]>
Commit | Line | Data |
---|---|---|
85e174ba RL |
1 | /* |
2 | * pNFS functions to call and manage layout drivers. | |
3 | * | |
4 | * Copyright (c) 2002 [year of first publication] | |
5 | * The Regents of the University of Michigan | |
6 | * All Rights Reserved | |
7 | * | |
8 | * Dean Hildebrand <dhildebz@umich.edu> | |
9 | * | |
10 | * Permission is granted to use, copy, create derivative works, and | |
11 | * redistribute this software and such derivative works for any purpose, | |
12 | * so long as the name of the University of Michigan is not used in | |
13 | * any advertising or publicity pertaining to the use or distribution | |
14 | * of this software without specific, written prior authorization. If | |
15 | * the above copyright notice or any other identification of the | |
16 | * University of Michigan is included in any copy of any portion of | |
17 | * this software, then the disclaimer below must also be included. | |
18 | * | |
19 | * This software is provided as is, without representation or warranty | |
20 | * of any kind either express or implied, including without limitation | |
21 | * the implied warranties of merchantability, fitness for a particular | |
22 | * purpose, or noninfringement. The Regents of the University of | |
23 | * Michigan shall not be liable for any damages, including special, | |
24 | * indirect, incidental, or consequential damages, with respect to any | |
25 | * claim arising out of or in connection with the use of the software, | |
26 | * even if it has been or is hereafter advised of the possibility of | |
27 | * such damages. | |
28 | */ | |
29 | ||
30 | #include <linux/nfs_fs.h> | |
974cec8c | 31 | #include "internal.h" |
85e174ba | 32 | #include "pnfs.h" |
64419a9b | 33 | #include "iostat.h" |
85e174ba RL |
34 | |
35 | #define NFSDBG_FACILITY NFSDBG_PNFS | |
36 | ||
02c35fca FI |
37 | /* Locking: |
38 | * | |
39 | * pnfs_spinlock: | |
40 | * protects pnfs_modules_tbl. | |
41 | */ | |
42 | static DEFINE_SPINLOCK(pnfs_spinlock); | |
43 | ||
44 | /* | |
45 | * pnfs_modules_tbl holds all pnfs modules | |
46 | */ | |
47 | static LIST_HEAD(pnfs_modules_tbl); | |
48 | ||
49 | /* Return the registered pnfs layout driver module matching given id */ | |
50 | static struct pnfs_layoutdriver_type * | |
51 | find_pnfs_driver_locked(u32 id) | |
52 | { | |
53 | struct pnfs_layoutdriver_type *local; | |
54 | ||
55 | list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid) | |
56 | if (local->id == id) | |
57 | goto out; | |
58 | local = NULL; | |
59 | out: | |
60 | dprintk("%s: Searching for id %u, found %p\n", __func__, id, local); | |
61 | return local; | |
62 | } | |
63 | ||
85e174ba RL |
64 | static struct pnfs_layoutdriver_type * |
65 | find_pnfs_driver(u32 id) | |
66 | { | |
02c35fca FI |
67 | struct pnfs_layoutdriver_type *local; |
68 | ||
69 | spin_lock(&pnfs_spinlock); | |
70 | local = find_pnfs_driver_locked(id); | |
71 | spin_unlock(&pnfs_spinlock); | |
72 | return local; | |
85e174ba RL |
73 | } |
74 | ||
75 | void | |
76 | unset_pnfs_layoutdriver(struct nfs_server *nfss) | |
77 | { | |
02c35fca | 78 | if (nfss->pnfs_curr_ld) { |
1c787096 | 79 | nfss->pnfs_curr_ld->clear_layoutdriver(nfss); |
02c35fca FI |
80 | module_put(nfss->pnfs_curr_ld->owner); |
81 | } | |
85e174ba RL |
82 | nfss->pnfs_curr_ld = NULL; |
83 | } | |
84 | ||
85 | /* | |
86 | * Try to set the server's pnfs module to the pnfs layout type specified by id. | |
87 | * Currently only one pNFS layout driver per filesystem is supported. | |
88 | * | |
89 | * @id layout type. Zero (illegal layout type) indicates pNFS not in use. | |
90 | */ | |
91 | void | |
92 | set_pnfs_layoutdriver(struct nfs_server *server, u32 id) | |
93 | { | |
94 | struct pnfs_layoutdriver_type *ld_type = NULL; | |
95 | ||
96 | if (id == 0) | |
97 | goto out_no_driver; | |
98 | if (!(server->nfs_client->cl_exchange_flags & | |
99 | (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) { | |
100 | printk(KERN_ERR "%s: id %u cl_exchange_flags 0x%x\n", __func__, | |
101 | id, server->nfs_client->cl_exchange_flags); | |
102 | goto out_no_driver; | |
103 | } | |
104 | ld_type = find_pnfs_driver(id); | |
105 | if (!ld_type) { | |
106 | request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id); | |
107 | ld_type = find_pnfs_driver(id); | |
108 | if (!ld_type) { | |
109 | dprintk("%s: No pNFS module found for %u.\n", | |
110 | __func__, id); | |
111 | goto out_no_driver; | |
112 | } | |
113 | } | |
02c35fca FI |
114 | if (!try_module_get(ld_type->owner)) { |
115 | dprintk("%s: Could not grab reference on module\n", __func__); | |
116 | goto out_no_driver; | |
117 | } | |
85e174ba | 118 | server->pnfs_curr_ld = ld_type; |
1c787096 | 119 | if (ld_type->set_layoutdriver(server)) { |
02c35fca FI |
120 | printk(KERN_ERR |
121 | "%s: Error initializing mount point for layout driver %u.\n", | |
122 | __func__, id); | |
