xref: /openbmc/linux/fs/nfs/pagelist.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
1 /*
2  * linux/fs/nfs/pagelist.c
3  *
4  * A set of helper functions for managing NFS read and write requests.
5  * The main purpose of these routines is to provide support for the
6  * coalescing of several requests into a single RPC call.
7  *
8  * Copyright 2000, 2001 (c) Trond Myklebust <trond.myklebust@fys.uio.no>
9  *
10  */
11 
12 #include <linux/slab.h>
13 #include <linux/file.h>
14 #include <linux/sunrpc/clnt.h>
15 #include <linux/nfs3.h>
16 #include <linux/nfs4.h>
17 #include <linux/nfs_page.h>
18 #include <linux/nfs_fs.h>
19 #include <linux/nfs_mount.h>
20 
21 #include "internal.h"
22 
23 static struct kmem_cache *nfs_page_cachep;
24 
25 static inline struct nfs_page *
26 nfs_page_alloc(void)
27 {
28 	struct nfs_page	*p;
29 	p = kmem_cache_alloc(nfs_page_cachep, GFP_KERNEL);
30 	if (p) {
31 		memset(p, 0, sizeof(*p));
32 		INIT_LIST_HEAD(&p->wb_list);
33 	}
34 	return p;
35 }
36 
37 static inline void
38 nfs_page_free(struct nfs_page *p)
39 {
40 	kmem_cache_free(nfs_page_cachep, p);
41 }
42 
43 /**
44  * nfs_create_request - Create an NFS read/write request.
45  * @file: file descriptor to use
46  * @inode: inode to which the request is attached
47  * @page: page to write
48  * @offset: starting offset within the page for the write
49  * @count: number of bytes to read/write
50  *
51  * The page must be locked by the caller. This makes sure we never
52  * create two different requests for the same page.
53  * User should ensure it is safe to sleep in this function.
54  */
55 struct nfs_page *
56 nfs_create_request(struct nfs_open_context *ctx, struct inode *inode,
57 		   struct page *page,
58 		   unsigned int offset, unsigned int count)
59 {
60 	struct nfs_server *server = NFS_SERVER(inode);
61 	struct nfs_page		*req;
62 
63 	for (;;) {
64 		/* try to allocate the request struct */
65 		req = nfs_page_alloc();
66 		if (req != NULL)
67 			break;
68 
69 		if (signalled() && (server->flags & NFS_MOUNT_INTR))
70 			return ERR_PTR(-ERESTARTSYS);
71 		yield();
72 	}
73 
74 	/* Initialize the request struct. Initially, we assume a
75 	 * long write-back delay. This will be adjusted in
76 	 * update_nfs_request below if the region is not locked. */
77 	req->wb_page    = page;
78 	atomic_set(&req->wb_complete, 0);
79 	req->wb_index	= page->index;
80 	page_cache_get(page);
81 	BUG_ON(PagePrivate(page));
82 	BUG_ON(!PageLocked(page));
83 	BUG_ON(page->mapping->host != inode);
84 	req->wb_offset  = offset;
85 	req->wb_pgbase	= offset;
86 	req->wb_bytes   = count;
87 	atomic_set(&req->wb_count, 1);
88 	req->wb_context = get_nfs_open_context(ctx);
89 
90 	return req;
91 }
92 
93 /**
94  * nfs_unlock_request - Unlock request and wake up sleepers.
95  * @req:
96  */
97 void nfs_unlock_request(struct nfs_page *req)
98 {
99 	if (!NFS_WBACK_BUSY(req)) {
100 		printk(KERN_ERR "NFS: Invalid unlock attempted\n");
101 		BUG();
102 	}
103 	smp_mb__before_clear_bit();
104 	clear_bit(PG_BUSY, &req->wb_flags);
105 	smp_mb__after_clear_bit();
106 	wake_up_bit(&req->wb_flags, PG_BUSY);
107 	nfs_release_request(req);
108 }
109 
110 /**
111  * nfs_set_page_writeback_locked - Lock a request for writeback
112  * @req:
113  */
114 int nfs_set_page_writeback_locked(struct nfs_page *req)
115 {
116 	struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode);
117 
118 	if (!nfs_lock_request(req))
119 		return 0;
120 	radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_WRITEBACK);
121 	return 1;
122 }
123 
124 /**
125  * nfs_clear_page_writeback - Unlock request and wake up sleepers
126  */
127 void nfs_clear_page_writeback(struct nfs_page *req)
128 {
129 	struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode);
130 
131 	if (req->wb_page != NULL) {
132 		spin_lock(&nfsi->req_lock);
133 		radix_tree_tag_clear(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_WRITEBACK);
134 		spin_unlock(&nfsi->req_lock);
135 	}
136 	nfs_unlock_request(req);
137 }
138 
139 /**
140  * nfs_clear_request - Free up all resources allocated to the request
141  * @req:
142  *
143  * Release page resources associated with a write request after it
144  * has completed.
