1457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/fs/nfs/write.c 41da177e4SLinus Torvalds * 57c85d900STrond Myklebust * Write file data over NFS. 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> 81da177e4SLinus Torvalds */ 91da177e4SLinus Torvalds 101da177e4SLinus Torvalds #include <linux/types.h> 111da177e4SLinus Torvalds #include <linux/slab.h> 121da177e4SLinus Torvalds #include <linux/mm.h> 131da177e4SLinus Torvalds #include <linux/pagemap.h> 141da177e4SLinus Torvalds #include <linux/file.h> 151da177e4SLinus Torvalds #include <linux/writeback.h> 1689a09141SPeter Zijlstra #include <linux/swap.h> 17074cc1deSTrond Myklebust #include <linux/migrate.h> 181da177e4SLinus Torvalds 191da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h> 201da177e4SLinus Torvalds #include <linux/nfs_fs.h> 211da177e4SLinus Torvalds #include <linux/nfs_mount.h> 221da177e4SLinus Torvalds #include <linux/nfs_page.h> 233fcfab16SAndrew Morton #include <linux/backing-dev.h> 24afeacc8cSPaul Gortmaker #include <linux/export.h> 25af7cf057STrond Myklebust #include <linux/freezer.h> 26af7cf057STrond Myklebust #include <linux/wait.h> 271eb5d98fSJeff Layton #include <linux/iversion.h> 283fcfab16SAndrew Morton 297c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 30875bc3fbSTrond Myklebust #include <linux/sched/mm.h> 311da177e4SLinus Torvalds 321da177e4SLinus Torvalds #include "delegation.h" 3349a70f27STrond Myklebust #include "internal.h" 3491d5b470SChuck Lever #include "iostat.h" 35def6ed7eSAndy Adamson #include "nfs4_fs.h" 36074cc1deSTrond Myklebust #include "fscache.h" 3794ad1c80SFred Isaman #include "pnfs.h" 381da177e4SLinus Torvalds 39f4ce1299STrond Myklebust #include "nfstrace.h" 40f4ce1299STrond Myklebust 411da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 421da177e4SLinus Torvalds 431da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 441da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 451da177e4SLinus Torvalds 46919e3bd9STrond Myklebust struct nfs_io_completion { 47919e3bd9STrond Myklebust void (*complete)(void *data); 48919e3bd9STrond Myklebust void *data; 49919e3bd9STrond Myklebust struct kref refcount; 50919e3bd9STrond Myklebust }; 51919e3bd9STrond Myklebust 521da177e4SLinus Torvalds /* 531da177e4SLinus Torvalds * Local function declarations 541da177e4SLinus Torvalds */ 55f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 56788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 57061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; 58f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops; 594a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops; 60d4581383SWeston Andros Adamson static void nfs_clear_request_commit(struct nfs_page *req); 6102d1426cSWeston Andros Adamson static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 6202d1426cSWeston Andros Adamson struct inode *inode); 633a3908c8STrond Myklebust static struct nfs_page * 643a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 653a3908c8STrond Myklebust struct page *page); 661da177e4SLinus Torvalds 67e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 683feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 690b7c0153SFred Isaman static struct kmem_cache *nfs_cdata_cachep; 701da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 711da177e4SLinus Torvalds 72518662e0SNeilBrown struct nfs_commit_data *nfs_commitdata_alloc(bool never_fail) 731da177e4SLinus Torvalds { 74518662e0SNeilBrown struct nfs_commit_data *p; 7540859d7eSChuck Lever 76518662e0SNeilBrown if (never_fail) 77518662e0SNeilBrown p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); 78518662e0SNeilBrown else { 79518662e0SNeilBrown /* It is OK to do some reclaim, not no safe to wait 80518662e0SNeilBrown * for anything to be returned to the pool. 81518662e0SNeilBrown * mempool_alloc() cannot handle that particular combination, 82518662e0SNeilBrown * so we need two separate attempts. 83518662e0SNeilBrown */ 84518662e0SNeilBrown p = mempool_alloc(nfs_commit_mempool, GFP_NOWAIT); 85518662e0SNeilBrown if (!p) 86518662e0SNeilBrown p = kmem_cache_alloc(nfs_cdata_cachep, GFP_NOIO | 87518662e0SNeilBrown __GFP_NOWARN | __GFP_NORETRY); 88518662e0SNeilBrown if (!p) 89518662e0SNeilBrown return NULL; 90518662e0SNeilBrown } 91518662e0SNeilBrown 921da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 931da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 941da177e4SLinus Torvalds return p; 951da177e4SLinus Torvalds } 96e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 971da177e4SLinus Torvalds 980b7c0153SFred Isaman void nfs_commit_free(struct nfs_commit_data *p) 991da177e4SLinus Torvalds { 1001da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 1011da177e4SLinus Torvalds } 102e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 1031da177e4SLinus Torvalds 1041e7f3a48SWeston Andros Adamson static struct nfs_pgio_header *nfs_writehdr_alloc(void) 1053feb2d49STrond Myklebust { 1062b17d725STrond Myklebust struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_KERNEL); 1073feb2d49STrond Myklebust 1083feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 109fbe77c30SBenjamin Coddington p->rw_mode = FMODE_WRITE; 1103feb2d49STrond Myklebust return p; 1113feb2d49STrond Myklebust } 1123feb2d49STrond Myklebust 1131e7f3a48SWeston Andros Adamson static void nfs_writehdr_free(struct nfs_pgio_header *hdr) 1146c75dc0dSFred Isaman { 1151e7f3a48SWeston Andros Adamson mempool_free(hdr, nfs_wdata_mempool); 1163feb2d49STrond Myklebust } 1171da177e4SLinus Torvalds 118919e3bd9STrond Myklebust static struct nfs_io_completion *nfs_io_completion_alloc(gfp_t gfp_flags) 119919e3bd9STrond Myklebust { 120919e3bd9STrond Myklebust return kmalloc(sizeof(struct nfs_io_completion), gfp_flags); 121919e3bd9STrond Myklebust } 122919e3bd9STrond Myklebust 123919e3bd9STrond Myklebust static void nfs_io_completion_init(struct nfs_io_completion *ioc, 124919e3bd9STrond Myklebust void (*complete)(void *), void *data) 125919e3bd9STrond Myklebust { 126919e3bd9STrond Myklebust ioc->complete = complete; 127919e3bd9STrond Myklebust ioc->data = data; 128919e3bd9STrond Myklebust kref_init(&ioc->refcount); 129919e3bd9STrond Myklebust } 130919e3bd9STrond Myklebust 131919e3bd9STrond Myklebust static void nfs_io_completion_release(struct kref *kref) 132919e3bd9STrond Myklebust { 133919e3bd9STrond Myklebust struct nfs_io_completion *ioc = container_of(kref, 134919e3bd9STrond Myklebust struct nfs_io_completion, refcount); 135919e3bd9STrond Myklebust ioc->complete(ioc->data); 136919e3bd9STrond Myklebust kfree(ioc); 137919e3bd9STrond Myklebust } 138919e3bd9STrond Myklebust 139919e3bd9STrond Myklebust static void nfs_io_completion_get(struct nfs_io_completion *ioc) 140919e3bd9STrond Myklebust { 141919e3bd9STrond Myklebust if (ioc != NULL) 142919e3bd9STrond Myklebust kref_get(&ioc->refcount); 143919e3bd9STrond Myklebust } 144919e3bd9STrond Myklebust 145919e3bd9STrond Myklebust static void nfs_io_completion_put(struct nfs_io_completion *ioc) 146919e3bd9STrond Myklebust { 147919e3bd9STrond Myklebust if (ioc != NULL) 148919e3bd9STrond Myklebust kref_put(&ioc->refcount, nfs_io_completion_release); 149919e3bd9STrond Myklebust } 150919e3bd9STrond Myklebust 151bd37d6fcSTrond Myklebust static struct nfs_page * 152bd37d6fcSTrond Myklebust nfs_page_private_request(struct page *page) 153bd37d6fcSTrond Myklebust { 154bd37d6fcSTrond Myklebust if (!PagePrivate(page)) 155bd37d6fcSTrond Myklebust return NULL; 156bd37d6fcSTrond Myklebust return (struct nfs_page *)page_private(page); 157bd37d6fcSTrond Myklebust } 158bd37d6fcSTrond Myklebust 15984d3a9a9SWeston Andros Adamson /* 16084d3a9a9SWeston Andros Adamson * nfs_page_find_head_request_locked - find head request associated with @page 16184d3a9a9SWeston Andros Adamson * 16284d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 16384d3a9a9SWeston Andros Adamson * 16484d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 16584d3a9a9SWeston Andros Adamson */ 16629418aa4SMel Gorman static struct nfs_page * 167b30d2f04STrond Myklebust nfs_page_find_private_request(struct page *page) 168277459d2STrond Myklebust { 1694b9bb25bSTrond Myklebust struct address_space *mapping = page_file_mapping(page); 170bd37d6fcSTrond Myklebust struct nfs_page *req; 171277459d2STrond Myklebust 172b30d2f04STrond Myklebust if (!PagePrivate(page)) 173b30d2f04STrond Myklebust return NULL; 1744b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 175bd37d6fcSTrond Myklebust req = nfs_page_private_request(page); 17684d3a9a9SWeston Andros Adamson if (req) { 17784d3a9a9SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 17829418aa4SMel Gorman kref_get(&req->wb_kref); 17984d3a9a9SWeston Andros Adamson } 1804b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 181b30d2f04STrond Myklebust return req; 182b30d2f04STrond Myklebust } 18329418aa4SMel Gorman 184b30d2f04STrond Myklebust static struct nfs_page * 185b30d2f04STrond Myklebust nfs_page_find_swap_request(struct page *page) 186b30d2f04STrond Myklebust { 187b30d2f04STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 188b30d2f04STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 189b30d2f04STrond Myklebust struct nfs_page *req = NULL; 190b30d2f04STrond Myklebust if (!PageSwapCache(page)) 191b30d2f04STrond Myklebust return NULL; 192e824f99aSTrond Myklebust mutex_lock(&nfsi->commit_mutex); 193b30d2f04STrond Myklebust if (PageSwapCache(page)) { 194b30d2f04STrond Myklebust req = nfs_page_search_commits_for_head_request_locked(nfsi, 195b30d2f04STrond Myklebust page); 196b30d2f04STrond Myklebust if (req) { 197b30d2f04STrond Myklebust WARN_ON_ONCE(req->wb_head != req); 198b30d2f04STrond Myklebust kref_get(&req->wb_kref); 199b30d2f04STrond Myklebust } 200b30d2f04STrond Myklebust } 201e824f99aSTrond Myklebust mutex_unlock(&nfsi->commit_mutex); 202277459d2STrond Myklebust return req; 203277459d2STrond Myklebust } 204277459d2STrond Myklebust 20584d3a9a9SWeston Andros Adamson /* 20684d3a9a9SWeston Andros Adamson * nfs_page_find_head_request - find head request associated with @page 20784d3a9a9SWeston Andros Adamson * 20884d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 20984d3a9a9SWeston Andros Adamson */ 21084d3a9a9SWeston Andros Adamson static struct nfs_page *nfs_page_find_head_request(struct page *page) 211277459d2STrond Myklebust { 212b30d2f04STrond Myklebust struct nfs_page *req; 213277459d2STrond Myklebust 214b30d2f04STrond Myklebust req = nfs_page_find_private_request(page); 215b30d2f04STrond Myklebust if (!req) 216b30d2f04STrond Myklebust req = nfs_page_find_swap_request(page); 217277459d2STrond Myklebust return req; 218277459d2STrond Myklebust } 219277459d2STrond Myklebust 2201da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 2211da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 2221da177e4SLinus Torvalds { 223d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 224a3d01454STrond Myklebust loff_t end, i_size; 225a3d01454STrond Myklebust pgoff_t end_index; 2261da177e4SLinus Torvalds 227a3d01454STrond Myklebust spin_lock(&inode->i_lock); 228a3d01454STrond Myklebust i_size = i_size_read(inode); 22909cbfeafSKirill A. Shutemov end_index = (i_size - 1) >> PAGE_SHIFT; 2308cd79788SHuang Ying if (i_size > 0 && page_index(page) < end_index) 231a3d01454STrond Myklebust goto out; 232d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 2331da177e4SLinus Torvalds if (i_size >= end) 234a3d01454STrond Myklebust goto out; 2351da177e4SLinus Torvalds i_size_write(inode, end); 236f6cdfa6dSTrond Myklebust NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE; 237a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 238a3d01454STrond Myklebust out: 239a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 2401da177e4SLinus Torvalds } 2411da177e4SLinus Torvalds 242a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 243d2ceb7e5SBenjamin Coddington static void nfs_set_pageerror(struct address_space *mapping) 244a301b777STrond Myklebust { 245d2ceb7e5SBenjamin Coddington nfs_zap_mapping(mapping->host, mapping); 246a301b777STrond Myklebust } 247a301b777STrond Myklebust 2486fbda89bSTrond Myklebust static void nfs_mapping_set_error(struct page *page, int error) 2496fbda89bSTrond Myklebust { 2506fbda89bSTrond Myklebust SetPageError(page); 2516fbda89bSTrond Myklebust mapping_set_error(page_file_mapping(page), error); 2526fbda89bSTrond Myklebust } 2536fbda89bSTrond Myklebust 254d72ddcbaSWeston Andros Adamson /* 255d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 256d72ddcbaSWeston Andros Adamson * @head - head request of page group 257d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 258d72ddcbaSWeston Andros Adamson * 259d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 260d72ddcbaSWeston Andros Adamson * page offset @page_offset. 261d72ddcbaSWeston Andros Adamson * 262d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 263d72ddcbaSWeston Andros Adamson * match is found. 