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; 49d4581383SWeston Andros Adamson static void nfs_clear_request_commit(struct nfs_page *req); 50*02d1426cSWeston Andros Adamson static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 51*02d1426cSWeston Andros Adamson struct inode *inode); 521da177e4SLinus Torvalds 53e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 543feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 550b7c0153SFred Isaman static struct kmem_cache *nfs_cdata_cachep; 561da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 571da177e4SLinus Torvalds 580b7c0153SFred Isaman struct nfs_commit_data *nfs_commitdata_alloc(void) 591da177e4SLinus Torvalds { 60192e501bSMel Gorman struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); 6140859d7eSChuck Lever 621da177e4SLinus Torvalds if (p) { 631da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 641da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 651da177e4SLinus Torvalds } 661da177e4SLinus Torvalds return p; 671da177e4SLinus Torvalds } 68e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 691da177e4SLinus Torvalds 700b7c0153SFred Isaman void nfs_commit_free(struct nfs_commit_data *p) 711da177e4SLinus Torvalds { 721da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 731da177e4SLinus Torvalds } 74e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 751da177e4SLinus Torvalds 761e7f3a48SWeston Andros Adamson static struct nfs_pgio_header *nfs_writehdr_alloc(void) 773feb2d49STrond Myklebust { 781e7f3a48SWeston Andros Adamson struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); 793feb2d49STrond Myklebust 804a0de55cSAnna Schumaker if (p) 813feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 823feb2d49STrond Myklebust return p; 833feb2d49STrond Myklebust } 843feb2d49STrond Myklebust 851e7f3a48SWeston Andros Adamson static void nfs_writehdr_free(struct nfs_pgio_header *hdr) 866c75dc0dSFred Isaman { 871e7f3a48SWeston Andros Adamson mempool_free(hdr, nfs_wdata_mempool); 883feb2d49STrond Myklebust } 891da177e4SLinus Torvalds 907b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 917b159fc1STrond Myklebust { 927b159fc1STrond Myklebust ctx->error = error; 937b159fc1STrond Myklebust smp_wmb(); 947b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 957b159fc1STrond Myklebust } 967b159fc1STrond Myklebust 9784d3a9a9SWeston Andros Adamson /* 98*02d1426cSWeston Andros Adamson * nfs_page_search_commits_for_head_request_locked 99*02d1426cSWeston Andros Adamson * 100*02d1426cSWeston Andros Adamson * Search through commit lists on @inode for the head request for @page. 101*02d1426cSWeston Andros Adamson * Must be called while holding the inode (which is cinfo) lock. 102*02d1426cSWeston Andros Adamson * 103*02d1426cSWeston Andros Adamson * Returns the head request if found, or NULL if not found. 104*02d1426cSWeston Andros Adamson */ 105*02d1426cSWeston Andros Adamson static struct nfs_page * 106*02d1426cSWeston Andros Adamson nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 107*02d1426cSWeston Andros Adamson struct page *page) 108*02d1426cSWeston Andros Adamson { 109*02d1426cSWeston Andros Adamson struct nfs_page *freq, *t; 110*02d1426cSWeston Andros Adamson struct nfs_commit_info cinfo; 111*02d1426cSWeston Andros Adamson struct inode *inode = &nfsi->vfs_inode; 112*02d1426cSWeston Andros Adamson 113*02d1426cSWeston Andros Adamson nfs_init_cinfo_from_inode(&cinfo, inode); 114*02d1426cSWeston Andros Adamson 115*02d1426cSWeston Andros Adamson /* search through pnfs commit lists */ 116*02d1426cSWeston Andros Adamson freq = pnfs_search_commit_reqs(inode, &cinfo, page); 117*02d1426cSWeston Andros Adamson if (freq) 118*02d1426cSWeston Andros Adamson return freq->wb_head; 119*02d1426cSWeston Andros Adamson 120*02d1426cSWeston Andros Adamson /* Linearly search the commit list for the correct request */ 121*02d1426cSWeston Andros Adamson list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 122*02d1426cSWeston Andros Adamson if (freq->wb_page == page) 123*02d1426cSWeston Andros Adamson return freq->wb_head; 124*02d1426cSWeston Andros Adamson } 125*02d1426cSWeston Andros Adamson 126*02d1426cSWeston Andros Adamson return NULL; 127*02d1426cSWeston Andros Adamson } 128*02d1426cSWeston Andros Adamson 129*02d1426cSWeston Andros Adamson /* 13084d3a9a9SWeston Andros Adamson * nfs_page_find_head_request_locked - find head request associated with @page 13184d3a9a9SWeston Andros Adamson * 13284d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 13384d3a9a9SWeston Andros Adamson * 13484d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 13584d3a9a9SWeston Andros Adamson */ 13629418aa4SMel Gorman static struct nfs_page * 13784d3a9a9SWeston Andros Adamson nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page) 138277459d2STrond Myklebust { 139277459d2STrond Myklebust struct nfs_page *req = NULL; 140277459d2STrond Myklebust 14129418aa4SMel Gorman if (PagePrivate(page)) 142277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 143*02d1426cSWeston Andros Adamson else if (unlikely(PageSwapCache(page))) 144*02d1426cSWeston Andros Adamson req = nfs_page_search_commits_for_head_request_locked(nfsi, 145*02d1426cSWeston Andros Adamson page); 14629418aa4SMel Gorman 14784d3a9a9SWeston Andros Adamson if (req) { 14884d3a9a9SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 14929418aa4SMel Gorman kref_get(&req->wb_kref); 15084d3a9a9SWeston Andros Adamson } 15129418aa4SMel Gorman 152277459d2STrond Myklebust return req; 153277459d2STrond Myklebust } 154277459d2STrond Myklebust 15584d3a9a9SWeston Andros Adamson /* 15684d3a9a9SWeston Andros Adamson * nfs_page_find_head_request - find head request associated with @page 15784d3a9a9SWeston Andros Adamson * 15884d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 15984d3a9a9SWeston Andros Adamson */ 16084d3a9a9SWeston Andros Adamson static struct nfs_page *nfs_page_find_head_request(struct page *page) 161277459d2STrond Myklebust { 162d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 163277459d2STrond Myklebust struct nfs_page *req = NULL; 164277459d2STrond Myklebust 165587142f8STrond Myklebust spin_lock(&inode->i_lock); 16684d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 167587142f8STrond Myklebust spin_unlock(&inode->i_lock); 168277459d2STrond Myklebust return req; 169277459d2STrond Myklebust } 170277459d2STrond Myklebust 1711da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1721da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1731da177e4SLinus Torvalds { 174d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 175a3d01454STrond Myklebust loff_t end, i_size; 176a3d01454STrond Myklebust pgoff_t end_index; 1771da177e4SLinus Torvalds 178a3d01454STrond Myklebust spin_lock(&inode->i_lock); 179a3d01454STrond Myklebust i_size = i_size_read(inode); 180a3d01454STrond Myklebust end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 181d56b4ddfSMel Gorman if (i_size > 0 && page_file_index(page) < end_index) 182a3d01454STrond Myklebust goto out; 183d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 1841da177e4SLinus Torvalds if (i_size >= end) 185a3d01454STrond Myklebust goto out; 1861da177e4SLinus Torvalds i_size_write(inode, end); 187a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 188a3d01454STrond Myklebust out: 189a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 1901da177e4SLinus Torvalds } 1911da177e4SLinus Torvalds 192a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 193a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 194a301b777STrond Myklebust { 195d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 196a301b777STrond Myklebust } 197a301b777STrond Myklebust 198d72ddcbaSWeston Andros Adamson /* 199d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 200d72ddcbaSWeston Andros Adamson * @head - head request of page group 201d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 202d72ddcbaSWeston Andros Adamson * 203d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 204d72ddcbaSWeston Andros Adamson * page offset @page_offset. 205d72ddcbaSWeston Andros Adamson * 206d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 207d72ddcbaSWeston Andros Adamson * match is found. 208d72ddcbaSWeston Andros Adamson * 209d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 210d72ddcbaSWeston Andros Adamson */ 211d72ddcbaSWeston Andros Adamson static struct nfs_page * 212d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 213d72ddcbaSWeston Andros Adamson { 214d72ddcbaSWeston Andros Adamson struct nfs_page *req; 215d72ddcbaSWeston Andros Adamson 216d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(head != head->wb_head); 217d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags)); 218d72ddcbaSWeston Andros Adamson 219d72ddcbaSWeston Andros Adamson req = head; 220d72ddcbaSWeston Andros Adamson do { 221d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 222d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 223d72ddcbaSWeston Andros Adamson return req; 224d72ddcbaSWeston Andros Adamson 225d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 226d72ddcbaSWeston Andros Adamson } while (req != head); 227d72ddcbaSWeston Andros Adamson 228d72ddcbaSWeston Andros Adamson return NULL; 229d72ddcbaSWeston Andros Adamson } 230d72ddcbaSWeston Andros Adamson 231d72ddcbaSWeston Andros Adamson /* 232d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 233d72ddcbaSWeston Andros Adamson * @head - head request of page group 234d72ddcbaSWeston Andros Adamson * 235d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 236d72ddcbaSWeston Andros Adamson * returns false otherwise 237d72ddcbaSWeston Andros Adamson */ 238d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 239d72ddcbaSWeston Andros Adamson { 240d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 241d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 242d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 243d72ddcbaSWeston Andros Adamson 244e7029206SWeston Andros Adamson nfs_page_group_lock(req, true); 245d72ddcbaSWeston Andros Adamson 246d72ddcbaSWeston Andros Adamson do { 247d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 248d72ddcbaSWeston Andros Adamson if (tmp) { 249d72ddcbaSWeston Andros Adamson /* no way this should happen */ 250d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(tmp->wb_pgbase != pos); 251d72ddcbaSWeston Andros Adamson pos += tmp->wb_bytes - (pos - tmp->wb_pgbase); 252d72ddcbaSWeston Andros Adamson } 253d72ddcbaSWeston Andros Adamson } while (tmp && pos < len); 254d72ddcbaSWeston Andros Adamson 255d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 256d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(pos > len); 257d72ddcbaSWeston Andros Adamson return pos == len; 258d72ddcbaSWeston Andros Adamson } 259d72ddcbaSWeston Andros Adamson 2601da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 2611da177e4SLinus Torvalds * covers the full page. 2621da177e4SLinus Torvalds */ 263d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 2641da177e4SLinus Torvalds { 265d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 2661da177e4SLinus Torvalds return; 267d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 2681da177e4SLinus Torvalds return; 269d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 2701da177e4SLinus Torvalds } 2711da177e4SLinus Torvalds 2721da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 2731da177e4SLinus Torvalds { 2741da177e4SLinus Torvalds if (wbc->for_reclaim) 275c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 276b17621feSWu Fengguang if (wbc->for_kupdate || wbc->for_background) 277b31268acSTrond Myklebust return FLUSH_LOWPRI | FLUSH_COND_STABLE; 278b31268acSTrond Myklebust return FLUSH_COND_STABLE; 2791da177e4SLinus