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 WAA |
249 | /* Helper function for pnfs_generic_commit_pagelist to catch an empty |
250 | * page list. This can happen when two commits race. */ | |
251 | static bool | |
252 | pnfs_generic_commit_cancel_empty_pagelist(struct list_head *pages, | |
253 | struct nfs_commit_data *data, | |
254 | struct nfs_commit_info *cinfo) | |
255 | { | |
256 | if (list_empty(pages)) { | |
257 | if (atomic_dec_and_test(&cinfo->mds->rpcs_out)) | |
258 | wake_up_atomic_t(&cinfo->mds->rpcs_out); | |
259 | nfs_commitdata_release(data); | |
260 | return true; | |
261 | } | |
262 | ||
263 | return false; | |
264 | } | |
265 | ||
f54bcf2e TH |
266 | /* This follows nfs_commit_list pretty closely */ |
267 | int | |
268 | pnfs_generic_commit_pagelist(struct inode *inode, struct list_head *mds_pages, | |
269 | int how, struct nfs_commit_info *cinfo, | |
270 | int (*initiate_commit)(struct nfs_commit_data *data, | |
271 | int how)) | |
272 | { | |
273 | struct nfs_commit_data *data, *tmp; | |
274 | LIST_HEAD(list); | |
275 | unsigned int nreq = 0; | |
276 | ||
277 | if (!list_empty(mds_pages)) { | |
278 | data = nfs_commitdata_alloc(); | |
279 | if (data != NULL) { | |
27571297 | 280 | data->ds_commit_index = -1; |
f54bcf2e TH |
281 | list_add(&data->pages, &list); |
282 | nreq++; | |
283 | } else { | |
b57ff130 | 284 | nfs_retry_commit(mds_pages, NULL, cinfo, 0); |
f54bcf2e | 285 | pnfs_generic_retry_commit(cinfo, 0); |
f54bcf2e TH |
286 | return -ENOMEM; |
287 | } | |
288 | } | |
289 | ||
290 | nreq += pnfs_generic_alloc_ds_commits(cinfo, &list); | |
291 | ||
af7cf057 | 292 | if (nreq == 0) |
f54bcf2e | 293 | goto out; |
f54bcf2e TH |
294 | |
295 | atomic_add(nreq, &cinfo->mds->rpcs_out); | |
296 | ||
297 | list_for_each_entry_safe(data, tmp, &list, pages) { | |
298 | list_del_init(&data->pages); | |
27571297 | 299 | if (data->ds_commit_index < 0) { |
ade8febd WAA |
300 | /* another commit raced with us */ |
301 | if (pnfs_generic_commit_cancel_empty_pagelist(mds_pages, | |
302 | data, cinfo)) | |
303 | continue; | |
304 | ||
f54bcf2e TH |
305 | nfs_init_commit(data, mds_pages, NULL, cinfo); |
306 | nfs_initiate_commit(NFS_CLIENT(inode), data, | |
c36aae9a | 307 | NFS_PROTO(data->inode), |
f54bcf2e TH |
308 | data->mds_ops, how, 0); |
309 | } else { | |
27571297 | 310 | LIST_HEAD(pages); |
f54bcf2e | 311 | |
27571297 | 312 | pnfs_fetch_commit_bucket_list(&pages, data, cinfo); |
ade8febd WAA |
313 | |
314 | /* another commit raced with us */ | |
315 | if (pnfs_generic_commit_cancel_empty_pagelist(&pages, | |
316 | data, cinfo)) | |
317 | continue; | |
318 | ||
27571297 | 319 | nfs_init_commit(data, &pages, data->lseg, cinfo); |
f54bcf2e TH |
320 | initiate_commit(data, how); |
321 | } | |
322 | } | |
323 | out: | |
324 | cinfo->ds->ncommitting = 0; | |
325 | return PNFS_ATTEMPTED; | |
326 | } | |
327 | EXPORT_SYMBOL_GPL(pnfs_generic_commit_pagelist); | |
875ae069 PT |
328 | |
329 | /* | |
330 | * Data server cache | |
331 | * | |
332 | * Data servers can be mapped to different device ids. | |
333 | * nfs4_pnfs_ds reference counting | |
334 | * - set to 1 on allocation | |
335 | * - incremented when a device id maps a data server already in the cache. | |
336 | * - decremented when deviceid is removed from the cache. | |
