11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/nfs/write.c 31da177e4SLinus Torvalds * 47c85d900STrond Myklebust * Write file data over NFS. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> 71da177e4SLinus Torvalds */ 81da177e4SLinus Torvalds 91da177e4SLinus Torvalds #include <linux/types.h> 101da177e4SLinus Torvalds #include <linux/slab.h> 111da177e4SLinus Torvalds #include <linux/mm.h> 121da177e4SLinus Torvalds #include <linux/pagemap.h> 131da177e4SLinus Torvalds #include <linux/file.h> 141da177e4SLinus Torvalds #include <linux/writeback.h> 1589a09141SPeter Zijlstra #include <linux/swap.h> 16074cc1deSTrond Myklebust #include <linux/migrate.h> 171da177e4SLinus Torvalds 181da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h> 191da177e4SLinus Torvalds #include <linux/nfs_fs.h> 201da177e4SLinus Torvalds #include <linux/nfs_mount.h> 211da177e4SLinus Torvalds #include <linux/nfs_page.h> 223fcfab16SAndrew Morton #include <linux/backing-dev.h> 23afeacc8cSPaul Gortmaker #include <linux/export.h> 243fcfab16SAndrew Morton 251da177e4SLinus Torvalds #include <asm/uaccess.h> 261da177e4SLinus Torvalds 271da177e4SLinus Torvalds #include "delegation.h" 2849a70f27STrond Myklebust #include "internal.h" 2991d5b470SChuck Lever #include "iostat.h" 30def6ed7eSAndy Adamson #include "nfs4_fs.h" 31074cc1deSTrond Myklebust #include "fscache.h" 3294ad1c80SFred Isaman #include "pnfs.h" 331da177e4SLinus Torvalds 34f4ce1299STrond Myklebust #include "nfstrace.h" 35f4ce1299STrond Myklebust 361da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 371da177e4SLinus Torvalds 381da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 391da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 401da177e4SLinus Torvalds 411da177e4SLinus Torvalds /* 421da177e4SLinus Torvalds * Local function declarations 431da177e4SLinus Torvalds */ 44f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 45788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 46061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; 47f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops; 484a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops; 49d4581383SWeston Andros Adamson static void nfs_clear_request_commit(struct nfs_page *req); 501da177e4SLinus Torvalds 51e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 523feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 530b7c0153SFred Isaman static struct kmem_cache *nfs_cdata_cachep; 541da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 551da177e4SLinus Torvalds 560b7c0153SFred Isaman struct nfs_commit_data *nfs_commitdata_alloc(void) 571da177e4SLinus Torvalds { 58192e501bSMel Gorman struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); 5940859d7eSChuck Lever 601da177e4SLinus Torvalds if (p) { 611da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 621da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 631da177e4SLinus Torvalds } 641da177e4SLinus Torvalds return p; 651da177e4SLinus Torvalds } 66e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 671da177e4SLinus Torvalds 680b7c0153SFred Isaman void nfs_commit_free(struct nfs_commit_data *p) 691da177e4SLinus Torvalds { 701da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 711da177e4SLinus Torvalds } 72e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 731da177e4SLinus Torvalds 741e7f3a48SWeston Andros Adamson static struct nfs_pgio_header *nfs_writehdr_alloc(void) 753feb2d49STrond Myklebust { 761e7f3a48SWeston Andros Adamson struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); 773feb2d49STrond Myklebust 784a0de55cSAnna Schumaker if (p) 793feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 803feb2d49STrond Myklebust return p; 813feb2d49STrond Myklebust } 823feb2d49STrond Myklebust 831e7f3a48SWeston Andros Adamson static void nfs_writehdr_free(struct nfs_pgio_header *hdr) 846c75dc0dSFred Isaman { 851e7f3a48SWeston Andros Adamson mempool_free(hdr, nfs_wdata_mempool); 863feb2d49STrond Myklebust } 871da177e4SLinus Torvalds 887b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 897b159fc1STrond Myklebust { 907b159fc1STrond Myklebust ctx->error = error; 917b159fc1STrond Myklebust smp_wmb(); 927b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 937b159fc1STrond Myklebust } 947b159fc1STrond Myklebust 9584d3a9a9SWeston Andros Adamson /* 9684d3a9a9SWeston Andros Adamson * nfs_page_find_head_request_locked - find head request associated with @page 9784d3a9a9SWeston Andros Adamson * 9884d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 9984d3a9a9SWeston Andros Adamson * 10084d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 10184d3a9a9SWeston Andros Adamson */ 10229418aa4SMel Gorman static struct nfs_page * 10384d3a9a9SWeston Andros Adamson nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page) 104277459d2STrond Myklebust { 105277459d2STrond Myklebust struct nfs_page *req = NULL; 106277459d2STrond Myklebust 10729418aa4SMel Gorman if (PagePrivate(page)) 108277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 10929418aa4SMel Gorman else if (unlikely(PageSwapCache(page))) { 11029418aa4SMel Gorman struct nfs_page *freq, *t; 11129418aa4SMel Gorman 11229418aa4SMel Gorman /* Linearly search the commit list for the correct req */ 11329418aa4SMel Gorman list_for_each_entry_safe(freq, t, &nfsi->commit_info.list, wb_list) { 11429418aa4SMel Gorman if (freq->wb_page == page) { 11584d3a9a9SWeston Andros Adamson req = freq->wb_head; 11629418aa4SMel Gorman break; 117277459d2STrond Myklebust } 11829418aa4SMel Gorman } 11929418aa4SMel Gorman } 12029418aa4SMel Gorman 12184d3a9a9SWeston Andros Adamson if (req) { 12284d3a9a9SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 12384d3a9a9SWeston Andros Adamson 12429418aa4SMel Gorman kref_get(&req->wb_kref); 12584d3a9a9SWeston Andros Adamson } 12629418aa4SMel Gorman 127277459d2STrond Myklebust return req; 128277459d2STrond Myklebust } 129277459d2STrond Myklebust 13084d3a9a9SWeston Andros Adamson /* 13184d3a9a9SWeston Andros Adamson * nfs_page_find_head_request - find head request associated with @page 13284d3a9a9SWeston Andros Adamson * 13384d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 13484d3a9a9SWeston Andros Adamson */ 13584d3a9a9SWeston Andros Adamson static struct nfs_page *nfs_page_find_head_request(struct page *page) 136277459d2STrond Myklebust { 137d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 138277459d2STrond Myklebust struct nfs_page *req = NULL; 139277459d2STrond Myklebust 140587142f8STrond Myklebust spin_lock(&inode->i_lock); 14184d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 142587142f8STrond Myklebust spin_unlock(&inode->i_lock); 143277459d2STrond Myklebust return req; 144277459d2STrond Myklebust } 145277459d2STrond Myklebust 1461da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 1471da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 1481da177e4SLinus Torvalds { 149d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 150a3d01454STrond Myklebust loff_t end, i_size; 151a3d01454STrond Myklebust pgoff_t end_index; 1521da177e4SLinus Torvalds 153a3d01454STrond Myklebust spin_lock(&inode->i_lock); 154a3d01454STrond Myklebust i_size = i_size_read(inode); 155a3d01454STrond Myklebust end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; 156d56b4ddfSMel Gorman if (i_size > 0 && page_file_index(page) < end_index) 157a3d01454STrond Myklebust goto out; 158d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 1591da177e4SLinus Torvalds if (i_size >= end) 160a3d01454STrond Myklebust goto out; 1611da177e4SLinus Torvalds i_size_write(inode, end); 162a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 163a3d01454STrond Myklebust out: 164a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 1651da177e4SLinus Torvalds } 1661da177e4SLinus Torvalds 167a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 168a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 169a301b777STrond Myklebust { 170d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 171a301b777STrond Myklebust } 172a301b777STrond Myklebust 173d72ddcbaSWeston Andros Adamson /* 174d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 175d72ddcbaSWeston Andros Adamson * @head - head request of page group 176d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 177d72ddcbaSWeston Andros Adamson * 178d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 179d72ddcbaSWeston Andros Adamson * page offset @page_offset. 180d72ddcbaSWeston Andros Adamson * 181d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 182d72ddcbaSWeston Andros Adamson * match is found. 183d72ddcbaSWeston Andros Adamson * 184d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 185d72ddcbaSWeston Andros Adamson */ 186d72ddcbaSWeston Andros Adamson static struct nfs_page * 187d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 188d72ddcbaSWeston Andros Adamson { 189d72ddcbaSWeston Andros Adamson struct nfs_page *req; 190d72ddcbaSWeston Andros Adamson 191d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(head != head->wb_head); 192d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags)); 193d72ddcbaSWeston Andros Adamson 194d72ddcbaSWeston Andros Adamson req = head; 195d72ddcbaSWeston Andros Adamson do { 196d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 197d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 198d72ddcbaSWeston Andros Adamson return req; 199d72ddcbaSWeston Andros Adamson 200d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 201d72ddcbaSWeston Andros Adamson } while (req != head); 202d72ddcbaSWeston Andros Adamson 203d72ddcbaSWeston Andros Adamson return NULL; 204d72ddcbaSWeston Andros Adamson } 205d72ddcbaSWeston Andros Adamson 206d72ddcbaSWeston Andros Adamson /* 207d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 208d72ddcbaSWeston Andros Adamson * @head - head request of page group 209d72ddcbaSWeston Andros Adamson * 210d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 211d72ddcbaSWeston Andros Adamson * returns false otherwise 212d72ddcbaSWeston Andros Adamson */ 213d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 214d72ddcbaSWeston Andros Adamson { 215d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 216d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 217d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 218d72ddcbaSWeston Andros Adamson 219*e7029206SWeston Andros Adamson nfs_page_group_lock(req, true); 220d72ddcbaSWeston Andros Adamson 221d72ddcbaSWeston Andros Adamson do { 222d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 223d72ddcbaSWeston Andros Adamson if (tmp) { 224d72ddcbaSWeston Andros Adamson /* no way this should happen */ 225d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(tmp->wb_pgbase != pos); 226d72ddcbaSWeston Andros Adamson pos += tmp->wb_bytes - (pos - tmp->wb_pgbase); 227d72ddcbaSWeston Andros Adamson } 228d72ddcbaSWeston Andros Adamson } while (tmp && pos < len); 229d72ddcbaSWeston Andros Adamson 230d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 231d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(pos > len); 232d72ddcbaSWeston Andros Adamson return pos == len; 233d72ddcbaSWeston Andros Adamson } 234d72ddcbaSWeston Andros Adamson 2351da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 2361da177e4SLinus Torvalds * covers the full page. 2371da177e4SLinus Torvalds */ 238d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 2391da177e4SLinus Torvalds { 240d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 2411da177e4SLinus Torvalds return; 242d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 2431da177e4SLinus Torvalds return; 244d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 2451da177e4SLinus Torvalds } 2461da177e4SLinus Torvalds 2471da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 2481da177e4SLinus Torvalds { 2491da177e4SLinus