11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/nfs/write.c 31da177e4SLinus Torvalds * 47c85d900STrond Myklebust * Write file data over NFS. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> 71da177e4SLinus Torvalds */ 81da177e4SLinus Torvalds 91da177e4SLinus Torvalds #include <linux/types.h> 101da177e4SLinus Torvalds #include <linux/slab.h> 111da177e4SLinus Torvalds #include <linux/mm.h> 121da177e4SLinus Torvalds #include <linux/pagemap.h> 131da177e4SLinus Torvalds #include <linux/file.h> 141da177e4SLinus Torvalds #include <linux/writeback.h> 1589a09141SPeter Zijlstra #include <linux/swap.h> 16074cc1deSTrond Myklebust #include <linux/migrate.h> 171da177e4SLinus Torvalds 181da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h> 191da177e4SLinus Torvalds #include <linux/nfs_fs.h> 201da177e4SLinus Torvalds #include <linux/nfs_mount.h> 211da177e4SLinus Torvalds #include <linux/nfs_page.h> 223fcfab16SAndrew Morton #include <linux/backing-dev.h> 23afeacc8cSPaul Gortmaker #include <linux/export.h> 24af7cf057STrond Myklebust #include <linux/freezer.h> 25af7cf057STrond Myklebust #include <linux/wait.h> 263fcfab16SAndrew Morton 277c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 281da177e4SLinus Torvalds 291da177e4SLinus Torvalds #include "delegation.h" 3049a70f27STrond Myklebust #include "internal.h" 3191d5b470SChuck Lever #include "iostat.h" 32def6ed7eSAndy Adamson #include "nfs4_fs.h" 33074cc1deSTrond Myklebust #include "fscache.h" 3494ad1c80SFred Isaman #include "pnfs.h" 351da177e4SLinus Torvalds 36f4ce1299STrond Myklebust #include "nfstrace.h" 37f4ce1299STrond Myklebust 381da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 391da177e4SLinus Torvalds 401da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 411da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 421da177e4SLinus Torvalds 43919e3bd9STrond Myklebust struct nfs_io_completion { 44919e3bd9STrond Myklebust void (*complete)(void *data); 45919e3bd9STrond Myklebust void *data; 46919e3bd9STrond Myklebust struct kref refcount; 47919e3bd9STrond Myklebust }; 48919e3bd9STrond Myklebust 491da177e4SLinus Torvalds /* 501da177e4SLinus Torvalds * Local function declarations 511da177e4SLinus Torvalds */ 52f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 53788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 54061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; 55f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops; 564a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops; 57d4581383SWeston Andros Adamson static void nfs_clear_request_commit(struct nfs_page *req); 5802d1426cSWeston Andros Adamson static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 5902d1426cSWeston Andros Adamson struct inode *inode); 603a3908c8STrond Myklebust static struct nfs_page * 613a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 623a3908c8STrond Myklebust struct page *page); 631da177e4SLinus Torvalds 64e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 653feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 660b7c0153SFred Isaman static struct kmem_cache *nfs_cdata_cachep; 671da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 681da177e4SLinus Torvalds 69518662e0SNeilBrown struct nfs_commit_data *nfs_commitdata_alloc(bool never_fail) 701da177e4SLinus Torvalds { 71518662e0SNeilBrown struct nfs_commit_data *p; 7240859d7eSChuck Lever 73518662e0SNeilBrown if (never_fail) 74518662e0SNeilBrown p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); 75518662e0SNeilBrown else { 76518662e0SNeilBrown /* It is OK to do some reclaim, not no safe to wait 77518662e0SNeilBrown * for anything to be returned to the pool. 78518662e0SNeilBrown * mempool_alloc() cannot handle that particular combination, 79518662e0SNeilBrown * so we need two separate attempts. 80518662e0SNeilBrown */ 81518662e0SNeilBrown p = mempool_alloc(nfs_commit_mempool, GFP_NOWAIT); 82518662e0SNeilBrown if (!p) 83518662e0SNeilBrown p = kmem_cache_alloc(nfs_cdata_cachep, GFP_NOIO | 84518662e0SNeilBrown __GFP_NOWARN | __GFP_NORETRY); 85518662e0SNeilBrown if (!p) 86518662e0SNeilBrown return NULL; 87518662e0SNeilBrown } 88518662e0SNeilBrown 891da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 901da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 911da177e4SLinus Torvalds return p; 921da177e4SLinus Torvalds } 93e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 941da177e4SLinus Torvalds 950b7c0153SFred Isaman void nfs_commit_free(struct nfs_commit_data *p) 961da177e4SLinus Torvalds { 971da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 981da177e4SLinus Torvalds } 99e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 1001da177e4SLinus Torvalds 1011e7f3a48SWeston Andros Adamson static struct nfs_pgio_header *nfs_writehdr_alloc(void) 1023feb2d49STrond Myklebust { 1031e7f3a48SWeston Andros Adamson struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); 1043feb2d49STrond Myklebust 105fbe77c30SBenjamin Coddington if (p) { 1063feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 107fbe77c30SBenjamin Coddington p->rw_mode = FMODE_WRITE; 108fbe77c30SBenjamin Coddington } 1093feb2d49STrond Myklebust return p; 1103feb2d49STrond Myklebust } 1113feb2d49STrond Myklebust 1121e7f3a48SWeston Andros Adamson static void nfs_writehdr_free(struct nfs_pgio_header *hdr) 1136c75dc0dSFred Isaman { 1141e7f3a48SWeston Andros Adamson mempool_free(hdr, nfs_wdata_mempool); 1153feb2d49STrond Myklebust } 1161da177e4SLinus Torvalds 117919e3bd9STrond Myklebust static struct nfs_io_completion *nfs_io_completion_alloc(gfp_t gfp_flags) 118919e3bd9STrond Myklebust { 119919e3bd9STrond Myklebust return kmalloc(sizeof(struct nfs_io_completion), gfp_flags); 120919e3bd9STrond Myklebust } 121919e3bd9STrond Myklebust 122919e3bd9STrond Myklebust static void nfs_io_completion_init(struct nfs_io_completion *ioc, 123919e3bd9STrond Myklebust void (*complete)(void *), void *data) 124919e3bd9STrond Myklebust { 125919e3bd9STrond Myklebust ioc->complete = complete; 126919e3bd9STrond Myklebust ioc->data = data; 127919e3bd9STrond Myklebust kref_init(&ioc->refcount); 128919e3bd9STrond Myklebust } 129919e3bd9STrond Myklebust 130919e3bd9STrond Myklebust static void nfs_io_completion_release(struct kref *kref) 131919e3bd9STrond Myklebust { 132919e3bd9STrond Myklebust struct nfs_io_completion *ioc = container_of(kref, 133919e3bd9STrond Myklebust struct nfs_io_completion, refcount); 134919e3bd9STrond Myklebust ioc->complete(ioc->data); 135919e3bd9STrond Myklebust kfree(ioc); 136919e3bd9STrond Myklebust } 137919e3bd9STrond Myklebust 138919e3bd9STrond Myklebust static void nfs_io_completion_get(struct nfs_io_completion *ioc) 139919e3bd9STrond Myklebust { 140919e3bd9STrond Myklebust if (ioc != NULL) 141919e3bd9STrond Myklebust kref_get(&ioc->refcount); 142919e3bd9STrond Myklebust } 143919e3bd9STrond Myklebust 144919e3bd9STrond Myklebust static void nfs_io_completion_put(struct nfs_io_completion *ioc) 145919e3bd9STrond Myklebust { 146919e3bd9STrond Myklebust if (ioc != NULL) 147919e3bd9STrond Myklebust kref_put(&ioc->refcount, nfs_io_completion_release); 148919e3bd9STrond Myklebust } 149919e3bd9STrond Myklebust 1507b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) 1517b159fc1STrond Myklebust { 1527b159fc1STrond Myklebust ctx->error = error; 1537b159fc1STrond Myklebust smp_wmb(); 1547b159fc1STrond Myklebust set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); 1557b159fc1STrond Myklebust } 1567b159fc1STrond Myklebust 157bd37d6fcSTrond Myklebust static struct nfs_page * 158bd37d6fcSTrond Myklebust nfs_page_private_request(struct page *page) 159bd37d6fcSTrond Myklebust { 160bd37d6fcSTrond Myklebust if (!PagePrivate(page)) 161bd37d6fcSTrond Myklebust return NULL; 162bd37d6fcSTrond Myklebust return (struct nfs_page *)page_private(page); 163bd37d6fcSTrond Myklebust } 164bd37d6fcSTrond Myklebust 16584d3a9a9SWeston Andros Adamson /* 16684d3a9a9SWeston Andros Adamson * nfs_page_find_head_request_locked - find head request associated with @page 16784d3a9a9SWeston Andros Adamson * 16884d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 16984d3a9a9SWeston Andros Adamson * 17084d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 17184d3a9a9SWeston Andros Adamson */ 17229418aa4SMel Gorman static struct nfs_page * 173b30d2f04STrond Myklebust nfs_page_find_private_request(struct page *page) 174277459d2STrond Myklebust { 175b30d2f04STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 176bd37d6fcSTrond Myklebust struct nfs_page *req; 177277459d2STrond Myklebust 178b30d2f04STrond Myklebust if (!PagePrivate(page)) 179b30d2f04STrond Myklebust return NULL; 180b30d2f04STrond Myklebust spin_lock(&inode->i_lock); 181bd37d6fcSTrond Myklebust req = nfs_page_private_request(page); 18284d3a9a9SWeston Andros Adamson if (req) { 18384d3a9a9SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 18429418aa4SMel Gorman kref_get(&req->wb_kref); 18584d3a9a9SWeston Andros Adamson } 186b30d2f04STrond Myklebust spin_unlock(&inode->i_lock); 187b30d2f04STrond Myklebust return req; 188b30d2f04STrond Myklebust } 18929418aa4SMel Gorman 190b30d2f04STrond Myklebust static struct nfs_page * 191b30d2f04STrond Myklebust nfs_page_find_swap_request(struct page *page) 192b30d2f04STrond Myklebust { 193b30d2f04STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 194b30d2f04STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 195b30d2f04STrond Myklebust struct nfs_page *req = NULL; 196b30d2f04STrond Myklebust if (!PageSwapCache(page)) 197b30d2f04STrond Myklebust return NULL; 198e824f99aSTrond Myklebust mutex_lock(&nfsi->commit_mutex); 199b30d2f04STrond Myklebust if (PageSwapCache(page)) { 200b30d2f04STrond Myklebust req = nfs_page_search_commits_for_head_request_locked(nfsi, 201b30d2f04STrond Myklebust page); 202b30d2f04STrond Myklebust if (req) { 203b30d2f04STrond Myklebust WARN_ON_ONCE(req->wb_head != req); 204b30d2f04STrond Myklebust kref_get(&req->wb_kref); 205b30d2f04STrond Myklebust } 206b30d2f04STrond Myklebust } 207e824f99aSTrond Myklebust mutex_unlock(&nfsi->commit_mutex); 208277459d2STrond Myklebust return req; 209277459d2STrond Myklebust } 210277459d2STrond Myklebust 21184d3a9a9SWeston Andros Adamson /* 21284d3a9a9SWeston Andros Adamson * nfs_page_find_head_request - find head request associated with @page 21384d3a9a9SWeston Andros Adamson * 21484d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 21584d3a9a9SWeston Andros Adamson */ 21684d3a9a9SWeston Andros Adamson static struct nfs_page *nfs_page_find_head_request(struct page *page) 217277459d2STrond Myklebust { 218b30d2f04STrond Myklebust struct nfs_page *req; 219277459d2STrond Myklebust 220b30d2f04STrond Myklebust req = nfs_page_find_private_request(page); 221b30d2f04STrond Myklebust if (!req) 222b30d2f04STrond Myklebust req = nfs_page_find_swap_request(page); 223277459d2STrond Myklebust return req; 224277459d2STrond Myklebust } 225277459d2STrond Myklebust 2261da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 2271da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 2281da177e4SLinus Torvalds { 229d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 230a3d01454STrond Myklebust loff_t end, i_size; 231a3d01454STrond Myklebust pgoff_t end_index; 2321da177e4SLinus Torvalds 233a3d01454STrond Myklebust spin_lock(&inode->i_lock); 234a3d01454STrond Myklebust i_size = i_size_read(inode); 23509cbfeafSKirill A. Shutemov end_index = (i_size - 1) >> PAGE_SHIFT; 2368cd79788SHuang Ying if (i_size > 0 && page_index(page) < end_index) 237a3d01454STrond Myklebust goto out; 238d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 2391da177e4SLinus Torvalds if (i_size >= end) 240a3d01454STrond Myklebust goto out; 2411da177e4SLinus Torvalds i_size_write(inode, end); 242a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 243a3d01454STrond Myklebust out: 244a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 2451da177e4SLinus Torvalds } 2461da177e4SLinus Torvalds 247a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 248a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 249a301b777STrond Myklebust { 250d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 251a301b777STrond Myklebust } 252a301b777STrond Myklebust 253d72ddcbaSWeston Andros Adamson /* 254d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 255d72ddcbaSWeston Andros Adamson * @head - head request of page group 256d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 257d72ddcbaSWeston Andros Adamson * 258d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 259d72ddcbaSWeston Andros Adamson * page offset @page_offset. 260d72ddcbaSWeston Andros Adamson * 261d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 262d72ddcbaSWeston Andros Adamson * match is found. 