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 271da177e4SLinus Torvalds #include <asm/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; 154d56b4ddfSMel Gorman if (i_size > 0 && page_file_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 606d56b4ddfSMel Gorman nfs_pageio_cond_complete(pgio, page_file_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); 62840f90271STrond Myklebust nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), 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; 66072cb77f4STrond Myklebust unsigned long *bitlock = &NFS_I(inode)->flags; 661c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 6621da177e4SLinus Torvalds int err; 6631da177e4SLinus Torvalds 66472cb77f4STrond Myklebust /* Stop dirtying of new pages while we sync */ 66574316201SNeilBrown err = wait_on_bit_lock_action(bitlock, NFS_INO_FLUSHING, 66672cb77f4STrond Myklebust nfs_wait_bit_killable, TASK_KILLABLE); 66772cb77f4STrond Myklebust if (err) 66872cb77f4STrond Myklebust goto out_err; 66972cb77f4STrond Myklebust 67091d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 67191d5b470SChuck Lever 672a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 673a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 674f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 675c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 67672cb77f4STrond Myklebust 67772cb77f4STrond Myklebust clear_bit_unlock(NFS_INO_FLUSHING, bitlock); 6784e857c58SPeter Zijlstra smp_mb__after_atomic(); 67972cb77f4STrond Myklebust wake_up_bit(bitlock, NFS_INO_FLUSHING); 68072cb77f4STrond Myklebust 681f758c885STrond Myklebust if (err < 0) 68272cb77f4STrond Myklebust goto out_err; 68372cb77f4STrond Myklebust err = pgio.pg_error; 68472cb77f4STrond Myklebust if (err < 0) 68572cb77f4STrond Myklebust goto out_err; 686c63c7b05STrond Myklebust return 0; 68772cb77f4STrond Myklebust out_err: 68872cb77f4STrond Myklebust return err; 6891da177e4SLinus Torvalds } 6901da177e4SLinus Torvalds 6911da177e4SLinus Torvalds /* 6921da177e4SLinus Torvalds * Insert a write request into an inode 6931da177e4SLinus Torvalds */ 694d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 6951da177e4SLinus Torvalds { 6961da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 697e7d39069STrond Myklebust 6982bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 6992bfc6e56SWeston Andros Adamson 700e7d39069STrond Myklebust /* Lock the request! */ 7017ad84aa9STrond Myklebust nfs_lock_request(req); 702e7d39069STrond Myklebust 703e7d39069STrond Myklebust spin_lock(&inode->i_lock); 704cb1410c7SWeston Andros Adamson if (!nfsi->nrequests && 705cb1410c7SWeston Andros Adamson NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 706a9a4a87aSTrond Myklebust inode->i_version++; 70729418aa4SMel Gorman /* 70829418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 70929418aa4SMel Gorman * with invalidate/truncate. 71029418aa4SMel Gorman */ 71129418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 7122df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 713deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 714277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 71529418aa4SMel Gorman } 716cb1410c7SWeston Andros Adamson nfsi->nrequests++; 71717089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 718cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 719cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 720cb1410c7SWeston Andros Adamson * adding subrequests. */ 72117089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 722c03b4024STrond Myklebust kref_get(&req->wb_kref); 723e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 7241da177e4SLinus Torvalds } 7251da177e4SLinus Torvalds 7261da177e4SLinus Torvalds /* 72789a09141SPeter Zijlstra * Remove a write request from an inode 7281da177e4SLinus Torvalds */ 7291da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7301da177e4SLinus Torvalds { 7312b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 7321da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 73320633f04SWeston Andros Adamson struct nfs_page *head; 73420633f04SWeston Andros Adamson 73520633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 73620633f04SWeston Andros Adamson head = req->wb_head; 7371da177e4SLinus Torvalds 738587142f8STrond Myklebust spin_lock(&inode->i_lock); 73967911c8fSAnna Schumaker if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { 74020633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 74120633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 74295905446SNeilBrown smp_mb__after_atomic(); 74395905446SNeilBrown wake_up_page(head->wb_page, PG_private); 74420633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 74529418aa4SMel Gorman } 746cb1410c7SWeston Andros Adamson nfsi->nrequests--; 747cb1410c7SWeston Andros Adamson spin_unlock(&inode->i_lock); 748cb1410c7SWeston Andros Adamson } else { 749cb1410c7SWeston Andros Adamson spin_lock(&inode->i_lock); 750cb1410c7SWeston Andros Adamson nfsi->nrequests--; 751587142f8STrond Myklebust spin_unlock(&inode->i_lock); 75220633f04SWeston Andros Adamson } 75317089a29SWeston Andros Adamson 75417089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 7551da177e4SLinus Torvalds nfs_release_request(req); 7561da177e4SLinus Torvalds } 7571da177e4SLinus Torvalds 75861822ab5STrond Myklebust static void 7596d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 76061822ab5STrond Myklebust { 76167911c8fSAnna Schumaker if (req->wb_page) 76261822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 76361822ab5STrond Myklebust } 76461822ab5STrond Myklebust 7653a3908c8STrond Myklebust /* 7663a3908c8STrond Myklebust * nfs_page_search_commits_for_head_request_locked 7673a3908c8STrond Myklebust * 7683a3908c8STrond Myklebust * Search through commit lists on @inode for the head request for @page. 7693a3908c8STrond Myklebust * Must be called while holding the inode (which is cinfo) lock. 7703a3908c8STrond Myklebust * 7713a3908c8STrond Myklebust * Returns the head request if found, or NULL if not found. 7723a3908c8STrond Myklebust */ 7733a3908c8STrond Myklebust static struct nfs_page * 7743a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 7753a3908c8STrond Myklebust struct page *page) 7763a3908c8STrond Myklebust { 7773a3908c8STrond Myklebust struct nfs_page *freq, *t; 7783a3908c8STrond Myklebust struct nfs_commit_info cinfo; 7793a3908c8STrond Myklebust struct inode *inode = &nfsi->vfs_inode; 7803a3908c8STrond Myklebust 7813a3908c8STrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 7823a3908c8STrond Myklebust 7833a3908c8STrond Myklebust /* search through pnfs commit lists */ 7843a3908c8STrond Myklebust freq = pnfs_search_commit_reqs(inode, &cinfo, page); 7853a3908c8STrond Myklebust if (freq) 7863a3908c8STrond Myklebust return freq->wb_head; 7873a3908c8STrond Myklebust 7883a3908c8STrond Myklebust /* Linearly search the commit list for the correct request */ 7893a3908c8STrond Myklebust list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 7903a3908c8STrond Myklebust if (freq->wb_page == page) 7913a3908c8STrond Myklebust return freq->wb_head; 7923a3908c8STrond Myklebust } 7933a3908c8STrond Myklebust 7943a3908c8STrond Myklebust return NULL; 7953a3908c8STrond Myklebust } 7963a3908c8STrond Myklebust 7978dd37758STrond Myklebust /** 79886d80f97STrond Myklebust * nfs_request_add_commit_list_locked - add request to a commit list 79986d80f97STrond Myklebust * @req: pointer to a struct nfs_page 80086d80f97STrond Myklebust * @dst: commit list head 80186d80f97STrond Myklebust * @cinfo: holds list lock and accounting info 80286d80f97STrond Myklebust * 80386d80f97STrond Myklebust * This sets the PG_CLEAN bit, updates the cinfo count of 80486d80f97STrond Myklebust * number of outstanding requests requiring a commit as well as 80586d80f97STrond Myklebust * the MM page stats. 80686d80f97STrond Myklebust * 807fe238e60SDave Wysochanski * The caller must hold cinfo->inode->i_lock, and the nfs_page lock. 80886d80f97STrond Myklebust */ 80986d80f97STrond Myklebust void 81086d80f97STrond Myklebust nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, 81186d80f97STrond Myklebust struct nfs_commit_info *cinfo) 81286d80f97STrond Myklebust { 81386d80f97STrond Myklebust set_bit(PG_CLEAN, &req->wb_flags); 81486d80f97STrond Myklebust nfs_list_add_request(req, dst); 81586d80f97STrond Myklebust cinfo->mds->ncommit++; 81686d80f97STrond Myklebust } 81786d80f97STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); 81886d80f97STrond Myklebust 81986d80f97STrond Myklebust /** 8208dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 8218dd37758STrond Myklebust * @req: pointer to a struct nfs_page 822ea2cf228SFred Isaman * @dst: commit list head 823ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8248dd37758STrond Myklebust * 825ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 8268dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8278dd37758STrond Myklebust * the MM page stats. 