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 | { | |
ea8eecdd | 78 | if (nfss->pnfs_curr_ld) |
02c35fca | 79 | module_put(nfss->pnfs_curr_ld->owner); |
85e174ba RL |
80 | nfss->pnfs_curr_ld = NULL; |
81 | } | |
82 | ||
83 | /* | |
84 | * Try to set the server's pnfs module to the pnfs layout type specified by id. | |
85 | * Currently only one pNFS layout driver per filesystem is supported. | |
86 | * | |
87 | * @id layout type. Zero (illegal layout type) indicates pNFS not in use. | |
88 | */ | |
89 | void | |
90 | set_pnfs_layoutdriver(struct nfs_server *server, u32 id) | |
91 | { | |
92 | struct pnfs_layoutdriver_type *ld_type = NULL; | |
93 | ||
94 | if (id == 0) | |
95 | goto out_no_driver; | |
96 | if (!(server->nfs_client->cl_exchange_flags & | |
97 | (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) { | |
98 | printk(KERN_ERR "%s: id %u cl_exchange_flags 0x%x\n", __func__, | |
99 | id, server->nfs_client->cl_exchange_flags); | |
100 | goto out_no_driver; | |
101 | } | |
102 | ld_type = find_pnfs_driver(id); | |
103 | if (!ld_type) { | |
104 | request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id); | |
105 | ld_type = find_pnfs_driver(id); | |
106 | if (!ld_type) { | |
107 | dprintk("%s: No pNFS module found for %u.\n", | |
108 | __func__, id); | |
109 | goto out_no_driver; | |
110 | } | |
111 | } | |
02c35fca FI |
112 | if (!try_module_get(ld_type->owner)) { |
113 | dprintk("%s: Could not grab reference on module\n", __func__); | |
114 | goto out_no_driver; | |
115 | } | |
85e174ba | 116 | server->pnfs_curr_ld = ld_type; |
ea8eecdd | 117 | |
85e174ba RL |
118 | dprintk("%s: pNFS module for %u set\n", __func__, id); |
119 | return; | |
120 | ||
121 | out_no_driver: | |
122 | dprintk("%s: Using NFSv4 I/O\n", __func__); | |
123 | server->pnfs_curr_ld = NULL; | |
124 | } | |
02c35fca FI |
125 | |
126 | int | |
127 | pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
128 | { | |
129 | int status = -EINVAL; | |
130 | struct pnfs_layoutdriver_type *tmp; | |
131 | ||
132 | if (ld_type->id == 0) { | |
133 | printk(KERN_ERR "%s id 0 is reserved\n", __func__); | |
134 | return status; | |
135 | } | |
b1f69b75 AA |
136 | if (!ld_type->alloc_lseg || !ld_type->free_lseg) { |
137 | printk(KERN_ERR "%s Layout driver must provide " | |
138 | "alloc_lseg and free_lseg.\n", __func__); | |
139 | return status; | |
140 | } | |
02c35fca FI |
141 | |
142 | spin_lock(&pnfs_spinlock); | |
143 | tmp = find_pnfs_driver_locked(ld_type->id); | |
144 | if (!tmp) { | |
145 | list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl); | |
146 | status = 0; | |
147 | dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id, | |
148 | ld_type->name); | |
149 | } else { | |
150 | printk(KERN_ERR "%s Module with id %d already loaded!\n", | |
151 | __func__, ld_type->id); | |
152 | } | |
153 | spin_unlock(&pnfs_spinlock); | |
154 | ||
155 | return status; | |
156 | } | |
157 | EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver); | |
158 | ||
159 | void | |
160 | pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
161 | { | |
162 | dprintk("%s Deregistering id:%u\n", __func__, ld_type->id); | |
163 | spin_lock(&pnfs_spinlock); | |
164 | list_del(&ld_type->pnfs_tblid); | |
165 | spin_unlock(&pnfs_spinlock); | |
166 | } | |
167 | EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver); | |
e5e94017 | 168 | |
b1f69b75 AA |
169 | /* |
170 | * pNFS client layout cache | |
171 | */ | |
172 | ||
cc6e5340 | 173 | /* Need to hold i_lock if caller does not already hold reference */ |
43f1b3da | 174 | void |
cc6e5340 | 175 | get_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 176 | { |
cc6e5340 | 177 | atomic_inc(&lo->plh_refcount); |
e5e94017 BH |
178 | } |
179 | ||
180 | static void | |
cc6e5340 | 181 | destroy_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 182 | { |
cc6e5340 FI |
183 | dprintk("%s: freeing layout cache %p\n", __func__, lo); |
184 | BUG_ON(!list_empty(&lo->plh_layouts)); | |
185 | NFS_I(lo->plh_inode)->layout = NULL; | |
186 | kfree(lo); | |
187 | } | |
e5e94017 | 188 | |
cc6e5340 FI |
