blob: ed0db61f8543f0ee4d68c3b457510cedad2187ca [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/slab.h>
13#include <linux/file.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040014#include <linux/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/sunrpc/clnt.h>
Trond Myklebust1313e602012-01-17 22:04:24 -050016#include <linux/nfs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/nfs3.h>
18#include <linux/nfs4.h>
19#include <linux/nfs_page.h>
20#include <linux/nfs_fs.h>
21#include <linux/nfs_mount.h>
Paul Gortmakerafeacc82011-05-26 16:00:52 -040022#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
Trond Myklebust8d5658c2007-04-10 09:26:35 -040024#include "internal.h"
Fred Isamanbae724e2011-03-01 01:34:15 +000025#include "pnfs.h"
Trond Myklebust8d5658c2007-04-10 09:26:35 -040026
Anna Schumaker0eecb212014-05-06 09:12:32 -040027#define NFSDBG_FACILITY NFSDBG_PAGECACHE
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
Linus Torvalds1da177e2005-04-16 15:20:36 -070029static struct kmem_cache *nfs_page_cachep;
Anna Schumakeref2c4882014-05-06 09:12:36 -040030static const struct rpc_call_ops nfs_pgio_common_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Anna Schumaker00bfa302014-05-06 09:12:29 -040032static bool nfs_pgarray_set(struct nfs_page_array *p, unsigned int pagecount)
Fred Isaman30dd3742012-04-20 14:47:45 -040033{
34 p->npages = pagecount;
35 if (pagecount <= ARRAY_SIZE(p->page_array))
36 p->pagevec = p->page_array;
37 else {
38 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
39 if (!p->pagevec)
40 p->npages = 0;
41 }
42 return p->pagevec != NULL;
43}
44
Fred Isaman4db6e0b2012-04-20 14:47:46 -040045void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
46 struct nfs_pgio_header *hdr,
47 void (*release)(struct nfs_pgio_header *hdr))
48{
49 hdr->req = nfs_list_entry(desc->pg_list.next);
50 hdr->inode = desc->pg_inode;
51 hdr->cred = hdr->req->wb_context->cred;
52 hdr->io_start = req_offset(hdr->req);
53 hdr->good_bytes = desc->pg_count;
Fred Isaman584aa812012-04-20 14:47:51 -040054 hdr->dreq = desc->pg_dreq;
Peng Taof6166382012-08-02 15:36:09 +030055 hdr->layout_private = desc->pg_layout_private;
Fred Isaman4db6e0b2012-04-20 14:47:46 -040056 hdr->release = release;
Fred Isaman061ae2e2012-04-20 14:47:48 -040057 hdr->completion_ops = desc->pg_completion_ops;
Fred Isaman584aa812012-04-20 14:47:51 -040058 if (hdr->completion_ops->init_hdr)
59 hdr->completion_ops->init_hdr(hdr);
Fred Isaman4db6e0b2012-04-20 14:47:46 -040060}
Bryan Schumaker89d77c82012-07-30 16:05:25 -040061EXPORT_SYMBOL_GPL(nfs_pgheader_init);
Fred Isaman4db6e0b2012-04-20 14:47:46 -040062
63void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
64{
65 spin_lock(&hdr->lock);
66 if (pos < hdr->io_start + hdr->good_bytes) {
67 set_bit(NFS_IOHDR_ERROR, &hdr->flags);
68 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
69 hdr->good_bytes = pos - hdr->io_start;
70 hdr->error = error;
71 }
72 spin_unlock(&hdr->lock);
73}
74
Linus Torvalds1da177e2005-04-16 15:20:36 -070075static inline struct nfs_page *
76nfs_page_alloc(void)
77{
Mel Gorman192e5012012-07-31 16:45:16 -070078 struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_NOIO);
Jesper Juhl72895b12010-12-09 23:17:15 +010079 if (p)
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 INIT_LIST_HEAD(&p->wb_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 return p;
82}
83
84static inline void
85nfs_page_free(struct nfs_page *p)
86{
87 kmem_cache_free(nfs_page_cachep, p);
88}
89
Trond Myklebust577b4232013-04-08 21:38:12 -040090static void
91nfs_iocounter_inc(struct nfs_io_counter *c)
92{
93 atomic_inc(&c->io_count);
94}
95
96static void
97nfs_iocounter_dec(struct nfs_io_counter *c)
98{
99 if (atomic_dec_and_test(&c->io_count)) {
100 clear_bit(NFS_IO_INPROGRESS, &c->flags);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100101 smp_mb__after_atomic();
Trond Myklebust577b4232013-04-08 21:38:12 -0400102 wake_up_bit(&c->flags, NFS_IO_INPROGRESS);
103 }
104}
105
106static int
107__nfs_iocounter_wait(struct nfs_io_counter *c)
108{
109 wait_queue_head_t *wq = bit_waitqueue(&c->flags, NFS_IO_INPROGRESS);
110 DEFINE_WAIT_BIT(q, &c->flags, NFS_IO_INPROGRESS);
111 int ret = 0;
112
113 do {
114 prepare_to_wait(wq, &q.wait, TASK_KILLABLE);
115 set_bit(NFS_IO_INPROGRESS, &c->flags);
116 if (atomic_read(&c->io_count) == 0)
117 break;
NeilBrownc1221322014-07-07 15:16:04 +1000118 ret = nfs_wait_bit_killable(&q.key);
David Jeffery92a56552014-08-05 11:19:42 -0400119 } while (atomic_read(&c->io_count) != 0 && !ret);
Trond Myklebust577b4232013-04-08 21:38:12 -0400120 finish_wait(wq, &q.wait);
121 return ret;
122}
123
124/**
125 * nfs_iocounter_wait - wait for i/o to complete
126 * @c: nfs_io_counter to use
127 *
128 * returns -ERESTARTSYS if interrupted by a fatal signal.
129 * Otherwise returns 0 once the io_count hits 0.
130 */
131int
132nfs_iocounter_wait(struct nfs_io_counter *c)
133{
134 if (atomic_read(&c->io_count) == 0)
135 return 0;
136 return __nfs_iocounter_wait(c);
137}
138
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400139/*
140 * nfs_page_group_lock - lock the head of the page group
141 * @req - request in group that is to be locked
Weston Andros Adamsonfd2f3a02014-08-08 11:00:53 -0400142 * @nonblock - if true don't block waiting for lock
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400143 *
144 * this lock must be held if modifying the page group list
Weston Andros Adamsone7029202014-07-17 20:42:15 -0400145 *
Weston Andros Adamsonbc8a3092014-08-08 11:00:54 -0400146 * return 0 on success, < 0 on error: -EDELAY if nonblocking or the
147 * result from wait_on_bit_lock
148 *
149 * NOTE: calling with nonblock=false should always have set the
150 * lock bit (see fs/buffer.c and other uses of wait_on_bit_lock
151 * with TASK_UNINTERRUPTIBLE), so there is no need to check the result.
