]>
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 RL |
32 | #include "pnfs.h" |
33 | ||
34 | #define NFSDBG_FACILITY NFSDBG_PNFS | |
35 | ||
02c35fca FI |
36 | /* Locking: |
37 | * | |
38 | * pnfs_spinlock: | |
39 | * protects pnfs_modules_tbl. | |
40 | */ | |
41 | static DEFINE_SPINLOCK(pnfs_spinlock); | |
42 | ||
43 | /* | |
44 | * pnfs_modules_tbl holds all pnfs modules | |
45 | */ | |
46 | static LIST_HEAD(pnfs_modules_tbl); | |
47 | ||
48 | /* Return the registered pnfs layout driver module matching given id */ | |
49 | static struct pnfs_layoutdriver_type * | |
50 | find_pnfs_driver_locked(u32 id) | |
51 | { | |
52 | struct pnfs_layoutdriver_type *local; | |
53 | ||
54 | list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid) | |
55 | if (local->id == id) | |
56 | goto out; | |
57 | local = NULL; | |
58 | out: | |
59 | dprintk("%s: Searching for id %u, found %p\n", __func__, id, local); | |
60 | return local; | |
61 | } | |
62 | ||
85e174ba RL |
63 | static struct pnfs_layoutdriver_type * |
64 | find_pnfs_driver(u32 id) | |
65 | { | |
02c35fca FI |
66 | struct pnfs_layoutdriver_type *local; |
67 | ||
68 | spin_lock(&pnfs_spinlock); | |
69 | local = find_pnfs_driver_locked(id); | |
70 | spin_unlock(&pnfs_spinlock); | |
71 | return local; | |
85e174ba RL |
72 | } |
73 | ||
74 | void | |
75 | unset_pnfs_layoutdriver(struct nfs_server *nfss) | |
76 | { | |
02c35fca | 77 | if (nfss->pnfs_curr_ld) { |
1c787096 | 78 | nfss->pnfs_curr_ld->clear_layoutdriver(nfss); |
02c35fca FI |
79 | module_put(nfss->pnfs_curr_ld->owner); |
80 | } | |
85e174ba RL |
81 | nfss->pnfs_curr_ld = NULL; |
82 | } | |
83 | ||
84 | /* | |
85 | * Try to set the server's pnfs module to the pnfs layout type specified by id. | |
86 | * Currently only one pNFS layout driver per filesystem is supported. | |
87 | * | |
88 | * @id layout type. Zero (illegal layout type) indicates pNFS not in use. | |
89 | */ | |
90 | void | |
91 | set_pnfs_layoutdriver(struct nfs_server *server, u32 id) | |
92 | { | |
93 | struct pnfs_layoutdriver_type *ld_type = NULL; | |
94 | ||
95 | if (id == 0) | |
96 | goto out_no_driver; | |
97 | if (!(server->nfs_client->cl_exchange_flags & | |
98 | (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) { | |
99 | printk(KERN_ERR "%s: id %u cl_exchange_flags 0x%x\n", __func__, | |
100 | id, server->nfs_client->cl_exchange_flags); | |
101 | goto out_no_driver; | |
102 | } | |
103 | ld_type = find_pnfs_driver(id); | |
104 | if (!ld_type) { | |
105 | request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id); | |
106 | ld_type = find_pnfs_driver(id); | |
107 | if (!ld_type) { | |
108 | dprintk("%s: No pNFS module found for %u.\n", | |
109 | __func__, id); | |
110 | goto out_no_driver; | |
111 | } | |
112 | } | |
02c35fca FI |
113 | if (!try_module_get(ld_type->owner)) { |
114 | dprintk("%s: Could not grab reference on module\n", __func__); | |
115 | goto out_no_driver; | |
116 | } | |
85e174ba | 117 | server->pnfs_curr_ld = ld_type; |
1c787096 | 118 | if (ld_type->set_layoutdriver(server)) { |
02c35fca FI |
119 | printk(KERN_ERR |
120 | "%s: Error initializing mount point for layout driver %u.\n", | |
121 | __func__, id); | |
122 | module_put(ld_type->owner); | |
123 | goto out_no_driver; | |
124 | } | |
85e174ba RL |
125 | dprintk("%s: pNFS module for %u set\n", __func__, id); |
126 | return; | |
127 | ||
128 | out_no_driver: | |
129 | dprintk("%s: Using NFSv4 I/O\n", __func__); | |
130 | server->pnfs_curr_ld = NULL; | |
131 | } | |
02c35fca FI |
132 | |
133 | int | |
134 | pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