8288dd37758STrond Myklebust * 829ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8308dd37758STrond Myklebust * holding the nfs_page lock. 8318dd37758STrond Myklebust */ 8328dd37758STrond Myklebust void 8336272dcc6SAnna Schumaker nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) 8348dd37758STrond Myklebust { 835fe238e60SDave Wysochanski spin_lock(&cinfo->inode->i_lock); 8366272dcc6SAnna Schumaker nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); 837fe238e60SDave Wysochanski spin_unlock(&cinfo->inode->i_lock); 83867911c8fSAnna Schumaker if (req->wb_page) 83986d80f97STrond Myklebust nfs_mark_page_unstable(req->wb_page, cinfo); 8408dd37758STrond Myklebust } 8418dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 8428dd37758STrond Myklebust 8438dd37758STrond Myklebust /** 8448dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 8458dd37758STrond Myklebust * @req: pointer to a nfs_page 846ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8478dd37758STrond Myklebust * 848ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 8498dd37758STrond Myklebust * number of outstanding requests requiring a commit 8508dd37758STrond Myklebust * It does not update the MM page stats. 8518dd37758STrond Myklebust * 852ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 8538dd37758STrond Myklebust */ 8548dd37758STrond Myklebust void 855ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 856ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 8578dd37758STrond Myklebust { 8588dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 8598dd37758STrond Myklebust return; 8608dd37758STrond Myklebust nfs_list_remove_request(req); 861ea2cf228SFred Isaman cinfo->mds->ncommit--; 8628dd37758STrond Myklebust } 8638dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 8648dd37758STrond Myklebust 865ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 866ea2cf228SFred Isaman struct inode *inode) 867ea2cf228SFred Isaman { 868fe238e60SDave Wysochanski cinfo->inode = inode; 869ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 870ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 871b359f9d0SFred Isaman cinfo->dreq = NULL; 872f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 873ea2cf228SFred Isaman } 874ea2cf228SFred Isaman 875ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 876ea2cf228SFred Isaman struct inode *inode, 877ea2cf228SFred Isaman struct nfs_direct_req *dreq) 878ea2cf228SFred Isaman { 8791763da12SFred Isaman if (dreq) 8801763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 8811763da12SFred Isaman else 882ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 883ea2cf228SFred Isaman } 884ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 8858dd37758STrond Myklebust 8861da177e4SLinus Torvalds /* 8871da177e4SLinus Torvalds * Add a request to the inode's commit list. 8881da177e4SLinus Torvalds */ 8891763da12SFred Isaman void 890ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 891b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 8921da177e4SLinus Torvalds { 893b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 8948dd37758STrond Myklebust return; 8956272dcc6SAnna Schumaker nfs_request_add_commit_list(req, cinfo); 8961da177e4SLinus Torvalds } 8978e821cadSTrond Myklebust 898d6d6dc7cSFred Isaman static void 899d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 900e468bae9STrond Myklebust { 901e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 90293f78d88STejun Heo dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, 90393f78d88STejun Heo WB_RECLAIMABLE); 904e468bae9STrond Myklebust } 905d6d6dc7cSFred Isaman 906411a99adSWeston Andros Adamson /* Called holding inode (/cinfo) lock */ 9078dd37758STrond Myklebust static void 908d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 909d6d6dc7cSFred Isaman { 9108dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 9112b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 912ea2cf228SFred Isaman struct nfs_commit_info cinfo; 913d6d6dc7cSFred Isaman 914ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 915ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 916ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 917d6d6dc7cSFred Isaman } 9188dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 9198dd37758STrond Myklebust } 920e468bae9STrond Myklebust } 921e468bae9STrond Myklebust 922d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 9238e821cadSTrond Myklebust { 924c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 925d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 926c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 9278e821cadSTrond Myklebust } 9288e821cadSTrond Myklebust 929061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 9306c75dc0dSFred Isaman { 931ea2cf228SFred Isaman struct nfs_commit_info cinfo; 9326c75dc0dSFred Isaman unsigned long bytes = 0; 9336c75dc0dSFred Isaman 9346c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 9356c75dc0dSFred Isaman goto out; 936ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 9376c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 9386c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 9396c75dc0dSFred Isaman 9406c75dc0dSFred Isaman bytes += req->wb_bytes; 9416c75dc0dSFred Isaman nfs_list_remove_request(req); 9426c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 9436c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 944d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 9456c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 9466c75dc0dSFred Isaman goto remove_req; 9476c75dc0dSFred Isaman } 948c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 949f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 950b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 951a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 9526c75dc0dSFred Isaman goto next; 9536c75dc0dSFred Isaman } 9546c75dc0dSFred Isaman remove_req: 9556c75dc0dSFred Isaman nfs_inode_remove_request(req); 9566c75dc0dSFred Isaman next: 9571d1afcbcSTrond Myklebust nfs_unlock_request(req); 95820633f04SWeston Andros Adamson nfs_end_page_writeback(req); 9593aff4ebbSTrond Myklebust nfs_release_request(req); 9606c75dc0dSFred Isaman } 9616c75dc0dSFred Isaman out: 9626c75dc0dSFred Isaman hdr->release(hdr); 9636c75dc0dSFred Isaman } 9646c75dc0dSFred Isaman 965ce59515cSAnna Schumaker unsigned long 966ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 967fb8a1f11STrond Myklebust { 968ea2cf228SFred Isaman return cinfo->mds->ncommit; 969fb8a1f11STrond Myklebust } 970fb8a1f11STrond Myklebust 971fe238e60SDave Wysochanski /* cinfo->inode->i_lock held by caller */ 9721763da12SFred Isaman int 973ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 974ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 975d6d6dc7cSFred Isaman { 976d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 977d6d6dc7cSFred Isaman int ret = 0; 978d6d6dc7cSFred Isaman 979d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 9808dd37758STrond Myklebust if (!nfs_lock_request(req)) 9818dd37758STrond Myklebust continue; 9827ad84aa9STrond Myklebust kref_get(&req->wb_kref); 983fe238e60SDave Wysochanski if (cond_resched_lock(&cinfo->inode->i_lock)) 9843b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 985ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 9868dd37758STrond Myklebust nfs_list_add_request(req, dst); 987d6d6dc7cSFred Isaman ret++; 9881763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 989d6d6dc7cSFred Isaman break; 990d6d6dc7cSFred Isaman } 991d6d6dc7cSFred Isaman return ret; 992d6d6dc7cSFred Isaman } 993d6d6dc7cSFred Isaman 9941da177e4SLinus Torvalds /* 9951da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 9961da177e4SLinus Torvalds * @inode: NFS inode to scan 997ea2cf228SFred Isaman * @dst: mds destination list 998ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 9991da177e4SLinus Torvalds * 10001da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 10011da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 10021da177e4SLinus Torvalds */ 10031763da12SFred Isaman int 1004ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 1005ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 10061da177e4SLinus Torvalds { 1007d6d6dc7cSFred Isaman int ret = 0; 1008fb8a1f11STrond Myklebust 1009fe238e60SDave Wysochanski spin_lock(&cinfo->inode->i_lock); 1010ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 10118dd37758STrond Myklebust const int max = INT_MAX; 1012d6d6dc7cSFred Isaman 1013ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 