]>
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 | ||
cc6e5340 | 180 | /* Need to hold i_lock if caller does not already hold reference */ |
43f1b3da | 181 | void |
cc6e5340 | 182 | get_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 183 | { |
cc6e5340 | 184 | atomic_inc(&lo->plh_refcount); |
e5e94017 BH |
185 | } |
186 | ||
187 | static void | |
cc6e5340 | 188 | destroy_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 189 | { |
cc6e5340 FI |
190 | dprintk("%s: freeing layout cache %p\n", __func__, lo); |
191 | BUG_ON(!list_empty(&lo->plh_layouts)); | |
192 | NFS_I(lo->plh_inode)->layout = NULL; | |
193 | kfree(lo); | |
194 | } | |
e5e94017 | 195 | |
cc6e5340 FI |
196 | static void |
197 | put_layout_hdr_locked(struct pnfs_layout_hdr *lo) | |
198 | { | |
199 | if (atomic_dec_and_test(&lo->plh_refcount)) | |
200 | destroy_layout_hdr(lo); | |
e5e94017 BH |
201 | } |
202 | ||
b1f69b75 | 203 | void |
cc6e5340 | 204 | put_layout_hdr(struct pnfs_layout_hdr *lo) |
974cec8c | 205 | { |
cc6e5340 FI |
206 | struct inode *inode = lo->plh_inode; |
207 | ||
208 | if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) { | |
209 | destroy_layout_hdr(lo); | |
210 | spin_unlock(&inode->i_lock); | |
211 | } | |
974cec8c AA |
212 | } |
213 | ||
214 | static void | |
215 | init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg) | |
216 | { | |
566052c5 | 217 | INIT_LIST_HEAD(&lseg->pls_list); |
4541d16c FI |
218 | atomic_set(&lseg->pls_refcount, 1); |
219 | smp_mb(); | |
220 | set_bit(NFS_LSEG_VALID, &lseg->pls_flags); | |
566052c5 | 221 | lseg->pls_layout = lo; |
974cec8c AA |
222 | } |
223 | ||
4541d16c | 224 | static void free_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 225 | { |
b7edfaa1 | 226 | struct inode *ino = lseg->pls_layout->plh_inode; |
974cec8c | 227 | |
b1f69b75 | 228 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); |
52fabd73 | 229 | /* Matched by get_layout_hdr in pnfs_insert_layout */ |
cc6e5340 | 230 | put_layout_hdr(NFS_I(ino)->layout); |
974cec8c AA |
231 | } |
232 | ||
4541d16c FI |
233 | /* The use of tmp_list is necessary because pnfs_curr_ld->free_lseg |
234 | * could sleep, so must be called outside of the lock. | |
235 | * Returns 1 if object was removed, otherwise return 0. | |
236 | */ | |
237 | static int | |
238 | put_lseg_locked(struct pnfs_layout_segment *lseg, | |
239 | struct list_head *tmp_list) | |
974cec8c | 240 | { |
4541d16c FI |
241 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, |
242 | atomic_read(&lseg->pls_refcount), | |
243 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
244 | if (atomic_dec_and_test(&lseg->pls_refcount)) { | |
245 | struct inode *ino = lseg->pls_layout->plh_inode; | |
246 | ||
247 | BUG_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
248 | list_del(&lseg->pls_list); | |
249 | if (list_empty(&lseg->pls_layout->plh_segs)) { | |
250 | struct nfs_client *clp; | |
251 | ||
252 | clp = NFS_SERVER(ino)->nfs_client; | |
253 | spin_lock(&clp->cl_lock); | |
254 | /* List does not take a reference, so no need for put here */ | |
255 | list_del_init(&lseg->pls_layout->plh_layouts); | |
256 | spin_unlock(&clp->cl_lock); | |
43f1b3da | 257 | clear_bit(NFS_LAYOUT_BULK_RECALL, &lseg->pls_layout->plh_flags); |
4541d16c | 258 | } |
f7e8917a | 259 | rpc_wake_up(&NFS_SERVER(ino)->roc_rpcwaitq); |
4541d16c FI |
260 | list_add(&lseg->pls_list, tmp_list); |
261 | return 1; | |
262 | } | |
263 | return 0; | |
264 | } | |
974cec8c | 265 | |
4541d16c FI |
266 | static bool |
267 | should_free_lseg(u32 lseg_iomode, u32 recall_iomode) | |
268 | { | |
269 | return (recall_iomode == IOMODE_ANY || | |
270 | lseg_iomode == recall_iomode); | |