Torvalds } 2801da177e4SLinus Torvalds 2811da177e4SLinus Torvalds /* 28289a09141SPeter Zijlstra * NFS congestion control 28389a09141SPeter Zijlstra */ 28489a09141SPeter Zijlstra 28589a09141SPeter Zijlstra int nfs_congestion_kb; 28689a09141SPeter Zijlstra 28789a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 28889a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 28989a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 29089a09141SPeter Zijlstra 291deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 29289a09141SPeter Zijlstra { 293deed85e7STrond Myklebust struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); 2945a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2955a6d41b3STrond Myklebust 296deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 29789a09141SPeter Zijlstra 298277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 2998aa7e847SJens Axboe NFS_CONGESTION_ON_THRESH) { 3008aa7e847SJens Axboe set_bdi_congested(&nfss->backing_dev_info, 3018aa7e847SJens Axboe BLK_RW_ASYNC); 3028aa7e847SJens Axboe } 30389a09141SPeter Zijlstra } 30489a09141SPeter Zijlstra 30520633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 30689a09141SPeter Zijlstra { 30720633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 30889a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 30989a09141SPeter Zijlstra 31020633f04SWeston Andros Adamson if (!nfs_page_group_sync_on_bit(req, PG_WB_END)) 31120633f04SWeston Andros Adamson return; 31220633f04SWeston Andros Adamson 31320633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 314c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 3158aa7e847SJens Axboe clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 31689a09141SPeter Zijlstra } 31789a09141SPeter Zijlstra 318d4581383SWeston Andros Adamson 319d4581383SWeston Andros Adamson /* nfs_page_group_clear_bits 320d4581383SWeston Andros Adamson * @req - an nfs request 321d4581383SWeston Andros Adamson * clears all page group related bits from @req 322d4581383SWeston Andros Adamson */ 323d4581383SWeston Andros Adamson static void 324d4581383SWeston Andros Adamson nfs_page_group_clear_bits(struct nfs_page *req) 325e261f51fSTrond Myklebust { 326d4581383SWeston Andros Adamson clear_bit(PG_TEARDOWN, &req->wb_flags); 327d4581383SWeston Andros Adamson clear_bit(PG_UNLOCKPAGE, &req->wb_flags); 328d4581383SWeston Andros Adamson clear_bit(PG_UPTODATE, &req->wb_flags); 329d4581383SWeston Andros Adamson clear_bit(PG_WB_END, &req->wb_flags); 330d4581383SWeston Andros Adamson clear_bit(PG_REMOVE, &req->wb_flags); 331d4581383SWeston Andros Adamson } 332d4581383SWeston Andros Adamson 333d4581383SWeston Andros Adamson 334d4581383SWeston Andros Adamson /* 335d4581383SWeston Andros Adamson * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req 336d4581383SWeston Andros Adamson * 337d4581383SWeston Andros Adamson * this is a helper function for nfs_lock_and_join_requests 338d4581383SWeston Andros Adamson * 339d4581383SWeston Andros Adamson * @inode - inode associated with request page group, must be holding inode lock 340d4581383SWeston Andros Adamson * @head - head request of page group, must be holding head lock 341d4581383SWeston Andros Adamson * @req - request that couldn't lock and needs to wait on the req bit lock 342d4581383SWeston Andros Adamson * @nonblock - if true, don't actually wait 343d4581383SWeston Andros Adamson * 344d4581383SWeston Andros Adamson * NOTE: this must be called holding page_group bit lock and inode spin lock 345d4581383SWeston Andros Adamson * and BOTH will be released before returning. 346d4581383SWeston Andros Adamson * 347d4581383SWeston Andros Adamson * returns 0 on success, < 0 on error. 348d4581383SWeston Andros Adamson */ 349d4581383SWeston Andros Adamson static int 350d4581383SWeston Andros Adamson nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head, 351d4581383SWeston Andros Adamson struct nfs_page *req, bool nonblock) 352d4581383SWeston Andros Adamson __releases(&inode->i_lock) 353d4581383SWeston Andros Adamson { 354d4581383SWeston Andros Adamson struct nfs_page *tmp; 355e261f51fSTrond Myklebust int ret; 356e261f51fSTrond Myklebust 357d4581383SWeston Andros Adamson /* relinquish all the locks successfully grabbed this run */ 358d4581383SWeston Andros Adamson for (tmp = head ; tmp != req; tmp = tmp->wb_this_page) 359d4581383SWeston Andros Adamson nfs_unlock_request(tmp); 360d4581383SWeston Andros Adamson 361d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags)); 362d4581383SWeston Andros Adamson 363d4581383SWeston Andros Adamson /* grab a ref on the request that will be waited on */ 364d4581383SWeston Andros Adamson kref_get(&req->wb_kref); 365d4581383SWeston Andros Adamson 366d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 367587142f8STrond Myklebust spin_unlock(&inode->i_lock); 368d4581383SWeston Andros Adamson 369d4581383SWeston Andros Adamson /* release ref from nfs_page_find_head_request_locked */ 370d4581383SWeston Andros Adamson nfs_release_request(head); 371d4581383SWeston Andros Adamson 372cfb506e1STrond Myklebust if (!nonblock) 373e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 374cfb506e1STrond Myklebust else 375cfb506e1STrond Myklebust ret = -EAGAIN; 376e261f51fSTrond Myklebust nfs_release_request(req); 377d4581383SWeston Andros Adamson 378d4581383SWeston Andros Adamson return ret; 379e261f51fSTrond Myklebust } 380d4581383SWeston Andros Adamson 381d4581383SWeston Andros Adamson /* 382d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 383d4581383SWeston Andros Adamson * 384d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 385d4581383SWeston Andros Adamson * @old_head - the old head of the list 386d4581383SWeston Andros Adamson * 387d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 388d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 389d4581383SWeston Andros Adamson * with a reference held by some other context. 390d4581383SWeston Andros Adamson */ 391d4581383SWeston Andros Adamson static void 392d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 393d4581383SWeston Andros Adamson struct nfs_page *old_head) 394d4581383SWeston Andros Adamson { 395d4581383SWeston Andros Adamson while (destroy_list) { 396d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 397d4581383SWeston Andros Adamson 398d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 399d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 400d4581383SWeston Andros Adamson 401d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 402d4581383SWeston Andros Adamson 403d4581383SWeston Andros Adamson /* make sure old group is not used */ 404d4581383SWeston Andros Adamson subreq->wb_head = subreq; 405d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 406d4581383SWeston Andros Adamson 407d4581383SWeston Andros Adamson nfs_clear_request_commit(subreq); 408d4581383SWeston Andros Adamson 409d4581383SWeston Andros Adamson /* subreq is now totally disconnected from page group or any 410d4581383SWeston Andros Adamson * write / commit lists. last chance to wake any waiters */ 411d4581383SWeston Andros Adamson nfs_unlock_request(subreq); 412d4581383SWeston Andros Adamson 413d4581383SWeston Andros Adamson if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) { 414d4581383SWeston Andros Adamson /* release ref on old head request */ 415d4581383SWeston Andros Adamson nfs_release_request(old_head); 416d4581383SWeston Andros Adamson 417d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 418d4581383SWeston Andros Adamson 419d4581383SWeston Andros Adamson /* release the PG_INODE_REF reference */ 420d4581383SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) 421d4581383SWeston Andros Adamson nfs_release_request(subreq); 422d4581383SWeston Andros Adamson else 423d4581383SWeston Andros Adamson WARN_ON_ONCE(1); 424d4581383SWeston Andros Adamson } else { 425d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags)); 426d4581383SWeston Andros Adamson /* zombie requests have already released the last 427d4581383SWeston Andros Adamson * reference and were waiting on the rest of the 428d4581383SWeston Andros Adamson * group to complete. Since it's no longer part of a 429d4581383SWeston Andros Adamson * group, simply free the request */ 430d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 431d4581383SWeston Andros Adamson nfs_free_request(subreq); 432d4581383SWeston Andros Adamson } 433d4581383SWeston Andros Adamson } 434d4581383SWeston Andros Adamson } 435d4581383SWeston Andros Adamson 436d4581383SWeston Andros Adamson /* 437d4581383SWeston Andros Adamson * nfs_lock_and_join_requests - join all subreqs to the head req and return 438d4581383SWeston Andros Adamson * a locked reference, cancelling any pending 439d4581383SWeston Andros Adamson * operations for this page. 440d4581383SWeston Andros Adamson * 441d4581383SWeston Andros Adamson * @page - the page used to lookup the "page group" of nfs_page structures 442d4581383SWeston Andros Adamson * @nonblock - if true, don't block waiting for request locks 443d4581383SWeston Andros Adamson * 444d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 445d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 446d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 447d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 448d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 449d4581383SWeston Andros Adamson * 450d4581383SWeston Andros Adamson * Returns a locked, referenced pointer to the head request - which after 451d4581383SWeston Andros Adamson * this call is guaranteed to be the only request associated with the page. 452d4581383SWeston Andros Adamson * Returns NULL if no requests are found for @page, or a ERR_PTR if an 453d4581383SWeston Andros Adamson * error was encountered. 454d4581383SWeston Andros Adamson */ 455d4581383SWeston Andros Adamson static struct nfs_page * 456d4581383SWeston Andros Adamson nfs_lock_and_join_requests(struct page *page, bool nonblock) 457d4581383SWeston Andros Adamson { 458d4581383SWeston Andros Adamson struct inode *inode = page_file_mapping(page)->host; 459d4581383SWeston Andros Adamson struct nfs_page *head, *subreq; 460d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 461d4581383SWeston Andros Adamson unsigned int total_bytes; 462d4581383SWeston Andros Adamson int ret; 463d4581383SWeston Andros Adamson 464d4581383SWeston Andros Adamson try_again: 465d4581383SWeston Andros Adamson total_bytes = 0; 466d4581383SWeston Andros Adamson 467d4581383SWeston Andros Adamson WARN_ON_ONCE(destroy_list); 468d4581383SWeston Andros Adamson 469d4581383SWeston Andros Adamson spin_lock(&inode->i_lock); 470d4581383SWeston Andros Adamson 471d4581383SWeston Andros Adamson /* 472d4581383SWeston Andros Adamson * A reference is taken only on the head request which acts as a 473d4581383SWeston Andros Adamson * reference to the whole page group - the group will not be destroyed 474d4581383SWeston Andros Adamson * until the head reference is released. 475d4581383SWeston Andros Adamson */ 476d4581383SWeston Andros Adamson head = nfs_page_find_head_request_locked(NFS_I(inode), page); 477d4581383SWeston Andros Adamson 478d4581383SWeston Andros Adamson if (!head) { 479587142f8STrond Myklebust spin_unlock(&inode->i_lock); 480d4581383SWeston Andros Adamson return NULL; 481d4581383SWeston Andros Adamson } 482d4581383SWeston Andros Adamson 483d4581383SWeston Andros Adamson /* lock each request in the page group */ 484e7029206SWeston Andros Adamson ret = nfs_page_group_lock(head, false); 485e7029206SWeston Andros Adamson if (ret < 0) 486e7029206SWeston Andros Adamson return ERR_PTR(ret); 487d4581383SWeston Andros Adamson subreq = head; 488d4581383SWeston Andros Adamson do { 489d4581383SWeston Andros Adamson /* 490d4581383SWeston Andros Adamson * Subrequests are always contiguous, non overlapping 491d4581383SWeston Andros Adamson * and in order. If not, it's a programming error. 