189 | static void |
190 | put_layout_hdr_locked(struct pnfs_layout_hdr *lo) | |
191 | { | |
192 | if (atomic_dec_and_test(&lo->plh_refcount)) | |
193 | destroy_layout_hdr(lo); | |
e5e94017 BH |
194 | } |
195 | ||
b1f69b75 | 196 | void |
cc6e5340 | 197 | put_layout_hdr(struct pnfs_layout_hdr *lo) |
974cec8c | 198 | { |
cc6e5340 FI |
199 | struct inode *inode = lo->plh_inode; |
200 | ||
201 | if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) { | |
202 | destroy_layout_hdr(lo); | |
203 | spin_unlock(&inode->i_lock); | |
204 | } | |
974cec8c AA |
205 | } |
206 | ||
207 | static void | |
208 | init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg) | |
209 | { | |
566052c5 | 210 | INIT_LIST_HEAD(&lseg->pls_list); |
4541d16c FI |
211 | atomic_set(&lseg->pls_refcount, 1); |
212 | smp_mb(); | |
213 | set_bit(NFS_LSEG_VALID, &lseg->pls_flags); | |
566052c5 | 214 | lseg->pls_layout = lo; |
974cec8c AA |
215 | } |
216 | ||
4541d16c | 217 | static void free_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 218 | { |
b7edfaa1 | 219 | struct inode *ino = lseg->pls_layout->plh_inode; |
974cec8c | 220 | |
b1f69b75 | 221 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); |
52fabd73 | 222 | /* Matched by get_layout_hdr in pnfs_insert_layout */ |
cc6e5340 | 223 | put_layout_hdr(NFS_I(ino)->layout); |
974cec8c AA |
224 | } |
225 | ||
d684d2ae FI |
226 | static void |
227 | put_lseg_common(struct pnfs_layout_segment *lseg) | |
228 | { | |
229 | struct inode *inode = lseg->pls_layout->plh_inode; | |
230 | ||
231 | BUG_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
232 | list_del_init(&lseg->pls_list); | |
233 | if (list_empty(&lseg->pls_layout->plh_segs)) { | |
234 | set_bit(NFS_LAYOUT_DESTROYED, &lseg->pls_layout->plh_flags); | |
235 | /* Matched by initial refcount set in alloc_init_layout_hdr */ | |
236 | put_layout_hdr_locked(lseg->pls_layout); | |
237 | } | |
238 | rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq); | |
239 | } | |
240 | ||
bae724ef | 241 | void |
d684d2ae | 242 | put_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 243 | { |
d684d2ae FI |
244 | struct inode *inode; |
245 | ||
246 | if (!lseg) | |
247 | return; | |
248 | ||
4541d16c FI |
249 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, |
250 | atomic_read(&lseg->pls_refcount), | |
251 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
d684d2ae FI |
252 | inode = lseg->pls_layout->plh_inode; |
253 | if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) { | |
254 | LIST_HEAD(free_me); | |
4541d16c | 255 | |
d684d2ae FI |
256 | put_lseg_common(lseg); |
257 | list_add(&lseg->pls_list, &free_me); | |
258 | spin_unlock(&inode->i_lock); | |
259 | pnfs_free_lseg_list(&free_me); | |
4541d16c | 260 | } |
4541d16c | 261 | } |
e0c2b380 | 262 | EXPORT_SYMBOL_GPL(put_lseg); |
974cec8c | 263 | |
4541d16c FI |
264 | static bool |
265 | should_free_lseg(u32 lseg_iomode, u32 recall_iomode) | |
266 | { | |
267 | return (recall_iomode == IOMODE_ANY || | |
268 | lseg_iomode == recall_iomode); | |
974cec8c AA |
269 | } |
270 | ||
4541d16c FI |
271 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
272 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
273 | struct list_head *tmp_list) | |
274 | { | |
275 | int rv = 0; | |
276 | ||
277 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
278 | /* Remove the reference keeping the lseg in the | |
279 | * list. It will now be removed when all | |
280 | * outstanding io is finished. | |
281 | */ | |
d684d2ae FI |
282 | dprintk("%s: lseg %p ref %d\n", __func__, lseg, |
283 | atomic_read(&lseg->pls_refcount)); | |
284 | if (atomic_dec_and_test(&lseg->pls_refcount)) { | |
285 | put_lseg_common(lseg); | |
286 | list_add(&lseg->pls_list, tmp_list); | |
287 | rv = 1; | |
288 | } | |
4541d16c FI |
289 | } |
290 | return rv; | |
291 | } | |
292 | ||
293 | /* Returns count of number of matching invalid lsegs remaining in list | |
294 | * after call. | |
295 | */ | |
43f1b3da | 296 | int |
4541d16c FI |
297 | mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, |
298 | struct list_head *tmp_list, | |