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400152 */
Weston Andros Adamsone7029202014-07-17 20:42:15 -0400153int
Weston Andros Adamsonfd2f3a02014-08-08 11:00:53 -0400154nfs_page_group_lock(struct nfs_page *req, bool nonblock)
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400155{
156 struct nfs_page *head = req->wb_head;
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400157
158 WARN_ON_ONCE(head != head->wb_head);
159
Weston Andros Adamsonbc8a3092014-08-08 11:00:54 -0400160 if (!test_and_set_bit(PG_HEADLOCK, &head->wb_flags))
161 return 0;
Weston Andros Adamsone7029202014-07-17 20:42:15 -0400162
Weston Andros Adamsonbc8a3092014-08-08 11:00:54 -0400163 if (!nonblock)
164 return wait_on_bit_lock(&head->wb_flags, PG_HEADLOCK,
165 TASK_UNINTERRUPTIBLE);
166
167 return -EAGAIN;
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400168}
169
170/*
Weston Andros Adamson7c3af972014-08-08 11:00:57 -0400171 * nfs_page_group_lock_wait - wait for the lock to clear, but don't grab it
172 * @req - a request in the group
173 *
174 * This is a blocking call to wait for the group lock to be cleared.
175 */
176void
177nfs_page_group_lock_wait(struct nfs_page *req)
178{
179 struct nfs_page *head = req->wb_head;
180
181 WARN_ON_ONCE(head != head->wb_head);
182
183 wait_on_bit(&head->wb_flags, PG_HEADLOCK,
184 TASK_UNINTERRUPTIBLE);
185}
186
187/*
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400188 * nfs_page_group_unlock - unlock the head of the page group
189 * @req - request in group that is to be unlocked
190 */
191void
192nfs_page_group_unlock(struct nfs_page *req)
193{
194 struct nfs_page *head = req->wb_head;
195
196 WARN_ON_ONCE(head != head->wb_head);
197
Linus Torvaldsd1e1cda2014-06-10 15:02:42 -0700198 smp_mb__before_atomic();
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400199 clear_bit(PG_HEADLOCK, &head->wb_flags);
Linus Torvaldsd1e1cda2014-06-10 15:02:42 -0700200 smp_mb__after_atomic();
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400201 wake_up_bit(&head->wb_flags, PG_HEADLOCK);
202}
203
204/*
205 * nfs_page_group_sync_on_bit_locked
206 *
207 * must be called with page group lock held
208 */
209static bool
210nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
211{
212 struct nfs_page *head = req->wb_head;
213 struct nfs_page *tmp;
214
215 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags));
216 WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags));
217
218 tmp = req->wb_this_page;
219 while (tmp != req) {
220 if (!test_bit(bit, &tmp->wb_flags))
221 return false;
222 tmp = tmp->wb_this_page;
223 }
224
225 /* true! reset all bits */
226 tmp = req;
227 do {
228 clear_bit(bit, &tmp->wb_flags);
229 tmp = tmp->wb_this_page;
230 } while (tmp != req);
231
232 return true;
233}
234
235/*
236 * nfs_page_group_sync_on_bit - set bit on current request, but only
237 * return true if the bit is set for all requests in page group
238 * @req - request in page group
239 * @bit - PG_* bit that is used to sync page group
240 */
241bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
242{
243 bool ret;
244
Weston Andros Adamsonfd2f3a02014-08-08 11:00:53 -0400245 nfs_page_group_lock(req, false);
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400246 ret = nfs_page_group_sync_on_bit_locked(req, bit);
247 nfs_page_group_unlock(req);
248
249 return ret;
250}
251
252/*
253 * nfs_page_group_init - Initialize the page group linkage for @req
254 * @req - a new nfs request
255 * @prev - the previous request in page group, or NULL if @req is the first
256 * or only request in the group (the head).
257 */
258static inline void
259nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
260{
261 WARN_ON_ONCE(prev == req);
262
263 if (!prev) {
Weston Andros Adamson85710a82014-07-11 10:20:46 -0400264 /* a head request */
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400265 req->wb_head = req;
266 req->wb_this_page = req;
267 } else {
Weston Andros Adamson85710a82014-07-11 10:20:46 -0400268 /* a subrequest */
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400269 WARN_ON_ONCE(prev->wb_this_page != prev->wb_head);
270 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags));
271 req->wb_head = prev->wb_head;
272 req->wb_this_page = prev->wb_this_page;
273 prev->wb_this_page = req;
274
Weston Andros Adamson85710a82014-07-11 10:20:46 -0400275 /* All subrequests take a ref on the head request until
276 * nfs_page_group_destroy is called */
277 kref_get(&req->wb_head->wb_kref);
278
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400279 /* grab extra ref if head request has extra ref from
280 * the write/commit path to handle handoff between write
281 * and commit lists */
Weston Andros Adamson17089a22014-07-11 10:20:45 -0400282 if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags)) {
283 set_bit(PG_INODE_REF, &req->wb_flags);
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400284 kref_get(&req->wb_kref);
Weston Andros Adamson17089a22014-07-11 10:20:45 -0400285 }
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400286 }
287}
288
289/*
290 * nfs_page_group_destroy - sync the destruction of page groups
291 * @req - request that no longer needs the page group
292 *
293 * releases the page group reference from each member once all
294 * members have called this function.
295 */
296static void
297nfs_page_group_destroy(struct kref *kref)
298{
299 struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
300 struct nfs_page *tmp, *next;
301
Weston Andros Adamson85710a82014-07-11 10:20:46 -0400302 /* subrequests must release the ref on the head request */
303 if (req->wb_head != req)
304 nfs_release_request(req->wb_head);
305
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400306 if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN))
307 return;
308
309 tmp = req;
310 do {
311 next = tmp->wb_this_page;
312 /* unlink and free */
313 tmp->wb_this_page = tmp;
314 tmp->wb_head = tmp;
315 nfs_free_request(tmp);
316 tmp = next;
317 } while (tmp != req);
318}
319
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320/**
321 * nfs_create_request - Create an NFS read/write request.