337 | */ | |
338 | static DEFINE_SPINLOCK(nfs4_ds_cache_lock); | |
339 | static LIST_HEAD(nfs4_data_server_cache); | |
340 | ||
341 | /* Debug routines */ | |
342 | static void | |
343 | print_ds(struct nfs4_pnfs_ds *ds) | |
344 | { | |
345 | if (ds == NULL) { | |
346 | printk(KERN_WARNING "%s NULL device\n", __func__); | |
347 | return; | |
348 | } | |
349 | printk(KERN_WARNING " ds %s\n" | |
350 | " ref count %d\n" | |
351 | " client %p\n" | |
352 | " cl_exchange_flags %x\n", | |
353 | ds->ds_remotestr, | |
354 | atomic_read(&ds->ds_count), ds->ds_clp, | |
355 | ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0); | |
356 | } | |
357 | ||
358 | static bool | |
359 | same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2) | |
360 | { | |
361 | struct sockaddr_in *a, *b; | |
362 | struct sockaddr_in6 *a6, *b6; | |
363 | ||
364 | if (addr1->sa_family != addr2->sa_family) | |
365 | return false; | |
366 | ||
367 | switch (addr1->sa_family) { | |
368 | case AF_INET: | |
369 | a = (struct sockaddr_in *)addr1; | |
370 | b = (struct sockaddr_in *)addr2; | |
371 | ||
372 | if (a->sin_addr.s_addr == b->sin_addr.s_addr && | |
373 | a->sin_port == b->sin_port) | |
374 | return true; | |
375 | break; | |
376 | ||
377 | case AF_INET6: | |
378 | a6 = (struct sockaddr_in6 *)addr1; | |
379 | b6 = (struct sockaddr_in6 *)addr2; | |
380 | ||
381 | /* LINKLOCAL addresses must have matching scope_id */ | |
382 | if (ipv6_addr_src_scope(&a6->sin6_addr) == | |
383 | IPV6_ADDR_SCOPE_LINKLOCAL && | |
384 | a6->sin6_scope_id != b6->sin6_scope_id) | |
385 | return false; | |
386 | ||
387 | if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) && | |
388 | a6->sin6_port == b6->sin6_port) | |
389 | return true; | |
390 | break; | |
391 | ||
392 | default: | |
393 | dprintk("%s: unhandled address family: %u\n", | |
394 | __func__, addr1->sa_family); | |
395 | return false; | |
396 | } | |
397 | ||
398 | return false; | |
399 | } | |
400 | ||
6f536936 TM |
401 | /* |
402 | * Checks if 'dsaddrs1' contains a subset of 'dsaddrs2'. If it does, | |
403 | * declare a match. | |
404 | */ | |
875ae069 PT |
405 | static bool |
406 | _same_data_server_addrs_locked(const struct list_head *dsaddrs1, | |
407 | const struct list_head *dsaddrs2) | |
408 | { | |
409 | struct nfs4_pnfs_ds_addr *da1, *da2; | |
6f536936 TM |
410 | struct sockaddr *sa1, *sa2; |
411 | bool match = false; | |
412 | ||
413 | list_for_each_entry(da1, dsaddrs1, da_node) { | |
414 | sa1 = (struct sockaddr *)&da1->da_addr; | |
415 | match = false; | |
416 | list_for_each_entry(da2, dsaddrs2, da_node) { | |
417 | sa2 = (struct sockaddr *)&da2->da_addr; | |
418 | match = same_sockaddr(sa1, sa2); | |
419 | if (match) | |
420 | break; | |
421 | } | |
422 | if (!match) | |
423 | break; | |
875ae069 | 424 | } |
6f536936 | 425 | return match; |
875ae069 PT |
426 | } |
427 | ||
428 | /* | |
429 | * Lookup DS by addresses. nfs4_ds_cache_lock is held | |
430 | */ | |
431 | static struct nfs4_pnfs_ds * | |
432 | _data_server_lookup_locked(const struct list_head *dsaddrs) | |
433 | { | |
434 | struct nfs4_pnfs_ds *ds; | |
435 | ||
436 | list_for_each_entry(ds, &nfs4_data_server_cache, ds_node) | |
437 | if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs)) | |
438 | return ds; | |
439 | return NULL; | |
440 | } | |
441 | ||
