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 193587142f8STrond Myklebust spin_lock(&inode->i_lock); 19484d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 195587142f8STrond Myklebust spin_unlock(&inode->i_lock); 196277459d2STrond Myklebust return req; 197277459d2STrond Myklebust } 198277459d2STrond Myklebust 1991da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 2001da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 2011da177e4SLinus Torvalds { 202d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 203a3d01454STrond Myklebust loff_t end, i_size; 204a3d01454STrond Myklebust pgoff_t end_index; 2051da177e4SLinus Torvalds 206a3d01454STrond Myklebust spin_lock(&inode->i_lock); 207a3d01454STrond Myklebust i_size = i_size_read(inode); 20809cbfeafSKirill A. Shutemov end_index = (i_size - 1) >> PAGE_SHIFT; 2098cd79788SHuang Ying if (i_size > 0 && page_index(page) < end_index) 210a3d01454STrond Myklebust goto out; 211d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 2121da177e4SLinus Torvalds if (i_size >= end) 213a3d01454STrond Myklebust goto out; 2141da177e4SLinus Torvalds i_size_write(inode, end); 215a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 216a3d01454STrond Myklebust out: 217a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 2181da177e4SLinus Torvalds } 2191da177e4SLinus Torvalds 220a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 221a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 222a301b777STrond Myklebust { 223d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 224a301b777STrond Myklebust } 225a301b777STrond Myklebust 226d72ddcbaSWeston Andros Adamson /* 227d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 228d72ddcbaSWeston Andros Adamson * @head - head request of page group 229d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 230d72ddcbaSWeston Andros Adamson * 231d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 232d72ddcbaSWeston Andros Adamson * page offset @page_offset. 233d72ddcbaSWeston Andros Adamson * 234d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 235d72ddcbaSWeston Andros Adamson * match is found. 236d72ddcbaSWeston Andros Adamson * 237d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 238d72ddcbaSWeston Andros Adamson */ 239d72ddcbaSWeston Andros Adamson static struct nfs_page * 240d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 241d72ddcbaSWeston Andros Adamson { 242d72ddcbaSWeston Andros Adamson struct nfs_page *req; 243d72ddcbaSWeston Andros Adamson 244d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(head != head->wb_head); 245d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags)); 246d72ddcbaSWeston Andros Adamson 247d72ddcbaSWeston Andros Adamson req = head; 248d72ddcbaSWeston Andros Adamson do { 249d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 250d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 251d72ddcbaSWeston Andros Adamson return req; 252d72ddcbaSWeston Andros Adamson 253d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 254d72ddcbaSWeston Andros Adamson } while (req != head); 255d72ddcbaSWeston Andros Adamson 256d72ddcbaSWeston Andros Adamson return NULL; 257d72ddcbaSWeston Andros Adamson } 258d72ddcbaSWeston Andros Adamson 259d72ddcbaSWeston Andros Adamson /* 260d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 261d72ddcbaSWeston Andros Adamson * @head - head request of page group 262d72ddcbaSWeston Andros Adamson * 263d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 264d72ddcbaSWeston Andros Adamson * returns false otherwise 265d72ddcbaSWeston Andros Adamson */ 266d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 267d72ddcbaSWeston Andros Adamson { 268d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 269d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 270d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 271d72ddcbaSWeston Andros Adamson 272fd2f3a06SWeston Andros Adamson nfs_page_group_lock(req, false); 273d72ddcbaSWeston Andros Adamson 274d72ddcbaSWeston Andros Adamson do { 275d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 276d72ddcbaSWeston Andros Adamson if (tmp) { 277d72ddcbaSWeston Andros Adamson /* no way this should happen */ 278d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(tmp->wb_pgbase != pos); 279d72ddcbaSWeston Andros Adamson pos += tmp->wb_bytes - (pos - tmp->wb_pgbase); 280d72ddcbaSWeston Andros Adamson } 281d72ddcbaSWeston Andros Adamson } while (tmp && pos < len); 282d72ddcbaSWeston Andros Adamson 283d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 284d72ddcbaSWeston Andros Adamson WARN_ON_ONCE(pos > len); 285d72ddcbaSWeston Andros Adamson return pos == len; 286d72ddcbaSWeston Andros Adamson } 287d72ddcbaSWeston Andros Adamson 2881da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 2891da177e4SLinus Torvalds * covers the full page. 2901da177e4SLinus Torvalds */ 291d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 2921da177e4SLinus Torvalds { 293d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 2941da177e4SLinus Torvalds return; 295d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 2961da177e4SLinus Torvalds return; 297d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 2981da177e4SLinus Torvalds } 2991da177e4SLinus Torvalds 3001da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 3011da177e4SLinus Torvalds { 302e87b4c7aSNeilBrown int ret = 0; 303cca588d6STrond Myklebust 304e87b4c7aSNeilBrown if (wbc->sync_mode == WB_SYNC_ALL) 305e87b4c7aSNeilBrown ret = FLUSH_COND_STABLE; 306e87b4c7aSNeilBrown return ret; 3071da177e4SLinus Torvalds } 3081da177e4SLinus Torvalds 3091da177e4SLinus Torvalds /* 31089a09141SPeter Zijlstra * NFS congestion control 31189a09141SPeter Zijlstra */ 31289a09141SPeter Zijlstra 31389a09141SPeter Zijlstra int nfs_congestion_kb; 31489a09141SPeter Zijlstra 31589a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 31689a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 31789a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 31889a09141SPeter Zijlstra 319deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 32089a09141SPeter Zijlstra { 3210db10944SJan Kara struct inode *inode = page_file_mapping(page)->host; 3220db10944SJan Kara struct nfs_server *nfss = NFS_SERVER(inode); 3235a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 3245a6d41b3STrond Myklebust 325deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 32689a09141SPeter Zijlstra 327277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 3280db10944SJan Kara NFS_CONGESTION_ON_THRESH) 3290db10944SJan Kara set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 33089a09141SPeter Zijlstra } 33189a09141SPeter Zijlstra 33220633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 33389a09141SPeter Zijlstra { 33420633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 33589a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 33689a09141SPeter Zijlstra 33720633f04SWeston Andros Adamson if (!nfs_page_group_sync_on_bit(req, PG_WB_END)) 33820633f04SWeston Andros Adamson return; 33920633f04SWeston Andros Adamson 34020633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 341c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 3420db10944SJan Kara clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 34389a09141SPeter Zijlstra } 34489a09141SPeter Zijlstra 345d4581383SWeston Andros Adamson 346d4581383SWeston Andros Adamson /* nfs_page_group_clear_bits 347d4581383SWeston Andros Adamson * @req - an nfs request 348d4581383SWeston Andros Adamson * clears all page group related bits from @req 349d4581383SWeston Andros Adamson */ 350d4581383SWeston Andros Adamson static void 351d4581383SWeston Andros Adamson nfs_page_group_clear_bits(struct nfs_page *req) 352e261f51fSTrond Myklebust { 353d4581383SWeston Andros Adamson clear_bit(PG_TEARDOWN, &req->wb_flags); 354d4581383SWeston Andros Adamson clear_bit(PG_UNLOCKPAGE, &req->wb_flags); 355d4581383SWeston Andros Adamson clear_bit(PG_UPTODATE, &req->wb_flags); 356d4581383SWeston Andros Adamson clear_bit(PG_WB_END, &req->wb_flags); 357d4581383SWeston Andros Adamson clear_bit(PG_REMOVE, &req->wb_flags); 358d4581383SWeston Andros Adamson } 359d4581383SWeston Andros Adamson 360d4581383SWeston Andros Adamson 361d4581383SWeston Andros Adamson /* 362d4581383SWeston Andros Adamson * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req 363d4581383SWeston Andros Adamson * 364d4581383SWeston Andros Adamson * this is a helper function for nfs_lock_and_join_requests 365d4581383SWeston Andros Adamson * 366d4581383SWeston Andros Adamson * @inode - inode associated with request page group, must be holding inode lock 367d4581383SWeston Andros Adamson * @head - head request of page group, must be holding head lock 368d4581383SWeston Andros Adamson * @req - request that couldn't lock and needs to wait on the req bit lock 369d4581383SWeston Andros Adamson * 370d4581383SWeston Andros Adamson * NOTE: this must be called holding page_group bit lock and inode spin lock 371d4581383SWeston Andros Adamson * and BOTH will be released before returning. 372d4581383SWeston Andros Adamson * 373d4581383SWeston Andros Adamson * returns 0 on success, < 0 on error. 374d4581383SWeston Andros Adamson */ 375d4581383SWeston Andros Adamson static int 376d4581383SWeston Andros Adamson nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head, 377*6d17d653STrond Myklebust struct nfs_page *req) 378d4581383SWeston Andros Adamson __releases(&inode->i_lock) 379d4581383SWeston Andros Adamson { 380d4581383SWeston Andros Adamson struct nfs_page *tmp; 381e261f51fSTrond Myklebust int ret; 382e261f51fSTrond Myklebust 383d4581383SWeston Andros Adamson /* relinquish all the locks successfully grabbed this run */ 384d4581383SWeston Andros Adamson for (tmp = head ; tmp != req; tmp = tmp->wb_this_page) 385d4581383SWeston Andros Adamson nfs_unlock_request(tmp); 386d4581383SWeston Andros Adamson 387d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags)); 388d4581383SWeston Andros Adamson 389d4581383SWeston Andros Adamson /* grab a ref on the request that will be waited on */ 390d4581383SWeston Andros Adamson kref_get(&req->wb_kref); 391d4581383SWeston Andros Adamson 392d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 393587142f8STrond Myklebust spin_unlock(&inode->i_lock); 394d4581383SWeston Andros Adamson 395d4581383SWeston Andros Adamson /* release ref from nfs_page_find_head_request_locked */ 396d4581383SWeston Andros Adamson nfs_release_request(head); 397d4581383SWeston Andros Adamson 398e261f51fSTrond Myklebust ret = nfs_wait_on_request(req); 399e261f51fSTrond Myklebust nfs_release_request(req); 400d4581383SWeston Andros Adamson 401d4581383SWeston Andros Adamson return ret; 402e261f51fSTrond Myklebust } 403d4581383SWeston Andros Adamson 404d4581383SWeston Andros Adamson /* 405d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 406d4581383SWeston Andros Adamson * 407d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 408d4581383SWeston Andros Adamson * @old_head - the old head of the list 409d4581383SWeston Andros Adamson * 410d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 411d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 412d4581383SWeston Andros Adamson * with a reference held by some other context. 