135 | { | |
136 | int status = -EINVAL; | |
137 | struct pnfs_layoutdriver_type *tmp; | |
138 | ||
139 | if (ld_type->id == 0) { | |
140 | printk(KERN_ERR "%s id 0 is reserved\n", __func__); | |
141 | return status; | |
142 | } | |
b1f69b75 AA |
143 | if (!ld_type->alloc_lseg || !ld_type->free_lseg) { |
144 | printk(KERN_ERR "%s Layout driver must provide " | |
145 | "alloc_lseg and free_lseg.\n", __func__); | |
146 | return status; | |
147 | } | |
02c35fca FI |
148 | |
149 | spin_lock(&pnfs_spinlock); | |
150 | tmp = find_pnfs_driver_locked(ld_type->id); | |
151 | if (!tmp) { | |
152 | list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl); | |
153 | status = 0; | |
154 | dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id, | |
155 | ld_type->name); | |
156 | } else { | |
157 | printk(KERN_ERR "%s Module with id %d already loaded!\n", | |
158 | __func__, ld_type->id); | |
159 | } | |
160 | spin_unlock(&pnfs_spinlock); | |
161 | ||
162 | return status; | |
163 | } | |
164 | EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver); | |
165 | ||
166 | void | |
167 | pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
168 | { | |
169 | dprintk("%s Deregistering id:%u\n", __func__, ld_type->id); | |
170 | spin_lock(&pnfs_spinlock); | |
171 | list_del(&ld_type->pnfs_tblid); | |
172 | spin_unlock(&pnfs_spinlock); | |
173 | } | |
174 | EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver); | |
e5e94017 | 175 | |
b1f69b75 AA |
176 | /* |
177 | * pNFS client layout cache | |
178 | */ | |
179 | ||
e5e94017 BH |
180 | static void |
181 | get_layout_hdr_locked(struct pnfs_layout_hdr *lo) | |
182 | { | |
b7edfaa1 FI |
183 | assert_spin_locked(&lo->plh_inode->i_lock); |
184 | lo->plh_refcount++; | |
e5e94017 BH |
185 | } |
186 | ||
187 | static void | |
188 | put_layout_hdr_locked(struct pnfs_layout_hdr *lo) | |
189 | { | |
b7edfaa1 FI |
190 | assert_spin_locked(&lo->plh_inode->i_lock); |
191 | BUG_ON(lo->plh_refcount == 0); | |
e5e94017 | 192 | |
b7edfaa1 FI |
193 | lo->plh_refcount--; |
194 | if (!lo->plh_refcount) { | |
e5e94017 | 195 | dprintk("%s: freeing layout cache %p\n", __func__, lo); |
b7edfaa1 FI |
196 | BUG_ON(!list_empty(&lo->plh_layouts)); |
197 | NFS_I(lo->plh_inode)->layout = NULL; | |
e5e94017 BH |
198 | kfree(lo); |
199 | } | |
200 | } | |
201 | ||
b1f69b75 | 202 | void |
974cec8c AA |
203 | put_layout_hdr(struct inode *inode) |
204 | { | |
205 | spin_lock(&inode->i_lock); | |
206 | put_layout_hdr_locked(NFS_I(inode)->layout); | |
207 | spin_unlock(&inode->i_lock); | |
208 | } | |
209 | ||
210 | static void | |
211 | init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg) | |
212 | { | |
566052c5 | 213 | INIT_LIST_HEAD(&lseg->pls_list); |
4541d16c FI |
214 | atomic_set(&lseg->pls_refcount, 1); |
215 | smp_mb(); | |
216 | set_bit(NFS_LSEG_VALID, &lseg->pls_flags); | |
566052c5 | 217 | lseg->pls_layout = lo; |
974cec8c AA |
218 | } |
219 | ||
4541d16c | 220 | static void free_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 221 | { |
b7edfaa1 | 222 | struct inode *ino = lseg->pls_layout->plh_inode; |
974cec8c | 223 | |
b1f69b75 | 224 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); |
52fabd73 | 225 | /* Matched by get_layout_hdr in pnfs_insert_layout */ |
974cec8c AA |
226 | put_layout_hdr(ino); |
227 | } | |
228 | ||
4541d16c FI |
229 | /* The use of tmp_list is necessary because pnfs_curr_ld->free_lseg |
230 | * could sleep, so must be called outside of the lock. | |
231 | * Returns 1 if object was removed, otherwise return 0. | |
232 | */ | |
233 | static int | |