442 | static void destroy_ds(struct nfs4_pnfs_ds *ds) | |
443 | { | |
444 | struct nfs4_pnfs_ds_addr *da; | |
445 | ||
446 | dprintk("--> %s\n", __func__); | |
447 | ifdebug(FACILITY) | |
448 | print_ds(ds); | |
449 | ||
450 | nfs_put_client(ds->ds_clp); | |
451 | ||
452 | while (!list_empty(&ds->ds_addrs)) { | |
453 | da = list_first_entry(&ds->ds_addrs, | |
454 | struct nfs4_pnfs_ds_addr, | |
455 | da_node); | |
456 | list_del_init(&da->da_node); | |
457 | kfree(da->da_remotestr); | |
458 | kfree(da); | |
459 | } | |
460 | ||
461 | kfree(ds->ds_remotestr); | |
462 | kfree(ds); | |
463 | } | |
464 | ||
465 | void nfs4_pnfs_ds_put(struct nfs4_pnfs_ds *ds) | |
466 | { | |
467 | if (atomic_dec_and_lock(&ds->ds_count, | |
468 | &nfs4_ds_cache_lock)) { | |
469 | list_del_init(&ds->ds_node); | |
470 | spin_unlock(&nfs4_ds_cache_lock); | |
471 | destroy_ds(ds); | |
472 | } | |
473 | } | |
474 | EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_put); | |
475 | ||
476 | /* | |
477 | * Create a string with a human readable address and port to avoid | |
478 | * complicated setup around many dprinks. | |
479 | */ | |
480 | static char * | |
481 | nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags) | |
482 | { | |
483 | struct nfs4_pnfs_ds_addr *da; | |
484 | char *remotestr; | |
485 | size_t len; | |
486 | char *p; | |
487 | ||
488 | len = 3; /* '{', '}' and eol */ | |
489 | list_for_each_entry(da, dsaddrs, da_node) { | |
490 | len += strlen(da->da_remotestr) + 1; /* string plus comma */ | |
491 | } | |
492 | ||
493 | remotestr = kzalloc(len, gfp_flags); | |
494 | if (!remotestr) | |
495 | return NULL; | |
496 | ||
497 | p = remotestr; | |
498 | *(p++) = '{'; | |
499 | len--; | |
500 | list_for_each_entry(da, dsaddrs, da_node) { | |
501 | size_t ll = strlen(da->da_remotestr); | |
502 | ||
503 | if (ll > len) | |
504 | goto out_err; | |
505 | ||
506 | memcpy(p, da->da_remotestr, ll); | |
507 | p += ll; | |
508 | len -= ll; | |
509 | ||
510 | if (len < 1) | |
511 | goto out_err; | |
512 | (*p++) = ','; | |
513 | len--; | |
514 | } | |
515 | if (len < 2) | |
516 | goto out_err; | |
517 | *(p++) = '}'; | |
518 | *p = '\0'; | |
519 | return remotestr; | |
520 | out_err: | |
521 | kfree(remotestr); | |
522 | return NULL; | |
523 | } | |
524 | ||
525 | /* | |
526 | * Given a list of multipath struct nfs4_pnfs_ds_addr, add it to ds cache if | |
527 | * uncached and return cached struct nfs4_pnfs_ds. | |
528 | */ | |
529 | struct nfs4_pnfs_ds * | |
530 | nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags) | |
531 | { | |
532 | struct nfs4_pnfs_ds *tmp_ds, *ds = NULL; | |
533 | char *remotestr; | |
534 | ||
535 | if (list_empty(dsaddrs)) { | |
536 | dprintk("%s: no addresses defined\n", __func__); | |
537 | goto out; | |
538 | } | |
539 | ||
540 | ds = kzalloc(sizeof(*ds), gfp_flags); | |
541 | if (!ds) | |
542 | goto out; | |
543 | ||
544 | /* this is only used for debugging, so it's ok if its NULL */ | |
545 | remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags); | |
546 | ||
547 | spin_lock(&nfs4_ds_cache_lock); | |
548 | tmp_ds = _data_server_lookup_locked(dsaddrs); | |
549 | if (tmp_ds == NULL) { | |
550 | INIT_LIST_HEAD(&ds->ds_addrs); | |
551 | list_splice_init(dsaddrs, &ds->ds_addrs); | |
552 | ds->ds_remotestr = remotestr; | |
