Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | ||
1da177e4 LT |
12 | #include <linux/slab.h> |
13 | #include <linux/file.h> | |
e8edc6e0 | 14 | #include <linux/sched.h> |
1da177e4 | 15 | #include <linux/sunrpc/clnt.h> |
1313e603 | 16 | #include <linux/nfs.h> |
1da177e4 LT |
17 | #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> | |
afeacc8c | 22 | #include <linux/export.h> |
1da177e4 | 23 | |
8d5658c9 | 24 | #include "internal.h" |
bae724ef | 25 | #include "pnfs.h" |
8d5658c9 | 26 | |
0eecb214 AS |
27 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
28 | ||
e18b890b | 29 | static struct kmem_cache *nfs_page_cachep; |
ef2c488c | 30 | static const struct rpc_call_ops nfs_pgio_common_ops; |
1da177e4 | 31 | |
2bfc6e56 WAA |
32 | static void nfs_free_request(struct nfs_page *); |
33 | ||
00bfa30a | 34 | static bool nfs_pgarray_set(struct nfs_page_array *p, unsigned int pagecount) |
30dd374f FI |
35 | { |
36 | p->npages = pagecount; | |
37 | if (pagecount <= ARRAY_SIZE(p->page_array)) | |
38 | p->pagevec = p->page_array; | |
39 | else { | |
40 | p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL); | |
41 | if (!p->pagevec) | |
42 | p->npages = 0; | |
43 | } | |
44 | return p->pagevec != NULL; | |
45 | } | |
46 | ||
4db6e0b7 FI |
47 | void nfs_pgheader_init(struct nfs_pageio_descriptor *desc, |
48 | struct nfs_pgio_header *hdr, | |
49 | void (*release)(struct nfs_pgio_header *hdr)) | |
50 | { | |
51 | hdr->req = nfs_list_entry(desc->pg_list.next); | |
52 | hdr->inode = desc->pg_inode; | |
53 | hdr->cred = hdr->req->wb_context->cred; | |
54 | hdr->io_start = req_offset(hdr->req); | |
55 | hdr->good_bytes = desc->pg_count; | |
584aa810 | 56 | hdr->dreq = desc->pg_dreq; |
f6166384 | 57 | hdr->layout_private = desc->pg_layout_private; |
4db6e0b7 | 58 | hdr->release = release; |
061ae2ed | 59 | hdr->completion_ops = desc->pg_completion_ops; |
584aa810 FI |
60 | if (hdr->completion_ops->init_hdr) |
61 | hdr->completion_ops->init_hdr(hdr); | |
4db6e0b7 | 62 | } |
89d77c8f | 63 | EXPORT_SYMBOL_GPL(nfs_pgheader_init); |
4db6e0b7 FI |
64 | |
65 | void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos) | |
66 | { | |
67 | spin_lock(&hdr->lock); | |
68 | if (pos < hdr->io_start + hdr->good_bytes) { | |
69 | set_bit(NFS_IOHDR_ERROR, &hdr->flags); | |
70 | clear_bit(NFS_IOHDR_EOF, &hdr->flags); | |
71 | hdr->good_bytes = pos - hdr->io_start; | |
72 | hdr->error = error; | |
73 | } | |
74 | spin_unlock(&hdr->lock); | |
75 | } | |
76 | ||
1da177e4 LT |
77 | static inline struct nfs_page * |
78 | nfs_page_alloc(void) | |
79 | { | |
192e501b | 80 | struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_NOIO); |
72895b1a | 81 | if (p) |
1da177e4 | 82 | INIT_LIST_HEAD(&p->wb_list); |
1da177e4 LT |
83 | return p; |
84 | } | |
85 | ||
86 | static inline void | |
87 | nfs_page_free(struct nfs_page *p) | |
88 | { | |
89 | kmem_cache_free(nfs_page_cachep, p); | |
90 | } | |
91 | ||
577b4232 TM |
92 | static void |
93 | nfs_iocounter_inc(struct nfs_io_counter *c) | |
94 | { | |
95 | atomic_inc(&c->io_count); | |
96 | } | |
97 | ||
98 | static void | |
99 | nfs_iocounter_dec(struct nfs_io_counter *c) | |
100 | { | |
101 | if (atomic_dec_and_test(&c->io_count)) { | |
102 | clear_bit(NFS_IO_INPROGRESS, &c->flags); | |
103 | smp_mb__after_clear_bit(); | |
104 | wake_up_bit(&c->flags, NFS_IO_INPROGRESS); | |
105 | } | |
106 | } | |
107 | ||
108 | static int | |
109 | __nfs_iocounter_wait(struct nfs_io_counter *c) | |
110 | { | |
111 | wait_queue_head_t *wq = bit_waitqueue(&c->flags, NFS_IO_INPROGRESS); | |
112 | DEFINE_WAIT_BIT(q, &c->flags, NFS_IO_INPROGRESS); | |
113 | int ret = 0; | |
114 | ||
115 | do { | |
116 | prepare_to_wait(wq, &q.wait, TASK_KILLABLE); | |
117 | set_bit(NFS_IO_INPROGRESS, &c->flags); | |
118 | if (atomic_read(&c->io_count) == 0) | |
119 | break; | |
120 | ret = nfs_wait_bit_killable(&c->flags); | |
121 | } while (atomic_read(&c->io_count) != 0); | |
122 | finish_wait(wq, &q.wait); | |
123 | return ret; | |
124 | } | |
125 | ||
126 | /** | |
127 | * nfs_iocounter_wait - wait for i/o to complete | |
128 | * @c: nfs_io_counter to use | |
129 | * | |
130 | * returns -ERESTARTSYS if interrupted by a fatal signal. | |
131 | * Otherwise returns 0 once the io_count hits 0. | |
132 | */ | |
133 | int | |
134 | nfs_iocounter_wait(struct nfs_io_counter *c) | |
135 | { | |
136 | if (atomic_read(&c->io_count) == 0) | |
137 | return 0; | |
138 | return __nfs_iocounter_wait(c); | |
139 | } | |
140 | ||
2bfc6e56 WAA |
141 | /* |
142 | * nfs_page_group_lock - lock the head of the page group | |
143 | * @req - request in group that is to be locked | |
144 | * | |
145 | * this lock must be held if modifying the page group list | |
146 | */ | |
147 | void | |
148 | nfs_page_group_lock(struct nfs_page *req) | |
149 | { | |
150 | struct nfs_page *head = req->wb_head; | |
151 | int err = -EAGAIN; | |
152 | ||
153 | WARN_ON_ONCE(head != head->wb_head); | |
154 | ||
155 | while (err) | |
156 | err = wait_on_bit_lock(&head->wb_flags, PG_HEADLOCK, | |
157 | nfs_wait_bit_killable, TASK_KILLABLE); | |
158 | } | |
159 | ||
160 | /* | |
161 | * nfs_page_group_unlock - unlock the head of the page group | |
162 | * @req - request in group that is to be unlocked | |