263d72ddcbaSWeston Andros Adamson * 264d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 265d72ddcbaSWeston Andros Adamson */ 266d72ddcbaSWeston Andros Adamson static struct nfs_page * 267d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 268d72ddcbaSWeston Andros Adamson { 269d72ddcbaSWeston Andros Adamson struct nfs_page *req; 270d72ddcbaSWeston Andros Adamson 271d72ddcbaSWeston Andros Adamson req = head; 272d72ddcbaSWeston Andros Adamson do { 273d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 274d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 275d72ddcbaSWeston Andros Adamson return req; 276d72ddcbaSWeston Andros Adamson 277d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 278d72ddcbaSWeston Andros Adamson } while (req != head); 279d72ddcbaSWeston Andros Adamson 280d72ddcbaSWeston Andros Adamson return NULL; 281d72ddcbaSWeston Andros Adamson } 282d72ddcbaSWeston Andros Adamson 283d72ddcbaSWeston Andros Adamson /* 284d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 285d72ddcbaSWeston Andros Adamson * @head - head request of page group 286d72ddcbaSWeston Andros Adamson * 287d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 288d72ddcbaSWeston Andros Adamson * returns false otherwise 289d72ddcbaSWeston Andros Adamson */ 290d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 291d72ddcbaSWeston Andros Adamson { 292d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 293d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 294d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 295d72ddcbaSWeston Andros Adamson 2961344b7eaSTrond Myklebust nfs_page_group_lock(req); 297d72ddcbaSWeston Andros Adamson 2987e8a30f8STrond Myklebust for (;;) { 299d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 3007e8a30f8STrond Myklebust if (!tmp) 3017e8a30f8STrond Myklebust break; 3027e8a30f8STrond Myklebust pos = tmp->wb_pgbase + tmp->wb_bytes; 303d72ddcbaSWeston Andros Adamson } 304d72ddcbaSWeston Andros Adamson 305d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 3067e8a30f8STrond Myklebust return pos >= len; 307d72ddcbaSWeston Andros Adamson } 308d72ddcbaSWeston Andros Adamson 3091da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 3101da177e4SLinus Torvalds * covers the full page. 3111da177e4SLinus Torvalds */ 312d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 3131da177e4SLinus Torvalds { 314d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 3151da177e4SLinus Torvalds return; 316d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 3171da177e4SLinus Torvalds return; 318d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 3191da177e4SLinus Torvalds } 3201da177e4SLinus Torvalds 3211da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 3221da177e4SLinus Torvalds { 323e87b4c7aSNeilBrown int ret = 0; 324cca588d6STrond Myklebust 325e87b4c7aSNeilBrown if (wbc->sync_mode == WB_SYNC_ALL) 326e87b4c7aSNeilBrown ret = FLUSH_COND_STABLE; 327e87b4c7aSNeilBrown return ret; 3281da177e4SLinus Torvalds } 3291da177e4SLinus Torvalds 3301da177e4SLinus Torvalds /* 33189a09141SPeter Zijlstra * NFS congestion control 33289a09141SPeter Zijlstra */ 33389a09141SPeter Zijlstra 33489a09141SPeter Zijlstra int nfs_congestion_kb; 33589a09141SPeter Zijlstra 33689a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 33789a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 33889a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 33989a09141SPeter Zijlstra 340deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 34189a09141SPeter Zijlstra { 3420db10944SJan Kara struct inode *inode = page_file_mapping(page)->host; 3430db10944SJan Kara struct nfs_server *nfss = NFS_SERVER(inode); 3445a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 3455a6d41b3STrond Myklebust 346deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 34789a09141SPeter Zijlstra 348277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 3490db10944SJan Kara NFS_CONGESTION_ON_THRESH) 3500db10944SJan Kara set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 35189a09141SPeter Zijlstra } 35289a09141SPeter Zijlstra 35320633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 35489a09141SPeter Zijlstra { 35520633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 35689a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 35731a01f09STrond Myklebust bool is_done; 35889a09141SPeter Zijlstra 35931a01f09STrond Myklebust is_done = nfs_page_group_sync_on_bit(req, PG_WB_END); 36031a01f09STrond Myklebust nfs_unlock_request(req); 36131a01f09STrond Myklebust if (!is_done) 36220633f04SWeston Andros Adamson return; 36320633f04SWeston Andros Adamson 36420633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 365c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 3660db10944SJan Kara clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 36789a09141SPeter Zijlstra } 36889a09141SPeter Zijlstra 369d4581383SWeston Andros Adamson /* 370d4581383SWeston Andros Adamson * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req 371d4581383SWeston Andros Adamson * 372d4581383SWeston Andros Adamson * this is a helper function for nfs_lock_and_join_requests 373d4581383SWeston Andros Adamson * 374d4581383SWeston Andros Adamson * @inode - inode associated with request page group, must be holding inode lock 375d4581383SWeston Andros Adamson * @head - head request of page group, must be holding head lock 376d4581383SWeston Andros Adamson * @req - request that couldn't lock and needs to wait on the req bit lock 377d4581383SWeston Andros Adamson * 378b5bab9bfSTrond Myklebust * NOTE: this must be called holding page_group bit lock 379b5bab9bfSTrond Myklebust * which will be released before returning. 380d4581383SWeston Andros Adamson * 381d4581383SWeston Andros Adamson * returns 0 on success, < 0 on error. 382d4581383SWeston Andros Adamson */ 38374a6d4b5STrond Myklebust static void 38474a6d4b5STrond Myklebust nfs_unroll_locks(struct inode *inode, struct nfs_page *head, 3856d17d653STrond Myklebust struct nfs_page *req) 386d4581383SWeston Andros Adamson { 387d4581383SWeston Andros Adamson struct nfs_page *tmp; 388e261f51fSTrond Myklebust 389d4581383SWeston Andros Adamson /* relinquish all the locks successfully grabbed this run */ 3905b2b5187STrond Myklebust for (tmp = head->wb_this_page ; tmp != req; tmp = tmp->wb_this_page) { 3915b2b5187STrond Myklebust if (!kref_read(&tmp->wb_kref)) 3925b2b5187STrond Myklebust continue; 3935b2b5187STrond Myklebust nfs_unlock_and_release_request(tmp); 3945b2b5187STrond Myklebust } 395e261f51fSTrond Myklebust } 396d4581383SWeston Andros Adamson 397d4581383SWeston Andros Adamson /* 398d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 399d4581383SWeston Andros Adamson * 400d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 401d4581383SWeston Andros Adamson * @old_head - the old head of the list 402d4581383SWeston Andros Adamson * 403d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 404d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 405d4581383SWeston Andros Adamson * with a reference held by some other context. 406d4581383SWeston Andros Adamson */ 407d4581383SWeston Andros Adamson static void 408d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 409b66aaa8dSTrond Myklebust struct nfs_page *old_head, 410b66aaa8dSTrond Myklebust struct inode *inode) 411d4581383SWeston Andros Adamson { 412d4581383SWeston Andros Adamson while (destroy_list) { 413d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 414d4581383SWeston Andros Adamson 415d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 416d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 417d4581383SWeston Andros Adamson 418d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 419d4581383SWeston Andros Adamson 420d4581383SWeston Andros Adamson /* make sure old group is not used */ 421d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 422d4581383SWeston Andros Adamson 423902a4c00STrond Myklebust clear_bit(PG_REMOVE, &subreq->wb_flags); 424902a4c00STrond Myklebust 4255b2b5187STrond Myklebust /* Note: races with nfs_page_group_destroy() */ 4265b2b5187STrond Myklebust if (!kref_read(&subreq->wb_kref)) { 4275b2b5187STrond Myklebust /* Check if we raced with nfs_page_group_destroy() */ 428902a4c00STrond Myklebust if (test_and_clear_bit(PG_TEARDOWN, &subreq->wb_flags)) 4295b2b5187STrond Myklebust nfs_free_request(subreq); 4305b2b5187STrond Myklebust continue; 4315b2b5187STrond Myklebust } 432d4581383SWeston Andros Adamson 4335b2b5187STrond Myklebust subreq->wb_head = subreq; 4345b2b5187STrond Myklebust 435b66aaa8dSTrond Myklebust if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) { 436d4581383SWeston Andros Adamson nfs_release_request(subreq); 437a6b6d5b8STrond Myklebust atomic_long_dec(&NFS_I(inode)->nrequests); 438d4581383SWeston Andros Adamson } 4395b2b5187STrond Myklebust 4405b2b5187STrond Myklebust /* subreq is now totally disconnected from page group or any 4415b2b5187STrond Myklebust * write / commit lists. last chance to wake any waiters */ 4425b2b5187STrond Myklebust nfs_unlock_and_release_request(subreq); 443d4581383SWeston Andros Adamson } 444d4581383SWeston Andros Adamson } 445d4581383SWeston Andros Adamson 446d4581383SWeston Andros Adamson /* 447d4581383SWeston Andros Adamson * nfs_lock_and_join_requests - join all subreqs to the head req and return 448d4581383SWeston Andros Adamson * a locked reference, cancelling any pending 449d4581383SWeston Andros Adamson * operations for this page. 450d4581383SWeston Andros Adamson * 451d4581383SWeston Andros Adamson * @page - the page used to lookup the "page group" of nfs_page structures 452d4581383SWeston Andros Adamson * 453d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 454d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 455d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 456d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 457d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 458d4581383SWeston Andros Adamson * 459d4581383SWeston Andros Adamson * Returns a locked, referenced pointer to the head request - which after 460d4581383SWeston Andros Adamson * this call is guaranteed to be the only request associated with the page. 461d4581383SWeston Andros Adamson * Returns NULL if no requests are found for @page, or a ERR_PTR if an 462d4581383SWeston Andros Adamson * error was encountered. 463d4581383SWeston Andros Adamson */ 464d4581383SWeston Andros Adamson static struct nfs_page * 4656d17d653STrond Myklebust nfs_lock_and_join_requests(struct page *page) 466d4581383SWeston Andros Adamson { 467d4581383SWeston Andros Adamson struct inode *inode = page_file_mapping(page)->host; 468d4581383SWeston Andros Adamson struct nfs_page *head, *subreq; 469d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 470d4581383SWeston Andros Adamson unsigned int total_bytes; 471d4581383SWeston Andros Adamson int ret; 472d4581383SWeston Andros Adamson 473d4581383SWeston Andros Adamson try_again: 474d4581383SWeston Andros Adamson /* 475d4581383SWeston Andros Adamson * A reference is taken only on the head request which acts as a 476d4581383SWeston Andros Adamson * reference to the whole page group - the group will not be destroyed 477d4581383SWeston Andros Adamson * until the head reference is released. 478d4581383SWeston Andros Adamson */ 479bd37d6fcSTrond Myklebust head = nfs_page_find_head_request(page); 480bd37d6fcSTrond Myklebust if (!head) 481d4581383SWeston Andros Adamson return NULL; 482d4581383SWeston Andros Adamson 483a0e265bcSTrond Myklebust /* lock the page head first in order to avoid an ABBA inefficiency */ 484a0e265bcSTrond Myklebust if (!nfs_lock_request(head)) { 485a0e265bcSTrond Myklebust ret = nfs_wait_on_request(head); 486a0e265bcSTrond Myklebust nfs_release_request(head); 487a0e265bcSTrond Myklebust if (ret < 0) 488a0e265bcSTrond Myklebust return ERR_PTR(ret); 489a0e265bcSTrond Myklebust goto try_again; 490a0e265bcSTrond Myklebust } 491bd37d6fcSTrond Myklebust 492bd37d6fcSTrond Myklebust /* Ensure that nobody removed the request before we locked it */ 493bd37d6fcSTrond Myklebust if (head != nfs_page_private_request(page) && !PageSwapCache(page)) { 494bd37d6fcSTrond Myklebust nfs_unlock_and_release_request(head); 495bd37d6fcSTrond Myklebust goto try_again; 496bd37d6fcSTrond Myklebust } 4977c3af975SWeston Andros Adamson 4981344b7eaSTrond Myklebust ret = nfs_page_group_lock(head); 499b5bab9bfSTrond Myklebust if (ret < 0) { 500a0e265bcSTrond Myklebust nfs_unlock_and_release_request(head); 501b5bab9bfSTrond Myklebust return ERR_PTR(ret); 5027c3af975SWeston Andros Adamson } 5037c3af975SWeston Andros Adamson 5047c3af975SWeston Andros Adamson /* lock each request in the page group */ 505a0e265bcSTrond Myklebust total_bytes = head->wb_bytes; 506a0e265bcSTrond Myklebust for (subreq = head->wb_this_page; subreq != head; 507a0e265bcSTrond Myklebust subreq = subreq->wb_this_page) { 50874a6d4b5STrond Myklebust 5095b2b5187STrond Myklebust if (!kref_get_unless_zero(&subreq->wb_kref)) 5105b2b5187STrond Myklebust continue; 51174a6d4b5STrond Myklebust while (!nfs_lock_request(subreq)) { 512b5bab9bfSTrond Myklebust /* 51374a6d4b5STrond Myklebust * Unlock page to allow nfs_page_group_sync_on_bit() 51474a6d4b5STrond Myklebust * to succeed 515b5bab9bfSTrond Myklebust */ 51674a6d4b5STrond Myklebust nfs_page_group_unlock(head); 51774a6d4b5STrond Myklebust ret = nfs_wait_on_request(subreq); 51874a6d4b5STrond Myklebust if (!ret) 5191344b7eaSTrond Myklebust ret = nfs_page_group_lock(head); 52074a6d4b5STrond Myklebust if (ret < 0) { 52174a6d4b5STrond Myklebust nfs_unroll_locks(inode, head, subreq); 5225b2b5187STrond Myklebust nfs_release_request(subreq); 52374a6d4b5STrond Myklebust nfs_unlock_and_release_request(head); 524d4581383SWeston Andros Adamson return ERR_PTR(ret); 525d4581383SWeston Andros Adamson } 52674a6d4b5STrond Myklebust } 527e14bebf6STrond Myklebust /* 528e14bebf6STrond Myklebust * Subrequests are always contiguous, non overlapping 529e14bebf6STrond Myklebust * and in order - but may be repeated (mirrored writes). 