Commit | Line | Data |
---|---|---|
f54bcf2e TH |
1 | /* |
2 | * Common NFS I/O operations for the pnfs file based | |
3 | * layout drivers. | |
4 | * | |
5 | * Copyright (c) 2014, Primary Data, Inc. All rights reserved. | |
6 | * | |
7 | * Tom Haynes <loghyr@primarydata.com> | |
8 | */ | |
9 | ||
10 | #include <linux/nfs_fs.h> | |
11 | #include <linux/nfs_page.h> | |
6b7f3cf9 | 12 | #include <linux/sunrpc/addr.h> |
5f01d953 | 13 | #include <linux/module.h> |
f54bcf2e | 14 | |
7405f9e1 | 15 | #include "nfs4session.h" |
f54bcf2e TH |
16 | #include "internal.h" |
17 | #include "pnfs.h" | |
18 | ||
875ae069 PT |
19 | #define NFSDBG_FACILITY NFSDBG_PNFS |
20 | ||
f54bcf2e TH |
21 | void pnfs_generic_rw_release(void *data) |
22 | { | |
23 | struct nfs_pgio_header *hdr = data; | |
f54bcf2e | 24 | |
f54bcf2e TH |
25 | nfs_put_client(hdr->ds_clp); |
26 | hdr->mds_ops->rpc_release(data); | |
27 | } | |
28 | EXPORT_SYMBOL_GPL(pnfs_generic_rw_release); | |
29 | ||
30 | /* Fake up some data that will cause nfs_commit_release to retry the writes. */ | |
31 | void pnfs_generic_prepare_to_resend_writes(struct nfs_commit_data *data) | |
32 | { | |
33 | struct nfs_page *first = nfs_list_entry(data->pages.next); | |
34 | ||
35 | data->task.tk_status = 0; | |
36 | memcpy(&data->verf.verifier, &first->wb_verf, | |
37 | sizeof(data->verf.verifier)); | |
38 | data->verf.verifier.data[0]++; /* ensure verifier mismatch */ | |
39 | } | |
40 | EXPORT_SYMBOL_GPL(pnfs_generic_prepare_to_resend_writes); | |
41 | ||
42 | void pnfs_generic_write_commit_done(struct rpc_task *task, void *data) | |
43 | { | |
44 | struct nfs_commit_data *wdata = data; | |
45 | ||
46 | /* Note this may cause RPC to be resent */ | |
47 | wdata->mds_ops->rpc_call_done(task, data); | |
48 | } | |
49 | EXPORT_SYMBOL_GPL(pnfs_generic_write_commit_done); | |
50 | ||
51 | void pnfs_generic_commit_release(void *calldata) | |
52 | { | |
53 | struct nfs_commit_data *data = calldata; | |
54 | ||
55 | data->completion_ops->completion(data); | |
56 | pnfs_put_lseg(data->lseg); | |
57 | nfs_put_client(data->ds_clp); | |
58 | nfs_commitdata_release(data); | |
59 | } | |
60 | EXPORT_SYMBOL_GPL(pnfs_generic_commit_release); | |
61 | ||
62 | /* The generic layer is about to remove the req from the commit list. | |
63 | * If this will make the bucket empty, it will need to put the lseg reference. | |
fe238e60 | 64 | * Note this must be called holding i_lock |
f54bcf2e TH |
65 | */ |
66 | void | |
67 | pnfs_generic_clear_request_commit(struct nfs_page *req, | |
68 | struct nfs_commit_info *cinfo) | |
69 | { | |
70 | struct pnfs_layout_segment *freeme = NULL; | |
71 | ||
72 | if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags)) | |
73 | goto out; | |
74 | cinfo->ds->nwritten--; | |
75 | if (list_is_singular(&req->wb_list)) { | |
76 | struct pnfs_commit_bucket *bucket; | |
77 | ||
78 | bucket = list_first_entry(&req->wb_list, | |
79 | struct pnfs_commit_bucket, | |
80 | written); | |
81 | freeme = bucket->wlseg; | |
82 | bucket->wlseg = NULL; | |
83 | } | |
84 | out: | |
85 | nfs_request_remove_commit_list(req, cinfo); | |
86 | pnfs_put_lseg_locked(freeme); | |
87 | } | |
88 | EXPORT_SYMBOL_GPL(pnfs_generic_clear_request_commit); | |
89 | ||
90 | static int | |
91 | pnfs_generic_transfer_commit_list(struct list_head *src, struct list_head *dst, | |
92 | struct nfs_commit_info *cinfo, int max) | |
93 | { | |
94 | struct nfs_page *req, *tmp; | |
95 | int ret = 0; | |
96 | ||
97 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
98 | if (!nfs_lock_request(req)) | |
99 | continue; | |
100 | kref_get(&req->wb_kref); | |
fe238e60 | 101 | if (cond_resched_lock(&cinfo->inode->i_lock)) |
f54bcf2e TH |
102 | list_safe_reset_next(req, tmp, wb_list); |
103 | nfs_request_remove_commit_list(req, cinfo); | |
104 | clear_bit(PG_COMMIT_TO_DS, &req->wb_flags); | |
105 | nfs_list_add_request(req, dst); | |
106 | ret++; | |
107 | if ((ret == max) && !cinfo->dreq) | |
108 | break; | |
109 | } | |
110 | return ret; | |
111 | } | |
112 | ||
f54bcf2e TH |
113 | static int |
