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> 24af7cf057STrond Myklebust #include <linux/freezer.h> 25af7cf057STrond Myklebust #include <linux/wait.h> 263fcfab16SAndrew Morton 277c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 281da177e4SLinus Torvalds 291da177e4SLinus Torvalds #include "delegation.h" 3049a70f27STrond Myklebust #include "internal.h" 3191d5b470SChuck Lever #include "iostat.h" 32def6ed7eSAndy Adamson #include "nfs4_fs.h" 33074cc1deSTrond Myklebust #include "fscache.h" 3494ad1c80SFred Isaman #include "pnfs.h" 351da177e4SLinus Torvalds 36f4ce1299STrond Myklebust #include "nfstrace.h" 37f4ce1299STrond Myklebust 381da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 391da177e4SLinus Torvalds 401da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 411da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 421da177e4SLinus Torvalds 431da177e4SLinus Torvalds /* 441da177e4SLinus Torvalds * Local function declarations 451da177e4SLinus Torvalds */ 46f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 47788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 48061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; 49f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops; 504a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops; 51d4581383SWeston Andros Adamson static void nfs_clear_request_commit(struct nfs_page *req); 5202d1426cSWeston Andros Adamson static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 5302d1426cSWeston Andros Adamson struct inode *inode); 543a3908c8STrond Myklebust static struct nfs_page * 553a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 563a3908c8STrond Myklebust struct page *page); 571da177e4SLinus Torvalds 58e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 593feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 600b7c0153SFred Isaman static struct kmem_cache *nfs_cdata_cachep; 611da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 621da177e4SLinus Torvalds 630b7c0153SFred Isaman struct nfs_commit_data *nfs_commitdata_alloc(void) 641da177e4SLinus Torvalds { 65192e501bSMel Gorman struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); 6640859d7eSChuck Lever 671da177e4SLinus Torvalds if (p) { 681da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 691da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 701da177e4SLinus Torvalds } 711da177e4SLinus Torvalds return p; 721da177e4SLinus Torvalds } 73e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 741da177e4SLinus Torvalds 750b7c0153SFred Isaman void nfs_commit_free(struct nfs_commit_data *p) 761da177e4SLinus Torvalds { 771da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 781da177e4SLinus Torvalds } 79e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 801da177e4SLinus Torvalds 811e7f3a48SWeston Andros Adamson static struct nfs_pgio_header *nfs_writehdr_alloc(void) 823feb2d49STrond Myklebust { 831e7f3a48SWeston Andros Adamson struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); 843feb2d49STrond Myklebust 854a0de55cSAnna Schumaker if (p) 863feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 873feb2d49STrond Myklebust return p; 883feb2d49STrond Myklebust } 893feb2d49STrond Myklebust 901e7f3a48SWeston Andros Adamson static void nfs_writehdr_free(struct nfs_pgio_header *hdr) 916c75dc0dSFred Isaman { 921e7f3a48SWeston Andros Adamson mempool_free(hdr, nfs_wdata_mempool); 933feb2d49STrond Myklebust } 941da177e4SLinus Torvalds 957b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 967b159fc1STrond Myklebust { 977b159fc1STrond Myklebust ctx->error = error; 987b159fc1STrond Myklebust smp_wmb(); 997b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 1007b159fc1STrond Myklebust } 1017b159fc1STrond Myklebust 10284d3a9a9SWeston Andros Adamson /* 10384d3a9a9SWeston Andros Adamson * nfs_page_find_head_request_locked - find head request associated with @page 10484d3a9a9SWeston Andros Adamson * 10584d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 10684d3a9a9SWeston Andros Adamson * 10784d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 10884d3a9a9SWeston Andros Adamson */ 10929418aa4SMel Gorman static struct nfs_page * 11084d3a9a9SWeston Andros Adamson nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page) 111277459d2STrond Myklebust { 112277459d2STrond Myklebust struct nfs_page *req = NULL; 113277459d2STrond Myklebust 11429418aa4SMel Gorman if (PagePrivate(page)) 115277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 11602d1426cSWeston Andros Adamson else if (unlikely(PageSwapCache(page))) 11702d1426cSWeston Andros Adamson req = nfs_page_search_commits_for_head_request_locked(nfsi, 11802d1426cSWeston Andros Adamson page); 11929418aa4SMel Gorman 12084d3a9a9SWeston Andros Adamson if (req) { 12184d3a9a9SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 12229418aa4SMel Gorman kref_get(&req->wb_kref); 12384d3a9a9SWeston Andros Adamson } 12429418aa4SMel Gorman 125277459d2STrond Myklebust return req; 126277459d2STrond Myklebust } 127277459d2STrond Myklebust 12884d3a9a9SWeston Andros Adamson /* 12984d3a9a9SWeston Andros Adamson * nfs_page_find_head_request - find head request associated with @page 13084d3a9a9SWeston Andros Adamson * 13184d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 13284d3a9a9SWeston Andros Adamson */ 13384d3a9a9SWeston Andros Adamson static struct nfs_page *nfs_page_find_head_request(struct page *page) 134277459d2STrond Myklebust { 135d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 136277459d2STrond Myklebust struct nfs_page *req = NULL; 137277459d2STrond Myklebust 138587142f8STrond Myklebust spin_lock(&inode->i_lock); 13984d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 140587142f8STrond Myklebust spin_unlock(&inode->i_lock); 141277459d2STrond Myklebust return req; 142277459d2STrond Myklebust } 143277459d2STrond Myklebust 1441da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1451da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1461da177e4SLinus Torvalds { 147d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 148a3d01454STrond Myklebust loff_t end, i_size; 149a3d01454STrond Myklebust pgoff_t end_index; 1501da177e4SLinus Torvalds 151a3d01454STrond Myklebust spin_lock(&inode->i_lock); 152a3d01454STrond Myklebust i_size = i_size_read(inode); 15309cbfeafSKirill A. Shutemov end_index = (i_size - 1) >> PAGE_SHIFT; 1548cd79788SHuang Ying if (i_size > 0 && page_index(page) < end_index) 155a3d01454STrond Myklebust goto out; 156d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 1571da177e4SLinus Torvalds if (i_size >= end) 158a3d01454STrond Myklebust goto out; 1591da177e4SLinus Torvalds i_size_write(inode, end); 160a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 161a3d01454STrond Myklebust out: 162a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 1631da177e4SLinus Torvalds } 1641da177e4SLinus Torvalds 165a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 166a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 167a301b777STrond Myklebust { 168d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 169a301b777STrond Myklebust } 170a301b777STrond Myklebust 171d72ddcbaSWeston Andros Adamson /* 172d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 173d72ddcbaSWeston Andros Adamson * @head - head request of page group 174d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 175d72ddcbaSWeston Andros Adamson * 176d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 177d72ddcbaSWeston Andros Adamson * page offset @page_offset. 178d72ddcbaSWeston Andros Adamson * 179d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 180d72ddcbaSWeston Andros Adamson * match is found. 181d72ddcbaSWeston Andros Adamson * 182d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 183d72ddcbaSWeston Andros Adamson */ 184d72ddcbaSWeston Andros Adamson static struct nfs_page * 185d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 186d72ddcbaSWeston Andros Adamson { 187d72ddcbaSWeston Andros Adamson struct nfs_page *req; 188d72ddcbaSWeston Andros Adamson 189d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(head != head->wb_head); 190d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags)); 191d72ddcbaSWeston Andros Adamson 192d72ddcbaSWeston Andros Adamson req = head; 193d72ddcbaSWeston Andros Adamson do { 194d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 195d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 196d72ddcbaSWeston Andros Adamson return req; 197d72ddcbaSWeston Andros Adamson 198d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 199d72ddcbaSWeston Andros Adamson } while (req != head); 200d72ddcbaSWeston Andros Adamson 201d72ddcbaSWeston Andros Adamson return NULL; 202d72ddcbaSWeston Andros Adamson } 203d72ddcbaSWeston Andros Adamson 204d72ddcbaSWeston Andros Adamson /* 205d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 206d72ddcbaSWeston Andros Adamson * @head - head request of page group 207d72ddcbaSWeston Andros Adamson * 208d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 209d72ddcbaSWeston Andros Adamson * returns false otherwise 210d72ddcbaSWeston Andros Adamson */ 211d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 212d72ddcbaSWeston Andros Adamson { 213d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 214d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 215d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 216d72ddcbaSWeston Andros Adamson 217fd2f3a06SWeston Andros Adamson nfs_page_group_lock(req, false); 218d72ddcbaSWeston Andros Adamson 219d72ddcbaSWeston Andros Adamson do { 220d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 221d72ddcbaSWeston Andros Adamson if (tmp) { 222d72ddcbaSWeston Andros Adamson /* no way this should happen */ 223d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(tmp->wb_pgbase != pos); 224d72ddcbaSWeston Andros Adamson pos += tmp->wb_bytes - (pos - tmp->wb_pgbase); 225d72ddcbaSWeston Andros Adamson } 226d72ddcbaSWeston Andros Adamson } while (tmp && pos < len); 227d72ddcbaSWeston Andros Adamson 228d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 229d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(pos > len); 230d72ddcbaSWeston Andros Adamson return pos == len; 231d72ddcbaSWeston Andros Adamson } 232d72ddcbaSWeston Andros Adamson 2331da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 2341da177e4SLinus Torvalds * covers the full page. 2351da177e4SLinus Torvalds */ 236d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 2371da177e4SLinus Torvalds { 238d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 2391da177e4SLinus Torvalds return; 240d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 2411da177e4SLinus Torvalds return; 242d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 2431da177e4SLinus Torvalds } 2441da177e4SLinus Torvalds 2451da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 2461da177e4SLinus Torvalds { 247e87b4c7aSNeilBrown int ret = 0; 248cca588d6STrond Myklebust 249e87b4c7aSNeilBrown if (wbc->sync_mode == WB_SYNC_ALL) 250e87b4c7aSNeilBrown ret = FLUSH_COND_STABLE; 251e87b4c7aSNeilBrown return ret; 2521da177e4SLinus Torvalds } 2531da177e4SLinus Torvalds 2541da177e4SLinus Torvalds /* 25589a09141SPeter Zijlstra * NFS congestion control 25689a09141SPeter Zijlstra */ 25789a09141SPeter Zijlstra 25889a09141SPeter Zijlstra int nfs_congestion_kb; 25989a09141SPeter Zijlstra 26089a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 26189a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 26289a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 26389a09141SPeter Zijlstra 264deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 