163 | */ | |
164 | void | |
165 | nfs_page_group_unlock(struct nfs_page *req) | |
166 | { | |
167 | struct nfs_page *head = req->wb_head; | |
168 | ||
169 | WARN_ON_ONCE(head != head->wb_head); | |
170 | ||
171 | smp_mb__before_clear_bit(); | |
172 | clear_bit(PG_HEADLOCK, &head->wb_flags); | |
173 | smp_mb__after_clear_bit(); | |
174 | wake_up_bit(&head->wb_flags, PG_HEADLOCK); | |
175 | } | |
176 | ||
177 | /* | |
178 | * nfs_page_group_sync_on_bit_locked | |
179 | * | |
180 | * must be called with page group lock held | |
181 | */ | |
182 | static bool | |
183 | nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit) | |
184 | { | |
185 | struct nfs_page *head = req->wb_head; | |
186 | struct nfs_page *tmp; | |
187 | ||
188 | WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags)); | |
189 | WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags)); | |
190 | ||
191 | tmp = req->wb_this_page; | |
192 | while (tmp != req) { | |
193 | if (!test_bit(bit, &tmp->wb_flags)) | |
194 | return false; | |
195 | tmp = tmp->wb_this_page; | |
196 | } | |
197 | ||
198 | /* true! reset all bits */ | |
199 | tmp = req; | |
200 | do { | |
201 | clear_bit(bit, &tmp->wb_flags); | |
202 | tmp = tmp->wb_this_page; | |
203 | } while (tmp != req); | |
204 | ||
205 | return true; | |
206 | } | |
207 | ||
208 | /* | |
209 | * nfs_page_group_sync_on_bit - set bit on current request, but only | |
210 | * return true if the bit is set for all requests in page group | |
211 | * @req - request in page group | |
212 | * @bit - PG_* bit that is used to sync page group | |
213 | */ | |
214 | bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit) | |
215 | { | |
216 | bool ret; | |
217 | ||
218 | nfs_page_group_lock(req); | |
219 | ret = nfs_page_group_sync_on_bit_locked(req, bit); | |
220 | nfs_page_group_unlock(req); | |
221 | ||
222 | return ret; | |
223 | } | |
224 | ||
225 | /* | |
226 | * nfs_page_group_init - Initialize the page group linkage for @req | |
227 | * @req - a new nfs request | |
228 | * @prev - the previous request in page group, or NULL if @req is the first | |
229 | * or only request in the group (the head). | |
230 | */ | |
231 | static inline void | |
232 | nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev) | |
233 | { | |
234 | WARN_ON_ONCE(prev == req); | |
235 | ||
236 | if (!prev) { | |
237 | req->wb_head = req; | |
238 | req->wb_this_page = req; | |
239 | } else { | |
240 | WARN_ON_ONCE(prev->wb_this_page != prev->wb_head); | |
241 | WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags)); | |
242 | req->wb_head = prev->wb_head; | |
243 | req->wb_this_page = prev->wb_this_page; | |
244 | prev->wb_this_page = req; | |
245 | ||
246 | /* grab extra ref if head request has extra ref from | |
247 | * the write/commit path to handle handoff between write | |
248 | * and commit lists */ | |
249 | if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags)) | |
250 | kref_get(&req->wb_kref); | |
251 | } | |
252 | } | |
253 | ||
254 | /* | |
255 | * nfs_page_group_destroy - sync the destruction of page groups | |
256 | * @req - request that no longer needs the page group | |
257 | * | |
258 | * releases the page group reference from each member once all | |
259 | * members have called this function. | |
260 | */ | |
261 | static void | |
262 | nfs_page_group_destroy(struct kref *kref) | |
263 | { | |
264 | struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref); | |
265 | struct nfs_page *tmp, *next; | |
266 | ||
267 | if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN)) | |
268 | return; | |
269 | ||
270 | tmp = req; | |
271 | do { | |
272 | next = tmp->wb_this_page; | |
273 | /* unlink and free */ | |
274 | tmp->wb_this_page = tmp; | |
275 | tmp->wb_head = tmp; | |
276 | nfs_free_request(tmp); | |
277 | tmp = next; | |
278 | } while (tmp != req); | |
279 | } | |
280 | ||
1da177e4 LT |
281 | /** |
282 | * nfs_create_request - Create an NFS read/write request. | |
c02f557d | 283 | * @ctx: open context to use |
1da177e4 | 284 | * @page: page to write |
2bfc6e56 | 285 | * @last: last nfs request created for this page group or NULL if head |
1da177e4 LT |
286 | * @offset: starting offset within the page for the write |
287 | * @count: number of bytes to read/write | |
288 | * | |
289 | * The page must be locked by the caller. This makes sure we never | |
a19b89ca | 290 | * create two different requests for the same page. |
1da177e4 LT |
291 | * User should ensure it is safe to sleep in this function. |
292 | */ | |
293 | struct nfs_page * | |
8c8f1ac1 | 294 | nfs_create_request(struct nfs_open_context *ctx, struct page *page, |
2bfc6e56 WAA |
295 | struct nfs_page *last, unsigned int offset, |
296 | unsigned int count) | |
1da177e4 | 297 | { |
1da177e4 | 298 | struct nfs_page *req; |
b3c54de6 | 299 | struct nfs_lock_context *l_ctx; |
1da177e4 | 300 | |
c58c8441 TM |
301 | if (test_bit(NFS_CONTEXT_BAD, &ctx->flags)) |
302 | return ERR_PTR(-EBADF); | |
18eb8842 TM |
303 | /* try to allocate the request struct */ |
304 | req = nfs_page_alloc(); | |
305 | if (req == NULL) | |
306 | return ERR_PTR(-ENOMEM); | |
1da177e4 | 307 | |
015f0212 | 308 | /* get lock context early so we can deal with alloc failures */ |