264d72ddcbaSWeston Andros Adamson * 265d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 266d72ddcbaSWeston Andros Adamson */ 267d72ddcbaSWeston Andros Adamson static struct nfs_page * 268d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 269d72ddcbaSWeston Andros Adamson { 270d72ddcbaSWeston Andros Adamson struct nfs_page *req; 271d72ddcbaSWeston Andros Adamson 272d72ddcbaSWeston Andros Adamson req = head; 273d72ddcbaSWeston Andros Adamson do { 274d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 275d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 276d72ddcbaSWeston Andros Adamson return req; 277d72ddcbaSWeston Andros Adamson 278d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 279d72ddcbaSWeston Andros Adamson } while (req != head); 280d72ddcbaSWeston Andros Adamson 281d72ddcbaSWeston Andros Adamson return NULL; 282d72ddcbaSWeston Andros Adamson } 283d72ddcbaSWeston Andros Adamson 284d72ddcbaSWeston Andros Adamson /* 285d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 286d72ddcbaSWeston Andros Adamson * @head - head request of page group 287d72ddcbaSWeston Andros Adamson * 288d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 289d72ddcbaSWeston Andros Adamson * returns false otherwise 290d72ddcbaSWeston Andros Adamson */ 291d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 292d72ddcbaSWeston Andros Adamson { 293d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 294d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 295d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 296d72ddcbaSWeston Andros Adamson 2971344b7eaSTrond Myklebust nfs_page_group_lock(req); 298d72ddcbaSWeston Andros Adamson 2997e8a30f8STrond Myklebust for (;;) { 300d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 3017e8a30f8STrond Myklebust if (!tmp) 3027e8a30f8STrond Myklebust break; 3037e8a30f8STrond Myklebust pos = tmp->wb_pgbase + tmp->wb_bytes; 304d72ddcbaSWeston Andros Adamson } 305d72ddcbaSWeston Andros Adamson 306d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 3077e8a30f8STrond Myklebust return pos >= len; 308d72ddcbaSWeston Andros Adamson } 309d72ddcbaSWeston Andros Adamson 3101da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 3111da177e4SLinus Torvalds * covers the full page. 3121da177e4SLinus Torvalds */ 313d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 3141da177e4SLinus Torvalds { 315d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 3161da177e4SLinus Torvalds return; 317d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 3181da177e4SLinus Torvalds return; 319d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 3201da177e4SLinus Torvalds } 3211da177e4SLinus Torvalds 3221da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 3231da177e4SLinus Torvalds { 324e87b4c7aSNeilBrown int ret = 0; 325cca588d6STrond Myklebust 326e87b4c7aSNeilBrown if (wbc->sync_mode == WB_SYNC_ALL) 327e87b4c7aSNeilBrown ret = FLUSH_COND_STABLE; 328e87b4c7aSNeilBrown return ret; 3291da177e4SLinus Torvalds } 3301da177e4SLinus Torvalds 3311da177e4SLinus Torvalds /* 33289a09141SPeter Zijlstra * NFS congestion control 33389a09141SPeter Zijlstra */ 33489a09141SPeter Zijlstra 33589a09141SPeter Zijlstra int nfs_congestion_kb; 33689a09141SPeter Zijlstra 33789a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 33889a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 33989a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 34089a09141SPeter Zijlstra 341deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 34289a09141SPeter Zijlstra { 3430db10944SJan Kara struct inode *inode = page_file_mapping(page)->host; 3440db10944SJan Kara struct nfs_server *nfss = NFS_SERVER(inode); 3455a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 3465a6d41b3STrond Myklebust 347deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 34889a09141SPeter Zijlstra 349277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 3500db10944SJan Kara NFS_CONGESTION_ON_THRESH) 3510db10944SJan Kara set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 35289a09141SPeter Zijlstra } 35389a09141SPeter Zijlstra 35420633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 35589a09141SPeter Zijlstra { 35620633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 35789a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 35831a01f09STrond Myklebust bool is_done; 35989a09141SPeter Zijlstra 36031a01f09STrond Myklebust is_done = nfs_page_group_sync_on_bit(req, PG_WB_END); 36131a01f09STrond Myklebust nfs_unlock_request(req); 36231a01f09STrond Myklebust if (!is_done) 36320633f04SWeston Andros Adamson return; 36420633f04SWeston Andros Adamson 36520633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 366c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 3670db10944SJan Kara clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 36889a09141SPeter Zijlstra } 36989a09141SPeter Zijlstra 370d4581383SWeston Andros Adamson /* 371d4581383SWeston Andros Adamson * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req 372d4581383SWeston Andros Adamson * 373d4581383SWeston Andros Adamson * this is a helper function for nfs_lock_and_join_requests 374d4581383SWeston Andros Adamson * 375d4581383SWeston Andros Adamson * @inode - inode associated with request page group, must be holding inode lock 376d4581383SWeston Andros Adamson * @head - head request of page group, must be holding head lock 377d4581383SWeston Andros Adamson * @req - request that couldn't lock and needs to wait on the req bit lock 378d4581383SWeston Andros Adamson * 379b5bab9bfSTrond Myklebust * NOTE: this must be called holding page_group bit lock 380b5bab9bfSTrond Myklebust * which will be released before returning. 381d4581383SWeston Andros Adamson * 382d4581383SWeston Andros Adamson * returns 0 on success, < 0 on error. 383d4581383SWeston Andros Adamson */ 38474a6d4b5STrond Myklebust static void 38574a6d4b5STrond Myklebust nfs_unroll_locks(struct inode *inode, struct nfs_page *head, 3866d17d653STrond Myklebust struct nfs_page *req) 387d4581383SWeston Andros Adamson { 388d4581383SWeston Andros Adamson struct nfs_page *tmp; 389e261f51fSTrond Myklebust 390d4581383SWeston Andros Adamson /* relinquish all the locks successfully grabbed this run */ 3915b2b5187STrond Myklebust for (tmp = head->wb_this_page ; tmp != req; tmp = tmp->wb_this_page) { 3925b2b5187STrond Myklebust if (!kref_read(&tmp->wb_kref)) 3935b2b5187STrond Myklebust continue; 3945b2b5187STrond Myklebust nfs_unlock_and_release_request(tmp); 3955b2b5187STrond Myklebust } 396e261f51fSTrond Myklebust } 397d4581383SWeston Andros Adamson 398d4581383SWeston Andros Adamson /* 399d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 400d4581383SWeston Andros Adamson * 401d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 402d4581383SWeston Andros Adamson * @old_head - the old head of the list 403d4581383SWeston Andros Adamson * 404d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 405d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 406d4581383SWeston Andros Adamson * with a reference held by some other context. 407d4581383SWeston Andros Adamson */ 408d4581383SWeston Andros Adamson static void 409d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 410b66aaa8dSTrond Myklebust struct nfs_page *old_head, 411b66aaa8dSTrond Myklebust struct inode *inode) 412d4581383SWeston Andros Adamson { 413d4581383SWeston Andros Adamson while (destroy_list) { 414d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 415d4581383SWeston Andros Adamson 416d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 417d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 418d4581383SWeston Andros Adamson 419d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 420d4581383SWeston Andros Adamson 421d4581383SWeston Andros Adamson /* make sure old group is not used */ 422d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 423d4581383SWeston Andros Adamson 424902a4c00STrond Myklebust clear_bit(PG_REMOVE, &subreq->wb_flags); 425902a4c00STrond Myklebust 4265b2b5187STrond Myklebust /* Note: races with nfs_page_group_destroy() */ 4275b2b5187STrond Myklebust if (!kref_read(&subreq->wb_kref)) { 4285b2b5187STrond Myklebust /* Check if we raced with nfs_page_group_destroy() */ 429902a4c00STrond Myklebust if (test_and_clear_bit(PG_TEARDOWN, &subreq->wb_flags)) 4305b2b5187STrond Myklebust nfs_free_request(subreq); 4315b2b5187STrond Myklebust continue; 4325b2b5187STrond Myklebust } 433d4581383SWeston Andros Adamson 4345b2b5187STrond Myklebust subreq->wb_head = subreq; 4355b2b5187STrond Myklebust 436b66aaa8dSTrond Myklebust if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) { 437d4581383SWeston Andros Adamson nfs_release_request(subreq); 438a6b6d5b8STrond Myklebust atomic_long_dec(&NFS_I(inode)->nrequests); 439d4581383SWeston Andros Adamson } 4405b2b5187STrond Myklebust 4415b2b5187STrond Myklebust /* subreq is now totally disconnected from page group or any 4425b2b5187STrond Myklebust * write / commit lists. last chance to wake any waiters */ 4435b2b5187STrond Myklebust nfs_unlock_and_release_request(subreq); 444d4581383SWeston Andros Adamson } 445d4581383SWeston Andros Adamson } 446d4581383SWeston Andros Adamson 447d4581383SWeston Andros Adamson /* 448d4581383SWeston Andros Adamson * nfs_lock_and_join_requests - join all subreqs to the head req and return 449d4581383SWeston Andros Adamson * a locked reference, cancelling any pending 450d4581383SWeston Andros Adamson * operations for this page. 451d4581383SWeston Andros Adamson * 452d4581383SWeston Andros Adamson * @page - the page used to lookup the "page group" of nfs_page structures 453d4581383SWeston Andros Adamson * 454d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 455d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 456d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 457d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 458d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 459d4581383SWeston Andros Adamson * 460d4581383SWeston Andros Adamson * Returns a locked, referenced pointer to the head request - which after 461d4581383SWeston Andros Adamson * this call is guaranteed to be the only request associated with the page. 462d4581383SWeston Andros Adamson * Returns NULL if no requests are found for @page, or a ERR_PTR if an 463d4581383SWeston Andros Adamson * error was encountered. 464d4581383SWeston Andros Adamson */ 465d4581383SWeston Andros Adamson static struct nfs_page * 4666d17d653STrond Myklebust nfs_lock_and_join_requests(struct page *page) 467d4581383SWeston Andros Adamson { 468d4581383SWeston Andros Adamson struct inode *inode = page_file_mapping(page)->host; 469d4581383SWeston Andros Adamson struct nfs_page *head, *subreq; 470d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 471d4581383SWeston Andros Adamson unsigned int total_bytes; 472d4581383SWeston Andros Adamson int ret; 473d4581383SWeston Andros Adamson 474d4581383SWeston Andros Adamson try_again: 475d4581383SWeston Andros Adamson /* 476d4581383SWeston Andros Adamson * A reference is taken only on the head request which acts as a 477d4581383SWeston Andros Adamson * reference to the whole page group - the group will not be destroyed 478d4581383SWeston Andros Adamson * until the head reference is released. 479d4581383SWeston Andros Adamson */ 480bd37d6fcSTrond Myklebust head = nfs_page_find_head_request(page); 481bd37d6fcSTrond Myklebust if (!head) 482d4581383SWeston Andros Adamson return NULL; 483d4581383SWeston Andros Adamson 484a0e265bcSTrond Myklebust /* lock the page head first in order to avoid an ABBA inefficiency */ 485a0e265bcSTrond Myklebust if (!nfs_lock_request(head)) { 486a0e265bcSTrond Myklebust ret = nfs_wait_on_request(head); 487a0e265bcSTrond Myklebust nfs_release_request(head); 488a0e265bcSTrond Myklebust if (ret < 0) 489a0e265bcSTrond Myklebust return ERR_PTR(ret); 490a0e265bcSTrond Myklebust goto try_again; 491a0e265bcSTrond Myklebust } 492bd37d6fcSTrond Myklebust 493bd37d6fcSTrond Myklebust /* Ensure that nobody removed the request before we locked it */ 494bd37d6fcSTrond Myklebust if (head != nfs_page_private_request(page) && !PageSwapCache(page)) { 495bd37d6fcSTrond Myklebust nfs_unlock_and_release_request(head); 496bd37d6fcSTrond Myklebust goto try_again; 497bd37d6fcSTrond Myklebust } 4987c3af975SWeston Andros Adamson 4991344b7eaSTrond Myklebust ret = nfs_page_group_lock(head); 5000671d8f1SMarkus Elfring if (ret < 0) 5010671d8f1SMarkus Elfring goto release_request; 5027c3af975SWeston Andros Adamson 5037c3af975SWeston Andros Adamson /* lock each request in the page group */ 504a0e265bcSTrond Myklebust total_bytes = head->wb_bytes; 505a0e265bcSTrond Myklebust for (subreq = head->wb_this_page; subreq != head; 506a0e265bcSTrond Myklebust subreq = subreq->wb_this_page) { 50774a6d4b5STrond Myklebust 5081bd5d6d0STrond Myklebust if (!kref_get_unless_zero(&subreq->wb_kref)) { 5091bd5d6d0STrond Myklebust if (subreq->wb_offset == head->wb_offset + total_bytes) 5101bd5d6d0STrond Myklebust total_bytes += subreq->wb_bytes; 5115b2b5187STrond Myklebust continue; 5121bd5d6d0STrond Myklebust } 5131bd5d6d0STrond Myklebust 51474a6d4b5STrond Myklebust while (!nfs_lock_request(subreq)) { 515b5bab9bfSTrond Myklebust /* 51674a6d4b5STrond Myklebust * Unlock page to allow nfs_page_group_sync_on_bit() 51774a6d4b5STrond Myklebust * to succeed 518b5bab9bfSTrond Myklebust */ 51974a6d4b5STrond Myklebust nfs_page_group_unlock(head); 52074a6d4b5STrond Myklebust ret = nfs_wait_on_request(subreq); 52174a6d4b5STrond Myklebust if (!ret) 5221344b7eaSTrond Myklebust ret = nfs_page_group_lock(head); 52374a6d4b5STrond Myklebust if (ret < 0) { 52474a6d4b5STrond Myklebust nfs_unroll_locks(inode, head, subreq); 5255b2b5187STrond Myklebust nfs_release_request(subreq); 5260671d8f1SMarkus Elfring goto release_request; 527d4581383SWeston Andros Adamson } 52874a6d4b5STrond Myklebust } 529e14bebf6STrond Myklebust /* 530e14bebf6STrond Myklebust * Subrequests are always contiguous, non overlapping 531e14bebf6STrond Myklebust * and in order - but may be repeated (mirrored writes). 