Torvalds if (wbc->for_reclaim) 250c63c7b05STrond Myklebust return FLUSH_HIGHPRI | FLUSH_STABLE; 251b17621feSWu Fengguang if (wbc->for_kupdate || wbc->for_background) 252b31268acSTrond Myklebust return FLUSH_LOWPRI | FLUSH_COND_STABLE; 253b31268acSTrond Myklebust return FLUSH_COND_STABLE; 2541da177e4SLinus Torvalds } 2551da177e4SLinus Torvalds 2561da177e4SLinus Torvalds /* 25789a09141SPeter Zijlstra * NFS congestion control 25889a09141SPeter Zijlstra */ 25989a09141SPeter Zijlstra 26089a09141SPeter Zijlstra int nfs_congestion_kb; 26189a09141SPeter Zijlstra 26289a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 26389a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 26489a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 26589a09141SPeter Zijlstra 266deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 26789a09141SPeter Zijlstra { 268deed85e7STrond Myklebust struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); 2695a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 2705a6d41b3STrond Myklebust 271deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 27289a09141SPeter Zijlstra 273277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 2748aa7e847SJens Axboe NFS_CONGESTION_ON_THRESH) { 2758aa7e847SJens Axboe set_bdi_congested(&nfss->backing_dev_info, 2768aa7e847SJens Axboe BLK_RW_ASYNC); 2778aa7e847SJens Axboe } 27889a09141SPeter Zijlstra } 27989a09141SPeter Zijlstra 28020633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 28189a09141SPeter Zijlstra { 28220633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 28389a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 28489a09141SPeter Zijlstra 28520633f04SWeston Andros Adamson if (!nfs_page_group_sync_on_bit(req, PG_WB_END)) 28620633f04SWeston Andros Adamson return; 28720633f04SWeston Andros Adamson 28820633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 289c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 2908aa7e847SJens Axboe clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); 29189a09141SPeter Zijlstra } 29289a09141SPeter Zijlstra 293d4581383SWeston Andros Adamson 294d4581383SWeston Andros Adamson /* nfs_page_group_clear_bits 295d4581383SWeston Andros Adamson * @req - an nfs request 296d4581383SWeston Andros Adamson * clears all page group related bits from @req 297d4581383SWeston Andros Adamson */ 298d4581383SWeston Andros Adamson static void 299d4581383SWeston Andros Adamson nfs_page_group_clear_bits(struct nfs_page *req) 300e261f51fSTrond Myklebust { 301d4581383SWeston Andros Adamson clear_bit(PG_TEARDOWN, &req->wb_flags); 302d4581383SWeston Andros Adamson clear_bit(PG_UNLOCKPAGE, &req->wb_flags); 303d4581383SWeston Andros Adamson clear_bit(PG_UPTODATE, &req->wb_flags); 304d4581383SWeston Andros Adamson clear_bit(PG_WB_END, &req->wb_flags); 305d4581383SWeston Andros Adamson clear_bit(PG_REMOVE, &req->wb_flags); 306d4581383SWeston Andros Adamson } 307d4581383SWeston Andros Adamson 308d4581383SWeston Andros Adamson 309d4581383SWeston Andros Adamson /* 310d4581383SWeston Andros Adamson * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req 311d4581383SWeston Andros Adamson * 312d4581383SWeston Andros Adamson * this is a helper function for nfs_lock_and_join_requests 313d4581383SWeston Andros Adamson * 314d4581383SWeston Andros Adamson * @inode - inode associated with request page group, must be holding inode lock 315d4581383SWeston Andros Adamson * @head - head request of page group, must be holding head lock 316d4581383SWeston Andros Adamson * @req - request that couldn't lock and needs to wait on the req bit lock 317d4581383SWeston Andros Adamson * @nonblock - if true, don't actually wait 318d4581383SWeston Andros Adamson * 319d4581383SWeston Andros Adamson * NOTE: this must be called holding page_group bit lock and inode spin lock 320d4581383SWeston Andros Adamson * and BOTH will be released before returning. 321d4581383SWeston Andros Adamson * 322d4581383SWeston Andros Adamson * returns 0 on success, < 0 on error. 323d4581383SWeston Andros Adamson */ 324d4581383SWeston Andros Adamson static int 325d4581383SWeston Andros Adamson nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head, 326d4581383SWeston Andros Adamson struct nfs_page *req, bool nonblock) 327d4581383SWeston Andros Adamson __releases(&inode->i_lock) 328d4581383SWeston Andros Adamson { 329d4581383SWeston Andros Adamson struct nfs_page *tmp; 330e261f51fSTrond Myklebust int ret; 331e261f51fSTrond Myklebust 332d4581383SWeston Andros Adamson /* relinquish all the locks successfully grabbed this run */ 333d4581383SWeston Andros Adamson for (tmp = head ; tmp != req; tmp = tmp->wb_this_page) 334d4581383SWeston Andros Adamson nfs_unlock_request(tmp); 335d4581383SWeston Andros Adamson 336d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags)); 337d4581383SWeston Andros Adamson 338d4581383SWeston Andros Adamson /* grab a ref on the request that will be waited on */ 339d4581383SWeston Andros Adamson kref_get(&req->wb_kref); 340d4581383SWeston Andros Adamson 341d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 342587142f8STrond Myklebust spin_unlock(&inode->i_lock); 343d4581383SWeston Andros Adamson 344d4581383SWeston Andros Adamson /* release ref from nfs_page_find_head_request_locked */ 345d4581383SWeston Andros Adamson nfs_release_request(head); 346d4581383SWeston Andros Adamson 347cfb506e1STrond Myklebust if (!nonblock) 348e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 349cfb506e1STrond Myklebust else 350cfb506e1STrond Myklebust ret = -EAGAIN; 351e261f51fSTrond Myklebust nfs_release_request(req); 352d4581383SWeston Andros Adamson 353d4581383SWeston Andros Adamson return ret; 354e261f51fSTrond Myklebust } 355d4581383SWeston Andros Adamson 356d4581383SWeston Andros Adamson /* 357d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 358d4581383SWeston Andros Adamson * 359d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 360d4581383SWeston Andros Adamson * @old_head - the old head of the list 361d4581383SWeston Andros Adamson * 362d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 363d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 364d4581383SWeston Andros Adamson * with a reference held by some other context. 365d4581383SWeston Andros Adamson */ 366d4581383SWeston Andros Adamson static void 367d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 368d4581383SWeston Andros Adamson struct nfs_page *old_head) 369d4581383SWeston Andros Adamson { 370d4581383SWeston Andros Adamson while (destroy_list) { 371d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 372d4581383SWeston Andros Adamson 373d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 374d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 375d4581383SWeston Andros Adamson 376d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 377d4581383SWeston Andros Adamson 378d4581383SWeston Andros Adamson /* make sure old group is not used */ 379d4581383SWeston Andros Adamson subreq->wb_head = subreq; 380d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 381d4581383SWeston Andros Adamson 382d4581383SWeston Andros Adamson nfs_clear_request_commit(subreq); 383d4581383SWeston Andros Adamson 384d4581383SWeston Andros Adamson /* subreq is now totally disconnected from page group or any 385d4581383SWeston Andros Adamson * write / commit lists. last chance to wake any waiters */ 386d4581383SWeston Andros Adamson nfs_unlock_request(subreq); 387d4581383SWeston Andros Adamson 388d4581383SWeston Andros Adamson if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) { 389d4581383SWeston Andros Adamson /* release ref on old head request */ 390d4581383SWeston Andros Adamson nfs_release_request(old_head); 391d4581383SWeston Andros Adamson 392d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 393d4581383SWeston Andros Adamson 394d4581383SWeston Andros Adamson /* release the PG_INODE_REF reference */ 395d4581383SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) 396d4581383SWeston Andros Adamson nfs_release_request(subreq); 397d4581383SWeston Andros Adamson else 398d4581383SWeston Andros Adamson WARN_ON_ONCE(1); 399d4581383SWeston Andros Adamson } else { 400d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags)); 401d4581383SWeston Andros Adamson /* zombie requests have already released the last 402d4581383SWeston Andros Adamson * reference and were waiting on the rest of the 403d4581383SWeston Andros Adamson * group to complete. Since it's no longer part of a 404d4581383SWeston Andros Adamson * group, simply free the request */ 405d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 406d4581383SWeston Andros Adamson nfs_free_request(subreq); 407d4581383SWeston Andros Adamson } 408d4581383SWeston Andros Adamson } 409d4581383SWeston Andros Adamson } 410d4581383SWeston Andros Adamson 411d4581383SWeston Andros Adamson /* 412d4581383SWeston Andros Adamson * nfs_lock_and_join_requests - join all subreqs to the head req and return 413d4581383SWeston Andros Adamson * a locked reference, cancelling any pending 414d4581383SWeston Andros Adamson * operations for this page. 415d4581383SWeston Andros Adamson * 416d4581383SWeston Andros Adamson * @page - the page used to lookup the "page group" of nfs_page structures 417d4581383SWeston Andros Adamson * @nonblock - if true, don't block waiting for request locks 418d4581383SWeston Andros Adamson * 419d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 420d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 421d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 422d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 423d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 424d4581383SWeston Andros Adamson * 425d4581383SWeston Andros Adamson * Returns a locked, referenced pointer to the head request - which after 426d4581383SWeston Andros Adamson * this call is guaranteed to be the only request associated with the page. 427d4581383SWeston Andros Adamson * Returns NULL if no requests are found for @page, or a ERR_PTR if an 428d4581383SWeston Andros Adamson * error was encountered. 429d4581383SWeston Andros Adamson */ 430d4581383SWeston Andros Adamson static struct nfs_page * 431d4581383SWeston Andros Adamson nfs_lock_and_join_requests(struct page *page, bool nonblock) 432d4581383SWeston Andros Adamson { 433d4581383SWeston Andros Adamson struct inode *inode = page_file_mapping(page)->host; 434d4581383SWeston Andros Adamson struct nfs_page *head, *subreq; 435d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 436d4581383SWeston Andros Adamson unsigned int total_bytes; 437d4581383SWeston Andros Adamson int ret; 438d4581383SWeston Andros Adamson 439d4581383SWeston Andros Adamson try_again: 440d4581383SWeston Andros Adamson total_bytes = 0; 441d4581383SWeston Andros Adamson 442d4581383SWeston Andros Adamson WARN_ON_ONCE(destroy_list); 443d4581383SWeston Andros Adamson 444d4581383SWeston Andros Adamson spin_lock(&inode->i_lock); 445d4581383SWeston Andros Adamson 446d4581383SWeston Andros Adamson /* 447d4581383SWeston Andros Adamson * A reference is taken only on the head request which acts as a 448d4581383SWeston Andros Adamson * reference to the whole page group - the group will not be destroyed 449d4581383SWeston Andros Adamson * until the head reference is released. 450d4581383SWeston Andros Adamson */ 451d4581383SWeston Andros Adamson head = nfs_page_find_head_request_locked(NFS_I(inode), page); 452d4581383SWeston Andros Adamson 453d4581383SWeston Andros Adamson if (!head) { 454587142f8STrond Myklebust spin_unlock(&inode->i_lock); 455d4581383SWeston Andros Adamson return NULL; 456d4581383SWeston Andros Adamson } 457d4581383SWeston Andros Adamson 458d4581383SWeston Andros Adamson /* lock each request in the page group */ 459*e7029206SWeston Andros Adamson ret = nfs_page_group_lock(head, false); 460*e7029206SWeston Andros Adamson if (ret < 0) 461*e7029206SWeston Andros Adamson return ERR_PTR(ret); 462d4581383SWeston Andros Adamson subreq = head; 463d4581383SWeston Andros Adamson do { 464d4581383SWeston Andros Adamson /* 465d4581383SWeston Andros Adamson * Subrequests are always contiguous, non overlapping 466d4581383SWeston Andros Adamson * and in order. If not, it's a programming error. 