11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/nfs/write.c 31da177e4SLinus Torvalds * 47c85d900STrond Myklebust * Write file data over NFS. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> 71da177e4SLinus Torvalds */ 81da177e4SLinus Torvalds 91da177e4SLinus Torvalds #include <linux/types.h> 101da177e4SLinus Torvalds #include <linux/slab.h> 111da177e4SLinus Torvalds #include <linux/mm.h> 121da177e4SLinus Torvalds #include <linux/pagemap.h> 131da177e4SLinus Torvalds #include <linux/file.h> 141da177e4SLinus Torvalds #include <linux/writeback.h> 1589a09141SPeter Zijlstra #include <linux/swap.h> 16074cc1deSTrond Myklebust #include <linux/migrate.h> 171da177e4SLinus Torvalds 181da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h> 191da177e4SLinus Torvalds #include <linux/nfs_fs.h> 201da177e4SLinus Torvalds #include <linux/nfs_mount.h> 211da177e4SLinus Torvalds #include <linux/nfs_page.h> 223fcfab16SAndrew Morton #include <linux/backing-dev.h> 23afeacc8cSPaul Gortmaker #include <linux/export.h> 243fcfab16SAndrew Morton 251da177e4SLinus Torvalds #include <asm/uaccess.h> 261da177e4SLinus Torvalds 271da177e4SLinus Torvalds #include "delegation.h" 2849a70f27STrond Myklebust #include "internal.h" 2991d5b470SChuck Lever #include "iostat.h" 30def6ed7eSAndy Adamson #include "nfs4_fs.h" 31074cc1deSTrond Myklebust #include "fscache.h" 3294ad1c80SFred Isaman #include "pnfs.h" 331da177e4SLinus Torvalds 34f4ce1299STrond Myklebust #include "nfstrace.h" 35f4ce1299STrond Myklebust 361da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 371da177e4SLinus Torvalds 381da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 391da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 401da177e4SLinus Torvalds 411da177e4SLinus Torvalds /* 421da177e4SLinus Torvalds * Local function declarations 431da177e4SLinus Torvalds */ 44f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 45788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 46061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; 47f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops; 484a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops; 49d4581383SWeston Andros Adamson static void nfs_clear_request_commit(struct nfs_page *req); 5002d1426cSWeston Andros Adamson static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 5102d1426cSWeston Andros Adamson struct inode *inode); 523a3908c8STrond Myklebust static struct nfs_page * 533a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 543a3908c8STrond Myklebust struct page *page); 551da177e4SLinus Torvalds 56e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 573feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 580b7c0153SFred Isaman static struct kmem_cache *nfs_cdata_cachep; 591da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 601da177e4SLinus Torvalds 610b7c0153SFred Isaman struct nfs_commit_data *nfs_commitdata_alloc(void) 621da177e4SLinus Torvalds { 63192e501bSMel Gorman struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); 6440859d7eSChuck Lever 651da177e4SLinus Torvalds if (p) { 661da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 671da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 681da177e4SLinus Torvalds } 691da177e4SLinus Torvalds return p; 701da177e4SLinus Torvalds } 71e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 721da177e4SLinus Torvalds 730b7c0153SFred Isaman void nfs_commit_free(struct nfs_commit_data *p) 741da177e4SLinus Torvalds { 751da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 761da177e4SLinus Torvalds } 77e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 781da177e4SLinus Torvalds 791e7f3a48SWeston Andros Adamson static struct nfs_pgio_header *nfs_writehdr_alloc(void) 803feb2d49STrond Myklebust { 811e7f3a48SWeston Andros Adamson struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); 823feb2d49STrond Myklebust 834a0de55cSAnna Schumaker if (p) 843feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 853feb2d49STrond Myklebust return p; 863feb2d49STrond Myklebust } 873feb2d49STrond Myklebust 881e7f3a48SWeston Andros Adamson static void nfs_writehdr_free(struct nfs_pgio_header *hdr) 896c75dc0dSFred Isaman { 901e7f3a48SWeston Andros Adamson mempool_free(hdr, nfs_wdata_mempool); 913feb2d49STrond Myklebust } 921da177e4SLinus Torvalds 937b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 947b159fc1STrond Myklebust { 957b159fc1STrond Myklebust ctx->error = error; 967b159fc1STrond Myklebust smp_wmb(); 977b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 987b159fc1STrond Myklebust } 997b159fc1STrond Myklebust 10084d3a9a9SWeston Andros Adamson /* 10184d3a9a9SWeston Andros Adamson * nfs_page_find_head_request_locked - find head request associated with @page 10284d3a9a9SWeston Andros Adamson * 10384d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 10484d3a9a9SWeston Andros Adamson * 10584d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 10684d3a9a9SWeston Andros Adamson */ 10729418aa4SMel Gorman static struct nfs_page * 10884d3a9a9SWeston Andros Adamson nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page) 109277459d2STrond Myklebust { 110277459d2STrond Myklebust struct nfs_page *req = NULL; 111277459d2STrond Myklebust 11229418aa4SMel Gorman if (PagePrivate(page)) 113277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 11402d1426cSWeston Andros Adamson else if (unlikely(PageSwapCache(page))) 11502d1426cSWeston Andros Adamson req = nfs_page_search_commits_for_head_request_locked(nfsi, 11602d1426cSWeston Andros Adamson page); 11729418aa4SMel Gorman 11884d3a9a9SWeston Andros Adamson if (req) { 11984d3a9a9SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 12029418aa4SMel Gorman kref_get(&req->wb_kref); 12184d3a9a9SWeston Andros Adamson } 12229418aa4SMel Gorman 123277459d2STrond Myklebust return req; 124277459d2STrond Myklebust } 125277459d2STrond Myklebust 12684d3a9a9SWeston Andros Adamson /* 12784d3a9a9SWeston Andros Adamson * nfs_page_find_head_request - find head request associated with @page 12884d3a9a9SWeston Andros Adamson * 12984d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 13084d3a9a9SWeston Andros Adamson */ 13184d3a9a9SWeston Andros Adamson static struct nfs_page *nfs_page_find_head_request(struct page *page) 132277459d2STrond Myklebust { 133d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 134277459d2STrond Myklebust struct nfs_page *req = NULL; 135277459d2STrond Myklebust 136587142f8STrond Myklebust spin_lock(&inode->i_lock); 13784d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 138587142f8STrond Myklebust spin_unlock(&inode->i_lock); 139277459d2STrond Myklebust return req; 140277459d2STrond Myklebust } 141277459d2STrond Myklebust 1421da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1431da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1441da177e4SLinus Torvalds { 145d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 146a3d01454STrond Myklebust loff_t end, i_size; 147a3d01454STrond Myklebust pgoff_t end_index; 1481da177e4SLinus Torvalds 149a3d01454STrond Myklebust spin_lock(&inode->i_lock); 150a3d01454STrond Myklebust i_size = i_size_read(inode); 151a3d01454STrond Myklebust end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 152d56b4ddfSMel Gorman if (i_size > 0 && page_file_index(page) < end_index) 153a3d01454STrond Myklebust goto out; 154d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 1551da177e4SLinus Torvalds if (i_size >= end) 156a3d01454STrond Myklebust goto out; 1571da177e4SLinus Torvalds i_size_write(inode, end); 158a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 159a3d01454STrond Myklebust out: 160a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 1611da177e4SLinus Torvalds } 1621da177e4SLinus Torvalds 163a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 164a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 165a301b777STrond Myklebust { 166d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 167a301b777STrond Myklebust } 168a301b777STrond Myklebust 169d72ddcbaSWeston Andros Adamson /* 170d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 171d72ddcbaSWeston Andros Adamson * @head - head request of page group 172d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 173d72ddcbaSWeston Andros Adamson * 174d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 175d72ddcbaSWeston Andros Adamson * page offset @page_offset. 176d72ddcbaSWeston Andros Adamson * 177d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 178d72ddcbaSWeston Andros Adamson * match is found. 179d72ddcbaSWeston Andros Adamson * 180d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 181d72ddcbaSWeston Andros Adamson */ 182d72ddcbaSWeston Andros Adamson static struct nfs_page * 183d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 184d72ddcbaSWeston Andros Adamson { 185d72ddcbaSWeston Andros Adamson struct nfs_page *req; 186d72ddcbaSWeston Andros Adamson 187d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(head != head->wb_head); 188d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags)); 189d72ddcbaSWeston Andros Adamson 190d72ddcbaSWeston Andros Adamson req = head; 191d72ddcbaSWeston Andros Adamson do { 192d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 193d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 194d72ddcbaSWeston Andros Adamson return req; 195d72ddcbaSWeston Andros Adamson 196d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 197d72ddcbaSWeston Andros Adamson } while (req != head); 198d72ddcbaSWeston Andros Adamson 199d72ddcbaSWeston Andros Adamson return NULL; 200d72ddcbaSWeston Andros Adamson } 201d72ddcbaSWeston Andros Adamson 202d72ddcbaSWeston Andros Adamson /* 203d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 204d72ddcbaSWeston Andros Adamson * @head - head request of page group 205d72ddcbaSWeston Andros Adamson * 206d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 207d72ddcbaSWeston Andros Adamson * returns false otherwise 208d72ddcbaSWeston Andros Adamson */ 209d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 210d72ddcbaSWeston Andros Adamson { 211d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 212d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 213d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 214d72ddcbaSWeston Andros Adamson 215fd2f3a06SWeston Andros Adamson nfs_page_group_lock(req, false); 216d72ddcbaSWeston Andros Adamson 217d72ddcbaSWeston Andros Adamson do { 218d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 219d72ddcbaSWeston Andros Adamson if (tmp) { 220d72ddcbaSWeston Andros Adamson /* no way this should happen */ 221d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(tmp->wb_pgbase != pos); 222d72ddcbaSWeston Andros Adamson pos += tmp->wb_bytes - (pos - tmp->wb_pgbase); 223d72ddcbaSWeston Andros Adamson } 224d72ddcbaSWeston Andros Adamson } while (tmp && pos < len); 225d72ddcbaSWeston Andros Adamson 226d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 227d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(pos > len); 228d72ddcbaSWeston Andros Adamson return pos == len; 229d72ddcbaSWeston Andros Adamson } 230d72ddcbaSWeston Andros Adamson 2311da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 2321da177e4SLinus Torvalds * covers the full page. 2331da177e4SLinus Torvalds */ 234d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 2351da177e4SLinus Torvalds { 236d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 2371da177e4SLinus Torvalds return; 238d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 2391da177e4SLinus Torvalds return; 240d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 2411da177e4SLinus Torvalds } 2421da177e4SLinus Torvalds 2431da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 2441da177e4SLinus Torvalds { 245e87b4c7aSNeilBrown int ret = 0; 2461da177e4SLinus Torvalds if (wbc->for_reclaim) 247c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 248e87b4c7aSNeilBrown if (wbc->sync_mode == WB_SYNC_ALL) 249e87b4c7aSNeilBrown ret = FLUSH_COND_STABLE; 250b17621feSWu Fengguang if (wbc->for_kupdate || wbc->for_background) 251e87b4c7aSNeilBrown ret |= FLUSH_LOWPRI; 252e87b4c7aSNeilBrown return ret; 2531da177e4SLinus Torvalds } 2541da177e4SLinus Torvalds 2551da177e4SLinus Torvalds /* 25689a09141SPeter Zijlstra * NFS congestion control 25789a09141SPeter Zijlstra */ 25889a09141SPeter Zijlstra 25989a09141SPeter Zijlstra int nfs_congestion_kb; 26089a09141SPeter Zijlstra 26189a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 26289a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 26389a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 26489a09141SPeter Zijlstra 265deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 26689a09141SPeter Zijlstra { 267deed85e7STrond Myklebust struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); 2685a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2695a6d41b3STrond Myklebust 270deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 27189a09141SPeter Zijlstra 272277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 2738aa7e847SJens Axboe NFS_CONGESTION_ON_THRESH) { 2748aa7e847SJens Axboe set_bdi_congested(&nfss->backing_dev_info, 2758aa7e847SJens Axboe BLK_RW_ASYNC); 2768aa7e847SJens Axboe } 27789a09141SPeter Zijlstra } 27889a09141SPeter Zijlstra 27920633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 28089a09141SPeter Zijlstra { 28120633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 28289a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 28389a09141SPeter Zijlstra 28420633f04SWeston Andros Adamson if (!nfs_page_group_sync_on_bit(req, PG_WB_END)) 28520633f04SWeston Andros Adamson return; 28620633f04SWeston Andros Adamson 28720633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 288c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 2898aa7e847SJens Axboe clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 29089a09141SPeter Zijlstra } 29189a09141SPeter Zijlstra 292d4581383SWeston Andros Adamson 293d4581383SWeston Andros Adamson /* nfs_page_group_clear_bits 294d4581383SWeston Andros Adamson * @req - an nfs request 295d4581383SWeston Andros Adamson * clears all page group related bits from @req 296d4581383SWeston Andros Adamson */ 297d4581383SWeston Andros Adamson static void 298d4581383SWeston Andros Adamson nfs_page_group_clear_bits(struct nfs_page *req) 299e261f51fSTrond Myklebust { 300d4581383SWeston Andros Adamson clear_bit(PG_TEARDOWN, &req->wb_flags); 301d4581383SWeston Andros Adamson clear_bit(PG_UNLOCKPAGE, &req->wb_flags); 302d4581383SWeston Andros Adamson clear_bit(PG_UPTODATE, &req->wb_flags); 303d4581383SWeston Andros Adamson clear_bit(PG_WB_END, &req->wb_flags); 304d4581383SWeston Andros Adamson clear_bit(PG_REMOVE, &req->wb_flags); 305d4581383SWeston Andros Adamson } 306d4581383SWeston Andros Adamson 307d4581383SWeston Andros Adamson 308d4581383SWeston Andros Adamson /* 309d4581383SWeston Andros Adamson * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req 310d4581383SWeston Andros Adamson * 311d4581383SWeston Andros Adamson * this is a helper function for nfs_lock_and_join_requests 312d4581383SWeston Andros Adamson * 313d4581383SWeston Andros Adamson * @inode - inode associated with request page group, must be holding inode lock 314d4581383SWeston Andros Adamson * @head - head request of page group, must be holding head lock 315d4581383SWeston Andros Adamson * @req - request that couldn't lock and needs to wait on the req bit lock 316d4581383SWeston Andros Adamson * @nonblock - if true, don't actually wait 317d4581383SWeston Andros Adamson * 318d4581383SWeston Andros Adamson * NOTE: this must be called holding page_group bit lock and inode spin lock 319d4581383SWeston Andros Adamson * and BOTH will be released before returning. 320d4581383SWeston Andros Adamson * 321d4581383SWeston Andros Adamson * returns 0 on success, < 0 on error. 322d4581383SWeston Andros Adamson */ 323d4581383SWeston Andros Adamson static int 324d4581383SWeston Andros Adamson nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head, 325d4581383SWeston Andros Adamson struct nfs_page *req, bool nonblock) 326d4581383SWeston Andros Adamson __releases(&inode->i_lock) 327d4581383SWeston Andros Adamson { 328d4581383SWeston Andros Adamson struct nfs_page *tmp; 329e261f51fSTrond Myklebust int ret; 330e261f51fSTrond Myklebust 331d4581383SWeston Andros Adamson /* relinquish all the locks successfully grabbed this run */ 332d4581383SWeston Andros Adamson for (tmp = head ; tmp != req; tmp = tmp->wb_this_page) 333d4581383SWeston Andros Adamson nfs_unlock_request(tmp); 334d4581383SWeston Andros Adamson 335d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags)); 336d4581383SWeston Andros Adamson 337d4581383SWeston Andros Adamson /* grab a ref on the request that will be waited on */ 338d4581383SWeston Andros Adamson kref_get(&req->wb_kref); 339d4581383SWeston Andros Adamson 340d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 341587142f8STrond Myklebust spin_unlock(&inode->i_lock); 342d4581383SWeston Andros Adamson 343d4581383SWeston Andros Adamson /* release ref from nfs_page_find_head_request_locked */ 344d4581383SWeston Andros Adamson nfs_release_request(head); 345d4581383SWeston Andros Adamson 346cfb506e1STrond Myklebust if (!nonblock) 347e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 348cfb506e1STrond Myklebust else 349cfb506e1STrond Myklebust ret = -EAGAIN; 350e261f51fSTrond Myklebust nfs_release_request(req); 351d4581383SWeston Andros Adamson 352d4581383SWeston Andros Adamson return ret; 353e261f51fSTrond Myklebust } 354d4581383SWeston Andros Adamson 355d4581383SWeston Andros Adamson /* 356d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 357d4581383SWeston Andros Adamson * 358d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 359d4581383SWeston Andros Adamson * @old_head - the old head of the list 360d4581383SWeston Andros Adamson * 361d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 362d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 363d4581383SWeston Andros Adamson * with a reference held by some other context. 364d4581383SWeston Andros Adamson */ 365d4581383SWeston Andros Adamson static void 366d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 367d4581383SWeston Andros Adamson struct nfs_page *old_head) 368d4581383SWeston Andros Adamson { 369d4581383SWeston Andros Adamson while (destroy_list) { 370d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 371d4581383SWeston Andros Adamson 372d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 373d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 374d4581383SWeston Andros Adamson 375d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 376d4581383SWeston Andros Adamson 377d4581383SWeston Andros Adamson /* make sure old group is not used */ 378d4581383SWeston Andros Adamson subreq->wb_head = subreq; 379d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 380d4581383SWeston Andros Adamson 381d4581383SWeston Andros Adamson /* subreq is now totally disconnected from page group or any 382d4581383SWeston Andros Adamson * write / commit lists. last chance to wake any waiters */ 383d4581383SWeston Andros Adamson nfs_unlock_request(subreq); 384d4581383SWeston Andros Adamson 385d4581383SWeston Andros Adamson if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) { 386d4581383SWeston Andros Adamson /* release ref on old head request */ 387d4581383SWeston Andros Adamson nfs_release_request(old_head); 388d4581383SWeston Andros Adamson 389d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 390d4581383SWeston Andros Adamson 391d4581383SWeston Andros Adamson /* release the PG_INODE_REF reference */ 392d4581383SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) 393d4581383SWeston Andros Adamson nfs_release_request(subreq); 394d4581383SWeston Andros Adamson else 395d4581383SWeston Andros Adamson WARN_ON_ONCE(1); 396d4581383SWeston Andros Adamson } else { 397d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags)); 398d4581383SWeston Andros Adamson /* zombie requests have already released the last 399d4581383SWeston Andros Adamson * reference and were waiting on the rest of the 400d4581383SWeston Andros Adamson * group to complete. Since it's no longer part of a 401d4581383SWeston Andros Adamson * group, simply free the request */ 402d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 403d4581383SWeston Andros Adamson nfs_free_request(subreq); 404d4581383SWeston Andros Adamson } 405d4581383SWeston Andros Adamson } 406d4581383SWeston Andros Adamson } 407d4581383SWeston Andros Adamson 408d4581383SWeston Andros Adamson /* 409d4581383SWeston Andros Adamson * nfs_lock_and_join_requests - join all subreqs to the head req and return 410d4581383SWeston Andros Adamson * a locked reference, cancelling any pending 411d4581383SWeston Andros Adamson * operations for this page. 412d4581383SWeston Andros Adamson * 413d4581383SWeston Andros Adamson * @page - the page used to lookup the "page group" of nfs_page structures 414d4581383SWeston Andros Adamson * @nonblock - if true, don't block waiting for request locks 415d4581383SWeston Andros Adamson * 416d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 417d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 418d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 419d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 420d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 421d4581383SWeston Andros Adamson * 422d4581383SWeston Andros Adamson * Returns a locked, referenced pointer to the head request - which after 423d4581383SWeston Andros Adamson * this call is guaranteed to be the only request associated with the page. 424d4581383SWeston Andros Adamson * Returns NULL if no requests are found for @page, or a ERR_PTR if an 425d4581383SWeston Andros Adamson * error was encountered. 426d4581383SWeston Andros Adamson */ 427d4581383SWeston Andros Adamson static struct nfs_page * 428d4581383SWeston Andros Adamson nfs_lock_and_join_requests(struct page *page, bool nonblock) 429d4581383SWeston Andros Adamson { 430d4581383SWeston Andros Adamson struct inode *inode = page_file_mapping(page)->host; 431d4581383SWeston Andros Adamson struct nfs_page *head, *subreq; 432d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 433d4581383SWeston Andros Adamson unsigned int total_bytes; 434d4581383SWeston Andros Adamson int ret; 435d4581383SWeston Andros Adamson 436d4581383SWeston Andros Adamson try_again: 437d4581383SWeston Andros Adamson total_bytes = 0; 438d4581383SWeston Andros Adamson 439d4581383SWeston Andros Adamson WARN_ON_ONCE(destroy_list); 440d4581383SWeston Andros Adamson 441d4581383SWeston Andros Adamson spin_lock(&inode->i_lock); 442d4581383SWeston Andros Adamson 443d4581383SWeston Andros Adamson /* 444d4581383SWeston Andros Adamson * A reference is taken only on the head request which acts as a 445d4581383SWeston Andros Adamson * reference to the whole page group - the group will not be destroyed 446d4581383SWeston Andros Adamson * until the head reference is released. 447d4581383SWeston Andros Adamson */ 448d4581383SWeston Andros Adamson head = nfs_page_find_head_request_locked(NFS_I(inode), page); 449d4581383SWeston Andros Adamson 450d4581383SWeston Andros Adamson if (!head) { 451587142f8STrond Myklebust spin_unlock(&inode->i_lock); 452d4581383SWeston Andros Adamson return NULL; 453d4581383SWeston Andros Adamson } 454d4581383SWeston Andros Adamson 4557c3af975SWeston Andros Adamson /* holding inode lock, so always make a non-blocking call to try the 4567c3af975SWeston Andros Adamson * page group lock */ 457fd2f3a06SWeston Andros Adamson ret = nfs_page_group_lock(head, true); 45894970014SWeston Andros Adamson if (ret < 0) { 45994970014SWeston Andros Adamson spin_unlock(&inode->i_lock); 4607c3af975SWeston Andros Adamson 4617c3af975SWeston Andros Adamson if (!nonblock && ret == -EAGAIN) { 4627c3af975SWeston Andros Adamson nfs_page_group_lock_wait(head); 4637c3af975SWeston Andros Adamson nfs_release_request(head); 4647c3af975SWeston Andros Adamson goto try_again; 4657c3af975SWeston Andros Adamson } 4667c3af975SWeston Andros Adamson 46794970014SWeston Andros Adamson nfs_release_request(head); 468e7029206SWeston Andros Adamson return ERR_PTR(ret); 46994970014SWeston Andros Adamson } 4707c3af975SWeston Andros Adamson 4717c3af975SWeston Andros Adamson /* lock each request in the page group */ 472d4581383SWeston Andros Adamson subreq = head; 473d4581383SWeston Andros Adamson do { 474d4581383SWeston Andros Adamson /* 475d4581383SWeston Andros Adamson * Subrequests are always contiguous, non overlapping 476309a1d65SWeston Andros Adamson * and in order - but may be repeated (mirrored writes). 