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 15784d3a9a9SWeston Andros Adamson /* 15884d3a9a9SWeston Andros Adamson * nfs_page_find_head_request_locked - find head request associated with @page 15984d3a9a9SWeston Andros Adamson * 16084d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 16184d3a9a9SWeston Andros Adamson * 16284d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 16384d3a9a9SWeston Andros Adamson */ 16429418aa4SMel Gorman static struct nfs_page * 16584d3a9a9SWeston Andros Adamson nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page) 166277459d2STrond Myklebust { 167277459d2STrond Myklebust struct nfs_page *req = NULL; 168277459d2STrond Myklebust 16929418aa4SMel Gorman if (PagePrivate(page)) 170277459d2STrond Myklebust req = (struct nfs_page *)page_private(page); 17102d1426cSWeston Andros Adamson else if (unlikely(PageSwapCache(page))) 17202d1426cSWeston Andros Adamson req = nfs_page_search_commits_for_head_request_locked(nfsi, 17302d1426cSWeston Andros Adamson page); 17429418aa4SMel Gorman 17584d3a9a9SWeston Andros Adamson if (req) { 17684d3a9a9SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 17729418aa4SMel Gorman kref_get(&req->wb_kref); 17884d3a9a9SWeston Andros Adamson } 17929418aa4SMel Gorman 180277459d2STrond Myklebust return req; 181277459d2STrond Myklebust } 182277459d2STrond Myklebust 18384d3a9a9SWeston Andros Adamson /* 18484d3a9a9SWeston Andros Adamson * nfs_page_find_head_request - find head request associated with @page 18584d3a9a9SWeston Andros Adamson * 18684d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 18784d3a9a9SWeston Andros Adamson */ 18884d3a9a9SWeston Andros Adamson static struct nfs_page *nfs_page_find_head_request(struct page *page) 189277459d2STrond Myklebust { 190d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 191277459d2STrond Myklebust struct nfs_page *req = NULL; 192277459d2STrond Myklebust 19382749dd4STrond Myklebust if (PagePrivate(page) || PageSwapCache(page)) { 194587142f8STrond Myklebust spin_lock(&inode->i_lock); 19584d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 196587142f8STrond Myklebust spin_unlock(&inode->i_lock); 19782749dd4STrond Myklebust } 198277459d2STrond Myklebust return req; 199277459d2STrond Myklebust } 200277459d2STrond Myklebust 2011da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 2021da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 2031da177e4SLinus Torvalds { 204d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 205a3d01454STrond Myklebust loff_t end, i_size; 206a3d01454STrond Myklebust pgoff_t end_index; 2071da177e4SLinus Torvalds 208a3d01454STrond Myklebust spin_lock(&inode->i_lock); 209a3d01454STrond Myklebust i_size = i_size_read(inode); 21009cbfeafSKirill A. Shutemov end_index = (i_size - 1) >> PAGE_SHIFT; 2118cd79788SHuang Ying if (i_size > 0 && page_index(page) < end_index) 212a3d01454STrond Myklebust goto out; 213d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 2141da177e4SLinus Torvalds if (i_size >= end) 215a3d01454STrond Myklebust goto out; 2161da177e4SLinus Torvalds i_size_write(inode, end); 217a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 218a3d01454STrond Myklebust out: 219a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 2201da177e4SLinus Torvalds } 2211da177e4SLinus Torvalds 222a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 223a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 224a301b777STrond Myklebust { 225d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 226a301b777STrond Myklebust } 227a301b777STrond Myklebust 228d72ddcbaSWeston Andros Adamson /* 229d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 230d72ddcbaSWeston Andros Adamson * @head - head request of page group 231d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 232d72ddcbaSWeston Andros Adamson * 233d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 234d72ddcbaSWeston Andros Adamson * page offset @page_offset. 235d72ddcbaSWeston Andros Adamson * 236d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 237d72ddcbaSWeston Andros Adamson * match is found. 238d72ddcbaSWeston Andros Adamson * 239d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 240d72ddcbaSWeston Andros Adamson */ 241d72ddcbaSWeston Andros Adamson static struct nfs_page * 242d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 243d72ddcbaSWeston Andros Adamson { 244d72ddcbaSWeston Andros Adamson struct nfs_page *req; 245d72ddcbaSWeston Andros Adamson 246d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(head != head->wb_head); 247d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags)); 248d72ddcbaSWeston Andros Adamson 249d72ddcbaSWeston Andros Adamson req = head; 250d72ddcbaSWeston Andros Adamson do { 251d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 252d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 253d72ddcbaSWeston Andros Adamson return req; 254d72ddcbaSWeston Andros Adamson 255d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 256d72ddcbaSWeston Andros Adamson } while (req != head); 257d72ddcbaSWeston Andros Adamson 258d72ddcbaSWeston Andros Adamson return NULL; 259d72ddcbaSWeston Andros Adamson } 260d72ddcbaSWeston Andros Adamson 261d72ddcbaSWeston Andros Adamson /* 262d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 263d72ddcbaSWeston Andros Adamson * @head - head request of page group 264d72ddcbaSWeston Andros Adamson * 265d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 266d72ddcbaSWeston Andros Adamson * returns false otherwise 267d72ddcbaSWeston Andros Adamson */ 268d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 269d72ddcbaSWeston Andros Adamson { 270d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 271d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 272d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 273d72ddcbaSWeston Andros Adamson 274fd2f3a06SWeston Andros Adamson nfs_page_group_lock(req, false); 275d72ddcbaSWeston Andros Adamson 276d72ddcbaSWeston Andros Adamson do { 277d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 278d72ddcbaSWeston Andros Adamson if (tmp) { 279d72ddcbaSWeston Andros Adamson /* no way this should happen */ 280d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(tmp->wb_pgbase != pos); 281d72ddcbaSWeston Andros Adamson pos += tmp->wb_bytes - (pos - tmp->wb_pgbase); 282d72ddcbaSWeston Andros Adamson } 283d72ddcbaSWeston Andros Adamson } while (tmp && pos < len); 284d72ddcbaSWeston Andros Adamson 285d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 286d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(pos > len); 287d72ddcbaSWeston Andros Adamson return pos == len; 288d72ddcbaSWeston Andros Adamson } 289d72ddcbaSWeston Andros Adamson 2901da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 2911da177e4SLinus Torvalds * covers the full page. 2921da177e4SLinus Torvalds */ 293d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 2941da177e4SLinus Torvalds { 295d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 2961da177e4SLinus Torvalds return; 297d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 2981da177e4SLinus Torvalds return; 299d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 3001da177e4SLinus Torvalds } 3011da177e4SLinus Torvalds 3021da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 3031da177e4SLinus Torvalds { 304e87b4c7aSNeilBrown int ret = 0; 305cca588d6STrond Myklebust 306e87b4c7aSNeilBrown if (wbc->sync_mode == WB_SYNC_ALL) 307e87b4c7aSNeilBrown ret = FLUSH_COND_STABLE; 308e87b4c7aSNeilBrown return ret; 3091da177e4SLinus Torvalds } 3101da177e4SLinus Torvalds 3111da177e4SLinus Torvalds /* 31289a09141SPeter Zijlstra * NFS congestion control 31389a09141SPeter Zijlstra */ 31489a09141SPeter Zijlstra 31589a09141SPeter Zijlstra int nfs_congestion_kb; 31689a09141SPeter Zijlstra 31789a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 31889a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 31989a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 32089a09141SPeter Zijlstra 321deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 32289a09141SPeter Zijlstra { 3230db10944SJan Kara struct inode *inode = page_file_mapping(page)->host; 3240db10944SJan Kara struct nfs_server *nfss = NFS_SERVER(inode); 3255a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 3265a6d41b3STrond Myklebust 327deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 32889a09141SPeter Zijlstra 329277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 3300db10944SJan Kara NFS_CONGESTION_ON_THRESH) 3310db10944SJan Kara set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 33289a09141SPeter Zijlstra } 33389a09141SPeter Zijlstra 33420633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 33589a09141SPeter Zijlstra { 33620633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 33789a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 33889a09141SPeter Zijlstra 33920633f04SWeston Andros Adamson if (!nfs_page_group_sync_on_bit(req, PG_WB_END)) 34020633f04SWeston Andros Adamson return; 34120633f04SWeston Andros Adamson 34220633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 343c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 3440db10944SJan Kara clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 34589a09141SPeter Zijlstra } 34689a09141SPeter Zijlstra 347d4581383SWeston Andros Adamson 348d4581383SWeston Andros Adamson /* nfs_page_group_clear_bits 349d4581383SWeston Andros Adamson * @req - an nfs request 350d4581383SWeston Andros Adamson * clears all page group related bits from @req 351d4581383SWeston Andros Adamson */ 352d4581383SWeston Andros Adamson static void 353d4581383SWeston Andros Adamson nfs_page_group_clear_bits(struct nfs_page *req) 354e261f51fSTrond Myklebust { 355d4581383SWeston Andros Adamson clear_bit(PG_TEARDOWN, &req->wb_flags); 356d4581383SWeston Andros Adamson clear_bit(PG_UNLOCKPAGE, &req->wb_flags); 357d4581383SWeston Andros Adamson clear_bit(PG_UPTODATE, &req->wb_flags); 358d4581383SWeston Andros Adamson clear_bit(PG_WB_END, &req->wb_flags); 359d4581383SWeston Andros Adamson clear_bit(PG_REMOVE, &req->wb_flags); 360d4581383SWeston Andros Adamson } 361d4581383SWeston Andros Adamson 362d4581383SWeston Andros Adamson 363d4581383SWeston Andros Adamson /* 364d4581383SWeston Andros Adamson * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req 365d4581383SWeston Andros Adamson * 366d4581383SWeston Andros Adamson * this is a helper function for nfs_lock_and_join_requests 367d4581383SWeston Andros Adamson * 368d4581383SWeston Andros Adamson * @inode - inode associated with request page group, must be holding inode lock 369d4581383SWeston Andros Adamson * @head - head request of page group, must be holding head lock 370d4581383SWeston Andros Adamson * @req - request that couldn't lock and needs to wait on the req bit lock 371d4581383SWeston Andros Adamson * 372d4581383SWeston Andros Adamson * NOTE: this must be called holding page_group bit lock and inode spin lock 373d4581383SWeston Andros Adamson * and BOTH will be released before returning. 374d4581383SWeston Andros Adamson * 375d4581383SWeston Andros Adamson * returns 0 on success, < 0 on error. 376d4581383SWeston Andros Adamson */ 377d4581383SWeston Andros Adamson static int 378d4581383SWeston Andros Adamson nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head, 3796d17d653STrond Myklebust struct nfs_page *req) 380d4581383SWeston Andros Adamson __releases(&inode->i_lock) 381d4581383SWeston Andros Adamson { 382d4581383SWeston Andros Adamson struct nfs_page *tmp; 383e261f51fSTrond Myklebust int ret; 384e261f51fSTrond Myklebust 385d4581383SWeston Andros Adamson /* relinquish all the locks successfully grabbed this run */ 386a0e265bcSTrond Myklebust for (tmp = head->wb_this_page ; tmp != req; tmp = tmp->wb_this_page) 387d4581383SWeston Andros Adamson nfs_unlock_request(tmp); 388d4581383SWeston Andros Adamson 389d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags)); 390d4581383SWeston Andros Adamson 391d4581383SWeston Andros Adamson /* grab a ref on the request that will be waited on */ 392d4581383SWeston Andros Adamson kref_get(&req->wb_kref); 393d4581383SWeston Andros Adamson 394d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 395587142f8STrond Myklebust spin_unlock(&inode->i_lock); 396d4581383SWeston Andros Adamson 397d4581383SWeston Andros Adamson /* release ref from nfs_page_find_head_request_locked */ 398a0e265bcSTrond Myklebust nfs_unlock_and_release_request(head); 399d4581383SWeston Andros Adamson 400e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 401e261f51fSTrond Myklebust nfs_release_request(req); 402d4581383SWeston Andros Adamson 403d4581383SWeston Andros Adamson return ret; 404e261f51fSTrond Myklebust } 405d4581383SWeston Andros Adamson 406d4581383SWeston Andros Adamson /* 407d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 408d4581383SWeston Andros Adamson * 409d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 410d4581383SWeston Andros Adamson * @old_head - the old head of the list 411d4581383SWeston Andros Adamson * 412d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 413d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 414d4581383SWeston Andros Adamson * with a reference held by some other context. 