]>
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> | |
493292dd | 31 | #include <linux/nfs_page.h> |
143cb494 | 32 | #include <linux/module.h> |
974cec8c | 33 | #include "internal.h" |
85e174ba | 34 | #include "pnfs.h" |
64419a9b | 35 | #include "iostat.h" |
85e174ba RL |
36 | |
37 | #define NFSDBG_FACILITY NFSDBG_PNFS | |
38 | ||
02c35fca FI |
39 | /* Locking: |
40 | * | |
41 | * pnfs_spinlock: | |
42 | * protects pnfs_modules_tbl. | |
43 | */ | |
44 | static DEFINE_SPINLOCK(pnfs_spinlock); | |
45 | ||
46 | /* | |
47 | * pnfs_modules_tbl holds all pnfs modules | |
48 | */ | |
49 | static LIST_HEAD(pnfs_modules_tbl); | |
50 | ||
51 | /* Return the registered pnfs layout driver module matching given id */ | |
52 | static struct pnfs_layoutdriver_type * | |
53 | find_pnfs_driver_locked(u32 id) | |
54 | { | |
55 | struct pnfs_layoutdriver_type *local; | |
56 | ||
57 | list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid) | |
58 | if (local->id == id) | |
59 | goto out; | |
60 | local = NULL; | |
61 | out: | |
62 | dprintk("%s: Searching for id %u, found %p\n", __func__, id, local); | |
63 | return local; | |
64 | } | |
65 | ||
85e174ba RL |
66 | static struct pnfs_layoutdriver_type * |
67 | find_pnfs_driver(u32 id) | |
68 | { | |
02c35fca FI |
69 | struct pnfs_layoutdriver_type *local; |
70 | ||
71 | spin_lock(&pnfs_spinlock); | |
72 | local = find_pnfs_driver_locked(id); | |
0a9c63fa TM |
73 | if (local != NULL && !try_module_get(local->owner)) { |
74 | dprintk("%s: Could not grab reference on module\n", __func__); | |
75 | local = NULL; | |
76 | } | |
02c35fca FI |
77 | spin_unlock(&pnfs_spinlock); |
78 | return local; | |
85e174ba RL |
79 | } |
80 | ||
81 | void | |
82 | unset_pnfs_layoutdriver(struct nfs_server *nfss) | |
83 | { | |
738fd0f3 BH |
84 | if (nfss->pnfs_curr_ld) { |
85 | if (nfss->pnfs_curr_ld->clear_layoutdriver) | |
86 | nfss->pnfs_curr_ld->clear_layoutdriver(nfss); | |
2a4c8994 TM |
87 | /* Decrement the MDS count. Purge the deviceid cache if zero */ |
88 | if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count)) | |
89 | nfs4_deviceid_purge_client(nfss->nfs_client); | |
02c35fca | 90 | module_put(nfss->pnfs_curr_ld->owner); |
738fd0f3 | 91 | } |
85e174ba RL |
92 | nfss->pnfs_curr_ld = NULL; |
93 | } | |
94 | ||
95 | /* | |
96 | * Try to set the server's pnfs module to the pnfs layout type specified by id. | |
97 | * Currently only one pNFS layout driver per filesystem is supported. | |
98 | * | |
99 | * @id layout type. Zero (illegal layout type) indicates pNFS not in use. | |
100 | */ | |
101 | void | |
738fd0f3 BH |
102 | set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh, |
103 | u32 id) | |
85e174ba RL |
104 | { |
105 | struct pnfs_layoutdriver_type *ld_type = NULL; | |
106 | ||
107 | if (id == 0) | |
108 | goto out_no_driver; | |
109 | if (!(server->nfs_client->cl_exchange_flags & | |
110 | (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) { | |
a030889a WAA |
111 | printk(KERN_ERR "NFS: %s: id %u cl_exchange_flags 0x%x\n", |
112 | __func__, id, server->nfs_client->cl_exchange_flags); | |
85e174ba RL |
113 | goto out_no_driver; |
114 | } | |
115 | ld_type = find_pnfs_driver(id); | |
116 | if (!ld_type) { | |
117 | request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id); | |
118 | ld_type = find_pnfs_driver(id); | |
119 | if (!ld_type) { | |
120 | dprintk("%s: No pNFS module found for %u.\n", | |
121 | __func__, id); | |
122 | goto out_no_driver; | |
123 | } | |
124 | } | |
125 | server->pnfs_curr_ld = ld_type; | |
738fd0f3 BH |
126 | if (ld_type->set_layoutdriver |
127 | && ld_type->set_layoutdriver(server, mntfh)) { | |
a030889a WAA |
128 | printk(KERN_ERR "NFS: %s: Error initializing pNFS layout " |
129 | "driver %u.\n", __func__, id); | |
738fd0f3 BH |