467d4581383SWeston Andros Adamson */ 468d4581383SWeston Andros Adamson WARN_ON_ONCE(subreq->wb_offset != 469d4581383SWeston Andros Adamson (head->wb_offset + total_bytes)); 470d4581383SWeston Andros Adamson 471d4581383SWeston Andros Adamson /* keep track of how many bytes this group covers */ 472d4581383SWeston Andros Adamson total_bytes += subreq->wb_bytes; 473d4581383SWeston Andros Adamson 474d4581383SWeston Andros Adamson if (!nfs_lock_request(subreq)) { 475d4581383SWeston Andros Adamson /* releases page group bit lock and 476d4581383SWeston Andros Adamson * inode spin lock and all references */ 477d4581383SWeston Andros Adamson ret = nfs_unroll_locks_and_wait(inode, head, 478d4581383SWeston Andros Adamson subreq, nonblock); 479d4581383SWeston Andros Adamson 480d4581383SWeston Andros Adamson if (ret == 0) 481d4581383SWeston Andros Adamson goto try_again; 482d4581383SWeston Andros Adamson 483d4581383SWeston Andros Adamson return ERR_PTR(ret); 484d4581383SWeston Andros Adamson } 485d4581383SWeston Andros Adamson 486d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 487d4581383SWeston Andros Adamson } while (subreq != head); 488d4581383SWeston Andros Adamson 489d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 490d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 491d4581383SWeston Andros Adamson subreq = head; 492d4581383SWeston Andros Adamson do { 493d4581383SWeston Andros Adamson nfs_list_remove_request(subreq); 494d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 495d4581383SWeston Andros Adamson } while (subreq != head); 496d4581383SWeston Andros Adamson 497d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 498d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 499d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 500d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 501d4581383SWeston Andros Adamson head->wb_this_page = head; 502d4581383SWeston Andros Adamson 503d4581383SWeston Andros Adamson /* change head request to cover whole range that 504d4581383SWeston Andros Adamson * the former page group covered */ 505d4581383SWeston Andros Adamson head->wb_bytes = total_bytes; 506d4581383SWeston Andros Adamson } 507d4581383SWeston Andros Adamson 508d4581383SWeston Andros Adamson /* 509d4581383SWeston Andros Adamson * prepare head request to be added to new pgio descriptor 510d4581383SWeston Andros Adamson */ 511d4581383SWeston Andros Adamson nfs_page_group_clear_bits(head); 512d4581383SWeston Andros Adamson 513d4581383SWeston Andros Adamson /* 514d4581383SWeston Andros Adamson * some part of the group was still on the inode list - otherwise 515d4581383SWeston Andros Adamson * the group wouldn't be involved in async write. 516d4581383SWeston Andros Adamson * grab a reference for the head request, iff it needs one. 517d4581383SWeston Andros Adamson */ 518d4581383SWeston Andros Adamson if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags)) 519d4581383SWeston Andros Adamson kref_get(&head->wb_kref); 520d4581383SWeston Andros Adamson 521d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 522d4581383SWeston Andros Adamson 523d4581383SWeston Andros Adamson /* drop lock to clear_request_commit the head req and clean up 524d4581383SWeston Andros Adamson * requests on destroy list */ 525d4581383SWeston Andros Adamson spin_unlock(&inode->i_lock); 526d4581383SWeston Andros Adamson 527d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(destroy_list, head); 528d4581383SWeston Andros Adamson 529d4581383SWeston Andros Adamson /* clean up commit list state */ 530d4581383SWeston Andros Adamson nfs_clear_request_commit(head); 531d4581383SWeston Andros Adamson 532d4581383SWeston Andros Adamson /* still holds ref on head from nfs_page_find_head_request_locked 533d4581383SWeston Andros Adamson * and still has lock on head from lock loop */ 534d4581383SWeston Andros Adamson return head; 535612c9384STrond Myklebust } 536074cc1deSTrond Myklebust 537074cc1deSTrond Myklebust /* 538074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 539074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 540074cc1deSTrond Myklebust */ 541074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 542cfb506e1STrond Myklebust struct page *page, bool nonblock) 543074cc1deSTrond Myklebust { 544074cc1deSTrond Myklebust struct nfs_page *req; 545074cc1deSTrond Myklebust int ret = 0; 546074cc1deSTrond Myklebust 547d4581383SWeston Andros Adamson req = nfs_lock_and_join_requests(page, nonblock); 548074cc1deSTrond Myklebust if (!req) 549074cc1deSTrond Myklebust goto out; 550074cc1deSTrond Myklebust ret = PTR_ERR(req); 551074cc1deSTrond Myklebust if (IS_ERR(req)) 552074cc1deSTrond Myklebust goto out; 553074cc1deSTrond Myklebust 554deed85e7STrond Myklebust nfs_set_page_writeback(page); 555deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 556074cc1deSTrond Myklebust 557deed85e7STrond Myklebust ret = 0; 558f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 559f8512ad0SFred Isaman nfs_redirty_request(req); 560074cc1deSTrond Myklebust ret = pgio->pg_error; 561f8512ad0SFred Isaman } 562074cc1deSTrond Myklebust out: 563074cc1deSTrond Myklebust return ret; 564e261f51fSTrond Myklebust } 565e261f51fSTrond Myklebust 566f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) 567f758c885STrond Myklebust { 568d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 569cfb506e1STrond Myklebust int ret; 570f758c885STrond Myklebust 571f758c885STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 572f758c885STrond Myklebust nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); 573f758c885STrond Myklebust 574d56b4ddfSMel Gorman nfs_pageio_cond_complete(pgio, page_file_index(page)); 5751b430beeSWu Fengguang ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); 576cfb506e1STrond Myklebust if (ret == -EAGAIN) { 577cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 578cfb506e1STrond Myklebust ret = 0; 579cfb506e1STrond Myklebust } 580cfb506e1STrond Myklebust return ret; 581f758c885STrond Myklebust } 582f758c885STrond Myklebust 583e261f51fSTrond Myklebust /* 5841da177e4SLinus Torvalds * Write an mmapped page to the server. 5851da177e4SLinus Torvalds */ 5864d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) 5871da177e4SLinus Torvalds { 588f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 589e261f51fSTrond Myklebust int err; 5901da177e4SLinus Torvalds 591a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc), 592a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 593f758c885STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 594f758c885STrond Myklebust nfs_pageio_complete(&pgio); 595f758c885STrond Myklebust if (err < 0) 5964d770ccfSTrond Myklebust return err; 597f758c885STrond Myklebust if (pgio.pg_error < 0) 598f758c885STrond Myklebust return pgio.pg_error; 599f758c885STrond Myklebust return 0; 6004d770ccfSTrond Myklebust } 6014d770ccfSTrond Myklebust 6024d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 6034d770ccfSTrond Myklebust { 604f758c885STrond Myklebust int ret; 6054d770ccfSTrond Myklebust 606f758c885STrond Myklebust ret = nfs_writepage_locked(page, wbc); 6071da177e4SLinus Torvalds unlock_page(page); 608f758c885STrond Myklebust return ret; 609f758c885STrond Myklebust } 610f758c885STrond Myklebust 611f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 612f758c885STrond Myklebust { 613f758c885STrond Myklebust int ret; 614f758c885STrond Myklebust 615f758c885STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 616f758c885STrond Myklebust unlock_page(page); 617f758c885STrond Myklebust return ret; 6181da177e4SLinus Torvalds } 6191da177e4SLinus Torvalds 6201da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 6211da177e4SLinus Torvalds { 6221da177e4SLinus Torvalds struct inode *inode = mapping->host; 62372cb77f4STrond Myklebust unsigned long *bitlock = &NFS_I(inode)->flags; 624c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 6251da177e4SLinus Torvalds int err; 6261da177e4SLinus Torvalds 62772cb77f4STrond Myklebust /* Stop dirtying of new pages while we sync */ 62872cb77f4STrond Myklebust err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, 62972cb77f4STrond Myklebust nfs_wait_bit_killable, TASK_KILLABLE); 63072cb77f4STrond Myklebust if (err) 63172cb77f4STrond Myklebust goto out_err; 63272cb77f4STrond Myklebust 63391d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 63491d5b470SChuck Lever 635a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 636a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 637f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 638c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 63972cb77f4STrond Myklebust 64072cb77f4STrond Myklebust clear_bit_unlock(NFS_INO_FLUSHING, bitlock); 6414e857c58SPeter Zijlstra smp_mb__after_atomic(); 64272cb77f4STrond Myklebust wake_up_bit(bitlock, NFS_INO_FLUSHING); 64372cb77f4STrond Myklebust 644f758c885STrond Myklebust if (err < 0) 64572cb77f4STrond Myklebust goto out_err; 64672cb77f4STrond Myklebust err = pgio.pg_error; 64772cb77f4STrond Myklebust if (err < 0) 64872cb77f4STrond Myklebust goto out_err; 649c63c7b05STrond Myklebust return 0; 65072cb77f4STrond Myklebust out_err: 65172cb77f4STrond Myklebust return err; 6521da177e4SLinus Torvalds } 6531da177e4SLinus Torvalds 6541da177e4SLinus Torvalds /* 6551da177e4SLinus Torvalds * Insert a write request into an inode 6561da177e4SLinus Torvalds */ 657d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 6581da177e4SLinus Torvalds { 6591da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 660e7d39069STrond Myklebust 6612bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 6622bfc6e56SWeston Andros Adamson 663e7d39069STrond Myklebust /* Lock the request! */ 6647ad84aa9STrond Myklebust nfs_lock_request(req); 665e7d39069STrond Myklebust 666e7d39069STrond Myklebust spin_lock(&inode->i_lock); 667011e2a7fSBryan Schumaker if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 668a9a4a87aSTrond Myklebust inode->i_version++; 66929418aa4SMel Gorman /* 67029418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 67129418aa4SMel Gorman * with invalidate/truncate. 