26589a09141SPeter Zijlstra { 266deed85e7STrond Myklebust struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); 2675a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2685a6d41b3STrond Myklebust 269deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 27089a09141SPeter Zijlstra 271277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 2728aa7e847SJens Axboe NFS_CONGESTION_ON_THRESH) { 2738aa7e847SJens Axboe set_bdi_congested(&nfss->backing_dev_info, 2748aa7e847SJens Axboe BLK_RW_ASYNC); 2758aa7e847SJens Axboe } 27689a09141SPeter Zijlstra } 27789a09141SPeter Zijlstra 27820633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 27989a09141SPeter Zijlstra { 28020633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 28189a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 28289a09141SPeter Zijlstra 28320633f04SWeston Andros Adamson if (!nfs_page_group_sync_on_bit(req, PG_WB_END)) 28420633f04SWeston Andros Adamson return; 28520633f04SWeston Andros Adamson 28620633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 287c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 2888aa7e847SJens Axboe clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 28989a09141SPeter Zijlstra } 29089a09141SPeter Zijlstra 291d4581383SWeston Andros Adamson 292d4581383SWeston Andros Adamson /* nfs_page_group_clear_bits 293d4581383SWeston Andros Adamson * @req - an nfs request 294d4581383SWeston Andros Adamson * clears all page group related bits from @req 295d4581383SWeston Andros Adamson */ 296d4581383SWeston Andros Adamson static void 297d4581383SWeston Andros Adamson nfs_page_group_clear_bits(struct nfs_page *req) 298e261f51fSTrond Myklebust { 299d4581383SWeston Andros Adamson clear_bit(PG_TEARDOWN, &req->wb_flags); 300d4581383SWeston Andros Adamson clear_bit(PG_UNLOCKPAGE, &req->wb_flags); 301d4581383SWeston Andros Adamson clear_bit(PG_UPTODATE, &req->wb_flags); 302d4581383SWeston Andros Adamson clear_bit(PG_WB_END, &req->wb_flags); 303d4581383SWeston Andros Adamson clear_bit(PG_REMOVE, &req->wb_flags); 304d4581383SWeston Andros Adamson } 305d4581383SWeston Andros Adamson 306d4581383SWeston Andros Adamson 307d4581383SWeston Andros Adamson /* 308d4581383SWeston Andros Adamson * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req 309d4581383SWeston Andros Adamson * 310d4581383SWeston Andros Adamson * this is a helper function for nfs_lock_and_join_requests 311d4581383SWeston Andros Adamson * 312d4581383SWeston Andros Adamson * @inode - inode associated with request page group, must be holding inode lock 313d4581383SWeston Andros Adamson * @head - head request of page group, must be holding head lock 314d4581383SWeston Andros Adamson * @req - request that couldn't lock and needs to wait on the req bit lock 315d4581383SWeston Andros Adamson * @nonblock - if true, don't actually wait 316d4581383SWeston Andros Adamson * 317d4581383SWeston Andros Adamson * NOTE: this must be called holding page_group bit lock and inode spin lock 318d4581383SWeston Andros Adamson * and BOTH will be released before returning. 319d4581383SWeston Andros Adamson * 320d4581383SWeston Andros Adamson * returns 0 on success, < 0 on error. 321d4581383SWeston Andros Adamson */ 322d4581383SWeston Andros Adamson static int 323d4581383SWeston Andros Adamson nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head, 324d4581383SWeston Andros Adamson struct nfs_page *req, bool nonblock) 325d4581383SWeston Andros Adamson __releases(&inode->i_lock) 326d4581383SWeston Andros Adamson { 327d4581383SWeston Andros Adamson struct nfs_page *tmp; 328e261f51fSTrond Myklebust int ret; 329e261f51fSTrond Myklebust 330d4581383SWeston Andros Adamson /* relinquish all the locks successfully grabbed this run */ 331d4581383SWeston Andros Adamson for (tmp = head ; tmp != req; tmp = tmp->wb_this_page) 332d4581383SWeston Andros Adamson nfs_unlock_request(tmp); 333d4581383SWeston Andros Adamson 334d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags)); 335d4581383SWeston Andros Adamson 336d4581383SWeston Andros Adamson /* grab a ref on the request that will be waited on */ 337d4581383SWeston Andros Adamson kref_get(&req->wb_kref); 338d4581383SWeston Andros Adamson 339d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 340587142f8STrond Myklebust spin_unlock(&inode->i_lock); 341d4581383SWeston Andros Adamson 342d4581383SWeston Andros Adamson /* release ref from nfs_page_find_head_request_locked */ 343d4581383SWeston Andros Adamson nfs_release_request(head); 344d4581383SWeston Andros Adamson 345cfb506e1STrond Myklebust if (!nonblock) 346e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 347cfb506e1STrond Myklebust else 348cfb506e1STrond Myklebust ret = -EAGAIN; 349e261f51fSTrond Myklebust nfs_release_request(req); 350d4581383SWeston Andros Adamson 351d4581383SWeston Andros Adamson return ret; 352e261f51fSTrond Myklebust } 353d4581383SWeston Andros Adamson 354d4581383SWeston Andros Adamson /* 355d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 356d4581383SWeston Andros Adamson * 357d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 358d4581383SWeston Andros Adamson * @old_head - the old head of the list 359d4581383SWeston Andros Adamson * 360d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 361d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 362d4581383SWeston Andros Adamson * with a reference held by some other context. 363d4581383SWeston Andros Adamson */ 364d4581383SWeston Andros Adamson static void 365d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 366d4581383SWeston Andros Adamson struct nfs_page *old_head) 367d4581383SWeston Andros Adamson { 368d4581383SWeston Andros Adamson while (destroy_list) { 369d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 370d4581383SWeston Andros Adamson 371d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 372d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 373d4581383SWeston Andros Adamson 374d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 375d4581383SWeston Andros Adamson 376d4581383SWeston Andros Adamson /* make sure old group is not used */ 377d4581383SWeston Andros Adamson subreq->wb_head = subreq; 378d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 379d4581383SWeston Andros Adamson 380d4581383SWeston Andros Adamson /* subreq is now totally disconnected from page group or any 381d4581383SWeston Andros Adamson * write / commit lists. last chance to wake any waiters */ 382d4581383SWeston Andros Adamson nfs_unlock_request(subreq); 383d4581383SWeston Andros Adamson 384d4581383SWeston Andros Adamson if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) { 385d4581383SWeston Andros Adamson /* release ref on old head request */ 386d4581383SWeston Andros Adamson nfs_release_request(old_head); 387d4581383SWeston Andros Adamson 388d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 389d4581383SWeston Andros Adamson 390d4581383SWeston Andros Adamson /* release the PG_INODE_REF reference */ 391d4581383SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) 392d4581383SWeston Andros Adamson nfs_release_request(subreq); 393d4581383SWeston Andros Adamson else 394d4581383SWeston Andros Adamson WARN_ON_ONCE(1); 395d4581383SWeston Andros Adamson } else { 396d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags)); 397d4581383SWeston Andros Adamson /* zombie requests have already released the last 398d4581383SWeston Andros Adamson * reference and were waiting on the rest of the 399d4581383SWeston Andros Adamson * group to complete. Since it's no longer part of a 400d4581383SWeston Andros Adamson * group, simply free the request */ 401d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 402d4581383SWeston Andros Adamson nfs_free_request(subreq); 403d4581383SWeston Andros Adamson } 404d4581383SWeston Andros Adamson } 405d4581383SWeston Andros Adamson } 406d4581383SWeston Andros Adamson 407d4581383SWeston Andros Adamson /* 408d4581383SWeston Andros Adamson * nfs_lock_and_join_requests - join all subreqs to the head req and return 409d4581383SWeston Andros Adamson * a locked reference, cancelling any pending 410d4581383SWeston Andros Adamson * operations for this page. 411d4581383SWeston Andros Adamson * 412d4581383SWeston Andros Adamson * @page - the page used to lookup the "page group" of nfs_page structures 413d4581383SWeston Andros Adamson * @nonblock - if true, don't block waiting for request locks 414d4581383SWeston Andros Adamson * 415d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 416d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 417d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 418d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 419d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 420d4581383SWeston Andros Adamson * 421d4581383SWeston Andros Adamson * Returns a locked, referenced pointer to the head request - which after 422d4581383SWeston Andros Adamson * this call is guaranteed to be the only request associated with the page. 423d4581383SWeston Andros Adamson * Returns NULL if no requests are found for @page, or a ERR_PTR if an 424d4581383SWeston Andros Adamson * error was encountered. 425d4581383SWeston Andros Adamson */ 426d4581383SWeston Andros Adamson static struct nfs_page * 427d4581383SWeston Andros Adamson nfs_lock_and_join_requests(struct page *page, bool nonblock) 428d4581383SWeston Andros Adamson { 429d4581383SWeston Andros Adamson struct inode *inode = page_file_mapping(page)->host; 430d4581383SWeston Andros Adamson struct nfs_page *head, *subreq; 431d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 432d4581383SWeston Andros Adamson unsigned int total_bytes; 433d4581383SWeston Andros Adamson int ret; 434d4581383SWeston Andros Adamson 435d4581383SWeston Andros Adamson try_again: 436d4581383SWeston Andros Adamson total_bytes = 0; 437d4581383SWeston Andros Adamson 438d4581383SWeston Andros Adamson WARN_ON_ONCE(destroy_list); 439d4581383SWeston Andros Adamson 440d4581383SWeston Andros Adamson spin_lock(&inode->i_lock); 441d4581383SWeston Andros Adamson 442d4581383SWeston Andros Adamson /* 443d4581383SWeston Andros Adamson * A reference is taken only on the head request which acts as a 444d4581383SWeston Andros Adamson * reference to the whole page group - the group will not be destroyed 445d4581383SWeston Andros Adamson * until the head reference is released. 446d4581383SWeston Andros Adamson */ 447d4581383SWeston Andros Adamson head = nfs_page_find_head_request_locked(NFS_I(inode), page); 448d4581383SWeston Andros Adamson 449d4581383SWeston Andros Adamson if (!head) { 450587142f8STrond Myklebust spin_unlock(&inode->i_lock); 451d4581383SWeston Andros Adamson return NULL; 452d4581383SWeston Andros Adamson } 453d4581383SWeston Andros Adamson 4547c3af975SWeston Andros Adamson /* holding inode lock, so always make a non-blocking call to try the 4557c3af975SWeston Andros Adamson * page group lock */ 456fd2f3a06SWeston Andros Adamson ret = nfs_page_group_lock(head, true); 45794970014SWeston Andros Adamson if (ret < 0) { 45894970014SWeston Andros Adamson spin_unlock(&inode->i_lock); 4597c3af975SWeston Andros Adamson 4607c3af975SWeston Andros Adamson if (!nonblock && ret == -EAGAIN) { 4617c3af975SWeston Andros Adamson nfs_page_group_lock_wait(head); 4627c3af975SWeston Andros Adamson nfs_release_request(head); 4637c3af975SWeston Andros Adamson goto try_again; 4647c3af975SWeston Andros Adamson } 4657c3af975SWeston Andros Adamson 46694970014SWeston Andros Adamson nfs_release_request(head); 467e7029206SWeston Andros Adamson return ERR_PTR(ret); 46894970014SWeston Andros Adamson } 4697c3af975SWeston Andros Adamson 4707c3af975SWeston Andros Adamson /* lock each request in the page group */ 471d4581383SWeston Andros Adamson subreq = head; 472d4581383SWeston Andros Adamson do { 473d4581383SWeston Andros Adamson /* 474d4581383SWeston Andros Adamson * Subrequests are always contiguous, non overlapping 475309a1d65SWeston Andros Adamson * and in order - but may be repeated (mirrored writes). 