413d4581383SWeston Andros Adamson */ 414d4581383SWeston Andros Adamson static void 415d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 416d4581383SWeston Andros Adamson struct nfs_page *old_head) 417d4581383SWeston Andros Adamson { 418d4581383SWeston Andros Adamson while (destroy_list) { 419d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 420d4581383SWeston Andros Adamson 421d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 422d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 423d4581383SWeston Andros Adamson 424d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 425d4581383SWeston Andros Adamson 426d4581383SWeston Andros Adamson /* make sure old group is not used */ 427d4581383SWeston Andros Adamson subreq->wb_head = subreq; 428d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 429d4581383SWeston Andros Adamson 430d4581383SWeston Andros Adamson /* subreq is now totally disconnected from page group or any 431d4581383SWeston Andros Adamson * write / commit lists. last chance to wake any waiters */ 432d4581383SWeston Andros Adamson nfs_unlock_request(subreq); 433d4581383SWeston Andros Adamson 434d4581383SWeston Andros Adamson if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) { 435d4581383SWeston Andros Adamson /* release ref on old head request */ 436d4581383SWeston Andros Adamson nfs_release_request(old_head); 437d4581383SWeston Andros Adamson 438d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 439d4581383SWeston Andros Adamson 440d4581383SWeston Andros Adamson /* release the PG_INODE_REF reference */ 441d4581383SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) 442d4581383SWeston Andros Adamson nfs_release_request(subreq); 443d4581383SWeston Andros Adamson else 444d4581383SWeston Andros Adamson WARN_ON_ONCE(1); 445d4581383SWeston Andros Adamson } else { 446d4581383SWeston Andros Adamson WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags)); 447d4581383SWeston Andros Adamson /* zombie requests have already released the last 448d4581383SWeston Andros Adamson * reference and were waiting on the rest of the 449d4581383SWeston Andros Adamson * group to complete. Since it's no longer part of a 450d4581383SWeston Andros Adamson * group, simply free the request */ 451d4581383SWeston Andros Adamson nfs_page_group_clear_bits(subreq); 452d4581383SWeston Andros Adamson nfs_free_request(subreq); 453d4581383SWeston Andros Adamson } 454d4581383SWeston Andros Adamson } 455d4581383SWeston Andros Adamson } 456d4581383SWeston Andros Adamson 457d4581383SWeston Andros Adamson /* 458d4581383SWeston Andros Adamson * nfs_lock_and_join_requests - join all subreqs to the head req and return 459d4581383SWeston Andros Adamson * a locked reference, cancelling any pending 460d4581383SWeston Andros Adamson * operations for this page. 461d4581383SWeston Andros Adamson * 462d4581383SWeston Andros Adamson * @page - the page used to lookup the "page group" of nfs_page structures 463d4581383SWeston Andros Adamson * 464d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 465d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 466d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 467d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 468d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 469d4581383SWeston Andros Adamson * 470d4581383SWeston Andros Adamson * Returns a locked, referenced pointer to the head request - which after 471d4581383SWeston Andros Adamson * this call is guaranteed to be the only request associated with the page. 472d4581383SWeston Andros Adamson * Returns NULL if no requests are found for @page, or a ERR_PTR if an 473d4581383SWeston Andros Adamson * error was encountered. 474d4581383SWeston Andros Adamson */ 475d4581383SWeston Andros Adamson static struct nfs_page * 476*6d17d653STrond Myklebust nfs_lock_and_join_requests(struct page *page) 477d4581383SWeston Andros Adamson { 478d4581383SWeston Andros Adamson struct inode *inode = page_file_mapping(page)->host; 479d4581383SWeston Andros Adamson struct nfs_page *head, *subreq; 480d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 481d4581383SWeston Andros Adamson unsigned int total_bytes; 482d4581383SWeston Andros Adamson int ret; 483d4581383SWeston Andros Adamson 484d4581383SWeston Andros Adamson try_again: 485d4581383SWeston Andros Adamson total_bytes = 0; 486d4581383SWeston Andros Adamson 487d4581383SWeston Andros Adamson WARN_ON_ONCE(destroy_list); 488d4581383SWeston Andros Adamson 489d4581383SWeston Andros Adamson spin_lock(&inode->i_lock); 490d4581383SWeston Andros Adamson 491d4581383SWeston Andros Adamson /* 492d4581383SWeston Andros Adamson * A reference is taken only on the head request which acts as a 493d4581383SWeston Andros Adamson * reference to the whole page group - the group will not be destroyed 494d4581383SWeston Andros Adamson * until the head reference is released. 495d4581383SWeston Andros Adamson */ 496d4581383SWeston Andros Adamson head = nfs_page_find_head_request_locked(NFS_I(inode), page); 497d4581383SWeston Andros Adamson 498d4581383SWeston Andros Adamson if (!head) { 499587142f8STrond Myklebust spin_unlock(&inode->i_lock); 500d4581383SWeston Andros Adamson return NULL; 501d4581383SWeston Andros Adamson } 502d4581383SWeston Andros Adamson 5037c3af975SWeston Andros Adamson /* holding inode lock, so always make a non-blocking call to try the 5047c3af975SWeston Andros Adamson * page group lock */ 505fd2f3a06SWeston Andros Adamson ret = nfs_page_group_lock(head, true); 50694970014SWeston Andros Adamson if (ret < 0) { 50794970014SWeston Andros Adamson spin_unlock(&inode->i_lock); 5087c3af975SWeston Andros Adamson 5097c3af975SWeston Andros Adamson nfs_page_group_lock_wait(head); 5107c3af975SWeston Andros Adamson nfs_release_request(head); 5117c3af975SWeston Andros Adamson goto try_again; 5127c3af975SWeston Andros Adamson } 5137c3af975SWeston Andros Adamson 5147c3af975SWeston Andros Adamson /* lock each request in the page group */ 515d4581383SWeston Andros Adamson subreq = head; 516d4581383SWeston Andros Adamson do { 517d4581383SWeston Andros Adamson /* 518d4581383SWeston Andros Adamson * Subrequests are always contiguous, non overlapping 519309a1d65SWeston Andros Adamson * and in order - but may be repeated (mirrored writes). 520d4581383SWeston Andros Adamson */ 521309a1d65SWeston Andros Adamson if (subreq->wb_offset == (head->wb_offset + total_bytes)) { 522d4581383SWeston Andros Adamson /* keep track of how many bytes this group covers */ 523d4581383SWeston Andros Adamson total_bytes += subreq->wb_bytes; 524309a1d65SWeston Andros Adamson } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || 525309a1d65SWeston Andros Adamson ((subreq->wb_offset + subreq->wb_bytes) > 526309a1d65SWeston Andros Adamson (head->wb_offset + total_bytes)))) { 527309a1d65SWeston Andros Adamson nfs_page_group_unlock(head); 528309a1d65SWeston Andros Adamson spin_unlock(&inode->i_lock); 529309a1d65SWeston Andros Adamson return ERR_PTR(-EIO); 530309a1d65SWeston Andros Adamson } 531d4581383SWeston Andros Adamson 532d4581383SWeston Andros Adamson if (!nfs_lock_request(subreq)) { 533d4581383SWeston Andros Adamson /* releases page group bit lock and 534d4581383SWeston Andros Adamson * inode spin lock and all references */ 535d4581383SWeston Andros Adamson ret = nfs_unroll_locks_and_wait(inode, head, 536*6d17d653STrond Myklebust subreq); 537d4581383SWeston Andros Adamson 538d4581383SWeston Andros Adamson if (ret == 0) 539d4581383SWeston Andros Adamson goto try_again; 540d4581383SWeston Andros Adamson 541d4581383SWeston Andros Adamson return ERR_PTR(ret); 542d4581383SWeston Andros Adamson } 543d4581383SWeston Andros Adamson 544d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 545d4581383SWeston Andros Adamson } while (subreq != head); 546d4581383SWeston Andros Adamson 547d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 548d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 549d4581383SWeston Andros Adamson subreq = head; 550d4581383SWeston Andros Adamson do { 551411a99adSWeston Andros Adamson nfs_clear_request_commit(subreq); 552d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 553d4581383SWeston Andros Adamson } while (subreq != head); 554d4581383SWeston Andros Adamson 555d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 556d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 557d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 558d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 559d4581383SWeston Andros Adamson head->wb_this_page = head; 560d4581383SWeston Andros Adamson 561d4581383SWeston Andros Adamson /* change head request to cover whole range that 562d4581383SWeston Andros Adamson * the former page group covered */ 563d4581383SWeston Andros Adamson head->wb_bytes = total_bytes; 564d4581383SWeston Andros Adamson } 565d4581383SWeston Andros Adamson 566d4581383SWeston Andros Adamson /* 567d4581383SWeston Andros Adamson * prepare head request to be added to new pgio descriptor 568d4581383SWeston Andros Adamson */ 569d4581383SWeston Andros Adamson nfs_page_group_clear_bits(head); 570d4581383SWeston Andros Adamson 571d4581383SWeston Andros Adamson /* 572d4581383SWeston Andros Adamson * some part of the group was still on the inode list - otherwise 573d4581383SWeston Andros Adamson * the group wouldn't be involved in async write. 574d4581383SWeston Andros Adamson * grab a reference for the head request, iff it needs one. 575d4581383SWeston Andros Adamson */ 576d4581383SWeston Andros Adamson if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags)) 577d4581383SWeston Andros Adamson kref_get(&head->wb_kref); 578d4581383SWeston Andros Adamson 579d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 580d4581383SWeston Andros Adamson 581411a99adSWeston Andros Adamson /* drop lock to clean uprequests on destroy list */ 582d4581383SWeston Andros Adamson spin_unlock(&inode->i_lock); 583d4581383SWeston Andros Adamson 584d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(destroy_list, head); 585d4581383SWeston Andros Adamson 586d4581383SWeston Andros Adamson /* still holds ref on head from nfs_page_find_head_request_locked 587d4581383SWeston Andros Adamson * and still has lock on head from lock loop */ 588d4581383SWeston Andros Adamson return head; 589612c9384STrond Myklebust } 590074cc1deSTrond Myklebust 5910bcbf039SPeng Tao static void nfs_write_error_remove_page(struct nfs_page *req) 5920bcbf039SPeng Tao { 5930bcbf039SPeng Tao nfs_unlock_request(req); 5940bcbf039SPeng Tao nfs_end_page_writeback(req); 5950bcbf039SPeng Tao generic_error_remove_page(page_file_mapping(req->wb_page), 5960bcbf039SPeng Tao req->wb_page); 5971f84ccdfSFred Isaman nfs_release_request(req); 5980bcbf039SPeng Tao } 5990bcbf039SPeng Tao 600a6598813STrond Myklebust static bool 601a6598813STrond Myklebust nfs_error_is_fatal_on_server(int err) 602a6598813STrond Myklebust { 603a6598813STrond Myklebust switch (err) { 604a6598813STrond Myklebust case 0: 605a6598813STrond Myklebust case -ERESTARTSYS: 606a6598813STrond Myklebust case -EINTR: 607a6598813STrond Myklebust return false; 608a6598813STrond Myklebust } 609a6598813STrond Myklebust return nfs_error_is_fatal(err); 610e261f51fSTrond Myklebust } 611074cc1deSTrond Myklebust 612074cc1deSTrond Myklebust /* 613074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 614074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 615074cc1deSTrond Myklebust */ 616074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 617*6d17d653STrond Myklebust struct page *page) 618074cc1deSTrond Myklebust { 619074cc1deSTrond Myklebust struct nfs_page *req; 620074cc1deSTrond Myklebust int ret = 0; 621074cc1deSTrond Myklebust 622*6d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 623074cc1deSTrond Myklebust if (!req) 624074cc1deSTrond Myklebust goto out; 625074cc1deSTrond Myklebust ret = PTR_ERR(req); 626074cc1deSTrond Myklebust if (IS_ERR(req)) 627074cc1deSTrond Myklebust goto out; 628074cc1deSTrond Myklebust 629deed85e7STrond Myklebust nfs_set_page_writeback(page); 630deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 631074cc1deSTrond Myklebust 632deed85e7STrond Myklebust ret = 0; 633a6598813STrond Myklebust /* If there is a fatal error that covers this write, just exit */ 634a6598813STrond Myklebust if (nfs_error_is_fatal_on_server(req->wb_context->error)) 635a6598813STrond Myklebust goto out_launder; 636a6598813STrond Myklebust 637f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 638074cc1deSTrond Myklebust ret = pgio->pg_error; 6390bcbf039SPeng Tao /* 640c373fff7STrond Myklebust * Remove the problematic req upon fatal errors on the server 6410bcbf039SPeng Tao */ 6420bcbf039SPeng Tao if (nfs_error_is_fatal(ret)) { 6430bcbf039SPeng Tao nfs_context_set_write_error(req->wb_context, ret); 644c373fff7STrond Myklebust if (nfs_error_is_fatal_on_server(ret)) 645a6598813STrond Myklebust goto out_launder; 646d6c843b9SPeng Tao } 6470bcbf039SPeng Tao nfs_redirty_request(req); 6480bcbf039SPeng Tao ret = -EAGAIN; 64940f90271STrond Myklebust } else 65040f90271STrond Myklebust nfs_add_stats(page_file_mapping(page)->host, 65140f90271STrond Myklebust NFSIOS_WRITEPAGES, 1); 652074cc1deSTrond Myklebust out: 653074cc1deSTrond Myklebust return ret; 654a6598813STrond Myklebust out_launder: 655a6598813STrond Myklebust nfs_write_error_remove_page(req); 656a6598813STrond Myklebust return ret; 657e261f51fSTrond Myklebust } 658e261f51fSTrond Myklebust 659d6c843b9SPeng Tao static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, 660c373fff7STrond Myklebust struct nfs_pageio_descriptor *pgio) 661f758c885STrond Myklebust { 662cfb506e1STrond Myklebust int ret; 663f758c885STrond Myklebust 6648cd79788SHuang Ying nfs_pageio_cond_complete(pgio, page_index(page)); 665*6d17d653STrond Myklebust ret = nfs_page_async_flush(pgio, page); 666cfb506e1STrond Myklebust if (ret == -EAGAIN) { 667cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 668cfb506e1STrond Myklebust ret = 0; 669cfb506e1STrond Myklebust } 670cfb506e1STrond Myklebust return ret; 671f758c885STrond Myklebust } 672f758c885STrond Myklebust 673e261f51fSTrond Myklebust /* 6741da177e4SLinus Torvalds * Write an mmapped page to the server. 