67229418aa4SMel Gorman */ 67329418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 6742df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 675deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 676277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 67729418aa4SMel Gorman } 6781da177e4SLinus Torvalds nfsi->npages++; 67917089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 68017089a29SWeston Andros Adamson * extra reference so sub groups can follow suit */ 68117089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 682c03b4024STrond Myklebust kref_get(&req->wb_kref); 683e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 6841da177e4SLinus Torvalds } 6851da177e4SLinus Torvalds 6861da177e4SLinus Torvalds /* 68789a09141SPeter Zijlstra * Remove a write request from an inode 6881da177e4SLinus Torvalds */ 6891da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 6901da177e4SLinus Torvalds { 6913d4ff43dSAl Viro struct inode *inode = req->wb_context->dentry->d_inode; 6921da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 69320633f04SWeston Andros Adamson struct nfs_page *head; 69420633f04SWeston Andros Adamson 69520633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 69620633f04SWeston Andros Adamson head = req->wb_head; 6971da177e4SLinus Torvalds 698587142f8STrond Myklebust spin_lock(&inode->i_lock); 69920633f04SWeston Andros Adamson if (likely(!PageSwapCache(head->wb_page))) { 70020633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 70120633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 70220633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 70329418aa4SMel Gorman } 7041da177e4SLinus Torvalds nfsi->npages--; 705587142f8STrond Myklebust spin_unlock(&inode->i_lock); 70620633f04SWeston Andros Adamson } 70717089a29SWeston Andros Adamson 70817089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 7091da177e4SLinus Torvalds nfs_release_request(req); 7101da177e4SLinus Torvalds } 7111da177e4SLinus Torvalds 71261822ab5STrond Myklebust static void 7136d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 71461822ab5STrond Myklebust { 71561822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 71661822ab5STrond Myklebust } 71761822ab5STrond Myklebust 71889d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 7198dd37758STrond Myklebust /** 7208dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 7218dd37758STrond Myklebust * @req: pointer to a struct nfs_page 722ea2cf228SFred Isaman * @dst: commit list head 723ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 7248dd37758STrond Myklebust * 725ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 7268dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 7278dd37758STrond Myklebust * the MM page stats. 7288dd37758STrond Myklebust * 729ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 7308dd37758STrond Myklebust * holding the nfs_page lock. 7318dd37758STrond Myklebust */ 7328dd37758STrond Myklebust void 733ea2cf228SFred Isaman nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, 734ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 7358dd37758STrond Myklebust { 7368dd37758STrond Myklebust set_bit(PG_CLEAN, &(req)->wb_flags); 737ea2cf228SFred Isaman spin_lock(cinfo->lock); 738ea2cf228SFred Isaman nfs_list_add_request(req, dst); 739ea2cf228SFred Isaman cinfo->mds->ncommit++; 740ea2cf228SFred Isaman spin_unlock(cinfo->lock); 74156f9cd68SFred Isaman if (!cinfo->dreq) { 7428dd37758STrond Myklebust inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 743d56b4ddfSMel Gorman inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, 74456f9cd68SFred Isaman BDI_RECLAIMABLE); 74556f9cd68SFred Isaman __mark_inode_dirty(req->wb_context->dentry->d_inode, 74656f9cd68SFred Isaman I_DIRTY_DATASYNC); 74756f9cd68SFred Isaman } 7488dd37758STrond Myklebust } 7498dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 7508dd37758STrond Myklebust 7518dd37758STrond Myklebust /** 7528dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 7538dd37758STrond Myklebust * @req: pointer to a nfs_page 754ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 7558dd37758STrond Myklebust * 756ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 7578dd37758STrond Myklebust * number of outstanding requests requiring a commit 7588dd37758STrond Myklebust * It does not update the MM page stats. 7598dd37758STrond Myklebust * 760ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 7618dd37758STrond Myklebust */ 7628dd37758STrond Myklebust void 763ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 764ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 7658dd37758STrond Myklebust { 7668dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 7678dd37758STrond Myklebust return; 7688dd37758STrond Myklebust nfs_list_remove_request(req); 769ea2cf228SFred Isaman cinfo->mds->ncommit--; 7708dd37758STrond Myklebust } 7718dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 7728dd37758STrond Myklebust 773ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 774ea2cf228SFred Isaman struct inode *inode) 775ea2cf228SFred Isaman { 776ea2cf228SFred Isaman cinfo->lock = &inode->i_lock; 777ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 778ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 779b359f9d0SFred Isaman cinfo->dreq = NULL; 780f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 781ea2cf228SFred Isaman } 782ea2cf228SFred Isaman 783ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 784ea2cf228SFred Isaman struct inode *inode, 785ea2cf228SFred Isaman struct nfs_direct_req *dreq) 786ea2cf228SFred Isaman { 7871763da12SFred Isaman if (dreq) 7881763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 7891763da12SFred Isaman else 790ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 791ea2cf228SFred Isaman } 792ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 7938dd37758STrond Myklebust 7941da177e4SLinus Torvalds /* 7951da177e4SLinus Torvalds * Add a request to the inode's commit list. 7961da177e4SLinus Torvalds */ 7971763da12SFred Isaman void 798ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 799ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 8001da177e4SLinus Torvalds { 801ea2cf228SFred Isaman if (pnfs_mark_request_commit(req, lseg, cinfo)) 8028dd37758STrond Myklebust return; 803ea2cf228SFred Isaman nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); 8041da177e4SLinus Torvalds } 8058e821cadSTrond Myklebust 806d6d6dc7cSFred Isaman static void 807d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 808e468bae9STrond Myklebust { 809e468bae9STrond Myklebust dec_zone_page_state(page, NR_UNSTABLE_NFS); 810d56b4ddfSMel Gorman dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE); 811e468bae9STrond Myklebust } 812d6d6dc7cSFred Isaman 8138dd37758STrond Myklebust static void 814d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 815d6d6dc7cSFred Isaman { 8168dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 8178dd37758STrond Myklebust struct inode *inode = req->wb_context->dentry->d_inode; 818ea2cf228SFred Isaman struct nfs_commit_info cinfo; 819d6d6dc7cSFred Isaman 820ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 821ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 822ea2cf228SFred Isaman spin_lock(cinfo.lock); 823ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 824ea2cf228SFred Isaman spin_unlock(cinfo.lock); 825d6d6dc7cSFred Isaman } 8268dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 8278dd37758STrond Myklebust } 828e468bae9STrond Myklebust } 829e468bae9STrond Myklebust 830d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 8318e821cadSTrond Myklebust { 832c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 833d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 834c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 8358e821cadSTrond Myklebust } 8368e821cadSTrond Myklebust 8378e821cadSTrond Myklebust #else 83868cd6fa4SBryan Schumaker static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 83968cd6fa4SBryan Schumaker struct inode *inode) 84068cd6fa4SBryan Schumaker { 84168cd6fa4SBryan Schumaker } 84268cd6fa4SBryan Schumaker 84368cd6fa4SBryan Schumaker void nfs_init_cinfo(struct nfs_commit_info *cinfo, 84468cd6fa4SBryan Schumaker struct inode *inode, 84568cd6fa4SBryan Schumaker struct nfs_direct_req *dreq) 84668cd6fa4SBryan Schumaker { 84768cd6fa4SBryan Schumaker } 84868cd6fa4SBryan Schumaker 8491763da12SFred Isaman void 850ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 851ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 8528e821cadSTrond Myklebust { 8538e821cadSTrond Myklebust } 8548e821cadSTrond Myklebust 8558dd37758STrond Myklebust static void 856e468bae9STrond Myklebust nfs_clear_request_commit(struct nfs_page *req) 857e468bae9STrond Myklebust { 858e468bae9STrond Myklebust } 859e468bae9STrond Myklebust 860d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 8618e821cadSTrond Myklebust { 8628e821cadSTrond Myklebust return 0; 8638e821cadSTrond Myklebust } 8648e821cadSTrond Myklebust 8651da177e4SLinus Torvalds #endif 8661da177e4SLinus Torvalds 867061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 8686c75dc0dSFred Isaman { 869ea2cf228SFred Isaman struct nfs_commit_info cinfo; 8706c75dc0dSFred Isaman unsigned long bytes = 0; 8716c75dc0dSFred Isaman 8726c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 8736c75dc0dSFred Isaman goto out; 874ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 8756c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 8766c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 8776c75dc0dSFred Isaman 8786c75dc0dSFred Isaman bytes += req->wb_bytes; 8796c75dc0dSFred Isaman nfs_list_remove_request(req); 8806c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 8816c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 882d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 8836c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 8846c75dc0dSFred Isaman goto remove_req; 8856c75dc0dSFred Isaman } 886c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 887f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 888ea2cf228SFred Isaman nfs_mark_request_commit(req, hdr->lseg, &cinfo); 8896c75dc0dSFred Isaman goto next; 8906c75dc0dSFred Isaman } 8916c75dc0dSFred Isaman remove_req: 8926c75dc0dSFred Isaman nfs_inode_remove_request(req); 8936c75dc0dSFred Isaman next: 8941d1afcbcSTrond Myklebust nfs_unlock_request(req); 89520633f04SWeston Andros Adamson nfs_end_page_writeback(req); 8963aff4ebbSTrond Myklebust nfs_release_request(req); 8976c75dc0dSFred Isaman } 8986c75dc0dSFred Isaman out: 8996c75dc0dSFred Isaman hdr->release(hdr); 9006c75dc0dSFred Isaman } 9016c75dc0dSFred Isaman 90289d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 903ce59515cSAnna Schumaker unsigned long 904ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 905fb8a1f11STrond Myklebust { 906ea2cf228SFred Isaman return cinfo->mds->ncommit; 907fb8a1f11STrond Myklebust } 908fb8a1f11STrond Myklebust 909ea2cf228SFred Isaman /* cinfo->lock held by caller */ 9101763da12SFred Isaman int 911ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 912ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 913d6d6dc7cSFred Isaman { 914d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 915d6d6dc7cSFred Isaman int ret = 0; 916d6d6dc7cSFred Isaman 917d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 9188dd37758STrond Myklebust if (!nfs_lock_request(req)) 9198dd37758STrond Myklebust continue; 9207ad84aa9STrond Myklebust kref_get(&req->wb_kref); 921ea2cf228SFred Isaman if (cond_resched_lock(cinfo->lock)) 9223b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 923ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 9248dd37758STrond Myklebust nfs_list_add_request(req, dst); 925d6d6dc7cSFred Isaman ret++; 9261763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 927d6d6dc7cSFred Isaman break; 928d6d6dc7cSFred Isaman } 929d6d6dc7cSFred Isaman return ret; 930d6d6dc7cSFred Isaman } 931d6d6dc7cSFred Isaman 9321da177e4SLinus Torvalds /* 9331da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 9341da177e4SLinus Torvalds * @inode: NFS inode to scan 935ea2cf228SFred Isaman * @dst: mds destination list 936ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 9371da177e4SLinus Torvalds * 9381da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 9391da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 9401da177e4SLinus Torvalds */ 9411763da12SFred Isaman int 942ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 943ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9441da177e4SLinus Torvalds { 945d6d6dc7cSFred Isaman int ret = 0; 946fb8a1f11STrond Myklebust 947ea2cf228SFred Isaman spin_lock(cinfo->lock); 948ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 9498dd37758STrond Myklebust const int max = INT_MAX; 950d6d6dc7cSFred Isaman 951ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 952ea2cf228SFred Isaman cinfo, max); 953ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 954d6d6dc7cSFred Isaman } 955ea2cf228SFred Isaman spin_unlock(cinfo->lock); 956ff778d02STrond Myklebust return ret; 9571da177e4SLinus Torvalds } 958d6d6dc7cSFred Isaman 