6 * Partial copy of Linus' read cache modifications to fs/nfs/file.c
10 #include <linux/time.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/fcntl.h>
14 #include <linux/stat.h>
16 #include <linux/slab.h>
17 #include <linux/pagemap.h>
18 #include <linux/sunrpc/clnt.h>
19 #include <linux/nfs_fs.h>
20 #include <linux/nfs_page.h>
22 #include <asm/system.h>
29 #define NFSDBG_FACILITY NFSDBG_PAGECACHE
31 static int nfs_pagein_multi(struct inode *, struct list_head *, unsigned int, size_t, int);
32 static int nfs_pagein_one(struct inode *, struct list_head *, unsigned int, size_t, int);
33 static const struct rpc_call_ops nfs_read_partial_ops;
34 static const struct rpc_call_ops nfs_read_full_ops;
36 static struct kmem_cache *nfs_rdata_cachep;
37 static mempool_t *nfs_rdata_mempool;
39 #define MIN_POOL_READ (32)
41 struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount)
43 struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, GFP_NOFS);
46 memset(p, 0, sizeof(*p));
47 INIT_LIST_HEAD(&p->pages);
48 p->npages = pagecount;
49 p->res.seq_res.sr_slotid = NFS4_MAX_SLOT_TABLE;
50 if (pagecount <= ARRAY_SIZE(p->page_array))
51 p->pagevec = p->page_array;
53 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
55 mempool_free(p, nfs_rdata_mempool);
63 void nfs_readdata_free(struct nfs_read_data *p)
65 if (p && (p->pagevec != &p->page_array[0]))
67 mempool_free(p, nfs_rdata_mempool);
70 static void nfs_readdata_release(struct nfs_read_data *rdata)
72 put_nfs_open_context(rdata->args.context);
73 nfs_readdata_free(rdata);
77 int nfs_return_empty_page(struct page *page)
79 zero_user(page, 0, PAGE_CACHE_SIZE);
80 SetPageUptodate(page);
85 static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data *data)
87 unsigned int remainder = data->args.count - data->res.count;
88 unsigned int base = data->args.pgbase + data->res.count;
92 if (data->res.eof == 0 || remainder == 0)
95 * Note: "remainder" can never be negative, since we check for
96 * this in the XDR code.
98 pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
99 base &= ~PAGE_CACHE_MASK;
100 pglen = PAGE_CACHE_SIZE - base;
102 if (remainder <= pglen) {
103 zero_user(*pages, base, remainder);
106 zero_user(*pages, base, pglen);
109 pglen = PAGE_CACHE_SIZE;
114 int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
117 LIST_HEAD(one_request);
118 struct nfs_page *new;
121 len = nfs_page_length(page);
123 return nfs_return_empty_page(page);
124 new = nfs_create_request(ctx, inode, page, 0, len);
129 if (len < PAGE_CACHE_SIZE)
130 zero_user_segment(page, len, PAGE_CACHE_SIZE);
132 nfs_list_add_request(new, &one_request);
133 if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE)
134 nfs_pagein_multi(inode, &one_request, 1, len, 0);
136 nfs_pagein_one(inode, &one_request, 1, len, 0);
140 static void nfs_readpage_release(struct nfs_page *req)
142 struct inode *d_inode = req->wb_context->path.dentry->d_inode;
144 if (PageUptodate(req->wb_page))
145 nfs_readpage_to_fscache(d_inode, req->wb_page, 0);
147 unlock_page(req->wb_page);
149 dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
150 req->wb_context->path.dentry->d_inode->i_sb->s_id,
151 (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
153 (long long)req_offset(req));
154 nfs_clear_request(req);
155 nfs_release_request(req);
159 * Set up the NFS read request struct
161 static int nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data,
162 const struct rpc_call_ops *call_ops,
163 unsigned int count, unsigned int offset)
165 struct inode *inode = req->wb_context->path.dentry->d_inode;
166 int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0;
