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 | |
48d635f1 PT |
32 | struct nfs_pgio_mirror * |
33 | nfs_pgio_current_mirror(struct nfs_pageio_descriptor *desc) | |
34 | { | |
35 | return nfs_pgio_has_mirroring(desc) ? | |
36 | &desc->pg_mirrors[desc->pg_mirror_idx] : | |
37 | &desc->pg_mirrors[0]; | |
38 | } | |
39 | EXPORT_SYMBOL_GPL(nfs_pgio_current_mirror); | |
40 | ||
4db6e0b7 FI |
41 | void nfs_pgheader_init(struct nfs_pageio_descriptor *desc, |
42 | struct nfs_pgio_header *hdr, | |
43 | void (*release)(struct nfs_pgio_header *hdr)) | |
44 | { | |
48d635f1 | 45 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb WAA |
46 | |
47 | ||
48 | hdr->req = nfs_list_entry(mirror->pg_list.next); | |
4db6e0b7 FI |
49 | hdr->inode = desc->pg_inode; |
50 | hdr->cred = hdr->req->wb_context->cred; | |
51 | hdr->io_start = req_offset(hdr->req); | |
a7d42ddb | 52 | hdr->good_bytes = mirror->pg_count; |
919e3bd9 | 53 | hdr->io_completion = desc->pg_io_completion; |
584aa810 | 54 | hdr->dreq = desc->pg_dreq; |
4db6e0b7 | 55 | hdr->release = release; |
061ae2ed | 56 | hdr->completion_ops = desc->pg_completion_ops; |
584aa810 FI |
57 | if (hdr->completion_ops->init_hdr) |
58 | hdr->completion_ops->init_hdr(hdr); | |
a7d42ddb WAA |
59 | |
60 | hdr->pgio_mirror_idx = desc->pg_mirror_idx; | |
4db6e0b7 | 61 | } |
89d77c8f | 62 | EXPORT_SYMBOL_GPL(nfs_pgheader_init); |
4db6e0b7 FI |
63 | |
64 | void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos) | |
65 | { | |
66 | spin_lock(&hdr->lock); | |
e9ae58ae TM |
67 | if (!test_and_set_bit(NFS_IOHDR_ERROR, &hdr->flags) |
68 | || pos < hdr->io_start + hdr->good_bytes) { | |
4db6e0b7 FI |
69 | clear_bit(NFS_IOHDR_EOF, &hdr->flags); |
70 | hdr->good_bytes = pos - hdr->io_start; | |
71 | hdr->error = error; | |
72 | } | |
73 | spin_unlock(&hdr->lock); | |
74 | } | |
75 | ||
1da177e4 LT |
76 | static inline struct nfs_page * |
77 | nfs_page_alloc(void) | |
78 | { | |
192e501b | 79 | struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_NOIO); |
72895b1a | 80 | if (p) |
1da177e4 | 81 | INIT_LIST_HEAD(&p->wb_list); |
1da177e4 LT |
82 | return p; |
83 | } | |
84 | ||
85 | static inline void | |
86 | nfs_page_free(struct nfs_page *p) | |
87 | { | |
88 | kmem_cache_free(nfs_page_cachep, p); | |
89 | } | |
90 | ||
577b4232 TM |
91 | /** |
92 | * nfs_iocounter_wait - wait for i/o to complete | |
210c7c17 | 93 | * @l_ctx: nfs_lock_context with io_counter to use |
577b4232 TM |
94 | * |
95 | * returns -ERESTARTSYS if interrupted by a fatal signal. | |
96 | * Otherwise returns 0 once the io_count hits 0. | |
97 | */ | |
98 | int | |
210c7c17 | 99 | nfs_iocounter_wait(struct nfs_lock_context *l_ctx) |
577b4232 | 100 | { |
723c921e PZ |
101 | return wait_var_event_killable(&l_ctx->io_count, |
102 | !atomic_read(&l_ctx->io_count)); | |
577b4232 TM |
103 | } |
104 | ||
7d6ddf88 BC |
105 | /** |
106 | * nfs_async_iocounter_wait - wait on a rpc_waitqueue for I/O | |
107 | * to complete | |
108 | * @task: the rpc_task that should wait | |
109 | * @l_ctx: nfs_lock_context with io_counter to check | |
110 | * | |
111 | * Returns true if there is outstanding I/O to wait on and the | |
112 | * task has been put to sleep. | |
113 | */ | |
114 | bool | |
115 | nfs_async_iocounter_wait(struct rpc_task *task, struct nfs_lock_context *l_ctx) | |
116 | { | |
117 | struct inode *inode = d_inode(l_ctx->open_context->dentry); | |
118 | bool ret = false; | |
119 | ||
120 | if (atomic_read(&l_ctx->io_count) > 0) { | |
121 | rpc_sleep_on(&NFS_SERVER(inode)->uoc_rpcwaitq, task, NULL); | |
122 | ret = true; | |
123 | } | |
124 | ||
125 | if (atomic_read(&l_ctx->io_count) == 0) { | |
126 | rpc_wake_up_queued_task(&NFS_SERVER(inode)->uoc_rpcwaitq, task); | |
127 | ret = false; | |
128 | } | |
129 | ||
130 | return ret; | |
131 | } | |
132 | EXPORT_SYMBOL_GPL(nfs_async_iocounter_wait); | |
133 | ||
2bfc6e56 WAA |
134 | /* |
135 | * nfs_page_group_lock - lock the head of the page group | |
136 | * @req - request in group that is to be locked | |
137 | * | |
1344b7ea TM |
138 | * this lock must be held when traversing or modifying the page |
139 | * group list | |
e7029206 | 140 | * |
1344b7ea | 141 | * return 0 on success, < 0 on error |
2bfc6e56 | 142 | */ |
e7029206 | 143 | int |
1344b7ea | 144 | nfs_page_group_lock(struct nfs_page *req) |
2bfc6e56 WAA |
145 | { |
146 | struct nfs_page *head = req->wb_head; | |
2bfc6e56 WAA |
147 | |
148 | WARN_ON_ONCE(head != head->wb_head); | |
149 | ||
bc8a309e WAA |
150 | if (!test_and_set_bit(PG_HEADLOCK, &head->wb_flags)) |
151 | return 0; | |
e7029206 | 152 | |
1344b7ea TM |
153 | set_bit(PG_CONTENDED1, &head->wb_flags); |
154 | smp_mb__after_atomic(); | |
155 | return wait_on_bit_lock(&head->wb_flags, PG_HEADLOCK, | |
bc8a309e | 156 | TASK_UNINTERRUPTIBLE); |
2bfc6e56 WAA |
157 | } |
158 | ||
159 | /* | |
160 | * nfs_page_group_unlock - unlock the head of the page group | |
161 | * @req - request in group that is to be unlocked | |
162 | */ | |
163 | void | |
164 | nfs_page_group_unlock(struct nfs_page *req) | |
165 | { | |
166 | struct nfs_page *head = req->wb_head; | |
167 | ||
168 | WARN_ON_ONCE(head != head->wb_head); | |
169 | ||
d1e1cda8 | 170 | smp_mb__before_atomic(); |
2bfc6e56 | 171 | clear_bit(PG_HEADLOCK, &head->wb_flags); |
d1e1cda8 | 172 | smp_mb__after_atomic(); |
b4f937cf TM |
173 | if (!test_bit(PG_CONTENDED1, &head->wb_flags)) |
174 | return; | |
2bfc6e56 WAA |
175 | wake_up_bit(&head->wb_flags, PG_HEADLOCK); |
176 | } | |
177 | ||
178 | /* | |
179 | * nfs_page_group_sync_on_bit_locked | |
180 | * | |
181 | * must be called with page group lock held | |
182 | */ | |
183 | static bool | |
184 | nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit) | |
185 | { | |
186 | struct nfs_page *head = req->wb_head; | |
187 | struct nfs_page *tmp; | |
188 | ||
189 | WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags)); | |
190 | WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags)); | |
191 | ||
192 | tmp = req->wb_this_page; | |
193 | while (tmp != req) { | |
194 | if (!test_bit(bit, &tmp->wb_flags)) | |
195 | return false; | |
196 | tmp = tmp->wb_this_page; | |
197 | } | |
198 | ||
199 | /* true! reset all bits */ | |
200 | tmp = req; | |
201 | do { | |