299 | u32 iomode) | |
974cec8c AA |
300 | { |
301 | struct pnfs_layout_segment *lseg, *next; | |
4541d16c | 302 | int invalid = 0, removed = 0; |
974cec8c AA |
303 | |
304 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
305 | ||
38511722 FI |
306 | if (list_empty(&lo->plh_segs)) { |
307 | if (!test_and_set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags)) | |
308 | put_layout_hdr_locked(lo); | |
309 | return 0; | |
310 | } | |
4541d16c FI |
311 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
312 | if (should_free_lseg(lseg->pls_range.iomode, iomode)) { | |
313 | dprintk("%s: freeing lseg %p iomode %d " | |
314 | "offset %llu length %llu\n", __func__, | |
315 | lseg, lseg->pls_range.iomode, lseg->pls_range.offset, | |
316 | lseg->pls_range.length); | |
317 | invalid++; | |
318 | removed += mark_lseg_invalid(lseg, tmp_list); | |
319 | } | |
320 | dprintk("%s:Return %i\n", __func__, invalid - removed); | |
321 | return invalid - removed; | |
974cec8c AA |
322 | } |
323 | ||
f49f9baa | 324 | /* note free_me must contain lsegs from a single layout_hdr */ |
43f1b3da | 325 | void |
4541d16c | 326 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 327 | { |
4541d16c | 328 | struct pnfs_layout_segment *lseg, *tmp; |
f49f9baa FI |
329 | struct pnfs_layout_hdr *lo; |
330 | ||
331 | if (list_empty(free_me)) | |
332 | return; | |
333 | ||
334 | lo = list_first_entry(free_me, struct pnfs_layout_segment, | |
335 | pls_list)->pls_layout; | |
974cec8c | 336 | |
f49f9baa FI |
337 | if (test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags)) { |
338 | struct nfs_client *clp; | |
339 | ||
340 | clp = NFS_SERVER(lo->plh_inode)->nfs_client; | |
341 | spin_lock(&clp->cl_lock); | |
342 | list_del_init(&lo->plh_layouts); | |
343 | spin_unlock(&clp->cl_lock); | |
344 | } | |
4541d16c | 345 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 346 | list_del(&lseg->pls_list); |
4541d16c | 347 | free_lseg(lseg); |
974cec8c AA |
348 | } |
349 | } | |
350 | ||
e5e94017 BH |
351 | void |
352 | pnfs_destroy_layout(struct nfs_inode *nfsi) | |
353 | { | |
354 | struct pnfs_layout_hdr *lo; | |
974cec8c | 355 | LIST_HEAD(tmp_list); |
e5e94017 BH |
356 | |
357 | spin_lock(&nfsi->vfs_inode.i_lock); | |
358 | lo = nfsi->layout; | |
359 | if (lo) { | |
38511722 | 360 | lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ |
4541d16c | 361 | mark_matching_lsegs_invalid(lo, &tmp_list, IOMODE_ANY); |
e5e94017 BH |
362 | } |
363 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
974cec8c AA |
364 | pnfs_free_lseg_list(&tmp_list); |
365 | } | |
366 | ||
367 | /* | |
368 | * Called by the state manger to remove all layouts established under an | |
369 | * expired lease. | |
370 | */ | |
371 | void | |
372 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
373 | { | |
374 | struct pnfs_layout_hdr *lo; | |
375 | LIST_HEAD(tmp_list); | |
376 | ||
377 | spin_lock(&clp->cl_lock); | |
378 | list_splice_init(&clp->cl_layouts, &tmp_list); | |
379 | spin_unlock(&clp->cl_lock); | |
380 | ||
381 | while (!list_empty(&tmp_list)) { | |
382 | lo = list_entry(tmp_list.next, struct pnfs_layout_hdr, | |
b7edfaa1 | 383 | plh_layouts); |
974cec8c | 384 | dprintk("%s freeing layout for inode %lu\n", __func__, |
b7edfaa1 FI |
385 | lo->plh_inode->i_ino); |
386 | pnfs_destroy_layout(NFS_I(lo->plh_inode)); | |
974cec8c | 387 | } |
e5e94017 BH |
388 | } |
389 | ||
fd6002e9 | 390 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
391 | void |
392 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
393 | bool update_barrier) | |
b1f69b75 | 394 | { |
fd6002e9 | 395 | u32 oldseq, newseq; |
b1f69b75 | 396 | |
fd6002e9 FI |
397 | oldseq = be32_to_cpu(lo->plh_stateid.stateid.seqid); |
398 | newseq = be32_to_cpu(new->stateid.seqid); | |
43f1b3da | 399 | if ((int)(newseq - oldseq) > 0) { |
fd6002e9 | 400 | memcpy(&lo->plh_stateid, &new->stateid, sizeof(new->stateid)); |
43f1b3da FI |
401 | if (update_barrier) { |
402 | u32 new_barrier = be32_to_cpu(new->stateid.seqid); | |
403 | ||
404 | if ((int)(new_barrier - lo->plh_barrier)) | |