6751da177e4SLinus Torvalds */ 676d6c843b9SPeng Tao static int nfs_writepage_locked(struct page *page, 677c373fff7STrond Myklebust struct writeback_control *wbc) 6781da177e4SLinus Torvalds { 679f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 68040f90271STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 681e261f51fSTrond Myklebust int err; 6821da177e4SLinus Torvalds 68340f90271STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 684811ed92eSTrond Myklebust nfs_pageio_init_write(&pgio, inode, 0, 685a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 686c373fff7STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 687f758c885STrond Myklebust nfs_pageio_complete(&pgio); 688f758c885STrond Myklebust if (err < 0) 6894d770ccfSTrond Myklebust return err; 690f758c885STrond Myklebust if (pgio.pg_error < 0) 691f758c885STrond Myklebust return pgio.pg_error; 692f758c885STrond Myklebust return 0; 6934d770ccfSTrond Myklebust } 6944d770ccfSTrond Myklebust 6954d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 6964d770ccfSTrond Myklebust { 697f758c885STrond Myklebust int ret; 6984d770ccfSTrond Myklebust 699c373fff7STrond Myklebust ret = nfs_writepage_locked(page, wbc); 7001da177e4SLinus Torvalds unlock_page(page); 701f758c885STrond Myklebust return ret; 702f758c885STrond Myklebust } 703f758c885STrond Myklebust 704f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 705f758c885STrond Myklebust { 706f758c885STrond Myklebust int ret; 707f758c885STrond Myklebust 708c373fff7STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 709f758c885STrond Myklebust unlock_page(page); 710f758c885STrond Myklebust return ret; 7111da177e4SLinus Torvalds } 7121da177e4SLinus Torvalds 713919e3bd9STrond Myklebust static void nfs_io_completion_commit(void *inode) 714919e3bd9STrond Myklebust { 715919e3bd9STrond Myklebust nfs_commit_inode(inode, 0); 716919e3bd9STrond Myklebust } 717919e3bd9STrond Myklebust 7181da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 7191da177e4SLinus Torvalds { 7201da177e4SLinus Torvalds struct inode *inode = mapping->host; 721c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 722919e3bd9STrond Myklebust struct nfs_io_completion *ioc = nfs_io_completion_alloc(GFP_NOFS); 7231da177e4SLinus Torvalds int err; 7241da177e4SLinus Torvalds 72591d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 72691d5b470SChuck Lever 727919e3bd9STrond Myklebust if (ioc) 728919e3bd9STrond Myklebust nfs_io_completion_init(ioc, nfs_io_completion_commit, inode); 729919e3bd9STrond Myklebust 730a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 731a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 732919e3bd9STrond Myklebust pgio.pg_io_completion = ioc; 733f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 734c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 735919e3bd9STrond Myklebust nfs_io_completion_put(ioc); 73672cb77f4STrond Myklebust 737f758c885STrond Myklebust if (err < 0) 73872cb77f4STrond Myklebust goto out_err; 73972cb77f4STrond Myklebust err = pgio.pg_error; 74072cb77f4STrond Myklebust if (err < 0) 74172cb77f4STrond Myklebust goto out_err; 742c63c7b05STrond Myklebust return 0; 74372cb77f4STrond Myklebust out_err: 74472cb77f4STrond Myklebust return err; 7451da177e4SLinus Torvalds } 7461da177e4SLinus Torvalds 7471da177e4SLinus Torvalds /* 7481da177e4SLinus Torvalds * Insert a write request into an inode 7491da177e4SLinus Torvalds */ 750d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 7511da177e4SLinus Torvalds { 7521da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 753e7d39069STrond Myklebust 7542bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 7552bfc6e56SWeston Andros Adamson 756e7d39069STrond Myklebust /* Lock the request! */ 7577ad84aa9STrond Myklebust nfs_lock_request(req); 758e7d39069STrond Myklebust 759e7d39069STrond Myklebust spin_lock(&inode->i_lock); 760cb1410c7SWeston Andros Adamson if (!nfsi->nrequests && 761cb1410c7SWeston Andros Adamson NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 762a9a4a87aSTrond Myklebust inode->i_version++; 76329418aa4SMel Gorman /* 76429418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 76529418aa4SMel Gorman * with invalidate/truncate. 76629418aa4SMel Gorman */ 76729418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 7682df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 769deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 770277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 77129418aa4SMel Gorman } 772cb1410c7SWeston Andros Adamson nfsi->nrequests++; 77317089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 774cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 775cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 776cb1410c7SWeston Andros Adamson * adding subrequests. */ 77717089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 778c03b4024STrond Myklebust kref_get(&req->wb_kref); 779e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 7801da177e4SLinus Torvalds } 7811da177e4SLinus Torvalds 7821da177e4SLinus Torvalds /* 78389a09141SPeter Zijlstra * Remove a write request from an inode 7841da177e4SLinus Torvalds */ 7851da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7861da177e4SLinus Torvalds { 7872b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 7881da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 78920633f04SWeston Andros Adamson struct nfs_page *head; 79020633f04SWeston Andros Adamson 79120633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 79220633f04SWeston Andros Adamson head = req->wb_head; 7931da177e4SLinus Torvalds 794587142f8STrond Myklebust spin_lock(&inode->i_lock); 79567911c8fSAnna Schumaker if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { 79620633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 79720633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 79820633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 79929418aa4SMel Gorman } 800cb1410c7SWeston Andros Adamson nfsi->nrequests--; 801cb1410c7SWeston Andros Adamson spin_unlock(&inode->i_lock); 802cb1410c7SWeston Andros Adamson } else { 803cb1410c7SWeston Andros Adamson spin_lock(&inode->i_lock); 804cb1410c7SWeston Andros Adamson nfsi->nrequests--; 805587142f8STrond Myklebust spin_unlock(&inode->i_lock); 80620633f04SWeston Andros Adamson } 80717089a29SWeston Andros Adamson 80817089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 8091da177e4SLinus Torvalds nfs_release_request(req); 8101da177e4SLinus Torvalds } 8111da177e4SLinus Torvalds 81261822ab5STrond Myklebust static void 8136d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 81461822ab5STrond Myklebust { 81567911c8fSAnna Schumaker if (req->wb_page) 81661822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 81761822ab5STrond Myklebust } 81861822ab5STrond Myklebust 8193a3908c8STrond Myklebust /* 8203a3908c8STrond Myklebust * nfs_page_search_commits_for_head_request_locked 8213a3908c8STrond Myklebust * 8223a3908c8STrond Myklebust * Search through commit lists on @inode for the head request for @page. 8233a3908c8STrond Myklebust * Must be called while holding the inode (which is cinfo) lock. 8243a3908c8STrond Myklebust * 8253a3908c8STrond Myklebust * Returns the head request if found, or NULL if not found. 8263a3908c8STrond Myklebust */ 8273a3908c8STrond Myklebust static struct nfs_page * 8283a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 8293a3908c8STrond Myklebust struct page *page) 8303a3908c8STrond Myklebust { 8313a3908c8STrond Myklebust struct nfs_page *freq, *t; 8323a3908c8STrond Myklebust struct nfs_commit_info cinfo; 8333a3908c8STrond Myklebust struct inode *inode = &nfsi->vfs_inode; 8343a3908c8STrond Myklebust 8353a3908c8STrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 8363a3908c8STrond Myklebust 8373a3908c8STrond Myklebust /* search through pnfs commit lists */ 8383a3908c8STrond Myklebust freq = pnfs_search_commit_reqs(inode, &cinfo, page); 8393a3908c8STrond Myklebust if (freq) 8403a3908c8STrond Myklebust return freq->wb_head; 8413a3908c8STrond Myklebust 8423a3908c8STrond Myklebust /* Linearly search the commit list for the correct request */ 8433a3908c8STrond Myklebust list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 8443a3908c8STrond Myklebust if (freq->wb_page == page) 8453a3908c8STrond Myklebust return freq->wb_head; 8463a3908c8STrond Myklebust } 8473a3908c8STrond Myklebust 8483a3908c8STrond Myklebust return NULL; 8493a3908c8STrond Myklebust } 8503a3908c8STrond Myklebust 8518dd37758STrond Myklebust /** 85286d80f97STrond Myklebust * nfs_request_add_commit_list_locked - add request to a commit list 85386d80f97STrond Myklebust * @req: pointer to a struct nfs_page 85486d80f97STrond Myklebust * @dst: commit list head 85586d80f97STrond Myklebust * @cinfo: holds list lock and accounting info 85686d80f97STrond Myklebust * 85786d80f97STrond Myklebust * This sets the PG_CLEAN bit, updates the cinfo count of 85886d80f97STrond Myklebust * number of outstanding requests requiring a commit as well as 85986d80f97STrond Myklebust * the MM page stats. 86086d80f97STrond Myklebust * 861fe238e60SDave Wysochanski * The caller must hold cinfo->inode->i_lock, and the nfs_page lock. 86286d80f97STrond Myklebust */ 86386d80f97STrond Myklebust void 86486d80f97STrond Myklebust nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, 86586d80f97STrond Myklebust struct nfs_commit_info *cinfo) 86686d80f97STrond Myklebust { 86786d80f97STrond Myklebust set_bit(PG_CLEAN, &req->wb_flags); 86886d80f97STrond Myklebust nfs_list_add_request(req, dst); 86986d80f97STrond Myklebust cinfo->mds->ncommit++; 87086d80f97STrond Myklebust } 87186d80f97STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); 87286d80f97STrond Myklebust 87386d80f97STrond Myklebust /** 8748dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 8758dd37758STrond Myklebust * @req: pointer to a struct nfs_page 876ea2cf228SFred Isaman * @dst: commit list head 877ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8788dd37758STrond Myklebust * 879ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 8808dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8818dd37758STrond Myklebust * the MM page stats. 