123 | module_put(ld_type->owner); | |
124 | goto out_no_driver; | |
125 | } | |
85e174ba RL |
126 | dprintk("%s: pNFS module for %u set\n", __func__, id); |
127 | return; | |
128 | ||
129 | out_no_driver: | |
130 | dprintk("%s: Using NFSv4 I/O\n", __func__); | |
131 | server->pnfs_curr_ld = NULL; | |
132 | } | |
02c35fca FI |
133 | |
134 | int | |
135 | pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
136 | { | |
137 | int status = -EINVAL; | |
138 | struct pnfs_layoutdriver_type *tmp; | |
139 | ||
140 | if (ld_type->id == 0) { | |
141 | printk(KERN_ERR "%s id 0 is reserved\n", __func__); | |
142 | return status; | |
143 | } | |
b1f69b75 AA |
144 | if (!ld_type->alloc_lseg || !ld_type->free_lseg) { |
145 | printk(KERN_ERR "%s Layout driver must provide " | |
146 | "alloc_lseg and free_lseg.\n", __func__); | |
147 | return status; | |
148 | } | |
02c35fca FI |
149 | |
150 | spin_lock(&pnfs_spinlock); | |
151 | tmp = find_pnfs_driver_locked(ld_type->id); | |
152 | if (!tmp) { | |
153 | list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl); | |
154 | status = 0; | |
155 | dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id, | |
156 | ld_type->name); | |
157 | } else { | |
158 | printk(KERN_ERR "%s Module with id %d already loaded!\n", | |
159 | __func__, ld_type->id); | |
160 | } | |
161 | spin_unlock(&pnfs_spinlock); | |
162 | ||
163 | return status; | |
164 | } | |
165 | EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver); | |
166 | ||
167 | void | |
168 | pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
169 | { | |
170 | dprintk("%s Deregistering id:%u\n", __func__, ld_type->id); | |
171 | spin_lock(&pnfs_spinlock); | |
172 | list_del(&ld_type->pnfs_tblid); | |
173 | spin_unlock(&pnfs_spinlock); | |
174 | } | |
175 | EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver); | |
e5e94017 | 176 | |
b1f69b75 AA |
177 | /* |
178 | * pNFS client layout cache | |
179 | */ | |
180 | ||
cc6e5340 | 181 | /* Need to hold i_lock if caller does not already hold reference */ |
43f1b3da | 182 | void |
cc6e5340 | 183 | get_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 184 | { |
cc6e5340 | 185 | atomic_inc(&lo->plh_refcount); |
e5e94017 BH |
186 | } |
187 | ||
188 | static void | |
cc6e5340 | 189 | destroy_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 190 | { |
cc6e5340 FI |
191 | dprintk("%s: freeing layout cache %p\n", __func__, lo); |
192 | BUG_ON(!list_empty(&lo->plh_layouts)); | |
193 | NFS_I(lo->plh_inode)->layout = NULL; | |
194 | kfree(lo); | |
195 | } | |
e5e94017 | 196 | |
cc6e5340 FI |
197 | static void |
198 | put_layout_hdr_locked(struct pnfs_layout_hdr *lo) | |
199 | { | |
200 | if (atomic_dec_and_test(&lo->plh_refcount)) | |
201 | destroy_layout_hdr(lo); | |
e5e94017 BH |
202 | } |
203 | ||
b1f69b75 | 204 | void |
cc6e5340 | 205 | put_layout_hdr(struct pnfs_layout_hdr *lo) |
974cec8c | 206 | { |
cc6e5340 FI |
207 | struct inode *inode = lo->plh_inode; |
208 | ||
209 | if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) { | |
210 | destroy_layout_hdr(lo); | |
211 | spin_unlock(&inode->i_lock); | |
212 | } | |
974cec8c AA |
213 | } |
214 | ||
215 | static void | |
216 | init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg) | |
217 | { | |
566052c5 | 218 | INIT_LIST_HEAD(&lseg->pls_list); |
4541d16c FI |
219 | atomic_set(&lseg->pls_refcount, 1); |
220 | smp_mb(); | |
221 | set_bit(NFS_LSEG_VALID, &lseg->pls_flags); | |
566052c5 | 222 | lseg->pls_layout = lo; |
974cec8c AA |
223 | } |
224 | ||
4541d16c | 225 | static void free_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 226 | { |
b7edfaa1 | 227 | struct inode *ino = lseg->pls_layout->plh_inode; |
974cec8c | 228 | |
b1f69b75 | 229 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); |
52fabd73 | 230 | /* Matched by get_layout_hdr in pnfs_insert_layout */ |
cc6e5340 | 231 | put_layout_hdr(NFS_I(ino)->layout); |
974cec8c AA |
232 | } |
233 | ||
d684d2ae FI |
234 | static void |
235 | put_lseg_common(struct pnfs_layout_segment *lseg) | |
236 | { | |
237 | struct inode *inode = lseg->pls_layout->plh_inode; | |
238 | ||
239 | BUG_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
240 | list_del_init(&lseg->pls_list); | |
241 | if (list_empty(&lseg->pls_layout->plh_segs)) { | |
242 | set_bit(NFS_LAYOUT_DESTROYED, &lseg->pls_layout->plh_flags); | |
243 | /* Matched by initial refcount set in alloc_init_layout_hdr */ | |
244 | put_layout_hdr_locked(lseg->pls_layout); | |
245 | } | |
246 | rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq); | |
247 | } | |
248 | ||
bae724ef | 249 | void |
d684d2ae | 250 | put_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 251 | { |
d684d2ae FI |
252 | struct inode *inode; |
253 | ||
254 | if (!lseg) | |
255 | return; | |
256 | ||
4541d16c FI |
257 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, |
258 | atomic_read(&lseg->pls_refcount), | |