202 | clear_bit(bit, &tmp->wb_flags); | |
203 | tmp = tmp->wb_this_page; | |
204 | } while (tmp != req); | |
205 | ||
206 | return true; | |
207 | } | |
208 | ||
209 | /* | |
210 | * nfs_page_group_sync_on_bit - set bit on current request, but only | |
211 | * return true if the bit is set for all requests in page group | |
212 | * @req - request in page group | |
213 | * @bit - PG_* bit that is used to sync page group | |
214 | */ | |
215 | bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit) | |
216 | { | |
217 | bool ret; | |
218 | ||
1344b7ea | 219 | nfs_page_group_lock(req); |
2bfc6e56 WAA |
220 | ret = nfs_page_group_sync_on_bit_locked(req, bit); |
221 | nfs_page_group_unlock(req); | |
222 | ||
223 | return ret; | |
224 | } | |
225 | ||
226 | /* | |
227 | * nfs_page_group_init - Initialize the page group linkage for @req | |
228 | * @req - a new nfs request | |
229 | * @prev - the previous request in page group, or NULL if @req is the first | |
230 | * or only request in the group (the head). | |
231 | */ | |
232 | static inline void | |
233 | nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev) | |
234 | { | |
cb1410c7 | 235 | struct inode *inode; |
2bfc6e56 WAA |
236 | WARN_ON_ONCE(prev == req); |
237 | ||
238 | if (!prev) { | |
85710a83 | 239 | /* a head request */ |
2bfc6e56 WAA |
240 | req->wb_head = req; |
241 | req->wb_this_page = req; | |
242 | } else { | |
85710a83 | 243 | /* a subrequest */ |
2bfc6e56 WAA |
244 | WARN_ON_ONCE(prev->wb_this_page != prev->wb_head); |
245 | WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags)); | |
246 | req->wb_head = prev->wb_head; | |
247 | req->wb_this_page = prev->wb_this_page; | |
248 | prev->wb_this_page = req; | |
249 | ||
85710a83 WAA |
250 | /* All subrequests take a ref on the head request until |
251 | * nfs_page_group_destroy is called */ | |
252 | kref_get(&req->wb_head->wb_kref); | |
253 | ||
cb1410c7 WAA |
254 | /* grab extra ref and bump the request count if head request |
255 | * has extra ref from the write/commit path to handle handoff | |
256 | * between write and commit lists. */ | |
17089a29 | 257 | if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags)) { |
cb1410c7 | 258 | inode = page_file_mapping(req->wb_page)->host; |
17089a29 | 259 | set_bit(PG_INODE_REF, &req->wb_flags); |
2bfc6e56 | 260 | kref_get(&req->wb_kref); |
a6b6d5b8 | 261 | atomic_long_inc(&NFS_I(inode)->nrequests); |
17089a29 | 262 | } |
2bfc6e56 WAA |
263 | } |
264 | } | |
265 | ||
266 | /* | |
267 | * nfs_page_group_destroy - sync the destruction of page groups | |
268 | * @req - request that no longer needs the page group | |
269 | * | |
270 | * releases the page group reference from each member once all | |
271 | * members have called this function. | |
272 | */ | |
273 | static void | |
274 | nfs_page_group_destroy(struct kref *kref) | |
275 | { | |
276 | struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref); | |
08fead2a | 277 | struct nfs_page *head = req->wb_head; |
2bfc6e56 WAA |
278 | struct nfs_page *tmp, *next; |
279 | ||
280 | if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN)) | |
08fead2a | 281 | goto out; |
2bfc6e56 WAA |
282 | |
283 | tmp = req; | |
284 | do { | |
285 | next = tmp->wb_this_page; | |
286 | /* unlink and free */ | |
287 | tmp->wb_this_page = tmp; | |
288 | tmp->wb_head = tmp; | |
289 | nfs_free_request(tmp); | |
290 | tmp = next; | |
291 | } while (tmp != req); | |
08fead2a TM |
292 | out: |
293 | /* subrequests must release the ref on the head request */ | |
294 | if (head != req) | |
295 | nfs_release_request(head); | |
2bfc6e56 WAA |
296 | } |
297 | ||
1da177e4 LT |
298 | /** |
299 | * nfs_create_request - Create an NFS read/write request. | |
c02f557d | 300 | * @ctx: open context to use |
1da177e4 | 301 | * @page: page to write |
2bfc6e56 | 302 | * @last: last nfs request created for this page group or NULL if head |
1da177e4 LT |
303 | * @offset: starting offset within the page for the write |
304 | * @count: number of bytes to read/write | |
305 | * | |
306 | * The page must be locked by the caller. This makes sure we never | |
a19b89ca | 307 | * create two different requests for the same page. |
1da177e4 LT |
308 | * User should ensure it is safe to sleep in this function. |
309 | */ | |
310 | struct nfs_page * | |
8c8f1ac1 | 311 | nfs_create_request(struct nfs_open_context *ctx, struct page *page, |
2bfc6e56 WAA |
312 | struct nfs_page *last, unsigned int offset, |
313 | unsigned int count) | |
1da177e4 | 314 | { |
1da177e4 | 315 | struct nfs_page *req; |
b3c54de6 | 316 | struct nfs_lock_context *l_ctx; |
1da177e4 | 317 | |
c58c8441 TM |
318 | if (test_bit(NFS_CONTEXT_BAD, &ctx->flags)) |
319 | return ERR_PTR(-EBADF); | |
18eb8842 TM |
320 | /* try to allocate the request struct */ |
321 | req = nfs_page_alloc(); | |
322 | if (req == NULL) | |
323 | return ERR_PTR(-ENOMEM); | |
1da177e4 | 324 | |
015f0212 | 325 | /* get lock context early so we can deal with alloc failures */ |
b3c54de6 TM |
326 | l_ctx = nfs_get_lock_context(ctx); |
327 | if (IS_ERR(l_ctx)) { | |
015f0212 | 328 | nfs_page_free(req); |
b3c54de6 | 329 | return ERR_CAST(l_ctx); |
015f0212 | 330 | } |
b3c54de6 | 331 | req->wb_lock_context = l_ctx; |
210c7c17 | 332 | atomic_inc(&l_ctx->io_count); |
015f0212 | 333 | |
1da177e4 LT |
334 | /* Initialize the request struct. Initially, we assume a |
335 | * long write-back delay. This will be adjusted in | |
336 | * update_nfs_request below if the region is not locked. */ | |
337 | req->wb_page = page; | |
67911c8f | 338 | if (page) { |
8cd79788 | 339 | req->wb_index = page_index(page); |
67911c8f AS |
340 | get_page(page); |
341 | } | |
1da177e4 LT |
342 | req->wb_offset = offset; |
343 | req->wb_pgbase = offset; | |
344 | req->wb_bytes = count; | |
1da177e4 | 345 | req->wb_context = get_nfs_open_context(ctx); |
c03b4024 | 346 | kref_init(&req->wb_kref); |
2bfc6e56 | 347 | nfs_page_group_init(req, last); |
1da177e4 LT |
348 | return req; |
349 | } | |
350 | ||
351 | /** | |
1d1afcbc | 352 | * nfs_unlock_request - Unlock request and wake up sleepers. |
1da177e4 LT |
353 | * @req: |
354 | */ | |
1d1afcbc | 355 | void nfs_unlock_request(struct nfs_page *req) |
1da177e4 LT |
356 | { |
