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 | |
00bfa30a | 32 | static bool nfs_pgarray_set(struct nfs_page_array *p, unsigned int pagecount) |
30dd374f FI |
33 | { |
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 | ||
48d635f1 PT |
45 | struct nfs_pgio_mirror * |
46 | nfs_pgio_current_mirror(struct nfs_pageio_descriptor *desc) | |
47 | { | |
48 | return nfs_pgio_has_mirroring(desc) ? | |
49 | &desc->pg_mirrors[desc->pg_mirror_idx] : | |
50 | &desc->pg_mirrors[0]; | |
51 | } | |
52 | EXPORT_SYMBOL_GPL(nfs_pgio_current_mirror); | |
53 | ||
4db6e0b7 FI |
54 | void nfs_pgheader_init(struct nfs_pageio_descriptor *desc, |
55 | struct nfs_pgio_header *hdr, | |
56 | void (*release)(struct nfs_pgio_header *hdr)) | |
57 | { | |
48d635f1 | 58 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb WAA |
59 | |
60 | ||
61 | hdr->req = nfs_list_entry(mirror->pg_list.next); | |
4db6e0b7 FI |
62 | hdr->inode = desc->pg_inode; |
63 | hdr->cred = hdr->req->wb_context->cred; | |
64 | hdr->io_start = req_offset(hdr->req); | |
a7d42ddb | 65 | hdr->good_bytes = mirror->pg_count; |
584aa810 | 66 | hdr->dreq = desc->pg_dreq; |
f6166384 | 67 | hdr->layout_private = desc->pg_layout_private; |
4db6e0b7 | 68 | hdr->release = release; |
061ae2ed | 69 | hdr->completion_ops = desc->pg_completion_ops; |
584aa810 FI |
70 | if (hdr->completion_ops->init_hdr) |
71 | hdr->completion_ops->init_hdr(hdr); | |
a7d42ddb WAA |
72 | |
73 | hdr->pgio_mirror_idx = desc->pg_mirror_idx; | |
4db6e0b7 | 74 | } |
89d77c8f | 75 | EXPORT_SYMBOL_GPL(nfs_pgheader_init); |
4db6e0b7 FI |
76 | |
77 | void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos) | |
78 | { | |
79 | spin_lock(&hdr->lock); | |
80 | if (pos < hdr->io_start + hdr->good_bytes) { | |
81 | set_bit(NFS_IOHDR_ERROR, &hdr->flags); | |
82 | clear_bit(NFS_IOHDR_EOF, &hdr->flags); | |
83 | hdr->good_bytes = pos - hdr->io_start; | |
84 | hdr->error = error; | |
85 | } | |
86 | spin_unlock(&hdr->lock); | |
87 | } | |
88 | ||
1da177e4 LT |
89 | static inline struct nfs_page * |
90 | nfs_page_alloc(void) | |
91 | { | |
192e501b | 92 | struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_NOIO); |
72895b1a | 93 | if (p) |
1da177e4 | 94 | INIT_LIST_HEAD(&p->wb_list); |
1da177e4 LT |
95 | return p; |
96 | } | |
97 | ||
98 | static inline void | |
99 | nfs_page_free(struct nfs_page *p) | |
100 | { | |
101 | kmem_cache_free(nfs_page_cachep, p); | |
102 | } | |
103 | ||
577b4232 TM |
104 | static void |
105 | nfs_iocounter_inc(struct nfs_io_counter *c) | |
106 | { | |
107 | atomic_inc(&c->io_count); | |
108 | } | |
109 | ||
110 | static void | |
111 | nfs_iocounter_dec(struct nfs_io_counter *c) | |
112 | { | |
113 | if (atomic_dec_and_test(&c->io_count)) { | |
114 | clear_bit(NFS_IO_INPROGRESS, &c->flags); | |
4e857c58 | 115 | smp_mb__after_atomic(); |
577b4232 TM |
116 | wake_up_bit(&c->flags, NFS_IO_INPROGRESS); |
117 | } | |
118 | } | |
119 | ||
120 | static int | |
121 | __nfs_iocounter_wait(struct nfs_io_counter *c) | |
122 | { | |
123 | wait_queue_head_t *wq = bit_waitqueue(&c->flags, NFS_IO_INPROGRESS); | |
124 | DEFINE_WAIT_BIT(q, &c->flags, NFS_IO_INPROGRESS); | |
125 | int ret = 0; | |
126 | ||
127 | do { | |
128 | prepare_to_wait(wq, &q.wait, TASK_KILLABLE); | |
129 | set_bit(NFS_IO_INPROGRESS, &c->flags); | |
130 | if (atomic_read(&c->io_count) == 0) | |
131 | break; | |
c1221321 | 132 | ret = nfs_wait_bit_killable(&q.key); |
92a56555 | 133 | } while (atomic_read(&c->io_count) != 0 && !ret); |
577b4232 TM |
134 | finish_wait(wq, &q.wait); |
135 | return ret; | |
136 | } | |
137 | ||
138 | /** | |
139 | * nfs_iocounter_wait - wait for i/o to complete | |
140 | * @c: nfs_io_counter to use | |
141 | * | |
142 | * returns -ERESTARTSYS if interrupted by a fatal signal. | |
143 | * Otherwise returns 0 once the io_count hits 0. | |
144 | */ | |
145 | int | |
146 | nfs_iocounter_wait(struct nfs_io_counter *c) | |
147 | { | |
148 | if (atomic_read(&c->io_count) == 0) | |
149 | return 0; | |
150 | return __nfs_iocounter_wait(c); | |
151 | } | |
152 | ||
2bfc6e56 WAA |
153 | /* |
154 | * nfs_page_group_lock - lock the head of the page group | |
155 | * @req - request in group that is to be locked | |
fd2f3a06 | 156 | * @nonblock - if true don't block waiting for lock |
2bfc6e56 WAA |
157 | * |
158 | * this lock must be held if modifying the page group list | |
e7029206 | 159 | * |
bc8a309e WAA |
160 | * return 0 on success, < 0 on error: -EDELAY if nonblocking or the |
161 | * result from wait_on_bit_lock | |
162 | * | |
163 | * NOTE: calling with nonblock=false should always have set the | |
164 | * lock bit (see fs/buffer.c and other uses of wait_on_bit_lock | |
165 | * with TASK_UNINTERRUPTIBLE), so there is no need to check the result. | |
2bfc6e56 | 166 | */ |
e7029206 | 167 | int |
fd2f3a06 | 168 | nfs_page_group_lock(struct nfs_page *req, bool nonblock) |
2bfc6e56 WAA |
169 | { |
170 | struct nfs_page *head = req->wb_head; | |
2bfc6e56 WAA |
171 | |
172 | WARN_ON_ONCE(head != head->wb_head); | |
173 | ||
bc8a309e WAA |
174 | if (!test_and_set_bit(PG_HEADLOCK, &head->wb_flags)) |
175 | return 0; | |
e7029206 | 176 | |
bc8a309e WAA |
177 | if (!nonblock) |
178 | return wait_on_bit_lock(&head->wb_flags, PG_HEADLOCK, | |
179 | TASK_UNINTERRUPTIBLE); | |
180 | ||
181 | return -EAGAIN; | |
2bfc6e56 WAA |
182 | } |
183 | ||
7c3af975 WAA |
184 | /* |
185 | * nfs_page_group_lock_wait - wait for the lock to clear, but don't grab it | |
186 | * @req - a request in the group | |
187 | * | |
188 | * This is a blocking call to wait for the group lock to be cleared. | |
189 | */ | |
190 | void | |
191 | nfs_page_group_lock_wait(struct nfs_page *req) | |
192 | { | |
193 | struct nfs_page *head = req->wb_head; | |
194 | ||
195 | WARN_ON_ONCE(head != head->wb_head); | |
196 | ||
197 | wait_on_bit(&head->wb_flags, PG_HEADLOCK, | |
198 | TASK_UNINTERRUPTIBLE); | |
199 | } | |
200 | ||
2bfc6e56 WAA |
201 | /* |
202 | * nfs_page_group_unlock - unlock the head of the page group | |
203 | * @req - request in group that is to be unlocked | |
204 | */ | |
205 | void | |
206 | nfs_page_group_unlock(struct nfs_page *req) | |