1014ea2cf228SFred Isaman cinfo, max); 1015ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 1016d6d6dc7cSFred Isaman } 1017fe238e60SDave Wysochanski spin_unlock(&cinfo->inode->i_lock); 1018ff778d02STrond Myklebust return ret; 10191da177e4SLinus Torvalds } 1020d6d6dc7cSFred Isaman 10211da177e4SLinus Torvalds /* 1022e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1023e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 10241da177e4SLinus Torvalds * 1025e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1026e7d39069STrond Myklebust * to disk. 10271da177e4SLinus Torvalds */ 1028e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 1029e7d39069STrond Myklebust struct page *page, 1030e7d39069STrond Myklebust unsigned int offset, 1031e7d39069STrond Myklebust unsigned int bytes) 10321da177e4SLinus Torvalds { 1033e7d39069STrond Myklebust struct nfs_page *req; 1034e7d39069STrond Myklebust unsigned int rqend; 1035e7d39069STrond Myklebust unsigned int end; 10361da177e4SLinus Torvalds int error; 1037277459d2STrond Myklebust 1038e7d39069STrond Myklebust if (!PagePrivate(page)) 1039e7d39069STrond Myklebust return NULL; 1040e7d39069STrond Myklebust 1041e7d39069STrond Myklebust end = offset + bytes; 1042e7d39069STrond Myklebust spin_lock(&inode->i_lock); 1043e7d39069STrond Myklebust 1044e7d39069STrond Myklebust for (;;) { 104584d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 1046e7d39069STrond Myklebust if (req == NULL) 1047e7d39069STrond Myklebust goto out_unlock; 1048e7d39069STrond Myklebust 10492bfc6e56SWeston Andros Adamson /* should be handled by nfs_flush_incompatible */ 10502bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 10512bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 10522bfc6e56SWeston Andros Adamson 1053e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1054e7d39069STrond Myklebust /* 1055e7d39069STrond Myklebust * Tell the caller to flush out the request if 1056e7d39069STrond Myklebust * the offsets are non-contiguous. 1057e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1058e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1059e7d39069STrond Myklebust */ 1060e468bae9STrond Myklebust if (offset > rqend 1061e7d39069STrond Myklebust || end < req->wb_offset) 1062e7d39069STrond Myklebust goto out_flushme; 1063e7d39069STrond Myklebust 10647ad84aa9STrond Myklebust if (nfs_lock_request(req)) 1065e7d39069STrond Myklebust break; 1066e7d39069STrond Myklebust 1067e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 1068587142f8STrond Myklebust spin_unlock(&inode->i_lock); 10691da177e4SLinus Torvalds error = nfs_wait_on_request(req); 10701da177e4SLinus Torvalds nfs_release_request(req); 1071e7d39069STrond Myklebust if (error != 0) 1072e7d39069STrond Myklebust goto out_err; 1073e7d39069STrond Myklebust spin_lock(&inode->i_lock); 10741da177e4SLinus Torvalds } 10751da177e4SLinus Torvalds 10761da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 10771da177e4SLinus Torvalds if (offset < req->wb_offset) { 10781da177e4SLinus Torvalds req->wb_offset = offset; 10791da177e4SLinus Torvalds req->wb_pgbase = offset; 10801da177e4SLinus Torvalds } 10811da177e4SLinus Torvalds if (end > rqend) 10821da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1083e7d39069STrond Myklebust else 1084e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 1085e7d39069STrond Myklebust out_unlock: 1086ca138f36SFred Isaman if (req) 10878dd37758STrond Myklebust nfs_clear_request_commit(req); 1088411a99adSWeston Andros Adamson spin_unlock(&inode->i_lock); 1089e7d39069STrond Myklebust return req; 1090e7d39069STrond Myklebust out_flushme: 1091e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 1092e7d39069STrond Myklebust nfs_release_request(req); 1093e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1094e7d39069STrond Myklebust out_err: 1095e7d39069STrond Myklebust return ERR_PTR(error); 1096e7d39069STrond Myklebust } 10971da177e4SLinus Torvalds 1098e7d39069STrond Myklebust /* 1099e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1100e7d39069STrond Myklebust * 1101e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1102e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1103e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1104e7d39069STrond Myklebust */ 1105e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1106e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1107e7d39069STrond Myklebust { 1108d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1109e7d39069STrond Myklebust struct nfs_page *req; 1110e7d39069STrond Myklebust 1111e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1112e7d39069STrond Myklebust if (req != NULL) 1113e7d39069STrond Myklebust goto out; 11142bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 1115e7d39069STrond Myklebust if (IS_ERR(req)) 1116e7d39069STrond Myklebust goto out; 1117d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1118efc91ed0STrond Myklebust out: 111961e930a9STrond Myklebust return req; 11201da177e4SLinus Torvalds } 11211da177e4SLinus Torvalds 1122e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1123e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1124e7d39069STrond Myklebust { 1125e7d39069STrond Myklebust struct nfs_page *req; 1126e7d39069STrond Myklebust 1127e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1128e7d39069STrond Myklebust if (IS_ERR(req)) 1129e7d39069STrond Myklebust return PTR_ERR(req); 1130e7d39069STrond Myklebust /* Update file length */ 1131e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1132d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1133a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11341d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1135e7d39069STrond Myklebust return 0; 1136e7d39069STrond Myklebust } 1137e7d39069STrond Myklebust 11381da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 11391da177e4SLinus Torvalds { 1140cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 11412a369153STrond Myklebust struct nfs_lock_context *l_ctx; 1142bd61e0a9SJeff Layton struct file_lock_context *flctx = file_inode(file)->i_flctx; 11431da177e4SLinus Torvalds struct nfs_page *req; 11441a54533eSTrond Myklebust int do_flush, status; 11451da177e4SLinus Torvalds /* 11461da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 11471da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 11481da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 11491da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11501da177e4SLinus Torvalds * Also do the same if we find a request from an existing 11511da177e4SLinus Torvalds * dropped page. 11521da177e4SLinus Torvalds */ 11531a54533eSTrond Myklebust do { 115484d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 11551a54533eSTrond Myklebust if (req == NULL) 11561a54533eSTrond Myklebust return 0; 11572a369153STrond Myklebust l_ctx = req->wb_lock_context; 1158138a2935STrond Myklebust do_flush = req->wb_page != page || 1159138a2935STrond Myklebust !nfs_match_open_context(req->wb_context, ctx); 11602bfc6e56SWeston Andros Adamson /* for now, flush if more than 1 request in page_group */ 11612bfc6e56SWeston Andros Adamson do_flush |= req->wb_this_page != req; 1162bd61e0a9SJeff Layton if (l_ctx && flctx && 1163bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1164bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 11655263e31eSJeff Layton do_flush |= l_ctx->lockowner.l_owner != current->files 11665263e31eSJeff Layton || l_ctx->lockowner.l_pid != current->tgid; 11675263e31eSJeff Layton } 11681da177e4SLinus Torvalds nfs_release_request(req); 11691a54533eSTrond Myklebust if (!do_flush) 11701a54533eSTrond Myklebust return 0; 1171d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 11721a54533eSTrond Myklebust } while (status == 0); 11731a54533eSTrond Myklebust return status; 11741da177e4SLinus Torvalds } 11751da177e4SLinus Torvalds 11761da177e4SLinus Torvalds /* 1177dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1178dc24826bSAndy Adamson * expire soon. 1179dc24826bSAndy Adamson * 1180dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1181dc24826bSAndy Adamson * 1182dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1183dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1184dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1185dc24826bSAndy Adamson */ 1186dc24826bSAndy Adamson int 1187dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1188dc24826bSAndy Adamson { 1189dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1190dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1191dc24826bSAndy Adamson 1192dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 1193dc24826bSAndy Adamson } 1194dc24826bSAndy Adamson 1195dc24826bSAndy Adamson /* 1196dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1197dc24826bSAndy