8828dd37758STrond Myklebust * 883ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8848dd37758STrond Myklebust * holding the nfs_page lock. 8858dd37758STrond Myklebust */ 8868dd37758STrond Myklebust void 8876272dcc6SAnna Schumaker nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) 8888dd37758STrond Myklebust { 889fe238e60SDave Wysochanski spin_lock(&cinfo->inode->i_lock); 8906272dcc6SAnna Schumaker nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); 891fe238e60SDave Wysochanski spin_unlock(&cinfo->inode->i_lock); 89267911c8fSAnna Schumaker if (req->wb_page) 89386d80f97STrond Myklebust nfs_mark_page_unstable(req->wb_page, cinfo); 8948dd37758STrond Myklebust } 8958dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 8968dd37758STrond Myklebust 8978dd37758STrond Myklebust /** 8988dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 8998dd37758STrond Myklebust * @req: pointer to a nfs_page 900ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 9018dd37758STrond Myklebust * 902ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 9038dd37758STrond Myklebust * number of outstanding requests requiring a commit 9048dd37758STrond Myklebust * It does not update the MM page stats. 9058dd37758STrond Myklebust * 906ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 9078dd37758STrond Myklebust */ 9088dd37758STrond Myklebust void 909ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 910ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9118dd37758STrond Myklebust { 9128dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 9138dd37758STrond Myklebust return; 9148dd37758STrond Myklebust nfs_list_remove_request(req); 915ea2cf228SFred Isaman cinfo->mds->ncommit--; 9168dd37758STrond Myklebust } 9178dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 9188dd37758STrond Myklebust 919ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 920ea2cf228SFred Isaman struct inode *inode) 921ea2cf228SFred Isaman { 922fe238e60SDave Wysochanski cinfo->inode = inode; 923ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 924ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 925b359f9d0SFred Isaman cinfo->dreq = NULL; 926f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 927ea2cf228SFred Isaman } 928ea2cf228SFred Isaman 929ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 930ea2cf228SFred Isaman struct inode *inode, 931ea2cf228SFred Isaman struct nfs_direct_req *dreq) 932ea2cf228SFred Isaman { 9331763da12SFred Isaman if (dreq) 9341763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 9351763da12SFred Isaman else 936ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 937ea2cf228SFred Isaman } 938ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 9398dd37758STrond Myklebust 9401da177e4SLinus Torvalds /* 9411da177e4SLinus Torvalds * Add a request to the inode's commit list. 9421da177e4SLinus Torvalds */ 9431763da12SFred Isaman void 944ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 945b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 9461da177e4SLinus Torvalds { 947b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 9488dd37758STrond Myklebust return; 9496272dcc6SAnna Schumaker nfs_request_add_commit_list(req, cinfo); 9501da177e4SLinus Torvalds } 9518e821cadSTrond Myklebust 952d6d6dc7cSFred Isaman static void 953d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 954e468bae9STrond Myklebust { 95511fb9989SMel Gorman dec_node_page_state(page, NR_UNSTABLE_NFS); 95693f78d88STejun Heo dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, 95793f78d88STejun Heo WB_RECLAIMABLE); 958e468bae9STrond Myklebust } 959d6d6dc7cSFred Isaman 960411a99adSWeston Andros Adamson /* Called holding inode (/cinfo) lock */ 9618dd37758STrond Myklebust static void 962d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 963d6d6dc7cSFred Isaman { 9648dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 9652b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 966ea2cf228SFred Isaman struct nfs_commit_info cinfo; 967d6d6dc7cSFred Isaman 968ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 969ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 970ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 971d6d6dc7cSFred Isaman } 9728dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 9738dd37758STrond Myklebust } 974e468bae9STrond Myklebust } 975e468bae9STrond Myklebust 976d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 9778e821cadSTrond Myklebust { 978c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 979d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 980c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 9818e821cadSTrond Myklebust } 9828e821cadSTrond Myklebust 983919e3bd9STrond Myklebust static void nfs_async_write_init(struct nfs_pgio_header *hdr) 984919e3bd9STrond Myklebust { 985919e3bd9STrond Myklebust nfs_io_completion_get(hdr->io_completion); 986919e3bd9STrond Myklebust } 987919e3bd9STrond Myklebust 988061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 9896c75dc0dSFred Isaman { 990ea2cf228SFred Isaman struct nfs_commit_info cinfo; 9916c75dc0dSFred Isaman unsigned long bytes = 0; 9926c75dc0dSFred Isaman 9936c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 9946c75dc0dSFred Isaman goto out; 995ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 9966c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 9976c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 9986c75dc0dSFred Isaman 9996c75dc0dSFred Isaman bytes += req->wb_bytes; 10006c75dc0dSFred Isaman nfs_list_remove_request(req); 10016c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 10026c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 1003d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 10046c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 10056c75dc0dSFred Isaman goto remove_req; 10066c75dc0dSFred Isaman } 1007c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 1008f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 1009b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 1010a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 10116c75dc0dSFred Isaman goto next; 10126c75dc0dSFred Isaman } 10136c75dc0dSFred Isaman remove_req: 10146c75dc0dSFred Isaman nfs_inode_remove_request(req); 10156c75dc0dSFred Isaman next: 10161d1afcbcSTrond Myklebust nfs_unlock_request(req); 101720633f04SWeston Andros Adamson nfs_end_page_writeback(req); 10183aff4ebbSTrond Myklebust nfs_release_request(req); 10196c75dc0dSFred Isaman } 10206c75dc0dSFred Isaman out: 1021919e3bd9STrond Myklebust nfs_io_completion_put(hdr->io_completion); 10226c75dc0dSFred Isaman hdr->release(hdr); 10236c75dc0dSFred Isaman } 10246c75dc0dSFred Isaman 1025ce59515cSAnna Schumaker unsigned long 1026ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 1027fb8a1f11STrond Myklebust { 1028ea2cf228SFred Isaman return cinfo->mds->ncommit; 1029fb8a1f11STrond Myklebust } 1030fb8a1f11STrond Myklebust 1031fe238e60SDave Wysochanski /* cinfo->inode->i_lock held by caller */ 10321763da12SFred Isaman int 1033ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 1034ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 1035d6d6dc7cSFred Isaman { 1036d6d6dc7cSFred Isaman struct nfs_page *req, *tmp; 1037d6d6dc7cSFred Isaman int ret = 0; 1038d6d6dc7cSFred Isaman 1039d6d6dc7cSFred Isaman list_for_each_entry_safe(req, tmp, src, wb_list) { 10408dd37758STrond Myklebust if (!nfs_lock_request(req)) 10418dd37758STrond Myklebust continue; 10427ad84aa9STrond Myklebust kref_get(&req->wb_kref); 1043fe238e60SDave Wysochanski if (cond_resched_lock(&cinfo->inode->i_lock)) 10443b3be88dSTrond Myklebust list_safe_reset_next(req, tmp, wb_list); 1045ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 10468dd37758STrond Myklebust nfs_list_add_request(req, dst); 1047d6d6dc7cSFred Isaman ret++; 10481763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 1049d6d6dc7cSFred Isaman break; 1050d6d6dc7cSFred Isaman } 1051d6d6dc7cSFred Isaman return ret; 1052d6d6dc7cSFred Isaman } 1053d6d6dc7cSFred Isaman 10541da177e4SLinus Torvalds /* 10551da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 10561da177e4SLinus Torvalds * @inode: NFS inode to scan 1057ea2cf228SFred Isaman * @dst: mds destination list 1058ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 10591da177e4SLinus Torvalds * 10601da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 10611da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 10621da177e4SLinus Torvalds */ 10631763da12SFred Isaman int 1064ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 1065ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 10661da177e4SLinus Torvalds { 1067d6d6dc7cSFred Isaman int ret = 0; 1068fb8a1f11STrond Myklebust 1069fe238e60SDave Wysochanski spin_lock(&cinfo->inode->i_lock); 1070ea2cf228SFred Isaman if (cinfo->mds->ncommit > 0) { 10718dd37758STrond Myklebust const int max = INT_MAX; 1072d6d6dc7cSFred Isaman 1073ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 1074ea2cf228SFred Isaman cinfo, max); 1075ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 1076d6d6dc7cSFred Isaman } 1077fe238e60SDave Wysochanski spin_unlock(&cinfo->inode->i_lock); 1078ff778d02STrond Myklebust return ret; 10791da177e4SLinus Torvalds } 1080d6d6dc7cSFred Isaman 10811da177e4SLinus Torvalds /* 1082e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1083e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 10841da177e4SLinus Torvalds * 1085e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1086e7d39069STrond Myklebust * to disk. 10871da177e4SLinus Torvalds */ 1088e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 1089e7d39069STrond Myklebust struct page *page, 1090e7d39069STrond Myklebust unsigned int offset, 1091e7d39069STrond Myklebust unsigned int bytes) 10921da177e4SLinus Torvalds { 1093e7d39069STrond Myklebust struct nfs_page *req; 1094e7d39069STrond Myklebust unsigned int rqend; 1095e7d39069STrond Myklebust unsigned int end; 10961da177e4SLinus Torvalds int error; 1097277459d2STrond Myklebust 1098e7d39069STrond Myklebust if (!PagePrivate(page)) 1099e7d39069STrond Myklebust return NULL; 1100e7d39069STrond Myklebust 1101e7d39069STrond Myklebust end = offset + bytes; 1102e7d39069STrond Myklebust spin_lock(&inode->i_lock); 1103e7d39069STrond Myklebust 1104e7d39069STrond Myklebust for (;;) { 110584d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request_locked(NFS_I(inode), page); 1106e7d39069STrond Myklebust if (req == NULL) 1107e7d39069STrond Myklebust goto out_unlock; 1108e7d39069STrond Myklebust 11092bfc6e56SWeston Andros Adamson /* should be handled by nfs_flush_incompatible */ 11102bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 11112bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 11122bfc6e56SWeston Andros Adamson 1113e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1114e7d39069STrond Myklebust /* 1115e7d39069STrond Myklebust * Tell the caller to flush out the request if 1116e7d39069STrond Myklebust * the offsets are non-contiguous. 