477d4581383SWeston Andros Adamson */ 478309a1d65SWeston Andros Adamson if (subreq->wb_offset == (head->wb_offset + total_bytes)) { 479d4581383SWeston Andros Adamson /* keep track of how many bytes this group covers */ 480d4581383SWeston Andros Adamson total_bytes += subreq->wb_bytes; 481309a1d65SWeston Andros Adamson } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || 482309a1d65SWeston Andros Adamson ((subreq->wb_offset + subreq->wb_bytes) > 483309a1d65SWeston Andros Adamson (head->wb_offset + total_bytes)))) { 484309a1d65SWeston Andros Adamson nfs_page_group_unlock(head); 485309a1d65SWeston Andros Adamson spin_unlock(&inode->i_lock); 486309a1d65SWeston Andros Adamson return ERR_PTR(-EIO); 487309a1d65SWeston Andros Adamson } 488d4581383SWeston Andros Adamson 489d4581383SWeston Andros Adamson if (!nfs_lock_request(subreq)) { 490d4581383SWeston Andros Adamson /* releases page group bit lock and 491d4581383SWeston Andros Adamson * inode spin lock and all references */ 492d4581383SWeston Andros Adamson ret = nfs_unroll_locks_and_wait(inode, head, 493d4581383SWeston Andros Adamson subreq, nonblock); 494d4581383SWeston Andros Adamson 495d4581383SWeston Andros Adamson if (ret == 0) 496d4581383SWeston Andros Adamson goto try_again; 497d4581383SWeston Andros Adamson 498d4581383SWeston Andros Adamson return ERR_PTR(ret); 499d4581383SWeston Andros Adamson } 500d4581383SWeston Andros Adamson 501d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 502d4581383SWeston Andros Adamson } while (subreq != head); 503d4581383SWeston Andros Adamson 504d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 505d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 506d4581383SWeston Andros Adamson subreq = head; 507d4581383SWeston Andros Adamson do { 508411a99adSWeston Andros Adamson nfs_clear_request_commit(subreq); 509d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 510d4581383SWeston Andros Adamson } while (subreq != head); 511d4581383SWeston Andros Adamson 512d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 513d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 514d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 515d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 516d4581383SWeston Andros Adamson head->wb_this_page = head; 517d4581383SWeston Andros Adamson 518d4581383SWeston Andros Adamson /* change head request to cover whole range that 519d4581383SWeston Andros Adamson * the former page group covered */ 520d4581383SWeston Andros Adamson head->wb_bytes = total_bytes; 521d4581383SWeston Andros Adamson } 522d4581383SWeston Andros Adamson 523d4581383SWeston Andros Adamson /* 524d4581383SWeston Andros Adamson * prepare head request to be added to new pgio descriptor 525d4581383SWeston Andros Adamson */ 526d4581383SWeston Andros Adamson nfs_page_group_clear_bits(head); 527d4581383SWeston Andros Adamson 528d4581383SWeston Andros Adamson /* 529d4581383SWeston Andros Adamson * some part of the group was still on the inode list - otherwise 530d4581383SWeston Andros Adamson * the group wouldn't be involved in async write. 531d4581383SWeston Andros Adamson * grab a reference for the head request, iff it needs one. 532d4581383SWeston Andros Adamson */ 533d4581383SWeston Andros Adamson if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags)) 534d4581383SWeston Andros Adamson kref_get(&head->wb_kref); 535d4581383SWeston Andros Adamson 536d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 537d4581383SWeston Andros Adamson 538411a99adSWeston Andros Adamson /* drop lock to clean uprequests on destroy list */ 539d4581383SWeston Andros Adamson spin_unlock(&inode->i_lock); 540d4581383SWeston Andros Adamson 541d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(destroy_list, head); 542d4581383SWeston Andros Adamson 543d4581383SWeston Andros Adamson /* still holds ref on head from nfs_page_find_head_request_locked 544d4581383SWeston Andros Adamson * and still has lock on head from lock loop */ 545d4581383SWeston Andros Adamson return head; 546612c9384STrond Myklebust } 547074cc1deSTrond Myklebust 5480bcbf039SPeng Tao static void nfs_write_error_remove_page(struct nfs_page *req) 5490bcbf039SPeng Tao { 5500bcbf039SPeng Tao nfs_unlock_request(req); 5510bcbf039SPeng Tao nfs_end_page_writeback(req); 5520bcbf039SPeng Tao nfs_release_request(req); 5530bcbf039SPeng Tao generic_error_remove_page(page_file_mapping(req->wb_page), 5540bcbf039SPeng Tao req->wb_page); 5550bcbf039SPeng Tao } 5560bcbf039SPeng Tao 557074cc1deSTrond Myklebust /* 558074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 559074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 560074cc1deSTrond Myklebust */ 561074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 562*d6c843b9SPeng Tao struct page *page, bool nonblock, 563*d6c843b9SPeng Tao bool launder) 564074cc1deSTrond Myklebust { 565074cc1deSTrond Myklebust struct nfs_page *req; 566074cc1deSTrond Myklebust int ret = 0; 567074cc1deSTrond Myklebust 568d4581383SWeston Andros Adamson req = nfs_lock_and_join_requests(page, nonblock); 569074cc1deSTrond Myklebust if (!req) 570074cc1deSTrond Myklebust goto out; 571074cc1deSTrond Myklebust ret = PTR_ERR(req); 572074cc1deSTrond Myklebust if (IS_ERR(req)) 573074cc1deSTrond Myklebust goto out; 574074cc1deSTrond Myklebust 575deed85e7STrond Myklebust nfs_set_page_writeback(page); 576deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 577074cc1deSTrond Myklebust 578deed85e7STrond Myklebust ret = 0; 579f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 580074cc1deSTrond Myklebust ret = pgio->pg_error; 5810bcbf039SPeng Tao /* 582*d6c843b9SPeng Tao * Remove the problematic req upon fatal errors 583*d6c843b9SPeng Tao * in launder case, while other dirty pages can 584*d6c843b9SPeng Tao * still be around until they get flushed. 5850bcbf039SPeng Tao */ 5860bcbf039SPeng Tao if (nfs_error_is_fatal(ret)) { 5870bcbf039SPeng Tao nfs_context_set_write_error(req->wb_context, ret); 588*d6c843b9SPeng Tao if (launder) { 5890bcbf039SPeng Tao nfs_write_error_remove_page(req); 590*d6c843b9SPeng Tao goto out; 591*d6c843b9SPeng Tao } 592*d6c843b9SPeng Tao } 5930bcbf039SPeng Tao nfs_redirty_request(req); 5940bcbf039SPeng Tao ret = -EAGAIN; 59540f90271STrond Myklebust } else 59640f90271STrond Myklebust nfs_add_stats(page_file_mapping(page)->host, 59740f90271STrond Myklebust NFSIOS_WRITEPAGES, 1); 598074cc1deSTrond Myklebust out: 599074cc1deSTrond Myklebust return ret; 600e261f51fSTrond Myklebust } 601e261f51fSTrond Myklebust 602*d6c843b9SPeng Tao static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, 603*d6c843b9SPeng Tao struct nfs_pageio_descriptor *pgio, bool launder) 604f758c885STrond Myklebust { 605cfb506e1STrond Myklebust int ret; 606f758c885STrond Myklebust 607d56b4ddfSMel Gorman nfs_pageio_cond_complete(pgio, page_file_index(page)); 608*d6c843b9SPeng Tao ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE, 609*d6c843b9SPeng Tao launder); 610cfb506e1STrond Myklebust if (ret == -EAGAIN) { 611cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 612cfb506e1STrond Myklebust ret = 0; 613cfb506e1STrond Myklebust } 614cfb506e1STrond Myklebust return ret; 615f758c885STrond Myklebust } 616f758c885STrond Myklebust 617e261f51fSTrond Myklebust /* 6181da177e4SLinus Torvalds * Write an mmapped page to the server. 6191da177e4SLinus Torvalds */ 620*d6c843b9SPeng Tao static int nfs_writepage_locked(struct page *page, 621*d6c843b9SPeng Tao struct writeback_control *wbc, 622*d6c843b9SPeng Tao bool launder) 6231da177e4SLinus Torvalds { 624f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 62540f90271STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 626e261f51fSTrond Myklebust int err; 6271da177e4SLinus Torvalds 62840f90271STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 62940f90271STrond Myklebust nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), 630a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 631*d6c843b9SPeng Tao err = nfs_do_writepage(page, wbc, &pgio, launder); 632f758c885STrond Myklebust nfs_pageio_complete(&pgio); 633f758c885STrond Myklebust if (err < 0) 6344d770ccfSTrond Myklebust return err; 635f758c885STrond Myklebust if (pgio.pg_error < 0) 636f758c885STrond Myklebust return pgio.pg_error; 637f758c885STrond Myklebust return 0; 6384d770ccfSTrond Myklebust } 6394d770ccfSTrond Myklebust 6404d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 6414d770ccfSTrond Myklebust { 642f758c885STrond Myklebust int ret; 6434d770ccfSTrond Myklebust 644*d6c843b9SPeng Tao ret = nfs_writepage_locked(page, wbc, false); 6451da177e4SLinus Torvalds unlock_page(page); 646f758c885STrond Myklebust return ret; 647f758c885STrond Myklebust } 648f758c885STrond Myklebust 649f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 650f758c885STrond Myklebust { 651f758c885STrond Myklebust int ret; 652f758c885STrond Myklebust 653*d6c843b9SPeng Tao ret = nfs_do_writepage(page, wbc, data, false); 654f758c885STrond Myklebust unlock_page(page); 655f758c885STrond Myklebust return ret; 6561da177e4SLinus Torvalds } 6571da177e4SLinus Torvalds 6581da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 6591da177e4SLinus Torvalds { 6601da177e4SLinus Torvalds struct inode *inode = mapping->host; 66172cb77f4STrond Myklebust unsigned long *bitlock = &NFS_I(inode)->flags; 662c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 6631da177e4SLinus Torvalds int err; 6641da177e4SLinus Torvalds 66572cb77f4STrond Myklebust /* Stop dirtying of new pages while we sync */ 66674316201SNeilBrown err = wait_on_bit_lock_action(bitlock, NFS_INO_FLUSHING, 66772cb77f4STrond Myklebust nfs_wait_bit_killable, TASK_KILLABLE); 66872cb77f4STrond Myklebust if (err) 66972cb77f4STrond Myklebust goto out_err; 67072cb77f4STrond Myklebust 67191d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 67291d5b470SChuck Lever 673a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 674a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 675f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 676c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 67772cb77f4STrond Myklebust 67872cb77f4STrond Myklebust clear_bit_unlock(NFS_INO_FLUSHING, bitlock); 6794e857c58SPeter Zijlstra smp_mb__after_atomic(); 68072cb77f4STrond Myklebust wake_up_bit(bitlock, NFS_INO_FLUSHING); 68172cb77f4STrond Myklebust 682f758c885STrond Myklebust if (err < 0) 68372cb77f4STrond Myklebust goto out_err; 68472cb77f4STrond Myklebust err = pgio.pg_error; 68572cb77f4STrond Myklebust if (err < 0) 68672cb77f4STrond Myklebust goto out_err; 687c63c7b05STrond Myklebust return 0; 68872cb77f4STrond Myklebust out_err: 68972cb77f4STrond Myklebust return err; 6901da177e4SLinus Torvalds } 6911da177e4SLinus Torvalds 6921da177e4SLinus Torvalds /* 6931da177e4SLinus Torvalds * Insert a write request into an inode 6941da177e4SLinus Torvalds */ 695d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 6961da177e4SLinus Torvalds { 6971da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 698e7d39069STrond Myklebust 6992bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 7002bfc6e56SWeston Andros Adamson 701e7d39069STrond Myklebust /* Lock the request! */ 7027ad84aa9STrond Myklebust nfs_lock_request(req); 703e7d39069STrond Myklebust 704e7d39069STrond Myklebust spin_lock(&inode->i_lock); 705cb1410c7SWeston Andros Adamson if (!nfsi->nrequests && 706cb1410c7SWeston Andros Adamson NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 707a9a4a87aSTrond Myklebust inode->i_version++; 70829418aa4SMel Gorman /* 70929418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 71029418aa4SMel Gorman * with invalidate/truncate. 71129418aa4SMel Gorman */ 71229418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 7132df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 714deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 715277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 71629418aa4SMel Gorman } 717cb1410c7SWeston Andros Adamson nfsi->nrequests++; 71817089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 719cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 720cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 721cb1410c7SWeston Andros Adamson * adding subrequests. */ 72217089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 723c03b4024STrond Myklebust kref_get(&req->wb_kref); 724e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 7251da177e4SLinus Torvalds } 7261da177e4SLinus Torvalds 7271da177e4SLinus Torvalds /* 72889a09141SPeter Zijlstra * Remove a write request from an inode 7291da177e4SLinus Torvalds */ 7301da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7311da177e4SLinus Torvalds { 7322b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 7331da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 73420633f04SWeston Andros Adamson struct nfs_page *head; 73520633f04SWeston Andros Adamson 73620633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 73720633f04SWeston Andros Adamson head = req->wb_head; 7381da177e4SLinus Torvalds 739587142f8STrond Myklebust spin_lock(&inode->i_lock); 74020633f04SWeston Andros Adamson if (likely(!PageSwapCache(head->wb_page))) { 74120633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 74220633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 74395905446SNeilBrown smp_mb__after_atomic(); 74495905446SNeilBrown wake_up_page(head->wb_page, PG_private); 74520633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 74629418aa4SMel Gorman } 747cb1410c7SWeston Andros Adamson nfsi->nrequests--; 748cb1410c7SWeston Andros Adamson spin_unlock(&inode->i_lock); 749cb1410c7SWeston Andros Adamson } else { 750cb1410c7SWeston Andros Adamson spin_lock(&inode->i_lock); 751cb1410c7SWeston Andros Adamson nfsi->nrequests--; 752587142f8STrond Myklebust spin_unlock(&inode->i_lock); 75320633f04SWeston Andros Adamson } 75417089a29SWeston Andros Adamson 75517089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 7561da177e4SLinus Torvalds nfs_release_request(req); 7571da177e4SLinus Torvalds } 7581da177e4SLinus Torvalds 75961822ab5STrond Myklebust static void 7606d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 76161822ab5STrond Myklebust { 76261822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 76361822ab5STrond Myklebust } 76461822ab5STrond Myklebust 7653a3908c8STrond Myklebust /* 7663a3908c8STrond Myklebust * nfs_page_search_commits_for_head_request_locked 7673a3908c8STrond Myklebust * 7683a3908c8STrond Myklebust * Search through commit lists on @inode for the head request for @page. 7693a3908c8STrond Myklebust * Must be called while holding the inode (which is cinfo) lock. 7703a3908c8STrond Myklebust * 7713a3908c8STrond Myklebust * Returns the head request if found, or NULL if not found. 