Adamson */ 1198ce52914eSScott Mayhew bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) 1199dc24826bSAndy Adamson { 1200ce52914eSScott Mayhew struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1201ce52914eSScott Mayhew 1202ce52914eSScott Mayhew return rpcauth_cred_key_to_expire(auth, ctx->cred); 1203dc24826bSAndy Adamson } 1204dc24826bSAndy Adamson 1205dc24826bSAndy Adamson /* 12065d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 12075d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 12085d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 12095d47a356STrond Myklebust */ 12108d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 12115d47a356STrond Myklebust { 1212d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1213d529ef83SJeff Layton 12148d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 12158d197a56STrond Myklebust goto out; 121618dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1217d529ef83SJeff Layton return false; 12184db72b40SJeff Layton smp_rmb(); 1219d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 12208d197a56STrond Myklebust return false; 12218d197a56STrond Myklebust out: 122218dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 122318dd78c4SScott Mayhew return false; 12248d197a56STrond Myklebust return PageUptodate(page) != 0; 12255d47a356STrond Myklebust } 12265d47a356STrond Myklebust 12275263e31eSJeff Layton static bool 12285263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 12295263e31eSJeff Layton { 12305263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 12315263e31eSJeff Layton fl->fl_type == F_WRLCK; 12325263e31eSJeff Layton } 12335263e31eSJeff Layton 1234c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1235c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1236c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1237c7559663SScott Mayhew * inefficiencies. 1238c7559663SScott Mayhew * 1239263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1240263b4509SScott Mayhew * of the checks. 1241c7559663SScott Mayhew */ 1242c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1243c7559663SScott Mayhew { 12445263e31eSJeff Layton int ret; 12455263e31eSJeff Layton struct file_lock_context *flctx = inode->i_flctx; 12465263e31eSJeff Layton struct file_lock *fl; 12475263e31eSJeff Layton 1248c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1249c7559663SScott Mayhew return 0; 1250263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1251263b4509SScott Mayhew return 0; 1252c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1253c7559663SScott Mayhew return 1; 1254bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1255bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 12568fa4592aSTrond Myklebust return 1; 12575263e31eSJeff Layton 12585263e31eSJeff Layton /* Check to see if there are whole file write locks */ 12595263e31eSJeff Layton ret = 0; 12606109c850SJeff Layton spin_lock(&flctx->flc_lock); 1261bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1262bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1263bd61e0a9SJeff Layton fl_list); 1264bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 12655263e31eSJeff Layton ret = 1; 1266bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 12675263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 12685263e31eSJeff Layton fl_list); 12695263e31eSJeff Layton if (fl->fl_type == F_WRLCK) 12705263e31eSJeff Layton ret = 1; 12715263e31eSJeff Layton } 12726109c850SJeff Layton spin_unlock(&flctx->flc_lock); 12735263e31eSJeff Layton return ret; 1274c7559663SScott Mayhew } 1275c7559663SScott Mayhew 12765d47a356STrond Myklebust /* 12771da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 12781da177e4SLinus Torvalds * 12791da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 12801da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 12811da177e4SLinus Torvalds */ 12821da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 12831da177e4SLinus Torvalds unsigned int offset, unsigned int count) 12841da177e4SLinus Torvalds { 1285cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1286d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 12871da177e4SLinus Torvalds int status = 0; 12881da177e4SLinus Torvalds 128991d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 129091d5b470SChuck Lever 12916de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 12926de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 12931da177e4SLinus Torvalds 1294*149a4fddSBenjamin Coddington if (!count) 1295*149a4fddSBenjamin Coddington goto out; 1296*149a4fddSBenjamin Coddington 1297c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 129849a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 12991da177e4SLinus Torvalds offset = 0; 13001da177e4SLinus Torvalds } 13011da177e4SLinus Torvalds 1302e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 130303fa9e84STrond Myklebust if (status < 0) 130403fa9e84STrond Myklebust nfs_set_pageerror(page); 130559b7c05fSTrond Myklebust else 130659b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 1307*149a4fddSBenjamin Coddington out: 130848186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 13091da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 13101da177e4SLinus Torvalds return status; 13111da177e4SLinus Torvalds } 13121da177e4SLinus Torvalds 13133ff7576dSTrond Myklebust static int flush_task_priority(int how) 13141da177e4SLinus Torvalds { 13151da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 13161da177e4SLinus Torvalds case FLUSH_HIGHPRI: 13171da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 13181da177e4SLinus Torvalds case FLUSH_LOWPRI: 13191da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 13201da177e4SLinus Torvalds } 13211da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 13221da177e4SLinus Torvalds } 13231da177e4SLinus Torvalds 1324d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1325d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1326abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 13271ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 13281da177e4SLinus Torvalds { 13293ff7576dSTrond Myklebust int priority = flush_task_priority(how); 13301da177e4SLinus Torvalds 13311ed26f33SAnna Schumaker task_setup_data->priority = priority; 1332abde71f4STom Haynes rpc_ops->write_setup(hdr, msg); 1333d138d5d1SAndy Adamson 1334abde71f4STom Haynes nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client, 1335d45f60c6SWeston Andros Adamson &task_setup_data->rpc_client, msg, hdr); 1336275acaafSTrond Myklebust } 1337275acaafSTrond Myklebust 13386d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 13396d884e8fSFred * call this on each, which will prepare them to be retried on next 13406d884e8fSFred * writeback using standard nfs. 13416d884e8fSFred */ 13426d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 13436d884e8fSFred { 13446d884e8fSFred nfs_mark_request_dirty(req); 1345c7070113STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 13461d1afcbcSTrond Myklebust nfs_unlock_request(req); 134720633f04SWeston Andros Adamson nfs_end_page_writeback(req); 13483aff4ebbSTrond Myklebust nfs_release_request(req); 13496d884e8fSFred } 13506d884e8fSFred 1351061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 13526c75dc0dSFred Isaman { 13536c75dc0dSFred Isaman struct nfs_page *req; 13546c75dc0dSFred Isaman 13556c75dc0dSFred Isaman while (!list_empty(head)) { 13566c75dc0dSFred Isaman req = nfs_list_entry(head->next); 13576c75dc0dSFred Isaman nfs_list_remove_request(req); 13586c75dc0dSFred Isaman nfs_redirty_request(req); 13596c75dc0dSFred Isaman } 13606c75dc0dSFred Isaman } 13616c75dc0dSFred Isaman 1362dc602dd7STrond Myklebust static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) 1363dc602dd7STrond Myklebust { 1364dc602dd7STrond Myklebust nfs_async_write_error(&hdr->pages); 1365dc602dd7STrond Myklebust } 1366dc602dd7STrond Myklebust 1367061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1368061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1369061ae2edSFred Isaman .completion = nfs_write_completion, 1370dc602dd7STrond Myklebust .reschedule_io = nfs_async_write_reschedule_io, 1371061ae2edSFred Isaman }; 1372061ae2edSFred Isaman 137357208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1374a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1375061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 13761751c363STrond Myklebust { 1377a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 137841d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1379a20c93e3SChristoph Hellwig 1380a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1381a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1382a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1383a20c93e3SChristoph