959c42de9ddSTrond Myklebust #else 960ce59515cSAnna Schumaker unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 961fb8a1f11STrond Myklebust { 962fb8a1f11STrond Myklebust return 0; 963fb8a1f11STrond Myklebust } 964fb8a1f11STrond Myklebust 9651763da12SFred Isaman int nfs_scan_commit(struct inode *inode, struct list_head *dst, 966ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 967c42de9ddSTrond Myklebust { 968c42de9ddSTrond Myklebust return 0; 969c42de9ddSTrond Myklebust } 9701da177e4SLinus Torvalds #endif 9711da177e4SLinus Torvalds 9721da177e4SLinus Torvalds /* 973e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 974e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 9751da177e4SLinus Torvalds * 976e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 977e7d39069STrond Myklebust * to disk. 9781da177e4SLinus Torvalds */ 979e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 980e7d39069STrond Myklebust struct page *page, 981e7d39069STrond Myklebust unsigned int offset, 982e7d39069STrond Myklebust unsigned int bytes) 9831da177e4SLinus Torvalds { 984e7d39069STrond Myklebust struct nfs_page *req; 985e7d39069STrond Myklebust unsigned int rqend; 986e7d39069STrond Myklebust unsigned int end; 9871da177e4SLinus Torvalds int error; 988277459d2STrond Myklebust 989e7d39069STrond Myklebust if (!PagePrivate(page)) 990e7d39069STrond Myklebust return NULL; 991e7d39069STrond Myklebust 992e7d39069STrond Myklebust end = offset + bytes; 993e7d39069STrond Myklebust spin_lock(&inode->i_lock); 994e7d39069STrond Myklebust 995e7d39069STrond Myklebust for (;;) { 99684d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 997e7d39069STrond Myklebust if (req == NULL) 998e7d39069STrond Myklebust goto out_unlock; 999e7d39069STrond Myklebust 10002bfc6e56SWeston Andros Adamson /* should be handled by nfs_flush_incompatible */ 10012bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 10022bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 10032bfc6e56SWeston Andros Adamson 1004e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1005e7d39069STrond Myklebust /* 1006e7d39069STrond Myklebust * Tell the caller to flush out the request if 1007e7d39069STrond Myklebust * the offsets are non-contiguous. 1008e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1009e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1010e7d39069STrond Myklebust */ 1011e468bae9STrond Myklebust if (offset > rqend 1012e7d39069STrond Myklebust || end < req->wb_offset) 1013e7d39069STrond Myklebust goto out_flushme; 1014e7d39069STrond Myklebust 10157ad84aa9STrond Myklebust if (nfs_lock_request(req)) 1016e7d39069STrond Myklebust break; 1017e7d39069STrond Myklebust 1018e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 1019587142f8STrond Myklebust spin_unlock(&inode->i_lock); 10201da177e4SLinus Torvalds error = nfs_wait_on_request(req); 10211da177e4SLinus Torvalds nfs_release_request(req); 1022e7d39069STrond Myklebust if (error != 0) 1023e7d39069STrond Myklebust goto out_err; 1024e7d39069STrond Myklebust spin_lock(&inode->i_lock); 10251da177e4SLinus Torvalds } 10261da177e4SLinus Torvalds 10271da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 10281da177e4SLinus Torvalds if (offset < req->wb_offset) { 10291da177e4SLinus Torvalds req->wb_offset = offset; 10301da177e4SLinus Torvalds req->wb_pgbase = offset; 10311da177e4SLinus Torvalds } 10321da177e4SLinus Torvalds if (end > rqend) 10331da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1034e7d39069STrond Myklebust else 1035e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 1036e7d39069STrond Myklebust out_unlock: 1037e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 1038ca138f36SFred Isaman if (req) 10398dd37758STrond Myklebust nfs_clear_request_commit(req); 1040e7d39069STrond Myklebust return req; 1041e7d39069STrond Myklebust out_flushme: 1042e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 1043e7d39069STrond Myklebust nfs_release_request(req); 1044e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1045e7d39069STrond Myklebust out_err: 1046e7d39069STrond Myklebust return ERR_PTR(error); 1047e7d39069STrond Myklebust } 10481da177e4SLinus Torvalds 1049e7d39069STrond Myklebust /* 1050e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1051e7d39069STrond Myklebust * 1052e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1053e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1054e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1055e7d39069STrond Myklebust */ 1056e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1057e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1058e7d39069STrond Myklebust { 1059d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1060e7d39069STrond Myklebust struct nfs_page *req; 1061e7d39069STrond Myklebust 1062e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1063e7d39069STrond Myklebust if (req != NULL) 1064e7d39069STrond Myklebust goto out; 10652bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 1066e7d39069STrond Myklebust if (IS_ERR(req)) 1067e7d39069STrond Myklebust goto out; 1068d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1069efc91ed0STrond Myklebust out: 107061e930a9STrond Myklebust return req; 10711da177e4SLinus Torvalds } 10721da177e4SLinus Torvalds 1073e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1074e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1075e7d39069STrond Myklebust { 1076e7d39069STrond Myklebust struct nfs_page *req; 1077e7d39069STrond Myklebust 1078e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1079e7d39069STrond Myklebust if (IS_ERR(req)) 1080e7d39069STrond Myklebust return PTR_ERR(req); 1081e7d39069STrond Myklebust /* Update file length */ 1082e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1083d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1084a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 10851d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1086e7d39069STrond Myklebust return 0; 1087e7d39069STrond Myklebust } 1088e7d39069STrond Myklebust 10891da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 10901da177e4SLinus Torvalds { 1091cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 10922a369153STrond Myklebust struct nfs_lock_context *l_ctx; 10931da177e4SLinus Torvalds struct nfs_page *req; 10941a54533eSTrond Myklebust int do_flush, status; 10951da177e4SLinus Torvalds /* 10961da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 10971da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 10981da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 10991da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11001da177e4SLinus Torvalds * Also do the same if we find a request from an existing 11011da177e4SLinus Torvalds * dropped page. 11021da177e4SLinus Torvalds */ 11031a54533eSTrond Myklebust do { 110484d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 11051a54533eSTrond Myklebust if (req == NULL) 11061a54533eSTrond Myklebust return 0; 11072a369153STrond Myklebust l_ctx = req->wb_lock_context; 11082a369153STrond Myklebust do_flush = req->wb_page != page || req->wb_context != ctx; 11092bfc6e56SWeston Andros Adamson /* for now, flush if more than 1 request in page_group */ 11102bfc6e56SWeston Andros Adamson do_flush |= req->wb_this_page != req; 11110f1d2605STrond Myklebust if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) { 11122a369153STrond Myklebust do_flush |= l_ctx->lockowner.l_owner != current->files 11132a369153STrond Myklebust || l_ctx->lockowner.l_pid != current->tgid; 11142a369153STrond Myklebust } 11151da177e4SLinus Torvalds nfs_release_request(req); 11161a54533eSTrond Myklebust if (!do_flush) 11171a54533eSTrond Myklebust return 0; 1118d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 11191a54533eSTrond Myklebust } while (status == 0); 11201a54533eSTrond Myklebust return status; 11211da177e4SLinus Torvalds } 11221da177e4SLinus Torvalds 11231da177e4SLinus Torvalds /* 1124dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1125dc24826bSAndy Adamson * expire soon. 1126dc24826bSAndy Adamson * 1127dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1128dc24826bSAndy Adamson * 1129dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1130dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1131dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1132dc24826bSAndy Adamson */ 1133dc24826bSAndy Adamson int 1134dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1135dc24826bSAndy Adamson { 1136dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1137dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1138dc24826bSAndy Adamson 1139dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 1140dc24826bSAndy Adamson } 1141dc24826bSAndy Adamson 1142dc24826bSAndy Adamson /* 1143dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1144dc24826bSAndy Adamson */ 1145dc24826bSAndy Adamson bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx) 1146dc24826bSAndy Adamson { 1147dc24826bSAndy Adamson return rpcauth_cred_key_to_expire(ctx->cred); 1148dc24826bSAndy Adamson } 1149dc24826bSAndy Adamson 1150dc24826bSAndy Adamson /* 11515d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 11525d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 11535d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 11545d47a356STrond Myklebust */ 11558d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 11565d47a356STrond Myklebust { 1157d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1158d529ef83SJeff Layton 11598d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 11608d197a56STrond Myklebust goto out; 116118dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1162d529ef83SJeff Layton return false; 11634db72b40SJeff Layton smp_rmb(); 1164d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 11658d197a56STrond Myklebust return false; 11668d197a56STrond Myklebust out: 116718dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 116818dd78c4SScott Mayhew return false; 11698d197a56STrond Myklebust return PageUptodate(page) != 0; 11705d47a356STrond Myklebust } 11715d47a356STrond Myklebust 1172c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1173c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1174c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1175c7559663SScott Mayhew * inefficiencies. 1176c7559663SScott Mayhew * 1177263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1178263b4509SScott Mayhew * of the checks. 