167 struct rpc_task *task;
168 struct rpc_message msg = {
169 .rpc_argp = &data->args,
170 .rpc_resp = &data->res,
171 .rpc_cred = req->wb_context->cred,
173 struct rpc_task_setup task_setup_data = {
175 .rpc_client = NFS_CLIENT(inode),
177 .callback_ops = call_ops,
178 .callback_data = data,
179 .workqueue = nfsiod_workqueue,
180 .flags = RPC_TASK_ASYNC | swap_flags,
185 data->cred = msg.rpc_cred;
187 data->args.fh = NFS_FH(inode);
188 data->args.offset = req_offset(req) + offset;
189 data->args.pgbase = req->wb_pgbase + offset;
190 data->args.pages = data->pagevec;
191 data->args.count = count;
192 data->args.context = get_nfs_open_context(req->wb_context);
194 data->res.fattr = &data->fattr;
195 data->res.count = count;
197 nfs_fattr_init(&data->fattr);
199 /* Set up the initial task struct. */
200 NFS_PROTO(inode)->read_setup(data, &msg);
202 dprintk("NFS: %5u initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n",
205 (long long)NFS_FILEID(inode),
207 (unsigned long long)data->args.offset);
209 task = rpc_run_task(&task_setup_data);
211 return PTR_ERR(task);
217 nfs_async_read_error(struct list_head *head)
219 struct nfs_page *req;
221 while (!list_empty(head)) {
222 req = nfs_list_entry(head->next);
223 nfs_list_remove_request(req);
224 SetPageError(req->wb_page);
225 nfs_readpage_release(req);
230 * Generate multiple requests to fill a single page.
232 * We optimize to reduce the number of read operations on the wire. If we
233 * detect that we're reading a page, or an area of a page, that is past the
234 * end of file, we do not generate NFS read operations but just clear the
235 * parts of the page that would have come back zero from the server anyway.
237 * We rely on the cached value of i_size to make this determination; another
238 * client can fill pages on the server past our cached end-of-file, but we
239 * won't see the new data until our attribute cache is updated. This is more
240 * or less conventional NFS client behavior.
242 static int nfs_pagein_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int flags)
244 struct nfs_page *req = nfs_list_entry(head->next);
245 struct page *page = req->wb_page;
246 struct nfs_read_data *data;
247 size_t rsize = NFS_SERVER(inode)->rsize, nbytes;
253 nfs_list_remove_request(req);
257 size_t len = min(nbytes,rsize);
259 data = nfs_readdata_alloc(1);
262 list_add(&data->pages, &list);
265 } while(nbytes != 0);
266 atomic_set(&req->wb_complete, requests);
268 ClearPageError(page);
274 data = list_entry(list.next, struct nfs_read_data, pages);
275 list_del_init(&data->pages);
277 data->pagevec[0] = page;
281 ret2 = nfs_read_rpcsetup(req, data, &nfs_read_partial_ops,
287 } while (nbytes != 0);
292 while (!list_empty(&list)) {
293 data = list_entry(list.next, struct nfs_read_data, pages);
294 list_del(&data->pages);
295 nfs_readdata_free(data);
298 nfs_readpage_release(req);
302 static int nfs_pagein_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int flags)
304 struct nfs_page *req;
306 struct nfs_read_data *data;
309 data = nfs_readdata_alloc(npages);
313 pages = data->pagevec;
314 while (!list_empty(head)) {
315 req = nfs_list_entry(head->next);
316 nfs_list_remove_request(req);
317 nfs_list_add_request(req, &data->pages);
318 ClearPageError(req->wb_page);
319 *pages++ = req->wb_page;
321 req = nfs_list_entry(data->pages.next);
323 return nfs_read_rpcsetup(req, data, &nfs_read_full_ops, count, 0);
325 nfs_async_read_error(head);
330 * This is the callback from RPC telling us whether a reply was
331 * received or some error occurred (timeout or socket shutdown).