114 | pnfs_generic_scan_ds_commit_list(struct pnfs_commit_bucket *bucket, | |
115 | struct nfs_commit_info *cinfo, | |
116 | int max) | |
117 | { | |
118 | struct list_head *src = &bucket->written; | |
119 | struct list_head *dst = &bucket->committing; | |
120 | int ret; | |
121 | ||
fe238e60 | 122 | lockdep_assert_held(&cinfo->inode->i_lock); |
f54bcf2e TH |
123 | ret = pnfs_generic_transfer_commit_list(src, dst, cinfo, max); |
124 | if (ret) { | |
125 | cinfo->ds->nwritten -= ret; | |
126 | cinfo->ds->ncommitting += ret; | |
27571297 TM |
127 | if (bucket->clseg == NULL) |
128 | bucket->clseg = pnfs_get_lseg(bucket->wlseg); | |
129 | if (list_empty(src)) { | |
130 | pnfs_put_lseg_locked(bucket->wlseg); | |
f54bcf2e | 131 | bucket->wlseg = NULL; |
27571297 | 132 | } |
f54bcf2e TH |
133 | } |
134 | return ret; | |
135 | } | |
136 | ||
085d1e33 TH |
137 | /* Move reqs from written to committing lists, returning count |
138 | * of number moved. | |
f54bcf2e TH |
139 | */ |
140 | int pnfs_generic_scan_commit_lists(struct nfs_commit_info *cinfo, | |
141 | int max) | |
142 | { | |
143 | int i, rv = 0, cnt; | |
144 | ||
fe238e60 | 145 | lockdep_assert_held(&cinfo->inode->i_lock); |
f54bcf2e TH |
146 | for (i = 0; i < cinfo->ds->nbuckets && max != 0; i++) { |
147 | cnt = pnfs_generic_scan_ds_commit_list(&cinfo->ds->buckets[i], | |
148 | cinfo, max); | |
149 | max -= cnt; | |
150 | rv += cnt; | |
151 | } | |
152 | return rv; | |
153 | } | |
154 | EXPORT_SYMBOL_GPL(pnfs_generic_scan_commit_lists); | |
155 | ||
085d1e33 | 156 | /* Pull everything off the committing lists and dump into @dst. */ |
f54bcf2e TH |
157 | void pnfs_generic_recover_commit_reqs(struct list_head *dst, |
158 | struct nfs_commit_info *cinfo) | |
159 | { | |
160 | struct pnfs_commit_bucket *b; | |
161 | struct pnfs_layout_segment *freeme; | |
162 | int i; | |
163 | ||
fe238e60 | 164 | lockdep_assert_held(&cinfo->inode->i_lock); |
f54bcf2e | 165 | restart: |
f54bcf2e TH |
166 | for (i = 0, b = cinfo->ds->buckets; i < cinfo->ds->nbuckets; i++, b++) { |
167 | if (pnfs_generic_transfer_commit_list(&b->written, dst, | |
168 | cinfo, 0)) { | |
169 | freeme = b->wlseg; | |
170 | b->wlseg = NULL; | |
fe238e60 | 171 | spin_unlock(&cinfo->inode->i_lock); |
f54bcf2e | 172 | pnfs_put_lseg(freeme); |
fe238e60 | 173 | spin_lock(&cinfo->inode->i_lock); |
f54bcf2e TH |
174 | goto restart; |
175 | } | |
176 | } | |
177 | cinfo->ds->nwritten = 0; | |
f54bcf2e TH |
178 | } |
179 | EXPORT_SYMBOL_GPL(pnfs_generic_recover_commit_reqs); | |
180 | ||
181 | static void pnfs_generic_retry_commit(struct nfs_commit_info *cinfo, int idx) | |
182 | { | |
183 | struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds; | |
184 | struct pnfs_commit_bucket *bucket; | |
185 | struct pnfs_layout_segment *freeme; | |
27571297 | 186 | LIST_HEAD(pages); |
f54bcf2e TH |
187 | int i; |
188 | ||
fe238e60 | 189 | spin_lock(&cinfo->inode->i_lock); |
f54bcf2e TH |
190 | for (i = idx; i < fl_cinfo->nbuckets; i++) { |
191 | bucket = &fl_cinfo->buckets[i]; | |
192 | if (list_empty(&bucket->committing)) | |
193 | continue; | |
f54bcf2e TH |
194 | freeme = bucket->clseg; |
195 | bucket->clseg = NULL; | |
27571297 | 196 | list_splice_init(&bucket->committing, &pages); |
fe238e60 | 197 | spin_unlock(&cinfo->inode->i_lock); |
27571297 | 198 | nfs_retry_commit(&pages, freeme, cinfo, i); |
f54bcf2e | 199 | pnfs_put_lseg(freeme); |
fe238e60 | 200 | spin_lock(&cinfo->inode->i_lock); |
f54bcf2e | 201 | } |
fe238e60 | 202 | spin_unlock(&cinfo->inode->i_lock); |
f54bcf2e TH |
203 | } |
204 | ||
205 | static unsigned int | |
206 | pnfs_generic_alloc_ds_commits(struct nfs_commit_info *cinfo, | |
207 | struct list_head *list) | |
208 | { | |
209 | struct pnfs_ds_commit_info *fl_cinfo; | |
210 | struct pnfs_commit_bucket *bucket; | |
211 | struct nfs_commit_data *data; | |
212 | int i; | |
213 | unsigned int nreq = 0; | |
214 | ||
215 | fl_cinfo = cinfo->ds; | |
216 | bucket = fl_cinfo->buckets; | |
217 | for (i = 0; i < fl_cinfo->nbuckets; i++, bucket++) { | |
218 | if (list_empty(&bucket->committing)) | |
219 | continue; | |