Chuck Leverc02f5572011-10-25 12:17:43 -0400322 * @ctx: open context to use
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 * @page: page to write
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400324 * @last: last nfs request created for this page group or NULL if head
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 * @offset: starting offset within the page for the write
326 * @count: number of bytes to read/write
327 *
328 * The page must be locked by the caller. This makes sure we never
Jason Uhlenkotta19b89c2007-04-26 17:25:51 -0700329 * create two different requests for the same page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 * User should ensure it is safe to sleep in this function.
331 */
332struct nfs_page *
Weston Andros Adamson8c8f1ac12014-05-15 11:56:42 -0400333nfs_create_request(struct nfs_open_context *ctx, struct page *page,
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400334 struct nfs_page *last, unsigned int offset,
335 unsigned int count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 struct nfs_page *req;
Trond Myklebustb3c54de2012-08-13 17:15:50 -0400338 struct nfs_lock_context *l_ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
Trond Myklebustc58c8442013-03-18 19:45:14 -0400340 if (test_bit(NFS_CONTEXT_BAD, &ctx->flags))
341 return ERR_PTR(-EBADF);
Trond Myklebust18eb8842010-05-13 12:51:02 -0400342 /* try to allocate the request struct */
343 req = nfs_page_alloc();
344 if (req == NULL)
345 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346
Jeff Layton015f0212010-10-28 10:10:37 -0400347 /* get lock context early so we can deal with alloc failures */
Trond Myklebustb3c54de2012-08-13 17:15:50 -0400348 l_ctx = nfs_get_lock_context(ctx);
349 if (IS_ERR(l_ctx)) {
Jeff Layton015f0212010-10-28 10:10:37 -0400350 nfs_page_free(req);
Trond Myklebustb3c54de2012-08-13 17:15:50 -0400351 return ERR_CAST(l_ctx);
Jeff Layton015f0212010-10-28 10:10:37 -0400352 }
Trond Myklebustb3c54de2012-08-13 17:15:50 -0400353 req->wb_lock_context = l_ctx;
Trond Myklebust577b4232013-04-08 21:38:12 -0400354 nfs_iocounter_inc(&l_ctx->io_count);
Jeff Layton015f0212010-10-28 10:10:37 -0400355
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 /* Initialize the request struct. Initially, we assume a
357 * long write-back delay. This will be adjusted in
358 * update_nfs_request below if the region is not locked. */
359 req->wb_page = page;
Mel Gormand56b4dd2012-07-31 16:45:06 -0700360 req->wb_index = page_file_index(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 page_cache_get(page);
362 req->wb_offset = offset;
363 req->wb_pgbase = offset;
364 req->wb_bytes = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 req->wb_context = get_nfs_open_context(ctx);
Trond Myklebustc03b4022007-06-17 13:26:38 -0400366 kref_init(&req->wb_kref);
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400367 nfs_page_group_init(req, last);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 return req;
369}
370
371/**
Trond Myklebust1d1afcb2012-05-09 14:04:55 -0400372 * nfs_unlock_request - Unlock request and wake up sleepers.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 * @req:
374 */
Trond Myklebust1d1afcb2012-05-09 14:04:55 -0400375void nfs_unlock_request(struct nfs_page *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376{
377 if (!NFS_WBACK_BUSY(req)) {
378 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
379 BUG();
380 }
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100381 smp_mb__before_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 clear_bit(PG_BUSY, &req->wb_flags);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100383 smp_mb__after_atomic();
Trond Myklebust464a98b2005-06-22 17:16:21 +0000384 wake_up_bit(&req->wb_flags, PG_BUSY);
Trond Myklebust3aff4eb2012-05-09 14:30:35 -0400385}
386
387/**
Trond Myklebust1d1afcb2012-05-09 14:04:55 -0400388 * nfs_unlock_and_release_request - Unlock request and release the nfs_page
389 * @req:
Trond Myklebust3aff4eb2012-05-09 14:30:35 -0400390 */
Trond Myklebust1d1afcb2012-05-09 14:04:55 -0400391void nfs_unlock_and_release_request(struct nfs_page *req)
Trond Myklebust3aff4eb2012-05-09 14:30:35 -0400392{
Trond Myklebust1d1afcb2012-05-09 14:04:55 -0400393 nfs_unlock_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 nfs_release_request(req);
395}
396
Trond Myklebust4d65c522011-03-25 14:15:11 -0400397/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 * nfs_clear_request - Free up all resources allocated to the request
399 * @req:
400 *
Trond Myklebustbb6fbc42010-03-11 09:19:35 -0500401 * Release page and open context resources associated with a read/write
402 * request after it has completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 */
Trond Myklebust4d65c522011-03-25 14:15:11 -0400404static void nfs_clear_request(struct nfs_page *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405{
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500406 struct page *page = req->wb_page;
Trond Myklebustbb6fbc42010-03-11 09:19:35 -0500407 struct nfs_open_context *ctx = req->wb_context;
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400408 struct nfs_lock_context *l_ctx = req->wb_lock_context;
Trond Myklebustbb6fbc42010-03-11 09:19:35 -0500409
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500410 if (page != NULL) {
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500411 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 req->wb_page = NULL;
413 }
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400414 if (l_ctx != NULL) {
Trond Myklebust577b4232013-04-08 21:38:12 -0400415 nfs_iocounter_dec(&l_ctx->io_count);
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400416 nfs_put_lock_context(l_ctx);
417 req->wb_lock_context = NULL;
418 }
Trond Myklebustbb6fbc42010-03-11 09:19:35 -0500419 if (ctx != NULL) {
420 put_nfs_open_context(ctx);
421 req->wb_context = NULL;
422 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423}
424
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425/**
426 * nfs_release_request - Release the count on an NFS read/write request
427 * @req: request to release
428 *
429 * Note: Should never be called with the spinlock held!
430 */
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400431void nfs_free_request(struct nfs_page *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432{
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400433 WARN_ON_ONCE(req->wb_this_page != req);
434
435 /* extra debug: make sure no sync bits are still set */
436 WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
Weston Andros Adamson67d03382014-05-15 11:56:46 -0400437 WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
438 WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400439 WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
440 WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441
Trond Myklebustbb6fbc42010-03-11 09:19:35 -0500442 /* Release struct file and open context */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 nfs_clear_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 nfs_page_free(req);
445}
446
Trond Myklebustc03b4022007-06-17 13:26:38 -0400447void nfs_release_request(struct nfs_page *req)
448{
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400449 kref_put(&req->wb_kref, nfs_page_group_destroy);
Trond Myklebust9f557cd2010-02-03 08:27:22 -0500450}
451
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452/**
453 * nfs_wait_on_request - Wait for a request to complete.
454 * @req: request to wait upon.
455 *
Matthew Wilcox150030b2007-12-06 16:24:39 -0500456 * Interruptible by fatal signals only.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 * The user is responsible for holding a count on the request.