333 int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data)
337 dprintk("NFS: %s: %5u, (status %d)\n", __func__, task->tk_pid,
340 status = NFS_PROTO(data->inode)->read_done(task, data);
344 nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, data->res.count);
346 if (task->tk_status == -ESTALE) {
347 set_bit(NFS_INO_STALE, &NFS_I(data->inode)->flags);
348 nfs_mark_for_revalidate(data->inode);
353 static void nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data)
355 struct nfs_readargs *argp = &data->args;
356 struct nfs_readres *resp = &data->res;
358 if (resp->eof || resp->count == argp->count)
361 /* This is a short read! */
362 nfs_inc_stats(data->inode, NFSIOS_SHORTREAD);
363 /* Has the server at least made some progress? */
364 if (resp->count == 0)
367 /* Yes, so retry the read at the end of the data */
368 argp->offset += resp->count;
369 argp->pgbase += resp->count;
370 argp->count -= resp->count;
371 nfs4_restart_rpc(task, NFS_SERVER(data->inode)->nfs_client);
374 nfs4_sequence_free_slot(NFS_SERVER(data->inode)->nfs_client,
381 * Handle a read reply that fills part of a page.
383 static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata)
385 struct nfs_read_data *data = calldata;
387 if (nfs_readpage_result(task, data) != 0)
389 if (task->tk_status < 0)
392 nfs_readpage_truncate_uninitialised_page(data);
393 nfs_readpage_retry(task, data);
396 static void nfs_readpage_release_partial(void *calldata)
398 struct nfs_read_data *data = calldata;
399 struct nfs_page *req = data->req;
400 struct page *page = req->wb_page;
401 int status = data->task.tk_status;
406 if (atomic_dec_and_test(&req->wb_complete)) {
407 if (!PageError(page))
408 SetPageUptodate(page);
409 nfs_readpage_release(req);
411 nfs_readdata_release(calldata);
414 #if defined(CONFIG_NFS_V4_1)
415 void nfs_read_prepare(struct rpc_task *task, void *calldata)
417 struct nfs_read_data *data = calldata;
419 if (nfs4_setup_sequence(NFS_SERVER(data->inode)->nfs_client,
420 &data->args.seq_args, &data->res.seq_res,
423 rpc_call_start(task);
425 #endif /* CONFIG_NFS_V4_1 */
427 static const struct rpc_call_ops nfs_read_partial_ops = {
428 #if defined(CONFIG_NFS_V4_1)
429 .rpc_call_prepare = nfs_read_prepare,
430 #endif /* CONFIG_NFS_V4_1 */
431 .rpc_call_done = nfs_readpage_result_partial,
432 .rpc_release = nfs_readpage_release_partial,
435 static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data)
437 unsigned int count = data->res.count;
438 unsigned int base = data->args.pgbase;
442 count = data->args.count;
443 if (unlikely(count == 0))
445 pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
446 base &= ~PAGE_CACHE_MASK;
448 for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++)
449 SetPageUptodate(*pages);
452 /* Was this a short read? */
453 if (data->res.eof || data->res.count == data->args.count)
454 SetPageUptodate(*pages);
458 * This is the callback from RPC telling us whether a reply was
459 * received or some error occurred (timeout or socket shutdown).
461 static void nfs_readpage_result_full(struct rpc_task *task, void *calldata)
463 struct nfs_read_data *data = calldata;
465 if (nfs_readpage_result(task, data) != 0)
467 if (task->tk_status < 0)
470 * Note: nfs_readpage_retry may change the values of
471 * data->args. In the multi-page case, we therefore need
472 * to ensure that we call nfs_readpage_set_pages_uptodate()
475 nfs_readpage_truncate_uninitialised_page(data);
476 nfs_readpage_set_pages_uptodate(data);
477 nfs_readpage_retry(task, data);
480 static void nfs_readpage_release_full(void *calldata)
482 struct nfs_read_data *data = calldata;
484 while (!list_empty(&data->pages)) {
485 struct nfs_page *req = nfs_list_entry(data->pages.next);
487 nfs_list_remove_request(req);
488 nfs_readpage_release(req);
490 nfs_readdata_release(calldata);
493 static const struct rpc_call_ops nfs_read_full_ops = {
494 #if defined(CONFIG_NFS_V4_1)
495 .rpc_call_prepare = nfs_read_prepare,
496 #endif /* CONFIG_NFS_V4_1 */
497 .rpc_call_done = nfs_readpage_result_full,
498 .rpc_release = nfs_readpage_release_full,
502 * Read a page over NFS.