220 | data = nfs_commitdata_alloc(); | |
221 | if (!data) | |
222 | break; | |
223 | data->ds_commit_index = i; | |
f54bcf2e TH |
224 | list_add(&data->pages, list); |
225 | nreq++; | |
226 | } | |
227 | ||
228 | /* Clean up on error */ | |
229 | pnfs_generic_retry_commit(cinfo, i); | |
230 | return nreq; | |
231 | } | |
232 | ||
27571297 TM |
233 | static inline |
234 | void pnfs_fetch_commit_bucket_list(struct list_head *pages, | |
235 | struct nfs_commit_data *data, | |
236 | struct nfs_commit_info *cinfo) | |
237 | { | |
238 | struct pnfs_commit_bucket *bucket; | |
239 | ||
240 | bucket = &cinfo->ds->buckets[data->ds_commit_index]; | |
fe238e60 | 241 | spin_lock(&cinfo->inode->i_lock); |
046be74d | 242 | list_splice_init(&bucket->committing, pages); |
27571297 TM |
243 | data->lseg = bucket->clseg; |
244 | bucket->clseg = NULL; | |
fe238e60 | 245 | spin_unlock(&cinfo->inode->i_lock); |
27571297 TM |
246 | |
247 | } | |
248 | ||
ade8febd | 249 | /* Helper function for pnfs_generic_commit_pagelist to catch an empty |
5e3a9888 WAA |
250 | * page list. This can happen when two commits race. |
251 | * | |
252 | * This must be called instead of nfs_init_commit - call one or the other, but | |
253 | * not both! | |
254 | */ | |
ade8febd WAA |
255 | static bool |
256 | pnfs_generic_commit_cancel_empty_pagelist(struct list_head *pages, | |
257 | struct nfs_commit_data *data, | |
258 | struct nfs_commit_info *cinfo) | |
259 | { | |
260 | if (list_empty(pages)) { | |
261 | if (atomic_dec_and_test(&cinfo->mds->rpcs_out)) | |
262 | wake_up_atomic_t(&cinfo->mds->rpcs_out); | |
5e3a9888 WAA |
263 | /* don't call nfs_commitdata_release - it tries to put |
264 | * the open_context which is not acquired until nfs_init_commit | |
265 | * which has not been called on @data */ | |
266 | WARN_ON_ONCE(data->context); | |
267 | nfs_commit_free(data); | |
ade8febd WAA |
268 | return true; |
269 | } | |
270 | ||
271 | return false; | |
272 | } | |
273 | ||
f54bcf2e TH |
274 | /* This follows nfs_commit_list pretty closely */ |
275 | int | |
276 | pnfs_generic_commit_pagelist(struct inode *inode, struct list_head *mds_pages, | |
277 | int how, struct nfs_commit_info *cinfo, | |
278 | int (*initiate_commit)(struct nfs_commit_data *data, | |
279 | int how)) | |
280 | { | |
281 | struct nfs_commit_data *data, *tmp; | |
282 | LIST_HEAD(list); | |
283 | unsigned int nreq = 0; | |
284 | ||
285 | if (!list_empty(mds_pages)) { | |
286 | data = nfs_commitdata_alloc(); | |
287 | if (data != NULL) { | |
27571297 | 288 | data->ds_commit_index = -1; |
f54bcf2e TH |
289 | list_add(&data->pages, &list); |
290 | nreq++; | |
291 | } else { | |
b57ff130 | 292 | nfs_retry_commit(mds_pages, NULL, cinfo, 0); |
f54bcf2e | 293 | pnfs_generic_retry_commit(cinfo, 0); |
f54bcf2e TH |
294 | return -ENOMEM; |
295 | } | |
296 | } | |
297 | ||
298 | nreq += pnfs_generic_alloc_ds_commits(cinfo, &list); | |
299 | ||
af7cf057 | 300 | if (nreq == 0) |
f54bcf2e | 301 | goto out; |
f54bcf2e TH |
302 | |
303 | atomic_add(nreq, &cinfo->mds->rpcs_out); | |
304 | ||
305 | list_for_each_entry_safe(data, tmp, &list, pages) { | |
306 | list_del_init(&data->pages); | |
27571297 | 307 | if (data->ds_commit_index < 0) { |
ade8febd WAA |
308 | /* another commit raced with us */ |
309 | if (pnfs_generic_commit_cancel_empty_pagelist(mds_pages, | |
310 | data, cinfo)) | |
311 | continue; | |
312 | ||
f54bcf2e TH |
313 | nfs_init_commit(data, mds_pages, NULL, cinfo); |
314 | nfs_initiate_commit(NFS_CLIENT(inode), data, | |
c36aae9a | 315 | NFS_PROTO(data->inode), |
f54bcf2e TH |
316 | data->mds_ops, how, 0); |
317 | } else { | |
27571297 | 318 | LIST_HEAD(pages); |
f54bcf2e | 319 | |
27571297 | 320 | pnfs_fetch_commit_bucket_list(&pages, data, cinfo); |
ade8febd WAA |
321 | |
322 | /* another commit raced with us */ | |
323 | if (pnfs_generic_commit_cancel_empty_pagelist(&pages, | |
324 | data, cinfo)) | |
325 | continue; | |
326 | ||
27571297 | 327 | nfs_init_commit(data, &pages, data->lseg, cinfo); |
f54bcf2e TH |
328 | initiate_commit(data, how); |
329 | } | |
330 | } | |