1117e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1118e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1119e7d39069STrond Myklebust */ 1120e468bae9STrond Myklebust if (offset > rqend 1121e7d39069STrond Myklebust || end < req->wb_offset) 1122e7d39069STrond Myklebust goto out_flushme; 1123e7d39069STrond Myklebust 11247ad84aa9STrond Myklebust if (nfs_lock_request(req)) 1125e7d39069STrond Myklebust break; 1126e7d39069STrond Myklebust 1127e7d39069STrond Myklebust /* The request is locked, so wait and then retry */ 1128587142f8STrond Myklebust spin_unlock(&inode->i_lock); 11291da177e4SLinus Torvalds error = nfs_wait_on_request(req); 11301da177e4SLinus Torvalds nfs_release_request(req); 1131e7d39069STrond Myklebust if (error != 0) 1132e7d39069STrond Myklebust goto out_err; 1133e7d39069STrond Myklebust spin_lock(&inode->i_lock); 11341da177e4SLinus Torvalds } 11351da177e4SLinus Torvalds 11361da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 11371da177e4SLinus Torvalds if (offset < req->wb_offset) { 11381da177e4SLinus Torvalds req->wb_offset = offset; 11391da177e4SLinus Torvalds req->wb_pgbase = offset; 11401da177e4SLinus Torvalds } 11411da177e4SLinus Torvalds if (end > rqend) 11421da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1143e7d39069STrond Myklebust else 1144e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 1145e7d39069STrond Myklebust out_unlock: 1146ca138f36SFred Isaman if (req) 11478dd37758STrond Myklebust nfs_clear_request_commit(req); 1148411a99adSWeston Andros Adamson spin_unlock(&inode->i_lock); 1149e7d39069STrond Myklebust return req; 1150e7d39069STrond Myklebust out_flushme: 1151e7d39069STrond Myklebust spin_unlock(&inode->i_lock); 1152e7d39069STrond Myklebust nfs_release_request(req); 1153e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1154e7d39069STrond Myklebust out_err: 1155e7d39069STrond Myklebust return ERR_PTR(error); 1156e7d39069STrond Myklebust } 11571da177e4SLinus Torvalds 1158e7d39069STrond Myklebust /* 1159e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1160e7d39069STrond Myklebust * 1161e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1162e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1163e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1164e7d39069STrond Myklebust */ 1165e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1166e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1167e7d39069STrond Myklebust { 1168d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1169e7d39069STrond Myklebust struct nfs_page *req; 1170e7d39069STrond Myklebust 1171e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1172e7d39069STrond Myklebust if (req != NULL) 1173e7d39069STrond Myklebust goto out; 11742bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 1175e7d39069STrond Myklebust if (IS_ERR(req)) 1176e7d39069STrond Myklebust goto out; 1177d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1178efc91ed0STrond Myklebust out: 117961e930a9STrond Myklebust return req; 11801da177e4SLinus Torvalds } 11811da177e4SLinus Torvalds 1182e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1183e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1184e7d39069STrond Myklebust { 1185e7d39069STrond Myklebust struct nfs_page *req; 1186e7d39069STrond Myklebust 1187e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1188e7d39069STrond Myklebust if (IS_ERR(req)) 1189e7d39069STrond Myklebust return PTR_ERR(req); 1190e7d39069STrond Myklebust /* Update file length */ 1191e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1192d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1193a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11941d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1195e7d39069STrond Myklebust return 0; 1196e7d39069STrond Myklebust } 1197e7d39069STrond Myklebust 11981da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 11991da177e4SLinus Torvalds { 1200cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 12012a369153STrond Myklebust struct nfs_lock_context *l_ctx; 1202bd61e0a9SJeff Layton struct file_lock_context *flctx = file_inode(file)->i_flctx; 12031da177e4SLinus Torvalds struct nfs_page *req; 12041a54533eSTrond Myklebust int do_flush, status; 12051da177e4SLinus Torvalds /* 12061da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 12071da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 12081da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 12091da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 12101da177e4SLinus Torvalds * Also do the same if we find a request from an existing 12111da177e4SLinus Torvalds * dropped page. 12121da177e4SLinus Torvalds */ 12131a54533eSTrond Myklebust do { 121484d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 12151a54533eSTrond Myklebust if (req == NULL) 12161a54533eSTrond Myklebust return 0; 12172a369153STrond Myklebust l_ctx = req->wb_lock_context; 1218138a2935STrond Myklebust do_flush = req->wb_page != page || 1219138a2935STrond Myklebust !nfs_match_open_context(req->wb_context, ctx); 12202bfc6e56SWeston Andros Adamson /* for now, flush if more than 1 request in page_group */ 12212bfc6e56SWeston Andros Adamson do_flush |= req->wb_this_page != req; 1222bd61e0a9SJeff Layton if (l_ctx && flctx && 1223bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1224bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 1225d51fdb87SNeilBrown do_flush |= l_ctx->lockowner != current->files; 12265263e31eSJeff Layton } 12271da177e4SLinus Torvalds nfs_release_request(req); 12281a54533eSTrond Myklebust if (!do_flush) 12291a54533eSTrond Myklebust return 0; 1230d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 12311a54533eSTrond Myklebust } while (status == 0); 12321a54533eSTrond Myklebust return status; 12331da177e4SLinus Torvalds } 12341da177e4SLinus Torvalds 12351da177e4SLinus Torvalds /* 1236dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1237dc24826bSAndy Adamson * expire soon. 1238dc24826bSAndy Adamson * 1239dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1240dc24826bSAndy Adamson * 1241dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1242dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1243dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1244dc24826bSAndy Adamson */ 1245dc24826bSAndy Adamson int 1246dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1247dc24826bSAndy Adamson { 1248dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1249dc24826bSAndy Adamson struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1250dc24826bSAndy Adamson 1251dc24826bSAndy Adamson return rpcauth_key_timeout_notify(auth, ctx->cred); 1252dc24826bSAndy Adamson } 1253dc24826bSAndy Adamson 1254dc24826bSAndy Adamson /* 1255dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1256dc24826bSAndy Adamson */ 1257ce52914eSScott Mayhew bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) 1258dc24826bSAndy Adamson { 1259ce52914eSScott Mayhew struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1260ce52914eSScott Mayhew 1261ce52914eSScott Mayhew return rpcauth_cred_key_to_expire(auth, ctx->cred); 1262dc24826bSAndy Adamson } 1263dc24826bSAndy Adamson 1264dc24826bSAndy Adamson /* 12655d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 12665d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 12675d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 12685d47a356STrond Myklebust */ 12698d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 12705d47a356STrond Myklebust { 1271d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1272d529ef83SJeff Layton 12738d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 12748d197a56STrond Myklebust goto out; 127518dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1276d529ef83SJeff Layton return false; 12774db72b40SJeff Layton smp_rmb(); 1278d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 12798d197a56STrond Myklebust return false; 12808d197a56STrond Myklebust out: 128118dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 128218dd78c4SScott Mayhew return false; 12838d197a56STrond Myklebust return PageUptodate(page) != 0; 12845d47a356STrond Myklebust } 12855d47a356STrond Myklebust 12865263e31eSJeff Layton static bool 12875263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 12885263e31eSJeff Layton { 12895263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 12905263e31eSJeff Layton fl->fl_type == F_WRLCK; 12915263e31eSJeff Layton } 12925263e31eSJeff Layton 1293c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1294c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1295c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1296c7559663SScott Mayhew * inefficiencies. 1297c7559663SScott Mayhew * 1298263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1299263b4509SScott Mayhew * of the checks. 1300c7559663SScott Mayhew */ 1301c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1302c7559663SScott Mayhew { 13035263e31eSJeff Layton int ret; 13045263e31eSJeff Layton struct file_lock_context *flctx = inode->i_flctx; 13055263e31eSJeff Layton struct file_lock *fl; 13065263e31eSJeff Layton 1307c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1308c7559663SScott Mayhew return 0; 1309263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1310263b4509SScott Mayhew return 0; 1311c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1312c7559663SScott Mayhew return 1; 1313bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1314bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 13158fa4592aSTrond Myklebust return 1; 13165263e31eSJeff Layton 13175263e31eSJeff Layton /* Check to see if there are whole file write locks */ 13185263e31eSJeff Layton ret = 0; 13196109c850SJeff Layton spin_lock(&flctx->flc_lock); 1320bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1321bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1322bd61e0a9SJeff Layton fl_list); 1323bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 13245263e31eSJeff Layton ret = 1; 1325bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 13265263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 13275263e31eSJeff Layton fl_list); 13285263e31eSJeff Layton if (fl->fl_type == F_WRLCK) 13295263e31eSJeff Layton ret = 1; 13305263e31eSJeff Layton } 13316109c850SJeff Layton spin_unlock(&flctx->flc_lock); 13325263e31eSJeff Layton return ret; 1333c7559663SScott Mayhew } 1334c7559663SScott Mayhew 13355d47a356STrond Myklebust /* 13361da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 13371da177e4SLinus Torvalds * 13381da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 13391da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 13401da177e4SLinus Torvalds */ 13411da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 13421da177e4SLinus Torvalds unsigned int offset, unsigned int count) 13431da177e4SLinus Torvalds { 1344cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1345d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 13461da177e4SLinus Torvalds int status = 0; 13471da177e4SLinus Torvalds 134891d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 134991d5b470SChuck Lever 13506de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 13516de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 13521da177e4SLinus Torvalds 1353149a4fddSBenjamin Coddington if (!count) 1354149a4fddSBenjamin Coddington goto out; 1355149a4fddSBenjamin Coddington 1356c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 135749a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 13581da177e4SLinus Torvalds offset = 0; 13591da177e4SLinus Torvalds } 13601da177e4SLinus Torvalds 1361e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 136203fa9e84STrond Myklebust if (status < 0) 136303fa9e84STrond Myklebust nfs_set_pageerror(page); 136459b7c05fSTrond Myklebust else 136559b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 1366149a4fddSBenjamin Coddington out: 136748186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 13681da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 13691da177e4SLinus Torvalds return status; 13701da177e4SLinus Torvalds } 13711da177e4SLinus Torvalds 13723ff7576dSTrond Myklebust static int flush_task_priority(int how) 13731da177e4SLinus Torvalds { 13741da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 13751da177e4SLinus Torvalds case FLUSH_HIGHPRI: 13761da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 13771da177e4SLinus Torvalds case FLUSH_LOWPRI: 13781da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 13791da177e4SLinus Torvalds } 13801da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 13811da177e4SLinus Torvalds } 13821da177e4SLinus Torvalds 1383d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1384d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1385abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 13861ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 13871da177e4SLinus Torvalds { 13883ff7576dSTrond Myklebust int priority = flush_task_priority(how); 13891da177e4SLinus Torvalds 13901ed26f33SAnna Schumaker task_setup_data->priority = priority; 1391abde71f4STom Haynes rpc_ops->write_setup(hdr, msg); 1392d138d5d1SAndy Adamson 1393abde71f4STom Haynes nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client, 1394d45f60c6SWeston Andros Adamson &task_setup_data->rpc_client, msg, hdr); 1395275acaafSTrond Myklebust } 1396275acaafSTrond Myklebust 13976d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 13986d884e8fSFred * call this on each, which will prepare them to be retried on next 13996d884e8fSFred * writeback using standard nfs. 14006d884e8fSFred */ 14016d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 14026d884e8fSFred { 14036d884e8fSFred nfs_mark_request_dirty(req); 1404c7070113STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 