553 | atomic_set(&ds->ds_count, 1); | |
554 | INIT_LIST_HEAD(&ds->ds_node); | |
555 | ds->ds_clp = NULL; | |
556 | list_add(&ds->ds_node, &nfs4_data_server_cache); | |
557 | dprintk("%s add new data server %s\n", __func__, | |
558 | ds->ds_remotestr); | |
559 | } else { | |
560 | kfree(remotestr); | |
561 | kfree(ds); | |
562 | atomic_inc(&tmp_ds->ds_count); | |
563 | dprintk("%s data server %s found, inc'ed ds_count to %d\n", | |
564 | __func__, tmp_ds->ds_remotestr, | |
565 | atomic_read(&tmp_ds->ds_count)); | |
566 | ds = tmp_ds; | |
567 | } | |
568 | spin_unlock(&nfs4_ds_cache_lock); | |
569 | out: | |
570 | return ds; | |
571 | } | |
572 | EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_add); | |
6b7f3cf9 | 573 | |
7405f9e1 PT |
574 | static void nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds) |
575 | { | |
576 | might_sleep(); | |
577 | wait_on_bit(&ds->ds_state, NFS4DS_CONNECTING, | |
578 | TASK_KILLABLE); | |
579 | } | |
580 | ||
581 | static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds) | |
582 | { | |
583 | smp_mb__before_atomic(); | |
584 | clear_bit(NFS4DS_CONNECTING, &ds->ds_state); | |
585 | smp_mb__after_atomic(); | |
586 | wake_up_bit(&ds->ds_state, NFS4DS_CONNECTING); | |
587 | } | |
588 | ||
5f01d953 PT |
589 | static struct nfs_client *(*get_v3_ds_connect)( |
590 | struct nfs_client *mds_clp, | |
591 | const struct sockaddr *ds_addr, | |
592 | int ds_addrlen, | |
593 | int ds_proto, | |
594 | unsigned int ds_timeo, | |
595 | unsigned int ds_retrans, | |
596 | rpc_authflavor_t au_flavor); | |
597 | ||
598 | static bool load_v3_ds_connect(void) | |
599 | { | |
600 | if (!get_v3_ds_connect) { | |
601 | get_v3_ds_connect = symbol_request(nfs3_set_ds_client); | |
602 | WARN_ON_ONCE(!get_v3_ds_connect); | |
603 | } | |
604 | ||
605 | return(get_v3_ds_connect != NULL); | |
606 | } | |
607 | ||
df137bc1 | 608 | void nfs4_pnfs_v3_ds_connect_unload(void) |
5f01d953 PT |
609 | { |
610 | if (get_v3_ds_connect) { | |
611 | symbol_put(nfs3_set_ds_client); | |
612 | get_v3_ds_connect = NULL; | |
613 | } | |
614 | } | |
615 | EXPORT_SYMBOL_GPL(nfs4_pnfs_v3_ds_connect_unload); | |
616 | ||
617 | static int _nfs4_pnfs_v3_ds_connect(struct nfs_server *mds_srv, | |
618 | struct nfs4_pnfs_ds *ds, | |
619 | unsigned int timeo, | |
620 | unsigned int retrans, | |
621 | rpc_authflavor_t au_flavor) | |
622 | { | |
623 | struct nfs_client *clp = ERR_PTR(-EIO); | |
624 | struct nfs4_pnfs_ds_addr *da; | |
625 | int status = 0; | |
626 | ||
627 | dprintk("--> %s DS %s au_flavor %d\n", __func__, | |
628 | ds->ds_remotestr, au_flavor); | |
629 | ||
630 | if (!load_v3_ds_connect()) | |
631 | goto out; | |
632 | ||
633 | list_for_each_entry(da, &ds->ds_addrs, da_node) { | |
634 | dprintk("%s: DS %s: trying address %s\n", | |
635 | __func__, ds->ds_remotestr, da->da_remotestr); | |
636 | ||
fc821d59 TM |
637 | if (!IS_ERR(clp)) { |
638 | struct xprt_create xprt_args = { | |
639 | .ident = XPRT_TRANSPORT_TCP, | |
640 | .net = clp->cl_net, | |
641 | .dstaddr = (struct sockaddr *)&da->da_addr, | |
642 | .addrlen = da->da_addrlen, | |
643 | .servername = clp->cl_hostname, | |
644 | }; | |
645 | /* Add this address as an alias */ | |
646 | rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args, | |
647 | rpc_clnt_test_and_add_xprt, NULL); | |