Hellwig #endif 13844a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 13854a0de55cSAnna Schumaker server->wsize, ioflags); 13861751c363STrond Myklebust } 1387ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 13881751c363STrond Myklebust 1389dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1390dce81290STrond Myklebust { 1391a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1392a7d42ddbSWeston Andros Adamson 13936f29b9bbSKinglong Mee if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) 13946f29b9bbSKinglong Mee pgio->pg_ops->pg_cleanup(pgio); 13956f29b9bbSKinglong Mee 139641d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1397a7d42ddbSWeston Andros Adamson 1398a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1399a7d42ddbSWeston Andros Adamson 1400a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1401a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1402dce81290STrond Myklebust } 14031f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1404dce81290STrond Myklebust 14051da177e4SLinus Torvalds 14060b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 14070b7c0153SFred Isaman { 14080b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 14090b7c0153SFred Isaman 14100b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 14110b7c0153SFred Isaman } 14120b7c0153SFred Isaman 14131f2edbe3STrond Myklebust /* 14141f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 14151f2edbe3STrond Myklebust */ 14161f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 14171f2edbe3STrond Myklebust { 14181f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 14191f2edbe3STrond Myklebust int kill = 0; 1420788e7a89STrond Myklebust 14211f2edbe3STrond Myklebust /* suid always must be killed */ 14221f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 14231f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 14241f2edbe3STrond Myklebust 14251f2edbe3STrond Myklebust /* 14261f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 14271f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 14281f2edbe3STrond Myklebust */ 14291f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 14301f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 14311f2edbe3STrond Myklebust 14321f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 14331f2edbe3STrond Myklebust return kill; 14341f2edbe3STrond Myklebust 14351f2edbe3STrond Myklebust return 0; 14361f2edbe3STrond Myklebust } 1437788e7a89STrond Myklebust 1438a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1439a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1440a08a8cd3STrond Myklebust { 1441a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1442a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 14432b83d3deSTrond Myklebust u64 size = argp->offset + resp->count; 1444a08a8cd3STrond Myklebust 1445a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 14462b83d3deSTrond Myklebust fattr->size = size; 14472b83d3deSTrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { 14482b83d3deSTrond Myklebust fattr->valid &= ~NFS_ATTR_FATTR_SIZE; 1449a08a8cd3STrond Myklebust return; 14502b83d3deSTrond Myklebust } 14512b83d3deSTrond Myklebust if (size != fattr->size) 1452a08a8cd3STrond Myklebust return; 1453a08a8cd3STrond Myklebust /* Set attribute barrier */ 1454a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 14552b83d3deSTrond Myklebust /* ...and update size */ 14562b83d3deSTrond Myklebust fattr->valid |= NFS_ATTR_FATTR_SIZE; 1457a08a8cd3STrond Myklebust } 1458a08a8cd3STrond Myklebust 1459a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1460a08a8cd3STrond Myklebust { 14612b83d3deSTrond Myklebust struct nfs_fattr *fattr = &hdr->fattr; 1462a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1463a08a8cd3STrond Myklebust 1464a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1465a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1466a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1467a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1468a08a8cd3STrond Myklebust } 1469a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1470a08a8cd3STrond Myklebust 14711da177e4SLinus Torvalds /* 14721da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 14731da177e4SLinus Torvalds */ 1474d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1475d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 14760eecb214SAnna Schumaker struct inode *inode) 14771da177e4SLinus Torvalds { 1478788e7a89STrond Myklebust int status; 14791da177e4SLinus Torvalds 1480f551e44fSChuck Lever /* 1481f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1482f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1483f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1484f551e44fSChuck Lever * another writer had changed the file, but some applications 1485f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1486f551e44fSChuck Lever */ 1487d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1488788e7a89STrond Myklebust if (status != 0) 14890eecb214SAnna Schumaker return status; 1490d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 149191d5b470SChuck Lever 1492d45f60c6SWeston Andros Adamson if (hdr->res.verf->committed < hdr->args.stable && 1493d45f60c6SWeston Andros Adamson task->tk_status >= 0) { 14941da177e4SLinus Torvalds /* We tried a write call, but the server did not 14951da177e4SLinus Torvalds * commit data to stable storage even though we 14961da177e4SLinus Torvalds * requested it. 14971da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 14981da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 14991da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 15001da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 15011da177e4SLinus Torvalds */ 15021da177e4SLinus Torvalds static unsigned long complain; 15031da177e4SLinus Torvalds 1504a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 15051da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 15061da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 15071da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1508cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1509d45f60c6SWeston Andros Adamson hdr->res.verf->committed, hdr->args.stable); 15101da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 15111da177e4SLinus Torvalds } 15121da177e4SLinus Torvalds } 15131f2edbe3STrond Myklebust 15141f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 15151f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 15161f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 15170eecb214SAnna Schumaker return 0; 15180eecb214SAnna Schumaker } 15190eecb214SAnna Schumaker 15200eecb214SAnna Schumaker /* 15210eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 15220eecb214SAnna Schumaker */ 1523d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1524d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 15250eecb214SAnna Schumaker { 1526d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1527d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 15281f2edbe3STrond Myklebust 15291f2edbe3STrond Myklebust if (resp->count < argp->count) { 15301da177e4SLinus Torvalds static unsigned long complain; 15311da177e4SLinus Torvalds 15326c75dc0dSFred Isaman /* This a short write! */ 1533d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 153491d5b470SChuck Lever 15351da177e4SLinus Torvalds /* Has the server at least made some progress? */ 15366c75dc0dSFred Isaman if (resp->count == 0) { 15376c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 15386c75dc0dSFred Isaman printk(KERN_WARNING 15396c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 15406c75dc0dSFred Isaman argp->count); 15416c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 15426c75dc0dSFred Isaman } 1543d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 15446c75dc0dSFred Isaman task->tk_status = -EIO; 15456c75dc0dSFred Isaman return; 15466c75dc0dSFred Isaman } 1547f8417b48SKinglong Mee 1548f8417b48SKinglong Mee /* For non rpc-based layout drivers, retry-through-MDS */ 1549f8417b48SKinglong Mee if (!task->tk_ops) { 1550f8417b48SKinglong Mee hdr->pnfs_error = -EAGAIN; 1551f8417b48SKinglong Mee return; 1552f8417b48SKinglong Mee } 1553f8417b48SKinglong Mee 15541da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 15551da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 15561da177e4SLinus Torvalds /* Resend from where the server left off */ 1557d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 15581da177e4SLinus Torvalds argp->offset += resp->count; 15591da177e4SLinus Torvalds argp->pgbase += resp->count; 15601da177e4SLinus Torvalds argp->count -= resp->count; 15611da177e4SLinus Torvalds } else { 15621da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 15631da177e4SLinus Torvalds * headaches in the case of a server crash. 