207 | { | |
208 | struct nfs_page *head = req->wb_head; | |
209 | ||
210 | WARN_ON_ONCE(head != head->wb_head); | |
211 | ||
d1e1cda8 | 212 | smp_mb__before_atomic(); |
2bfc6e56 | 213 | clear_bit(PG_HEADLOCK, &head->wb_flags); |
d1e1cda8 | 214 | smp_mb__after_atomic(); |
2bfc6e56 WAA |
215 | wake_up_bit(&head->wb_flags, PG_HEADLOCK); |
216 | } | |
217 | ||
218 | /* | |
219 | * nfs_page_group_sync_on_bit_locked | |
220 | * | |
221 | * must be called with page group lock held | |
222 | */ | |
223 | static bool | |
224 | nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit) | |
225 | { | |
226 | struct nfs_page *head = req->wb_head; | |
227 | struct nfs_page *tmp; | |
228 | ||
229 | WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags)); | |
230 | WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags)); | |
231 | ||
232 | tmp = req->wb_this_page; | |
233 | while (tmp != req) { | |
234 | if (!test_bit(bit, &tmp->wb_flags)) | |
235 | return false; | |
236 | tmp = tmp->wb_this_page; | |
237 | } | |
238 | ||
239 | /* true! reset all bits */ | |
240 | tmp = req; | |
241 | do { | |
242 | clear_bit(bit, &tmp->wb_flags); | |
243 | tmp = tmp->wb_this_page; | |
244 | } while (tmp != req); | |
245 | ||
246 | return true; | |
247 | } | |
248 | ||
249 | /* | |
250 | * nfs_page_group_sync_on_bit - set bit on current request, but only | |
251 | * return true if the bit is set for all requests in page group | |
252 | * @req - request in page group | |
253 | * @bit - PG_* bit that is used to sync page group | |
254 | */ | |
255 | bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit) | |
256 | { | |
257 | bool ret; | |
258 | ||
fd2f3a06 | 259 | nfs_page_group_lock(req, false); |
2bfc6e56 WAA |
260 | ret = nfs_page_group_sync_on_bit_locked(req, bit); |
261 | nfs_page_group_unlock(req); | |
262 | ||
263 | return ret; | |
264 | } | |
265 | ||
266 | /* | |
267 | * nfs_page_group_init - Initialize the page group linkage for @req | |
268 | * @req - a new nfs request | |
269 | * @prev - the previous request in page group, or NULL if @req is the first | |
270 | * or only request in the group (the head). | |
271 | */ | |
272 | static inline void | |
273 | nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev) | |
274 | { | |
cb1410c7 | 275 | struct inode *inode; |
2bfc6e56 WAA |
276 | WARN_ON_ONCE(prev == req); |
277 | ||
278 | if (!prev) { | |
85710a83 | 279 | /* a head request */ |
2bfc6e56 WAA |
280 | req->wb_head = req; |
281 | req->wb_this_page = req; | |
282 | } else { | |
85710a83 | 283 | /* a subrequest */ |
2bfc6e56 WAA |
284 | WARN_ON_ONCE(prev->wb_this_page != prev->wb_head); |
285 | WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags)); | |
286 | req->wb_head = prev->wb_head; | |
287 | req->wb_this_page = prev->wb_this_page; | |
288 | prev->wb_this_page = req; | |
289 | ||
85710a83 WAA |
290 | /* All subrequests take a ref on the head request until |
291 | * nfs_page_group_destroy is called */ | |
292 | kref_get(&req->wb_head->wb_kref); | |
293 | ||
cb1410c7 WAA |
294 | /* grab extra ref and bump the request count if head request |
295 | * has extra ref from the write/commit path to handle handoff | |
296 | * between write and commit lists. */ | |
17089a29 | 297 | if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags)) { |
cb1410c7 | 298 | inode = page_file_mapping(req->wb_page)->host; |
17089a29 | 299 | set_bit(PG_INODE_REF, &req->wb_flags); |
2bfc6e56 | 300 | kref_get(&req->wb_kref); |
cb1410c7 WAA |
301 | spin_lock(&inode->i_lock); |
302 | NFS_I(inode)->nrequests++; | |
303 | spin_unlock(&inode->i_lock); | |
17089a29 | 304 | } |
2bfc6e56 WAA |
305 | } |
306 | } | |
307 | ||
308 | /* | |
309 | * nfs_page_group_destroy - sync the destruction of page groups | |
310 | * @req - request that no longer needs the page group | |
311 | * | |
312 | * releases the page group reference from each member once all | |
313 | * members have called this function. | |
314 | */ | |
315 | static void | |
316 | nfs_page_group_destroy(struct kref *kref) | |
317 | { | |
318 | struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref); | |
319 | struct nfs_page *tmp, *next; | |
320 | ||
85710a83 WAA |
321 | /* subrequests must release the ref on the head request */ |
322 | if (req->wb_head != req) | |
323 | nfs_release_request(req->wb_head); | |
324 | ||
2bfc6e56 WAA |
325 | if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN)) |
326 | return; | |
327 | ||
328 | tmp = req; | |
329 | do { | |
330 | next = tmp->wb_this_page; | |
331 | /* unlink and free */ | |
332 | tmp->wb_this_page = tmp; | |
333 | tmp->wb_head = tmp; | |
334 | nfs_free_request(tmp); | |
335 | tmp = next; | |
336 | } while (tmp != req); | |
337 | } | |
338 | ||
1da177e4 LT |
339 | /** |
340 | * nfs_create_request - Create an NFS read/write request. | |
c02f557d | 341 | * @ctx: open context to use |
1da177e4 | 342 | * @page: page to write |
2bfc6e56 | 343 | * @last: last nfs request created for this page group or NULL if head |
1da177e4 LT |
344 | * @offset: starting offset within the page for the write |
345 | * @count: number of bytes to read/write | |
346 | * | |
347 | * The page must be locked by the caller. This makes sure we never | |
a19b89ca | 348 | * create two different requests for the same page. |
1da177e4 LT |
349 | * User should ensure it is safe to sleep in this function. |
350 | */ | |
351 | struct nfs_page * | |
8c8f1ac1 | 352 | nfs_create_request(struct nfs_open_context *ctx, struct page *page, |
2bfc6e56 WAA |
353 | struct nfs_page *last, unsigned int offset, |
354 | unsigned int count) | |
1da177e4 | 355 | { |
1da177e4 | 356 | struct nfs_page *req; |
b3c54de6 | 357 | struct nfs_lock_context *l_ctx; |
1da177e4 | 358 | |
c58c8441 TM |
359 | if (test_bit(NFS_CONTEXT_BAD, &ctx->flags)) |
360 | return ERR_PTR(-EBADF); | |
18eb8842 TM |
361 | /* try to allocate the request struct */ |
362 | req = nfs_page_alloc(); | |
363 | if (req == NULL) | |
364 | return ERR_PTR(-ENOMEM); | |
1da177e4 | 365 | |
015f0212 | 366 | /* get lock context early so we can deal with alloc failures */ |
b3c54de6 TM |
367 | l_ctx = nfs_get_lock_context(ctx); |