b3c54de6 TM |
309 | l_ctx = nfs_get_lock_context(ctx); |
310 | if (IS_ERR(l_ctx)) { | |
015f0212 | 311 | nfs_page_free(req); |
b3c54de6 | 312 | return ERR_CAST(l_ctx); |
015f0212 | 313 | } |
b3c54de6 | 314 | req->wb_lock_context = l_ctx; |
577b4232 | 315 | nfs_iocounter_inc(&l_ctx->io_count); |
015f0212 | 316 | |
1da177e4 LT |
317 | /* Initialize the request struct. Initially, we assume a |
318 | * long write-back delay. This will be adjusted in | |
319 | * update_nfs_request below if the region is not locked. */ | |
320 | req->wb_page = page; | |
d56b4ddf | 321 | req->wb_index = page_file_index(page); |
1da177e4 LT |
322 | page_cache_get(page); |
323 | req->wb_offset = offset; | |
324 | req->wb_pgbase = offset; | |
325 | req->wb_bytes = count; | |
1da177e4 | 326 | req->wb_context = get_nfs_open_context(ctx); |
c03b4024 | 327 | kref_init(&req->wb_kref); |
2bfc6e56 | 328 | nfs_page_group_init(req, last); |
1da177e4 LT |
329 | return req; |
330 | } | |
331 | ||
332 | /** | |
1d1afcbc | 333 | * nfs_unlock_request - Unlock request and wake up sleepers. |
1da177e4 LT |
334 | * @req: |
335 | */ | |
1d1afcbc | 336 | void nfs_unlock_request(struct nfs_page *req) |
1da177e4 LT |
337 | { |
338 | if (!NFS_WBACK_BUSY(req)) { | |
339 | printk(KERN_ERR "NFS: Invalid unlock attempted\n"); | |
340 | BUG(); | |
341 | } | |
342 | smp_mb__before_clear_bit(); | |
343 | clear_bit(PG_BUSY, &req->wb_flags); | |
344 | smp_mb__after_clear_bit(); | |
464a98bd | 345 | wake_up_bit(&req->wb_flags, PG_BUSY); |
3aff4ebb TM |
346 | } |
347 | ||
348 | /** | |
1d1afcbc TM |
349 | * nfs_unlock_and_release_request - Unlock request and release the nfs_page |
350 | * @req: | |
3aff4ebb | 351 | */ |
1d1afcbc | 352 | void nfs_unlock_and_release_request(struct nfs_page *req) |
3aff4ebb | 353 | { |
1d1afcbc | 354 | nfs_unlock_request(req); |
1da177e4 LT |
355 | nfs_release_request(req); |
356 | } | |
357 | ||
4d65c520 | 358 | /* |
1da177e4 LT |
359 | * nfs_clear_request - Free up all resources allocated to the request |
360 | * @req: | |
361 | * | |
bb6fbc45 TM |
362 | * Release page and open context resources associated with a read/write |
363 | * request after it has completed. | |
1da177e4 | 364 | */ |
4d65c520 | 365 | static void nfs_clear_request(struct nfs_page *req) |
1da177e4 | 366 | { |
cd52ed35 | 367 | struct page *page = req->wb_page; |
bb6fbc45 | 368 | struct nfs_open_context *ctx = req->wb_context; |
f11ac8db | 369 | struct nfs_lock_context *l_ctx = req->wb_lock_context; |
bb6fbc45 | 370 | |
cd52ed35 | 371 | if (page != NULL) { |
cd52ed35 | 372 | page_cache_release(page); |
1da177e4 LT |
373 | req->wb_page = NULL; |
374 | } | |
f11ac8db | 375 | if (l_ctx != NULL) { |
577b4232 | 376 | nfs_iocounter_dec(&l_ctx->io_count); |
f11ac8db TM |
377 | nfs_put_lock_context(l_ctx); |
378 | req->wb_lock_context = NULL; | |
379 | } | |
bb6fbc45 TM |
380 | if (ctx != NULL) { |
381 | put_nfs_open_context(ctx); | |
382 | req->wb_context = NULL; | |
383 | } | |
1da177e4 LT |
384 | } |
385 | ||
1da177e4 LT |
386 | /** |
387 | * nfs_release_request - Release the count on an NFS read/write request | |
388 | * @req: request to release | |
389 | * | |
390 | * Note: Should never be called with the spinlock held! | |
391 | */ | |
2bfc6e56 | 392 | static void nfs_free_request(struct nfs_page *req) |
1da177e4 | 393 | { |
2bfc6e56 WAA |
394 | WARN_ON_ONCE(req->wb_this_page != req); |
395 | ||
396 | /* extra debug: make sure no sync bits are still set */ | |
397 | WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags)); | |
67d0338e WAA |
398 | WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags)); |
399 | WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags)); | |
20633f04 WAA |
400 | WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags)); |
401 | WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags)); | |
1da177e4 | 402 | |
bb6fbc45 | 403 | /* Release struct file and open context */ |
1da177e4 | 404 | nfs_clear_request(req); |
1da177e4 LT |
405 | nfs_page_free(req); |
406 | } | |
407 | ||
c03b4024 TM |
408 | void nfs_release_request(struct nfs_page *req) |
409 | { | |
2bfc6e56 | 410 | kref_put(&req->wb_kref, nfs_page_group_destroy); |
c03b4024 TM |
411 | } |
412 | ||
9f557cd8 TM |
413 | static int nfs_wait_bit_uninterruptible(void *word) |
414 | { | |
415 | io_schedule(); | |
416 | return 0; | |
417 | } | |
418 | ||
1da177e4 LT |
419 | /** |
420 | * nfs_wait_on_request - Wait for a request to complete. | |
421 | * @req: request to wait upon. | |
422 | * | |
150030b7 | 423 | * Interruptible by fatal signals only. |
1da177e4 LT |
424 | * The user is responsible for holding a count on the request. |
425 | */ | |
426 | int | |
427 | nfs_wait_on_request(struct nfs_page *req) | |
428 | { | |
9f557cd8 TM |
429 | return wait_on_bit(&req->wb_flags, PG_BUSY, |
430 | nfs_wait_bit_uninterruptible, | |
431 | TASK_UNINTERRUPTIBLE); | |
1da177e4 LT |
432 | } |
433 | ||
b4fdac1a WAA |
434 | /* |
435 | * nfs_generic_pg_test - determine if requests can be coalesced | |
436 | * @desc: pointer to descriptor | |
437 | * @prev: previous request in desc, or NULL | |
438 | * @req: this request | |
439 | * | |
440 | * Returns zero if @req can be coalesced into @desc, otherwise it returns | |