331 | out: | |
332 | cinfo->ds->ncommitting = 0; | |
333 | return PNFS_ATTEMPTED; | |
334 | } | |
335 | EXPORT_SYMBOL_GPL(pnfs_generic_commit_pagelist); | |
875ae069 PT |
336 | |
337 | /* | |
338 | * Data server cache | |
339 | * | |
340 | * Data servers can be mapped to different device ids. | |
341 | * nfs4_pnfs_ds reference counting | |
342 | * - set to 1 on allocation | |
343 | * - incremented when a device id maps a data server already in the cache. | |
344 | * - decremented when deviceid is removed from the cache. | |
345 | */ | |
346 | static DEFINE_SPINLOCK(nfs4_ds_cache_lock); | |
347 | static LIST_HEAD(nfs4_data_server_cache); | |
348 | ||
349 | /* Debug routines */ | |
350 | static void | |
351 | print_ds(struct nfs4_pnfs_ds *ds) | |
352 | { | |
353 | if (ds == NULL) { | |
354 | printk(KERN_WARNING "%s NULL device\n", __func__); | |
355 | return; | |
356 | } | |
357 | printk(KERN_WARNING " ds %s\n" | |
358 | " ref count %d\n" | |
359 | " client %p\n" | |
360 | " cl_exchange_flags %x\n", | |
361 | ds->ds_remotestr, | |
362 | atomic_read(&ds->ds_count), ds->ds_clp, | |
363 | ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0); | |
364 | } | |
365 | ||
366 | static bool | |
367 | same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2) | |
368 | { | |
369 | struct sockaddr_in *a, *b; | |
370 | struct sockaddr_in6 *a6, *b6; | |
371 | ||
372 | if (addr1->sa_family != addr2->sa_family) | |
373 | return false; | |
374 | ||
375 | switch (addr1->sa_family) { | |
376 | case AF_INET: | |
377 | a = (struct sockaddr_in *)addr1; | |
378 | b = (struct sockaddr_in *)addr2; | |
379 | ||
380 | if (a->sin_addr.s_addr == b->sin_addr.s_addr && | |
381 | a->sin_port == b->sin_port) | |
382 | return true; | |
383 | break; | |
384 | ||
385 | case AF_INET6: | |
386 | a6 = (struct sockaddr_in6 *)addr1; | |
387 | b6 = (struct sockaddr_in6 *)addr2; | |
388 | ||
389 | /* LINKLOCAL addresses must have matching scope_id */ | |
390 | if (ipv6_addr_src_scope(&a6->sin6_addr) == | |
391 | IPV6_ADDR_SCOPE_LINKLOCAL && | |
392 | a6->sin6_scope_id != b6->sin6_scope_id) | |
393 | return false; | |
394 | ||
395 | if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) && | |
396 | a6->sin6_port == b6->sin6_port) | |
397 | return true; | |
398 | break; | |
399 | ||
400 | default: | |
401 | dprintk("%s: unhandled address family: %u\n", | |
402 | __func__, addr1->sa_family); | |
403 | return false; | |
404 | } | |
405 | ||
406 | return false; | |
407 | } | |
408 | ||
6f536936 TM |
409 | /* |
410 | * Checks if 'dsaddrs1' contains a subset of 'dsaddrs2'. If it does, | |
411 | * declare a match. | |
412 | */ | |
875ae069 PT |
413 | static bool |
414 | _same_data_server_addrs_locked(const struct list_head *dsaddrs1, | |
415 | const struct list_head *dsaddrs2) | |
416 | { | |
417 | struct nfs4_pnfs_ds_addr *da1, *da2; | |
6f536936 TM |
418 | struct sockaddr *sa1, *sa2; |
419 | bool match = false; | |
420 | ||
421 | list_for_each_entry(da1, dsaddrs1, da_node) { | |
422 | sa1 = (struct sockaddr *)&da1->da_addr; | |
423 | match = false; | |
424 | list_for_each_entry(da2, dsaddrs2, da_node) { | |
425 | sa2 = (struct sockaddr *)&da2->da_addr; | |
426 | match = same_sockaddr(sa1, sa2); | |
427 | if (match) | |
428 | break; | |
429 | } | |
430 | if (!match) | |
431 | break; | |
875ae069 | 432 | } |
6f536936 | 433 | return match; |
875ae069 PT |
434 | } |
435 | ||
436 | /* | |
437 | * Lookup DS by addresses. nfs4_ds_cache_lock is held | |
438 | */ | |
439 | static struct nfs4_pnfs_ds * | |
440 | _data_server_lookup_locked(const struct list_head *dsaddrs) | |
441 | { | |
442 | struct nfs4_pnfs_ds *ds; | |
443 | ||
444 | list_for_each_entry(ds, &nfs4_data_server_cache, ds_node) | |
445 | if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs)) | |
446 | return ds; | |
447 | return NULL; | |
448 | } | |
449 | ||
450 | static void destroy_ds(struct nfs4_pnfs_ds *ds) | |
451 | { | |
452 | struct nfs4_pnfs_ds_addr *da; | |
453 | ||
454 | dprintk("--> %s\n", __func__); | |
455 | ifdebug(FACILITY) | |
456 | print_ds(ds); | |
457 | ||