530e14bebf6STrond Myklebust */ 531e14bebf6STrond Myklebust if (subreq->wb_offset == (head->wb_offset + total_bytes)) { 532e14bebf6STrond Myklebust /* keep track of how many bytes this group covers */ 533e14bebf6STrond Myklebust total_bytes += subreq->wb_bytes; 534e14bebf6STrond Myklebust } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || 535e14bebf6STrond Myklebust ((subreq->wb_offset + subreq->wb_bytes) > 536e14bebf6STrond Myklebust (head->wb_offset + total_bytes)))) { 53774a6d4b5STrond Myklebust nfs_unroll_locks(inode, head, subreq); 5385b2b5187STrond Myklebust nfs_unlock_and_release_request(subreq); 53974a6d4b5STrond Myklebust nfs_page_group_unlock(head); 54074a6d4b5STrond Myklebust nfs_unlock_and_release_request(head); 541e14bebf6STrond Myklebust return ERR_PTR(-EIO); 542e14bebf6STrond Myklebust } 543a0e265bcSTrond Myklebust } 544d4581383SWeston Andros Adamson 545d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 546d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 547d4581383SWeston Andros Adamson subreq = head; 548d4581383SWeston Andros Adamson do { 549411a99adSWeston Andros Adamson nfs_clear_request_commit(subreq); 550d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 551d4581383SWeston Andros Adamson } while (subreq != head); 552d4581383SWeston Andros Adamson 553d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 554d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 555d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 556d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 557d4581383SWeston Andros Adamson head->wb_this_page = head; 558d4581383SWeston Andros Adamson 559d4581383SWeston Andros Adamson /* change head request to cover whole range that 560d4581383SWeston Andros Adamson * the former page group covered */ 561d4581383SWeston Andros Adamson head->wb_bytes = total_bytes; 562d4581383SWeston Andros Adamson } 563d4581383SWeston Andros Adamson 564b66aaa8dSTrond Myklebust /* Postpone destruction of this request */ 565b66aaa8dSTrond Myklebust if (test_and_clear_bit(PG_REMOVE, &head->wb_flags)) { 566b66aaa8dSTrond Myklebust set_bit(PG_INODE_REF, &head->wb_flags); 567b66aaa8dSTrond Myklebust kref_get(&head->wb_kref); 568a6b6d5b8STrond Myklebust atomic_long_inc(&NFS_I(inode)->nrequests); 569b66aaa8dSTrond Myklebust } 570b66aaa8dSTrond Myklebust 571d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 572d4581383SWeston Andros Adamson 573b66aaa8dSTrond Myklebust nfs_destroy_unlinked_subrequests(destroy_list, head, inode); 574d4581383SWeston Andros Adamson 575b5bab9bfSTrond Myklebust /* Did we lose a race with nfs_inode_remove_request()? */ 576b5bab9bfSTrond Myklebust if (!(PagePrivate(page) || PageSwapCache(page))) { 577b5bab9bfSTrond Myklebust nfs_unlock_and_release_request(head); 578b5bab9bfSTrond Myklebust return NULL; 579b5bab9bfSTrond Myklebust } 580b5bab9bfSTrond Myklebust 581bd37d6fcSTrond Myklebust /* still holds ref on head from nfs_page_find_head_request 582d4581383SWeston Andros Adamson * and still has lock on head from lock loop */ 583d4581383SWeston Andros Adamson return head; 584612c9384STrond Myklebust } 585074cc1deSTrond Myklebust 5860bcbf039SPeng Tao static void nfs_write_error_remove_page(struct nfs_page *req) 5870bcbf039SPeng Tao { 5880bcbf039SPeng Tao nfs_end_page_writeback(req); 5890bcbf039SPeng Tao generic_error_remove_page(page_file_mapping(req->wb_page), 5900bcbf039SPeng Tao req->wb_page); 5911f84ccdfSFred Isaman nfs_release_request(req); 5920bcbf039SPeng Tao } 5930bcbf039SPeng Tao 594a6598813STrond Myklebust static bool 595a6598813STrond Myklebust nfs_error_is_fatal_on_server(int err) 596a6598813STrond Myklebust { 597a6598813STrond Myklebust switch (err) { 598a6598813STrond Myklebust case 0: 599a6598813STrond Myklebust case -ERESTARTSYS: 600a6598813STrond Myklebust case -EINTR: 601a6598813STrond Myklebust return false; 602a6598813STrond Myklebust } 603a6598813STrond Myklebust return nfs_error_is_fatal(err); 604e261f51fSTrond Myklebust } 605074cc1deSTrond Myklebust 606074cc1deSTrond Myklebust /* 607074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 608074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 609074cc1deSTrond Myklebust */ 610074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 6116d17d653STrond Myklebust struct page *page) 612074cc1deSTrond Myklebust { 613074cc1deSTrond Myklebust struct nfs_page *req; 614074cc1deSTrond Myklebust int ret = 0; 615074cc1deSTrond Myklebust 6166d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 617074cc1deSTrond Myklebust if (!req) 618074cc1deSTrond Myklebust goto out; 619074cc1deSTrond Myklebust ret = PTR_ERR(req); 620074cc1deSTrond Myklebust if (IS_ERR(req)) 621074cc1deSTrond Myklebust goto out; 622074cc1deSTrond Myklebust 623deed85e7STrond Myklebust nfs_set_page_writeback(page); 624deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 625074cc1deSTrond Myklebust 626deed85e7STrond Myklebust ret = 0; 627a6598813STrond Myklebust /* If there is a fatal error that covers this write, just exit */ 628a6598813STrond Myklebust if (nfs_error_is_fatal_on_server(req->wb_context->error)) 629a6598813STrond Myklebust goto out_launder; 630a6598813STrond Myklebust 631f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 632074cc1deSTrond Myklebust ret = pgio->pg_error; 6330bcbf039SPeng Tao /* 634c373fff7STrond Myklebust * Remove the problematic req upon fatal errors on the server 6350bcbf039SPeng Tao */ 6360bcbf039SPeng Tao if (nfs_error_is_fatal(ret)) { 6370bcbf039SPeng Tao nfs_context_set_write_error(req->wb_context, ret); 638c373fff7STrond Myklebust if (nfs_error_is_fatal_on_server(ret)) 639a6598813STrond Myklebust goto out_launder; 640d6c843b9SPeng Tao } 6410bcbf039SPeng Tao nfs_redirty_request(req); 6420bcbf039SPeng Tao ret = -EAGAIN; 64340f90271STrond Myklebust } else 64440f90271STrond Myklebust nfs_add_stats(page_file_mapping(page)->host, 64540f90271STrond Myklebust NFSIOS_WRITEPAGES, 1); 646074cc1deSTrond Myklebust out: 647074cc1deSTrond Myklebust return ret; 648a6598813STrond Myklebust out_launder: 649a6598813STrond Myklebust nfs_write_error_remove_page(req); 650a6598813STrond Myklebust return ret; 651e261f51fSTrond Myklebust } 652e261f51fSTrond Myklebust 653d6c843b9SPeng Tao static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, 654c373fff7STrond Myklebust struct nfs_pageio_descriptor *pgio) 655f758c885STrond Myklebust { 656cfb506e1STrond Myklebust int ret; 657f758c885STrond Myklebust 6588cd79788SHuang Ying nfs_pageio_cond_complete(pgio, page_index(page)); 6596d17d653STrond Myklebust ret = nfs_page_async_flush(pgio, page); 660cfb506e1STrond Myklebust if (ret == -EAGAIN) { 661cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 662cfb506e1STrond Myklebust ret = 0; 663cfb506e1STrond Myklebust } 664cfb506e1STrond Myklebust return ret; 665f758c885STrond Myklebust } 666f758c885STrond Myklebust 667e261f51fSTrond Myklebust /* 6681da177e4SLinus Torvalds * Write an mmapped page to the server. 6691da177e4SLinus Torvalds */ 670d6c843b9SPeng Tao static int nfs_writepage_locked(struct page *page, 671c373fff7STrond Myklebust struct writeback_control *wbc) 6721da177e4SLinus Torvalds { 673f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 67440f90271STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 675e261f51fSTrond Myklebust int err; 6761da177e4SLinus Torvalds 67740f90271STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 678811ed92eSTrond Myklebust nfs_pageio_init_write(&pgio, inode, 0, 679a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 680c373fff7STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 681f758c885STrond Myklebust nfs_pageio_complete(&pgio); 682f758c885STrond Myklebust if (err < 0) 6834d770ccfSTrond Myklebust return err; 684f758c885STrond Myklebust if (pgio.pg_error < 0) 685f758c885STrond Myklebust return pgio.pg_error; 686f758c885STrond Myklebust return 0; 6874d770ccfSTrond Myklebust } 6884d770ccfSTrond Myklebust 6894d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 6904d770ccfSTrond Myklebust { 691f758c885STrond Myklebust int ret; 6924d770ccfSTrond Myklebust 693c373fff7STrond Myklebust ret = nfs_writepage_locked(page, wbc); 6941da177e4SLinus Torvalds unlock_page(page); 695f758c885STrond Myklebust return ret; 696f758c885STrond Myklebust } 697f758c885STrond Myklebust 698f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 699f758c885STrond Myklebust { 700f758c885STrond Myklebust int ret; 701f758c885STrond Myklebust 702c373fff7STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 703f758c885STrond Myklebust unlock_page(page); 704f758c885STrond Myklebust return ret; 7051da177e4SLinus Torvalds } 7061da177e4SLinus Torvalds 707919e3bd9STrond Myklebust static void nfs_io_completion_commit(void *inode) 708919e3bd9STrond Myklebust { 709919e3bd9STrond Myklebust nfs_commit_inode(inode, 0); 710919e3bd9STrond Myklebust } 711919e3bd9STrond Myklebust 7121da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 7131da177e4SLinus Torvalds { 7141da177e4SLinus Torvalds struct inode *inode = mapping->host; 715c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 716919e3bd9STrond Myklebust struct nfs_io_completion *ioc = nfs_io_completion_alloc(GFP_NOFS); 7171da177e4SLinus Torvalds int err; 7181da177e4SLinus Torvalds 71991d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 72091d5b470SChuck Lever 721919e3bd9STrond Myklebust if (ioc) 722919e3bd9STrond Myklebust nfs_io_completion_init(ioc, nfs_io_completion_commit, inode); 723919e3bd9STrond Myklebust 724a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 725a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 726919e3bd9STrond Myklebust pgio.pg_io_completion = ioc; 727f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 728c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 729919e3bd9STrond Myklebust nfs_io_completion_put(ioc); 73072cb77f4STrond Myklebust 731f758c885STrond Myklebust if (err < 0) 73272cb77f4STrond Myklebust goto out_err; 73372cb77f4STrond Myklebust err = pgio.pg_error; 73472cb77f4STrond Myklebust if (err < 0) 73572cb77f4STrond Myklebust goto out_err; 736c63c7b05STrond Myklebust return 0; 73772cb77f4STrond Myklebust out_err: 73872cb77f4STrond Myklebust return err; 7391da177e4SLinus Torvalds } 7401da177e4SLinus Torvalds 7411da177e4SLinus Torvalds /* 7421da177e4SLinus Torvalds * Insert a write request into an inode 7431da177e4SLinus Torvalds */ 744d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 7451da177e4SLinus Torvalds { 7461da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 747e7d39069STrond Myklebust 7482bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 7492bfc6e56SWeston Andros Adamson 750e7d39069STrond Myklebust /* Lock the request! */ 7517ad84aa9STrond Myklebust nfs_lock_request(req); 752e7d39069STrond Myklebust 753e7d39069STrond Myklebust spin_lock(&inode->i_lock); 754a6b6d5b8STrond Myklebust if (!nfs_have_writebacks(inode) && 755cb1410c7SWeston Andros Adamson NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 756a9a4a87aSTrond Myklebust inode->i_version++; 75729418aa4SMel Gorman /* 75829418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 75929418aa4SMel Gorman * with invalidate/truncate. 