974cec8c AA |
271 | } |
272 | ||
4541d16c FI |
273 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
274 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
275 | struct list_head *tmp_list) | |
276 | { | |
277 | int rv = 0; | |
278 | ||
279 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
280 | /* Remove the reference keeping the lseg in the | |
281 | * list. It will now be removed when all | |
282 | * outstanding io is finished. | |
283 | */ | |
284 | rv = put_lseg_locked(lseg, tmp_list); | |
285 | } | |
286 | return rv; | |
287 | } | |
288 | ||
289 | /* Returns count of number of matching invalid lsegs remaining in list | |
290 | * after call. | |
291 | */ | |
43f1b3da | 292 | int |
4541d16c FI |
293 | mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, |
294 | struct list_head *tmp_list, | |
295 | u32 iomode) | |
974cec8c AA |
296 | { |
297 | struct pnfs_layout_segment *lseg, *next; | |
4541d16c | 298 | int invalid = 0, removed = 0; |
974cec8c AA |
299 | |
300 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
301 | ||
4541d16c FI |
302 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
303 | if (should_free_lseg(lseg->pls_range.iomode, iomode)) { | |
304 | dprintk("%s: freeing lseg %p iomode %d " | |
305 | "offset %llu length %llu\n", __func__, | |
306 | lseg, lseg->pls_range.iomode, lseg->pls_range.offset, | |
307 | lseg->pls_range.length); | |
308 | invalid++; | |
309 | removed += mark_lseg_invalid(lseg, tmp_list); | |
310 | } | |
311 | dprintk("%s:Return %i\n", __func__, invalid - removed); | |
312 | return invalid - removed; | |
974cec8c AA |
313 | } |
314 | ||
43f1b3da | 315 | void |
4541d16c | 316 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 317 | { |
4541d16c | 318 | struct pnfs_layout_segment *lseg, *tmp; |
974cec8c | 319 | |
4541d16c | 320 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 321 | list_del(&lseg->pls_list); |
4541d16c | 322 | free_lseg(lseg); |
974cec8c AA |
323 | } |
324 | } | |
325 | ||
e5e94017 BH |
326 | void |
327 | pnfs_destroy_layout(struct nfs_inode *nfsi) | |
328 | { | |
329 | struct pnfs_layout_hdr *lo; | |
974cec8c | 330 | LIST_HEAD(tmp_list); |
e5e94017 BH |
331 | |
332 | spin_lock(&nfsi->vfs_inode.i_lock); | |
333 | lo = nfsi->layout; | |
334 | if (lo) { | |
4541d16c FI |
335 | set_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags); |
336 | mark_matching_lsegs_invalid(lo, &tmp_list, IOMODE_ANY); | |
e5e94017 BH |
337 | /* Matched by refcount set to 1 in alloc_init_layout_hdr */ |
338 | put_layout_hdr_locked(lo); | |
339 | } | |
340 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
974cec8c AA |
341 | pnfs_free_lseg_list(&tmp_list); |
342 | } | |
343 | ||
344 | /* | |
345 | * Called by the state manger to remove all layouts established under an | |
346 | * expired lease. | |
347 | */ | |
348 | void | |
349 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
350 | { | |
351 | struct pnfs_layout_hdr *lo; | |
352 | LIST_HEAD(tmp_list); | |
353 | ||
354 | spin_lock(&clp->cl_lock); | |
355 | list_splice_init(&clp->cl_layouts, &tmp_list); | |
356 | spin_unlock(&clp->cl_lock); | |
357 | ||
358 | while (!list_empty(&tmp_list)) { | |
359 | lo = list_entry(tmp_list.next, struct pnfs_layout_hdr, | |
b7edfaa1 | 360 | plh_layouts); |
974cec8c | 361 | dprintk("%s freeing layout for inode %lu\n", __func__, |
b7edfaa1 FI |
362 | lo->plh_inode->i_ino); |
363 | pnfs_destroy_layout(NFS_I(lo->plh_inode)); | |
974cec8c | 364 | } |
e5e94017 BH |
365 | } |
366 | ||
fd6002e9 | 367 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
368 | void |
369 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
370 | bool update_barrier) | |
b1f69b75 | 371 | { |