476d4581383SWeston Andros Adamson */ 477309a1d65SWeston Andros Adamson if (subreq->wb_offset == (head->wb_offset + total_bytes)) { 478d4581383SWeston Andros Adamson /* keep track of how many bytes this group covers */ 479d4581383SWeston Andros Adamson total_bytes += subreq->wb_bytes; 480309a1d65SWeston Andros Adamson } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || 481309a1d65SWeston Andros Adamson ((subreq->wb_offset + subreq->wb_bytes) > 482309a1d65SWeston Andros Adamson (head->wb_offset + total_bytes)))) { 483309a1d65SWeston Andros Adamson nfs_page_group_unlock(head); 484309a1d65SWeston Andros Adamson spin_unlock(&inode->i_lock); 485309a1d65SWeston Andros Adamson return ERR_PTR(-EIO); 486309a1d65SWeston Andros Adamson } 487d4581383SWeston Andros Adamson 488d4581383SWeston Andros Adamson if (!nfs_lock_request(subreq)) { 489d4581383SWeston Andros Adamson /* releases page group bit lock and 490d4581383SWeston Andros Adamson * inode spin lock and all references */ 491d4581383SWeston Andros Adamson ret = nfs_unroll_locks_and_wait(inode, head, 492d4581383SWeston Andros Adamson subreq, nonblock); 493d4581383SWeston Andros Adamson 494d4581383SWeston Andros Adamson if (ret == 0) 495d4581383SWeston Andros Adamson goto try_again; 496d4581383SWeston Andros Adamson 497d4581383SWeston Andros Adamson return ERR_PTR(ret); 498d4581383SWeston Andros Adamson } 499d4581383SWeston Andros Adamson 500d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 501d4581383SWeston Andros Adamson } while (subreq != head); 502d4581383SWeston Andros Adamson 503d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 504d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 505d4581383SWeston Andros Adamson subreq = head; 506d4581383SWeston Andros Adamson do { 507411a99adSWeston Andros Adamson nfs_clear_request_commit(subreq); 508d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 509d4581383SWeston Andros Adamson } while (subreq != head); 510d4581383SWeston Andros Adamson 511d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 512d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 513d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 514d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 515d4581383SWeston Andros Adamson head->wb_this_page = head; 516d4581383SWeston Andros Adamson 517d4581383SWeston Andros Adamson /* change head request to cover whole range that 518d4581383SWeston Andros Adamson * the former page group covered */ 519d4581383SWeston Andros Adamson head->wb_bytes = total_bytes; 520d4581383SWeston Andros Adamson } 521d4581383SWeston Andros Adamson 522d4581383SWeston Andros Adamson /* 523d4581383SWeston Andros Adamson * prepare head request to be added to new pgio descriptor 524d4581383SWeston Andros Adamson */ 525d4581383SWeston Andros Adamson nfs_page_group_clear_bits(head); 526d4581383SWeston Andros Adamson 527d4581383SWeston Andros Adamson /* 528d4581383SWeston Andros Adamson * some part of the group was still on the inode list - otherwise 529d4581383SWeston Andros Adamson * the group wouldn't be involved in async write. 530d4581383SWeston Andros Adamson * grab a reference for the head request, iff it needs one. 531d4581383SWeston Andros Adamson */ 532d4581383SWeston Andros Adamson if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags)) 533d4581383SWeston Andros Adamson kref_get(&head->wb_kref); 534d4581383SWeston Andros Adamson 535d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 536d4581383SWeston Andros Adamson 537411a99adSWeston Andros Adamson /* drop lock to clean uprequests on destroy list */ 538d4581383SWeston Andros Adamson spin_unlock(&inode->i_lock); 539d4581383SWeston Andros Adamson 540d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(destroy_list, head); 541d4581383SWeston Andros Adamson 542d4581383SWeston Andros Adamson /* still holds ref on head from nfs_page_find_head_request_locked 543d4581383SWeston Andros Adamson * and still has lock on head from lock loop */ 544d4581383SWeston Andros Adamson return head; 545612c9384STrond Myklebust } 546074cc1deSTrond Myklebust 5470bcbf039SPeng Tao static void nfs_write_error_remove_page(struct nfs_page *req) 5480bcbf039SPeng Tao { 5490bcbf039SPeng Tao nfs_unlock_request(req); 5500bcbf039SPeng Tao nfs_end_page_writeback(req); 5510bcbf039SPeng Tao nfs_release_request(req); 5520bcbf039SPeng Tao generic_error_remove_page(page_file_mapping(req->wb_page), 5530bcbf039SPeng Tao req->wb_page); 5540bcbf039SPeng Tao } 5550bcbf039SPeng Tao 556074cc1deSTrond Myklebust /* 557074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 558074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 559074cc1deSTrond Myklebust */ 560074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 561d6c843b9SPeng Tao struct page *page, bool nonblock, 562d6c843b9SPeng Tao bool launder) 563074cc1deSTrond Myklebust { 564074cc1deSTrond Myklebust struct nfs_page *req; 565074cc1deSTrond Myklebust int ret = 0; 566074cc1deSTrond Myklebust 567d4581383SWeston Andros Adamson req = nfs_lock_and_join_requests(page, nonblock); 568074cc1deSTrond Myklebust if (!req) 569074cc1deSTrond Myklebust goto out; 570074cc1deSTrond Myklebust ret = PTR_ERR(req); 571074cc1deSTrond Myklebust if (IS_ERR(req)) 572074cc1deSTrond Myklebust goto out; 573074cc1deSTrond Myklebust 574deed85e7STrond Myklebust nfs_set_page_writeback(page); 575deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 576074cc1deSTrond Myklebust 577deed85e7STrond Myklebust ret = 0; 578f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 579074cc1deSTrond Myklebust ret = pgio->pg_error; 5800bcbf039SPeng Tao /* 581d6c843b9SPeng Tao * Remove the problematic req upon fatal errors 582d6c843b9SPeng Tao * in launder case, while other dirty pages can 583d6c843b9SPeng Tao * still be around until they get flushed. 5840bcbf039SPeng Tao */ 5850bcbf039SPeng Tao if (nfs_error_is_fatal(ret)) { 5860bcbf039SPeng Tao nfs_context_set_write_error(req->wb_context, ret); 587d6c843b9SPeng Tao if (launder) { 5880bcbf039SPeng Tao nfs_write_error_remove_page(req); 589d6c843b9SPeng Tao goto out; 590d6c843b9SPeng Tao } 591d6c843b9SPeng Tao } 5920bcbf039SPeng Tao nfs_redirty_request(req); 5930bcbf039SPeng Tao ret = -EAGAIN; 59440f90271STrond Myklebust } else 59540f90271STrond Myklebust nfs_add_stats(page_file_mapping(page)->host, 59640f90271STrond Myklebust NFSIOS_WRITEPAGES, 1); 597074cc1deSTrond Myklebust out: 598074cc1deSTrond Myklebust return ret; 599e261f51fSTrond Myklebust } 600e261f51fSTrond Myklebust 601d6c843b9SPeng Tao static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, 602d6c843b9SPeng Tao struct nfs_pageio_descriptor *pgio, bool launder) 603f758c885STrond Myklebust { 604cfb506e1STrond Myklebust int ret; 605f758c885STrond Myklebust 6068cd79788SHuang Ying nfs_pageio_cond_complete(pgio, page_index(page)); 607d6c843b9SPeng Tao ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE, 608d6c843b9SPeng Tao launder); 609cfb506e1STrond Myklebust if (ret == -EAGAIN) { 610cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 611cfb506e1STrond Myklebust ret = 0; 612cfb506e1STrond Myklebust } 613cfb506e1STrond Myklebust return ret; 614f758c885STrond Myklebust } 615f758c885STrond Myklebust 616e261f51fSTrond Myklebust /* 6171da177e4SLinus Torvalds * Write an mmapped page to the server. 6181da177e4SLinus Torvalds */ 619d6c843b9SPeng Tao static int nfs_writepage_locked(struct page *page, 620d6c843b9SPeng Tao struct writeback_control *wbc, 621d6c843b9SPeng Tao bool launder) 6221da177e4SLinus Torvalds { 623f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 62440f90271STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 625e261f51fSTrond Myklebust int err; 6261da177e4SLinus Torvalds 62740f90271STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 628811ed92eSTrond Myklebust nfs_pageio_init_write(&pgio, inode, 0, 629a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 630d6c843b9SPeng Tao err = nfs_do_writepage(page, wbc, &pgio, launder); 631f758c885STrond Myklebust nfs_pageio_complete(&pgio); 632f758c885STrond Myklebust if (err < 0) 6334d770ccfSTrond Myklebust return err; 634f758c885STrond Myklebust if (pgio.pg_error < 0) 635f758c885STrond Myklebust return pgio.pg_error; 636f758c885STrond Myklebust return 0; 6374d770ccfSTrond Myklebust } 6384d770ccfSTrond Myklebust 6394d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 6404d770ccfSTrond Myklebust { 641f758c885STrond Myklebust int ret; 6424d770ccfSTrond Myklebust 643d6c843b9SPeng Tao ret = nfs_writepage_locked(page, wbc, false); 6441da177e4SLinus Torvalds unlock_page(page); 645f758c885STrond Myklebust return ret; 646f758c885STrond Myklebust } 647f758c885STrond Myklebust 648f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 649f758c885STrond Myklebust { 650f758c885STrond Myklebust int ret; 651f758c885STrond Myklebust 652d6c843b9SPeng Tao ret = nfs_do_writepage(page, wbc, data, false); 653f758c885STrond Myklebust unlock_page(page); 654f758c885STrond Myklebust return ret; 6551da177e4SLinus Torvalds } 6561da177e4SLinus Torvalds 6571da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 6581da177e4SLinus Torvalds { 6591da177e4SLinus Torvalds struct inode *inode = mapping->host; 660c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 6611da177e4SLinus Torvalds int err; 6621da177e4SLinus Torvalds 66391d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 66491d5b470SChuck Lever 665a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 666a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 667f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 668c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 66972cb77f4STrond Myklebust 670f758c885STrond Myklebust if (err < 0) 67172cb77f4STrond Myklebust goto out_err; 67272cb77f4STrond Myklebust err = pgio.pg_error; 67372cb77f4STrond Myklebust if (err < 0) 67472cb77f4STrond Myklebust goto out_err; 675c63c7b05STrond Myklebust return 0; 67672cb77f4STrond Myklebust out_err: 67772cb77f4STrond Myklebust return err; 6781da177e4SLinus Torvalds } 6791da177e4SLinus Torvalds 6801da177e4SLinus Torvalds /* 6811da177e4SLinus Torvalds * Insert a write request into an inode 6821da177e4SLinus Torvalds */ 683d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 6841da177e4SLinus Torvalds { 6851da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 686e7d39069STrond Myklebust 6872bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 6882bfc6e56SWeston Andros Adamson 689e7d39069STrond Myklebust /* Lock the request! */ 6907ad84aa9STrond Myklebust nfs_lock_request(req); 691e7d39069STrond Myklebust 692e7d39069STrond Myklebust spin_lock(&inode->i_lock); 693cb1410c7SWeston Andros Adamson if (!nfsi->nrequests && 694cb1410c7SWeston Andros Adamson NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 695a9a4a87aSTrond Myklebust inode->i_version++; 69629418aa4SMel Gorman /* 69729418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 69829418aa4SMel Gorman * with invalidate/truncate. 