415d4581383SWeston Andros Adamson */ 416d4581383SWeston Andros Adamson static void 417d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 418d4581383SWeston Andros Adamson struct nfs_page *old_head) 419d4581383SWeston Andros Adamson { 420d4581383SWeston Andros Adamson while (destroy_list) { 421d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 422d4581383SWeston Andros Adamson 423d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 424d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 425d4581383SWeston Andros Adamson 426d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 427d4581383SWeston Andros Adamson 428d4581383SWeston Andros Adamson /* make sure old group is not used */ 429d4581383SWeston Andros Adamson subreq->wb_head = subreq; 430d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 431d4581383SWeston Andros Adamson 432d4581383SWeston Andros Adamson /* subreq is now totally disconnected from page group or any 433d4581383SWeston Andros Adamson * write / commit lists. last chance to wake any waiters */ 434d4581383SWeston Andros Adamson nfs_unlock_request(subreq); 435d4581383SWeston Andros Adamson 436d4581383SWeston Andros Adamson if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) { 437d4581383SWeston Andros Adamson /* release ref on old head request */ 438d4581383SWeston Andros Adamson nfs_release_request(old_head); 439d4581383SWeston Andros Adamson 440d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 441d4581383SWeston Andros Adamson 442d4581383SWeston Andros Adamson /* release the PG_INODE_REF reference */ 443d4581383SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) 444d4581383SWeston Andros Adamson nfs_release_request(subreq); 445d4581383SWeston Andros Adamson else 446d4581383SWeston Andros Adamson WARN_ON_ONCE(1); 447d4581383SWeston Andros Adamson } else { 448d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags)); 449d4581383SWeston Andros Adamson /* zombie requests have already released the last 450d4581383SWeston Andros Adamson * reference and were waiting on the rest of the 451d4581383SWeston Andros Adamson * group to complete. Since it's no longer part of a 452d4581383SWeston Andros Adamson * group, simply free the request */ 453d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 454d4581383SWeston Andros Adamson nfs_free_request(subreq); 455d4581383SWeston Andros Adamson } 456d4581383SWeston Andros Adamson } 457d4581383SWeston Andros Adamson } 458d4581383SWeston Andros Adamson 459d4581383SWeston Andros Adamson /* 460d4581383SWeston Andros Adamson * nfs_lock_and_join_requests - join all subreqs to the head req and return 461d4581383SWeston Andros Adamson * a locked reference, cancelling any pending 462d4581383SWeston Andros Adamson * operations for this page. 463d4581383SWeston Andros Adamson * 464d4581383SWeston Andros Adamson * @page - the page used to lookup the "page group" of nfs_page structures 465d4581383SWeston Andros Adamson * 466d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 467d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 468d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 469d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 470d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 471d4581383SWeston Andros Adamson * 472d4581383SWeston Andros Adamson * Returns a locked, referenced pointer to the head request - which after 473d4581383SWeston Andros Adamson * this call is guaranteed to be the only request associated with the page. 474d4581383SWeston Andros Adamson * Returns NULL if no requests are found for @page, or a ERR_PTR if an 475d4581383SWeston Andros Adamson * error was encountered. 476d4581383SWeston Andros Adamson */ 477d4581383SWeston Andros Adamson static struct nfs_page * 4786d17d653STrond Myklebust nfs_lock_and_join_requests(struct page *page) 479d4581383SWeston Andros Adamson { 480d4581383SWeston Andros Adamson struct inode *inode = page_file_mapping(page)->host; 481d4581383SWeston Andros Adamson struct nfs_page *head, *subreq; 482d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 483d4581383SWeston Andros Adamson unsigned int total_bytes; 484d4581383SWeston Andros Adamson int ret; 485d4581383SWeston Andros Adamson 486d4581383SWeston Andros Adamson try_again: 487d4581383SWeston Andros Adamson spin_lock(&inode->i_lock); 488d4581383SWeston Andros Adamson 489d4581383SWeston Andros Adamson /* 490d4581383SWeston Andros Adamson * A reference is taken only on the head request which acts as a 491d4581383SWeston Andros Adamson * reference to the whole page group - the group will not be destroyed 492d4581383SWeston Andros Adamson * until the head reference is released. 493d4581383SWeston Andros Adamson */ 494d4581383SWeston Andros Adamson head = nfs_page_find_head_request_locked(NFS_I(inode), page); 495d4581383SWeston Andros Adamson 496d4581383SWeston Andros Adamson if (!head) { 497587142f8STrond Myklebust spin_unlock(&inode->i_lock); 498d4581383SWeston Andros Adamson return NULL; 499d4581383SWeston Andros Adamson } 500d4581383SWeston Andros Adamson 501a0e265bcSTrond Myklebust /* lock the page head first in order to avoid an ABBA inefficiency */ 502a0e265bcSTrond Myklebust if (!nfs_lock_request(head)) { 503a0e265bcSTrond Myklebust spin_unlock(&inode->i_lock); 504a0e265bcSTrond Myklebust ret = nfs_wait_on_request(head); 505a0e265bcSTrond Myklebust nfs_release_request(head); 506a0e265bcSTrond Myklebust if (ret < 0) 507a0e265bcSTrond Myklebust return ERR_PTR(ret); 508a0e265bcSTrond Myklebust goto try_again; 509a0e265bcSTrond Myklebust } 510a0e265bcSTrond Myklebust 5117c3af975SWeston Andros Adamson /* holding inode lock, so always make a non-blocking call to try the 5127c3af975SWeston Andros Adamson * page group lock */ 513fd2f3a06SWeston Andros Adamson ret = nfs_page_group_lock(head, true); 51494970014SWeston Andros Adamson if (ret < 0) { 51594970014SWeston Andros Adamson spin_unlock(&inode->i_lock); 5167c3af975SWeston Andros Adamson 5177c3af975SWeston Andros Adamson nfs_page_group_lock_wait(head); 518a0e265bcSTrond Myklebust nfs_unlock_and_release_request(head); 5197c3af975SWeston Andros Adamson goto try_again; 5207c3af975SWeston Andros Adamson } 5217c3af975SWeston Andros Adamson 5227c3af975SWeston Andros Adamson /* lock each request in the page group */ 523a0e265bcSTrond Myklebust total_bytes = head->wb_bytes; 524a0e265bcSTrond Myklebust for (subreq = head->wb_this_page; subreq != head; 525a0e265bcSTrond Myklebust subreq = subreq->wb_this_page) { 526d4581383SWeston Andros Adamson if (!nfs_lock_request(subreq)) { 527d4581383SWeston Andros Adamson /* releases page group bit lock and 528d4581383SWeston Andros Adamson * inode spin lock and all references */ 529d4581383SWeston Andros Adamson ret = nfs_unroll_locks_and_wait(inode, head, 5306d17d653STrond Myklebust subreq); 531d4581383SWeston Andros Adamson 532d4581383SWeston Andros Adamson if (ret == 0) 533d4581383SWeston Andros Adamson goto try_again; 534d4581383SWeston Andros Adamson 535d4581383SWeston Andros Adamson return ERR_PTR(ret); 536d4581383SWeston Andros Adamson } 537*e14bebf6STrond Myklebust /* 538*e14bebf6STrond Myklebust * Subrequests are always contiguous, non overlapping 539*e14bebf6STrond Myklebust * and in order - but may be repeated (mirrored writes). 540*e14bebf6STrond Myklebust */ 541*e14bebf6STrond Myklebust if (subreq->wb_offset == (head->wb_offset + total_bytes)) { 542*e14bebf6STrond Myklebust /* keep track of how many bytes this group covers */ 543*e14bebf6STrond Myklebust total_bytes += subreq->wb_bytes; 544*e14bebf6STrond Myklebust } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || 545*e14bebf6STrond Myklebust ((subreq->wb_offset + subreq->wb_bytes) > 546*e14bebf6STrond Myklebust (head->wb_offset + total_bytes)))) { 547*e14bebf6STrond Myklebust nfs_unlock_request(subreq); 548*e14bebf6STrond Myklebust nfs_unroll_locks_and_wait(inode, head, subreq); 549*e14bebf6STrond Myklebust return ERR_PTR(-EIO); 550*e14bebf6STrond Myklebust } 551a0e265bcSTrond Myklebust } 552d4581383SWeston Andros Adamson 553d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 554d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 555d4581383SWeston Andros Adamson subreq = head; 556d4581383SWeston Andros Adamson do { 557411a99adSWeston Andros Adamson nfs_clear_request_commit(subreq); 558d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 559d4581383SWeston Andros Adamson } while (subreq != head); 560d4581383SWeston Andros Adamson 561d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 562d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 563d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 564d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 565d4581383SWeston Andros Adamson head->wb_this_page = head; 566d4581383SWeston Andros Adamson 567d4581383SWeston Andros Adamson /* change head request to cover whole range that 568d4581383SWeston Andros Adamson * the former page group covered */ 569d4581383SWeston Andros Adamson head->wb_bytes = total_bytes; 570d4581383SWeston Andros Adamson } 571d4581383SWeston Andros Adamson 572d4581383SWeston Andros Adamson /* 573d4581383SWeston Andros Adamson * prepare head request to be added to new pgio descriptor 574d4581383SWeston Andros Adamson */ 575d4581383SWeston Andros Adamson nfs_page_group_clear_bits(head); 576d4581383SWeston Andros Adamson 577d4581383SWeston Andros Adamson /* 578d4581383SWeston Andros Adamson * some part of the group was still on the inode list - otherwise 579d4581383SWeston Andros Adamson * the group wouldn't be involved in async write. 