357 | if (!NFS_WBACK_BUSY(req)) { | |
358 | printk(KERN_ERR "NFS: Invalid unlock attempted\n"); | |
359 | BUG(); | |
360 | } | |
4e857c58 | 361 | smp_mb__before_atomic(); |
1da177e4 | 362 | clear_bit(PG_BUSY, &req->wb_flags); |
4e857c58 | 363 | smp_mb__after_atomic(); |
b4f937cf TM |
364 | if (!test_bit(PG_CONTENDED2, &req->wb_flags)) |
365 | return; | |
464a98bd | 366 | wake_up_bit(&req->wb_flags, PG_BUSY); |
3aff4ebb TM |
367 | } |
368 | ||
369 | /** | |
1d1afcbc TM |
370 | * nfs_unlock_and_release_request - Unlock request and release the nfs_page |
371 | * @req: | |
3aff4ebb | 372 | */ |
1d1afcbc | 373 | void nfs_unlock_and_release_request(struct nfs_page *req) |
3aff4ebb | 374 | { |
1d1afcbc | 375 | nfs_unlock_request(req); |
1da177e4 LT |
376 | nfs_release_request(req); |
377 | } | |
378 | ||
4d65c520 | 379 | /* |
1da177e4 LT |
380 | * nfs_clear_request - Free up all resources allocated to the request |
381 | * @req: | |
382 | * | |
bb6fbc45 TM |
383 | * Release page and open context resources associated with a read/write |
384 | * request after it has completed. | |
1da177e4 | 385 | */ |
4d65c520 | 386 | static void nfs_clear_request(struct nfs_page *req) |
1da177e4 | 387 | { |
cd52ed35 | 388 | struct page *page = req->wb_page; |
bb6fbc45 | 389 | struct nfs_open_context *ctx = req->wb_context; |
f11ac8db | 390 | struct nfs_lock_context *l_ctx = req->wb_lock_context; |
bb6fbc45 | 391 | |
cd52ed35 | 392 | if (page != NULL) { |
09cbfeaf | 393 | put_page(page); |
1da177e4 LT |
394 | req->wb_page = NULL; |
395 | } | |
f11ac8db | 396 | if (l_ctx != NULL) { |
7d6ddf88 | 397 | if (atomic_dec_and_test(&l_ctx->io_count)) { |
723c921e | 398 | wake_up_var(&l_ctx->io_count); |
7d6ddf88 BC |
399 | if (test_bit(NFS_CONTEXT_UNLOCK, &ctx->flags)) |
400 | rpc_wake_up(&NFS_SERVER(d_inode(ctx->dentry))->uoc_rpcwaitq); | |
401 | } | |
f11ac8db TM |
402 | nfs_put_lock_context(l_ctx); |
403 | req->wb_lock_context = NULL; | |
404 | } | |
bb6fbc45 TM |
405 | if (ctx != NULL) { |
406 | put_nfs_open_context(ctx); | |
407 | req->wb_context = NULL; | |
408 | } | |
1da177e4 LT |
409 | } |
410 | ||
1da177e4 LT |
411 | /** |
412 | * nfs_release_request - Release the count on an NFS read/write request | |
413 | * @req: request to release | |
414 | * | |
415 | * Note: Should never be called with the spinlock held! | |
416 | */ | |
d4581383 | 417 | void nfs_free_request(struct nfs_page *req) |
1da177e4 | 418 | { |
2bfc6e56 WAA |
419 | WARN_ON_ONCE(req->wb_this_page != req); |
420 | ||
421 | /* extra debug: make sure no sync bits are still set */ | |
422 | WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags)); | |
67d0338e WAA |
423 | WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags)); |
424 | WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags)); | |
20633f04 WAA |
425 | WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags)); |
426 | WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags)); | |
1da177e4 | 427 | |
bb6fbc45 | 428 | /* Release struct file and open context */ |
1da177e4 | 429 | nfs_clear_request(req); |
1da177e4 LT |
430 | nfs_page_free(req); |
431 | } | |
432 | ||
c03b4024 TM |
433 | void nfs_release_request(struct nfs_page *req) |
434 | { | |
2bfc6e56 | 435 | kref_put(&req->wb_kref, nfs_page_group_destroy); |
9f557cd8 | 436 | } |
2ce209c4 | 437 | EXPORT_SYMBOL_GPL(nfs_release_request); |
9f557cd8 | 438 | |
1da177e4 LT |
439 | /** |
440 | * nfs_wait_on_request - Wait for a request to complete. | |
441 | * @req: request to wait upon. | |
442 | * | |
150030b7 | 443 | * Interruptible by fatal signals only. |
1da177e4 LT |
444 | * The user is responsible for holding a count on the request. |
445 | */ | |
446 | int | |
447 | nfs_wait_on_request(struct nfs_page *req) | |
448 | { | |
b4f937cf TM |
449 | if (!test_bit(PG_BUSY, &req->wb_flags)) |
450 | return 0; | |
451 | set_bit(PG_CONTENDED2, &req->wb_flags); | |
452 | smp_mb__after_atomic(); | |
74316201 N |
453 | return wait_on_bit_io(&req->wb_flags, PG_BUSY, |
454 | TASK_UNINTERRUPTIBLE); | |
1da177e4 | 455 | } |
2ce209c4 | 456 | EXPORT_SYMBOL_GPL(nfs_wait_on_request); |
1da177e4 | 457 | |
b4fdac1a WAA |
458 | /* |
459 | * nfs_generic_pg_test - determine if requests can be coalesced | |
460 | * @desc: pointer to descriptor | |
461 | * @prev: previous request in desc, or NULL | |
462 | * @req: this request | |
463 | * | |
464 | * Returns zero if @req can be coalesced into @desc, otherwise it returns | |
465 | * the size of the request. | |
466 | */ | |
467 | size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc, | |
468 | struct nfs_page *prev, struct nfs_page *req) | |
5b36c7dc | 469 | { |
48d635f1 | 470 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb WAA |
471 | |
472 | ||
473 | if (mirror->pg_count > mirror->pg_bsize) { | |
f0cb9ab8 WAA |
474 | /* should never happen */ |
475 | WARN_ON_ONCE(1); | |
5b36c7dc | 476 | return 0; |
f0cb9ab8 | 477 | } |
5b36c7dc | 478 | |
2e11f829 CH |
479 | /* |
480 | * Limit the request size so that we can still allocate a page array | |
481 | * for it without upsetting the slab allocator. | |
482 | */ | |
a7d42ddb | 483 | if (((mirror->pg_count + req->wb_bytes) >> PAGE_SHIFT) * |
048883e0 | 484 | sizeof(struct page *) > PAGE_SIZE) |
2e11f829 CH |
485 | return 0; |
486 | ||
a7d42ddb | 487 | return min(mirror->pg_bsize - mirror->pg_count, (size_t)req->wb_bytes); |
5b36c7dc | 488 | } |
19345cb2 | 489 | EXPORT_SYMBOL_GPL(nfs_generic_pg_test); |
5b36c7dc | 490 | |
1e7f3a48 | 491 | struct nfs_pgio_header *nfs_pgio_header_alloc(const struct nfs_rw_ops *ops) |
4a0de55c | 492 | { |
1e7f3a48 | 493 | struct nfs_pgio_header *hdr = ops->rw_alloc_header(); |
4a0de55c | 494 | |
1e7f3a48 | 495 | if (hdr) { |
4a0de55c | 496 | INIT_LIST_HEAD(&hdr->pages); |
4a0de55c | 497 | spin_lock_init(&hdr->lock); |
4a0de55c AS |
498 | hdr->rw_ops = ops; |
499 | } | |
1e7f3a48 | 500 | return hdr; |
4a0de55c | 501 | } |
1e7f3a48 | 502 | EXPORT_SYMBOL_GPL(nfs_pgio_header_alloc); |
4a0de55c | 503 | |
00bfa30a | 504 | /** |
4714fb51 WAA |
505 | * nfs_pgio_data_destroy - make @hdr suitable for reuse |
506 | * | |
507 | * Frees memory and releases refs from nfs_generic_pgio, so that it may | |
508 | * be called again. | |
509 | * | |
510 | * @hdr: A header that has had nfs_generic_pgio called | |