405 | lo->plh_barrier = new_barrier; | |
406 | } else { | |
407 | /* Because of wraparound, we want to keep the barrier | |
408 | * "close" to the current seqids. It needs to be | |
409 | * within 2**31 to count as "behind", so if it | |
410 | * gets too near that limit, give us a litle leeway | |
411 | * and bring it to within 2**30. | |
412 | * NOTE - and yes, this is all unsigned arithmetic. | |
413 | */ | |
414 | if (unlikely((newseq - lo->plh_barrier) > (3 << 29))) | |
415 | lo->plh_barrier = newseq - (1 << 30); | |
416 | } | |
417 | } | |
b1f69b75 AA |
418 | } |
419 | ||
cf7d63f1 FI |
420 | /* lget is set to 1 if called from inside send_layoutget call chain */ |
421 | static bool | |
43f1b3da FI |
422 | pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid, |
423 | int lget) | |
424 | { | |
425 | if ((stateid) && | |
426 | (int)(lo->plh_barrier - be32_to_cpu(stateid->stateid.seqid)) >= 0) | |
427 | return true; | |
f7e8917a | 428 | return lo->plh_block_lgets || |
38511722 | 429 | test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags) || |
f7e8917a | 430 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) || |
43f1b3da | 431 | (list_empty(&lo->plh_segs) && |
cf7d63f1 FI |
432 | (atomic_read(&lo->plh_outstanding) > lget)); |
433 | } | |
434 | ||
fd6002e9 FI |
435 | int |
436 | pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, | |
437 | struct nfs4_state *open_state) | |
b1f69b75 | 438 | { |
fd6002e9 | 439 | int status = 0; |
974cec8c | 440 | |
b1f69b75 | 441 | dprintk("--> %s\n", __func__); |
fd6002e9 | 442 | spin_lock(&lo->plh_inode->i_lock); |
43f1b3da | 443 | if (pnfs_layoutgets_blocked(lo, NULL, 1)) { |
cf7d63f1 FI |
444 | status = -EAGAIN; |
445 | } else if (list_empty(&lo->plh_segs)) { | |
fd6002e9 FI |
446 | int seq; |
447 | ||
448 | do { | |
449 | seq = read_seqbegin(&open_state->seqlock); | |
450 | memcpy(dst->data, open_state->stateid.data, | |
451 | sizeof(open_state->stateid.data)); | |
452 | } while (read_seqretry(&open_state->seqlock, seq)); | |
453 | } else | |
454 | memcpy(dst->data, lo->plh_stateid.data, sizeof(lo->plh_stateid.data)); | |
455 | spin_unlock(&lo->plh_inode->i_lock); | |
b1f69b75 | 456 | dprintk("<-- %s\n", __func__); |
fd6002e9 | 457 | return status; |
b1f69b75 AA |
458 | } |
459 | ||
460 | /* | |
461 | * Get layout from server. | |
462 | * for now, assume that whole file layouts are requested. | |
463 | * arg->offset: 0 | |
464 | * arg->length: all ones | |
465 | */ | |
e5e94017 BH |
466 | static struct pnfs_layout_segment * |
467 | send_layoutget(struct pnfs_layout_hdr *lo, | |
468 | struct nfs_open_context *ctx, | |
469 | u32 iomode) | |
470 | { | |
b7edfaa1 | 471 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
472 | struct nfs_server *server = NFS_SERVER(ino); |
473 | struct nfs4_layoutget *lgp; | |
474 | struct pnfs_layout_segment *lseg = NULL; | |
475 | ||
476 | dprintk("--> %s\n", __func__); | |
e5e94017 | 477 | |
b1f69b75 AA |
478 | BUG_ON(ctx == NULL); |
479 | lgp = kzalloc(sizeof(*lgp), GFP_KERNEL); | |
cf7d63f1 | 480 | if (lgp == NULL) |
b1f69b75 | 481 | return NULL; |
b1f69b75 AA |
482 | lgp->args.minlength = NFS4_MAX_UINT64; |
483 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; | |
484 | lgp->args.range.iomode = iomode; | |
485 | lgp->args.range.offset = 0; | |
486 | lgp->args.range.length = NFS4_MAX_UINT64; | |
487 | lgp->args.type = server->pnfs_curr_ld->id; | |
488 | lgp->args.inode = ino; | |
489 | lgp->args.ctx = get_nfs_open_context(ctx); | |
490 | lgp->lsegpp = &lseg; | |
491 | ||
492 | /* Synchronously retrieve layout information from server and | |
493 | * store in lseg. | |
494 | */ | |
495 | nfs4_proc_layoutget(lgp); | |
974cec8c | 496 | if (!lseg) { |
b1f69b75 | 497 | /* remember that LAYOUTGET failed and suspend trying */ |
566052c5 | 498 | set_bit(lo_fail_bit(iomode), &lo->plh_flags); |
974cec8c | 499 | } |
974cec8c AA |
500 | return lseg; |
501 | } | |
502 | ||
f7e8917a FI |
503 | bool pnfs_roc(struct inode *ino) |
504 | { | |
505 | struct pnfs_layout_hdr *lo; | |
506 | struct pnfs_layout_segment *lseg, *tmp; | |
507 | LIST_HEAD(tmp_list); | |
