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; 49*d4581383SWeston Andros Adamson static void nfs_clear_request_commit(struct nfs_page *req); 501da177e4SLinus Torvalds 51e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 523feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 530b7c0153SFred Isaman static struct kmem_cache *nfs_cdata_cachep; 541da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 551da177e4SLinus Torvalds 560b7c0153SFred Isaman struct nfs_commit_data *nfs_commitdata_alloc(void) 571da177e4SLinus Torvalds { 58192e501bSMel Gorman struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); 5940859d7eSChuck Lever 601da177e4SLinus Torvalds if (p) { 611da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 621da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 631da177e4SLinus Torvalds } 641da177e4SLinus Torvalds return p; 651da177e4SLinus Torvalds } 66e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 671da177e4SLinus Torvalds 680b7c0153SFred Isaman void nfs_commit_free(struct nfs_commit_data *p) 691da177e4SLinus Torvalds { 701da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 711da177e4SLinus Torvalds } 72e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 731da177e4SLinus Torvalds 744a0de55cSAnna Schumaker static struct nfs_rw_header *nfs_writehdr_alloc(void) 753feb2d49STrond Myklebust { 76c0752cdfSAnna Schumaker struct nfs_rw_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); 773feb2d49STrond Myklebust 784a0de55cSAnna Schumaker if (p) 793feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 803feb2d49STrond Myklebust return p; 813feb2d49STrond Myklebust } 823feb2d49STrond Myklebust 834a0de55cSAnna Schumaker static void nfs_writehdr_free(struct nfs_rw_header *whdr) 846c75dc0dSFred Isaman { 85cd841605SFred Isaman mempool_free(whdr, nfs_wdata_mempool); 863feb2d49STrond Myklebust } 871da177e4SLinus Torvalds 887b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 897b159fc1STrond Myklebust { 907b159fc1STrond Myklebust ctx->error = error; 917b159fc1STrond Myklebust smp_wmb(); 927b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 937b159fc1STrond Myklebust } 947b159fc1STrond Myklebust 9584d3a9a9SWeston Andros Adamson /* 9684d3a9a9SWeston Andros Adamson * nfs_page_find_head_request_locked - find head request associated with @page 9784d3a9a9SWeston Andros Adamson * 9884d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 9984d3a9a9SWeston Andros Adamson * 10084d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 10184d3a9a9SWeston Andros Adamson */ 10229418aa4SMel Gorman static struct nfs_page * 10384d3a9a9SWeston Andros Adamson nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page) 104277459d2STrond Myklebust { 105277459d2STrond Myklebust struct nfs_page *req = NULL; 106277459d2STrond Myklebust 10729418aa4SMel Gorman if (PagePrivate(page)) 108277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 10929418aa4SMel Gorman else if (unlikely(PageSwapCache(page))) { 11029418aa4SMel Gorman struct nfs_page *freq, *t; 11129418aa4SMel Gorman 11229418aa4SMel Gorman /* Linearly search the commit list for the correct req */ 11329418aa4SMel Gorman list_for_each_entry_safe(freq, t, &nfsi->commit_info.list, wb_list) { 11429418aa4SMel Gorman if (freq->wb_page == page) { 11584d3a9a9SWeston Andros Adamson req = freq->wb_head; 11629418aa4SMel Gorman break; 117277459d2STrond Myklebust } 11829418aa4SMel Gorman } 11929418aa4SMel Gorman } 12029418aa4SMel Gorman 12184d3a9a9SWeston Andros Adamson if (req) { 12284d3a9a9SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 12384d3a9a9SWeston Andros Adamson 12429418aa4SMel Gorman kref_get(&req->wb_kref); 12584d3a9a9SWeston Andros Adamson } 12629418aa4SMel Gorman 127277459d2STrond Myklebust return req; 128277459d2STrond Myklebust } 129277459d2STrond Myklebust 13084d3a9a9SWeston Andros Adamson /* 13184d3a9a9SWeston Andros Adamson * nfs_page_find_head_request - find head request associated with @page 13284d3a9a9SWeston Andros Adamson * 13384d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 13484d3a9a9SWeston Andros Adamson */ 13584d3a9a9SWeston Andros Adamson static struct nfs_page *nfs_page_find_head_request(struct page *page) 136277459d2STrond Myklebust { 137d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 138277459d2STrond Myklebust struct nfs_page *req = NULL; 139277459d2STrond Myklebust 140587142f8STrond Myklebust spin_lock(&inode->i_lock); 14184d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 142587142f8STrond Myklebust spin_unlock(&inode->i_lock); 143277459d2STrond Myklebust return req; 144277459d2STrond Myklebust } 145277459d2STrond Myklebust 1461da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1471da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1481da177e4SLinus Torvalds { 149d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 150a3d01454STrond Myklebust loff_t end, i_size; 151a3d01454STrond Myklebust pgoff_t end_index; 1521da177e4SLinus Torvalds 153a3d01454STrond Myklebust spin_lock(&inode->i_lock); 154a3d01454STrond Myklebust i_size = i_size_read(inode); 155a3d01454STrond Myklebust end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 156d56b4ddfSMel Gorman if (i_size > 0 && page_file_index(page) < end_index) 157a3d01454STrond Myklebust goto out; 158d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 1591da177e4SLinus Torvalds if (i_size >= end) 160a3d01454STrond Myklebust goto out; 1611da177e4SLinus Torvalds i_size_write(inode, end); 162a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 163a3d01454STrond Myklebust out: 164a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 1651da177e4SLinus Torvalds } 1661da177e4SLinus Torvalds 167a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 168a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 169a301b777STrond Myklebust { 170d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 171a301b777STrond Myklebust } 172a301b777STrond Myklebust 173d72ddcbaSWeston Andros Adamson /* 174d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 175d72ddcbaSWeston Andros Adamson * @head - head request of page group 176d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 177d72ddcbaSWeston Andros Adamson * 178d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 179d72ddcbaSWeston Andros Adamson * page offset @page_offset. 180d72ddcbaSWeston Andros Adamson * 181d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 182d72ddcbaSWeston Andros Adamson * match is found. 183d72ddcbaSWeston Andros Adamson * 184d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 185d72ddcbaSWeston Andros Adamson */ 186d72ddcbaSWeston Andros Adamson static struct nfs_page * 187d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 188d72ddcbaSWeston Andros Adamson { 189d72ddcbaSWeston Andros Adamson struct nfs_page *req; 190d72ddcbaSWeston Andros Adamson 191d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(head != head->wb_head); 192d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags)); 193d72ddcbaSWeston Andros Adamson 194d72ddcbaSWeston Andros Adamson req = head; 195d72ddcbaSWeston Andros Adamson do { 196d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 197d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 198d72ddcbaSWeston Andros Adamson return req; 199d72ddcbaSWeston Andros Adamson 200d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 201d72ddcbaSWeston Andros Adamson } while (req != head); 202d72ddcbaSWeston Andros Adamson 203d72ddcbaSWeston Andros Adamson return NULL; 204d72ddcbaSWeston Andros Adamson } 205d72ddcbaSWeston Andros Adamson 206d72ddcbaSWeston Andros Adamson /* 207d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 208d72ddcbaSWeston Andros Adamson * @head - head request of page group 209d72ddcbaSWeston Andros Adamson * 210d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 211d72ddcbaSWeston Andros Adamson * returns false otherwise 212d72ddcbaSWeston Andros Adamson */ 213d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 214d72ddcbaSWeston Andros Adamson { 215d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 216d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 217d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 218d72ddcbaSWeston Andros Adamson 219d72ddcbaSWeston Andros Adamson nfs_page_group_lock(req); 220d72ddcbaSWeston Andros Adamson 221d72ddcbaSWeston Andros Adamson do { 222d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 223d72ddcbaSWeston Andros Adamson if (tmp) { 224d72ddcbaSWeston Andros Adamson /* no way this should happen */ 225d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(tmp->wb_pgbase != pos); 226d72ddcbaSWeston Andros Adamson pos += tmp->wb_bytes - (pos - tmp->wb_pgbase); 227d72ddcbaSWeston Andros Adamson } 228d72ddcbaSWeston Andros Adamson } while (tmp && pos < len); 229d72ddcbaSWeston Andros Adamson 230d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 231d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(pos > len); 232d72ddcbaSWeston Andros Adamson return pos == len; 233d72ddcbaSWeston Andros Adamson } 234d72ddcbaSWeston Andros Adamson 2351da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 2361da177e4SLinus Torvalds * covers the full page. 2371da177e4SLinus Torvalds */ 238d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 2391da177e4SLinus Torvalds { 240d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 2411da177e4SLinus Torvalds return; 242d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 2431da177e4SLinus Torvalds return; 244d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 2451da177e4SLinus Torvalds } 2461da177e4SLinus Torvalds 2471da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 2481da177e4SLinus Torvalds { 2491da177e4SLinus Torvalds if (wbc->for_reclaim) 250c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 251b17621feSWu Fengguang if (wbc->for_kupdate || wbc->for_background) 252b31268acSTrond Myklebust return FLUSH_LOWPRI | FLUSH_COND_STABLE; 253b31268acSTrond Myklebust return FLUSH_COND_STABLE; 2541da177e4SLinus Torvalds } 2551da177e4SLinus Torvalds 2561da177e4SLinus Torvalds /* 25789a09141SPeter Zijlstra * NFS congestion control 25889a09141SPeter Zijlstra */ 25989a09141SPeter Zijlstra 26089a09141SPeter Zijlstra int nfs_congestion_kb; 26189a09141SPeter Zijlstra 26289a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 26389a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 26489a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 26589a09141SPeter Zijlstra 266deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 26789a09141SPeter Zijlstra { 268deed85e7STrond Myklebust struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); 2695a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2705a6d41b3STrond Myklebust 271deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 27289a09141SPeter Zijlstra 273277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 2748aa7e847SJens Axboe NFS_CONGESTION_ON_THRESH) { 2758aa7e847SJens Axboe set_bdi_congested(&nfss->backing_dev_info, 2768aa7e847SJens Axboe BLK_RW_ASYNC); 2778aa7e847SJens Axboe } 27889a09141SPeter Zijlstra } 27989a09141SPeter Zijlstra 28020633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 28189a09141SPeter Zijlstra { 28220633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 28389a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 28489a09141SPeter Zijlstra 28520633f04SWeston Andros Adamson if (!nfs_page_group_sync_on_bit(req, PG_WB_END)) 28620633f04SWeston Andros Adamson return; 28720633f04SWeston Andros Adamson 28820633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 289c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 2908aa7e847SJens Axboe clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 