fd6002e9 | 372 | u32 oldseq, newseq; |
b1f69b75 | 373 | |
fd6002e9 FI |
374 | oldseq = be32_to_cpu(lo->plh_stateid.stateid.seqid); |
375 | newseq = be32_to_cpu(new->stateid.seqid); | |
43f1b3da | 376 | if ((int)(newseq - oldseq) > 0) { |
fd6002e9 | 377 | memcpy(&lo->plh_stateid, &new->stateid, sizeof(new->stateid)); |
43f1b3da FI |
378 | if (update_barrier) { |
379 | u32 new_barrier = be32_to_cpu(new->stateid.seqid); | |
380 | ||
381 | if ((int)(new_barrier - lo->plh_barrier)) | |
382 | lo->plh_barrier = new_barrier; | |
383 | } else { | |
384 | /* Because of wraparound, we want to keep the barrier | |
385 | * "close" to the current seqids. It needs to be | |
386 | * within 2**31 to count as "behind", so if it | |
387 | * gets too near that limit, give us a litle leeway | |
388 | * and bring it to within 2**30. | |
389 | * NOTE - and yes, this is all unsigned arithmetic. | |
390 | */ | |
391 | if (unlikely((newseq - lo->plh_barrier) > (3 << 29))) | |
392 | lo->plh_barrier = newseq - (1 << 30); | |
393 | } | |
394 | } | |
b1f69b75 AA |
395 | } |
396 | ||
cf7d63f1 FI |
397 | /* lget is set to 1 if called from inside send_layoutget call chain */ |
398 | static bool | |
43f1b3da FI |
399 | pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid, |
400 | int lget) | |
401 | { | |
402 | if ((stateid) && | |
403 | (int)(lo->plh_barrier - be32_to_cpu(stateid->stateid.seqid)) >= 0) | |
404 | return true; | |
f7e8917a FI |
405 | return lo->plh_block_lgets || |
406 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) || | |
43f1b3da | 407 | (list_empty(&lo->plh_segs) && |
cf7d63f1 FI |
408 | (atomic_read(&lo->plh_outstanding) > lget)); |
409 | } | |
410 | ||
fd6002e9 FI |
411 | int |
412 | pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, | |
413 | struct nfs4_state *open_state) | |
b1f69b75 | 414 | { |
fd6002e9 | 415 | int status = 0; |
974cec8c | 416 | |
b1f69b75 | 417 | dprintk("--> %s\n", __func__); |
fd6002e9 | 418 | spin_lock(&lo->plh_inode->i_lock); |
43f1b3da | 419 | if (pnfs_layoutgets_blocked(lo, NULL, 1)) { |
cf7d63f1 FI |
420 | status = -EAGAIN; |
421 | } else if (list_empty(&lo->plh_segs)) { | |
fd6002e9 FI |
422 | int seq; |
423 | ||
424 | do { | |
425 | seq = read_seqbegin(&open_state->seqlock); | |
426 | memcpy(dst->data, open_state->stateid.data, | |
427 | sizeof(open_state->stateid.data)); | |
428 | } while (read_seqretry(&open_state->seqlock, seq)); | |
429 | } else | |
430 | memcpy(dst->data, lo->plh_stateid.data, sizeof(lo->plh_stateid.data)); | |
431 | spin_unlock(&lo->plh_inode->i_lock); | |
b1f69b75 | 432 | dprintk("<-- %s\n", __func__); |
fd6002e9 | 433 | return status; |
b1f69b75 AA |
434 | } |
435 | ||
436 | /* | |
437 | * Get layout from server. | |
438 | * for now, assume that whole file layouts are requested. | |
439 | * arg->offset: 0 | |
440 | * arg->length: all ones | |
441 | */ | |
e5e94017 BH |
442 | static struct pnfs_layout_segment * |
443 | send_layoutget(struct pnfs_layout_hdr *lo, | |
444 | struct nfs_open_context *ctx, | |
445 | u32 iomode) | |
446 | { | |
b7edfaa1 | 447 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
448 | struct nfs_server *server = NFS_SERVER(ino); |
449 | struct nfs4_layoutget *lgp; | |
450 | struct pnfs_layout_segment *lseg = NULL; | |
451 | ||
452 | dprintk("--> %s\n", __func__); | |
e5e94017 | 453 | |
b1f69b75 AA |
454 | BUG_ON(ctx == NULL); |
455 | lgp = kzalloc(sizeof(*lgp), GFP_KERNEL); | |
cf7d63f1 | 456 | if (lgp == NULL) |
b1f69b75 | 457 | return NULL; |
b1f69b75 AA |
458 | lgp->args.minlength = NFS4_MAX_UINT64; |
459 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; | |
460 | lgp->args.range.iomode = iomode; | |
461 | lgp->args.range.offset = 0; | |