648 | } else | |
649 | clp = get_v3_ds_connect(mds_srv->nfs_client, | |
5f01d953 PT |
650 | (struct sockaddr *)&da->da_addr, |
651 | da->da_addrlen, IPPROTO_TCP, | |
652 | timeo, retrans, au_flavor); | |
5f01d953 PT |
653 | } |
654 | ||
655 | if (IS_ERR(clp)) { | |
656 | status = PTR_ERR(clp); | |
657 | goto out; | |
658 | } | |
659 | ||
660 | smp_wmb(); | |
661 | ds->ds_clp = clp; | |
662 | dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr); | |
663 | out: | |
664 | return status; | |
665 | } | |
666 | ||
667 | static int _nfs4_pnfs_v4_ds_connect(struct nfs_server *mds_srv, | |
7405f9e1 PT |
668 | struct nfs4_pnfs_ds *ds, |
669 | unsigned int timeo, | |
064172f3 | 670 | unsigned int retrans, |
30626f9c | 671 | u32 minor_version, |
064172f3 | 672 | rpc_authflavor_t au_flavor) |
7405f9e1 PT |
673 | { |
674 | struct nfs_client *clp = ERR_PTR(-EIO); | |
675 | struct nfs4_pnfs_ds_addr *da; | |
676 | int status = 0; | |
677 | ||
678 | dprintk("--> %s DS %s au_flavor %d\n", __func__, ds->ds_remotestr, | |
5f01d953 | 679 | au_flavor); |
7405f9e1 PT |
680 | |
681 | list_for_each_entry(da, &ds->ds_addrs, da_node) { | |
682 | dprintk("%s: DS %s: trying address %s\n", | |
683 | __func__, ds->ds_remotestr, da->da_remotestr); | |
684 | ||
685 | clp = nfs4_set_ds_client(mds_srv->nfs_client, | |
686 | (struct sockaddr *)&da->da_addr, | |
687 | da->da_addrlen, IPPROTO_TCP, | |
30626f9c PT |
688 | timeo, retrans, minor_version, |
689 | au_flavor); | |
7405f9e1 PT |
690 | if (!IS_ERR(clp)) |
691 | break; | |
692 | } | |
693 | ||
694 | if (IS_ERR(clp)) { | |
695 | status = PTR_ERR(clp); | |
696 | goto out; | |
697 | } | |
698 | ||
699 | status = nfs4_init_ds_session(clp, mds_srv->nfs_client->cl_lease_time); | |
700 | if (status) | |
701 | goto out_put; | |
702 | ||
703 | smp_wmb(); | |
704 | ds->ds_clp = clp; | |
705 | dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr); | |
706 | out: | |
707 | return status; | |
708 | out_put: | |
709 | nfs_put_client(clp); | |
710 | goto out; | |
711 | } | |
712 | ||
713 | /* | |
714 | * Create an rpc connection to the nfs4_pnfs_ds data server. | |
715 | * Currently only supports IPv4 and IPv6 addresses. | |
716 | * If connection fails, make devid unavailable. | |
717 | */ | |
718 | void nfs4_pnfs_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds, | |
719 | struct nfs4_deviceid_node *devid, unsigned int timeo, | |
30626f9c PT |
720 | unsigned int retrans, u32 version, |
721 | u32 minor_version, rpc_authflavor_t au_flavor) | |
7405f9e1 PT |
722 | { |
723 | if (test_and_set_bit(NFS4DS_CONNECTING, &ds->ds_state) == 0) { | |
724 | int err = 0; | |
725 | ||
5f01d953 PT |
726 | if (version == 3) { |
727 | err = _nfs4_pnfs_v3_ds_connect(mds_srv, ds, timeo, | |
728 | retrans, au_flavor); | |
729 | } else if (version == 4) { | |
730 | err = _nfs4_pnfs_v4_ds_connect(mds_srv, ds, timeo, | |
731 | retrans, minor_version, | |
732 | au_flavor); | |
733 | } else { | |
734 | dprintk("%s: unsupported DS version %d\n", __func__, | |
735 | version); | |
736 | err = -EPROTONOSUPPORT; | |
737 | } | |
738 | ||
7405f9e1 PT |
739 | if (err) |
740 | nfs4_mark_deviceid_unavailable(devid); | |
741 | nfs4_clear_ds_conn_bit(ds); | |
742 | } else { | |
743 | nfs4_wait_ds_connect(ds); | |
744 | } | |
745 | } | |
746 | EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_connect); | |