532e14bebf6STrond Myklebust */ 533e14bebf6STrond Myklebust if (subreq->wb_offset == (head->wb_offset + total_bytes)) { 534e14bebf6STrond Myklebust /* keep track of how many bytes this group covers */ 535e14bebf6STrond Myklebust total_bytes += subreq->wb_bytes; 536e14bebf6STrond Myklebust } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || 537e14bebf6STrond Myklebust ((subreq->wb_offset + subreq->wb_bytes) > 538e14bebf6STrond Myklebust (head->wb_offset + total_bytes)))) { 5398b77484fSTrond Myklebust nfs_page_group_unlock(head); 54074a6d4b5STrond Myklebust nfs_unroll_locks(inode, head, subreq); 5415b2b5187STrond Myklebust nfs_unlock_and_release_request(subreq); 5420671d8f1SMarkus Elfring ret = -EIO; 5430671d8f1SMarkus Elfring goto release_request; 544e14bebf6STrond Myklebust } 545a0e265bcSTrond Myklebust } 546d4581383SWeston Andros Adamson 547d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 548d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 549d4581383SWeston Andros Adamson subreq = head; 550d4581383SWeston Andros Adamson do { 551411a99adSWeston Andros Adamson nfs_clear_request_commit(subreq); 552d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 553d4581383SWeston Andros Adamson } while (subreq != head); 554d4581383SWeston Andros Adamson 555d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 556d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 557d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 558d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 559d4581383SWeston Andros Adamson head->wb_this_page = head; 560d4581383SWeston Andros Adamson 561d4581383SWeston Andros Adamson /* change head request to cover whole range that 562d4581383SWeston Andros Adamson * the former page group covered */ 563d4581383SWeston Andros Adamson head->wb_bytes = total_bytes; 564d4581383SWeston Andros Adamson } 565d4581383SWeston Andros Adamson 566b66aaa8dSTrond Myklebust /* Postpone destruction of this request */ 567b66aaa8dSTrond Myklebust if (test_and_clear_bit(PG_REMOVE, &head->wb_flags)) { 568b66aaa8dSTrond Myklebust set_bit(PG_INODE_REF, &head->wb_flags); 569b66aaa8dSTrond Myklebust kref_get(&head->wb_kref); 570a6b6d5b8STrond Myklebust atomic_long_inc(&NFS_I(inode)->nrequests); 571b66aaa8dSTrond Myklebust } 572b66aaa8dSTrond Myklebust 573d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 574d4581383SWeston Andros Adamson 575b66aaa8dSTrond Myklebust nfs_destroy_unlinked_subrequests(destroy_list, head, inode); 576d4581383SWeston Andros Adamson 577b5bab9bfSTrond Myklebust /* Did we lose a race with nfs_inode_remove_request()? */ 578b5bab9bfSTrond Myklebust if (!(PagePrivate(page) || PageSwapCache(page))) { 579b5bab9bfSTrond Myklebust nfs_unlock_and_release_request(head); 580b5bab9bfSTrond Myklebust return NULL; 581b5bab9bfSTrond Myklebust } 582b5bab9bfSTrond Myklebust 583bd37d6fcSTrond Myklebust /* still holds ref on head from nfs_page_find_head_request 584d4581383SWeston Andros Adamson * and still has lock on head from lock loop */ 585d4581383SWeston Andros Adamson return head; 5860671d8f1SMarkus Elfring 5870671d8f1SMarkus Elfring release_request: 5880671d8f1SMarkus Elfring nfs_unlock_and_release_request(head); 5890671d8f1SMarkus Elfring return ERR_PTR(ret); 590612c9384STrond Myklebust } 591074cc1deSTrond Myklebust 5926fbda89bSTrond Myklebust static void nfs_write_error(struct nfs_page *req, int error) 5930bcbf039SPeng Tao { 5946fbda89bSTrond Myklebust nfs_mapping_set_error(req->wb_page, error); 5950bcbf039SPeng Tao nfs_end_page_writeback(req); 5961f84ccdfSFred Isaman nfs_release_request(req); 5970bcbf039SPeng Tao } 5980bcbf039SPeng Tao 599a6598813STrond Myklebust static bool 600a6598813STrond Myklebust nfs_error_is_fatal_on_server(int err) 601a6598813STrond Myklebust { 602a6598813STrond Myklebust switch (err) { 603a6598813STrond Myklebust case 0: 604a6598813STrond Myklebust case -ERESTARTSYS: 605a6598813STrond Myklebust case -EINTR: 606a6598813STrond Myklebust return false; 607a6598813STrond Myklebust } 608a6598813STrond Myklebust return nfs_error_is_fatal(err); 609e261f51fSTrond Myklebust } 610074cc1deSTrond Myklebust 611074cc1deSTrond Myklebust /* 612074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 613074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 614074cc1deSTrond Myklebust */ 615074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 6166d17d653STrond Myklebust struct page *page) 617074cc1deSTrond Myklebust { 6186fbda89bSTrond Myklebust struct address_space *mapping; 619074cc1deSTrond Myklebust struct nfs_page *req; 620074cc1deSTrond Myklebust int ret = 0; 621074cc1deSTrond Myklebust 6226d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 623074cc1deSTrond Myklebust if (!req) 624074cc1deSTrond Myklebust goto out; 625074cc1deSTrond Myklebust ret = PTR_ERR(req); 626074cc1deSTrond Myklebust if (IS_ERR(req)) 627074cc1deSTrond Myklebust goto out; 628074cc1deSTrond Myklebust 629deed85e7STrond Myklebust nfs_set_page_writeback(page); 630deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 631074cc1deSTrond Myklebust 632a6598813STrond Myklebust /* If there is a fatal error that covers this write, just exit */ 6336fbda89bSTrond Myklebust ret = 0; 6346fbda89bSTrond Myklebust mapping = page_file_mapping(page); 6356fbda89bSTrond Myklebust if (test_bit(AS_ENOSPC, &mapping->flags) || 6366fbda89bSTrond Myklebust test_bit(AS_EIO, &mapping->flags)) 637a6598813STrond Myklebust goto out_launder; 638a6598813STrond Myklebust 639f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 640074cc1deSTrond Myklebust ret = pgio->pg_error; 6410bcbf039SPeng Tao /* 642c373fff7STrond Myklebust * Remove the problematic req upon fatal errors on the server 6430bcbf039SPeng Tao */ 6440bcbf039SPeng Tao if (nfs_error_is_fatal(ret)) { 645c373fff7STrond Myklebust if (nfs_error_is_fatal_on_server(ret)) 646a6598813STrond Myklebust goto out_launder; 6478fc75bedSTrond Myklebust } else 6480bcbf039SPeng Tao ret = -EAGAIN; 6498fc75bedSTrond Myklebust nfs_redirty_request(req); 65040f90271STrond Myklebust } else 65140f90271STrond Myklebust nfs_add_stats(page_file_mapping(page)->host, 65240f90271STrond Myklebust NFSIOS_WRITEPAGES, 1); 653074cc1deSTrond Myklebust out: 654074cc1deSTrond Myklebust return ret; 655a6598813STrond Myklebust out_launder: 6566fbda89bSTrond Myklebust nfs_write_error(req, ret); 65714bebe3cSTrond Myklebust return 0; 658e261f51fSTrond Myklebust } 659e261f51fSTrond Myklebust 660d6c843b9SPeng Tao static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, 661c373fff7STrond Myklebust struct nfs_pageio_descriptor *pgio) 662f758c885STrond Myklebust { 663cfb506e1STrond Myklebust int ret; 664f758c885STrond Myklebust 6658cd79788SHuang Ying nfs_pageio_cond_complete(pgio, page_index(page)); 6666d17d653STrond Myklebust ret = nfs_page_async_flush(pgio, page); 667cfb506e1STrond Myklebust if (ret == -EAGAIN) { 668cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 669cfb506e1STrond Myklebust ret = 0; 670cfb506e1STrond Myklebust } 671cfb506e1STrond Myklebust return ret; 672f758c885STrond Myklebust } 673f758c885STrond Myklebust 674e261f51fSTrond Myklebust /* 6751da177e4SLinus Torvalds * Write an mmapped page to the server. 6761da177e4SLinus Torvalds */ 677d6c843b9SPeng Tao static int nfs_writepage_locked(struct page *page, 678c373fff7STrond Myklebust struct writeback_control *wbc) 6791da177e4SLinus Torvalds { 680f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 68140f90271STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 682e261f51fSTrond Myklebust int err; 6831da177e4SLinus Torvalds 68440f90271STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 685811ed92eSTrond Myklebust nfs_pageio_init_write(&pgio, inode, 0, 686a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 687c373fff7STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 688f758c885STrond Myklebust nfs_pageio_complete(&pgio); 689f758c885STrond Myklebust if (err < 0) 6904d770ccfSTrond Myklebust return err; 691f758c885STrond Myklebust if (pgio.pg_error < 0) 692f758c885STrond Myklebust return pgio.pg_error; 693f758c885STrond Myklebust return 0; 6944d770ccfSTrond Myklebust } 6954d770ccfSTrond Myklebust 6964d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 6974d770ccfSTrond Myklebust { 698f758c885STrond Myklebust int ret; 6994d770ccfSTrond Myklebust 700c373fff7STrond Myklebust ret = nfs_writepage_locked(page, wbc); 7011da177e4SLinus Torvalds unlock_page(page); 702f758c885STrond Myklebust return ret; 703f758c885STrond Myklebust } 704f758c885STrond Myklebust 705f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 706f758c885STrond Myklebust { 707f758c885STrond Myklebust int ret; 708f758c885STrond Myklebust 709c373fff7STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 710f758c885STrond Myklebust unlock_page(page); 711f758c885STrond Myklebust return ret; 7121da177e4SLinus Torvalds } 7131da177e4SLinus Torvalds 714919e3bd9STrond Myklebust static void nfs_io_completion_commit(void *inode) 715919e3bd9STrond Myklebust { 716919e3bd9STrond Myklebust nfs_commit_inode(inode, 0); 717919e3bd9STrond Myklebust } 718919e3bd9STrond Myklebust 7191da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 7201da177e4SLinus Torvalds { 7211da177e4SLinus Torvalds struct inode *inode = mapping->host; 722c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 723875bc3fbSTrond Myklebust struct nfs_io_completion *ioc; 7241da177e4SLinus Torvalds int err; 7251da177e4SLinus Torvalds 72691d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 72791d5b470SChuck Lever 7282b17d725STrond Myklebust ioc = nfs_io_completion_alloc(GFP_KERNEL); 729919e3bd9STrond Myklebust if (ioc) 730919e3bd9STrond Myklebust nfs_io_completion_init(ioc, nfs_io_completion_commit, inode); 731919e3bd9STrond Myklebust 732a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 733a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 734919e3bd9STrond Myklebust pgio.pg_io_completion = ioc; 735f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 736c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 737919e3bd9STrond Myklebust nfs_io_completion_put(ioc); 73872cb77f4STrond Myklebust 739f758c885STrond Myklebust if (err < 0) 74072cb77f4STrond Myklebust goto out_err; 74172cb77f4STrond Myklebust err = pgio.pg_error; 74272cb77f4STrond Myklebust if (err < 0) 74372cb77f4STrond Myklebust goto out_err; 744c63c7b05STrond Myklebust return 0; 74572cb77f4STrond Myklebust out_err: 74672cb77f4STrond Myklebust return err; 7471da177e4SLinus Torvalds } 7481da177e4SLinus Torvalds 7491da177e4SLinus Torvalds /* 7501da177e4SLinus Torvalds * Insert a write request into an inode 7511da177e4SLinus Torvalds */ 752d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 7531da177e4SLinus Torvalds { 7544b9bb25bSTrond Myklebust struct address_space *mapping = page_file_mapping(req->wb_page); 7551da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 756e7d39069STrond Myklebust 7572bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 7582bfc6e56SWeston Andros Adamson 759e7d39069STrond Myklebust /* Lock the request! */ 7607ad84aa9STrond Myklebust nfs_lock_request(req); 761e7d39069STrond Myklebust 76229418aa4SMel Gorman /* 76329418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 76429418aa4SMel Gorman * with invalidate/truncate. 76529418aa4SMel Gorman */ 7664b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 7674b9bb25bSTrond Myklebust if (!nfs_have_writebacks(inode) && 7681eb5d98fSJeff Layton NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 7691eb5d98fSJeff Layton inode_inc_iversion_raw(inode); 77029418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 7712df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 772deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 773277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 77429418aa4SMel Gorman } 7754b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 776a6b6d5b8STrond Myklebust atomic_long_inc(&nfsi->nrequests); 77717089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 778cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 779cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 780cb1410c7SWeston Andros Adamson * adding subrequests. */ 78117089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 782c03b4024STrond Myklebust kref_get(&req->wb_kref); 7831da177e4SLinus Torvalds } 7841da177e4SLinus Torvalds 7851da177e4SLinus Torvalds /* 78689a09141SPeter Zijlstra * Remove a write request from an inode 7871da177e4SLinus Torvalds */ 7881da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7891da177e4SLinus Torvalds { 7904b9bb25bSTrond Myklebust struct address_space *mapping = page_file_mapping(req->wb_page); 7914b9bb25bSTrond Myklebust struct inode *inode = mapping->host; 7921da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 79320633f04SWeston Andros Adamson struct nfs_page *head; 79420633f04SWeston Andros Adamson 795a6b6d5b8STrond Myklebust atomic_long_dec(&nfsi->nrequests); 79620633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 79720633f04SWeston Andros Adamson head = req->wb_head; 7981da177e4SLinus Torvalds 7994b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 80067911c8fSAnna Schumaker if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { 80120633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 80220633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 80320633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 80429418aa4SMel Gorman } 8054b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 80620633f04SWeston Andros Adamson } 80717089a29SWeston Andros Adamson 80817089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 8091da177e4SLinus Torvalds nfs_release_request(req); 8101da177e4SLinus Torvalds } 8111da177e4SLinus Torvalds 81261822ab5STrond Myklebust static void 8136d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 81461822ab5STrond Myklebust { 81567911c8fSAnna Schumaker if (req->wb_page) 81661822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 81761822ab5STrond Myklebust } 81861822ab5STrond Myklebust 8193a3908c8STrond Myklebust /* 8203a3908c8STrond Myklebust * nfs_page_search_commits_for_head_request_locked 8213a3908c8STrond Myklebust * 8223a3908c8STrond Myklebust * Search through commit lists on @inode for the head request for @page. 8233a3908c8STrond Myklebust * Must be called while holding the inode (which is cinfo) lock. 8243a3908c8STrond Myklebust * 8253a3908c8STrond Myklebust * Returns the head request if found, or NULL if not found. 