29189a09141SPeter Zijlstra } 29289a09141SPeter Zijlstra 293*d4581383SWeston Andros Adamson 294*d4581383SWeston Andros Adamson /* nfs_page_group_clear_bits 295*d4581383SWeston Andros Adamson * @req - an nfs request 296*d4581383SWeston Andros Adamson * clears all page group related bits from @req 297*d4581383SWeston Andros Adamson */ 298*d4581383SWeston Andros Adamson static void 299*d4581383SWeston Andros Adamson nfs_page_group_clear_bits(struct nfs_page *req) 300e261f51fSTrond Myklebust { 301*d4581383SWeston Andros Adamson clear_bit(PG_TEARDOWN, &req->wb_flags); 302*d4581383SWeston Andros Adamson clear_bit(PG_UNLOCKPAGE, &req->wb_flags); 303*d4581383SWeston Andros Adamson clear_bit(PG_UPTODATE, &req->wb_flags); 304*d4581383SWeston Andros Adamson clear_bit(PG_WB_END, &req->wb_flags); 305*d4581383SWeston Andros Adamson clear_bit(PG_REMOVE, &req->wb_flags); 306*d4581383SWeston Andros Adamson } 307*d4581383SWeston Andros Adamson 308*d4581383SWeston Andros Adamson 309*d4581383SWeston Andros Adamson /* 310*d4581383SWeston Andros Adamson * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req 311*d4581383SWeston Andros Adamson * 312*d4581383SWeston Andros Adamson * this is a helper function for nfs_lock_and_join_requests 313*d4581383SWeston Andros Adamson * 314*d4581383SWeston Andros Adamson * @inode - inode associated with request page group, must be holding inode lock 315*d4581383SWeston Andros Adamson * @head - head request of page group, must be holding head lock 316*d4581383SWeston Andros Adamson * @req - request that couldn't lock and needs to wait on the req bit lock 317*d4581383SWeston Andros Adamson * @nonblock - if true, don't actually wait 318*d4581383SWeston Andros Adamson * 319*d4581383SWeston Andros Adamson * NOTE: this must be called holding page_group bit lock and inode spin lock 320*d4581383SWeston Andros Adamson * and BOTH will be released before returning. 321*d4581383SWeston Andros Adamson * 322*d4581383SWeston Andros Adamson * returns 0 on success, < 0 on error. 323*d4581383SWeston Andros Adamson */ 324*d4581383SWeston Andros Adamson static int 325*d4581383SWeston Andros Adamson nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head, 326*d4581383SWeston Andros Adamson struct nfs_page *req, bool nonblock) 327*d4581383SWeston Andros Adamson __releases(&inode->i_lock) 328*d4581383SWeston Andros Adamson { 329*d4581383SWeston Andros Adamson struct nfs_page *tmp; 330e261f51fSTrond Myklebust int ret; 331e261f51fSTrond Myklebust 332*d4581383SWeston Andros Adamson /* relinquish all the locks successfully grabbed this run */ 333*d4581383SWeston Andros Adamson for (tmp = head ; tmp != req; tmp = tmp->wb_this_page) 334*d4581383SWeston Andros Adamson nfs_unlock_request(tmp); 335*d4581383SWeston Andros Adamson 336*d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags)); 337*d4581383SWeston Andros Adamson 338*d4581383SWeston Andros Adamson /* grab a ref on the request that will be waited on */ 339*d4581383SWeston Andros Adamson kref_get(&req->wb_kref); 340*d4581383SWeston Andros Adamson 341*d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 342587142f8STrond Myklebust spin_unlock(&inode->i_lock); 343*d4581383SWeston Andros Adamson 344*d4581383SWeston Andros Adamson /* release ref from nfs_page_find_head_request_locked */ 345*d4581383SWeston Andros Adamson nfs_release_request(head); 346*d4581383SWeston Andros Adamson 347cfb506e1STrond Myklebust if (!nonblock) 348e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 349cfb506e1STrond Myklebust else 350cfb506e1STrond Myklebust ret = -EAGAIN; 351e261f51fSTrond Myklebust nfs_release_request(req); 352*d4581383SWeston Andros Adamson 353*d4581383SWeston Andros Adamson return ret; 354e261f51fSTrond Myklebust } 355*d4581383SWeston Andros Adamson 356*d4581383SWeston Andros Adamson /* 357*d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 358*d4581383SWeston Andros Adamson * 359*d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 360*d4581383SWeston Andros Adamson * @old_head - the old head of the list 361*d4581383SWeston Andros Adamson * 362*d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 363*d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 364*d4581383SWeston Andros Adamson * with a reference held by some other context. 365*d4581383SWeston Andros Adamson */ 366*d4581383SWeston Andros Adamson static void 367*d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 368*d4581383SWeston Andros Adamson struct nfs_page *old_head) 369*d4581383SWeston Andros Adamson { 370*d4581383SWeston Andros Adamson while (destroy_list) { 371*d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 372*d4581383SWeston Andros Adamson 373*d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 374*d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 375*d4581383SWeston Andros Adamson 376*d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 377*d4581383SWeston Andros Adamson 378*d4581383SWeston Andros Adamson /* make sure old group is not used */ 379*d4581383SWeston Andros Adamson subreq->wb_head = subreq; 380*d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 381*d4581383SWeston Andros Adamson 382*d4581383SWeston Andros Adamson nfs_clear_request_commit(subreq); 383*d4581383SWeston Andros Adamson 384*d4581383SWeston Andros Adamson /* subreq is now totally disconnected from page group or any 385*d4581383SWeston Andros Adamson * write / commit lists. last chance to wake any waiters */ 386*d4581383SWeston Andros Adamson nfs_unlock_request(subreq); 387*d4581383SWeston Andros Adamson 388*d4581383SWeston Andros Adamson if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) { 389*d4581383SWeston Andros Adamson /* release ref on old head request */ 390*d4581383SWeston Andros Adamson nfs_release_request(old_head); 391*d4581383SWeston Andros Adamson 392*d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 393*d4581383SWeston Andros Adamson 394*d4581383SWeston Andros Adamson /* release the PG_INODE_REF reference */ 395*d4581383SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) 396*d4581383SWeston Andros Adamson nfs_release_request(subreq); 397*d4581383SWeston Andros Adamson else 398*d4581383SWeston Andros Adamson WARN_ON_ONCE(1); 399*d4581383SWeston Andros Adamson } else { 400*d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags)); 401*d4581383SWeston Andros Adamson /* zombie requests have already released the last 402*d4581383SWeston Andros Adamson * reference and were waiting on the rest of the 403*d4581383SWeston Andros Adamson * group to complete. Since it's no longer part of a 404*d4581383SWeston Andros Adamson * group, simply free the request */ 405*d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 406*d4581383SWeston Andros Adamson nfs_free_request(subreq); 407*d4581383SWeston Andros Adamson } 408*d4581383SWeston Andros Adamson } 409*d4581383SWeston Andros Adamson } 410*d4581383SWeston Andros Adamson 411*d4581383SWeston Andros Adamson /* 412*d4581383SWeston Andros Adamson * nfs_lock_and_join_requests - join all subreqs to the head req and return 413*d4581383SWeston Andros Adamson * a locked reference, cancelling any pending 414*d4581383SWeston Andros Adamson * operations for this page. 415*d4581383SWeston Andros Adamson * 416*d4581383SWeston Andros Adamson * @page - the page used to lookup the "page group" of nfs_page structures 417*d4581383SWeston Andros Adamson * @nonblock - if true, don't block waiting for request locks 418*d4581383SWeston Andros Adamson * 419*d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 420*d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 421*d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 422*d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 423*d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 424*d4581383SWeston Andros Adamson * 425*d4581383SWeston Andros Adamson * Returns a locked, referenced pointer to the head request - which after 426*d4581383SWeston Andros Adamson * this call is guaranteed to be the only request associated with the page. 427*d4581383SWeston Andros Adamson * Returns NULL if no requests are found for @page, or a ERR_PTR if an 428*d4581383SWeston Andros Adamson * error was encountered. 429*d4581383SWeston Andros Adamson */ 430*d4581383SWeston Andros Adamson static struct nfs_page * 431*d4581383SWeston Andros Adamson nfs_lock_and_join_requests(struct page *page, bool nonblock) 432*d4581383SWeston Andros Adamson { 433*d4581383SWeston Andros Adamson struct inode *inode = page_file_mapping(page)->host; 434*d4581383SWeston Andros Adamson struct nfs_page *head, *subreq; 435*d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 436*d4581383SWeston Andros Adamson unsigned int total_bytes; 437*d4581383SWeston Andros Adamson int ret; 438*d4581383SWeston Andros Adamson 439*d4581383SWeston Andros Adamson try_again: 440*d4581383SWeston Andros Adamson total_bytes = 0; 441*d4581383SWeston Andros Adamson 442*d4581383SWeston Andros Adamson WARN_ON_ONCE(destroy_list); 443*d4581383SWeston Andros Adamson 444*d4581383SWeston Andros Adamson spin_lock(&inode->i_lock); 445*d4581383SWeston Andros Adamson 446*d4581383SWeston Andros Adamson /* 447*d4581383SWeston Andros Adamson * A reference is taken only on the head request which acts as a 448*d4581383SWeston Andros Adamson * reference to the whole page group - the group will not be destroyed 449*d4581383SWeston Andros Adamson * until the head reference is released. 450*d4581383SWeston Andros Adamson */ 451*d4581383SWeston Andros Adamson head = nfs_page_find_head_request_locked(NFS_I(inode), page); 452*d4581383SWeston Andros Adamson 453*d4581383SWeston Andros Adamson if (!head) { 454587142f8STrond Myklebust spin_unlock(&inode->i_lock); 455*d4581383SWeston Andros Adamson return NULL; 456*d4581383SWeston Andros Adamson } 457*d4581383SWeston Andros Adamson 458*d4581383SWeston Andros Adamson /* lock each request in the page group */ 459*d4581383SWeston Andros Adamson nfs_page_group_lock(head); 460*d4581383SWeston Andros Adamson subreq = head; 461*d4581383SWeston Andros Adamson do { 462*d4581383SWeston Andros Adamson /* 463*d4581383SWeston Andros Adamson * Subrequests are always contiguous, non overlapping 464*d4581383SWeston Andros Adamson * and in order. If not, it's a programming error. 465*d4581383SWeston Andros Adamson */ 466*d4581383SWeston Andros Adamson WARN_ON_ONCE(subreq->wb_offset != 467*d4581383SWeston Andros Adamson (head->wb_offset + total_bytes)); 468*d4581383SWeston Andros Adamson 469*d4581383SWeston Andros Adamson /* keep track of how many bytes this group covers */ 470*d4581383SWeston Andros Adamson total_bytes += subreq->wb_bytes; 471*d4581383SWeston Andros Adamson 472*d4581383SWeston Andros Adamson if (!nfs_lock_request(subreq)) { 473*d4581383SWeston Andros Adamson /* releases page group bit lock and 474*d4581383SWeston Andros Adamson * inode spin lock and all references */ 475*d4581383SWeston Andros Adamson ret = nfs_unroll_locks_and_wait(inode, head, 476*d4581383SWeston Andros Adamson subreq, nonblock); 477*d4581383SWeston Andros Adamson 478*d4581383SWeston Andros Adamson if (ret == 0) 479*d4581383SWeston Andros Adamson goto try_again; 480*d4581383SWeston Andros Adamson 481*d4581383SWeston Andros Adamson return ERR_PTR(ret); 482*d4581383SWeston Andros Adamson } 483*d4581383SWeston Andros Adamson 484*d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 485*d4581383SWeston Andros Adamson } while (subreq != head); 486*d4581383SWeston Andros Adamson 487*d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 488*d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 489*d4581383SWeston Andros Adamson subreq = head; 490*d4581383SWeston Andros Adamson do { 491*d4581383SWeston Andros Adamson nfs_list_remove_request(subreq); 492*d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 493*d4581383SWeston Andros Adamson } while (subreq != head); 494*d4581383SWeston Andros Adamson 495*d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 496*d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 497*d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 498*d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 499*d4581383SWeston Andros Adamson head->wb_this_page = head; 500*d4581383SWeston Andros Adamson 501*d4581383SWeston Andros Adamson /* change head request to cover whole range that 502*d4581383SWeston Andros Adamson * the former page group covered */ 503*d4581383SWeston Andros Adamson head->wb_bytes = total_bytes; 504*d4581383SWeston Andros Adamson } 505*d4581383SWeston Andros Adamson 506*d4581383SWeston Andros Adamson /* 507*d4581383SWeston Andros Adamson * prepare head request to be added to new pgio descriptor 508*d4581383SWeston Andros Adamson */ 509*d4581383SWeston Andros Adamson nfs_page_group_clear_bits(head); 510*d4581383SWeston Andros Adamson 511*d4581383SWeston Andros Adamson /* 512*d4581383SWeston Andros Adamson * some part of the group was still on the inode list - otherwise 513*d4581383SWeston Andros Adamson * the group wouldn't be involved in async write. 