15641da177e4SLinus Torvalds */ 15651da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 15661da177e4SLinus Torvalds } 1567d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 15681da177e4SLinus Torvalds } 15691da177e4SLinus Torvalds } 15701da177e4SLinus Torvalds 1571af7cf057STrond Myklebust static int wait_on_commit(struct nfs_mds_commit_info *cinfo) 157271d0a611STrond Myklebust { 1573af7cf057STrond Myklebust return wait_on_atomic_t(&cinfo->rpcs_out, 1574af7cf057STrond Myklebust nfs_wait_atomic_killable, TASK_KILLABLE); 1575af7cf057STrond Myklebust } 1576af7cf057STrond Myklebust 1577af7cf057STrond Myklebust static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) 1578af7cf057STrond Myklebust { 1579af7cf057STrond Myklebust atomic_inc(&cinfo->rpcs_out); 1580af7cf057STrond Myklebust } 1581af7cf057STrond Myklebust 1582af7cf057STrond Myklebust static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) 1583af7cf057STrond Myklebust { 1584af7cf057STrond Myklebust if (atomic_dec_and_test(&cinfo->rpcs_out)) 1585af7cf057STrond Myklebust wake_up_atomic_t(&cinfo->rpcs_out); 158671d0a611STrond Myklebust } 158771d0a611STrond Myklebust 15880b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 15891da177e4SLinus Torvalds { 15900b7c0153SFred Isaman put_nfs_open_context(data->context); 15910b7c0153SFred Isaman nfs_commit_free(data); 15921da177e4SLinus Torvalds } 1593e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 15941da177e4SLinus Torvalds 15950b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1596c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 15979ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 15989f0ec176SAndy Adamson int how, int flags) 15991da177e4SLinus Torvalds { 160007737691STrond Myklebust struct rpc_task *task; 16019ace33cdSFred Isaman int priority = flush_task_priority(how); 1602bdc7f021STrond Myklebust struct rpc_message msg = { 1603bdc7f021STrond Myklebust .rpc_argp = &data->args, 1604bdc7f021STrond Myklebust .rpc_resp = &data->res, 16059ace33cdSFred Isaman .rpc_cred = data->cred, 1606bdc7f021STrond Myklebust }; 160784115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 160807737691STrond Myklebust .task = &data->task, 16099ace33cdSFred Isaman .rpc_client = clnt, 1610bdc7f021STrond Myklebust .rpc_message = &msg, 16119ace33cdSFred Isaman .callback_ops = call_ops, 161284115e1cSTrond Myklebust .callback_data = data, 1613101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 16149f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 16153ff7576dSTrond Myklebust .priority = priority, 161684115e1cSTrond Myklebust }; 1617788e7a89STrond Myklebust /* Set up the initial task struct. */ 1618c36aae9aSPeng Tao nfs_ops->commit_setup(data, &msg); 16191da177e4SLinus Torvalds 1620b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1621bdc7f021STrond Myklebust 16228c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 16238c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 16248c21c62cSWeston Andros Adamson 162507737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1626dbae4c73STrond Myklebust if (IS_ERR(task)) 1627dbae4c73STrond Myklebust return PTR_ERR(task); 1628d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1629d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 163007737691STrond Myklebust rpc_put_task(task); 1631dbae4c73STrond Myklebust return 0; 16321da177e4SLinus Torvalds } 1633e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 16341da177e4SLinus Torvalds 1635378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1636378520b8SPeng Tao { 1637378520b8SPeng Tao loff_t lwb = 0; 1638378520b8SPeng Tao struct nfs_page *req; 1639378520b8SPeng Tao 1640378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1641378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1642378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1643378520b8SPeng Tao 1644378520b8SPeng Tao return lwb; 1645378520b8SPeng Tao } 1646378520b8SPeng Tao 16471da177e4SLinus Torvalds /* 16489ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 16499ace33cdSFred Isaman */ 16500b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1651988b6dceSFred Isaman struct list_head *head, 1652f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1653f453a54aSFred Isaman struct nfs_commit_info *cinfo) 16549ace33cdSFred Isaman { 16559ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 16562b0143b5SDavid Howells struct inode *inode = d_inode(first->wb_context->dentry); 16579ace33cdSFred Isaman 16589ace33cdSFred Isaman /* Set up the RPC argument and reply structs 16599ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 16609ace33cdSFred Isaman 16619ace33cdSFred Isaman list_splice_init(head, &data->pages); 16629ace33cdSFred Isaman 16639ace33cdSFred Isaman data->inode = inode; 16649ace33cdSFred Isaman data->cred = first->wb_context->cred; 1665988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1666378520b8SPeng Tao /* only set lwb for pnfs commit */ 1667378520b8SPeng Tao if (lseg) 1668378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 16699ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1670f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1671b359f9d0SFred Isaman data->dreq = cinfo->dreq; 16729ace33cdSFred Isaman 16739ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 16749ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 16759ace33cdSFred Isaman data->args.offset = 0; 16769ace33cdSFred Isaman data->args.count = 0; 16770b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 16789ace33cdSFred Isaman data->res.fattr = &data->fattr; 16799ace33cdSFred Isaman data->res.verf = &data->verf; 16809ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 16819ace33cdSFred Isaman } 1682e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 16839ace33cdSFred Isaman 1684e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1685ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1686b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1687b57ff130SWeston Andros Adamson u32 ds_commit_idx) 168864bfeb49SFred Isaman { 168964bfeb49SFred Isaman struct nfs_page *req; 169064bfeb49SFred Isaman 169164bfeb49SFred Isaman while (!list_empty(page_list)) { 169264bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 169364bfeb49SFred Isaman nfs_list_remove_request(req); 1694b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 1695487b9b8aSTom Haynes if (!cinfo->dreq) 1696487b9b8aSTom Haynes nfs_clear_page_commit(req->wb_page); 16971d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 169864bfeb49SFred Isaman } 169964bfeb49SFred Isaman } 1700e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 170164bfeb49SFred Isaman 1702b20135d0STrond Myklebust static void 1703b20135d0STrond Myklebust nfs_commit_resched_write(struct nfs_commit_info *cinfo, 1704b20135d0STrond Myklebust struct nfs_page *req) 1705b20135d0STrond Myklebust { 1706b20135d0STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 1707b20135d0STrond Myklebust } 1708b20135d0STrond Myklebust 17099ace33cdSFred Isaman /* 17101da177e4SLinus Torvalds * Commit dirty pages 17111da177e4SLinus Torvalds */ 17121da177e4SLinus Torvalds static int 1713ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1714ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 17151da177e4SLinus Torvalds { 17160b7c0153SFred Isaman struct nfs_commit_data *data; 17171da177e4SLinus Torvalds 1718ade8febdSWeston Andros Adamson /* another commit raced with us */ 1719ade8febdSWeston Andros Adamson if (list_empty(head)) 1720ade8febdSWeston Andros Adamson return 0; 1721ade8febdSWeston Andros Adamson 1722c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 17231da177e4SLinus Torvalds 17241da177e4SLinus Torvalds if (!data) 17251da177e4SLinus Torvalds goto out_bad; 17261da177e4SLinus Torvalds 17271da177e4SLinus Torvalds /* Set up the argument struct */ 1728f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1729f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 1730c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 1731c36aae9aSPeng Tao data->mds_ops, how, 0); 17321da177e4SLinus Torvalds out_bad: 1733b57ff130SWeston Andros Adamson nfs_retry_commit(head, NULL, cinfo, 0); 17341da177e4SLinus Torvalds return -ENOMEM; 17351da177e4SLinus Torvalds } 17361da177e4SLinus Torvalds 173767911c8fSAnna Schumaker int nfs_commit_file(struct file *file, struct nfs_write_verifier *verf) 173867911c8fSAnna Schumaker { 173967911c8fSAnna Schumaker struct inode *inode = file_inode(file); 174067911c8fSAnna Schumaker struct nfs_open_context *open; 174167911c8fSAnna Schumaker struct nfs_commit_info cinfo; 174267911c8fSAnna Schumaker struct nfs_page *req; 174367911c8fSAnna Schumaker int ret; 