130 | module_put(ld_type->owner); |
131 | goto out_no_driver; | |
132 | } | |
2a4c8994 TM |
133 | /* Bump the MDS count */ |
134 | atomic_inc(&server->nfs_client->cl_mds_count); | |
ea8eecdd | 135 | |
85e174ba RL |
136 | dprintk("%s: pNFS module for %u set\n", __func__, id); |
137 | return; | |
138 | ||
139 | out_no_driver: | |
140 | dprintk("%s: Using NFSv4 I/O\n", __func__); | |
141 | server->pnfs_curr_ld = NULL; | |
142 | } | |
02c35fca FI |
143 | |
144 | int | |
145 | pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
146 | { | |
147 | int status = -EINVAL; | |
148 | struct pnfs_layoutdriver_type *tmp; | |
149 | ||
150 | if (ld_type->id == 0) { | |
a030889a | 151 | printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__); |
02c35fca FI |
152 | return status; |
153 | } | |
b1f69b75 | 154 | if (!ld_type->alloc_lseg || !ld_type->free_lseg) { |
a030889a | 155 | printk(KERN_ERR "NFS: %s Layout driver must provide " |
b1f69b75 AA |
156 | "alloc_lseg and free_lseg.\n", __func__); |
157 | return status; | |
158 | } | |
02c35fca FI |
159 | |
160 | spin_lock(&pnfs_spinlock); | |
161 | tmp = find_pnfs_driver_locked(ld_type->id); | |
162 | if (!tmp) { | |
163 | list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl); | |
164 | status = 0; | |
165 | dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id, | |
166 | ld_type->name); | |
167 | } else { | |
a030889a | 168 | printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n", |
02c35fca FI |
169 | __func__, ld_type->id); |
170 | } | |
171 | spin_unlock(&pnfs_spinlock); | |
172 | ||
173 | return status; | |
174 | } | |
175 | EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver); | |
176 | ||
177 | void | |
178 | pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
179 | { | |
180 | dprintk("%s Deregistering id:%u\n", __func__, ld_type->id); | |
181 | spin_lock(&pnfs_spinlock); | |
182 | list_del(&ld_type->pnfs_tblid); | |
183 | spin_unlock(&pnfs_spinlock); | |
184 | } | |
185 | EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver); | |
e5e94017 | 186 | |
b1f69b75 AA |
187 | /* |
188 | * pNFS client layout cache | |
189 | */ | |
190 | ||
cc6e5340 | 191 | /* Need to hold i_lock if caller does not already hold reference */ |
43f1b3da | 192 | void |
70c3bd2b | 193 | pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 194 | { |
cc6e5340 | 195 | atomic_inc(&lo->plh_refcount); |
e5e94017 BH |
196 | } |
197 | ||
636fb9c8 BH |
198 | static struct pnfs_layout_hdr * |
199 | pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags) | |
200 | { | |
201 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld; | |
202 | return ld->alloc_layout_hdr ? ld->alloc_layout_hdr(ino, gfp_flags) : | |
203 | kzalloc(sizeof(struct pnfs_layout_hdr), gfp_flags); | |
204 | } | |
205 | ||
206 | static void | |
207 | pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo) | |
208 | { | |
209 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(lo->plh_inode)->pnfs_curr_ld; | |
9fa40758 | 210 | put_rpccred(lo->plh_lc_cred); |
636fb9c8 BH |
211 | return ld->alloc_layout_hdr ? ld->free_layout_hdr(lo) : kfree(lo); |
212 | } | |
213 | ||
e5e94017 | 214 | static void |
cc6e5340 | 215 | destroy_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 216 | { |
cc6e5340 FI |
217 | dprintk("%s: freeing layout cache %p\n", __func__, lo); |
218 | BUG_ON(!list_empty(&lo->plh_layouts)); | |
219 | NFS_I(lo->plh_inode)->layout = NULL; | |
636fb9c8 | 220 | pnfs_free_layout_hdr(lo); |
cc6e5340 | 221 | } |
e5e94017 | 222 | |
cc6e5340 | 223 | static void |
70c3bd2b | 224 | pnfs_put_layout_hdr_locked(struct pnfs_layout_hdr *lo) |
cc6e5340 FI |
225 | { |
226 | if (atomic_dec_and_test(&lo->plh_refcount)) | |
227 | destroy_layout_hdr(lo); | |
e5e94017 BH |
228 | } |
229 | ||
b1f69b75 | 230 | void |