508 | bool found = false; | |
509 | ||
510 | spin_lock(&ino->i_lock); | |
511 | lo = NFS_I(ino)->layout; | |
512 | if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) || | |
513 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) | |
514 | goto out_nolayout; | |
515 | list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) | |
516 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
517 | mark_lseg_invalid(lseg, &tmp_list); | |
518 | found = true; | |
519 | } | |
520 | if (!found) | |
521 | goto out_nolayout; | |
522 | lo->plh_block_lgets++; | |
523 | get_layout_hdr(lo); /* matched in pnfs_roc_release */ | |
524 | spin_unlock(&ino->i_lock); | |
525 | pnfs_free_lseg_list(&tmp_list); | |
526 | return true; | |
527 | ||
528 | out_nolayout: | |
529 | spin_unlock(&ino->i_lock); | |
530 | return false; | |
531 | } | |
532 | ||
533 | void pnfs_roc_release(struct inode *ino) | |
534 | { | |
535 | struct pnfs_layout_hdr *lo; | |
536 | ||
537 | spin_lock(&ino->i_lock); | |
538 | lo = NFS_I(ino)->layout; | |
539 | lo->plh_block_lgets--; | |
540 | put_layout_hdr_locked(lo); | |
541 | spin_unlock(&ino->i_lock); | |
542 | } | |
543 | ||
544 | void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) | |
545 | { | |
546 | struct pnfs_layout_hdr *lo; | |
547 | ||
548 | spin_lock(&ino->i_lock); | |
549 | lo = NFS_I(ino)->layout; | |
550 | if ((int)(barrier - lo->plh_barrier) > 0) | |
551 | lo->plh_barrier = barrier; | |
552 | spin_unlock(&ino->i_lock); | |
553 | } | |
554 | ||
555 | bool pnfs_roc_drain(struct inode *ino, u32 *barrier) | |
556 | { | |
557 | struct nfs_inode *nfsi = NFS_I(ino); | |
558 | struct pnfs_layout_segment *lseg; | |
559 | bool found = false; | |
560 | ||
561 | spin_lock(&ino->i_lock); | |
562 | list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list) | |
563 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
564 | found = true; | |
565 | break; | |
566 | } | |
567 | if (!found) { | |
568 | struct pnfs_layout_hdr *lo = nfsi->layout; | |
569 | u32 current_seqid = be32_to_cpu(lo->plh_stateid.stateid.seqid); | |
570 | ||
571 | /* Since close does not return a layout stateid for use as | |
572 | * a barrier, we choose the worst-case barrier. | |
573 | */ | |
574 | *barrier = current_seqid + atomic_read(&lo->plh_outstanding); | |
575 | } | |
576 | spin_unlock(&ino->i_lock); | |
577 | return found; | |
578 | } | |
579 | ||
b1f69b75 AA |
580 | /* |
581 | * Compare two layout segments for sorting into layout cache. | |
582 | * We want to preferentially return RW over RO layouts, so ensure those | |
583 | * are seen first. | |
584 | */ | |
585 | static s64 | |
586 | cmp_layout(u32 iomode1, u32 iomode2) | |
587 | { | |
588 | /* read > read/write */ | |
589 | return (int)(iomode2 == IOMODE_READ) - (int)(iomode1 == IOMODE_READ); | |
590 | } | |
591 | ||
974cec8c AA |
592 | static void |
593 | pnfs_insert_layout(struct pnfs_layout_hdr *lo, | |
594 | struct pnfs_layout_segment *lseg) | |
595 | { | |
b1f69b75 AA |
596 | struct pnfs_layout_segment *lp; |
597 | int found = 0; | |
598 | ||
974cec8c AA |
599 | dprintk("%s:Begin\n", __func__); |
600 | ||
b7edfaa1 | 601 | assert_spin_locked(&lo->plh_inode->i_lock); |
b7edfaa1 | 602 | list_for_each_entry(lp, &lo->plh_segs, pls_list) { |
566052c5 | 603 | if (cmp_layout(lp->pls_range.iomode, lseg->pls_range.iomode) > 0) |
b1f69b75 | 604 | continue; |
566052c5 | 605 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
606 | dprintk("%s: inserted lseg %p " |
607 | "iomode %d offset %llu length %llu before " | |
608 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
609 | __func__, lseg, lseg->pls_range.iomode, |
610 | lseg->pls_range.offset, lseg->pls_range.length, | |
611 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
612 | lp->pls_range.length); | |
b1f69b75 AA |
613 | found = 1; |
614 | break; | |
615 | } | |
616 | if (!found) { | |
b7edfaa1 | 617 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
b1f69b75 AA |
618 | dprintk("%s: inserted lseg %p " |
619 | "iomode %d offset %llu length %llu at tail\n", | |
566052c5 FI |