514*d4581383SWeston Andros Adamson * grab a reference for the head request, iff it needs one. 515*d4581383SWeston Andros Adamson */ 516*d4581383SWeston Andros Adamson if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags)) 517*d4581383SWeston Andros Adamson kref_get(&head->wb_kref); 518*d4581383SWeston Andros Adamson 519*d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 520*d4581383SWeston Andros Adamson 521*d4581383SWeston Andros Adamson /* drop lock to clear_request_commit the head req and clean up 522*d4581383SWeston Andros Adamson * requests on destroy list */ 523*d4581383SWeston Andros Adamson spin_unlock(&inode->i_lock); 524*d4581383SWeston Andros Adamson 525*d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(destroy_list, head); 526*d4581383SWeston Andros Adamson 527*d4581383SWeston Andros Adamson /* clean up commit list state */ 528*d4581383SWeston Andros Adamson nfs_clear_request_commit(head); 529*d4581383SWeston Andros Adamson 530*d4581383SWeston Andros Adamson /* still holds ref on head from nfs_page_find_head_request_locked 531*d4581383SWeston Andros Adamson * and still has lock on head from lock loop */ 532*d4581383SWeston Andros Adamson return head; 533612c9384STrond Myklebust } 534074cc1deSTrond Myklebust 535074cc1deSTrond Myklebust /* 536074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 537074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 538074cc1deSTrond Myklebust */ 539074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 540cfb506e1STrond Myklebust struct page *page, bool nonblock) 541074cc1deSTrond Myklebust { 542074cc1deSTrond Myklebust struct nfs_page *req; 543074cc1deSTrond Myklebust int ret = 0; 544074cc1deSTrond Myklebust 545*d4581383SWeston Andros Adamson req = nfs_lock_and_join_requests(page, nonblock); 546074cc1deSTrond Myklebust if (!req) 547074cc1deSTrond Myklebust goto out; 548074cc1deSTrond Myklebust ret = PTR_ERR(req); 549074cc1deSTrond Myklebust if (IS_ERR(req)) 550074cc1deSTrond Myklebust goto out; 551074cc1deSTrond Myklebust 552deed85e7STrond Myklebust nfs_set_page_writeback(page); 553deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 554074cc1deSTrond Myklebust 555deed85e7STrond Myklebust ret = 0; 556f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 557f8512ad0SFred Isaman nfs_redirty_request(req); 558074cc1deSTrond Myklebust ret = pgio->pg_error; 559f8512ad0SFred Isaman } 560074cc1deSTrond Myklebust out: 561074cc1deSTrond Myklebust return ret; 562e261f51fSTrond Myklebust } 563e261f51fSTrond Myklebust 564f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 565f758c885STrond Myklebust { 566d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 567cfb506e1STrond Myklebust int ret; 568f758c885STrond Myklebust 569f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 570f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 571f758c885STrond Myklebust 572d56b4ddfSMel Gorman nfs_pageio_cond_complete(pgio, page_file_index(page)); 5731b430beeSWu Fengguang ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); 574cfb506e1STrond Myklebust if (ret == -EAGAIN) { 575cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 576cfb506e1STrond Myklebust ret = 0; 577cfb506e1STrond Myklebust } 578cfb506e1STrond Myklebust return ret; 579f758c885STrond Myklebust } 580f758c885STrond Myklebust 581e261f51fSTrond Myklebust /* 5821da177e4SLinus Torvalds * Write an mmapped page to the server. 5831da177e4SLinus Torvalds */ 5844d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 5851da177e4SLinus Torvalds { 586f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 587e261f51fSTrond Myklebust int err; 5881da177e4SLinus Torvalds 589a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc), 590a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 591f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 592f758c885STrond Myklebust nfs_pageio_complete(&pgio); 593f758c885STrond Myklebust if (err < 0) 5944d770ccfSTrond Myklebust return err; 595f758c885STrond Myklebust if (pgio.pg_error < 0) 596f758c885STrond Myklebust return pgio.pg_error; 597f758c885STrond Myklebust return 0; 5984d770ccfSTrond Myklebust } 5994d770ccfSTrond Myklebust 6004d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 6014d770ccfSTrond Myklebust { 602f758c885STrond Myklebust int ret; 6034d770ccfSTrond Myklebust 604f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 6051da177e4SLinus Torvalds unlock_page(page); 606f758c885STrond Myklebust return ret; 607f758c885STrond Myklebust } 608f758c885STrond Myklebust 609f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 610f758c885STrond Myklebust { 611f758c885STrond Myklebust int ret; 612f758c885STrond Myklebust 613f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 614f758c885STrond Myklebust unlock_page(page); 615f758c885STrond Myklebust return ret; 6161da177e4SLinus Torvalds } 6171da177e4SLinus Torvalds 6181da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 6191da177e4SLinus Torvalds { 6201da177e4SLinus Torvalds struct inode *inode = mapping->host; 62172cb77f4STrond Myklebust unsigned long *bitlock = &NFS_I(inode)->flags; 622c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 6231da177e4SLinus Torvalds int err; 6241da177e4SLinus Torvalds 62572cb77f4STrond Myklebust /* Stop dirtying of new pages while we sync */ 62672cb77f4STrond Myklebust err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, 62772cb77f4STrond Myklebust nfs_wait_bit_killable, TASK_KILLABLE); 62872cb77f4STrond Myklebust if (err) 62972cb77f4STrond Myklebust goto out_err; 63072cb77f4STrond Myklebust 63191d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 63291d5b470SChuck Lever 633a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 634a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 635f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 636c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 63772cb77f4STrond Myklebust 63872cb77f4STrond Myklebust clear_bit_unlock(NFS_INO_FLUSHING, bitlock); 6394e857c58SPeter Zijlstra smp_mb__after_atomic(); 64072cb77f4STrond Myklebust wake_up_bit(bitlock, NFS_INO_FLUSHING); 64172cb77f4STrond Myklebust 642f758c885STrond Myklebust if (err < 0) 64372cb77f4STrond Myklebust goto out_err; 64472cb77f4STrond Myklebust err = pgio.pg_error; 64572cb77f4STrond Myklebust if (err < 0) 64672cb77f4STrond Myklebust goto out_err; 647c63c7b05STrond Myklebust return 0; 64872cb77f4STrond Myklebust out_err: 64972cb77f4STrond Myklebust return err; 6501da177e4SLinus Torvalds } 6511da177e4SLinus Torvalds 6521da177e4SLinus Torvalds /* 6531da177e4SLinus Torvalds * Insert a write request into an inode 6541da177e4SLinus Torvalds */ 655d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 6561da177e4SLinus Torvalds { 6571da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 658e7d39069STrond Myklebust 6592bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 6602bfc6e56SWeston Andros Adamson 661e7d39069STrond Myklebust /* Lock the request! */ 6627ad84aa9STrond Myklebust nfs_lock_request(req); 663e7d39069STrond Myklebust 664e7d39069STrond Myklebust spin_lock(&inode->i_lock); 665011e2a7fSBryan Schumaker if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 666a9a4a87aSTrond Myklebust inode->i_version++; 66729418aa4SMel Gorman /* 66829418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 66929418aa4SMel Gorman * with invalidate/truncate. 67029418aa4SMel Gorman */ 67129418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 6722df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 673deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 674277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 67529418aa4SMel Gorman } 6761da177e4SLinus Torvalds nfsi->npages++; 67717089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 67817089a29SWeston Andros Adamson * extra reference so sub groups can follow suit */ 67917089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 680c03b4024STrond Myklebust kref_get(&req->wb_kref); 681e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 6821da177e4SLinus Torvalds } 6831da177e4SLinus Torvalds 6841da177e4SLinus Torvalds /* 68589a09141SPeter Zijlstra * Remove a write request from an inode 6861da177e4SLinus Torvalds */ 6871da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 6881da177e4SLinus Torvalds { 6893d4ff43dSAl Viro struct inode *inode = req->wb_context->dentry->d_inode; 6901da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 69120633f04SWeston Andros Adamson struct nfs_page *head; 69220633f04SWeston Andros Adamson 69320633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 69420633f04SWeston Andros Adamson head = req->wb_head; 6951da177e4SLinus Torvalds 696587142f8STrond Myklebust spin_lock(&inode->i_lock); 69720633f04SWeston Andros Adamson if (likely(!PageSwapCache(head->wb_page))) { 69820633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 69920633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 70020633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 70129418aa4SMel Gorman } 7021da177e4SLinus Torvalds nfsi->npages--; 703587142f8STrond Myklebust spin_unlock(&inode->i_lock); 70420633f04SWeston Andros Adamson } 70517089a29SWeston Andros Adamson 70617089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 7071da177e4SLinus Torvalds nfs_release_request(req); 7081da177e4SLinus Torvalds } 7091da177e4SLinus Torvalds 71061822ab5STrond Myklebust static void 7116d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 71261822ab5STrond Myklebust { 71361822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 71461822ab5STrond Myklebust } 71561822ab5STrond Myklebust 71689d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 7178dd37758STrond Myklebust /** 7188dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 7198dd37758STrond Myklebust * @req: pointer to a struct nfs_page 720ea2cf228SFred Isaman * @dst: commit list head 721ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 7228dd37758STrond Myklebust * 723ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 7248dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 7258dd37758STrond Myklebust * the MM page stats. 