259 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
d684d2ae FI |
260 | inode = lseg->pls_layout->plh_inode; |
261 | if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) { | |
262 | LIST_HEAD(free_me); | |
4541d16c | 263 | |
d684d2ae FI |
264 | put_lseg_common(lseg); |
265 | list_add(&lseg->pls_list, &free_me); | |
266 | spin_unlock(&inode->i_lock); | |
267 | pnfs_free_lseg_list(&free_me); | |
4541d16c | 268 | } |
4541d16c | 269 | } |
974cec8c | 270 | |
4541d16c FI |
271 | static bool |
272 | should_free_lseg(u32 lseg_iomode, u32 recall_iomode) | |
273 | { | |
274 | return (recall_iomode == IOMODE_ANY || | |
275 | lseg_iomode == recall_iomode); | |
974cec8c AA |
276 | } |
277 | ||
4541d16c FI |
278 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
279 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
280 | struct list_head *tmp_list) | |
281 | { | |
282 | int rv = 0; | |
283 | ||
284 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
285 | /* Remove the reference keeping the lseg in the | |
286 | * list. It will now be removed when all | |
287 | * outstanding io is finished. | |
288 | */ | |
d684d2ae FI |
289 | dprintk("%s: lseg %p ref %d\n", __func__, lseg, |
290 | atomic_read(&lseg->pls_refcount)); | |
291 | if (atomic_dec_and_test(&lseg->pls_refcount)) { | |
292 | put_lseg_common(lseg); | |
293 | list_add(&lseg->pls_list, tmp_list); | |
294 | rv = 1; | |
295 | } | |
4541d16c FI |
296 | } |
297 | return rv; | |
298 | } | |
299 | ||
300 | /* Returns count of number of matching invalid lsegs remaining in list | |
301 | * after call. | |
302 | */ | |
43f1b3da | 303 | int |
4541d16c FI |
304 | mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, |
305 | struct list_head *tmp_list, | |
306 | u32 iomode) | |
974cec8c AA |
307 | { |
308 | struct pnfs_layout_segment *lseg, *next; | |
4541d16c | 309 | int invalid = 0, removed = 0; |
974cec8c AA |
310 | |
311 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
312 | ||
38511722 FI |
313 | if (list_empty(&lo->plh_segs)) { |
314 | if (!test_and_set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags)) | |
315 | put_layout_hdr_locked(lo); | |
316 | return 0; | |
317 | } | |
4541d16c FI |
318 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
319 | if (should_free_lseg(lseg->pls_range.iomode, iomode)) { | |
320 | dprintk("%s: freeing lseg %p iomode %d " | |
321 | "offset %llu length %llu\n", __func__, | |
322 | lseg, lseg->pls_range.iomode, lseg->pls_range.offset, | |
323 | lseg->pls_range.length); | |
324 | invalid++; | |
325 | removed += mark_lseg_invalid(lseg, tmp_list); | |
326 | } | |
327 | dprintk("%s:Return %i\n", __func__, invalid - removed); | |
328 | return invalid - removed; | |
974cec8c AA |
329 | } |
330 | ||
f49f9baa | 331 | /* note free_me must contain lsegs from a single layout_hdr */ |
43f1b3da | 332 | void |
4541d16c | 333 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 334 | { |
4541d16c | 335 | struct pnfs_layout_segment *lseg, *tmp; |
f49f9baa FI |
336 | struct pnfs_layout_hdr *lo; |
337 | ||
338 | if (list_empty(free_me)) | |
339 | return; | |
340 | ||
341 | lo = list_first_entry(free_me, struct pnfs_layout_segment, | |
342 | pls_list)->pls_layout; | |
974cec8c | 343 | |
f49f9baa FI |
344 | if (test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags)) { |
345 | struct nfs_client *clp; | |
346 | ||
347 | clp = NFS_SERVER(lo->plh_inode)->nfs_client; | |
348 | spin_lock(&clp->cl_lock); | |
349 | list_del_init(&lo->plh_layouts); | |
350 | spin_unlock(&clp->cl_lock); | |
351 | } | |
4541d16c | 352 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 353 | list_del(&lseg->pls_list); |
4541d16c | 354 | free_lseg(lseg); |
974cec8c AA |
355 | } |
356 | } | |
357 | ||
e5e94017 BH |
358 | void |
359 | pnfs_destroy_layout(struct nfs_inode *nfsi) | |
360 | { | |
361 | struct pnfs_layout_hdr *lo; | |
974cec8c | 362 | LIST_HEAD(tmp_list); |
e5e94017 BH |
363 | |
364 | spin_lock(&nfsi->vfs_inode.i_lock); | |
365 | lo = nfsi->layout; | |
366 | if (lo) { | |
38511722 | 367 | lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ |
4541d16c | 368 | mark_matching_lsegs_invalid(lo, &tmp_list, IOMODE_ANY); |
e5e94017 BH |
369 | } |
370 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
974cec8c AA |
371 | pnfs_free_lseg_list(&tmp_list); |
372 | } | |
373 | ||
374 | /* | |
375 | * Called by the state manger to remove all layouts established under an | |
376 | * expired lease. | |
377 | */ | |
378 | void | |
379 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
380 | { | |
381 | struct pnfs_layout_hdr *lo; | |
382 | LIST_HEAD(tmp_list); | |
383 | ||
384 | spin_lock(&clp->cl_lock); | |
385 | list_splice_init(&clp->cl_layouts, &tmp_list); | |
386 | spin_unlock(&clp->cl_lock); | |
387 | ||
388 | while (!list_empty(&tmp_list)) { | |
389 | lo = list_entry(tmp_list.next, struct pnfs_layout_hdr, | |
b7edfaa1 | 390 | plh_layouts); |