76029418aa4SMel Gorman */ 76129418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 7622df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 763deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 764277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 76529418aa4SMel Gorman } 766a6b6d5b8STrond Myklebust atomic_long_inc(&nfsi->nrequests); 76717089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 768cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 769cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 770cb1410c7SWeston Andros Adamson * adding subrequests. */ 77117089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 772c03b4024STrond Myklebust kref_get(&req->wb_kref); 773e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 7741da177e4SLinus Torvalds } 7751da177e4SLinus Torvalds 7761da177e4SLinus Torvalds /* 77789a09141SPeter Zijlstra * Remove a write request from an inode 7781da177e4SLinus Torvalds */ 7791da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7801da177e4SLinus Torvalds { 7812b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 7821da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 78320633f04SWeston Andros Adamson struct nfs_page *head; 78420633f04SWeston Andros Adamson 785a6b6d5b8STrond Myklebust atomic_long_dec(&nfsi->nrequests); 78620633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 78720633f04SWeston Andros Adamson head = req->wb_head; 7881da177e4SLinus Torvalds 789587142f8STrond Myklebust spin_lock(&inode->i_lock); 79067911c8fSAnna Schumaker if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { 79120633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 79220633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 79320633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 79429418aa4SMel Gorman } 795587142f8STrond Myklebust spin_unlock(&inode->i_lock); 79620633f04SWeston Andros Adamson } 79717089a29SWeston Andros Adamson 79817089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 7991da177e4SLinus Torvalds nfs_release_request(req); 8001da177e4SLinus Torvalds } 8011da177e4SLinus Torvalds 80261822ab5STrond Myklebust static void 8036d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 80461822ab5STrond Myklebust { 80567911c8fSAnna Schumaker if (req->wb_page) 80661822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 80761822ab5STrond Myklebust } 80861822ab5STrond Myklebust 8093a3908c8STrond Myklebust /* 8103a3908c8STrond Myklebust * nfs_page_search_commits_for_head_request_locked 8113a3908c8STrond Myklebust * 8123a3908c8STrond Myklebust * Search through commit lists on @inode for the head request for @page. 8133a3908c8STrond Myklebust * Must be called while holding the inode (which is cinfo) lock. 8143a3908c8STrond Myklebust * 8153a3908c8STrond Myklebust * Returns the head request if found, or NULL if not found. 8163a3908c8STrond Myklebust */ 8173a3908c8STrond Myklebust static struct nfs_page * 8183a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 8193a3908c8STrond Myklebust struct page *page) 8203a3908c8STrond Myklebust { 8213a3908c8STrond Myklebust struct nfs_page *freq, *t; 8223a3908c8STrond Myklebust struct nfs_commit_info cinfo; 8233a3908c8STrond Myklebust struct inode *inode = &nfsi->vfs_inode; 8243a3908c8STrond Myklebust 8253a3908c8STrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 8263a3908c8STrond Myklebust 8273a3908c8STrond Myklebust /* search through pnfs commit lists */ 8283a3908c8STrond Myklebust freq = pnfs_search_commit_reqs(inode, &cinfo, page); 8293a3908c8STrond Myklebust if (freq) 8303a3908c8STrond Myklebust return freq->wb_head; 8313a3908c8STrond Myklebust 8323a3908c8STrond Myklebust /* Linearly search the commit list for the correct request */ 8333a3908c8STrond Myklebust list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 8343a3908c8STrond Myklebust if (freq->wb_page == page) 8353a3908c8STrond Myklebust return freq->wb_head; 8363a3908c8STrond Myklebust } 8373a3908c8STrond Myklebust 8383a3908c8STrond Myklebust return NULL; 8393a3908c8STrond Myklebust } 8403a3908c8STrond Myklebust 8418dd37758STrond Myklebust /** 84286d80f97STrond Myklebust * nfs_request_add_commit_list_locked - add request to a commit list 84386d80f97STrond Myklebust * @req: pointer to a struct nfs_page 84486d80f97STrond Myklebust * @dst: commit list head 84586d80f97STrond Myklebust * @cinfo: holds list lock and accounting info 84686d80f97STrond Myklebust * 84786d80f97STrond Myklebust * This sets the PG_CLEAN bit, updates the cinfo count of 84886d80f97STrond Myklebust * number of outstanding requests requiring a commit as well as 84986d80f97STrond Myklebust * the MM page stats. 85086d80f97STrond Myklebust * 851e824f99aSTrond Myklebust * The caller must hold NFS_I(cinfo->inode)->commit_mutex, and the 852e824f99aSTrond Myklebust * nfs_page lock. 85386d80f97STrond Myklebust */ 85486d80f97STrond Myklebust void 85586d80f97STrond Myklebust nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, 85686d80f97STrond Myklebust struct nfs_commit_info *cinfo) 85786d80f97STrond Myklebust { 85886d80f97STrond Myklebust set_bit(PG_CLEAN, &req->wb_flags); 85986d80f97STrond Myklebust nfs_list_add_request(req, dst); 86086d80f97STrond Myklebust cinfo->mds->ncommit++; 86186d80f97STrond Myklebust } 86286d80f97STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); 86386d80f97STrond Myklebust 86486d80f97STrond Myklebust /** 8658dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 8668dd37758STrond Myklebust * @req: pointer to a struct nfs_page 867ea2cf228SFred Isaman * @dst: commit list head 868ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8698dd37758STrond Myklebust * 870ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 8718dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8728dd37758STrond Myklebust * the MM page stats. 8738dd37758STrond Myklebust * 874ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8758dd37758STrond Myklebust * holding the nfs_page lock. 8768dd37758STrond Myklebust */ 8778dd37758STrond Myklebust void 8786272dcc6SAnna Schumaker nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) 8798dd37758STrond Myklebust { 880e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 8816272dcc6SAnna Schumaker nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); 882e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 88367911c8fSAnna Schumaker if (req->wb_page) 88486d80f97STrond Myklebust nfs_mark_page_unstable(req->wb_page, cinfo); 8858dd37758STrond Myklebust } 8868dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 8878dd37758STrond Myklebust 8888dd37758STrond Myklebust /** 8898dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 8908dd37758STrond Myklebust * @req: pointer to a nfs_page 891ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8928dd37758STrond Myklebust * 893ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 8948dd37758STrond Myklebust * number of outstanding requests requiring a commit 8958dd37758STrond Myklebust * It does not update the MM page stats. 8968dd37758STrond Myklebust * 897ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 8988dd37758STrond Myklebust */ 8998dd37758STrond Myklebust void 900ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 901ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9028dd37758STrond Myklebust { 9038dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 9048dd37758STrond Myklebust return; 9058dd37758STrond Myklebust nfs_list_remove_request(req); 906ea2cf228SFred Isaman cinfo->mds->ncommit--; 9078dd37758STrond Myklebust } 9088dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 9098dd37758STrond Myklebust 910ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 911ea2cf228SFred Isaman struct inode *inode) 912ea2cf228SFred Isaman { 913fe238e60SDave Wysochanski cinfo->inode = inode; 914ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 915ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 916b359f9d0SFred Isaman cinfo->dreq = NULL; 917f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 918ea2cf228SFred Isaman } 919ea2cf228SFred Isaman 920ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 921ea2cf228SFred Isaman struct inode *inode, 922ea2cf228SFred Isaman struct nfs_direct_req *dreq) 923ea2cf228SFred Isaman { 9241763da12SFred Isaman if (dreq) 9251763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 9261763da12SFred Isaman else 927ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 928ea2cf228SFred Isaman } 929ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 9308dd37758STrond Myklebust 9311da177e4SLinus Torvalds /* 9321da177e4SLinus Torvalds * Add a request to the inode's commit list. 9331da177e4SLinus Torvalds */ 9341763da12SFred Isaman void 935ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 936b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 9371da177e4SLinus Torvalds { 938b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 9398dd37758STrond Myklebust return; 9406272dcc6SAnna Schumaker nfs_request_add_commit_list(req, cinfo); 9411da177e4SLinus Torvalds } 9428e821cadSTrond Myklebust 943d6d6dc7cSFred Isaman static void 944d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 945e468bae9STrond Myklebust { 94611fb9989SMel Gorman dec_node_page_state(page, NR_UNSTABLE_NFS); 94793f78d88STejun Heo dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, 94893f78d88STejun Heo WB_RECLAIMABLE); 949e468bae9STrond Myklebust } 950d6d6dc7cSFred Isaman 951b5bab9bfSTrond Myklebust /* Called holding the request lock on @req */ 9528dd37758STrond Myklebust static void 953d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 954d6d6dc7cSFred Isaman { 9558dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 9562b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 957ea2cf228SFred Isaman struct nfs_commit_info cinfo; 958d6d6dc7cSFred Isaman 959ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 960e824f99aSTrond Myklebust mutex_lock(&NFS_I(inode)->commit_mutex); 961ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 962ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 963d6d6dc7cSFred Isaman } 964e824f99aSTrond Myklebust mutex_unlock(&NFS_I(inode)->commit_mutex); 9658dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 9668dd37758STrond Myklebust } 967e468bae9STrond Myklebust } 968e468bae9STrond Myklebust 969d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 9708e821cadSTrond Myklebust { 971c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 972d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 973c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 9748e821cadSTrond Myklebust } 9758e821cadSTrond Myklebust 976919e3bd9STrond Myklebust static void nfs_async_write_init(struct nfs_pgio_header *hdr) 977919e3bd9STrond Myklebust { 978919e3bd9STrond Myklebust nfs_io_completion_get(hdr->io_completion); 979919e3bd9STrond Myklebust } 980919e3bd9STrond Myklebust 981061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 9826c75dc0dSFred Isaman { 983ea2cf228SFred Isaman struct nfs_commit_info cinfo; 9846c75dc0dSFred Isaman unsigned long bytes = 0; 9856c75dc0dSFred Isaman 9866c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 9876c75dc0dSFred Isaman goto out; 988ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 9896c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 9906c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 9916c75dc0dSFred Isaman 9926c75dc0dSFred Isaman bytes += req->wb_bytes; 9936c75dc0dSFred Isaman nfs_list_remove_request(req); 9946c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 9956c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 996d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 9976c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 9986c75dc0dSFred Isaman goto remove_req; 9996c75dc0dSFred Isaman } 1000c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 1001f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 1002b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 1003a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 10046c75dc0dSFred Isaman goto next; 10056c75dc0dSFred Isaman } 10066c75dc0dSFred Isaman remove_req: 10076c75dc0dSFred Isaman nfs_inode_remove_request(req); 10086c75dc0dSFred Isaman next: 100920633f04SWeston Andros Adamson nfs_end_page_writeback(req); 10103aff4ebbSTrond Myklebust nfs_release_request(req); 10116c75dc0dSFred Isaman } 10126c75dc0dSFred Isaman out: 1013919e3bd9STrond Myklebust nfs_io_completion_put(hdr->io_completion); 10146c75dc0dSFred Isaman hdr->release(hdr); 10156c75dc0dSFred Isaman } 10166c75dc0dSFred Isaman 1017ce59515cSAnna Schumaker unsigned long 1018ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 1019fb8a1f11STrond Myklebust { 1020ea2cf228SFred Isaman return cinfo->mds->ncommit; 1021fb8a1f11STrond Myklebust } 1022fb8a1f11STrond Myklebust 1023e824f99aSTrond Myklebust /* NFS_I(cinfo->inode)->commit_mutex held by caller */ 10241763da12SFred Isaman int 1025ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 1026ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 1027d6d6dc7cSFred Isaman { 1028d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 1029d6d6dc7cSFred Isaman int ret = 0; 1030d6d6dc7cSFred Isaman 1031d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 10328dd37758STrond Myklebust if (!nfs_lock_request(req)) 10338dd37758STrond Myklebust continue; 10347ad84aa9STrond Myklebust kref_get(&req->wb_kref); 1035ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 10368dd37758STrond Myklebust nfs_list_add_request(req, dst); 1037d6d6dc7cSFred Isaman ret++; 10381763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 1039d6d6dc7cSFred Isaman break; 1040e824f99aSTrond