7268dd37758STrond Myklebust * 727ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 7288dd37758STrond Myklebust * holding the nfs_page lock. 7298dd37758STrond Myklebust */ 7308dd37758STrond Myklebust void 731ea2cf228SFred Isaman nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, 732ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 7338dd37758STrond Myklebust { 7348dd37758STrond Myklebust set_bit(PG_CLEAN, &(req)->wb_flags); 735ea2cf228SFred Isaman spin_lock(cinfo->lock); 736ea2cf228SFred Isaman nfs_list_add_request(req, dst); 737ea2cf228SFred Isaman cinfo->mds->ncommit++; 738ea2cf228SFred Isaman spin_unlock(cinfo->lock); 73956f9cd68SFred Isaman if (!cinfo->dreq) { 7408dd37758STrond Myklebust inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 741d56b4ddfSMel Gorman inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, 74256f9cd68SFred Isaman BDI_RECLAIMABLE); 74356f9cd68SFred Isaman __mark_inode_dirty(req->wb_context->dentry->d_inode, 74456f9cd68SFred Isaman I_DIRTY_DATASYNC); 74556f9cd68SFred Isaman } 7468dd37758STrond Myklebust } 7478dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 7488dd37758STrond Myklebust 7498dd37758STrond Myklebust /** 7508dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 7518dd37758STrond Myklebust * @req: pointer to a nfs_page 752ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 7538dd37758STrond Myklebust * 754ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 7558dd37758STrond Myklebust * number of outstanding requests requiring a commit 7568dd37758STrond Myklebust * It does not update the MM page stats. 7578dd37758STrond Myklebust * 758ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 7598dd37758STrond Myklebust */ 7608dd37758STrond Myklebust void 761ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 762ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 7638dd37758STrond Myklebust { 7648dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 7658dd37758STrond Myklebust return; 7668dd37758STrond Myklebust nfs_list_remove_request(req); 767ea2cf228SFred Isaman cinfo->mds->ncommit--; 7688dd37758STrond Myklebust } 7698dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 7708dd37758STrond Myklebust 771ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 772ea2cf228SFred Isaman struct inode *inode) 773ea2cf228SFred Isaman { 774ea2cf228SFred Isaman cinfo->lock = &inode->i_lock; 775ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 776ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 777b359f9d0SFred Isaman cinfo->dreq = NULL; 778f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 779ea2cf228SFred Isaman } 780ea2cf228SFred Isaman 781ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 782ea2cf228SFred Isaman struct inode *inode, 783ea2cf228SFred Isaman struct nfs_direct_req *dreq) 784ea2cf228SFred Isaman { 7851763da12SFred Isaman if (dreq) 7861763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 7871763da12SFred Isaman else 788ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 789ea2cf228SFred Isaman } 790ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 7918dd37758STrond Myklebust 7921da177e4SLinus Torvalds /* 7931da177e4SLinus Torvalds * Add a request to the inode's commit list. 7941da177e4SLinus Torvalds */ 7951763da12SFred Isaman void 796ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 797ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 7981da177e4SLinus Torvalds { 799ea2cf228SFred Isaman if (pnfs_mark_request_commit(req, lseg, cinfo)) 8008dd37758STrond Myklebust return; 801ea2cf228SFred Isaman nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); 8021da177e4SLinus Torvalds } 8038e821cadSTrond Myklebust 804d6d6dc7cSFred Isaman static void 805d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 806e468bae9STrond Myklebust { 807e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 808d56b4ddfSMel Gorman dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE); 809e468bae9STrond Myklebust } 810d6d6dc7cSFred Isaman 8118dd37758STrond Myklebust static void 812d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 813d6d6dc7cSFred Isaman { 8148dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 8158dd37758STrond Myklebust struct inode *inode = req->wb_context->dentry->d_inode; 816ea2cf228SFred Isaman struct nfs_commit_info cinfo; 817d6d6dc7cSFred Isaman 818ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 819ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 820ea2cf228SFred Isaman spin_lock(cinfo.lock); 821ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 822ea2cf228SFred Isaman spin_unlock(cinfo.lock); 823d6d6dc7cSFred Isaman } 8248dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 8258dd37758STrond Myklebust } 826e468bae9STrond Myklebust } 827e468bae9STrond Myklebust 8288e821cadSTrond Myklebust static inline 8299c7e1b3dSAnna Schumaker int nfs_write_need_commit(struct nfs_pgio_data *data) 8308e821cadSTrond Myklebust { 831465d5243SFred Isaman if (data->verf.committed == NFS_DATA_SYNC) 832cd841605SFred Isaman return data->header->lseg == NULL; 8338e821cadSTrond Myklebust return data->verf.committed != NFS_FILE_SYNC; 8348e821cadSTrond Myklebust } 8358e821cadSTrond Myklebust 8368e821cadSTrond Myklebust #else 83768cd6fa4SBryan Schumaker static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 83868cd6fa4SBryan Schumaker struct inode *inode) 83968cd6fa4SBryan Schumaker { 84068cd6fa4SBryan Schumaker } 84168cd6fa4SBryan Schumaker 84268cd6fa4SBryan Schumaker void nfs_init_cinfo(struct nfs_commit_info *cinfo, 84368cd6fa4SBryan Schumaker struct inode *inode, 84468cd6fa4SBryan Schumaker struct nfs_direct_req *dreq) 84568cd6fa4SBryan Schumaker { 84668cd6fa4SBryan Schumaker } 84768cd6fa4SBryan Schumaker 8481763da12SFred Isaman void 849ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 850ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 8518e821cadSTrond Myklebust { 8528e821cadSTrond Myklebust } 8538e821cadSTrond Myklebust 8548dd37758STrond Myklebust static void 855e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 856e468bae9STrond Myklebust { 857e468bae9STrond Myklebust } 858e468bae9STrond Myklebust 8598e821cadSTrond Myklebust static inline 8609c7e1b3dSAnna Schumaker int nfs_write_need_commit(struct nfs_pgio_data *data) 8618e821cadSTrond Myklebust { 8628e821cadSTrond Myklebust return 0; 8638e821cadSTrond Myklebust } 8648e821cadSTrond Myklebust 8651da177e4SLinus Torvalds #endif 8661da177e4SLinus Torvalds 867061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 8686c75dc0dSFred Isaman { 869ea2cf228SFred Isaman struct nfs_commit_info cinfo; 8706c75dc0dSFred Isaman unsigned long bytes = 0; 8712bfc6e56SWeston Andros Adamson bool do_destroy; 8726c75dc0dSFred Isaman 8736c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 8746c75dc0dSFred Isaman goto out; 875ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 8766c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 8776c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 8786c75dc0dSFred Isaman 8796c75dc0dSFred Isaman bytes += req->wb_bytes; 8806c75dc0dSFred Isaman nfs_list_remove_request(req); 8816c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 8826c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 883d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 8846c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 8856c75dc0dSFred Isaman goto remove_req; 8866c75dc0dSFred Isaman } 8876c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) { 8886c75dc0dSFred Isaman nfs_mark_request_dirty(req); 8896c75dc0dSFred Isaman goto next; 8906c75dc0dSFred Isaman } 8916c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) { 892f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 893ea2cf228SFred Isaman nfs_mark_request_commit(req, hdr->lseg, &cinfo); 8946c75dc0dSFred Isaman goto next; 8956c75dc0dSFred Isaman } 8966c75dc0dSFred Isaman remove_req: 8976c75dc0dSFred Isaman nfs_inode_remove_request(req); 8986c75dc0dSFred Isaman next: 8991d1afcbcSTrond Myklebust nfs_unlock_request(req); 90020633f04SWeston Andros Adamson nfs_end_page_writeback(req); 9012bfc6e56SWeston Andros Adamson do_destroy = !test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags); 9023aff4ebbSTrond Myklebust nfs_release_request(req); 9036c75dc0dSFred Isaman } 9046c75dc0dSFred Isaman out: 9056c75dc0dSFred Isaman hdr->release(hdr); 9066c75dc0dSFred Isaman } 9076c75dc0dSFred Isaman 90889d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 909ce59515cSAnna Schumaker unsigned long 910ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 911fb8a1f11STrond Myklebust { 912ea2cf228SFred Isaman return cinfo->mds->ncommit; 913fb8a1f11STrond Myklebust } 914fb8a1f11STrond Myklebust 915ea2cf228SFred Isaman /* cinfo->lock held by caller */ 9161763da12SFred Isaman int 917ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 918ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 919d6d6dc7cSFred Isaman { 920d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 921d6d6dc7cSFred Isaman int ret = 0; 922d6d6dc7cSFred Isaman 923d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 9248dd37758STrond Myklebust if (!nfs_lock_request(req)) 9258dd37758STrond Myklebust continue; 9267ad84aa9STrond Myklebust kref_get(&req->wb_kref); 927ea2cf228SFred Isaman if (cond_resched_lock(cinfo->lock)) 9283b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 929ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 9308dd37758STrond Myklebust nfs_list_add_request(req, dst); 931d6d6dc7cSFred Isaman ret++; 9321763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 933d6d6dc7cSFred Isaman break; 934d6d6dc7cSFred Isaman } 935d6d6dc7cSFred Isaman return ret; 936d6d6dc7cSFred Isaman } 937d6d6dc7cSFred Isaman 9381da177e4SLinus Torvalds /* 9391da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 9401da177e4SLinus Torvalds * @inode: NFS inode to scan 941ea2cf228SFred Isaman * @dst: mds destination list 942ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 9431da177e4SLinus Torvalds * 9441da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 9451da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 9461da177e4SLinus Torvalds */ 9471763da12SFred Isaman int 948ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 949ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9501da177e4SLinus Torvalds { 951d6d6dc7cSFred Isaman int ret = 0; 952fb8a1f11STrond Myklebust 953ea2cf228SFred Isaman spin_lock(cinfo->lock); 954ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 9558dd37758STrond Myklebust const int max = INT_MAX; 956d6d6dc7cSFred Isaman 957ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 958ea2cf228SFred Isaman cinfo, max); 959ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 960d6d6dc7cSFred Isaman } 961ea2cf228SFred Isaman spin_unlock(cinfo->lock); 962ff778d02STrond Myklebust return ret; 9631da177e4SLinus Torvalds } 964d6d6dc7cSFred Isaman 965c42de9ddSTrond Myklebust #else 966ce59515cSAnna Schumaker unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 967fb8a1f11STrond Myklebust { 968fb8a1f11STrond Myklebust return 0; 969fb8a1f11STrond Myklebust } 970fb8a1f11STrond Myklebust 9711763da12SFred Isaman int nfs_scan_commit(struct inode *inode, struct list_head *dst, 972ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 973c42de9ddSTrond Myklebust { 974c42de9ddSTrond Myklebust return 0; 975c42de9ddSTrond Myklebust } 9761da177e4SLinus Torvalds #endif 9771da177e4SLinus Torvalds 9781da177e4SLinus Torvalds /* 979e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 980e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 9811da177e4SLinus Torvalds * 982e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 983e7d39069STrond Myklebust * to disk. 9841da177e4SLinus Torvalds */ 985e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 986e7d39069STrond Myklebust struct page *page, 987e7d39069STrond Myklebust unsigned int offset, 988e7d39069STrond Myklebust unsigned int bytes) 9891da177e4SLinus Torvalds { 990e7d39069STrond Myklebust struct nfs_page *req; 991e7d39069STrond Myklebust unsigned int rqend; 992e7d39069STrond Myklebust unsigned int end; 9931da177e4SLinus Torvalds int error; 994277459d2STrond Myklebust 995e7d39069STrond Myklebust if (!PagePrivate(page)) 996e7d39069STrond Myklebust return NULL; 997e7d39069STrond Myklebust 998e7d39069STrond Myklebust end = offset + bytes; 999e7d39069STrond Myklebust spin_lock(&inode->i_lock); 1000e7d39069STrond Myklebust 1001e7d39069STrond Myklebust for (;;) { 100284d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 1003e7d39069STrond Myklebust if (req == NULL) 1004e7d39069STrond Myklebust goto out_unlock; 1005e7d39069STrond Myklebust 10062bfc6e56SWeston Andros Adamson /* should be handled by nfs_flush_incompatible */ 10072bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 10082bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 10092bfc6e56SWeston Andros Adamson 1010e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1011e7d39069STrond Myklebust /* 1012e7d39069STrond Myklebust * Tell the caller to flush out the request if 1013e7d39069STrond Myklebust * the offsets are non-contiguous. 