7723a3908c8STrond Myklebust */ 7733a3908c8STrond Myklebust static struct nfs_page * 7743a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 7753a3908c8STrond Myklebust struct page *page) 7763a3908c8STrond Myklebust { 7773a3908c8STrond Myklebust struct nfs_page *freq, *t; 7783a3908c8STrond Myklebust struct nfs_commit_info cinfo; 7793a3908c8STrond Myklebust struct inode *inode = &nfsi->vfs_inode; 7803a3908c8STrond Myklebust 7813a3908c8STrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 7823a3908c8STrond Myklebust 7833a3908c8STrond Myklebust /* search through pnfs commit lists */ 7843a3908c8STrond Myklebust freq = pnfs_search_commit_reqs(inode, &cinfo, page); 7853a3908c8STrond Myklebust if (freq) 7863a3908c8STrond Myklebust return freq->wb_head; 7873a3908c8STrond Myklebust 7883a3908c8STrond Myklebust /* Linearly search the commit list for the correct request */ 7893a3908c8STrond Myklebust list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 7903a3908c8STrond Myklebust if (freq->wb_page == page) 7913a3908c8STrond Myklebust return freq->wb_head; 7923a3908c8STrond Myklebust } 7933a3908c8STrond Myklebust 7943a3908c8STrond Myklebust return NULL; 7953a3908c8STrond Myklebust } 7963a3908c8STrond Myklebust 7978dd37758STrond Myklebust /** 79886d80f97STrond Myklebust * nfs_request_add_commit_list_locked - add request to a commit list 79986d80f97STrond Myklebust * @req: pointer to a struct nfs_page 80086d80f97STrond Myklebust * @dst: commit list head 80186d80f97STrond Myklebust * @cinfo: holds list lock and accounting info 80286d80f97STrond Myklebust * 80386d80f97STrond Myklebust * This sets the PG_CLEAN bit, updates the cinfo count of 80486d80f97STrond Myklebust * number of outstanding requests requiring a commit as well as 80586d80f97STrond Myklebust * the MM page stats. 80686d80f97STrond Myklebust * 80786d80f97STrond Myklebust * The caller must hold the cinfo->lock, and the nfs_page lock. 80886d80f97STrond Myklebust */ 80986d80f97STrond Myklebust void 81086d80f97STrond Myklebust nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, 81186d80f97STrond Myklebust struct nfs_commit_info *cinfo) 81286d80f97STrond Myklebust { 81386d80f97STrond Myklebust set_bit(PG_CLEAN, &req->wb_flags); 81486d80f97STrond Myklebust nfs_list_add_request(req, dst); 81586d80f97STrond Myklebust cinfo->mds->ncommit++; 81686d80f97STrond Myklebust } 81786d80f97STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); 81886d80f97STrond Myklebust 81986d80f97STrond Myklebust /** 8208dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 8218dd37758STrond Myklebust * @req: pointer to a struct nfs_page 822ea2cf228SFred Isaman * @dst: commit list head 823ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8248dd37758STrond Myklebust * 825ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 8268dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8278dd37758STrond Myklebust * the MM page stats. 8288dd37758STrond Myklebust * 829ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8308dd37758STrond Myklebust * holding the nfs_page lock. 8318dd37758STrond Myklebust */ 8328dd37758STrond Myklebust void 833ea2cf228SFred Isaman nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, 834ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 8358dd37758STrond Myklebust { 836ea2cf228SFred Isaman spin_lock(cinfo->lock); 83786d80f97STrond Myklebust nfs_request_add_commit_list_locked(req, dst, cinfo); 838ea2cf228SFred Isaman spin_unlock(cinfo->lock); 83986d80f97STrond Myklebust nfs_mark_page_unstable(req->wb_page, cinfo); 8408dd37758STrond Myklebust } 8418dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 8428dd37758STrond Myklebust 8438dd37758STrond Myklebust /** 8448dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 8458dd37758STrond Myklebust * @req: pointer to a nfs_page 846ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8478dd37758STrond Myklebust * 848ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 8498dd37758STrond Myklebust * number of outstanding requests requiring a commit 8508dd37758STrond Myklebust * It does not update the MM page stats. 8518dd37758STrond Myklebust * 852ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 8538dd37758STrond Myklebust */ 8548dd37758STrond Myklebust void 855ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 856ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 8578dd37758STrond Myklebust { 8588dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 8598dd37758STrond Myklebust return; 8608dd37758STrond Myklebust nfs_list_remove_request(req); 861ea2cf228SFred Isaman cinfo->mds->ncommit--; 8628dd37758STrond Myklebust } 8638dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 8648dd37758STrond Myklebust 865ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 866ea2cf228SFred Isaman struct inode *inode) 867ea2cf228SFred Isaman { 868ea2cf228SFred Isaman cinfo->lock = &inode->i_lock; 869ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 870ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 871b359f9d0SFred Isaman cinfo->dreq = NULL; 872f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 873ea2cf228SFred Isaman } 874ea2cf228SFred Isaman 875ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 876ea2cf228SFred Isaman struct inode *inode, 877ea2cf228SFred Isaman struct nfs_direct_req *dreq) 878ea2cf228SFred Isaman { 8791763da12SFred Isaman if (dreq) 8801763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 8811763da12SFred Isaman else 882ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 883ea2cf228SFred Isaman } 884ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 8858dd37758STrond Myklebust 8861da177e4SLinus Torvalds /* 8871da177e4SLinus Torvalds * Add a request to the inode's commit list. 8881da177e4SLinus Torvalds */ 8891763da12SFred Isaman void 890ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 891b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 8921da177e4SLinus Torvalds { 893b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 8948dd37758STrond Myklebust return; 895ea2cf228SFred Isaman nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); 8961da177e4SLinus Torvalds } 8978e821cadSTrond Myklebust 898d6d6dc7cSFred Isaman static void 899d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 900e468bae9STrond Myklebust { 901e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 90293f78d88STejun Heo dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, 90393f78d88STejun Heo WB_RECLAIMABLE); 904e468bae9STrond Myklebust } 905d6d6dc7cSFred Isaman 906411a99adSWeston Andros Adamson /* Called holding inode (/cinfo) lock */ 9078dd37758STrond Myklebust static void 908d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 909d6d6dc7cSFred Isaman { 9108dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 9112b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 912ea2cf228SFred Isaman struct nfs_commit_info cinfo; 913d6d6dc7cSFred Isaman 914ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 915ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 916ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 917d6d6dc7cSFred Isaman } 9188dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 9198dd37758STrond Myklebust } 920e468bae9STrond Myklebust } 921e468bae9STrond Myklebust 922d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 9238e821cadSTrond Myklebust { 924c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 925d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 926c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 9278e821cadSTrond Myklebust } 9288e821cadSTrond Myklebust 929061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 9306c75dc0dSFred Isaman { 931ea2cf228SFred Isaman struct nfs_commit_info cinfo; 9326c75dc0dSFred Isaman unsigned long bytes = 0; 9336c75dc0dSFred Isaman 9346c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 9356c75dc0dSFred Isaman goto out; 936ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 9376c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 9386c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 9396c75dc0dSFred Isaman 9406c75dc0dSFred Isaman bytes += req->wb_bytes; 9416c75dc0dSFred Isaman nfs_list_remove_request(req); 9426c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 9436c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 944d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 9456c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 9466c75dc0dSFred Isaman goto remove_req; 9476c75dc0dSFred Isaman } 948c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 949f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 950b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 951a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 9526c75dc0dSFred Isaman goto next; 9536c75dc0dSFred Isaman } 9546c75dc0dSFred Isaman remove_req: 9556c75dc0dSFred Isaman nfs_inode_remove_request(req); 9566c75dc0dSFred Isaman next: 9571d1afcbcSTrond Myklebust nfs_unlock_request(req); 95820633f04SWeston Andros Adamson nfs_end_page_writeback(req); 9593aff4ebbSTrond Myklebust nfs_release_request(req); 9606c75dc0dSFred Isaman } 9616c75dc0dSFred Isaman out: 9626c75dc0dSFred Isaman hdr->release(hdr); 9636c75dc0dSFred Isaman } 9646c75dc0dSFred Isaman 965ce59515cSAnna Schumaker unsigned long 966ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 967fb8a1f11STrond Myklebust { 968ea2cf228SFred Isaman return cinfo->mds->ncommit; 969fb8a1f11STrond Myklebust } 970fb8a1f11STrond Myklebust 971ea2cf228SFred Isaman /* cinfo->lock held by caller */ 9721763da12SFred Isaman int 973ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 974ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 975d6d6dc7cSFred Isaman { 976d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 977d6d6dc7cSFred Isaman int ret = 0; 978d6d6dc7cSFred Isaman 979d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 9808dd37758STrond Myklebust if (!nfs_lock_request(req)) 9818dd37758STrond Myklebust continue; 9827ad84aa9STrond Myklebust kref_get(&req->wb_kref); 983ea2cf228SFred Isaman if (cond_resched_lock(cinfo->lock)) 9843b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 985ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 9868dd37758STrond Myklebust nfs_list_add_request(req, dst); 987d6d6dc7cSFred Isaman ret++; 9881763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 989d6d6dc7cSFred Isaman break; 990d6d6dc7cSFred Isaman } 991d6d6dc7cSFred Isaman return ret; 992d6d6dc7cSFred Isaman } 993d6d6dc7cSFred Isaman 9941da177e4SLinus Torvalds /* 9951da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 9961da177e4SLinus Torvalds * @inode: NFS inode to scan 997ea2cf228SFred Isaman * @dst: mds destination list 998ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 9991da177e4SLinus Torvalds * 10001da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 10011da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 10021da177e4SLinus Torvalds */ 10031763da12SFred Isaman int 1004ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 1005ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 10061da177e4SLinus Torvalds { 1007d6d6dc7cSFred Isaman int ret = 0; 1008fb8a1f11STrond Myklebust 