620 | __func__, lseg, lseg->pls_range.iomode, |
621 | lseg->pls_range.offset, lseg->pls_range.length); | |
974cec8c | 622 | } |
cc6e5340 | 623 | get_layout_hdr(lo); |
974cec8c AA |
624 | |
625 | dprintk("%s:Return\n", __func__); | |
e5e94017 BH |
626 | } |
627 | ||
628 | static struct pnfs_layout_hdr * | |
629 | alloc_init_layout_hdr(struct inode *ino) | |
630 | { | |
631 | struct pnfs_layout_hdr *lo; | |
632 | ||
633 | lo = kzalloc(sizeof(struct pnfs_layout_hdr), GFP_KERNEL); | |
634 | if (!lo) | |
635 | return NULL; | |
cc6e5340 | 636 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
637 | INIT_LIST_HEAD(&lo->plh_layouts); |
638 | INIT_LIST_HEAD(&lo->plh_segs); | |
43f1b3da | 639 | INIT_LIST_HEAD(&lo->plh_bulk_recall); |
b7edfaa1 | 640 | lo->plh_inode = ino; |
e5e94017 BH |
641 | return lo; |
642 | } | |
643 | ||
644 | static struct pnfs_layout_hdr * | |
645 | pnfs_find_alloc_layout(struct inode *ino) | |
646 | { | |
647 | struct nfs_inode *nfsi = NFS_I(ino); | |
648 | struct pnfs_layout_hdr *new = NULL; | |
649 | ||
650 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
651 | ||
652 | assert_spin_locked(&ino->i_lock); | |
4541d16c FI |
653 | if (nfsi->layout) { |
654 | if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags)) | |
655 | return NULL; | |
656 | else | |
657 | return nfsi->layout; | |
658 | } | |
e5e94017 BH |
659 | spin_unlock(&ino->i_lock); |
660 | new = alloc_init_layout_hdr(ino); | |
661 | spin_lock(&ino->i_lock); | |
662 | ||
663 | if (likely(nfsi->layout == NULL)) /* Won the race? */ | |
664 | nfsi->layout = new; | |
665 | else | |
666 | kfree(new); | |
667 | return nfsi->layout; | |
668 | } | |
669 | ||
b1f69b75 AA |
670 | /* |
671 | * iomode matching rules: | |
672 | * iomode lseg match | |
673 | * ----- ----- ----- | |
674 | * ANY READ true | |
675 | * ANY RW true | |
676 | * RW READ false | |
677 | * RW RW true | |
678 | * READ READ true | |
679 | * READ RW true | |
680 | */ | |
681 | static int | |
682 | is_matching_lseg(struct pnfs_layout_segment *lseg, u32 iomode) | |
683 | { | |
566052c5 | 684 | return (iomode != IOMODE_RW || lseg->pls_range.iomode == IOMODE_RW); |
b1f69b75 AA |
685 | } |
686 | ||
687 | /* | |
688 | * lookup range in layout | |
689 | */ | |
e5e94017 | 690 | static struct pnfs_layout_segment * |
43f1b3da | 691 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, u32 iomode) |
e5e94017 | 692 | { |
b1f69b75 AA |
693 | struct pnfs_layout_segment *lseg, *ret = NULL; |
694 | ||
695 | dprintk("%s:Begin\n", __func__); | |
696 | ||
b7edfaa1 FI |
697 | assert_spin_locked(&lo->plh_inode->i_lock); |
698 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { | |
4541d16c FI |
699 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
700 | is_matching_lseg(lseg, iomode)) { | |
d684d2ae | 701 | ret = get_lseg(lseg); |
b1f69b75 AA |
702 | break; |
703 | } | |
566052c5 | 704 | if (cmp_layout(iomode, lseg->pls_range.iomode) > 0) |
b1f69b75 AA |
705 | break; |
706 | } | |
707 | ||
708 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 709 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 710 | return ret; |
e5e94017 BH |
711 | } |
712 | ||
713 | /* | |
714 | * Layout segment is retreived from the server if not cached. | |
715 | * The appropriate layout segment is referenced and returned to the caller. | |
716 | */ | |
717 | struct pnfs_layout_segment * | |
718 | pnfs_update_layout(struct inode *ino, | |
719 | struct nfs_open_context *ctx, | |
720 | enum pnfs_iomode iomode) | |
721 | { | |
722 | struct nfs_inode *nfsi = NFS_I(ino); | |
2130ff66 | 723 | struct nfs_client *clp = NFS_SERVER(ino)->nfs_client; |
e5e94017 BH |
724 | struct pnfs_layout_hdr *lo; |
725 | struct pnfs_layout_segment *lseg = NULL; | |
f49f9baa | 726 | bool first = false; |
e5e94017 BH |
727 | |
728 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) | |
729 | return NULL; | |
730 | spin_lock(&ino->i_lock); | |
731 | lo = pnfs_find_alloc_layout(ino); | |
732 | if (lo == NULL) { | |
733 | dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__); | |
734 | goto out_unlock; | |
735 | } | |
736 | ||
43f1b3da FI |