974cec8c | 391 | dprintk("%s freeing layout for inode %lu\n", __func__, |
b7edfaa1 FI |
392 | lo->plh_inode->i_ino); |
393 | pnfs_destroy_layout(NFS_I(lo->plh_inode)); | |
974cec8c | 394 | } |
e5e94017 BH |
395 | } |
396 | ||
fd6002e9 | 397 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
398 | void |
399 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
400 | bool update_barrier) | |
b1f69b75 | 401 | { |
fd6002e9 | 402 | u32 oldseq, newseq; |
b1f69b75 | 403 | |
fd6002e9 FI |
404 | oldseq = be32_to_cpu(lo->plh_stateid.stateid.seqid); |
405 | newseq = be32_to_cpu(new->stateid.seqid); | |
43f1b3da | 406 | if ((int)(newseq - oldseq) > 0) { |
fd6002e9 | 407 | memcpy(&lo->plh_stateid, &new->stateid, sizeof(new->stateid)); |
43f1b3da FI |
408 | if (update_barrier) { |
409 | u32 new_barrier = be32_to_cpu(new->stateid.seqid); | |
410 | ||
411 | if ((int)(new_barrier - lo->plh_barrier)) | |
412 | lo->plh_barrier = new_barrier; | |
413 | } else { | |
414 | /* Because of wraparound, we want to keep the barrier | |
415 | * "close" to the current seqids. It needs to be | |
416 | * within 2**31 to count as "behind", so if it | |
417 | * gets too near that limit, give us a litle leeway | |
418 | * and bring it to within 2**30. | |
419 | * NOTE - and yes, this is all unsigned arithmetic. | |
420 | */ | |
421 | if (unlikely((newseq - lo->plh_barrier) > (3 << 29))) | |
422 | lo->plh_barrier = newseq - (1 << 30); | |
423 | } | |
424 | } | |
b1f69b75 AA |
425 | } |
426 | ||
cf7d63f1 FI |
427 | /* lget is set to 1 if called from inside send_layoutget call chain */ |
428 | static bool | |
43f1b3da FI |
429 | pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid, |
430 | int lget) | |
431 | { | |
432 | if ((stateid) && | |
433 | (int)(lo->plh_barrier - be32_to_cpu(stateid->stateid.seqid)) >= 0) | |
434 | return true; | |
f7e8917a | 435 | return lo->plh_block_lgets || |
38511722 | 436 | test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags) || |
f7e8917a | 437 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) || |
43f1b3da | 438 | (list_empty(&lo->plh_segs) && |
cf7d63f1 FI |
439 | (atomic_read(&lo->plh_outstanding) > lget)); |
440 | } | |
441 | ||
fd6002e9 FI |
442 | int |
443 | pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, | |
444 | struct nfs4_state *open_state) | |
b1f69b75 | 445 | { |
fd6002e9 | 446 | int status = 0; |
974cec8c | 447 | |
b1f69b75 | 448 | dprintk("--> %s\n", __func__); |
fd6002e9 | 449 | spin_lock(&lo->plh_inode->i_lock); |
43f1b3da | 450 | if (pnfs_layoutgets_blocked(lo, NULL, 1)) { |
cf7d63f1 FI |
451 | status = -EAGAIN; |
452 | } else if (list_empty(&lo->plh_segs)) { | |
fd6002e9 FI |
453 | int seq; |
454 | ||
455 | do { | |
456 | seq = read_seqbegin(&open_state->seqlock); | |
457 | memcpy(dst->data, open_state->stateid.data, | |
458 | sizeof(open_state->stateid.data)); | |
459 | } while (read_seqretry(&open_state->seqlock, seq)); | |
460 | } else | |
461 | memcpy(dst->data, lo->plh_stateid.data, sizeof(lo->plh_stateid.data)); | |
462 | spin_unlock(&lo->plh_inode->i_lock); | |
b1f69b75 | 463 | dprintk("<-- %s\n", __func__); |
fd6002e9 | 464 | return status; |
b1f69b75 AA |
465 | } |
466 | ||
467 | /* | |
468 | * Get layout from server. | |
469 | * for now, assume that whole file layouts are requested. | |
470 | * arg->offset: 0 | |
471 | * arg->length: all ones | |
472 | */ | |
e5e94017 BH |
473 | static struct pnfs_layout_segment * |
474 | send_layoutget(struct pnfs_layout_hdr *lo, | |
475 | struct nfs_open_context *ctx, | |
476 | u32 iomode) | |
477 | { | |
b7edfaa1 | 478 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
479 | struct nfs_server *server = NFS_SERVER(ino); |
480 | struct nfs4_layoutget *lgp; | |
481 | struct pnfs_layout_segment *lseg = NULL; | |
482 | ||
483 | dprintk("--> %s\n", __func__); | |
e5e94017 | 484 | |
b1f69b75 AA |
485 | BUG_ON(ctx == NULL); |
486 | lgp = kzalloc(sizeof(*lgp), GFP_KERNEL); | |
cf7d63f1 | 487 | if (lgp == NULL) |
b1f69b75 | 488 | return NULL; |
b1f69b75 AA |
489 | lgp->args.minlength = NFS4_MAX_UINT64; |
490 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; | |
491 | lgp->args.range.iomode = iomode; | |
492 | lgp->args.range.offset = 0; | |
493 | lgp->args.range.length = NFS4_MAX_UINT64; | |
494 | lgp->args.type = server->pnfs_curr_ld->id; | |
495 | lgp->args.inode = ino; | |
496 | lgp->args.ctx = get_nfs_open_context(ctx); | |
497 | lgp->lsegpp = &lseg; | |
498 | ||
499 | /* Synchronously retrieve layout information from server and | |
500 | * store in lseg. | |
501 | */ | |
502 | nfs4_proc_layoutget(lgp); | |
974cec8c | 503 | if (!lseg) { |
b1f69b75 | 504 | /* remember that LAYOUTGET failed and suspend trying */ |
566052c5 | 505 | set_bit(lo_fail_bit(iomode), &lo->plh_flags); |
974cec8c | 506 | } |
974cec8c AA |
507 | return lseg; |
508 | } | |
509 | ||
f7e8917a FI |
510 | bool pnfs_roc(struct inode *ino) |
511 | { | |
512 | struct pnfs_layout_hdr *lo; | |