368 | if (IS_ERR(l_ctx)) { | |
015f0212 | 369 | nfs_page_free(req); |
b3c54de6 | 370 | return ERR_CAST(l_ctx); |
015f0212 | 371 | } |
b3c54de6 | 372 | req->wb_lock_context = l_ctx; |
577b4232 | 373 | nfs_iocounter_inc(&l_ctx->io_count); |
015f0212 | 374 | |
1da177e4 LT |
375 | /* Initialize the request struct. Initially, we assume a |
376 | * long write-back delay. This will be adjusted in | |
377 | * update_nfs_request below if the region is not locked. */ | |
378 | req->wb_page = page; | |
d56b4ddf | 379 | req->wb_index = page_file_index(page); |
1da177e4 LT |
380 | page_cache_get(page); |
381 | req->wb_offset = offset; | |
382 | req->wb_pgbase = offset; | |
383 | req->wb_bytes = count; | |
1da177e4 | 384 | req->wb_context = get_nfs_open_context(ctx); |
c03b4024 | 385 | kref_init(&req->wb_kref); |
2bfc6e56 | 386 | nfs_page_group_init(req, last); |
1da177e4 LT |
387 | return req; |
388 | } | |
389 | ||
390 | /** | |
1d1afcbc | 391 | * nfs_unlock_request - Unlock request and wake up sleepers. |
1da177e4 LT |
392 | * @req: |
393 | */ | |
1d1afcbc | 394 | void nfs_unlock_request(struct nfs_page *req) |
1da177e4 LT |
395 | { |
396 | if (!NFS_WBACK_BUSY(req)) { | |
397 | printk(KERN_ERR "NFS: Invalid unlock attempted\n"); | |
398 | BUG(); | |
399 | } | |
4e857c58 | 400 | smp_mb__before_atomic(); |
1da177e4 | 401 | clear_bit(PG_BUSY, &req->wb_flags); |
4e857c58 | 402 | smp_mb__after_atomic(); |
464a98bd | 403 | wake_up_bit(&req->wb_flags, PG_BUSY); |
3aff4ebb TM |
404 | } |
405 | ||
406 | /** | |
1d1afcbc TM |
407 | * nfs_unlock_and_release_request - Unlock request and release the nfs_page |
408 | * @req: | |
3aff4ebb | 409 | */ |
1d1afcbc | 410 | void nfs_unlock_and_release_request(struct nfs_page *req) |
3aff4ebb | 411 | { |
1d1afcbc | 412 | nfs_unlock_request(req); |
1da177e4 LT |
413 | nfs_release_request(req); |
414 | } | |
415 | ||
4d65c520 | 416 | /* |
1da177e4 LT |
417 | * nfs_clear_request - Free up all resources allocated to the request |
418 | * @req: | |
419 | * | |
bb6fbc45 TM |
420 | * Release page and open context resources associated with a read/write |
421 | * request after it has completed. | |
1da177e4 | 422 | */ |
4d65c520 | 423 | static void nfs_clear_request(struct nfs_page *req) |
1da177e4 | 424 | { |
cd52ed35 | 425 | struct page *page = req->wb_page; |
bb6fbc45 | 426 | struct nfs_open_context *ctx = req->wb_context; |
f11ac8db | 427 | struct nfs_lock_context *l_ctx = req->wb_lock_context; |
bb6fbc45 | 428 | |
cd52ed35 | 429 | if (page != NULL) { |
cd52ed35 | 430 | page_cache_release(page); |
1da177e4 LT |
431 | req->wb_page = NULL; |
432 | } | |
f11ac8db | 433 | if (l_ctx != NULL) { |
577b4232 | 434 | nfs_iocounter_dec(&l_ctx->io_count); |
f11ac8db TM |
435 | nfs_put_lock_context(l_ctx); |
436 | req->wb_lock_context = NULL; | |
437 | } | |
bb6fbc45 TM |
438 | if (ctx != NULL) { |
439 | put_nfs_open_context(ctx); | |
440 | req->wb_context = NULL; | |
441 | } | |
1da177e4 LT |
442 | } |
443 | ||
1da177e4 LT |
444 | /** |
445 | * nfs_release_request - Release the count on an NFS read/write request | |
446 | * @req: request to release | |
447 | * | |
448 | * Note: Should never be called with the spinlock held! | |
449 | */ | |
d4581383 | 450 | void nfs_free_request(struct nfs_page *req) |
1da177e4 | 451 | { |
2bfc6e56 WAA |
452 | WARN_ON_ONCE(req->wb_this_page != req); |
453 | ||
454 | /* extra debug: make sure no sync bits are still set */ | |
455 | WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags)); | |
67d0338e WAA |
456 | WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags)); |
457 | WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags)); | |
20633f04 WAA |
458 | WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags)); |
459 | WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags)); | |
1da177e4 | 460 | |
bb6fbc45 | 461 | /* Release struct file and open context */ |
1da177e4 | 462 | nfs_clear_request(req); |
1da177e4 LT |
463 | nfs_page_free(req); |
464 | } | |
465 | ||
c03b4024 TM |
466 | void nfs_release_request(struct nfs_page *req) |
467 | { | |
2bfc6e56 | 468 | kref_put(&req->wb_kref, nfs_page_group_destroy); |
9f557cd8 TM |
469 | } |
470 | ||
1da177e4 LT |
471 | /** |
472 | * nfs_wait_on_request - Wait for a request to complete. | |
473 | * @req: request to wait upon. | |
474 | * | |
150030b7 | 475 | * Interruptible by fatal signals only. |
1da177e4 LT |
476 | * The user is responsible for holding a count on the request. |
477 | */ | |
478 | int | |
479 | nfs_wait_on_request(struct nfs_page *req) | |
480 | { | |
74316201 N |
481 | return wait_on_bit_io(&req->wb_flags, PG_BUSY, |
482 | TASK_UNINTERRUPTIBLE); | |
1da177e4 LT |
483 | } |
484 | ||
b4fdac1a WAA |
485 | /* |
486 | * nfs_generic_pg_test - determine if requests can be coalesced | |
487 | * @desc: pointer to descriptor | |
488 | * @prev: previous request in desc, or NULL | |
489 | * @req: this request | |
490 | * | |
491 | * Returns zero if @req can be coalesced into @desc, otherwise it returns | |
492 | * the size of the request. | |
493 | */ | |
494 | size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc, | |
495 | struct nfs_page *prev, struct nfs_page *req) | |
5b36c7dc | 496 | { |
48d635f1 | 497 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb WAA |
498 | |
499 | ||
500 | if (mirror->pg_count > mirror->pg_bsize) { | |
f0cb9ab8 WAA |
501 | /* should never happen */ |
502 | WARN_ON_ONCE(1); | |
5b36c7dc | 503 | return 0; |
f0cb9ab8 | 504 | } |
5b36c7dc | 505 | |
2e11f829 CH |
506 | /* |
507 | * Limit the request size so that we can still allocate a page array | |
508 | * for it without upsetting the slab allocator. | |
509 | */ | |
a7d42ddb | 510 | if (((mirror->pg_count + req->wb_bytes) >> PAGE_SHIFT) * |
2e11f829 CH |
511 | sizeof(struct page) > PAGE_SIZE) |
512 | return 0; | |
513 | ||
a7d42ddb | 514 | return min(mirror->pg_bsize - mirror->pg_count, (size_t)req->wb_bytes); |
5b36c7dc | 515 | } |
19345cb2 | 516 | EXPORT_SYMBOL_GPL(nfs_generic_pg_test); |
5b36c7dc | 517 | |
1e7f3a48 | 518 | struct nfs_pgio_header *nfs_pgio_header_alloc(const struct nfs_rw_ops *ops) |
4a0de55c | 519 | { |