8263a3908c8STrond Myklebust */ 8273a3908c8STrond Myklebust static struct nfs_page * 8283a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 8293a3908c8STrond Myklebust struct page *page) 8303a3908c8STrond Myklebust { 8313a3908c8STrond Myklebust struct nfs_page *freq, *t; 8323a3908c8STrond Myklebust struct nfs_commit_info cinfo; 8333a3908c8STrond Myklebust struct inode *inode = &nfsi->vfs_inode; 8343a3908c8STrond Myklebust 8353a3908c8STrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 8363a3908c8STrond Myklebust 8373a3908c8STrond Myklebust /* search through pnfs commit lists */ 8383a3908c8STrond Myklebust freq = pnfs_search_commit_reqs(inode, &cinfo, page); 8393a3908c8STrond Myklebust if (freq) 8403a3908c8STrond Myklebust return freq->wb_head; 8413a3908c8STrond Myklebust 8423a3908c8STrond Myklebust /* Linearly search the commit list for the correct request */ 8433a3908c8STrond Myklebust list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 8443a3908c8STrond Myklebust if (freq->wb_page == page) 8453a3908c8STrond Myklebust return freq->wb_head; 8463a3908c8STrond Myklebust } 8473a3908c8STrond Myklebust 8483a3908c8STrond Myklebust return NULL; 8493a3908c8STrond Myklebust } 8503a3908c8STrond Myklebust 8518dd37758STrond Myklebust /** 85286d80f97STrond Myklebust * nfs_request_add_commit_list_locked - add request to a commit list 85386d80f97STrond Myklebust * @req: pointer to a struct nfs_page 85486d80f97STrond Myklebust * @dst: commit list head 85586d80f97STrond Myklebust * @cinfo: holds list lock and accounting info 85686d80f97STrond Myklebust * 85786d80f97STrond Myklebust * This sets the PG_CLEAN bit, updates the cinfo count of 85886d80f97STrond Myklebust * number of outstanding requests requiring a commit as well as 85986d80f97STrond Myklebust * the MM page stats. 86086d80f97STrond Myklebust * 861e824f99aSTrond Myklebust * The caller must hold NFS_I(cinfo->inode)->commit_mutex, and the 862e824f99aSTrond Myklebust * nfs_page lock. 86386d80f97STrond Myklebust */ 86486d80f97STrond Myklebust void 86586d80f97STrond Myklebust nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, 86686d80f97STrond Myklebust struct nfs_commit_info *cinfo) 86786d80f97STrond Myklebust { 86886d80f97STrond Myklebust set_bit(PG_CLEAN, &req->wb_flags); 86986d80f97STrond Myklebust nfs_list_add_request(req, dst); 8705cb953d4STrond Myklebust atomic_long_inc(&cinfo->mds->ncommit); 87186d80f97STrond Myklebust } 87286d80f97STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); 87386d80f97STrond Myklebust 87486d80f97STrond Myklebust /** 8758dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 8768dd37758STrond Myklebust * @req: pointer to a struct nfs_page 877ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8788dd37758STrond Myklebust * 879ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 8808dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8818dd37758STrond Myklebust * the MM page stats. 8828dd37758STrond Myklebust * 883ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8848dd37758STrond Myklebust * holding the nfs_page lock. 8858dd37758STrond Myklebust */ 8868dd37758STrond Myklebust void 8876272dcc6SAnna Schumaker nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) 8888dd37758STrond Myklebust { 889e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 8906272dcc6SAnna Schumaker nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); 891e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 89267911c8fSAnna Schumaker if (req->wb_page) 89386d80f97STrond Myklebust nfs_mark_page_unstable(req->wb_page, cinfo); 8948dd37758STrond Myklebust } 8958dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 8968dd37758STrond Myklebust 8978dd37758STrond Myklebust /** 8988dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 8998dd37758STrond Myklebust * @req: pointer to a nfs_page 900ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 9018dd37758STrond Myklebust * 902ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 9038dd37758STrond Myklebust * number of outstanding requests requiring a commit 9048dd37758STrond Myklebust * It does not update the MM page stats. 9058dd37758STrond Myklebust * 906ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 9078dd37758STrond Myklebust */ 9088dd37758STrond Myklebust void 909ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 910ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9118dd37758STrond Myklebust { 9128dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 9138dd37758STrond Myklebust return; 9148dd37758STrond Myklebust nfs_list_remove_request(req); 9155cb953d4STrond Myklebust atomic_long_dec(&cinfo->mds->ncommit); 9168dd37758STrond Myklebust } 9178dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 9188dd37758STrond Myklebust 919ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 920ea2cf228SFred Isaman struct inode *inode) 921ea2cf228SFred Isaman { 922fe238e60SDave Wysochanski cinfo->inode = inode; 923ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 924ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 925b359f9d0SFred Isaman cinfo->dreq = NULL; 926f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 927ea2cf228SFred Isaman } 928ea2cf228SFred Isaman 929ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 930ea2cf228SFred Isaman struct inode *inode, 931ea2cf228SFred Isaman struct nfs_direct_req *dreq) 932ea2cf228SFred Isaman { 9331763da12SFred Isaman if (dreq) 9341763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 9351763da12SFred Isaman else 936ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 937ea2cf228SFred Isaman } 938ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 9398dd37758STrond Myklebust 9401da177e4SLinus Torvalds /* 9411da177e4SLinus Torvalds * Add a request to the inode's commit list. 9421da177e4SLinus Torvalds */ 9431763da12SFred Isaman void 944ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 945b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 9461da177e4SLinus Torvalds { 947b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 9488dd37758STrond Myklebust return; 9496272dcc6SAnna Schumaker nfs_request_add_commit_list(req, cinfo); 9501da177e4SLinus Torvalds } 9518e821cadSTrond Myklebust 952d6d6dc7cSFred Isaman static void 953d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 954e468bae9STrond Myklebust { 95511fb9989SMel Gorman dec_node_page_state(page, NR_UNSTABLE_NFS); 95693f78d88STejun Heo dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, 95793f78d88STejun Heo WB_RECLAIMABLE); 958e468bae9STrond Myklebust } 959d6d6dc7cSFred Isaman 960b5bab9bfSTrond Myklebust /* Called holding the request lock on @req */ 9618dd37758STrond Myklebust static void 962d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 963d6d6dc7cSFred Isaman { 9648dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 9659fcd5960STrond Myklebust struct nfs_open_context *ctx = nfs_req_openctx(req); 9669fcd5960STrond Myklebust struct inode *inode = d_inode(ctx->dentry); 967ea2cf228SFred Isaman struct nfs_commit_info cinfo; 968d6d6dc7cSFred Isaman 969ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 970e824f99aSTrond Myklebust mutex_lock(&NFS_I(inode)->commit_mutex); 971ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 972ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 973d6d6dc7cSFred Isaman } 974e824f99aSTrond Myklebust mutex_unlock(&NFS_I(inode)->commit_mutex); 9758dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 9768dd37758STrond Myklebust } 977e468bae9STrond Myklebust } 978e468bae9STrond Myklebust 979d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 9808e821cadSTrond Myklebust { 981c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 982d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 983c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 9848e821cadSTrond Myklebust } 9858e821cadSTrond Myklebust 986919e3bd9STrond Myklebust static void nfs_async_write_init(struct nfs_pgio_header *hdr) 987919e3bd9STrond Myklebust { 988919e3bd9STrond Myklebust nfs_io_completion_get(hdr->io_completion); 989919e3bd9STrond Myklebust } 990919e3bd9STrond Myklebust 991061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 9926c75dc0dSFred Isaman { 993ea2cf228SFred Isaman struct nfs_commit_info cinfo; 9946c75dc0dSFred Isaman unsigned long bytes = 0; 9956c75dc0dSFred Isaman 9966c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 9976c75dc0dSFred Isaman goto out; 998ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 9996c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 10006c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 10016c75dc0dSFred Isaman 10026c75dc0dSFred Isaman bytes += req->wb_bytes; 10036c75dc0dSFred Isaman nfs_list_remove_request(req); 10046c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 10056c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 1006d2ceb7e5SBenjamin Coddington nfs_set_pageerror(page_file_mapping(req->wb_page)); 10076fbda89bSTrond Myklebust nfs_mapping_set_error(req->wb_page, hdr->error); 10086c75dc0dSFred Isaman goto remove_req; 10096c75dc0dSFred Isaman } 1010c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 101133344e0fSTrond Myklebust /* Reset wb_nio, since the write was successful. */ 101233344e0fSTrond Myklebust req->wb_nio = 0; 1013f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 1014b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 1015a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 10166c75dc0dSFred Isaman goto next; 10176c75dc0dSFred Isaman } 10186c75dc0dSFred Isaman remove_req: 10196c75dc0dSFred Isaman nfs_inode_remove_request(req); 10206c75dc0dSFred Isaman next: 102120633f04SWeston Andros Adamson nfs_end_page_writeback(req); 10223aff4ebbSTrond Myklebust nfs_release_request(req); 10236c75dc0dSFred Isaman } 10246c75dc0dSFred Isaman out: 1025919e3bd9STrond Myklebust nfs_io_completion_put(hdr->io_completion); 10266c75dc0dSFred Isaman hdr->release(hdr); 10276c75dc0dSFred Isaman } 10286c75dc0dSFred Isaman 1029ce59515cSAnna Schumaker unsigned long 1030ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 1031fb8a1f11STrond Myklebust { 10325cb953d4STrond Myklebust return atomic_long_read(&cinfo->mds->ncommit); 1033fb8a1f11STrond Myklebust } 1034fb8a1f11STrond Myklebust 1035e824f99aSTrond Myklebust /* NFS_I(cinfo->inode)->commit_mutex held by caller */ 10361763da12SFred Isaman int 1037ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 1038ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 1039d6d6dc7cSFred Isaman { 1040137da553STrond Myklebust struct nfs_page *req, *tmp; 1041d6d6dc7cSFred Isaman int ret = 0; 1042d6d6dc7cSFred Isaman 1043137da553STrond Myklebust restart: 1044137da553STrond Myklebust list_for_each_entry_safe(req, tmp, src, wb_list) { 10457ad84aa9STrond Myklebust kref_get(&req->wb_kref); 10462ce209c4STrond Myklebust if (!nfs_lock_request(req)) { 10472ce209c4STrond Myklebust int status; 1048137da553STrond Myklebust 1049137da553STrond Myklebust /* Prevent deadlock with nfs_lock_and_join_requests */ 1050137da553STrond Myklebust if (!list_empty(dst)) { 1051137da553STrond Myklebust nfs_release_request(req); 1052137da553STrond Myklebust continue; 1053137da553STrond Myklebust } 1054137da553STrond Myklebust /* Ensure we make progress to prevent livelock */ 10552ce209c4STrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 10562ce209c4STrond Myklebust status = nfs_wait_on_request(req); 10572ce209c4STrond Myklebust nfs_release_request(req); 10582ce209c4STrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 10592ce209c4STrond Myklebust if (status < 0) 10602ce209c4STrond Myklebust break; 1061137da553STrond Myklebust goto restart; 10622ce209c4STrond Myklebust } 1063ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 10645d2a9d9dSTrond Myklebust clear_bit(PG_COMMIT_TO_DS, &req->wb_flags); 10658dd37758STrond Myklebust nfs_list_add_request(req, dst); 1066d6d6dc7cSFred Isaman ret++; 10671763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 1068d6d6dc7cSFred Isaman break; 1069e824f99aSTrond Myklebust cond_resched(); 1070d6d6dc7cSFred Isaman } 1071d6d6dc7cSFred Isaman return ret; 1072d6d6dc7cSFred Isaman } 10735d2a9d9dSTrond Myklebust EXPORT_SYMBOL_GPL(nfs_scan_commit_list); 1074d6d6dc7cSFred Isaman 10751da177e4SLinus Torvalds /* 10761da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 10771da177e4SLinus Torvalds * @inode: NFS inode to scan 1078ea2cf228SFred Isaman * @dst: mds destination list 