69929418aa4SMel Gorman */ 70029418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 7012df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 702deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 703277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 70429418aa4SMel Gorman } 705cb1410c7SWeston Andros Adamson nfsi->nrequests++; 70617089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 707cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 708cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 709cb1410c7SWeston Andros Adamson * adding subrequests. */ 71017089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 711c03b4024STrond Myklebust kref_get(&req->wb_kref); 712e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 7131da177e4SLinus Torvalds } 7141da177e4SLinus Torvalds 7151da177e4SLinus Torvalds /* 71689a09141SPeter Zijlstra * Remove a write request from an inode 7171da177e4SLinus Torvalds */ 7181da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7191da177e4SLinus Torvalds { 7202b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 7211da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 72220633f04SWeston Andros Adamson struct nfs_page *head; 72320633f04SWeston Andros Adamson 72420633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 72520633f04SWeston Andros Adamson head = req->wb_head; 7261da177e4SLinus Torvalds 727587142f8STrond Myklebust spin_lock(&inode->i_lock); 72867911c8fSAnna Schumaker if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { 72920633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 73020633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 73195905446SNeilBrown smp_mb__after_atomic(); 73295905446SNeilBrown wake_up_page(head->wb_page, PG_private); 73320633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 73429418aa4SMel Gorman } 735cb1410c7SWeston Andros Adamson nfsi->nrequests--; 736cb1410c7SWeston Andros Adamson spin_unlock(&inode->i_lock); 737cb1410c7SWeston Andros Adamson } else { 738cb1410c7SWeston Andros Adamson spin_lock(&inode->i_lock); 739cb1410c7SWeston Andros Adamson nfsi->nrequests--; 740587142f8STrond Myklebust spin_unlock(&inode->i_lock); 74120633f04SWeston Andros Adamson } 74217089a29SWeston Andros Adamson 74317089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 7441da177e4SLinus Torvalds nfs_release_request(req); 7451da177e4SLinus Torvalds } 7461da177e4SLinus Torvalds 74761822ab5STrond Myklebust static void 7486d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 74961822ab5STrond Myklebust { 75067911c8fSAnna Schumaker if (req->wb_page) 75161822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 75261822ab5STrond Myklebust } 75361822ab5STrond Myklebust 7543a3908c8STrond Myklebust /* 7553a3908c8STrond Myklebust * nfs_page_search_commits_for_head_request_locked 7563a3908c8STrond Myklebust * 7573a3908c8STrond Myklebust * Search through commit lists on @inode for the head request for @page. 7583a3908c8STrond Myklebust * Must be called while holding the inode (which is cinfo) lock. 7593a3908c8STrond Myklebust * 7603a3908c8STrond Myklebust * Returns the head request if found, or NULL if not found. 7613a3908c8STrond Myklebust */ 7623a3908c8STrond Myklebust static struct nfs_page * 7633a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 7643a3908c8STrond Myklebust struct page *page) 7653a3908c8STrond Myklebust { 7663a3908c8STrond Myklebust struct nfs_page *freq, *t; 7673a3908c8STrond Myklebust struct nfs_commit_info cinfo; 7683a3908c8STrond Myklebust struct inode *inode = &nfsi->vfs_inode; 7693a3908c8STrond Myklebust 7703a3908c8STrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 7713a3908c8STrond Myklebust 7723a3908c8STrond Myklebust /* search through pnfs commit lists */ 7733a3908c8STrond Myklebust freq = pnfs_search_commit_reqs(inode, &cinfo, page); 7743a3908c8STrond Myklebust if (freq) 7753a3908c8STrond Myklebust return freq->wb_head; 7763a3908c8STrond Myklebust 7773a3908c8STrond Myklebust /* Linearly search the commit list for the correct request */ 7783a3908c8STrond Myklebust list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 7793a3908c8STrond Myklebust if (freq->wb_page == page) 7803a3908c8STrond Myklebust return freq->wb_head; 7813a3908c8STrond Myklebust } 7823a3908c8STrond Myklebust 7833a3908c8STrond Myklebust return NULL; 7843a3908c8STrond Myklebust } 7853a3908c8STrond Myklebust 7868dd37758STrond Myklebust /** 78786d80f97STrond Myklebust * nfs_request_add_commit_list_locked - add request to a commit list 78886d80f97STrond Myklebust * @req: pointer to a struct nfs_page 78986d80f97STrond Myklebust * @dst: commit list head 79086d80f97STrond Myklebust * @cinfo: holds list lock and accounting info 79186d80f97STrond Myklebust * 79286d80f97STrond Myklebust * This sets the PG_CLEAN bit, updates the cinfo count of 79386d80f97STrond Myklebust * number of outstanding requests requiring a commit as well as 79486d80f97STrond Myklebust * the MM page stats. 79586d80f97STrond Myklebust * 796fe238e60SDave Wysochanski * The caller must hold cinfo->inode->i_lock, and the nfs_page lock. 79786d80f97STrond Myklebust */ 79886d80f97STrond Myklebust void 79986d80f97STrond Myklebust nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, 80086d80f97STrond Myklebust struct nfs_commit_info *cinfo) 80186d80f97STrond Myklebust { 80286d80f97STrond Myklebust set_bit(PG_CLEAN, &req->wb_flags); 80386d80f97STrond Myklebust nfs_list_add_request(req, dst); 80486d80f97STrond Myklebust cinfo->mds->ncommit++; 80586d80f97STrond Myklebust } 80686d80f97STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); 80786d80f97STrond Myklebust 80886d80f97STrond Myklebust /** 8098dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 8108dd37758STrond Myklebust * @req: pointer to a struct nfs_page 811ea2cf228SFred Isaman * @dst: commit list head 812ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8138dd37758STrond Myklebust * 814ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 8158dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8168dd37758STrond Myklebust * the MM page stats. 8178dd37758STrond Myklebust * 818ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8198dd37758STrond Myklebust * holding the nfs_page lock. 8208dd37758STrond Myklebust */ 8218dd37758STrond Myklebust void 8226272dcc6SAnna Schumaker nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) 8238dd37758STrond Myklebust { 824fe238e60SDave Wysochanski spin_lock(&cinfo->inode->i_lock); 8256272dcc6SAnna Schumaker nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); 826fe238e60SDave Wysochanski spin_unlock(&cinfo->inode->i_lock); 82767911c8fSAnna Schumaker if (req->wb_page) 82886d80f97STrond Myklebust nfs_mark_page_unstable(req->wb_page, cinfo); 8298dd37758STrond Myklebust } 8308dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 8318dd37758STrond Myklebust 8328dd37758STrond Myklebust /** 8338dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 8348dd37758STrond Myklebust * @req: pointer to a nfs_page 835ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8368dd37758STrond Myklebust * 837ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 8388dd37758STrond Myklebust * number of outstanding requests requiring a commit 8398dd37758STrond Myklebust * It does not update the MM page stats. 8408dd37758STrond Myklebust * 841ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 8428dd37758STrond Myklebust */ 8438dd37758STrond Myklebust void 844ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 845ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 8468dd37758STrond Myklebust { 8478dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 8488dd37758STrond Myklebust return; 8498dd37758STrond Myklebust nfs_list_remove_request(req); 850ea2cf228SFred Isaman cinfo->mds->ncommit--; 8518dd37758STrond Myklebust } 8528dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 8538dd37758STrond Myklebust 854ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 855ea2cf228SFred Isaman struct inode *inode) 856ea2cf228SFred Isaman { 857fe238e60SDave Wysochanski cinfo->inode = inode; 858ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 859ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 860b359f9d0SFred Isaman cinfo->dreq = NULL; 861f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 862ea2cf228SFred Isaman } 863ea2cf228SFred Isaman 864ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 865ea2cf228SFred Isaman struct inode *inode, 866ea2cf228SFred Isaman struct nfs_direct_req *dreq) 867ea2cf228SFred Isaman { 8681763da12SFred Isaman if (dreq) 8691763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 8701763da12SFred Isaman else 871ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 872ea2cf228SFred Isaman } 873ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 8748dd37758STrond Myklebust 8751da177e4SLinus Torvalds /* 8761da177e4SLinus Torvalds * Add a request to the inode's commit list. 8771da177e4SLinus Torvalds */ 8781763da12SFred Isaman void 879ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 880b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 8811da177e4SLinus Torvalds { 882b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 8838dd37758STrond Myklebust return; 8846272dcc6SAnna Schumaker nfs_request_add_commit_list(req, cinfo); 8851da177e4SLinus Torvalds } 8868e821cadSTrond Myklebust 887d6d6dc7cSFred Isaman static void 888d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 889e468bae9STrond Myklebust { 89011fb9989SMel Gorman dec_node_page_state(page, NR_UNSTABLE_NFS); 89193f78d88STejun Heo dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, 89293f78d88STejun Heo WB_RECLAIMABLE); 893e468bae9STrond Myklebust } 894d6d6dc7cSFred Isaman 895411a99adSWeston Andros Adamson /* Called holding inode (/cinfo) lock */ 8968dd37758STrond Myklebust static void 897d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 898d6d6dc7cSFred Isaman { 8998dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 9002b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 901ea2cf228SFred Isaman struct nfs_commit_info cinfo; 902d6d6dc7cSFred Isaman 903ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 904ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 905ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 906d6d6dc7cSFred Isaman } 9078dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 9088dd37758STrond Myklebust } 909e468bae9STrond Myklebust } 910e468bae9STrond Myklebust 911d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 9128e821cadSTrond Myklebust { 913c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 914d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 915c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 9168e821cadSTrond Myklebust } 9178e821cadSTrond Myklebust 918061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 9196c75dc0dSFred Isaman { 920ea2cf228SFred Isaman struct nfs_commit_info cinfo; 9216c75dc0dSFred Isaman unsigned long bytes = 0; 9226c75dc0dSFred Isaman 9236c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 9246c75dc0dSFred Isaman goto out; 925ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 9266c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 9276c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 9286c75dc0dSFred Isaman 9296c75dc0dSFred Isaman bytes += req->wb_bytes; 9306c75dc0dSFred Isaman nfs_list_remove_request(req); 9316c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 9326c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 933d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 9346c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 9356c75dc0dSFred Isaman goto remove_req; 9366c75dc0dSFred Isaman } 937c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 938f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 939b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 940a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 9416c75dc0dSFred