580d4581383SWeston Andros Adamson * grab a reference for the head request, iff it needs one. 581d4581383SWeston Andros Adamson */ 582d4581383SWeston Andros Adamson if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags)) 583d4581383SWeston Andros Adamson kref_get(&head->wb_kref); 584d4581383SWeston Andros Adamson 585d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 586d4581383SWeston Andros Adamson 587411a99adSWeston Andros Adamson /* drop lock to clean uprequests on destroy list */ 588d4581383SWeston Andros Adamson spin_unlock(&inode->i_lock); 589d4581383SWeston Andros Adamson 590d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(destroy_list, head); 591d4581383SWeston Andros Adamson 592d4581383SWeston Andros Adamson /* still holds ref on head from nfs_page_find_head_request_locked 593d4581383SWeston Andros Adamson * and still has lock on head from lock loop */ 594d4581383SWeston Andros Adamson return head; 595612c9384STrond Myklebust } 596074cc1deSTrond Myklebust 5970bcbf039SPeng Tao static void nfs_write_error_remove_page(struct nfs_page *req) 5980bcbf039SPeng Tao { 5990bcbf039SPeng Tao nfs_unlock_request(req); 6000bcbf039SPeng Tao nfs_end_page_writeback(req); 6010bcbf039SPeng Tao generic_error_remove_page(page_file_mapping(req->wb_page), 6020bcbf039SPeng Tao req->wb_page); 6031f84ccdfSFred Isaman nfs_release_request(req); 6040bcbf039SPeng Tao } 6050bcbf039SPeng Tao 606a6598813STrond Myklebust static bool 607a6598813STrond Myklebust nfs_error_is_fatal_on_server(int err) 608a6598813STrond Myklebust { 609a6598813STrond Myklebust switch (err) { 610a6598813STrond Myklebust case 0: 611a6598813STrond Myklebust case -ERESTARTSYS: 612a6598813STrond Myklebust case -EINTR: 613a6598813STrond Myklebust return false; 614a6598813STrond Myklebust } 615a6598813STrond Myklebust return nfs_error_is_fatal(err); 616e261f51fSTrond Myklebust } 617074cc1deSTrond Myklebust 618074cc1deSTrond Myklebust /* 619074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 620074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 621074cc1deSTrond Myklebust */ 622074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 6236d17d653STrond Myklebust struct page *page) 624074cc1deSTrond Myklebust { 625074cc1deSTrond Myklebust struct nfs_page *req; 626074cc1deSTrond Myklebust int ret = 0; 627074cc1deSTrond Myklebust 6286d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 629074cc1deSTrond Myklebust if (!req) 630074cc1deSTrond Myklebust goto out; 631074cc1deSTrond Myklebust ret = PTR_ERR(req); 632074cc1deSTrond Myklebust if (IS_ERR(req)) 633074cc1deSTrond Myklebust goto out; 634074cc1deSTrond Myklebust 635deed85e7STrond Myklebust nfs_set_page_writeback(page); 636deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 637074cc1deSTrond Myklebust 638deed85e7STrond Myklebust ret = 0; 639a6598813STrond Myklebust /* If there is a fatal error that covers this write, just exit */ 640a6598813STrond Myklebust if (nfs_error_is_fatal_on_server(req->wb_context->error)) 641a6598813STrond Myklebust goto out_launder; 642a6598813STrond Myklebust 643f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 644074cc1deSTrond Myklebust ret = pgio->pg_error; 6450bcbf039SPeng Tao /* 646c373fff7STrond Myklebust * Remove the problematic req upon fatal errors on the server 6470bcbf039SPeng Tao */ 6480bcbf039SPeng Tao if (nfs_error_is_fatal(ret)) { 6490bcbf039SPeng Tao nfs_context_set_write_error(req->wb_context, ret); 650c373fff7STrond Myklebust if (nfs_error_is_fatal_on_server(ret)) 651a6598813STrond Myklebust goto out_launder; 652d6c843b9SPeng Tao } 6530bcbf039SPeng Tao nfs_redirty_request(req); 6540bcbf039SPeng Tao ret = -EAGAIN; 65540f90271STrond Myklebust } else 65640f90271STrond Myklebust nfs_add_stats(page_file_mapping(page)->host, 65740f90271STrond Myklebust NFSIOS_WRITEPAGES, 1); 658074cc1deSTrond Myklebust out: 659074cc1deSTrond Myklebust return ret; 660a6598813STrond Myklebust out_launder: 661a6598813STrond Myklebust nfs_write_error_remove_page(req); 662a6598813STrond Myklebust return ret; 663e261f51fSTrond Myklebust } 664e261f51fSTrond Myklebust 665d6c843b9SPeng Tao static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, 666c373fff7STrond Myklebust struct nfs_pageio_descriptor *pgio) 667f758c885STrond Myklebust { 668cfb506e1STrond Myklebust int ret; 669f758c885STrond Myklebust 6708cd79788SHuang Ying nfs_pageio_cond_complete(pgio, page_index(page)); 6716d17d653STrond Myklebust ret = nfs_page_async_flush(pgio, page); 672cfb506e1STrond Myklebust if (ret == -EAGAIN) { 673cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 674cfb506e1STrond Myklebust ret = 0; 675cfb506e1STrond Myklebust } 676cfb506e1STrond Myklebust return ret; 677f758c885STrond Myklebust } 678f758c885STrond Myklebust 679e261f51fSTrond Myklebust /* 6801da177e4SLinus Torvalds * Write an mmapped page to the server. 6811da177e4SLinus Torvalds */ 682d6c843b9SPeng Tao static int nfs_writepage_locked(struct page *page, 683c373fff7STrond Myklebust struct writeback_control *wbc) 6841da177e4SLinus Torvalds { 685f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 68640f90271STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 687e261f51fSTrond Myklebust int err; 6881da177e4SLinus Torvalds 68940f90271STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 690811ed92eSTrond Myklebust nfs_pageio_init_write(&pgio, inode, 0, 691a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 692c373fff7STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 693f758c885STrond Myklebust nfs_pageio_complete(&pgio); 694f758c885STrond Myklebust if (err < 0) 6954d770ccfSTrond Myklebust return err; 696f758c885STrond Myklebust if (pgio.pg_error < 0) 697f758c885STrond Myklebust return pgio.pg_error; 698f758c885STrond Myklebust return 0; 6994d770ccfSTrond Myklebust } 7004d770ccfSTrond Myklebust 7014d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 7024d770ccfSTrond Myklebust { 703f758c885STrond Myklebust int ret; 7044d770ccfSTrond Myklebust 705c373fff7STrond Myklebust ret = nfs_writepage_locked(page, wbc); 7061da177e4SLinus Torvalds unlock_page(page); 707f758c885STrond Myklebust return ret; 708f758c885STrond Myklebust } 709f758c885STrond Myklebust 710f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 711f758c885STrond Myklebust { 712f758c885STrond Myklebust int ret; 713f758c885STrond Myklebust 714c373fff7STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 715f758c885STrond Myklebust unlock_page(page); 716f758c885STrond Myklebust return ret; 7171da177e4SLinus Torvalds } 7181da177e4SLinus Torvalds 719919e3bd9STrond Myklebust static void nfs_io_completion_commit(void *inode) 720919e3bd9STrond Myklebust { 721919e3bd9STrond Myklebust nfs_commit_inode(inode, 0); 722919e3bd9STrond Myklebust } 723919e3bd9STrond Myklebust 7241da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 7251da177e4SLinus Torvalds { 7261da177e4SLinus Torvalds struct inode *inode = mapping->host; 727c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 728919e3bd9STrond Myklebust struct nfs_io_completion *ioc = nfs_io_completion_alloc(GFP_NOFS); 7291da177e4SLinus Torvalds int err; 7301da177e4SLinus Torvalds 73191d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 73291d5b470SChuck Lever 733919e3bd9STrond Myklebust if (ioc) 734919e3bd9STrond Myklebust nfs_io_completion_init(ioc, nfs_io_completion_commit, inode); 735919e3bd9STrond Myklebust 736a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 737a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 738919e3bd9STrond Myklebust pgio.pg_io_completion = ioc; 739f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 740c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 741919e3bd9STrond Myklebust nfs_io_completion_put(ioc); 74272cb77f4STrond Myklebust 743f758c885STrond Myklebust if (err < 0) 74472cb77f4STrond Myklebust goto out_err; 74572cb77f4STrond Myklebust err = pgio.pg_error; 74672cb77f4STrond Myklebust if (err < 0) 74772cb77f4STrond Myklebust goto out_err; 748c63c7b05STrond Myklebust return 0; 74972cb77f4STrond Myklebust out_err: 75072cb77f4STrond Myklebust return err; 7511da177e4SLinus Torvalds } 7521da177e4SLinus Torvalds 7531da177e4SLinus Torvalds /* 7541da177e4SLinus Torvalds * Insert a write request into an inode 7551da177e4SLinus Torvalds */ 756d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 7571da177e4SLinus Torvalds { 7581da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 759e7d39069STrond Myklebust 7602bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 7612bfc6e56SWeston Andros Adamson 762e7d39069STrond Myklebust /* Lock the request! */ 7637ad84aa9STrond Myklebust nfs_lock_request(req); 764e7d39069STrond Myklebust 765e7d39069STrond Myklebust spin_lock(&inode->i_lock); 766cb1410c7SWeston Andros Adamson if (!nfsi->nrequests && 767cb1410c7SWeston Andros Adamson NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 768a9a4a87aSTrond Myklebust inode->i_version++; 76929418aa4SMel Gorman /* 77029418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 77129418aa4SMel Gorman * with invalidate/truncate. 77229418aa4SMel Gorman */ 77329418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 7742df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 775deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 776277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 77729418aa4SMel Gorman } 778cb1410c7SWeston Andros Adamson nfsi->nrequests++; 77917089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 780cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 781cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 782cb1410c7SWeston Andros Adamson * adding subrequests. */ 78317089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 784c03b4024STrond Myklebust kref_get(&req->wb_kref); 785e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 7861da177e4SLinus Torvalds } 7871da177e4SLinus Torvalds 7881da177e4SLinus Torvalds /* 78989a09141SPeter Zijlstra * Remove a write request from an inode 7901da177e4SLinus Torvalds */ 7911da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7921da177e4SLinus Torvalds { 7932b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 7941da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 79520633f04SWeston Andros Adamson struct nfs_page *head; 79620633f04SWeston Andros Adamson 79720633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 79820633f04SWeston Andros Adamson head = req->wb_head; 7991da177e4SLinus Torvalds 800587142f8STrond Myklebust spin_lock(&inode->i_lock); 80167911c8fSAnna Schumaker if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { 80220633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 80320633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 80420633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 80529418aa4SMel Gorman } 806cb1410c7SWeston Andros Adamson nfsi->nrequests--; 807cb1410c7SWeston Andros Adamson spin_unlock(&inode->i_lock); 808cb1410c7SWeston Andros Adamson } else { 809cb1410c7SWeston Andros Adamson spin_lock(&inode->i_lock); 810cb1410c7SWeston Andros Adamson nfsi->nrequests--; 811587142f8STrond Myklebust spin_unlock(&inode->i_lock); 81220633f04SWeston Andros Adamson } 81317089a29SWeston Andros Adamson 81417089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 8151da177e4SLinus Torvalds nfs_release_request(req); 8161da177e4SLinus Torvalds } 8171da177e4SLinus Torvalds 81861822ab5STrond Myklebust static void 8196d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 82061822ab5STrond Myklebust { 82167911c8fSAnna Schumaker if (req->wb_page) 82261822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 82361822ab5STrond Myklebust } 82461822ab5STrond Myklebust 8253a3908c8STrond Myklebust /* 8263a3908c8STrond Myklebust * nfs_page_search_commits_for_head_request_locked 8273a3908c8STrond Myklebust * 8283a3908c8STrond Myklebust * Search through commit lists on @inode for the head request for @page. 8293a3908c8STrond Myklebust * Must be called while holding the inode (which is cinfo) lock. 8303a3908c8STrond Myklebust * 8313a3908c8STrond Myklebust * Returns the head request if found, or NULL if not found. 8323a3908c8STrond Myklebust */ 8333a3908c8STrond Myklebust static struct nfs_page * 8343a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 8353a3908c8STrond Myklebust struct page *page) 8363a3908c8STrond Myklebust { 8373a3908c8STrond Myklebust struct nfs_page *freq, *t; 8383a3908c8STrond Myklebust struct nfs_commit_info cinfo; 8393a3908c8STrond Myklebust struct inode *inode = &nfsi->vfs_inode; 8403a3908c8STrond Myklebust 8413a3908c8STrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 8423a3908c8STrond Myklebust 8433a3908c8STrond Myklebust /* search through pnfs commit lists */ 8443a3908c8STrond Myklebust freq = pnfs_search_commit_reqs(inode, &cinfo, page); 8453a3908c8STrond Myklebust if (freq) 8463a3908c8STrond Myklebust return freq->wb_head; 8473a3908c8STrond Myklebust 8483a3908c8STrond Myklebust /* Linearly search the commit list for the correct request */ 8493a3908c8STrond Myklebust list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 8503a3908c8STrond Myklebust if (freq->wb_page == page) 8513a3908c8STrond Myklebust return freq->wb_head; 8523a3908c8STrond Myklebust } 8533a3908c8STrond Myklebust 8543a3908c8STrond Myklebust return NULL; 8553a3908c8STrond Myklebust } 8563a3908c8STrond Myklebust 8578dd37758STrond Myklebust /** 85886d80f97STrond Myklebust * nfs_request_add_commit_list_locked - add request to a commit list 85986d80f97STrond Myklebust * @req: pointer to a struct nfs_page 86086d80f97STrond Myklebust * @dst: commit list head 86186d80f97STrond Myklebust * @cinfo: holds list lock and accounting info 86286d80f97STrond Myklebust * 86386d80f97STrond Myklebust * This sets the PG_CLEAN bit, updates the cinfo count of 86486d80f97STrond Myklebust * number of outstanding requests requiring a commit as well as 86586d80f97STrond Myklebust * the MM page stats. 