ab5STrond Myklebust } 43061822ab5STrond Myklebust 4311da177e4SLinus Torvalds /* 4321da177e4SLinus Torvalds * Check if a request is dirty 4331da177e4SLinus Torvalds */ 4341da177e4SLinus Torvalds static inline int 4351da177e4SLinus Torvalds nfs_dirty_request(struct nfs_page *req) 4361da177e4SLinus Torvalds { 4375a6d41b3STrond Myklebust struct page *page = req->wb_page; 4385a6d41b3STrond Myklebust 439612c9384STrond Myklebust if (page == NULL || test_bit(PG_NEED_COMMIT, &req->wb_flags)) 4405a6d41b3STrond Myklebust return 0; 4415a6d41b3STrond Myklebust return !PageWriteback(req->wb_page); 4421da177e4SLinus Torvalds } 4431da177e4SLinus Torvalds 4441da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 4451da177e4SLinus Torvalds /* 4461da177e4SLinus Torvalds * Add a request to the inode's commit list. 4471da177e4SLinus Torvalds */ 4481da177e4SLinus Torvalds static void 4491da177e4SLinus Torvalds nfs_mark_request_commit(struct nfs_page *req) 4501da177e4SLinus Torvalds { 4511da177e4SLinus Torvalds struct inode *inode = req->wb_context->dentry->d_inode; 4521da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 4531da177e4SLinus Torvalds 4541da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 4551da177e4SLinus Torvalds nfs_list_add_request(req, &nfsi->commit); 4561da177e4SLinus Torvalds nfsi->ncommit++; 457612c9384STrond Myklebust set_bit(PG_NEED_COMMIT, &(req)->wb_flags); 4581da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 459fd39fc85SChristoph Lameter inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 460a1803044STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 4611da177e4SLinus Torvalds } 4628e821cadSTrond Myklebust 4638e821cadSTrond Myklebust static inline 4648e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4658e821cadSTrond Myklebust { 4668e821cadSTrond Myklebust return data->verf.committed != NFS_FILE_SYNC; 4678e821cadSTrond Myklebust } 4688e821cadSTrond Myklebust 4698e821cadSTrond Myklebust static inline 4708e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req) 4718e821cadSTrond Myklebust { 472612c9384STrond Myklebust if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) { 4738e821cadSTrond Myklebust nfs_mark_request_commit(req); 4748e821cadSTrond Myklebust return 1; 4758e821cadSTrond Myklebust } 4768e821cadSTrond Myklebust if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) { 4778e821cadSTrond Myklebust nfs_redirty_request(req); 4788e821cadSTrond Myklebust return 1; 4798e821cadSTrond Myklebust } 4808e821cadSTrond Myklebust return 0; 4818e821cadSTrond Myklebust } 4828e821cadSTrond Myklebust #else 4838e821cadSTrond Myklebust static inline void 4848e821cadSTrond Myklebust nfs_mark_request_commit(struct nfs_page *req) 4858e821cadSTrond Myklebust { 4868e821cadSTrond Myklebust } 4878e821cadSTrond Myklebust 4888e821cadSTrond Myklebust static inline 4898e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data) 4908e821cadSTrond Myklebust { 4918e821cadSTrond Myklebust return 0; 4928e821cadSTrond Myklebust } 4938e821cadSTrond Myklebust 4948e821cadSTrond Myklebust static inline 4958e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req) 4968e821cadSTrond Myklebust { 4978e821cadSTrond Myklebust return 0; 4988e821cadSTrond Myklebust } 4991da177e4SLinus Torvalds #endif 5001da177e4SLinus Torvalds 5011da177e4SLinus Torvalds /* 5021da177e4SLinus Torvalds * Wait for a request to complete. 5031da177e4SLinus Torvalds * 5041da177e4SLinus Torvalds * Interruptible by signals only if mounted with intr flag. 5051da177e4SLinus Torvalds */ 506c42de9ddSTrond Myklebust static int nfs_wait_on_requests_locked(struct inode *inode, unsigned long idx_start, unsigned int npages) 5071da177e4SLinus Torvalds { 5081da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5091da177e4SLinus Torvalds struct nfs_page *req; 5101da177e4SLinus Torvalds unsigned long idx_end, next; 5111da177e4SLinus Torvalds unsigned int res = 0; 5121da177e4SLinus Torvalds int error; 5131da177e4SLinus Torvalds 5141da177e4SLinus Torvalds if (npages == 0) 5151da177e4SLinus Torvalds idx_end = ~0; 5161da177e4SLinus Torvalds else 5171da177e4SLinus Torvalds idx_end = idx_start + npages - 1; 5181da177e4SLinus Torvalds 5191da177e4SLinus Torvalds next = idx_start; 520c6a556b8STrond Myklebust while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_WRITEBACK)) { 5211da177e4SLinus Torvalds if (req->wb_index > idx_end) 5221da177e4SLinus Torvalds break; 5231da177e4SLinus Torvalds 5241da177e4SLinus Torvalds next = req->wb_index + 1; 525c6a556b8STrond Myklebust BUG_ON(!NFS_WBACK_BUSY(req)); 5261da177e4SLinus Torvalds 5271da177e4SLinus Torvalds atomic_inc(&req->wb_count); 5281da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 5291da177e4SLinus Torvalds error = nfs_wait_on_request(req); 5301da177e4SLinus Torvalds nfs_release_request(req); 531c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 5321da177e4SLinus Torvalds if (error < 0) 5331da177e4SLinus Torvalds return error; 5341da177e4SLinus Torvalds res++; 5351da177e4SLinus Torvalds } 5361da177e4SLinus Torvalds return res; 5371da177e4SLinus Torvalds } 5381da177e4SLinus Torvalds 53983715ad5STrond Myklebust static void nfs_cancel_dirty_list(struct list_head *head) 540d2ccddf0STrond Myklebust { 541d2ccddf0STrond Myklebust struct nfs_page *req; 542d2ccddf0STrond Myklebust while(!list_empty(head)) { 543d2ccddf0STrond Myklebust req = nfs_list_entry(head->next); 544d2ccddf0STrond Myklebust nfs_list_remove_request(req); 5455a6d41b3STrond Myklebust nfs_end_page_writeback(req->wb_page); 546d2ccddf0STrond Myklebust nfs_inode_remove_request(req); 547d2ccddf0STrond Myklebust nfs_clear_page_writeback(req); 548d2ccddf0STrond Myklebust } 549d2ccddf0STrond Myklebust } 550d2ccddf0STrond Myklebust 55183715ad5STrond Myklebust static void nfs_cancel_commit_list(struct list_head *head) 55283715ad5STrond Myklebust { 55383715ad5STrond Myklebust struct nfs_page *req; 55483715ad5STrond Myklebust 55583715ad5STrond Myklebust while(!list_empty(head)) { 55683715ad5STrond Myklebust req = nfs_list_entry(head->next); 557b6dff26aSTrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 55883715ad5STrond Myklebust nfs_list_remove_request(req); 559612c9384STrond Myklebust clear_bit(PG_NEED_COMMIT, &(req)->wb_flags); 56083715ad5STrond Myklebust nfs_inode_remove_request(req); 561b6dff26aSTrond Myklebust nfs_unlock_request(req); 56283715ad5STrond Myklebust } 56383715ad5STrond Myklebust } 56483715ad5STrond Myklebust 5651da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 5661da177e4SLinus Torvalds /* 5671da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 5681da177e4SLinus Torvalds * @inode: NFS inode to scan 5691da177e4SLinus Torvalds * @dst: destination list 5701da177e4SLinus Torvalds * @idx_start: lower bound of page->index to scan. 5711da177e4SLinus Torvalds * @npages: idx_start + npages sets the upper bound to scan. 5721da177e4SLinus Torvalds * 5731da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 5741da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 5751da177e4SLinus Torvalds */ 5761da177e4SLinus Torvalds static int 5771da177e4SLinus Torvalds nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) 5781da177e4SLinus Torvalds { 5791da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 5803da28eb1STrond Myklebust int res = 0; 5813da28eb1STrond Myklebust 5823da28eb1STrond Myklebust if (nfsi->ncommit != 0) { 583d2ccddf0STrond Myklebust res = nfs_scan_list(nfsi, &nfsi->commit, dst, idx_start, npages); 5841da177e4SLinus Torvalds nfsi->ncommit -= res; 5851da177e4SLinus Torvalds if ((nfsi->ncommit == 0) != list_empty(&nfsi->commit)) 5861da177e4SLinus Torvalds printk(KERN_ERR "NFS: desynchronized value of nfs_i.ncommit.