Myklebust cond_resched(); 1041d6d6dc7cSFred Isaman } 1042d6d6dc7cSFred Isaman return ret; 1043d6d6dc7cSFred Isaman } 1044d6d6dc7cSFred Isaman 10451da177e4SLinus Torvalds /* 10461da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 10471da177e4SLinus Torvalds * @inode: NFS inode to scan 1048ea2cf228SFred Isaman * @dst: mds destination list 1049ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 10501da177e4SLinus Torvalds * 10511da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 10521da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 10531da177e4SLinus Torvalds */ 10541763da12SFred Isaman int 1055ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 1056ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 10571da177e4SLinus Torvalds { 1058d6d6dc7cSFred Isaman int ret = 0; 1059fb8a1f11STrond Myklebust 1060e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 1061ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 10628dd37758STrond Myklebust const int max = INT_MAX; 1063d6d6dc7cSFred Isaman 1064ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 1065ea2cf228SFred Isaman cinfo, max); 1066ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 1067d6d6dc7cSFred Isaman } 1068e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 1069ff778d02STrond Myklebust return ret; 10701da177e4SLinus Torvalds } 1071d6d6dc7cSFred Isaman 10721da177e4SLinus Torvalds /* 1073e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1074e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 10751da177e4SLinus Torvalds * 1076e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1077e7d39069STrond Myklebust * to disk. 10781da177e4SLinus Torvalds */ 1079e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 1080e7d39069STrond Myklebust struct page *page, 1081e7d39069STrond Myklebust unsigned int offset, 1082e7d39069STrond Myklebust unsigned int bytes) 10831da177e4SLinus Torvalds { 1084e7d39069STrond Myklebust struct nfs_page *req; 1085e7d39069STrond Myklebust unsigned int rqend; 1086e7d39069STrond Myklebust unsigned int end; 10871da177e4SLinus Torvalds int error; 1088277459d2STrond Myklebust 1089e7d39069STrond Myklebust end = offset + bytes; 1090e7d39069STrond Myklebust 1091f6032f21STrond Myklebust req = nfs_lock_and_join_requests(page); 1092f6032f21STrond Myklebust if (IS_ERR_OR_NULL(req)) 1093f6032f21STrond Myklebust return req; 10942bfc6e56SWeston Andros Adamson 1095e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1096e7d39069STrond Myklebust /* 1097e7d39069STrond Myklebust * Tell the caller to flush out the request if 1098e7d39069STrond Myklebust * the offsets are non-contiguous. 1099e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1100e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1101e7d39069STrond Myklebust */ 1102f6032f21STrond Myklebust if (offset > rqend || end < req->wb_offset) 1103e7d39069STrond Myklebust goto out_flushme; 1104e7d39069STrond Myklebust 11051da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 11061da177e4SLinus Torvalds if (offset < req->wb_offset) { 11071da177e4SLinus Torvalds req->wb_offset = offset; 11081da177e4SLinus Torvalds req->wb_pgbase = offset; 11091da177e4SLinus Torvalds } 11101da177e4SLinus Torvalds if (end > rqend) 11111da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1112e7d39069STrond Myklebust else 1113e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 1114e7d39069STrond Myklebust return req; 1115e7d39069STrond Myklebust out_flushme: 1116f6032f21STrond Myklebust /* 1117f6032f21STrond Myklebust * Note: we mark the request dirty here because 1118f6032f21STrond Myklebust * nfs_lock_and_join_requests() cannot preserve 1119f6032f21STrond Myklebust * commit flags, so we have to replay the write. 1120f6032f21STrond Myklebust */ 1121f6032f21STrond Myklebust nfs_mark_request_dirty(req); 1122f6032f21STrond Myklebust nfs_unlock_and_release_request(req); 1123e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1124f6032f21STrond Myklebust return (error < 0) ? ERR_PTR(error) : NULL; 1125e7d39069STrond Myklebust } 11261da177e4SLinus Torvalds 1127e7d39069STrond Myklebust /* 1128e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1129e7d39069STrond Myklebust * 1130e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1131e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1132e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1133e7d39069STrond Myklebust */ 1134e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1135e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1136e7d39069STrond Myklebust { 1137d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1138e7d39069STrond Myklebust struct nfs_page *req; 1139e7d39069STrond Myklebust 1140e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1141e7d39069STrond Myklebust if (req != NULL) 1142e7d39069STrond Myklebust goto out; 11432bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 1144e7d39069STrond Myklebust if (IS_ERR(req)) 1145e7d39069STrond Myklebust goto out; 1146d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1147efc91ed0STrond Myklebust out: 114861e930a9STrond Myklebust return req; 11491da177e4SLinus Torvalds } 11501da177e4SLinus Torvalds 1151e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1152e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1153e7d39069STrond Myklebust { 1154e7d39069STrond Myklebust struct nfs_page *req; 1155e7d39069STrond Myklebust 1156e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1157e7d39069STrond Myklebust if (IS_ERR(req)) 1158e7d39069STrond Myklebust return PTR_ERR(req); 1159e7d39069STrond Myklebust /* Update file length */ 1160e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1161d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1162a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11631d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1164e7d39069STrond Myklebust return 0; 1165e7d39069STrond Myklebust } 1166e7d39069STrond Myklebust 11671da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 11681da177e4SLinus Torvalds { 1169cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 11702a369153STrond Myklebust struct nfs_lock_context *l_ctx; 1171bd61e0a9SJeff Layton struct file_lock_context *flctx = file_inode(file)->i_flctx; 11721da177e4SLinus Torvalds struct nfs_page *req; 11731a54533eSTrond Myklebust int do_flush, status; 11741da177e4SLinus Torvalds /* 11751da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 11761da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 11771da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 11781da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11791da177e4SLinus Torvalds * Also do the same if we find a request from an existing 11801da177e4SLinus Torvalds * dropped page. 11811da177e4SLinus Torvalds */ 11821a54533eSTrond Myklebust do { 118384d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 11841a54533eSTrond Myklebust if (req == NULL) 11851a54533eSTrond Myklebust return 0; 11862a369153STrond Myklebust l_ctx = req->wb_lock_context; 1187138a2935STrond Myklebust do_flush = req->wb_page != page || 1188138a2935STrond Myklebust !nfs_match_open_context(req->wb_context, ctx); 1189bd61e0a9SJeff Layton if (l_ctx && flctx && 1190bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1191bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 1192d51fdb87SNeilBrown do_flush |= l_ctx->lockowner != current->files; 11935263e31eSJeff Layton } 11941da177e4SLinus Torvalds nfs_release_request(req); 11951a54533eSTrond Myklebust if (!do_flush) 11961a54533eSTrond Myklebust return 0; 1197d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 11981a54533eSTrond Myklebust } while (status == 0); 11991a54533eSTrond Myklebust return status; 12001da177e4SLinus Torvalds } 12011da177e4SLinus Torvalds 12021da177e4SLinus Torvalds /* 1203dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1204dc24826bSAndy Adamson * expire soon. 1205dc24826bSAndy Adamson * 1206dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1207dc24826bSAndy Adamson * 1208dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1209dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1210dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1211dc24826bSAndy Adamson */ 1212dc24826bSAndy Adamson int 1213dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1214dc24826bSAndy Adamson { 1215dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1216dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1217dc24826bSAndy Adamson 1218dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 1219dc24826bSAndy Adamson } 1220dc24826bSAndy Adamson 1221dc24826bSAndy Adamson /* 1222dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1223dc24826bSAndy Adamson */ 1224ce52914eSScott Mayhew bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) 1225dc24826bSAndy Adamson { 1226ce52914eSScott Mayhew struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1227ce52914eSScott Mayhew 1228ce52914eSScott Mayhew return rpcauth_cred_key_to_expire(auth, ctx->cred); 1229dc24826bSAndy Adamson } 1230dc24826bSAndy Adamson 1231dc24826bSAndy Adamson /* 12325d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 12335d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 12345d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 12355d47a356STrond Myklebust */ 12368d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 12375d47a356STrond Myklebust { 1238d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1239d529ef83SJeff Layton 12408d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 12418d197a56STrond Myklebust goto out; 124218dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1243d529ef83SJeff Layton return false; 12444db72b40SJeff Layton smp_rmb(); 1245d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 12468d197a56STrond Myklebust return false; 12478d197a56STrond Myklebust out: 124818dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 124918dd78c4SScott Mayhew return false; 12508d197a56STrond Myklebust return PageUptodate(page) != 0; 12515d47a356STrond Myklebust } 12525d47a356STrond Myklebust 12535263e31eSJeff Layton static bool 12545263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 12555263e31eSJeff Layton { 12565263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 12575263e31eSJeff Layton fl->fl_type == F_WRLCK; 12585263e31eSJeff Layton } 12595263e31eSJeff Layton 1260c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1261c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1262c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1263c7559663SScott Mayhew * inefficiencies. 1264c7559663SScott Mayhew * 1265263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1266263b4509SScott Mayhew * of the checks. 1267c7559663SScott Mayhew */ 1268c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1269c7559663SScott Mayhew { 12705263e31eSJeff Layton int ret; 12715263e31eSJeff Layton struct file_lock_context *flctx = inode->i_flctx; 12725263e31eSJeff Layton struct file_lock *fl; 12735263e31eSJeff Layton 1274c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1275c7559663SScott Mayhew return 0; 1276263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1277263b4509SScott Mayhew return 0; 1278c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1279c7559663SScott Mayhew return 1; 1280bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1281bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 12828fa4592aSTrond Myklebust return 1; 12835263e31eSJeff Layton 12845263e31eSJeff Layton /* Check to see if there are whole file write locks */ 12855263e31eSJeff Layton ret = 0; 12866109c850SJeff Layton spin_lock(&flctx->flc_lock); 1287bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1288bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1289bd61e0a9SJeff Layton fl_list); 1290bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 12915263e31eSJeff Layton ret = 1; 1292bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 12935263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 12945263e31eSJeff Layton fl_list); 12955263e31eSJeff Layton if (fl->fl_type == F_WRLCK) 12965263e31eSJeff Layton ret = 1; 12975263e31eSJeff Layton } 12986109c850SJeff Layton spin_unlock(&flctx->flc_lock); 12995263e31eSJeff Layton return ret; 1300c7559663SScott Mayhew } 1301c7559663SScott Mayhew 13025d47a356STrond Myklebust /* 13031da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 13041da177e4SLinus Torvalds * 13051da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 13061da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 13071da177e4SLinus Torvalds */ 13081da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 13091da177e4SLinus Torvalds unsigned int offset, unsigned int count) 13101da177e4SLinus Torvalds { 1311cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1312d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 13131da177e4SLinus Torvalds int status = 0; 13141da177e4SLinus Torvalds 