513 | struct pnfs_layout_segment *lseg, *tmp; | |
514 | LIST_HEAD(tmp_list); | |
515 | bool found = false; | |
516 | ||
517 | spin_lock(&ino->i_lock); | |
518 | lo = NFS_I(ino)->layout; | |
519 | if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) || | |
520 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) | |
521 | goto out_nolayout; | |
522 | list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) | |
523 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
524 | mark_lseg_invalid(lseg, &tmp_list); | |
525 | found = true; | |
526 | } | |
527 | if (!found) | |
528 | goto out_nolayout; | |
529 | lo->plh_block_lgets++; | |
530 | get_layout_hdr(lo); /* matched in pnfs_roc_release */ | |
531 | spin_unlock(&ino->i_lock); | |
532 | pnfs_free_lseg_list(&tmp_list); | |
533 | return true; | |
534 | ||
535 | out_nolayout: | |
536 | spin_unlock(&ino->i_lock); | |
537 | return false; | |
538 | } | |
539 | ||
540 | void pnfs_roc_release(struct inode *ino) | |
541 | { | |
542 | struct pnfs_layout_hdr *lo; | |
543 | ||
544 | spin_lock(&ino->i_lock); | |
545 | lo = NFS_I(ino)->layout; | |
546 | lo->plh_block_lgets--; | |
547 | put_layout_hdr_locked(lo); | |
548 | spin_unlock(&ino->i_lock); | |
549 | } | |
550 | ||
551 | void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) | |
552 | { | |
553 | struct pnfs_layout_hdr *lo; | |
554 | ||
555 | spin_lock(&ino->i_lock); | |
556 | lo = NFS_I(ino)->layout; | |
557 | if ((int)(barrier - lo->plh_barrier) > 0) | |
558 | lo->plh_barrier = barrier; | |
559 | spin_unlock(&ino->i_lock); | |
560 | } | |
561 | ||
562 | bool pnfs_roc_drain(struct inode *ino, u32 *barrier) | |
563 | { | |
564 | struct nfs_inode *nfsi = NFS_I(ino); | |
565 | struct pnfs_layout_segment *lseg; | |
566 | bool found = false; | |
567 | ||
568 | spin_lock(&ino->i_lock); | |
569 | list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list) | |
570 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
571 | found = true; | |
572 | break; | |
573 | } | |
574 | if (!found) { | |
575 | struct pnfs_layout_hdr *lo = nfsi->layout; | |
576 | u32 current_seqid = be32_to_cpu(lo->plh_stateid.stateid.seqid); | |
577 | ||
578 | /* Since close does not return a layout stateid for use as | |
579 | * a barrier, we choose the worst-case barrier. | |
580 | */ | |
581 | *barrier = current_seqid + atomic_read(&lo->plh_outstanding); | |
582 | } | |
583 | spin_unlock(&ino->i_lock); | |
584 | return found; | |
585 | } | |
586 | ||
b1f69b75 AA |
587 | /* |
588 | * Compare two layout segments for sorting into layout cache. | |
589 | * We want to preferentially return RW over RO layouts, so ensure those | |
590 | * are seen first. | |
591 | */ | |
592 | static s64 | |
593 | cmp_layout(u32 iomode1, u32 iomode2) | |
594 | { | |
595 | /* read > read/write */ | |
596 | return (int)(iomode2 == IOMODE_READ) - (int)(iomode1 == IOMODE_READ); | |
597 | } | |
598 | ||
974cec8c AA |
599 | static void |
600 | pnfs_insert_layout(struct pnfs_layout_hdr *lo, | |
601 | struct pnfs_layout_segment *lseg) | |
602 | { | |
b1f69b75 AA |
603 | struct pnfs_layout_segment *lp; |
604 | int found = 0; | |
605 | ||
974cec8c AA |
606 | dprintk("%s:Begin\n", __func__); |
607 | ||
b7edfaa1 | 608 | assert_spin_locked(&lo->plh_inode->i_lock); |
b7edfaa1 | 609 | list_for_each_entry(lp, &lo->plh_segs, pls_list) { |
566052c5 | 610 | if (cmp_layout(lp->pls_range.iomode, lseg->pls_range.iomode) > 0) |
b1f69b75 | 611 | continue; |
566052c5 | 612 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
613 | dprintk("%s: inserted lseg %p " |
614 | "iomode %d offset %llu length %llu before " | |
615 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
616 | __func__, lseg, lseg->pls_range.iomode, |
617 | lseg->pls_range.offset, lseg->pls_range.length, | |
618 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
619 | lp->pls_range.length); | |
b1f69b75 AA |
620 | found = 1; |
621 | break; | |
622 | } | |
623 | if (!found) { | |
b7edfaa1 | 624 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
b1f69b75 AA |
625 | dprintk("%s: inserted lseg %p " |
626 | "iomode %d offset %llu length %llu at tail\n", | |
566052c5 FI |
627 | __func__, lseg, lseg->pls_range.iomode, |
628 | lseg->pls_range.offset, lseg->pls_range.length); | |
974cec8c | 629 | } |
cc6e5340 | 630 | get_layout_hdr(lo); |
974cec8c AA |
631 | |
632 | dprintk("%s:Return\n", __func__); | |
e5e94017 BH |
633 | } |
634 | ||
635 | static struct pnfs_layout_hdr * | |
636 | alloc_init_layout_hdr(struct inode *ino) | |
637 | { | |
638 | struct pnfs_layout_hdr *lo; | |
639 | ||
640 | lo = kzalloc(sizeof(struct pnfs_layout_hdr), GFP_KERNEL); | |
641 | if (!lo) | |
642 | return NULL; | |
cc6e5340 | 643 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
644 | INIT_LIST_HEAD(&lo->plh_layouts); |
645 | INIT_LIST_HEAD(&lo->plh_segs); | |
43f1b3da | 646 | INIT_LIST_HEAD(&lo->plh_bulk_recall); |