\n"); 5873da28eb1STrond Myklebust } 5881da177e4SLinus Torvalds return res; 5891da177e4SLinus Torvalds } 590c42de9ddSTrond Myklebust #else 591c42de9ddSTrond Myklebust static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages) 592c42de9ddSTrond Myklebust { 593c42de9ddSTrond Myklebust return 0; 594c42de9ddSTrond Myklebust } 5951da177e4SLinus Torvalds #endif 5961da177e4SLinus Torvalds 59789a09141SPeter Zijlstra static int nfs_wait_on_write_congestion(struct address_space *mapping) 5981da177e4SLinus Torvalds { 59989a09141SPeter Zijlstra struct inode *inode = mapping->host; 6001da177e4SLinus Torvalds struct backing_dev_info *bdi = mapping->backing_dev_info; 6011da177e4SLinus Torvalds int ret = 0; 6021da177e4SLinus Torvalds 6031da177e4SLinus Torvalds might_sleep(); 6041da177e4SLinus Torvalds 6051da177e4SLinus Torvalds if (!bdi_write_congested(bdi)) 6061da177e4SLinus Torvalds return 0; 60791d5b470SChuck Lever 60889a09141SPeter Zijlstra nfs_inc_stats(inode, NFSIOS_CONGESTIONWAIT); 60991d5b470SChuck Lever 61089a09141SPeter Zijlstra do { 61189a09141SPeter Zijlstra struct rpc_clnt *clnt = NFS_CLIENT(inode); 6121da177e4SLinus Torvalds sigset_t oldset; 6131da177e4SLinus Torvalds 6141da177e4SLinus Torvalds rpc_clnt_sigmask(clnt, &oldset); 61589a09141SPeter Zijlstra ret = congestion_wait_interruptible(WRITE, HZ/10); 6161da177e4SLinus Torvalds rpc_clnt_sigunmask(clnt, &oldset); 61789a09141SPeter Zijlstra if (ret == -ERESTARTSYS) 61889a09141SPeter Zijlstra break; 61989a09141SPeter Zijlstra ret = 0; 62089a09141SPeter Zijlstra } while (bdi_write_congested(bdi)); 62189a09141SPeter Zijlstra 6221da177e4SLinus Torvalds return ret; 6231da177e4SLinus Torvalds } 6241da177e4SLinus Torvalds 6251da177e4SLinus Torvalds /* 6261da177e4SLinus Torvalds * Try to update any existing write request, or create one if there is none. 6271da177e4SLinus Torvalds * In order to match, the request's credentials must match those of 6281da177e4SLinus Torvalds * the calling process. 6291da177e4SLinus Torvalds * 6301da177e4SLinus Torvalds * Note: Should always be called with the Page Lock held! 6311da177e4SLinus Torvalds */ 6321da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx, 633e21195a7STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 6341da177e4SLinus Torvalds { 63589a09141SPeter Zijlstra struct address_space *mapping = page->mapping; 63689a09141SPeter Zijlstra struct inode *inode = mapping->host; 6371da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 6381da177e4SLinus Torvalds struct nfs_page *req, *new = NULL; 6391da177e4SLinus Torvalds unsigned long rqend, end; 6401da177e4SLinus Torvalds 6411da177e4SLinus Torvalds end = offset + bytes; 6421da177e4SLinus Torvalds 64389a09141SPeter Zijlstra if (nfs_wait_on_write_congestion(mapping)) 6441da177e4SLinus Torvalds return ERR_PTR(-ERESTARTSYS); 6451da177e4SLinus Torvalds for (;;) { 6461da177e4SLinus Torvalds /* Loop over all inode entries and see if we find 6471da177e4SLinus Torvalds * A request for the page we wish to update 6481da177e4SLinus Torvalds */ 6491da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 650277459d2STrond Myklebust req = nfs_page_find_request_locked(page); 6511da177e4SLinus Torvalds if (req) { 6521da177e4SLinus Torvalds if (!nfs_lock_request_dontget(req)) { 6531da177e4SLinus Torvalds int error; 654277459d2STrond Myklebust 6551da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 6561da177e4SLinus Torvalds error = nfs_wait_on_request(req); 6571da177e4SLinus Torvalds nfs_release_request(req); 6581dd594b2SNeil Brown if (error < 0) { 6591dd594b2SNeil Brown if (new) 6601dd594b2SNeil Brown nfs_release_request(new); 6611da177e4SLinus Torvalds return ERR_PTR(error); 6621dd594b2SNeil Brown } 6631da177e4SLinus Torvalds continue; 6641da177e4SLinus Torvalds } 6651da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 6661da177e4SLinus Torvalds if (new) 6671da177e4SLinus Torvalds nfs_release_request(new); 6681da177e4SLinus Torvalds break; 6691da177e4SLinus Torvalds } 6701da177e4SLinus Torvalds 6711da177e4SLinus Torvalds if (new) { 6721da177e4SLinus Torvalds int error; 6731da177e4SLinus Torvalds nfs_lock_request_dontget(new); 6741da177e4SLinus Torvalds error = nfs_inode_add_request(inode, new); 6751da177e4SLinus Torvalds if (error) { 6761da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 6771da177e4SLinus Torvalds nfs_unlock_request(new); 6781da177e4SLinus Torvalds return ERR_PTR(error); 6791da177e4SLinus Torvalds } 6801da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 6811da177e4SLinus Torvalds return new; 6821da177e4SLinus Torvalds } 6831da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 6841da177e4SLinus Torvalds 6851da177e4SLinus Torvalds new = nfs_create_request(ctx, inode, page, offset, bytes); 6861da177e4SLinus Torvalds if (IS_ERR(new)) 6871da177e4SLinus Torvalds return new; 6881da177e4SLinus Torvalds } 6891da177e4SLinus Torvalds 6901da177e4SLinus Torvalds /* We have a request for our page. 6911da177e4SLinus Torvalds * If the creds don't match, or the 6921da177e4SLinus Torvalds * page addresses don't match, 6931da177e4SLinus Torvalds * tell the caller to wait on the conflicting 6941da177e4SLinus Torvalds * request. 6951da177e4SLinus Torvalds */ 6961da177e4SLinus Torvalds rqend = req->wb_offset + req->wb_bytes; 6971da177e4SLinus Torvalds if (req->wb_context != ctx 6981da177e4SLinus Torvalds || req->wb_page != page 6991da177e4SLinus Torvalds || !nfs_dirty_request(req) 7001da177e4SLinus Torvalds || offset > rqend || end < req->wb_offset) { 7011da177e4SLinus Torvalds nfs_unlock_request(req); 7021da177e4SLinus Torvalds return ERR_PTR(-EBUSY); 7031da177e4SLinus Torvalds } 7041da177e4SLinus Torvalds 7051da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 7061da177e4SLinus Torvalds if (offset < req->wb_offset) { 7071da177e4SLinus Torvalds req->wb_offset = offset; 7081da177e4SLinus Torvalds req->wb_pgbase = offset; 7091da177e4SLinus Torvalds req->wb_bytes = rqend - req->wb_offset; 7101da177e4SLinus Torvalds } 7111da177e4SLinus Torvalds 7121da177e4SLinus Torvalds if (end > rqend) 7131da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 7141da177e4SLinus Torvalds 7151da177e4SLinus Torvalds return req; 7161da177e4SLinus Torvalds } 7171da177e4SLinus Torvalds 7181da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 7191da177e4SLinus Torvalds { 7201da177e4SLinus Torvalds struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; 7211da177e4SLinus Torvalds struct nfs_page *req; 7221a54533eSTrond Myklebust int do_flush, status; 7231da177e4SLinus Torvalds /* 7241da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 7251da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 7261da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 7271da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 7281da177e4SLinus Torvalds * Also do the same if we find a request from an existing 7291da177e4SLinus Torvalds * dropped page. 7301da177e4SLinus Torvalds */ 7311a54533eSTrond Myklebust do { 732277459d2STrond Myklebust req = nfs_page_find_request(page); 7331a54533eSTrond Myklebust if (req == NULL) 7341a54533eSTrond Myklebust return 0; 