145  */
146 void nfs_clear_request(struct nfs_page *req)
147 {
148 	struct page *page = req->wb_page;
149 	if (page != NULL) {
150 		page_cache_release(page);
151 		req->wb_page = NULL;
152 	}
153 }
154 
155 
156 /**
157  * nfs_release_request - Release the count on an NFS read/write request
158  * @req: request to release
159  *
160  * Note: Should never be called with the spinlock held!
161  */
162 void
163 nfs_release_request(struct nfs_page *req)
164 {
165 	if (!atomic_dec_and_test(&req->wb_count))
166 		return;
167 
168 	/* Release struct file or cached credential */
169 	nfs_clear_request(req);
170 	put_nfs_open_context(req->wb_context);
171 	nfs_page_free(req);
172 }
173 
174 static int nfs_wait_bit_interruptible(void *word)
175 {
176 	int ret = 0;
177 
178 	if (signal_pending(current))
179 		ret = -ERESTARTSYS;
180 	else
181 		schedule();
182 	return ret;
183 }
184 
185 /**
186  * nfs_wait_on_request - Wait for a request to complete.
187  * @req: request to wait upon.
188  *
189  * Interruptible by signals only if mounted with intr flag.
190  * The user is responsible for holding a count on the request.
191  */
192 int
193 nfs_wait_on_request(struct nfs_page *req)
194 {
195         struct rpc_clnt	*clnt = NFS_CLIENT(req->wb_context->dentry->d_inode);
196 	sigset_t oldmask;
197 	int ret = 0;
198 
199 	if (!test_bit(PG_BUSY, &req->wb_flags))
200 		goto out;
201 	/*
202 	 * Note: the call to rpc_clnt_sigmask() suffices to ensure that we
203 	 *	 are not interrupted if intr flag is not set
204 	 */
205 	rpc_clnt_sigmask(clnt, &oldmask);
206 	ret = out_of_line_wait_on_bit(&req->wb_flags, PG_BUSY,
207 			nfs_wait_bit_interruptible, TASK_INTERRUPTIBLE);
208 	rpc_clnt_sigunmask(clnt, &oldmask);
209 out:
210 	return ret;
211 }
212 
213 /**
214  * nfs_pageio_init - initialise a page io descriptor
215  * @desc: pointer to descriptor
216  * @inode: pointer to inode
217  * @doio: pointer to io function
218  * @bsize: io block size
219  * @io_flags: extra parameters for the io function
220  */
221 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
222 		     struct inode *inode,
223 		     int (*doio)(struct inode *, struct list_head *, unsigned int, size_t, int),
224 		     size_t bsize,
225 		     int io_flags)
226 {
227 	INIT_LIST_HEAD(&desc->pg_list);
228 	desc->pg_bytes_written = 0;
229 	desc->pg_count = 0;
230 	desc->pg_bsize = bsize;
231 	desc->pg_base = 0;
232 	desc->pg_inode = inode;
233 	desc->pg_doio = doio;
234 	desc->pg_ioflags = io_flags;
235 	desc->pg_error = 0;
236 }
237 
238 /**
239  * nfs_can_coalesce_requests - test two requests for compatibility
240  * @prev: pointer to nfs_page
241  * @req: pointer to nfs_page
242  *
243  * The nfs_page structures 'prev' and 'req' are compared to ensure that the
244  * page data area they describe is contiguous, and that their RPC
245  * credentials, NFSv4 open state, and lockowners are the same.
246  *
247  * Return 'true' if this is the case, else return 'false'.
248  */
249 static int nfs_can_coalesce_requests(struct nfs_page *prev,
250 				     struct nfs_page *req)
251 {
252 	if (req->wb_context->cred != prev->wb_context->cred)
253 		return 0;
254 	if (req->wb_context->lockowner != prev->wb_context->lockowner)
255 		return 0;
256 	if (req->wb_context->state != prev->wb_context->state)
257 		return 0;
258 	if (req->wb_index != (prev->wb_index + 1))
259 		return 0;
260 	if (req->wb_pgbase != 0)
261 		return 0;
262 	if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
263 		return 0;
264 	return 1;
265 }
266 
267 /**
268  * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
269  * @desc: destination io descriptor
270  * @req: request
271  *
272  * Returns true if the request 'req' was successfully coalesced into the
273  * existing list of pages 'desc'.