462 | lgp->args.range.length = NFS4_MAX_UINT64; | |
463 | lgp->args.type = server->pnfs_curr_ld->id; | |
464 | lgp->args.inode = ino; | |
465 | lgp->args.ctx = get_nfs_open_context(ctx); | |
466 | lgp->lsegpp = &lseg; | |
467 | ||
468 | /* Synchronously retrieve layout information from server and | |
469 | * store in lseg. | |
470 | */ | |
471 | nfs4_proc_layoutget(lgp); | |
974cec8c | 472 | if (!lseg) { |
b1f69b75 | 473 | /* remember that LAYOUTGET failed and suspend trying */ |
566052c5 | 474 | set_bit(lo_fail_bit(iomode), &lo->plh_flags); |
974cec8c | 475 | } |
974cec8c AA |
476 | return lseg; |
477 | } | |
478 | ||
f7e8917a FI |
479 | bool pnfs_roc(struct inode *ino) |
480 | { | |
481 | struct pnfs_layout_hdr *lo; | |
482 | struct pnfs_layout_segment *lseg, *tmp; | |
483 | LIST_HEAD(tmp_list); | |
484 | bool found = false; | |
485 | ||
486 | spin_lock(&ino->i_lock); | |
487 | lo = NFS_I(ino)->layout; | |
488 | if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) || | |
489 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) | |
490 | goto out_nolayout; | |
491 | list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) | |
492 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
493 | mark_lseg_invalid(lseg, &tmp_list); | |
494 | found = true; | |
495 | } | |
496 | if (!found) | |
497 | goto out_nolayout; | |
498 | lo->plh_block_lgets++; | |
499 | get_layout_hdr(lo); /* matched in pnfs_roc_release */ | |
500 | spin_unlock(&ino->i_lock); | |
501 | pnfs_free_lseg_list(&tmp_list); | |
502 | return true; | |
503 | ||
504 | out_nolayout: | |
505 | spin_unlock(&ino->i_lock); | |
506 | return false; | |
507 | } | |
508 | ||
509 | void pnfs_roc_release(struct inode *ino) | |
510 | { | |
511 | struct pnfs_layout_hdr *lo; | |
512 | ||
513 | spin_lock(&ino->i_lock); | |
514 | lo = NFS_I(ino)->layout; | |
515 | lo->plh_block_lgets--; | |
516 | put_layout_hdr_locked(lo); | |
517 | spin_unlock(&ino->i_lock); | |
518 | } | |
519 | ||
520 | void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) | |
521 | { | |
522 | struct pnfs_layout_hdr *lo; | |
523 | ||
524 | spin_lock(&ino->i_lock); | |
525 | lo = NFS_I(ino)->layout; | |
526 | if ((int)(barrier - lo->plh_barrier) > 0) | |
527 | lo->plh_barrier = barrier; | |
528 | spin_unlock(&ino->i_lock); | |
529 | } | |
530 | ||
531 | bool pnfs_roc_drain(struct inode *ino, u32 *barrier) | |
532 | { | |
533 | struct nfs_inode *nfsi = NFS_I(ino); | |
534 | struct pnfs_layout_segment *lseg; | |
535 | bool found = false; | |
536 | ||
537 | spin_lock(&ino->i_lock); | |
538 | list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list) | |
539 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
540 | found = true; | |
541 | break; | |
542 | } | |
543 | if (!found) { | |
544 | struct pnfs_layout_hdr *lo = nfsi->layout; | |
545 | u32 current_seqid = be32_to_cpu(lo->plh_stateid.stateid.seqid); | |
546 | ||
547 | /* Since close does not return a layout stateid for use as | |
548 | * a barrier, we choose the worst-case barrier. | |
549 | */ | |
550 | *barrier = current_seqid + atomic_read(&lo->plh_outstanding); | |
551 | } | |
552 | spin_unlock(&ino->i_lock); | |
553 | return found; | |
554 | } | |
555 | ||
b1f69b75 AA |
556 | /* |
557 | * Compare two layout segments for sorting into layout cache. | |
558 | * We want to preferentially return RW over RO layouts, so ensure those | |
559 | * are seen first. | |
560 | */ | |
561 | static s64 | |
562 | cmp_layout(u32 iomode1, u32 iomode2) | |
563 | { | |
564 | /* read > read/write */ | |
565 | return (int)(iomode2 == IOMODE_READ) - (int)(iomode1 == IOMODE_READ); | |
566 | } | |
567 | ||
974cec8c AA |
568 | static void |