00bfa30a | 511 | */ |
196639eb | 512 | static void nfs_pgio_data_destroy(struct nfs_pgio_header *hdr) |
00bfa30a | 513 | { |
3caa0c6e TM |
514 | if (hdr->args.context) |
515 | put_nfs_open_context(hdr->args.context); | |
d45f60c6 WAA |
516 | if (hdr->page_array.pagevec != hdr->page_array.page_array) |
517 | kfree(hdr->page_array.pagevec); | |
00bfa30a | 518 | } |
196639eb TM |
519 | |
520 | /* | |
521 | * nfs_pgio_header_free - Free a read or write header | |
522 | * @hdr: The header to free | |
523 | */ | |
524 | void nfs_pgio_header_free(struct nfs_pgio_header *hdr) | |
525 | { | |
526 | nfs_pgio_data_destroy(hdr); | |
527 | hdr->rw_ops->rw_free_header(hdr); | |
528 | } | |
529 | EXPORT_SYMBOL_GPL(nfs_pgio_header_free); | |
00bfa30a | 530 | |
ce59515c AS |
531 | /** |
532 | * nfs_pgio_rpcsetup - Set up arguments for a pageio call | |
d45f60c6 | 533 | * @hdr: The pageio hdr |
ce59515c AS |
534 | * @count: Number of bytes to read |
535 | * @offset: Initial offset | |
536 | * @how: How to commit data (writes only) | |
537 | * @cinfo: Commit information for the call (writes only) | |
538 | */ | |
d45f60c6 | 539 | static void nfs_pgio_rpcsetup(struct nfs_pgio_header *hdr, |
e545735a | 540 | unsigned int count, |
ce59515c AS |
541 | int how, struct nfs_commit_info *cinfo) |
542 | { | |
d45f60c6 | 543 | struct nfs_page *req = hdr->req; |
ce59515c AS |
544 | |
545 | /* Set up the RPC argument and reply structs | |
d45f60c6 | 546 | * NB: take care not to mess about with hdr->commit et al. */ |
ce59515c | 547 | |
d45f60c6 | 548 | hdr->args.fh = NFS_FH(hdr->inode); |
e545735a | 549 | hdr->args.offset = req_offset(req); |
ce59515c | 550 | /* pnfs_set_layoutcommit needs this */ |
d45f60c6 | 551 | hdr->mds_offset = hdr->args.offset; |
e545735a | 552 | hdr->args.pgbase = req->wb_pgbase; |
d45f60c6 WAA |
553 | hdr->args.pages = hdr->page_array.pagevec; |
554 | hdr->args.count = count; | |
555 | hdr->args.context = get_nfs_open_context(req->wb_context); | |
556 | hdr->args.lock_context = req->wb_lock_context; | |
557 | hdr->args.stable = NFS_UNSTABLE; | |
ce59515c AS |
558 | switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { |
559 | case 0: | |
560 | break; | |
561 | case FLUSH_COND_STABLE: | |
562 | if (nfs_reqs_to_commit(cinfo)) | |
563 | break; | |
01e03bdc | 564 | /* fall through */ |
ce59515c | 565 | default: |
d45f60c6 | 566 | hdr->args.stable = NFS_FILE_SYNC; |
ce59515c AS |
567 | } |
568 | ||
d45f60c6 WAA |
569 | hdr->res.fattr = &hdr->fattr; |
570 | hdr->res.count = count; | |
571 | hdr->res.eof = 0; | |
c65e6254 | 572 | hdr->res.verf = &hdr->verf; |
d45f60c6 | 573 | nfs_fattr_init(&hdr->fattr); |
ce59515c AS |
574 | } |
575 | ||
a4cdda59 | 576 | /** |
d45f60c6 | 577 | * nfs_pgio_prepare - Prepare pageio hdr to go over the wire |
a4cdda59 | 578 | * @task: The current task |
d45f60c6 | 579 | * @calldata: pageio header to prepare |
a4cdda59 | 580 | */ |
6f92fa45 | 581 | static void nfs_pgio_prepare(struct rpc_task *task, void *calldata) |
a4cdda59 | 582 | { |
d45f60c6 | 583 | struct nfs_pgio_header *hdr = calldata; |
a4cdda59 | 584 | int err; |
d45f60c6 | 585 | err = NFS_PROTO(hdr->inode)->pgio_rpc_prepare(task, hdr); |
a4cdda59 AS |
586 | if (err) |
587 | rpc_exit(task, err); | |
588 | } | |
589 | ||
d45f60c6 | 590 | int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr, |
46a5ab47 | 591 | struct rpc_cred *cred, const struct nfs_rpc_ops *rpc_ops, |
1ed26f33 AS |
592 | const struct rpc_call_ops *call_ops, int how, int flags) |
593 | { | |
594 | struct rpc_task *task; | |
595 | struct rpc_message msg = { | |
d45f60c6 WAA |
596 | .rpc_argp = &hdr->args, |
597 | .rpc_resp = &hdr->res, | |
46a5ab47 | 598 | .rpc_cred = cred, |
1ed26f33 AS |
599 | }; |
600 | struct rpc_task_setup task_setup_data = { | |
601 | .rpc_client = clnt, | |
d45f60c6 | 602 | .task = &hdr->task, |
1ed26f33 AS |
603 | .rpc_message = &msg, |
604 | .callback_ops = call_ops, | |
d45f60c6 | 605 | .callback_data = hdr, |
1ed26f33 AS |
606 | .workqueue = nfsiod_workqueue, |
607 | .flags = RPC_TASK_ASYNC | flags, | |
608 | }; | |
609 | int ret = 0; | |
610 | ||
abde71f4 | 611 | hdr->rw_ops->rw_initiate(hdr, &msg, rpc_ops, &task_setup_data, how); |
1ed26f33 | 612 | |
b4839ebe | 613 | dprintk("NFS: initiated pgio call " |
1ed26f33 | 614 | "(req %s/%llu, %u bytes @ offset %llu)\n", |
343ae531 AS |
615 | hdr->inode->i_sb->s_id, |
616 | (unsigned long long)NFS_FILEID(hdr->inode), | |
d45f60c6 WAA |
617 | hdr->args.count, |
618 | (unsigned long long)hdr->args.offset); | |
1ed26f33 AS |
619 | |
620 | task = rpc_run_task(&task_setup_data); | |
621 | if (IS_ERR(task)) { | |
622 | ret = PTR_ERR(task); | |
623 | goto out; | |
624 | } | |
625 | if (how & FLUSH_SYNC) { | |
626 | ret = rpc_wait_for_completion_task(task); | |
627 | if (ret == 0) | |
628 | ret = task->tk_status; | |
629 | } | |
630 | rpc_put_task(task); | |
631 | out: | |
632 | return ret; | |
633 | } | |
634 | EXPORT_SYMBOL_GPL(nfs_initiate_pgio); | |
635 | ||
844c9e69 AS |
636 | /** |
637 | * nfs_pgio_error - Clean up from a pageio error | |
638 | * @desc: IO descriptor | |
639 | * @hdr: pageio header | |
640 | */ | |
2bff2288 | 641 | static void nfs_pgio_error(struct nfs_pgio_header *hdr) |
844c9e69 | 642 | { |
844c9e69 | 643 | set_bit(NFS_IOHDR_REDO, &hdr->flags); |
4714fb51 | 644 | hdr->completion_ops->completion(hdr); |
844c9e69 AS |
645 | } |
646 | ||
a4cdda59 AS |
647 | /** |
648 | * nfs_pgio_release - Release pageio data | |
d45f60c6 | 649 | * @calldata: The pageio header to release |
a4cdda59 | 650 | */ |
6f92fa45 | 651 | static void nfs_pgio_release(void *calldata) |
a4cdda59 | 652 | { |
d45f60c6 | 653 | struct nfs_pgio_header *hdr = calldata; |
4714fb51 | 654 | hdr->completion_ops->completion(hdr); |
a4cdda59 AS |
655 | } |
656 | ||
a7d42ddb WAA |
657 | static void nfs_pageio_mirror_init(struct nfs_pgio_mirror *mirror, |
658 | unsigned int bsize) | |
659 | { | |
660 | INIT_LIST_HEAD(&mirror->pg_list); | |
661 | mirror->pg_bytes_written = 0; | |
662 | mirror->pg_count = 0; | |
663 | mirror->pg_bsize = bsize; | |
664 | mirror->pg_base = 0; | |
665 | mirror->pg_recoalesce = 0; | |
666 | } | |
667 | ||
1da177e4 | 668 | /** |
d8a5ad75 TM |
669 | * nfs_pageio_init - initialise a page io descriptor |