1179c7559663SScott Mayhew */ 1180c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1181c7559663SScott Mayhew { 1182c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1183c7559663SScott Mayhew return 0; 1184263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1185263b4509SScott Mayhew return 0; 1186c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1187c7559663SScott Mayhew return 1; 1188263b4509SScott Mayhew if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 && 1189c7559663SScott Mayhew inode->i_flock->fl_end == OFFSET_MAX && 1190263b4509SScott Mayhew inode->i_flock->fl_type != F_RDLCK)) 1191c7559663SScott Mayhew return 1; 1192c7559663SScott Mayhew return 0; 1193c7559663SScott Mayhew } 1194c7559663SScott Mayhew 11955d47a356STrond Myklebust /* 11961da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 11971da177e4SLinus Torvalds * 11981da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 11991da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 12001da177e4SLinus Torvalds */ 12011da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 12021da177e4SLinus Torvalds unsigned int offset, unsigned int count) 12031da177e4SLinus Torvalds { 1204cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1205d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 12061da177e4SLinus Torvalds int status = 0; 12071da177e4SLinus Torvalds 120891d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 120991d5b470SChuck Lever 12106de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 12116de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 12121da177e4SLinus Torvalds 1213c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 121449a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 12151da177e4SLinus Torvalds offset = 0; 12161da177e4SLinus Torvalds } 12171da177e4SLinus Torvalds 1218e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 121903fa9e84STrond Myklebust if (status < 0) 122003fa9e84STrond Myklebust nfs_set_pageerror(page); 122159b7c05fSTrond Myklebust else 122259b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 12231da177e4SLinus Torvalds 122448186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 12251da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 12261da177e4SLinus Torvalds return status; 12271da177e4SLinus Torvalds } 12281da177e4SLinus Torvalds 12293ff7576dSTrond Myklebust static int flush_task_priority(int how) 12301da177e4SLinus Torvalds { 12311da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 12321da177e4SLinus Torvalds case FLUSH_HIGHPRI: 12331da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 12341da177e4SLinus Torvalds case FLUSH_LOWPRI: 12351da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 12361da177e4SLinus Torvalds } 12371da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 12381da177e4SLinus Torvalds } 12391da177e4SLinus Torvalds 1240d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1241d45f60c6SWeston Andros Adamson struct rpc_message *msg, 12421ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 12431da177e4SLinus Torvalds { 1244d45f60c6SWeston Andros Adamson struct inode *inode = hdr->inode; 12453ff7576dSTrond Myklebust int priority = flush_task_priority(how); 12461da177e4SLinus Torvalds 12471ed26f33SAnna Schumaker task_setup_data->priority = priority; 1248d45f60c6SWeston Andros Adamson NFS_PROTO(inode)->write_setup(hdr, msg); 1249d138d5d1SAndy Adamson 12508c21c62cSWeston Andros Adamson nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client, 1251d45f60c6SWeston Andros Adamson &task_setup_data->rpc_client, msg, hdr); 1252275acaafSTrond Myklebust } 1253275acaafSTrond Myklebust 12546d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 12556d884e8fSFred * call this on each, which will prepare them to be retried on next 12566d884e8fSFred * writeback using standard nfs. 12576d884e8fSFred */ 12586d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 12596d884e8fSFred { 12606d884e8fSFred nfs_mark_request_dirty(req); 12611d1afcbcSTrond Myklebust nfs_unlock_request(req); 126220633f04SWeston Andros Adamson nfs_end_page_writeback(req); 12633aff4ebbSTrond Myklebust nfs_release_request(req); 12646d884e8fSFred } 12656d884e8fSFred 1266061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 12676c75dc0dSFred Isaman { 12686c75dc0dSFred Isaman struct nfs_page *req; 12696c75dc0dSFred Isaman 12706c75dc0dSFred Isaman while (!list_empty(head)) { 12716c75dc0dSFred Isaman req = nfs_list_entry(head->next); 12726c75dc0dSFred Isaman nfs_list_remove_request(req); 12736c75dc0dSFred Isaman nfs_redirty_request(req); 12746c75dc0dSFred Isaman } 12756c75dc0dSFred Isaman } 12766c75dc0dSFred Isaman 1277061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1278061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1279061ae2edSFred Isaman .completion = nfs_write_completion, 1280061ae2edSFred Isaman }; 1281061ae2edSFred Isaman 128257208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1283a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1284061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 12851751c363STrond Myklebust { 1286a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 128741d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1288a20c93e3SChristoph Hellwig 1289a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1290a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1291a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1292a20c93e3SChristoph Hellwig #endif 12934a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 12944a0de55cSAnna Schumaker server->wsize, ioflags); 12951751c363STrond Myklebust } 1296ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 12971751c363STrond Myklebust 1298dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1299dce81290STrond Myklebust { 130041d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1301dce81290STrond Myklebust pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1302dce81290STrond Myklebust } 13031f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1304dce81290STrond Myklebust 13051da177e4SLinus Torvalds 13060b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 13070b7c0153SFred Isaman { 13080b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 13090b7c0153SFred Isaman 13100b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 13110b7c0153SFred Isaman } 13120b7c0153SFred Isaman 1313d45f60c6SWeston Andros Adamson static void nfs_writeback_release_common(struct nfs_pgio_header *hdr) 13141da177e4SLinus Torvalds { 1315c65e6254SWeston Andros Adamson /* do nothing! */ 13161da177e4SLinus Torvalds } 13171da177e4SLinus Torvalds 13181f2edbe3STrond Myklebust /* 13191f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 13201f2edbe3STrond Myklebust */ 13211f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 13221f2edbe3STrond Myklebust { 13231f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 13241f2edbe3STrond Myklebust int kill = 0; 1325788e7a89STrond Myklebust 13261f2edbe3STrond Myklebust /* suid always must be killed */ 13271f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 13281f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 13291f2edbe3STrond Myklebust 13301f2edbe3STrond Myklebust /* 13311f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 13321f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 13331f2edbe3STrond Myklebust */ 13341f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 13351f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 13361f2edbe3STrond Myklebust 13371f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 13381f2edbe3STrond Myklebust return kill; 13391f2edbe3STrond Myklebust 13401f2edbe3STrond Myklebust return 0; 13411f2edbe3STrond Myklebust } 1342788e7a89STrond Myklebust 13431da177e4SLinus Torvalds /* 13441da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 13451da177e4SLinus Torvalds */ 1346d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1347d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 13480eecb214SAnna Schumaker struct inode *inode) 13491da177e4SLinus Torvalds { 1350788e7a89STrond Myklebust int status; 13511da177e4SLinus Torvalds 1352f551e44fSChuck Lever /* 1353f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1354f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1355f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1356f551e44fSChuck Lever * another writer had changed the file, but some applications 1357f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1358f551e44fSChuck Lever */ 1359d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1360788e7a89STrond Myklebust if (status != 0) 13610eecb214SAnna Schumaker return status; 1362d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 136391d5b470SChuck Lever 136489d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 1365d45f60c6SWeston Andros Adamson if (hdr->res.verf->committed < hdr->args.stable && 1366d45f60c6SWeston Andros Adamson task->tk_status >= 0) { 13671da177e4SLinus Torvalds /* We tried a write call, but the server did not 13681da177e4SLinus Torvalds * commit data to stable storage even though we 13691da177e4SLinus Torvalds * requested it. 13701da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 13711da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 13721da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 13731da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 13741da177e4SLinus Torvalds */ 13751da177e4SLinus Torvalds static unsigned long complain; 13761da177e4SLinus Torvalds 1377a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 13781da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 13791da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 13801da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1381cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1382d45f60c6SWeston Andros Adamson hdr->res.verf->committed, hdr->args.stable); 13831da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 13841da177e4SLinus Torvalds } 13851da177e4SLinus Torvalds } 13861da177e4SLinus Torvalds #endif 13871f2edbe3STrond Myklebust 13881f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 13891f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 13901f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 13910eecb214SAnna Schumaker return 0; 13920eecb214SAnna Schumaker } 13930eecb214SAnna Schumaker 13940eecb214SAnna Schumaker /* 13950eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 13960eecb214SAnna Schumaker */ 1397d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1398d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 13990eecb214SAnna Schumaker { 1400d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1401d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 14021f2edbe3STrond Myklebust 14031f2edbe3STrond Myklebust if (resp->count < argp->count) { 14041da177e4SLinus Torvalds static unsigned long complain; 14051da177e4SLinus Torvalds 14066c75dc0dSFred Isaman /* This a short write! */ 1407d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 140891d5b470SChuck Lever 14091da177e4SLinus Torvalds /* Has the server at least made some progress? */ 14106c75dc0dSFred Isaman if (resp->count == 0) { 14116c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 14126c75dc0dSFred Isaman printk(KERN_WARNING 14136c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 14146c75dc0dSFred Isaman argp->count); 14156c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 14166c75dc0dSFred Isaman } 1417d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 14186c75dc0dSFred Isaman task->tk_status = -EIO; 14196c75dc0dSFred Isaman return; 14206c75dc0dSFred Isaman } 14211da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 14221da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 14231da177e4SLinus Torvalds /* Resend from where the server left off */ 1424d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 14251da177e4SLinus Torvalds argp->offset += resp->count; 14261da177e4SLinus Torvalds argp->pgbase += resp->count; 14271da177e4SLinus Torvalds argp->count -= resp->count; 14281da177e4SLinus Torvalds } else { 14291da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 14301da177e4SLinus Torvalds * headaches in the case of a server crash. 