7351a54533eSTrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx 736e261f51fSTrond Myklebust || !nfs_dirty_request(req); 7371da177e4SLinus Torvalds nfs_release_request(req); 7381a54533eSTrond Myklebust if (!do_flush) 7391a54533eSTrond Myklebust return 0; 740277459d2STrond Myklebust status = nfs_wb_page(page->mapping->host, page); 7411a54533eSTrond Myklebust } while (status == 0); 7421a54533eSTrond Myklebust return status; 7431da177e4SLinus Torvalds } 7441da177e4SLinus Torvalds 7451da177e4SLinus Torvalds /* 7461da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 7471da177e4SLinus Torvalds * 7481da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 7491da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 7501da177e4SLinus Torvalds */ 7511da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 7521da177e4SLinus Torvalds unsigned int offset, unsigned int count) 7531da177e4SLinus Torvalds { 7541da177e4SLinus Torvalds struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data; 7551da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 7561da177e4SLinus Torvalds int status = 0; 7571da177e4SLinus Torvalds 75891d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 75991d5b470SChuck Lever 7601da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage(%s/%s %d@%Ld)\n", 76101cce933SJosef "Jeff" Sipek file->f_path.dentry->d_parent->d_name.name, 76201cce933SJosef "Jeff" Sipek file->f_path.dentry->d_name.name, count, 7630bbacc40SChuck Lever (long long)(page_offset(page) +offset)); 7641da177e4SLinus Torvalds 7651da177e4SLinus Torvalds /* If we're not using byte range locks, and we know the page 7661da177e4SLinus Torvalds * is entirely in cache, it may be more efficient to avoid 7671da177e4SLinus Torvalds * fragmenting write requests. 7681da177e4SLinus Torvalds */ 769ab0a3dbeSTrond Myklebust if (PageUptodate(page) && inode->i_flock == NULL && !(file->f_mode & O_SYNC)) { 77049a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 7711da177e4SLinus Torvalds offset = 0; 7721da177e4SLinus Torvalds } 7731da177e4SLinus Torvalds 774e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 775e261f51fSTrond Myklebust __set_page_dirty_nobuffers(page); 7761da177e4SLinus Torvalds 7771da177e4SLinus Torvalds dprintk("NFS: nfs_updatepage returns %d (isize %Ld)\n", 7781da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 7791da177e4SLinus Torvalds if (status < 0) 780a301b777STrond Myklebust nfs_set_pageerror(page); 7811da177e4SLinus Torvalds return status; 7821da177e4SLinus Torvalds } 7831da177e4SLinus Torvalds 7841da177e4SLinus Torvalds static void nfs_writepage_release(struct nfs_page *req) 7851da177e4SLinus Torvalds { 7868e821cadSTrond Myklebust 7878e821cadSTrond Myklebust if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req)) { 78889a09141SPeter Zijlstra nfs_end_page_writeback(req->wb_page); 7891da177e4SLinus Torvalds nfs_inode_remove_request(req); 7908e821cadSTrond Myklebust } else 7918e821cadSTrond Myklebust nfs_end_page_writeback(req->wb_page); 792c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 7931da177e4SLinus Torvalds } 7941da177e4SLinus Torvalds 7951da177e4SLinus Torvalds static inline int flush_task_priority(int how) 7961da177e4SLinus Torvalds { 7971da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 7981da177e4SLinus Torvalds case FLUSH_HIGHPRI: 7991da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 8001da177e4SLinus Torvalds case FLUSH_LOWPRI: 8011da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 8021da177e4SLinus Torvalds } 8031da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 8041da177e4SLinus Torvalds } 8051da177e4SLinus Torvalds 8061da177e4SLinus Torvalds /* 8071da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 8081da177e4SLinus Torvalds */ 8091da177e4SLinus Torvalds static void nfs_write_rpcsetup(struct nfs_page *req, 8101da177e4SLinus Torvalds struct nfs_write_data *data, 811788e7a89STrond Myklebust const struct rpc_call_ops *call_ops, 8121da177e4SLinus Torvalds unsigned int count, unsigned int offset, 8131da177e4SLinus Torvalds int how) 8141da177e4SLinus Torvalds { 8151da177e4SLinus Torvalds struct inode *inode; 816788e7a89STrond Myklebust int flags; 8171da177e4SLinus Torvalds 8181da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 8191da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 8201da177e4SLinus Torvalds 8211da177e4SLinus Torvalds data->req = req; 8221da177e4SLinus Torvalds data->inode = inode = req->wb_context->dentry->d_inode; 8231da177e4SLinus Torvalds data->cred = req->wb_context->cred; 8241da177e4SLinus Torvalds 8251da177e4SLinus Torvalds data->args.fh = NFS_FH(inode); 8261da177e4SLinus Torvalds data->args.offset = req_offset(req) + offset; 8271da177e4SLinus Torvalds data->args.pgbase = req->wb_pgbase + offset; 8281da177e4SLinus Torvalds data->args.pages = data->pagevec; 8291da177e4SLinus Torvalds data->args.count = count; 8301da177e4SLinus Torvalds data->args.context = req->wb_context; 8311da177e4SLinus Torvalds 8321da177e4SLinus Torvalds data->res.fattr = &data->fattr; 8331da177e4SLinus Torvalds data->res.count = count; 8341da177e4SLinus Torvalds data->res.verf = &data->verf; 8350e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 8361da177e4SLinus Torvalds 837788e7a89STrond Myklebust /* Set up the initial task struct. */ 838788e7a89STrond Myklebust flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 839788e7a89STrond Myklebust rpc_init_task(&data->task, NFS_CLIENT(inode), flags, call_ops, data); 8401da177e4SLinus Torvalds NFS_PROTO(inode)->write_setup(data, how); 8411da177e4SLinus Torvalds 8421da177e4SLinus Torvalds data->task.tk_priority = flush_task_priority(how); 8431da177e4SLinus Torvalds data->task.tk_cookie = (unsigned long)inode; 8441da177e4SLinus Torvalds 845a3f565b1SChuck Lever dprintk("NFS: %5u initiated write call " 846a3f565b1SChuck Lever "(req %s/%Ld, %u bytes @ offset %Lu)\n", 8470bbacc40SChuck Lever data->task.tk_pid, 8481da177e4SLinus Torvalds inode->i_sb->s_id, 8491da177e4SLinus Torvalds (long long)NFS_FILEID(inode), 8501da177e4SLinus Torvalds count, 8511da177e4SLinus Torvalds (unsigned long long)data->args.offset); 8521da177e4SLinus Torvalds } 8531da177e4SLinus Torvalds 8541da177e4SLinus Torvalds static void nfs_execute_write(struct nfs_write_data *data) 8551da177e4SLinus Torvalds { 8561da177e4SLinus Torvalds struct rpc_clnt *clnt = NFS_CLIENT(data->inode); 8571da177e4SLinus Torvalds sigset_t oldset; 8581da177e4SLinus Torvalds 8591da177e4SLinus Torvalds rpc_clnt_sigmask(clnt, &oldset); 8601da177e4SLinus Torvalds rpc_execute(&data->task); 8611da177e4SLinus Torvalds rpc_clnt_sigunmask(clnt, &oldset); 8621da177e4SLinus Torvalds } 8631da177e4SLinus Torvalds 8641da177e4SLinus Torvalds /* 8651da177e4SLinus Torvalds * Generate multiple small requests to write out a single 8661da177e4SLinus Torvalds * contiguous dirty area on one page. 8671da177e4SLinus Torvalds */ 8687d46a49fSTrond Myklebust static int nfs_flush_multi(struct inode *inode, struct list_head *head, int how) 8691da177e4SLinus Torvalds { 8701da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 8711da177e4SLinus Torvalds struct page *page = req->wb_page; 8721da177e4SLinus Torvalds struct nfs_write_data *data; 873e9f7bee1STrond Myklebust size_t wsize = NFS_SERVER(inode)->wsize, nbytes; 874e9f7bee1STrond Myklebust unsigned int offset; 8751da177e4SLinus Torvalds int requests = 0; 8761da177e4SLinus Torvalds LIST_HEAD(list); 