131591d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 131691d5b470SChuck Lever 13176de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 13186de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 13191da177e4SLinus Torvalds 1320149a4fddSBenjamin Coddington if (!count) 1321149a4fddSBenjamin Coddington goto out; 1322149a4fddSBenjamin Coddington 1323c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 132449a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 13251da177e4SLinus Torvalds offset = 0; 13261da177e4SLinus Torvalds } 13271da177e4SLinus Torvalds 1328e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 132903fa9e84STrond Myklebust if (status < 0) 133003fa9e84STrond Myklebust nfs_set_pageerror(page); 133159b7c05fSTrond Myklebust else 133259b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 1333149a4fddSBenjamin Coddington out: 133448186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 13351da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 13361da177e4SLinus Torvalds return status; 13371da177e4SLinus Torvalds } 13381da177e4SLinus Torvalds 13393ff7576dSTrond Myklebust static int flush_task_priority(int how) 13401da177e4SLinus Torvalds { 13411da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 13421da177e4SLinus Torvalds case FLUSH_HIGHPRI: 13431da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 13441da177e4SLinus Torvalds case FLUSH_LOWPRI: 13451da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 13461da177e4SLinus Torvalds } 13471da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 13481da177e4SLinus Torvalds } 13491da177e4SLinus Torvalds 1350d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1351d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1352abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 13531ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 13541da177e4SLinus Torvalds { 13553ff7576dSTrond Myklebust int priority = flush_task_priority(how); 13561da177e4SLinus Torvalds 13571ed26f33SAnna Schumaker task_setup_data->priority = priority; 1358abde71f4STom Haynes rpc_ops->write_setup(hdr, msg); 1359d138d5d1SAndy Adamson 1360abde71f4STom Haynes nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client, 1361d45f60c6SWeston Andros Adamson &task_setup_data->rpc_client, msg, hdr); 1362275acaafSTrond Myklebust } 1363275acaafSTrond Myklebust 13646d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 13656d884e8fSFred * call this on each, which will prepare them to be retried on next 13666d884e8fSFred * writeback using standard nfs. 13676d884e8fSFred */ 13686d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 13696d884e8fSFred { 13706d884e8fSFred nfs_mark_request_dirty(req); 1371c7070113STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 137220633f04SWeston Andros Adamson nfs_end_page_writeback(req); 13733aff4ebbSTrond Myklebust nfs_release_request(req); 13746d884e8fSFred } 13756d884e8fSFred 1376061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 13776c75dc0dSFred Isaman { 13786c75dc0dSFred Isaman struct nfs_page *req; 13796c75dc0dSFred Isaman 13806c75dc0dSFred Isaman while (!list_empty(head)) { 13816c75dc0dSFred Isaman req = nfs_list_entry(head->next); 13826c75dc0dSFred Isaman nfs_list_remove_request(req); 13836c75dc0dSFred Isaman nfs_redirty_request(req); 13846c75dc0dSFred Isaman } 13856c75dc0dSFred Isaman } 13866c75dc0dSFred Isaman 1387dc602dd7STrond Myklebust static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) 1388dc602dd7STrond Myklebust { 1389dc602dd7STrond Myklebust nfs_async_write_error(&hdr->pages); 1390dc602dd7STrond Myklebust } 1391dc602dd7STrond Myklebust 1392061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1393919e3bd9STrond Myklebust .init_hdr = nfs_async_write_init, 1394061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1395061ae2edSFred Isaman .completion = nfs_write_completion, 1396dc602dd7STrond Myklebust .reschedule_io = nfs_async_write_reschedule_io, 1397061ae2edSFred Isaman }; 1398061ae2edSFred Isaman 139957208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1400a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1401061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 14021751c363STrond Myklebust { 1403a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 140441d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1405a20c93e3SChristoph Hellwig 1406a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1407a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1408a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1409a20c93e3SChristoph Hellwig #endif 14104a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 1411fbe77c30SBenjamin Coddington server->wsize, ioflags, GFP_NOIO); 14121751c363STrond Myklebust } 1413ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 14141751c363STrond Myklebust 1415dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1416dce81290STrond Myklebust { 1417a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1418a7d42ddbSWeston Andros Adamson 14196f29b9bbSKinglong Mee if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) 14206f29b9bbSKinglong Mee pgio->pg_ops->pg_cleanup(pgio); 14216f29b9bbSKinglong Mee 142241d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1423a7d42ddbSWeston Andros Adamson 1424a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1425a7d42ddbSWeston Andros Adamson 1426a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1427a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1428dce81290STrond Myklebust } 14291f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1430dce81290STrond Myklebust 14311da177e4SLinus Torvalds 14320b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 14330b7c0153SFred Isaman { 14340b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 14350b7c0153SFred Isaman 14360b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 14370b7c0153SFred Isaman } 14380b7c0153SFred Isaman 14391f2edbe3STrond Myklebust /* 14401f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 14411f2edbe3STrond Myklebust */ 14421f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 14431f2edbe3STrond Myklebust { 14441f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 14451f2edbe3STrond Myklebust int kill = 0; 1446788e7a89STrond Myklebust 14471f2edbe3STrond Myklebust /* suid always must be killed */ 14481f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 14491f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 14501f2edbe3STrond Myklebust 14511f2edbe3STrond Myklebust /* 14521f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 14531f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 14541f2edbe3STrond Myklebust */ 14551f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 14561f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 14571f2edbe3STrond Myklebust 14581f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 14591f2edbe3STrond Myklebust return kill; 14601f2edbe3STrond Myklebust 14611f2edbe3STrond Myklebust return 0; 14621f2edbe3STrond Myklebust } 1463788e7a89STrond Myklebust 1464a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1465a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1466a08a8cd3STrond Myklebust { 1467a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1468a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 14692b83d3deSTrond Myklebust u64 size = argp->offset + resp->count; 1470a08a8cd3STrond Myklebust 1471a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 14722b83d3deSTrond Myklebust fattr->size = size; 14732b83d3deSTrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { 14742b83d3deSTrond Myklebust fattr->valid &= ~NFS_ATTR_FATTR_SIZE; 1475a08a8cd3STrond Myklebust return; 14762b83d3deSTrond Myklebust } 14772b83d3deSTrond Myklebust if (size != fattr->size) 1478a08a8cd3STrond Myklebust return; 1479a08a8cd3STrond Myklebust /* Set attribute barrier */ 1480a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 14812b83d3deSTrond Myklebust /* ...and update size */ 14822b83d3deSTrond Myklebust fattr->valid |= NFS_ATTR_FATTR_SIZE; 1483a08a8cd3STrond Myklebust } 1484a08a8cd3STrond Myklebust 1485a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1486a08a8cd3STrond Myklebust { 14872b83d3deSTrond Myklebust struct nfs_fattr *fattr = &hdr->fattr; 1488a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1489a08a8cd3STrond Myklebust 1490a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1491a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1492a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1493a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1494a08a8cd3STrond Myklebust } 1495a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1496a08a8cd3STrond Myklebust 14971da177e4SLinus Torvalds /* 14981da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 14991da177e4SLinus Torvalds */ 1500d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1501d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 15020eecb214SAnna Schumaker struct inode *inode) 15031da177e4SLinus Torvalds { 1504788e7a89STrond Myklebust int status; 15051da177e4SLinus Torvalds 1506f551e44fSChuck Lever /* 1507f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1508f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1509f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1510f551e44fSChuck Lever * another writer had changed the file, but some applications 1511f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1512f551e44fSChuck Lever */ 1513d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1514788e7a89STrond Myklebust if (status != 0) 15150eecb214SAnna Schumaker return status; 1516d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 151791d5b470SChuck Lever 1518d45f60c6SWeston Andros Adamson if (hdr->res.verf->committed < hdr->args.stable && 1519d45f60c6SWeston Andros Adamson task->tk_status >= 0) { 15201da177e4SLinus Torvalds /* We tried a write call, but the server did not 15211da177e4SLinus Torvalds * commit data to stable storage even though we 15221da177e4SLinus Torvalds * requested it. 15231da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 15241da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 15251da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 15261da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 15271da177e4SLinus Torvalds */ 15281da177e4SLinus Torvalds static unsigned long complain; 15291da177e4SLinus Torvalds 1530a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 15311da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 15321da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 15331da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1534cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1535d45f60c6SWeston Andros Adamson hdr->res.verf->committed, hdr->args.stable); 15361da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 15371da177e4SLinus Torvalds } 15381da177e4SLinus Torvalds } 15391f2edbe3STrond Myklebust 15401f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 15411f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 15421f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 15430eecb214SAnna Schumaker return 0; 15440eecb214SAnna Schumaker } 15450eecb214SAnna Schumaker 15460eecb214SAnna Schumaker /* 15470eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 15480eecb214SAnna Schumaker */ 1549d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1550d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 15510eecb214SAnna Schumaker { 1552d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1553d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 15541f2edbe3STrond Myklebust 15551f2edbe3STrond Myklebust if (resp->count < argp->count) { 15561da177e4SLinus Torvalds static unsigned long complain; 15571da177e4SLinus Torvalds 15586c75dc0dSFred Isaman /* This a short write! */ 1559d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 156091d5b470SChuck Lever 15611da177e4SLinus Torvalds /* Has the server at least made some progress? */ 15626c75dc0dSFred Isaman if (resp->count == 0) { 15636c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 15646c75dc0dSFred Isaman printk(KERN_WARNING 15656c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 15666c75dc0dSFred Isaman argp->count); 15676c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 15686c75dc0dSFred Isaman } 1569d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 15706c75dc0dSFred Isaman task->tk_status = -EIO; 15716c75dc0dSFred Isaman return; 15726c75dc0dSFred Isaman } 1573f8417b48SKinglong Mee 1574f8417b48SKinglong Mee /* For non rpc-based layout drivers, retry-through-MDS */ 1575f8417b48SKinglong Mee if (!task->tk_ops) { 1576f8417b48SKinglong Mee hdr->pnfs_error = -EAGAIN; 1577f8417b48SKinglong Mee return; 1578f8417b48SKinglong Mee } 1579f8417b48SKinglong Mee 15801da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 15811da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 15821da177e4SLinus Torvalds /* Resend from where the server left off */ 1583d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 15841da177e4SLinus Torvalds argp->offset += resp->count; 15851da177e4SLinus Torvalds argp->pgbase += resp->count; 15861da177e4SLinus Torvalds argp->count -= resp->count; 15871da177e4SLinus Torvalds } else { 15881da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 15891da177e4SLinus Torvalds * headaches in the case of a server crash. 