b7edfaa1 | 647 | lo->plh_inode = ino; |
e5e94017 BH |
648 | return lo; |
649 | } | |
650 | ||
651 | static struct pnfs_layout_hdr * | |
652 | pnfs_find_alloc_layout(struct inode *ino) | |
653 | { | |
654 | struct nfs_inode *nfsi = NFS_I(ino); | |
655 | struct pnfs_layout_hdr *new = NULL; | |
656 | ||
657 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
658 | ||
659 | assert_spin_locked(&ino->i_lock); | |
4541d16c FI |
660 | if (nfsi->layout) { |
661 | if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags)) | |
662 | return NULL; | |
663 | else | |
664 | return nfsi->layout; | |
665 | } | |
e5e94017 BH |
666 | spin_unlock(&ino->i_lock); |
667 | new = alloc_init_layout_hdr(ino); | |
668 | spin_lock(&ino->i_lock); | |
669 | ||
670 | if (likely(nfsi->layout == NULL)) /* Won the race? */ | |
671 | nfsi->layout = new; | |
672 | else | |
673 | kfree(new); | |
674 | return nfsi->layout; | |
675 | } | |
676 | ||
b1f69b75 AA |
677 | /* |
678 | * iomode matching rules: | |
679 | * iomode lseg match | |
680 | * ----- ----- ----- | |
681 | * ANY READ true | |
682 | * ANY RW true | |
683 | * RW READ false | |
684 | * RW RW true | |
685 | * READ READ true | |
686 | * READ RW true | |
687 | */ | |
688 | static int | |
689 | is_matching_lseg(struct pnfs_layout_segment *lseg, u32 iomode) | |
690 | { | |
566052c5 | 691 | return (iomode != IOMODE_RW || lseg->pls_range.iomode == IOMODE_RW); |
b1f69b75 AA |
692 | } |
693 | ||
694 | /* | |
695 | * lookup range in layout | |
696 | */ | |
e5e94017 | 697 | static struct pnfs_layout_segment * |
43f1b3da | 698 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, u32 iomode) |
e5e94017 | 699 | { |
b1f69b75 AA |
700 | struct pnfs_layout_segment *lseg, *ret = NULL; |
701 | ||
702 | dprintk("%s:Begin\n", __func__); | |
703 | ||
b7edfaa1 FI |
704 | assert_spin_locked(&lo->plh_inode->i_lock); |
705 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { | |
4541d16c FI |
706 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
707 | is_matching_lseg(lseg, iomode)) { | |
d684d2ae | 708 | ret = get_lseg(lseg); |
b1f69b75 AA |
709 | break; |
710 | } | |
566052c5 | 711 | if (cmp_layout(iomode, lseg->pls_range.iomode) > 0) |
b1f69b75 AA |
712 | break; |
713 | } | |
714 | ||
715 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 716 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 717 | return ret; |
e5e94017 BH |
718 | } |
719 | ||
720 | /* | |
721 | * Layout segment is retreived from the server if not cached. | |
722 | * The appropriate layout segment is referenced and returned to the caller. | |
723 | */ | |
724 | struct pnfs_layout_segment * | |
725 | pnfs_update_layout(struct inode *ino, | |
726 | struct nfs_open_context *ctx, | |
727 | enum pnfs_iomode iomode) | |
728 | { | |
729 | struct nfs_inode *nfsi = NFS_I(ino); | |
2130ff66 | 730 | struct nfs_client *clp = NFS_SERVER(ino)->nfs_client; |
e5e94017 BH |
731 | struct pnfs_layout_hdr *lo; |
732 | struct pnfs_layout_segment *lseg = NULL; | |
f49f9baa | 733 | bool first = false; |
e5e94017 BH |
734 | |
735 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) | |
736 | return NULL; | |
737 | spin_lock(&ino->i_lock); | |
738 | lo = pnfs_find_alloc_layout(ino); | |
739 | if (lo == NULL) { | |
740 | dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__); | |
741 | goto out_unlock; | |
742 | } | |
743 | ||
43f1b3da FI |
744 | /* Do we even need to bother with this? */ |
745 | if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) || | |
746 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { | |
747 | dprintk("%s matches recall, use MDS\n", __func__); | |
e5e94017 BH |
748 | goto out_unlock; |
749 | } | |
43f1b3da FI |
750 | /* Check to see if the layout for the given range already exists */ |
751 | lseg = pnfs_find_lseg(lo, iomode); | |
752 | if (lseg) | |
753 | goto out_unlock; | |
e5e94017 BH |
754 | |
755 | /* if LAYOUTGET already failed once we don't try again */ | |
566052c5 | 756 | if (test_bit(lo_fail_bit(iomode), &nfsi->layout->plh_flags)) |
e5e94017 BH |
757 | goto out_unlock; |
758 | ||
43f1b3da | 759 | if (pnfs_layoutgets_blocked(lo, NULL, 0)) |
cf7d63f1 FI |
760 | goto out_unlock; |
761 | atomic_inc(&lo->plh_outstanding); | |
762 | ||
cc6e5340 | 763 | get_layout_hdr(lo); |
f49f9baa FI |
764 | if (list_empty(&lo->plh_segs)) |
765 | first = true; | |
766 | spin_unlock(&ino->i_lock); | |
767 | if (first) { | |
2130ff66 FI |
768 | /* The lo must be on the clp list if there is any |
769 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
770 | */ | |
771 | spin_lock(&clp->cl_lock); | |
772 | BUG_ON(!list_empty(&lo->plh_layouts)); | |
773 | list_add_tail(&lo->plh_layouts, &clp->cl_layouts); | |
774 | spin_unlock(&clp->cl_lock); | |
775 | } | |
e5e94017 BH |
776 | |
777 | lseg = send_layoutget(lo, ctx, iomode); | |
f49f9baa FI |
778 | if (!lseg && first) { |
779 | spin_lock(&clp->cl_lock); | |
780 | list_del_init(&lo->plh_layouts); | |