1e7f3a48 | 520 | struct nfs_pgio_header *hdr = ops->rw_alloc_header(); |
4a0de55c | 521 | |
1e7f3a48 | 522 | if (hdr) { |
4a0de55c | 523 | INIT_LIST_HEAD(&hdr->pages); |
4a0de55c | 524 | spin_lock_init(&hdr->lock); |
4a0de55c AS |
525 | hdr->rw_ops = ops; |
526 | } | |
1e7f3a48 | 527 | return hdr; |
4a0de55c | 528 | } |
1e7f3a48 | 529 | EXPORT_SYMBOL_GPL(nfs_pgio_header_alloc); |
4a0de55c AS |
530 | |
531 | /* | |
1e7f3a48 | 532 | * nfs_pgio_header_free - Free a read or write header |
4a0de55c AS |
533 | * @hdr: The header to free |
534 | */ | |
1e7f3a48 | 535 | void nfs_pgio_header_free(struct nfs_pgio_header *hdr) |
4a0de55c | 536 | { |
1e7f3a48 | 537 | hdr->rw_ops->rw_free_header(hdr); |
4a0de55c | 538 | } |
1e7f3a48 | 539 | EXPORT_SYMBOL_GPL(nfs_pgio_header_free); |
4a0de55c | 540 | |
00bfa30a | 541 | /** |
4714fb51 WAA |
542 | * nfs_pgio_data_destroy - make @hdr suitable for reuse |
543 | * | |
544 | * Frees memory and releases refs from nfs_generic_pgio, so that it may | |
545 | * be called again. | |
546 | * | |
547 | * @hdr: A header that has had nfs_generic_pgio called | |
00bfa30a | 548 | */ |
d45f60c6 | 549 | void nfs_pgio_data_destroy(struct nfs_pgio_header *hdr) |
00bfa30a | 550 | { |
3caa0c6e TM |
551 | if (hdr->args.context) |
552 | put_nfs_open_context(hdr->args.context); | |
d45f60c6 WAA |
553 | if (hdr->page_array.pagevec != hdr->page_array.page_array) |
554 | kfree(hdr->page_array.pagevec); | |
00bfa30a | 555 | } |
1e7f3a48 | 556 | EXPORT_SYMBOL_GPL(nfs_pgio_data_destroy); |
00bfa30a | 557 | |
ce59515c AS |
558 | /** |
559 | * nfs_pgio_rpcsetup - Set up arguments for a pageio call | |
d45f60c6 | 560 | * @hdr: The pageio hdr |
ce59515c AS |
561 | * @count: Number of bytes to read |
562 | * @offset: Initial offset | |
563 | * @how: How to commit data (writes only) | |
564 | * @cinfo: Commit information for the call (writes only) | |
565 | */ | |
d45f60c6 | 566 | static void nfs_pgio_rpcsetup(struct nfs_pgio_header *hdr, |
ce59515c AS |
567 | unsigned int count, unsigned int offset, |
568 | int how, struct nfs_commit_info *cinfo) | |
569 | { | |
d45f60c6 | 570 | struct nfs_page *req = hdr->req; |
ce59515c AS |
571 | |
572 | /* Set up the RPC argument and reply structs | |
d45f60c6 | 573 | * NB: take care not to mess about with hdr->commit et al. */ |
ce59515c | 574 | |
d45f60c6 WAA |
575 | hdr->args.fh = NFS_FH(hdr->inode); |
576 | hdr->args.offset = req_offset(req) + offset; | |
ce59515c | 577 | /* pnfs_set_layoutcommit needs this */ |
d45f60c6 WAA |
578 | hdr->mds_offset = hdr->args.offset; |
579 | hdr->args.pgbase = req->wb_pgbase + offset; | |
580 | hdr->args.pages = hdr->page_array.pagevec; | |
581 | hdr->args.count = count; | |
582 | hdr->args.context = get_nfs_open_context(req->wb_context); | |
583 | hdr->args.lock_context = req->wb_lock_context; | |
584 | hdr->args.stable = NFS_UNSTABLE; | |
ce59515c AS |
585 | switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { |
586 | case 0: | |
587 | break; | |
588 | case FLUSH_COND_STABLE: | |
589 | if (nfs_reqs_to_commit(cinfo)) | |
590 | break; | |
591 | default: | |
d45f60c6 | 592 | hdr->args.stable = NFS_FILE_SYNC; |
ce59515c AS |
593 | } |
594 | ||
d45f60c6 WAA |
595 | hdr->res.fattr = &hdr->fattr; |
596 | hdr->res.count = count; | |
597 | hdr->res.eof = 0; | |
c65e6254 | 598 | hdr->res.verf = &hdr->verf; |
d45f60c6 | 599 | nfs_fattr_init(&hdr->fattr); |
ce59515c AS |
600 | } |
601 | ||
a4cdda59 | 602 | /** |
d45f60c6 | 603 | * nfs_pgio_prepare - Prepare pageio hdr to go over the wire |
a4cdda59 | 604 | * @task: The current task |
d45f60c6 | 605 | * @calldata: pageio header to prepare |
a4cdda59 | 606 | */ |
6f92fa45 | 607 | static void nfs_pgio_prepare(struct rpc_task *task, void *calldata) |
a4cdda59 | 608 | { |
d45f60c6 | 609 | struct nfs_pgio_header *hdr = calldata; |
a4cdda59 | 610 | int err; |
d45f60c6 | 611 | err = NFS_PROTO(hdr->inode)->pgio_rpc_prepare(task, hdr); |
a4cdda59 AS |
612 | if (err) |
613 | rpc_exit(task, err); | |
614 | } | |
615 | ||
d45f60c6 | 616 | int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr, |
46a5ab47 | 617 | struct rpc_cred *cred, const struct nfs_rpc_ops *rpc_ops, |
1ed26f33 AS |
618 | const struct rpc_call_ops *call_ops, int how, int flags) |
619 | { | |
620 | struct rpc_task *task; | |
621 | struct rpc_message msg = { | |
d45f60c6 WAA |
622 | .rpc_argp = &hdr->args, |
623 | .rpc_resp = &hdr->res, | |
46a5ab47 | 624 | .rpc_cred = cred, |
1ed26f33 AS |
625 | }; |
626 | struct rpc_task_setup task_setup_data = { | |
627 | .rpc_client = clnt, | |
d45f60c6 | 628 | .task = &hdr->task, |
1ed26f33 AS |
629 | .rpc_message = &msg, |
630 | .callback_ops = call_ops, | |
d45f60c6 | 631 | .callback_data = hdr, |
1ed26f33 AS |
632 | .workqueue = nfsiod_workqueue, |
633 | .flags = RPC_TASK_ASYNC | flags, | |
634 | }; | |
635 | int ret = 0; | |
636 | ||
abde71f4 | 637 | hdr->rw_ops->rw_initiate(hdr, &msg, rpc_ops, &task_setup_data, how); |
1ed26f33 | 638 | |
b4839ebe | 639 | dprintk("NFS: initiated pgio call " |
1ed26f33 | 640 | "(req %s/%llu, %u bytes @ offset %llu)\n", |
343ae531 AS |
641 | hdr->inode->i_sb->s_id, |
642 | (unsigned long long)NFS_FILEID(hdr->inode), | |
d45f60c6 WAA |
643 | hdr->args.count, |
644 | (unsigned long long)hdr->args.offset); | |
1ed26f33 AS |
645 | |
646 | task = rpc_run_task(&task_setup_data); | |
647 | if (IS_ERR(task)) { | |
648 | ret = PTR_ERR(task); | |
649 | goto out; | |
650 | } | |
651 | if (how & FLUSH_SYNC) { | |
652 | ret = rpc_wait_for_completion_task(task); | |
653 | if (ret == 0) | |
654 | ret = task->tk_status; | |
655 | } | |
656 | rpc_put_task(task); | |
657 | out: | |
658 | return ret; | |
659 | } | |
660 | EXPORT_SYMBOL_GPL(nfs_initiate_pgio); | |
661 | ||
844c9e69 AS |
662 | /** |
663 | * nfs_pgio_error - Clean up from a pageio error | |
664 | * @desc: IO descriptor | |
665 | * @hdr: pageio header | |
666 | */ | |
ef2c488c | 667 | static int nfs_pgio_error(struct nfs_pageio_descriptor *desc, |
844c9e69 AS |
668 | struct nfs_pgio_header *hdr) |
669 | { | |
a7d42ddb WAA |
670 | struct nfs_pgio_mirror *mirror; |
671 | u32 midx; | |
672 | ||
844c9e69 | 673 | set_bit(NFS_IOHDR_REDO, &hdr->flags); |