458 */
459int
460nfs_wait_on_request(struct nfs_page *req)
461{
NeilBrown74316202014-07-07 15:16:04 +1000462 return wait_on_bit_io(&req->wb_flags, PG_BUSY,
463 TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464}
465
Weston Andros Adamsonb4fdac12014-05-15 11:56:43 -0400466/*
467 * nfs_generic_pg_test - determine if requests can be coalesced
468 * @desc: pointer to descriptor
469 * @prev: previous request in desc, or NULL
470 * @req: this request
471 *
472 * Returns zero if @req can be coalesced into @desc, otherwise it returns
473 * the size of the request.
474 */
475size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
476 struct nfs_page *prev, struct nfs_page *req)
Boaz Harrosh5b36c7d2011-05-29 11:45:39 +0300477{
Weston Andros Adamsonf0cb9ab2014-05-15 11:56:52 -0400478 if (desc->pg_count > desc->pg_bsize) {
479 /* should never happen */
480 WARN_ON_ONCE(1);
Boaz Harrosh5b36c7d2011-05-29 11:45:39 +0300481 return 0;
Weston Andros Adamsonf0cb9ab2014-05-15 11:56:52 -0400482 }
Boaz Harrosh5b36c7d2011-05-29 11:45:39 +0300483
Christoph Hellwig2e11f822014-08-21 11:09:17 -0500484 /*
485 * Limit the request size so that we can still allocate a page array
486 * for it without upsetting the slab allocator.
487 */
488 if (((desc->pg_count + req->wb_bytes) >> PAGE_SHIFT) *
489 sizeof(struct page) > PAGE_SIZE)
490 return 0;
491
Weston Andros Adamsonf0cb9ab2014-05-15 11:56:52 -0400492 return min(desc->pg_bsize - desc->pg_count, (size_t)req->wb_bytes);
Boaz Harrosh5b36c7d2011-05-29 11:45:39 +0300493}
Benny Halevy19345cb2011-06-19 18:33:46 -0400494EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
Boaz Harrosh5b36c7d2011-05-29 11:45:39 +0300495
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -0400496struct nfs_pgio_header *nfs_pgio_header_alloc(const struct nfs_rw_ops *ops)
Anna Schumaker00bfa302014-05-06 09:12:29 -0400497{
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -0400498 struct nfs_pgio_header *hdr = ops->rw_alloc_header();
Anna Schumaker00bfa302014-05-06 09:12:29 -0400499
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -0400500 if (hdr) {
Anna Schumaker4a0de552014-05-06 09:12:30 -0400501 INIT_LIST_HEAD(&hdr->pages);
Anna Schumaker4a0de552014-05-06 09:12:30 -0400502 spin_lock_init(&hdr->lock);
Anna Schumaker4a0de552014-05-06 09:12:30 -0400503 hdr->rw_ops = ops;
504 }
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -0400505 return hdr;
Anna Schumaker4a0de552014-05-06 09:12:30 -0400506}
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -0400507EXPORT_SYMBOL_GPL(nfs_pgio_header_alloc);
Anna Schumaker4a0de552014-05-06 09:12:30 -0400508
509/*
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -0400510 * nfs_pgio_header_free - Free a read or write header
Anna Schumaker4a0de552014-05-06 09:12:30 -0400511 * @hdr: The header to free
512 */
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -0400513void nfs_pgio_header_free(struct nfs_pgio_header *hdr)
Anna Schumaker4a0de552014-05-06 09:12:30 -0400514{
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -0400515 hdr->rw_ops->rw_free_header(hdr);
Anna Schumaker4a0de552014-05-06 09:12:30 -0400516}
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -0400517EXPORT_SYMBOL_GPL(nfs_pgio_header_free);
Anna Schumaker4a0de552014-05-06 09:12:30 -0400518
519/**
Weston Andros Adamson4714fb52014-06-09 11:48:37 -0400520 * nfs_pgio_data_destroy - make @hdr suitable for reuse
521 *
522 * Frees memory and releases refs from nfs_generic_pgio, so that it may
523 * be called again.
524 *
525 * @hdr: A header that has had nfs_generic_pgio called
Anna Schumaker00bfa302014-05-06 09:12:29 -0400526 */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400527void nfs_pgio_data_destroy(struct nfs_pgio_header *hdr)
Anna Schumaker00bfa302014-05-06 09:12:29 -0400528{
Trond Myklebust3caa0c62014-10-13 10:26:43 -0400529 if (hdr->args.context)
530 put_nfs_open_context(hdr->args.context);
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400531 if (hdr->page_array.pagevec != hdr->page_array.page_array)
532 kfree(hdr->page_array.pagevec);
Anna Schumaker00bfa302014-05-06 09:12:29 -0400533}
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -0400534EXPORT_SYMBOL_GPL(nfs_pgio_data_destroy);
Anna Schumaker00bfa302014-05-06 09:12:29 -0400535
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536/**
Anna Schumakerce595152014-05-06 09:12:34 -0400537 * nfs_pgio_rpcsetup - Set up arguments for a pageio call
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400538 * @hdr: The pageio hdr
Anna Schumakerce595152014-05-06 09:12:34 -0400539 * @count: Number of bytes to read
540 * @offset: Initial offset
541 * @how: How to commit data (writes only)
542 * @cinfo: Commit information for the call (writes only)
543 */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400544static void nfs_pgio_rpcsetup(struct nfs_pgio_header *hdr,
Anna Schumakerce595152014-05-06 09:12:34 -0400545 unsigned int count, unsigned int offset,
546 int how, struct nfs_commit_info *cinfo)
547{
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400548 struct nfs_page *req = hdr->req;
Anna Schumakerce595152014-05-06 09:12:34 -0400549
550 /* Set up the RPC argument and reply structs
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400551 * NB: take care not to mess about with hdr->commit et al. */
Anna Schumakerce595152014-05-06 09:12:34 -0400552
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400553 hdr->args.fh = NFS_FH(hdr->inode);
554 hdr->args.offset = req_offset(req) + offset;
Anna Schumakerce595152014-05-06 09:12:34 -0400555 /* pnfs_set_layoutcommit needs this */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400556 hdr->mds_offset = hdr->args.offset;
557 hdr->args.pgbase = req->wb_pgbase + offset;
558 hdr->args.pages = hdr->page_array.pagevec;
559 hdr->args.count = count;
560 hdr->args.context = get_nfs_open_context(req->wb_context);
561 hdr->args.lock_context = req->wb_lock_context;
562 hdr->args.stable = NFS_UNSTABLE;
Anna Schumakerce595152014-05-06 09:12:34 -0400563 switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
564 case 0:
565 break;
566 case FLUSH_COND_STABLE:
567 if (nfs_reqs_to_commit(cinfo))
568 break;
569 default:
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400570 hdr->args.stable = NFS_FILE_SYNC;
Anna Schumakerce595152014-05-06 09:12:34 -0400571 }
572
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400573 hdr->res.fattr = &hdr->fattr;
574 hdr->res.count = count;
575 hdr->res.eof = 0;