8771da177e4SLinus Torvalds 8781da177e4SLinus Torvalds nfs_list_remove_request(req); 8791da177e4SLinus Torvalds 8801da177e4SLinus Torvalds nbytes = req->wb_bytes; 881e9f7bee1STrond Myklebust do { 882e9f7bee1STrond Myklebust size_t len = min(nbytes, wsize); 883e9f7bee1STrond Myklebust 884e9f7bee1STrond Myklebust data = nfs_writedata_alloc(len); 8851da177e4SLinus Torvalds if (!data) 8861da177e4SLinus Torvalds goto out_bad; 8871da177e4SLinus Torvalds list_add(&data->pages, &list); 8881da177e4SLinus Torvalds requests++; 889e9f7bee1STrond Myklebust nbytes -= len; 890e9f7bee1STrond Myklebust } while (nbytes != 0); 8911da177e4SLinus Torvalds atomic_set(&req->wb_complete, requests); 8921da177e4SLinus Torvalds 8931da177e4SLinus Torvalds ClearPageError(page); 8941da177e4SLinus Torvalds offset = 0; 8951da177e4SLinus Torvalds nbytes = req->wb_bytes; 8961da177e4SLinus Torvalds do { 8971da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 8981da177e4SLinus Torvalds list_del_init(&data->pages); 8991da177e4SLinus Torvalds 9001da177e4SLinus Torvalds data->pagevec[0] = page; 9011da177e4SLinus Torvalds 9021da177e4SLinus Torvalds if (nbytes > wsize) { 903788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 904788e7a89STrond Myklebust wsize, offset, how); 9051da177e4SLinus Torvalds offset += wsize; 9061da177e4SLinus Torvalds nbytes -= wsize; 9071da177e4SLinus Torvalds } else { 908788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_partial_ops, 909788e7a89STrond Myklebust nbytes, offset, how); 9101da177e4SLinus Torvalds nbytes = 0; 9111da177e4SLinus Torvalds } 9121da177e4SLinus Torvalds nfs_execute_write(data); 9131da177e4SLinus Torvalds } while (nbytes != 0); 9141da177e4SLinus Torvalds 9151da177e4SLinus Torvalds return 0; 9161da177e4SLinus Torvalds 9171da177e4SLinus Torvalds out_bad: 9181da177e4SLinus Torvalds while (!list_empty(&list)) { 9191da177e4SLinus Torvalds data = list_entry(list.next, struct nfs_write_data, pages); 9201da177e4SLinus Torvalds list_del(&data->pages); 9218aca67f0STrond Myklebust nfs_writedata_release(data); 9221da177e4SLinus Torvalds } 92361822ab5STrond Myklebust nfs_redirty_request(req); 9246d677e35STrond Myklebust nfs_end_page_writeback(req->wb_page); 925c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 9261da177e4SLinus Torvalds return -ENOMEM; 9271da177e4SLinus Torvalds } 9281da177e4SLinus Torvalds 9291da177e4SLinus Torvalds /* 9301da177e4SLinus Torvalds * Create an RPC task for the given write request and kick it. 9311da177e4SLinus Torvalds * The page must have been locked by the caller. 9321da177e4SLinus Torvalds * 9331da177e4SLinus Torvalds * It may happen that the page we're passed is not marked dirty. 9341da177e4SLinus Torvalds * This is the case if nfs_updatepage detects a conflicting request 9351da177e4SLinus Torvalds * that has been written but not committed. 9361da177e4SLinus Torvalds */ 9377d46a49fSTrond Myklebust static int nfs_flush_one(struct inode *inode, struct list_head *head, int how) 9381da177e4SLinus Torvalds { 9391da177e4SLinus Torvalds struct nfs_page *req; 9401da177e4SLinus Torvalds struct page **pages; 9411da177e4SLinus Torvalds struct nfs_write_data *data; 9421da177e4SLinus Torvalds unsigned int count; 9431da177e4SLinus Torvalds 944e9f7bee1STrond Myklebust data = nfs_writedata_alloc(NFS_SERVER(inode)->wsize); 9451da177e4SLinus Torvalds if (!data) 9461da177e4SLinus Torvalds goto out_bad; 9471da177e4SLinus Torvalds 9481da177e4SLinus Torvalds pages = data->pagevec; 9491da177e4SLinus Torvalds count = 0; 9501da177e4SLinus Torvalds while (!list_empty(head)) { 9511da177e4SLinus Torvalds req = nfs_list_entry(head->next); 9521da177e4SLinus Torvalds nfs_list_remove_request(req); 9531da177e4SLinus Torvalds nfs_list_add_request(req, &data->pages); 9541da177e4SLinus Torvalds ClearPageError(req->wb_page); 9551da177e4SLinus Torvalds *pages++ = req->wb_page; 9561da177e4SLinus Torvalds count += req->wb_bytes; 9571da177e4SLinus Torvalds } 9581da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 9591da177e4SLinus Torvalds 9601da177e4SLinus Torvalds /* Set up the argument struct */ 961788e7a89STrond Myklebust nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how); 9621da177e4SLinus Torvalds 9631da177e4SLinus Torvalds nfs_execute_write(data); 9641da177e4SLinus Torvalds return 0; 9651da177e4SLinus Torvalds out_bad: 9661da177e4SLinus Torvalds while (!list_empty(head)) { 9671da177e4SLinus Torvalds struct nfs_page *req = nfs_list_entry(head->next); 9681da177e4SLinus Torvalds nfs_list_remove_request(req); 96961822ab5STrond Myklebust nfs_redirty_request(req); 9706d677e35STrond Myklebust nfs_end_page_writeback(req->wb_page); 971c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 9721da177e4SLinus Torvalds } 9731da177e4SLinus Torvalds return -ENOMEM; 9741da177e4SLinus Torvalds } 9751da177e4SLinus Torvalds 9767d46a49fSTrond Myklebust static int nfs_flush_list(struct inode *inode, struct list_head *head, int npages, int how) 9771da177e4SLinus Torvalds { 9781da177e4SLinus Torvalds LIST_HEAD(one_request); 9797d46a49fSTrond Myklebust int (*flush_one)(struct inode *, struct list_head *, int); 9801da177e4SLinus Torvalds struct nfs_page *req; 9817d46a49fSTrond Myklebust int wpages = NFS_SERVER(inode)->wpages; 9827d46a49fSTrond Myklebust int wsize = NFS_SERVER(inode)->wsize; 9837d46a49fSTrond Myklebust int error; 9841da177e4SLinus Torvalds 9857d46a49fSTrond Myklebust flush_one = nfs_flush_one; 9867d46a49fSTrond Myklebust if (wsize < PAGE_CACHE_SIZE) 9877d46a49fSTrond Myklebust flush_one = nfs_flush_multi; 9887d46a49fSTrond Myklebust /* For single writes, FLUSH_STABLE is more efficient */ 9897d46a49fSTrond Myklebust if (npages <= wpages && npages == NFS_I(inode)->npages 9907d46a49fSTrond Myklebust && nfs_list_entry(head->next)->wb_bytes <= wsize) 9917d46a49fSTrond Myklebust how |= FLUSH_STABLE; 9927d46a49fSTrond Myklebust 9937d46a49fSTrond Myklebust do { 9947d46a49fSTrond Myklebust nfs_coalesce_requests(head, &one_request, wpages); 9951da177e4SLinus Torvalds req = nfs_list_entry(one_request.next); 9967d46a49fSTrond Myklebust error = flush_one(inode, &one_request, how); 9971da177e4SLinus Torvalds if (error < 0) 9987d46a49fSTrond Myklebust goto out_err; 9997d46a49fSTrond Myklebust } while (!list_empty(head)); 10007d46a49fSTrond Myklebust return 0; 10017d46a49fSTrond Myklebust out_err: 10021da177e4SLinus Torvalds while (!list_empty(head)) { 10031da177e4SLinus Torvalds req = nfs_list_entry(head->next); 10041da177e4SLinus Torvalds nfs_list_remove_request(req); 100561822ab5STrond Myklebust nfs_redirty_request(req); 10066d677e35STrond Myklebust nfs_end_page_writeback(req->wb_page); 1007c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 10081da177e4SLinus Torvalds } 10091da177e4SLinus Torvalds return error; 10101da177e4SLinus Torvalds } 10111da177e4SLinus Torvalds 10121da177e4SLinus Torvalds /* 10131da177e4SLinus Torvalds * Handle a write reply that flushed part of a page. 10141da177e4SLinus Torvalds */ 1015788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata) 10161da177e4SLinus Torvalds { 1017788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 10181da177e4SLinus Torvalds struct nfs_page *req = data->req; 10191da177e4SLinus Torvalds struct page *page = req->wb_page; 10201da177e4SLinus Torvalds 10211da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 10221da177e4SLinus Torvalds req->wb_context->dentry->d_inode->i_sb->s_id, 10231da177e4SLinus