503 * We read the page synchronously in the following case:
504 * - The error flag is set for this page. This happens only when a
505 * previous async read operation failed.
507 int nfs_readpage(struct file *file, struct page *page)
509 struct nfs_open_context *ctx;
510 struct inode *inode = page->mapping->host;
513 dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
514 page, PAGE_CACHE_SIZE, page->index);
515 nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
516 nfs_add_stats(inode, NFSIOS_READPAGES, 1);
519 * Try to flush any pending writes to the file..
521 * NOTE! Because we own the page lock, there cannot
522 * be any new pending writes generated at this point
523 * for this page (other pages can be written to).
525 error = nfs_wb_page(inode, page);
528 if (PageUptodate(page))
532 if (NFS_STALE(inode))
537 ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
541 ctx = get_nfs_open_context(nfs_file_open_context(file));
543 if (!IS_SYNC(inode)) {
544 error = nfs_readpage_from_fscache(ctx, inode, page);
549 error = nfs_readpage_async(ctx, inode, page);
552 put_nfs_open_context(ctx);
559 struct nfs_readdesc {
560 struct nfs_pageio_descriptor *pgio;
561 struct nfs_open_context *ctx;
565 readpage_async_filler(void *data, struct page *page)
567 struct nfs_readdesc *desc = (struct nfs_readdesc *)data;
568 struct inode *inode = page->mapping->host;
569 struct nfs_page *new;
573 len = nfs_page_length(page);
575 return nfs_return_empty_page(page);
577 new = nfs_create_request(desc->ctx, inode, page, 0, len);
581 if (len < PAGE_CACHE_SIZE)
582 zero_user_segment(page, len, PAGE_CACHE_SIZE);
583 if (!nfs_pageio_add_request(desc->pgio, new)) {
584 error = desc->pgio->pg_error;
589 error = PTR_ERR(new);
596 int nfs_readpages(struct file *filp, struct address_space *mapping,
597 struct list_head *pages, unsigned nr_pages)
599 struct nfs_pageio_descriptor pgio;
600 struct nfs_readdesc desc = {
603 struct inode *inode = mapping->host;
604 struct nfs_server *server = NFS_SERVER(inode);
605 size_t rsize = server->rsize;
606 unsigned long npages;
609 dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
611 (long long)NFS_FILEID(inode),
613 nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
615 if (NFS_STALE(inode))
619 desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
620 if (desc.ctx == NULL)
623 desc.ctx = get_nfs_open_context(nfs_file_open_context(filp));
625 /* attempt to read as many of the pages as possible from the cache
626 * - this returns -ENOBUFS immediately if the cookie is negative
628 ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping,
631 goto read_complete; /* all pages were read */
633 if (rsize < PAGE_CACHE_SIZE)
634 nfs_pageio_init(&pgio, inode, nfs_pagein_multi, rsize, 0);
636 nfs_pageio_init(&pgio, inode, nfs_pagein_one, rsize, 0);
638 ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
640 nfs_pageio_complete(&pgio);
641 npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
642 nfs_add_stats(inode, NFSIOS_READPAGES, npages);
644 put_nfs_open_context(desc.ctx);
649 int __init nfs_init_readpagecache(void)
651 nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
652 sizeof(struct nfs_read_data),
653 0, SLAB_HWCACHE_ALIGN,
655 if (nfs_rdata_cachep == NULL)
658 nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ,
660 if (nfs_rdata_mempool == NULL)
666 void nfs_destroy_readpagecache(void)
668 mempool_destroy(nfs_rdata_mempool);
669 kmem_cache_destroy(nfs_rdata_cachep);