14051d1afcbcSTrond Myklebust nfs_unlock_request(req); 140620633f04SWeston Andros Adamson nfs_end_page_writeback(req); 14073aff4ebbSTrond Myklebust nfs_release_request(req); 14086d884e8fSFred } 14096d884e8fSFred 1410061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 14116c75dc0dSFred Isaman { 14126c75dc0dSFred Isaman struct nfs_page *req; 14136c75dc0dSFred Isaman 14146c75dc0dSFred Isaman while (!list_empty(head)) { 14156c75dc0dSFred Isaman req = nfs_list_entry(head->next); 14166c75dc0dSFred Isaman nfs_list_remove_request(req); 14176c75dc0dSFred Isaman nfs_redirty_request(req); 14186c75dc0dSFred Isaman } 14196c75dc0dSFred Isaman } 14206c75dc0dSFred Isaman 1421dc602dd7STrond Myklebust static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) 1422dc602dd7STrond Myklebust { 1423dc602dd7STrond Myklebust nfs_async_write_error(&hdr->pages); 1424dc602dd7STrond Myklebust } 1425dc602dd7STrond Myklebust 1426061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1427919e3bd9STrond Myklebust .init_hdr = nfs_async_write_init, 1428061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1429061ae2edSFred Isaman .completion = nfs_write_completion, 1430dc602dd7STrond Myklebust .reschedule_io = nfs_async_write_reschedule_io, 1431061ae2edSFred Isaman }; 1432061ae2edSFred Isaman 143357208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1434a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1435061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 14361751c363STrond Myklebust { 1437a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 143841d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1439a20c93e3SChristoph Hellwig 1440a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1441a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1442a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1443a20c93e3SChristoph Hellwig #endif 14444a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 1445fbe77c30SBenjamin Coddington server->wsize, ioflags, GFP_NOIO); 14461751c363STrond Myklebust } 1447ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 14481751c363STrond Myklebust 1449dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1450dce81290STrond Myklebust { 1451a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1452a7d42ddbSWeston Andros Adamson 14536f29b9bbSKinglong Mee if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) 14546f29b9bbSKinglong Mee pgio->pg_ops->pg_cleanup(pgio); 14556f29b9bbSKinglong Mee 145641d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1457a7d42ddbSWeston Andros Adamson 1458a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1459a7d42ddbSWeston Andros Adamson 1460a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1461a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1462dce81290STrond Myklebust } 14631f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1464dce81290STrond Myklebust 14651da177e4SLinus Torvalds 14660b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 14670b7c0153SFred Isaman { 14680b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 14690b7c0153SFred Isaman 14700b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 14710b7c0153SFred Isaman } 14720b7c0153SFred Isaman 14731f2edbe3STrond Myklebust /* 14741f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 14751f2edbe3STrond Myklebust */ 14761f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 14771f2edbe3STrond Myklebust { 14781f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 14791f2edbe3STrond Myklebust int kill = 0; 1480788e7a89STrond Myklebust 14811f2edbe3STrond Myklebust /* suid always must be killed */ 14821f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 14831f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 14841f2edbe3STrond Myklebust 14851f2edbe3STrond Myklebust /* 14861f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 14871f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 14881f2edbe3STrond Myklebust */ 14891f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 14901f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 14911f2edbe3STrond Myklebust 14921f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 14931f2edbe3STrond Myklebust return kill; 14941f2edbe3STrond Myklebust 14951f2edbe3STrond Myklebust return 0; 14961f2edbe3STrond Myklebust } 1497788e7a89STrond Myklebust 1498a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1499a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1500a08a8cd3STrond Myklebust { 1501a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1502a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 15032b83d3deSTrond Myklebust u64 size = argp->offset + resp->count; 1504a08a8cd3STrond Myklebust 1505a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 15062b83d3deSTrond Myklebust fattr->size = size; 15072b83d3deSTrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { 15082b83d3deSTrond Myklebust fattr->valid &= ~NFS_ATTR_FATTR_SIZE; 1509a08a8cd3STrond Myklebust return; 15102b83d3deSTrond Myklebust } 15112b83d3deSTrond Myklebust if (size != fattr->size) 1512a08a8cd3STrond Myklebust return; 1513a08a8cd3STrond Myklebust /* Set attribute barrier */ 1514a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 15152b83d3deSTrond Myklebust /* ...and update size */ 15162b83d3deSTrond Myklebust fattr->valid |= NFS_ATTR_FATTR_SIZE; 1517a08a8cd3STrond Myklebust } 1518a08a8cd3STrond Myklebust 1519a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1520a08a8cd3STrond Myklebust { 15212b83d3deSTrond Myklebust struct nfs_fattr *fattr = &hdr->fattr; 1522a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1523a08a8cd3STrond Myklebust 1524a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1525a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1526a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1527a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1528a08a8cd3STrond Myklebust } 1529a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1530a08a8cd3STrond Myklebust 15311da177e4SLinus Torvalds /* 15321da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 15331da177e4SLinus Torvalds */ 1534d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1535d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 15360eecb214SAnna Schumaker struct inode *inode) 15371da177e4SLinus Torvalds { 1538788e7a89STrond Myklebust int status; 15391da177e4SLinus Torvalds 1540f551e44fSChuck Lever /* 1541f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1542f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1543f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1544f551e44fSChuck Lever * another writer had changed the file, but some applications 1545f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1546f551e44fSChuck Lever */ 1547d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1548788e7a89STrond Myklebust if (status != 0) 15490eecb214SAnna Schumaker return status; 1550d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 155191d5b470SChuck Lever 1552d45f60c6SWeston Andros Adamson if (hdr->res.verf->committed < hdr->args.stable && 1553d45f60c6SWeston Andros Adamson task->tk_status >= 0) { 15541da177e4SLinus Torvalds /* We tried a write call, but the server did not 15551da177e4SLinus Torvalds * commit data to stable storage even though we 15561da177e4SLinus Torvalds * requested it. 15571da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 15581da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 15591da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 15601da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 15611da177e4SLinus Torvalds */ 15621da177e4SLinus Torvalds static unsigned long complain; 15631da177e4SLinus Torvalds 1564a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 15651da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 15661da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 15671da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1568cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1569d45f60c6SWeston Andros Adamson hdr->res.verf->committed, hdr->args.stable); 15701da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 15711da177e4SLinus Torvalds } 15721da177e4SLinus Torvalds } 15731f2edbe3STrond Myklebust 15741f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 15751f2edbe3STrond Myklebust if (nfs_should_remove_suid(inode)) 15761f2edbe3STrond Myklebust nfs_mark_for_revalidate(inode); 15770eecb214SAnna Schumaker return 0; 15780eecb214SAnna Schumaker } 15790eecb214SAnna Schumaker 15800eecb214SAnna Schumaker /* 15810eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 15820eecb214SAnna Schumaker */ 1583d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1584d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 15850eecb214SAnna Schumaker { 1586d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1587d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 15881f2edbe3STrond Myklebust 15891f2edbe3STrond Myklebust if (resp->count < argp->count) { 15901da177e4SLinus Torvalds static unsigned long complain; 15911da177e4SLinus Torvalds 15926c75dc0dSFred Isaman /* This a short write! */ 1593d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 159491d5b470SChuck Lever 15951da177e4SLinus Torvalds /* Has the server at least made some progress? */ 15966c75dc0dSFred Isaman if (resp->count == 0) { 15976c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 15986c75dc0dSFred Isaman printk(KERN_WARNING 15996c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 16006c75dc0dSFred Isaman argp->count); 16016c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 16026c75dc0dSFred Isaman } 1603d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 16046c75dc0dSFred Isaman task->tk_status = -EIO; 16056c75dc0dSFred Isaman return; 16066c75dc0dSFred Isaman } 1607f8417b48SKinglong Mee 1608f8417b48SKinglong Mee /* For non rpc-based layout drivers, retry-through-MDS */ 1609f8417b48SKinglong Mee if (!task->tk_ops) { 1610f8417b48SKinglong Mee hdr->pnfs_error = -EAGAIN; 1611f8417b48SKinglong Mee return; 1612f8417b48SKinglong Mee } 1613f8417b48SKinglong Mee 16141da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 16151da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 16161da177e4SLinus Torvalds /* Resend from where the server left off */ 1617d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 16181da177e4SLinus Torvalds argp->offset += resp->count; 16191da177e4SLinus Torvalds argp->pgbase += resp->count; 16201da177e4SLinus Torvalds argp->count -= resp->count; 16211da177e4SLinus Torvalds } else { 16221da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 16231da177e4SLinus Torvalds * headaches in the case of a server crash. 