1009ea2cf228SFred Isaman spin_lock(cinfo->lock); 1010ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 10118dd37758STrond Myklebust const int max = INT_MAX; 1012d6d6dc7cSFred Isaman 1013ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 1014ea2cf228SFred Isaman cinfo, max); 1015ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 1016d6d6dc7cSFred Isaman } 1017ea2cf228SFred Isaman spin_unlock(cinfo->lock); 1018ff778d02STrond Myklebust return ret; 10191da177e4SLinus Torvalds } 1020d6d6dc7cSFred Isaman 10211da177e4SLinus Torvalds /* 1022e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1023e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 10241da177e4SLinus Torvalds * 1025e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1026e7d39069STrond Myklebust * to disk. 10271da177e4SLinus Torvalds */ 1028e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 1029e7d39069STrond Myklebust struct page *page, 1030e7d39069STrond Myklebust unsigned int offset, 1031e7d39069STrond Myklebust unsigned int bytes) 10321da177e4SLinus Torvalds { 1033e7d39069STrond Myklebust struct nfs_page *req; 1034e7d39069STrond Myklebust unsigned int rqend; 1035e7d39069STrond Myklebust unsigned int end; 10361da177e4SLinus Torvalds int error; 1037277459d2STrond Myklebust 1038e7d39069STrond Myklebust if (!PagePrivate(page)) 1039e7d39069STrond Myklebust return NULL; 1040e7d39069STrond Myklebust 1041e7d39069STrond Myklebust end = offset + bytes; 1042e7d39069STrond Myklebust spin_lock(&inode->i_lock); 1043e7d39069STrond Myklebust 1044e7d39069STrond Myklebust for (;;) { 104584d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 1046e7d39069STrond Myklebust if (req == NULL) 1047e7d39069STrond Myklebust goto out_unlock; 1048e7d39069STrond Myklebust 10492bfc6e56SWeston Andros Adamson /* should be handled by nfs_flush_incompatible */ 10502bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 10512bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 10522bfc6e56SWeston Andros Adamson 1053e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1054e7d39069STrond Myklebust /* 1055e7d39069STrond Myklebust * Tell the caller to flush out the request if 1056e7d39069STrond Myklebust * the offsets are non-contiguous. 1057e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1058e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1059e7d39069STrond Myklebust */ 1060e468bae9STrond Myklebust if (offset > rqend 1061e7d39069STrond Myklebust || end < req->wb_offset) 1062e7d39069STrond Myklebust goto out_flushme; 1063e7d39069STrond Myklebust 10647ad84aa9STrond Myklebust if (nfs_lock_request(req)) 1065e7d39069STrond Myklebust break; 1066e7d39069STrond Myklebust 1067e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 1068587142f8STrond Myklebust spin_unlock(&inode->i_lock); 10691da177e4SLinus Torvalds error = nfs_wait_on_request(req); 10701da177e4SLinus Torvalds nfs_release_request(req); 1071e7d39069STrond Myklebust if (error != 0) 1072e7d39069STrond Myklebust goto out_err; 1073e7d39069STrond Myklebust spin_lock(&inode->i_lock); 10741da177e4SLinus Torvalds } 10751da177e4SLinus Torvalds 10761da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 10771da177e4SLinus Torvalds if (offset < req->wb_offset) { 10781da177e4SLinus Torvalds req->wb_offset = offset; 10791da177e4SLinus Torvalds req->wb_pgbase = offset; 10801da177e4SLinus Torvalds } 10811da177e4SLinus Torvalds if (end > rqend) 10821da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1083e7d39069STrond Myklebust else 1084e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 1085e7d39069STrond Myklebust out_unlock: 1086ca138f36SFred Isaman if (req) 10878dd37758STrond Myklebust nfs_clear_request_commit(req); 1088411a99adSWeston Andros Adamson spin_unlock(&inode->i_lock); 1089e7d39069STrond Myklebust return req; 1090e7d39069STrond Myklebust out_flushme: 1091e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 1092e7d39069STrond Myklebust nfs_release_request(req); 1093e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1094e7d39069STrond Myklebust out_err: 1095e7d39069STrond Myklebust return ERR_PTR(error); 1096e7d39069STrond Myklebust } 10971da177e4SLinus Torvalds 1098e7d39069STrond Myklebust /* 1099e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1100e7d39069STrond Myklebust * 1101e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1102e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1103e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1104e7d39069STrond Myklebust */ 1105e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1106e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1107e7d39069STrond Myklebust { 1108d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1109e7d39069STrond Myklebust struct nfs_page *req; 1110e7d39069STrond Myklebust 1111e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1112e7d39069STrond Myklebust if (req != NULL) 1113e7d39069STrond Myklebust goto out; 11142bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 1115e7d39069STrond Myklebust if (IS_ERR(req)) 1116e7d39069STrond Myklebust goto out; 1117d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1118efc91ed0STrond Myklebust out: 111961e930a9STrond Myklebust return req; 11201da177e4SLinus Torvalds } 11211da177e4SLinus Torvalds 1122e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1123e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1124e7d39069STrond Myklebust { 1125e7d39069STrond Myklebust struct nfs_page *req; 1126e7d39069STrond Myklebust 1127e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1128e7d39069STrond Myklebust if (IS_ERR(req)) 1129e7d39069STrond Myklebust return PTR_ERR(req); 1130e7d39069STrond Myklebust /* Update file length */ 1131e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1132d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1133a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11341d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1135e7d39069STrond Myklebust return 0; 1136e7d39069STrond Myklebust } 1137e7d39069STrond Myklebust 11381da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 11391da177e4SLinus Torvalds { 1140cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 11412a369153STrond Myklebust struct nfs_lock_context *l_ctx; 1142bd61e0a9SJeff Layton struct file_lock_context *flctx = file_inode(file)->i_flctx; 11431da177e4SLinus Torvalds struct nfs_page *req; 11441a54533eSTrond Myklebust int do_flush, status; 11451da177e4SLinus Torvalds /* 11461da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 11471da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 11481da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 11491da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11501da177e4SLinus Torvalds * Also do the same if we find a request from an existing 11511da177e4SLinus Torvalds * dropped page. 11521da177e4SLinus Torvalds */ 11531a54533eSTrond Myklebust do { 115484d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 11551a54533eSTrond Myklebust if (req == NULL) 11561a54533eSTrond Myklebust return 0; 11572a369153STrond Myklebust l_ctx = req->wb_lock_context; 11582a369153STrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx; 11592bfc6e56SWeston Andros Adamson /* for now, flush if more than 1 request in page_group */ 11602bfc6e56SWeston Andros Adamson do_flush |= req->wb_this_page != req; 1161bd61e0a9SJeff Layton if (l_ctx && flctx && 1162bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1163bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 11645263e31eSJeff Layton do_flush |= l_ctx->lockowner.l_owner != current->files 11655263e31eSJeff Layton || l_ctx->lockowner.l_pid != current->tgid; 11665263e31eSJeff Layton } 11671da177e4SLinus Torvalds nfs_release_request(req); 11681a54533eSTrond Myklebust if (!do_flush) 11691a54533eSTrond Myklebust return 0; 1170d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 11711a54533eSTrond Myklebust } while (status == 0); 11721a54533eSTrond Myklebust return status; 11731da177e4SLinus Torvalds } 11741da177e4SLinus Torvalds 11751da177e4SLinus Torvalds /* 1176dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1177dc24826bSAndy Adamson * expire soon. 1178dc24826bSAndy Adamson * 1179dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1180dc24826bSAndy Adamson * 1181dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1182dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1183dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1184dc24826bSAndy Adamson */ 1185dc24826bSAndy Adamson int 1186dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1187dc24826bSAndy Adamson { 1188dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1189dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1190dc24826bSAndy Adamson 1191dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 1192dc24826bSAndy Adamson } 1193dc24826bSAndy Adamson 1194dc24826bSAndy Adamson /* 1195dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1196dc24826bSAndy Adamson */ 1197dc24826bSAndy Adamson bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) 1198dc24826bSAndy Adamson { 1199dc24826bSAndy Adamson return rpcauth_cred_key_to_expire(ctx->cred); 1200dc24826bSAndy Adamson } 1201dc24826bSAndy Adamson 1202dc24826bSAndy Adamson /* 12035d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 12045d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 12055d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 12065d47a356STrond Myklebust */ 12078d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 12085d47a356STrond Myklebust { 1209d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1210d529ef83SJeff Layton 12118d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 12128d197a56STrond Myklebust goto out; 121318dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1214d529ef83SJeff Layton return false; 12154db72b40SJeff Layton smp_rmb(); 1216d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 12178d197a56STrond Myklebust return false; 12188d197a56STrond Myklebust out: 121918dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 122018dd78c4SScott Mayhew return false; 12218d197a56STrond Myklebust return PageUptodate(page) != 0; 12225d47a356STrond Myklebust } 12235d47a356STrond Myklebust 12245263e31eSJeff Layton static bool 12255263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 12265263e31eSJeff Layton { 12275263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 12285263e31eSJeff Layton fl->fl_type == F_WRLCK; 12295263e31eSJeff Layton } 12305263e31eSJeff Layton 1231c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1232c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1233c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1234c7559663SScott Mayhew * inefficiencies. 1235c7559663SScott Mayhew * 1236263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1237263b4509SScott Mayhew * of the checks. 