441 | * the size of the request. | |
442 | */ | |
443 | size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc, | |
444 | struct nfs_page *prev, struct nfs_page *req) | |
5b36c7dc | 445 | { |
ab75e417 WAA |
446 | if (!prev) |
447 | return req->wb_bytes; | |
5b36c7dc BH |
448 | /* |
449 | * FIXME: ideally we should be able to coalesce all requests | |
450 | * that are not block boundary aligned, but currently this | |
451 | * is problematic for the case of bsize < PAGE_CACHE_SIZE, | |
452 | * since nfs_flush_multi and nfs_pagein_multi assume you | |
453 | * can have only one struct nfs_page. | |
454 | */ | |
455 | if (desc->pg_bsize < PAGE_SIZE) | |
456 | return 0; | |
457 | ||
b4fdac1a WAA |
458 | if (desc->pg_count + req->wb_bytes <= desc->pg_bsize) |
459 | return req->wb_bytes; | |
460 | return 0; | |
5b36c7dc | 461 | } |
19345cb2 | 462 | EXPORT_SYMBOL_GPL(nfs_generic_pg_test); |
5b36c7dc | 463 | |
00bfa30a AS |
464 | static inline struct nfs_rw_header *NFS_RW_HEADER(struct nfs_pgio_header *hdr) |
465 | { | |
466 | return container_of(hdr, struct nfs_rw_header, header); | |
467 | } | |
468 | ||
4a0de55c AS |
469 | /** |
470 | * nfs_rw_header_alloc - Allocate a header for a read or write | |
471 | * @ops: Read or write function vector | |
472 | */ | |
473 | struct nfs_rw_header *nfs_rw_header_alloc(const struct nfs_rw_ops *ops) | |
474 | { | |
475 | struct nfs_rw_header *header = ops->rw_alloc_header(); | |
476 | ||
477 | if (header) { | |
478 | struct nfs_pgio_header *hdr = &header->header; | |
479 | ||
480 | INIT_LIST_HEAD(&hdr->pages); | |
481 | INIT_LIST_HEAD(&hdr->rpc_list); | |
482 | spin_lock_init(&hdr->lock); | |
483 | atomic_set(&hdr->refcnt, 0); | |
484 | hdr->rw_ops = ops; | |
485 | } | |
486 | return header; | |
487 | } | |
488 | EXPORT_SYMBOL_GPL(nfs_rw_header_alloc); | |
489 | ||
490 | /* | |
491 | * nfs_rw_header_free - Free a read or write header | |
492 | * @hdr: The header to free | |
493 | */ | |
494 | void nfs_rw_header_free(struct nfs_pgio_header *hdr) | |
495 | { | |
496 | hdr->rw_ops->rw_free_header(NFS_RW_HEADER(hdr)); | |
497 | } | |
498 | EXPORT_SYMBOL_GPL(nfs_rw_header_free); | |
499 | ||
00bfa30a AS |
500 | /** |
501 | * nfs_pgio_data_alloc - Allocate pageio data | |
502 | * @hdr: The header making a request | |
503 | * @pagecount: Number of pages to create | |
504 | */ | |
ef2c488c AS |
505 | static struct nfs_pgio_data *nfs_pgio_data_alloc(struct nfs_pgio_header *hdr, |
506 | unsigned int pagecount) | |
00bfa30a AS |
507 | { |
508 | struct nfs_pgio_data *data, *prealloc; | |
509 | ||
510 | prealloc = &NFS_RW_HEADER(hdr)->rpc_data; | |
511 | if (prealloc->header == NULL) | |
512 | data = prealloc; | |
513 | else | |
514 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
515 | if (!data) | |
516 | goto out; | |
517 | ||
518 | if (nfs_pgarray_set(&data->pages, pagecount)) { | |
519 | data->header = hdr; | |
520 | atomic_inc(&hdr->refcnt); | |
521 | } else { | |
522 | if (data != prealloc) | |
523 | kfree(data); | |
524 | data = NULL; | |
525 | } | |
526 | out: | |
527 | return data; | |
528 | } | |
529 | ||
530 | /** | |
531 | * nfs_pgio_data_release - Properly free pageio data | |
532 | * @data: The data to release | |
533 | */ | |
534 | void nfs_pgio_data_release(struct nfs_pgio_data *data) | |
535 | { | |
536 | struct nfs_pgio_header *hdr = data->header; | |
537 | struct nfs_rw_header *pageio_header = NFS_RW_HEADER(hdr); | |
538 | ||
539 | put_nfs_open_context(data->args.context); | |
540 | if (data->pages.pagevec != data->pages.page_array) | |
541 | kfree(data->pages.pagevec); | |
542 | if (data == &pageio_header->rpc_data) { | |
543 | data->header = NULL; | |
544 | data = NULL; | |
545 | } | |
546 | if (atomic_dec_and_test(&hdr->refcnt)) | |
547 | hdr->completion_ops->completion(hdr); | |
548 | /* Note: we only free the rpc_task after callbacks are done. | |
549 | * See the comment in rpc_free_task() for why | |
550 | */ | |
551 | kfree(data); | |
552 | } | |
553 | EXPORT_SYMBOL_GPL(nfs_pgio_data_release); | |
554 | ||
ce59515c AS |
555 | /** |
556 | * nfs_pgio_rpcsetup - Set up arguments for a pageio call | |
557 | * @data: The pageio data | |
558 | * @count: Number of bytes to read | |
559 | * @offset: Initial offset | |
560 | * @how: How to commit data (writes only) | |
561 | * @cinfo: Commit information for the call (writes only) | |
562 | */ | |
ef2c488c | 563 | static void nfs_pgio_rpcsetup(struct nfs_pgio_data *data, |
ce59515c AS |
564 | unsigned int count, unsigned int offset, |
565 | int how, struct nfs_commit_info *cinfo) | |
566 | { | |
567 | struct nfs_page *req = data->header->req; | |
568 | ||
569 | /* Set up the RPC argument and reply structs | |
570 | * NB: take care not to mess about with data->commit et al. */ | |
571 | ||
572 | data->args.fh = NFS_FH(data->header->inode); | |
573 | data->args.offset = req_offset(req) + offset; | |
574 | /* pnfs_set_layoutcommit needs this */ | |
575 | data->mds_offset = data->args.offset; | |
576 | data->args.pgbase = req->wb_pgbase + offset; | |
577 | data->args.pages = data->pages.pagevec; | |
578 | data->args.count = count; | |
579 | data->args.context = get_nfs_open_context(req->wb_context); | |