86686d80f97STrond Myklebust * 867fe238e60SDave Wysochanski * The caller must hold cinfo->inode->i_lock, and the nfs_page lock. 86886d80f97STrond Myklebust */ 86986d80f97STrond Myklebust void 87086d80f97STrond Myklebust nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, 87186d80f97STrond Myklebust struct nfs_commit_info *cinfo) 87286d80f97STrond Myklebust { 87386d80f97STrond Myklebust set_bit(PG_CLEAN, &req->wb_flags); 87486d80f97STrond Myklebust nfs_list_add_request(req, dst); 87586d80f97STrond Myklebust cinfo->mds->ncommit++; 87686d80f97STrond Myklebust } 87786d80f97STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); 87886d80f97STrond Myklebust 87986d80f97STrond Myklebust /** 8808dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 8818dd37758STrond Myklebust * @req: pointer to a struct nfs_page 882ea2cf228SFred Isaman * @dst: commit list head 883ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8848dd37758STrond Myklebust * 885ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 8868dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8878dd37758STrond Myklebust * the MM page stats. 8888dd37758STrond Myklebust * 889ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8908dd37758STrond Myklebust * holding the nfs_page lock. 8918dd37758STrond Myklebust */ 8928dd37758STrond Myklebust void 8936272dcc6SAnna Schumaker nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) 8948dd37758STrond Myklebust { 895fe238e60SDave Wysochanski spin_lock(&cinfo->inode->i_lock); 8966272dcc6SAnna Schumaker nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); 897fe238e60SDave Wysochanski spin_unlock(&cinfo->inode->i_lock); 89867911c8fSAnna Schumaker if (req->wb_page) 89986d80f97STrond Myklebust nfs_mark_page_unstable(req->wb_page, cinfo); 9008dd37758STrond Myklebust } 9018dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 9028dd37758STrond Myklebust 9038dd37758STrond Myklebust /** 9048dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 9058dd37758STrond Myklebust * @req: pointer to a nfs_page 906ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 9078dd37758STrond Myklebust * 908ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 9098dd37758STrond Myklebust * number of outstanding requests requiring a commit 9108dd37758STrond Myklebust * It does not update the MM page stats. 9118dd37758STrond Myklebust * 912ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 9138dd37758STrond Myklebust */ 9148dd37758STrond Myklebust void 915ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 916ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9178dd37758STrond Myklebust { 9188dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 9198dd37758STrond Myklebust return; 9208dd37758STrond Myklebust nfs_list_remove_request(req); 921ea2cf228SFred Isaman cinfo->mds->ncommit--; 9228dd37758STrond Myklebust } 9238dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 9248dd37758STrond Myklebust 925ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 926ea2cf228SFred Isaman struct inode *inode) 927ea2cf228SFred Isaman { 928fe238e60SDave Wysochanski cinfo->inode = inode; 929ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 930ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 931b359f9d0SFred Isaman cinfo->dreq = NULL; 932f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 933ea2cf228SFred Isaman } 934ea2cf228SFred Isaman 935ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 936ea2cf228SFred Isaman struct inode *inode, 937ea2cf228SFred Isaman struct nfs_direct_req *dreq) 938ea2cf228SFred Isaman { 9391763da12SFred Isaman if (dreq) 9401763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 9411763da12SFred Isaman else 942ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 943ea2cf228SFred Isaman } 944ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 9458dd37758STrond Myklebust 9461da177e4SLinus Torvalds /* 9471da177e4SLinus Torvalds * Add a request to the inode's commit list. 9481da177e4SLinus Torvalds */ 9491763da12SFred Isaman void 950ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 951b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 9521da177e4SLinus Torvalds { 953b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 9548dd37758STrond Myklebust return; 9556272dcc6SAnna Schumaker nfs_request_add_commit_list(req, cinfo); 9561da177e4SLinus Torvalds } 9578e821cadSTrond Myklebust 958d6d6dc7cSFred Isaman static void 959d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 960e468bae9STrond Myklebust { 96111fb9989SMel Gorman dec_node_page_state(page, NR_UNSTABLE_NFS); 96293f78d88STejun Heo dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, 96393f78d88STejun Heo WB_RECLAIMABLE); 964e468bae9STrond Myklebust } 965d6d6dc7cSFred Isaman 966411a99adSWeston Andros Adamson /* Called holding inode (/cinfo) lock */ 9678dd37758STrond Myklebust static void 968d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 969d6d6dc7cSFred Isaman { 9708dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 9712b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 972ea2cf228SFred Isaman struct nfs_commit_info cinfo; 973d6d6dc7cSFred Isaman 974ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 975ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 976ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 977d6d6dc7cSFred Isaman } 9788dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 9798dd37758STrond Myklebust } 980e468bae9STrond Myklebust } 981e468bae9STrond Myklebust 982d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 9838e821cadSTrond Myklebust { 984c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 985d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 986c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 9878e821cadSTrond Myklebust } 9888e821cadSTrond Myklebust 989919e3bd9STrond Myklebust static void nfs_async_write_init(struct nfs_pgio_header *hdr) 990919e3bd9STrond Myklebust { 991919e3bd9STrond Myklebust nfs_io_completion_get(hdr->io_completion); 992919e3bd9STrond Myklebust } 993919e3bd9STrond Myklebust 994061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 9956c75dc0dSFred Isaman { 996ea2cf228SFred Isaman struct nfs_commit_info cinfo; 9976c75dc0dSFred Isaman unsigned long bytes = 0; 9986c75dc0dSFred Isaman 9996c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 10006c75dc0dSFred Isaman goto out; 1001ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 10026c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 10036c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 10046c75dc0dSFred Isaman 10056c75dc0dSFred Isaman bytes += req->wb_bytes; 10066c75dc0dSFred Isaman nfs_list_remove_request(req); 10076c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 10086c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 1009d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 10106c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 10116c75dc0dSFred Isaman goto remove_req; 10126c75dc0dSFred Isaman } 1013c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 1014f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 1015b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 1016a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 10176c75dc0dSFred Isaman goto next; 10186c75dc0dSFred Isaman } 10196c75dc0dSFred Isaman remove_req: 10206c75dc0dSFred Isaman nfs_inode_remove_request(req); 10216c75dc0dSFred Isaman next: 10221d1afcbcSTrond Myklebust nfs_unlock_request(req); 102320633f04SWeston Andros Adamson nfs_end_page_writeback(req); 10243aff4ebbSTrond Myklebust nfs_release_request(req); 10256c75dc0dSFred Isaman } 10266c75dc0dSFred Isaman out: 1027919e3bd9STrond Myklebust nfs_io_completion_put(hdr->io_completion); 10286c75dc0dSFred Isaman hdr->release(hdr); 10296c75dc0dSFred Isaman } 10306c75dc0dSFred Isaman 1031ce59515cSAnna Schumaker unsigned long 1032ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 1033fb8a1f11STrond Myklebust { 1034ea2cf228SFred Isaman return cinfo->mds->ncommit; 1035fb8a1f11STrond Myklebust } 1036fb8a1f11STrond Myklebust 1037fe238e60SDave Wysochanski /* cinfo->inode->i_lock held by caller */ 10381763da12SFred Isaman int 1039ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 1040ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 1041d6d6dc7cSFred Isaman { 1042d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 1043d6d6dc7cSFred Isaman int ret = 0; 1044d6d6dc7cSFred Isaman 1045d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 10468dd37758STrond Myklebust if (!nfs_lock_request(req)) 10478dd37758STrond Myklebust continue; 10487ad84aa9STrond Myklebust kref_get(&req->wb_kref); 1049fe238e60SDave Wysochanski if (cond_resched_lock(&cinfo->inode->i_lock)) 10503b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 1051ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 10528dd37758STrond Myklebust nfs_list_add_request(req, dst); 1053d6d6dc7cSFred Isaman ret++; 10541763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 1055d6d6dc7cSFred Isaman break; 1056d6d6dc7cSFred Isaman } 1057d6d6dc7cSFred Isaman return ret; 1058d6d6dc7cSFred Isaman } 1059d6d6dc7cSFred Isaman 10601da177e4SLinus Torvalds /* 10611da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 10621da177e4SLinus Torvalds * @inode: NFS inode to scan 1063ea2cf228SFred Isaman * @dst: mds destination list 1064ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 10651da177e4SLinus Torvalds * 10661da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 10671da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 10681da177e4SLinus Torvalds */ 10691763da12SFred Isaman int 1070ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 1071ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 10721da177e4SLinus Torvalds { 1073d6d6dc7cSFred Isaman int ret = 0; 1074fb8a1f11STrond Myklebust 1075fe238e60SDave Wysochanski spin_lock(&cinfo->inode->i_lock); 1076ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 10778dd37758STrond Myklebust const int max = INT_MAX; 1078d6d6dc7cSFred Isaman 1079ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 1080ea2cf228SFred Isaman cinfo, max); 1081ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 1082d6d6dc7cSFred Isaman } 1083fe238e60SDave Wysochanski spin_unlock(&cinfo->inode->i_lock); 1084ff778d02STrond Myklebust return ret; 10851da177e4SLinus Torvalds } 1086d6d6dc7cSFred Isaman 10871da177e4SLinus Torvalds /* 1088e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1089e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 10901da177e4SLinus Torvalds * 1091e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1092e7d39069STrond Myklebust * to disk. 10931da177e4SLinus Torvalds */ 1094e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 1095e7d39069STrond Myklebust struct page *page, 1096e7d39069STrond Myklebust unsigned int offset, 1097e7d39069STrond Myklebust unsigned int bytes) 10981da177e4SLinus Torvalds { 1099e7d39069STrond Myklebust struct nfs_page *req; 1100e7d39069STrond Myklebust unsigned int rqend; 1101e7d39069STrond Myklebust unsigned int end; 11021da177e4SLinus Torvalds int error; 1103277459d2STrond Myklebust 1104e7d39069STrond Myklebust end = offset + bytes; 1105e7d39069STrond Myklebust 1106e7d39069STrond Myklebust for (;;) { 11071403390dSTrond Myklebust if (!(PagePrivate(page) || PageSwapCache(page))) 11081403390dSTrond Myklebust return NULL; 11091403390dSTrond Myklebust spin_lock(&inode->i_lock); 111084d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 1111e7d39069STrond Myklebust if (req == NULL) 1112e7d39069STrond Myklebust goto out_unlock; 1113e7d39069STrond Myklebust 11142bfc6e56SWeston Andros Adamson /* should be handled by nfs_flush_incompatible */ 11152bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 11162bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 11172bfc6e56SWeston Andros Adamson 1118e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1119e7d39069STrond Myklebust /* 1120e7d39069STrond Myklebust * Tell the caller to flush out the request if 1121e7d39069STrond Myklebust * the offsets are non-contiguous. 