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; 2481da177e4SLinus Torvalds if (wbc->for_reclaim) 249494f74a2STrond Myklebust return FLUSH_HIGHPRI | FLUSH_COND_STABLE; 250e87b4c7aSNeilBrown if (wbc->sync_mode == WB_SYNC_ALL) 251e87b4c7aSNeilBrown ret = FLUSH_COND_STABLE; 252e87b4c7aSNeilBrown return ret; 2531da177e4SLinus Torvalds } 2541da177e4SLinus Torvalds 2551da177e4SLinus Torvalds /* 25689a09141SPeter Zijlstra * NFS congestion control 25789a09141SPeter Zijlstra */ 25889a09141SPeter Zijlstra 25989a09141SPeter Zijlstra int nfs_congestion_kb; 26089a09141SPeter Zijlstra 26189a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 26289a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 26389a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 26489a09141SPeter Zijlstra 265deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 26689a09141SPeter Zijlstra { 267deed85e7STrond Myklebust struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); 2685a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2695a6d41b3STrond Myklebust 270deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 27189a09141SPeter Zijlstra 272277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 2738aa7e847SJens Axboe NFS_CONGESTION_ON_THRESH) { 2748aa7e847SJens Axboe set_bdi_congested(&nfss->backing_dev_info, 2758aa7e847SJens Axboe BLK_RW_ASYNC); 2768aa7e847SJens Axboe } 27789a09141SPeter Zijlstra } 27889a09141SPeter Zijlstra 27920633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 28089a09141SPeter Zijlstra { 28120633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 28289a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 28389a09141SPeter Zijlstra 28420633f04SWeston Andros Adamson if (!nfs_page_group_sync_on_bit(req, PG_WB_END)) 28520633f04SWeston Andros Adamson return; 28620633f04SWeston Andros Adamson 28720633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 288c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 2898aa7e847SJens Axboe clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 29089a09141SPeter Zijlstra } 29189a09141SPeter Zijlstra 292d4581383SWeston Andros Adamson 293d4581383SWeston Andros Adamson /* nfs_page_group_clear_bits 294d4581383SWeston Andros Adamson * @req - an nfs request 295d4581383SWeston Andros Adamson * clears all page group related bits from @req 296d4581383SWeston Andros Adamson */ 297d4581383SWeston Andros Adamson static void 298d4581383SWeston Andros Adamson nfs_page_group_clear_bits(struct nfs_page *req) 299e261f51fSTrond Myklebust { 300d4581383SWeston Andros Adamson clear_bit(PG_TEARDOWN, &req->wb_flags); 301d4581383SWeston Andros Adamson clear_bit(PG_UNLOCKPAGE, &req->wb_flags); 302d4581383SWeston Andros Adamson clear_bit(PG_UPTODATE, &req->wb_flags); 303d4581383SWeston Andros Adamson clear_bit(PG_WB_END, &req->wb_flags); 304d4581383SWeston Andros Adamson clear_bit(PG_REMOVE, &req->wb_flags); 305d4581383SWeston Andros Adamson } 306d4581383SWeston Andros Adamson 307d4581383SWeston Andros Adamson 308d4581383SWeston Andros Adamson /* 309d4581383SWeston Andros Adamson * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req 310d4581383SWeston Andros Adamson * 311d4581383SWeston Andros Adamson * this is a helper function for nfs_lock_and_join_requests 312d4581383SWeston Andros Adamson * 313d4581383SWeston Andros Adamson * @inode - inode associated with request page group, must be holding inode lock 314d4581383SWeston Andros Adamson * @head - head request of page group, must be holding head lock 315d4581383SWeston Andros Adamson * @req - request that couldn't lock and needs to wait on the req bit lock 316d4581383SWeston Andros Adamson * @nonblock - if true, don't actually wait 317d4581383SWeston Andros Adamson * 318d4581383SWeston Andros Adamson * NOTE: this must be called holding page_group bit lock and inode spin lock 319d4581383SWeston Andros Adamson * and BOTH will be released before returning. 320d4581383SWeston Andros Adamson * 321d4581383SWeston Andros Adamson * returns 0 on success, < 0 on error. 322d4581383SWeston Andros Adamson */ 323d4581383SWeston Andros Adamson static int 324d4581383SWeston Andros Adamson nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head, 325d4581383SWeston Andros Adamson struct nfs_page *req, bool nonblock) 326d4581383SWeston Andros Adamson __releases(&inode->i_lock) 327d4581383SWeston Andros Adamson { 328d4581383SWeston Andros Adamson struct nfs_page *tmp; 329e261f51fSTrond Myklebust int ret; 330e261f51fSTrond Myklebust 331d4581383SWeston Andros Adamson /* relinquish all the locks successfully grabbed this run */ 332d4581383SWeston Andros Adamson for (tmp = head ; tmp != req; tmp = tmp->wb_this_page) 333d4581383SWeston Andros Adamson nfs_unlock_request(tmp); 334d4581383SWeston Andros Adamson 335d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags)); 336d4581383SWeston Andros Adamson 337d4581383SWeston Andros Adamson /* grab a ref on the request that will be waited on */ 338d4581383SWeston Andros Adamson kref_get(&req->wb_kref); 339d4581383SWeston Andros Adamson 340d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 341587142f8STrond Myklebust spin_unlock(&inode->i_lock); 342d4581383SWeston Andros Adamson 343d4581383SWeston Andros Adamson /* release ref from nfs_page_find_head_request_locked */ 344d4581383SWeston Andros Adamson nfs_release_request(head); 345d4581383SWeston Andros Adamson 346cfb506e1STrond Myklebust if (!nonblock) 347e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 348cfb506e1STrond Myklebust else 349cfb506e1STrond Myklebust ret = -EAGAIN; 350e261f51fSTrond Myklebust nfs_release_request(req); 351d4581383SWeston Andros Adamson 352d4581383SWeston Andros Adamson return ret; 353e261f51fSTrond Myklebust } 354d4581383SWeston Andros Adamson 355d4581383SWeston Andros Adamson /* 356d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 357d4581383SWeston Andros Adamson * 358d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 359d4581383SWeston Andros Adamson * @old_head - the old head of the list 360d4581383SWeston Andros Adamson * 361d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 362d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 363d4581383SWeston Andros Adamson * with a reference held by some other context. 364d4581383SWeston Andros Adamson */ 365d4581383SWeston Andros Adamson static void 366d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 367d4581383SWeston Andros Adamson struct nfs_page *old_head) 368d4581383SWeston Andros Adamson { 369d4581383SWeston Andros Adamson while (destroy_list) { 370d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 371d4581383SWeston Andros Adamson 372d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 373d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 374d4581383SWeston Andros Adamson 375d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 376d4581383SWeston Andros Adamson 377d4581383SWeston Andros Adamson /* make sure old group is not used */ 378d4581383SWeston Andros Adamson subreq->wb_head = subreq; 379d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 380d4581383SWeston Andros Adamson 381d4581383SWeston Andros Adamson /* subreq is now totally disconnected from page group or any 382d4581383SWeston Andros Adamson * write / commit lists. last chance to wake any waiters */ 383d4581383SWeston Andros Adamson nfs_unlock_request(subreq); 384d4581383SWeston Andros Adamson 385d4581383SWeston Andros Adamson if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) { 386d4581383SWeston Andros Adamson /* release ref on old head request */ 387d4581383SWeston Andros Adamson nfs_release_request(old_head); 388d4581383SWeston Andros Adamson 389d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 390d4581383SWeston Andros Adamson 391d4581383SWeston Andros Adamson /* release the PG_INODE_REF reference */ 392d4581383SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) 393d4581383SWeston Andros Adamson nfs_release_request(subreq); 394d4581383SWeston Andros Adamson else 395d4581383SWeston Andros Adamson WARN_ON_ONCE(1); 396d4581383SWeston Andros Adamson } else { 397d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags)); 398d4581383SWeston Andros Adamson /* zombie requests have already released the last 399d4581383SWeston Andros Adamson * reference and were waiting on the rest of the 400d4581383SWeston Andros Adamson * group to complete. Since it's no longer part of a 401d4581383SWeston Andros Adamson * group, simply free the request */ 402d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 403d4581383SWeston Andros Adamson nfs_free_request(subreq); 404d4581383SWeston Andros Adamson } 405d4581383SWeston Andros Adamson } 406d4581383SWeston Andros Adamson } 407d4581383SWeston Andros Adamson 408d4581383SWeston Andros Adamson /* 409d4581383SWeston Andros Adamson * nfs_lock_and_join_requests - join all subreqs to the head req and return 410d4581383SWeston Andros Adamson * a locked reference, cancelling any pending 411d4581383SWeston Andros Adamson * operations for this page. 412d4581383SWeston Andros Adamson * 413d4581383SWeston Andros Adamson * @page - the page used to lookup the "page group" of nfs_page structures 414d4581383SWeston Andros Adamson * @nonblock - if true, don't block waiting for request locks 415d4581383SWeston Andros Adamson * 416d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 417d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 418d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 419d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 420d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 421d4581383SWeston Andros Adamson * 422d4581383SWeston Andros Adamson * Returns a locked, referenced pointer to the head request - which after 423d4581383SWeston Andros Adamson * this call is guaranteed to be the only request associated with the page. 424d4581383SWeston Andros Adamson * Returns NULL if no requests are found for @page, or a ERR_PTR if an 425d4581383SWeston Andros Adamson * error was encountered. 426d4581383SWeston Andros Adamson */ 427d4581383SWeston Andros Adamson static struct nfs_page * 428d4581383SWeston Andros Adamson nfs_lock_and_join_requests(struct page *page, bool nonblock) 429d4581383SWeston Andros Adamson { 430d4581383SWeston Andros Adamson struct inode *inode = page_file_mapping(page)->host; 431d4581383SWeston Andros Adamson struct nfs_page *head, *subreq; 432d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 433d4581383SWeston Andros Adamson unsigned int total_bytes; 434d4581383SWeston Andros Adamson int ret; 435d4581383SWeston Andros Adamson 436d4581383SWeston Andros Adamson try_again: 437d4581383SWeston Andros Adamson total_bytes = 0; 438d4581383SWeston Andros Adamson 439d4581383SWeston Andros Adamson WARN_ON_ONCE(destroy_list); 440d4581383SWeston Andros Adamson 441d4581383SWeston Andros Adamson spin_lock(&inode->i_lock); 442d4581383SWeston Andros Adamson 443d4581383SWeston Andros Adamson /* 444d4581383SWeston Andros Adamson * A reference is taken only on the head request which acts as a 445d4581383SWeston Andros Adamson * reference to the whole page group - the group will not be destroyed 446d4581383SWeston Andros Adamson * until the head reference is released. 