234 | put_lseg_locked(struct pnfs_layout_segment *lseg, | |
235 | struct list_head *tmp_list) | |
974cec8c | 236 | { |
4541d16c FI |
237 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, |
238 | atomic_read(&lseg->pls_refcount), | |
239 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
240 | if (atomic_dec_and_test(&lseg->pls_refcount)) { | |
241 | struct inode *ino = lseg->pls_layout->plh_inode; | |
242 | ||
243 | BUG_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
244 | list_del(&lseg->pls_list); | |
245 | if (list_empty(&lseg->pls_layout->plh_segs)) { | |
246 | struct nfs_client *clp; | |
247 | ||
248 | clp = NFS_SERVER(ino)->nfs_client; | |
249 | spin_lock(&clp->cl_lock); | |
250 | /* List does not take a reference, so no need for put here */ | |
251 | list_del_init(&lseg->pls_layout->plh_layouts); | |
252 | spin_unlock(&clp->cl_lock); | |
253 | } | |
254 | list_add(&lseg->pls_list, tmp_list); | |
255 | return 1; | |
256 | } | |
257 | return 0; | |
258 | } | |
974cec8c | 259 | |
4541d16c FI |
260 | static bool |
261 | should_free_lseg(u32 lseg_iomode, u32 recall_iomode) | |
262 | { | |
263 | return (recall_iomode == IOMODE_ANY || | |
264 | lseg_iomode == recall_iomode); | |
974cec8c AA |
265 | } |
266 | ||
4541d16c FI |
267 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
268 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
269 | struct list_head *tmp_list) | |
270 | { | |
271 | int rv = 0; | |
272 | ||
273 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
274 | /* Remove the reference keeping the lseg in the | |
275 | * list. It will now be removed when all | |
276 | * outstanding io is finished. | |
277 | */ | |
278 | rv = put_lseg_locked(lseg, tmp_list); | |
279 | } | |
280 | return rv; | |
281 | } | |
282 | ||
283 | /* Returns count of number of matching invalid lsegs remaining in list | |
284 | * after call. | |
285 | */ | |
286 | static int | |
287 | mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, | |
288 | struct list_head *tmp_list, | |
289 | u32 iomode) | |
974cec8c AA |
290 | { |
291 | struct pnfs_layout_segment *lseg, *next; | |
4541d16c | 292 | int invalid = 0, removed = 0; |
974cec8c AA |
293 | |
294 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
295 | ||
4541d16c FI |
296 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
297 | if (should_free_lseg(lseg->pls_range.iomode, iomode)) { | |
298 | dprintk("%s: freeing lseg %p iomode %d " | |
299 | "offset %llu length %llu\n", __func__, | |
300 | lseg, lseg->pls_range.iomode, lseg->pls_range.offset, | |
301 | lseg->pls_range.length); | |
302 | invalid++; | |
303 | removed += mark_lseg_invalid(lseg, tmp_list); | |
304 | } | |
305 | dprintk("%s:Return %i\n", __func__, invalid - removed); | |
306 | return invalid - removed; | |
974cec8c AA |
307 | } |
308 | ||
309 | static void | |
4541d16c | 310 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 311 | { |
4541d16c | 312 | struct pnfs_layout_segment *lseg, *tmp; |
974cec8c | 313 | |
4541d16c | 314 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 315 | list_del(&lseg->pls_list); |
4541d16c | 316 | free_lseg(lseg); |
974cec8c AA |
317 | } |
318 | } | |
319 | ||
e5e94017 BH |
320 | void |
321 | pnfs_destroy_layout(struct nfs_inode *nfsi) | |
322 | { | |
323 | struct pnfs_layout_hdr *lo; | |
974cec8c | 324 | LIST_HEAD(tmp_list); |
e5e94017 BH |
325 | |
326 | spin_lock(&nfsi->vfs_inode.i_lock); | |
327 | lo = nfsi->layout; | |
328 | if (lo) { | |
4541d16c FI |
329 | set_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags); |
330 | mark_matching_lsegs_invalid(lo, &tmp_list, IOMODE_ANY); | |