580 | data->args.lock_context = req->wb_lock_context; | |
581 | data->args.stable = NFS_UNSTABLE; | |
582 | switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { | |
583 | case 0: | |
584 | break; | |
585 | case FLUSH_COND_STABLE: | |
586 | if (nfs_reqs_to_commit(cinfo)) | |
587 | break; | |
588 | default: | |
589 | data->args.stable = NFS_FILE_SYNC; | |
590 | } | |
591 | ||
592 | data->res.fattr = &data->fattr; | |
593 | data->res.count = count; | |
594 | data->res.eof = 0; | |
595 | data->res.verf = &data->verf; | |
596 | nfs_fattr_init(&data->fattr); | |
597 | } | |
598 | ||
a4cdda59 AS |
599 | /** |
600 | * nfs_pgio_prepare - Prepare pageio data to go over the wire | |
601 | * @task: The current task | |
602 | * @calldata: pageio data to prepare | |
603 | */ | |
6f92fa45 | 604 | static void nfs_pgio_prepare(struct rpc_task *task, void *calldata) |
a4cdda59 AS |
605 | { |
606 | struct nfs_pgio_data *data = calldata; | |
607 | int err; | |
608 | err = NFS_PROTO(data->header->inode)->pgio_rpc_prepare(task, data); | |
609 | if (err) | |
610 | rpc_exit(task, err); | |
611 | } | |
612 | ||
1ed26f33 AS |
613 | int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_data *data, |
614 | const struct rpc_call_ops *call_ops, int how, int flags) | |
615 | { | |
616 | struct rpc_task *task; | |
617 | struct rpc_message msg = { | |
618 | .rpc_argp = &data->args, | |
619 | .rpc_resp = &data->res, | |
620 | .rpc_cred = data->header->cred, | |
621 | }; | |
622 | struct rpc_task_setup task_setup_data = { | |
623 | .rpc_client = clnt, | |
624 | .task = &data->task, | |
625 | .rpc_message = &msg, | |
626 | .callback_ops = call_ops, | |
627 | .callback_data = data, | |
628 | .workqueue = nfsiod_workqueue, | |
629 | .flags = RPC_TASK_ASYNC | flags, | |
630 | }; | |
631 | int ret = 0; | |
632 | ||
633 | data->header->rw_ops->rw_initiate(data, &msg, &task_setup_data, how); | |
634 | ||
635 | dprintk("NFS: %5u initiated pgio call " | |
636 | "(req %s/%llu, %u bytes @ offset %llu)\n", | |
637 | data->task.tk_pid, | |
638 | data->header->inode->i_sb->s_id, | |
639 | (unsigned long long)NFS_FILEID(data->header->inode), | |
640 | data->args.count, | |
641 | (unsigned long long)data->args.offset); | |
642 | ||
643 | task = rpc_run_task(&task_setup_data); | |
644 | if (IS_ERR(task)) { | |
645 | ret = PTR_ERR(task); | |
646 | goto out; | |
647 | } | |
648 | if (how & FLUSH_SYNC) { | |
649 | ret = rpc_wait_for_completion_task(task); | |
650 | if (ret == 0) | |
651 | ret = task->tk_status; | |
652 | } | |
653 | rpc_put_task(task); | |
654 | out: | |
655 | return ret; | |
656 | } | |
657 | EXPORT_SYMBOL_GPL(nfs_initiate_pgio); | |
658 | ||
cf485fcd | 659 | static int nfs_do_multiple_pgios(struct list_head *head, |
c3766276 AS |
660 | const struct rpc_call_ops *call_ops, |
661 | int how) | |
662 | { | |
663 | struct nfs_pgio_data *data; | |
664 | int ret = 0; | |
665 | ||
666 | while (!list_empty(head)) { | |
667 | int ret2; | |
668 | ||
669 | data = list_first_entry(head, struct nfs_pgio_data, list); | |
670 | list_del_init(&data->list); | |
671 | ||
672 | ret2 = nfs_initiate_pgio(NFS_CLIENT(data->header->inode), | |
673 | data, call_ops, how, 0); | |
674 | if (ret == 0) | |
675 | ret = ret2; | |
676 | } | |
677 | return ret; | |
678 | } | |
679 | ||
844c9e69 AS |
680 | /** |
681 | * nfs_pgio_error - Clean up from a pageio error | |
682 | * @desc: IO descriptor | |
683 | * @hdr: pageio header | |
684 | */ | |
ef2c488c | 685 | static int nfs_pgio_error(struct nfs_pageio_descriptor *desc, |
844c9e69 AS |
686 | struct nfs_pgio_header *hdr) |
687 | { | |
688 | struct nfs_pgio_data *data; | |
689 | ||
690 | set_bit(NFS_IOHDR_REDO, &hdr->flags); | |
691 | while (!list_empty(&hdr->rpc_list)) { | |
692 | data = list_first_entry(&hdr->rpc_list, struct nfs_pgio_data, list); | |
693 | list_del(&data->list); | |
694 | nfs_pgio_data_release(data); | |
695 | } | |
696 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); | |
697 | return -ENOMEM; | |
698 | } | |
699 | ||
a4cdda59 AS |
700 | /** |
701 | * nfs_pgio_release - Release pageio data | |
702 | * @calldata: The pageio data to release | |
703 | */ | |
6f92fa45 | 704 | static void nfs_pgio_release(void *calldata) |
a4cdda59 AS |
705 | { |
706 | struct nfs_pgio_data *data = calldata; | |
707 | if (data->header->rw_ops->rw_release) | |
708 | data->header->rw_ops->rw_release(data); | |
709 | nfs_pgio_data_release(data); | |
710 | } | |
711 | ||
1da177e4 | 712 | /** |
d8a5ad75 TM |
713 | * nfs_pageio_init - initialise a page io descriptor |
714 | * @desc: pointer to descriptor | |
bcb71bba TM |
715 | * @inode: pointer to inode |
716 | * @doio: pointer to io function | |
717 | * @bsize: io block size | |
718 | * @io_flags: extra parameters for the io function | |
d8a5ad75 | 719 | */ |
bcb71bba TM |
720 | void nfs_pageio_init(struct nfs_pageio_descriptor *desc, |
721 | struct inode *inode, | |
1751c363 | 722 | const struct nfs_pageio_ops *pg_ops, |
061ae2ed | 723 | const struct nfs_pgio_completion_ops *compl_ops, |
4a0de55c | 724 | const struct nfs_rw_ops *rw_ops, |
84dde76c | 725 | size_t bsize, |
bcb71bba | 726 | int io_flags) |
d8a5ad75 TM |
727 | { |
728 | INIT_LIST_HEAD(&desc->pg_list); | |