Isaman goto next; 9426c75dc0dSFred Isaman } 9436c75dc0dSFred Isaman remove_req: 9446c75dc0dSFred Isaman nfs_inode_remove_request(req); 9456c75dc0dSFred Isaman next: 9461d1afcbcSTrond Myklebust nfs_unlock_request(req); 94720633f04SWeston Andros Adamson nfs_end_page_writeback(req); 9483aff4ebbSTrond Myklebust nfs_release_request(req); 9496c75dc0dSFred Isaman } 9506c75dc0dSFred Isaman out: 9516c75dc0dSFred Isaman hdr->release(hdr); 9526c75dc0dSFred Isaman } 9536c75dc0dSFred Isaman 954ce59515cSAnna Schumaker unsigned long 955ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 956fb8a1f11STrond Myklebust { 957ea2cf228SFred Isaman return cinfo->mds->ncommit; 958fb8a1f11STrond Myklebust } 959fb8a1f11STrond Myklebust 960fe238e60SDave Wysochanski /* cinfo->inode->i_lock held by caller */ 9611763da12SFred Isaman int 962ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 963ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 964d6d6dc7cSFred Isaman { 965d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 966d6d6dc7cSFred Isaman int ret = 0; 967d6d6dc7cSFred Isaman 968d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 9698dd37758STrond Myklebust if (!nfs_lock_request(req)) 9708dd37758STrond Myklebust continue; 9717ad84aa9STrond Myklebust kref_get(&req->wb_kref); 972fe238e60SDave Wysochanski if (cond_resched_lock(&cinfo->inode->i_lock)) 9733b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 974ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 9758dd37758STrond Myklebust nfs_list_add_request(req, dst); 976d6d6dc7cSFred Isaman ret++; 9771763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 978d6d6dc7cSFred Isaman break; 979d6d6dc7cSFred Isaman } 980d6d6dc7cSFred Isaman return ret; 981d6d6dc7cSFred Isaman } 982d6d6dc7cSFred Isaman 9831da177e4SLinus Torvalds /* 9841da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 9851da177e4SLinus Torvalds * @inode: NFS inode to scan 986ea2cf228SFred Isaman * @dst: mds destination list 987ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 9881da177e4SLinus Torvalds * 9891da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 9901da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 9911da177e4SLinus Torvalds */ 9921763da12SFred Isaman int 993ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 994ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9951da177e4SLinus Torvalds { 996d6d6dc7cSFred Isaman int ret = 0; 997fb8a1f11STrond Myklebust 998fe238e60SDave Wysochanski spin_lock(&cinfo->inode->i_lock); 999ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 10008dd37758STrond Myklebust const int max = INT_MAX; 1001d6d6dc7cSFred Isaman 1002ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 1003ea2cf228SFred Isaman cinfo, max); 1004ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 1005d6d6dc7cSFred Isaman } 1006fe238e60SDave Wysochanski spin_unlock(&cinfo->inode->i_lock); 1007ff778d02STrond Myklebust return ret; 10081da177e4SLinus Torvalds } 1009d6d6dc7cSFred Isaman 10101da177e4SLinus Torvalds /* 1011e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1012e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 10131da177e4SLinus Torvalds * 1014e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1015e7d39069STrond Myklebust * to disk. 10161da177e4SLinus Torvalds */ 1017e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 1018e7d39069STrond Myklebust struct page *page, 1019e7d39069STrond Myklebust unsigned int offset, 1020e7d39069STrond Myklebust unsigned int bytes) 10211da177e4SLinus Torvalds { 1022e7d39069STrond Myklebust struct nfs_page *req; 1023e7d39069STrond Myklebust unsigned int rqend; 1024e7d39069STrond Myklebust unsigned int end; 10251da177e4SLinus Torvalds int error; 1026277459d2STrond Myklebust 1027e7d39069STrond Myklebust if (!PagePrivate(page)) 1028e7d39069STrond Myklebust return NULL; 1029e7d39069STrond Myklebust 1030e7d39069STrond Myklebust end = offset + bytes; 1031e7d39069STrond Myklebust spin_lock(&inode->i_lock); 1032e7d39069STrond Myklebust 1033e7d39069STrond Myklebust for (;;) { 103484d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 1035e7d39069STrond Myklebust if (req == NULL) 1036e7d39069STrond Myklebust goto out_unlock; 1037e7d39069STrond Myklebust 10382bfc6e56SWeston Andros Adamson /* should be handled by nfs_flush_incompatible */ 10392bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 10402bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 10412bfc6e56SWeston Andros Adamson 1042e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1043e7d39069STrond Myklebust /* 1044e7d39069STrond Myklebust * Tell the caller to flush out the request if 1045e7d39069STrond Myklebust * the offsets are non-contiguous. 1046e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1047e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1048e7d39069STrond Myklebust */ 1049e468bae9STrond Myklebust if (offset > rqend 1050e7d39069STrond Myklebust || end < req->wb_offset) 1051e7d39069STrond Myklebust goto out_flushme; 1052e7d39069STrond Myklebust 10537ad84aa9STrond Myklebust if (nfs_lock_request(req)) 1054e7d39069STrond Myklebust break; 1055e7d39069STrond Myklebust 1056e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 1057587142f8STrond Myklebust spin_unlock(&inode->i_lock); 10581da177e4SLinus Torvalds error = nfs_wait_on_request(req); 10591da177e4SLinus Torvalds nfs_release_request(req); 1060e7d39069STrond Myklebust if (error != 0) 1061e7d39069STrond Myklebust goto out_err; 1062e7d39069STrond Myklebust spin_lock(&inode->i_lock); 10631da177e4SLinus Torvalds } 10641da177e4SLinus Torvalds 10651da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 10661da177e4SLinus Torvalds if (offset < req->wb_offset) { 10671da177e4SLinus Torvalds req->wb_offset = offset; 10681da177e4SLinus Torvalds req->wb_pgbase = offset; 10691da177e4SLinus Torvalds } 10701da177e4SLinus Torvalds if (end > rqend) 10711da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1072e7d39069STrond Myklebust else 1073e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 1074e7d39069STrond Myklebust out_unlock: 1075ca138f36SFred Isaman if (req) 10768dd37758STrond Myklebust nfs_clear_request_commit(req); 1077411a99adSWeston Andros Adamson spin_unlock(&inode->i_lock); 1078e7d39069STrond Myklebust return req; 1079e7d39069STrond Myklebust out_flushme: 1080e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 1081e7d39069STrond Myklebust nfs_release_request(req); 1082e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1083e7d39069STrond Myklebust out_err: 1084e7d39069STrond Myklebust return ERR_PTR(error); 1085e7d39069STrond Myklebust } 10861da177e4SLinus Torvalds 1087e7d39069STrond Myklebust /* 1088e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1089e7d39069STrond Myklebust * 1090e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1091e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1092e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1093e7d39069STrond Myklebust */ 1094e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1095e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1096e7d39069STrond Myklebust { 1097d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1098e7d39069STrond Myklebust struct nfs_page *req; 1099e7d39069STrond Myklebust 1100e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1101e7d39069STrond Myklebust if (req != NULL) 1102e7d39069STrond Myklebust goto out; 11032bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 1104e7d39069STrond Myklebust if (IS_ERR(req)) 1105e7d39069STrond Myklebust goto out; 1106d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1107efc91ed0STrond Myklebust out: 110861e930a9STrond Myklebust return req; 11091da177e4SLinus Torvalds } 11101da177e4SLinus Torvalds 1111e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1112e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1113e7d39069STrond Myklebust { 1114e7d39069STrond Myklebust struct nfs_page *req; 1115e7d39069STrond Myklebust 1116e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1117e7d39069STrond Myklebust if (IS_ERR(req)) 1118e7d39069STrond Myklebust return PTR_ERR(req); 1119e7d39069STrond Myklebust /* Update file length */ 1120e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1121d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1122a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11231d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1124e7d39069STrond Myklebust return 0; 1125e7d39069STrond Myklebust } 1126e7d39069STrond Myklebust 11271da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 11281da177e4SLinus Torvalds { 1129cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 11302a369153STrond Myklebust struct nfs_lock_context *l_ctx; 1131bd61e0a9SJeff Layton struct file_lock_context *flctx = file_inode(file)->i_flctx; 11321da177e4SLinus Torvalds struct nfs_page *req; 11331a54533eSTrond Myklebust int do_flush, status; 11341da177e4SLinus Torvalds /* 11351da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 11361da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 11371da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 11381da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11391da177e4SLinus Torvalds * Also do the same if we find a request from an existing 11401da177e4SLinus Torvalds * dropped page. 11411da177e4SLinus Torvalds */ 11421a54533eSTrond Myklebust do { 114384d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 11441a54533eSTrond Myklebust if (req == NULL) 11451a54533eSTrond Myklebust return 0; 11462a369153STrond Myklebust l_ctx = req->wb_lock_context; 1147138a2935STrond Myklebust do_flush = req->wb_page != page || 1148138a2935STrond Myklebust !nfs_match_open_context(req->wb_context, ctx); 11492bfc6e56SWeston Andros Adamson /* for now, flush if more than 1 request in page_group */ 11502bfc6e56SWeston Andros Adamson do_flush |= req->wb_this_page != req; 1151bd61e0a9SJeff Layton if (l_ctx && flctx && 1152bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1153bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 1154d51fdb87SNeilBrown do_flush |= l_ctx->lockowner != current->files; 11555263e31eSJeff Layton } 11561da177e4SLinus Torvalds nfs_release_request(req); 11571a54533eSTrond Myklebust if (!do_flush) 11581a54533eSTrond Myklebust return 0; 1159d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 11601a54533eSTrond Myklebust } while (status == 0); 11611a54533eSTrond Myklebust return status; 11621da177e4SLinus Torvalds } 11631da177e4SLinus Torvalds 11641da177e4SLinus Torvalds /* 1165dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1166dc24826bSAndy Adamson * expire soon. 1167dc24826bSAndy Adamson * 1168dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1169dc24826bSAndy Adamson * 1170dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1171dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1172dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1173dc24826bSAndy Adamson */ 1174dc24826bSAndy Adamson int 1175dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1176dc24826bSAndy Adamson { 1177dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1178dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1179dc24826bSAndy Adamson 1180dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 1181dc24826bSAndy Adamson } 1182dc24826bSAndy Adamson 1183dc24826bSAndy Adamson /* 1184dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1185dc24826bSAndy Adamson */ 1186ce52914eSScott Mayhew bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) 1187dc24826bSAndy Adamson { 1188ce52914eSScott Mayhew struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1189ce52914eSScott Mayhew 1190ce52914eSScott Mayhew return rpcauth_cred_key_to_expire(auth, ctx->cred); 1191dc24826bSAndy Adamson } 1192dc24826bSAndy Adamson 1193dc24826bSAndy Adamson /* 11945d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 11955d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 11965d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 11975d47a356STrond Myklebust */ 11988d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 11995d47a356STrond Myklebust { 1200d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1201d529ef83SJeff Layton 12028d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 12038d197a56STrond Myklebust goto out; 120418dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1205d529ef83SJeff Layton return false; 12064db72b40SJeff Layton smp_rmb(); 1207d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 12088d197a56STrond Myklebust return false; 12098d197a56STrond Myklebust out: 121018dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 121118dd78c4SScott Mayhew return false; 12128d197a56STrond Myklebust return PageUptodate(page) != 0; 12135d47a356STrond Myklebust } 12145d47a356STrond Myklebust 12155263e31eSJeff Layton static bool 12165263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 12175263e31eSJeff Layton { 12185263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 12195263e31eSJeff Layton fl->fl_type == F_WRLCK; 12205263e31eSJeff Layton } 12215263e31eSJeff Layton 1222c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1223c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1224c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1225c7559663SScott Mayhew * inefficiencies. 