1238c7559663SScott Mayhew */ 1239c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1240c7559663SScott Mayhew { 12415263e31eSJeff Layton int ret; 12425263e31eSJeff Layton struct file_lock_context *flctx = inode->i_flctx; 12435263e31eSJeff Layton struct file_lock *fl; 12445263e31eSJeff Layton 1245c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1246c7559663SScott Mayhew return 0; 1247263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1248263b4509SScott Mayhew return 0; 1249c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1250c7559663SScott Mayhew return 1; 1251bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1252bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 12538fa4592aSTrond Myklebust return 1; 12545263e31eSJeff Layton 12555263e31eSJeff Layton /* Check to see if there are whole file write locks */ 12565263e31eSJeff Layton ret = 0; 12576109c850SJeff Layton spin_lock(&flctx->flc_lock); 1258bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1259bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1260bd61e0a9SJeff Layton fl_list); 1261bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 12625263e31eSJeff Layton ret = 1; 1263bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 12645263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 12655263e31eSJeff Layton fl_list); 12665263e31eSJeff Layton if (fl->fl_type == F_WRLCK) 12675263e31eSJeff Layton ret = 1; 12685263e31eSJeff Layton } 12696109c850SJeff Layton spin_unlock(&flctx->flc_lock); 12705263e31eSJeff Layton return ret; 1271c7559663SScott Mayhew } 1272c7559663SScott Mayhew 12735d47a356STrond Myklebust /* 12741da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 12751da177e4SLinus Torvalds * 12761da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 12771da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 12781da177e4SLinus Torvalds */ 12791da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 12801da177e4SLinus Torvalds unsigned int offset, unsigned int count) 12811da177e4SLinus Torvalds { 1282cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1283d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 12841da177e4SLinus Torvalds int status = 0; 12851da177e4SLinus Torvalds 128691d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 128791d5b470SChuck Lever 12886de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 12896de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 12901da177e4SLinus Torvalds 1291c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 129249a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 12931da177e4SLinus Torvalds offset = 0; 12941da177e4SLinus Torvalds } 12951da177e4SLinus Torvalds 1296e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 129703fa9e84STrond Myklebust if (status < 0) 129803fa9e84STrond Myklebust nfs_set_pageerror(page); 129959b7c05fSTrond Myklebust else 130059b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 13011da177e4SLinus Torvalds 130248186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 13031da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 13041da177e4SLinus Torvalds return status; 13051da177e4SLinus Torvalds } 13061da177e4SLinus Torvalds 13073ff7576dSTrond Myklebust static int flush_task_priority(int how) 13081da177e4SLinus Torvalds { 13091da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 13101da177e4SLinus Torvalds case FLUSH_HIGHPRI: 13111da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 13121da177e4SLinus Torvalds case FLUSH_LOWPRI: 13131da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 13141da177e4SLinus Torvalds } 13151da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 13161da177e4SLinus Torvalds } 13171da177e4SLinus Torvalds 1318d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1319d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1320abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 13211ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 13221da177e4SLinus Torvalds { 13233ff7576dSTrond Myklebust int priority = flush_task_priority(how); 13241da177e4SLinus Torvalds 13251ed26f33SAnna Schumaker task_setup_data->priority = priority; 1326abde71f4STom Haynes rpc_ops->write_setup(hdr, msg); 1327d138d5d1SAndy Adamson 1328abde71f4STom Haynes nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client, 1329d45f60c6SWeston Andros Adamson &task_setup_data->rpc_client, msg, hdr); 1330275acaafSTrond Myklebust } 1331275acaafSTrond Myklebust 13326d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 13336d884e8fSFred * call this on each, which will prepare them to be retried on next 13346d884e8fSFred * writeback using standard nfs. 13356d884e8fSFred */ 13366d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 13376d884e8fSFred { 13386d884e8fSFred nfs_mark_request_dirty(req); 1339c7070113STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 13401d1afcbcSTrond Myklebust nfs_unlock_request(req); 134120633f04SWeston Andros Adamson nfs_end_page_writeback(req); 13423aff4ebbSTrond Myklebust nfs_release_request(req); 13436d884e8fSFred } 13446d884e8fSFred 1345061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 13466c75dc0dSFred Isaman { 13476c75dc0dSFred Isaman struct nfs_page *req; 13486c75dc0dSFred Isaman 13496c75dc0dSFred Isaman while (!list_empty(head)) { 13506c75dc0dSFred Isaman req = nfs_list_entry(head->next); 13516c75dc0dSFred Isaman nfs_list_remove_request(req); 13526c75dc0dSFred Isaman nfs_redirty_request(req); 13536c75dc0dSFred Isaman } 13546c75dc0dSFred Isaman } 13556c75dc0dSFred Isaman 1356061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1357061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1358061ae2edSFred Isaman .completion = nfs_write_completion, 1359061ae2edSFred Isaman }; 1360061ae2edSFred Isaman 136157208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1362a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1363061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 13641751c363STrond Myklebust { 1365a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 136641d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1367a20c93e3SChristoph Hellwig 1368a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1369a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1370a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1371a20c93e3SChristoph Hellwig #endif 13724a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 13734a0de55cSAnna Schumaker server->wsize, ioflags); 13741751c363STrond Myklebust } 1375ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 13761751c363STrond Myklebust 1377dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1378dce81290STrond Myklebust { 1379a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1380a7d42ddbSWeston Andros Adamson 13816f29b9bbSKinglong Mee if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) 13826f29b9bbSKinglong Mee pgio->pg_ops->pg_cleanup(pgio); 13836f29b9bbSKinglong Mee 138441d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1385a7d42ddbSWeston Andros Adamson 1386a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1387a7d42ddbSWeston Andros Adamson 1388a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1389a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1390dce81290STrond Myklebust } 13911f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1392dce81290STrond Myklebust 13931da177e4SLinus Torvalds 13940b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 13950b7c0153SFred Isaman { 13960b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 13970b7c0153SFred Isaman 13980b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 13990b7c0153SFred Isaman } 14000b7c0153SFred Isaman 14011f2edbe3STrond Myklebust /* 14021f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 14031f2edbe3STrond Myklebust */ 14041f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 14051f2edbe3STrond Myklebust { 14061f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 14071f2edbe3STrond Myklebust int kill = 0; 1408788e7a89STrond Myklebust 14091f2edbe3STrond Myklebust /* suid always must be killed */ 14101f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 14111f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 14121f2edbe3STrond Myklebust 14131f2edbe3STrond Myklebust /* 14141f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 14151f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 14161f2edbe3STrond Myklebust */ 14171f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 14181f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 14191f2edbe3STrond Myklebust 14201f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 14211f2edbe3STrond Myklebust return kill; 14221f2edbe3STrond Myklebust 14231f2edbe3STrond Myklebust return 0; 14241f2edbe3STrond Myklebust } 1425788e7a89STrond Myklebust 1426a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1427a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1428a08a8cd3STrond Myklebust { 1429a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1430a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 14312b83d3deSTrond Myklebust u64 size = argp->offset + resp->count; 1432a08a8cd3STrond Myklebust 1433a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 14342b83d3deSTrond Myklebust fattr->size = size; 14352b83d3deSTrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { 14362b83d3deSTrond Myklebust fattr->valid &= ~NFS_ATTR_FATTR_SIZE; 1437a08a8cd3STrond Myklebust return; 14382b83d3deSTrond Myklebust } 14392b83d3deSTrond Myklebust if (size != fattr->size) 1440a08a8cd3STrond Myklebust return; 1441a08a8cd3STrond Myklebust /* Set attribute barrier */ 1442a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 14432b83d3deSTrond Myklebust /* ...and update size */ 14442b83d3deSTrond Myklebust fattr->valid |= NFS_ATTR_FATTR_SIZE; 1445a08a8cd3STrond Myklebust } 1446a08a8cd3STrond Myklebust 1447a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1448a08a8cd3STrond Myklebust { 14492b83d3deSTrond Myklebust struct nfs_fattr *fattr = &hdr->fattr; 1450a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1451a08a8cd3STrond Myklebust 1452a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1453a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1454a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1455a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1456a08a8cd3STrond Myklebust } 1457a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1458a08a8cd3STrond Myklebust 14591da177e4SLinus Torvalds /* 14601da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 14611da177e4SLinus Torvalds */ 1462d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1463d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 14640eecb214SAnna Schumaker struct inode *inode) 14651da177e4SLinus Torvalds { 1466788e7a89STrond Myklebust int status; 14671da177e4SLinus Torvalds 1468f551e44fSChuck Lever /* 1469f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1470f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1471f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1472f551e44fSChuck Lever * another writer had changed the file, but some applications 1473f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1474f551e44fSChuck Lever */ 1475d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1476788e7a89STrond Myklebust if (status != 0) 14770eecb214SAnna Schumaker return status; 1478d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 147991d5b470SChuck Lever 1480d45f60c6SWeston Andros Adamson if (hdr->res.verf->committed < hdr->args.stable && 1481d45f60c6SWeston Andros Adamson task->tk_status >= 0) { 14821da177e4SLinus Torvalds /* We tried a write call, but the server did not 14831da177e4SLinus Torvalds * commit data to stable storage even though we 14841da177e4SLinus Torvalds * requested it. 14851da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 14861da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 14871da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 14881da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 14891da177e4SLinus Torvalds */ 14901da177e4SLinus Torvalds static unsigned long complain; 14911da177e4SLinus Torvalds 1492a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 14931da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 14941da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 14951da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1496cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1497d45f60c6SWeston Andros Adamson hdr->res.verf->committed, hdr->args.stable); 14981da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 14991da177e4SLinus Torvalds } 15001da177e4SLinus Torvalds } 15011f2edbe3STrond Myklebust 15021f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 15031f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 15041f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 15050eecb214SAnna Schumaker return 0; 15060eecb214SAnna Schumaker } 15070eecb214SAnna Schumaker 15080eecb214SAnna Schumaker /* 15090eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 15100eecb214SAnna Schumaker */ 1511d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1512d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 15130eecb214SAnna Schumaker { 1514d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1515d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 15161f2edbe3STrond Myklebust 15171f2edbe3STrond Myklebust if (resp->count < argp->count) { 15181da177e4SLinus Torvalds static unsigned long complain; 15191da177e4SLinus Torvalds 15206c75dc0dSFred Isaman /* This a short write! */ 1521d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 152291d5b470SChuck Lever 15231da177e4SLinus Torvalds /* Has the server at least made some progress? */ 15246c75dc0dSFred Isaman if (resp->count == 0) { 15256c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 15266c75dc0dSFred Isaman printk(KERN_WARNING 15276c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 15286c75dc0dSFred Isaman argp->count); 15296c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 15306c75dc0dSFred Isaman } 1531d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 15326c75dc0dSFred Isaman task->tk_status = -EIO; 15336c75dc0dSFred Isaman return; 15346c75dc0dSFred Isaman } 1535f8417b48SKinglong Mee 1536f8417b48SKinglong Mee /* For non rpc-based layout drivers, retry-through-MDS */ 1537f8417b48SKinglong Mee if (!task->tk_ops) { 1538f8417b48SKinglong Mee hdr->pnfs_error = -EAGAIN; 1539f8417b48SKinglong Mee return; 1540f8417b48SKinglong Mee } 1541f8417b48SKinglong Mee 15421da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 15431da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 15441da177e4SLinus Torvalds /* Resend from where the server left off */ 1545d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 15461da177e4SLinus Torvalds argp->offset += resp->count; 15471da177e4SLinus Torvalds argp->pgbase += resp->count; 15481da177e4SLinus Torvalds argp->count -= resp->count; 15491da177e4SLinus Torvalds } else { 15501da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 15511da177e4SLinus Torvalds * headaches in the case of a server crash. 