569 | pnfs_insert_layout(struct pnfs_layout_hdr *lo, | |
570 | struct pnfs_layout_segment *lseg) | |
571 | { | |
b1f69b75 AA |
572 | struct pnfs_layout_segment *lp; |
573 | int found = 0; | |
574 | ||
974cec8c AA |
575 | dprintk("%s:Begin\n", __func__); |
576 | ||
b7edfaa1 | 577 | assert_spin_locked(&lo->plh_inode->i_lock); |
b7edfaa1 | 578 | list_for_each_entry(lp, &lo->plh_segs, pls_list) { |
566052c5 | 579 | if (cmp_layout(lp->pls_range.iomode, lseg->pls_range.iomode) > 0) |
b1f69b75 | 580 | continue; |
566052c5 | 581 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
582 | dprintk("%s: inserted lseg %p " |
583 | "iomode %d offset %llu length %llu before " | |
584 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
585 | __func__, lseg, lseg->pls_range.iomode, |
586 | lseg->pls_range.offset, lseg->pls_range.length, | |
587 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
588 | lp->pls_range.length); | |
b1f69b75 AA |
589 | found = 1; |
590 | break; | |
591 | } | |
592 | if (!found) { | |
b7edfaa1 | 593 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
b1f69b75 AA |
594 | dprintk("%s: inserted lseg %p " |
595 | "iomode %d offset %llu length %llu at tail\n", | |
566052c5 FI |
596 | __func__, lseg, lseg->pls_range.iomode, |
597 | lseg->pls_range.offset, lseg->pls_range.length); | |
974cec8c | 598 | } |
cc6e5340 | 599 | get_layout_hdr(lo); |
974cec8c AA |
600 | |
601 | dprintk("%s:Return\n", __func__); | |
e5e94017 BH |
602 | } |
603 | ||
604 | static struct pnfs_layout_hdr * | |
605 | alloc_init_layout_hdr(struct inode *ino) | |
606 | { | |
607 | struct pnfs_layout_hdr *lo; | |
608 | ||
609 | lo = kzalloc(sizeof(struct pnfs_layout_hdr), GFP_KERNEL); | |
610 | if (!lo) | |
611 | return NULL; | |
cc6e5340 | 612 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
613 | INIT_LIST_HEAD(&lo->plh_layouts); |
614 | INIT_LIST_HEAD(&lo->plh_segs); | |
43f1b3da | 615 | INIT_LIST_HEAD(&lo->plh_bulk_recall); |
b7edfaa1 | 616 | lo->plh_inode = ino; |
e5e94017 BH |
617 | return lo; |
618 | } | |
619 | ||
620 | static struct pnfs_layout_hdr * | |
621 | pnfs_find_alloc_layout(struct inode *ino) | |
622 | { | |
623 | struct nfs_inode *nfsi = NFS_I(ino); | |
624 | struct pnfs_layout_hdr *new = NULL; | |
625 | ||
626 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
627 | ||
628 | assert_spin_locked(&ino->i_lock); | |
4541d16c FI |
629 | if (nfsi->layout) { |
630 | if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags)) | |
631 | return NULL; | |
632 | else | |
633 | return nfsi->layout; | |
634 | } | |
e5e94017 BH |
635 | spin_unlock(&ino->i_lock); |
636 | new = alloc_init_layout_hdr(ino); | |
637 | spin_lock(&ino->i_lock); | |
638 | ||
639 | if (likely(nfsi->layout == NULL)) /* Won the race? */ | |
640 | nfsi->layout = new; | |
641 | else | |
642 | kfree(new); | |
643 | return nfsi->layout; | |
644 | } | |
645 | ||
b1f69b75 AA |
646 | /* |
647 | * iomode matching rules: | |
648 | * iomode lseg match | |
649 | * ----- ----- ----- | |
650 | * ANY READ true | |
651 | * ANY RW true | |
652 | * RW READ false | |
653 | * RW RW true | |
654 | * READ READ true | |
655 | * READ RW true | |
656 | */ | |
657 | static int | |
658 | is_matching_lseg(struct pnfs_layout_segment *lseg, u32 iomode) | |
659 | { | |
566052c5 | 660 | return (iomode != IOMODE_RW || lseg->pls_range.iomode == IOMODE_RW); |
b1f69b75 AA |
661 | } |
662 | ||
663 | /* | |
664 | * lookup range in layout | |
665 | */ | |
e5e94017 | 666 | static struct pnfs_layout_segment * |
43f1b3da | 667 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, u32 iomode) |
e5e94017 | 668 | { |
b1f69b75 AA |
669 | struct pnfs_layout_segment *lseg, *ret = NULL; |
670 | ||
671 | dprintk("%s:Begin\n", __func__); | |