e5e94017 BH |
331 | /* Matched by refcount set to 1 in alloc_init_layout_hdr */ |
332 | put_layout_hdr_locked(lo); | |
333 | } | |
334 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
974cec8c AA |
335 | pnfs_free_lseg_list(&tmp_list); |
336 | } | |
337 | ||
338 | /* | |
339 | * Called by the state manger to remove all layouts established under an | |
340 | * expired lease. | |
341 | */ | |
342 | void | |
343 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
344 | { | |
345 | struct pnfs_layout_hdr *lo; | |
346 | LIST_HEAD(tmp_list); | |
347 | ||
348 | spin_lock(&clp->cl_lock); | |
349 | list_splice_init(&clp->cl_layouts, &tmp_list); | |
350 | spin_unlock(&clp->cl_lock); | |
351 | ||
352 | while (!list_empty(&tmp_list)) { | |
353 | lo = list_entry(tmp_list.next, struct pnfs_layout_hdr, | |
b7edfaa1 | 354 | plh_layouts); |
974cec8c | 355 | dprintk("%s freeing layout for inode %lu\n", __func__, |
b7edfaa1 FI |
356 | lo->plh_inode->i_ino); |
357 | pnfs_destroy_layout(NFS_I(lo->plh_inode)); | |
974cec8c | 358 | } |
e5e94017 BH |
359 | } |
360 | ||
fd6002e9 | 361 | /* update lo->plh_stateid with new if is more recent */ |
b1f69b75 AA |
362 | static void |
363 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, | |
364 | const nfs4_stateid *new) | |
365 | { | |
fd6002e9 | 366 | u32 oldseq, newseq; |
b1f69b75 | 367 | |
fd6002e9 FI |
368 | oldseq = be32_to_cpu(lo->plh_stateid.stateid.seqid); |
369 | newseq = be32_to_cpu(new->stateid.seqid); | |
370 | if ((int)(newseq - oldseq) > 0) | |
371 | memcpy(&lo->plh_stateid, &new->stateid, sizeof(new->stateid)); | |
b1f69b75 AA |
372 | } |
373 | ||
cf7d63f1 FI |
374 | /* lget is set to 1 if called from inside send_layoutget call chain */ |
375 | static bool | |
376 | pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, int lget) | |
377 | { | |
378 | return (list_empty(&lo->plh_segs) && | |
379 | (atomic_read(&lo->plh_outstanding) > lget)); | |
380 | } | |
381 | ||
fd6002e9 FI |
382 | int |
383 | pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, | |
384 | struct nfs4_state *open_state) | |
b1f69b75 | 385 | { |
fd6002e9 | 386 | int status = 0; |
974cec8c | 387 | |
b1f69b75 | 388 | dprintk("--> %s\n", __func__); |
fd6002e9 | 389 | spin_lock(&lo->plh_inode->i_lock); |
cf7d63f1 FI |
390 | if (pnfs_layoutgets_blocked(lo, 1)) { |
391 | status = -EAGAIN; | |
392 | } else if (list_empty(&lo->plh_segs)) { | |
fd6002e9 FI |
393 | int seq; |
394 | ||
395 | do { | |
396 | seq = read_seqbegin(&open_state->seqlock); | |
397 | memcpy(dst->data, open_state->stateid.data, | |
398 | sizeof(open_state->stateid.data)); | |
399 | } while (read_seqretry(&open_state->seqlock, seq)); | |
400 | } else | |
401 | memcpy(dst->data, lo->plh_stateid.data, sizeof(lo->plh_stateid.data)); | |
402 | spin_unlock(&lo->plh_inode->i_lock); | |
b1f69b75 | 403 | dprintk("<-- %s\n", __func__); |
fd6002e9 | 404 | return status; |
b1f69b75 AA |
405 | } |
406 | ||
407 | /* | |
408 | * Get layout from server. | |
409 | * for now, assume that whole file layouts are requested. | |
410 | * arg->offset: 0 | |
411 | * arg->length: all ones | |
412 | */ | |
e5e94017 BH |
413 | static struct pnfs_layout_segment * |
414 | send_layoutget(struct pnfs_layout_hdr *lo, | |
415 | struct nfs_open_context *ctx, | |
416 | u32 iomode) | |
417 | { | |
b7edfaa1 | 418 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
419 | struct nfs_server *server = NFS_SERVER(ino); |
420 | struct nfs4_layoutget *lgp; | |
421 | struct pnfs_layout_segment *lseg = NULL; | |
422 | ||
423 | dprintk("--> %s\n", __func__); | |
e5e94017 | 424 | |