d45f60c6 | 674 | nfs_pgio_data_destroy(hdr); |
4714fb51 | 675 | hdr->completion_ops->completion(hdr); |
a7d42ddb WAA |
676 | /* TODO: Make sure it's right to clean up all mirrors here |
677 | * and not just hdr->pgio_mirror_idx */ | |
678 | for (midx = 0; midx < desc->pg_mirror_count; midx++) { | |
679 | mirror = &desc->pg_mirrors[midx]; | |
680 | desc->pg_completion_ops->error_cleanup(&mirror->pg_list); | |
681 | } | |
844c9e69 AS |
682 | return -ENOMEM; |
683 | } | |
684 | ||
a4cdda59 AS |
685 | /** |
686 | * nfs_pgio_release - Release pageio data | |
d45f60c6 | 687 | * @calldata: The pageio header to release |
a4cdda59 | 688 | */ |
6f92fa45 | 689 | static void nfs_pgio_release(void *calldata) |
a4cdda59 | 690 | { |
d45f60c6 | 691 | struct nfs_pgio_header *hdr = calldata; |
d45f60c6 | 692 | nfs_pgio_data_destroy(hdr); |
4714fb51 | 693 | hdr->completion_ops->completion(hdr); |
a4cdda59 AS |
694 | } |
695 | ||
a7d42ddb WAA |
696 | static void nfs_pageio_mirror_init(struct nfs_pgio_mirror *mirror, |
697 | unsigned int bsize) | |
698 | { | |
699 | INIT_LIST_HEAD(&mirror->pg_list); | |
700 | mirror->pg_bytes_written = 0; | |
701 | mirror->pg_count = 0; | |
702 | mirror->pg_bsize = bsize; | |
703 | mirror->pg_base = 0; | |
704 | mirror->pg_recoalesce = 0; | |
705 | } | |
706 | ||
1da177e4 | 707 | /** |
d8a5ad75 TM |
708 | * nfs_pageio_init - initialise a page io descriptor |
709 | * @desc: pointer to descriptor | |
bcb71bba | 710 | * @inode: pointer to inode |
dfad7000 YW |
711 | * @pg_ops: pointer to pageio operations |
712 | * @compl_ops: pointer to pageio completion operations | |
713 | * @rw_ops: pointer to nfs read/write operations | |
bcb71bba TM |
714 | * @bsize: io block size |
715 | * @io_flags: extra parameters for the io function | |
d8a5ad75 | 716 | */ |
bcb71bba TM |
717 | void nfs_pageio_init(struct nfs_pageio_descriptor *desc, |
718 | struct inode *inode, | |
1751c363 | 719 | const struct nfs_pageio_ops *pg_ops, |
061ae2ed | 720 | const struct nfs_pgio_completion_ops *compl_ops, |
4a0de55c | 721 | const struct nfs_rw_ops *rw_ops, |
84dde76c | 722 | size_t bsize, |
bcb71bba | 723 | int io_flags) |
d8a5ad75 | 724 | { |
a7d42ddb WAA |
725 | struct nfs_pgio_mirror *new; |
726 | int i; | |
727 | ||
b31268ac | 728 | desc->pg_moreio = 0; |
bcb71bba | 729 | desc->pg_inode = inode; |
1751c363 | 730 | desc->pg_ops = pg_ops; |
061ae2ed | 731 | desc->pg_completion_ops = compl_ops; |
4a0de55c | 732 | desc->pg_rw_ops = rw_ops; |
bcb71bba TM |
733 | desc->pg_ioflags = io_flags; |
734 | desc->pg_error = 0; | |
94ad1c80 | 735 | desc->pg_lseg = NULL; |
584aa810 | 736 | desc->pg_dreq = NULL; |
f6166384 | 737 | desc->pg_layout_private = NULL; |
a7d42ddb WAA |
738 | desc->pg_bsize = bsize; |
739 | ||
740 | desc->pg_mirror_count = 1; | |
741 | desc->pg_mirror_idx = 0; | |
742 | ||
743 | if (pg_ops->pg_get_mirror_count) { | |
744 | /* until we have a request, we don't have an lseg and no | |
745 | * idea how many mirrors there will be */ | |
746 | new = kcalloc(NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX, | |
747 | sizeof(struct nfs_pgio_mirror), GFP_KERNEL); | |
748 | desc->pg_mirrors_dynamic = new; | |
749 | desc->pg_mirrors = new; | |
750 | ||
751 | for (i = 0; i < NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX; i++) | |
752 | nfs_pageio_mirror_init(&desc->pg_mirrors[i], bsize); | |
753 | } else { | |
754 | desc->pg_mirrors_dynamic = NULL; | |
755 | desc->pg_mirrors = desc->pg_mirrors_static; | |
756 | nfs_pageio_mirror_init(&desc->pg_mirrors[0], bsize); | |
757 | } | |
d8a5ad75 | 758 | } |
89d77c8f | 759 | EXPORT_SYMBOL_GPL(nfs_pageio_init); |
d8a5ad75 | 760 | |
0eecb214 AS |
761 | /** |
762 | * nfs_pgio_result - Basic pageio error handling | |
763 | * @task: The task that ran | |
d45f60c6 | 764 | * @calldata: Pageio header to check |
0eecb214 | 765 | */ |
6f92fa45 | 766 | static void nfs_pgio_result(struct rpc_task *task, void *calldata) |
0eecb214 | 767 | { |
d45f60c6 WAA |
768 | struct nfs_pgio_header *hdr = calldata; |
769 | struct inode *inode = hdr->inode; | |
0eecb214 AS |
770 | |
771 | dprintk("NFS: %s: %5u, (status %d)\n", __func__, | |
772 | task->tk_pid, task->tk_status); | |
773 | ||
d45f60c6 | 774 | if (hdr->rw_ops->rw_done(task, hdr, inode) != 0) |
0eecb214 AS |
775 | return; |
776 | if (task->tk_status < 0) | |
d45f60c6 | 777 | nfs_set_pgio_error(hdr, task->tk_status, hdr->args.offset); |
0eecb214 | 778 | else |
d45f60c6 | 779 | hdr->rw_ops->rw_result(task, hdr); |
0eecb214 AS |
780 | } |
781 | ||
ef2c488c AS |
782 | /* |
783 | * Create an RPC task for the given read or write request and kick it. | |
784 | * The page must have been locked by the caller. | |
785 | * | |
786 | * It may happen that the page we're passed is not marked dirty. | |
787 | * This is the case if nfs_updatepage detects a conflicting request | |
788 | * that has been written but not committed. | |
789 | */ | |
f0cb9ab8 WAA |
790 | int nfs_generic_pgio(struct nfs_pageio_descriptor *desc, |
791 | struct nfs_pgio_header *hdr) | |
ef2c488c | 792 | { |
48d635f1 | 793 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 794 | |
ef2c488c | 795 | struct nfs_page *req; |
bba5c188 WAA |
796 | struct page **pages, |
797 | *last_page; | |
a7d42ddb | 798 | struct list_head *head = &mirror->pg_list; |
ef2c488c | 799 | struct nfs_commit_info cinfo; |
bba5c188 | 800 | unsigned int pagecount, pageused; |
ef2c488c | 801 | |
a7d42ddb | 802 | pagecount = nfs_page_array_len(mirror->pg_base, mirror->pg_count); |
4714fb51 | 803 | if (!nfs_pgarray_set(&hdr->page_array, pagecount)) |
ef2c488c AS |
804 | return nfs_pgio_error(desc, hdr); |
805 | ||
806 | nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); | |
d45f60c6 | 807 | pages = hdr->page_array.pagevec; |
bba5c188 WAA |
808 | last_page = NULL; |
809 | pageused = 0; | |
ef2c488c AS |
810 | while (!list_empty(head)) { |
811 | req = nfs_list_entry(head->next); | |
812 | nfs_list_remove_request(req); | |
813 | nfs_list_add_request(req, &hdr->pages); | |
bba5c188 | 814 | |
bba5c188 | 815 | if (!last_page || last_page != req->wb_page) { |
bba5c188 | 816 | pageused++; |
b8fb9c30 TM |
817 | if (pageused > pagecount) |