Weston Andros Adamsonc65e6252014-06-09 11:48:36 -0400576 hdr->res.verf = &hdr->verf;
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400577 nfs_fattr_init(&hdr->fattr);
Anna Schumakerce595152014-05-06 09:12:34 -0400578}
579
580/**
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400581 * nfs_pgio_prepare - Prepare pageio hdr to go over the wire
Anna Schumakera4cdda52014-05-06 09:12:31 -0400582 * @task: The current task
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400583 * @calldata: pageio header to prepare
Anna Schumakera4cdda52014-05-06 09:12:31 -0400584 */
Anna Schumaker6f92fa42014-05-06 09:12:33 -0400585static void nfs_pgio_prepare(struct rpc_task *task, void *calldata)
Anna Schumakera4cdda52014-05-06 09:12:31 -0400586{
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400587 struct nfs_pgio_header *hdr = calldata;
Anna Schumakera4cdda52014-05-06 09:12:31 -0400588 int err;
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400589 err = NFS_PROTO(hdr->inode)->pgio_rpc_prepare(task, hdr);
Anna Schumakera4cdda52014-05-06 09:12:31 -0400590 if (err)
591 rpc_exit(task, err);
592}
593
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400594int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr,
Anna Schumaker1ed26f32014-05-06 09:12:37 -0400595 const struct rpc_call_ops *call_ops, int how, int flags)
596{
597 struct rpc_task *task;
598 struct rpc_message msg = {
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400599 .rpc_argp = &hdr->args,
600 .rpc_resp = &hdr->res,
601 .rpc_cred = hdr->cred,
Anna Schumaker1ed26f32014-05-06 09:12:37 -0400602 };
603 struct rpc_task_setup task_setup_data = {
604 .rpc_client = clnt,
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400605 .task = &hdr->task,
Anna Schumaker1ed26f32014-05-06 09:12:37 -0400606 .rpc_message = &msg,
607 .callback_ops = call_ops,
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400608 .callback_data = hdr,
Anna Schumaker1ed26f32014-05-06 09:12:37 -0400609 .workqueue = nfsiod_workqueue,
610 .flags = RPC_TASK_ASYNC | flags,
611 };
612 int ret = 0;
613
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400614 hdr->rw_ops->rw_initiate(hdr, &msg, &task_setup_data, how);
Anna Schumaker1ed26f32014-05-06 09:12:37 -0400615
616 dprintk("NFS: %5u initiated pgio call "
617 "(req %s/%llu, %u bytes @ offset %llu)\n",
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400618 hdr->task.tk_pid,
Anna Schumaker343ae532014-06-20 13:30:26 -0400619 hdr->inode->i_sb->s_id,
620 (unsigned long long)NFS_FILEID(hdr->inode),
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400621 hdr->args.count,
622 (unsigned long long)hdr->args.offset);
Anna Schumaker1ed26f32014-05-06 09:12:37 -0400623
624 task = rpc_run_task(&task_setup_data);
625 if (IS_ERR(task)) {
626 ret = PTR_ERR(task);
627 goto out;
628 }
629 if (how & FLUSH_SYNC) {
630 ret = rpc_wait_for_completion_task(task);
631 if (ret == 0)
632 ret = task->tk_status;
633 }
634 rpc_put_task(task);
635out:
636 return ret;
637}
638EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
639
Anna Schumakera4cdda52014-05-06 09:12:31 -0400640/**
Anna Schumaker844c9e62014-05-06 09:12:35 -0400641 * nfs_pgio_error - Clean up from a pageio error
642 * @desc: IO descriptor
643 * @hdr: pageio header
644 */
Anna Schumakeref2c4882014-05-06 09:12:36 -0400645static int nfs_pgio_error(struct nfs_pageio_descriptor *desc,
Anna Schumaker844c9e62014-05-06 09:12:35 -0400646 struct nfs_pgio_header *hdr)
647{
Anna Schumaker844c9e62014-05-06 09:12:35 -0400648 set_bit(NFS_IOHDR_REDO, &hdr->flags);
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400649 nfs_pgio_data_destroy(hdr);
Weston Andros Adamson4714fb52014-06-09 11:48:37 -0400650 hdr->completion_ops->completion(hdr);
Anna Schumaker844c9e62014-05-06 09:12:35 -0400651 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
652 return -ENOMEM;
653}
654
655/**
Anna Schumakera4cdda52014-05-06 09:12:31 -0400656 * nfs_pgio_release - Release pageio data
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400657 * @calldata: The pageio header to release
Anna Schumakera4cdda52014-05-06 09:12:31 -0400658 */
Anna Schumaker6f92fa42014-05-06 09:12:33 -0400659static void nfs_pgio_release(void *calldata)
Anna Schumakera4cdda52014-05-06 09:12:31 -0400660{
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400661 struct nfs_pgio_header *hdr = calldata;
662 if (hdr->rw_ops->rw_release)
663 hdr->rw_ops->rw_release(hdr);
664 nfs_pgio_data_destroy(hdr);
Weston Andros Adamson4714fb52014-06-09 11:48:37 -0400665 hdr->completion_ops->completion(hdr);
Anna Schumakera4cdda52014-05-06 09:12:31 -0400666}
667
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668/**
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400669 * nfs_pageio_init - initialise a page io descriptor
670 * @desc: pointer to descriptor
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400671 * @inode: pointer to inode
672 * @doio: pointer to io function
673 * @bsize: io block size
674 * @io_flags: extra parameters for the io function
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400675 */
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400676void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
677 struct inode *inode,
Trond Myklebust1751c362011-06-10 13:30:23 -0400678 const struct nfs_pageio_ops *pg_ops,
Fred Isaman061ae2e2012-04-20 14:47:48 -0400679 const struct nfs_pgio_completion_ops *compl_ops,
Anna Schumaker4a0de552014-05-06 09:12:30 -0400680 const struct nfs_rw_ops *rw_ops,
Trond Myklebust84dde762007-05-04 14:44:06 -0400681 size_t bsize,
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400682 int io_flags)
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400683{
684 INIT_LIST_HEAD(&desc->pg_list);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400685 desc->pg_bytes_written = 0;
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400686 desc->pg_count = 0;
687 desc->pg_bsize = bsize;
688 desc->pg_base = 0;
Trond Myklebustb31268a2011-03-21 17:02:00 -0400689 desc->pg_moreio = 0;
Trond Myklebustd9156f92011-07-12 13:42:02 -0400690 desc->pg_recoalesce = 0;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400691 desc->pg_inode = inode;
Trond Myklebust1751c362011-06-10 13:30:23 -0400692 desc->pg_ops = pg_ops;
Fred Isaman061ae2e2012-04-20 14:47:48 -0400693 desc->pg_completion_ops = compl_ops;
Anna Schumaker4a0de552014-05-06 09:12:30 -0400694 desc->pg_rw_ops = rw_ops;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400695 desc->pg_ioflags = io_flags;
696 desc->pg_error = 0;