b1f69b75 AA |
425 | BUG_ON(ctx == NULL); |
426 | lgp = kzalloc(sizeof(*lgp), GFP_KERNEL); | |
cf7d63f1 | 427 | if (lgp == NULL) |
b1f69b75 | 428 | return NULL; |
b1f69b75 AA |
429 | lgp->args.minlength = NFS4_MAX_UINT64; |
430 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; | |
431 | lgp->args.range.iomode = iomode; | |
432 | lgp->args.range.offset = 0; | |
433 | lgp->args.range.length = NFS4_MAX_UINT64; | |
434 | lgp->args.type = server->pnfs_curr_ld->id; | |
435 | lgp->args.inode = ino; | |
436 | lgp->args.ctx = get_nfs_open_context(ctx); | |
437 | lgp->lsegpp = &lseg; | |
438 | ||
439 | /* Synchronously retrieve layout information from server and | |
440 | * store in lseg. | |
441 | */ | |
442 | nfs4_proc_layoutget(lgp); | |
974cec8c | 443 | if (!lseg) { |
b1f69b75 | 444 | /* remember that LAYOUTGET failed and suspend trying */ |
566052c5 | 445 | set_bit(lo_fail_bit(iomode), &lo->plh_flags); |
974cec8c | 446 | } |
974cec8c AA |
447 | return lseg; |
448 | } | |
449 | ||
b1f69b75 AA |
450 | /* |
451 | * Compare two layout segments for sorting into layout cache. | |
452 | * We want to preferentially return RW over RO layouts, so ensure those | |
453 | * are seen first. | |
454 | */ | |
455 | static s64 | |
456 | cmp_layout(u32 iomode1, u32 iomode2) | |
457 | { | |
458 | /* read > read/write */ | |
459 | return (int)(iomode2 == IOMODE_READ) - (int)(iomode1 == IOMODE_READ); | |
460 | } | |
461 | ||
974cec8c AA |
462 | static void |
463 | pnfs_insert_layout(struct pnfs_layout_hdr *lo, | |
464 | struct pnfs_layout_segment *lseg) | |
465 | { | |
b1f69b75 AA |
466 | struct pnfs_layout_segment *lp; |
467 | int found = 0; | |
468 | ||
974cec8c AA |
469 | dprintk("%s:Begin\n", __func__); |
470 | ||
b7edfaa1 | 471 | assert_spin_locked(&lo->plh_inode->i_lock); |
b7edfaa1 | 472 | list_for_each_entry(lp, &lo->plh_segs, pls_list) { |
566052c5 | 473 | if (cmp_layout(lp->pls_range.iomode, lseg->pls_range.iomode) > 0) |
b1f69b75 | 474 | continue; |
566052c5 | 475 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
476 | dprintk("%s: inserted lseg %p " |
477 | "iomode %d offset %llu length %llu before " | |
478 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
479 | __func__, lseg, lseg->pls_range.iomode, |
480 | lseg->pls_range.offset, lseg->pls_range.length, | |
481 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
482 | lp->pls_range.length); | |
b1f69b75 AA |
483 | found = 1; |
484 | break; | |
485 | } | |
486 | if (!found) { | |
b7edfaa1 | 487 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
b1f69b75 AA |
488 | dprintk("%s: inserted lseg %p " |
489 | "iomode %d offset %llu length %llu at tail\n", | |
566052c5 FI |
490 | __func__, lseg, lseg->pls_range.iomode, |
491 | lseg->pls_range.offset, lseg->pls_range.length); | |
974cec8c | 492 | } |
b1f69b75 | 493 | get_layout_hdr_locked(lo); |
974cec8c AA |
494 | |
495 | dprintk("%s:Return\n", __func__); | |
e5e94017 BH |
496 | } |
497 | ||
498 | static struct pnfs_layout_hdr * | |
499 | alloc_init_layout_hdr(struct inode *ino) | |
500 | { | |
501 | struct pnfs_layout_hdr *lo; | |
502 | ||
503 | lo = kzalloc(sizeof(struct pnfs_layout_hdr), GFP_KERNEL); | |
504 | if (!lo) | |
505 | return NULL; | |
b7edfaa1 FI |
506 | lo->plh_refcount = 1; |
507 | INIT_LIST_HEAD(&lo->plh_layouts); | |
508 | INIT_LIST_HEAD(&lo->plh_segs); | |
b7edfaa1 | 509 | lo->plh_inode = ino; |
e5e94017 BH |
510 | return lo; |
511 | } | |
512 | ||
513 | static struct pnfs_layout_hdr * | |
514 | pnfs_find_alloc_layout(struct inode *ino) | |
515 | { | |
516 | struct nfs_inode *nfsi = NFS_I(ino); | |