Torvalds (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 10241da177e4SLinus Torvalds req->wb_bytes, 10251da177e4SLinus Torvalds (long long)req_offset(req)); 10261da177e4SLinus Torvalds 1027788e7a89STrond Myklebust if (nfs_writeback_done(task, data) != 0) 1028788e7a89STrond Myklebust return; 1029788e7a89STrond Myklebust 1030788e7a89STrond Myklebust if (task->tk_status < 0) { 1031a301b777STrond Myklebust nfs_set_pageerror(page); 1032788e7a89STrond Myklebust req->wb_context->error = task->tk_status; 1033788e7a89STrond Myklebust dprintk(", error = %d\n", task->tk_status); 10348e821cadSTrond Myklebust goto out; 10358e821cadSTrond Myklebust } 10368e821cadSTrond Myklebust 10378e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 10388e821cadSTrond Myklebust spinlock_t *req_lock = &NFS_I(page->mapping->host)->req_lock; 10398e821cadSTrond Myklebust 10408e821cadSTrond Myklebust spin_lock(req_lock); 10418e821cadSTrond Myklebust if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) { 10428e821cadSTrond Myklebust /* Do nothing we need to resend the writes */ 10438e821cadSTrond Myklebust } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) { 10441da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 10451da177e4SLinus Torvalds dprintk(" defer commit\n"); 10461da177e4SLinus Torvalds } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) { 10478e821cadSTrond Myklebust set_bit(PG_NEED_RESCHED, &req->wb_flags); 10488e821cadSTrond Myklebust clear_bit(PG_NEED_COMMIT, &req->wb_flags); 10491da177e4SLinus Torvalds dprintk(" server reboot detected\n"); 10501da177e4SLinus Torvalds } 10518e821cadSTrond Myklebust spin_unlock(req_lock); 10521da177e4SLinus Torvalds } else 10531da177e4SLinus Torvalds dprintk(" OK\n"); 10541da177e4SLinus Torvalds 10558e821cadSTrond Myklebust out: 10561da177e4SLinus Torvalds if (atomic_dec_and_test(&req->wb_complete)) 10571da177e4SLinus Torvalds nfs_writepage_release(req); 10581da177e4SLinus Torvalds } 10591da177e4SLinus Torvalds 1060788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops = { 1061788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_partial, 1062788e7a89STrond Myklebust .rpc_release = nfs_writedata_release, 1063788e7a89STrond Myklebust }; 1064788e7a89STrond Myklebust 10651da177e4SLinus Torvalds /* 10661da177e4SLinus Torvalds * Handle a write reply that flushes a whole page. 10671da177e4SLinus Torvalds * 10681da177e4SLinus Torvalds * FIXME: There is an inherent race with invalidate_inode_pages and 10691da177e4SLinus Torvalds * writebacks since the page->count is kept > 1 for as long 10701da177e4SLinus Torvalds * as the page has a write request pending. 10711da177e4SLinus Torvalds */ 1072788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata) 10731da177e4SLinus Torvalds { 1074788e7a89STrond Myklebust struct nfs_write_data *data = calldata; 10751da177e4SLinus Torvalds struct nfs_page *req; 10761da177e4SLinus Torvalds struct page *page; 10771da177e4SLinus Torvalds 1078788e7a89STrond Myklebust if (nfs_writeback_done(task, data) != 0) 1079788e7a89STrond Myklebust return; 1080788e7a89STrond Myklebust 10811da177e4SLinus Torvalds /* Update attributes as result of writeback. */ 10821da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 10831da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 10841da177e4SLinus Torvalds nfs_list_remove_request(req); 10851da177e4SLinus Torvalds page = req->wb_page; 10861da177e4SLinus Torvalds 10871da177e4SLinus Torvalds dprintk("NFS: write (%s/%Ld %d@%Ld)", 10881da177e4SLinus Torvalds req->wb_context->dentry->d_inode->i_sb->s_id, 10891da177e4SLinus Torvalds (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 10901da177e4SLinus Torvalds req->wb_bytes, 10911da177e4SLinus Torvalds (long long)req_offset(req)); 10921da177e4SLinus Torvalds 1093788e7a89STrond Myklebust if (task->tk_status < 0) { 1094a301b777STrond Myklebust nfs_set_pageerror(page); 1095788e7a89STrond Myklebust req->wb_context->error = task->tk_status; 1096788e7a89STrond Myklebust dprintk(", error = %d\n", task->tk_status); 10978e821cadSTrond Myklebust goto remove_request; 10981da177e4SLinus Torvalds } 10991da177e4SLinus Torvalds 11008e821cadSTrond Myklebust if (nfs_write_need_commit(data)) { 11011da177e4SLinus Torvalds memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf)); 11021da177e4SLinus Torvalds nfs_mark_request_commit(req); 11038e821cadSTrond Myklebust nfs_end_page_writeback(page); 11041da177e4SLinus Torvalds dprintk(" marked for commit\n"); 11058e821cadSTrond Myklebust goto next; 11068e821cadSTrond Myklebust } 11078e821cadSTrond Myklebust dprintk(" OK\n"); 11088e821cadSTrond Myklebust remove_request: 11098e821cadSTrond Myklebust nfs_end_page_writeback(page); 11101da177e4SLinus Torvalds nfs_inode_remove_request(req); 11111da177e4SLinus Torvalds next: 1112c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 11131da177e4SLinus Torvalds } 11141da177e4SLinus Torvalds } 11151da177e4SLinus Torvalds 1116788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = { 1117788e7a89STrond Myklebust .rpc_call_done = nfs_writeback_done_full, 1118788e7a89STrond Myklebust .rpc_release = nfs_writedata_release, 1119788e7a89STrond Myklebust }; 1120788e7a89STrond Myklebust 1121788e7a89STrond Myklebust 11221da177e4SLinus Torvalds /* 11231da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 11241da177e4SLinus Torvalds */ 1125462d5b32SChuck Lever int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) 11261da177e4SLinus Torvalds { 11271da177e4SLinus Torvalds struct nfs_writeargs *argp = &data->args; 11281da177e4SLinus Torvalds struct nfs_writeres *resp = &data->res; 1129788e7a89STrond Myklebust int status; 11301da177e4SLinus Torvalds 1131a3f565b1SChuck Lever dprintk("NFS: %5u nfs_writeback_done (status %d)\n", 11321da177e4SLinus Torvalds task->tk_pid, task->tk_status); 11331da177e4SLinus Torvalds 1134f551e44fSChuck Lever /* 1135f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1136f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1137f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1138f551e44fSChuck Lever * another writer had changed the file, but some applications 1139f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1140f551e44fSChuck Lever */ 1141788e7a89STrond Myklebust status = NFS_PROTO(data->inode)->write_done(task, data); 1142788e7a89STrond Myklebust if (status != 0) 1143788e7a89STrond Myklebust return status; 114491d5b470SChuck Lever nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count); 114591d5b470SChuck Lever 11461da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 11471da177e4SLinus Torvalds if (resp->verf->committed < argp->stable && task->tk_status >= 0) { 11481da177e4SLinus Torvalds /* We tried a write call, but the server did not 11491da177e4SLinus Torvalds * commit data to stable storage even though we 11501da177e4SLinus Torvalds * requested it. 11511da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 11521da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 11531da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 11541da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 11551da177e4SLinus Torvalds */ 11561da177e4SLinus Torvalds static unsigned long complain; 11571da177e4SLinus Torvalds 11581da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 11591da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 11601da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 