737 | /* Do we even need to bother with this? */ |
738 | if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) || | |
739 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { | |
740 | dprintk("%s matches recall, use MDS\n", __func__); | |
e5e94017 BH |
741 | goto out_unlock; |
742 | } | |
743 | ||
744 | /* if LAYOUTGET already failed once we don't try again */ | |
566052c5 | 745 | if (test_bit(lo_fail_bit(iomode), &nfsi->layout->plh_flags)) |
e5e94017 BH |
746 | goto out_unlock; |
747 | ||
568e8c49 AA |
748 | /* Check to see if the layout for the given range already exists */ |
749 | lseg = pnfs_find_lseg(lo, iomode); | |
750 | if (lseg) | |
751 | goto out_unlock; | |
752 | ||
43f1b3da | 753 | if (pnfs_layoutgets_blocked(lo, NULL, 0)) |
cf7d63f1 FI |
754 | goto out_unlock; |
755 | atomic_inc(&lo->plh_outstanding); | |
756 | ||
cc6e5340 | 757 | get_layout_hdr(lo); |
f49f9baa FI |
758 | if (list_empty(&lo->plh_segs)) |
759 | first = true; | |
760 | spin_unlock(&ino->i_lock); | |
761 | if (first) { | |
2130ff66 FI |
762 | /* The lo must be on the clp list if there is any |
763 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
764 | */ | |
765 | spin_lock(&clp->cl_lock); | |
766 | BUG_ON(!list_empty(&lo->plh_layouts)); | |
767 | list_add_tail(&lo->plh_layouts, &clp->cl_layouts); | |
768 | spin_unlock(&clp->cl_lock); | |
769 | } | |
e5e94017 BH |
770 | |
771 | lseg = send_layoutget(lo, ctx, iomode); | |
f49f9baa FI |
772 | if (!lseg && first) { |
773 | spin_lock(&clp->cl_lock); | |
774 | list_del_init(&lo->plh_layouts); | |
775 | spin_unlock(&clp->cl_lock); | |
2130ff66 | 776 | } |
cf7d63f1 | 777 | atomic_dec(&lo->plh_outstanding); |
cc6e5340 | 778 | put_layout_hdr(lo); |
e5e94017 BH |
779 | out: |
780 | dprintk("%s end, state 0x%lx lseg %p\n", __func__, | |
bf9c1387 | 781 | nfsi->layout ? nfsi->layout->plh_flags : -1, lseg); |
e5e94017 BH |
782 | return lseg; |
783 | out_unlock: | |
784 | spin_unlock(&ino->i_lock); | |
785 | goto out; | |
786 | } | |
b1f69b75 AA |
787 | |
788 | int | |
789 | pnfs_layout_process(struct nfs4_layoutget *lgp) | |
790 | { | |
791 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
792 | struct nfs4_layoutget_res *res = &lgp->res; | |
793 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 794 | struct inode *ino = lo->plh_inode; |
43f1b3da | 795 | struct nfs_client *clp = NFS_SERVER(ino)->nfs_client; |
b1f69b75 AA |
796 | int status = 0; |
797 | ||
fc1794c5 FI |
798 | /* Verify we got what we asked for. |
799 | * Note that because the xdr parsing only accepts a single | |
800 | * element array, this can fail even if the server is behaving | |
801 | * correctly. | |
802 | */ | |
803 | if (lgp->args.range.iomode > res->range.iomode || | |
804 | res->range.offset != 0 || | |
805 | res->range.length != NFS4_MAX_UINT64) { | |
806 | status = -EINVAL; | |
807 | goto out; | |
808 | } | |
b1f69b75 AA |
809 | /* Inject layout blob into I/O device driver */ |
810 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res); | |
811 | if (!lseg || IS_ERR(lseg)) { | |
812 | if (!lseg) | |
813 | status = -ENOMEM; | |
814 | else | |
815 | status = PTR_ERR(lseg); | |
816 | dprintk("%s: Could not allocate layout: error %d\n", | |
817 | __func__, status); | |
818 | goto out; | |
819 | } | |
820 | ||
821 | spin_lock(&ino->i_lock); | |
43f1b3da FI |
822 | if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) || |
823 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { | |
824 | dprintk("%s forget reply due to recall\n", __func__); | |
825 | goto out_forget_reply; | |
826 | } | |
827 | ||
828 | if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) { | |
829 | dprintk("%s forget reply due to state\n", __func__); | |
830 | goto out_forget_reply; | |
831 | } | |
b1f69b75 | 832 | init_lseg(lo, lseg); |
566052c5 | 833 | lseg->pls_range = res->range; |
d684d2ae | 834 | *lgp->lsegpp = get_lseg(lseg); |
b1f69b75 AA |
835 | pnfs_insert_layout(lo, lseg); |
836 | ||
f7e8917a FI |
837 | if (res->return_on_close) { |
838 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); | |
839 | set_bit(NFS_LAYOUT_ROC, &lo->plh_flags); | |