14311da177e4SLinus Torvalds */ 14321da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 14331da177e4SLinus Torvalds } 1434d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 14351da177e4SLinus Torvalds } 14361da177e4SLinus Torvalds } 14371da177e4SLinus Torvalds 14381da177e4SLinus Torvalds 143989d77c8fSBryan Schumaker #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) 144071d0a611STrond Myklebust static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) 144171d0a611STrond Myklebust { 1442b8413f98STrond Myklebust int ret; 1443b8413f98STrond Myklebust 144471d0a611STrond Myklebust if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) 144571d0a611STrond Myklebust return 1; 1446b8413f98STrond Myklebust if (!may_wait) 144771d0a611STrond Myklebust return 0; 1448b8413f98STrond Myklebust ret = out_of_line_wait_on_bit_lock(&nfsi->flags, 1449b8413f98STrond Myklebust NFS_INO_COMMIT, 1450b8413f98STrond Myklebust nfs_wait_bit_killable, 1451b8413f98STrond Myklebust TASK_KILLABLE); 1452b8413f98STrond Myklebust return (ret < 0) ? ret : 1; 145371d0a611STrond Myklebust } 145471d0a611STrond Myklebust 1455f453a54aSFred Isaman static void nfs_commit_clear_lock(struct nfs_inode *nfsi) 145671d0a611STrond Myklebust { 145771d0a611STrond Myklebust clear_bit(NFS_INO_COMMIT, &nfsi->flags); 14584e857c58SPeter Zijlstra smp_mb__after_atomic(); 145971d0a611STrond Myklebust wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); 146071d0a611STrond Myklebust } 146171d0a611STrond Myklebust 14620b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 14631da177e4SLinus Torvalds { 14640b7c0153SFred Isaman put_nfs_open_context(data->context); 14650b7c0153SFred Isaman nfs_commit_free(data); 14661da177e4SLinus Torvalds } 1467e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 14681da177e4SLinus Torvalds 14690b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 14709ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 14719f0ec176SAndy Adamson int how, int flags) 14721da177e4SLinus Torvalds { 147307737691STrond Myklebust struct rpc_task *task; 14749ace33cdSFred Isaman int priority = flush_task_priority(how); 1475bdc7f021STrond Myklebust struct rpc_message msg = { 1476bdc7f021STrond Myklebust .rpc_argp = &data->args, 1477bdc7f021STrond Myklebust .rpc_resp = &data->res, 14789ace33cdSFred Isaman .rpc_cred = data->cred, 1479bdc7f021STrond Myklebust }; 148084115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 148107737691STrond Myklebust .task = &data->task, 14829ace33cdSFred Isaman .rpc_client = clnt, 1483bdc7f021STrond Myklebust .rpc_message = &msg, 14849ace33cdSFred Isaman .callback_ops = call_ops, 148584115e1cSTrond Myklebust .callback_data = data, 1486101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 14879f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 14883ff7576dSTrond Myklebust .priority = priority, 148984115e1cSTrond Myklebust }; 1490788e7a89STrond Myklebust /* Set up the initial task struct. */ 14919ace33cdSFred Isaman NFS_PROTO(data->inode)->commit_setup(data, &msg); 14921da177e4SLinus Torvalds 1493a3f565b1SChuck Lever dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1494bdc7f021STrond Myklebust 14958c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 14968c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 14978c21c62cSWeston Andros Adamson 149807737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1499dbae4c73STrond Myklebust if (IS_ERR(task)) 1500dbae4c73STrond Myklebust return PTR_ERR(task); 1501d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1502d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 150307737691STrond Myklebust rpc_put_task(task); 1504dbae4c73STrond Myklebust return 0; 15051da177e4SLinus Torvalds } 1506e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 15071da177e4SLinus Torvalds 15081da177e4SLinus Torvalds /* 15099ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 15109ace33cdSFred Isaman */ 15110b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1512988b6dceSFred Isaman struct list_head *head, 1513f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1514f453a54aSFred Isaman struct nfs_commit_info *cinfo) 15159ace33cdSFred Isaman { 15169ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 15173d4ff43dSAl Viro struct inode *inode = first->wb_context->dentry->d_inode; 15189ace33cdSFred Isaman 15199ace33cdSFred Isaman /* Set up the RPC argument and reply structs 15209ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 15219ace33cdSFred Isaman 15229ace33cdSFred Isaman list_splice_init(head, &data->pages); 15239ace33cdSFred Isaman 15249ace33cdSFred Isaman data->inode = inode; 15259ace33cdSFred Isaman data->cred = first->wb_context->cred; 1526988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 15279ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1528f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1529b359f9d0SFred Isaman data->dreq = cinfo->dreq; 15309ace33cdSFred Isaman 15319ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 15329ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 15339ace33cdSFred Isaman data->args.offset = 0; 15349ace33cdSFred Isaman data->args.count = 0; 15350b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 15369ace33cdSFred Isaman data->res.fattr = &data->fattr; 15379ace33cdSFred Isaman data->res.verf = &data->verf; 15389ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 15399ace33cdSFred Isaman } 1540e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 15419ace33cdSFred Isaman 1542e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1543ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1544ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 154564bfeb49SFred Isaman { 154664bfeb49SFred Isaman struct nfs_page *req; 154764bfeb49SFred Isaman 154864bfeb49SFred Isaman while (!list_empty(page_list)) { 154964bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 155064bfeb49SFred Isaman nfs_list_remove_request(req); 1551ea2cf228SFred Isaman nfs_mark_request_commit(req, lseg, cinfo); 155256f9cd68SFred Isaman if (!cinfo->dreq) { 155364bfeb49SFred Isaman dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1554d56b4ddfSMel Gorman dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, 155564bfeb49SFred Isaman BDI_RECLAIMABLE); 155656f9cd68SFred Isaman } 15571d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 155864bfeb49SFred Isaman } 155964bfeb49SFred Isaman } 1560e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 156164bfeb49SFred Isaman 15629ace33cdSFred Isaman /* 15631da177e4SLinus Torvalds * Commit dirty pages 15641da177e4SLinus Torvalds */ 15651da177e4SLinus Torvalds static int 1566ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1567ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 15681da177e4SLinus Torvalds { 15690b7c0153SFred Isaman struct nfs_commit_data *data; 15701da177e4SLinus Torvalds 1571c9d8f89dSTrond Myklebust data = nfs_commitdata_alloc(); 15721da177e4SLinus Torvalds 15731da177e4SLinus Torvalds if (!data) 15741da177e4SLinus Torvalds goto out_bad; 15751da177e4SLinus Torvalds 15761da177e4SLinus Torvalds /* Set up the argument struct */ 1577f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1578f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 15799f0ec176SAndy Adamson return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, 15809f0ec176SAndy Adamson how, 0); 15811da177e4SLinus Torvalds out_bad: 1582ea2cf228SFred Isaman nfs_retry_commit(head, NULL, cinfo); 1583f453a54aSFred Isaman cinfo->completion_ops->error_cleanup(NFS_I(inode)); 15841da177e4SLinus Torvalds return -ENOMEM; 15851da177e4SLinus Torvalds } 15861da177e4SLinus Torvalds 15871da177e4SLinus Torvalds /* 15881da177e4SLinus Torvalds * COMMIT call returned 15891da177e4SLinus Torvalds */ 1590788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 15911da177e4SLinus Torvalds { 15920b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 15931da177e4SLinus Torvalds 1594a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 15951da177e4SLinus Torvalds task->tk_pid, task->tk_status); 15961da177e4SLinus Torvalds 1597788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1598c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1599c9d8f89dSTrond Myklebust } 1600c9d8f89dSTrond Myklebust 1601f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1602c9d8f89dSTrond Myklebust { 1603c9d8f89dSTrond Myklebust struct nfs_page *req; 1604c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1605f453a54aSFred Isaman struct nfs_commit_info cinfo; 1606788e7a89STrond Myklebust 16071da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 16081da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 16091da177e4SLinus Torvalds nfs_list_remove_request(req); 1610d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 16111da177e4SLinus Torvalds 16121e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 16133d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 16141e8968c5SNiels de Vos (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode), 16151da177e4SLinus Torvalds req->wb_bytes, 16161da177e4SLinus Torvalds (long long)req_offset(req)); 1617c9d8f89dSTrond Myklebust if (status < 0) { 1618c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 16191da177e4SLinus Torvalds nfs_inode_remove_request(req); 1620c9d8f89dSTrond Myklebust dprintk(", error = %d\n", status); 16211da177e4SLinus Torvalds goto next; 16221da177e4SLinus Torvalds } 16231da177e4SLinus Torvalds 16241da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 16251da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 16262f2c63bcSTrond Myklebust if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { 16271da177e4SLinus Torvalds /* We have a match */ 16281da177e4SLinus Torvalds nfs_inode_remove_request(req); 16291da177e4SLinus Torvalds dprintk(" OK\n"); 16301da177e4SLinus Torvalds goto next; 16311da177e4SLinus Torvalds } 16321da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 16331da177e4SLinus Torvalds dprintk(" mismatch\n"); 16346d884e8fSFred nfs_mark_request_dirty(req); 163505990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 16361da177e4SLinus Torvalds next: 16371d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 16381da177e4SLinus Torvalds } 1639f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1640f453a54aSFred Isaman if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) 1641f453a54aSFred Isaman nfs_commit_clear_lock(NFS_I(data->inode)); 16425917ce84SFred Isaman } 16435917ce84SFred Isaman 16445917ce84SFred Isaman static void nfs_commit_release(void *calldata) 16455917ce84SFred Isaman { 16460b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 16475917ce84SFred Isaman 1648f453a54aSFred Isaman data->completion_ops->completion(data); 1649c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 16501da177e4SLinus Torvalds } 1651788e7a89STrond Myklebust 1652788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 16530b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1654788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1655788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1656788e7a89STrond Myklebust }; 16571da177e4SLinus Torvalds 1658f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1659f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1660f453a54aSFred Isaman .error_cleanup = nfs_commit_clear_lock, 1661f453a54aSFred Isaman }; 1662f453a54aSFred Isaman 16631763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1664ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 166584c53ab5SFred Isaman { 166684c53ab5SFred Isaman int status; 166784c53ab5SFred Isaman 1668ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 166984c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1670ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 167184c53ab5SFred Isaman return status; 167284c53ab5SFred Isaman } 167384c53ab5SFred Isaman 1674b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 16751da177e4SLinus Torvalds { 16761da177e4SLinus Torvalds LIST_HEAD(head); 1677ea2cf228SFred Isaman struct nfs_commit_info cinfo; 167871d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1679b8413f98STrond Myklebust int res; 16801da177e4SLinus Torvalds 1681b8413f98STrond Myklebust res = nfs_commit_set_lock(NFS_I(inode), may_wait); 1682b8413f98STrond Myklebust if (res <= 0) 1683c5efa5fcSTrond Myklebust goto out_mark_dirty; 1684ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1685ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 16861da177e4SLinus Torvalds if (res) { 1687a861a1e1SFred Isaman int error; 1688a861a1e1SFred Isaman 1689ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 16901da177e4SLinus Torvalds if (error < 0) 16911da177e4SLinus Torvalds return error; 1692b8413f98STrond Myklebust if (!may_wait) 1693b8413f98STrond Myklebust goto out_mark_dirty; 1694b8413f98STrond Myklebust error = wait_on_bit(&NFS_I(inode)->flags, 1695b8413f98STrond Myklebust NFS_INO_COMMIT, 