670 | * @desc: pointer to descriptor | |
bcb71bba | 671 | * @inode: pointer to inode |
dfad7000 YW |
672 | * @pg_ops: pointer to pageio operations |
673 | * @compl_ops: pointer to pageio completion operations | |
674 | * @rw_ops: pointer to nfs read/write operations | |
bcb71bba TM |
675 | * @bsize: io block size |
676 | * @io_flags: extra parameters for the io function | |
d8a5ad75 | 677 | */ |
bcb71bba TM |
678 | void nfs_pageio_init(struct nfs_pageio_descriptor *desc, |
679 | struct inode *inode, | |
1751c363 | 680 | const struct nfs_pageio_ops *pg_ops, |
061ae2ed | 681 | const struct nfs_pgio_completion_ops *compl_ops, |
4a0de55c | 682 | const struct nfs_rw_ops *rw_ops, |
84dde76c | 683 | size_t bsize, |
3bde7afd | 684 | int io_flags) |
d8a5ad75 | 685 | { |
b31268ac | 686 | desc->pg_moreio = 0; |
bcb71bba | 687 | desc->pg_inode = inode; |
1751c363 | 688 | desc->pg_ops = pg_ops; |
061ae2ed | 689 | desc->pg_completion_ops = compl_ops; |
4a0de55c | 690 | desc->pg_rw_ops = rw_ops; |
bcb71bba TM |
691 | desc->pg_ioflags = io_flags; |
692 | desc->pg_error = 0; | |
94ad1c80 | 693 | desc->pg_lseg = NULL; |
919e3bd9 | 694 | desc->pg_io_completion = NULL; |
584aa810 | 695 | desc->pg_dreq = NULL; |
a7d42ddb WAA |
696 | desc->pg_bsize = bsize; |
697 | ||
698 | desc->pg_mirror_count = 1; | |
699 | desc->pg_mirror_idx = 0; | |
700 | ||
14abcb0b TM |
701 | desc->pg_mirrors_dynamic = NULL; |
702 | desc->pg_mirrors = desc->pg_mirrors_static; | |
703 | nfs_pageio_mirror_init(&desc->pg_mirrors[0], bsize); | |
d8a5ad75 TM |
704 | } |
705 | ||
0eecb214 AS |
706 | /** |
707 | * nfs_pgio_result - Basic pageio error handling | |
708 | * @task: The task that ran | |
d45f60c6 | 709 | * @calldata: Pageio header to check |
0eecb214 | 710 | */ |
6f92fa45 | 711 | static void nfs_pgio_result(struct rpc_task *task, void *calldata) |
0eecb214 | 712 | { |
d45f60c6 WAA |
713 | struct nfs_pgio_header *hdr = calldata; |
714 | struct inode *inode = hdr->inode; | |
0eecb214 AS |
715 | |
716 | dprintk("NFS: %s: %5u, (status %d)\n", __func__, | |
717 | task->tk_pid, task->tk_status); | |
718 | ||
d45f60c6 | 719 | if (hdr->rw_ops->rw_done(task, hdr, inode) != 0) |
0eecb214 AS |
720 | return; |
721 | if (task->tk_status < 0) | |
d45f60c6 | 722 | nfs_set_pgio_error(hdr, task->tk_status, hdr->args.offset); |
0eecb214 | 723 | else |
d45f60c6 | 724 | hdr->rw_ops->rw_result(task, hdr); |
0eecb214 AS |
725 | } |
726 | ||
ef2c488c AS |
727 | /* |
728 | * Create an RPC task for the given read or write request and kick it. | |
729 | * The page must have been locked by the caller. | |
730 | * | |
731 | * It may happen that the page we're passed is not marked dirty. | |
732 | * This is the case if nfs_updatepage detects a conflicting request | |
733 | * that has been written but not committed. | |
734 | */ | |
f0cb9ab8 WAA |
735 | int nfs_generic_pgio(struct nfs_pageio_descriptor *desc, |
736 | struct nfs_pgio_header *hdr) | |
ef2c488c | 737 | { |
48d635f1 | 738 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 739 | |
ef2c488c | 740 | struct nfs_page *req; |
bba5c188 WAA |
741 | struct page **pages, |
742 | *last_page; | |
a7d42ddb | 743 | struct list_head *head = &mirror->pg_list; |
ef2c488c | 744 | struct nfs_commit_info cinfo; |
8ef9b0b9 | 745 | struct nfs_page_array *pg_array = &hdr->page_array; |
bba5c188 | 746 | unsigned int pagecount, pageused; |
ae97aa52 | 747 | gfp_t gfp_flags = GFP_KERNEL; |
ef2c488c | 748 | |
a7d42ddb | 749 | pagecount = nfs_page_array_len(mirror->pg_base, mirror->pg_count); |
2eb3aea7 | 750 | pg_array->npages = pagecount; |
8ef9b0b9 BC |
751 | |
752 | if (pagecount <= ARRAY_SIZE(pg_array->page_array)) | |
753 | pg_array->pagevec = pg_array->page_array; | |
754 | else { | |
fbe77c30 | 755 | if (hdr->rw_mode == FMODE_WRITE) |
8ef9b0b9 BC |
756 | gfp_flags = GFP_NOIO; |
757 | pg_array->pagevec = kcalloc(pagecount, sizeof(struct page *), gfp_flags); | |
758 | if (!pg_array->pagevec) { | |
759 | pg_array->npages = 0; | |
760 | nfs_pgio_error(hdr); | |
761 | desc->pg_error = -ENOMEM; | |
762 | return desc->pg_error; | |
763 | } | |
2bff2288 | 764 | } |
ef2c488c AS |
765 | |
766 | nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); | |
d45f60c6 | 767 | pages = hdr->page_array.pagevec; |
bba5c188 WAA |
768 | last_page = NULL; |
769 | pageused = 0; | |
ef2c488c AS |
770 | while (!list_empty(head)) { |
771 | req = nfs_list_entry(head->next); | |
772 | nfs_list_remove_request(req); | |
773 | nfs_list_add_request(req, &hdr->pages); | |
bba5c188 | 774 | |
bba5c188 | 775 | if (!last_page || last_page != req->wb_page) { |
bba5c188 | 776 | pageused++; |
b8fb9c30 TM |
777 | if (pageused > pagecount) |
778 | break; | |
779 | *pages++ = last_page = req->wb_page; | |
bba5c188 | 780 | } |
ef2c488c | 781 | } |
2bff2288 PT |
782 | if (WARN_ON_ONCE(pageused != pagecount)) { |
783 | nfs_pgio_error(hdr); | |
784 | desc->pg_error = -EINVAL; | |
785 | return desc->pg_error; | |
786 | } | |
ef2c488c AS |
787 | |
788 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && | |
789 | (desc->pg_moreio || nfs_reqs_to_commit(&cinfo))) | |
790 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
791 | ||
792 | /* Set up the argument struct */ | |
e545735a | 793 | nfs_pgio_rpcsetup(hdr, mirror->pg_count, desc->pg_ioflags, &cinfo); |
ef2c488c AS |
794 | desc->pg_rpc_callops = &nfs_pgio_common_ops; |
795 | return 0; | |
796 | } | |
f0cb9ab8 | 797 | EXPORT_SYMBOL_GPL(nfs_generic_pgio); |
ef2c488c | 798 | |
41d8d5b7 | 799 | static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc) |
cf485fcd | 800 | { |
cf485fcd AS |
801 | struct nfs_pgio_header *hdr; |
802 | int ret; | |
803 | ||
1e7f3a48 WAA |
804 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
805 | if (!hdr) { | |
2bff2288 PT |
806 | desc->pg_error = -ENOMEM; |
807 | return desc->pg_error; | |
cf485fcd | 808 | } |
1e7f3a48 | 809 | nfs_pgheader_init(desc, hdr, nfs_pgio_header_free); |
cf485fcd AS |
810 | ret = nfs_generic_pgio(desc, hdr); |
811 | if (ret == 0) | |
7f714720 | 812 | ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode), |
46a5ab47 PT |
813 | hdr, |
814 | hdr->cred, | |
815 | NFS_PROTO(hdr->inode), | |
abde71f4 | 816 | desc->pg_rpc_callops, |
7f714720 | 817 | desc->pg_ioflags, 0); |