16241da177e4SLinus Torvalds */ 16251da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 16261da177e4SLinus Torvalds } 1627d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 16281da177e4SLinus Torvalds } 16291da177e4SLinus Torvalds } 16301da177e4SLinus Torvalds 1631af7cf057STrond Myklebust static int wait_on_commit(struct nfs_mds_commit_info *cinfo) 163271d0a611STrond Myklebust { 1633af7cf057STrond Myklebust return wait_on_atomic_t(&cinfo->rpcs_out, 1634af7cf057STrond Myklebust nfs_wait_atomic_killable, TASK_KILLABLE); 1635af7cf057STrond Myklebust } 1636af7cf057STrond Myklebust 1637af7cf057STrond Myklebust static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) 1638af7cf057STrond Myklebust { 1639af7cf057STrond Myklebust atomic_inc(&cinfo->rpcs_out); 1640af7cf057STrond Myklebust } 1641af7cf057STrond Myklebust 1642af7cf057STrond Myklebust static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) 1643af7cf057STrond Myklebust { 1644af7cf057STrond Myklebust if (atomic_dec_and_test(&cinfo->rpcs_out)) 1645af7cf057STrond Myklebust wake_up_atomic_t(&cinfo->rpcs_out); 164671d0a611STrond Myklebust } 164771d0a611STrond Myklebust 16480b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 16491da177e4SLinus Torvalds { 16500b7c0153SFred Isaman put_nfs_open_context(data->context); 16510b7c0153SFred Isaman nfs_commit_free(data); 16521da177e4SLinus Torvalds } 1653e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 16541da177e4SLinus Torvalds 16550b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1656c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 16579ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 16589f0ec176SAndy Adamson int how, int flags) 16591da177e4SLinus Torvalds { 166007737691STrond Myklebust struct rpc_task *task; 16619ace33cdSFred Isaman int priority = flush_task_priority(how); 1662bdc7f021STrond Myklebust struct rpc_message msg = { 1663bdc7f021STrond Myklebust .rpc_argp = &data->args, 1664bdc7f021STrond Myklebust .rpc_resp = &data->res, 16659ace33cdSFred Isaman .rpc_cred = data->cred, 1666bdc7f021STrond Myklebust }; 166784115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 166807737691STrond Myklebust .task = &data->task, 16699ace33cdSFred Isaman .rpc_client = clnt, 1670bdc7f021STrond Myklebust .rpc_message = &msg, 16719ace33cdSFred Isaman .callback_ops = call_ops, 167284115e1cSTrond Myklebust .callback_data = data, 1673101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 16749f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 16753ff7576dSTrond Myklebust .priority = priority, 167684115e1cSTrond Myklebust }; 1677788e7a89STrond Myklebust /* Set up the initial task struct. */ 1678c36aae9aSPeng Tao nfs_ops->commit_setup(data, &msg); 16791da177e4SLinus Torvalds 1680b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1681bdc7f021STrond Myklebust 16828c21c62cSWeston Andros Adamson nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client, 16838c21c62cSWeston Andros Adamson NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg); 16848c21c62cSWeston Andros Adamson 168507737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1686dbae4c73STrond Myklebust if (IS_ERR(task)) 1687dbae4c73STrond Myklebust return PTR_ERR(task); 1688d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1689d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 169007737691STrond Myklebust rpc_put_task(task); 1691dbae4c73STrond Myklebust return 0; 16921da177e4SLinus Torvalds } 1693e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 16941da177e4SLinus Torvalds 1695378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1696378520b8SPeng Tao { 1697378520b8SPeng Tao loff_t lwb = 0; 1698378520b8SPeng Tao struct nfs_page *req; 1699378520b8SPeng Tao 1700378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1701378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1702378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1703378520b8SPeng Tao 1704378520b8SPeng Tao return lwb; 1705378520b8SPeng Tao } 1706378520b8SPeng Tao 17071da177e4SLinus Torvalds /* 17089ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 17099ace33cdSFred Isaman */ 17100b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1711988b6dceSFred Isaman struct list_head *head, 1712f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1713f453a54aSFred Isaman struct nfs_commit_info *cinfo) 17149ace33cdSFred Isaman { 17159ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 17162b0143b5SDavid Howells struct inode *inode = d_inode(first->wb_context->dentry); 17179ace33cdSFred Isaman 17189ace33cdSFred Isaman /* Set up the RPC argument and reply structs 17199ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 17209ace33cdSFred Isaman 17219ace33cdSFred Isaman list_splice_init(head, &data->pages); 17229ace33cdSFred Isaman 17239ace33cdSFred Isaman data->inode = inode; 17249ace33cdSFred Isaman data->cred = first->wb_context->cred; 1725988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1726378520b8SPeng Tao /* only set lwb for pnfs commit */ 1727378520b8SPeng Tao if (lseg) 1728378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 17299ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1730f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1731b359f9d0SFred Isaman data->dreq = cinfo->dreq; 17329ace33cdSFred Isaman 17339ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 17349ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 17359ace33cdSFred Isaman data->args.offset = 0; 17369ace33cdSFred Isaman data->args.count = 0; 17370b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 17389ace33cdSFred Isaman data->res.fattr = &data->fattr; 17399ace33cdSFred Isaman data->res.verf = &data->verf; 17409ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 17419ace33cdSFred Isaman } 1742e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 17439ace33cdSFred Isaman 1744e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1745ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1746b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1747b57ff130SWeston Andros Adamson u32 ds_commit_idx) 174864bfeb49SFred Isaman { 174964bfeb49SFred Isaman struct nfs_page *req; 175064bfeb49SFred Isaman 175164bfeb49SFred Isaman while (!list_empty(page_list)) { 175264bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 175364bfeb49SFred Isaman nfs_list_remove_request(req); 1754b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 1755487b9b8aSTom Haynes if (!cinfo->dreq) 1756487b9b8aSTom Haynes nfs_clear_page_commit(req->wb_page); 17571d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 175864bfeb49SFred Isaman } 175964bfeb49SFred Isaman } 1760e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 176164bfeb49SFred Isaman 1762b20135d0STrond Myklebust static void 1763b20135d0STrond Myklebust nfs_commit_resched_write(struct nfs_commit_info *cinfo, 1764b20135d0STrond Myklebust struct nfs_page *req) 1765b20135d0STrond Myklebust { 1766b20135d0STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 1767b20135d0STrond Myklebust } 1768b20135d0STrond Myklebust 17699ace33cdSFred Isaman /* 17701da177e4SLinus Torvalds * Commit dirty pages 17711da177e4SLinus Torvalds */ 17721da177e4SLinus Torvalds static int 1773ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1774ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 17751da177e4SLinus Torvalds { 17760b7c0153SFred Isaman struct nfs_commit_data *data; 17771da177e4SLinus Torvalds 1778ade8febdSWeston Andros Adamson /* another commit raced with us */ 1779ade8febdSWeston Andros Adamson if (list_empty(head)) 1780ade8febdSWeston Andros Adamson return 0; 1781ade8febdSWeston Andros Adamson 1782518662e0SNeilBrown data = nfs_commitdata_alloc(true); 17831da177e4SLinus Torvalds 17841da177e4SLinus Torvalds /* Set up the argument struct */ 1785f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1786f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 1787c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 1788c36aae9aSPeng Tao data->mds_ops, how, 0); 17891da177e4SLinus Torvalds } 17901da177e4SLinus Torvalds 17911da177e4SLinus Torvalds /* 17921da177e4SLinus Torvalds * COMMIT call returned 17931da177e4SLinus Torvalds */ 1794788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 17951da177e4SLinus Torvalds { 17960b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 17971da177e4SLinus Torvalds 1798a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 17991da177e4SLinus Torvalds task->tk_pid, task->tk_status); 18001da177e4SLinus Torvalds 1801788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1802c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 1803c9d8f89dSTrond Myklebust } 1804c9d8f89dSTrond Myklebust 1805f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1806c9d8f89dSTrond Myklebust { 1807c9d8f89dSTrond Myklebust struct nfs_page *req; 1808c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1809f453a54aSFred Isaman struct nfs_commit_info cinfo; 1810353db796SNeilBrown struct nfs_server *nfss; 1811788e7a89STrond Myklebust 18121da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 18131da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 18141da177e4SLinus Torvalds nfs_list_remove_request(req); 181567911c8fSAnna Schumaker if (req->wb_page) 1816d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 18171da177e4SLinus Torvalds 18181e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 18193d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 18202b0143b5SDavid Howells (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)), 18211da177e4SLinus Torvalds req->wb_bytes, 18221da177e4SLinus Torvalds (long long)req_offset(req)); 1823c9d8f89dSTrond Myklebust if (status < 0) { 1824c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 182538a33101SKinglong Mee if (req->wb_page) 18261da177e4SLinus Torvalds nfs_inode_remove_request(req); 1827ddeaa637SJoe Perches dprintk_cont(", error = %d\n", status); 18281da177e4SLinus Torvalds goto next; 18291da177e4SLinus Torvalds } 18301da177e4SLinus Torvalds 18311da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 18321da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 18338fc3c386STrond Myklebust if (!nfs_write_verifier_cmp(&req->wb_verf, &data->verf.verifier)) { 18341da177e4SLinus Torvalds /* We have a match */ 183538a33101SKinglong Mee if (req->wb_page) 18361da177e4SLinus Torvalds nfs_inode_remove_request(req); 1837ddeaa637SJoe Perches dprintk_cont(" OK\n"); 18381da177e4SLinus Torvalds goto next; 18391da177e4SLinus Torvalds } 18401da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 1841ddeaa637SJoe Perches dprintk_cont(" mismatch\n"); 18426d884e8fSFred nfs_mark_request_dirty(req); 184305990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 18441da177e4SLinus Torvalds next: 18451d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18461da177e4SLinus Torvalds } 1847353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1848353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 18490db10944SJan Kara clear_bdi_congested(inode_to_bdi(data->inode), BLK_RW_ASYNC); 1850353db796SNeilBrown 1851f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1852af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18535917ce84SFred Isaman } 18545917ce84SFred Isaman 18555917ce84SFred Isaman static void nfs_commit_release(void *calldata) 18565917ce84SFred Isaman { 18570b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18585917ce84SFred Isaman 1859f453a54aSFred Isaman data->completion_ops->completion(data); 1860c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 18611da177e4SLinus Torvalds } 1862788e7a89STrond Myklebust 1863788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 18640b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1865788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1866788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1867788e7a89STrond Myklebust }; 18681da177e4SLinus Torvalds 1869f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1870f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1871b20135d0STrond Myklebust .resched_write = nfs_commit_resched_write, 1872f453a54aSFred Isaman }; 1873f453a54aSFred Isaman 18741763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1875ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 187684c53ab5SFred Isaman { 187784c53ab5SFred Isaman int status; 187884c53ab5SFred Isaman 1879ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 188084c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1881ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 188284c53ab5SFred Isaman return status; 188384c53ab5SFred Isaman } 188484c53ab5SFred Isaman 1885b608b283STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 18861da177e4SLinus Torvalds { 18871da177e4SLinus Torvalds LIST_HEAD(head); 1888ea2cf228SFred Isaman struct nfs_commit_info cinfo; 188971d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1890af7cf057STrond Myklebust int error = 0; 1891b8413f98STrond Myklebust int res; 18921da177e4SLinus Torvalds 1893ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1894af7cf057STrond Myklebust nfs_commit_begin(cinfo.mds); 1895ea2cf228SFred Isaman res = nfs_scan_commit(inode, &head, &cinfo); 1896af7cf057STrond Myklebust if (res) 1897ea2cf228SFred Isaman error = nfs_generic_commit_list(inode, &head, how, &cinfo); 1898af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18991da177e4SLinus Torvalds if (error < 0) 1900af7cf057STrond Myklebust goto out_error; 1901b8413f98STrond Myklebust if (!may_wait) 