447d4581383SWeston Andros Adamson */ 448d4581383SWeston Andros Adamson head = nfs_page_find_head_request_locked(NFS_I(inode), page); 449d4581383SWeston Andros Adamson 450d4581383SWeston Andros Adamson if (!head) { 451587142f8STrond Myklebust spin_unlock(&inode->i_lock); 452d4581383SWeston Andros Adamson return NULL; 453d4581383SWeston Andros Adamson } 454d4581383SWeston Andros Adamson 4557c3af975SWeston Andros Adamson /* holding inode lock, so always make a non-blocking call to try the 4567c3af975SWeston Andros Adamson * page group lock */ 457fd2f3a06SWeston Andros Adamson ret = nfs_page_group_lock(head, true); 45894970014SWeston Andros Adamson if (ret < 0) { 45994970014SWeston Andros Adamson spin_unlock(&inode->i_lock); 4607c3af975SWeston Andros Adamson 4617c3af975SWeston Andros Adamson if (!nonblock && ret == -EAGAIN) { 4627c3af975SWeston Andros Adamson nfs_page_group_lock_wait(head); 4637c3af975SWeston Andros Adamson nfs_release_request(head); 4647c3af975SWeston Andros Adamson goto try_again; 4657c3af975SWeston Andros Adamson } 4667c3af975SWeston Andros Adamson 46794970014SWeston Andros Adamson nfs_release_request(head); 468e7029206SWeston Andros Adamson return ERR_PTR(ret); 46994970014SWeston Andros Adamson } 4707c3af975SWeston Andros Adamson 4717c3af975SWeston Andros Adamson /* lock each request in the page group */ 472d4581383SWeston Andros Adamson subreq = head; 473d4581383SWeston Andros Adamson do { 474d4581383SWeston Andros Adamson /* 475d4581383SWeston Andros Adamson * Subrequests are always contiguous, non overlapping 476309a1d65SWeston Andros Adamson * and in order - but may be repeated (mirrored writes). 477d4581383SWeston Andros Adamson */ 478309a1d65SWeston Andros Adamson if (subreq->wb_offset == (head->wb_offset + total_bytes)) { 479d4581383SWeston Andros Adamson /* keep track of how many bytes this group covers */ 480d4581383SWeston Andros Adamson total_bytes += subreq->wb_bytes; 481309a1d65SWeston Andros Adamson } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || 482309a1d65SWeston Andros Adamson ((subreq->wb_offset + subreq->wb_bytes) > 483309a1d65SWeston Andros Adamson (head->wb_offset + total_bytes)))) { 484309a1d65SWeston Andros Adamson nfs_page_group_unlock(head); 485309a1d65SWeston Andros Adamson spin_unlock(&inode->i_lock); 486309a1d65SWeston Andros Adamson return ERR_PTR(-EIO); 487309a1d65SWeston Andros Adamson } 488d4581383SWeston Andros Adamson 489d4581383SWeston Andros Adamson if (!nfs_lock_request(subreq)) { 490d4581383SWeston Andros Adamson /* releases page group bit lock and 491d4581383SWeston Andros Adamson * inode spin lock and all references */ 492d4581383SWeston Andros Adamson ret = nfs_unroll_locks_and_wait(inode, head, 493d4581383SWeston Andros Adamson subreq, nonblock); 494d4581383SWeston Andros Adamson 495d4581383SWeston Andros Adamson if (ret == 0) 496d4581383SWeston Andros Adamson goto try_again; 497d4581383SWeston Andros Adamson 498d4581383SWeston Andros Adamson return ERR_PTR(ret); 499d4581383SWeston Andros Adamson } 500d4581383SWeston Andros Adamson 501d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 502d4581383SWeston Andros Adamson } while (subreq != head); 503d4581383SWeston Andros Adamson 504d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 505d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 506d4581383SWeston Andros Adamson subreq = head; 507d4581383SWeston Andros Adamson do { 508411a99adSWeston Andros Adamson nfs_clear_request_commit(subreq); 509d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 510d4581383SWeston Andros Adamson } while (subreq != head); 511d4581383SWeston Andros Adamson 512d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 513d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 514d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 515d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 516d4581383SWeston Andros Adamson head->wb_this_page = head; 517d4581383SWeston Andros Adamson 518d4581383SWeston Andros Adamson /* change head request to cover whole range that 519d4581383SWeston Andros Adamson * the former page group covered */ 520d4581383SWeston Andros Adamson head->wb_bytes = total_bytes; 521d4581383SWeston Andros Adamson } 522d4581383SWeston Andros Adamson 523d4581383SWeston Andros Adamson /* 524d4581383SWeston Andros Adamson * prepare head request to be added to new pgio descriptor 525d4581383SWeston Andros Adamson */ 526d4581383SWeston Andros Adamson nfs_page_group_clear_bits(head); 527d4581383SWeston Andros Adamson 528d4581383SWeston Andros Adamson /* 529d4581383SWeston Andros Adamson * some part of the group was still on the inode list - otherwise 530d4581383SWeston Andros Adamson * the group wouldn't be involved in async write. 531d4581383SWeston Andros Adamson * grab a reference for the head request, iff it needs one. 532d4581383SWeston Andros Adamson */ 533d4581383SWeston Andros Adamson if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags)) 534d4581383SWeston Andros Adamson kref_get(&head->wb_kref); 535d4581383SWeston Andros Adamson 536d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 537d4581383SWeston Andros Adamson 538411a99adSWeston Andros Adamson /* drop lock to clean uprequests on destroy list */ 539d4581383SWeston Andros Adamson spin_unlock(&inode->i_lock); 540d4581383SWeston Andros Adamson 541d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(destroy_list, head); 542d4581383SWeston Andros Adamson 543d4581383SWeston Andros Adamson /* still holds ref on head from nfs_page_find_head_request_locked 544d4581383SWeston Andros Adamson * and still has lock on head from lock loop */ 545d4581383SWeston Andros Adamson return head; 546612c9384STrond Myklebust } 547074cc1deSTrond Myklebust 5480bcbf039SPeng Tao static void nfs_write_error_remove_page(struct nfs_page *req) 5490bcbf039SPeng Tao { 5500bcbf039SPeng Tao nfs_unlock_request(req); 5510bcbf039SPeng Tao nfs_end_page_writeback(req); 5520bcbf039SPeng Tao nfs_release_request(req); 5530bcbf039SPeng Tao generic_error_remove_page(page_file_mapping(req->wb_page), 5540bcbf039SPeng Tao req->wb_page); 5550bcbf039SPeng Tao } 5560bcbf039SPeng Tao 557074cc1deSTrond Myklebust /* 558074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 559074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 560074cc1deSTrond Myklebust */ 561074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 562d6c843b9SPeng Tao struct page *page, bool nonblock, 563d6c843b9SPeng Tao bool launder) 564074cc1deSTrond Myklebust { 565074cc1deSTrond Myklebust struct nfs_page *req; 566074cc1deSTrond Myklebust int ret = 0; 567074cc1deSTrond Myklebust 568d4581383SWeston Andros Adamson req = nfs_lock_and_join_requests(page, nonblock); 569074cc1deSTrond Myklebust if (!req) 570074cc1deSTrond Myklebust goto out; 571074cc1deSTrond Myklebust ret = PTR_ERR(req); 572074cc1deSTrond Myklebust if (IS_ERR(req)) 573074cc1deSTrond Myklebust goto out; 574074cc1deSTrond Myklebust 575deed85e7STrond Myklebust nfs_set_page_writeback(page); 576deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 577074cc1deSTrond Myklebust 578deed85e7STrond Myklebust ret = 0; 579f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 580074cc1deSTrond Myklebust ret = pgio->pg_error; 5810bcbf039SPeng Tao /* 582d6c843b9SPeng Tao * Remove the problematic req upon fatal errors 583d6c843b9SPeng Tao * in launder case, while other dirty pages can 584d6c843b9SPeng Tao * still be around until they get flushed. 5850bcbf039SPeng Tao */ 5860bcbf039SPeng Tao if (nfs_error_is_fatal(ret)) { 5870bcbf039SPeng Tao nfs_context_set_write_error(req->wb_context, ret); 588d6c843b9SPeng Tao if (launder) { 5890bcbf039SPeng Tao nfs_write_error_remove_page(req); 590d6c843b9SPeng Tao goto out; 591d6c843b9SPeng Tao } 592d6c843b9SPeng Tao } 5930bcbf039SPeng Tao nfs_redirty_request(req); 5940bcbf039SPeng Tao ret = -EAGAIN; 59540f90271STrond Myklebust } else 59640f90271STrond Myklebust nfs_add_stats(page_file_mapping(page)->host, 59740f90271STrond Myklebust NFSIOS_WRITEPAGES, 1); 598074cc1deSTrond Myklebust out: 599074cc1deSTrond Myklebust return ret; 600e261f51fSTrond Myklebust } 601e261f51fSTrond Myklebust 602d6c843b9SPeng Tao static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, 603d6c843b9SPeng Tao struct nfs_pageio_descriptor *pgio, bool launder) 604f758c885STrond Myklebust { 605cfb506e1STrond Myklebust int ret; 606f758c885STrond Myklebust 607d56b4ddfSMel Gorman nfs_pageio_cond_complete(pgio, page_file_index(page)); 608d6c843b9SPeng Tao ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE, 609d6c843b9SPeng Tao launder); 610cfb506e1STrond Myklebust if (ret == -EAGAIN) { 611cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 612cfb506e1STrond Myklebust ret = 0; 613cfb506e1STrond Myklebust } 614cfb506e1STrond Myklebust return ret; 615f758c885STrond Myklebust } 616f758c885STrond Myklebust 617e261f51fSTrond Myklebust /* 6181da177e4SLinus Torvalds * Write an mmapped page to the server. 6191da177e4SLinus Torvalds */ 620d6c843b9SPeng Tao static int nfs_writepage_locked(struct page *page, 621d6c843b9SPeng Tao struct writeback_control *wbc, 622d6c843b9SPeng Tao bool launder) 6231da177e4SLinus Torvalds { 624f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 62540f90271STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 626e261f51fSTrond Myklebust int err; 6271da177e4SLinus Torvalds 62840f90271STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 62940f90271STrond Myklebust nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), 630a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 631d6c843b9SPeng Tao err = nfs_do_writepage(page, wbc, &pgio, launder); 632f758c885STrond Myklebust nfs_pageio_complete(&pgio); 633f758c885STrond Myklebust if (err < 0) 6344d770ccfSTrond Myklebust return err; 635f758c885STrond Myklebust if (pgio.pg_error < 0) 636f758c885STrond Myklebust return pgio.pg_error; 637f758c885STrond Myklebust return 0; 6384d770ccfSTrond Myklebust } 6394d770ccfSTrond Myklebust 6404d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 6414d770ccfSTrond Myklebust { 642f758c885STrond Myklebust int ret; 6434d770ccfSTrond Myklebust 644d6c843b9SPeng Tao ret = nfs_writepage_locked(page, wbc, false); 6451da177e4SLinus Torvalds unlock_page(page); 646f758c885STrond Myklebust return ret; 647f758c885STrond Myklebust } 648f758c885STrond Myklebust 649f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 650f758c885STrond Myklebust { 651f758c885STrond Myklebust int ret; 652f758c885STrond Myklebust 653d6c843b9SPeng Tao ret = nfs_do_writepage(page, wbc, data, false); 654f758c885STrond Myklebust unlock_page(page); 655f758c885STrond Myklebust return ret; 6561da177e4SLinus Torvalds } 6571da177e4SLinus Torvalds 6581da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 6591da177e4SLinus Torvalds { 6601da177e4SLinus Torvalds struct inode *inode = mapping->host; 66172cb77f4STrond Myklebust unsigned long *bitlock = &NFS_I(inode)->flags; 662c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 6631da177e4SLinus Torvalds int err; 6641da177e4SLinus Torvalds 66572cb77f4STrond Myklebust /* Stop dirtying of new pages while we sync */ 66674316201SNeilBrown err = wait_on_bit_lock_action(bitlock, NFS_INO_FLUSHING, 66772cb77f4STrond Myklebust nfs_wait_bit_killable, TASK_KILLABLE); 66872cb77f4STrond Myklebust if (err) 66972cb77f4STrond Myklebust goto out_err; 67072cb77f4STrond Myklebust 67191d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 67291d5b470SChuck Lever 673a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 674a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 675f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 676c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 67772cb77f4STrond Myklebust 67872cb77f4STrond Myklebust clear_bit_unlock(NFS_INO_FLUSHING, bitlock); 6794e857c58SPeter Zijlstra smp_mb__after_atomic(); 68072cb77f4STrond Myklebust wake_up_bit(bitlock, NFS_INO_FLUSHING); 68172cb77f4STrond Myklebust 682f758c885STrond Myklebust if (err < 0) 68372cb77f4STrond Myklebust goto out_err; 68472cb77f4STrond Myklebust err = pgio.pg_error; 68572cb77f4STrond Myklebust if (err < 0) 68672cb77f4STrond Myklebust goto out_err; 687c63c7b05STrond Myklebust return 0; 68872cb77f4STrond Myklebust out_err: 68972cb77f4STrond Myklebust return err; 6901da177e4SLinus Torvalds } 6911da177e4SLinus Torvalds 6921da177e4SLinus Torvalds /* 6931da177e4SLinus Torvalds * Insert a write request into an inode 6941da177e4SLinus Torvalds */ 695d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 6961da177e4SLinus Torvalds { 6971da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 698e7d39069STrond Myklebust 6992bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 7002bfc6e56SWeston Andros Adamson 701e7d39069STrond Myklebust /* Lock the request! */ 7027ad84aa9STrond Myklebust nfs_lock_request(req); 703e7d39069STrond Myklebust 704e7d39069STrond Myklebust spin_lock(&inode->i_lock); 705cb1410c7SWeston Andros Adamson if (!nfsi->nrequests && 706cb1410c7SWeston Andros Adamson NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 707a9a4a87aSTrond Myklebust inode->i_version++; 70829418aa4SMel Gorman /* 70929418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 71029418aa4SMel Gorman * with invalidate/truncate. 