672 | ||
b7edfaa1 FI |
673 | assert_spin_locked(&lo->plh_inode->i_lock); |
674 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { | |
4541d16c FI |
675 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
676 | is_matching_lseg(lseg, iomode)) { | |
b1f69b75 AA |
677 | ret = lseg; |
678 | break; | |
679 | } | |
566052c5 | 680 | if (cmp_layout(iomode, lseg->pls_range.iomode) > 0) |
b1f69b75 AA |
681 | break; |
682 | } | |
683 | ||
684 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 685 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 686 | return ret; |
e5e94017 BH |
687 | } |
688 | ||
689 | /* | |
690 | * Layout segment is retreived from the server if not cached. | |
691 | * The appropriate layout segment is referenced and returned to the caller. | |
692 | */ | |
693 | struct pnfs_layout_segment * | |
694 | pnfs_update_layout(struct inode *ino, | |
695 | struct nfs_open_context *ctx, | |
696 | enum pnfs_iomode iomode) | |
697 | { | |
698 | struct nfs_inode *nfsi = NFS_I(ino); | |
2130ff66 | 699 | struct nfs_client *clp = NFS_SERVER(ino)->nfs_client; |
e5e94017 BH |
700 | struct pnfs_layout_hdr *lo; |
701 | struct pnfs_layout_segment *lseg = NULL; | |
702 | ||
703 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) | |
704 | return NULL; | |
705 | spin_lock(&ino->i_lock); | |
706 | lo = pnfs_find_alloc_layout(ino); | |
707 | if (lo == NULL) { | |
708 | dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__); | |
709 | goto out_unlock; | |
710 | } | |
711 | ||
43f1b3da FI |
712 | /* Do we even need to bother with this? */ |
713 | if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) || | |
714 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { | |
715 | dprintk("%s matches recall, use MDS\n", __func__); | |
e5e94017 BH |
716 | goto out_unlock; |
717 | } | |
43f1b3da FI |
718 | /* Check to see if the layout for the given range already exists */ |
719 | lseg = pnfs_find_lseg(lo, iomode); | |
720 | if (lseg) | |
721 | goto out_unlock; | |
e5e94017 BH |
722 | |
723 | /* if LAYOUTGET already failed once we don't try again */ | |
566052c5 | 724 | if (test_bit(lo_fail_bit(iomode), &nfsi->layout->plh_flags)) |
e5e94017 BH |
725 | goto out_unlock; |
726 | ||
43f1b3da | 727 | if (pnfs_layoutgets_blocked(lo, NULL, 0)) |
cf7d63f1 FI |
728 | goto out_unlock; |
729 | atomic_inc(&lo->plh_outstanding); | |
730 | ||
cc6e5340 | 731 | get_layout_hdr(lo); |
2130ff66 FI |
732 | if (list_empty(&lo->plh_segs)) { |
733 | /* The lo must be on the clp list if there is any | |
734 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
735 | */ | |
736 | spin_lock(&clp->cl_lock); | |
737 | BUG_ON(!list_empty(&lo->plh_layouts)); | |
738 | list_add_tail(&lo->plh_layouts, &clp->cl_layouts); | |
739 | spin_unlock(&clp->cl_lock); | |
740 | } | |
e5e94017 BH |
741 | spin_unlock(&ino->i_lock); |
742 | ||
743 | lseg = send_layoutget(lo, ctx, iomode); | |
2130ff66 FI |
744 | if (!lseg) { |
745 | spin_lock(&ino->i_lock); | |
746 | if (list_empty(&lo->plh_segs)) { | |
747 | spin_lock(&clp->cl_lock); | |
748 | list_del_init(&lo->plh_layouts); | |
749 | spin_unlock(&clp->cl_lock); | |
43f1b3da | 750 | clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); |
2130ff66 FI |
751 | } |
752 | spin_unlock(&ino->i_lock); | |
753 | } | |
cf7d63f1 | 754 | atomic_dec(&lo->plh_outstanding); |
cc6e5340 | 755 | put_layout_hdr(lo); |
e5e94017 BH |
756 | out: |
757 | dprintk("%s end, state 0x%lx lseg %p\n", __func__, | |
566052c5 | 758 | nfsi->layout->plh_flags, lseg); |
e5e94017 BH |
759 | return lseg; |
760 | out_unlock: | |
761 | spin_unlock(&ino->i_lock); | |
762 | goto out; | |
763 | } | |
b1f69b75 AA |
764 | |
765 | int | |