818 | break; | |
819 | *pages++ = last_page = req->wb_page; | |
bba5c188 | 820 | } |
ef2c488c | 821 | } |
bba5c188 WAA |
822 | if (WARN_ON_ONCE(pageused != pagecount)) |
823 | return nfs_pgio_error(desc, hdr); | |
ef2c488c AS |
824 | |
825 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && | |
826 | (desc->pg_moreio || nfs_reqs_to_commit(&cinfo))) | |
827 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
828 | ||
829 | /* Set up the argument struct */ | |
a7d42ddb | 830 | nfs_pgio_rpcsetup(hdr, mirror->pg_count, 0, desc->pg_ioflags, &cinfo); |
ef2c488c AS |
831 | desc->pg_rpc_callops = &nfs_pgio_common_ops; |
832 | return 0; | |
833 | } | |
f0cb9ab8 | 834 | EXPORT_SYMBOL_GPL(nfs_generic_pgio); |
ef2c488c | 835 | |
41d8d5b7 | 836 | static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc) |
cf485fcd | 837 | { |
a7d42ddb | 838 | struct nfs_pgio_mirror *mirror; |
cf485fcd AS |
839 | struct nfs_pgio_header *hdr; |
840 | int ret; | |
841 | ||
48d635f1 | 842 | mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 843 | |
1e7f3a48 WAA |
844 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
845 | if (!hdr) { | |
a7d42ddb WAA |
846 | /* TODO: make sure this is right with mirroring - or |
847 | * should it back out all mirrors? */ | |
848 | desc->pg_completion_ops->error_cleanup(&mirror->pg_list); | |
cf485fcd AS |
849 | return -ENOMEM; |
850 | } | |
1e7f3a48 | 851 | nfs_pgheader_init(desc, hdr, nfs_pgio_header_free); |
cf485fcd AS |
852 | ret = nfs_generic_pgio(desc, hdr); |
853 | if (ret == 0) | |
7f714720 | 854 | ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode), |
46a5ab47 PT |
855 | hdr, |
856 | hdr->cred, | |
857 | NFS_PROTO(hdr->inode), | |
abde71f4 | 858 | desc->pg_rpc_callops, |
7f714720 | 859 | desc->pg_ioflags, 0); |
cf485fcd AS |
860 | return ret; |
861 | } | |
862 | ||
a7d42ddb WAA |
863 | /* |
864 | * nfs_pageio_setup_mirroring - determine if mirroring is to be used | |
865 | * by calling the pg_get_mirror_count op | |
866 | */ | |
867 | static int nfs_pageio_setup_mirroring(struct nfs_pageio_descriptor *pgio, | |
868 | struct nfs_page *req) | |
869 | { | |
870 | int mirror_count = 1; | |
871 | ||
872 | if (!pgio->pg_ops->pg_get_mirror_count) | |
873 | return 0; | |
874 | ||
875 | mirror_count = pgio->pg_ops->pg_get_mirror_count(pgio, req); | |
876 | ||
877 | if (!mirror_count || mirror_count > NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX) | |
878 | return -EINVAL; | |
879 | ||
880 | if (WARN_ON_ONCE(!pgio->pg_mirrors_dynamic)) | |
881 | return -EINVAL; | |
882 | ||
883 | pgio->pg_mirror_count = mirror_count; | |
884 | ||
885 | return 0; | |
886 | } | |
887 | ||
888 | /* | |
889 | * nfs_pageio_stop_mirroring - stop using mirroring (set mirror count to 1) | |
890 | */ | |
891 | void nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor *pgio) | |
892 | { | |
893 | pgio->pg_mirror_count = 1; | |
894 | pgio->pg_mirror_idx = 0; | |
895 | } | |
896 | ||
897 | static void nfs_pageio_cleanup_mirroring(struct nfs_pageio_descriptor *pgio) | |
898 | { | |
899 | pgio->pg_mirror_count = 1; | |
900 | pgio->pg_mirror_idx = 0; | |
901 | pgio->pg_mirrors = pgio->pg_mirrors_static; | |
902 | kfree(pgio->pg_mirrors_dynamic); | |
903 | pgio->pg_mirrors_dynamic = NULL; | |
904 | } | |
905 | ||
4109bb74 TM |
906 | static bool nfs_match_open_context(const struct nfs_open_context *ctx1, |
907 | const struct nfs_open_context *ctx2) | |
908 | { | |
909 | return ctx1->cred == ctx2->cred && ctx1->state == ctx2->state; | |
910 | } | |
911 | ||
912 | static bool nfs_match_lock_context(const struct nfs_lock_context *l1, | |
913 | const struct nfs_lock_context *l2) | |
914 | { | |
915 | return l1->lockowner.l_owner == l2->lockowner.l_owner | |
916 | && l1->lockowner.l_pid == l2->lockowner.l_pid; | |
917 | } | |
918 | ||
d8a5ad75 TM |
919 | /** |
920 | * nfs_can_coalesce_requests - test two requests for compatibility | |
921 | * @prev: pointer to nfs_page | |
922 | * @req: pointer to nfs_page | |
923 | * | |
924 | * The nfs_page structures 'prev' and 'req' are compared to ensure that the | |
925 | * page data area they describe is contiguous, and that their RPC | |
926 | * credentials, NFSv4 open state, and lockowners are the same. | |
927 | * | |
928 | * Return 'true' if this is the case, else return 'false'. | |
929 | */ | |
18ad0a9f BH |
930 | static bool nfs_can_coalesce_requests(struct nfs_page *prev, |
931 | struct nfs_page *req, | |
932 | struct nfs_pageio_descriptor *pgio) | |
d8a5ad75 | 933 | { |
b4fdac1a | 934 | size_t size; |
5263e31e | 935 | struct file_lock_context *flctx; |
b4fdac1a | 936 | |
ab75e417 WAA |
937 | if (prev) { |
938 | if (!nfs_match_open_context(req->wb_context, prev->wb_context)) | |
939 | return false; | |
2b0143b5 | 940 | flctx = d_inode(req->wb_context->dentry)->i_flctx; |
bd61e0a9 JL |
941 | if (flctx != NULL && |
942 | !(list_empty_careful(&flctx->flc_posix) && | |
943 | list_empty_careful(&flctx->flc_flock)) && | |
5263e31e JL |
944 | !nfs_match_lock_context(req->wb_lock_context, |
945 | prev->wb_lock_context)) | |
946 | return false; | |
ab75e417 WAA |
947 | if (req_offset(req) != req_offset(prev) + prev->wb_bytes) |
948 | return false; | |
78270e8f WAA |
949 | if (req->wb_page == prev->wb_page) { |
950 | if (req->wb_pgbase != prev->wb_pgbase + prev->wb_bytes) | |
951 | return false; | |
952 | } else { | |
953 | if (req->wb_pgbase != 0 || | |
954 | prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE) | |
955 | return false; | |
956 | } | |
ab75e417 | 957 | } |
b4fdac1a | 958 | size = pgio->pg_ops->pg_test(pgio, prev, req); |
f0cb9ab8 WAA |
959 | WARN_ON_ONCE(size > req->wb_bytes); |
960 | if (size && size < req->wb_bytes) | |
961 | req->wb_bytes = size; | |
b4fdac1a | 962 | return size > 0; |
d8a5ad75 TM |
963 | } |
964 | ||
965 | /** | |
bcb71bba | 966 | * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list. |
d8a5ad75 TM |
967 | * @desc: destination io descriptor |
968 | * @req: request | |
969 | * | |
970 | * Returns true if the request 'req' was successfully coalesced into the | |
971 | * existing list of pages 'desc'. | |
972 | */ | |
bcb71bba TM |