1902b8413f98STrond Myklebust goto out_mark_dirty; 1903af7cf057STrond Myklebust error = wait_on_commit(cinfo.mds); 1904b8413f98STrond Myklebust if (error < 0) 1905b8413f98STrond Myklebust return error; 1906c5efa5fcSTrond Myklebust return res; 1907af7cf057STrond Myklebust out_error: 1908af7cf057STrond Myklebust res = error; 1909c5efa5fcSTrond Myklebust /* Note: If we exit without ensuring that the commit is complete, 1910c5efa5fcSTrond Myklebust * we must mark the inode as dirty. Otherwise, future calls to 1911c5efa5fcSTrond Myklebust * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure 1912c5efa5fcSTrond Myklebust * that the data is on the disk. 1913c5efa5fcSTrond Myklebust */ 1914c5efa5fcSTrond Myklebust out_mark_dirty: 1915c5efa5fcSTrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 19161da177e4SLinus Torvalds return res; 19171da177e4SLinus Torvalds } 1918b20135d0STrond Myklebust EXPORT_SYMBOL_GPL(nfs_commit_inode); 19198fc795f7STrond Myklebust 1920ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 19218fc795f7STrond Myklebust { 1922420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1923420e3646STrond Myklebust int flags = FLUSH_SYNC; 1924420e3646STrond Myklebust int ret = 0; 19258fc795f7STrond Myklebust 19263236c3e1SJeff Layton /* no commits means nothing needs to be done */ 1927ea2cf228SFred Isaman if (!nfsi->commit_info.ncommit) 19283236c3e1SJeff Layton return ret; 19293236c3e1SJeff Layton 1930a00dd6c0SJeff Layton if (wbc->sync_mode == WB_SYNC_NONE) { 1931a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1932a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1933420e3646STrond Myklebust */ 19341a4edf0fSTrond Myklebust if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)) 1935420e3646STrond Myklebust goto out_mark_dirty; 1936420e3646STrond Myklebust 1937a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1938420e3646STrond Myklebust flags = 0; 1939a00dd6c0SJeff Layton } 1940a00dd6c0SJeff Layton 1941420e3646STrond Myklebust ret = nfs_commit_inode(inode, flags); 1942420e3646STrond Myklebust if (ret >= 0) { 1943420e3646STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 1944420e3646STrond Myklebust if (ret < wbc->nr_to_write) 1945420e3646STrond Myklebust wbc->nr_to_write -= ret; 1946420e3646STrond Myklebust else 1947420e3646STrond Myklebust wbc->nr_to_write = 0; 1948420e3646STrond Myklebust } 19498fc795f7STrond Myklebust return 0; 1950420e3646STrond Myklebust } 1951420e3646STrond Myklebust out_mark_dirty: 19528fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 19538fc795f7STrond Myklebust return ret; 19548fc795f7STrond Myklebust } 195589d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1956863a3c6cSAndy Adamson 1957acdc53b2STrond Myklebust /* 1958837bb1d7STrond Myklebust * Wrapper for filemap_write_and_wait_range() 1959837bb1d7STrond Myklebust * 1960837bb1d7STrond Myklebust * Needed for pNFS in order to ensure data becomes visible to the 1961837bb1d7STrond Myklebust * client. 1962837bb1d7STrond Myklebust */ 1963837bb1d7STrond Myklebust int nfs_filemap_write_and_wait_range(struct address_space *mapping, 1964837bb1d7STrond Myklebust loff_t lstart, loff_t lend) 1965837bb1d7STrond Myklebust { 1966837bb1d7STrond Myklebust int ret; 1967837bb1d7STrond Myklebust 1968837bb1d7STrond Myklebust ret = filemap_write_and_wait_range(mapping, lstart, lend); 1969837bb1d7STrond Myklebust if (ret == 0) 1970837bb1d7STrond Myklebust ret = pnfs_sync_inode(mapping->host, true); 1971837bb1d7STrond Myklebust return ret; 1972837bb1d7STrond Myklebust } 1973837bb1d7STrond Myklebust EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); 1974837bb1d7STrond Myklebust 1975837bb1d7STrond Myklebust /* 1976acdc53b2STrond Myklebust * flush the inode to disk. 1977acdc53b2STrond Myklebust */ 1978acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 197934901f70STrond Myklebust { 1980f4ce1299STrond Myklebust int ret; 198134901f70STrond Myklebust 1982f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 1983f4ce1299STrond Myklebust 19845bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 19856b196875SChuck Lever if (ret) 19866b196875SChuck Lever goto out; 19875bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 19886b196875SChuck Lever if (ret < 0) 19896b196875SChuck Lever goto out; 19905bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 19916b196875SChuck Lever ret = 0; 1992f4ce1299STrond Myklebust 19936b196875SChuck Lever out: 1994f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 1995f4ce1299STrond Myklebust return ret; 19961c75950bSTrond Myklebust } 1997ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 19981c75950bSTrond Myklebust 19991b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 20001b3b4a1aSTrond Myklebust { 20011b3b4a1aSTrond Myklebust struct nfs_page *req; 20021b3b4a1aSTrond Myklebust int ret = 0; 20031b3b4a1aSTrond Myklebust 2004ba8b06e6STrond Myklebust wait_on_page_writeback(page); 20053e217045SWeston Andros Adamson 20063e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 20073e217045SWeston Andros Adamson * request */ 2008*6d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 20093e217045SWeston Andros Adamson 20103e217045SWeston Andros Adamson if (IS_ERR(req)) { 20113e217045SWeston Andros Adamson ret = PTR_ERR(req); 20123e217045SWeston Andros Adamson } else if (req) { 20133e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 20143e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 20153e217045SWeston Andros Adamson * request from the inode / page_private pointer and 20163e217045SWeston Andros Adamson * release it */ 20171b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 20181d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 20191b3b4a1aSTrond Myklebust } 20203e217045SWeston Andros Adamson 20211b3b4a1aSTrond Myklebust return ret; 20221b3b4a1aSTrond Myklebust } 20231b3b4a1aSTrond Myklebust 20241c75950bSTrond Myklebust /* 20251c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 20261c75950bSTrond Myklebust */ 2027c373fff7STrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 20281c75950bSTrond Myklebust { 202929418aa4SMel Gorman loff_t range_start = page_file_offset(page); 203009cbfeafSKirill A. Shutemov loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); 20317f2f12d9STrond Myklebust struct writeback_control wbc = { 20327f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 20337f2f12d9STrond Myklebust .nr_to_write = 0, 20347f2f12d9STrond Myklebust .range_start = range_start, 20357f2f12d9STrond Myklebust .range_end = range_end, 20367f2f12d9STrond Myklebust }; 20377f2f12d9STrond Myklebust int ret; 20387f2f12d9STrond Myklebust 2039f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 2040f4ce1299STrond Myklebust 20410522f6adSTrond Myklebust for (;;) { 2042ba8b06e6STrond Myklebust wait_on_page_writeback(page); 20437f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 2044c373fff7STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 20457f2f12d9STrond Myklebust if (ret < 0) 20467f2f12d9STrond Myklebust goto out_error; 20470522f6adSTrond Myklebust continue; 20487f2f12d9STrond Myklebust } 2049f4ce1299STrond Myklebust ret = 0; 20500522f6adSTrond Myklebust if (!PagePrivate(page)) 20510522f6adSTrond Myklebust break; 20520522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20537f2f12d9STrond Myklebust if (ret < 0) 20547f2f12d9STrond Myklebust goto out_error; 20557f2f12d9STrond Myklebust } 20567f2f12d9STrond Myklebust out_error: 2057f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 20587f2f12d9STrond Myklebust return ret; 20591c75950bSTrond Myklebust } 20601c75950bSTrond Myklebust 2061074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 2062074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 2063a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 2064074cc1deSTrond Myklebust { 20652da95652SJeff Layton /* 20662da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 20672da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 20682da95652SJeff Layton * 20692da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 20702da95652SJeff Layton * that we can safely release the inode reference while holding 20712da95652SJeff Layton * the page lock. 20722da95652SJeff Layton */ 20732da95652SJeff Layton if (PagePrivate(page)) 20742da95652SJeff Layton return -EBUSY; 2075074cc1deSTrond Myklebust 20768c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 20778c209ce7SDavid Howells return -EBUSY; 2078074cc1deSTrond Myklebust 2079a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 2080074cc1deSTrond Myklebust } 2081074cc1deSTrond Myklebust #endif 2082074cc1deSTrond Myklebust 2083f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 20841da177e4SLinus Torvalds { 20851da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 20861e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 20871da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 208820c2df83SPaul Mundt NULL); 20891da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 20901da177e4SLinus Torvalds return -ENOMEM; 20911da177e4SLinus Torvalds 209293d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 20931da177e4SLinus Torvalds nfs_wdata_cachep); 20941da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 20953dd4765fSJeff Layton goto out_destroy_write_cache; 20961da177e4SLinus Torvalds 20970b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 20980b7c0153SFred Isaman sizeof(struct nfs_commit_data), 20990b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 21000b7c0153SFred Isaman NULL); 21010b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 21023dd4765fSJeff Layton goto out_destroy_write_mempool; 21030b7c0153SFred Isaman 210493d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 21054c100210SYanchuan Nian nfs_cdata_cachep); 21061da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 21073dd4765fSJeff Layton goto out_destroy_commit_cache; 21081da177e4SLinus Torvalds 210989a09141SPeter Zijlstra /* 211089a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 211189a09141SPeter Zijlstra * 211289a09141SPeter Zijlstra * 64MB: 8192k 211389a09141SPeter Zijlstra * 128MB: 11585k 211489a09141SPeter Zijlstra * 256MB: 16384k 211589a09141SPeter Zijlstra * 512MB: 23170k 211689a09141SPeter Zijlstra * 1GB: 32768k 211789a09141SPeter Zijlstra * 2GB: 46340k 211889a09141SPeter Zijlstra * 4GB: 65536k 211989a09141SPeter Zijlstra * 8GB: 92681k 212089a09141SPeter Zijlstra * 16GB: 131072k 212189a09141SPeter Zijlstra * 212289a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 212389a09141SPeter Zijlstra * Limit the default to 256M 212489a09141SPeter Zijlstra */ 212589a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 212689a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 212789a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 212889a09141SPeter Zijlstra 21291da177e4SLinus Torvalds return 0; 21303dd4765fSJeff Layton 21313dd4765fSJeff Layton out_destroy_commit_cache: 21323dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21333dd4765fSJeff Layton out_destroy_write_mempool: 21343dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 21353dd4765fSJeff Layton out_destroy_write_cache: 21363dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 21373dd4765fSJeff Layton return -ENOMEM; 21381da177e4SLinus Torvalds } 21391da177e4SLinus Torvalds 2140266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 21411da177e4SLinus Torvalds { 21421da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 21433dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21441da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 21451a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 21461da177e4SLinus Torvalds } 21471da177e4SLinus Torvalds 21484a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 21494a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 21504a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 21510eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 21520eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 21531ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 21544a0de55cSAnna Schumaker }; 2155