1079ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 10801da177e4SLinus Torvalds * 10811da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 10821da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 10831da177e4SLinus Torvalds */ 10841763da12SFred Isaman int 1085ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 1086ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 10871da177e4SLinus Torvalds { 1088d6d6dc7cSFred Isaman int ret = 0; 1089fb8a1f11STrond Myklebust 10905cb953d4STrond Myklebust if (!atomic_long_read(&cinfo->mds->ncommit)) 10915cb953d4STrond Myklebust return 0; 1092e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 10935cb953d4STrond Myklebust if (atomic_long_read(&cinfo->mds->ncommit) > 0) { 10948dd37758STrond Myklebust const int max = INT_MAX; 1095d6d6dc7cSFred Isaman 1096ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 1097ea2cf228SFred Isaman cinfo, max); 1098ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 1099d6d6dc7cSFred Isaman } 1100e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 1101ff778d02STrond Myklebust return ret; 11021da177e4SLinus Torvalds } 1103d6d6dc7cSFred Isaman 11041da177e4SLinus Torvalds /* 1105e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1106e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 11071da177e4SLinus Torvalds * 1108e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1109e7d39069STrond Myklebust * to disk. 11101da177e4SLinus Torvalds */ 1111e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 1112e7d39069STrond Myklebust struct page *page, 1113e7d39069STrond Myklebust unsigned int offset, 1114e7d39069STrond Myklebust unsigned int bytes) 11151da177e4SLinus Torvalds { 1116e7d39069STrond Myklebust struct nfs_page *req; 1117e7d39069STrond Myklebust unsigned int rqend; 1118e7d39069STrond Myklebust unsigned int end; 11191da177e4SLinus Torvalds int error; 1120277459d2STrond Myklebust 1121e7d39069STrond Myklebust end = offset + bytes; 1122e7d39069STrond Myklebust 1123f6032f21STrond Myklebust req = nfs_lock_and_join_requests(page); 1124f6032f21STrond Myklebust if (IS_ERR_OR_NULL(req)) 1125f6032f21STrond Myklebust return req; 11262bfc6e56SWeston Andros Adamson 1127e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1128e7d39069STrond Myklebust /* 1129e7d39069STrond Myklebust * Tell the caller to flush out the request if 1130e7d39069STrond Myklebust * the offsets are non-contiguous. 1131e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1132e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1133e7d39069STrond Myklebust */ 1134f6032f21STrond Myklebust if (offset > rqend || end < req->wb_offset) 1135e7d39069STrond Myklebust goto out_flushme; 1136e7d39069STrond Myklebust 11371da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 11381da177e4SLinus Torvalds if (offset < req->wb_offset) { 11391da177e4SLinus Torvalds req->wb_offset = offset; 11401da177e4SLinus Torvalds req->wb_pgbase = offset; 11411da177e4SLinus Torvalds } 11421da177e4SLinus Torvalds if (end > rqend) 11431da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1144e7d39069STrond Myklebust else 1145e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 114633344e0fSTrond Myklebust req->wb_nio = 0; 1147e7d39069STrond Myklebust return req; 1148e7d39069STrond Myklebust out_flushme: 1149f6032f21STrond Myklebust /* 1150f6032f21STrond Myklebust * Note: we mark the request dirty here because 1151f6032f21STrond Myklebust * nfs_lock_and_join_requests() cannot preserve 1152f6032f21STrond Myklebust * commit flags, so we have to replay the write. 1153f6032f21STrond Myklebust */ 1154f6032f21STrond Myklebust nfs_mark_request_dirty(req); 1155f6032f21STrond Myklebust nfs_unlock_and_release_request(req); 1156e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1157f6032f21STrond Myklebust return (error < 0) ? ERR_PTR(error) : NULL; 1158e7d39069STrond Myklebust } 11591da177e4SLinus Torvalds 1160e7d39069STrond Myklebust /* 1161e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1162e7d39069STrond Myklebust * 1163e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1164e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1165e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1166e7d39069STrond Myklebust */ 1167e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1168e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1169e7d39069STrond Myklebust { 1170d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1171e7d39069STrond Myklebust struct nfs_page *req; 1172e7d39069STrond Myklebust 1173e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1174e7d39069STrond Myklebust if (req != NULL) 1175e7d39069STrond Myklebust goto out; 117628b1d3f5STrond Myklebust req = nfs_create_request(ctx, page, offset, bytes); 1177e7d39069STrond Myklebust if (IS_ERR(req)) 1178e7d39069STrond Myklebust goto out; 1179d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1180efc91ed0STrond Myklebust out: 118161e930a9STrond Myklebust return req; 11821da177e4SLinus Torvalds } 11831da177e4SLinus Torvalds 1184e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1185e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1186e7d39069STrond Myklebust { 1187e7d39069STrond Myklebust struct nfs_page *req; 1188e7d39069STrond Myklebust 1189e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1190e7d39069STrond Myklebust if (IS_ERR(req)) 1191e7d39069STrond Myklebust return PTR_ERR(req); 1192e7d39069STrond Myklebust /* Update file length */ 1193e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1194d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1195a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11961d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1197e7d39069STrond Myklebust return 0; 1198e7d39069STrond Myklebust } 1199e7d39069STrond Myklebust 12001da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 12011da177e4SLinus Torvalds { 1202cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 12032a369153STrond Myklebust struct nfs_lock_context *l_ctx; 1204bd61e0a9SJeff Layton struct file_lock_context *flctx = file_inode(file)->i_flctx; 12051da177e4SLinus Torvalds struct nfs_page *req; 12061a54533eSTrond Myklebust int do_flush, status; 12071da177e4SLinus Torvalds /* 12081da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 12091da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 12101da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 12111da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 12121da177e4SLinus Torvalds * Also do the same if we find a request from an existing 12131da177e4SLinus Torvalds * dropped page. 12141da177e4SLinus Torvalds */ 12151a54533eSTrond Myklebust do { 121684d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 12171a54533eSTrond Myklebust if (req == NULL) 12181a54533eSTrond Myklebust return 0; 12192a369153STrond Myklebust l_ctx = req->wb_lock_context; 1220138a2935STrond Myklebust do_flush = req->wb_page != page || 12219fcd5960STrond Myklebust !nfs_match_open_context(nfs_req_openctx(req), ctx); 1222bd61e0a9SJeff Layton if (l_ctx && flctx && 1223bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1224bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 1225d51fdb87SNeilBrown do_flush |= l_ctx->lockowner != current->files; 12265263e31eSJeff Layton } 12271da177e4SLinus Torvalds nfs_release_request(req); 12281a54533eSTrond Myklebust if (!do_flush) 12291a54533eSTrond Myklebust return 0; 1230d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 12311a54533eSTrond Myklebust } while (status == 0); 12321a54533eSTrond Myklebust return status; 12331da177e4SLinus Torvalds } 12341da177e4SLinus Torvalds 12351da177e4SLinus Torvalds /* 1236dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1237dc24826bSAndy Adamson * expire soon. 1238dc24826bSAndy Adamson * 1239dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1240dc24826bSAndy Adamson * 1241dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1242dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1243dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1244dc24826bSAndy Adamson */ 1245dc24826bSAndy Adamson int 1246dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1247dc24826bSAndy Adamson { 1248dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1249dc24826bSAndy Adamson 1250ddf529eeSNeilBrown if (nfs_ctx_key_to_expire(ctx, inode) && 1251ddf529eeSNeilBrown !ctx->ll_cred) 1252ddf529eeSNeilBrown /* Already expired! */ 1253ddf529eeSNeilBrown return -EACCES; 1254ddf529eeSNeilBrown return 0; 1255dc24826bSAndy Adamson } 1256dc24826bSAndy Adamson 1257dc24826bSAndy Adamson /* 1258dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1259dc24826bSAndy Adamson */ 1260ce52914eSScott Mayhew bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) 1261dc24826bSAndy Adamson { 1262ce52914eSScott Mayhew struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1263ddf529eeSNeilBrown struct rpc_cred *cred = ctx->ll_cred; 1264ddf529eeSNeilBrown struct auth_cred acred = { 1265a52458b4SNeilBrown .cred = ctx->cred, 1266ddf529eeSNeilBrown }; 1267ce52914eSScott Mayhew 1268ddf529eeSNeilBrown if (cred && !cred->cr_ops->crmatch(&acred, cred, 0)) { 1269ddf529eeSNeilBrown put_rpccred(cred); 1270ddf529eeSNeilBrown ctx->ll_cred = NULL; 1271ddf529eeSNeilBrown cred = NULL; 1272ddf529eeSNeilBrown } 1273ddf529eeSNeilBrown if (!cred) 1274ddf529eeSNeilBrown cred = auth->au_ops->lookup_cred(auth, &acred, 0); 1275ddf529eeSNeilBrown if (!cred || IS_ERR(cred)) 1276ddf529eeSNeilBrown return true; 1277ddf529eeSNeilBrown ctx->ll_cred = cred; 1278ddf529eeSNeilBrown return !!(cred->cr_ops->crkey_timeout && 1279ddf529eeSNeilBrown cred->cr_ops->crkey_timeout(cred)); 1280dc24826bSAndy Adamson } 1281dc24826bSAndy Adamson 1282dc24826bSAndy Adamson /* 12835d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 12845d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 12855d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 12865d47a356STrond Myklebust */ 12878d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 12885d47a356STrond Myklebust { 1289d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1290d529ef83SJeff Layton 12918d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 12928d197a56STrond Myklebust goto out; 129318dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1294d529ef83SJeff Layton return false; 12954db72b40SJeff Layton smp_rmb(); 1296d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 12978d197a56STrond Myklebust return false; 12988d197a56STrond Myklebust out: 129918dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 130018dd78c4SScott Mayhew return false; 13018d197a56STrond Myklebust return PageUptodate(page) != 0; 13025d47a356STrond Myklebust } 13035d47a356STrond Myklebust 13045263e31eSJeff Layton static bool 13055263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 13065263e31eSJeff Layton { 13075263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 13085263e31eSJeff Layton fl->fl_type == F_WRLCK; 13095263e31eSJeff Layton } 13105263e31eSJeff Layton 1311c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1312c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1313c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1314c7559663SScott Mayhew * inefficiencies. 1315c7559663SScott Mayhew * 1316263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1317263b4509SScott Mayhew * of the checks. 1318c7559663SScott Mayhew */ 1319c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1320c7559663SScott Mayhew { 13215263e31eSJeff Layton int ret; 13225263e31eSJeff Layton struct file_lock_context *flctx = inode->i_flctx; 13235263e31eSJeff Layton struct file_lock *fl; 13245263e31eSJeff Layton 1325c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1326c7559663SScott Mayhew return 0; 1327263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1328263b4509SScott Mayhew return 0; 1329c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1330c7559663SScott Mayhew return 1; 1331bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1332bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 13338fa4592aSTrond Myklebust return 1; 13345263e31eSJeff Layton 13355263e31eSJeff Layton /* Check to see if there are whole file write locks */ 13365263e31eSJeff Layton ret = 0; 13376109c850SJeff Layton spin_lock(&flctx->flc_lock); 1338bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1339bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1340bd61e0a9SJeff Layton fl_list); 1341bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 13425263e31eSJeff Layton ret = 1; 1343bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 13445263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 13455263e31eSJeff Layton fl_list); 13465263e31eSJeff Layton if (fl->fl_type == F_WRLCK) 13475263e31eSJeff Layton ret = 1; 13485263e31eSJeff Layton } 13496109c850SJeff Layton spin_unlock(&flctx->flc_lock); 13505263e31eSJeff Layton return ret; 1351c7559663SScott Mayhew } 1352c7559663SScott Mayhew 13535d47a356STrond Myklebust /* 13541da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 13551da177e4SLinus Torvalds * 13561da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 13571da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 13581da177e4SLinus Torvalds */ 13591da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 13601da177e4SLinus Torvalds unsigned int offset, unsigned int count) 13611da177e4SLinus Torvalds { 1362cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1363d2ceb7e5SBenjamin Coddington struct address_space *mapping = page_file_mapping(page); 1364d2ceb7e5SBenjamin Coddington struct inode *inode = mapping->host; 13651da177e4SLinus Torvalds int status = 0; 13661da177e4SLinus Torvalds 136791d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 136891d5b470SChuck Lever 13696de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 13706de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 13711da177e4SLinus Torvalds 1372149a4fddSBenjamin Coddington if (!count) 1373149a4fddSBenjamin Coddington goto out; 1374149a4fddSBenjamin Coddington 1375c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 137649a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 13771da177e4SLinus Torvalds offset = 0; 13781da177e4SLinus Torvalds } 13791da177e4SLinus Torvalds 1380e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 138103fa9e84STrond Myklebust if (status < 0) 1382d2ceb7e5SBenjamin Coddington nfs_set_pageerror(mapping); 138359b7c05fSTrond Myklebust else 138459b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 1385149a4fddSBenjamin Coddington out: 138648186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 13871da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 13881da177e4SLinus Torvalds return status; 13891da177e4SLinus Torvalds } 13901da177e4SLinus Torvalds 13913ff7576dSTrond Myklebust static int flush_task_priority(int how) 13921da177e4SLinus Torvalds { 13931da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 13941da177e4SLinus Torvalds case FLUSH_HIGHPRI: 13951da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 13961da177e4SLinus Torvalds case FLUSH_LOWPRI: 13971da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 13981da177e4SLinus Torvalds } 13991da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 14001da177e4SLinus Torvalds } 14011da177e4SLinus Torvalds 1402d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1403d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1404abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 14051ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 14061da177e4SLinus Torvalds { 14073ff7576dSTrond Myklebust int priority = flush_task_priority(how); 14081da177e4SLinus Torvalds 14091ed26f33SAnna Schumaker task_setup_data->priority = priority; 1410fb91fb0eSAnna Schumaker rpc_ops->write_setup(hdr, msg, &task_setup_data->rpc_client); 14118224b273SChuck Lever trace_nfs_initiate_write(hdr->inode, hdr->io_start, hdr->good_bytes, 14128224b273SChuck Lever hdr->args.stable); 1413275acaafSTrond Myklebust } 1414275acaafSTrond Myklebust 14156d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 14166d884e8fSFred * call this on each, which will prepare them to be retried on next 14176d884e8fSFred * writeback using standard nfs. 14186d884e8fSFred */ 14196d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 14206d884e8fSFred { 142133344e0fSTrond Myklebust /* Bump the transmission count */ 142233344e0fSTrond Myklebust req->wb_nio++; 14236d884e8fSFred nfs_mark_request_dirty(req); 14249fcd5960STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &nfs_req_openctx(req)->flags); 142520633f04SWeston Andros Adamson nfs_end_page_writeback(req); 14263aff4ebbSTrond Myklebust nfs_release_request(req); 14276d884e8fSFred } 14286d884e8fSFred 1429df3accb8STrond Myklebust static void nfs_async_write_error(struct list_head *head, int error) 14306c75dc0dSFred Isaman { 14316c75dc0dSFred Isaman struct nfs_page *req; 14326c75dc0dSFred Isaman 14336c75dc0dSFred Isaman while (!list_empty(head)) { 14346c75dc0dSFred Isaman req = nfs_list_entry(head->next); 14356c75dc0dSFred Isaman nfs_list_remove_request(req); 14366fbda89bSTrond Myklebust if (nfs_error_is_fatal(error)) 14376fbda89bSTrond Myklebust nfs_write_error(req, error); 14386fbda89bSTrond Myklebust else 14396c75dc0dSFred Isaman nfs_redirty_request(req); 14406c75dc0dSFred Isaman } 14416c75dc0dSFred Isaman } 14426c75dc0dSFred Isaman 1443dc602dd7STrond Myklebust static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) 1444dc602dd7STrond Myklebust { 1445df3accb8STrond Myklebust nfs_async_write_error(&hdr->pages, 0); 14467be7b3caSTrond Myklebust filemap_fdatawrite_range(hdr->inode->i_mapping, hdr->args.offset, 14477be7b3caSTrond Myklebust hdr->args.offset + hdr->args.count - 1); 1448dc602dd7STrond Myklebust } 1449dc602dd7STrond Myklebust 1450061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1451919e3bd9STrond Myklebust .init_hdr = nfs_async_write_init, 1452061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1453061ae2edSFred Isaman .completion = nfs_write_completion, 1454dc602dd7STrond Myklebust .reschedule_io = nfs_async_write_reschedule_io, 1455061ae2edSFred Isaman }; 1456061ae2edSFred Isaman 145757208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1458a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1459061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 14601751c363STrond Myklebust { 1461a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 146241d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1463a20c93e3SChristoph Hellwig 1464a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1465a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1466a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1467a20c93e3SChristoph Hellwig #endif 14684a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 14693bde7afdSTrond Myklebust server->wsize, ioflags); 14701751c363STrond Myklebust } 1471ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 14721751c363STrond Myklebust 1473dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1474dce81290STrond Myklebust { 1475a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1476a7d42ddbSWeston Andros Adamson 14776f29b9bbSKinglong Mee if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) 14786f29b9bbSKinglong Mee pgio->pg_ops->pg_cleanup(pgio); 14796f29b9bbSKinglong Mee 148041d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1481a7d42ddbSWeston Andros Adamson 1482a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1483a7d42ddbSWeston Andros Adamson 1484a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1485a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1486dce81290STrond Myklebust } 14871f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1488dce81290STrond Myklebust 14891da177e4SLinus Torvalds 14900b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 14910b7c0153SFred Isaman { 14920b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 14930b7c0153SFred Isaman 14940b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 14950b7c0153SFred Isaman } 14960b7c0153SFred Isaman 14971f2edbe3STrond Myklebust /* 14981f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 14991f2edbe3STrond Myklebust */ 15001f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 15011f2edbe3STrond Myklebust { 15021f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 15031f2edbe3STrond Myklebust int kill = 0; 1504788e7a89STrond Myklebust 15051f2edbe3STrond Myklebust /* suid always must be killed */ 15061f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 15071f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 15081f2edbe3STrond Myklebust 15091f2edbe3STrond Myklebust /* 15101f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 15111f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 15121f2edbe3STrond Myklebust */ 15131f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 15141f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 15151f2edbe3STrond Myklebust 15161f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 15171f2edbe3STrond Myklebust return kill; 15181f2edbe3STrond Myklebust 15191f2edbe3STrond Myklebust return 0; 15201f2edbe3STrond Myklebust } 1521788e7a89STrond Myklebust 1522a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1523a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1524a08a8cd3STrond Myklebust { 1525a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1526a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 15272b83d3deSTrond Myklebust u64 size = argp->offset + resp->count; 1528a08a8cd3STrond Myklebust 1529a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 15302b83d3deSTrond Myklebust fattr->size = size; 15312b83d3deSTrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { 15322b83d3deSTrond Myklebust fattr->valid &= ~NFS_ATTR_FATTR_SIZE; 1533a08a8cd3STrond Myklebust return; 15342b83d3deSTrond Myklebust } 15352b83d3deSTrond Myklebust if (size != fattr->size) 1536a08a8cd3STrond Myklebust return; 1537a08a8cd3STrond Myklebust /* Set attribute barrier */ 1538a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 15392b83d3deSTrond Myklebust /* ...and update size */ 15402b83d3deSTrond Myklebust fattr->valid |= NFS_ATTR_FATTR_SIZE; 1541a08a8cd3STrond Myklebust } 1542a08a8cd3STrond Myklebust 1543a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1544a08a8cd3STrond Myklebust { 15452b83d3deSTrond Myklebust struct nfs_fattr *fattr = &hdr->fattr; 1546a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1547a08a8cd3STrond Myklebust 1548a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1549a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1550a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1551a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1552a08a8cd3STrond Myklebust } 1553a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1554a08a8cd3STrond Myklebust 15551da177e4SLinus Torvalds /* 15561da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 15571da177e4SLinus Torvalds */ 1558d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1559d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 15600eecb214SAnna Schumaker struct inode *inode) 15611da177e4SLinus Torvalds { 1562788e7a89STrond Myklebust int status; 15631da177e4SLinus Torvalds 1564f551e44fSChuck Lever /* 1565f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1566f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1567f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1568f551e44fSChuck Lever * another writer had changed the file, but some applications 1569f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1570f551e44fSChuck Lever */ 1571d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1572788e7a89STrond Myklebust if (status != 0) 15730eecb214SAnna Schumaker return status; 15748224b273SChuck Lever 1575d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 15768224b273SChuck Lever trace_nfs_writeback_done(inode, task->tk_status, 15778224b273SChuck Lever hdr->args.offset, hdr->res.verf); 157891d5b470SChuck Lever 1579d45f60c6SWeston Andros Adamson if (hdr->res.verf->committed < hdr->args.stable && 1580d45f60c6SWeston Andros Adamson task->tk_status >= 0) { 15811da177e4SLinus Torvalds /* We tried a write call, but the server did not 15821da177e4SLinus Torvalds * commit data to stable storage even though we 15831da177e4SLinus Torvalds * requested it. 15841da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 15851da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 15861da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 15871da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 15881da177e4SLinus Torvalds */ 15891da177e4SLinus Torvalds static unsigned long complain; 15901da177e4SLinus Torvalds 1591a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 15921da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 15931da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 15941da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1595cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1596d45f60c6SWeston Andros Adamson hdr->res.verf->committed, hdr->args.stable); 15971da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 15981da177e4SLinus Torvalds } 15991da177e4SLinus Torvalds } 16001f2edbe3STrond Myklebust 16011f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 160216e14375STrond Myklebust if (nfs_should_remove_suid(inode)) { 160316e14375STrond Myklebust spin_lock(&inode->i_lock); 160416e14375STrond Myklebust NFS_I(inode)->cache_validity |= NFS_INO_INVALID_OTHER; 160516e14375STrond Myklebust spin_unlock(&inode->i_lock); 160616e14375STrond Myklebust } 16070eecb214SAnna Schumaker return 0; 16080eecb214SAnna Schumaker } 16090eecb214SAnna Schumaker 16100eecb214SAnna Schumaker /* 16110eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 16120eecb214SAnna Schumaker */ 1613d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1614d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 16150eecb214SAnna Schumaker { 1616d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1617d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 16181f2edbe3STrond Myklebust 16191f2edbe3STrond Myklebust if (resp->count < argp->count) { 16201da177e4SLinus Torvalds static unsigned long complain; 16211da177e4SLinus Torvalds 16226c75dc0dSFred Isaman /* This a short write! */ 1623d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 162491d5b470SChuck Lever 16251da177e4SLinus Torvalds /* Has the server at least made some progress? */ 16266c75dc0dSFred Isaman if (resp->count == 0) { 16276c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 16286c75dc0dSFred Isaman printk(KERN_WARNING 16296c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 16306c75dc0dSFred Isaman argp->count); 16316c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 16326c75dc0dSFred Isaman } 1633d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 16346c75dc0dSFred Isaman task->tk_status = -EIO; 16356c75dc0dSFred Isaman return; 16366c75dc0dSFred Isaman } 1637f8417b48SKinglong Mee 1638f8417b48SKinglong Mee /* For non rpc-based layout drivers, retry-through-MDS */ 1639f8417b48SKinglong Mee if (!task->tk_ops) { 1640f8417b48SKinglong Mee hdr->pnfs_error = -EAGAIN; 1641f8417b48SKinglong Mee return; 1642f8417b48SKinglong Mee } 1643f8417b48SKinglong Mee 16441da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 16451da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 16461da177e4SLinus Torvalds /* Resend from where the server left off */ 1647d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 16481da177e4SLinus Torvalds argp->offset += resp->count; 16491da177e4SLinus Torvalds argp->pgbase += resp->count; 16501da177e4SLinus Torvalds argp->count -= resp->count; 16511da177e4SLinus Torvalds } else { 16521da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 16531da177e4SLinus Torvalds * headaches in the case of a server crash. 