973 | static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc, |
974 | struct nfs_page *req) | |
d8a5ad75 | 975 | { |
48d635f1 | 976 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 977 | |
ab75e417 | 978 | struct nfs_page *prev = NULL; |
a7d42ddb WAA |
979 | |
980 | if (mirror->pg_count != 0) { | |
981 | prev = nfs_list_entry(mirror->pg_list.prev); | |
5b36c7dc | 982 | } else { |
d8007d4d TM |
983 | if (desc->pg_ops->pg_init) |
984 | desc->pg_ops->pg_init(desc, req); | |
a7d42ddb | 985 | mirror->pg_base = req->wb_pgbase; |
5b36c7dc | 986 | } |
ab75e417 WAA |
987 | if (!nfs_can_coalesce_requests(prev, req, desc)) |
988 | return 0; | |
d8a5ad75 | 989 | nfs_list_remove_request(req); |
a7d42ddb WAA |
990 | nfs_list_add_request(req, &mirror->pg_list); |
991 | mirror->pg_count += req->wb_bytes; | |
d8a5ad75 TM |
992 | return 1; |
993 | } | |
994 | ||
bcb71bba TM |
995 | /* |
996 | * Helper for nfs_pageio_add_request and nfs_pageio_complete | |
997 | */ | |
998 | static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc) | |
999 | { | |
48d635f1 | 1000 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb WAA |
1001 | |
1002 | ||
1003 | if (!list_empty(&mirror->pg_list)) { | |
1751c363 | 1004 | int error = desc->pg_ops->pg_doio(desc); |
bcb71bba TM |
1005 | if (error < 0) |
1006 | desc->pg_error = error; | |
1007 | else | |
a7d42ddb | 1008 | mirror->pg_bytes_written += mirror->pg_count; |
bcb71bba | 1009 | } |
a7d42ddb WAA |
1010 | if (list_empty(&mirror->pg_list)) { |
1011 | mirror->pg_count = 0; | |
1012 | mirror->pg_base = 0; | |
bcb71bba TM |
1013 | } |
1014 | } | |
1015 | ||
1016 | /** | |
1017 | * nfs_pageio_add_request - Attempt to coalesce a request into a page list. | |
1018 | * @desc: destination io descriptor | |
1019 | * @req: request | |
1020 | * | |
2bfc6e56 WAA |
1021 | * This may split a request into subrequests which are all part of the |
1022 | * same page group. | |
1023 | * | |
bcb71bba TM |
1024 | * Returns true if the request 'req' was successfully coalesced into the |
1025 | * existing list of pages 'desc'. | |
1026 | */ | |
d9156f9f | 1027 | static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc, |
8b09bee3 | 1028 | struct nfs_page *req) |
bcb71bba | 1029 | { |
48d635f1 | 1030 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 1031 | |
2bfc6e56 WAA |
1032 | struct nfs_page *subreq; |
1033 | unsigned int bytes_left = 0; | |
1034 | unsigned int offset, pgbase; | |
1035 | ||
bfd484a5 | 1036 | nfs_page_group_lock(req, false); |
2bfc6e56 WAA |
1037 | |
1038 | subreq = req; | |
1039 | bytes_left = subreq->wb_bytes; | |
1040 | offset = subreq->wb_offset; | |
1041 | pgbase = subreq->wb_pgbase; | |
1042 | ||
1043 | do { | |
1044 | if (!nfs_pageio_do_add_request(desc, subreq)) { | |
1045 | /* make sure pg_test call(s) did nothing */ | |
1046 | WARN_ON_ONCE(subreq->wb_bytes != bytes_left); | |
1047 | WARN_ON_ONCE(subreq->wb_offset != offset); | |
1048 | WARN_ON_ONCE(subreq->wb_pgbase != pgbase); | |
1049 | ||
1050 | nfs_page_group_unlock(req); | |
1051 | desc->pg_moreio = 1; | |
1052 | nfs_pageio_doio(desc); | |
1053 | if (desc->pg_error < 0) | |
1054 | return 0; | |
a7d42ddb | 1055 | if (mirror->pg_recoalesce) |
2bfc6e56 WAA |
1056 | return 0; |
1057 | /* retry add_request for this subreq */ | |
bfd484a5 | 1058 | nfs_page_group_lock(req, false); |
2bfc6e56 WAA |
1059 | continue; |
1060 | } | |
1061 | ||
1062 | /* check for buggy pg_test call(s) */ | |
1063 | WARN_ON_ONCE(subreq->wb_bytes + subreq->wb_pgbase > PAGE_SIZE); | |
1064 | WARN_ON_ONCE(subreq->wb_bytes > bytes_left); | |
1065 | WARN_ON_ONCE(subreq->wb_bytes == 0); | |
1066 | ||
1067 | bytes_left -= subreq->wb_bytes; | |
1068 | offset += subreq->wb_bytes; | |
1069 | pgbase += subreq->wb_bytes; | |
1070 | ||
1071 | if (bytes_left) { | |
1072 | subreq = nfs_create_request(req->wb_context, | |
1073 | req->wb_page, | |
1074 | subreq, pgbase, bytes_left); | |
c1109558 TM |
1075 | if (IS_ERR(subreq)) |
1076 | goto err_ptr; | |
2bfc6e56 WAA |
1077 | nfs_lock_request(subreq); |
1078 | subreq->wb_offset = offset; | |
1079 | subreq->wb_index = req->wb_index; | |
1080 | } | |
1081 | } while (bytes_left > 0); | |
1082 | ||
1083 | nfs_page_group_unlock(req); | |
bcb71bba | 1084 | return 1; |
c1109558 TM |
1085 | err_ptr: |
1086 | desc->pg_error = PTR_ERR(subreq); | |
1087 | nfs_page_group_unlock(req); | |
1088 | return 0; | |
bcb71bba TM |
1089 | } |
1090 | ||
d9156f9f TM |
1091 | static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc) |
1092 | { | |
48d635f1 | 1093 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
d9156f9f TM |
1094 | LIST_HEAD(head); |
1095 | ||
1096 | do { | |
a7d42ddb WAA |
1097 | list_splice_init(&mirror->pg_list, &head); |
1098 | mirror->pg_bytes_written -= mirror->pg_count; | |
1099 | mirror->pg_count = 0; | |
1100 | mirror->pg_base = 0; | |
1101 | mirror->pg_recoalesce = 0; | |
1102 | ||
d9156f9f TM |
1103 | while (!list_empty(&head)) { |
1104 | struct nfs_page *req; | |
1105 | ||
1106 | req = list_first_entry(&head, struct nfs_page, wb_list); | |
1107 | nfs_list_remove_request(req); | |
1108 | if (__nfs_pageio_add_request(desc, req)) | |
1109 | continue; | |
03d5eb65 TM |
1110 | if (desc->pg_error < 0) { |
1111 | list_splice_tail(&head, &mirror->pg_list); | |
1112 | mirror->pg_recoalesce = 1; | |
d9156f9f | 1113 | return 0; |
03d5eb65 | 1114 | } |
d9156f9f TM |
1115 | break; |
1116 | } | |
a7d42ddb | 1117 | } while (mirror->pg_recoalesce); |
d9156f9f TM |
1118 | return 1; |
1119 | } | |
1120 | ||
a7d42ddb | 1121 | static int nfs_pageio_add_request_mirror(struct nfs_pageio_descriptor *desc, |
d9156f9f TM |
1122 | struct nfs_page *req) |
1123 | { | |
1124 | int ret; | |
1125 | ||
1126 | do { | |
1127 | ret = __nfs_pageio_add_request(desc, req); | |
1128 | if (ret) | |
1129 | break; | |
1130 | if (desc->pg_error < 0) | |
1131 | break; | |
1132 | ret = nfs_do_recoalesce(desc); | |
1133 | } while (ret); | |
a7d42ddb | 1134 | |
d9156f9f TM |
1135 | return ret; |
1136 | } | |
1137 | ||
a7d42ddb WAA |
1138 | int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc, |
1139 | struct nfs_page *req) | |
1140 | { | |
1141 | u32 midx; | |