cf485fcd AS |
818 | return ret; |
819 | } | |
820 | ||
14abcb0b TM |
821 | static struct nfs_pgio_mirror * |
822 | nfs_pageio_alloc_mirrors(struct nfs_pageio_descriptor *desc, | |
823 | unsigned int mirror_count) | |
824 | { | |
825 | struct nfs_pgio_mirror *ret; | |
826 | unsigned int i; | |
827 | ||
828 | kfree(desc->pg_mirrors_dynamic); | |
829 | desc->pg_mirrors_dynamic = NULL; | |
830 | if (mirror_count == 1) | |
831 | return desc->pg_mirrors_static; | |
832 | ret = kmalloc_array(mirror_count, sizeof(*ret), GFP_NOFS); | |
833 | if (ret != NULL) { | |
834 | for (i = 0; i < mirror_count; i++) | |
835 | nfs_pageio_mirror_init(&ret[i], desc->pg_bsize); | |
836 | desc->pg_mirrors_dynamic = ret; | |
837 | } | |
838 | return ret; | |
839 | } | |
840 | ||
a7d42ddb WAA |
841 | /* |
842 | * nfs_pageio_setup_mirroring - determine if mirroring is to be used | |
843 | * by calling the pg_get_mirror_count op | |
844 | */ | |
14abcb0b | 845 | static void nfs_pageio_setup_mirroring(struct nfs_pageio_descriptor *pgio, |
a7d42ddb WAA |
846 | struct nfs_page *req) |
847 | { | |
14abcb0b | 848 | unsigned int mirror_count = 1; |
a7d42ddb | 849 | |
14abcb0b TM |
850 | if (pgio->pg_ops->pg_get_mirror_count) |
851 | mirror_count = pgio->pg_ops->pg_get_mirror_count(pgio, req); | |
852 | if (mirror_count == pgio->pg_mirror_count || pgio->pg_error < 0) | |
853 | return; | |
a7d42ddb | 854 | |
14abcb0b TM |
855 | if (!mirror_count || mirror_count > NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX) { |
856 | pgio->pg_error = -EINVAL; | |
857 | return; | |
858 | } | |
a7d42ddb | 859 | |
14abcb0b TM |
860 | pgio->pg_mirrors = nfs_pageio_alloc_mirrors(pgio, mirror_count); |
861 | if (pgio->pg_mirrors == NULL) { | |
862 | pgio->pg_error = -ENOMEM; | |
863 | pgio->pg_mirrors = pgio->pg_mirrors_static; | |
864 | mirror_count = 1; | |
865 | } | |
a7d42ddb | 866 | pgio->pg_mirror_count = mirror_count; |
a7d42ddb WAA |
867 | } |
868 | ||
869 | /* | |
870 | * nfs_pageio_stop_mirroring - stop using mirroring (set mirror count to 1) | |
871 | */ | |
872 | void nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor *pgio) | |
873 | { | |
874 | pgio->pg_mirror_count = 1; | |
875 | pgio->pg_mirror_idx = 0; | |
876 | } | |
877 | ||
878 | static void nfs_pageio_cleanup_mirroring(struct nfs_pageio_descriptor *pgio) | |
879 | { | |
880 | pgio->pg_mirror_count = 1; | |
881 | pgio->pg_mirror_idx = 0; | |
882 | pgio->pg_mirrors = pgio->pg_mirrors_static; | |
883 | kfree(pgio->pg_mirrors_dynamic); | |
884 | pgio->pg_mirrors_dynamic = NULL; | |
885 | } | |
886 | ||
4109bb74 TM |
887 | static bool nfs_match_lock_context(const struct nfs_lock_context *l1, |
888 | const struct nfs_lock_context *l2) | |
889 | { | |
d51fdb87 | 890 | return l1->lockowner == l2->lockowner; |
4109bb74 TM |
891 | } |
892 | ||
d8a5ad75 TM |
893 | /** |
894 | * nfs_can_coalesce_requests - test two requests for compatibility | |
895 | * @prev: pointer to nfs_page | |
896 | * @req: pointer to nfs_page | |
897 | * | |
898 | * The nfs_page structures 'prev' and 'req' are compared to ensure that the | |
899 | * page data area they describe is contiguous, and that their RPC | |
900 | * credentials, NFSv4 open state, and lockowners are the same. | |
901 | * | |
902 | * Return 'true' if this is the case, else return 'false'. | |
903 | */ | |
18ad0a9f BH |
904 | static bool nfs_can_coalesce_requests(struct nfs_page *prev, |
905 | struct nfs_page *req, | |
906 | struct nfs_pageio_descriptor *pgio) | |
d8a5ad75 | 907 | { |
b4fdac1a | 908 | size_t size; |
5263e31e | 909 | struct file_lock_context *flctx; |
b4fdac1a | 910 | |
ab75e417 WAA |
911 | if (prev) { |
912 | if (!nfs_match_open_context(req->wb_context, prev->wb_context)) | |
913 | return false; | |
2b0143b5 | 914 | flctx = d_inode(req->wb_context->dentry)->i_flctx; |
bd61e0a9 JL |
915 | if (flctx != NULL && |
916 | !(list_empty_careful(&flctx->flc_posix) && | |
917 | list_empty_careful(&flctx->flc_flock)) && | |
5263e31e JL |
918 | !nfs_match_lock_context(req->wb_lock_context, |
919 | prev->wb_lock_context)) | |
920 | return false; | |
ab75e417 WAA |
921 | if (req_offset(req) != req_offset(prev) + prev->wb_bytes) |
922 | return false; | |
78270e8f WAA |
923 | if (req->wb_page == prev->wb_page) { |
924 | if (req->wb_pgbase != prev->wb_pgbase + prev->wb_bytes) | |
925 | return false; | |
926 | } else { | |
927 | if (req->wb_pgbase != 0 || | |
09cbfeaf | 928 | prev->wb_pgbase + prev->wb_bytes != PAGE_SIZE) |
78270e8f WAA |
929 | return false; |
930 | } | |
ab75e417 | 931 | } |
b4fdac1a | 932 | size = pgio->pg_ops->pg_test(pgio, prev, req); |
f0cb9ab8 WAA |
933 | WARN_ON_ONCE(size > req->wb_bytes); |
934 | if (size && size < req->wb_bytes) | |
935 | req->wb_bytes = size; | |
b4fdac1a | 936 | return size > 0; |
d8a5ad75 TM |
937 | } |
938 | ||
939 | /** | |
bcb71bba | 940 | * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list. |
d8a5ad75 TM |
941 | * @desc: destination io descriptor |
942 | * @req: request | |
943 | * | |
944 | * Returns true if the request 'req' was successfully coalesced into the | |
945 | * existing list of pages 'desc'. | |
946 | */ | |
bcb71bba TM |
947 | static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc, |
948 | struct nfs_page *req) | |
d8a5ad75 | 949 | { |
48d635f1 | 950 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 951 | |
ab75e417 | 952 | struct nfs_page *prev = NULL; |
a7d42ddb WAA |
953 | |
954 | if (mirror->pg_count != 0) { | |
955 | prev = nfs_list_entry(mirror->pg_list.prev); | |
5b36c7dc | 956 | } else { |
d8007d4d TM |
957 | if (desc->pg_ops->pg_init) |
958 | desc->pg_ops->pg_init(desc, req); | |
d600ad1f PT |
959 | if (desc->pg_error < 0) |
960 | return 0; | |
a7d42ddb | 961 | mirror->pg_base = req->wb_pgbase; |
5b36c7dc | 962 | } |
ab75e417 WAA |
963 | if (!nfs_can_coalesce_requests(prev, req, desc)) |
964 | return 0; | |
d8a5ad75 | 965 | nfs_list_remove_request(req); |
a7d42ddb WAA |
966 | nfs_list_add_request(req, &mirror->pg_list); |
967 | mirror->pg_count += req->wb_bytes; | |
d8a5ad75 TM |
968 | return 1; |
969 | } | |
970 | ||
bcb71bba TM |
971 | /* |
972 | * Helper for nfs_pageio_add_request and nfs_pageio_complete | |
973 | */ | |