116154ceac45SDavid Howells NFS_SERVER(data->inode)->nfs_client->cl_hostname, 11621da177e4SLinus Torvalds resp->verf->committed, argp->stable); 11631da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11641da177e4SLinus Torvalds } 11651da177e4SLinus Torvalds } 11661da177e4SLinus Torvalds #endif 11671da177e4SLinus Torvalds /* Is this a short write? */ 11681da177e4SLinus Torvalds if (task->tk_status >= 0 && resp->count < argp->count) { 11691da177e4SLinus Torvalds static unsigned long complain; 11701da177e4SLinus Torvalds 117191d5b470SChuck Lever nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE); 117291d5b470SChuck Lever 11731da177e4SLinus Torvalds /* Has the server at least made some progress? */ 11741da177e4SLinus Torvalds if (resp->count != 0) { 11751da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 11761da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 11771da177e4SLinus Torvalds /* Resend from where the server left off */ 11781da177e4SLinus Torvalds argp->offset += resp->count; 11791da177e4SLinus Torvalds argp->pgbase += resp->count; 11801da177e4SLinus Torvalds argp->count -= resp->count; 11811da177e4SLinus Torvalds } else { 11821da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 11831da177e4SLinus Torvalds * headaches in the case of a server crash. 11841da177e4SLinus Torvalds */ 11851da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 11861da177e4SLinus Torvalds } 11871da177e4SLinus Torvalds rpc_restart_call(task); 1188788e7a89STrond Myklebust return -EAGAIN; 11891da177e4SLinus Torvalds } 11901da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 11911da177e4SLinus Torvalds printk(KERN_WARNING 11921da177e4SLinus Torvalds "NFS: Server wrote zero bytes, expected %u.\n", 11931da177e4SLinus Torvalds argp->count); 11941da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 11951da177e4SLinus Torvalds } 11961da177e4SLinus Torvalds /* Can't do anything about it except throw an error. */ 11971da177e4SLinus Torvalds task->tk_status = -EIO; 11981da177e4SLinus Torvalds } 1199788e7a89STrond Myklebust return 0; 12001da177e4SLinus Torvalds } 12011da177e4SLinus Torvalds 12021da177e4SLinus Torvalds 12031da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 1204963d8fe5STrond Myklebust void nfs_commit_release(void *wdata) 12051da177e4SLinus Torvalds { 12061da177e4SLinus Torvalds nfs_commit_free(wdata); 12071da177e4SLinus Torvalds } 12081da177e4SLinus Torvalds 12091da177e4SLinus Torvalds /* 12101da177e4SLinus Torvalds * Set up the argument/result storage required for the RPC call. 12111da177e4SLinus Torvalds */ 12121da177e4SLinus Torvalds static void nfs_commit_rpcsetup(struct list_head *head, 1213788e7a89STrond Myklebust struct nfs_write_data *data, 1214788e7a89STrond Myklebust int how) 12151da177e4SLinus Torvalds { 12163da28eb1STrond Myklebust struct nfs_page *first; 12171da177e4SLinus Torvalds struct inode *inode; 1218788e7a89STrond Myklebust int flags; 12191da177e4SLinus Torvalds 12201da177e4SLinus Torvalds /* Set up the RPC argument and reply structs 12211da177e4SLinus Torvalds * NB: take care not to mess about with data->commit et al. */ 12221da177e4SLinus Torvalds 12231da177e4SLinus Torvalds list_splice_init(head, &data->pages); 12241da177e4SLinus Torvalds first = nfs_list_entry(data->pages.next); 12251da177e4SLinus Torvalds inode = first->wb_context->dentry->d_inode; 12261da177e4SLinus Torvalds 12271da177e4SLinus Torvalds data->inode = inode; 12281da177e4SLinus Torvalds data->cred = first->wb_context->cred; 12291da177e4SLinus Torvalds 12301da177e4SLinus Torvalds data->args.fh = NFS_FH(data->inode); 12313da28eb1STrond Myklebust /* Note: we always request a commit of the entire inode */ 12323da28eb1STrond Myklebust data->args.offset = 0; 12333da28eb1STrond Myklebust data->args.count = 0; 12343da28eb1STrond Myklebust data->res.count = 0; 12351da177e4SLinus Torvalds data->res.fattr = &data->fattr; 12361da177e4SLinus Torvalds data->res.verf = &data->verf; 12370e574af1STrond Myklebust nfs_fattr_init(&data->fattr); 12381da177e4SLinus Torvalds 1239788e7a89STrond Myklebust /* Set up the initial task struct. */ 1240788e7a89STrond Myklebust flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC; 1241788e7a89STrond Myklebust rpc_init_task(&data->task, NFS_CLIENT(inode), flags, &nfs_commit_ops, data); 12421da177e4SLinus Torvalds NFS_PROTO(inode)->commit_setup(data, how); 12431da177e4SLinus Torvalds 12441da177e4SLinus Torvalds data->task.tk_priority = flush_task_priority(how); 12451da177e4SLinus Torvalds data->task.tk_cookie = (unsigned long)inode; 12461da177e4SLinus Torvalds 1247a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 12481da177e4SLinus Torvalds } 12491da177e4SLinus Torvalds 12501da177e4SLinus Torvalds /* 12511da177e4SLinus Torvalds * Commit dirty pages 12521da177e4SLinus Torvalds */ 12531da177e4SLinus Torvalds static int 125440859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how) 12551da177e4SLinus Torvalds { 12561da177e4SLinus Torvalds struct nfs_write_data *data; 12571da177e4SLinus Torvalds struct nfs_page *req; 12581da177e4SLinus Torvalds 1259e9f7bee1STrond Myklebust data = nfs_commit_alloc(); 12601da177e4SLinus Torvalds 12611da177e4SLinus Torvalds if (!data) 12621da177e4SLinus Torvalds goto out_bad; 12631da177e4SLinus Torvalds 12641da177e4SLinus Torvalds /* Set up the argument struct */ 12651da177e4SLinus Torvalds nfs_commit_rpcsetup(head, data, how); 12661da177e4SLinus Torvalds 12671da177e4SLinus Torvalds nfs_execute_write(data); 12681da177e4SLinus Torvalds return 0; 12691da177e4SLinus Torvalds out_bad: 12701da177e4SLinus Torvalds while (!list_empty(head)) { 12711da177e4SLinus Torvalds req = nfs_list_entry(head->next); 12721da177e4SLinus Torvalds nfs_list_remove_request(req); 12731da177e4SLinus Torvalds nfs_mark_request_commit(req); 127483715ad5STrond Myklebust dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 12755c2d97cbSTrond Myklebust nfs_clear_page_writeback(req); 12761da177e4SLinus Torvalds } 12771da177e4SLinus Torvalds return -ENOMEM; 12781da177e4SLinus Torvalds } 12791da177e4SLinus Torvalds 12801da177e4SLinus Torvalds /* 12811da177e4SLinus Torvalds * COMMIT call returned 12821da177e4SLinus Torvalds */ 1283788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 12841da177e4SLinus Torvalds { 1285963d8fe5STrond Myklebust struct nfs_write_data *data = calldata; 12861da177e4SLinus Torvalds struct nfs_page *req; 12871da177e4SLinus Torvalds 1288a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 12891da177e4SLinus Torvalds task->tk_pid, task->tk_status); 12901da177e4SLinus Torvalds 1291788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1292788e7a89STrond Myklebust if (NFS_PROTO(data->inode)->commit_done(task, data) != 0) 1293788e7a89STrond Myklebust return; 1294788e7a89STrond Myklebust 12951da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 12961da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 12971da177e4SLinus Torvalds nfs_list_remove_request(req); 1298612c9384STrond Myklebust clear_bit(PG_NEED_COMMIT, &(req)->wb_flags); 1299fd39fc85SChristoph Lameter dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 13001da177e4SLinus Torvalds 13011da177e4SLinus Torvalds dprintk("NFS: commit (%s/%Ld %d@%Ld)", 13021da177e4SLinus Torvalds req->wb_context->dentry->d_inode->i_sb->s_id, 13031da177e4SLinus Torvalds (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 13041da177e4SLinus Torvalds