Fred Isaman94ad1c82011-03-01 01:34:14 +0000697 desc->pg_lseg = NULL;
Fred Isaman584aa812012-04-20 14:47:51 -0400698 desc->pg_dreq = NULL;
Peng Taof6166382012-08-02 15:36:09 +0300699 desc->pg_layout_private = NULL;
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400700}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400701EXPORT_SYMBOL_GPL(nfs_pageio_init);
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400702
Anna Schumaker0eecb212014-05-06 09:12:32 -0400703/**
704 * nfs_pgio_result - Basic pageio error handling
705 * @task: The task that ran
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400706 * @calldata: Pageio header to check
Anna Schumaker0eecb212014-05-06 09:12:32 -0400707 */
Anna Schumaker6f92fa42014-05-06 09:12:33 -0400708static void nfs_pgio_result(struct rpc_task *task, void *calldata)
Anna Schumaker0eecb212014-05-06 09:12:32 -0400709{
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400710 struct nfs_pgio_header *hdr = calldata;
711 struct inode *inode = hdr->inode;
Anna Schumaker0eecb212014-05-06 09:12:32 -0400712
713 dprintk("NFS: %s: %5u, (status %d)\n", __func__,
714 task->tk_pid, task->tk_status);
715
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400716 if (hdr->rw_ops->rw_done(task, hdr, inode) != 0)
Anna Schumaker0eecb212014-05-06 09:12:32 -0400717 return;
718 if (task->tk_status < 0)
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400719 nfs_set_pgio_error(hdr, task->tk_status, hdr->args.offset);
Anna Schumaker0eecb212014-05-06 09:12:32 -0400720 else
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400721 hdr->rw_ops->rw_result(task, hdr);
Anna Schumaker0eecb212014-05-06 09:12:32 -0400722}
723
Anna Schumakeref2c4882014-05-06 09:12:36 -0400724/*
Anna Schumakeref2c4882014-05-06 09:12:36 -0400725 * Create an RPC task for the given read or write request and kick it.
726 * The page must have been locked by the caller.
727 *
728 * It may happen that the page we're passed is not marked dirty.
729 * This is the case if nfs_updatepage detects a conflicting request
730 * that has been written but not committed.
731 */
Weston Andros Adamsonf0cb9ab2014-05-15 11:56:52 -0400732int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
733 struct nfs_pgio_header *hdr)
Anna Schumakeref2c4882014-05-06 09:12:36 -0400734{
735 struct nfs_page *req;
Weston Andros Adamsonbba5c182014-08-14 17:39:32 -0400736 struct page **pages,
737 *last_page;
Anna Schumakeref2c4882014-05-06 09:12:36 -0400738 struct list_head *head = &desc->pg_list;
739 struct nfs_commit_info cinfo;
Weston Andros Adamsonbba5c182014-08-14 17:39:32 -0400740 unsigned int pagecount, pageused;
Anna Schumakeref2c4882014-05-06 09:12:36 -0400741
Weston Andros Adamson4714fb52014-06-09 11:48:37 -0400742 pagecount = nfs_page_array_len(desc->pg_base, desc->pg_count);
743 if (!nfs_pgarray_set(&hdr->page_array, pagecount))
Anna Schumakeref2c4882014-05-06 09:12:36 -0400744 return nfs_pgio_error(desc, hdr);
745
746 nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400747 pages = hdr->page_array.pagevec;
Weston Andros Adamsonbba5c182014-08-14 17:39:32 -0400748 last_page = NULL;
749 pageused = 0;
Anna Schumakeref2c4882014-05-06 09:12:36 -0400750 while (!list_empty(head)) {
751 req = nfs_list_entry(head->next);
752 nfs_list_remove_request(req);
753 nfs_list_add_request(req, &hdr->pages);
Weston Andros Adamsonbba5c182014-08-14 17:39:32 -0400754
Weston Andros Adamsonbba5c182014-08-14 17:39:32 -0400755 if (!last_page || last_page != req->wb_page) {
Weston Andros Adamsonbba5c182014-08-14 17:39:32 -0400756 pageused++;
Trond Myklebustb8fb9c302014-10-13 10:56:12 -0400757 if (pageused > pagecount)
758 break;
759 *pages++ = last_page = req->wb_page;
Weston Andros Adamsonbba5c182014-08-14 17:39:32 -0400760 }
Anna Schumakeref2c4882014-05-06 09:12:36 -0400761 }
Weston Andros Adamsonbba5c182014-08-14 17:39:32 -0400762 if (WARN_ON_ONCE(pageused != pagecount))
763 return nfs_pgio_error(desc, hdr);
Anna Schumakeref2c4882014-05-06 09:12:36 -0400764
765 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
766 (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
767 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
768
769 /* Set up the argument struct */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400770 nfs_pgio_rpcsetup(hdr, desc->pg_count, 0, desc->pg_ioflags, &cinfo);
Anna Schumakeref2c4882014-05-06 09:12:36 -0400771 desc->pg_rpc_callops = &nfs_pgio_common_ops;
772 return 0;
773}
Weston Andros Adamsonf0cb9ab2014-05-15 11:56:52 -0400774EXPORT_SYMBOL_GPL(nfs_generic_pgio);
Anna Schumakeref2c4882014-05-06 09:12:36 -0400775
Anna Schumaker41d8d5b2014-05-06 09:12:40 -0400776static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
Anna Schumakercf485fc2014-05-06 09:12:39 -0400777{
Anna Schumakercf485fc2014-05-06 09:12:39 -0400778 struct nfs_pgio_header *hdr;
779 int ret;
780
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -0400781 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
782 if (!hdr) {
Anna Schumakercf485fc2014-05-06 09:12:39 -0400783 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
784 return -ENOMEM;
785 }
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -0400786 nfs_pgheader_init(desc, hdr, nfs_pgio_header_free);
Anna Schumakercf485fc2014-05-06 09:12:39 -0400787 ret = nfs_generic_pgio(desc, hdr);
788 if (ret == 0)
Weston Andros Adamson7f714722014-05-15 11:56:53 -0400789 ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400790 hdr, desc->pg_rpc_callops,
Weston Andros Adamson7f714722014-05-15 11:56:53 -0400791 desc->pg_ioflags, 0);
Anna Schumakercf485fc2014-05-06 09:12:39 -0400792 return ret;
793}
794
Trond Myklebust4109bb72013-09-06 11:09:38 -0400795static bool nfs_match_open_context(const struct nfs_open_context *ctx1,
796 const struct nfs_open_context *ctx2)
797{
798 return ctx1->cred == ctx2->cred && ctx1->state == ctx2->state;
799}
800
801static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
802 const struct nfs_lock_context *l2)
803{
804 return l1->lockowner.l_owner == l2->lockowner.l_owner
805 && l1->lockowner.l_pid == l2->lockowner.l_pid;
806}
807
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400808/**
809 * nfs_can_coalesce_requests - test two requests for compatibility
810 * @prev: pointer to nfs_page
811 * @req: pointer to nfs_page
812 *
813 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
814 * page data area they describe is contiguous, and that their RPC
815 * credentials, NFSv4 open state, and lockowners are the same.