1014e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1015e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1016e7d39069STrond Myklebust */ 1017e468bae9STrond Myklebust if (offset > rqend 1018e7d39069STrond Myklebust || end < req->wb_offset) 1019e7d39069STrond Myklebust goto out_flushme; 1020e7d39069STrond Myklebust 10217ad84aa9STrond Myklebust if (nfs_lock_request(req)) 1022e7d39069STrond Myklebust break; 1023e7d39069STrond Myklebust 1024e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 1025587142f8STrond Myklebust spin_unlock(&inode->i_lock); 10261da177e4SLinus Torvalds error = nfs_wait_on_request(req); 10271da177e4SLinus Torvalds nfs_release_request(req); 1028e7d39069STrond Myklebust if (error != 0) 1029e7d39069STrond Myklebust goto out_err; 1030e7d39069STrond Myklebust spin_lock(&inode->i_lock); 10311da177e4SLinus Torvalds } 10321da177e4SLinus Torvalds 10331da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 10341da177e4SLinus Torvalds if (offset < req->wb_offset) { 10351da177e4SLinus Torvalds req->wb_offset = offset; 10361da177e4SLinus Torvalds req->wb_pgbase = offset; 10371da177e4SLinus Torvalds } 10381da177e4SLinus Torvalds if (end > rqend) 10391da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1040e7d39069STrond Myklebust else 1041e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 1042e7d39069STrond Myklebust out_unlock: 1043e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 1044ca138f36SFred Isaman if (req) 10458dd37758STrond Myklebust nfs_clear_request_commit(req); 1046e7d39069STrond Myklebust return req; 1047e7d39069STrond Myklebust out_flushme: 1048e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 1049e7d39069STrond Myklebust nfs_release_request(req); 1050e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1051e7d39069STrond Myklebust out_err: 1052e7d39069STrond Myklebust return ERR_PTR(error); 1053e7d39069STrond Myklebust } 10541da177e4SLinus Torvalds 1055e7d39069STrond Myklebust /* 1056e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1057e7d39069STrond Myklebust * 1058e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1059e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1060e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1061e7d39069STrond Myklebust */ 1062e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1063e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1064e7d39069STrond Myklebust { 1065d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1066e7d39069STrond Myklebust struct nfs_page *req; 1067e7d39069STrond Myklebust 1068e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1069e7d39069STrond Myklebust if (req != NULL) 1070e7d39069STrond Myklebust goto out; 10712bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 1072e7d39069STrond Myklebust if (IS_ERR(req)) 1073e7d39069STrond Myklebust goto out; 1074d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1075efc91ed0STrond Myklebust out: 107661e930a9STrond Myklebust return req; 10771da177e4SLinus Torvalds } 10781da177e4SLinus Torvalds 1079e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1080e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1081e7d39069STrond Myklebust { 1082e7d39069STrond Myklebust struct nfs_page *req; 1083e7d39069STrond Myklebust 1084e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1085e7d39069STrond Myklebust if (IS_ERR(req)) 1086e7d39069STrond Myklebust return PTR_ERR(req); 1087e7d39069STrond Myklebust /* Update file length */ 1088e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1089d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1090a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 10911d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1092e7d39069STrond Myklebust return 0; 1093e7d39069STrond Myklebust } 1094e7d39069STrond Myklebust 10951da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 10961da177e4SLinus Torvalds { 1097cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 10982a369153STrond Myklebust struct nfs_lock_context *l_ctx; 10991da177e4SLinus Torvalds struct nfs_page *req; 11001a54533eSTrond Myklebust int do_flush, status; 11011da177e4SLinus Torvalds /* 11021da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 11031da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 11041da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 11051da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11061da177e4SLinus Torvalds * Also do the same if we find a request from an existing 11071da177e4SLinus Torvalds * dropped page. 11081da177e4SLinus Torvalds */ 11091a54533eSTrond Myklebust do { 111084d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 11111a54533eSTrond Myklebust if (req == NULL) 11121a54533eSTrond Myklebust return 0; 11132a369153STrond Myklebust l_ctx = req->wb_lock_context; 11142a369153STrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx; 11152bfc6e56SWeston Andros Adamson /* for now, flush if more than 1 request in page_group */ 11162bfc6e56SWeston Andros Adamson do_flush |= req->wb_this_page != req; 11170f1d2605STrond Myklebust if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) { 11182a369153STrond Myklebust do_flush |= l_ctx->lockowner.l_owner != current->files 11192a369153STrond Myklebust || l_ctx->lockowner.l_pid != current->tgid; 11202a369153STrond Myklebust } 11211da177e4SLinus Torvalds nfs_release_request(req); 11221a54533eSTrond Myklebust if (!do_flush) 11231a54533eSTrond Myklebust return 0; 1124d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 11251a54533eSTrond Myklebust } while (status == 0); 11261a54533eSTrond Myklebust return status; 11271da177e4SLinus Torvalds } 11281da177e4SLinus Torvalds 11291da177e4SLinus Torvalds /* 1130dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1131dc24826bSAndy Adamson * expire soon. 1132dc24826bSAndy Adamson * 1133dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1134dc24826bSAndy Adamson * 1135dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1136dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1137dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1138dc24826bSAndy Adamson */ 1139dc24826bSAndy Adamson int 1140dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1141dc24826bSAndy Adamson { 1142dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1143dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1144dc24826bSAndy Adamson 1145dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 1146dc24826bSAndy Adamson } 1147dc24826bSAndy Adamson 1148dc24826bSAndy Adamson /* 1149dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1150dc24826bSAndy Adamson */ 1151dc24826bSAndy Adamson bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) 1152dc24826bSAndy Adamson { 1153dc24826bSAndy Adamson return rpcauth_cred_key_to_expire(ctx->cred); 1154dc24826bSAndy Adamson } 1155dc24826bSAndy Adamson 1156dc24826bSAndy Adamson /* 11575d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 11585d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 11595d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 11605d47a356STrond Myklebust */ 11618d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 11625d47a356STrond Myklebust { 1163d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1164d529ef83SJeff Layton 11658d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 11668d197a56STrond Myklebust goto out; 116718dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1168d529ef83SJeff Layton return false; 11694db72b40SJeff Layton smp_rmb(); 1170d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 11718d197a56STrond Myklebust return false; 11728d197a56STrond Myklebust out: 117318dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 117418dd78c4SScott Mayhew return false; 11758d197a56STrond Myklebust return PageUptodate(page) != 0; 11765d47a356STrond Myklebust } 11775d47a356STrond Myklebust 1178c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1179c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1180c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1181c7559663SScott Mayhew * inefficiencies. 1182c7559663SScott Mayhew * 1183263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1184263b4509SScott Mayhew * of the checks. 1185c7559663SScott Mayhew */ 1186c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1187c7559663SScott Mayhew { 1188c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1189c7559663SScott Mayhew return 0; 1190263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1191263b4509SScott Mayhew return 0; 1192c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1193c7559663SScott Mayhew return 1; 1194263b4509SScott Mayhew if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 && 1195c7559663SScott Mayhew inode->i_flock->fl_end == OFFSET_MAX && 1196263b4509SScott Mayhew inode->i_flock->fl_type != F_RDLCK)) 1197c7559663SScott Mayhew return 1; 1198c7559663SScott Mayhew return 0; 1199c7559663SScott Mayhew } 1200c7559663SScott Mayhew 12015d47a356STrond Myklebust /* 12021da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 12031da177e4SLinus Torvalds * 12041da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 12051da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 12061da177e4SLinus Torvalds */ 12071da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 12081da177e4SLinus Torvalds unsigned int offset, unsigned int count) 12091da177e4SLinus Torvalds { 1210cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1211d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 12121da177e4SLinus Torvalds int status = 0; 12131da177e4SLinus Torvalds 121491d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 121591d5b470SChuck Lever 12166de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 12176de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 12181da177e4SLinus Torvalds 1219c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 122049a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 12211da177e4SLinus Torvalds offset = 0; 12221da177e4SLinus Torvalds } 12231da177e4SLinus Torvalds 1224e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 122503fa9e84STrond Myklebust if (status < 0) 122603fa9e84STrond Myklebust nfs_set_pageerror(page); 122759b7c05fSTrond Myklebust else 122859b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 12291da177e4SLinus Torvalds 123048186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 12311da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 12321da177e4SLinus Torvalds return status; 12331da177e4SLinus Torvalds } 12341da177e4SLinus Torvalds 12353ff7576dSTrond Myklebust