71129418aa4SMel Gorman */ 71229418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 7132df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 714deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 715277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 71629418aa4SMel Gorman } 717cb1410c7SWeston Andros Adamson nfsi->nrequests++; 71817089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 719cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 720cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 721cb1410c7SWeston Andros Adamson * adding subrequests. */ 72217089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 723c03b4024STrond Myklebust kref_get(&req->wb_kref); 724e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 7251da177e4SLinus Torvalds } 7261da177e4SLinus Torvalds 7271da177e4SLinus Torvalds /* 72889a09141SPeter Zijlstra * Remove a write request from an inode 7291da177e4SLinus Torvalds */ 7301da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7311da177e4SLinus Torvalds { 7322b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 7331da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 73420633f04SWeston Andros Adamson struct nfs_page *head; 73520633f04SWeston Andros Adamson 73620633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 73720633f04SWeston Andros Adamson head = req->wb_head; 7381da177e4SLinus Torvalds 739587142f8STrond Myklebust spin_lock(&inode->i_lock); 74067911c8fSAnna Schumaker if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { 74120633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 74220633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 74395905446SNeilBrown smp_mb__after_atomic(); 74495905446SNeilBrown wake_up_page(head->wb_page, PG_private); 74520633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 74629418aa4SMel Gorman } 747cb1410c7SWeston Andros Adamson nfsi->nrequests--; 748cb1410c7SWeston Andros Adamson spin_unlock(&inode->i_lock); 749cb1410c7SWeston Andros Adamson } else { 750cb1410c7SWeston Andros Adamson spin_lock(&inode->i_lock); 751cb1410c7SWeston Andros Adamson nfsi->nrequests--; 752587142f8STrond Myklebust spin_unlock(&inode->i_lock); 75320633f04SWeston Andros Adamson } 75417089a29SWeston Andros Adamson 75517089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 7561da177e4SLinus Torvalds nfs_release_request(req); 7571da177e4SLinus Torvalds } 7581da177e4SLinus Torvalds 75961822ab5STrond Myklebust static void 7606d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 76161822ab5STrond Myklebust { 76267911c8fSAnna Schumaker if (req->wb_page) 76361822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 76461822ab5STrond Myklebust } 76561822ab5STrond Myklebust 7663a3908c8STrond Myklebust /* 7673a3908c8STrond Myklebust * nfs_page_search_commits_for_head_request_locked 7683a3908c8STrond Myklebust * 7693a3908c8STrond Myklebust * Search through commit lists on @inode for the head request for @page. 7703a3908c8STrond Myklebust * Must be called while holding the inode (which is cinfo) lock. 7713a3908c8STrond Myklebust * 7723a3908c8STrond Myklebust * Returns the head request if found, or NULL if not found. 7733a3908c8STrond Myklebust */ 7743a3908c8STrond Myklebust static struct nfs_page * 7753a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 7763a3908c8STrond Myklebust struct page *page) 7773a3908c8STrond Myklebust { 7783a3908c8STrond Myklebust struct nfs_page *freq, *t; 7793a3908c8STrond Myklebust struct nfs_commit_info cinfo; 7803a3908c8STrond Myklebust struct inode *inode = &nfsi->vfs_inode; 7813a3908c8STrond Myklebust 7823a3908c8STrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 7833a3908c8STrond Myklebust 7843a3908c8STrond Myklebust /* search through pnfs commit lists */ 7853a3908c8STrond Myklebust freq = pnfs_search_commit_reqs(inode, &cinfo, page); 7863a3908c8STrond Myklebust if (freq) 7873a3908c8STrond Myklebust return freq->wb_head; 7883a3908c8STrond Myklebust 7893a3908c8STrond Myklebust /* Linearly search the commit list for the correct request */ 7903a3908c8STrond Myklebust list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 7913a3908c8STrond Myklebust if (freq->wb_page == page) 7923a3908c8STrond Myklebust return freq->wb_head; 7933a3908c8STrond Myklebust } 7943a3908c8STrond Myklebust 7953a3908c8STrond Myklebust return NULL; 7963a3908c8STrond Myklebust } 7973a3908c8STrond Myklebust 7988dd37758STrond Myklebust /** 79986d80f97STrond Myklebust * nfs_request_add_commit_list_locked - add request to a commit list 80086d80f97STrond Myklebust * @req: pointer to a struct nfs_page 80186d80f97STrond Myklebust * @dst: commit list head 80286d80f97STrond Myklebust * @cinfo: holds list lock and accounting info 80386d80f97STrond Myklebust * 80486d80f97STrond Myklebust * This sets the PG_CLEAN bit, updates the cinfo count of 80586d80f97STrond Myklebust * number of outstanding requests requiring a commit as well as 80686d80f97STrond Myklebust * the MM page stats. 80786d80f97STrond Myklebust * 808fe238e60SDave Wysochanski * The caller must hold cinfo->inode->i_lock, and the nfs_page lock. 80986d80f97STrond Myklebust */ 81086d80f97STrond Myklebust void 81186d80f97STrond Myklebust nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, 81286d80f97STrond Myklebust struct nfs_commit_info *cinfo) 81386d80f97STrond Myklebust { 81486d80f97STrond Myklebust set_bit(PG_CLEAN, &req->wb_flags); 81586d80f97STrond Myklebust nfs_list_add_request(req, dst); 81686d80f97STrond Myklebust cinfo->mds->ncommit++; 81786d80f97STrond Myklebust } 81886d80f97STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); 81986d80f97STrond Myklebust 82086d80f97STrond Myklebust /** 8218dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 8228dd37758STrond Myklebust * @req: pointer to a struct nfs_page 823ea2cf228SFred Isaman * @dst: commit list head 824ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8258dd37758STrond Myklebust * 826ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 8278dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8288dd37758STrond Myklebust * the MM page stats. 8298dd37758STrond Myklebust * 830ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8318dd37758STrond Myklebust * holding the nfs_page lock. 8328dd37758STrond Myklebust */ 8338dd37758STrond Myklebust void 8346272dcc6SAnna Schumaker nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) 8358dd37758STrond Myklebust { 836fe238e60SDave Wysochanski spin_lock(&cinfo->inode->i_lock); 8376272dcc6SAnna Schumaker nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); 838fe238e60SDave Wysochanski spin_unlock(&cinfo->inode->i_lock); 83967911c8fSAnna Schumaker if (req->wb_page) 84086d80f97STrond Myklebust nfs_mark_page_unstable(req->wb_page, cinfo); 8418dd37758STrond Myklebust } 8428dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 8438dd37758STrond Myklebust 8448dd37758STrond Myklebust /** 8458dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 8468dd37758STrond Myklebust * @req: pointer to a nfs_page 847ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8488dd37758STrond Myklebust * 849ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 8508dd37758STrond Myklebust * number of outstanding requests requiring a commit 8518dd37758STrond Myklebust * It does not update the MM page stats. 8528dd37758STrond Myklebust * 853ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 8548dd37758STrond Myklebust */ 8558dd37758STrond Myklebust void 856ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 857ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 8588dd37758STrond Myklebust { 8598dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 8608dd37758STrond Myklebust return; 8618dd37758STrond Myklebust nfs_list_remove_request(req); 862ea2cf228SFred Isaman cinfo->mds->ncommit--; 8638dd37758STrond Myklebust } 8648dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 8658dd37758STrond Myklebust 866ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 867ea2cf228SFred Isaman struct inode *inode) 868ea2cf228SFred Isaman { 869fe238e60SDave Wysochanski cinfo->inode = inode; 870ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 871ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 872b359f9d0SFred Isaman cinfo->dreq = NULL; 873f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 874ea2cf228SFred Isaman } 875ea2cf228SFred Isaman 876ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 877ea2cf228SFred Isaman struct inode *inode, 878ea2cf228SFred Isaman struct nfs_direct_req *dreq) 879ea2cf228SFred Isaman { 8801763da12SFred Isaman if (dreq) 8811763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 8821763da12SFred Isaman else 883ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 884ea2cf228SFred Isaman } 885ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 8868dd37758STrond Myklebust 8871da177e4SLinus Torvalds /* 8881da177e4SLinus Torvalds * Add a request to the inode's commit list. 8891da177e4SLinus Torvalds */ 8901763da12SFred Isaman void 891ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 892b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 8931da177e4SLinus Torvalds { 894b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 8958dd37758STrond Myklebust return; 8966272dcc6SAnna Schumaker nfs_request_add_commit_list(req, cinfo); 8971da177e4SLinus Torvalds } 8988e821cadSTrond Myklebust 899d6d6dc7cSFred Isaman static void 900d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 901e468bae9STrond Myklebust { 902e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 90393f78d88STejun Heo dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, 90493f78d88STejun Heo WB_RECLAIMABLE); 905e468bae9STrond Myklebust } 906d6d6dc7cSFred Isaman 907411a99adSWeston Andros Adamson /* Called