816 *
817 * Return 'true' if this is the case, else return 'false'.
818 */
Benny Halevy18ad0a92011-05-25 21:03:56 +0300819static bool nfs_can_coalesce_requests(struct nfs_page *prev,
820 struct nfs_page *req,
821 struct nfs_pageio_descriptor *pgio)
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400822{
Weston Andros Adamsonb4fdac12014-05-15 11:56:43 -0400823 size_t size;
824
Weston Andros Adamsonab75e412014-05-15 11:56:44 -0400825 if (prev) {
826 if (!nfs_match_open_context(req->wb_context, prev->wb_context))
827 return false;
828 if (req->wb_context->dentry->d_inode->i_flock != NULL &&
829 !nfs_match_lock_context(req->wb_lock_context,
830 prev->wb_lock_context))
831 return false;
Weston Andros Adamsonab75e412014-05-15 11:56:44 -0400832 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
833 return false;
Weston Andros Adamson78270e82014-08-14 17:39:33 -0400834 if (req->wb_page == prev->wb_page) {
835 if (req->wb_pgbase != prev->wb_pgbase + prev->wb_bytes)
836 return false;
837 } else {
838 if (req->wb_pgbase != 0 ||
839 prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
840 return false;
841 }
Weston Andros Adamsonab75e412014-05-15 11:56:44 -0400842 }
Weston Andros Adamsonb4fdac12014-05-15 11:56:43 -0400843 size = pgio->pg_ops->pg_test(pgio, prev, req);
Weston Andros Adamsonf0cb9ab2014-05-15 11:56:52 -0400844 WARN_ON_ONCE(size > req->wb_bytes);
845 if (size && size < req->wb_bytes)
846 req->wb_bytes = size;
Weston Andros Adamsonb4fdac12014-05-15 11:56:43 -0400847 return size > 0;
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400848}
849
850/**
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400851 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400852 * @desc: destination io descriptor
853 * @req: request
854 *
855 * Returns true if the request 'req' was successfully coalesced into the
856 * existing list of pages 'desc'.
857 */
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400858static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
859 struct nfs_page *req)
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400860{
Weston Andros Adamsonab75e412014-05-15 11:56:44 -0400861 struct nfs_page *prev = NULL;
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400862 if (desc->pg_count != 0) {
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400863 prev = nfs_list_entry(desc->pg_list.prev);
Boaz Harrosh5b36c7d2011-05-29 11:45:39 +0300864 } else {
Trond Myklebustd8007d42011-06-10 13:30:23 -0400865 if (desc->pg_ops->pg_init)
866 desc->pg_ops->pg_init(desc, req);
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400867 desc->pg_base = req->wb_pgbase;
Boaz Harrosh5b36c7d2011-05-29 11:45:39 +0300868 }
Weston Andros Adamsonab75e412014-05-15 11:56:44 -0400869 if (!nfs_can_coalesce_requests(prev, req, desc))
870 return 0;
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400871 nfs_list_remove_request(req);
872 nfs_list_add_request(req, &desc->pg_list);
Boaz Harrosh5b36c7d2011-05-29 11:45:39 +0300873 desc->pg_count += req->wb_bytes;
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400874 return 1;
875}
876
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400877/*
878 * Helper for nfs_pageio_add_request and nfs_pageio_complete
879 */
880static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
881{
882 if (!list_empty(&desc->pg_list)) {
Trond Myklebust1751c362011-06-10 13:30:23 -0400883 int error = desc->pg_ops->pg_doio(desc);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400884 if (error < 0)
885 desc->pg_error = error;
886 else
887 desc->pg_bytes_written += desc->pg_count;
888 }
889 if (list_empty(&desc->pg_list)) {
890 desc->pg_count = 0;
891 desc->pg_base = 0;
892 }
893}
894
895/**
896 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
897 * @desc: destination io descriptor
898 * @req: request
899 *
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400900 * This may split a request into subrequests which are all part of the
901 * same page group.
902 *
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400903 * Returns true if the request 'req' was successfully coalesced into the
904 * existing list of pages 'desc'.