1142 | unsigned int pgbase, offset, bytes; | |
1143 | struct nfs_page *dupreq, *lastreq; | |
1144 | ||
1145 | pgbase = req->wb_pgbase; | |
1146 | offset = req->wb_offset; | |
1147 | bytes = req->wb_bytes; | |
1148 | ||
1149 | nfs_pageio_setup_mirroring(desc, req); | |
1150 | ||
1151 | for (midx = 0; midx < desc->pg_mirror_count; midx++) { | |
1152 | if (midx) { | |
1153 | nfs_page_group_lock(req, false); | |
1154 | ||
1155 | /* find the last request */ | |
1156 | for (lastreq = req->wb_head; | |
1157 | lastreq->wb_this_page != req->wb_head; | |
1158 | lastreq = lastreq->wb_this_page) | |
1159 | ; | |
1160 | ||
1161 | dupreq = nfs_create_request(req->wb_context, | |
1162 | req->wb_page, lastreq, pgbase, bytes); | |
1163 | ||
1164 | if (IS_ERR(dupreq)) { | |
1165 | nfs_page_group_unlock(req); | |
1166 | return 0; | |
1167 | } | |
1168 | ||
1169 | nfs_lock_request(dupreq); | |
1170 | nfs_page_group_unlock(req); | |
1171 | dupreq->wb_offset = offset; | |
1172 | dupreq->wb_index = req->wb_index; | |
1173 | } else | |
1174 | dupreq = req; | |
1175 | ||
47af81f2 PT |
1176 | if (nfs_pgio_has_mirroring(desc)) |
1177 | desc->pg_mirror_idx = midx; | |
a7d42ddb WAA |
1178 | if (!nfs_pageio_add_request_mirror(desc, dupreq)) |
1179 | return 0; | |
1180 | } | |
1181 | ||
1182 | return 1; | |
1183 | } | |
1184 | ||
1185 | /* | |
1186 | * nfs_pageio_complete_mirror - Complete I/O on the current mirror of an | |
1187 | * nfs_pageio_descriptor | |
1188 | * @desc: pointer to io descriptor | |
dfad7000 | 1189 | * @mirror_idx: pointer to mirror index |
a7d42ddb WAA |
1190 | */ |
1191 | static void nfs_pageio_complete_mirror(struct nfs_pageio_descriptor *desc, | |
1192 | u32 mirror_idx) | |
1193 | { | |
1194 | struct nfs_pgio_mirror *mirror = &desc->pg_mirrors[mirror_idx]; | |
1195 | u32 restore_idx = desc->pg_mirror_idx; | |
1196 | ||
47af81f2 PT |
1197 | if (nfs_pgio_has_mirroring(desc)) |
1198 | desc->pg_mirror_idx = mirror_idx; | |
a7d42ddb WAA |
1199 | for (;;) { |
1200 | nfs_pageio_doio(desc); | |
1201 | if (!mirror->pg_recoalesce) | |
1202 | break; | |
1203 | if (!nfs_do_recoalesce(desc)) | |
1204 | break; | |
1205 | } | |
1206 | desc->pg_mirror_idx = restore_idx; | |
1207 | } | |
1208 | ||
53113ad3 WAA |
1209 | /* |
1210 | * nfs_pageio_resend - Transfer requests to new descriptor and resend | |
1211 | * @hdr - the pgio header to move request from | |
1212 | * @desc - the pageio descriptor to add requests to | |
1213 | * | |
1214 | * Try to move each request (nfs_page) from @hdr to @desc then attempt | |
1215 | * to send them. | |
1216 | * | |
1217 | * Returns 0 on success and < 0 on error. | |
1218 | */ | |
1219 | int nfs_pageio_resend(struct nfs_pageio_descriptor *desc, | |
1220 | struct nfs_pgio_header *hdr) | |
1221 | { | |
1222 | LIST_HEAD(failed); | |
1223 | ||
1224 | desc->pg_dreq = hdr->dreq; | |
1225 | while (!list_empty(&hdr->pages)) { | |
1226 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
1227 | ||
1228 | nfs_list_remove_request(req); | |
1229 | if (!nfs_pageio_add_request(desc, req)) | |
1230 | nfs_list_add_request(req, &failed); | |
1231 | } | |
1232 | nfs_pageio_complete(desc); | |
1233 | if (!list_empty(&failed)) { | |
1234 | list_move(&failed, &hdr->pages); | |
1235 | return -EIO; | |
1236 | } | |
1237 | return 0; | |
1238 | } | |
1239 | EXPORT_SYMBOL_GPL(nfs_pageio_resend); | |
d9156f9f | 1240 | |
bcb71bba | 1241 | /** |
2176bf42 | 1242 | * nfs_pageio_complete - Complete I/O then cleanup an nfs_pageio_descriptor |
bcb71bba TM |
1243 | * @desc: pointer to io descriptor |
1244 | */ | |
1245 | void nfs_pageio_complete(struct nfs_pageio_descriptor *desc) | |
1246 | { | |
a7d42ddb WAA |
1247 | u32 midx; |
1248 | ||
1249 | for (midx = 0; midx < desc->pg_mirror_count; midx++) | |
1250 | nfs_pageio_complete_mirror(desc, midx); | |
2176bf42 WAA |
1251 | |
1252 | if (desc->pg_ops->pg_cleanup) | |
1253 | desc->pg_ops->pg_cleanup(desc); | |
a7d42ddb | 1254 | nfs_pageio_cleanup_mirroring(desc); |
bcb71bba TM |
1255 | } |
1256 | ||
7fe7f848 TM |
1257 | /** |
1258 | * nfs_pageio_cond_complete - Conditional I/O completion | |
1259 | * @desc: pointer to io descriptor | |
1260 | * @index: page index | |
1261 | * | |
1262 | * It is important to ensure that processes don't try to take locks | |
1263 | * on non-contiguous ranges of pages as that might deadlock. This | |
1264 | * function should be called before attempting to wait on a locked | |
1265 | * nfs_page. It will complete the I/O if the page index 'index' | |
1266 | * is not contiguous with the existing list of pages in 'desc'. | |
1267 | */ | |
1268 | void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index) | |
1269 | { | |
a7d42ddb WAA |
1270 | struct nfs_pgio_mirror *mirror; |
1271 | struct nfs_page *prev; | |
1272 | u32 midx; | |
1273 | ||
1274 | for (midx = 0; midx < desc->pg_mirror_count; midx++) { | |
1275 | mirror = &desc->pg_mirrors[midx]; | |
1276 | if (!list_empty(&mirror->pg_list)) { | |
1277 | prev = nfs_list_entry(mirror->pg_list.prev); | |
1278 | if (index != prev->wb_index + 1) | |
1279 | nfs_pageio_complete_mirror(desc, midx); | |
1280 | } | |
7fe7f848 TM |
1281 | } |
1282 | } | |
1283 | ||
f7b422b1 | 1284 | int __init nfs_init_nfspagecache(void) |
1da177e4 LT |
1285 | { |
1286 | nfs_page_cachep = kmem_cache_create("nfs_page", | |
1287 | sizeof(struct nfs_page), | |
1288 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1289 | NULL); |
1da177e4 LT |
1290 | if (nfs_page_cachep == NULL) |
1291 | return -ENOMEM; | |
1292 | ||
1293 | return 0; | |
1294 | } | |
1295 | ||
266bee88 | 1296 | void nfs_destroy_nfspagecache(void) |
1da177e4 | 1297 | { |
1a1d92c1 | 1298 | kmem_cache_destroy(nfs_page_cachep); |
1da177e4 LT |
1299 | } |
1300 | ||
ef2c488c | 1301 | static const struct rpc_call_ops nfs_pgio_common_ops = { |
6f92fa45 AS |
1302 | .rpc_call_prepare = nfs_pgio_prepare, |
1303 | .rpc_call_done = nfs_pgio_result, | |
1304 | .rpc_release = nfs_pgio_release, | |
1305 | }; | |
41d8d5b7 AS |
1306 | |
1307 | const struct nfs_pageio_ops nfs_pgio_rw_ops = { | |
1308 | .pg_test = nfs_generic_pg_test, | |
1309 | .pg_doio = nfs_generic_pg_pgios, | |
1310 | }; |