174467911c8fSAnna Schumaker 174567911c8fSAnna Schumaker open = get_nfs_open_context(nfs_file_open_context(file)); 174667911c8fSAnna Schumaker req = nfs_create_request(open, NULL, NULL, 0, i_size_read(inode)); 17472997bfd0SDan Carpenter if (IS_ERR(req)) { 17482997bfd0SDan Carpenter ret = PTR_ERR(req); 174967911c8fSAnna Schumaker goto out_put; 175067911c8fSAnna Schumaker } 175167911c8fSAnna Schumaker 175267911c8fSAnna Schumaker nfs_init_cinfo_from_inode(&cinfo, inode); 175367911c8fSAnna Schumaker 175467911c8fSAnna Schumaker memcpy(&req->wb_verf, verf, sizeof(struct nfs_write_verifier)); 175567911c8fSAnna Schumaker nfs_request_add_commit_list(req, &cinfo); 175667911c8fSAnna Schumaker ret = nfs_commit_inode(inode, FLUSH_SYNC); 175767911c8fSAnna Schumaker if (ret > 0) 175867911c8fSAnna Schumaker ret = 0; 175967911c8fSAnna Schumaker 176067911c8fSAnna Schumaker nfs_free_request(req); 176167911c8fSAnna Schumaker out_put: 176267911c8fSAnna Schumaker put_nfs_open_context(open); 176367911c8fSAnna Schumaker return ret; 176467911c8fSAnna Schumaker } 176567911c8fSAnna Schumaker EXPORT_SYMBOL_GPL(nfs_commit_file); 176667911c8fSAnna Schumaker 17671da177e4SLinus Torvalds /* 17681da177e4SLinus Torvalds * COMMIT call returned 17691da177e4SLinus Torvalds */ 1770788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 17711da177e4SLinus Torvalds { 17720b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 17731da177e4SLinus Torvalds 1774a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 17751da177e4SLinus Torvalds task->tk_pid, task->tk_status); 17761da177e4SLinus Torvalds 1777788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1778c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1779c9d8f89dSTrond Myklebust } 1780c9d8f89dSTrond Myklebust 1781f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1782c9d8f89dSTrond Myklebust { 1783c9d8f89dSTrond Myklebust struct nfs_page *req; 1784c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1785f453a54aSFred Isaman struct nfs_commit_info cinfo; 1786353db796SNeilBrown struct nfs_server *nfss; 1787788e7a89STrond Myklebust 17881da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 17891da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 17901da177e4SLinus Torvalds nfs_list_remove_request(req); 179167911c8fSAnna Schumaker if (req->wb_page) 1792d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 17931da177e4SLinus Torvalds 17941e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 17953d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 17962b0143b5SDavid Howells (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)), 17971da177e4SLinus Torvalds req->wb_bytes, 17981da177e4SLinus Torvalds (long long)req_offset(req)); 1799c9d8f89dSTrond Myklebust if (status < 0) { 1800c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 18011da177e4SLinus Torvalds nfs_inode_remove_request(req); 1802c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 18031da177e4SLinus Torvalds goto next; 18041da177e4SLinus Torvalds } 18051da177e4SLinus Torvalds 18061da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 18071da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 18082f2c63bcSTrond Myklebust if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { 18091da177e4SLinus Torvalds /* We have a match */ 18101da177e4SLinus Torvalds nfs_inode_remove_request(req); 18111da177e4SLinus Torvalds dprintk(" OK\n"); 18121da177e4SLinus Torvalds goto next; 18131da177e4SLinus Torvalds } 18141da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 18151da177e4SLinus Torvalds dprintk(" mismatch\n"); 18166d884e8fSFred nfs_mark_request_dirty(req); 181705990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 18181da177e4SLinus Torvalds next: 18191d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18201da177e4SLinus Torvalds } 1821353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1822353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 1823353db796SNeilBrown clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 1824353db796SNeilBrown 1825f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1826af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18275917ce84SFred Isaman } 18285917ce84SFred Isaman 18295917ce84SFred Isaman static void nfs_commit_release(void *calldata) 18305917ce84SFred Isaman { 18310b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18325917ce84SFred Isaman 1833f453a54aSFred Isaman data->completion_ops->completion(data); 1834c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 18351da177e4SLinus Torvalds } 1836788e7a89STrond Myklebust 1837788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 18380b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1839788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1840788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1841788e7a89STrond Myklebust }; 18421da177e4SLinus Torvalds 1843f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1844f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1845b20135d0STrond Myklebust .resched_write = nfs_commit_resched_write, 1846f453a54aSFred Isaman }; 1847f453a54aSFred Isaman 18481763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1849ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 185084c53ab5SFred Isaman { 185184c53ab5SFred Isaman int status; 185284c53ab5SFred Isaman 1853ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 185484c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1855ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 185684c53ab5SFred Isaman return status; 185784c53ab5SFred Isaman } 185884c53ab5SFred Isaman 1859b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 18601da177e4SLinus Torvalds { 18611da177e4SLinus Torvalds LIST_HEAD(head); 1862ea2cf228SFred Isaman struct nfs_commit_info cinfo; 186371d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1864af7cf057STrond Myklebust int error = 0; 1865b8413f98STrond Myklebust int res; 18661da177e4SLinus Torvalds 1867ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1868af7cf057STrond Myklebust nfs_commit_begin(cinfo.mds); 1869ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 1870af7cf057STrond Myklebust if (res) 1871ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 1872af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18731da177e4SLinus Torvalds if (error < 0) 1874af7cf057STrond Myklebust goto out_error; 1875b8413f98STrond Myklebust if (!may_wait) 1876b8413f98STrond Myklebust goto out_mark_dirty; 1877af7cf057STrond Myklebust error = wait_on_commit(cinfo.mds); 1878b8413f98STrond Myklebust if (error < 0) 1879b8413f98STrond Myklebust return error; 1880c5efa5fcSTrond Myklebust return res; 1881af7cf057STrond Myklebust out_error: 1882af7cf057STrond Myklebust res = error; 1883c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1884c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1885c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1886c5efa5fcSTrond Myklebust * that the data is on the disk. 1887c5efa5fcSTrond Myklebust */ 1888c5efa5fcSTrond Myklebust out_mark_dirty: 1889c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 18901da177e4SLinus Torvalds return res; 18911da177e4SLinus Torvalds } 1892b20135d0STrond Myklebust EXPORT_SYMBOL_GPL(nfs_commit_inode); 18938fc795f7STrond Myklebust 1894ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 18958fc795f7STrond Myklebust { 1896420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1897420e3646STrond Myklebust int flags = FLUSH_SYNC; 1898420e3646STrond Myklebust int ret = 0; 18998fc795f7STrond Myklebust 19003236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1901ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 19023236c3e1SJeff Layton return ret; 19033236c3e1SJeff Layton 1904a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1905a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1906a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1907420e3646STrond Myklebust */ 1908cb1410c7SWeston Andros Adamson if (nfsi->commit_info.ncommit <= (nfsi->nrequests >> 1)) 1909420e3646STrond Myklebust goto out_mark_dirty; 1910420e3646STrond Myklebust 1911a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1912420e3646STrond Myklebust flags = 0; 1913a00dd6c0SJeff Layton } 1914a00dd6c0SJeff Layton 1915420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1916420e3646STrond Myklebust if (ret >= 0) { 1917420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1918420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1919420e3646STrond Myklebust wbc->nr_to_write -= ret; 1920420e3646STrond Myklebust else 1921420e3646STrond Myklebust wbc->nr_to_write = 0; 1922420e3646STrond Myklebust } 19238fc795f7STrond Myklebust return 0; 1924420e3646STrond Myklebust } 1925420e3646STrond Myklebust out_mark_dirty: 