1226c7559663SScott Mayhew * 1227263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1228263b4509SScott Mayhew * of the checks. 1229c7559663SScott Mayhew */ 1230c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1231c7559663SScott Mayhew { 12325263e31eSJeff Layton int ret; 12335263e31eSJeff Layton struct file_lock_context *flctx = inode->i_flctx; 12345263e31eSJeff Layton struct file_lock *fl; 12355263e31eSJeff Layton 1236c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1237c7559663SScott Mayhew return 0; 1238263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1239263b4509SScott Mayhew return 0; 1240c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1241c7559663SScott Mayhew return 1; 1242bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1243bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 12448fa4592aSTrond Myklebust return 1; 12455263e31eSJeff Layton 12465263e31eSJeff Layton /* Check to see if there are whole file write locks */ 12475263e31eSJeff Layton ret = 0; 12486109c850SJeff Layton spin_lock(&flctx->flc_lock); 1249bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1250bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1251bd61e0a9SJeff Layton fl_list); 1252bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 12535263e31eSJeff Layton ret = 1; 1254bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 12555263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 12565263e31eSJeff Layton fl_list); 12575263e31eSJeff Layton if (fl->fl_type == F_WRLCK) 12585263e31eSJeff Layton ret = 1; 12595263e31eSJeff Layton } 12606109c850SJeff Layton spin_unlock(&flctx->flc_lock); 12615263e31eSJeff Layton return ret; 1262c7559663SScott Mayhew } 1263c7559663SScott Mayhew 12645d47a356STrond Myklebust /* 12651da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 12661da177e4SLinus Torvalds * 12671da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 12681da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 12691da177e4SLinus Torvalds */ 12701da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 12711da177e4SLinus Torvalds unsigned int offset, unsigned int count) 12721da177e4SLinus Torvalds { 1273cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1274d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 12751da177e4SLinus Torvalds int status = 0; 12761da177e4SLinus Torvalds 127791d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 127891d5b470SChuck Lever 12796de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 12806de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 12811da177e4SLinus Torvalds 1282149a4fddSBenjamin Coddington if (!count) 1283149a4fddSBenjamin Coddington goto out; 1284149a4fddSBenjamin Coddington 1285c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 128649a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 12871da177e4SLinus Torvalds offset = 0; 12881da177e4SLinus Torvalds } 12891da177e4SLinus Torvalds 1290e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 129103fa9e84STrond Myklebust if (status < 0) 129203fa9e84STrond Myklebust nfs_set_pageerror(page); 129359b7c05fSTrond Myklebust else 129459b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 1295149a4fddSBenjamin Coddington out: 129648186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 12971da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 12981da177e4SLinus Torvalds return status; 12991da177e4SLinus Torvalds } 13001da177e4SLinus Torvalds 13013ff7576dSTrond Myklebust static int flush_task_priority(int how) 13021da177e4SLinus Torvalds { 13031da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 13041da177e4SLinus Torvalds case FLUSH_HIGHPRI: 13051da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 13061da177e4SLinus Torvalds case FLUSH_LOWPRI: 13071da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 13081da177e4SLinus Torvalds } 13091da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 13101da177e4SLinus Torvalds } 13111da177e4SLinus Torvalds 1312d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1313d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1314abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 13151ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 13161da177e4SLinus Torvalds { 13173ff7576dSTrond Myklebust int priority = flush_task_priority(how); 13181da177e4SLinus Torvalds 13191ed26f33SAnna Schumaker task_setup_data->priority = priority; 1320abde71f4STom Haynes rpc_ops->write_setup(hdr, msg); 1321d138d5d1SAndy Adamson 1322abde71f4STom Haynes nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client, 1323d45f60c6SWeston Andros Adamson &task_setup_data->rpc_client, msg, hdr); 1324275acaafSTrond Myklebust } 1325275acaafSTrond Myklebust 13266d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 13276d884e8fSFred * call this on each, which will prepare them to be retried on next 13286d884e8fSFred * writeback using standard nfs. 13296d884e8fSFred */ 13306d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 13316d884e8fSFred { 13326d884e8fSFred nfs_mark_request_dirty(req); 1333c7070113STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 13341d1afcbcSTrond Myklebust nfs_unlock_request(req); 133520633f04SWeston Andros Adamson nfs_end_page_writeback(req); 13363aff4ebbSTrond Myklebust nfs_release_request(req); 13376d884e8fSFred } 13386d884e8fSFred 1339061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 13406c75dc0dSFred Isaman { 13416c75dc0dSFred Isaman struct nfs_page *req; 13426c75dc0dSFred Isaman 13436c75dc0dSFred Isaman while (!list_empty(head)) { 13446c75dc0dSFred Isaman req = nfs_list_entry(head->next); 13456c75dc0dSFred Isaman nfs_list_remove_request(req); 13466c75dc0dSFred Isaman nfs_redirty_request(req); 13476c75dc0dSFred Isaman } 13486c75dc0dSFred Isaman } 13496c75dc0dSFred Isaman 1350dc602dd7STrond Myklebust static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) 1351dc602dd7STrond Myklebust { 1352dc602dd7STrond Myklebust nfs_async_write_error(&hdr->pages); 1353dc602dd7STrond Myklebust } 1354dc602dd7STrond Myklebust 1355061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1356061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1357061ae2edSFred Isaman .completion = nfs_write_completion, 1358dc602dd7STrond Myklebust .reschedule_io = nfs_async_write_reschedule_io, 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 1559af7cf057STrond Myklebust static int wait_on_commit(struct nfs_mds_commit_info *cinfo) 156071d0a611STrond Myklebust { 1561af7cf057STrond Myklebust return wait_on_atomic_t(&cinfo->rpcs_out, 1562af7cf057STrond Myklebust nfs_wait_atomic_killable, TASK_KILLABLE); 1563af7cf057STrond Myklebust } 1564af7cf057STrond Myklebust 1565af7cf057STrond Myklebust static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) 1566af7cf057STrond Myklebust { 1567af7cf057STrond Myklebust atomic_inc(&cinfo->rpcs_out); 1568af7cf057STrond Myklebust } 1569af7cf057STrond Myklebust 1570af7cf057STrond Myklebust static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) 1571af7cf057STrond Myklebust { 1572af7cf057STrond Myklebust if (atomic_dec_and_test(&cinfo->rpcs_out)) 1573af7cf057STrond Myklebust wake_up_atomic_t(&cinfo->rpcs_out); 157471d0a611STrond Myklebust } 157571d0a611STrond Myklebust 15760b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 15771da177e4SLinus Torvalds { 15780b7c0153SFred Isaman put_nfs_open_context(data->context); 15790b7c0153SFred Isaman nfs_commit_free(data); 15801da177e4SLinus Torvalds } 1581e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 15821da177e4SLinus Torvalds 15830b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1584c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 15859ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 15869f0ec176SAndy Adamson int how, int flags) 15871da177e4SLinus Torvalds { 158807737691STrond Myklebust struct rpc_task *task; 15899ace33cdSFred Isaman int priority = flush_task_priority(how); 1590bdc7f021STrond Myklebust struct rpc_message msg = { 1591bdc7f021STrond Myklebust .rpc_argp = &data->args, 1592bdc7f021STrond Myklebust .rpc_resp = &data->res, 15939ace33cdSFred Isaman .rpc_cred = data->cred, 1594bdc7f021STrond Myklebust }; 159584115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 159607737691STrond Myklebust .task = &data->task, 15979ace33cdSFred Isaman .rpc_client = clnt, 1598bdc7f021STrond Myklebust .rpc_message = &msg, 15999ace33cdSFred Isaman .callback_ops = call_ops, 160084115e1cSTrond Myklebust .callback_data = data, 1601101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 16029f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 16033ff7576dSTrond Myklebust .priority = priority, 160484115e1cSTrond Myklebust }; 1605788e7a89STrond Myklebust /* Set up the initial task struct. */ 1606c36aae9aSPeng Tao nfs_ops->commit_setup(data, &msg); 16071da177e4SLinus Torvalds 1608b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1609bdc7f021STrond Myklebust 16108c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 16118c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 16128c21c62cSWeston Andros Adamson 161307737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1614dbae4c73STrond Myklebust if (IS_ERR(task)) 1615dbae4c73STrond Myklebust return PTR_ERR(task); 1616d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1617d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 161807737691STrond Myklebust rpc_put_task(task); 1619dbae4c73STrond Myklebust return 0; 16201da177e4SLinus Torvalds } 1621e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 16221da177e4SLinus Torvalds 1623378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1624378520b8SPeng Tao { 1625378520b8SPeng Tao loff_t lwb = 0; 1626378520b8SPeng Tao struct nfs_page *req; 1627378520b8SPeng Tao 1628378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1629378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1630378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1631378520b8SPeng Tao 1632378520b8SPeng Tao return lwb; 