bcb71bba | 729 | desc->pg_bytes_written = 0; |
d8a5ad75 TM |
730 | desc->pg_count = 0; |
731 | desc->pg_bsize = bsize; | |
732 | desc->pg_base = 0; | |
b31268ac | 733 | desc->pg_moreio = 0; |
d9156f9f | 734 | desc->pg_recoalesce = 0; |
bcb71bba | 735 | desc->pg_inode = inode; |
1751c363 | 736 | desc->pg_ops = pg_ops; |
061ae2ed | 737 | desc->pg_completion_ops = compl_ops; |
4a0de55c | 738 | desc->pg_rw_ops = rw_ops; |
bcb71bba TM |
739 | desc->pg_ioflags = io_flags; |
740 | desc->pg_error = 0; | |
94ad1c80 | 741 | desc->pg_lseg = NULL; |
584aa810 | 742 | desc->pg_dreq = NULL; |
f6166384 | 743 | desc->pg_layout_private = NULL; |
d8a5ad75 | 744 | } |
89d77c8f | 745 | EXPORT_SYMBOL_GPL(nfs_pageio_init); |
d8a5ad75 | 746 | |
0eecb214 AS |
747 | /** |
748 | * nfs_pgio_result - Basic pageio error handling | |
749 | * @task: The task that ran | |
750 | * @calldata: Pageio data to check | |
751 | */ | |
6f92fa45 | 752 | static void nfs_pgio_result(struct rpc_task *task, void *calldata) |
0eecb214 AS |
753 | { |
754 | struct nfs_pgio_data *data = calldata; | |
755 | struct inode *inode = data->header->inode; | |
756 | ||
757 | dprintk("NFS: %s: %5u, (status %d)\n", __func__, | |
758 | task->tk_pid, task->tk_status); | |
759 | ||
760 | if (data->header->rw_ops->rw_done(task, data, inode) != 0) | |
761 | return; | |
762 | if (task->tk_status < 0) | |
763 | nfs_set_pgio_error(data->header, task->tk_status, data->args.offset); | |
764 | else | |
765 | data->header->rw_ops->rw_result(task, data); | |
766 | } | |
767 | ||
ef2c488c AS |
768 | /* |
769 | * Generate multiple small requests to read or write a single | |
770 | * contiguous dirty on one page. | |
771 | */ | |
772 | static int nfs_pgio_multi(struct nfs_pageio_descriptor *desc, | |
773 | struct nfs_pgio_header *hdr) | |
774 | { | |
775 | struct nfs_page *req = hdr->req; | |
776 | struct page *page = req->wb_page; | |
777 | struct nfs_pgio_data *data; | |
778 | size_t wsize = desc->pg_bsize, nbytes; | |
779 | unsigned int offset; | |
780 | int requests = 0; | |
781 | struct nfs_commit_info cinfo; | |
782 | ||
783 | nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); | |
784 | ||
785 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && | |
786 | (desc->pg_moreio || nfs_reqs_to_commit(&cinfo) || | |
787 | desc->pg_count > wsize)) | |
788 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
789 | ||
790 | offset = 0; | |
791 | nbytes = desc->pg_count; | |
792 | do { | |
793 | size_t len = min(nbytes, wsize); | |
794 | ||
795 | data = nfs_pgio_data_alloc(hdr, 1); | |
796 | if (!data) | |
797 | return nfs_pgio_error(desc, hdr); | |
798 | data->pages.pagevec[0] = page; | |
799 | nfs_pgio_rpcsetup(data, len, offset, desc->pg_ioflags, &cinfo); | |
800 | list_add(&data->list, &hdr->rpc_list); | |
801 | requests++; | |
802 | nbytes -= len; | |
803 | offset += len; | |
804 | } while (nbytes != 0); | |
805 | ||
806 | nfs_list_remove_request(req); | |
807 | nfs_list_add_request(req, &hdr->pages); | |
808 | desc->pg_rpc_callops = &nfs_pgio_common_ops; | |
809 | return 0; | |
810 | } | |
811 | ||
812 | /* | |
813 | * Create an RPC task for the given read or write request and kick it. | |
814 | * The page must have been locked by the caller. | |
815 | * | |
816 | * It may happen that the page we're passed is not marked dirty. | |
817 | * This is the case if nfs_updatepage detects a conflicting request | |
818 | * that has been written but not committed. | |
819 | */ | |
820 | static int nfs_pgio_one(struct nfs_pageio_descriptor *desc, | |
821 | struct nfs_pgio_header *hdr) | |
822 | { | |
823 | struct nfs_page *req; | |
824 | struct page **pages; | |
825 | struct nfs_pgio_data *data; | |
826 | struct list_head *head = &desc->pg_list; | |
827 | struct nfs_commit_info cinfo; | |
828 | ||
829 | data = nfs_pgio_data_alloc(hdr, nfs_page_array_len(desc->pg_base, | |
830 | desc->pg_count)); | |
831 | if (!data) | |
832 | return nfs_pgio_error(desc, hdr); | |
833 | ||
834 | nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); | |
835 | pages = data->pages.pagevec; | |
836 | while (!list_empty(head)) { | |
837 | req = nfs_list_entry(head->next); | |
838 | nfs_list_remove_request(req); | |
839 | nfs_list_add_request(req, &hdr->pages); | |
840 | *pages++ = req->wb_page; | |
841 | } | |
842 | ||
843 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && | |
844 | (desc->pg_moreio || nfs_reqs_to_commit(&cinfo))) | |
845 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
846 | ||
847 | /* Set up the argument struct */ | |
848 | nfs_pgio_rpcsetup(data, desc->pg_count, 0, desc->pg_ioflags, &cinfo); | |
849 | list_add(&data->list, &hdr->rpc_list); | |
850 | desc->pg_rpc_callops = &nfs_pgio_common_ops; | |
851 | return 0; | |
852 | } | |
853 | ||
41d8d5b7 | 854 | static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc) |
cf485fcd AS |
855 | { |
856 | struct nfs_rw_header *rw_hdr; | |
857 | struct nfs_pgio_header *hdr; | |
858 | int ret; | |
859 | ||
860 | rw_hdr = nfs_rw_header_alloc(desc->pg_rw_ops); | |
861 | if (!rw_hdr) { | |
862 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); | |
863 | return -ENOMEM; | |
864 | } | |
865 | hdr = &rw_hdr->header; | |
866 | nfs_pgheader_init(desc, hdr, nfs_rw_header_free); | |