req->wb_bytes, 13051da177e4SLinus Torvalds (long long)req_offset(req)); 13061da177e4SLinus Torvalds if (task->tk_status < 0) { 13071da177e4SLinus Torvalds req->wb_context->error = task->tk_status; 13081da177e4SLinus Torvalds nfs_inode_remove_request(req); 13091da177e4SLinus Torvalds dprintk(", error = %d\n", task->tk_status); 13101da177e4SLinus Torvalds goto next; 13111da177e4SLinus Torvalds } 13121da177e4SLinus Torvalds 13131da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 13141da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 13151da177e4SLinus Torvalds if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) { 13161da177e4SLinus Torvalds /* We have a match */ 13171da177e4SLinus Torvalds nfs_inode_remove_request(req); 13181da177e4SLinus Torvalds dprintk(" OK\n"); 13191da177e4SLinus Torvalds goto next; 13201da177e4SLinus Torvalds } 13211da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 13221da177e4SLinus Torvalds dprintk(" mismatch\n"); 132361822ab5STrond Myklebust nfs_redirty_request(req); 13241da177e4SLinus Torvalds next: 1325c6a556b8STrond Myklebust nfs_clear_page_writeback(req); 13261da177e4SLinus Torvalds } 13271da177e4SLinus Torvalds } 1328788e7a89STrond Myklebust 1329788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 1330788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1331788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1332788e7a89STrond Myklebust }; 1333c42de9ddSTrond Myklebust #else 1334c42de9ddSTrond Myklebust static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how) 1335c42de9ddSTrond Myklebust { 1336c42de9ddSTrond Myklebust return 0; 1337c42de9ddSTrond Myklebust } 13381da177e4SLinus Torvalds #endif 13391da177e4SLinus Torvalds 13403f442547STrond Myklebust static long nfs_flush_mapping(struct address_space *mapping, struct writeback_control *wbc, int how) 13411da177e4SLinus Torvalds { 134228c6925fSTrond Myklebust struct nfs_inode *nfsi = NFS_I(mapping->host); 13431da177e4SLinus Torvalds LIST_HEAD(head); 13443f442547STrond Myklebust long res; 13451da177e4SLinus Torvalds 13461da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 13473f442547STrond Myklebust res = nfs_scan_dirty(mapping, wbc, &head); 13481da177e4SLinus Torvalds spin_unlock(&nfsi->req_lock); 1349ab0a3dbeSTrond Myklebust if (res) { 135028c6925fSTrond Myklebust int error = nfs_flush_list(mapping->host, &head, res, how); 13511da177e4SLinus Torvalds if (error < 0) 13521da177e4SLinus Torvalds return error; 13537d46a49fSTrond Myklebust } 13541da177e4SLinus Torvalds return res; 13551da177e4SLinus Torvalds } 13561da177e4SLinus Torvalds 13571da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 13583da28eb1STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 13591da177e4SLinus Torvalds { 13601da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 13611da177e4SLinus Torvalds LIST_HEAD(head); 13627d46a49fSTrond Myklebust int res; 13631da177e4SLinus Torvalds 13641da177e4SLinus Torvalds spin_lock(&nfsi->req_lock); 13653da28eb1STrond Myklebust res = nfs_scan_commit(inode, &head, 0, 0); 13663da28eb1STrond Myklebust spin_unlock(&nfsi->req_lock); 13671da177e4SLinus Torvalds if (res) { 13687d46a49fSTrond Myklebust int error = nfs_commit_list(inode, &head, how); 13691da177e4SLinus Torvalds if (error < 0) 13701da177e4SLinus Torvalds return error; 13713da28eb1STrond Myklebust } 13721da177e4SLinus Torvalds return res; 13731da177e4SLinus Torvalds } 13741da177e4SLinus Torvalds #endif 13751da177e4SLinus Torvalds 13761c75950bSTrond Myklebust long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how) 13771da177e4SLinus Torvalds { 13781c75950bSTrond Myklebust struct inode *inode = mapping->host; 1379c42de9ddSTrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 13801c75950bSTrond Myklebust unsigned long idx_start, idx_end; 13811c75950bSTrond Myklebust unsigned int npages = 0; 1382c42de9ddSTrond Myklebust LIST_HEAD(head); 138370b9ecbdSTrond Myklebust int nocommit = how & FLUSH_NOCOMMIT; 13843f442547STrond Myklebust long pages, ret; 13851da177e4SLinus Torvalds 13861c75950bSTrond Myklebust /* FIXME */ 13871c75950bSTrond Myklebust if (wbc->range_cyclic) 13881c75950bSTrond Myklebust idx_start = 0; 13891c75950bSTrond Myklebust else { 13901c75950bSTrond Myklebust idx_start = wbc->range_start >> PAGE_CACHE_SHIFT; 13911c75950bSTrond Myklebust idx_end = wbc->range_end >> PAGE_CACHE_SHIFT; 13921c75950bSTrond Myklebust if (idx_end > idx_start) { 13931c75950bSTrond Myklebust unsigned long l_npages = 1 + idx_end - idx_start; 13941c75950bSTrond Myklebust npages = l_npages; 13951c75950bSTrond Myklebust if (sizeof(npages) != sizeof(l_npages) && 13961c75950bSTrond Myklebust (unsigned long)npages != l_npages) 13971c75950bSTrond Myklebust npages = 0; 13981c75950bSTrond Myklebust } 13991c75950bSTrond Myklebust } 1400c42de9ddSTrond Myklebust how &= ~FLUSH_NOCOMMIT; 1401c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 14021da177e4SLinus Torvalds do { 14031c75950bSTrond Myklebust wbc->pages_skipped = 0; 1404c42de9ddSTrond Myklebust ret = nfs_wait_on_requests_locked(inode, idx_start, npages); 1405c42de9ddSTrond Myklebust if (ret != 0) 1406c42de9ddSTrond Myklebust continue; 14071c75950bSTrond Myklebust pages = nfs_scan_dirty(mapping, wbc, &head); 1408c42de9ddSTrond Myklebust if (pages != 0) { 1409c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 1410e8e058e8STrond Myklebust if (how & FLUSH_INVALIDATE) { 141183715ad5STrond Myklebust nfs_cancel_dirty_list(&head); 1412e8e058e8STrond Myklebust ret = pages; 1413e8e058e8STrond Myklebust } else 1414c42de9ddSTrond Myklebust ret = nfs_flush_list(inode, &head, pages, how); 1415c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 141670b9ecbdSTrond Myklebust continue; 141770b9ecbdSTrond Myklebust } 14181c75950bSTrond Myklebust if (wbc->pages_skipped != 0) 14191c75950bSTrond Myklebust continue; 1420c42de9ddSTrond Myklebust if (nocommit) 1421c42de9ddSTrond Myklebust break; 1422d2ccddf0STrond Myklebust pages = nfs_scan_commit(inode, &head, idx_start, npages); 14231c75950bSTrond Myklebust if (pages == 0) { 14241c75950bSTrond Myklebust if (wbc->pages_skipped != 0) 14251c75950bSTrond Myklebust continue; 1426c42de9ddSTrond Myklebust break; 14271c75950bSTrond Myklebust } 1428d2ccddf0STrond Myklebust if (how & FLUSH_INVALIDATE) { 1429d2ccddf0STrond Myklebust spin_unlock(&nfsi->req_lock); 143083715ad5STrond Myklebust nfs_cancel_commit_list(&head); 1431e8e058e8STrond Myklebust ret = pages; 1432d2ccddf0STrond Myklebust spin_lock(&nfsi->req_lock); 1433d2ccddf0STrond Myklebust continue; 1434d2ccddf0STrond Myklebust } 1435d2ccddf0STrond Myklebust pages += nfs_scan_commit(inode, &head, 0, 0); 1436c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 1437c42de9ddSTrond Myklebust ret = nfs_commit_list(inode, &head, how); 1438c42de9ddSTrond Myklebust spin_lock(&nfsi->req_lock); 1439c42de9ddSTrond Myklebust } while (ret >= 0); 1440c42de9ddSTrond Myklebust spin_unlock(&nfsi->req_lock); 1441c42de9ddSTrond Myklebust return ret; 14421da177e4SLinus Torvalds } 14431da177e4SLinus Torvalds 14441c75950bSTrond Myklebust /* 14451c75950bSTrond Myklebust * flush the inode to disk. 14461c75950bSTrond Myklebust */ 14471c75950bSTrond Myklebust int nfs_wb_all(struct inode *inode) 14481c75950bSTrond Myklebust { 14491c75950bSTrond Myklebust struct address_space *mapping = inode->i_mapping; 14501c75950bSTrond Myklebust struct writeback_control wbc = { 14511c75950bSTrond