holding inode (/cinfo) lock */ 9088dd37758STrond Myklebust static void 909d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 910d6d6dc7cSFred Isaman { 9118dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 9122b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 913ea2cf228SFred Isaman struct nfs_commit_info cinfo; 914d6d6dc7cSFred Isaman 915ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 916ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 917ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 918d6d6dc7cSFred Isaman } 9198dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 9208dd37758STrond Myklebust } 921e468bae9STrond Myklebust } 922e468bae9STrond Myklebust 923d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 9248e821cadSTrond Myklebust { 925c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 926d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 927c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 9288e821cadSTrond Myklebust } 9298e821cadSTrond Myklebust 930061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 9316c75dc0dSFred Isaman { 932ea2cf228SFred Isaman struct nfs_commit_info cinfo; 9336c75dc0dSFred Isaman unsigned long bytes = 0; 9346c75dc0dSFred Isaman 9356c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 9366c75dc0dSFred Isaman goto out; 937ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 9386c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 9396c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 9406c75dc0dSFred Isaman 9416c75dc0dSFred Isaman bytes += req->wb_bytes; 9426c75dc0dSFred Isaman nfs_list_remove_request(req); 9436c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 9446c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 945d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 9466c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 9476c75dc0dSFred Isaman goto remove_req; 9486c75dc0dSFred Isaman } 949c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 950f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 951b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 952a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 9536c75dc0dSFred Isaman goto next; 9546c75dc0dSFred Isaman } 9556c75dc0dSFred Isaman remove_req: 9566c75dc0dSFred Isaman nfs_inode_remove_request(req); 9576c75dc0dSFred Isaman next: 9581d1afcbcSTrond Myklebust nfs_unlock_request(req); 95920633f04SWeston Andros Adamson nfs_end_page_writeback(req); 9603aff4ebbSTrond Myklebust nfs_release_request(req); 9616c75dc0dSFred Isaman } 9626c75dc0dSFred Isaman out: 9636c75dc0dSFred Isaman hdr->release(hdr); 9646c75dc0dSFred Isaman } 9656c75dc0dSFred Isaman 966ce59515cSAnna Schumaker unsigned long 967ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 968fb8a1f11STrond Myklebust { 969ea2cf228SFred Isaman return cinfo->mds->ncommit; 970fb8a1f11STrond Myklebust } 971fb8a1f11STrond Myklebust 972fe238e60SDave Wysochanski /* cinfo->inode->i_lock held by caller */ 9731763da12SFred Isaman int 974ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 975ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 976d6d6dc7cSFred Isaman { 977d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 978d6d6dc7cSFred Isaman int ret = 0; 979d6d6dc7cSFred Isaman 980d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 9818dd37758STrond Myklebust if (!nfs_lock_request(req)) 9828dd37758STrond Myklebust continue; 9837ad84aa9STrond Myklebust kref_get(&req->wb_kref); 984fe238e60SDave Wysochanski if (cond_resched_lock(&cinfo->inode->i_lock)) 9853b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 986ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 9878dd37758STrond Myklebust nfs_list_add_request(req, dst); 988d6d6dc7cSFred Isaman ret++; 9891763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 990d6d6dc7cSFred Isaman break; 991d6d6dc7cSFred Isaman } 992d6d6dc7cSFred Isaman return ret; 993d6d6dc7cSFred Isaman } 994d6d6dc7cSFred Isaman 9951da177e4SLinus Torvalds /* 9961da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 9971da177e4SLinus Torvalds * @inode: NFS inode to scan 998ea2cf228SFred Isaman * @dst: mds destination list 999ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 10001da177e4SLinus Torvalds * 10011da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 10021da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 10031da177e4SLinus Torvalds */ 10041763da12SFred Isaman int 1005ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 1006ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 10071da177e4SLinus Torvalds { 1008d6d6dc7cSFred Isaman int ret = 0; 1009fb8a1f11STrond Myklebust 1010fe238e60SDave Wysochanski spin_lock(&cinfo->inode->i_lock); 1011ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 10128dd37758STrond Myklebust const int max = INT_MAX; 1013d6d6dc7cSFred Isaman 1014ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 1015ea2cf228SFred Isaman cinfo, max); 1016ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 1017d6d6dc7cSFred Isaman } 1018fe238e60SDave Wysochanski spin_unlock(&cinfo->inode->i_lock); 1019ff778d02STrond Myklebust return ret; 10201da177e4SLinus Torvalds } 1021d6d6dc7cSFred Isaman 10221da177e4SLinus Torvalds /* 1023e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1024e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 10251da177e4SLinus Torvalds * 1026e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1027e7d39069STrond Myklebust * to disk. 10281da177e4SLinus Torvalds */ 1029e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 1030e7d39069STrond Myklebust struct page *page, 1031e7d39069STrond Myklebust unsigned int offset, 1032e7d39069STrond Myklebust unsigned int bytes) 10331da177e4SLinus Torvalds { 1034e7d39069STrond Myklebust struct nfs_page *req; 1035e7d39069STrond Myklebust unsigned int rqend; 1036e7d39069STrond Myklebust unsigned int end; 10371da177e4SLinus Torvalds int error; 1038277459d2STrond Myklebust 1039e7d39069STrond Myklebust if (!PagePrivate(page)) 1040e7d39069STrond Myklebust return NULL; 1041e7d39069STrond Myklebust 1042e7d39069STrond Myklebust end = offset + bytes; 1043e7d39069STrond Myklebust spin_lock(&inode->i_lock); 1044e7d39069STrond Myklebust 1045e7d39069STrond Myklebust for (;;) { 104684d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 1047e7d39069STrond Myklebust if (req == NULL) 1048e7d39069STrond Myklebust goto out_unlock; 1049e7d39069STrond Myklebust 10502bfc6e56SWeston Andros Adamson /* should be handled by nfs_flush_incompatible */ 10512bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 10522bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 10532bfc6e56SWeston Andros Adamson 1054e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1055e7d39069STrond Myklebust /* 1056e7d39069STrond Myklebust * Tell the caller to flush out the request if 1057e7d39069STrond Myklebust * the offsets are non-contiguous. 1058e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1059e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1060e7d39069STrond Myklebust */ 1061e468bae9STrond Myklebust if (offset > rqend 1062e7d39069STrond Myklebust || end < req->wb_offset) 1063e7d39069STrond Myklebust goto out_flushme; 1064e7d39069STrond Myklebust 10657ad84aa9STrond Myklebust if (nfs_lock_request(req)) 1066e7d39069STrond Myklebust break; 1067e7d39069STrond Myklebust 1068e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 1069587142f8STrond Myklebust spin_unlock(&inode->i_lock); 10701da177e4SLinus Torvalds error = nfs_wait_on_request(req); 10711da177e4SLinus Torvalds nfs_release_request(req); 1072e7d39069STrond Myklebust if (error != 0) 1073e7d39069STrond Myklebust goto out_err; 1074e7d39069STrond Myklebust spin_lock(&inode->i_lock); 10751da177e4SLinus Torvalds } 10761da177e4SLinus Torvalds 10771da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 10781da177e4SLinus Torvalds if (offset < req->wb_offset) { 10791da177e4SLinus Torvalds req->wb_offset = offset; 10801da177e4SLinus Torvalds req->wb_pgbase = offset; 10811da177e4SLinus Torvalds } 10821da177e4SLinus Torvalds if (end > rqend) 10831da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1084e7d39069STrond Myklebust else 1085e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 1086e7d39069STrond Myklebust out_unlock: 1087ca138f36SFred Isaman if (req) 10888dd37758STrond Myklebust nfs_clear_request_commit(req); 1089411a99adSWeston Andros Adamson spin_unlock(&inode->i_lock); 1090e7d39069STrond Myklebust return req; 1091e7d39069STrond Myklebust out_flushme: 1092e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 1093e7d39069STrond Myklebust nfs_release_request(req); 1094e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1095e7d39069STrond Myklebust out_err: 1096e7d39069STrond Myklebust return ERR_PTR(error); 1097e7d39069STrond Myklebust } 10981da177e4SLinus Torvalds 1099e7d39069STrond Myklebust /* 1100e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1101e7d39069STrond Myklebust * 1102e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1103e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1104e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1105e7d39069STrond Myklebust */ 1106e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1107e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1108e7d39069STrond Myklebust { 1109d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1110e7d39069STrond Myklebust struct nfs_page *req; 1111e7d39069STrond Myklebust 1112e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1113e7d39069STrond Myklebust if (req != NULL) 1114e7d39069STrond Myklebust goto out; 11152bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 1116e7d39069STrond Myklebust if (IS_ERR(req)) 1117e7d39069STrond Myklebust goto out; 1118d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1119efc91ed0STrond Myklebust out: 112061e930a9STrond Myklebust return req; 11211da177e4SLinus Torvalds } 11221da177e4SLinus Torvalds 1123e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1124e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1125e7d39069STrond Myklebust { 1126e7d39069STrond Myklebust struct nfs_page *req; 1127e7d39069STrond Myklebust 1128e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1129e7d39069STrond Myklebust if (IS_ERR(req)) 1130e7d39069STrond Myklebust return PTR_ERR(req); 1131e7d39069STrond Myklebust /* Update file length */ 1132e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1133d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1134a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11351d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1136e7d39069STrond Myklebust return 0; 1137e7d39069STrond Myklebust } 1138e7d39069STrond Myklebust 11391da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 11401da177e4SLinus Torvalds { 1141cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 11422a369153STrond Myklebust struct nfs_lock_context *l_ctx; 1143bd61e0a9SJeff Layton struct file_lock_context *flctx = file_inode(file)->i_flctx; 11441da177e4SLinus Torvalds struct nfs_page *req; 11451a54533eSTrond Myklebust int do_flush, status; 11461da177e4SLinus Torvalds /* 11471da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 11481da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 11491da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 11501da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11511da177e4SLinus Torvalds * Also