781 | spin_unlock(&clp->cl_lock); | |
2130ff66 | 782 | } |
cf7d63f1 | 783 | atomic_dec(&lo->plh_outstanding); |
cc6e5340 | 784 | put_layout_hdr(lo); |
e5e94017 BH |
785 | out: |
786 | dprintk("%s end, state 0x%lx lseg %p\n", __func__, | |
bf9c1387 | 787 | nfsi->layout ? nfsi->layout->plh_flags : -1, lseg); |
e5e94017 BH |
788 | return lseg; |
789 | out_unlock: | |
790 | spin_unlock(&ino->i_lock); | |
791 | goto out; | |
792 | } | |
b1f69b75 AA |
793 | |
794 | int | |
795 | pnfs_layout_process(struct nfs4_layoutget *lgp) | |
796 | { | |
797 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
798 | struct nfs4_layoutget_res *res = &lgp->res; | |
799 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 800 | struct inode *ino = lo->plh_inode; |
43f1b3da | 801 | struct nfs_client *clp = NFS_SERVER(ino)->nfs_client; |
b1f69b75 AA |
802 | int status = 0; |
803 | ||
fc1794c5 FI |
804 | /* Verify we got what we asked for. |
805 | * Note that because the xdr parsing only accepts a single | |
806 | * element array, this can fail even if the server is behaving | |
807 | * correctly. | |
808 | */ | |
809 | if (lgp->args.range.iomode > res->range.iomode || | |
810 | res->range.offset != 0 || | |
811 | res->range.length != NFS4_MAX_UINT64) { | |
812 | status = -EINVAL; | |
813 | goto out; | |
814 | } | |
b1f69b75 AA |
815 | /* Inject layout blob into I/O device driver */ |
816 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res); | |
817 | if (!lseg || IS_ERR(lseg)) { | |
818 | if (!lseg) | |
819 | status = -ENOMEM; | |
820 | else | |
821 | status = PTR_ERR(lseg); | |
822 | dprintk("%s: Could not allocate layout: error %d\n", | |
823 | __func__, status); | |
824 | goto out; | |
825 | } | |
826 | ||
827 | spin_lock(&ino->i_lock); | |
43f1b3da FI |
828 | if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) || |
829 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { | |
830 | dprintk("%s forget reply due to recall\n", __func__); | |
831 | goto out_forget_reply; | |
832 | } | |
833 | ||
834 | if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) { | |
835 | dprintk("%s forget reply due to state\n", __func__); | |
836 | goto out_forget_reply; | |
837 | } | |
b1f69b75 | 838 | init_lseg(lo, lseg); |
566052c5 | 839 | lseg->pls_range = res->range; |
d684d2ae | 840 | *lgp->lsegpp = get_lseg(lseg); |
b1f69b75 AA |
841 | pnfs_insert_layout(lo, lseg); |
842 | ||
f7e8917a FI |
843 | if (res->return_on_close) { |
844 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); | |
845 | set_bit(NFS_LAYOUT_ROC, &lo->plh_flags); | |
846 | } | |
847 | ||
b1f69b75 | 848 | /* Done processing layoutget. Set the layout stateid */ |
43f1b3da | 849 | pnfs_set_layout_stateid(lo, &res->stateid, false); |
b1f69b75 AA |
850 | spin_unlock(&ino->i_lock); |
851 | out: | |
852 | return status; | |
43f1b3da FI |
853 | |
854 | out_forget_reply: | |
855 | spin_unlock(&ino->i_lock); | |
856 | lseg->pls_layout = lo; | |
857 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
858 | goto out; | |
b1f69b75 AA |
859 | } |
860 | ||
bae724ef FI |
861 | static int pnfs_read_pg_test(struct nfs_pageio_descriptor *pgio, |
862 | struct nfs_page *prev, | |
863 | struct nfs_page *req) | |
94ad1c80 | 864 | { |
bae724ef FI |
865 | if (pgio->pg_count == prev->wb_bytes) { |
866 | /* This is first coelesce call for a series of nfs_pages */ | |
867 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
868 | prev->wb_context, | |
869 | IOMODE_READ); | |
870 | } | |
871 | return NFS_SERVER(pgio->pg_inode)->pnfs_curr_ld->pg_test(pgio, prev, req); | |
94ad1c80 FI |
872 | } |
873 | ||
874 | void | |
bae724ef | 875 | pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode) |
94ad1c80 | 876 | { |
bae724ef FI |
877 | struct pnfs_layoutdriver_type *ld; |
878 | ||
879 | ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
880 | pgio->pg_test = (ld && ld->pg_test) ? pnfs_read_pg_test : NULL; | |
94ad1c80 FI |
881 | } |
882 | ||
64419a9b AA |
883 | /* |
884 | * Call the appropriate parallel I/O subsystem read function. | |
885 | */ | |
886 | enum pnfs_try_status | |
887 | pnfs_try_to_read_data(struct nfs_read_data *rdata, | |
888 | const struct rpc_call_ops *call_ops) | |
889 | { | |
890 | struct inode *inode = rdata->inode; | |
891 | struct nfs_server *nfss = NFS_SERVER(inode); | |
892 | enum pnfs_try_status trypnfs; | |
893 | ||
894 | rdata->mds_ops = call_ops; | |
895 | ||
896 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
897 | __func__, inode->i_ino, rdata->args.count, rdata->args.offset); | |
898 | ||
899 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(rdata); | |
900 | if (trypnfs == PNFS_NOT_ATTEMPTED) { | |
901 | put_lseg(rdata->lseg); | |
902 | rdata->lseg = NULL; | |
903 | } else { | |
904 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); | |
905 | } | |
906 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); | |
907 | return trypnfs; | |