1633378520b8SPeng Tao } 1634378520b8SPeng Tao 16351da177e4SLinus Torvalds /* 16369ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 16379ace33cdSFred Isaman */ 16380b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1639988b6dceSFred Isaman struct list_head *head, 1640f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1641f453a54aSFred Isaman struct nfs_commit_info *cinfo) 16429ace33cdSFred Isaman { 16439ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 16442b0143b5SDavid Howells struct inode *inode = d_inode(first->wb_context->dentry); 16459ace33cdSFred Isaman 16469ace33cdSFred Isaman /* Set up the RPC argument and reply structs 16479ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 16489ace33cdSFred Isaman 16499ace33cdSFred Isaman list_splice_init(head, &data->pages); 16509ace33cdSFred Isaman 16519ace33cdSFred Isaman data->inode = inode; 16529ace33cdSFred Isaman data->cred = first->wb_context->cred; 1653988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1654378520b8SPeng Tao /* only set lwb for pnfs commit */ 1655378520b8SPeng Tao if (lseg) 1656378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 16579ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1658f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1659b359f9d0SFred Isaman data->dreq = cinfo->dreq; 16609ace33cdSFred Isaman 16619ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 16629ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 16639ace33cdSFred Isaman data->args.offset = 0; 16649ace33cdSFred Isaman data->args.count = 0; 16650b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 16669ace33cdSFred Isaman data->res.fattr = &data->fattr; 16679ace33cdSFred Isaman data->res.verf = &data->verf; 16689ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 16699ace33cdSFred Isaman } 1670e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 16719ace33cdSFred Isaman 1672e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1673ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1674b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1675b57ff130SWeston Andros Adamson u32 ds_commit_idx) 167664bfeb49SFred Isaman { 167764bfeb49SFred Isaman struct nfs_page *req; 167864bfeb49SFred Isaman 167964bfeb49SFred Isaman while (!list_empty(page_list)) { 168064bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 168164bfeb49SFred Isaman nfs_list_remove_request(req); 1682b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 1683487b9b8aSTom Haynes if (!cinfo->dreq) 1684487b9b8aSTom Haynes nfs_clear_page_commit(req->wb_page); 16851d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 168664bfeb49SFred Isaman } 168764bfeb49SFred Isaman } 1688e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 168964bfeb49SFred Isaman 1690b20135d0STrond Myklebust static void 1691b20135d0STrond Myklebust nfs_commit_resched_write(struct nfs_commit_info *cinfo, 1692b20135d0STrond Myklebust struct nfs_page *req) 1693b20135d0STrond Myklebust { 1694b20135d0STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 1695b20135d0STrond Myklebust } 1696b20135d0STrond Myklebust 16979ace33cdSFred Isaman /* 16981da177e4SLinus Torvalds * Commit dirty pages 16991da177e4SLinus Torvalds */ 17001da177e4SLinus Torvalds static int 1701ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1702ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 17031da177e4SLinus Torvalds { 17040b7c0153SFred Isaman struct nfs_commit_data *data; 17051da177e4SLinus Torvalds 1706ade8febdSWeston Andros Adamson /* another commit raced with us */ 1707ade8febdSWeston Andros Adamson if (list_empty(head)) 1708ade8febdSWeston Andros Adamson return 0; 1709ade8febdSWeston Andros Adamson 1710c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 17111da177e4SLinus Torvalds 17121da177e4SLinus Torvalds if (!data) 17131da177e4SLinus Torvalds goto out_bad; 17141da177e4SLinus Torvalds 17151da177e4SLinus Torvalds /* Set up the argument struct */ 1716f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1717f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 1718c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 1719c36aae9aSPeng Tao data->mds_ops, how, 0); 17201da177e4SLinus Torvalds out_bad: 1721b57ff130SWeston Andros Adamson nfs_retry_commit(head, NULL, cinfo, 0); 17221da177e4SLinus Torvalds return -ENOMEM; 17231da177e4SLinus Torvalds } 17241da177e4SLinus Torvalds 172567911c8fSAnna Schumaker int nfs_commit_file(struct file *file, struct nfs_write_verifier *verf) 172667911c8fSAnna Schumaker { 172767911c8fSAnna Schumaker struct inode *inode = file_inode(file); 172867911c8fSAnna Schumaker struct nfs_open_context *open; 172967911c8fSAnna Schumaker struct nfs_commit_info cinfo; 173067911c8fSAnna Schumaker struct nfs_page *req; 173167911c8fSAnna Schumaker int ret; 173267911c8fSAnna Schumaker 173367911c8fSAnna Schumaker open = get_nfs_open_context(nfs_file_open_context(file)); 173467911c8fSAnna Schumaker req = nfs_create_request(open, NULL, NULL, 0, i_size_read(inode)); 17352997bfd0SDan Carpenter if (IS_ERR(req)) { 17362997bfd0SDan Carpenter ret = PTR_ERR(req); 173767911c8fSAnna Schumaker goto out_put; 173867911c8fSAnna Schumaker } 173967911c8fSAnna Schumaker 174067911c8fSAnna Schumaker nfs_init_cinfo_from_inode(&cinfo, inode); 174167911c8fSAnna Schumaker 174267911c8fSAnna Schumaker memcpy(&req->wb_verf, verf, sizeof(struct nfs_write_verifier)); 174367911c8fSAnna Schumaker nfs_request_add_commit_list(req, &cinfo); 174467911c8fSAnna Schumaker ret = nfs_commit_inode(inode, FLUSH_SYNC); 174567911c8fSAnna Schumaker if (ret > 0) 174667911c8fSAnna Schumaker ret = 0; 174767911c8fSAnna Schumaker 174867911c8fSAnna Schumaker nfs_free_request(req); 174967911c8fSAnna Schumaker out_put: 175067911c8fSAnna Schumaker put_nfs_open_context(open); 175167911c8fSAnna Schumaker return ret; 175267911c8fSAnna Schumaker } 175367911c8fSAnna Schumaker EXPORT_SYMBOL_GPL(nfs_commit_file); 175467911c8fSAnna Schumaker 17551da177e4SLinus Torvalds /* 17561da177e4SLinus Torvalds * COMMIT call returned 17571da177e4SLinus Torvalds */ 1758788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 17591da177e4SLinus Torvalds { 17600b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 17611da177e4SLinus Torvalds 1762a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 17631da177e4SLinus Torvalds task->tk_pid, task->tk_status); 17641da177e4SLinus Torvalds 1765788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1766c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1767c9d8f89dSTrond Myklebust } 1768c9d8f89dSTrond Myklebust 1769f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1770c9d8f89dSTrond Myklebust { 1771c9d8f89dSTrond Myklebust struct nfs_page *req; 1772c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1773f453a54aSFred Isaman struct nfs_commit_info cinfo; 1774353db796SNeilBrown struct nfs_server *nfss; 1775788e7a89STrond Myklebust 17761da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 17771da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 17781da177e4SLinus Torvalds nfs_list_remove_request(req); 177967911c8fSAnna Schumaker if (req->wb_page) 1780d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 17811da177e4SLinus Torvalds 17821e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 17833d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 17842b0143b5SDavid Howells (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)), 17851da177e4SLinus Torvalds req->wb_bytes, 17861da177e4SLinus Torvalds (long long)req_offset(req)); 1787c9d8f89dSTrond Myklebust if (status < 0) { 1788c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 17891da177e4SLinus Torvalds nfs_inode_remove_request(req); 1790*ddeaa637SJoe Perches dprintk_cont(", error = %d\n", status); 17911da177e4SLinus Torvalds goto next; 17921da177e4SLinus Torvalds } 17931da177e4SLinus Torvalds 17941da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 17951da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 17968fc3c386STrond Myklebust if (!nfs_write_verifier_cmp(&req->wb_verf, &data->verf.verifier)) { 17971da177e4SLinus Torvalds /* We have a match */ 17981da177e4SLinus Torvalds nfs_inode_remove_request(req); 1799*ddeaa637SJoe Perches dprintk_cont(" OK\n"); 18001da177e4SLinus Torvalds goto next; 18011da177e4SLinus Torvalds } 18021da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 1803*ddeaa637SJoe Perches dprintk_cont(" mismatch\n"); 18046d884e8fSFred nfs_mark_request_dirty(req); 180505990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 18061da177e4SLinus Torvalds next: 18071d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18081da177e4SLinus Torvalds } 1809353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1810353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 1811353db796SNeilBrown clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 1812353db796SNeilBrown 1813f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1814af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18155917ce84SFred Isaman } 18165917ce84SFred Isaman 18175917ce84SFred Isaman static void nfs_commit_release(void *calldata) 18185917ce84SFred Isaman { 18190b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18205917ce84SFred Isaman 1821f453a54aSFred Isaman data->completion_ops->completion(data); 1822c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 18231da177e4SLinus Torvalds } 1824788e7a89STrond Myklebust 1825788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 18260b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1827788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1828788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1829788e7a89STrond Myklebust }; 18301da177e4SLinus Torvalds 1831f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1832f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1833b20135d0STrond Myklebust .resched_write = nfs_commit_resched_write, 1834f453a54aSFred Isaman }; 1835f453a54aSFred Isaman 18361763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1837ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 183884c53ab5SFred Isaman { 183984c53ab5SFred Isaman int status; 184084c53ab5SFred Isaman 1841ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 184284c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1843ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 184484c53ab5SFred Isaman return status; 184584c53ab5SFred Isaman } 184684c53ab5SFred Isaman 1847b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 18481da177e4SLinus Torvalds { 18491da177e4SLinus Torvalds LIST_HEAD(head); 1850ea2cf228SFred Isaman struct nfs_commit_info cinfo; 185171d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1852af7cf057STrond Myklebust int error = 0; 1853b8413f98STrond Myklebust int res; 18541da177e4SLinus Torvalds 1855ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1856af7cf057STrond Myklebust nfs_commit_begin(cinfo.mds); 1857ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 1858af7cf057STrond Myklebust if (res) 1859ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 1860af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18611da177e4SLinus Torvalds if (error < 0) 1862af7cf057STrond Myklebust goto out_error; 1863b8413f98STrond Myklebust if (!may_wait) 1864b8413f98STrond Myklebust goto out_mark_dirty; 1865af7cf057STrond Myklebust error = wait_on_commit(cinfo.mds); 1866b8413f98STrond Myklebust if (error < 0) 1867b8413f98STrond Myklebust return error; 1868c5efa5fcSTrond Myklebust return res; 1869af7cf057STrond Myklebust out_error: 1870af7cf057STrond Myklebust res = error; 1871c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1872c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1873c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1874c5efa5fcSTrond