do the same if we find a request from an existing 11521da177e4SLinus Torvalds * dropped page. 11531da177e4SLinus Torvalds */ 11541a54533eSTrond Myklebust do { 115584d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 11561a54533eSTrond Myklebust if (req == NULL) 11571a54533eSTrond Myklebust return 0; 11582a369153STrond Myklebust l_ctx = req->wb_lock_context; 1159138a2935STrond Myklebust do_flush = req->wb_page != page || 1160138a2935STrond Myklebust !nfs_match_open_context(req->wb_context, ctx); 11612bfc6e56SWeston Andros Adamson /* for now, flush if more than 1 request in page_group */ 11622bfc6e56SWeston Andros Adamson do_flush |= req->wb_this_page != req; 1163bd61e0a9SJeff Layton if (l_ctx && flctx && 1164bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1165bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 11665263e31eSJeff Layton do_flush |= l_ctx->lockowner.l_owner != current->files 11675263e31eSJeff Layton || l_ctx->lockowner.l_pid != current->tgid; 11685263e31eSJeff Layton } 11691da177e4SLinus Torvalds nfs_release_request(req); 11701a54533eSTrond Myklebust if (!do_flush) 11711a54533eSTrond Myklebust return 0; 1172d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 11731a54533eSTrond Myklebust } while (status == 0); 11741a54533eSTrond Myklebust return status; 11751da177e4SLinus Torvalds } 11761da177e4SLinus Torvalds 11771da177e4SLinus Torvalds /* 1178dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1179dc24826bSAndy Adamson * expire soon. 1180dc24826bSAndy Adamson * 1181dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1182dc24826bSAndy Adamson * 1183dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1184dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1185dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1186dc24826bSAndy Adamson */ 1187dc24826bSAndy Adamson int 1188dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1189dc24826bSAndy Adamson { 1190dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1191dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1192dc24826bSAndy Adamson 1193dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 1194dc24826bSAndy Adamson } 1195dc24826bSAndy Adamson 1196dc24826bSAndy Adamson /* 1197dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1198dc24826bSAndy Adamson */ 1199dc24826bSAndy Adamson bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) 1200dc24826bSAndy Adamson { 1201dc24826bSAndy Adamson return rpcauth_cred_key_to_expire(ctx->cred); 1202dc24826bSAndy Adamson } 1203dc24826bSAndy Adamson 1204dc24826bSAndy Adamson /* 12055d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 12065d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 12075d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 12085d47a356STrond Myklebust */ 12098d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 12105d47a356STrond Myklebust { 1211d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1212d529ef83SJeff Layton 12138d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 12148d197a56STrond Myklebust goto out; 121518dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1216d529ef83SJeff Layton return false; 12174db72b40SJeff Layton smp_rmb(); 1218d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 12198d197a56STrond Myklebust return false; 12208d197a56STrond Myklebust out: 122118dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 122218dd78c4SScott Mayhew return false; 12238d197a56STrond Myklebust return PageUptodate(page) != 0; 12245d47a356STrond Myklebust } 12255d47a356STrond Myklebust 12265263e31eSJeff Layton static bool 12275263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 12285263e31eSJeff Layton { 12295263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 12305263e31eSJeff Layton fl->fl_type == F_WRLCK; 12315263e31eSJeff Layton } 12325263e31eSJeff Layton 1233c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1234c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1235c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1236c7559663SScott Mayhew * inefficiencies. 1237c7559663SScott Mayhew * 1238263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1239263b4509SScott Mayhew * of the checks. 1240c7559663SScott Mayhew */ 1241c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1242c7559663SScott Mayhew { 12435263e31eSJeff Layton int ret; 12445263e31eSJeff Layton struct file_lock_context *flctx = inode->i_flctx; 12455263e31eSJeff Layton struct file_lock *fl; 12465263e31eSJeff Layton 1247c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1248c7559663SScott Mayhew return 0; 1249263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1250263b4509SScott Mayhew return 0; 1251c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1252c7559663SScott Mayhew return 1; 1253bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1254bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 12558fa4592aSTrond Myklebust return 1; 12565263e31eSJeff Layton 12575263e31eSJeff Layton /* Check to see if there are whole file write locks */ 12585263e31eSJeff Layton ret = 0; 12596109c850SJeff Layton spin_lock(&flctx->flc_lock); 1260bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1261bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1262bd61e0a9SJeff Layton fl_list); 1263bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 12645263e31eSJeff Layton ret = 1; 1265bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 12665263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 12675263e31eSJeff Layton fl_list); 12685263e31eSJeff Layton if (fl->fl_type == F_WRLCK) 12695263e31eSJeff Layton ret = 1; 12705263e31eSJeff Layton } 12716109c850SJeff Layton spin_unlock(&flctx->flc_lock); 12725263e31eSJeff Layton return ret; 1273c7559663SScott Mayhew } 1274c7559663SScott Mayhew 12755d47a356STrond Myklebust /* 12761da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 12771da177e4SLinus Torvalds * 12781da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 12791da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 12801da177e4SLinus Torvalds */ 12811da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 12821da177e4SLinus Torvalds unsigned int offset, unsigned int count) 12831da177e4SLinus Torvalds { 1284cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1285d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 12861da177e4SLinus Torvalds int status = 0; 12871da177e4SLinus Torvalds 128891d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 128991d5b470SChuck Lever 12906de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 12916de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 12921da177e4SLinus Torvalds 1293c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 129449a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 12951da177e4SLinus Torvalds offset = 0; 12961da177e4SLinus Torvalds } 12971da177e4SLinus Torvalds 1298e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 129903fa9e84STrond Myklebust if (status < 0) 130003fa9e84STrond Myklebust nfs_set_pageerror(page); 130159b7c05fSTrond Myklebust else 130259b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 13031da177e4SLinus Torvalds 130448186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 13051da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 13061da177e4SLinus Torvalds return status; 13071da177e4SLinus Torvalds } 13081da177e4SLinus Torvalds 13093ff7576dSTrond Myklebust static int flush_task_priority(int how) 13101da177e4SLinus Torvalds { 13111da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 13121da177e4SLinus Torvalds case FLUSH_HIGHPRI: 13131da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 13141da177e4SLinus Torvalds case FLUSH_LOWPRI: 13151da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 13161da177e4SLinus Torvalds } 13171da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 13181da177e4SLinus Torvalds } 13191da177e4SLinus Torvalds 1320d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1321d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1322abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 13231ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 13241da177e4SLinus Torvalds { 13253ff7576dSTrond Myklebust int priority = flush_task_priority(how); 13261da177e4SLinus Torvalds 13271ed26f33SAnna Schumaker task_setup_data->priority = priority; 1328abde71f4STom Haynes rpc_ops->write_setup(hdr, msg); 1329d138d5d1SAndy Adamson 1330abde71f4STom Haynes nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client, 1331d45f60c6SWeston Andros Adamson &task_setup_data->rpc_client, msg, hdr); 1332275acaafSTrond Myklebust } 1333275acaafSTrond Myklebust 13346d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 13356d884e8fSFred * call this on each, which will prepare them to be retried on next 13366d884e8fSFred * writeback using standard nfs. 13376d884e8fSFred */ 13386d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 13396d884e8fSFred { 13406d884e8fSFred nfs_mark_request_dirty(req); 1341c7070113STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 13421d1afcbcSTrond Myklebust nfs_unlock_request(req); 134320633f04SWeston Andros Adamson nfs_end_page_writeback(req); 13443aff4ebbSTrond Myklebust nfs_release_request(req); 13456d884e8fSFred } 13466d884e8fSFred 1347061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 13486c75dc0dSFred Isaman { 13496c75dc0dSFred Isaman struct nfs_page *req; 13506c75dc0dSFred Isaman 13516c75dc0dSFred Isaman while (!list_empty(head)) { 13526c75dc0dSFred Isaman req = nfs_list_entry(head->next); 13536c75dc0dSFred Isaman nfs_list_remove_request(req); 13546c75dc0dSFred Isaman nfs_redirty_request(req); 13556c75dc0dSFred Isaman } 13566c75dc0dSFred Isaman } 13576c75dc0dSFred Isaman 1358dc602dd7STrond Myklebust static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) 1359dc602dd7STrond Myklebust { 1360dc602dd7STrond Myklebust nfs_async_write_error(&hdr->pages); 1361dc602dd7STrond Myklebust } 1362dc602dd7STrond Myklebust 1363061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1364061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1365061ae2edSFred Isaman .completion = nfs_write_completion, 1366dc602dd7STrond Myklebust .reschedule_io = nfs_async_write_reschedule_io, 1367061ae2edSFred Isaman }; 1368061ae2edSFred Isaman 136957208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1370a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1371061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 13721751c363STrond Myklebust { 1373a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 137441d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1375a20c93e3SChristoph Hellwig 1376a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1377a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1378a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1379a20c93e3SChristoph Hellwig #endif 13804a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 13814a0de55cSAnna Schumaker server->wsize, ioflags); 13821751c363STrond Myklebust } 1383ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 13841751c363STrond Myklebust 1385dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1386dce81290STrond Myklebust { 1387a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1388a7d42ddbSWeston Andros Adamson 13896f29b9bbSKinglong Mee if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) 13906f29b9bbSKinglong Mee pgio->pg_ops->pg_cleanup(pgio); 13916f29b9bbSKinglong