458 | nfs_put_client(ds->ds_clp); | |
459 | ||
460 | while (!list_empty(&ds->ds_addrs)) { | |
461 | da = list_first_entry(&ds->ds_addrs, | |
462 | struct nfs4_pnfs_ds_addr, | |
463 | da_node); | |
464 | list_del_init(&da->da_node); | |
465 | kfree(da->da_remotestr); | |
466 | kfree(da); | |
467 | } | |
468 | ||
469 | kfree(ds->ds_remotestr); | |
470 | kfree(ds); | |
471 | } | |
472 | ||
473 | void nfs4_pnfs_ds_put(struct nfs4_pnfs_ds *ds) | |
474 | { | |
475 | if (atomic_dec_and_lock(&ds->ds_count, | |
476 | &nfs4_ds_cache_lock)) { | |
477 | list_del_init(&ds->ds_node); | |
478 | spin_unlock(&nfs4_ds_cache_lock); | |
479 | destroy_ds(ds); | |
480 | } | |
481 | } | |
482 | EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_put); | |
483 | ||
484 | /* | |
485 | * Create a string with a human readable address and port to avoid | |
486 | * complicated setup around many dprinks. | |
487 | */ | |
488 | static char * | |
489 | nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags) | |
490 | { | |
491 | struct nfs4_pnfs_ds_addr *da; | |
492 | char *remotestr; | |
493 | size_t len; | |
494 | char *p; | |
495 | ||
496 | len = 3; /* '{', '}' and eol */ | |
497 | list_for_each_entry(da, dsaddrs, da_node) { | |
498 | len += strlen(da->da_remotestr) + 1; /* string plus comma */ | |
499 | } | |
500 | ||
501 | remotestr = kzalloc(len, gfp_flags); | |
502 | if (!remotestr) | |
503 | return NULL; | |
504 | ||
505 | p = remotestr; | |
506 | *(p++) = '{'; | |
507 | len--; | |
508 | list_for_each_entry(da, dsaddrs, da_node) { | |
509 | size_t ll = strlen(da->da_remotestr); | |
510 | ||
511 | if (ll > len) | |
512 | goto out_err; | |
513 | ||
514 | memcpy(p, da->da_remotestr, ll); | |
515 | p += ll; | |
516 | len -= ll; | |
517 | ||
518 | if (len < 1) | |
519 | goto out_err; | |
520 | (*p++) = ','; | |
521 | len--; | |
522 | } | |
523 | if (len < 2) | |
524 | goto out_err; | |
525 | *(p++) = '}'; | |
526 | *p = '\0'; | |
527 | return remotestr; | |
528 | out_err: | |
529 | kfree(remotestr); | |
530 | return NULL; | |
531 | } | |
532 | ||
533 | /* | |
534 | * Given a list of multipath struct nfs4_pnfs_ds_addr, add it to ds cache if | |
535 | * uncached and return cached struct nfs4_pnfs_ds. | |
536 | */ | |
537 | struct nfs4_pnfs_ds * | |
538 | nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags) | |
539 | { | |
540 | struct nfs4_pnfs_ds *tmp_ds, *ds = NULL; | |
541 | char *remotestr; | |
542 | ||
543 | if (list_empty(dsaddrs)) { | |
544 | dprintk("%s: no addresses defined\n", __func__); | |
545 | goto out; | |
546 | } | |
547 | ||
548 | ds = kzalloc(sizeof(*ds), gfp_flags); | |
549 | if (!ds) | |
550 | goto out; | |
551 | ||
552 | /* this is only used for debugging, so it's ok if its NULL */ | |
553 | remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags); | |
554 | ||
555 | spin_lock(&nfs4_ds_cache_lock); | |
556 | tmp_ds = _data_server_lookup_locked(dsaddrs); | |
557 | if (tmp_ds == NULL) { | |
558 | INIT_LIST_HEAD(&ds->ds_addrs); | |
559 | list_splice_init(dsaddrs, &ds->ds_addrs); | |
560 | ds->ds_remotestr = remotestr; | |
561 | atomic_set(&ds->ds_count, 1); | |
562 | INIT_LIST_HEAD(&ds->ds_node); | |
563 | ds->ds_clp = NULL; | |
564 | list_add(&ds->ds_node, &nfs4_data_server_cache); | |
565 | dprintk("%s add new data server %s\n", __func__, | |
566 | ds->ds_remotestr); | |
567 | } else { | |
568 | kfree(remotestr); | |
569 | kfree(ds); | |
570 | atomic_inc(&tmp_ds->ds_count); | |
571 | dprintk("%s data server %s found, inc'ed ds_count to %d\n", | |
572 | __func__, tmp_ds->ds_remotestr, | |
573 | atomic_read(&tmp_ds->ds_count)); | |
574 | ds = tmp_ds; | |
575 | } | |
576 | spin_unlock(&nfs4_ds_cache_lock); | |
577 | out: | |
578 | return ds; | |
579 | } | |
580 | EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_add); | |
6b7f3cf9 | 581 | |
7405f9e1 PT |
582 | static void nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds) |
583 | { | |
584 | might_sleep(); | |
585 | wait_on_bit(&ds->ds_state, NFS4DS_CONNECTING, | |
586 | TASK_KILLABLE); | |
587 | } | |
588 | ||
589 | static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds) | |