492d4581383SWeston Andros Adamson */ 493d4581383SWeston Andros Adamson WARN_ON_ONCE(subreq->wb_offset != 494d4581383SWeston Andros Adamson (head->wb_offset + total_bytes)); 495d4581383SWeston Andros Adamson 496d4581383SWeston Andros Adamson /* keep track of how many bytes this group covers */ 497d4581383SWeston Andros Adamson total_bytes += subreq->wb_bytes; 498d4581383SWeston Andros Adamson 499d4581383SWeston Andros Adamson if (!nfs_lock_request(subreq)) { 500d4581383SWeston Andros Adamson /* releases page group bit lock and 501d4581383SWeston Andros Adamson * inode spin lock and all references */ 502d4581383SWeston Andros Adamson ret = nfs_unroll_locks_and_wait(inode, head, 503d4581383SWeston Andros Adamson subreq, nonblock); 504d4581383SWeston Andros Adamson 505d4581383SWeston Andros Adamson if (ret == 0) 506d4581383SWeston Andros Adamson goto try_again; 507d4581383SWeston Andros Adamson 508d4581383SWeston Andros Adamson return ERR_PTR(ret); 509d4581383SWeston Andros Adamson } 510d4581383SWeston Andros Adamson 511d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 512d4581383SWeston Andros Adamson } while (subreq != head); 513d4581383SWeston Andros Adamson 514d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 515d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 516d4581383SWeston Andros Adamson subreq = head; 517d4581383SWeston Andros Adamson do { 518d4581383SWeston Andros Adamson nfs_list_remove_request(subreq); 519d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 520d4581383SWeston Andros Adamson } while (subreq != head); 521d4581383SWeston Andros Adamson 522d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 523d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 524d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 525d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 526d4581383SWeston Andros Adamson head->wb_this_page = head; 527d4581383SWeston Andros Adamson 528d4581383SWeston Andros Adamson /* change head request to cover whole range that 529d4581383SWeston Andros Adamson * the former page group covered */ 530d4581383SWeston Andros Adamson head->wb_bytes = total_bytes; 531d4581383SWeston Andros Adamson } 532d4581383SWeston Andros Adamson 533d4581383SWeston Andros Adamson /* 534d4581383SWeston Andros Adamson * prepare head request to be added to new pgio descriptor 535d4581383SWeston Andros Adamson */ 536d4581383SWeston Andros Adamson nfs_page_group_clear_bits(head); 537d4581383SWeston Andros Adamson 538d4581383SWeston Andros Adamson /* 539d4581383SWeston Andros Adamson * some part of the group was still on the inode list - otherwise 540d4581383SWeston Andros Adamson * the group wouldn't be involved in async write. 541d4581383SWeston Andros Adamson * grab a reference for the head request, iff it needs one. 542d4581383SWeston Andros Adamson */ 543d4581383SWeston Andros Adamson if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags)) 544d4581383SWeston Andros Adamson kref_get(&head->wb_kref); 545d4581383SWeston Andros Adamson 546d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 547d4581383SWeston Andros Adamson 548d4581383SWeston Andros Adamson /* drop lock to clear_request_commit the head req and clean up 549d4581383SWeston Andros Adamson * requests on destroy list */ 550d4581383SWeston Andros Adamson spin_unlock(&inode->i_lock); 551d4581383SWeston Andros Adamson 552d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(destroy_list, head); 553d4581383SWeston Andros Adamson 554d4581383SWeston Andros Adamson /* clean up commit list state */ 555d4581383SWeston Andros Adamson nfs_clear_request_commit(head); 556d4581383SWeston Andros Adamson 557d4581383SWeston Andros Adamson /* still holds ref on head from nfs_page_find_head_request_locked 558d4581383SWeston Andros Adamson * and still has lock on head from lock loop */ 559d4581383SWeston Andros Adamson return head; 560612c9384STrond Myklebust } 561074cc1deSTrond Myklebust 562074cc1deSTrond Myklebust /* 563074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 564074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 565074cc1deSTrond Myklebust */ 566074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 567cfb506e1STrond Myklebust struct page *page, bool nonblock) 568074cc1deSTrond Myklebust { 569074cc1deSTrond Myklebust struct nfs_page *req; 570074cc1deSTrond Myklebust int ret = 0; 571074cc1deSTrond Myklebust 572d4581383SWeston Andros Adamson req = nfs_lock_and_join_requests(page, nonblock); 573074cc1deSTrond Myklebust if (!req) 574074cc1deSTrond Myklebust goto out; 575074cc1deSTrond Myklebust ret = PTR_ERR(req); 576074cc1deSTrond Myklebust if (IS_ERR(req)) 577074cc1deSTrond Myklebust goto out; 578074cc1deSTrond Myklebust 579deed85e7STrond Myklebust nfs_set_page_writeback(page); 580deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 581074cc1deSTrond Myklebust 582deed85e7STrond Myklebust ret = 0; 583f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 584f8512ad0SFred Isaman nfs_redirty_request(req); 585074cc1deSTrond Myklebust ret = pgio->pg_error; 586f8512ad0SFred Isaman } 587074cc1deSTrond Myklebust out: 588074cc1deSTrond Myklebust return ret; 589e261f51fSTrond Myklebust } 590e261f51fSTrond Myklebust 591f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 592f758c885STrond Myklebust { 593d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 594cfb506e1STrond Myklebust int ret; 595f758c885STrond Myklebust 596f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 597f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 598f758c885STrond Myklebust 599d56b4ddfSMel Gorman nfs_pageio_cond_complete(pgio, page_file_index(page)); 6001b430beeSWu Fengguang ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); 601cfb506e1STrond Myklebust if (ret == -EAGAIN) { 602cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 603cfb506e1STrond Myklebust ret = 0; 604cfb506e1STrond Myklebust } 605cfb506e1STrond Myklebust return ret; 606f758c885STrond Myklebust } 607f758c885STrond Myklebust 608e261f51fSTrond Myklebust /* 6091da177e4SLinus Torvalds * Write an mmapped page to the server. 6101da177e4SLinus Torvalds */ 6114d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 6121da177e4SLinus Torvalds { 613f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 614e261f51fSTrond Myklebust int err; 6151da177e4SLinus Torvalds 616a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc), 617a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 618f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 619f758c885STrond Myklebust nfs_pageio_complete(&pgio); 620f758c885STrond Myklebust if (err < 0) 6214d770ccfSTrond Myklebust return err; 622f758c885STrond Myklebust if (pgio.pg_error < 0) 623f758c885STrond Myklebust return pgio.pg_error; 624f758c885STrond Myklebust return 0; 6254d770ccfSTrond Myklebust } 6264d770ccfSTrond Myklebust 6274d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 6284d770ccfSTrond Myklebust { 629f758c885STrond Myklebust int ret; 6304d770ccfSTrond Myklebust 631f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 6321da177e4SLinus Torvalds unlock_page(page); 633f758c885STrond Myklebust return ret; 634f758c885STrond Myklebust } 635f758c885STrond Myklebust 636f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 637f758c885STrond Myklebust { 638f758c885STrond Myklebust int ret; 639f758c885STrond Myklebust 640f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 641f758c885STrond Myklebust unlock_page(page); 642f758c885STrond Myklebust return ret; 6431da177e4SLinus Torvalds } 6441da177e4SLinus Torvalds 6451da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 6461da177e4SLinus Torvalds { 6471da177e4SLinus Torvalds struct inode *inode = mapping->host; 64872cb77f4STrond Myklebust unsigned long *bitlock = &NFS_I(inode)->flags; 649c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 6501da177e4SLinus Torvalds int err; 6511da177e4SLinus Torvalds 65272cb77f4STrond Myklebust /* Stop dirtying of new pages while we sync */ 65372cb77f4STrond Myklebust err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, 65472cb77f4STrond Myklebust nfs_wait_bit_killable, TASK_KILLABLE); 65572cb77f4STrond Myklebust if (err) 65672cb77f4STrond Myklebust goto out_err; 65772cb77f4STrond Myklebust 65891d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 65991d5b470SChuck Lever 660a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 661a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 662f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 663c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 66472cb77f4STrond Myklebust 66572cb77f4STrond Myklebust clear_bit_unlock(NFS_INO_FLUSHING, bitlock); 6664e857c58SPeter Zijlstra smp_mb__after_atomic(); 66772cb77f4STrond Myklebust wake_up_bit(bitlock, NFS_INO_FLUSHING); 66872cb77f4STrond Myklebust 669f758c885STrond Myklebust if (err < 0) 67072cb77f4STrond Myklebust goto out_err; 67172cb77f4STrond Myklebust err = pgio.pg_error; 67272cb77f4STrond Myklebust if (err < 0) 67372cb77f4STrond Myklebust goto out_err; 674c63c7b05STrond Myklebust return 0; 67572cb77f4STrond Myklebust out_err: 67672cb77f4STrond Myklebust return err; 6771da177e4SLinus Torvalds } 6781da177e4SLinus Torvalds 6791da177e4SLinus Torvalds /* 6801da177e4SLinus Torvalds * Insert a write request into an inode 6811da177e4SLinus Torvalds */ 682d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 6831da177e4SLinus Torvalds { 6841da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 685e7d39069STrond Myklebust 6862bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 6872bfc6e56SWeston Andros Adamson 688e7d39069STrond Myklebust /* Lock the request! */ 6897ad84aa9STrond Myklebust nfs_lock_request(req); 690e7d39069STrond Myklebust 691e7d39069STrond Myklebust spin_lock(&inode->i_lock); 692011e2a7fSBryan Schumaker if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 693a9a4a87aSTrond Myklebust inode->i_version++; 69429418aa4SMel Gorman /* 69529418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 69629418aa4SMel Gorman * with invalidate/truncate. 69729418aa4SMel Gorman */ 69829418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 6992df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 700deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 701277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 70229418aa4SMel Gorman } 7031da177e4SLinus Torvalds nfsi->npages++; 70417089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 70517089a29SWeston Andros Adamson * extra reference so sub groups can follow suit */ 70617089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 707c03b4024STrond Myklebust kref_get(&req->wb_kref); 708e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 7091da177e4SLinus Torvalds } 7101da177e4SLinus Torvalds 7111da177e4SLinus Torvalds /* 71289a09141SPeter Zijlstra * Remove a write request from an inode 7131da177e4SLinus Torvalds */ 7141da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7151da177e4SLinus Torvalds { 7163d4ff43dSAl Viro struct inode *inode = req->wb_context->dentry->d_inode; 7171da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 71820633f04SWeston Andros Adamson struct nfs_page *head; 71920633f04SWeston Andros Adamson 72020633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 72120633f04SWeston Andros Adamson head = req->wb_head; 7221da177e4SLinus Torvalds 723587142f8STrond Myklebust spin_lock(&inode->i_lock); 72420633f04SWeston Andros Adamson if (likely(!PageSwapCache(head->wb_page))) { 72520633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 72620633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 72720633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 72829418aa4SMel Gorman } 7291da177e4SLinus Torvalds nfsi->npages--; 730587142f8STrond Myklebust spin_unlock(&inode->i_lock); 73120633f04SWeston Andros Adamson } 73217089a29SWeston Andros Adamson 73317089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 7341da177e4SLinus Torvalds nfs_release_request(req); 735b412ddf0SWeston Andros Adamson else 736b412ddf0SWeston Andros Adamson WARN_ON_ONCE(1); 7371da177e4SLinus Torvalds } 7381da177e4SLinus Torvalds 73961822ab5STrond Myklebust static void 7406d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 74161822ab5STrond Myklebust { 74261822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 74361822ab5STrond Myklebust } 74461822ab5STrond Myklebust 74589d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 7468dd37758STrond Myklebust /** 7478dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 7488dd37758STrond Myklebust * @req: pointer to a struct nfs_page 749ea2cf228SFred Isaman * @dst: commit list head 750ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 7518dd37758STrond Myklebust * 752ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 7538dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 7548dd37758STrond Myklebust * the MM page stats. 