Mee 139241d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1393a7d42ddbSWeston Andros Adamson 1394a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1395a7d42ddbSWeston Andros Adamson 1396a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1397a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1398dce81290STrond Myklebust } 13991f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1400dce81290STrond Myklebust 14011da177e4SLinus Torvalds 14020b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 14030b7c0153SFred Isaman { 14040b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 14050b7c0153SFred Isaman 14060b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 14070b7c0153SFred Isaman } 14080b7c0153SFred Isaman 14091f2edbe3STrond Myklebust /* 14101f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 14111f2edbe3STrond Myklebust */ 14121f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 14131f2edbe3STrond Myklebust { 14141f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 14151f2edbe3STrond Myklebust int kill = 0; 1416788e7a89STrond Myklebust 14171f2edbe3STrond Myklebust /* suid always must be killed */ 14181f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 14191f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 14201f2edbe3STrond Myklebust 14211f2edbe3STrond Myklebust /* 14221f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 14231f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 14241f2edbe3STrond Myklebust */ 14251f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 14261f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 14271f2edbe3STrond Myklebust 14281f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 14291f2edbe3STrond Myklebust return kill; 14301f2edbe3STrond Myklebust 14311f2edbe3STrond Myklebust return 0; 14321f2edbe3STrond Myklebust } 1433788e7a89STrond Myklebust 1434a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1435a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1436a08a8cd3STrond Myklebust { 1437a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1438a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 14392b83d3deSTrond Myklebust u64 size = argp->offset + resp->count; 1440a08a8cd3STrond Myklebust 1441a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 14422b83d3deSTrond Myklebust fattr->size = size; 14432b83d3deSTrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { 14442b83d3deSTrond Myklebust fattr->valid &= ~NFS_ATTR_FATTR_SIZE; 1445a08a8cd3STrond Myklebust return; 14462b83d3deSTrond Myklebust } 14472b83d3deSTrond Myklebust if (size != fattr->size) 1448a08a8cd3STrond Myklebust return; 1449a08a8cd3STrond Myklebust /* Set attribute barrier */ 1450a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 14512b83d3deSTrond Myklebust /* ...and update size */ 14522b83d3deSTrond Myklebust fattr->valid |= NFS_ATTR_FATTR_SIZE; 1453a08a8cd3STrond Myklebust } 1454a08a8cd3STrond Myklebust 1455a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1456a08a8cd3STrond Myklebust { 14572b83d3deSTrond Myklebust struct nfs_fattr *fattr = &hdr->fattr; 1458a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1459a08a8cd3STrond Myklebust 1460a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1461a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1462a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1463a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1464a08a8cd3STrond Myklebust } 1465a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1466a08a8cd3STrond Myklebust 14671da177e4SLinus Torvalds /* 14681da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 14691da177e4SLinus Torvalds */ 1470d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1471d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 14720eecb214SAnna Schumaker struct inode *inode) 14731da177e4SLinus Torvalds { 1474788e7a89STrond Myklebust int status; 14751da177e4SLinus Torvalds 1476f551e44fSChuck Lever /* 1477f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1478f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1479f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1480f551e44fSChuck Lever * another writer had changed the file, but some applications 1481f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1482f551e44fSChuck Lever */ 1483d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1484788e7a89STrond Myklebust if (status != 0) 14850eecb214SAnna Schumaker return status; 1486d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 148791d5b470SChuck Lever 1488d45f60c6SWeston Andros Adamson if (hdr->res.verf->committed < hdr->args.stable && 1489d45f60c6SWeston Andros Adamson task->tk_status >= 0) { 14901da177e4SLinus Torvalds /* We tried a write call, but the server did not 14911da177e4SLinus Torvalds * commit data to stable storage even though we 14921da177e4SLinus Torvalds * requested it. 14931da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 14941da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 14951da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 14961da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 14971da177e4SLinus Torvalds */ 14981da177e4SLinus Torvalds static unsigned long complain; 14991da177e4SLinus Torvalds 1500a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 15011da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 15021da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 15031da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1504cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1505d45f60c6SWeston Andros Adamson hdr->res.verf->committed, hdr->args.stable); 15061da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 15071da177e4SLinus Torvalds } 15081da177e4SLinus Torvalds } 15091f2edbe3STrond Myklebust 15101f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 15111f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 15121f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 15130eecb214SAnna Schumaker return 0; 15140eecb214SAnna Schumaker } 15150eecb214SAnna Schumaker 15160eecb214SAnna Schumaker /* 15170eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 15180eecb214SAnna Schumaker */ 1519d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1520d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 15210eecb214SAnna Schumaker { 1522d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1523d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 15241f2edbe3STrond Myklebust 15251f2edbe3STrond Myklebust if (resp->count < argp->count) { 15261da177e4SLinus Torvalds static unsigned long complain; 15271da177e4SLinus Torvalds 15286c75dc0dSFred Isaman /* This a short write! */ 1529d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 153091d5b470SChuck Lever 15311da177e4SLinus Torvalds /* Has the server at least made some progress? */ 15326c75dc0dSFred Isaman if (resp->count == 0) { 15336c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 15346c75dc0dSFred Isaman printk(KERN_WARNING 15356c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 15366c75dc0dSFred Isaman argp->count); 15376c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 15386c75dc0dSFred Isaman } 1539d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 15406c75dc0dSFred Isaman task->tk_status = -EIO; 15416c75dc0dSFred Isaman return; 15426c75dc0dSFred Isaman } 1543f8417b48SKinglong Mee 1544f8417b48SKinglong Mee /* For non rpc-based layout drivers, retry-through-MDS */ 1545f8417b48SKinglong Mee if (!task->tk_ops) { 1546f8417b48SKinglong Mee hdr->pnfs_error = -EAGAIN; 1547f8417b48SKinglong Mee return; 1548f8417b48SKinglong Mee } 1549f8417b48SKinglong Mee 15501da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 15511da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 15521da177e4SLinus Torvalds /* Resend from where the server left off */ 1553d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 15541da177e4SLinus Torvalds argp->offset += resp->count; 15551da177e4SLinus Torvalds argp->pgbase += resp->count; 15561da177e4SLinus Torvalds argp->count -= resp->count; 15571da177e4SLinus Torvalds } else { 15581da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 15591da177e4SLinus Torvalds * headaches in the case of a server crash. 15601da177e4SLinus Torvalds */ 15611da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 15621da177e4SLinus Torvalds } 1563d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 15641da177e4SLinus Torvalds } 15651da177e4SLinus Torvalds } 15661da177e4SLinus Torvalds 1567af7cf057STrond Myklebust static int wait_on_commit(struct nfs_mds_commit_info *cinfo) 156871d0a611STrond Myklebust { 1569af7cf057STrond Myklebust return wait_on_atomic_t(&cinfo->rpcs_out, 1570af7cf057STrond Myklebust nfs_wait_atomic_killable, TASK_KILLABLE); 1571af7cf057STrond Myklebust } 1572af7cf057STrond Myklebust 1573af7cf057STrond Myklebust static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) 1574af7cf057STrond Myklebust { 1575af7cf057STrond Myklebust atomic_inc(&cinfo->rpcs_out); 1576af7cf057STrond Myklebust } 1577af7cf057STrond Myklebust 1578af7cf057STrond Myklebust static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) 1579af7cf057STrond Myklebust { 1580af7cf057STrond Myklebust if (atomic_dec_and_test(&cinfo->rpcs_out)) 1581af7cf057STrond Myklebust wake_up_atomic_t(&cinfo->rpcs_out); 158271d0a611STrond Myklebust } 158371d0a611STrond Myklebust 15840b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 15851da177e4SLinus Torvalds { 15860b7c0153SFred Isaman put_nfs_open_context(data->context); 15870b7c0153SFred Isaman nfs_commit_free(data); 15881da177e4SLinus Torvalds } 1589e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 15901da177e4SLinus Torvalds 15910b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1592c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 15939ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 15949f0ec176SAndy Adamson int how, int flags) 15951da177e4SLinus Torvalds { 159607737691STrond Myklebust struct rpc_task *task; 15979ace33cdSFred Isaman int priority = flush_task_priority(how); 1598bdc7f021STrond Myklebust struct rpc_message msg = { 1599bdc7f021STrond Myklebust .rpc_argp = &data->args, 1600bdc7f021STrond Myklebust .rpc_resp = &data->res, 16019ace33cdSFred Isaman .rpc_cred = data->cred, 1602bdc7f021STrond Myklebust }; 160384115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 160407737691STrond Myklebust .task = &data->task, 16059ace33cdSFred Isaman .rpc_client = clnt, 1606bdc7f021STrond Myklebust .rpc_message = &msg, 16079ace33cdSFred Isaman .callback_ops = call_ops, 160884115e1cSTrond Myklebust .callback_data = data, 1609101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 16109f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 16113ff7576dSTrond Myklebust .priority = priority, 161284115e1cSTrond Myklebust }; 1613788e7a89STrond Myklebust /* Set up the initial task struct. */ 1614c36aae9aSPeng Tao nfs_ops->commit_setup(data, &msg); 16151da177e4SLinus Torvalds 1616b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1617bdc7f021STrond Myklebust 16188c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 16198c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 16208c21c62cSWeston Andros Adamson 162107737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1622dbae4c73STrond Myklebust if (IS_ERR(task)) 1623dbae4c73STrond Myklebust return PTR_ERR(task); 