867 | atomic_inc(&hdr->refcnt); | |
868 | ret = nfs_generic_pgio(desc, hdr); | |
869 | if (ret == 0) | |
870 | ret = nfs_do_multiple_pgios(&hdr->rpc_list, | |
871 | desc->pg_rpc_callops, | |
872 | desc->pg_ioflags); | |
873 | if (atomic_dec_and_test(&hdr->refcnt)) | |
874 | hdr->completion_ops->completion(hdr); | |
875 | return ret; | |
876 | } | |
877 | ||
ef2c488c AS |
878 | int nfs_generic_pgio(struct nfs_pageio_descriptor *desc, |
879 | struct nfs_pgio_header *hdr) | |
880 | { | |
881 | if (desc->pg_bsize < PAGE_CACHE_SIZE) | |
882 | return nfs_pgio_multi(desc, hdr); | |
883 | return nfs_pgio_one(desc, hdr); | |
884 | } | |
885 | EXPORT_SYMBOL_GPL(nfs_generic_pgio); | |
886 | ||
4109bb74 TM |
887 | static bool nfs_match_open_context(const struct nfs_open_context *ctx1, |
888 | const struct nfs_open_context *ctx2) | |
889 | { | |
890 | return ctx1->cred == ctx2->cred && ctx1->state == ctx2->state; | |
891 | } | |
892 | ||
893 | static bool nfs_match_lock_context(const struct nfs_lock_context *l1, | |
894 | const struct nfs_lock_context *l2) | |
895 | { | |
896 | return l1->lockowner.l_owner == l2->lockowner.l_owner | |
897 | && l1->lockowner.l_pid == l2->lockowner.l_pid; | |
898 | } | |
899 | ||
d8a5ad75 TM |
900 | /** |
901 | * nfs_can_coalesce_requests - test two requests for compatibility | |
902 | * @prev: pointer to nfs_page | |
903 | * @req: pointer to nfs_page | |
904 | * | |
905 | * The nfs_page structures 'prev' and 'req' are compared to ensure that the | |
906 | * page data area they describe is contiguous, and that their RPC | |
907 | * credentials, NFSv4 open state, and lockowners are the same. | |
908 | * | |
909 | * Return 'true' if this is the case, else return 'false'. | |
910 | */ | |
18ad0a9f BH |
911 | static bool nfs_can_coalesce_requests(struct nfs_page *prev, |
912 | struct nfs_page *req, | |
913 | struct nfs_pageio_descriptor *pgio) | |
d8a5ad75 | 914 | { |
b4fdac1a WAA |
915 | size_t size; |
916 | ||
ab75e417 WAA |
917 | if (prev) { |
918 | if (!nfs_match_open_context(req->wb_context, prev->wb_context)) | |
919 | return false; | |
920 | if (req->wb_context->dentry->d_inode->i_flock != NULL && | |
921 | !nfs_match_lock_context(req->wb_lock_context, | |
922 | prev->wb_lock_context)) | |
923 | return false; | |
ab75e417 WAA |
924 | if (req_offset(req) != req_offset(prev) + prev->wb_bytes) |
925 | return false; | |
926 | } | |
b4fdac1a WAA |
927 | size = pgio->pg_ops->pg_test(pgio, prev, req); |
928 | WARN_ON_ONCE(size && size != req->wb_bytes); | |
929 | return size > 0; | |
d8a5ad75 TM |
930 | } |
931 | ||
932 | /** | |
bcb71bba | 933 | * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list. |
d8a5ad75 TM |
934 | * @desc: destination io descriptor |
935 | * @req: request | |
936 | * | |
937 | * Returns true if the request 'req' was successfully coalesced into the | |
938 | * existing list of pages 'desc'. | |
939 | */ | |
bcb71bba TM |
940 | static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc, |
941 | struct nfs_page *req) | |
d8a5ad75 | 942 | { |
ab75e417 | 943 | struct nfs_page *prev = NULL; |
d8a5ad75 | 944 | if (desc->pg_count != 0) { |
d8a5ad75 | 945 | prev = nfs_list_entry(desc->pg_list.prev); |
5b36c7dc | 946 | } else { |
d8007d4d TM |
947 | if (desc->pg_ops->pg_init) |
948 | desc->pg_ops->pg_init(desc, req); | |
d8a5ad75 | 949 | desc->pg_base = req->wb_pgbase; |
5b36c7dc | 950 | } |
ab75e417 WAA |
951 | if (!nfs_can_coalesce_requests(prev, req, desc)) |
952 | return 0; | |
d8a5ad75 TM |
953 | nfs_list_remove_request(req); |
954 | nfs_list_add_request(req, &desc->pg_list); | |
5b36c7dc | 955 | desc->pg_count += req->wb_bytes; |
d8a5ad75 TM |
956 | return 1; |
957 | } | |
958 | ||
bcb71bba TM |
959 | /* |
960 | * Helper for nfs_pageio_add_request and nfs_pageio_complete | |
961 | */ | |
962 | static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc) | |
963 | { | |
964 | if (!list_empty(&desc->pg_list)) { | |
1751c363 | 965 | int error = desc->pg_ops->pg_doio(desc); |
bcb71bba TM |
966 | if (error < 0) |
967 | desc->pg_error = error; | |
968 | else | |
969 | desc->pg_bytes_written += desc->pg_count; | |
970 | } | |
971 | if (list_empty(&desc->pg_list)) { | |
972 | desc->pg_count = 0; | |
973 | desc->pg_base = 0; | |
974 | } | |
975 | } | |
976 | ||
977 | /** | |
978 | * nfs_pageio_add_request - Attempt to coalesce a request into a page list. | |
979 | * @desc: destination io descriptor | |
980 | * @req: request | |
981 | * | |
2bfc6e56 WAA |
982 | * This may split a request into subrequests which are all part of the |
983 | * same page group. | |
984 | * | |
bcb71bba TM |
985 | * Returns true if the request 'req' was successfully coalesced into the |
986 | * existing list of pages 'desc'. | |
987 | */ | |
d9156f9f | 988 | static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc, |
8b09bee3 | 989 | struct nfs_page *req) |
bcb71bba | 990 | { |
2bfc6e56 WAA |
991 | struct nfs_page *subreq; |
992 | unsigned int bytes_left = 0; | |
993 | unsigned int offset, pgbase; | |
994 | ||
995 | nfs_page_group_lock(req); | |
996 | ||
997 | subreq = req; | |
998 | bytes_left = subreq->wb_bytes; | |
999 | offset = subreq->wb_offset; | |
1000 | pgbase = subreq->wb_pgbase; | |