517 | struct pnfs_layout_hdr *new = NULL; | |
518 | ||
519 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
520 | ||
521 | assert_spin_locked(&ino->i_lock); | |
4541d16c FI |
522 | if (nfsi->layout) { |
523 | if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags)) | |
524 | return NULL; | |
525 | else | |
526 | return nfsi->layout; | |
527 | } | |
e5e94017 BH |
528 | spin_unlock(&ino->i_lock); |
529 | new = alloc_init_layout_hdr(ino); | |
530 | spin_lock(&ino->i_lock); | |
531 | ||
532 | if (likely(nfsi->layout == NULL)) /* Won the race? */ | |
533 | nfsi->layout = new; | |
534 | else | |
535 | kfree(new); | |
536 | return nfsi->layout; | |
537 | } | |
538 | ||
b1f69b75 AA |
539 | /* |
540 | * iomode matching rules: | |
541 | * iomode lseg match | |
542 | * ----- ----- ----- | |
543 | * ANY READ true | |
544 | * ANY RW true | |
545 | * RW READ false | |
546 | * RW RW true | |
547 | * READ READ true | |
548 | * READ RW true | |
549 | */ | |
550 | static int | |
551 | is_matching_lseg(struct pnfs_layout_segment *lseg, u32 iomode) | |
552 | { | |
566052c5 | 553 | return (iomode != IOMODE_RW || lseg->pls_range.iomode == IOMODE_RW); |
b1f69b75 AA |
554 | } |
555 | ||
556 | /* | |
557 | * lookup range in layout | |
558 | */ | |
e5e94017 BH |
559 | static struct pnfs_layout_segment * |
560 | pnfs_has_layout(struct pnfs_layout_hdr *lo, u32 iomode) | |
561 | { | |
b1f69b75 AA |
562 | struct pnfs_layout_segment *lseg, *ret = NULL; |
563 | ||
564 | dprintk("%s:Begin\n", __func__); | |
565 | ||
b7edfaa1 FI |
566 | assert_spin_locked(&lo->plh_inode->i_lock); |
567 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { | |
4541d16c FI |
568 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
569 | is_matching_lseg(lseg, iomode)) { | |
b1f69b75 AA |
570 | ret = lseg; |
571 | break; | |
572 | } | |
566052c5 | 573 | if (cmp_layout(iomode, lseg->pls_range.iomode) > 0) |
b1f69b75 AA |
574 | break; |
575 | } | |
576 | ||
577 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 578 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 579 | return ret; |
e5e94017 BH |
580 | } |
581 | ||
582 | /* | |
583 | * Layout segment is retreived from the server if not cached. | |
584 | * The appropriate layout segment is referenced and returned to the caller. | |
585 | */ | |
586 | struct pnfs_layout_segment * | |
587 | pnfs_update_layout(struct inode *ino, | |
588 | struct nfs_open_context *ctx, | |
589 | enum pnfs_iomode iomode) | |
590 | { | |
591 | struct nfs_inode *nfsi = NFS_I(ino); | |
2130ff66 | 592 | struct nfs_client *clp = NFS_SERVER(ino)->nfs_client; |
e5e94017 BH |
593 | struct pnfs_layout_hdr *lo; |
594 | struct pnfs_layout_segment *lseg = NULL; | |
595 | ||
596 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) | |
597 | return NULL; | |
598 | spin_lock(&ino->i_lock); | |
599 | lo = pnfs_find_alloc_layout(ino); | |
600 | if (lo == NULL) { | |
601 | dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__); | |
602 | goto out_unlock; | |
603 | } | |
604 | ||
605 | /* Check to see if the layout for the given range already exists */ | |
606 | lseg = pnfs_has_layout(lo, iomode); | |
607 | if (lseg) { | |
608 | dprintk("%s: Using cached lseg %p for iomode %d)\n", | |
609 | __func__, lseg, iomode); | |
610 | goto out_unlock; | |
611 | } | |
612 | ||
613 | /* if LAYOUTGET already failed once we don't try again */ | |
566052c5 | 614 | if (test_bit(lo_fail_bit(iomode), &nfsi->layout->plh_flags)) |
e5e94017 BH |
615 | goto out_unlock; |
616 | ||
cf7d63f1 FI |
617 | if (pnfs_layoutgets_blocked(lo, 0)) |
618 | goto out_unlock; | |
619 | atomic_inc(&lo->plh_outstanding); | |