908 | } | |
909 | ||
b1f69b75 AA |
910 | /* |
911 | * Device ID cache. Currently supports one layout type per struct nfs_client. | |
912 | * Add layout type to the lookup key to expand to support multiple types. | |
913 | */ | |
914 | int | |
915 | pnfs_alloc_init_deviceid_cache(struct nfs_client *clp, | |
916 | void (*free_callback)(struct pnfs_deviceid_node *)) | |
917 | { | |
918 | struct pnfs_deviceid_cache *c; | |
919 | ||
920 | c = kzalloc(sizeof(struct pnfs_deviceid_cache), GFP_KERNEL); | |
921 | if (!c) | |
922 | return -ENOMEM; | |
923 | spin_lock(&clp->cl_lock); | |
924 | if (clp->cl_devid_cache != NULL) { | |
925 | atomic_inc(&clp->cl_devid_cache->dc_ref); | |
926 | dprintk("%s [kref [%d]]\n", __func__, | |
927 | atomic_read(&clp->cl_devid_cache->dc_ref)); | |
928 | kfree(c); | |
929 | } else { | |
930 | /* kzalloc initializes hlists */ | |
931 | spin_lock_init(&c->dc_lock); | |
932 | atomic_set(&c->dc_ref, 1); | |
933 | c->dc_free_callback = free_callback; | |
934 | clp->cl_devid_cache = c; | |
935 | dprintk("%s [new]\n", __func__); | |
936 | } | |
937 | spin_unlock(&clp->cl_lock); | |
938 | return 0; | |
939 | } | |
940 | EXPORT_SYMBOL_GPL(pnfs_alloc_init_deviceid_cache); | |
941 | ||
942 | /* | |
943 | * Called from pnfs_layoutdriver_type->free_lseg | |
944 | * last layout segment reference frees deviceid | |
945 | */ | |
946 | void | |
947 | pnfs_put_deviceid(struct pnfs_deviceid_cache *c, | |
948 | struct pnfs_deviceid_node *devid) | |
949 | { | |
950 | struct nfs4_deviceid *id = &devid->de_id; | |
951 | struct pnfs_deviceid_node *d; | |
952 | struct hlist_node *n; | |
953 | long h = nfs4_deviceid_hash(id); | |
954 | ||
955 | dprintk("%s [%d]\n", __func__, atomic_read(&devid->de_ref)); | |
956 | if (!atomic_dec_and_lock(&devid->de_ref, &c->dc_lock)) | |
957 | return; | |
958 | ||
959 | hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[h], de_node) | |
960 | if (!memcmp(&d->de_id, id, sizeof(*id))) { | |
961 | hlist_del_rcu(&d->de_node); | |
962 | spin_unlock(&c->dc_lock); | |
963 | synchronize_rcu(); | |
964 | c->dc_free_callback(devid); | |
965 | return; | |
966 | } | |
967 | spin_unlock(&c->dc_lock); | |
968 | /* Why wasn't it found in the list? */ | |
969 | BUG(); | |
970 | } | |
971 | EXPORT_SYMBOL_GPL(pnfs_put_deviceid); | |
972 | ||
973 | /* Find and reference a deviceid */ | |
974 | struct pnfs_deviceid_node * | |
975 | pnfs_find_get_deviceid(struct pnfs_deviceid_cache *c, struct nfs4_deviceid *id) | |
976 | { | |
977 | struct pnfs_deviceid_node *d; | |
978 | struct hlist_node *n; | |
979 | long hash = nfs4_deviceid_hash(id); | |
980 | ||
981 | dprintk("--> %s hash %ld\n", __func__, hash); | |
982 | rcu_read_lock(); | |
983 | hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[hash], de_node) { | |
984 | if (!memcmp(&d->de_id, id, sizeof(*id))) { | |
985 | if (!atomic_inc_not_zero(&d->de_ref)) { | |
986 | goto fail; | |
987 | } else { | |
988 | rcu_read_unlock(); | |
989 | return d; | |
990 | } | |
991 | } | |
992 | } | |
993 | fail: | |
994 | rcu_read_unlock(); | |
995 | return NULL; | |
996 | } | |
997 | EXPORT_SYMBOL_GPL(pnfs_find_get_deviceid); | |
998 | ||
999 | /* | |
1000 | * Add a deviceid to the cache. | |
1001 | * GETDEVICEINFOs for same deviceid can race. If deviceid is found, discard new | |
1002 | */ | |
1003 | struct pnfs_deviceid_node * | |
1004 | pnfs_add_deviceid(struct pnfs_deviceid_cache *c, struct pnfs_deviceid_node *new) | |
1005 | { | |
1006 | struct pnfs_deviceid_node *d; | |
1007 | long hash = nfs4_deviceid_hash(&new->de_id); | |
1008 | ||
1009 | dprintk("--> %s hash %ld\n", __func__, hash); | |
1010 | spin_lock(&c->dc_lock); | |
1011 | d = pnfs_find_get_deviceid(c, &new->de_id); | |
1012 | if (d) { | |
1013 | spin_unlock(&c->dc_lock); | |
1014 | dprintk("%s [discard]\n", __func__); | |
1015 | c->dc_free_callback(new); | |
1016 | return d; | |
1017 | } | |
1018 | INIT_HLIST_NODE(&new->de_node); | |
1019 | atomic_set(&new->de_ref, 1); | |
1020 | hlist_add_head_rcu(&new->de_node, &c->dc_deviceids[hash]); | |
1021 | spin_unlock(&c->dc_lock); | |
1022 | dprintk("%s [new]\n", __func__); | |
1023 | return new; | |
1024 | } | |
1025 | EXPORT_SYMBOL_GPL(pnfs_add_deviceid); | |
1026 | ||
1027 | void | |
1028 | pnfs_put_deviceid_cache(struct nfs_client *clp) | |
1029 | { | |
1030 | struct pnfs_deviceid_cache *local = clp->cl_devid_cache; | |
1031 | ||
b2a2897d | 1032 | dprintk("--> %s ({%d})\n", __func__, atomic_read(&local->dc_ref)); |
b1f69b75 AA |
1033 | if (atomic_dec_and_lock(&local->dc_ref, &clp->cl_lock)) { |
1034 | int i; | |
1035 | /* Verify cache is empty */ | |
1036 | for (i = 0; i < NFS4_DEVICE_ID_HASH_SIZE; i++) | |
1037 | BUG_ON(!hlist_empty(&local->dc_deviceids[i])); | |
1038 | clp->cl_devid_cache = NULL; | |
1039 | spin_unlock(&clp->cl_lock); | |
1040 | kfree(local); | |
1041 | } | |
1042 | } | |
1043 | EXPORT_SYMBOL_GPL(pnfs_put_deviceid_cache); |