7558dd37758STrond Myklebust * 756ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 7578dd37758STrond Myklebust * holding the nfs_page lock. 7588dd37758STrond Myklebust */ 7598dd37758STrond Myklebust void 760ea2cf228SFred Isaman nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, 761ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 7628dd37758STrond Myklebust { 7638dd37758STrond Myklebust set_bit(PG_CLEAN, &(req)->wb_flags); 764ea2cf228SFred Isaman spin_lock(cinfo->lock); 765ea2cf228SFred Isaman nfs_list_add_request(req, dst); 766ea2cf228SFred Isaman cinfo->mds->ncommit++; 767ea2cf228SFred Isaman spin_unlock(cinfo->lock); 76856f9cd68SFred Isaman if (!cinfo->dreq) { 7698dd37758STrond Myklebust inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 770d56b4ddfSMel Gorman inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, 77156f9cd68SFred Isaman BDI_RECLAIMABLE); 77256f9cd68SFred Isaman __mark_inode_dirty(req->wb_context->dentry->d_inode, 77356f9cd68SFred Isaman I_DIRTY_DATASYNC); 77456f9cd68SFred Isaman } 7758dd37758STrond Myklebust } 7768dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 7778dd37758STrond Myklebust 7788dd37758STrond Myklebust /** 7798dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 7808dd37758STrond Myklebust * @req: pointer to a nfs_page 781ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 7828dd37758STrond Myklebust * 783ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 7848dd37758STrond Myklebust * number of outstanding requests requiring a commit 7858dd37758STrond Myklebust * It does not update the MM page stats. 7868dd37758STrond Myklebust * 787ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 7888dd37758STrond Myklebust */ 7898dd37758STrond Myklebust void 790ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 791ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 7928dd37758STrond Myklebust { 7938dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 7948dd37758STrond Myklebust return; 7958dd37758STrond Myklebust nfs_list_remove_request(req); 796ea2cf228SFred Isaman cinfo->mds->ncommit--; 7978dd37758STrond Myklebust } 7988dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 7998dd37758STrond Myklebust 800ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 801ea2cf228SFred Isaman struct inode *inode) 802ea2cf228SFred Isaman { 803ea2cf228SFred Isaman cinfo->lock = &inode->i_lock; 804ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 805ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 806b359f9d0SFred Isaman cinfo->dreq = NULL; 807f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 808ea2cf228SFred Isaman } 809ea2cf228SFred Isaman 810ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 811ea2cf228SFred Isaman struct inode *inode, 812ea2cf228SFred Isaman struct nfs_direct_req *dreq) 813ea2cf228SFred Isaman { 8141763da12SFred Isaman if (dreq) 8151763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 8161763da12SFred Isaman else 817ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 818ea2cf228SFred Isaman } 819ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 8208dd37758STrond Myklebust 8211da177e4SLinus Torvalds /* 8221da177e4SLinus Torvalds * Add a request to the inode's commit list. 8231da177e4SLinus Torvalds */ 8241763da12SFred Isaman void 825ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 826ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 8271da177e4SLinus Torvalds { 828ea2cf228SFred Isaman if (pnfs_mark_request_commit(req, lseg, cinfo)) 8298dd37758STrond Myklebust return; 830ea2cf228SFred Isaman nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); 8311da177e4SLinus Torvalds } 8328e821cadSTrond Myklebust 833d6d6dc7cSFred Isaman static void 834d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 835e468bae9STrond Myklebust { 836e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 837d56b4ddfSMel Gorman dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE); 838e468bae9STrond Myklebust } 839d6d6dc7cSFred Isaman 8408dd37758STrond Myklebust static void 841d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 842d6d6dc7cSFred Isaman { 8438dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 8448dd37758STrond Myklebust struct inode *inode = req->wb_context->dentry->d_inode; 845ea2cf228SFred Isaman struct nfs_commit_info cinfo; 846d6d6dc7cSFred Isaman 847ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 848ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 849ea2cf228SFred Isaman spin_lock(cinfo.lock); 850ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 851ea2cf228SFred Isaman spin_unlock(cinfo.lock); 852d6d6dc7cSFred Isaman } 8538dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 8548dd37758STrond Myklebust } 855e468bae9STrond Myklebust } 856e468bae9STrond Myklebust 857d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 8588e821cadSTrond Myklebust { 859c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 860d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 861c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 8628e821cadSTrond Myklebust } 8638e821cadSTrond Myklebust 8648e821cadSTrond Myklebust #else 86568cd6fa4SBryan Schumaker static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 86668cd6fa4SBryan Schumaker struct inode *inode) 86768cd6fa4SBryan Schumaker { 86868cd6fa4SBryan Schumaker } 86968cd6fa4SBryan Schumaker 87068cd6fa4SBryan Schumaker void nfs_init_cinfo(struct nfs_commit_info *cinfo, 87168cd6fa4SBryan Schumaker struct inode *inode, 87268cd6fa4SBryan Schumaker struct nfs_direct_req *dreq) 87368cd6fa4SBryan Schumaker { 87468cd6fa4SBryan Schumaker } 87568cd6fa4SBryan Schumaker 8761763da12SFred Isaman void 877ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 878ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 8798e821cadSTrond Myklebust { 8808e821cadSTrond Myklebust } 8818e821cadSTrond Myklebust 8828dd37758STrond Myklebust static void 883e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 884e468bae9STrond Myklebust { 885e468bae9STrond Myklebust } 886e468bae9STrond Myklebust 887d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 8888e821cadSTrond Myklebust { 8898e821cadSTrond Myklebust return 0; 8908e821cadSTrond Myklebust } 8918e821cadSTrond Myklebust 8921da177e4SLinus Torvalds #endif 8931da177e4SLinus Torvalds 894061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 8956c75dc0dSFred Isaman { 896ea2cf228SFred Isaman struct nfs_commit_info cinfo; 8976c75dc0dSFred Isaman unsigned long bytes = 0; 8986c75dc0dSFred Isaman 8996c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 9006c75dc0dSFred Isaman goto out; 901ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 9026c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 9036c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 9046c75dc0dSFred Isaman 9056c75dc0dSFred Isaman bytes += req->wb_bytes; 9066c75dc0dSFred Isaman nfs_list_remove_request(req); 9076c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 9086c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 909d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 9106c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 9116c75dc0dSFred Isaman goto remove_req; 9126c75dc0dSFred Isaman } 913c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 914f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 915ea2cf228SFred Isaman nfs_mark_request_commit(req, hdr->lseg, &cinfo); 9166c75dc0dSFred Isaman goto next; 9176c75dc0dSFred Isaman } 9186c75dc0dSFred Isaman remove_req: 9196c75dc0dSFred Isaman nfs_inode_remove_request(req); 9206c75dc0dSFred Isaman next: 9211d1afcbcSTrond Myklebust nfs_unlock_request(req); 92220633f04SWeston Andros Adamson nfs_end_page_writeback(req); 9233aff4ebbSTrond Myklebust nfs_release_request(req); 9246c75dc0dSFred Isaman } 9256c75dc0dSFred Isaman out: 9266c75dc0dSFred Isaman hdr->release(hdr); 9276c75dc0dSFred Isaman } 9286c75dc0dSFred Isaman 92989d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 930ce59515cSAnna Schumaker unsigned long 931ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 932fb8a1f11STrond Myklebust { 933ea2cf228SFred Isaman return cinfo->mds->ncommit; 934fb8a1f11STrond Myklebust } 935fb8a1f11STrond Myklebust 936ea2cf228SFred Isaman /* cinfo->lock held by caller */ 9371763da12SFred Isaman int 938ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 939ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 940d6d6dc7cSFred Isaman { 941d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 942d6d6dc7cSFred Isaman int ret = 0; 943d6d6dc7cSFred Isaman 944d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 9458dd37758STrond Myklebust if (!nfs_lock_request(req)) 9468dd37758STrond Myklebust continue; 9477ad84aa9STrond Myklebust kref_get(&req->wb_kref); 948ea2cf228SFred Isaman if (cond_resched_lock(cinfo->lock)) 9493b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 950ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 9518dd37758STrond Myklebust nfs_list_add_request(req, dst); 952d6d6dc7cSFred Isaman ret++; 9531763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 954d6d6dc7cSFred Isaman break; 955d6d6dc7cSFred Isaman } 956d6d6dc7cSFred Isaman return ret; 957d6d6dc7cSFred Isaman } 958d6d6dc7cSFred Isaman 9591da177e4SLinus Torvalds /* 9601da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 9611da177e4SLinus Torvalds * @inode: NFS inode to scan 962ea2cf228SFred Isaman * @dst: mds destination list 963ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 9641da177e4SLinus Torvalds * 9651da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 9661da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 9671da177e4SLinus Torvalds */ 9681763da12SFred Isaman int 969ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 970ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9711da177e4SLinus Torvalds { 972d6d6dc7cSFred Isaman int ret = 0; 973fb8a1f11STrond Myklebust 974ea2cf228SFred Isaman spin_lock(cinfo->lock); 975ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 9768dd37758STrond Myklebust const int max = INT_MAX; 977d6d6dc7cSFred Isaman 978ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 979ea2cf228SFred Isaman cinfo, max); 980ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 981d6d6dc7cSFred Isaman } 982ea2cf228SFred Isaman spin_unlock(cinfo->lock); 983ff778d02STrond Myklebust return ret; 9841da177e4SLinus Torvalds } 985d6d6dc7cSFred Isaman 986c42de9ddSTrond Myklebust #else 987ce59515cSAnna Schumaker unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 988fb8a1f11STrond