Myklebust * that the data is on the disk. 1875c5efa5fcSTrond Myklebust */ 1876c5efa5fcSTrond Myklebust out_mark_dirty: 1877c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 18781da177e4SLinus Torvalds return res; 18791da177e4SLinus Torvalds } 1880b20135d0STrond Myklebust EXPORT_SYMBOL_GPL(nfs_commit_inode); 18818fc795f7STrond Myklebust 1882ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 18838fc795f7STrond Myklebust { 1884420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1885420e3646STrond Myklebust int flags = FLUSH_SYNC; 1886420e3646STrond Myklebust int ret = 0; 18878fc795f7STrond Myklebust 18883236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1889ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 18903236c3e1SJeff Layton return ret; 18913236c3e1SJeff Layton 1892a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1893a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1894a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1895420e3646STrond Myklebust */ 1896cb1410c7SWeston Andros Adamson if (nfsi->commit_info.ncommit <= (nfsi->nrequests >> 1)) 1897420e3646STrond Myklebust goto out_mark_dirty; 1898420e3646STrond Myklebust 1899a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1900420e3646STrond Myklebust flags = 0; 1901a00dd6c0SJeff Layton } 1902a00dd6c0SJeff Layton 1903420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1904420e3646STrond Myklebust if (ret >= 0) { 1905420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1906420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1907420e3646STrond Myklebust wbc->nr_to_write -= ret; 1908420e3646STrond Myklebust else 1909420e3646STrond Myklebust wbc->nr_to_write = 0; 1910420e3646STrond Myklebust } 19118fc795f7STrond Myklebust return 0; 1912420e3646STrond Myklebust } 1913420e3646STrond Myklebust out_mark_dirty: 19148fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 19158fc795f7STrond Myklebust return ret; 19168fc795f7STrond Myklebust } 191789d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1918863a3c6cSAndy Adamson 1919acdc53b2STrond Myklebust /* 1920837bb1d7STrond Myklebust * Wrapper for filemap_write_and_wait_range() 1921837bb1d7STrond Myklebust * 1922837bb1d7STrond Myklebust * Needed for pNFS in order to ensure data becomes visible to the 1923837bb1d7STrond Myklebust * client. 1924837bb1d7STrond Myklebust */ 1925837bb1d7STrond Myklebust int nfs_filemap_write_and_wait_range(struct address_space *mapping, 1926837bb1d7STrond Myklebust loff_t lstart, loff_t lend) 1927837bb1d7STrond Myklebust { 1928837bb1d7STrond Myklebust int ret; 1929837bb1d7STrond Myklebust 1930837bb1d7STrond Myklebust ret = filemap_write_and_wait_range(mapping, lstart, lend); 1931837bb1d7STrond Myklebust if (ret == 0) 1932837bb1d7STrond Myklebust ret = pnfs_sync_inode(mapping->host, true); 1933837bb1d7STrond Myklebust return ret; 1934837bb1d7STrond Myklebust } 1935837bb1d7STrond Myklebust EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); 1936837bb1d7STrond Myklebust 1937837bb1d7STrond Myklebust /* 1938acdc53b2STrond Myklebust * flush the inode to disk. 1939acdc53b2STrond Myklebust */ 1940acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 194134901f70STrond Myklebust { 1942f4ce1299STrond Myklebust int ret; 194334901f70STrond Myklebust 1944f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1945f4ce1299STrond Myklebust 19465bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 19476b196875SChuck Lever if (ret) 19486b196875SChuck Lever goto out; 19495bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 19506b196875SChuck Lever if (ret < 0) 19516b196875SChuck Lever goto out; 19525bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 19536b196875SChuck Lever ret = 0; 1954f4ce1299STrond Myklebust 19556b196875SChuck Lever out: 1956f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1957f4ce1299STrond Myklebust return ret; 19581c75950bSTrond Myklebust } 1959ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 19601c75950bSTrond Myklebust 19611b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 19621b3b4a1aSTrond Myklebust { 19631b3b4a1aSTrond Myklebust struct nfs_page *req; 19641b3b4a1aSTrond Myklebust int ret = 0; 19651b3b4a1aSTrond Myklebust 1966ba8b06e6STrond Myklebust wait_on_page_writeback(page); 19673e217045SWeston Andros Adamson 19683e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 19693e217045SWeston Andros Adamson * request */ 19703e217045SWeston Andros Adamson req = nfs_lock_and_join_requests(page, false); 19713e217045SWeston Andros Adamson 19723e217045SWeston Andros Adamson if (IS_ERR(req)) { 19733e217045SWeston Andros Adamson ret = PTR_ERR(req); 19743e217045SWeston Andros Adamson } else if (req) { 19753e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 19763e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 19773e217045SWeston Andros Adamson * request from the inode / page_private pointer and 19783e217045SWeston Andros Adamson * release it */ 19791b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 19801d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 19811b3b4a1aSTrond Myklebust } 19823e217045SWeston Andros Adamson 19831b3b4a1aSTrond Myklebust return ret; 19841b3b4a1aSTrond Myklebust } 19851b3b4a1aSTrond Myklebust 19861c75950bSTrond Myklebust /* 19871c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 19881c75950bSTrond Myklebust */ 1989d6c843b9SPeng Tao int nfs_wb_single_page(struct inode *inode, struct page *page, bool launder) 19901c75950bSTrond Myklebust { 199129418aa4SMel Gorman loff_t range_start = page_file_offset(page); 199209cbfeafSKirill A. Shutemov loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); 19937f2f12d9STrond Myklebust struct writeback_control wbc = { 19947f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 19957f2f12d9STrond Myklebust .nr_to_write = 0, 19967f2f12d9STrond Myklebust .range_start = range_start, 19977f2f12d9STrond Myklebust .range_end = range_end, 19987f2f12d9STrond Myklebust }; 19997f2f12d9STrond Myklebust int ret; 20007f2f12d9STrond Myklebust 2001f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 2002f4ce1299STrond Myklebust 20030522f6adSTrond Myklebust for (;;) { 2004ba8b06e6STrond Myklebust wait_on_page_writeback(page); 20057f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 2006d6c843b9SPeng Tao ret = nfs_writepage_locked(page, &wbc, launder); 20077f2f12d9STrond Myklebust if (ret < 0) 20087f2f12d9STrond Myklebust goto out_error; 20090522f6adSTrond Myklebust continue; 20107f2f12d9STrond Myklebust } 2011f4ce1299STrond Myklebust ret = 0; 20120522f6adSTrond Myklebust if (!PagePrivate(page)) 20130522f6adSTrond Myklebust break; 20140522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20157f2f12d9STrond Myklebust if (ret < 0) 20167f2f12d9STrond Myklebust goto out_error; 20177f2f12d9STrond Myklebust } 20187f2f12d9STrond Myklebust out_error: 2019f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 20207f2f12d9STrond Myklebust return ret; 20211c75950bSTrond Myklebust } 20221c75950bSTrond Myklebust 2023074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 2024074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 2025a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 2026074cc1deSTrond Myklebust { 20272da95652SJeff Layton /* 20282da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 20292da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 20302da95652SJeff Layton * 20312da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 20322da95652SJeff Layton * that we can safely release the inode reference while holding 20332da95652SJeff Layton * the page lock. 20342da95652SJeff Layton */ 20352da95652SJeff Layton if (PagePrivate(page)) 20362da95652SJeff Layton return -EBUSY; 2037074cc1deSTrond Myklebust 20388c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 20398c209ce7SDavid Howells return -EBUSY; 2040074cc1deSTrond Myklebust 2041a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 2042074cc1deSTrond Myklebust } 2043074cc1deSTrond Myklebust #endif 2044074cc1deSTrond Myklebust 2045f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 20461da177e4SLinus Torvalds { 20471da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 20481e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 20491da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 205020c2df83SPaul Mundt NULL); 20511da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 20521da177e4SLinus Torvalds return -ENOMEM; 20531da177e4SLinus Torvalds 205493d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 20551da177e4SLinus Torvalds nfs_wdata_cachep); 20561da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 20573dd4765fSJeff Layton goto out_destroy_write_cache; 20581da177e4SLinus Torvalds 20590b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 20600b7c0153SFred Isaman sizeof(struct nfs_commit_data), 20610b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 20620b7c0153SFred Isaman NULL); 20630b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 20643dd4765fSJeff Layton goto out_destroy_write_mempool; 20650b7c0153SFred Isaman 206693d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 20674c100210SYanchuan Nian nfs_cdata_cachep); 20681da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 20693dd4765fSJeff Layton goto out_destroy_commit_cache; 20701da177e4SLinus Torvalds 207189a09141SPeter Zijlstra /* 207289a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 207389a09141SPeter Zijlstra * 207489a09141SPeter Zijlstra * 64MB: 8192k 207589a09141SPeter Zijlstra * 128MB: 11585k 207689a09141SPeter Zijlstra * 256MB: 16384k 207789a09141SPeter Zijlstra * 512MB: 23170k 207889a09141SPeter Zijlstra * 1GB: 32768k 207989a09141SPeter Zijlstra * 2GB: 46340k 208089a09141SPeter Zijlstra * 4GB: 65536k 208189a09141SPeter Zijlstra * 8GB: 92681k 208289a09141SPeter Zijlstra * 16GB: 131072k 208389a09141SPeter Zijlstra * 208489a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 208589a09141SPeter Zijlstra * Limit the default to 256M 208689a09141SPeter Zijlstra */ 208789a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 208889a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 208989a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 209089a09141SPeter Zijlstra 20911da177e4SLinus Torvalds return 0; 20923dd4765fSJeff Layton 20933dd4765fSJeff Layton out_destroy_commit_cache: 20943dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 20953dd4765fSJeff Layton out_destroy_write_mempool: 20963dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 20973dd4765fSJeff Layton out_destroy_write_cache: 20983dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 20993dd4765fSJeff Layton return -ENOMEM; 21001da177e4SLinus Torvalds } 21011da177e4SLinus Torvalds 2102266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 21031da177e4SLinus Torvalds { 21041da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 21053dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21061da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 21071a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 21081da177e4SLinus Torvalds } 21091da177e4SLinus Torvalds 21104a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 2111a4cdda59SAnna Schumaker .rw_mode = FMODE_WRITE, 21124a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 21134a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 21140eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 21150eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 21161ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 21174a0de55cSAnna Schumaker }; 2118