620 | ||
621 | get_layout_hdr_locked(lo); | |
2130ff66 FI |
622 | if (list_empty(&lo->plh_segs)) { |
623 | /* The lo must be on the clp list if there is any | |
624 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
625 | */ | |
626 | spin_lock(&clp->cl_lock); | |
627 | BUG_ON(!list_empty(&lo->plh_layouts)); | |
628 | list_add_tail(&lo->plh_layouts, &clp->cl_layouts); | |
629 | spin_unlock(&clp->cl_lock); | |
630 | } | |
e5e94017 BH |
631 | spin_unlock(&ino->i_lock); |
632 | ||
633 | lseg = send_layoutget(lo, ctx, iomode); | |
2130ff66 FI |
634 | if (!lseg) { |
635 | spin_lock(&ino->i_lock); | |
636 | if (list_empty(&lo->plh_segs)) { | |
637 | spin_lock(&clp->cl_lock); | |
638 | list_del_init(&lo->plh_layouts); | |
639 | spin_unlock(&clp->cl_lock); | |
640 | } | |
641 | spin_unlock(&ino->i_lock); | |
642 | } | |
cf7d63f1 FI |
643 | atomic_dec(&lo->plh_outstanding); |
644 | put_layout_hdr(ino); | |
e5e94017 BH |
645 | out: |
646 | dprintk("%s end, state 0x%lx lseg %p\n", __func__, | |
566052c5 | 647 | nfsi->layout->plh_flags, lseg); |
e5e94017 BH |
648 | return lseg; |
649 | out_unlock: | |
650 | spin_unlock(&ino->i_lock); | |
651 | goto out; | |
652 | } | |
b1f69b75 AA |
653 | |
654 | int | |
655 | pnfs_layout_process(struct nfs4_layoutget *lgp) | |
656 | { | |
657 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
658 | struct nfs4_layoutget_res *res = &lgp->res; | |
659 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 660 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
661 | int status = 0; |
662 | ||
663 | /* Inject layout blob into I/O device driver */ | |
664 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res); | |
665 | if (!lseg || IS_ERR(lseg)) { | |
666 | if (!lseg) | |
667 | status = -ENOMEM; | |
668 | else | |
669 | status = PTR_ERR(lseg); | |
670 | dprintk("%s: Could not allocate layout: error %d\n", | |
671 | __func__, status); | |
672 | goto out; | |
673 | } | |
674 | ||
675 | spin_lock(&ino->i_lock); | |
676 | init_lseg(lo, lseg); | |
566052c5 | 677 | lseg->pls_range = res->range; |
b1f69b75 AA |
678 | *lgp->lsegpp = lseg; |
679 | pnfs_insert_layout(lo, lseg); | |
680 | ||
681 | /* Done processing layoutget. Set the layout stateid */ | |
682 | pnfs_set_layout_stateid(lo, &res->stateid); | |
683 | spin_unlock(&ino->i_lock); | |
684 | out: | |
685 | return status; | |
686 | } | |
687 | ||
688 | /* | |
689 | * Device ID cache. Currently supports one layout type per struct nfs_client. | |
690 | * Add layout type to the lookup key to expand to support multiple types. | |
691 | */ | |
692 | int | |
693 | pnfs_alloc_init_deviceid_cache(struct nfs_client *clp, | |
694 | void (*free_callback)(struct pnfs_deviceid_node *)) | |
695 | { | |
696 | struct pnfs_deviceid_cache *c; | |
697 | ||
698 | c = kzalloc(sizeof(struct pnfs_deviceid_cache), GFP_KERNEL); | |
699 | if (!c) | |
700 | return -ENOMEM; | |
701 | spin_lock(&clp->cl_lock); | |
702 | if (clp->cl_devid_cache != NULL) { | |
703 | atomic_inc(&clp->cl_devid_cache->dc_ref); | |
704 | dprintk("%s [kref [%d]]\n", __func__, | |
705 | atomic_read(&clp->cl_devid_cache->dc_ref)); | |
706 | kfree(c); | |
707 | } else { | |
708 | /* kzalloc initializes hlists */ | |
709 | spin_lock_init(&c->dc_lock); | |
710 | atomic_set(&c->dc_ref, 1); | |
711 | c->dc_free_callback = free_callback; | |
712 | clp->cl_devid_cache = c; | |
713 | dprintk("%s [new]\n", __func__); | |
714 | } | |
715 | spin_unlock(&clp->cl_lock); | |
716 | return 0; | |
717 | } | |
718 | EXPORT_SYMBOL_GPL(pnfs_alloc_init_deviceid_cache); | |
719 | ||
720 | /* | |