15521da177e4SLinus Torvalds */ 15531da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 15541da177e4SLinus Torvalds } 1555d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 15561da177e4SLinus Torvalds } 15571da177e4SLinus Torvalds } 15581da177e4SLinus Torvalds 15591da177e4SLinus Torvalds 156071d0a611STrond Myklebust static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) 156171d0a611STrond Myklebust { 1562b8413f98STrond Myklebust int ret; 1563b8413f98STrond Myklebust 156471d0a611STrond Myklebust if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) 156571d0a611STrond Myklebust return 1; 1566b8413f98STrond Myklebust if (!may_wait) 156771d0a611STrond Myklebust return 0; 1568b8413f98STrond Myklebust ret = out_of_line_wait_on_bit_lock(&nfsi->flags, 1569b8413f98STrond Myklebust NFS_INO_COMMIT, 1570b8413f98STrond Myklebust nfs_wait_bit_killable, 1571b8413f98STrond Myklebust TASK_KILLABLE); 1572b8413f98STrond Myklebust return (ret < 0) ? ret : 1; 157371d0a611STrond Myklebust } 157471d0a611STrond Myklebust 1575f453a54aSFred Isaman static void nfs_commit_clear_lock(struct nfs_inode *nfsi) 157671d0a611STrond Myklebust { 157771d0a611STrond Myklebust clear_bit(NFS_INO_COMMIT, &nfsi->flags); 15784e857c58SPeter Zijlstra smp_mb__after_atomic(); 157971d0a611STrond Myklebust wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); 158071d0a611STrond Myklebust } 158171d0a611STrond Myklebust 15820b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 15831da177e4SLinus Torvalds { 15840b7c0153SFred Isaman put_nfs_open_context(data->context); 15850b7c0153SFred Isaman nfs_commit_free(data); 15861da177e4SLinus Torvalds } 1587e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 15881da177e4SLinus Torvalds 15890b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1590c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 15919ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 15929f0ec176SAndy Adamson int how, int flags) 15931da177e4SLinus Torvalds { 159407737691STrond Myklebust struct rpc_task *task; 15959ace33cdSFred Isaman int priority = flush_task_priority(how); 1596bdc7f021STrond Myklebust struct rpc_message msg = { 1597bdc7f021STrond Myklebust .rpc_argp = &data->args, 1598bdc7f021STrond Myklebust .rpc_resp = &data->res, 15999ace33cdSFred Isaman .rpc_cred = data->cred, 1600bdc7f021STrond Myklebust }; 160184115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 160207737691STrond Myklebust .task = &data->task, 16039ace33cdSFred Isaman .rpc_client = clnt, 1604bdc7f021STrond Myklebust .rpc_message = &msg, 16059ace33cdSFred Isaman .callback_ops = call_ops, 160684115e1cSTrond Myklebust .callback_data = data, 1607101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 16089f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 16093ff7576dSTrond Myklebust .priority = priority, 161084115e1cSTrond Myklebust }; 1611788e7a89STrond Myklebust /* Set up the initial task struct. */ 1612c36aae9aSPeng Tao nfs_ops->commit_setup(data, &msg); 16131da177e4SLinus Torvalds 1614b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1615bdc7f021STrond Myklebust 16168c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 16178c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 16188c21c62cSWeston Andros Adamson 161907737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1620dbae4c73STrond Myklebust if (IS_ERR(task)) 1621dbae4c73STrond Myklebust return PTR_ERR(task); 1622d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1623d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 162407737691STrond Myklebust rpc_put_task(task); 1625dbae4c73STrond Myklebust return 0; 16261da177e4SLinus Torvalds } 1627e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 16281da177e4SLinus Torvalds 1629378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1630378520b8SPeng Tao { 1631378520b8SPeng Tao loff_t lwb = 0; 1632378520b8SPeng Tao struct nfs_page *req; 1633378520b8SPeng Tao 1634378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1635378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1636378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1637378520b8SPeng Tao 1638378520b8SPeng Tao return lwb; 1639378520b8SPeng Tao } 1640378520b8SPeng Tao 16411da177e4SLinus Torvalds /* 16429ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 16439ace33cdSFred Isaman */ 16440b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1645988b6dceSFred Isaman struct list_head *head, 1646f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1647f453a54aSFred Isaman struct nfs_commit_info *cinfo) 16489ace33cdSFred Isaman { 16499ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 16502b0143b5SDavid Howells struct inode *inode = d_inode(first->wb_context->dentry); 16519ace33cdSFred Isaman 16529ace33cdSFred Isaman /* Set up the RPC argument and reply structs 16539ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 16549ace33cdSFred Isaman 16559ace33cdSFred Isaman list_splice_init(head, &data->pages); 16569ace33cdSFred Isaman 16579ace33cdSFred Isaman data->inode = inode; 16589ace33cdSFred Isaman data->cred = first->wb_context->cred; 1659988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1660378520b8SPeng Tao /* only set lwb for pnfs commit */ 1661378520b8SPeng Tao if (lseg) 1662378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 16639ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1664f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1665b359f9d0SFred Isaman data->dreq = cinfo->dreq; 16669ace33cdSFred Isaman 16679ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 16689ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 16699ace33cdSFred Isaman data->args.offset = 0; 16709ace33cdSFred Isaman data->args.count = 0; 16710b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 16729ace33cdSFred Isaman data->res.fattr = &data->fattr; 16739ace33cdSFred Isaman data->res.verf = &data->verf; 16749ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 16759ace33cdSFred Isaman } 1676e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 16779ace33cdSFred Isaman 1678e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1679ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1680b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1681b57ff130SWeston Andros Adamson u32 ds_commit_idx) 168264bfeb49SFred Isaman { 168364bfeb49SFred Isaman struct nfs_page *req; 168464bfeb49SFred Isaman 168564bfeb49SFred Isaman while (!list_empty(page_list)) { 168664bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 168764bfeb49SFred Isaman nfs_list_remove_request(req); 1688b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 1689487b9b8aSTom Haynes if (!cinfo->dreq) 1690487b9b8aSTom Haynes nfs_clear_page_commit(req->wb_page); 16911d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 169264bfeb49SFred Isaman } 169364bfeb49SFred Isaman } 1694e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 169564bfeb49SFred Isaman 16969ace33cdSFred Isaman /* 16971da177e4SLinus Torvalds * Commit dirty pages 16981da177e4SLinus Torvalds */ 16991da177e4SLinus Torvalds static int 1700ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1701ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 17021da177e4SLinus Torvalds { 17030b7c0153SFred Isaman struct nfs_commit_data *data; 17041da177e4SLinus Torvalds 1705c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 17061da177e4SLinus Torvalds 17071da177e4SLinus Torvalds if (!data) 17081da177e4SLinus Torvalds goto out_bad; 17091da177e4SLinus Torvalds 17101da177e4SLinus Torvalds /* Set up the argument struct */ 1711f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1712f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 1713c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 1714c36aae9aSPeng Tao data->mds_ops, how, 0); 17151da177e4SLinus Torvalds out_bad: 1716b57ff130SWeston Andros Adamson nfs_retry_commit(head, NULL, cinfo, 0); 1717f453a54aSFred Isaman cinfo->completion_ops->error_cleanup(NFS_I(inode)); 17181da177e4SLinus Torvalds return -ENOMEM; 17191da177e4SLinus Torvalds } 17201da177e4SLinus Torvalds 17211da177e4SLinus Torvalds /* 17221da177e4SLinus Torvalds * COMMIT call returned 17231da177e4SLinus Torvalds */ 1724788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 17251da177e4SLinus Torvalds { 17260b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 17271da177e4SLinus Torvalds 1728a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 17291da177e4SLinus Torvalds task->tk_pid, task->tk_status); 17301da177e4SLinus Torvalds 1731788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1732c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1733c9d8f89dSTrond Myklebust } 1734c9d8f89dSTrond Myklebust 1735f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1736c9d8f89dSTrond Myklebust { 1737c9d8f89dSTrond Myklebust struct nfs_page *req; 1738c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1739f453a54aSFred Isaman struct nfs_commit_info cinfo; 1740353db796SNeilBrown struct nfs_server *nfss; 1741788e7a89STrond Myklebust 17421da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 17431da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 17441da177e4SLinus Torvalds nfs_list_remove_request(req); 1745d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 17461da177e4SLinus Torvalds 17471e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 17483d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 17492b0143b5SDavid Howells (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)), 17501da177e4SLinus Torvalds req->wb_bytes, 17511da177e4SLinus Torvalds (long long)req_offset(req)); 1752c9d8f89dSTrond Myklebust if (status < 0) { 1753c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 17541da177e4SLinus Torvalds nfs_inode_remove_request(req); 1755c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 17561da177e4SLinus Torvalds goto next; 17571da177e4SLinus Torvalds } 17581da177e4SLinus Torvalds 17591da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 17601da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 17612f2c63bcSTrond Myklebust if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { 17621da177e4SLinus Torvalds /* We have a match */ 17631da177e4SLinus Torvalds nfs_inode_remove_request(req); 17641da177e4SLinus Torvalds dprintk(" OK\n"); 17651da177e4SLinus Torvalds goto next; 17661da177e4SLinus Torvalds } 17671da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 17681da177e4SLinus Torvalds dprintk(" mismatch\n"); 17696d884e8fSFred nfs_mark_request_dirty(req); 177005990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 17711da177e4SLinus Torvalds next: 17721d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 17731da177e4SLinus Torvalds } 1774353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1775353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 1776353db796SNeilBrown clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 1777353db796SNeilBrown 1778f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1779f453a54aSFred Isaman if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) 1780f453a54aSFred Isaman nfs_commit_clear_lock(NFS_I(data->inode)); 17815917ce84SFred Isaman } 17825917ce84SFred Isaman 17835917ce84SFred Isaman static void nfs_commit_release(void *calldata) 17845917ce84SFred Isaman { 17850b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 17865917ce84SFred Isaman 1787f453a54aSFred Isaman data->completion_ops->completion(data); 1788c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 17891da177e4SLinus Torvalds } 1790788e7a89STrond Myklebust 1791788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 17920b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1793788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1794788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1795788e7a89STrond Myklebust }; 17961da177e4SLinus Torvalds 1797f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1798f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1799f453a54aSFred Isaman .error_cleanup = nfs_commit_clear_lock, 1800f453a54aSFred Isaman }; 1801f453a54aSFred Isaman 18021763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1803ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 180484c53ab5SFred Isaman { 180584c53ab5SFred Isaman int status; 180684c53ab5SFred Isaman 1807ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 180884c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1809ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 181084c53ab5SFred Isaman return status; 181184c53ab5SFred Isaman } 181284c53ab5SFred Isaman 1813b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 18141da177e4SLinus Torvalds { 18151da177e4SLinus Torvalds LIST_HEAD(head); 1816ea2cf228SFred Isaman struct nfs_commit_info cinfo; 181771d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1818b8413f98STrond Myklebust int res; 18191da177e4SLinus Torvalds 1820b8413f98STrond Myklebust res = nfs_commit_set_lock(NFS_I(inode), may_wait); 1821b8413f98STrond Myklebust if (res <= 0) 1822c5efa5fcSTrond Myklebust goto out_mark_dirty; 1823ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1824ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 18251da177e4SLinus Torvalds if (res) { 1826a861a1e1SFred Isaman int error; 1827a861a1e1SFred Isaman 1828ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 18291da177e4SLinus Torvalds if (error < 0) 18301da177e4SLinus Torvalds return error; 1831b8413f98STrond Myklebust if (!may_wait) 1832b8413f98STrond Myklebust goto out_mark_dirty; 183374316201SNeilBrown error = wait_on_bit_action(&NFS_I(inode)->flags, 1834b8413f98STrond Myklebust NFS_INO_COMMIT, 183571d0a611STrond Myklebust nfs_wait_bit_killable, 183671d0a611STrond Myklebust TASK_KILLABLE); 1837b8413f98STrond Myklebust if (error < 0) 1838b8413f98STrond Myklebust return error; 183971d0a611STrond Myklebust } else 184071d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(inode)); 1841c5efa5fcSTrond Myklebust return res; 1842c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1843c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1844c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1845c5efa5fcSTrond Myklebust * that the data is on the disk. 1846c5efa5fcSTrond Myklebust */ 1847c5efa5fcSTrond Myklebust out_mark_dirty: 1848c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 18491da177e4SLinus Torvalds return res; 18501da177e4SLinus Torvalds } 18518fc795f7STrond Myklebust 1852ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 18538fc795f7STrond Myklebust { 1854420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1855420e3646STrond Myklebust int flags = FLUSH_SYNC; 1856420e3646STrond Myklebust int ret = 0; 18578fc795f7STrond Myklebust 18583236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1859ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 18603236c3e1SJeff Layton return ret; 18613236c3e1SJeff Layton 1862a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1863a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1864a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1865420e3646STrond Myklebust */ 1866cb1410c7SWeston Andros Adamson if (nfsi->commit_info.ncommit <= (nfsi->nrequests >> 1)) 1867420e3646STrond Myklebust goto out_mark_dirty; 1868420e3646STrond Myklebust 1869a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1870420e3646STrond Myklebust flags = 0; 1871a00dd6c0SJeff Layton } 1872a00dd6c0SJeff Layton 1873420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1874420e3646STrond Myklebust if (ret >= 0) { 1875420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1876420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1877420e3646STrond Myklebust wbc->nr_to_write -= ret; 1878420e3646STrond Myklebust else 1879420e3646STrond Myklebust wbc->nr_to_write = 0; 1880420e3646STrond Myklebust } 18818fc795f7STrond Myklebust return 0; 1882420e3646STrond Myklebust } 1883420e3646STrond Myklebust out_mark_dirty: 18848fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 18858fc795f7STrond Myklebust return ret; 18868fc795f7STrond Myklebust } 188789d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1888863a3c6cSAndy Adamson 1889acdc53b2STrond Myklebust /* 1890acdc53b2STrond Myklebust * flush the inode to disk. 1891acdc53b2STrond Myklebust */ 1892acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 189334901f70STrond Myklebust { 1894f4ce1299STrond Myklebust int ret; 189534901f70STrond Myklebust 1896f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1897f4ce1299STrond Myklebust 18985bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 18996b196875SChuck Lever if (ret) 19006b196875SChuck Lever goto out; 19015bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 19026b196875SChuck Lever if (ret < 0) 19036b196875SChuck Lever goto out; 19045bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 19056b196875SChuck Lever ret = 0; 1906f4ce1299STrond Myklebust 19076b196875SChuck Lever out: 1908f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1909f4ce1299STrond Myklebust return ret; 19101c75950bSTrond Myklebust } 1911ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 19121c75950bSTrond Myklebust 19131b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 19141b3b4a1aSTrond Myklebust { 19151b3b4a1aSTrond Myklebust struct nfs_page *req; 19161b3b4a1aSTrond Myklebust int ret = 0; 19171b3b4a1aSTrond Myklebust 1918ba8b06e6STrond Myklebust wait_on_page_writeback(page); 19193e217045SWeston Andros Adamson 19203e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 19213e217045SWeston Andros Adamson * request */ 19223e217045SWeston Andros Adamson req = nfs_lock_and_join_requests(page, false); 19233e217045SWeston Andros Adamson 19243e217045SWeston Andros Adamson if (IS_ERR(req)) { 19253e217045SWeston Andros Adamson ret = PTR_ERR(req); 19263e217045SWeston Andros Adamson } else if (req) { 19273e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 19283e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 19293e217045SWeston Andros Adamson * request from the inode / page_private pointer and 19303e217045SWeston Andros Adamson * release it */ 19311b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 19321d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 19331b3b4a1aSTrond Myklebust } 19343e217045SWeston Andros Adamson 19351b3b4a1aSTrond Myklebust return ret; 19361b3b4a1aSTrond Myklebust } 19371b3b4a1aSTrond Myklebust 19381c75950bSTrond Myklebust /* 19391c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 19401c75950bSTrond Myklebust */ 1941*d6c843b9SPeng Tao int nfs_wb_single_page(struct inode *inode, struct page *page, bool launder) 19421c75950bSTrond Myklebust { 194329418aa4SMel Gorman loff_t range_start = page_file_offset(page); 19447f2f12d9STrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 19457f2f12d9STrond Myklebust struct writeback_control wbc = { 19467f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 19477f2f12d9STrond Myklebust .nr_to_write = 0, 19487f2f12d9STrond Myklebust .range_start = range_start, 19497f2f12d9STrond Myklebust .range_end = range_end, 19507f2f12d9STrond Myklebust }; 19517f2f12d9STrond Myklebust int ret; 19527f2f12d9STrond Myklebust 1953f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 1954f4ce1299STrond Myklebust 19550522f6adSTrond Myklebust for (;;) { 1956ba8b06e6STrond Myklebust wait_on_page_writeback(page); 19577f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 1958*d6c843b9SPeng Tao ret = nfs_writepage_locked(page, &wbc, launder); 19597f2f12d9STrond Myklebust if (ret < 0) 19607f2f12d9STrond Myklebust goto out_error; 19610522f6adSTrond Myklebust continue; 19627f2f12d9STrond Myklebust } 1963f4ce1299STrond Myklebust ret = 0; 19640522f6adSTrond Myklebust if (!PagePrivate(page)) 19650522f6adSTrond Myklebust break; 19660522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 19677f2f12d9STrond Myklebust if (ret < 0) 19687f2f12d9STrond Myklebust goto out_error; 19697f2f12d9STrond Myklebust } 19707f2f12d9STrond Myklebust out_error: 1971f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 19727f2f12d9STrond Myklebust return ret; 19731c75950bSTrond Myklebust } 19741c75950bSTrond Myklebust 1975074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 1976074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 1977a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 1978074cc1deSTrond Myklebust { 19792da95652SJeff Layton /* 19802da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 19812da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 19822da95652SJeff Layton * 19832da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 19842da95652SJeff Layton * that we can safely release the inode reference while holding 19852da95652SJeff Layton * the page lock. 19862da95652SJeff Layton */ 19872da95652SJeff Layton if (PagePrivate(page)) 19882da95652SJeff Layton return -EBUSY; 1989074cc1deSTrond Myklebust 19908c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 19918c209ce7SDavid Howells return -EBUSY; 1992074cc1deSTrond Myklebust 1993a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 1994074cc1deSTrond Myklebust } 1995074cc1deSTrond Myklebust #endif 1996074cc1deSTrond Myklebust 1997f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 19981da177e4SLinus Torvalds { 19991da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 20001e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 20011da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 200220c2df83SPaul Mundt NULL); 20031da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 20041da177e4SLinus Torvalds return -ENOMEM; 20051da177e4SLinus Torvalds 200693d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 20071da177e4SLinus Torvalds nfs_wdata_cachep); 20081da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 20093dd4765fSJeff Layton goto out_destroy_write_cache; 20101da177e4SLinus Torvalds 20110b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 20120b7c0153SFred Isaman sizeof(struct nfs_commit_data), 20130b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 20140b7c0153SFred Isaman NULL); 20150b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 20163dd4765fSJeff Layton goto out_destroy_write_mempool; 20170b7c0153SFred Isaman 201893d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 20194c100210SYanchuan Nian nfs_cdata_cachep); 20201da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 20213dd4765fSJeff Layton goto out_destroy_commit_cache; 20221da177e4SLinus Torvalds 202389a09141SPeter Zijlstra /* 202489a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 202589a09141SPeter Zijlstra * 202689a09141SPeter Zijlstra * 64MB: 8192k 202789a09141SPeter Zijlstra * 128MB: 11585k 202889a09141SPeter Zijlstra * 256MB: 16384k 202989a09141SPeter Zijlstra * 512MB: 23170k 203089a09141SPeter Zijlstra * 1GB: 32768k 203189a09141SPeter Zijlstra * 2GB: 46340k 203289a09141SPeter Zijlstra * 4GB: 65536k 203389a09141SPeter Zijlstra * 8GB: 92681k 203489a09141SPeter Zijlstra * 16GB: 131072k 203589a09141SPeter Zijlstra * 203689a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 203789a09141SPeter Zijlstra * Limit the default to 256M 203889a09141SPeter Zijlstra */ 203989a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 204089a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 204189a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 204289a09141SPeter Zijlstra 20431da177e4SLinus Torvalds return 0; 20443dd4765fSJeff Layton 20453dd4765fSJeff Layton out_destroy_commit_cache: 20463dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 20473dd4765fSJeff Layton out_destroy_write_mempool: 20483dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 20493dd4765fSJeff Layton out_destroy_write_cache: 20503dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 20513dd4765fSJeff Layton return -ENOMEM; 20521da177e4SLinus Torvalds } 20531da177e4SLinus Torvalds 2054266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 20551da177e4SLinus Torvalds { 20561da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 20573dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 20581da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 20591a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 20601da177e4SLinus Torvalds } 20611da177e4SLinus Torvalds 20624a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 2063a4cdda59SAnna Schumaker .rw_mode = FMODE_WRITE, 20644a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 20654a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 20660eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 20670eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 20681ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 20694a0de55cSAnna Schumaker }; 2070