Myklebust { 989fb8a1f11STrond Myklebust return 0; 990fb8a1f11STrond Myklebust } 991fb8a1f11STrond Myklebust 9921763da12SFred Isaman int nfs_scan_commit(struct inode *inode, struct list_head *dst, 993ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 994c42de9ddSTrond Myklebust { 995c42de9ddSTrond Myklebust return 0; 996c42de9ddSTrond Myklebust } 9971da177e4SLinus Torvalds #endif 9981da177e4SLinus Torvalds 9991da177e4SLinus Torvalds /* 1000e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1001e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 10021da177e4SLinus Torvalds * 1003e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1004e7d39069STrond Myklebust * to disk. 10051da177e4SLinus Torvalds */ 1006e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 1007e7d39069STrond Myklebust struct page *page, 1008e7d39069STrond Myklebust unsigned int offset, 1009e7d39069STrond Myklebust unsigned int bytes) 10101da177e4SLinus Torvalds { 1011e7d39069STrond Myklebust struct nfs_page *req; 1012e7d39069STrond Myklebust unsigned int rqend; 1013e7d39069STrond Myklebust unsigned int end; 10141da177e4SLinus Torvalds int error; 1015277459d2STrond Myklebust 1016e7d39069STrond Myklebust if (!PagePrivate(page)) 1017e7d39069STrond Myklebust return NULL; 1018e7d39069STrond Myklebust 1019e7d39069STrond Myklebust end = offset + bytes; 1020e7d39069STrond Myklebust spin_lock(&inode->i_lock); 1021e7d39069STrond Myklebust 1022e7d39069STrond Myklebust for (;;) { 102384d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 1024e7d39069STrond Myklebust if (req == NULL) 1025e7d39069STrond Myklebust goto out_unlock; 1026e7d39069STrond Myklebust 10272bfc6e56SWeston Andros Adamson /* should be handled by nfs_flush_incompatible */ 10282bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 10292bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 10302bfc6e56SWeston Andros Adamson 1031e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1032e7d39069STrond Myklebust /* 1033e7d39069STrond Myklebust * Tell the caller to flush out the request if 1034e7d39069STrond Myklebust * the offsets are non-contiguous. 1035e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1036e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1037e7d39069STrond Myklebust */ 1038e468bae9STrond Myklebust if (offset > rqend 1039e7d39069STrond Myklebust || end < req->wb_offset) 1040e7d39069STrond Myklebust goto out_flushme; 1041e7d39069STrond Myklebust 10427ad84aa9STrond Myklebust if (nfs_lock_request(req)) 1043e7d39069STrond Myklebust break; 1044e7d39069STrond Myklebust 1045e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 1046587142f8STrond Myklebust spin_unlock(&inode->i_lock); 10471da177e4SLinus Torvalds error = nfs_wait_on_request(req); 10481da177e4SLinus Torvalds nfs_release_request(req); 1049e7d39069STrond Myklebust if (error != 0) 1050e7d39069STrond Myklebust goto out_err; 1051e7d39069STrond Myklebust spin_lock(&inode->i_lock); 10521da177e4SLinus Torvalds } 10531da177e4SLinus Torvalds 10541da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 10551da177e4SLinus Torvalds if (offset < req->wb_offset) { 10561da177e4SLinus Torvalds req->wb_offset = offset; 10571da177e4SLinus Torvalds req->wb_pgbase = offset; 10581da177e4SLinus Torvalds } 10591da177e4SLinus Torvalds if (end > rqend) 10601da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1061e7d39069STrond Myklebust else 1062e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 1063e7d39069STrond Myklebust out_unlock: 1064e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 1065ca138f36SFred Isaman if (req) 10668dd37758STrond Myklebust nfs_clear_request_commit(req); 1067e7d39069STrond Myklebust return req; 1068e7d39069STrond Myklebust out_flushme: 1069e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 1070e7d39069STrond Myklebust nfs_release_request(req); 1071e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1072e7d39069STrond Myklebust out_err: 1073e7d39069STrond Myklebust return ERR_PTR(error); 1074e7d39069STrond Myklebust } 10751da177e4SLinus Torvalds 1076e7d39069STrond Myklebust /* 1077e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1078e7d39069STrond Myklebust * 1079e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1080e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1081e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1082e7d39069STrond Myklebust */ 1083e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1084e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1085e7d39069STrond Myklebust { 1086d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1087e7d39069STrond Myklebust struct nfs_page *req; 1088e7d39069STrond Myklebust 1089e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1090e7d39069STrond Myklebust if (req != NULL) 1091e7d39069STrond Myklebust goto out; 10922bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 1093e7d39069STrond Myklebust if (IS_ERR(req)) 1094e7d39069STrond Myklebust goto out; 1095d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1096efc91ed0STrond Myklebust out: 109761e930a9STrond Myklebust return req; 10981da177e4SLinus Torvalds } 10991da177e4SLinus Torvalds 1100e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1101e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1102e7d39069STrond Myklebust { 1103e7d39069STrond Myklebust struct nfs_page *req; 1104e7d39069STrond Myklebust 1105e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1106e7d39069STrond Myklebust if (IS_ERR(req)) 1107e7d39069STrond Myklebust return PTR_ERR(req); 1108e7d39069STrond Myklebust /* Update file length */ 1109e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1110d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1111a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11121d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1113e7d39069STrond Myklebust return 0; 1114e7d39069STrond Myklebust } 1115e7d39069STrond Myklebust 11161da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 11171da177e4SLinus Torvalds { 1118cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 11192a369153STrond Myklebust struct nfs_lock_context *l_ctx; 11201da177e4SLinus Torvalds struct nfs_page *req; 11211a54533eSTrond Myklebust int do_flush, status; 11221da177e4SLinus Torvalds /* 11231da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 11241da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 11251da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 11261da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11271da177e4SLinus Torvalds * Also do the same if we find a request from an existing 11281da177e4SLinus Torvalds * dropped page. 11291da177e4SLinus Torvalds */ 11301a54533eSTrond Myklebust do { 113184d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 11321a54533eSTrond Myklebust if (req == NULL) 11331a54533eSTrond Myklebust return 0; 11342a369153STrond Myklebust l_ctx = req->wb_lock_context; 11352a369153STrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx; 11362bfc6e56SWeston Andros Adamson /* for now, flush if more than 1 request in page_group */ 11372bfc6e56SWeston Andros Adamson do_flush |= req->wb_this_page != req; 11380f1d2605STrond Myklebust if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) { 11392a369153STrond Myklebust do_flush |= l_ctx->lockowner.l_owner != current->files 11402a369153STrond Myklebust || l_ctx->lockowner.l_pid != current->tgid; 11412a369153STrond Myklebust } 11421da177e4SLinus Torvalds nfs_release_request(req); 11431a54533eSTrond Myklebust if (!do_flush) 11441a54533eSTrond Myklebust return 0; 1145d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 11461a54533eSTrond Myklebust } while (status == 0); 11471a54533eSTrond Myklebust return status; 11481da177e4SLinus Torvalds } 11491da177e4SLinus Torvalds 11501da177e4SLinus Torvalds /* 1151dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1152dc24826bSAndy Adamson * expire soon. 1153dc24826bSAndy Adamson * 1154dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1155dc24826bSAndy Adamson * 1156dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1157dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1158dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1159dc24826bSAndy Adamson */ 1160dc24826bSAndy Adamson int 1161dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1162dc24826bSAndy Adamson { 1163dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1164dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1165dc24826bSAndy Adamson 1166dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 1167dc24826bSAndy Adamson } 1168dc24826bSAndy Adamson 1169dc24826bSAndy Adamson /* 1170dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1171dc24826bSAndy Adamson */ 1172dc24826bSAndy Adamson bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) 1173dc24826bSAndy Adamson { 1174dc24826bSAndy Adamson return rpcauth_cred_key_to_expire(ctx->cred); 1175dc24826bSAndy Adamson } 1176dc24826bSAndy Adamson 1177dc24826bSAndy Adamson /* 11785d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 11795d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 11805d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 11815d47a356STrond Myklebust */ 11828d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 11835d47a356STrond Myklebust { 1184d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1185d529ef83SJeff Layton 11868d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 11878d197a56STrond Myklebust goto out; 118818dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1189d529ef83SJeff Layton return false; 11904db72b40SJeff Layton smp_rmb(); 1191d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 11928d197a56STrond Myklebust return false; 11938d197a56STrond Myklebust out: 119418dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 119518dd78c4SScott Mayhew return false; 11968d197a56STrond Myklebust return PageUptodate(page) != 0; 11975d47a356STrond Myklebust } 11985d47a356STrond Myklebust 1199c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1200c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1201c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1202c7559663SScott Mayhew * inefficiencies. 1203c7559663SScott Mayhew * 1204263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1205263b4509SScott Mayhew * of the checks. 