747 | ||
6b7f3cf9 PT |
748 | /* |
749 | * Currently only supports ipv4, ipv6 and one multi-path address. | |
750 | */ | |
751 | struct nfs4_pnfs_ds_addr * | |
752 | nfs4_decode_mp_ds_addr(struct net *net, struct xdr_stream *xdr, gfp_t gfp_flags) | |
753 | { | |
754 | struct nfs4_pnfs_ds_addr *da = NULL; | |
755 | char *buf, *portstr; | |
756 | __be16 port; | |
757 | int nlen, rlen; | |
758 | int tmp[2]; | |
759 | __be32 *p; | |
760 | char *netid, *match_netid; | |
761 | size_t len, match_netid_len; | |
762 | char *startsep = ""; | |
763 | char *endsep = ""; | |
764 | ||
765 | ||
766 | /* r_netid */ | |
767 | p = xdr_inline_decode(xdr, 4); | |
768 | if (unlikely(!p)) | |
769 | goto out_err; | |
770 | nlen = be32_to_cpup(p++); | |
771 | ||
772 | p = xdr_inline_decode(xdr, nlen); | |
773 | if (unlikely(!p)) | |
774 | goto out_err; | |
775 | ||
776 | netid = kmalloc(nlen+1, gfp_flags); | |
777 | if (unlikely(!netid)) | |
778 | goto out_err; | |
779 | ||
780 | netid[nlen] = '\0'; | |
781 | memcpy(netid, p, nlen); | |
782 | ||
783 | /* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */ | |
784 | p = xdr_inline_decode(xdr, 4); | |
785 | if (unlikely(!p)) | |
786 | goto out_free_netid; | |
787 | rlen = be32_to_cpup(p); | |
788 | ||
789 | p = xdr_inline_decode(xdr, rlen); | |
790 | if (unlikely(!p)) | |
791 | goto out_free_netid; | |
792 | ||
793 | /* port is ".ABC.DEF", 8 chars max */ | |
794 | if (rlen > INET6_ADDRSTRLEN + IPV6_SCOPE_ID_LEN + 8) { | |
795 | dprintk("%s: Invalid address, length %d\n", __func__, | |
796 | rlen); | |
797 | goto out_free_netid; | |
798 | } | |
799 | buf = kmalloc(rlen + 1, gfp_flags); | |
800 | if (!buf) { | |
801 | dprintk("%s: Not enough memory\n", __func__); | |
802 | goto out_free_netid; | |
803 | } | |
804 | buf[rlen] = '\0'; | |
805 | memcpy(buf, p, rlen); | |
806 | ||
807 | /* replace port '.' with '-' */ | |
808 | portstr = strrchr(buf, '.'); | |
809 | if (!portstr) { | |
810 | dprintk("%s: Failed finding expected dot in port\n", | |
811 | __func__); | |
812 | goto out_free_buf; | |
813 | } | |
814 | *portstr = '-'; | |
815 | ||
816 | /* find '.' between address and port */ | |
817 | portstr = strrchr(buf, '.'); | |
818 | if (!portstr) { | |
819 | dprintk("%s: Failed finding expected dot between address and " | |
820 | "port\n", __func__); | |
821 | goto out_free_buf; | |
822 | } | |
823 | *portstr = '\0'; | |
824 | ||
825 | da = kzalloc(sizeof(*da), gfp_flags); | |
826 | if (unlikely(!da)) | |
827 | goto out_free_buf; | |
828 | ||
829 | INIT_LIST_HEAD(&da->da_node); | |
830 | ||
831 | if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr, | |
832 | sizeof(da->da_addr))) { | |
833 | dprintk("%s: error parsing address %s\n", __func__, buf); | |
834 | goto out_free_da; | |
835 | } | |
836 | ||
837 | portstr++; | |
838 | sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]); | |
839 | port = htons((tmp[0] << 8) | (tmp[1])); | |
840 | ||
841 | switch (da->da_addr.ss_family) { | |
842 | case AF_INET: | |
843 | ((struct sockaddr_in *)&da->da_addr)->sin_port = port; | |
844 | da->da_addrlen = sizeof(struct sockaddr_in); | |
845 | match_netid = "tcp"; | |
846 | match_netid_len = 3; | |
847 | break; | |
848 | ||
849 | case AF_INET6: | |
850 | ((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port; | |