1624d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1625d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 162607737691STrond Myklebust rpc_put_task(task); 1627dbae4c73STrond Myklebust return 0; 16281da177e4SLinus Torvalds } 1629e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 16301da177e4SLinus Torvalds 1631378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1632378520b8SPeng Tao { 1633378520b8SPeng Tao loff_t lwb = 0; 1634378520b8SPeng Tao struct nfs_page *req; 1635378520b8SPeng Tao 1636378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1637378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1638378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1639378520b8SPeng Tao 1640378520b8SPeng Tao return lwb; 1641378520b8SPeng Tao } 1642378520b8SPeng Tao 16431da177e4SLinus Torvalds /* 16449ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 16459ace33cdSFred Isaman */ 16460b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1647988b6dceSFred Isaman struct list_head *head, 1648f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1649f453a54aSFred Isaman struct nfs_commit_info *cinfo) 16509ace33cdSFred Isaman { 16519ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 16522b0143b5SDavid Howells struct inode *inode = d_inode(first->wb_context->dentry); 16539ace33cdSFred Isaman 16549ace33cdSFred Isaman /* Set up the RPC argument and reply structs 16559ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 16569ace33cdSFred Isaman 16579ace33cdSFred Isaman list_splice_init(head, &data->pages); 16589ace33cdSFred Isaman 16599ace33cdSFred Isaman data->inode = inode; 16609ace33cdSFred Isaman data->cred = first->wb_context->cred; 1661988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1662378520b8SPeng Tao /* only set lwb for pnfs commit */ 1663378520b8SPeng Tao if (lseg) 1664378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 16659ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1666f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1667b359f9d0SFred Isaman data->dreq = cinfo->dreq; 16689ace33cdSFred Isaman 16699ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 16709ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 16719ace33cdSFred Isaman data->args.offset = 0; 16729ace33cdSFred Isaman data->args.count = 0; 16730b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 16749ace33cdSFred Isaman data->res.fattr = &data->fattr; 16759ace33cdSFred Isaman data->res.verf = &data->verf; 16769ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 16779ace33cdSFred Isaman } 1678e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 16799ace33cdSFred Isaman 1680e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1681ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1682b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1683b57ff130SWeston Andros Adamson u32 ds_commit_idx) 168464bfeb49SFred Isaman { 168564bfeb49SFred Isaman struct nfs_page *req; 168664bfeb49SFred Isaman 168764bfeb49SFred Isaman while (!list_empty(page_list)) { 168864bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 168964bfeb49SFred Isaman nfs_list_remove_request(req); 1690b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 1691487b9b8aSTom Haynes if (!cinfo->dreq) 1692487b9b8aSTom Haynes nfs_clear_page_commit(req->wb_page); 16931d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 169464bfeb49SFred Isaman } 169564bfeb49SFred Isaman } 1696e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 169764bfeb49SFred Isaman 1698b20135d0STrond Myklebust static void 1699b20135d0STrond Myklebust nfs_commit_resched_write(struct nfs_commit_info *cinfo, 1700b20135d0STrond Myklebust struct nfs_page *req) 1701b20135d0STrond Myklebust { 1702b20135d0STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 1703b20135d0STrond Myklebust } 1704b20135d0STrond Myklebust 17059ace33cdSFred Isaman /* 17061da177e4SLinus Torvalds * Commit dirty pages 17071da177e4SLinus Torvalds */ 17081da177e4SLinus Torvalds static int 1709ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1710ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 17111da177e4SLinus Torvalds { 17120b7c0153SFred Isaman struct nfs_commit_data *data; 17131da177e4SLinus Torvalds 1714c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 17151da177e4SLinus Torvalds 17161da177e4SLinus Torvalds if (!data) 17171da177e4SLinus Torvalds goto out_bad; 17181da177e4SLinus Torvalds 17191da177e4SLinus Torvalds /* Set up the argument struct */ 1720f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1721f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 1722c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 1723c36aae9aSPeng Tao data->mds_ops, how, 0); 17241da177e4SLinus Torvalds out_bad: 1725b57ff130SWeston Andros Adamson nfs_retry_commit(head, NULL, cinfo, 0); 17261da177e4SLinus Torvalds return -ENOMEM; 17271da177e4SLinus Torvalds } 17281da177e4SLinus Torvalds 172967911c8fSAnna Schumaker int nfs_commit_file(struct file *file, struct nfs_write_verifier *verf) 173067911c8fSAnna Schumaker { 173167911c8fSAnna Schumaker struct inode *inode = file_inode(file); 173267911c8fSAnna Schumaker struct nfs_open_context *open; 173367911c8fSAnna Schumaker struct nfs_commit_info cinfo; 173467911c8fSAnna Schumaker struct nfs_page *req; 173567911c8fSAnna Schumaker int ret; 173667911c8fSAnna Schumaker 173767911c8fSAnna Schumaker open = get_nfs_open_context(nfs_file_open_context(file)); 173867911c8fSAnna Schumaker req = nfs_create_request(open, NULL, NULL, 0, i_size_read(inode)); 173967911c8fSAnna Schumaker if (!req) { 174067911c8fSAnna Schumaker ret = -ENOMEM; 174167911c8fSAnna Schumaker goto out_put; 174267911c8fSAnna Schumaker } 174367911c8fSAnna Schumaker 174467911c8fSAnna Schumaker nfs_init_cinfo_from_inode(&cinfo, inode); 174567911c8fSAnna Schumaker 174667911c8fSAnna Schumaker memcpy(&req->wb_verf, verf, sizeof(struct nfs_write_verifier)); 174767911c8fSAnna Schumaker nfs_request_add_commit_list(req, &cinfo); 174867911c8fSAnna Schumaker ret = nfs_commit_inode(inode, FLUSH_SYNC); 174967911c8fSAnna Schumaker if (ret > 0) 175067911c8fSAnna Schumaker ret = 0; 175167911c8fSAnna Schumaker 175267911c8fSAnna Schumaker nfs_free_request(req); 175367911c8fSAnna Schumaker out_put: 175467911c8fSAnna Schumaker put_nfs_open_context(open); 175567911c8fSAnna Schumaker return ret; 175667911c8fSAnna Schumaker } 175767911c8fSAnna Schumaker EXPORT_SYMBOL_GPL(nfs_commit_file); 175867911c8fSAnna Schumaker 17591da177e4SLinus Torvalds /* 17601da177e4SLinus Torvalds * COMMIT call returned 17611da177e4SLinus Torvalds */ 1762788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 17631da177e4SLinus Torvalds { 17640b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 17651da177e4SLinus Torvalds 1766a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 17671da177e4SLinus Torvalds task->tk_pid, task->tk_status); 17681da177e4SLinus Torvalds 1769788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1770c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1771c9d8f89dSTrond Myklebust } 1772c9d8f89dSTrond Myklebust 1773f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1774c9d8f89dSTrond Myklebust { 1775c9d8f89dSTrond Myklebust struct nfs_page *req; 1776c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1777f453a54aSFred Isaman struct nfs_commit_info cinfo; 1778353db796SNeilBrown struct nfs_server *nfss; 1779788e7a89STrond Myklebust 17801da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 17811da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 17821da177e4SLinus Torvalds nfs_list_remove_request(req); 178367911c8fSAnna Schumaker if (req->wb_page) 1784d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 17851da177e4SLinus Torvalds 17861e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 17873d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 17882b0143b5SDavid Howells (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)), 17891da177e4SLinus Torvalds req->wb_bytes, 17901da177e4SLinus Torvalds (long long)req_offset(req)); 1791c9d8f89dSTrond Myklebust if (status < 0) { 1792c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 17931da177e4SLinus Torvalds nfs_inode_remove_request(req); 1794c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 17951da177e4SLinus Torvalds goto next; 17961da177e4SLinus Torvalds } 17971da177e4SLinus Torvalds 17981da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 17991da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 18002f2c63bcSTrond Myklebust if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { 18011da177e4SLinus Torvalds /* We have a match */ 18021da177e4SLinus Torvalds nfs_inode_remove_request(req); 18031da177e4SLinus Torvalds dprintk(" OK\n"); 18041da177e4SLinus Torvalds goto next; 18051da177e4SLinus Torvalds } 18061da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 18071da177e4SLinus Torvalds dprintk(" mismatch\n"); 18086d884e8fSFred nfs_mark_request_dirty(req); 180905990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 18101da177e4SLinus Torvalds next: 18111d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18121da177e4SLinus Torvalds } 1813353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1814353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 1815353db796SNeilBrown clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 1816353db796SNeilBrown 1817f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1818af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18195917ce84SFred Isaman } 18205917ce84SFred Isaman 18215917ce84SFred Isaman static void nfs_commit_release(void *calldata) 18225917ce84SFred Isaman { 18230b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18245917ce84SFred Isaman 1825f453a54aSFred Isaman data->completion_ops->completion(data); 1826c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 18271da177e4SLinus Torvalds } 1828788e7a89STrond Myklebust 1829788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 18300b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1831788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1832788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1833788e7a89STrond Myklebust }; 18341da177e4SLinus Torvalds 1835f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1836f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1837b20135d0STrond Myklebust .resched_write = nfs_commit_resched_write, 1838f453a54aSFred Isaman }; 1839f453a54aSFred Isaman 18401763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1841ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 184284c53ab5SFred Isaman { 184384c53ab5SFred Isaman int status; 184484c53ab5SFred Isaman 1845ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 184684c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1847ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 184884c53ab5SFred Isaman return status; 184984c53ab5SFred Isaman } 185084c53ab5SFred Isaman 1851b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 18521da177e4SLinus Torvalds { 18531da177e4SLinus Torvalds LIST_HEAD(head); 1854ea2cf228SFred Isaman struct nfs_commit_info cinfo; 185571d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1856af7cf057STrond Myklebust int error = 0; 1857b8413f98STrond Myklebust int res; 18581da177e4SLinus Torvalds 1859ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1860af7cf057STrond Myklebust nfs_commit_begin(cinfo.mds); 1861ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 