905 */
Trond Myklebustd9156f92011-07-12 13:42:02 -0400906static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400907 struct nfs_page *req)
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400908{
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400909 struct nfs_page *subreq;
910 unsigned int bytes_left = 0;
911 unsigned int offset, pgbase;
912
Weston Andros Adamsonbfd484a2014-08-08 11:00:55 -0400913 nfs_page_group_lock(req, false);
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400914
915 subreq = req;
916 bytes_left = subreq->wb_bytes;
917 offset = subreq->wb_offset;
918 pgbase = subreq->wb_pgbase;
919
920 do {
921 if (!nfs_pageio_do_add_request(desc, subreq)) {
922 /* make sure pg_test call(s) did nothing */
923 WARN_ON_ONCE(subreq->wb_bytes != bytes_left);
924 WARN_ON_ONCE(subreq->wb_offset != offset);
925 WARN_ON_ONCE(subreq->wb_pgbase != pgbase);
926
927 nfs_page_group_unlock(req);
928 desc->pg_moreio = 1;
929 nfs_pageio_doio(desc);
930 if (desc->pg_error < 0)
931 return 0;
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400932 if (desc->pg_recoalesce)
933 return 0;
934 /* retry add_request for this subreq */
Weston Andros Adamsonbfd484a2014-08-08 11:00:55 -0400935 nfs_page_group_lock(req, false);
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400936 continue;
937 }
938
939 /* check for buggy pg_test call(s) */
940 WARN_ON_ONCE(subreq->wb_bytes + subreq->wb_pgbase > PAGE_SIZE);
941 WARN_ON_ONCE(subreq->wb_bytes > bytes_left);
942 WARN_ON_ONCE(subreq->wb_bytes == 0);
943
944 bytes_left -= subreq->wb_bytes;
945 offset += subreq->wb_bytes;
946 pgbase += subreq->wb_bytes;
947
948 if (bytes_left) {
949 subreq = nfs_create_request(req->wb_context,
950 req->wb_page,
951 subreq, pgbase, bytes_left);
Trond Myklebustc1109552014-05-29 11:38:15 -0400952 if (IS_ERR(subreq))
953 goto err_ptr;
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400954 nfs_lock_request(subreq);
955 subreq->wb_offset = offset;
956 subreq->wb_index = req->wb_index;
957 }
958 } while (bytes_left > 0);
959
960 nfs_page_group_unlock(req);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400961 return 1;
Trond Myklebustc1109552014-05-29 11:38:15 -0400962err_ptr:
963 desc->pg_error = PTR_ERR(subreq);
964 nfs_page_group_unlock(req);
965 return 0;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400966}
967
Trond Myklebustd9156f92011-07-12 13:42:02 -0400968static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
969{
970 LIST_HEAD(head);
971
972 do {
973 list_splice_init(&desc->pg_list, &head);
974 desc->pg_bytes_written -= desc->pg_count;
975 desc->pg_count = 0;
976 desc->pg_base = 0;
977 desc->pg_recoalesce = 0;
Trond Myklebustf563b892014-07-13 15:13:19 -0400978 desc->pg_moreio = 0;
Trond Myklebustd9156f92011-07-12 13:42:02 -0400979
980 while (!list_empty(&head)) {
981 struct nfs_page *req;
982
983 req = list_first_entry(&head, struct nfs_page, wb_list);
984 nfs_list_remove_request(req);
985 if (__nfs_pageio_add_request(desc, req))
986 continue;
987 if (desc->pg_error < 0)
988 return 0;
989 break;
990 }
991 } while (desc->pg_recoalesce);
992 return 1;
993}
994
995int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
996 struct nfs_page *req)
997{
998 int ret;
999
1000 do {
1001 ret = __nfs_pageio_add_request(desc, req);
1002 if (ret)
1003 break;
1004 if (desc->pg_error < 0)
1005 break;
1006 ret = nfs_do_recoalesce(desc);
1007 } while (ret);
1008 return ret;
1009}
1010
Weston Andros Adamson53113ad2014-06-09 11:48:38 -04001011/*
1012 * nfs_pageio_resend - Transfer requests to new descriptor and resend
1013 * @hdr - the pgio header to move request from
1014 * @desc - the pageio descriptor to add requests to
1015 *
1016 * Try to move each request (nfs_page) from @hdr to @desc then attempt
1017 * to send them.
1018 *
1019 * Returns 0 on success and < 0 on error.
1020 */
1021int nfs_pageio_resend(struct nfs_pageio_descriptor *desc,
1022 struct nfs_pgio_header *hdr)
1023{
1024 LIST_HEAD(failed);
1025
1026 desc->pg_dreq = hdr->dreq;
1027 while (!list_empty(&hdr->pages)) {
1028 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
1029
1030 nfs_list_remove_request(req);
1031 if (!nfs_pageio_add_request(desc, req))
1032 nfs_list_add_request(req, &failed);
1033 }
1034 nfs_pageio_complete(desc);
1035 if (!list_empty(&failed)) {
1036 list_move(&failed, &hdr->pages);
1037 return -EIO;
1038 }
1039 return 0;
1040}
1041EXPORT_SYMBOL_GPL(nfs_pageio_resend);
Trond Myklebustd8a5ad72007-04-02 18:48:28 -04001042
1043/**
Trond Myklebustbcb71bb2007-04-02 18:48:28 -04001044 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
1045 * @desc: pointer to io descriptor
1046 */
1047void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1048{
Trond Myklebustd9156f92011-07-12 13:42:02 -04001049 for (;;) {
1050 nfs_pageio_doio(desc);
1051 if (!desc->pg_recoalesce)
1052 break;
1053 if (!nfs_do_recoalesce(desc))
1054 break;
1055 }
Trond Myklebustbcb71bb2007-04-02 18:48:28 -04001056}
1057
Trond Myklebust7fe7f842007-05-20 10:18:27 -04001058/**
1059 * nfs_pageio_cond_complete - Conditional I/O completion
1060 * @desc: pointer to io descriptor
1061 * @index: page index
1062 *
1063 * It is important to ensure that processes don't try to take locks
1064 * on non-contiguous ranges of pages as that might deadlock. This
1065 * function should be called before attempting to wait on a locked
1066 * nfs_page. It will complete the I/O if the page index 'index'
1067 * is not contiguous with the existing list of pages in 'desc'.
1068 */
1069void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1070{
1071 if (!list_empty(&desc->pg_list)) {
1072 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
1073 if (index != prev->wb_index + 1)
Trond Myklebustd9156f92011-07-12 13:42:02 -04001074 nfs_pageio_complete(desc);
Trond Myklebust7fe7f842007-05-20 10:18:27 -04001075 }
1076}
1077
David Howellsf7b422b2006-06-09 09:34:33 -04001078int __init nfs_init_nfspagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079{
1080 nfs_page_cachep = kmem_cache_create("nfs_page",
1081 sizeof(struct nfs_page),
1082 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09001083 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 if (nfs_page_cachep == NULL)
1085 return -ENOMEM;
1086
1087 return 0;
1088}
1089
David Brownell266bee82006-06-27 12:59:15 -07001090void nfs_destroy_nfspagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07001092 kmem_cache_destroy(nfs_page_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093}
1094
Anna Schumakeref2c4882014-05-06 09:12:36 -04001095static const struct rpc_call_ops nfs_pgio_common_ops = {
Anna Schumaker6f92fa42014-05-06 09:12:33 -04001096 .rpc_call_prepare = nfs_pgio_prepare,
1097 .rpc_call_done = nfs_pgio_result,
1098 .rpc_release = nfs_pgio_release,
1099};
Anna Schumaker41d8d5b2014-05-06 09:12:40 -04001100
1101const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1102 .pg_test = nfs_generic_pg_test,
1103 .pg_doio = nfs_generic_pg_pgios,
1104};