static int flush_task_priority(int how) 12361da177e4SLinus Torvalds { 12371da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 12381da177e4SLinus Torvalds case FLUSH_HIGHPRI: 12391da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 12401da177e4SLinus Torvalds case FLUSH_LOWPRI: 12411da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 12421da177e4SLinus Torvalds } 12431da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 12441da177e4SLinus Torvalds } 12451da177e4SLinus Torvalds 12461ed26f33SAnna Schumaker static void nfs_initiate_write(struct nfs_pgio_data *data, struct rpc_message *msg, 12471ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 12481da177e4SLinus Torvalds { 1249cd841605SFred Isaman struct inode *inode = data->header->inode; 12503ff7576dSTrond Myklebust int priority = flush_task_priority(how); 12511da177e4SLinus Torvalds 12521ed26f33SAnna Schumaker task_setup_data->priority = priority; 12531ed26f33SAnna Schumaker NFS_PROTO(inode)->write_setup(data, msg); 1254d138d5d1SAndy Adamson 12558c21c62cSWeston Andros Adamson nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client, 12561ed26f33SAnna Schumaker &task_setup_data->rpc_client, msg, data); 1257275acaafSTrond Myklebust } 1258275acaafSTrond Myklebust 12596d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 12606d884e8fSFred * call this on each, which will prepare them to be retried on next 12616d884e8fSFred * writeback using standard nfs. 12626d884e8fSFred */ 12636d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 12646d884e8fSFred { 12656d884e8fSFred nfs_mark_request_dirty(req); 12661d1afcbcSTrond Myklebust nfs_unlock_request(req); 126720633f04SWeston Andros Adamson nfs_end_page_writeback(req); 12683aff4ebbSTrond Myklebust nfs_release_request(req); 12696d884e8fSFred } 12706d884e8fSFred 1271061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 12726c75dc0dSFred Isaman { 12736c75dc0dSFred Isaman struct nfs_page *req; 12746c75dc0dSFred Isaman 12756c75dc0dSFred Isaman while (!list_empty(head)) { 12766c75dc0dSFred Isaman req = nfs_list_entry(head->next); 12776c75dc0dSFred Isaman nfs_list_remove_request(req); 12786c75dc0dSFred Isaman nfs_redirty_request(req); 12796c75dc0dSFred Isaman } 12806c75dc0dSFred Isaman } 12816c75dc0dSFred Isaman 1282061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1283061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1284061ae2edSFred Isaman .completion = nfs_write_completion, 1285061ae2edSFred Isaman }; 1286061ae2edSFred Isaman 128757208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1288a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1289061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 12901751c363STrond Myklebust { 1291a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 129241d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1293a20c93e3SChristoph Hellwig 1294a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1295a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1296a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1297a20c93e3SChristoph Hellwig #endif 12984a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 12994a0de55cSAnna Schumaker server->wsize, ioflags); 13001751c363STrond Myklebust } 1301ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 13021751c363STrond Myklebust 1303dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1304dce81290STrond Myklebust { 130541d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1306dce81290STrond Myklebust pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1307dce81290STrond Myklebust } 13081f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1309dce81290STrond Myklebust 13101da177e4SLinus Torvalds 13110b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 13120b7c0153SFred Isaman { 13130b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 13140b7c0153SFred Isaman 13150b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 13160b7c0153SFred Isaman } 13170b7c0153SFred Isaman 1318a4cdda59SAnna Schumaker static void nfs_writeback_release_common(struct nfs_pgio_data *data) 13191da177e4SLinus Torvalds { 1320cd841605SFred Isaman struct nfs_pgio_header *hdr = data->header; 1321e2fecb21STrond Myklebust int status = data->task.tk_status; 1322788e7a89STrond Myklebust 13236c75dc0dSFred Isaman if ((status >= 0) && nfs_write_need_commit(data)) { 13246c75dc0dSFred Isaman spin_lock(&hdr->lock); 13256c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) 13266c75dc0dSFred Isaman ; /* Do nothing */ 13276c75dc0dSFred Isaman else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) 1328f79d06f5SAnna Schumaker memcpy(&hdr->verf, &data->verf, sizeof(hdr->verf)); 1329f79d06f5SAnna Schumaker else if (memcmp(&hdr->verf, &data->verf, sizeof(hdr->verf))) 13306c75dc0dSFred Isaman set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags); 13316c75dc0dSFred Isaman spin_unlock(&hdr->lock); 13321da177e4SLinus Torvalds } 13331da177e4SLinus Torvalds } 13341da177e4SLinus Torvalds 13351f2edbe3STrond Myklebust /* 13361f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 13371f2edbe3STrond Myklebust */ 13381f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 13391f2edbe3STrond Myklebust { 13401f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 13411f2edbe3STrond Myklebust int kill = 0; 1342788e7a89STrond Myklebust 13431f2edbe3STrond Myklebust /* suid always must be killed */ 13441f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 13451f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 13461f2edbe3STrond Myklebust 13471f2edbe3STrond Myklebust /* 13481f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 13491f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 13501f2edbe3STrond Myklebust */ 13511f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 13521f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 13531f2edbe3STrond Myklebust 13541f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 13551f2edbe3STrond Myklebust return kill; 13561f2edbe3STrond Myklebust 13571f2edbe3STrond Myklebust return 0; 13581f2edbe3STrond Myklebust } 1359788e7a89STrond Myklebust 13601da177e4SLinus Torvalds /* 13611da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 13621da177e4SLinus Torvalds */ 13630eecb214SAnna Schumaker static int nfs_writeback_done(struct rpc_task *task, struct nfs_pgio_data *data, 13640eecb214SAnna Schumaker struct inode *inode) 13651da177e4SLinus Torvalds { 1366788e7a89STrond Myklebust int status; 13671da177e4SLinus Torvalds 1368f551e44fSChuck Lever /* 1369f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1370f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1371f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1372f551e44fSChuck Lever * another writer had changed the file, but some applications 1373f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1374f551e44fSChuck Lever */ 1375cd841605SFred Isaman status = NFS_PROTO(inode)->write_done(task, data); 1376788e7a89STrond Myklebust if (status != 0) 13770eecb214SAnna Schumaker return status; 13780eecb214SAnna Schumaker nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, data->res.count); 137991d5b470SChuck Lever 138089d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 13810eecb214SAnna Schumaker if (data->res.verf->committed < data->args.stable && task->tk_status >= 0) { 13821da177e4SLinus Torvalds /* We tried a write call, but the server did not 13831da177e4SLinus Torvalds * commit data to stable storage even though we 13841da177e4SLinus Torvalds * requested it. 13851da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 13861da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 13871da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 13881da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 13891da177e4SLinus Torvalds */ 13901da177e4SLinus Torvalds static unsigned long complain; 13911da177e4SLinus Torvalds 1392a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 13931da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 13941da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 13951da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1396cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 13970eecb214SAnna Schumaker data->res.verf->committed, data->args.stable); 13981da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 13991da177e4SLinus Torvalds } 14001da177e4SLinus Torvalds } 14011da177e4SLinus Torvalds #endif 14021f2edbe3STrond Myklebust 14031f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 14041f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 14051f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 14060eecb214SAnna Schumaker return 0; 14070eecb214SAnna Schumaker } 14080eecb214SAnna Schumaker 14090eecb214SAnna Schumaker /* 14100eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 14110eecb214SAnna Schumaker */ 14120eecb214SAnna Schumaker static void nfs_writeback_result(struct rpc_task *task, struct nfs_pgio_data *data) 14130eecb214SAnna Schumaker { 14140eecb214SAnna Schumaker struct nfs_pgio_args *argp = &data->args; 14150eecb214SAnna Schumaker struct nfs_pgio_res *resp = &data->res; 14161f2edbe3STrond Myklebust 14171f2edbe3STrond Myklebust if (resp->count < argp->count) { 14181da177e4SLinus Torvalds static unsigned long complain; 14191da177e4SLinus Torvalds 14206c75dc0dSFred Isaman /* This a short write! */ 14210eecb214SAnna Schumaker nfs_inc_stats(data->header->inode, NFSIOS_SHORTWRITE); 142291d5b470SChuck Lever 14231da177e4SLinus Torvalds /* Has the server at least made some progress? */ 14246c75dc0dSFred Isaman if (resp->count == 0) { 14256c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 14266c75dc0dSFred Isaman printk(KERN_WARNING 14276c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 14286c75dc0dSFred Isaman argp->count); 14296c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 14306c75dc0dSFred Isaman } 14316c75dc0dSFred Isaman nfs_set_pgio_error(data->header, -EIO, argp->offset); 14326c75dc0dSFred Isaman task->tk_status = -EIO; 14336c75dc0dSFred Isaman return; 14346c75dc0dSFred Isaman } 14351da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 14361da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 14371da177e4SLinus Torvalds /* Resend from where the server left off */ 1438a69aef14SFred Isaman data->mds_offset += resp->count; 14391da177e4SLinus Torvalds argp->offset += resp->count; 14401da177e4SLinus Torvalds argp->pgbase += resp->count; 14411da177e4SLinus Torvalds argp->count -= resp->count; 14421da177e4SLinus Torvalds } else { 14431da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 14441da177e4SLinus Torvalds * headaches in the case of a server crash. 