590 | { | |
591 | smp_mb__before_atomic(); | |
592 | clear_bit(NFS4DS_CONNECTING, &ds->ds_state); | |
593 | smp_mb__after_atomic(); | |
594 | wake_up_bit(&ds->ds_state, NFS4DS_CONNECTING); | |
595 | } | |
596 | ||
5f01d953 PT |
597 | static struct nfs_client *(*get_v3_ds_connect)( |
598 | struct nfs_client *mds_clp, | |
599 | const struct sockaddr *ds_addr, | |
600 | int ds_addrlen, | |
601 | int ds_proto, | |
602 | unsigned int ds_timeo, | |
603 | unsigned int ds_retrans, | |
604 | rpc_authflavor_t au_flavor); | |
605 | ||
606 | static bool load_v3_ds_connect(void) | |
607 | { | |
608 | if (!get_v3_ds_connect) { | |
609 | get_v3_ds_connect = symbol_request(nfs3_set_ds_client); | |
610 | WARN_ON_ONCE(!get_v3_ds_connect); | |
611 | } | |
612 | ||
613 | return(get_v3_ds_connect != NULL); | |
614 | } | |
615 | ||
df137bc1 | 616 | void nfs4_pnfs_v3_ds_connect_unload(void) |
5f01d953 PT |
617 | { |
618 | if (get_v3_ds_connect) { | |
619 | symbol_put(nfs3_set_ds_client); | |
620 | get_v3_ds_connect = NULL; | |
621 | } | |
622 | } | |
623 | EXPORT_SYMBOL_GPL(nfs4_pnfs_v3_ds_connect_unload); | |
624 | ||
625 | static int _nfs4_pnfs_v3_ds_connect(struct nfs_server *mds_srv, | |
626 | struct nfs4_pnfs_ds *ds, | |
627 | unsigned int timeo, | |
628 | unsigned int retrans, | |
629 | rpc_authflavor_t au_flavor) | |
630 | { | |
631 | struct nfs_client *clp = ERR_PTR(-EIO); | |
632 | struct nfs4_pnfs_ds_addr *da; | |
633 | int status = 0; | |
634 | ||
635 | dprintk("--> %s DS %s au_flavor %d\n", __func__, | |
636 | ds->ds_remotestr, au_flavor); | |
637 | ||
638 | if (!load_v3_ds_connect()) | |
639 | goto out; | |
640 | ||
641 | list_for_each_entry(da, &ds->ds_addrs, da_node) { | |
642 | dprintk("%s: DS %s: trying address %s\n", | |
643 | __func__, ds->ds_remotestr, da->da_remotestr); | |
644 | ||
fc821d59 TM |
645 | if (!IS_ERR(clp)) { |
646 | struct xprt_create xprt_args = { | |
647 | .ident = XPRT_TRANSPORT_TCP, | |
648 | .net = clp->cl_net, | |
649 | .dstaddr = (struct sockaddr *)&da->da_addr, | |
650 | .addrlen = da->da_addrlen, | |
651 | .servername = clp->cl_hostname, | |
652 | }; | |
653 | /* Add this address as an alias */ | |
654 | rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args, | |
655 | rpc_clnt_test_and_add_xprt, NULL); | |
656 | } else | |
657 | clp = get_v3_ds_connect(mds_srv->nfs_client, | |
5f01d953 PT |
658 | (struct sockaddr *)&da->da_addr, |
659 | da->da_addrlen, IPPROTO_TCP, | |
660 | timeo, retrans, au_flavor); | |
5f01d953 PT |
661 | } |
662 | ||
663 | if (IS_ERR(clp)) { | |
664 | status = PTR_ERR(clp); | |
665 | goto out; | |
666 | } | |
667 | ||
668 | smp_wmb(); | |
669 | ds->ds_clp = clp; | |
670 | dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr); | |
671 | out: | |
672 | return status; | |
673 | } | |
674 | ||
675 | static int _nfs4_pnfs_v4_ds_connect(struct nfs_server *mds_srv, | |
7405f9e1 PT |
676 | struct nfs4_pnfs_ds *ds, |
677 | unsigned int timeo, | |
064172f3 | 678 | unsigned int retrans, |
30626f9c | 679 | u32 minor_version, |
064172f3 | 680 | rpc_authflavor_t au_flavor) |
7405f9e1 PT |
681 | { |
682 | struct nfs_client *clp = ERR_PTR(-EIO); | |
683 | struct nfs4_pnfs_ds_addr *da; | |
684 | int status = 0; | |
685 | ||
686 | dprintk("--> %s DS %s au_flavor %d\n", __func__, ds->ds_remotestr, | |
5f01d953 | 687 | au_flavor); |
7405f9e1 PT |
688 | |
689 | list_for_each_entry(da, &ds->ds_addrs, da_node) { | |
690 | dprintk("%s: DS %s: trying address %s\n", | |
691 | __func__, ds->ds_remotestr, da->da_remotestr); | |
692 | ||
693 | clp = nfs4_set_ds_client(mds_srv->nfs_client, | |
694 | (struct sockaddr *)&da->da_addr, | |
695 | da->da_addrlen, IPPROTO_TCP, | |
30626f9c PT |
696 | timeo, retrans, minor_version, |
697 | au_flavor); | |
7405f9e1 PT |
698 | if (!IS_ERR(clp)) |
699 | break; | |
700 | } | |
701 | ||
702 | if (IS_ERR(clp)) { | |
703 | status = PTR_ERR(clp); | |
704 | goto out; | |
705 | } | |
706 | ||
707 | status = nfs4_init_ds_session(clp, mds_srv->nfs_client->cl_lease_time); | |
708 | if (status) | |
709 | goto out_put; | |