766 | pnfs_layout_process(struct nfs4_layoutget *lgp) | |
767 | { | |
768 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
769 | struct nfs4_layoutget_res *res = &lgp->res; | |
770 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 771 | struct inode *ino = lo->plh_inode; |
43f1b3da | 772 | struct nfs_client *clp = NFS_SERVER(ino)->nfs_client; |
b1f69b75 AA |
773 | int status = 0; |
774 | ||
fc1794c5 FI |
775 | /* Verify we got what we asked for. |
776 | * Note that because the xdr parsing only accepts a single | |
777 | * element array, this can fail even if the server is behaving | |
778 | * correctly. | |
779 | */ | |
780 | if (lgp->args.range.iomode > res->range.iomode || | |
781 | res->range.offset != 0 || | |
782 | res->range.length != NFS4_MAX_UINT64) { | |
783 | status = -EINVAL; | |
784 | goto out; | |
785 | } | |
b1f69b75 AA |
786 | /* Inject layout blob into I/O device driver */ |
787 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res); | |
788 | if (!lseg || IS_ERR(lseg)) { | |
789 | if (!lseg) | |
790 | status = -ENOMEM; | |
791 | else | |
792 | status = PTR_ERR(lseg); | |
793 | dprintk("%s: Could not allocate layout: error %d\n", | |
794 | __func__, status); | |
795 | goto out; | |
796 | } | |
797 | ||
798 | spin_lock(&ino->i_lock); | |
43f1b3da FI |
799 | if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) || |
800 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { | |
801 | dprintk("%s forget reply due to recall\n", __func__); | |
802 | goto out_forget_reply; | |
803 | } | |
804 | ||
805 | if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) { | |
806 | dprintk("%s forget reply due to state\n", __func__); | |
807 | goto out_forget_reply; | |
808 | } | |
b1f69b75 | 809 | init_lseg(lo, lseg); |
566052c5 | 810 | lseg->pls_range = res->range; |
b1f69b75 AA |
811 | *lgp->lsegpp = lseg; |
812 | pnfs_insert_layout(lo, lseg); | |
813 | ||
f7e8917a FI |
814 | if (res->return_on_close) { |
815 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); | |
816 | set_bit(NFS_LAYOUT_ROC, &lo->plh_flags); | |
817 | } | |
818 | ||
b1f69b75 | 819 | /* Done processing layoutget. Set the layout stateid */ |
43f1b3da | 820 | pnfs_set_layout_stateid(lo, &res->stateid, false); |
b1f69b75 AA |
821 | spin_unlock(&ino->i_lock); |
822 | out: | |
823 | return status; | |
43f1b3da FI |
824 | |
825 | out_forget_reply: | |
826 | spin_unlock(&ino->i_lock); | |
827 | lseg->pls_layout = lo; | |
828 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
829 | goto out; | |
b1f69b75 AA |
830 | } |
831 | ||
832 | /* | |
833 | * Device ID cache. Currently supports one layout type per struct nfs_client. | |
834 | * Add layout type to the lookup key to expand to support multiple types. | |
835 | */ | |
836 | int | |
837 | pnfs_alloc_init_deviceid_cache(struct nfs_client *clp, | |
838 | void (*free_callback)(struct pnfs_deviceid_node *)) | |
839 | { | |
840 | struct pnfs_deviceid_cache *c; | |
841 | ||
842 | c = kzalloc(sizeof(struct pnfs_deviceid_cache), GFP_KERNEL); | |
843 | if (!c) | |
844 | return -ENOMEM; | |
845 | spin_lock(&clp->cl_lock); | |
846 | if (clp->cl_devid_cache != NULL) { | |
847 | atomic_inc(&clp->cl_devid_cache->dc_ref); | |
848 | dprintk("%s [kref [%d]]\n", __func__, | |
849 | atomic_read(&clp->cl_devid_cache->dc_ref)); | |
850 | kfree(c); | |
851 | } else { | |
852 | /* kzalloc initializes hlists */ | |
853 | spin_lock_init(&c->dc_lock); | |
854 | atomic_set(&c->dc_ref, 1); | |
855 | c->dc_free_callback = free_callback; | |
856 | clp->cl_devid_cache = c; | |
857 | dprintk("%s [new]\n", __func__); | |
858 | } | |
859 | spin_unlock(&clp->cl_lock); | |
860 | return 0; | |
861 | } | |