14451da177e4SLinus Torvalds */ 14461da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 14471da177e4SLinus Torvalds } 1448d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 14491da177e4SLinus Torvalds } 14501da177e4SLinus Torvalds } 14511da177e4SLinus Torvalds 14521da177e4SLinus Torvalds 145389d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 145471d0a611STrond Myklebust static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) 145571d0a611STrond Myklebust { 1456b8413f98STrond Myklebust int ret; 1457b8413f98STrond Myklebust 145871d0a611STrond Myklebust if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) 145971d0a611STrond Myklebust return 1; 1460b8413f98STrond Myklebust if (!may_wait) 146171d0a611STrond Myklebust return 0; 1462b8413f98STrond Myklebust ret = out_of_line_wait_on_bit_lock(&nfsi->flags, 1463b8413f98STrond Myklebust NFS_INO_COMMIT, 1464b8413f98STrond Myklebust nfs_wait_bit_killable, 1465b8413f98STrond Myklebust TASK_KILLABLE); 1466b8413f98STrond Myklebust return (ret < 0) ? ret : 1; 146771d0a611STrond Myklebust } 146871d0a611STrond Myklebust 1469f453a54aSFred Isaman static void nfs_commit_clear_lock(struct nfs_inode *nfsi) 147071d0a611STrond Myklebust { 147171d0a611STrond Myklebust clear_bit(NFS_INO_COMMIT, &nfsi->flags); 14724e857c58SPeter Zijlstra smp_mb__after_atomic(); 147371d0a611STrond Myklebust wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); 147471d0a611STrond Myklebust } 147571d0a611STrond Myklebust 14760b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 14771da177e4SLinus Torvalds { 14780b7c0153SFred Isaman put_nfs_open_context(data->context); 14790b7c0153SFred Isaman nfs_commit_free(data); 14801da177e4SLinus Torvalds } 1481e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 14821da177e4SLinus Torvalds 14830b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 14849ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 14859f0ec176SAndy Adamson int how, int flags) 14861da177e4SLinus Torvalds { 148707737691STrond Myklebust struct rpc_task *task; 14889ace33cdSFred Isaman int priority = flush_task_priority(how); 1489bdc7f021STrond Myklebust struct rpc_message msg = { 1490bdc7f021STrond Myklebust .rpc_argp = &data->args, 1491bdc7f021STrond Myklebust .rpc_resp = &data->res, 14929ace33cdSFred Isaman .rpc_cred = data->cred, 1493bdc7f021STrond Myklebust }; 149484115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 149507737691STrond Myklebust .task = &data->task, 14969ace33cdSFred Isaman .rpc_client = clnt, 1497bdc7f021STrond Myklebust .rpc_message = &msg, 14989ace33cdSFred Isaman .callback_ops = call_ops, 149984115e1cSTrond Myklebust .callback_data = data, 1500101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 15019f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 15023ff7576dSTrond Myklebust .priority = priority, 150384115e1cSTrond Myklebust }; 1504788e7a89STrond Myklebust /* Set up the initial task struct. */ 15059ace33cdSFred Isaman NFS_PROTO(data->inode)->commit_setup(data, &msg); 15061da177e4SLinus Torvalds 1507a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1508bdc7f021STrond Myklebust 15098c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 15108c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 15118c21c62cSWeston Andros Adamson 151207737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1513dbae4c73STrond Myklebust if (IS_ERR(task)) 1514dbae4c73STrond Myklebust return PTR_ERR(task); 1515d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1516d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 151707737691STrond Myklebust rpc_put_task(task); 1518dbae4c73STrond Myklebust return 0; 15191da177e4SLinus Torvalds } 1520e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 15211da177e4SLinus Torvalds 15221da177e4SLinus Torvalds /* 15239ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 15249ace33cdSFred Isaman */ 15250b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1526988b6dceSFred Isaman struct list_head *head, 1527f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1528f453a54aSFred Isaman struct nfs_commit_info *cinfo) 15299ace33cdSFred Isaman { 15309ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 15313d4ff43dSAl Viro struct inode *inode = first->wb_context->dentry->d_inode; 15329ace33cdSFred Isaman 15339ace33cdSFred Isaman /* Set up the RPC argument and reply structs 15349ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 15359ace33cdSFred Isaman 15369ace33cdSFred Isaman list_splice_init(head, &data->pages); 15379ace33cdSFred Isaman 15389ace33cdSFred Isaman data->inode = inode; 15399ace33cdSFred Isaman data->cred = first->wb_context->cred; 1540988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 15419ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1542f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1543b359f9d0SFred Isaman data->dreq = cinfo->dreq; 15449ace33cdSFred Isaman 15459ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 15469ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 15479ace33cdSFred Isaman data->args.offset = 0; 15489ace33cdSFred Isaman data->args.count = 0; 15490b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 15509ace33cdSFred Isaman data->res.fattr = &data->fattr; 15519ace33cdSFred Isaman data->res.verf = &data->verf; 15529ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 15539ace33cdSFred Isaman } 1554e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 15559ace33cdSFred Isaman 1556e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1557ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1558ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 155964bfeb49SFred Isaman { 156064bfeb49SFred Isaman struct nfs_page *req; 156164bfeb49SFred Isaman 156264bfeb49SFred Isaman while (!list_empty(page_list)) { 156364bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 156464bfeb49SFred Isaman nfs_list_remove_request(req); 1565ea2cf228SFred Isaman nfs_mark_request_commit(req, lseg, cinfo); 156656f9cd68SFred Isaman if (!cinfo->dreq) { 156764bfeb49SFred Isaman dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1568d56b4ddfSMel Gorman dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, 156964bfeb49SFred Isaman BDI_RECLAIMABLE); 157056f9cd68SFred Isaman } 15711d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 157264bfeb49SFred Isaman } 157364bfeb49SFred Isaman } 1574e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 157564bfeb49SFred Isaman 15769ace33cdSFred Isaman /* 15771da177e4SLinus Torvalds * Commit dirty pages 15781da177e4SLinus Torvalds */ 15791da177e4SLinus Torvalds static int 1580ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1581ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 15821da177e4SLinus Torvalds { 15830b7c0153SFred Isaman struct nfs_commit_data *data; 15841da177e4SLinus Torvalds 1585c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 15861da177e4SLinus Torvalds 15871da177e4SLinus Torvalds if (!data) 15881da177e4SLinus Torvalds goto out_bad; 15891da177e4SLinus Torvalds 15901da177e4SLinus Torvalds /* Set up the argument struct */ 1591f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1592f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 15939f0ec176SAndy Adamson return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, 15949f0ec176SAndy Adamson how, 0); 15951da177e4SLinus Torvalds out_bad: 1596ea2cf228SFred Isaman nfs_retry_commit(head, NULL, cinfo); 1597f453a54aSFred Isaman cinfo->completion_ops->error_cleanup(NFS_I(inode)); 15981da177e4SLinus Torvalds return -ENOMEM; 15991da177e4SLinus Torvalds } 16001da177e4SLinus Torvalds 16011da177e4SLinus Torvalds /* 16021da177e4SLinus Torvalds * COMMIT call returned 16031da177e4SLinus Torvalds */ 1604788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 16051da177e4SLinus Torvalds { 16060b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 16071da177e4SLinus Torvalds 1608a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 16091da177e4SLinus Torvalds task->tk_pid, task->tk_status); 16101da177e4SLinus Torvalds 1611788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1612c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1613c9d8f89dSTrond Myklebust } 1614c9d8f89dSTrond Myklebust 1615f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1616c9d8f89dSTrond Myklebust { 1617c9d8f89dSTrond Myklebust struct nfs_page *req; 1618c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1619f453a54aSFred Isaman struct nfs_commit_info cinfo; 1620788e7a89STrond Myklebust 16211da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 16221da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 16231da177e4SLinus Torvalds nfs_list_remove_request(req); 1624d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 16251da177e4SLinus Torvalds 16261e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 16273d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 16281e8968c5SNiels de Vos (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode), 16291da177e4SLinus Torvalds req->wb_bytes, 16301da177e4SLinus Torvalds (long long)req_offset(req)); 1631c9d8f89dSTrond Myklebust if (status < 0) { 1632c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 16331da177e4SLinus Torvalds nfs_inode_remove_request(req); 1634c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 16351da177e4SLinus Torvalds goto next; 16361da177e4SLinus Torvalds } 16371da177e4SLinus Torvalds 16381da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 16391da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 16402f2c63bcSTrond Myklebust if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { 16411da177e4SLinus Torvalds /* We have a match */ 16421da177e4SLinus Torvalds nfs_inode_remove_request(req); 16431da177e4SLinus Torvalds dprintk(" OK\n"); 16441da177e4SLinus Torvalds goto next; 16451da177e4SLinus Torvalds } 16461da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 16471da177e4SLinus Torvalds dprintk(" mismatch\n"); 16486d884e8fSFred nfs_mark_request_dirty(req); 164905990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 16501da177e4SLinus Torvalds next: 16511d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 16521da177e4SLinus Torvalds } 1653f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1654f453a54aSFred Isaman if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) 1655f453a54aSFred Isaman nfs_commit_clear_lock(NFS_I(data->inode)); 16565917ce84SFred Isaman } 16575917ce84SFred Isaman 16585917ce84SFred Isaman static void nfs_commit_release(void *calldata) 16595917ce84SFred Isaman { 16600b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 16615917ce84SFred Isaman 1662f453a54aSFred Isaman data->completion_ops->completion(data); 1663c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 16641da177e4SLinus Torvalds } 1665788e7a89STrond Myklebust 1666788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 16670b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1668788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1669788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1670788e7a89STrond Myklebust }; 16711da177e4SLinus Torvalds 1672f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1673f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1674f453a54aSFred Isaman .error_cleanup = nfs_commit_clear_lock, 1675f453a54aSFred Isaman }; 1676f453a54aSFred Isaman 16771763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1678ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 167984c53ab5SFred Isaman { 168084c53ab5SFred Isaman int status; 168184c53ab5SFred Isaman 1682ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 168384c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1684ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 168584c53ab5SFred Isaman return status; 168684c53ab5SFred Isaman } 168784c53ab5SFred Isaman 1688b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 16891da177e4SLinus Torvalds { 16901da177e4SLinus Torvalds LIST_HEAD(head); 1691ea2cf228SFred Isaman struct nfs_commit_info cinfo; 169271d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1693b8413f98STrond Myklebust int res; 16941da177e4SLinus Torvalds 1695b8413f98STrond Myklebust res = nfs_commit_set_lock(NFS_I(inode), may_wait); 1696b8413f98STrond Myklebust if (res <= 0) 1697c5efa5fcSTrond Myklebust goto out_mark_dirty; 1698ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1699ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 17001da177e4SLinus Torvalds if (res) { 1701a861a1e1SFred Isaman int error; 1702a861a1e1SFred Isaman 1703ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 17041da177e4SLinus Torvalds if (error < 0) 17051da177e4SLinus Torvalds return error; 1706b8413f98STrond Myklebust if (!may_wait) 1707b8413f98STrond Myklebust goto out_mark_dirty; 1708b8413f98STrond