710 | ||
711 | smp_wmb(); | |
712 | ds->ds_clp = clp; | |
713 | dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr); | |
714 | out: | |
715 | return status; | |
716 | out_put: | |
717 | nfs_put_client(clp); | |
718 | goto out; | |
719 | } | |
720 | ||
721 | /* | |
722 | * Create an rpc connection to the nfs4_pnfs_ds data server. | |
723 | * Currently only supports IPv4 and IPv6 addresses. | |
724 | * If connection fails, make devid unavailable. | |
725 | */ | |
726 | void nfs4_pnfs_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds, | |
727 | struct nfs4_deviceid_node *devid, unsigned int timeo, | |
30626f9c PT |
728 | unsigned int retrans, u32 version, |
729 | u32 minor_version, rpc_authflavor_t au_flavor) | |
7405f9e1 PT |
730 | { |
731 | if (test_and_set_bit(NFS4DS_CONNECTING, &ds->ds_state) == 0) { | |
732 | int err = 0; | |
733 | ||
5f01d953 PT |
734 | if (version == 3) { |
735 | err = _nfs4_pnfs_v3_ds_connect(mds_srv, ds, timeo, | |
736 | retrans, au_flavor); | |
737 | } else if (version == 4) { | |
738 | err = _nfs4_pnfs_v4_ds_connect(mds_srv, ds, timeo, | |
739 | retrans, minor_version, | |
740 | au_flavor); | |
741 | } else { | |
742 | dprintk("%s: unsupported DS version %d\n", __func__, | |
743 | version); | |
744 | err = -EPROTONOSUPPORT; | |
745 | } | |
746 | ||
7405f9e1 PT |
747 | if (err) |
748 | nfs4_mark_deviceid_unavailable(devid); | |
749 | nfs4_clear_ds_conn_bit(ds); | |
750 | } else { | |
751 | nfs4_wait_ds_connect(ds); | |
752 | } | |
753 | } | |
754 | EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_connect); | |
755 | ||
6b7f3cf9 PT |
756 | /* |
757 | * Currently only supports ipv4, ipv6 and one multi-path address. | |
758 | */ | |
759 | struct nfs4_pnfs_ds_addr * | |
760 | nfs4_decode_mp_ds_addr(struct net *net, struct xdr_stream *xdr, gfp_t gfp_flags) | |
761 | { | |
762 | struct nfs4_pnfs_ds_addr *da = NULL; | |
763 | char *buf, *portstr; | |
764 | __be16 port; | |
765 | int nlen, rlen; | |
766 | int tmp[2]; | |
767 | __be32 *p; | |
768 | char *netid, *match_netid; | |
769 | size_t len, match_netid_len; | |
770 | char *startsep = ""; | |
771 | char *endsep = ""; | |
772 | ||
773 | ||
774 | /* r_netid */ | |
775 | p = xdr_inline_decode(xdr, 4); | |
776 | if (unlikely(!p)) | |
777 | goto out_err; | |
778 | nlen = be32_to_cpup(p++); | |
779 | ||
780 | p = xdr_inline_decode(xdr, nlen); | |
781 | if (unlikely(!p)) | |
782 | goto out_err; | |
783 | ||
784 | netid = kmalloc(nlen+1, gfp_flags); | |
785 | if (unlikely(!netid)) | |
786 | goto out_err; | |
787 | ||
788 | netid[nlen] = '\0'; | |
789 | memcpy(netid, p, nlen); | |
790 | ||
791 | /* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */ | |
792 | p = xdr_inline_decode(xdr, 4); | |
793 | if (unlikely(!p)) | |
794 | goto out_free_netid; | |
795 | rlen = be32_to_cpup(p); | |
796 | ||
797 | p = xdr_inline_decode(xdr, rlen); | |
798 | if (unlikely(!p)) | |
799 | goto out_free_netid; | |
800 | ||
801 | /* port is ".ABC.DEF", 8 chars max */ | |
802 | if (rlen > INET6_ADDRSTRLEN + IPV6_SCOPE_ID_LEN + 8) { | |
803 | dprintk("%s: Invalid address, length %d\n", __func__, | |
804 | rlen); | |
805 | goto out_free_netid; | |
806 | } | |
807 | buf = kmalloc(rlen + 1, gfp_flags); | |
808 | if (!buf) { | |
809 | dprintk("%s: Not enough memory\n", __func__); | |
810 | goto out_free_netid; | |
811 | } | |
812 | buf[rlen] = '\0'; | |
813 | memcpy(buf, p, rlen); | |
814 | ||
815 | /* replace port '.' with '-' */ | |
816 | portstr = strrchr(buf, '.'); | |
817 | if (!portstr) { | |
818 | dprintk("%s: Failed finding expected dot in port\n", | |
819 | __func__); | |
820 | goto out_free_buf; | |
821 | } | |
822 | *portstr = '-'; | |
823 | ||
824 | /* find '.' between address and port */ | |
825 | portstr = strrchr(buf, '.'); | |
826 | if (!portstr) { | |
827 | dprintk("%s: Failed finding expected dot between address and " | |
828 | "port\n", __func__); | |
829 | goto out_free_buf; | |
830 | } | |
831 | *portstr = '\0'; | |
832 | ||
833 | da = kzalloc(sizeof(*da), gfp_flags); | |
834 | if (unlikely(!da)) | |
835 | goto out_free_buf; | |