19268fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 19278fc795f7STrond Myklebust return ret; 19288fc795f7STrond Myklebust } 192989d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1930863a3c6cSAndy Adamson 1931acdc53b2STrond Myklebust /* 1932acdc53b2STrond Myklebust * flush the inode to disk. 1933acdc53b2STrond Myklebust */ 1934acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 193534901f70STrond Myklebust { 1936f4ce1299STrond Myklebust int ret; 193734901f70STrond Myklebust 1938f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1939f4ce1299STrond Myklebust 19405bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 19416b196875SChuck Lever if (ret) 19426b196875SChuck Lever goto out; 19435bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 19446b196875SChuck Lever if (ret < 0) 19456b196875SChuck Lever goto out; 19465bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 19476b196875SChuck Lever ret = 0; 1948f4ce1299STrond Myklebust 19496b196875SChuck Lever out: 1950f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1951f4ce1299STrond Myklebust return ret; 19521c75950bSTrond Myklebust } 1953ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 19541c75950bSTrond Myklebust 19551b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 19561b3b4a1aSTrond Myklebust { 19571b3b4a1aSTrond Myklebust struct nfs_page *req; 19581b3b4a1aSTrond Myklebust int ret = 0; 19591b3b4a1aSTrond Myklebust 1960ba8b06e6STrond Myklebust wait_on_page_writeback(page); 19613e217045SWeston Andros Adamson 19623e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 19633e217045SWeston Andros Adamson * request */ 19643e217045SWeston Andros Adamson req = nfs_lock_and_join_requests(page, false); 19653e217045SWeston Andros Adamson 19663e217045SWeston Andros Adamson if (IS_ERR(req)) { 19673e217045SWeston Andros Adamson ret = PTR_ERR(req); 19683e217045SWeston Andros Adamson } else if (req) { 19693e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 19703e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 19713e217045SWeston Andros Adamson * request from the inode / page_private pointer and 19723e217045SWeston Andros Adamson * release it */ 19731b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 19741d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 19751b3b4a1aSTrond Myklebust } 19763e217045SWeston Andros Adamson 19771b3b4a1aSTrond Myklebust return ret; 19781b3b4a1aSTrond Myklebust } 19791b3b4a1aSTrond Myklebust 19801c75950bSTrond Myklebust /* 19811c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 19821c75950bSTrond Myklebust */ 1983d6c843b9SPeng Tao int nfs_wb_single_page(struct inode *inode, struct page *page, bool launder) 19841c75950bSTrond Myklebust { 198529418aa4SMel Gorman loff_t range_start = page_file_offset(page); 198609cbfeafSKirill A. Shutemov loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); 19877f2f12d9STrond Myklebust struct writeback_control wbc = { 19887f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 19897f2f12d9STrond Myklebust .nr_to_write = 0, 19907f2f12d9STrond Myklebust .range_start = range_start, 19917f2f12d9STrond Myklebust .range_end = range_end, 19927f2f12d9STrond Myklebust }; 19937f2f12d9STrond Myklebust int ret; 19947f2f12d9STrond Myklebust 1995f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 1996f4ce1299STrond Myklebust 19970522f6adSTrond Myklebust for (;;) { 1998ba8b06e6STrond Myklebust wait_on_page_writeback(page); 19997f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 2000d6c843b9SPeng Tao ret = nfs_writepage_locked(page, &wbc, launder); 20017f2f12d9STrond Myklebust if (ret < 0) 20027f2f12d9STrond Myklebust goto out_error; 20030522f6adSTrond Myklebust continue; 20047f2f12d9STrond Myklebust } 2005f4ce1299STrond Myklebust ret = 0; 20060522f6adSTrond Myklebust if (!PagePrivate(page)) 20070522f6adSTrond Myklebust break; 20080522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20097f2f12d9STrond Myklebust if (ret < 0) 20107f2f12d9STrond Myklebust goto out_error; 20117f2f12d9STrond Myklebust } 20127f2f12d9STrond Myklebust out_error: 2013f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 20147f2f12d9STrond Myklebust return ret; 20151c75950bSTrond Myklebust } 20161c75950bSTrond Myklebust 2017074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 2018074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 2019a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 2020074cc1deSTrond Myklebust { 20212da95652SJeff Layton /* 20222da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 20232da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 20242da95652SJeff Layton * 20252da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 20262da95652SJeff Layton * that we can safely release the inode reference while holding 20272da95652SJeff Layton * the page lock. 20282da95652SJeff Layton */ 20292da95652SJeff Layton if (PagePrivate(page)) 20302da95652SJeff Layton return -EBUSY; 2031074cc1deSTrond Myklebust 20328c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 20338c209ce7SDavid Howells return -EBUSY; 2034074cc1deSTrond Myklebust 2035a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 2036074cc1deSTrond Myklebust } 2037074cc1deSTrond Myklebust #endif 2038074cc1deSTrond Myklebust 2039f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 20401da177e4SLinus Torvalds { 20411da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 20421e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 20431da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 204420c2df83SPaul Mundt NULL); 20451da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 20461da177e4SLinus Torvalds return -ENOMEM; 20471da177e4SLinus Torvalds 204893d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 20491da177e4SLinus Torvalds nfs_wdata_cachep); 20501da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 20513dd4765fSJeff Layton goto out_destroy_write_cache; 20521da177e4SLinus Torvalds 20530b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 20540b7c0153SFred Isaman sizeof(struct nfs_commit_data), 20550b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 20560b7c0153SFred Isaman NULL); 20570b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 20583dd4765fSJeff Layton goto out_destroy_write_mempool; 20590b7c0153SFred Isaman 206093d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 20614c100210SYanchuan Nian nfs_cdata_cachep); 20621da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 20633dd4765fSJeff Layton goto out_destroy_commit_cache; 20641da177e4SLinus Torvalds 206589a09141SPeter Zijlstra /* 206689a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 206789a09141SPeter Zijlstra * 206889a09141SPeter Zijlstra * 64MB: 8192k 206989a09141SPeter Zijlstra * 128MB: 11585k 207089a09141SPeter Zijlstra * 256MB: 16384k 207189a09141SPeter Zijlstra * 512MB: 23170k 207289a09141SPeter Zijlstra * 1GB: 32768k 207389a09141SPeter Zijlstra * 2GB: 46340k 207489a09141SPeter Zijlstra * 4GB: 65536k 207589a09141SPeter Zijlstra * 8GB: 92681k 207689a09141SPeter Zijlstra * 16GB: 131072k 207789a09141SPeter Zijlstra * 207889a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 207989a09141SPeter Zijlstra * Limit the default to 256M 208089a09141SPeter Zijlstra */ 208189a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 208289a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 208389a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 208489a09141SPeter Zijlstra 20851da177e4SLinus Torvalds return 0; 20863dd4765fSJeff Layton 20873dd4765fSJeff Layton out_destroy_commit_cache: 20883dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 20893dd4765fSJeff Layton out_destroy_write_mempool: 20903dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 20913dd4765fSJeff Layton out_destroy_write_cache: 20923dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 20933dd4765fSJeff Layton return -ENOMEM; 20941da177e4SLinus Torvalds } 20951da177e4SLinus Torvalds 2096266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 20971da177e4SLinus Torvalds { 20981da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 20993dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21001da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 21011a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 21021da177e4SLinus Torvalds } 21031da177e4SLinus Torvalds 21044a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 2105a4cdda59SAnna Schumaker .rw_mode = FMODE_WRITE, 21064a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 21074a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 21080eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 21090eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 21101ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 21114a0de55cSAnna Schumaker }; 2112