Myklebust error = wait_on_bit(&NFS_I(inode)->flags, 1709b8413f98STrond Myklebust NFS_INO_COMMIT, 171071d0a611STrond Myklebust nfs_wait_bit_killable, 171171d0a611STrond Myklebust TASK_KILLABLE); 1712b8413f98STrond Myklebust if (error < 0) 1713b8413f98STrond Myklebust return error; 171471d0a611STrond Myklebust } else 171571d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(inode)); 1716c5efa5fcSTrond Myklebust return res; 1717c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1718c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1719c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1720c5efa5fcSTrond Myklebust * that the data is on the disk. 1721c5efa5fcSTrond Myklebust */ 1722c5efa5fcSTrond Myklebust out_mark_dirty: 1723c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 17241da177e4SLinus Torvalds return res; 17251da177e4SLinus Torvalds } 17268fc795f7STrond Myklebust 17278fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 17288fc795f7STrond Myklebust { 1729420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1730420e3646STrond Myklebust int flags = FLUSH_SYNC; 1731420e3646STrond Myklebust int ret = 0; 17328fc795f7STrond Myklebust 17333236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1734ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 17353236c3e1SJeff Layton return ret; 17363236c3e1SJeff Layton 1737a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1738a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1739a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1740420e3646STrond Myklebust */ 1741ea2cf228SFred Isaman if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1)) 1742420e3646STrond Myklebust goto out_mark_dirty; 1743420e3646STrond Myklebust 1744a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1745420e3646STrond Myklebust flags = 0; 1746a00dd6c0SJeff Layton } 1747a00dd6c0SJeff Layton 1748420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1749420e3646STrond Myklebust if (ret >= 0) { 1750420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1751420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1752420e3646STrond Myklebust wbc->nr_to_write -= ret; 1753420e3646STrond Myklebust else 1754420e3646STrond Myklebust wbc->nr_to_write = 0; 1755420e3646STrond Myklebust } 17568fc795f7STrond Myklebust return 0; 1757420e3646STrond Myklebust } 1758420e3646STrond Myklebust out_mark_dirty: 17598fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 17608fc795f7STrond Myklebust return ret; 17618fc795f7STrond Myklebust } 1762c63c7b05STrond Myklebust #else 17638fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 17648fc795f7STrond Myklebust { 17658fc795f7STrond Myklebust return 0; 17668fc795f7STrond Myklebust } 17671da177e4SLinus Torvalds #endif 17681da177e4SLinus Torvalds 17698fc795f7STrond Myklebust int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 17708fc795f7STrond Myklebust { 1771a8d8f02cSBryan Schumaker return nfs_commit_unstable_pages(inode, wbc); 1772a8d8f02cSBryan Schumaker } 177389d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1774863a3c6cSAndy Adamson 1775acdc53b2STrond Myklebust /* 1776acdc53b2STrond Myklebust * flush the inode to disk. 1777acdc53b2STrond Myklebust */ 1778acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 177934901f70STrond Myklebust { 178034901f70STrond Myklebust struct writeback_control wbc = { 178172cb77f4STrond Myklebust .sync_mode = WB_SYNC_ALL, 178234901f70STrond Myklebust .nr_to_write = LONG_MAX, 1783d7fb1207STrond Myklebust .range_start = 0, 1784d7fb1207STrond Myklebust .range_end = LLONG_MAX, 178534901f70STrond Myklebust }; 1786f4ce1299STrond Myklebust int ret; 178734901f70STrond Myklebust 1788f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1789f4ce1299STrond Myklebust 1790f4ce1299STrond Myklebust ret = sync_inode(inode, &wbc); 1791f4ce1299STrond Myklebust 1792f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1793f4ce1299STrond Myklebust return ret; 17941c75950bSTrond Myklebust } 1795ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 17961c75950bSTrond Myklebust 17971b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 17981b3b4a1aSTrond Myklebust { 17991b3b4a1aSTrond Myklebust struct nfs_page *req; 18001b3b4a1aSTrond Myklebust int ret = 0; 18011b3b4a1aSTrond Myklebust 18021b3b4a1aSTrond Myklebust for (;;) { 1803ba8b06e6STrond Myklebust wait_on_page_writeback(page); 180484d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 18051b3b4a1aSTrond Myklebust if (req == NULL) 18061b3b4a1aSTrond Myklebust break; 18077ad84aa9STrond Myklebust if (nfs_lock_request(req)) { 18088dd37758STrond Myklebust nfs_clear_request_commit(req); 18091b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 18101b3b4a1aSTrond Myklebust /* 18111b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 18121b3b4a1aSTrond Myklebust * page as being dirty 18131b3b4a1aSTrond Myklebust */ 18141b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 18151d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18161b3b4a1aSTrond Myklebust break; 18171b3b4a1aSTrond Myklebust } 18181b3b4a1aSTrond Myklebust ret = nfs_wait_on_request(req); 1819c9edda71STrond Myklebust nfs_release_request(req); 18201b3b4a1aSTrond Myklebust if (ret < 0) 1821c988950eSTrond Myklebust break; 18221b3b4a1aSTrond Myklebust } 18231b3b4a1aSTrond Myklebust return ret; 18241b3b4a1aSTrond Myklebust } 18251b3b4a1aSTrond Myklebust 18261c75950bSTrond Myklebust /* 18271c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 18281c75950bSTrond Myklebust */ 18291c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 18301c75950bSTrond Myklebust { 183129418aa4SMel Gorman loff_t range_start = page_file_offset(page); 18327f2f12d9STrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 18337f2f12d9STrond Myklebust struct writeback_control wbc = { 18347f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 18357f2f12d9STrond Myklebust .nr_to_write = 0, 18367f2f12d9STrond Myklebust .range_start = range_start, 18377f2f12d9STrond Myklebust .range_end = range_end, 18387f2f12d9STrond Myklebust }; 18397f2f12d9STrond Myklebust int ret; 18407f2f12d9STrond Myklebust 1841f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 1842f4ce1299STrond Myklebust 18430522f6adSTrond Myklebust for (;;) { 1844ba8b06e6STrond Myklebust wait_on_page_writeback(page); 18457f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 18467f2f12d9STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 18477f2f12d9STrond Myklebust if (ret < 0) 18487f2f12d9STrond Myklebust goto out_error; 18490522f6adSTrond Myklebust continue; 18507f2f12d9STrond Myklebust } 1851f4ce1299STrond Myklebust ret = 0; 18520522f6adSTrond Myklebust if (!PagePrivate(page)) 18530522f6adSTrond Myklebust break; 18540522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 18557f2f12d9STrond Myklebust if (ret < 0) 18567f2f12d9STrond Myklebust goto out_error; 18577f2f12d9STrond Myklebust } 18587f2f12d9STrond Myklebust out_error: 1859f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 18607f2f12d9STrond Myklebust return ret; 18611c75950bSTrond Myklebust } 18621c75950bSTrond Myklebust 1863074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 1864074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 1865a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 1866074cc1deSTrond Myklebust { 18672da95652SJeff Layton /* 18682da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 18692da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 18702da95652SJeff Layton * 18712da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 18722da95652SJeff Layton * that we can safely release the inode reference while holding 18732da95652SJeff Layton * the page lock. 18742da95652SJeff Layton */ 18752da95652SJeff Layton if (PagePrivate(page)) 18762da95652SJeff Layton return -EBUSY; 1877074cc1deSTrond Myklebust 18788c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 18798c209ce7SDavid Howells return -EBUSY; 1880074cc1deSTrond Myklebust 1881a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 1882074cc1deSTrond Myklebust } 1883074cc1deSTrond Myklebust #endif 1884074cc1deSTrond Myklebust 1885f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 18861da177e4SLinus Torvalds { 18871da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 1888c0752cdfSAnna Schumaker sizeof(struct nfs_rw_header), 18891da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 189020c2df83SPaul Mundt NULL); 18911da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 18921da177e4SLinus Torvalds return -ENOMEM; 18931da177e4SLinus Torvalds 189493d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 18951da177e4SLinus Torvalds nfs_wdata_cachep); 18961da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 18973dd4765fSJeff Layton goto out_destroy_write_cache; 18981da177e4SLinus Torvalds 18990b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 19000b7c0153SFred Isaman sizeof(struct nfs_commit_data), 19010b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 19020b7c0153SFred Isaman NULL); 19030b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 19043dd4765fSJeff Layton goto out_destroy_write_mempool; 19050b7c0153SFred Isaman 190693d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 19074c100210SYanchuan Nian nfs_cdata_cachep); 19081da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 19093dd4765fSJeff Layton goto out_destroy_commit_cache; 19101da177e4SLinus Torvalds 191189a09141SPeter Zijlstra /* 191289a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 191389a09141SPeter Zijlstra * 191489a09141SPeter Zijlstra * 64MB: 8192k 191589a09141SPeter Zijlstra * 128MB: 11585k 191689a09141SPeter Zijlstra * 256MB: 16384k 191789a09141SPeter Zijlstra * 512MB: 23170k 191889a09141SPeter Zijlstra * 1GB: 32768k 191989a09141SPeter Zijlstra * 2GB: 46340k 192089a09141SPeter Zijlstra * 4GB: 65536k 192189a09141SPeter Zijlstra * 8GB: 92681k 192289a09141SPeter Zijlstra * 16GB: 131072k 192389a09141SPeter Zijlstra * 192489a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 192589a09141SPeter Zijlstra * Limit the default to 256M 192689a09141SPeter Zijlstra */ 192789a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 192889a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 192989a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 193089a09141SPeter Zijlstra 19311da177e4SLinus Torvalds return 0; 19323dd4765fSJeff Layton 19333dd4765fSJeff Layton out_destroy_commit_cache: 19343dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 19353dd4765fSJeff Layton out_destroy_write_mempool: 19363dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 19373dd4765fSJeff Layton out_destroy_write_cache: 19383dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 19393dd4765fSJeff Layton return -ENOMEM; 19401da177e4SLinus Torvalds } 19411da177e4SLinus Torvalds 1942266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 19431da177e4SLinus Torvalds { 19441da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 19453dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 19461da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 19471a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 19481da177e4SLinus Torvalds } 19491da177e4SLinus Torvalds 19504a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 1951a4cdda59SAnna Schumaker .rw_mode = FMODE_WRITE, 19524a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 19534a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 1954a4cdda59SAnna Schumaker .rw_release = nfs_writeback_release_common, 19550eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 19560eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 19571ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 19584a0de55cSAnna Schumaker }; 1959