851 | da->da_addrlen = sizeof(struct sockaddr_in6); | |
852 | match_netid = "tcp6"; | |
853 | match_netid_len = 4; | |
854 | startsep = "["; | |
855 | endsep = "]"; | |
856 | break; | |
857 | ||
858 | default: | |
859 | dprintk("%s: unsupported address family: %u\n", | |
860 | __func__, da->da_addr.ss_family); | |
861 | goto out_free_da; | |
862 | } | |
863 | ||
864 | if (nlen != match_netid_len || strncmp(netid, match_netid, nlen)) { | |
865 | dprintk("%s: ERROR: r_netid \"%s\" != \"%s\"\n", | |
866 | __func__, netid, match_netid); | |
867 | goto out_free_da; | |
868 | } | |
869 | ||
870 | /* save human readable address */ | |
871 | len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7; | |
872 | da->da_remotestr = kzalloc(len, gfp_flags); | |
873 | ||
874 | /* NULL is ok, only used for dprintk */ | |
875 | if (da->da_remotestr) | |
876 | snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep, | |
877 | buf, endsep, ntohs(port)); | |
878 | ||
879 | dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr); | |
880 | kfree(buf); | |
881 | kfree(netid); | |
882 | return da; | |
883 | ||
884 | out_free_da: | |
885 | kfree(da); | |
886 | out_free_buf: | |
887 | dprintk("%s: Error parsing DS addr: %s\n", __func__, buf); | |
888 | kfree(buf); | |
889 | out_free_netid: | |
890 | kfree(netid); | |
891 | out_err: | |
892 | return NULL; | |
893 | } | |
894 | EXPORT_SYMBOL_GPL(nfs4_decode_mp_ds_addr); | |
338d00cf TH |
895 | |
896 | void | |
897 | pnfs_layout_mark_request_commit(struct nfs_page *req, | |
898 | struct pnfs_layout_segment *lseg, | |
899 | struct nfs_commit_info *cinfo, | |
900 | u32 ds_commit_idx) | |
901 | { | |
902 | struct list_head *list; | |
903 | struct pnfs_commit_bucket *buckets; | |
904 | ||
fe238e60 | 905 | spin_lock(&cinfo->inode->i_lock); |
338d00cf TH |
906 | buckets = cinfo->ds->buckets; |
907 | list = &buckets[ds_commit_idx].written; | |
908 | if (list_empty(list)) { | |
b20135d0 | 909 | if (!pnfs_is_valid_lseg(lseg)) { |
fe238e60 | 910 | spin_unlock(&cinfo->inode->i_lock); |
b20135d0 TM |
911 | cinfo->completion_ops->resched_write(cinfo, req); |
912 | return; | |
913 | } | |
338d00cf TH |
914 | /* Non-empty buckets hold a reference on the lseg. That ref |
915 | * is normally transferred to the COMMIT call and released | |
916 | * there. It could also be released if the last req is pulled | |
917 | * off due to a rewrite, in which case it will be done in | |
918 | * pnfs_common_clear_request_commit | |
919 | */ | |
920 | WARN_ON_ONCE(buckets[ds_commit_idx].wlseg != NULL); | |
921 | buckets[ds_commit_idx].wlseg = pnfs_get_lseg(lseg); | |
922 | } | |
923 | set_bit(PG_COMMIT_TO_DS, &req->wb_flags); | |
924 | cinfo->ds->nwritten++; | |
338d00cf | 925 | |
86d80f97 | 926 | nfs_request_add_commit_list_locked(req, list, cinfo); |
fe238e60 | 927 | spin_unlock(&cinfo->inode->i_lock); |
86d80f97 | 928 | nfs_mark_page_unstable(req->wb_page, cinfo); |
338d00cf TH |
929 | } |
930 | EXPORT_SYMBOL_GPL(pnfs_layout_mark_request_commit); | |
5bb89b47 TM |
931 | |
932 | int | |
933 | pnfs_nfs_generic_sync(struct inode *inode, bool datasync) | |
934 | { | |
935 | if (datasync) | |
936 | return 0; | |
937 | return pnfs_layoutcommit_inode(inode, true); | |
938 | } | |
939 | EXPORT_SYMBOL_GPL(pnfs_nfs_generic_sync); | |
940 |