Myklebust .bdi = mapping->backing_dev_info, 14521c75950bSTrond Myklebust .sync_mode = WB_SYNC_ALL, 14531c75950bSTrond Myklebust .nr_to_write = LONG_MAX, 145461822ab5STrond Myklebust .for_writepages = 1, 14551c75950bSTrond Myklebust .range_cyclic = 1, 14561c75950bSTrond Myklebust }; 14571c75950bSTrond Myklebust int ret; 14581c75950bSTrond Myklebust 145961822ab5STrond Myklebust ret = generic_writepages(mapping, &wbc); 146061822ab5STrond Myklebust if (ret < 0) 146161822ab5STrond Myklebust goto out; 14621c75950bSTrond Myklebust ret = nfs_sync_mapping_wait(mapping, &wbc, 0); 14631c75950bSTrond Myklebust if (ret >= 0) 14641c75950bSTrond Myklebust return 0; 146561822ab5STrond Myklebust out: 1466e507d9ebSTrond Myklebust __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 14671c75950bSTrond Myklebust return ret; 14681c75950bSTrond Myklebust } 14691c75950bSTrond Myklebust 14701c75950bSTrond Myklebust int nfs_sync_mapping_range(struct address_space *mapping, loff_t range_start, loff_t range_end, int how) 14711c75950bSTrond Myklebust { 14721c75950bSTrond Myklebust struct writeback_control wbc = { 14731c75950bSTrond Myklebust .bdi = mapping->backing_dev_info, 14741c75950bSTrond Myklebust .sync_mode = WB_SYNC_ALL, 14751c75950bSTrond Myklebust .nr_to_write = LONG_MAX, 14761c75950bSTrond Myklebust .range_start = range_start, 14771c75950bSTrond Myklebust .range_end = range_end, 147861822ab5STrond Myklebust .for_writepages = 1, 14791c75950bSTrond Myklebust }; 14801c75950bSTrond Myklebust int ret; 14811c75950bSTrond Myklebust 148261822ab5STrond Myklebust if (!(how & FLUSH_NOWRITEPAGE)) { 148361822ab5STrond Myklebust ret = generic_writepages(mapping, &wbc); 148461822ab5STrond Myklebust if (ret < 0) 148561822ab5STrond Myklebust goto out; 148661822ab5STrond Myklebust } 14871c75950bSTrond Myklebust ret = nfs_sync_mapping_wait(mapping, &wbc, how); 14881c75950bSTrond Myklebust if (ret >= 0) 14891c75950bSTrond Myklebust return 0; 149061822ab5STrond Myklebust out: 1491e507d9ebSTrond Myklebust __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 14921c75950bSTrond Myklebust return ret; 14931c75950bSTrond Myklebust } 14941c75950bSTrond Myklebust 149561822ab5STrond Myklebust int nfs_wb_page_priority(struct inode *inode, struct page *page, int how) 14961c75950bSTrond Myklebust { 14971c75950bSTrond Myklebust loff_t range_start = page_offset(page); 14981c75950bSTrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 14994d770ccfSTrond Myklebust struct writeback_control wbc = { 15004d770ccfSTrond Myklebust .bdi = page->mapping->backing_dev_info, 15014d770ccfSTrond Myklebust .sync_mode = WB_SYNC_ALL, 15024d770ccfSTrond Myklebust .nr_to_write = LONG_MAX, 15034d770ccfSTrond Myklebust .range_start = range_start, 15044d770ccfSTrond Myklebust .range_end = range_end, 15054d770ccfSTrond Myklebust }; 15064d770ccfSTrond Myklebust int ret; 15071c75950bSTrond Myklebust 15084d770ccfSTrond Myklebust BUG_ON(!PageLocked(page)); 15094d770ccfSTrond Myklebust if (!(how & FLUSH_NOWRITEPAGE) && clear_page_dirty_for_io(page)) { 15104d770ccfSTrond Myklebust ret = nfs_writepage_locked(page, &wbc); 15114d770ccfSTrond Myklebust if (ret < 0) 15124d770ccfSTrond Myklebust goto out; 15134d770ccfSTrond Myklebust } 1514f40313acSTrond Myklebust if (!PagePrivate(page)) 1515f40313acSTrond Myklebust return 0; 15164d770ccfSTrond Myklebust ret = nfs_sync_mapping_wait(page->mapping, &wbc, how); 15174d770ccfSTrond Myklebust if (ret >= 0) 15184d770ccfSTrond Myklebust return 0; 15194d770ccfSTrond Myklebust out: 1520e507d9ebSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_PAGES); 15214d770ccfSTrond Myklebust return ret; 15221c75950bSTrond Myklebust } 15231c75950bSTrond Myklebust 15241c75950bSTrond Myklebust /* 15251c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 15261c75950bSTrond Myklebust */ 15271c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page* page) 15281c75950bSTrond Myklebust { 15294d770ccfSTrond Myklebust return nfs_wb_page_priority(inode, page, FLUSH_STABLE); 15301c75950bSTrond Myklebust } 15311c75950bSTrond Myklebust 15321a54533eSTrond Myklebust int nfs_set_page_dirty(struct page *page) 15331a54533eSTrond Myklebust { 1534*2b82f190STrond Myklebust spinlock_t *req_lock = &NFS_I(page->mapping->host)->req_lock; 15351a54533eSTrond Myklebust struct nfs_page *req; 1536*2b82f190STrond Myklebust int ret; 15371a54533eSTrond Myklebust 1538*2b82f190STrond Myklebust spin_lock(req_lock); 1539*2b82f190STrond Myklebust req = nfs_page_find_request_locked(page); 15401a54533eSTrond Myklebust if (req != NULL) { 15411a54533eSTrond Myklebust /* Mark any existing write requests for flushing */ 1542*2b82f190STrond Myklebust ret = !test_and_set_bit(PG_NEED_FLUSH, &req->wb_flags); 1543*2b82f190STrond Myklebust spin_unlock(req_lock); 15441a54533eSTrond Myklebust nfs_release_request(req); 1545*2b82f190STrond Myklebust return ret; 15461a54533eSTrond Myklebust } 1547*2b82f190STrond Myklebust ret = __set_page_dirty_nobuffers(page); 1548*2b82f190STrond Myklebust spin_unlock(req_lock); 1549*2b82f190STrond Myklebust return ret; 15501a54533eSTrond Myklebust } 15511a54533eSTrond Myklebust 15521c75950bSTrond Myklebust 1553f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 15541da177e4SLinus Torvalds { 15551da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 15561da177e4SLinus Torvalds sizeof(struct nfs_write_data), 15571da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 15581da177e4SLinus Torvalds NULL, NULL); 15591da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 15601da177e4SLinus Torvalds return -ENOMEM; 15611da177e4SLinus Torvalds 156293d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 15631da177e4SLinus Torvalds nfs_wdata_cachep); 15641da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 15651da177e4SLinus Torvalds return -ENOMEM; 15661da177e4SLinus Torvalds 156793d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 15681da177e4SLinus Torvalds nfs_wdata_cachep); 15691da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 15701da177e4SLinus Torvalds return -ENOMEM; 15711da177e4SLinus Torvalds 157289a09141SPeter Zijlstra /* 157389a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 157489a09141SPeter Zijlstra * 157589a09141SPeter Zijlstra * 64MB: 8192k 157689a09141SPeter Zijlstra * 128MB: 11585k 157789a09141SPeter Zijlstra * 256MB: 16384k 157889a09141SPeter Zijlstra * 512MB: 23170k 157989a09141SPeter Zijlstra * 1GB: 32768k 158089a09141SPeter Zijlstra * 2GB: 46340k 158189a09141SPeter Zijlstra * 4GB: 65536k 158289a09141SPeter Zijlstra * 8GB: 92681k 158389a09141SPeter Zijlstra * 16GB: 131072k 158489a09141SPeter Zijlstra * 158589a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 158689a09141SPeter Zijlstra * Limit the default to 256M 158789a09141SPeter Zijlstra */ 158889a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 158989a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 159089a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 159189a09141SPeter Zijlstra 15921da177e4SLinus Torvalds return 0; 15931da177e4SLinus Torvalds } 15941da177e4SLinus Torvalds 1595266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 15961da177e4SLinus Torvalds { 15971da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 15981da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 15991a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 16001da177e4SLinus Torvalds } 16011da177e4SLinus Torvalds 1602