974 | static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc) | |
975 | { | |
48d635f1 | 976 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb WAA |
977 | |
978 | ||
979 | if (!list_empty(&mirror->pg_list)) { | |
1751c363 | 980 | int error = desc->pg_ops->pg_doio(desc); |
bcb71bba TM |
981 | if (error < 0) |
982 | desc->pg_error = error; | |
983 | else | |
a7d42ddb | 984 | mirror->pg_bytes_written += mirror->pg_count; |
bcb71bba | 985 | } |
a7d42ddb WAA |
986 | if (list_empty(&mirror->pg_list)) { |
987 | mirror->pg_count = 0; | |
988 | mirror->pg_base = 0; | |
bcb71bba TM |
989 | } |
990 | } | |
991 | ||
992 | /** | |
993 | * nfs_pageio_add_request - Attempt to coalesce a request into a page list. | |
994 | * @desc: destination io descriptor | |
995 | * @req: request | |
996 | * | |
2bfc6e56 WAA |
997 | * This may split a request into subrequests which are all part of the |
998 | * same page group. | |
999 | * | |
bcb71bba TM |
1000 | * Returns true if the request 'req' was successfully coalesced into the |
1001 | * existing list of pages 'desc'. | |
1002 | */ | |
d9156f9f | 1003 | static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc, |
8b09bee3 | 1004 | struct nfs_page *req) |
bcb71bba | 1005 | { |
48d635f1 | 1006 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 1007 | |
2bfc6e56 WAA |
1008 | struct nfs_page *subreq; |
1009 | unsigned int bytes_left = 0; | |
1010 | unsigned int offset, pgbase; | |
1011 | ||
1344b7ea | 1012 | nfs_page_group_lock(req); |
2bfc6e56 WAA |
1013 | |
1014 | subreq = req; | |
1015 | bytes_left = subreq->wb_bytes; | |
1016 | offset = subreq->wb_offset; | |
1017 | pgbase = subreq->wb_pgbase; | |
1018 | ||
1019 | do { | |
1020 | if (!nfs_pageio_do_add_request(desc, subreq)) { | |
1021 | /* make sure pg_test call(s) did nothing */ | |
1022 | WARN_ON_ONCE(subreq->wb_bytes != bytes_left); | |
1023 | WARN_ON_ONCE(subreq->wb_offset != offset); | |
1024 | WARN_ON_ONCE(subreq->wb_pgbase != pgbase); | |
1025 | ||
1026 | nfs_page_group_unlock(req); | |
1027 | desc->pg_moreio = 1; | |
1028 | nfs_pageio_doio(desc); | |
1029 | if (desc->pg_error < 0) | |
1030 | return 0; | |
a7d42ddb | 1031 | if (mirror->pg_recoalesce) |
2bfc6e56 WAA |
1032 | return 0; |
1033 | /* retry add_request for this subreq */ | |
1344b7ea | 1034 | nfs_page_group_lock(req); |
2bfc6e56 WAA |
1035 | continue; |
1036 | } | |
1037 | ||
1038 | /* check for buggy pg_test call(s) */ | |
1039 | WARN_ON_ONCE(subreq->wb_bytes + subreq->wb_pgbase > PAGE_SIZE); | |
1040 | WARN_ON_ONCE(subreq->wb_bytes > bytes_left); | |
1041 | WARN_ON_ONCE(subreq->wb_bytes == 0); | |
1042 | ||
1043 | bytes_left -= subreq->wb_bytes; | |
1044 | offset += subreq->wb_bytes; | |
1045 | pgbase += subreq->wb_bytes; | |
1046 | ||
1047 | if (bytes_left) { | |
1048 | subreq = nfs_create_request(req->wb_context, | |
1049 | req->wb_page, | |
1050 | subreq, pgbase, bytes_left); | |
c1109558 TM |
1051 | if (IS_ERR(subreq)) |
1052 | goto err_ptr; | |
2bfc6e56 WAA |
1053 | nfs_lock_request(subreq); |
1054 | subreq->wb_offset = offset; | |
1055 | subreq->wb_index = req->wb_index; | |
1056 | } | |
1057 | } while (bytes_left > 0); | |
1058 | ||
1059 | nfs_page_group_unlock(req); | |
bcb71bba | 1060 | return 1; |
c1109558 TM |
1061 | err_ptr: |
1062 | desc->pg_error = PTR_ERR(subreq); | |
1063 | nfs_page_group_unlock(req); | |
1064 | return 0; | |
bcb71bba TM |
1065 | } |
1066 | ||
d9156f9f TM |
1067 | static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc) |
1068 | { | |
48d635f1 | 1069 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
d9156f9f TM |
1070 | LIST_HEAD(head); |
1071 | ||
1072 | do { | |
a7d42ddb WAA |
1073 | list_splice_init(&mirror->pg_list, &head); |
1074 | mirror->pg_bytes_written -= mirror->pg_count; | |
1075 | mirror->pg_count = 0; | |
1076 | mirror->pg_base = 0; | |
1077 | mirror->pg_recoalesce = 0; | |
1078 | ||
d9156f9f TM |
1079 | while (!list_empty(&head)) { |
1080 | struct nfs_page *req; | |
1081 | ||
1082 | req = list_first_entry(&head, struct nfs_page, wb_list); | |
1083 | nfs_list_remove_request(req); | |
1084 | if (__nfs_pageio_add_request(desc, req)) | |
1085 | continue; | |
03d5eb65 TM |
1086 | if (desc->pg_error < 0) { |
1087 | list_splice_tail(&head, &mirror->pg_list); | |
1088 | mirror->pg_recoalesce = 1; | |
d9156f9f | 1089 | return 0; |
03d5eb65 | 1090 | } |
d9156f9f TM |
1091 | break; |
1092 | } | |
a7d42ddb | 1093 | } while (mirror->pg_recoalesce); |
d9156f9f TM |
1094 | return 1; |
1095 | } | |
1096 | ||
a7d42ddb | 1097 | static int nfs_pageio_add_request_mirror(struct nfs_pageio_descriptor *desc, |
d9156f9f TM |
1098 | struct nfs_page *req) |
1099 | { | |
1100 | int ret; | |
1101 | ||
1102 | do { | |
1103 | ret = __nfs_pageio_add_request(desc, req); | |
1104 | if (ret) | |
1105 | break; | |
1106 | if (desc->pg_error < 0) | |
1107 | break; | |
1108 | ret = nfs_do_recoalesce(desc); | |
1109 | } while (ret); | |
a7d42ddb | 1110 | |
d9156f9f TM |
1111 | return ret; |
1112 | } | |
1113 | ||
a7d42ddb WAA |
1114 | int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc, |
1115 | struct nfs_page *req) | |
1116 | { | |
1117 | u32 midx; | |
1118 | unsigned int pgbase, offset, bytes; | |
1119 | struct nfs_page *dupreq, *lastreq; | |
1120 | ||
1121 | pgbase = req->wb_pgbase; | |
1122 | offset = req->wb_offset; | |
1123 | bytes = req->wb_bytes; | |
1124 | ||
1125 | nfs_pageio_setup_mirroring(desc, req); | |
d600ad1f | 1126 | if (desc->pg_error < 0) |
c18b96a1 | 1127 | goto out_failed; |
a7d42ddb WAA |
1128 | |
1129 | for (midx = 0; midx < desc->pg_mirror_count; midx++) { | |
1130 | if (midx) { | |
1344b7ea | 1131 | nfs_page_group_lock(req); |
a7d42ddb WAA |
1132 | |
1133 | /* find the last request */ | |
1134 | for (lastreq = req->wb_head; | |
1135 | lastreq->wb_this_page != req->wb_head; | |
1136 | lastreq = lastreq->wb_this_page) | |
1137 | ; | |
1138 | ||
1139 | dupreq = nfs_create_request(req->wb_context, | |
1140 | req->wb_page, lastreq, pgbase, bytes); | |
1141 | ||
1142 | if (IS_ERR(dupreq)) { | |
1143 | nfs_page_group_unlock(req); | |
c18b96a1 PT |
1144 | desc->pg_error = PTR_ERR(dupreq); |
1145 | goto out_failed; | |
a7d42ddb WAA |
1146 | } |
1147 | ||
1148 | nfs_lock_request(dupreq); | |
1149 | nfs_page_group_unlock(req); | |