840 | } | |
841 | ||
b1f69b75 | 842 | /* Done processing layoutget. Set the layout stateid */ |
43f1b3da | 843 | pnfs_set_layout_stateid(lo, &res->stateid, false); |
b1f69b75 AA |
844 | spin_unlock(&ino->i_lock); |
845 | out: | |
846 | return status; | |
43f1b3da FI |
847 | |
848 | out_forget_reply: | |
849 | spin_unlock(&ino->i_lock); | |
850 | lseg->pls_layout = lo; | |
851 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
852 | goto out; | |
b1f69b75 AA |
853 | } |
854 | ||
bae724ef FI |
855 | static int pnfs_read_pg_test(struct nfs_pageio_descriptor *pgio, |
856 | struct nfs_page *prev, | |
857 | struct nfs_page *req) | |
94ad1c80 | 858 | { |
bae724ef FI |
859 | if (pgio->pg_count == prev->wb_bytes) { |
860 | /* This is first coelesce call for a series of nfs_pages */ | |
861 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
862 | prev->wb_context, | |
863 | IOMODE_READ); | |
864 | } | |
865 | return NFS_SERVER(pgio->pg_inode)->pnfs_curr_ld->pg_test(pgio, prev, req); | |
94ad1c80 FI |
866 | } |
867 | ||
868 | void | |
bae724ef | 869 | pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode) |
94ad1c80 | 870 | { |
bae724ef FI |
871 | struct pnfs_layoutdriver_type *ld; |
872 | ||
873 | ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
874 | pgio->pg_test = (ld && ld->pg_test) ? pnfs_read_pg_test : NULL; | |
94ad1c80 FI |
875 | } |
876 | ||
44b83799 FI |
877 | static int pnfs_write_pg_test(struct nfs_pageio_descriptor *pgio, |
878 | struct nfs_page *prev, | |
879 | struct nfs_page *req) | |
880 | { | |
881 | if (pgio->pg_count == prev->wb_bytes) { | |
882 | /* This is first coelesce call for a series of nfs_pages */ | |
883 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, | |
884 | prev->wb_context, | |
885 | IOMODE_RW); | |
886 | } | |
887 | return NFS_SERVER(pgio->pg_inode)->pnfs_curr_ld->pg_test(pgio, prev, req); | |
888 | } | |
889 | ||
890 | void | |
891 | pnfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, struct inode *inode) | |
892 | { | |
893 | struct pnfs_layoutdriver_type *ld; | |
894 | ||
895 | ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
896 | pgio->pg_test = (ld && ld->pg_test) ? pnfs_write_pg_test : NULL; | |
897 | } | |
898 | ||
0382b744 AA |
899 | enum pnfs_try_status |
900 | pnfs_try_to_write_data(struct nfs_write_data *wdata, | |
901 | const struct rpc_call_ops *call_ops, int how) | |
902 | { | |
903 | struct inode *inode = wdata->inode; | |
904 | enum pnfs_try_status trypnfs; | |
905 | struct nfs_server *nfss = NFS_SERVER(inode); | |
906 | ||
907 | wdata->mds_ops = call_ops; | |
908 | ||
909 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
910 | inode->i_ino, wdata->args.count, wdata->args.offset, how); | |
911 | ||
912 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(wdata, how); | |
913 | if (trypnfs == PNFS_NOT_ATTEMPTED) { | |
914 | put_lseg(wdata->lseg); | |
915 | wdata->lseg = NULL; | |
916 | } else | |
917 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); | |
918 | ||
919 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); | |
920 | return trypnfs; | |
921 | } | |
922 | ||
64419a9b AA |
923 | /* |
924 | * Call the appropriate parallel I/O subsystem read function. | |
925 | */ | |
926 | enum pnfs_try_status | |
927 | pnfs_try_to_read_data(struct nfs_read_data *rdata, | |
928 | const struct rpc_call_ops *call_ops) | |
929 | { | |
930 | struct inode *inode = rdata->inode; | |
931 | struct nfs_server *nfss = NFS_SERVER(inode); | |
932 | enum pnfs_try_status trypnfs; | |
933 | ||
934 | rdata->mds_ops = call_ops; | |
935 | ||
936 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
937 | __func__, inode->i_ino, rdata->args.count, rdata->args.offset); | |
938 | ||
939 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(rdata); | |
940 | if (trypnfs == PNFS_NOT_ATTEMPTED) { | |
941 | put_lseg(rdata->lseg); | |
942 | rdata->lseg = NULL; | |
943 | } else { | |
944 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); | |
945 | } | |
946 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); | |
947 | return trypnfs; | |
948 | } |