721 | * Called from pnfs_layoutdriver_type->free_lseg | |
722 | * last layout segment reference frees deviceid | |
723 | */ | |
724 | void | |
725 | pnfs_put_deviceid(struct pnfs_deviceid_cache *c, | |
726 | struct pnfs_deviceid_node *devid) | |
727 | { | |
728 | struct nfs4_deviceid *id = &devid->de_id; | |
729 | struct pnfs_deviceid_node *d; | |
730 | struct hlist_node *n; | |
731 | long h = nfs4_deviceid_hash(id); | |
732 | ||
733 | dprintk("%s [%d]\n", __func__, atomic_read(&devid->de_ref)); | |
734 | if (!atomic_dec_and_lock(&devid->de_ref, &c->dc_lock)) | |
735 | return; | |
736 | ||
737 | hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[h], de_node) | |
738 | if (!memcmp(&d->de_id, id, sizeof(*id))) { | |
739 | hlist_del_rcu(&d->de_node); | |
740 | spin_unlock(&c->dc_lock); | |
741 | synchronize_rcu(); | |
742 | c->dc_free_callback(devid); | |
743 | return; | |
744 | } | |
745 | spin_unlock(&c->dc_lock); | |
746 | /* Why wasn't it found in the list? */ | |
747 | BUG(); | |
748 | } | |
749 | EXPORT_SYMBOL_GPL(pnfs_put_deviceid); | |
750 | ||
751 | /* Find and reference a deviceid */ | |
752 | struct pnfs_deviceid_node * | |
753 | pnfs_find_get_deviceid(struct pnfs_deviceid_cache *c, struct nfs4_deviceid *id) | |
754 | { | |
755 | struct pnfs_deviceid_node *d; | |
756 | struct hlist_node *n; | |
757 | long hash = nfs4_deviceid_hash(id); | |
758 | ||
759 | dprintk("--> %s hash %ld\n", __func__, hash); | |
760 | rcu_read_lock(); | |
761 | hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[hash], de_node) { | |
762 | if (!memcmp(&d->de_id, id, sizeof(*id))) { | |
763 | if (!atomic_inc_not_zero(&d->de_ref)) { | |
764 | goto fail; | |
765 | } else { | |
766 | rcu_read_unlock(); | |
767 | return d; | |
768 | } | |
769 | } | |
770 | } | |
771 | fail: | |
772 | rcu_read_unlock(); | |
773 | return NULL; | |
774 | } | |
775 | EXPORT_SYMBOL_GPL(pnfs_find_get_deviceid); | |
776 | ||
777 | /* | |
778 | * Add a deviceid to the cache. | |
779 | * GETDEVICEINFOs for same deviceid can race. If deviceid is found, discard new | |
780 | */ | |
781 | struct pnfs_deviceid_node * | |
782 | pnfs_add_deviceid(struct pnfs_deviceid_cache *c, struct pnfs_deviceid_node *new) | |
783 | { | |
784 | struct pnfs_deviceid_node *d; | |
785 | long hash = nfs4_deviceid_hash(&new->de_id); | |
786 | ||
787 | dprintk("--> %s hash %ld\n", __func__, hash); | |
788 | spin_lock(&c->dc_lock); | |
789 | d = pnfs_find_get_deviceid(c, &new->de_id); | |
790 | if (d) { | |
791 | spin_unlock(&c->dc_lock); | |
792 | dprintk("%s [discard]\n", __func__); | |
793 | c->dc_free_callback(new); | |
794 | return d; | |
795 | } | |
796 | INIT_HLIST_NODE(&new->de_node); | |
797 | atomic_set(&new->de_ref, 1); | |
798 | hlist_add_head_rcu(&new->de_node, &c->dc_deviceids[hash]); | |
799 | spin_unlock(&c->dc_lock); | |
800 | dprintk("%s [new]\n", __func__); | |
801 | return new; | |
802 | } | |
803 | EXPORT_SYMBOL_GPL(pnfs_add_deviceid); | |
804 | ||
805 | void | |
806 | pnfs_put_deviceid_cache(struct nfs_client *clp) | |
807 | { | |
808 | struct pnfs_deviceid_cache *local = clp->cl_devid_cache; | |
809 | ||
810 | dprintk("--> %s cl_devid_cache %p\n", __func__, clp->cl_devid_cache); | |
811 | if (atomic_dec_and_lock(&local->dc_ref, &clp->cl_lock)) { | |
812 | int i; | |
813 | /* Verify cache is empty */ | |
814 | for (i = 0; i < NFS4_DEVICE_ID_HASH_SIZE; i++) | |
815 | BUG_ON(!hlist_empty(&local->dc_deviceids[i])); | |
816 | clp->cl_devid_cache = NULL; | |
817 | spin_unlock(&clp->cl_lock); | |
818 | kfree(local); | |
819 | } | |
820 | } | |
821 | EXPORT_SYMBOL_GPL(pnfs_put_deviceid_cache); |