1862af7cf057STrond Myklebust if (res) 1863ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 1864af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18651da177e4SLinus Torvalds if (error < 0) 1866af7cf057STrond Myklebust goto out_error; 1867b8413f98STrond Myklebust if (!may_wait) 1868b8413f98STrond Myklebust goto out_mark_dirty; 1869af7cf057STrond Myklebust error = wait_on_commit(cinfo.mds); 1870b8413f98STrond Myklebust if (error < 0) 1871b8413f98STrond Myklebust return error; 1872c5efa5fcSTrond Myklebust return res; 1873af7cf057STrond Myklebust out_error: 1874af7cf057STrond Myklebust res = error; 1875c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1876c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1877c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1878c5efa5fcSTrond Myklebust * that the data is on the disk. 1879c5efa5fcSTrond Myklebust */ 1880c5efa5fcSTrond Myklebust out_mark_dirty: 1881c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 18821da177e4SLinus Torvalds return res; 18831da177e4SLinus Torvalds } 1884b20135d0STrond Myklebust EXPORT_SYMBOL_GPL(nfs_commit_inode); 18858fc795f7STrond Myklebust 1886ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 18878fc795f7STrond Myklebust { 1888420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1889420e3646STrond Myklebust int flags = FLUSH_SYNC; 1890420e3646STrond Myklebust int ret = 0; 18918fc795f7STrond Myklebust 18923236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1893ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 18943236c3e1SJeff Layton return ret; 18953236c3e1SJeff Layton 1896a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1897a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1898a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1899420e3646STrond Myklebust */ 1900cb1410c7SWeston Andros Adamson if (nfsi->commit_info.ncommit <= (nfsi->nrequests >> 1)) 1901420e3646STrond Myklebust goto out_mark_dirty; 1902420e3646STrond Myklebust 1903a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1904420e3646STrond Myklebust flags = 0; 1905a00dd6c0SJeff Layton } 1906a00dd6c0SJeff Layton 1907420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1908420e3646STrond Myklebust if (ret >= 0) { 1909420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1910420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1911420e3646STrond Myklebust wbc->nr_to_write -= ret; 1912420e3646STrond Myklebust else 1913420e3646STrond Myklebust wbc->nr_to_write = 0; 1914420e3646STrond Myklebust } 19158fc795f7STrond Myklebust return 0; 1916420e3646STrond Myklebust } 1917420e3646STrond Myklebust out_mark_dirty: 19188fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 19198fc795f7STrond Myklebust return ret; 19208fc795f7STrond Myklebust } 192189d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1922863a3c6cSAndy Adamson 1923acdc53b2STrond Myklebust /* 1924acdc53b2STrond Myklebust * flush the inode to disk. 1925acdc53b2STrond Myklebust */ 1926acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 192734901f70STrond Myklebust { 1928f4ce1299STrond Myklebust int ret; 192934901f70STrond Myklebust 1930f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1931f4ce1299STrond Myklebust 19325bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 19336b196875SChuck Lever if (ret) 19346b196875SChuck Lever goto out; 19355bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 19366b196875SChuck Lever if (ret < 0) 19376b196875SChuck Lever goto out; 19385bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 19396b196875SChuck Lever ret = 0; 1940f4ce1299STrond Myklebust 19416b196875SChuck Lever out: 1942f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1943f4ce1299STrond Myklebust return ret; 19441c75950bSTrond Myklebust } 1945ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 19461c75950bSTrond Myklebust 19471b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 19481b3b4a1aSTrond Myklebust { 19491b3b4a1aSTrond Myklebust struct nfs_page *req; 19501b3b4a1aSTrond Myklebust int ret = 0; 19511b3b4a1aSTrond Myklebust 1952ba8b06e6STrond Myklebust wait_on_page_writeback(page); 19533e217045SWeston Andros Adamson 19543e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 19553e217045SWeston Andros Adamson * request */ 19563e217045SWeston Andros Adamson req = nfs_lock_and_join_requests(page, false); 19573e217045SWeston Andros Adamson 19583e217045SWeston Andros Adamson if (IS_ERR(req)) { 19593e217045SWeston Andros Adamson ret = PTR_ERR(req); 19603e217045SWeston Andros Adamson } else if (req) { 19613e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 19623e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 19633e217045SWeston Andros Adamson * request from the inode / page_private pointer and 19643e217045SWeston Andros Adamson * release it */ 19651b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 19661d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 19671b3b4a1aSTrond Myklebust } 19683e217045SWeston Andros Adamson 19691b3b4a1aSTrond Myklebust return ret; 19701b3b4a1aSTrond Myklebust } 19711b3b4a1aSTrond Myklebust 19721c75950bSTrond Myklebust /* 19731c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 19741c75950bSTrond Myklebust */ 1975d6c843b9SPeng Tao int nfs_wb_single_page(struct inode *inode, struct page *page, bool launder) 19761c75950bSTrond Myklebust { 197729418aa4SMel Gorman loff_t range_start = page_file_offset(page); 197809cbfeafSKirill A. Shutemov loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); 19797f2f12d9STrond Myklebust struct writeback_control wbc = { 19807f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 19817f2f12d9STrond Myklebust .nr_to_write = 0, 19827f2f12d9STrond Myklebust .range_start = range_start, 19837f2f12d9STrond Myklebust .range_end = range_end, 19847f2f12d9STrond Myklebust }; 19857f2f12d9STrond Myklebust int ret; 19867f2f12d9STrond Myklebust 1987f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 1988f4ce1299STrond Myklebust 19890522f6adSTrond Myklebust for (;;) { 1990ba8b06e6STrond Myklebust wait_on_page_writeback(page); 19917f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 1992d6c843b9SPeng Tao ret = nfs_writepage_locked(page, &wbc, launder); 19937f2f12d9STrond Myklebust if (ret < 0) 19947f2f12d9STrond Myklebust goto out_error; 19950522f6adSTrond Myklebust continue; 19967f2f12d9STrond Myklebust } 1997f4ce1299STrond Myklebust ret = 0; 19980522f6adSTrond Myklebust if (!PagePrivate(page)) 19990522f6adSTrond Myklebust break; 20000522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20017f2f12d9STrond Myklebust if (ret < 0) 20027f2f12d9STrond Myklebust goto out_error; 20037f2f12d9STrond Myklebust } 20047f2f12d9STrond Myklebust out_error: 2005f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 20067f2f12d9STrond Myklebust return ret; 20071c75950bSTrond Myklebust } 20081c75950bSTrond Myklebust 2009074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 2010074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 2011a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 2012074cc1deSTrond Myklebust { 20132da95652SJeff Layton /* 20142da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 20152da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 20162da95652SJeff Layton * 20172da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 20182da95652SJeff Layton * that we can safely release the inode reference while holding 20192da95652SJeff Layton * the page lock. 20202da95652SJeff Layton */ 20212da95652SJeff Layton if (PagePrivate(page)) 20222da95652SJeff Layton return -EBUSY; 2023074cc1deSTrond Myklebust 20248c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 20258c209ce7SDavid Howells return -EBUSY; 2026074cc1deSTrond Myklebust 2027a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 2028074cc1deSTrond Myklebust } 2029074cc1deSTrond Myklebust #endif 2030074cc1deSTrond Myklebust 2031f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 20321da177e4SLinus Torvalds { 20331da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 20341e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 20351da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 203620c2df83SPaul Mundt NULL); 20371da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 20381da177e4SLinus Torvalds return -ENOMEM; 20391da177e4SLinus Torvalds 204093d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 20411da177e4SLinus Torvalds nfs_wdata_cachep); 20421da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 20433dd4765fSJeff Layton goto out_destroy_write_cache; 20441da177e4SLinus Torvalds 20450b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 20460b7c0153SFred Isaman sizeof(struct nfs_commit_data), 20470b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 20480b7c0153SFred Isaman NULL); 20490b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 20503dd4765fSJeff Layton goto out_destroy_write_mempool; 20510b7c0153SFred Isaman 205293d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 20534c100210SYanchuan Nian nfs_cdata_cachep); 20541da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 20553dd4765fSJeff Layton goto out_destroy_commit_cache; 20561da177e4SLinus Torvalds 205789a09141SPeter Zijlstra /* 205889a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 205989a09141SPeter Zijlstra * 206089a09141SPeter Zijlstra * 64MB: 8192k 206189a09141SPeter Zijlstra * 128MB: 11585k 206289a09141SPeter Zijlstra * 256MB: 16384k 206389a09141SPeter Zijlstra * 512MB: 23170k 206489a09141SPeter Zijlstra * 1GB: 32768k 206589a09141SPeter Zijlstra * 2GB: 46340k 206689a09141SPeter Zijlstra * 4GB: 65536k 206789a09141SPeter Zijlstra * 8GB: 92681k 206889a09141SPeter Zijlstra * 16GB: 131072k 206989a09141SPeter Zijlstra * 207089a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 207189a09141SPeter Zijlstra * Limit the default to 256M 207289a09141SPeter Zijlstra */ 207389a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 207489a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 207589a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 207689a09141SPeter Zijlstra 20771da177e4SLinus Torvalds return 0; 20783dd4765fSJeff Layton 20793dd4765fSJeff Layton out_destroy_commit_cache: 20803dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 20813dd4765fSJeff Layton out_destroy_write_mempool: 20823dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 20833dd4765fSJeff Layton out_destroy_write_cache: 20843dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 20853dd4765fSJeff Layton return -ENOMEM; 20861da177e4SLinus Torvalds } 20871da177e4SLinus Torvalds 2088266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 20891da177e4SLinus Torvalds { 20901da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 20913dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 20921da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 20931a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 20941da177e4SLinus Torvalds } 20951da177e4SLinus Torvalds 20964a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 2097a4cdda59SAnna Schumaker .rw_mode = FMODE_WRITE, 20984a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 20994a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 21000eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 21010eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 21021ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 21034a0de55cSAnna Schumaker }; 2104