862 | EXPORT_SYMBOL_GPL(pnfs_alloc_init_deviceid_cache); | |
863 | ||
864 | /* | |
865 | * Called from pnfs_layoutdriver_type->free_lseg | |
866 | * last layout segment reference frees deviceid | |
867 | */ | |
868 | void | |
869 | pnfs_put_deviceid(struct pnfs_deviceid_cache *c, | |
870 | struct pnfs_deviceid_node *devid) | |
871 | { | |
872 | struct nfs4_deviceid *id = &devid->de_id; | |
873 | struct pnfs_deviceid_node *d; | |
874 | struct hlist_node *n; | |
875 | long h = nfs4_deviceid_hash(id); | |
876 | ||
877 | dprintk("%s [%d]\n", __func__, atomic_read(&devid->de_ref)); | |
878 | if (!atomic_dec_and_lock(&devid->de_ref, &c->dc_lock)) | |
879 | return; | |
880 | ||
881 | hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[h], de_node) | |
882 | if (!memcmp(&d->de_id, id, sizeof(*id))) { | |
883 | hlist_del_rcu(&d->de_node); | |
884 | spin_unlock(&c->dc_lock); | |
885 | synchronize_rcu(); | |
886 | c->dc_free_callback(devid); | |
887 | return; | |
888 | } | |
889 | spin_unlock(&c->dc_lock); | |
890 | /* Why wasn't it found in the list? */ | |
891 | BUG(); | |
892 | } | |
893 | EXPORT_SYMBOL_GPL(pnfs_put_deviceid); | |
894 | ||
895 | /* Find and reference a deviceid */ | |
896 | struct pnfs_deviceid_node * | |
897 | pnfs_find_get_deviceid(struct pnfs_deviceid_cache *c, struct nfs4_deviceid *id) | |
898 | { | |
899 | struct pnfs_deviceid_node *d; | |
900 | struct hlist_node *n; | |
901 | long hash = nfs4_deviceid_hash(id); | |
902 | ||
903 | dprintk("--> %s hash %ld\n", __func__, hash); | |
904 | rcu_read_lock(); | |
905 | hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[hash], de_node) { | |
906 | if (!memcmp(&d->de_id, id, sizeof(*id))) { | |
907 | if (!atomic_inc_not_zero(&d->de_ref)) { | |
908 | goto fail; | |
909 | } else { | |
910 | rcu_read_unlock(); | |
911 | return d; | |
912 | } | |
913 | } | |
914 | } | |
915 | fail: | |
916 | rcu_read_unlock(); | |
917 | return NULL; | |
918 | } | |
919 | EXPORT_SYMBOL_GPL(pnfs_find_get_deviceid); | |
920 | ||
921 | /* | |
922 | * Add a deviceid to the cache. | |
923 | * GETDEVICEINFOs for same deviceid can race. If deviceid is found, discard new | |
924 | */ | |
925 | struct pnfs_deviceid_node * | |
926 | pnfs_add_deviceid(struct pnfs_deviceid_cache *c, struct pnfs_deviceid_node *new) | |
927 | { | |
928 | struct pnfs_deviceid_node *d; | |
929 | long hash = nfs4_deviceid_hash(&new->de_id); | |
930 | ||
931 | dprintk("--> %s hash %ld\n", __func__, hash); | |
932 | spin_lock(&c->dc_lock); | |
933 | d = pnfs_find_get_deviceid(c, &new->de_id); | |
934 | if (d) { | |
935 | spin_unlock(&c->dc_lock); | |
936 | dprintk("%s [discard]\n", __func__); | |
937 | c->dc_free_callback(new); | |
938 | return d; | |
939 | } | |
940 | INIT_HLIST_NODE(&new->de_node); | |
941 | atomic_set(&new->de_ref, 1); | |
942 | hlist_add_head_rcu(&new->de_node, &c->dc_deviceids[hash]); | |
943 | spin_unlock(&c->dc_lock); | |
944 | dprintk("%s [new]\n", __func__); | |
945 | return new; | |
946 | } | |
947 | EXPORT_SYMBOL_GPL(pnfs_add_deviceid); | |
948 | ||
949 | void | |
950 | pnfs_put_deviceid_cache(struct nfs_client *clp) | |
951 | { | |
952 | struct pnfs_deviceid_cache *local = clp->cl_devid_cache; | |
953 | ||
b2a2897d | 954 | dprintk("--> %s ({%d})\n", __func__, atomic_read(&local->dc_ref)); |
b1f69b75 AA |
955 | if (atomic_dec_and_lock(&local->dc_ref, &clp->cl_lock)) { |
956 | int i; | |
957 | /* Verify cache is empty */ | |
958 | for (i = 0; i < NFS4_DEVICE_ID_HASH_SIZE; i++) | |
959 | BUG_ON(!hlist_empty(&local->dc_deviceids[i])); | |
960 | clp->cl_devid_cache = NULL; | |
961 | spin_unlock(&clp->cl_lock); | |
962 | kfree(local); | |
963 | } | |
964 | } | |
965 | EXPORT_SYMBOL_GPL(pnfs_put_deviceid_cache); |