1206c7559663SScott Mayhew */ 1207c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1208c7559663SScott Mayhew { 1209c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1210c7559663SScott Mayhew return 0; 1211263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1212263b4509SScott Mayhew return 0; 1213c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1214c7559663SScott Mayhew return 1; 1215263b4509SScott Mayhew if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 && 1216c7559663SScott Mayhew inode->i_flock->fl_end == OFFSET_MAX && 1217263b4509SScott Mayhew inode->i_flock->fl_type != F_RDLCK)) 1218c7559663SScott Mayhew return 1; 1219c7559663SScott Mayhew return 0; 1220c7559663SScott Mayhew } 1221c7559663SScott Mayhew 12225d47a356STrond Myklebust /* 12231da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 12241da177e4SLinus Torvalds * 12251da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 12261da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 12271da177e4SLinus Torvalds */ 12281da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 12291da177e4SLinus Torvalds unsigned int offset, unsigned int count) 12301da177e4SLinus Torvalds { 1231cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1232d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 12331da177e4SLinus Torvalds int status = 0; 12341da177e4SLinus Torvalds 123591d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 123691d5b470SChuck Lever 12376de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 12386de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 12391da177e4SLinus Torvalds 1240c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 124149a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 12421da177e4SLinus Torvalds offset = 0; 12431da177e4SLinus Torvalds } 12441da177e4SLinus Torvalds 1245e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 124603fa9e84STrond Myklebust if (status < 0) 124703fa9e84STrond Myklebust nfs_set_pageerror(page); 124859b7c05fSTrond Myklebust else 124959b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 12501da177e4SLinus Torvalds 125148186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 12521da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 12531da177e4SLinus Torvalds return status; 12541da177e4SLinus Torvalds } 12551da177e4SLinus Torvalds 12563ff7576dSTrond Myklebust static int flush_task_priority(int how) 12571da177e4SLinus Torvalds { 12581da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 12591da177e4SLinus Torvalds case FLUSH_HIGHPRI: 12601da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 12611da177e4SLinus Torvalds case FLUSH_LOWPRI: 12621da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 12631da177e4SLinus Torvalds } 12641da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 12651da177e4SLinus Torvalds } 12661da177e4SLinus Torvalds 1267d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1268d45f60c6SWeston Andros Adamson struct rpc_message *msg, 12691ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 12701da177e4SLinus Torvalds { 1271d45f60c6SWeston Andros Adamson struct inode *inode = hdr->inode; 12723ff7576dSTrond Myklebust int priority = flush_task_priority(how); 12731da177e4SLinus Torvalds 12741ed26f33SAnna Schumaker task_setup_data->priority = priority; 1275d45f60c6SWeston Andros Adamson NFS_PROTO(inode)->write_setup(hdr, msg); 1276d138d5d1SAndy Adamson 12778c21c62cSWeston Andros Adamson nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client, 1278d45f60c6SWeston Andros Adamson &task_setup_data->rpc_client, msg, hdr); 1279275acaafSTrond Myklebust } 1280275acaafSTrond Myklebust 12816d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 12826d884e8fSFred * call this on each, which will prepare them to be retried on next 12836d884e8fSFred * writeback using standard nfs. 12846d884e8fSFred */ 12856d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 12866d884e8fSFred { 12876d884e8fSFred nfs_mark_request_dirty(req); 12881d1afcbcSTrond Myklebust nfs_unlock_request(req); 128920633f04SWeston Andros Adamson nfs_end_page_writeback(req); 12903aff4ebbSTrond Myklebust nfs_release_request(req); 12916d884e8fSFred } 12926d884e8fSFred 1293061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 12946c75dc0dSFred Isaman { 12956c75dc0dSFred Isaman struct nfs_page *req; 12966c75dc0dSFred Isaman 12976c75dc0dSFred Isaman while (!list_empty(head)) { 12986c75dc0dSFred Isaman req = nfs_list_entry(head->next); 12996c75dc0dSFred Isaman nfs_list_remove_request(req); 13006c75dc0dSFred Isaman nfs_redirty_request(req); 13016c75dc0dSFred Isaman } 13026c75dc0dSFred Isaman } 13036c75dc0dSFred Isaman 1304061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1305061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1306061ae2edSFred Isaman .completion = nfs_write_completion, 1307061ae2edSFred Isaman }; 1308061ae2edSFred Isaman 130957208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1310a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1311061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 13121751c363STrond Myklebust { 1313a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 131441d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1315a20c93e3SChristoph Hellwig 1316a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1317a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1318a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1319a20c93e3SChristoph Hellwig #endif 13204a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 13214a0de55cSAnna Schumaker server->wsize, ioflags); 13221751c363STrond Myklebust } 1323ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 13241751c363STrond Myklebust 1325dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1326dce81290STrond Myklebust { 132741d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1328dce81290STrond Myklebust pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1329dce81290STrond Myklebust } 13301f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1331dce81290STrond Myklebust 13321da177e4SLinus Torvalds 13330b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 13340b7c0153SFred Isaman { 13350b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 13360b7c0153SFred Isaman 13370b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 13380b7c0153SFred Isaman } 13390b7c0153SFred Isaman 1340d45f60c6SWeston Andros Adamson static void nfs_writeback_release_common(struct nfs_pgio_header *hdr) 13411da177e4SLinus Torvalds { 1342c65e6254SWeston Andros Adamson /* do nothing! */ 13431da177e4SLinus Torvalds } 13441da177e4SLinus Torvalds 13451f2edbe3STrond Myklebust /* 13461f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 13471f2edbe3STrond Myklebust */ 13481f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 13491f2edbe3STrond Myklebust { 13501f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 13511f2edbe3STrond Myklebust int kill = 0; 1352788e7a89STrond Myklebust 13531f2edbe3STrond Myklebust /* suid always must be killed */ 13541f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 13551f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 13561f2edbe3STrond Myklebust 13571f2edbe3STrond Myklebust /* 13581f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 13591f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 13601f2edbe3STrond Myklebust */ 13611f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 13621f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 13631f2edbe3STrond Myklebust 13641f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 13651f2edbe3STrond Myklebust return kill; 13661f2edbe3STrond Myklebust 13671f2edbe3STrond Myklebust return 0; 13681f2edbe3STrond Myklebust } 1369788e7a89STrond Myklebust 13701da177e4SLinus Torvalds /* 13711da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 13721da177e4SLinus Torvalds */ 1373d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1374d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 13750eecb214SAnna Schumaker struct inode *inode) 13761da177e4SLinus Torvalds { 1377788e7a89STrond Myklebust int status; 13781da177e4SLinus Torvalds 1379f551e44fSChuck Lever /* 1380f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1381f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1382f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1383f551e44fSChuck Lever * another writer had changed the file, but some applications 1384f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1385f551e44fSChuck Lever */ 1386d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1387788e7a89STrond Myklebust if (status != 0) 13880eecb214SAnna Schumaker return status; 1389d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 139091d5b470SChuck Lever 139189d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 1392d45f60c6SWeston Andros Adamson if (hdr->res.verf->committed < hdr->args.stable && 1393d45f60c6SWeston Andros Adamson task->tk_status >= 0) { 13941da177e4SLinus Torvalds /* We tried a write call, but the server did not 13951da177e4SLinus Torvalds * commit data to stable storage even though we 13961da177e4SLinus Torvalds * requested it. 13971da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 13981da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 13991da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 14001da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 14011da177e4SLinus Torvalds */ 14021da177e4SLinus Torvalds static unsigned long complain; 14031da177e4SLinus Torvalds 1404a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 14051da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 14061da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 14071da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1408cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1409d45f60c6SWeston Andros Adamson hdr->res.verf->committed, hdr->args.stable); 14101da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 14111da177e4SLinus Torvalds } 14121da177e4SLinus Torvalds } 14131da177e4SLinus Torvalds #endif 14141f2edbe3STrond Myklebust 14151f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 14161f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 14171f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 14180eecb214SAnna Schumaker return 0; 14190eecb214SAnna Schumaker } 14200eecb214SAnna Schumaker 14210eecb214SAnna Schumaker /* 14220eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 14230eecb214SAnna Schumaker */ 1424d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1425d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 14260eecb214SAnna Schumaker { 1427d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1428d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 14291f2edbe3STrond Myklebust 14301f2edbe3STrond Myklebust if (resp->count < argp->count) { 14311da177e4SLinus Torvalds static unsigned long complain; 14321da177e4SLinus Torvalds 14336c75dc0dSFred Isaman /* This a short write! */ 1434d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 143591d5b470SChuck Lever 14361da177e4SLinus Torvalds /* Has the server at least made some progress? */ 14376c75dc0dSFred Isaman if (resp->count == 0) { 14386c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 14396c75dc0dSFred Isaman printk(KERN_WARNING 14406c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 14416c75dc0dSFred Isaman argp->count); 14426c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 14436c75dc0dSFred Isaman } 1444d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 14456c75dc0dSFred Isaman task->tk_status = -EIO; 14466c75dc0dSFred Isaman return; 14476c75dc0dSFred Isaman } 14481da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 14491da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 14501da177e4SLinus Torvalds /* Resend from where the server left off */ 1451d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 14521da177e4SLinus Torvalds argp->offset += resp->count; 14531da177e4SLinus Torvalds argp->pgbase += resp->count; 14541da177e4SLinus Torvalds argp->count -= resp->count; 14551da177e4SLinus Torvalds } else { 14561da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 14571da177e4SLinus Torvalds * headaches in the case of a server crash. 