16541da177e4SLinus Torvalds */ 16551da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 16561da177e4SLinus Torvalds } 1657d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 16581da177e4SLinus Torvalds } 16591da177e4SLinus Torvalds } 16601da177e4SLinus Torvalds 1661af7cf057STrond Myklebust static int wait_on_commit(struct nfs_mds_commit_info *cinfo) 166271d0a611STrond Myklebust { 1663723c921eSPeter Zijlstra return wait_var_event_killable(&cinfo->rpcs_out, 1664723c921eSPeter Zijlstra !atomic_read(&cinfo->rpcs_out)); 1665af7cf057STrond Myklebust } 1666af7cf057STrond Myklebust 1667af7cf057STrond Myklebust static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) 1668af7cf057STrond Myklebust { 1669af7cf057STrond Myklebust atomic_inc(&cinfo->rpcs_out); 1670af7cf057STrond Myklebust } 1671af7cf057STrond Myklebust 1672af7cf057STrond Myklebust static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) 1673af7cf057STrond Myklebust { 1674af7cf057STrond Myklebust if (atomic_dec_and_test(&cinfo->rpcs_out)) 1675723c921eSPeter Zijlstra wake_up_var(&cinfo->rpcs_out); 167671d0a611STrond Myklebust } 167771d0a611STrond Myklebust 16780b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 16791da177e4SLinus Torvalds { 16800b7c0153SFred Isaman put_nfs_open_context(data->context); 16810b7c0153SFred Isaman nfs_commit_free(data); 16821da177e4SLinus Torvalds } 1683e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 16841da177e4SLinus Torvalds 16850b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1686c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 16879ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 16889f0ec176SAndy Adamson int how, int flags) 16891da177e4SLinus Torvalds { 169007737691STrond Myklebust struct rpc_task *task; 16919ace33cdSFred Isaman int priority = flush_task_priority(how); 1692bdc7f021STrond Myklebust struct rpc_message msg = { 1693bdc7f021STrond Myklebust .rpc_argp = &data->args, 1694bdc7f021STrond Myklebust .rpc_resp = &data->res, 16959ace33cdSFred Isaman .rpc_cred = data->cred, 1696bdc7f021STrond Myklebust }; 169784115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 169807737691STrond Myklebust .task = &data->task, 16999ace33cdSFred Isaman .rpc_client = clnt, 1700bdc7f021STrond Myklebust .rpc_message = &msg, 17019ace33cdSFred Isaman .callback_ops = call_ops, 170284115e1cSTrond Myklebust .callback_data = data, 1703101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 17049f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 17053ff7576dSTrond Myklebust .priority = priority, 170684115e1cSTrond Myklebust }; 1707788e7a89STrond Myklebust /* Set up the initial task struct. */ 1708e9ae1ee2SAnna Schumaker nfs_ops->commit_setup(data, &msg, &task_setup_data.rpc_client); 17098224b273SChuck Lever trace_nfs_initiate_commit(data); 17101da177e4SLinus Torvalds 1711b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1712bdc7f021STrond Myklebust 171307737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1714dbae4c73STrond Myklebust if (IS_ERR(task)) 1715dbae4c73STrond Myklebust return PTR_ERR(task); 1716d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1717d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 171807737691STrond Myklebust rpc_put_task(task); 1719dbae4c73STrond Myklebust return 0; 17201da177e4SLinus Torvalds } 1721e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 17221da177e4SLinus Torvalds 1723378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1724378520b8SPeng Tao { 1725378520b8SPeng Tao loff_t lwb = 0; 1726378520b8SPeng Tao struct nfs_page *req; 1727378520b8SPeng Tao 1728378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1729378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1730378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1731378520b8SPeng Tao 1732378520b8SPeng Tao return lwb; 1733378520b8SPeng Tao } 1734378520b8SPeng Tao 17351da177e4SLinus Torvalds /* 17369ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 17379ace33cdSFred Isaman */ 17380b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1739988b6dceSFred Isaman struct list_head *head, 1740f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1741f453a54aSFred Isaman struct nfs_commit_info *cinfo) 17429ace33cdSFred Isaman { 17439ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 17449fcd5960STrond Myklebust struct nfs_open_context *ctx = nfs_req_openctx(first); 17459fcd5960STrond Myklebust struct inode *inode = d_inode(ctx->dentry); 17469ace33cdSFred Isaman 17479ace33cdSFred Isaman /* Set up the RPC argument and reply structs 17489ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 17499ace33cdSFred Isaman 17509ace33cdSFred Isaman list_splice_init(head, &data->pages); 17519ace33cdSFred Isaman 17529ace33cdSFred Isaman data->inode = inode; 17539fcd5960STrond Myklebust data->cred = ctx->cred; 1754988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1755378520b8SPeng Tao /* only set lwb for pnfs commit */ 1756378520b8SPeng Tao if (lseg) 1757378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 17589ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1759f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1760b359f9d0SFred Isaman data->dreq = cinfo->dreq; 17619ace33cdSFred Isaman 17629ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 17639ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 17649ace33cdSFred Isaman data->args.offset = 0; 17659ace33cdSFred Isaman data->args.count = 0; 17669fcd5960STrond Myklebust data->context = get_nfs_open_context(ctx); 17679ace33cdSFred Isaman data->res.fattr = &data->fattr; 17689ace33cdSFred Isaman data->res.verf = &data->verf; 17699ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 17709ace33cdSFred Isaman } 1771e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 17729ace33cdSFred Isaman 1773e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1774ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1775b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1776b57ff130SWeston Andros Adamson u32 ds_commit_idx) 177764bfeb49SFred Isaman { 177864bfeb49SFred Isaman struct nfs_page *req; 177964bfeb49SFred Isaman 178064bfeb49SFred Isaman while (!list_empty(page_list)) { 178164bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 178264bfeb49SFred Isaman nfs_list_remove_request(req); 1783b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 1784487b9b8aSTom Haynes if (!cinfo->dreq) 1785487b9b8aSTom Haynes nfs_clear_page_commit(req->wb_page); 17861d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 178764bfeb49SFred Isaman } 178864bfeb49SFred Isaman } 1789e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 179064bfeb49SFred Isaman 1791b20135d0STrond Myklebust static void 1792b20135d0STrond Myklebust nfs_commit_resched_write(struct nfs_commit_info *cinfo, 1793b20135d0STrond Myklebust struct nfs_page *req) 1794b20135d0STrond Myklebust { 1795b20135d0STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 1796b20135d0STrond Myklebust } 1797b20135d0STrond Myklebust 17989ace33cdSFred Isaman /* 17991da177e4SLinus Torvalds * Commit dirty pages 18001da177e4SLinus Torvalds */ 18011da177e4SLinus Torvalds static int 1802ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1803ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 18041da177e4SLinus Torvalds { 18050b7c0153SFred Isaman struct nfs_commit_data *data; 18061da177e4SLinus Torvalds 1807ade8febdSWeston Andros Adamson /* another commit raced with us */ 1808ade8febdSWeston Andros Adamson if (list_empty(head)) 1809ade8febdSWeston Andros Adamson return 0; 1810ade8febdSWeston Andros Adamson 1811518662e0SNeilBrown data = nfs_commitdata_alloc(true); 18121da177e4SLinus Torvalds 18131da177e4SLinus Torvalds /* Set up the argument struct */ 1814f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1815f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 1816c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 1817c36aae9aSPeng Tao data->mds_ops, how, 0); 18181da177e4SLinus Torvalds } 18191da177e4SLinus Torvalds 18201da177e4SLinus Torvalds /* 18211da177e4SLinus Torvalds * COMMIT call returned 18221da177e4SLinus Torvalds */ 1823788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 18241da177e4SLinus Torvalds { 18250b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18261da177e4SLinus Torvalds 1827a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 18281da177e4SLinus Torvalds task->tk_pid, task->tk_status); 18291da177e4SLinus Torvalds 1830788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1831c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 18328224b273SChuck Lever trace_nfs_commit_done(data); 1833c9d8f89dSTrond Myklebust } 1834c9d8f89dSTrond Myklebust 1835f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1836c9d8f89dSTrond Myklebust { 1837c9d8f89dSTrond Myklebust struct nfs_page *req; 1838c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1839f453a54aSFred Isaman struct nfs_commit_info cinfo; 1840353db796SNeilBrown struct nfs_server *nfss; 1841788e7a89STrond Myklebust 18421da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 18431da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 18441da177e4SLinus Torvalds nfs_list_remove_request(req); 184567911c8fSAnna Schumaker if (req->wb_page) 1846d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 18471da177e4SLinus Torvalds 18481e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 18499fcd5960STrond Myklebust nfs_req_openctx(req)->dentry->d_sb->s_id, 18509fcd5960STrond Myklebust (unsigned long long)NFS_FILEID(d_inode(nfs_req_openctx(req)->dentry)), 18511da177e4SLinus Torvalds req->wb_bytes, 18521da177e4SLinus Torvalds (long long)req_offset(req)); 1853c9d8f89dSTrond Myklebust if (status < 0) { 18546fbda89bSTrond Myklebust if (req->wb_page) { 18556fbda89bSTrond Myklebust nfs_mapping_set_error(req->wb_page, status); 18561da177e4SLinus Torvalds nfs_inode_remove_request(req); 18576fbda89bSTrond Myklebust } 1858ddeaa637SJoe Perches dprintk_cont(", error = %d\n", status); 18591da177e4SLinus Torvalds goto next; 18601da177e4SLinus Torvalds } 18611da177e4SLinus Torvalds 18621da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 18631da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 18648fc3c386STrond Myklebust if (!nfs_write_verifier_cmp(&req->wb_verf, &data->verf.verifier)) { 18651da177e4SLinus Torvalds /* We have a match */ 186638a33101SKinglong Mee if (req->wb_page) 18671da177e4SLinus Torvalds nfs_inode_remove_request(req); 1868ddeaa637SJoe Perches dprintk_cont(" OK\n"); 18691da177e4SLinus Torvalds goto next; 18701da177e4SLinus Torvalds } 18711da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 1872ddeaa637SJoe Perches dprintk_cont(" mismatch\n"); 18736d884e8fSFred nfs_mark_request_dirty(req); 18749fcd5960STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &nfs_req_openctx(req)->flags); 18751da177e4SLinus Torvalds next: 18761d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18777f1bda44STrond Myklebust /* Latency breaker */ 18787f1bda44STrond Myklebust cond_resched(); 18791da177e4SLinus Torvalds } 1880353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1881353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 18820db10944SJan Kara clear_bdi_congested(inode_to_bdi(data->inode), BLK_RW_ASYNC); 1883353db796SNeilBrown 1884f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1885af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18865917ce84SFred Isaman } 18875917ce84SFred Isaman 18885917ce84SFred Isaman static void nfs_commit_release(void *calldata) 18895917ce84SFred Isaman { 18900b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18915917ce84SFred Isaman 1892f453a54aSFred Isaman data->completion_ops->completion(data); 1893c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 18941da177e4SLinus Torvalds } 1895788e7a89STrond Myklebust 1896788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 18970b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1898788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1899788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1900788e7a89STrond Myklebust }; 19011da177e4SLinus Torvalds 1902f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1903f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1904b20135d0STrond Myklebust .resched_write = nfs_commit_resched_write, 1905f453a54aSFred Isaman }; 1906f453a54aSFred Isaman 19071763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1908ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 190984c53ab5SFred Isaman { 191084c53ab5SFred Isaman int status; 191184c53ab5SFred Isaman 1912ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 