15901da177e4SLinus Torvalds */ 15911da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 15921da177e4SLinus Torvalds } 1593d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 15941da177e4SLinus Torvalds } 15951da177e4SLinus Torvalds } 15961da177e4SLinus Torvalds 1597af7cf057STrond Myklebust static int wait_on_commit(struct nfs_mds_commit_info *cinfo) 159871d0a611STrond Myklebust { 1599af7cf057STrond Myklebust return wait_on_atomic_t(&cinfo->rpcs_out, 1600af7cf057STrond Myklebust nfs_wait_atomic_killable, TASK_KILLABLE); 1601af7cf057STrond Myklebust } 1602af7cf057STrond Myklebust 1603af7cf057STrond Myklebust static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) 1604af7cf057STrond Myklebust { 1605af7cf057STrond Myklebust atomic_inc(&cinfo->rpcs_out); 1606af7cf057STrond Myklebust } 1607af7cf057STrond Myklebust 1608af7cf057STrond Myklebust static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) 1609af7cf057STrond Myklebust { 1610af7cf057STrond Myklebust if (atomic_dec_and_test(&cinfo->rpcs_out)) 1611af7cf057STrond Myklebust wake_up_atomic_t(&cinfo->rpcs_out); 161271d0a611STrond Myklebust } 161371d0a611STrond Myklebust 16140b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 16151da177e4SLinus Torvalds { 16160b7c0153SFred Isaman put_nfs_open_context(data->context); 16170b7c0153SFred Isaman nfs_commit_free(data); 16181da177e4SLinus Torvalds } 1619e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 16201da177e4SLinus Torvalds 16210b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1622c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 16239ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 16249f0ec176SAndy Adamson int how, int flags) 16251da177e4SLinus Torvalds { 162607737691STrond Myklebust struct rpc_task *task; 16279ace33cdSFred Isaman int priority = flush_task_priority(how); 1628bdc7f021STrond Myklebust struct rpc_message msg = { 1629bdc7f021STrond Myklebust .rpc_argp = &data->args, 1630bdc7f021STrond Myklebust .rpc_resp = &data->res, 16319ace33cdSFred Isaman .rpc_cred = data->cred, 1632bdc7f021STrond Myklebust }; 163384115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 163407737691STrond Myklebust .task = &data->task, 16359ace33cdSFred Isaman .rpc_client = clnt, 1636bdc7f021STrond Myklebust .rpc_message = &msg, 16379ace33cdSFred Isaman .callback_ops = call_ops, 163884115e1cSTrond Myklebust .callback_data = data, 1639101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 16409f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 16413ff7576dSTrond Myklebust .priority = priority, 164284115e1cSTrond Myklebust }; 1643788e7a89STrond Myklebust /* Set up the initial task struct. */ 1644c36aae9aSPeng Tao nfs_ops->commit_setup(data, &msg); 16451da177e4SLinus Torvalds 1646b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1647bdc7f021STrond Myklebust 16488c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 16498c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 16508c21c62cSWeston Andros Adamson 165107737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1652dbae4c73STrond Myklebust if (IS_ERR(task)) 1653dbae4c73STrond Myklebust return PTR_ERR(task); 1654d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1655d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 165607737691STrond Myklebust rpc_put_task(task); 1657dbae4c73STrond Myklebust return 0; 16581da177e4SLinus Torvalds } 1659e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 16601da177e4SLinus Torvalds 1661378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1662378520b8SPeng Tao { 1663378520b8SPeng Tao loff_t lwb = 0; 1664378520b8SPeng Tao struct nfs_page *req; 1665378520b8SPeng Tao 1666378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1667378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1668378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1669378520b8SPeng Tao 1670378520b8SPeng Tao return lwb; 1671378520b8SPeng Tao } 1672378520b8SPeng Tao 16731da177e4SLinus Torvalds /* 16749ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 16759ace33cdSFred Isaman */ 16760b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1677988b6dceSFred Isaman struct list_head *head, 1678f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1679f453a54aSFred Isaman struct nfs_commit_info *cinfo) 16809ace33cdSFred Isaman { 16819ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 16822b0143b5SDavid Howells struct inode *inode = d_inode(first->wb_context->dentry); 16839ace33cdSFred Isaman 16849ace33cdSFred Isaman /* Set up the RPC argument and reply structs 16859ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 16869ace33cdSFred Isaman 16879ace33cdSFred Isaman list_splice_init(head, &data->pages); 16889ace33cdSFred Isaman 16899ace33cdSFred Isaman data->inode = inode; 16909ace33cdSFred Isaman data->cred = first->wb_context->cred; 1691988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1692378520b8SPeng Tao /* only set lwb for pnfs commit */ 1693378520b8SPeng Tao if (lseg) 1694378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 16959ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1696f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1697b359f9d0SFred Isaman data->dreq = cinfo->dreq; 16989ace33cdSFred Isaman 16999ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 17009ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 17019ace33cdSFred Isaman data->args.offset = 0; 17029ace33cdSFred Isaman data->args.count = 0; 17030b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 17049ace33cdSFred Isaman data->res.fattr = &data->fattr; 17059ace33cdSFred Isaman data->res.verf = &data->verf; 17069ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 17079ace33cdSFred Isaman } 1708e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 17099ace33cdSFred Isaman 1710e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1711ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1712b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1713b57ff130SWeston Andros Adamson u32 ds_commit_idx) 171464bfeb49SFred Isaman { 171564bfeb49SFred Isaman struct nfs_page *req; 171664bfeb49SFred Isaman 171764bfeb49SFred Isaman while (!list_empty(page_list)) { 171864bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 171964bfeb49SFred Isaman nfs_list_remove_request(req); 1720b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 1721487b9b8aSTom Haynes if (!cinfo->dreq) 1722487b9b8aSTom Haynes nfs_clear_page_commit(req->wb_page); 17231d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 172464bfeb49SFred Isaman } 172564bfeb49SFred Isaman } 1726e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 172764bfeb49SFred Isaman 1728b20135d0STrond Myklebust static void 1729b20135d0STrond Myklebust nfs_commit_resched_write(struct nfs_commit_info *cinfo, 1730b20135d0STrond Myklebust struct nfs_page *req) 1731b20135d0STrond Myklebust { 1732b20135d0STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 1733b20135d0STrond Myklebust } 1734b20135d0STrond Myklebust 17359ace33cdSFred Isaman /* 17361da177e4SLinus Torvalds * Commit dirty pages 17371da177e4SLinus Torvalds */ 17381da177e4SLinus Torvalds static int 1739ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1740ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 17411da177e4SLinus Torvalds { 17420b7c0153SFred Isaman struct nfs_commit_data *data; 17431da177e4SLinus Torvalds 1744ade8febdSWeston Andros Adamson /* another commit raced with us */ 1745ade8febdSWeston Andros Adamson if (list_empty(head)) 1746ade8febdSWeston Andros Adamson return 0; 1747ade8febdSWeston Andros Adamson 1748518662e0SNeilBrown data = nfs_commitdata_alloc(true); 17491da177e4SLinus Torvalds 17501da177e4SLinus Torvalds /* Set up the argument struct */ 1751f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1752f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 1753c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 1754c36aae9aSPeng Tao data->mds_ops, how, 0); 17551da177e4SLinus Torvalds } 17561da177e4SLinus Torvalds 17571da177e4SLinus Torvalds /* 17581da177e4SLinus Torvalds * COMMIT call returned 17591da177e4SLinus Torvalds */ 1760788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 17611da177e4SLinus Torvalds { 17620b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 17631da177e4SLinus Torvalds 1764a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 17651da177e4SLinus Torvalds task->tk_pid, task->tk_status); 17661da177e4SLinus Torvalds 1767788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1768c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1769c9d8f89dSTrond Myklebust } 1770c9d8f89dSTrond Myklebust 1771f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1772c9d8f89dSTrond Myklebust { 1773c9d8f89dSTrond Myklebust struct nfs_page *req; 1774c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1775f453a54aSFred Isaman struct nfs_commit_info cinfo; 1776353db796SNeilBrown struct nfs_server *nfss; 1777788e7a89STrond Myklebust 17781da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 17791da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 17801da177e4SLinus Torvalds nfs_list_remove_request(req); 178167911c8fSAnna Schumaker if (req->wb_page) 1782d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 17831da177e4SLinus Torvalds 17841e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 17853d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 17862b0143b5SDavid Howells (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)), 17871da177e4SLinus Torvalds req->wb_bytes, 17881da177e4SLinus Torvalds (long long)req_offset(req)); 1789c9d8f89dSTrond Myklebust if (status < 0) { 1790c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 179138a33101SKinglong Mee if (req->wb_page) 17921da177e4SLinus Torvalds nfs_inode_remove_request(req); 1793ddeaa637SJoe Perches dprintk_cont(", error = %d\n", status); 17941da177e4SLinus Torvalds goto next; 17951da177e4SLinus Torvalds } 17961da177e4SLinus Torvalds 17971da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 17981da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 17998fc3c386STrond Myklebust if (!nfs_write_verifier_cmp(&req->wb_verf, &data->verf.verifier)) { 18001da177e4SLinus Torvalds /* We have a match */ 180138a33101SKinglong Mee if (req->wb_page) 18021da177e4SLinus Torvalds nfs_inode_remove_request(req); 1803ddeaa637SJoe Perches dprintk_cont(" OK\n"); 18041da177e4SLinus Torvalds goto next; 18051da177e4SLinus Torvalds } 18061da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 1807ddeaa637SJoe Perches dprintk_cont(" mismatch\n"); 18086d884e8fSFred nfs_mark_request_dirty(req); 180905990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 18101da177e4SLinus Torvalds next: 18111d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18121da177e4SLinus Torvalds } 1813353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1814353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 18150db10944SJan Kara clear_bdi_congested(inode_to_bdi(data->inode), BLK_RW_ASYNC); 1816353db796SNeilBrown 1817f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1818af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18195917ce84SFred Isaman } 18205917ce84SFred Isaman 18215917ce84SFred Isaman static void nfs_commit_release(void *calldata) 18225917ce84SFred Isaman { 18230b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18245917ce84SFred Isaman 1825f453a54aSFred Isaman data->completion_ops->completion(data); 1826c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 18271da177e4SLinus Torvalds } 1828788e7a89STrond Myklebust 1829788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 18300b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1831788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1832788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1833788e7a89STrond Myklebust }; 18341da177e4SLinus Torvalds 1835f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1836f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1837b20135d0STrond Myklebust .resched_write = nfs_commit_resched_write, 1838f453a54aSFred Isaman }; 1839f453a54aSFred Isaman 18401763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1841ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 184284c53ab5SFred Isaman { 184384c53ab5SFred Isaman int status; 184484c53ab5SFred Isaman 1845ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 184684c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1847ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 