169671d0a611STrond Myklebust nfs_wait_bit_killable, 169771d0a611STrond Myklebust TASK_KILLABLE); 1698b8413f98STrond Myklebust if (error < 0) 1699b8413f98STrond Myklebust return error; 170071d0a611STrond Myklebust } else 170171d0a611STrond Myklebust nfs_commit_clear_lock(NFS_I(inode)); 1702c5efa5fcSTrond Myklebust return res; 1703c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1704c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1705c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1706c5efa5fcSTrond Myklebust * that the data is on the disk. 1707c5efa5fcSTrond Myklebust */ 1708c5efa5fcSTrond Myklebust out_mark_dirty: 1709c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 17101da177e4SLinus Torvalds return res; 17111da177e4SLinus Torvalds } 17128fc795f7STrond Myklebust 17138fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 17148fc795f7STrond Myklebust { 1715420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1716420e3646STrond Myklebust int flags = FLUSH_SYNC; 1717420e3646STrond Myklebust int ret = 0; 17188fc795f7STrond Myklebust 17193236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1720ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 17213236c3e1SJeff Layton return ret; 17223236c3e1SJeff Layton 1723a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1724a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1725a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1726420e3646STrond Myklebust */ 1727ea2cf228SFred Isaman if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1)) 1728420e3646STrond Myklebust goto out_mark_dirty; 1729420e3646STrond Myklebust 1730a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1731420e3646STrond Myklebust flags = 0; 1732a00dd6c0SJeff Layton } 1733a00dd6c0SJeff Layton 1734420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1735420e3646STrond Myklebust if (ret >= 0) { 1736420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1737420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1738420e3646STrond Myklebust wbc->nr_to_write -= ret; 1739420e3646STrond Myklebust else 1740420e3646STrond Myklebust wbc->nr_to_write = 0; 1741420e3646STrond Myklebust } 17428fc795f7STrond Myklebust return 0; 1743420e3646STrond Myklebust } 1744420e3646STrond Myklebust out_mark_dirty: 17458fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 17468fc795f7STrond Myklebust return ret; 17478fc795f7STrond Myklebust } 1748c63c7b05STrond Myklebust #else 17498fc795f7STrond Myklebust static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) 17508fc795f7STrond Myklebust { 17518fc795f7STrond Myklebust return 0; 17528fc795f7STrond Myklebust } 17531da177e4SLinus Torvalds #endif 17541da177e4SLinus Torvalds 17558fc795f7STrond Myklebust int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 17568fc795f7STrond Myklebust { 1757a8d8f02cSBryan Schumaker return nfs_commit_unstable_pages(inode, wbc); 1758a8d8f02cSBryan Schumaker } 175989d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1760863a3c6cSAndy Adamson 1761acdc53b2STrond Myklebust /* 1762acdc53b2STrond Myklebust * flush the inode to disk. 1763acdc53b2STrond Myklebust */ 1764acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 176534901f70STrond Myklebust { 176634901f70STrond Myklebust struct writeback_control wbc = { 176772cb77f4STrond Myklebust .sync_mode = WB_SYNC_ALL, 176834901f70STrond Myklebust .nr_to_write = LONG_MAX, 1769d7fb1207STrond Myklebust .range_start = 0, 1770d7fb1207STrond Myklebust .range_end = LLONG_MAX, 177134901f70STrond Myklebust }; 1772f4ce1299STrond Myklebust int ret; 177334901f70STrond Myklebust 1774f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1775f4ce1299STrond Myklebust 1776f4ce1299STrond Myklebust ret = sync_inode(inode, &wbc); 1777f4ce1299STrond Myklebust 1778f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1779f4ce1299STrond Myklebust return ret; 17801c75950bSTrond Myklebust } 1781ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 17821c75950bSTrond Myklebust 17831b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 17841b3b4a1aSTrond Myklebust { 17851b3b4a1aSTrond Myklebust struct nfs_page *req; 17861b3b4a1aSTrond Myklebust int ret = 0; 17871b3b4a1aSTrond Myklebust 1788ba8b06e6STrond Myklebust wait_on_page_writeback(page); 17893e217045SWeston Andros Adamson 17903e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 17913e217045SWeston Andros Adamson * request */ 17923e217045SWeston Andros Adamson req = nfs_lock_and_join_requests(page, false); 17933e217045SWeston Andros Adamson 17943e217045SWeston Andros Adamson if (IS_ERR(req)) { 17953e217045SWeston Andros Adamson ret = PTR_ERR(req); 17963e217045SWeston Andros Adamson } else if (req) { 17973e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 17983e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 17993e217045SWeston Andros Adamson * request from the inode / page_private pointer and 18003e217045SWeston Andros Adamson * release it */ 18011b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 18021b3b4a1aSTrond Myklebust /* 18031b3b4a1aSTrond Myklebust * In case nfs_inode_remove_request has marked the 18041b3b4a1aSTrond Myklebust * page as being dirty 18051b3b4a1aSTrond Myklebust */ 18061b3b4a1aSTrond Myklebust cancel_dirty_page(page, PAGE_CACHE_SIZE); 18071d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18081b3b4a1aSTrond Myklebust } 18093e217045SWeston Andros Adamson 18101b3b4a1aSTrond Myklebust return ret; 18111b3b4a1aSTrond Myklebust } 18121b3b4a1aSTrond Myklebust 18131c75950bSTrond Myklebust /* 18141c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 18151c75950bSTrond Myklebust */ 18161c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 18171c75950bSTrond Myklebust { 181829418aa4SMel Gorman loff_t range_start = page_file_offset(page); 18197f2f12d9STrond Myklebust loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); 18207f2f12d9STrond Myklebust struct writeback_control wbc = { 18217f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 18227f2f12d9STrond Myklebust .nr_to_write = 0, 18237f2f12d9STrond Myklebust .range_start = range_start, 18247f2f12d9STrond Myklebust .range_end = range_end, 18257f2f12d9STrond Myklebust }; 18267f2f12d9STrond Myklebust int ret; 18277f2f12d9STrond Myklebust 1828f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 1829f4ce1299STrond Myklebust 18300522f6adSTrond Myklebust for (;;) { 1831ba8b06e6STrond Myklebust wait_on_page_writeback(page); 18327f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 18337f2f12d9STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 18347f2f12d9STrond Myklebust if (ret < 0) 18357f2f12d9STrond Myklebust goto out_error; 18360522f6adSTrond Myklebust continue; 18377f2f12d9STrond Myklebust } 1838f4ce1299STrond Myklebust ret = 0; 18390522f6adSTrond Myklebust if (!PagePrivate(page)) 18400522f6adSTrond Myklebust break; 18410522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 18427f2f12d9STrond Myklebust if (ret < 0) 18437f2f12d9STrond Myklebust goto out_error; 18447f2f12d9STrond Myklebust } 18457f2f12d9STrond Myklebust out_error: 1846f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 18477f2f12d9STrond Myklebust return ret; 18481c75950bSTrond Myklebust } 18491c75950bSTrond Myklebust 1850074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 1851074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 1852a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 1853074cc1deSTrond Myklebust { 18542da95652SJeff Layton /* 18552da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 18562da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 18572da95652SJeff Layton * 18582da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 18592da95652SJeff Layton * that we can safely release the inode reference while holding 18602da95652SJeff Layton * the page lock. 18612da95652SJeff Layton */ 18622da95652SJeff Layton if (PagePrivate(page)) 18632da95652SJeff Layton return -EBUSY; 1864074cc1deSTrond Myklebust 18658c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 18668c209ce7SDavid Howells return -EBUSY; 1867074cc1deSTrond Myklebust 1868a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 1869074cc1deSTrond Myklebust } 1870074cc1deSTrond Myklebust #endif 1871074cc1deSTrond Myklebust 1872f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 18731da177e4SLinus Torvalds { 18741da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 18751e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 18761da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 187720c2df83SPaul Mundt NULL); 18781da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 18791da177e4SLinus Torvalds return -ENOMEM; 18801da177e4SLinus Torvalds 188193d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 18821da177e4SLinus Torvalds nfs_wdata_cachep); 18831da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 18843dd4765fSJeff Layton goto out_destroy_write_cache; 18851da177e4SLinus Torvalds 18860b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 18870b7c0153SFred Isaman sizeof(struct nfs_commit_data), 18880b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 18890b7c0153SFred Isaman NULL); 18900b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 18913dd4765fSJeff Layton goto out_destroy_write_mempool; 18920b7c0153SFred Isaman 189393d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 18944c100210SYanchuan Nian nfs_cdata_cachep); 18951da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 18963dd4765fSJeff Layton goto out_destroy_commit_cache; 18971da177e4SLinus Torvalds 189889a09141SPeter Zijlstra /* 189989a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 190089a09141SPeter Zijlstra * 190189a09141SPeter Zijlstra * 64MB: 8192k 190289a09141SPeter Zijlstra * 128MB: 11585k 190389a09141SPeter Zijlstra * 256MB: 16384k 190489a09141SPeter Zijlstra * 512MB: 23170k 190589a09141SPeter Zijlstra * 1GB: 32768k 190689a09141SPeter Zijlstra * 2GB: 46340k 190789a09141SPeter Zijlstra * 4GB: 65536k 190889a09141SPeter Zijlstra * 8GB: 92681k 190989a09141SPeter Zijlstra * 16GB: 131072k 191089a09141SPeter Zijlstra * 191189a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 191289a09141SPeter Zijlstra * Limit the default to 256M 191389a09141SPeter Zijlstra */ 191489a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 191589a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 191689a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 191789a09141SPeter Zijlstra 19181da177e4SLinus Torvalds return 0; 19193dd4765fSJeff Layton 19203dd4765fSJeff Layton out_destroy_commit_cache: 19213dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 19223dd4765fSJeff Layton out_destroy_write_mempool: 19233dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 19243dd4765fSJeff Layton out_destroy_write_cache: 19253dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 19263dd4765fSJeff Layton return -ENOMEM; 19271da177e4SLinus Torvalds } 19281da177e4SLinus Torvalds 1929266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 19301da177e4SLinus Torvalds { 19311da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 19323dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 19331da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 19341a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 19351da177e4SLinus Torvalds } 19361da177e4SLinus Torvalds 19374a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 1938a4cdda59SAnna Schumaker .rw_mode = FMODE_WRITE, 19394a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 19404a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 1941a4cdda59SAnna Schumaker .rw_release = nfs_writeback_release_common, 19420eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 19430eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 19441ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 19454a0de55cSAnna Schumaker }; 1946