14581da177e4SLinus Torvalds */ 14591da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 14601da177e4SLinus Torvalds } 1461d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 14621da177e4SLinus Torvalds } 14631da177e4SLinus Torvalds } 14641da177e4SLinus Torvalds 14651da177e4SLinus Torvalds 146689d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 146771d0a611STrond Myklebust static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) 146871d0a611STrond Myklebust { 1469b8413f98STrond Myklebust int ret; 1470b8413f98STrond Myklebust 147171d0a611STrond Myklebust if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) 147271d0a611STrond Myklebust return 1; 1473b8413f98STrond Myklebust if (!may_wait) 147471d0a611STrond Myklebust return 0; 1475b8413f98STrond Myklebust ret = out_of_line_wait_on_bit_lock(&nfsi->flags, 1476b8413f98STrond Myklebust NFS_INO_COMMIT, 1477b8413f98STrond Myklebust nfs_wait_bit_killable, 1478b8413f98STrond Myklebust TASK_KILLABLE); 1479b8413f98STrond Myklebust return (ret < 0) ? ret : 1; 148071d0a611STrond Myklebust } 148171d0a611STrond Myklebust 1482f453a54aSFred Isaman static void nfs_commit_clear_lock(struct nfs_inode *nfsi) 148371d0a611STrond Myklebust { 148471d0a611STrond Myklebust clear_bit(NFS_INO_COMMIT, &nfsi->flags); 14854e857c58SPeter Zijlstra smp_mb__after_atomic(); 148671d0a611STrond Myklebust wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); 148771d0a611STrond Myklebust } 148871d0a611STrond Myklebust 14890b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 14901da177e4SLinus Torvalds { 14910b7c0153SFred Isaman put_nfs_open_context(data->context); 14920b7c0153SFred Isaman nfs_commit_free(data); 14931da177e4SLinus Torvalds } 1494e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 14951da177e4SLinus Torvalds 14960b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 14979ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 14989f0ec176SAndy Adamson int how, int flags) 14991da177e4SLinus Torvalds { 150007737691STrond Myklebust struct rpc_task *task; 15019ace33cdSFred Isaman int priority = flush_task_priority(how); 1502bdc7f021STrond Myklebust struct rpc_message msg = { 1503bdc7f021STrond Myklebust .rpc_argp = &data->args, 1504bdc7f021STrond Myklebust .rpc_resp = &data->res, 15059ace33cdSFred Isaman .rpc_cred = data->cred, 1506bdc7f021STrond Myklebust }; 150784115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 150807737691STrond Myklebust .task = &data->task, 15099ace33cdSFred Isaman .rpc_client = clnt, 1510bdc7f021STrond Myklebust .rpc_message = &msg, 15119ace33cdSFred Isaman .callback_ops = call_ops, 151284115e1cSTrond Myklebust .callback_data = data, 1513101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 15149f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 15153ff7576dSTrond Myklebust .priority = priority, 151684115e1cSTrond Myklebust }; 1517788e7a89STrond Myklebust /* Set up the initial task struct. */ 15189ace33cdSFred Isaman NFS_PROTO(data->inode)->commit_setup(data, &msg); 15191da177e4SLinus Torvalds 1520a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1521bdc7f021STrond Myklebust 15228c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 15238c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 15248c21c62cSWeston Andros Adamson 152507737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1526dbae4c73STrond Myklebust if (IS_ERR(task)) 1527dbae4c73STrond Myklebust return PTR_ERR(task); 1528d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1529d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 153007737691STrond Myklebust rpc_put_task(task); 1531dbae4c73STrond Myklebust return 0; 15321da177e4SLinus Torvalds } 1533e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 15341da177e4SLinus Torvalds 15351da177e4SLinus Torvalds /* 15369ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 15379ace33cdSFred Isaman */ 15380b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1539988b6dceSFred Isaman struct list_head *head, 1540f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1541f453a54aSFred Isaman struct nfs_commit_info *cinfo) 15429ace33cdSFred Isaman { 15439ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 15443d4ff43dSAl Viro struct inode *inode = first->wb_context->dentry->d_inode; 15459ace33cdSFred Isaman 15469ace33cdSFred Isaman /* Set up the RPC argument and reply structs 15479ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 15489ace33cdSFred Isaman 15499ace33cdSFred Isaman list_splice_init(head, &data->pages); 15509ace33cdSFred Isaman 15519ace33cdSFred Isaman data->inode = inode; 15529ace33cdSFred Isaman data->cred = first->wb_context->cred; 1553988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 15549ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1555f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1556b359f9d0SFred Isaman data->dreq = cinfo->dreq; 15579ace33cdSFred Isaman 15589ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 15599ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 15609ace33cdSFred Isaman data->args.offset = 0; 15619ace33cdSFred Isaman data->args.count = 0; 15620b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 15639ace33cdSFred Isaman data->res.fattr = &data->fattr; 15649ace33cdSFred Isaman data->res.verf = &data->verf; 15659ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 15669ace33cdSFred Isaman } 1567e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 15689ace33cdSFred Isaman 1569e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1570ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1571ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 157264bfeb49SFred Isaman { 157364bfeb49SFred Isaman struct nfs_page *req; 157464bfeb49SFred Isaman 157564bfeb49SFred Isaman while (!list_empty(page_list)) { 157664bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 157764bfeb49SFred Isaman nfs_list_remove_request(req); 1578ea2cf228SFred Isaman nfs_mark_request_commit(req, lseg, cinfo); 157956f9cd68SFred Isaman if (!cinfo->dreq) { 158064bfeb49SFred Isaman dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1581d56b4ddfSMel Gorman dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, 158264bfeb49SFred Isaman BDI_RECLAIMABLE); 158356f9cd68SFred Isaman } 15841d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 158564bfeb49SFred Isaman } 158664bfeb49SFred Isaman } 1587e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 158864bfeb49SFred Isaman 15899ace33cdSFred Isaman /* 15901da177e4SLinus Torvalds * Commit dirty pages 15911da177e4SLinus Torvalds */ 15921da177e4SLinus Torvalds static int 1593ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1594ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 15951da177e4SLinus Torvalds { 15960b7c0153SFred Isaman struct nfs_commit_data *data; 15971da177e4SLinus Torvalds 1598c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 15991da177e4SLinus Torvalds 16001da177e4SLinus Torvalds if (!data) 16011da177e4SLinus Torvalds goto out_bad; 16021da177e4SLinus Torvalds 16031da177e4SLinus Torvalds /* Set up the argument struct */ 1604f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1605f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 16069f0ec176SAndy Adamson return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, 16079f0ec176SAndy Adamson how, 0); 16081da177e4SLinus Torvalds out_bad: 1609ea2cf228SFred Isaman nfs_retry_commit(head, NULL, cinfo); 1610f453a54aSFred Isaman cinfo->completion_ops->error_cleanup(NFS_I(inode)); 16111da177e4SLinus Torvalds return -ENOMEM; 16121da177e4SLinus Torvalds } 16131da177e4SLinus Torvalds 16141da177e4SLinus Torvalds /* 16151da177e4SLinus Torvalds * COMMIT call returned 16161da177e4SLinus Torvalds */ 1617788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 16181da177e4SLinus Torvalds { 16190b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 16201da177e4SLinus Torvalds 1621a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 16221da177e4SLinus Torvalds task->tk_pid, task->tk_status); 16231da177e4SLinus Torvalds 1624788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1625c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1626c9d8f89dSTrond Myklebust } 1627c9d8f89dSTrond Myklebust 1628f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1629c9d8f89dSTrond Myklebust { 1630c9d8f89dSTrond Myklebust struct nfs_page *req; 1631c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1632f453a54aSFred Isaman struct nfs_commit_info cinfo; 1633788e7a89STrond Myklebust 16341da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 16351da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 16361da177e4SLinus Torvalds nfs_list_remove_request(req); 1637d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 16381da177e4SLinus Torvalds 16391e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 16403d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 16411e8968c5SNiels de Vos (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode), 16421da177e4SLinus Torvalds req->wb_bytes, 16431da177e4SLinus Torvalds (long long)req_offset(req)); 1644c9d8f89dSTrond Myklebust if (status < 0) { 1645c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 16461da177e4SLinus Torvalds nfs_inode_remove_request(req); 1647c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 16481da177e4SLinus Torvalds goto next; 16491da177e4SLinus Torvalds } 16501da177e4SLinus Torvalds 16511da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 16521da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 16532f2c63bcSTrond Myklebust if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { 16541da177e4SLinus Torvalds /* We have a match */ 16551da177e4SLinus Torvalds nfs_inode_remove_request(req); 16561da177e4SLinus Torvalds dprintk(" OK\n"); 16571da177e4SLinus Torvalds goto next; 16581da177e4SLinus Torvalds } 16591da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 16601da177e4SLinus Torvalds dprintk(" mismatch\n"); 16616d884e8fSFred nfs_mark_request_dirty(req); 166205990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 16631da177e4SLinus Torvalds next: 16641d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 16651da177e4SLinus Torvalds } 1666f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1667f453a54aSFred Isaman if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) 1668f453a54aSFred Isaman nfs_commit_clear_lock(NFS_I(data->inode)); 16695917ce84SFred Isaman } 16705917ce84SFred Isaman 16715917ce84SFred Isaman static void nfs_commit_release(void *calldata) 16725917ce84SFred Isaman { 16730b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 16745917ce84SFred Isaman 1675f453a54aSFred Isaman data->completion_ops->completion(data); 1676c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 16771da177e4SLinus Torvalds } 1678788e7a89STrond Myklebust 1679788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 16800b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1681788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1682788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1683788e7a89STrond Myklebust }; 16841da177e4SLinus Torvalds 1685f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1686f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1687f453a54aSFred Isaman .error_cleanup = nfs_commit_clear_lock, 1688f453a54aSFred Isaman }; 1689f453a54aSFred Isaman 16901763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1691ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 169284c53ab5SFred Isaman { 169384c53ab5SFred Isaman int status; 169484c53ab5SFred Isaman 1695ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 169684c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1697ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 169884c53ab5SFred Isaman return status; 169984c53ab5SFred Isaman } 170084c53ab5SFred Isaman 1701b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 17021da177e4SLinus Torvalds { 17031da177e4SLinus Torvalds LIST_HEAD(head); 1704ea2cf228SFred Isaman struct nfs_commit_info cinfo; 170571d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1706b8413f98STrond Myklebust int res; 17071da177e4SLinus Torvalds 1708b8413f98STrond Myklebust res = nfs_commit_set_lock(NFS_I(inode), may_wait); 1709b8413f98STrond Myklebust if (res <= 0) 1710c5efa5fcSTrond Myklebust goto out_mark_dirty; 1711ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1712ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 17131da177e4SLinus Torvalds if (res) { 1714a861a1e1SFred Isaman int error; 1715a861a1e1SFred Isaman 1716ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 17171da177e4SLinus Torvalds if (error < 0) 17181da177e4SLinus Torvalds return error; 1719b8413f98STrond Myklebust if (!may_wait) 1720b8413f98STrond Myklebust goto out_mark_dirty; 1721b8413f98STrond Myklebust error = wait_on_bit(&NFS_I(inode)->flags, 1722b8413f98STrond Myklebust