836 | ||
837 | INIT_LIST_HEAD(&da->da_node); | |
838 | ||
839 | if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr, | |
840 | sizeof(da->da_addr))) { | |
841 | dprintk("%s: error parsing address %s\n", __func__, buf); | |
842 | goto out_free_da; | |
843 | } | |
844 | ||
845 | portstr++; | |
846 | sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]); | |
847 | port = htons((tmp[0] << 8) | (tmp[1])); | |
848 | ||
849 | switch (da->da_addr.ss_family) { | |
850 | case AF_INET: | |
851 | ((struct sockaddr_in *)&da->da_addr)->sin_port = port; | |
852 | da->da_addrlen = sizeof(struct sockaddr_in); | |
853 | match_netid = "tcp"; | |
854 | match_netid_len = 3; | |
855 | break; | |
856 | ||
857 | case AF_INET6: | |
858 | ((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port; | |
859 | da->da_addrlen = sizeof(struct sockaddr_in6); | |
860 | match_netid = "tcp6"; | |
861 | match_netid_len = 4; | |
862 | startsep = "["; | |
863 | endsep = "]"; | |
864 | break; | |
865 | ||
866 | default: | |
867 | dprintk("%s: unsupported address family: %u\n", | |
868 | __func__, da->da_addr.ss_family); | |
869 | goto out_free_da; | |
870 | } | |
871 | ||
872 | if (nlen != match_netid_len || strncmp(netid, match_netid, nlen)) { | |
873 | dprintk("%s: ERROR: r_netid \"%s\" != \"%s\"\n", | |
874 | __func__, netid, match_netid); | |
875 | goto out_free_da; | |
876 | } | |
877 | ||
878 | /* save human readable address */ | |
879 | len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7; | |
880 | da->da_remotestr = kzalloc(len, gfp_flags); | |
881 | ||
882 | /* NULL is ok, only used for dprintk */ | |
883 | if (da->da_remotestr) | |
884 | snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep, | |
885 | buf, endsep, ntohs(port)); | |
886 | ||
887 | dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr); | |
888 | kfree(buf); | |
889 | kfree(netid); | |
890 | return da; | |
891 | ||
892 | out_free_da: | |
893 | kfree(da); | |
894 | out_free_buf: | |
895 | dprintk("%s: Error parsing DS addr: %s\n", __func__, buf); | |
896 | kfree(buf); | |
897 | out_free_netid: | |
898 | kfree(netid); | |
899 | out_err: | |
900 | return NULL; | |
901 | } | |
902 | EXPORT_SYMBOL_GPL(nfs4_decode_mp_ds_addr); | |
338d00cf TH |
903 | |
904 | void | |
905 | pnfs_layout_mark_request_commit(struct nfs_page *req, | |
906 | struct pnfs_layout_segment *lseg, | |
907 | struct nfs_commit_info *cinfo, | |
908 | u32 ds_commit_idx) | |
909 | { | |
910 | struct list_head *list; | |
911 | struct pnfs_commit_bucket *buckets; | |
912 | ||
fe238e60 | 913 | spin_lock(&cinfo->inode->i_lock); |
338d00cf TH |
914 | buckets = cinfo->ds->buckets; |
915 | list = &buckets[ds_commit_idx].written; | |
916 | if (list_empty(list)) { | |
b20135d0 | 917 | if (!pnfs_is_valid_lseg(lseg)) { |
fe238e60 | 918 | spin_unlock(&cinfo->inode->i_lock); |
b20135d0 TM |
919 | cinfo->completion_ops->resched_write(cinfo, req); |
920 | return; | |
921 | } | |
338d00cf TH |
922 | /* Non-empty buckets hold a reference on the lseg. That ref |
923 | * is normally transferred to the COMMIT call and released | |
924 | * there. It could also be released if the last req is pulled | |
925 | * off due to a rewrite, in which case it will be done in | |
926 | * pnfs_common_clear_request_commit | |
927 | */ | |
928 | WARN_ON_ONCE(buckets[ds_commit_idx].wlseg != NULL); | |
929 | buckets[ds_commit_idx].wlseg = pnfs_get_lseg(lseg); | |
930 | } | |
931 | set_bit(PG_COMMIT_TO_DS, &req->wb_flags); | |
932 | cinfo->ds->nwritten++; | |
338d00cf | 933 | |
86d80f97 | 934 | nfs_request_add_commit_list_locked(req, list, cinfo); |
fe238e60 | 935 | spin_unlock(&cinfo->inode->i_lock); |
86d80f97 | 936 | nfs_mark_page_unstable(req->wb_page, cinfo); |
338d00cf TH |
937 | } |
938 | EXPORT_SYMBOL_GPL(pnfs_layout_mark_request_commit); | |
5bb89b47 TM |
939 | |
940 | int | |
941 | pnfs_nfs_generic_sync(struct inode *inode, bool datasync) | |
942 | { | |
943 | if (datasync) | |
944 | return 0; | |
945 | return pnfs_layoutcommit_inode(inode, true); | |
946 | } | |
947 | EXPORT_SYMBOL_GPL(pnfs_nfs_generic_sync); | |
948 |