1122e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1123e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1124e7d39069STrond Myklebust */ 1125e468bae9STrond Myklebust if (offset > rqend 1126e7d39069STrond Myklebust || end < req->wb_offset) 1127e7d39069STrond Myklebust goto out_flushme; 1128e7d39069STrond Myklebust 11297ad84aa9STrond Myklebust if (nfs_lock_request(req)) 1130e7d39069STrond Myklebust break; 1131e7d39069STrond Myklebust 1132e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 1133587142f8STrond Myklebust spin_unlock(&inode->i_lock); 11341da177e4SLinus Torvalds error = nfs_wait_on_request(req); 11351da177e4SLinus Torvalds nfs_release_request(req); 1136e7d39069STrond Myklebust if (error != 0) 1137e7d39069STrond Myklebust goto out_err; 11381da177e4SLinus Torvalds } 11391da177e4SLinus Torvalds 11401da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 11411da177e4SLinus Torvalds if (offset < req->wb_offset) { 11421da177e4SLinus Torvalds req->wb_offset = offset; 11431da177e4SLinus Torvalds req->wb_pgbase = offset; 11441da177e4SLinus Torvalds } 11451da177e4SLinus Torvalds if (end > rqend) 11461da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1147e7d39069STrond Myklebust else 1148e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 1149e7d39069STrond Myklebust out_unlock: 1150ca138f36SFred Isaman if (req) 11518dd37758STrond Myklebust nfs_clear_request_commit(req); 1152411a99adSWeston Andros Adamson spin_unlock(&inode->i_lock); 1153e7d39069STrond Myklebust return req; 1154e7d39069STrond Myklebust out_flushme: 1155e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 1156e7d39069STrond Myklebust nfs_release_request(req); 1157e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1158e7d39069STrond Myklebust out_err: 1159e7d39069STrond Myklebust return ERR_PTR(error); 1160e7d39069STrond Myklebust } 11611da177e4SLinus Torvalds 1162e7d39069STrond Myklebust /* 1163e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1164e7d39069STrond Myklebust * 1165e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1166e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1167e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1168e7d39069STrond Myklebust */ 1169e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1170e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1171e7d39069STrond Myklebust { 1172d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1173e7d39069STrond Myklebust struct nfs_page *req; 1174e7d39069STrond Myklebust 1175e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1176e7d39069STrond Myklebust if (req != NULL) 1177e7d39069STrond Myklebust goto out; 11782bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 1179e7d39069STrond Myklebust if (IS_ERR(req)) 1180e7d39069STrond Myklebust goto out; 1181d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1182efc91ed0STrond Myklebust out: 118361e930a9STrond Myklebust return req; 11841da177e4SLinus Torvalds } 11851da177e4SLinus Torvalds 1186e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1187e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1188e7d39069STrond Myklebust { 1189e7d39069STrond Myklebust struct nfs_page *req; 1190e7d39069STrond Myklebust 1191e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1192e7d39069STrond Myklebust if (IS_ERR(req)) 1193e7d39069STrond Myklebust return PTR_ERR(req); 1194e7d39069STrond Myklebust /* Update file length */ 1195e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1196d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1197a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11981d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1199e7d39069STrond Myklebust return 0; 1200e7d39069STrond Myklebust } 1201e7d39069STrond Myklebust 12021da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 12031da177e4SLinus Torvalds { 1204cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 12052a369153STrond Myklebust struct nfs_lock_context *l_ctx; 1206bd61e0a9SJeff Layton struct file_lock_context *flctx = file_inode(file)->i_flctx; 12071da177e4SLinus Torvalds struct nfs_page *req; 12081a54533eSTrond Myklebust int do_flush, status; 12091da177e4SLinus Torvalds /* 12101da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 12111da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 12121da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 12131da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 12141da177e4SLinus Torvalds * Also do the same if we find a request from an existing 12151da177e4SLinus Torvalds * dropped page. 12161da177e4SLinus Torvalds */ 12171a54533eSTrond Myklebust do { 121884d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 12191a54533eSTrond Myklebust if (req == NULL) 12201a54533eSTrond Myklebust return 0; 12212a369153STrond Myklebust l_ctx = req->wb_lock_context; 1222138a2935STrond Myklebust do_flush = req->wb_page != page || 1223138a2935STrond Myklebust !nfs_match_open_context(req->wb_context, ctx); 12242bfc6e56SWeston Andros Adamson /* for now, flush if more than 1 request in page_group */ 12252bfc6e56SWeston Andros Adamson do_flush |= req->wb_this_page != req; 1226bd61e0a9SJeff Layton if (l_ctx && flctx && 1227bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1228bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 1229d51fdb87SNeilBrown do_flush |= l_ctx->lockowner != current->files; 12305263e31eSJeff Layton } 12311da177e4SLinus Torvalds nfs_release_request(req); 12321a54533eSTrond Myklebust if (!do_flush) 12331a54533eSTrond Myklebust return 0; 1234d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 12351a54533eSTrond Myklebust } while (status == 0); 12361a54533eSTrond Myklebust return status; 12371da177e4SLinus Torvalds } 12381da177e4SLinus Torvalds 12391da177e4SLinus Torvalds /* 1240dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1241dc24826bSAndy Adamson * expire soon. 1242dc24826bSAndy Adamson * 1243dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1244dc24826bSAndy Adamson * 1245dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1246dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1247dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1248dc24826bSAndy Adamson */ 1249dc24826bSAndy Adamson int 1250dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1251dc24826bSAndy Adamson { 1252dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1253dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1254dc24826bSAndy Adamson 1255dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 1256dc24826bSAndy Adamson } 1257dc24826bSAndy Adamson 1258dc24826bSAndy Adamson /* 1259dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1260dc24826bSAndy Adamson */ 1261ce52914eSScott Mayhew bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) 1262dc24826bSAndy Adamson { 1263ce52914eSScott Mayhew struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1264ce52914eSScott Mayhew 1265ce52914eSScott Mayhew return rpcauth_cred_key_to_expire(auth, ctx->cred); 1266dc24826bSAndy Adamson } 1267dc24826bSAndy Adamson 1268dc24826bSAndy Adamson /* 12695d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 12705d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 12715d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 12725d47a356STrond Myklebust */ 12738d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 12745d47a356STrond Myklebust { 1275d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1276d529ef83SJeff Layton 12778d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 12788d197a56STrond Myklebust goto out; 127918dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1280d529ef83SJeff Layton return false; 12814db72b40SJeff Layton smp_rmb(); 1282d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 12838d197a56STrond Myklebust return false; 12848d197a56STrond Myklebust out: 128518dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 128618dd78c4SScott Mayhew return false; 12878d197a56STrond Myklebust return PageUptodate(page) != 0; 12885d47a356STrond Myklebust } 12895d47a356STrond Myklebust 12905263e31eSJeff Layton static bool 12915263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 12925263e31eSJeff Layton { 12935263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 12945263e31eSJeff Layton fl->fl_type == F_WRLCK; 12955263e31eSJeff Layton } 12965263e31eSJeff Layton 1297c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1298c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1299c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1300c7559663SScott Mayhew * inefficiencies. 1301c7559663SScott Mayhew * 1302263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1303263b4509SScott Mayhew * of the checks. 1304c7559663SScott Mayhew */ 1305c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1306c7559663SScott Mayhew { 13075263e31eSJeff Layton int ret; 13085263e31eSJeff Layton struct file_lock_context *flctx = inode->i_flctx; 13095263e31eSJeff Layton struct file_lock *fl; 13105263e31eSJeff Layton 1311c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1312c7559663SScott Mayhew return 0; 1313263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1314263b4509SScott Mayhew return 0; 1315c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1316c7559663SScott Mayhew return 1; 1317bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1318bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 13198fa4592aSTrond Myklebust return 1; 13205263e31eSJeff Layton 13215263e31eSJeff Layton /* Check to see if there are whole file write locks */ 13225263e31eSJeff Layton ret = 0; 13236109c850SJeff Layton spin_lock(&flctx->flc_lock); 1324bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1325bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1326bd61e0a9SJeff Layton fl_list); 1327bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 13285263e31eSJeff Layton ret = 1; 1329bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 13305263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 13315263e31eSJeff Layton fl_list); 13325263e31eSJeff Layton if (fl->fl_type == F_WRLCK) 13335263e31eSJeff Layton ret = 1; 13345263e31eSJeff Layton } 13356109c850SJeff Layton spin_unlock(&flctx->flc_lock); 13365263e31eSJeff Layton return ret; 1337c7559663SScott Mayhew } 1338c7559663SScott Mayhew 13395d47a356STrond Myklebust /* 13401da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 13411da177e4SLinus Torvalds * 13421da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 13431da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 13441da177e4SLinus Torvalds */ 13451da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 13461da177e4SLinus Torvalds unsigned int offset, unsigned int count) 13471da177e4SLinus Torvalds { 1348cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1349d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 13501da177e4SLinus Torvalds int status = 0; 13511da177e4SLinus Torvalds 135291d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 135391d5b470SChuck Lever 13546de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 13556de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 13561da177e4SLinus Torvalds 1357149a4fddSBenjamin Coddington if (!count) 1358149a4fddSBenjamin Coddington goto out; 1359149a4fddSBenjamin Coddington 1360c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 136149a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 13621da177e4SLinus Torvalds offset = 0; 13631da177e4SLinus Torvalds } 13641da177e4SLinus Torvalds 1365e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 136603fa9e84STrond Myklebust if (status < 0) 136703fa9e84STrond Myklebust nfs_set_pageerror(page); 136859b7c05fSTrond Myklebust else 136959b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 1370149a4fddSBenjamin Coddington out: 137148186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 13721da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 13731da177e4SLinus Torvalds return status; 13741da177e4SLinus Torvalds } 13751da177e4SLinus Torvalds 13763ff7576dSTrond Myklebust static int flush_task_priority(int how) 13771da177e4SLinus Torvalds { 13781da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 13791da177e4SLinus Torvalds case FLUSH_HIGHPRI: 13801da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 13811da177e4SLinus Torvalds case FLUSH_LOWPRI: 13821da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 13831da177e4SLinus Torvalds } 13841da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 13851da177e4SLinus Torvalds } 13861da177e4SLinus Torvalds 1387d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1388d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1389abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 13901ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 