184884c53ab5SFred Isaman return status; 184984c53ab5SFred Isaman } 185084c53ab5SFred Isaman 1851b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 18521da177e4SLinus Torvalds { 18531da177e4SLinus Torvalds LIST_HEAD(head); 1854ea2cf228SFred Isaman struct nfs_commit_info cinfo; 185571d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1856af7cf057STrond Myklebust int error = 0; 1857b8413f98STrond Myklebust int res; 18581da177e4SLinus Torvalds 1859ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1860af7cf057STrond Myklebust nfs_commit_begin(cinfo.mds); 1861ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 1862af7cf057STrond Myklebust if (res) 1863ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 1864af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18651da177e4SLinus Torvalds if (error < 0) 1866af7cf057STrond Myklebust goto out_error; 1867b8413f98STrond Myklebust if (!may_wait) 1868b8413f98STrond Myklebust goto out_mark_dirty; 1869af7cf057STrond Myklebust error = wait_on_commit(cinfo.mds); 1870b8413f98STrond Myklebust if (error < 0) 1871b8413f98STrond Myklebust return error; 1872c5efa5fcSTrond Myklebust return res; 1873af7cf057STrond Myklebust out_error: 1874af7cf057STrond Myklebust res = error; 1875c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1876c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1877c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1878c5efa5fcSTrond Myklebust * that the data is on the disk. 1879c5efa5fcSTrond Myklebust */ 1880c5efa5fcSTrond Myklebust out_mark_dirty: 1881c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 18821da177e4SLinus Torvalds return res; 18831da177e4SLinus Torvalds } 1884b20135d0STrond Myklebust EXPORT_SYMBOL_GPL(nfs_commit_inode); 18858fc795f7STrond Myklebust 1886ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 18878fc795f7STrond Myklebust { 1888420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1889420e3646STrond Myklebust int flags = FLUSH_SYNC; 1890420e3646STrond Myklebust int ret = 0; 18918fc795f7STrond Myklebust 18923236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1893ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 18943236c3e1SJeff Layton return ret; 18953236c3e1SJeff Layton 1896a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1897a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1898a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1899420e3646STrond Myklebust */ 19001a4edf0fSTrond Myklebust if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)) 1901420e3646STrond Myklebust goto out_mark_dirty; 1902420e3646STrond Myklebust 1903a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1904420e3646STrond Myklebust flags = 0; 1905a00dd6c0SJeff Layton } 1906a00dd6c0SJeff Layton 1907420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1908420e3646STrond Myklebust if (ret >= 0) { 1909420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1910420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1911420e3646STrond Myklebust wbc->nr_to_write -= ret; 1912420e3646STrond Myklebust else 1913420e3646STrond Myklebust wbc->nr_to_write = 0; 1914420e3646STrond Myklebust } 19158fc795f7STrond Myklebust return 0; 1916420e3646STrond Myklebust } 1917420e3646STrond Myklebust out_mark_dirty: 19188fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 19198fc795f7STrond Myklebust return ret; 19208fc795f7STrond Myklebust } 192189d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1922863a3c6cSAndy Adamson 1923acdc53b2STrond Myklebust /* 1924837bb1d7STrond Myklebust * Wrapper for filemap_write_and_wait_range() 1925837bb1d7STrond Myklebust * 1926837bb1d7STrond Myklebust * Needed for pNFS in order to ensure data becomes visible to the 1927837bb1d7STrond Myklebust * client. 1928837bb1d7STrond Myklebust */ 1929837bb1d7STrond Myklebust int nfs_filemap_write_and_wait_range(struct address_space *mapping, 1930837bb1d7STrond Myklebust loff_t lstart, loff_t lend) 1931837bb1d7STrond Myklebust { 1932837bb1d7STrond Myklebust int ret; 1933837bb1d7STrond Myklebust 1934837bb1d7STrond Myklebust ret = filemap_write_and_wait_range(mapping, lstart, lend); 1935837bb1d7STrond Myklebust if (ret == 0) 1936837bb1d7STrond Myklebust ret = pnfs_sync_inode(mapping->host, true); 1937837bb1d7STrond Myklebust return ret; 1938837bb1d7STrond Myklebust } 1939837bb1d7STrond Myklebust EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); 1940837bb1d7STrond Myklebust 1941837bb1d7STrond Myklebust /* 1942acdc53b2STrond Myklebust * flush the inode to disk. 1943acdc53b2STrond Myklebust */ 1944acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 194534901f70STrond Myklebust { 1946f4ce1299STrond Myklebust int ret; 194734901f70STrond Myklebust 1948f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1949f4ce1299STrond Myklebust 19505bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 19516b196875SChuck Lever if (ret) 19526b196875SChuck Lever goto out; 19535bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 19546b196875SChuck Lever if (ret < 0) 19556b196875SChuck Lever goto out; 19565bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 19576b196875SChuck Lever ret = 0; 1958f4ce1299STrond Myklebust 19596b196875SChuck Lever out: 1960f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1961f4ce1299STrond Myklebust return ret; 19621c75950bSTrond Myklebust } 1963ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 19641c75950bSTrond Myklebust 19651b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 19661b3b4a1aSTrond Myklebust { 19671b3b4a1aSTrond Myklebust struct nfs_page *req; 19681b3b4a1aSTrond Myklebust int ret = 0; 19691b3b4a1aSTrond Myklebust 1970ba8b06e6STrond Myklebust wait_on_page_writeback(page); 19713e217045SWeston Andros Adamson 19723e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 19733e217045SWeston Andros Adamson * request */ 19746d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 19753e217045SWeston Andros Adamson 19763e217045SWeston Andros Adamson if (IS_ERR(req)) { 19773e217045SWeston Andros Adamson ret = PTR_ERR(req); 19783e217045SWeston Andros Adamson } else if (req) { 19793e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 19803e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 19813e217045SWeston Andros Adamson * request from the inode / page_private pointer and 19823e217045SWeston Andros Adamson * release it */ 19831b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 19841d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 19851b3b4a1aSTrond Myklebust } 19863e217045SWeston Andros Adamson 19871b3b4a1aSTrond Myklebust return ret; 19881b3b4a1aSTrond Myklebust } 19891b3b4a1aSTrond Myklebust 19901c75950bSTrond Myklebust /* 19911c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 19921c75950bSTrond Myklebust */ 1993c373fff7STrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 19941c75950bSTrond Myklebust { 199529418aa4SMel Gorman loff_t range_start = page_file_offset(page); 199609cbfeafSKirill A. Shutemov loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); 19977f2f12d9STrond Myklebust struct writeback_control wbc = { 19987f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 19997f2f12d9STrond Myklebust .nr_to_write = 0, 20007f2f12d9STrond Myklebust .range_start = range_start, 20017f2f12d9STrond Myklebust .range_end = range_end, 20027f2f12d9STrond Myklebust }; 20037f2f12d9STrond Myklebust int ret; 20047f2f12d9STrond Myklebust 2005f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 2006f4ce1299STrond Myklebust 20070522f6adSTrond Myklebust for (;;) { 2008ba8b06e6STrond Myklebust wait_on_page_writeback(page); 20097f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 2010c373fff7STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 20117f2f12d9STrond Myklebust if (ret < 0) 20127f2f12d9STrond Myklebust goto out_error; 20130522f6adSTrond Myklebust continue; 20147f2f12d9STrond Myklebust } 2015f4ce1299STrond Myklebust ret = 0; 20160522f6adSTrond Myklebust if (!PagePrivate(page)) 20170522f6adSTrond Myklebust break; 20180522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20197f2f12d9STrond Myklebust if (ret < 0) 20207f2f12d9STrond Myklebust goto out_error; 20217f2f12d9STrond Myklebust } 20227f2f12d9STrond Myklebust out_error: 2023f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 20247f2f12d9STrond Myklebust return ret; 20251c75950bSTrond Myklebust } 20261c75950bSTrond Myklebust 2027074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 2028074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 2029a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 2030074cc1deSTrond Myklebust { 20312da95652SJeff Layton /* 20322da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 20332da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 20342da95652SJeff Layton * 20352da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 20362da95652SJeff Layton * that we can safely release the inode reference while holding 20372da95652SJeff Layton * the page lock. 20382da95652SJeff Layton */ 20392da95652SJeff Layton if (PagePrivate(page)) 20402da95652SJeff Layton return -EBUSY; 2041074cc1deSTrond Myklebust 20428c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 20438c209ce7SDavid Howells return -EBUSY; 2044074cc1deSTrond Myklebust 2045a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 2046074cc1deSTrond Myklebust } 2047074cc1deSTrond Myklebust #endif 2048074cc1deSTrond Myklebust 2049f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 20501da177e4SLinus Torvalds { 20511da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 20521e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 20531da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 205420c2df83SPaul Mundt NULL); 20551da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 20561da177e4SLinus Torvalds return -ENOMEM; 20571da177e4SLinus Torvalds 205893d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 20591da177e4SLinus Torvalds nfs_wdata_cachep); 20601da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 20613dd4765fSJeff Layton goto out_destroy_write_cache; 20621da177e4SLinus Torvalds 20630b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 20640b7c0153SFred Isaman sizeof(struct nfs_commit_data), 20650b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 20660b7c0153SFred Isaman NULL); 20670b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 20683dd4765fSJeff Layton goto out_destroy_write_mempool; 20690b7c0153SFred Isaman 207093d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 20714c100210SYanchuan Nian nfs_cdata_cachep); 20721da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 20733dd4765fSJeff Layton goto out_destroy_commit_cache; 20741da177e4SLinus Torvalds 207589a09141SPeter Zijlstra /* 207689a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 207789a09141SPeter Zijlstra * 207889a09141SPeter Zijlstra * 64MB: 8192k 207989a09141SPeter Zijlstra * 128MB: 11585k 208089a09141SPeter Zijlstra * 256MB: 16384k 208189a09141SPeter Zijlstra * 512MB: 23170k 208289a09141SPeter Zijlstra * 1GB: 32768k 208389a09141SPeter Zijlstra * 2GB: 46340k 208489a09141SPeter Zijlstra * 4GB: 65536k 208589a09141SPeter Zijlstra * 8GB: 92681k 208689a09141SPeter Zijlstra * 16GB: 131072k 208789a09141SPeter Zijlstra * 208889a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 208989a09141SPeter Zijlstra * Limit the default to 256M 209089a09141SPeter Zijlstra */ 209189a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 209289a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 209389a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 209489a09141SPeter Zijlstra 20951da177e4SLinus Torvalds return 0; 20963dd4765fSJeff Layton 20973dd4765fSJeff Layton out_destroy_commit_cache: 20983dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 20993dd4765fSJeff Layton out_destroy_write_mempool: 21003dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 21013dd4765fSJeff Layton out_destroy_write_cache: 21023dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 21033dd4765fSJeff Layton return -ENOMEM; 21041da177e4SLinus Torvalds } 21051da177e4SLinus Torvalds 2106266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 21071da177e4SLinus Torvalds { 21081da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 21093dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21101da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 21111a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 21121da177e4SLinus Torvalds } 21131da177e4SLinus Torvalds 21144a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 21154a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 21164a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 21170eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 21180eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 21191ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 21204a0de55cSAnna Schumaker }; 2121