1150 | dupreq->wb_offset = offset; | |
1151 | dupreq->wb_index = req->wb_index; | |
1152 | } else | |
1153 | dupreq = req; | |
1154 | ||
47af81f2 PT |
1155 | if (nfs_pgio_has_mirroring(desc)) |
1156 | desc->pg_mirror_idx = midx; | |
a7d42ddb | 1157 | if (!nfs_pageio_add_request_mirror(desc, dupreq)) |
c18b96a1 | 1158 | goto out_failed; |
a7d42ddb WAA |
1159 | } |
1160 | ||
1161 | return 1; | |
c18b96a1 PT |
1162 | |
1163 | out_failed: | |
1164 | /* | |
1165 | * We might have failed before sending any reqs over wire. | |
2bff2288 | 1166 | * Clean up rest of the reqs in mirror pg_list. |
c18b96a1 | 1167 | */ |
2bff2288 PT |
1168 | if (desc->pg_error) { |
1169 | struct nfs_pgio_mirror *mirror; | |
1170 | void (*func)(struct list_head *); | |
1171 | ||
1172 | /* remember fatal errors */ | |
1173 | if (nfs_error_is_fatal(desc->pg_error)) | |
bf4b4905 N |
1174 | nfs_context_set_write_error(req->wb_context, |
1175 | desc->pg_error); | |
2bff2288 PT |
1176 | |
1177 | func = desc->pg_completion_ops->error_cleanup; | |
1178 | for (midx = 0; midx < desc->pg_mirror_count; midx++) { | |
1179 | mirror = &desc->pg_mirrors[midx]; | |
1180 | func(&mirror->pg_list); | |
1181 | } | |
1182 | } | |
c18b96a1 | 1183 | return 0; |
a7d42ddb WAA |
1184 | } |
1185 | ||
1186 | /* | |
1187 | * nfs_pageio_complete_mirror - Complete I/O on the current mirror of an | |
1188 | * nfs_pageio_descriptor | |
1189 | * @desc: pointer to io descriptor | |
dfad7000 | 1190 | * @mirror_idx: pointer to mirror index |
a7d42ddb WAA |
1191 | */ |
1192 | static void nfs_pageio_complete_mirror(struct nfs_pageio_descriptor *desc, | |
1193 | u32 mirror_idx) | |
1194 | { | |
1195 | struct nfs_pgio_mirror *mirror = &desc->pg_mirrors[mirror_idx]; | |
1196 | u32 restore_idx = desc->pg_mirror_idx; | |
1197 | ||
47af81f2 PT |
1198 | if (nfs_pgio_has_mirroring(desc)) |
1199 | desc->pg_mirror_idx = mirror_idx; | |
a7d42ddb WAA |
1200 | for (;;) { |
1201 | nfs_pageio_doio(desc); | |
1202 | if (!mirror->pg_recoalesce) | |
1203 | break; | |
1204 | if (!nfs_do_recoalesce(desc)) | |
1205 | break; | |
1206 | } | |
1207 | desc->pg_mirror_idx = restore_idx; | |
1208 | } | |
1209 | ||
53113ad3 WAA |
1210 | /* |
1211 | * nfs_pageio_resend - Transfer requests to new descriptor and resend | |
1212 | * @hdr - the pgio header to move request from | |
1213 | * @desc - the pageio descriptor to add requests to | |
1214 | * | |
1215 | * Try to move each request (nfs_page) from @hdr to @desc then attempt | |
1216 | * to send them. | |
1217 | * | |
1218 | * Returns 0 on success and < 0 on error. | |
1219 | */ | |
1220 | int nfs_pageio_resend(struct nfs_pageio_descriptor *desc, | |
1221 | struct nfs_pgio_header *hdr) | |
1222 | { | |
1223 | LIST_HEAD(failed); | |
1224 | ||
919e3bd9 | 1225 | desc->pg_io_completion = hdr->io_completion; |
53113ad3 WAA |
1226 | desc->pg_dreq = hdr->dreq; |
1227 | while (!list_empty(&hdr->pages)) { | |
1228 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
1229 | ||
1230 | nfs_list_remove_request(req); | |
1231 | if (!nfs_pageio_add_request(desc, req)) | |
1232 | nfs_list_add_request(req, &failed); | |
1233 | } | |
1234 | nfs_pageio_complete(desc); | |
1235 | if (!list_empty(&failed)) { | |
1236 | list_move(&failed, &hdr->pages); | |
d600ad1f | 1237 | return desc->pg_error < 0 ? desc->pg_error : -EIO; |
53113ad3 WAA |
1238 | } |
1239 | return 0; | |
1240 | } | |
1241 | EXPORT_SYMBOL_GPL(nfs_pageio_resend); | |
d9156f9f | 1242 | |
bcb71bba | 1243 | /** |
2176bf42 | 1244 | * nfs_pageio_complete - Complete I/O then cleanup an nfs_pageio_descriptor |
bcb71bba TM |
1245 | * @desc: pointer to io descriptor |
1246 | */ | |
1247 | void nfs_pageio_complete(struct nfs_pageio_descriptor *desc) | |
1248 | { | |
a7d42ddb WAA |
1249 | u32 midx; |
1250 | ||
1251 | for (midx = 0; midx < desc->pg_mirror_count; midx++) | |
1252 | nfs_pageio_complete_mirror(desc, midx); | |
2176bf42 WAA |
1253 | |
1254 | if (desc->pg_ops->pg_cleanup) | |
1255 | desc->pg_ops->pg_cleanup(desc); | |
a7d42ddb | 1256 | nfs_pageio_cleanup_mirroring(desc); |
bcb71bba TM |
1257 | } |
1258 | ||
7fe7f848 TM |
1259 | /** |
1260 | * nfs_pageio_cond_complete - Conditional I/O completion | |
1261 | * @desc: pointer to io descriptor | |
1262 | * @index: page index | |
1263 | * | |
1264 | * It is important to ensure that processes don't try to take locks | |
1265 | * on non-contiguous ranges of pages as that might deadlock. This | |
1266 | * function should be called before attempting to wait on a locked | |
1267 | * nfs_page. It will complete the I/O if the page index 'index' | |
1268 | * is not contiguous with the existing list of pages in 'desc'. | |
1269 | */ | |
1270 | void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index) | |
1271 | { | |
a7d42ddb WAA |
1272 | struct nfs_pgio_mirror *mirror; |
1273 | struct nfs_page *prev; | |
1274 | u32 midx; | |
1275 | ||
1276 | for (midx = 0; midx < desc->pg_mirror_count; midx++) { | |
1277 | mirror = &desc->pg_mirrors[midx]; | |
1278 | if (!list_empty(&mirror->pg_list)) { | |
1279 | prev = nfs_list_entry(mirror->pg_list.prev); | |
43b7d964 BC |
1280 | if (index != prev->wb_index + 1) { |
1281 | nfs_pageio_complete(desc); | |
1282 | break; | |
1283 | } | |
a7d42ddb | 1284 | } |
7fe7f848 TM |
1285 | } |
1286 | } | |
1287 | ||
f7b422b1 | 1288 | int __init nfs_init_nfspagecache(void) |
1da177e4 LT |
1289 | { |
1290 | nfs_page_cachep = kmem_cache_create("nfs_page", | |
1291 | sizeof(struct nfs_page), | |
1292 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1293 | NULL); |
1da177e4 LT |
1294 | if (nfs_page_cachep == NULL) |
1295 | return -ENOMEM; | |
1296 | ||
1297 | return 0; | |
1298 | } | |
1299 | ||
266bee88 | 1300 | void nfs_destroy_nfspagecache(void) |
1da177e4 | 1301 | { |
1a1d92c1 | 1302 | kmem_cache_destroy(nfs_page_cachep); |
1da177e4 LT |
1303 | } |
1304 | ||
ef2c488c | 1305 | static const struct rpc_call_ops nfs_pgio_common_ops = { |
6f92fa45 AS |
1306 | .rpc_call_prepare = nfs_pgio_prepare, |
1307 | .rpc_call_done = nfs_pgio_result, | |
1308 | .rpc_release = nfs_pgio_release, | |
1309 | }; | |
41d8d5b7 AS |
1310 | |
1311 | const struct nfs_pageio_ops nfs_pgio_rw_ops = { | |
1312 | .pg_test = nfs_generic_pg_test, | |
1313 | .pg_doio = nfs_generic_pg_pgios, | |
1314 | }; |