13911da177e4SLinus Torvalds { 13923ff7576dSTrond Myklebust int priority = flush_task_priority(how); 13931da177e4SLinus Torvalds 13941ed26f33SAnna Schumaker task_setup_data->priority = priority; 1395abde71f4STom Haynes rpc_ops->write_setup(hdr, msg); 1396d138d5d1SAndy Adamson 1397abde71f4STom Haynes nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client, 1398d45f60c6SWeston Andros Adamson &task_setup_data->rpc_client, msg, hdr); 1399275acaafSTrond Myklebust } 1400275acaafSTrond Myklebust 14016d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 14026d884e8fSFred * call this on each, which will prepare them to be retried on next 14036d884e8fSFred * writeback using standard nfs. 14046d884e8fSFred */ 14056d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 14066d884e8fSFred { 14076d884e8fSFred nfs_mark_request_dirty(req); 1408c7070113STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 14091d1afcbcSTrond Myklebust nfs_unlock_request(req); 141020633f04SWeston Andros Adamson nfs_end_page_writeback(req); 14113aff4ebbSTrond Myklebust nfs_release_request(req); 14126d884e8fSFred } 14136d884e8fSFred 1414061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 14156c75dc0dSFred Isaman { 14166c75dc0dSFred Isaman struct nfs_page *req; 14176c75dc0dSFred Isaman 14186c75dc0dSFred Isaman while (!list_empty(head)) { 14196c75dc0dSFred Isaman req = nfs_list_entry(head->next); 14206c75dc0dSFred Isaman nfs_list_remove_request(req); 14216c75dc0dSFred Isaman nfs_redirty_request(req); 14226c75dc0dSFred Isaman } 14236c75dc0dSFred Isaman } 14246c75dc0dSFred Isaman 1425dc602dd7STrond Myklebust static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) 1426dc602dd7STrond Myklebust { 1427dc602dd7STrond Myklebust nfs_async_write_error(&hdr->pages); 1428dc602dd7STrond Myklebust } 1429dc602dd7STrond Myklebust 1430061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1431919e3bd9STrond Myklebust .init_hdr = nfs_async_write_init, 1432061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1433061ae2edSFred Isaman .completion = nfs_write_completion, 1434dc602dd7STrond Myklebust .reschedule_io = nfs_async_write_reschedule_io, 1435061ae2edSFred Isaman }; 1436061ae2edSFred Isaman 143757208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1438a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1439061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 14401751c363STrond Myklebust { 1441a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 144241d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1443a20c93e3SChristoph Hellwig 1444a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1445a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1446a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1447a20c93e3SChristoph Hellwig #endif 14484a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 1449fbe77c30SBenjamin Coddington server->wsize, ioflags, GFP_NOIO); 14501751c363STrond Myklebust } 1451ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 14521751c363STrond Myklebust 1453dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1454dce81290STrond Myklebust { 1455a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1456a7d42ddbSWeston Andros Adamson 14576f29b9bbSKinglong Mee if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) 14586f29b9bbSKinglong Mee pgio->pg_ops->pg_cleanup(pgio); 14596f29b9bbSKinglong Mee 146041d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1461a7d42ddbSWeston Andros Adamson 1462a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1463a7d42ddbSWeston Andros Adamson 1464a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1465a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1466dce81290STrond Myklebust } 14671f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1468dce81290STrond Myklebust 14691da177e4SLinus Torvalds 14700b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 14710b7c0153SFred Isaman { 14720b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 14730b7c0153SFred Isaman 14740b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 14750b7c0153SFred Isaman } 14760b7c0153SFred Isaman 14771f2edbe3STrond Myklebust /* 14781f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 14791f2edbe3STrond Myklebust */ 14801f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 14811f2edbe3STrond Myklebust { 14821f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 14831f2edbe3STrond Myklebust int kill = 0; 1484788e7a89STrond Myklebust 14851f2edbe3STrond Myklebust /* suid always must be killed */ 14861f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 14871f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 14881f2edbe3STrond Myklebust 14891f2edbe3STrond Myklebust /* 14901f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 14911f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 14921f2edbe3STrond Myklebust */ 14931f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 14941f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 14951f2edbe3STrond Myklebust 14961f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 14971f2edbe3STrond Myklebust return kill; 14981f2edbe3STrond Myklebust 14991f2edbe3STrond Myklebust return 0; 15001f2edbe3STrond Myklebust } 1501788e7a89STrond Myklebust 1502a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1503a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1504a08a8cd3STrond Myklebust { 1505a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1506a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 15072b83d3deSTrond Myklebust u64 size = argp->offset + resp->count; 1508a08a8cd3STrond Myklebust 1509a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 15102b83d3deSTrond Myklebust fattr->size = size; 15112b83d3deSTrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { 15122b83d3deSTrond Myklebust fattr->valid &= ~NFS_ATTR_FATTR_SIZE; 1513a08a8cd3STrond Myklebust return; 15142b83d3deSTrond Myklebust } 15152b83d3deSTrond Myklebust if (size != fattr->size) 1516a08a8cd3STrond Myklebust return; 1517a08a8cd3STrond Myklebust /* Set attribute barrier */ 1518a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 15192b83d3deSTrond Myklebust /* ...and update size */ 15202b83d3deSTrond Myklebust fattr->valid |= NFS_ATTR_FATTR_SIZE; 1521a08a8cd3STrond Myklebust } 1522a08a8cd3STrond Myklebust 1523a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1524a08a8cd3STrond Myklebust { 15252b83d3deSTrond Myklebust struct nfs_fattr *fattr = &hdr->fattr; 1526a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1527a08a8cd3STrond Myklebust 1528a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1529a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1530a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1531a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1532a08a8cd3STrond Myklebust } 1533a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1534a08a8cd3STrond Myklebust 15351da177e4SLinus Torvalds /* 15361da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 15371da177e4SLinus Torvalds */ 1538d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1539d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 15400eecb214SAnna Schumaker struct inode *inode) 15411da177e4SLinus Torvalds { 1542788e7a89STrond Myklebust int status; 15431da177e4SLinus Torvalds 1544f551e44fSChuck Lever /* 1545f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1546f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1547f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1548f551e44fSChuck Lever * another writer had changed the file, but some applications 1549f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1550f551e44fSChuck Lever */ 1551d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1552788e7a89STrond Myklebust if (status != 0) 15530eecb214SAnna Schumaker return status; 1554d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 155591d5b470SChuck Lever 1556d45f60c6SWeston Andros Adamson if (hdr->res.verf->committed < hdr->args.stable && 1557d45f60c6SWeston Andros Adamson task->tk_status >= 0) { 15581da177e4SLinus Torvalds /* We tried a write call, but the server did not 15591da177e4SLinus Torvalds * commit data to stable storage even though we 15601da177e4SLinus Torvalds * requested it. 15611da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 15621da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 15631da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 15641da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 15651da177e4SLinus Torvalds */ 15661da177e4SLinus Torvalds static unsigned long complain; 15671da177e4SLinus Torvalds 1568a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 15691da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 15701da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 15711da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1572cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1573d45f60c6SWeston Andros Adamson hdr->res.verf->committed, hdr->args.stable); 15741da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 15751da177e4SLinus Torvalds } 15761da177e4SLinus Torvalds } 15771f2edbe3STrond Myklebust 15781f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 15791f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 15801f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 15810eecb214SAnna Schumaker return 0; 15820eecb214SAnna Schumaker } 15830eecb214SAnna Schumaker 15840eecb214SAnna Schumaker /* 15850eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 15860eecb214SAnna Schumaker */ 1587d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1588d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 15890eecb214SAnna Schumaker { 1590d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1591d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 15921f2edbe3STrond Myklebust 15931f2edbe3STrond Myklebust if (resp->count < argp->count) { 15941da177e4SLinus Torvalds static unsigned long complain; 15951da177e4SLinus Torvalds 15966c75dc0dSFred Isaman /* This a short write! */ 1597d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 159891d5b470SChuck Lever 15991da177e4SLinus Torvalds /* Has the server at least made some progress? */ 16006c75dc0dSFred Isaman if (resp->count == 0) { 16016c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 16026c75dc0dSFred Isaman printk(KERN_WARNING 16036c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 16046c75dc0dSFred Isaman argp->count); 16056c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 16066c75dc0dSFred Isaman } 1607d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 16086c75dc0dSFred Isaman task->tk_status = -EIO; 16096c75dc0dSFred Isaman return; 16106c75dc0dSFred Isaman } 1611f8417b48SKinglong Mee 1612f8417b48SKinglong Mee /* For non rpc-based layout drivers, retry-through-MDS */ 1613f8417b48SKinglong Mee if (!task->tk_ops) { 1614f8417b48SKinglong Mee hdr->pnfs_error = -EAGAIN; 1615f8417b48SKinglong Mee return; 1616f8417b48SKinglong Mee } 1617f8417b48SKinglong Mee 16181da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 16191da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 16201da177e4SLinus Torvalds /* Resend from where the server left off */ 1621d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 16221da177e4SLinus Torvalds argp->offset += resp->count; 16231da177e4SLinus Torvalds argp->pgbase += resp->count; 16241da177e4SLinus Torvalds argp->count -= resp->count; 16251da177e4SLinus Torvalds } else { 16261da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 16271da177e4SLinus Torvalds * headaches in the case of a server crash. 