NFS_INO_COMMIT, 172371d0a611STrond Myklebust nfs_wait_bit_killable, 172471d0a611STrond Myklebust TASK_KILLABLE); 1725b8413f98STrond Myklebust if (error < 0) 1726b8413f98STrond Myklebust return error; 172771d0a611STrond Myklebust } else 172871d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(inode)); 1729c5efa5fcSTrond Myklebust return res; 1730c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1731c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1732c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1733c5efa5fcSTrond Myklebust * that the data is on the disk. 1734c5efa5fcSTrond Myklebust */ 1735c5efa5fcSTrond Myklebust out_mark_dirty: 1736c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 17371da177e4SLinus Torvalds return res; 17381da177e4SLinus Torvalds } 17398fc795f7STrond Myklebust 17408fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 17418fc795f7STrond Myklebust { 1742420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1743420e3646STrond Myklebust int flags = FLUSH_SYNC; 1744420e3646STrond Myklebust int ret = 0; 17458fc795f7STrond Myklebust 17463236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1747ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 17483236c3e1SJeff Layton return ret; 17493236c3e1SJeff Layton 1750a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1751a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1752a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1753420e3646STrond Myklebust */ 1754ea2cf228SFred Isaman if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1)) 1755420e3646STrond Myklebust goto out_mark_dirty; 1756420e3646STrond Myklebust 1757a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1758420e3646STrond Myklebust flags = 0; 1759a00dd6c0SJeff Layton } 1760a00dd6c0SJeff Layton 1761420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1762420e3646STrond Myklebust if (ret >= 0) { 1763420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1764420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1765420e3646STrond Myklebust wbc->nr_to_write -= ret; 1766420e3646STrond Myklebust else 1767420e3646STrond Myklebust wbc->nr_to_write = 0; 1768420e3646STrond Myklebust } 17698fc795f7STrond Myklebust return 0; 1770420e3646STrond Myklebust } 1771420e3646STrond Myklebust out_mark_dirty: 17728fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 17738fc795f7STrond Myklebust return ret; 17748fc795f7STrond Myklebust } 1775c63c7b05STrond Myklebust #else 17768fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 17778fc795f7STrond Myklebust { 17788fc795f7STrond Myklebust return 0; 17798fc795f7STrond Myklebust } 17801da177e4SLinus Torvalds #endif 17811da177e4SLinus Torvalds 17828fc795f7STrond Myklebust int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 17838fc795f7STrond Myklebust { 1784a8d8f02cSBryan Schumaker return nfs_commit_unstable_pages(inode, wbc); 1785a8d8f02cSBryan Schumaker } 178689d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1787863a3c6cSAndy Adamson 1788acdc53b2STrond Myklebust /* 1789acdc53b2STrond Myklebust * flush the inode to disk. 1790acdc53b2STrond Myklebust */ 1791acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 179234901f70STrond Myklebust { 179334901f70STrond Myklebust struct writeback_control wbc = { 179472cb77f4STrond Myklebust .sync_mode = WB_SYNC_ALL, 179534901f70STrond Myklebust .nr_to_write = LONG_MAX, 1796d7fb1207STrond Myklebust .range_start = 0, 1797d7fb1207STrond Myklebust .range_end = LLONG_MAX, 179834901f70STrond Myklebust }; 1799f4ce1299STrond Myklebust int ret; 180034901f70STrond Myklebust 1801f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1802f4ce1299STrond Myklebust 1803f4ce1299STrond Myklebust ret = sync_inode(inode, &wbc); 1804f4ce1299STrond Myklebust 1805f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1806f4ce1299STrond Myklebust return ret; 18071c75950bSTrond Myklebust } 1808ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 18091c75950bSTrond Myklebust 18101b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 18111b3b4a1aSTrond Myklebust { 18121b3b4a1aSTrond Myklebust struct nfs_page *req; 18131b3b4a1aSTrond Myklebust int ret = 0; 18141b3b4a1aSTrond Myklebust 1815ba8b06e6STrond Myklebust wait_on_page_writeback(page); 18163e217045SWeston Andros Adamson 18173e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 18183e217045SWeston Andros Adamson * request */ 18193e217045SWeston Andros Adamson req = nfs_lock_and_join_requests(page, false); 18203e217045SWeston Andros Adamson 18213e217045SWeston Andros Adamson if (IS_ERR(req)) { 18223e217045SWeston Andros Adamson ret = PTR_ERR(req); 18233e217045SWeston Andros Adamson } else if (req) { 18243e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 18253e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 18263e217045SWeston Andros Adamson * request from the inode / page_private pointer and 18273e217045SWeston Andros Adamson * release it */ 18281b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 18291b3b4a1aSTrond Myklebust /* 18301b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 18311b3b4a1aSTrond Myklebust * page as being dirty 18321b3b4a1aSTrond Myklebust */ 18331b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 18341d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18351b3b4a1aSTrond Myklebust } 18363e217045SWeston Andros Adamson 18371b3b4a1aSTrond Myklebust return ret; 18381b3b4a1aSTrond Myklebust } 18391b3b4a1aSTrond Myklebust 18401c75950bSTrond Myklebust /* 18411c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 18421c75950bSTrond Myklebust */ 18431c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 18441c75950bSTrond Myklebust { 184529418aa4SMel Gorman loff_t range_start = page_file_offset(page); 18467f2f12d9STrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 18477f2f12d9STrond Myklebust struct writeback_control wbc = { 18487f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 18497f2f12d9STrond Myklebust .nr_to_write = 0, 18507f2f12d9STrond Myklebust .range_start = range_start, 18517f2f12d9STrond Myklebust .range_end = range_end, 18527f2f12d9STrond Myklebust }; 18537f2f12d9STrond Myklebust int ret; 18547f2f12d9STrond Myklebust 1855f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 1856f4ce1299STrond Myklebust 18570522f6adSTrond Myklebust for (;;) { 1858ba8b06e6STrond Myklebust wait_on_page_writeback(page); 18597f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 18607f2f12d9STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 18617f2f12d9STrond Myklebust if (ret < 0) 18627f2f12d9STrond Myklebust goto out_error; 18630522f6adSTrond Myklebust continue; 18647f2f12d9STrond Myklebust } 1865f4ce1299STrond Myklebust ret = 0; 18660522f6adSTrond Myklebust if (!PagePrivate(page)) 18670522f6adSTrond Myklebust break; 18680522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 18697f2f12d9STrond Myklebust if (ret < 0) 18707f2f12d9STrond Myklebust goto out_error; 18717f2f12d9STrond Myklebust } 18727f2f12d9STrond Myklebust out_error: 1873f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 18747f2f12d9STrond Myklebust return ret; 18751c75950bSTrond Myklebust } 18761c75950bSTrond Myklebust 1877074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 1878074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 1879a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 1880074cc1deSTrond Myklebust { 18812da95652SJeff Layton /* 18822da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 18832da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 18842da95652SJeff Layton * 18852da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 18862da95652SJeff Layton * that we can safely release the inode reference while holding 18872da95652SJeff Layton * the page lock. 18882da95652SJeff Layton */ 18892da95652SJeff Layton if (PagePrivate(page)) 18902da95652SJeff Layton return -EBUSY; 1891074cc1deSTrond Myklebust 18928c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 18938c209ce7SDavid Howells return -EBUSY; 1894074cc1deSTrond Myklebust 1895a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 1896074cc1deSTrond Myklebust } 1897074cc1deSTrond Myklebust #endif 1898074cc1deSTrond Myklebust 1899f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 19001da177e4SLinus Torvalds { 19011da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 19021e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 19031da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 190420c2df83SPaul Mundt NULL); 19051da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 19061da177e4SLinus Torvalds return -ENOMEM; 19071da177e4SLinus Torvalds 190893d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 19091da177e4SLinus Torvalds nfs_wdata_cachep); 19101da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 19113dd4765fSJeff Layton goto out_destroy_write_cache; 19121da177e4SLinus Torvalds 19130b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 19140b7c0153SFred Isaman sizeof(struct nfs_commit_data), 19150b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 19160b7c0153SFred Isaman NULL); 19170b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 19183dd4765fSJeff Layton goto out_destroy_write_mempool; 19190b7c0153SFred Isaman 192093d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 19214c100210SYanchuan Nian nfs_cdata_cachep); 19221da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 19233dd4765fSJeff Layton goto out_destroy_commit_cache; 19241da177e4SLinus Torvalds 192589a09141SPeter Zijlstra /* 192689a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 192789a09141SPeter Zijlstra * 192889a09141SPeter Zijlstra * 64MB: 8192k 192989a09141SPeter Zijlstra * 128MB: 11585k 193089a09141SPeter Zijlstra * 256MB: 16384k 193189a09141SPeter Zijlstra * 512MB: 23170k 193289a09141SPeter Zijlstra * 1GB: 32768k 193389a09141SPeter Zijlstra * 2GB: 46340k 193489a09141SPeter Zijlstra * 4GB: 65536k 193589a09141SPeter Zijlstra * 8GB: 92681k 193689a09141SPeter Zijlstra * 16GB: 131072k 193789a09141SPeter Zijlstra * 193889a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 193989a09141SPeter Zijlstra * Limit the default to 256M 194089a09141SPeter Zijlstra */ 194189a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 194289a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 194389a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 194489a09141SPeter Zijlstra 19451da177e4SLinus Torvalds return 0; 19463dd4765fSJeff Layton 19473dd4765fSJeff Layton out_destroy_commit_cache: 19483dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 19493dd4765fSJeff Layton out_destroy_write_mempool: 19503dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 19513dd4765fSJeff Layton out_destroy_write_cache: 19523dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 19533dd4765fSJeff Layton return -ENOMEM; 19541da177e4SLinus Torvalds } 19551da177e4SLinus Torvalds 1956266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 19571da177e4SLinus Torvalds { 19581da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 19593dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 19601da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 19611a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 19621da177e4SLinus Torvalds } 19631da177e4SLinus Torvalds 19644a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 1965a4cdda59SAnna Schumaker .rw_mode = FMODE_WRITE, 19664a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 19674a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 1968a4cdda59SAnna Schumaker .rw_release = nfs_writeback_release_common, 19690eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 19700eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 19711ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 19724a0de55cSAnna Schumaker }; 1973