1001 | ||
1002 | do { | |
1003 | if (!nfs_pageio_do_add_request(desc, subreq)) { | |
1004 | /* make sure pg_test call(s) did nothing */ | |
1005 | WARN_ON_ONCE(subreq->wb_bytes != bytes_left); | |
1006 | WARN_ON_ONCE(subreq->wb_offset != offset); | |
1007 | WARN_ON_ONCE(subreq->wb_pgbase != pgbase); | |
1008 | ||
1009 | nfs_page_group_unlock(req); | |
1010 | desc->pg_moreio = 1; | |
1011 | nfs_pageio_doio(desc); | |
1012 | if (desc->pg_error < 0) | |
1013 | return 0; | |
1014 | desc->pg_moreio = 0; | |
1015 | if (desc->pg_recoalesce) | |
1016 | return 0; | |
1017 | /* retry add_request for this subreq */ | |
1018 | nfs_page_group_lock(req); | |
1019 | continue; | |
1020 | } | |
1021 | ||
1022 | /* check for buggy pg_test call(s) */ | |
1023 | WARN_ON_ONCE(subreq->wb_bytes + subreq->wb_pgbase > PAGE_SIZE); | |
1024 | WARN_ON_ONCE(subreq->wb_bytes > bytes_left); | |
1025 | WARN_ON_ONCE(subreq->wb_bytes == 0); | |
1026 | ||
1027 | bytes_left -= subreq->wb_bytes; | |
1028 | offset += subreq->wb_bytes; | |
1029 | pgbase += subreq->wb_bytes; | |
1030 | ||
1031 | if (bytes_left) { | |
1032 | subreq = nfs_create_request(req->wb_context, | |
1033 | req->wb_page, | |
1034 | subreq, pgbase, bytes_left); | |
1035 | nfs_lock_request(subreq); | |
1036 | subreq->wb_offset = offset; | |
1037 | subreq->wb_index = req->wb_index; | |
1038 | } | |
1039 | } while (bytes_left > 0); | |
1040 | ||
1041 | nfs_page_group_unlock(req); | |
bcb71bba TM |
1042 | return 1; |
1043 | } | |
1044 | ||
d9156f9f TM |
1045 | static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc) |
1046 | { | |
1047 | LIST_HEAD(head); | |
1048 | ||
1049 | do { | |
1050 | list_splice_init(&desc->pg_list, &head); | |
1051 | desc->pg_bytes_written -= desc->pg_count; | |
1052 | desc->pg_count = 0; | |
1053 | desc->pg_base = 0; | |
1054 | desc->pg_recoalesce = 0; | |
1055 | ||
1056 | while (!list_empty(&head)) { | |
1057 | struct nfs_page *req; | |
1058 | ||
1059 | req = list_first_entry(&head, struct nfs_page, wb_list); | |
1060 | nfs_list_remove_request(req); | |
1061 | if (__nfs_pageio_add_request(desc, req)) | |
1062 | continue; | |
1063 | if (desc->pg_error < 0) | |
1064 | return 0; | |
1065 | break; | |
1066 | } | |
1067 | } while (desc->pg_recoalesce); | |
1068 | return 1; | |
1069 | } | |
1070 | ||
1071 | int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc, | |
1072 | struct nfs_page *req) | |
1073 | { | |
1074 | int ret; | |
1075 | ||
1076 | do { | |
1077 | ret = __nfs_pageio_add_request(desc, req); | |
1078 | if (ret) | |
1079 | break; | |
1080 | if (desc->pg_error < 0) | |
1081 | break; | |
1082 | ret = nfs_do_recoalesce(desc); | |
1083 | } while (ret); | |
1084 | return ret; | |
1085 | } | |
89d77c8f | 1086 | EXPORT_SYMBOL_GPL(nfs_pageio_add_request); |
d9156f9f | 1087 | |
bcb71bba TM |
1088 | /** |
1089 | * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor | |
1090 | * @desc: pointer to io descriptor | |
1091 | */ | |
1092 | void nfs_pageio_complete(struct nfs_pageio_descriptor *desc) | |
1093 | { | |
d9156f9f TM |
1094 | for (;;) { |
1095 | nfs_pageio_doio(desc); | |
1096 | if (!desc->pg_recoalesce) | |
1097 | break; | |
1098 | if (!nfs_do_recoalesce(desc)) | |
1099 | break; | |
1100 | } | |
bcb71bba | 1101 | } |
89d77c8f | 1102 | EXPORT_SYMBOL_GPL(nfs_pageio_complete); |
bcb71bba | 1103 | |
7fe7f848 TM |
1104 | /** |
1105 | * nfs_pageio_cond_complete - Conditional I/O completion | |
1106 | * @desc: pointer to io descriptor | |
1107 | * @index: page index | |
1108 | * | |
1109 | * It is important to ensure that processes don't try to take locks | |
1110 | * on non-contiguous ranges of pages as that might deadlock. This | |
1111 | * function should be called before attempting to wait on a locked | |
1112 | * nfs_page. It will complete the I/O if the page index 'index' | |
1113 | * is not contiguous with the existing list of pages in 'desc'. | |
1114 | */ | |
1115 | void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index) | |
1116 | { | |
1117 | if (!list_empty(&desc->pg_list)) { | |
1118 | struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev); | |
1119 | if (index != prev->wb_index + 1) | |
d9156f9f | 1120 | nfs_pageio_complete(desc); |
7fe7f848 TM |
1121 | } |
1122 | } | |
1123 | ||
f7b422b1 | 1124 | int __init nfs_init_nfspagecache(void) |
1da177e4 LT |
1125 | { |
1126 | nfs_page_cachep = kmem_cache_create("nfs_page", | |
1127 | sizeof(struct nfs_page), | |
1128 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1129 | NULL); |
1da177e4 LT |
1130 | if (nfs_page_cachep == NULL) |
1131 | return -ENOMEM; | |
1132 | ||
1133 | return 0; | |
1134 | } | |
1135 | ||
266bee88 | 1136 | void nfs_destroy_nfspagecache(void) |
1da177e4 | 1137 | { |
1a1d92c1 | 1138 | kmem_cache_destroy(nfs_page_cachep); |
1da177e4 LT |
1139 | } |
1140 | ||
ef2c488c | 1141 | static const struct rpc_call_ops nfs_pgio_common_ops = { |
6f92fa45 AS |
1142 | .rpc_call_prepare = nfs_pgio_prepare, |
1143 | .rpc_call_done = nfs_pgio_result, | |
1144 | .rpc_release = nfs_pgio_release, | |
1145 | }; | |
41d8d5b7 AS |
1146 | |
1147 | const struct nfs_pageio_ops nfs_pgio_rw_ops = { | |
1148 | .pg_test = nfs_generic_pg_test, | |
1149 | .pg_doio = nfs_generic_pg_pgios, | |
1150 | }; |