16281da177e4SLinus Torvalds */ 16291da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 16301da177e4SLinus Torvalds } 1631d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 16321da177e4SLinus Torvalds } 16331da177e4SLinus Torvalds } 16341da177e4SLinus Torvalds 1635af7cf057STrond Myklebust static int wait_on_commit(struct nfs_mds_commit_info *cinfo) 163671d0a611STrond Myklebust { 1637af7cf057STrond Myklebust return wait_on_atomic_t(&cinfo->rpcs_out, 1638af7cf057STrond Myklebust nfs_wait_atomic_killable, TASK_KILLABLE); 1639af7cf057STrond Myklebust } 1640af7cf057STrond Myklebust 1641af7cf057STrond Myklebust static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) 1642af7cf057STrond Myklebust { 1643af7cf057STrond Myklebust atomic_inc(&cinfo->rpcs_out); 1644af7cf057STrond Myklebust } 1645af7cf057STrond Myklebust 1646af7cf057STrond Myklebust static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) 1647af7cf057STrond Myklebust { 1648af7cf057STrond Myklebust if (atomic_dec_and_test(&cinfo->rpcs_out)) 1649af7cf057STrond Myklebust wake_up_atomic_t(&cinfo->rpcs_out); 165071d0a611STrond Myklebust } 165171d0a611STrond Myklebust 16520b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 16531da177e4SLinus Torvalds { 16540b7c0153SFred Isaman put_nfs_open_context(data->context); 16550b7c0153SFred Isaman nfs_commit_free(data); 16561da177e4SLinus Torvalds } 1657e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 16581da177e4SLinus Torvalds 16590b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1660c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 16619ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 16629f0ec176SAndy Adamson int how, int flags) 16631da177e4SLinus Torvalds { 166407737691STrond Myklebust struct rpc_task *task; 16659ace33cdSFred Isaman int priority = flush_task_priority(how); 1666bdc7f021STrond Myklebust struct rpc_message msg = { 1667bdc7f021STrond Myklebust .rpc_argp = &data->args, 1668bdc7f021STrond Myklebust .rpc_resp = &data->res, 16699ace33cdSFred Isaman .rpc_cred = data->cred, 1670bdc7f021STrond Myklebust }; 167184115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 167207737691STrond Myklebust .task = &data->task, 16739ace33cdSFred Isaman .rpc_client = clnt, 1674bdc7f021STrond Myklebust .rpc_message = &msg, 16759ace33cdSFred Isaman .callback_ops = call_ops, 167684115e1cSTrond Myklebust .callback_data = data, 1677101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 16789f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 16793ff7576dSTrond Myklebust .priority = priority, 168084115e1cSTrond Myklebust }; 1681788e7a89STrond Myklebust /* Set up the initial task struct. */ 1682c36aae9aSPeng Tao nfs_ops->commit_setup(data, &msg); 16831da177e4SLinus Torvalds 1684b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1685bdc7f021STrond Myklebust 16868c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 16878c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 16888c21c62cSWeston Andros Adamson 168907737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1690dbae4c73STrond Myklebust if (IS_ERR(task)) 1691dbae4c73STrond Myklebust return PTR_ERR(task); 1692d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1693d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 169407737691STrond Myklebust rpc_put_task(task); 1695dbae4c73STrond Myklebust return 0; 16961da177e4SLinus Torvalds } 1697e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 16981da177e4SLinus Torvalds 1699378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1700378520b8SPeng Tao { 1701378520b8SPeng Tao loff_t lwb = 0; 1702378520b8SPeng Tao struct nfs_page *req; 1703378520b8SPeng Tao 1704378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1705378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1706378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1707378520b8SPeng Tao 1708378520b8SPeng Tao return lwb; 1709378520b8SPeng Tao } 1710378520b8SPeng Tao 17111da177e4SLinus Torvalds /* 17129ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 17139ace33cdSFred Isaman */ 17140b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1715988b6dceSFred Isaman struct list_head *head, 1716f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1717f453a54aSFred Isaman struct nfs_commit_info *cinfo) 17189ace33cdSFred Isaman { 17199ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 17202b0143b5SDavid Howells struct inode *inode = d_inode(first->wb_context->dentry); 17219ace33cdSFred Isaman 17229ace33cdSFred Isaman /* Set up the RPC argument and reply structs 17239ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 17249ace33cdSFred Isaman 17259ace33cdSFred Isaman list_splice_init(head, &data->pages); 17269ace33cdSFred Isaman 17279ace33cdSFred Isaman data->inode = inode; 17289ace33cdSFred Isaman data->cred = first->wb_context->cred; 1729988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1730378520b8SPeng Tao /* only set lwb for pnfs commit */ 1731378520b8SPeng Tao if (lseg) 1732378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 17339ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1734f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1735b359f9d0SFred Isaman data->dreq = cinfo->dreq; 17369ace33cdSFred Isaman 17379ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 17389ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 17399ace33cdSFred Isaman data->args.offset = 0; 17409ace33cdSFred Isaman data->args.count = 0; 17410b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 17429ace33cdSFred Isaman data->res.fattr = &data->fattr; 17439ace33cdSFred Isaman data->res.verf = &data->verf; 17449ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 17459ace33cdSFred Isaman } 1746e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 17479ace33cdSFred Isaman 1748e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1749ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1750b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1751b57ff130SWeston Andros Adamson u32 ds_commit_idx) 175264bfeb49SFred Isaman { 175364bfeb49SFred Isaman struct nfs_page *req; 175464bfeb49SFred Isaman 175564bfeb49SFred Isaman while (!list_empty(page_list)) { 175664bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 175764bfeb49SFred Isaman nfs_list_remove_request(req); 1758b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 1759487b9b8aSTom Haynes if (!cinfo->dreq) 1760487b9b8aSTom Haynes nfs_clear_page_commit(req->wb_page); 17611d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 176264bfeb49SFred Isaman } 176364bfeb49SFred Isaman } 1764e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 176564bfeb49SFred Isaman 1766b20135d0STrond Myklebust static void 1767b20135d0STrond Myklebust nfs_commit_resched_write(struct nfs_commit_info *cinfo, 1768b20135d0STrond Myklebust struct nfs_page *req) 1769b20135d0STrond Myklebust { 1770b20135d0STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 1771b20135d0STrond Myklebust } 1772b20135d0STrond Myklebust 17739ace33cdSFred Isaman /* 17741da177e4SLinus Torvalds * Commit dirty pages 17751da177e4SLinus Torvalds */ 17761da177e4SLinus Torvalds static int 1777ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1778ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 17791da177e4SLinus Torvalds { 17800b7c0153SFred Isaman struct nfs_commit_data *data; 17811da177e4SLinus Torvalds 1782ade8febdSWeston Andros Adamson /* another commit raced with us */ 1783ade8febdSWeston Andros Adamson if (list_empty(head)) 1784ade8febdSWeston Andros Adamson return 0; 1785ade8febdSWeston Andros Adamson 1786518662e0SNeilBrown data = nfs_commitdata_alloc(true); 17871da177e4SLinus Torvalds 17881da177e4SLinus Torvalds /* Set up the argument struct */ 1789f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1790f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 1791c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 1792c36aae9aSPeng Tao data->mds_ops, how, 0); 17931da177e4SLinus Torvalds } 17941da177e4SLinus Torvalds 17951da177e4SLinus Torvalds /* 17961da177e4SLinus Torvalds * COMMIT call returned 17971da177e4SLinus Torvalds */ 1798788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 17991da177e4SLinus Torvalds { 18000b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18011da177e4SLinus Torvalds 1802a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 18031da177e4SLinus Torvalds task->tk_pid, task->tk_status); 18041da177e4SLinus Torvalds 1805788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1806c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1807c9d8f89dSTrond Myklebust } 1808c9d8f89dSTrond Myklebust 1809f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1810c9d8f89dSTrond Myklebust { 1811c9d8f89dSTrond Myklebust struct nfs_page *req; 1812c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1813f453a54aSFred Isaman struct nfs_commit_info cinfo; 1814353db796SNeilBrown struct nfs_server *nfss; 1815788e7a89STrond Myklebust 18161da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 18171da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 18181da177e4SLinus Torvalds nfs_list_remove_request(req); 181967911c8fSAnna Schumaker if (req->wb_page) 1820d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 18211da177e4SLinus Torvalds 18221e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 18233d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 18242b0143b5SDavid Howells (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)), 18251da177e4SLinus Torvalds req->wb_bytes, 18261da177e4SLinus Torvalds (long long)req_offset(req)); 1827c9d8f89dSTrond Myklebust if (status < 0) { 1828c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 182938a33101SKinglong Mee if (req->wb_page) 18301da177e4SLinus Torvalds nfs_inode_remove_request(req); 1831ddeaa637SJoe Perches dprintk_cont(", error = %d\n", status); 18321da177e4SLinus Torvalds goto next; 18331da177e4SLinus Torvalds } 18341da177e4SLinus Torvalds 18351da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 18361da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 18378fc3c386STrond Myklebust if (!nfs_write_verifier_cmp(&req->wb_verf, &data->verf.verifier)) { 18381da177e4SLinus Torvalds /* We have a match */ 183938a33101SKinglong Mee if (req->wb_page) 18401da177e4SLinus Torvalds nfs_inode_remove_request(req); 1841ddeaa637SJoe Perches dprintk_cont(" OK\n"); 18421da177e4SLinus Torvalds goto next; 18431da177e4SLinus Torvalds } 18441da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 1845ddeaa637SJoe Perches dprintk_cont(" mismatch\n"); 18466d884e8fSFred nfs_mark_request_dirty(req); 184705990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 18481da177e4SLinus Torvalds next: 18491d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18501da177e4SLinus Torvalds } 1851353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1852353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 18530db10944SJan Kara clear_bdi_congested(inode_to_bdi(data->inode), BLK_RW_ASYNC); 1854353db796SNeilBrown 1855f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1856af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18575917ce84SFred Isaman } 18585917ce84SFred Isaman 18595917ce84SFred Isaman static void nfs_commit_release(void *calldata) 18605917ce84SFred Isaman { 18610b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18625917ce84SFred Isaman 1863f453a54aSFred Isaman data->completion_ops->completion(data); 1864c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 18651da177e4SLinus Torvalds } 1866788e7a89STrond Myklebust 1867788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 18680b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1869788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1870788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1871788e7a89STrond Myklebust }; 18721da177e4SLinus Torvalds 1873f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1874f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1875b20135d0STrond Myklebust .resched_write = nfs_commit_resched_write, 1876f453a54aSFred Isaman }; 1877f453a54aSFred Isaman 18781763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1879ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 188084c53ab5SFred Isaman { 188184c53ab5SFred Isaman int status; 188284c53ab5SFred Isaman 1883ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 188484c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1885ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 188684c53ab5SFred Isaman return status; 188784c53ab5SFred Isaman } 188884c53ab5SFred Isaman 1889b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 18901da177e4SLinus Torvalds { 18911da177e4SLinus Torvalds LIST_HEAD(head); 1892ea2cf228SFred Isaman struct nfs_commit_info cinfo; 189371d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1894af7cf057STrond Myklebust int error = 0; 1895b8413f98STrond Myklebust int res; 18961da177e4SLinus Torvalds 1897ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1898af7cf057STrond Myklebust nfs_commit_begin(cinfo.mds); 