191384c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1914ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 191584c53ab5SFred Isaman return status; 191684c53ab5SFred Isaman } 191784c53ab5SFred Isaman 1918c4f24df9STrond Myklebust static int __nfs_commit_inode(struct inode *inode, int how, 1919c4f24df9STrond Myklebust struct writeback_control *wbc) 19201da177e4SLinus Torvalds { 19211da177e4SLinus Torvalds LIST_HEAD(head); 1922ea2cf228SFred Isaman struct nfs_commit_info cinfo; 192371d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1924c4f24df9STrond Myklebust int ret, nscan; 19251da177e4SLinus Torvalds 1926ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1927af7cf057STrond Myklebust nfs_commit_begin(cinfo.mds); 1928c4f24df9STrond Myklebust for (;;) { 1929c4f24df9STrond Myklebust ret = nscan = nfs_scan_commit(inode, &head, &cinfo); 1930c4f24df9STrond Myklebust if (ret <= 0) 1931c4f24df9STrond Myklebust break; 1932c4f24df9STrond Myklebust ret = nfs_generic_commit_list(inode, &head, how, &cinfo); 1933c4f24df9STrond Myklebust if (ret < 0) 1934c4f24df9STrond Myklebust break; 1935c4f24df9STrond Myklebust ret = 0; 1936c4f24df9STrond Myklebust if (wbc && wbc->sync_mode == WB_SYNC_NONE) { 1937c4f24df9STrond Myklebust if (nscan < wbc->nr_to_write) 1938c4f24df9STrond Myklebust wbc->nr_to_write -= nscan; 1939c4f24df9STrond Myklebust else 1940c4f24df9STrond Myklebust wbc->nr_to_write = 0; 1941c4f24df9STrond Myklebust } 1942c4f24df9STrond Myklebust if (nscan < INT_MAX) 1943c4f24df9STrond Myklebust break; 1944c4f24df9STrond Myklebust cond_resched(); 1945c4f24df9STrond Myklebust } 1946af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 1947c4f24df9STrond Myklebust if (ret || !may_wait) 1948c4f24df9STrond Myklebust return ret; 1949c4f24df9STrond Myklebust return wait_on_commit(cinfo.mds); 1950c4f24df9STrond Myklebust } 1951c4f24df9STrond Myklebust 1952c4f24df9STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 1953c4f24df9STrond Myklebust { 1954c4f24df9STrond Myklebust return __nfs_commit_inode(inode, how, NULL); 19551da177e4SLinus Torvalds } 1956b20135d0STrond Myklebust EXPORT_SYMBOL_GPL(nfs_commit_inode); 19578fc795f7STrond Myklebust 1958ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 19598fc795f7STrond Myklebust { 1960420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1961420e3646STrond Myklebust int flags = FLUSH_SYNC; 1962420e3646STrond Myklebust int ret = 0; 19638fc795f7STrond Myklebust 1964c4f24df9STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 19653236c3e1SJeff Layton /* no commits means nothing needs to be done */ 19665cb953d4STrond Myklebust if (!atomic_long_read(&nfsi->commit_info.ncommit)) 1967c4f24df9STrond Myklebust goto check_requests_outstanding; 19683236c3e1SJeff Layton 1969a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1970a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1971420e3646STrond Myklebust */ 19721a4edf0fSTrond Myklebust if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)) 1973420e3646STrond Myklebust goto out_mark_dirty; 1974420e3646STrond Myklebust 1975a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1976420e3646STrond Myklebust flags = 0; 1977a00dd6c0SJeff Layton } 1978a00dd6c0SJeff Layton 1979c4f24df9STrond Myklebust ret = __nfs_commit_inode(inode, flags, wbc); 1980c4f24df9STrond Myklebust if (!ret) { 1981c4f24df9STrond Myklebust if (flags & FLUSH_SYNC) 19828fc795f7STrond Myklebust return 0; 1983c4f24df9STrond Myklebust } else if (atomic_long_read(&nfsi->commit_info.ncommit)) 1984c4f24df9STrond Myklebust goto out_mark_dirty; 1985c4f24df9STrond Myklebust 1986c4f24df9STrond Myklebust check_requests_outstanding: 1987c4f24df9STrond Myklebust if (!atomic_read(&nfsi->commit_info.rpcs_out)) 1988c4f24df9STrond Myklebust return ret; 1989420e3646STrond Myklebust out_mark_dirty: 19908fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 19918fc795f7STrond Myklebust return ret; 19928fc795f7STrond Myklebust } 199389d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1994863a3c6cSAndy Adamson 1995acdc53b2STrond Myklebust /* 1996837bb1d7STrond Myklebust * Wrapper for filemap_write_and_wait_range() 1997837bb1d7STrond Myklebust * 1998837bb1d7STrond Myklebust * Needed for pNFS in order to ensure data becomes visible to the 1999837bb1d7STrond Myklebust * client. 2000837bb1d7STrond Myklebust */ 2001837bb1d7STrond Myklebust int nfs_filemap_write_and_wait_range(struct address_space *mapping, 2002837bb1d7STrond Myklebust loff_t lstart, loff_t lend) 2003837bb1d7STrond Myklebust { 2004837bb1d7STrond Myklebust int ret; 2005837bb1d7STrond Myklebust 2006837bb1d7STrond Myklebust ret = filemap_write_and_wait_range(mapping, lstart, lend); 2007837bb1d7STrond Myklebust if (ret == 0) 2008837bb1d7STrond Myklebust ret = pnfs_sync_inode(mapping->host, true); 2009837bb1d7STrond Myklebust return ret; 2010837bb1d7STrond Myklebust } 2011837bb1d7STrond Myklebust EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); 2012837bb1d7STrond Myklebust 2013837bb1d7STrond Myklebust /* 2014acdc53b2STrond Myklebust * flush the inode to disk. 2015acdc53b2STrond Myklebust */ 2016acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 201734901f70STrond Myklebust { 2018f4ce1299STrond Myklebust int ret; 201934901f70STrond Myklebust 2020f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 2021f4ce1299STrond Myklebust 20225bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 20236b196875SChuck Lever if (ret) 20246b196875SChuck Lever goto out; 20255bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20266b196875SChuck Lever if (ret < 0) 20276b196875SChuck Lever goto out; 20285bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 20296b196875SChuck Lever ret = 0; 2030f4ce1299STrond Myklebust 20316b196875SChuck Lever out: 2032f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 2033f4ce1299STrond Myklebust return ret; 20341c75950bSTrond Myklebust } 2035ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 20361c75950bSTrond Myklebust 20371b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 20381b3b4a1aSTrond Myklebust { 20391b3b4a1aSTrond Myklebust struct nfs_page *req; 20401b3b4a1aSTrond Myklebust int ret = 0; 20411b3b4a1aSTrond Myklebust 2042ba8b06e6STrond Myklebust wait_on_page_writeback(page); 20433e217045SWeston Andros Adamson 20443e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 20453e217045SWeston Andros Adamson * request */ 20466d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 20473e217045SWeston Andros Adamson 20483e217045SWeston Andros Adamson if (IS_ERR(req)) { 20493e217045SWeston Andros Adamson ret = PTR_ERR(req); 20503e217045SWeston Andros Adamson } else if (req) { 20513e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 20523e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 20533e217045SWeston Andros Adamson * request from the inode / page_private pointer and 20543e217045SWeston Andros Adamson * release it */ 20551b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 20561d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 20571b3b4a1aSTrond Myklebust } 20583e217045SWeston Andros Adamson 20591b3b4a1aSTrond Myklebust return ret; 20601b3b4a1aSTrond Myklebust } 20611b3b4a1aSTrond Myklebust 20621c75950bSTrond Myklebust /* 20631c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 20641c75950bSTrond Myklebust */ 2065c373fff7STrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 20661c75950bSTrond Myklebust { 206729418aa4SMel Gorman loff_t range_start = page_file_offset(page); 206809cbfeafSKirill A. Shutemov loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); 20697f2f12d9STrond Myklebust struct writeback_control wbc = { 20707f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 20717f2f12d9STrond Myklebust .nr_to_write = 0, 20727f2f12d9STrond Myklebust .range_start = range_start, 20737f2f12d9STrond Myklebust .range_end = range_end, 20747f2f12d9STrond Myklebust }; 20757f2f12d9STrond Myklebust int ret; 20767f2f12d9STrond Myklebust 2077f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 2078f4ce1299STrond Myklebust 20790522f6adSTrond Myklebust for (;;) { 2080ba8b06e6STrond Myklebust wait_on_page_writeback(page); 20817f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 2082c373fff7STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 20837f2f12d9STrond Myklebust if (ret < 0) 20847f2f12d9STrond Myklebust goto out_error; 20850522f6adSTrond Myklebust continue; 20867f2f12d9STrond Myklebust } 2087f4ce1299STrond Myklebust ret = 0; 20880522f6adSTrond Myklebust if (!PagePrivate(page)) 20890522f6adSTrond Myklebust break; 20900522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20917f2f12d9STrond Myklebust if (ret < 0) 20927f2f12d9STrond Myklebust goto out_error; 20937f2f12d9STrond Myklebust } 20947f2f12d9STrond Myklebust out_error: 2095f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 20967f2f12d9STrond Myklebust return ret; 20971c75950bSTrond Myklebust } 20981c75950bSTrond Myklebust 2099074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 2100074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 2101a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 2102074cc1deSTrond Myklebust { 21032da95652SJeff Layton /* 21042da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 21052da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 21062da95652SJeff Layton * 21072da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 21082da95652SJeff Layton * that we can safely release the inode reference while holding 21092da95652SJeff Layton * the page lock. 21102da95652SJeff Layton */ 21112da95652SJeff Layton if (PagePrivate(page)) 21122da95652SJeff Layton return -EBUSY; 2113074cc1deSTrond Myklebust 21148c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 21158c209ce7SDavid Howells return -EBUSY; 2116074cc1deSTrond Myklebust 2117a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 2118074cc1deSTrond Myklebust } 2119074cc1deSTrond Myklebust #endif 2120074cc1deSTrond Myklebust 2121f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 21221da177e4SLinus Torvalds { 21231da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 21241e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 21251da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 212620c2df83SPaul Mundt NULL); 21271da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 21281da177e4SLinus Torvalds return -ENOMEM; 21291da177e4SLinus Torvalds 213093d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 21311da177e4SLinus Torvalds nfs_wdata_cachep); 21321da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 21333dd4765fSJeff Layton goto out_destroy_write_cache; 21341da177e4SLinus Torvalds 21350b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 21360b7c0153SFred Isaman sizeof(struct nfs_commit_data), 21370b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 21380b7c0153SFred Isaman NULL); 21390b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 21403dd4765fSJeff Layton goto out_destroy_write_mempool; 21410b7c0153SFred Isaman 214293d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 21434c100210SYanchuan Nian nfs_cdata_cachep); 21441da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 21453dd4765fSJeff Layton goto out_destroy_commit_cache; 21461da177e4SLinus Torvalds 214789a09141SPeter Zijlstra /* 214889a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 214989a09141SPeter Zijlstra * 215089a09141SPeter Zijlstra * 64MB: 8192k 215189a09141SPeter Zijlstra * 128MB: 11585k 215289a09141SPeter Zijlstra * 256MB: 16384k 215389a09141SPeter Zijlstra * 512MB: 23170k 215489a09141SPeter Zijlstra * 1GB: 32768k 215589a09141SPeter Zijlstra * 2GB: 46340k 215689a09141SPeter Zijlstra * 4GB: 65536k 215789a09141SPeter Zijlstra * 8GB: 92681k 215889a09141SPeter Zijlstra * 16GB: 131072k 215989a09141SPeter Zijlstra * 216089a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 216189a09141SPeter Zijlstra * Limit the default to 256M 216289a09141SPeter Zijlstra */ 2163ca79b0c2SArun KS nfs_congestion_kb = (16*int_sqrt(totalram_pages())) << (PAGE_SHIFT-10); 216489a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 216589a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 216689a09141SPeter Zijlstra 21671da177e4SLinus Torvalds return 0; 21683dd4765fSJeff Layton 21693dd4765fSJeff Layton out_destroy_commit_cache: 21703dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21713dd4765fSJeff Layton out_destroy_write_mempool: 21723dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 21733dd4765fSJeff Layton out_destroy_write_cache: 21743dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 21753dd4765fSJeff Layton return -ENOMEM; 21761da177e4SLinus Torvalds } 21771da177e4SLinus Torvalds 2178266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 21791da177e4SLinus Torvalds { 21801da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 21813dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21821da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 21831a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 21841da177e4SLinus Torvalds } 21851da177e4SLinus Torvalds 21864a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 21874a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 21884a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 21890eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 21900eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 21911ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 21924a0de55cSAnna Schumaker }; 2193