274  */
275 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
276 				     struct nfs_page *req)
277 {
278 	size_t newlen = req->wb_bytes;
279 
280 	if (desc->pg_count != 0) {
281 		struct nfs_page *prev;
282 
283 		/*
284 		 * FIXME: ideally we should be able to coalesce all requests
285 		 * that are not block boundary aligned, but currently this
286 		 * is problematic for the case of bsize < PAGE_CACHE_SIZE,
287 		 * since nfs_flush_multi and nfs_pagein_multi assume you
288 		 * can have only one struct nfs_page.
289 		 */
290 		if (desc->pg_bsize < PAGE_SIZE)
291 			return 0;
292 		newlen += desc->pg_count;
293 		if (newlen > desc->pg_bsize)
294 			return 0;
295 		prev = nfs_list_entry(desc->pg_list.prev);
296 		if (!nfs_can_coalesce_requests(prev, req))
297 			return 0;
298 	} else
299 		desc->pg_base = req->wb_pgbase;
300 	nfs_list_remove_request(req);
301 	nfs_list_add_request(req, &desc->pg_list);
302 	desc->pg_count = newlen;
303 	return 1;
304 }
305 
306 /*
307  * Helper for nfs_pageio_add_request and nfs_pageio_complete
308  */
309 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
310 {
311 	if (!list_empty(&desc->pg_list)) {
312 		int error = desc->pg_doio(desc->pg_inode,
313 					  &desc->pg_list,
314 					  nfs_page_array_len(desc->pg_base,
315 							     desc->pg_count),
316 					  desc->pg_count,
317 					  desc->pg_ioflags);
318 		if (error < 0)
319 			desc->pg_error = error;
320 		else
321 			desc->pg_bytes_written += desc->pg_count;
322 	}
323 	if (list_empty(&desc->pg_list)) {
324 		desc->pg_count = 0;
325 		desc->pg_base = 0;
326 	}
327 }
328 
329 /**
330  * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
331  * @desc: destination io descriptor
332  * @req: request
333  *
334  * Returns true if the request 'req' was successfully coalesced into the
335  * existing list of pages 'desc'.
336  */
337 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
338 			   struct nfs_page *req)
339 {
340 	while (!nfs_pageio_do_add_request(desc, req)) {
341 		nfs_pageio_doio(desc);
342 		if (desc->pg_error < 0)
343 			return 0;
344 	}
345 	return 1;
346 }
347 
348 /**
349  * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
350  * @desc: pointer to io descriptor
351  */
352 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
353 {
354 	nfs_pageio_doio(desc);
355 }
356 
357 #define NFS_SCAN_MAXENTRIES 16
358 /**
359  * nfs_scan_list - Scan a list for matching requests
360  * @nfsi: NFS inode
361  * @head: One of the NFS inode request lists
362  * @dst: Destination list
363  * @idx_start: lower bound of page->index to scan
364  * @npages: idx_start + npages sets the upper bound to scan.
365  *
366  * Moves elements from one of the inode request lists.
367  * If the number of requests is set to 0, the entire address_space
368  * starting at index idx_start, is scanned.
369  * The requests are *not* checked to ensure that they form a contiguous set.
370  * You must be holding the inode's req_lock when calling this function
371  */
372 int nfs_scan_list(struct nfs_inode *nfsi, struct list_head *head,
373 		struct list_head *dst, pgoff_t idx_start,
374 		unsigned int npages)
375 {
376 	struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
377 	struct nfs_page *req;
378 	pgoff_t idx_end;
379 	int found, i;
380 	int res;
381 
382 	res = 0;
383 	if (npages == 0)
384 		idx_end = ~0;
385 	else
386 		idx_end = idx_start + npages - 1;
387 
388 	for (;;) {
389 		found = radix_tree_gang_lookup(&nfsi->nfs_page_tree,
390 				(void **)&pgvec[0], idx_start,
391 				NFS_SCAN_MAXENTRIES);
392 		if (found <= 0)
393 			break;
394 		for (i = 0; i < found; i++) {
395 			req = pgvec[i];
396 			if (req->wb_index > idx_end)
397 				goto out;
398 			idx_start = req->wb_index + 1;
399 			if (req->wb_list_head != head)
400 				continue;
401 			if (nfs_set_page_writeback_locked(req)) {
402 				nfs_list_remove_request(req);
403 				nfs_list_add_request(req, dst);
404 				res++;
405 			}
406 		}
407 
408 	}
409 out:
410 	return res;
411 }
412 
413 int __init nfs_init_nfspagecache(void)
414 {
415 	nfs_page_cachep = kmem_cache_create("nfs_page",
416 					    sizeof(struct nfs_page),
417 					    0, SLAB_HWCACHE_ALIGN,
418 					    NULL, NULL);
419 	if (nfs_page_cachep == NULL)
420 		return -ENOMEM;
421 
422 	return 0;
423 }
424 
425 void nfs_destroy_nfspagecache(void)
426 {
427 	kmem_cache_destroy(nfs_page_cachep);
428 }
429 
430