1899ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 1900af7cf057STrond Myklebust if (res) 1901ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 1902af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 19031da177e4SLinus Torvalds if (error < 0) 1904af7cf057STrond Myklebust goto out_error; 1905b8413f98STrond Myklebust if (!may_wait) 1906b8413f98STrond Myklebust goto out_mark_dirty; 1907af7cf057STrond Myklebust error = wait_on_commit(cinfo.mds); 1908b8413f98STrond Myklebust if (error < 0) 1909b8413f98STrond Myklebust return error; 1910c5efa5fcSTrond Myklebust return res; 1911af7cf057STrond Myklebust out_error: 1912af7cf057STrond Myklebust res = error; 1913c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1914c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1915c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1916c5efa5fcSTrond Myklebust * that the data is on the disk. 1917c5efa5fcSTrond Myklebust */ 1918c5efa5fcSTrond Myklebust out_mark_dirty: 1919c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 19201da177e4SLinus Torvalds return res; 19211da177e4SLinus Torvalds } 1922b20135d0STrond Myklebust EXPORT_SYMBOL_GPL(nfs_commit_inode); 19238fc795f7STrond Myklebust 1924ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 19258fc795f7STrond Myklebust { 1926420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1927420e3646STrond Myklebust int flags = FLUSH_SYNC; 1928420e3646STrond Myklebust int ret = 0; 19298fc795f7STrond Myklebust 19303236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1931ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 19323236c3e1SJeff Layton return ret; 19333236c3e1SJeff Layton 1934a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1935a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1936a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1937420e3646STrond Myklebust */ 19381a4edf0fSTrond Myklebust if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)) 1939420e3646STrond Myklebust goto out_mark_dirty; 1940420e3646STrond Myklebust 1941a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1942420e3646STrond Myklebust flags = 0; 1943a00dd6c0SJeff Layton } 1944a00dd6c0SJeff Layton 1945420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1946420e3646STrond Myklebust if (ret >= 0) { 1947420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1948420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1949420e3646STrond Myklebust wbc->nr_to_write -= ret; 1950420e3646STrond Myklebust else 1951420e3646STrond Myklebust wbc->nr_to_write = 0; 1952420e3646STrond Myklebust } 19538fc795f7STrond Myklebust return 0; 1954420e3646STrond Myklebust } 1955420e3646STrond Myklebust out_mark_dirty: 19568fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 19578fc795f7STrond Myklebust return ret; 19588fc795f7STrond Myklebust } 195989d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1960863a3c6cSAndy Adamson 1961acdc53b2STrond Myklebust /* 1962837bb1d7STrond Myklebust * Wrapper for filemap_write_and_wait_range() 1963837bb1d7STrond Myklebust * 1964837bb1d7STrond Myklebust * Needed for pNFS in order to ensure data becomes visible to the 1965837bb1d7STrond Myklebust * client. 1966837bb1d7STrond Myklebust */ 1967837bb1d7STrond Myklebust int nfs_filemap_write_and_wait_range(struct address_space *mapping, 1968837bb1d7STrond Myklebust loff_t lstart, loff_t lend) 1969837bb1d7STrond Myklebust { 1970837bb1d7STrond Myklebust int ret; 1971837bb1d7STrond Myklebust 1972837bb1d7STrond Myklebust ret = filemap_write_and_wait_range(mapping, lstart, lend); 1973837bb1d7STrond Myklebust if (ret == 0) 1974837bb1d7STrond Myklebust ret = pnfs_sync_inode(mapping->host, true); 1975837bb1d7STrond Myklebust return ret; 1976837bb1d7STrond Myklebust } 1977837bb1d7STrond Myklebust EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); 1978837bb1d7STrond Myklebust 1979837bb1d7STrond Myklebust /* 1980acdc53b2STrond Myklebust * flush the inode to disk. 1981acdc53b2STrond Myklebust */ 1982acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 198334901f70STrond Myklebust { 1984f4ce1299STrond Myklebust int ret; 198534901f70STrond Myklebust 1986f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1987f4ce1299STrond Myklebust 19885bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 19896b196875SChuck Lever if (ret) 19906b196875SChuck Lever goto out; 19915bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 19926b196875SChuck Lever if (ret < 0) 19936b196875SChuck Lever goto out; 19945bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 19956b196875SChuck Lever ret = 0; 1996f4ce1299STrond Myklebust 19976b196875SChuck Lever out: 1998f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1999f4ce1299STrond Myklebust return ret; 20001c75950bSTrond Myklebust } 2001ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 20021c75950bSTrond Myklebust 20031b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 20041b3b4a1aSTrond Myklebust { 20051b3b4a1aSTrond Myklebust struct nfs_page *req; 20061b3b4a1aSTrond Myklebust int ret = 0; 20071b3b4a1aSTrond Myklebust 2008ba8b06e6STrond Myklebust wait_on_page_writeback(page); 20093e217045SWeston Andros Adamson 20103e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 20113e217045SWeston Andros Adamson * request */ 20126d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 20133e217045SWeston Andros Adamson 20143e217045SWeston Andros Adamson if (IS_ERR(req)) { 20153e217045SWeston Andros Adamson ret = PTR_ERR(req); 20163e217045SWeston Andros Adamson } else if (req) { 20173e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 20183e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 20193e217045SWeston Andros Adamson * request from the inode / page_private pointer and 20203e217045SWeston Andros Adamson * release it */ 20211b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 20221d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 20231b3b4a1aSTrond Myklebust } 20243e217045SWeston Andros Adamson 20251b3b4a1aSTrond Myklebust return ret; 20261b3b4a1aSTrond Myklebust } 20271b3b4a1aSTrond Myklebust 20281c75950bSTrond Myklebust /* 20291c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 20301c75950bSTrond Myklebust */ 2031c373fff7STrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 20321c75950bSTrond Myklebust { 203329418aa4SMel Gorman loff_t range_start = page_file_offset(page); 203409cbfeafSKirill A. Shutemov loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); 20357f2f12d9STrond Myklebust struct writeback_control wbc = { 20367f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 20377f2f12d9STrond Myklebust .nr_to_write = 0, 20387f2f12d9STrond Myklebust .range_start = range_start, 20397f2f12d9STrond Myklebust .range_end = range_end, 20407f2f12d9STrond Myklebust }; 20417f2f12d9STrond Myklebust int ret; 20427f2f12d9STrond Myklebust 2043f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 2044f4ce1299STrond Myklebust 20450522f6adSTrond Myklebust for (;;) { 2046ba8b06e6STrond Myklebust wait_on_page_writeback(page); 20477f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 2048c373fff7STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 20497f2f12d9STrond Myklebust if (ret < 0) 20507f2f12d9STrond Myklebust goto out_error; 20510522f6adSTrond Myklebust continue; 20527f2f12d9STrond Myklebust } 2053f4ce1299STrond Myklebust ret = 0; 20540522f6adSTrond Myklebust if (!PagePrivate(page)) 20550522f6adSTrond Myklebust break; 20560522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20577f2f12d9STrond Myklebust if (ret < 0) 20587f2f12d9STrond Myklebust goto out_error; 20597f2f12d9STrond Myklebust } 20607f2f12d9STrond Myklebust out_error: 2061f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 20627f2f12d9STrond Myklebust return ret; 20631c75950bSTrond Myklebust } 20641c75950bSTrond Myklebust 2065074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 2066074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 2067a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 2068074cc1deSTrond Myklebust { 20692da95652SJeff Layton /* 20702da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 20712da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 20722da95652SJeff Layton * 20732da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 20742da95652SJeff Layton * that we can safely release the inode reference while holding 20752da95652SJeff Layton * the page lock. 20762da95652SJeff Layton */ 20772da95652SJeff Layton if (PagePrivate(page)) 20782da95652SJeff Layton return -EBUSY; 2079074cc1deSTrond Myklebust 20808c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 20818c209ce7SDavid Howells return -EBUSY; 2082074cc1deSTrond Myklebust 2083a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 2084074cc1deSTrond Myklebust } 2085074cc1deSTrond Myklebust #endif 2086074cc1deSTrond Myklebust 2087f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 20881da177e4SLinus Torvalds { 20891da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 20901e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 20911da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 209220c2df83SPaul Mundt NULL); 20931da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 20941da177e4SLinus Torvalds return -ENOMEM; 20951da177e4SLinus Torvalds 209693d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 20971da177e4SLinus Torvalds nfs_wdata_cachep); 20981da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 20993dd4765fSJeff Layton goto out_destroy_write_cache; 21001da177e4SLinus Torvalds 21010b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 21020b7c0153SFred Isaman sizeof(struct nfs_commit_data), 21030b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 21040b7c0153SFred Isaman NULL); 21050b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 21063dd4765fSJeff Layton goto out_destroy_write_mempool; 21070b7c0153SFred Isaman 210893d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 21094c100210SYanchuan Nian nfs_cdata_cachep); 21101da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 21113dd4765fSJeff Layton goto out_destroy_commit_cache; 21121da177e4SLinus Torvalds 211389a09141SPeter Zijlstra /* 211489a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 211589a09141SPeter Zijlstra * 211689a09141SPeter Zijlstra * 64MB: 8192k 211789a09141SPeter Zijlstra * 128MB: 11585k 211889a09141SPeter Zijlstra * 256MB: 16384k 211989a09141SPeter Zijlstra * 512MB: 23170k 212089a09141SPeter Zijlstra * 1GB: 32768k 212189a09141SPeter Zijlstra * 2GB: 46340k 212289a09141SPeter Zijlstra * 4GB: 65536k 212389a09141SPeter Zijlstra * 8GB: 92681k 212489a09141SPeter Zijlstra * 16GB: 131072k 212589a09141SPeter Zijlstra * 212689a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 212789a09141SPeter Zijlstra * Limit the default to 256M 212889a09141SPeter Zijlstra */ 212989a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 213089a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 213189a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 213289a09141SPeter Zijlstra 21331da177e4SLinus Torvalds return 0; 21343dd4765fSJeff Layton 21353dd4765fSJeff Layton out_destroy_commit_cache: 21363dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21373dd4765fSJeff Layton out_destroy_write_mempool: 21383dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 21393dd4765fSJeff Layton out_destroy_write_cache: 21403dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 21413dd4765fSJeff Layton return -ENOMEM; 21421da177e4SLinus Torvalds } 21431da177e4SLinus Torvalds 2144266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 21451da177e4SLinus Torvalds { 21461da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 21473dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21481da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 21491a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 21501da177e4SLinus Torvalds } 21511da177e4SLinus Torvalds 21524a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 21534a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 21544a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 21550eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 21560eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 21571ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 21584a0de55cSAnna Schumaker }; 2159