70c3bd2b | 231 | pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo) |
974cec8c | 232 | { |
cc6e5340 FI |
233 | struct inode *inode = lo->plh_inode; |
234 | ||
235 | if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) { | |
236 | destroy_layout_hdr(lo); | |
237 | spin_unlock(&inode->i_lock); | |
238 | } | |
974cec8c AA |
239 | } |
240 | ||
241 | static void | |
242 | init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg) | |
243 | { | |
566052c5 | 244 | INIT_LIST_HEAD(&lseg->pls_list); |
a9bae566 | 245 | INIT_LIST_HEAD(&lseg->pls_lc_list); |
4541d16c FI |
246 | atomic_set(&lseg->pls_refcount, 1); |
247 | smp_mb(); | |
248 | set_bit(NFS_LSEG_VALID, &lseg->pls_flags); | |
566052c5 | 249 | lseg->pls_layout = lo; |
974cec8c AA |
250 | } |
251 | ||
4541d16c | 252 | static void free_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 253 | { |
b7edfaa1 | 254 | struct inode *ino = lseg->pls_layout->plh_inode; |
974cec8c | 255 | |
b1f69b75 | 256 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); |
70c3bd2b TM |
257 | /* Matched by pnfs_get_layout_hdr in pnfs_insert_layout */ |
258 | pnfs_put_layout_hdr(NFS_I(ino)->layout); | |
974cec8c AA |
259 | } |
260 | ||
d684d2ae | 261 | static void |
9369a431 | 262 | pnfs_put_lseg_common(struct pnfs_layout_segment *lseg) |
d684d2ae FI |
263 | { |
264 | struct inode *inode = lseg->pls_layout->plh_inode; | |
265 | ||
d20581aa | 266 | WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); |
d684d2ae FI |
267 | list_del_init(&lseg->pls_list); |
268 | if (list_empty(&lseg->pls_layout->plh_segs)) { | |
269 | set_bit(NFS_LAYOUT_DESTROYED, &lseg->pls_layout->plh_flags); | |
270 | /* Matched by initial refcount set in alloc_init_layout_hdr */ | |
70c3bd2b | 271 | pnfs_put_layout_hdr_locked(lseg->pls_layout); |
d684d2ae FI |
272 | } |
273 | rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq); | |
274 | } | |
275 | ||
bae724ef | 276 | void |
9369a431 | 277 | pnfs_put_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 278 | { |
d684d2ae FI |
279 | struct inode *inode; |
280 | ||
281 | if (!lseg) | |
282 | return; | |
283 | ||
4541d16c FI |
284 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, |
285 | atomic_read(&lseg->pls_refcount), | |
286 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); | |
d684d2ae FI |
287 | inode = lseg->pls_layout->plh_inode; |
288 | if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) { | |
289 | LIST_HEAD(free_me); | |
4541d16c | 290 | |
9369a431 | 291 | pnfs_put_lseg_common(lseg); |
d684d2ae FI |
292 | list_add(&lseg->pls_list, &free_me); |
293 | spin_unlock(&inode->i_lock); | |
294 | pnfs_free_lseg_list(&free_me); | |
4541d16c | 295 | } |
4541d16c | 296 | } |
9369a431 | 297 | EXPORT_SYMBOL_GPL(pnfs_put_lseg); |
974cec8c | 298 | |
fb3296eb BH |
299 | static inline u64 |
300 | end_offset(u64 start, u64 len) | |
301 | { | |
302 | u64 end; | |
303 | ||
304 | end = start + len; | |
305 | return end >= start ? end : NFS4_MAX_UINT64; | |
306 | } | |
307 | ||
308 | /* last octet in a range */ | |
309 | static inline u64 | |
310 | last_byte_offset(u64 start, u64 len) | |
311 | { | |
312 | u64 end; | |
313 | ||
314 | BUG_ON(!len); | |
315 | end = start + len; | |
316 | return end > start ? end - 1 : NFS4_MAX_UINT64; | |
317 | } | |
318 | ||
319 | /* | |
320 | * is l2 fully contained in l1? | |
321 | * start1 end1 | |
322 | * [----------------------------------) | |
323 | * start2 end2 | |
324 | * [----------------) | |
325 | */ | |
326 | static inline int | |
327 | lo_seg_contained(struct pnfs_layout_range *l1, | |
328 | struct pnfs_layout_range *l2) | |
329 | { | |
330 | u64 start1 = l1->offset; | |
331 | u64 end1 = end_offset(start1, l1->length); | |
332 | u64 start2 = l2->offset; | |
333 | u64 end2 = end_offset(start2, l2->length); | |
334 | ||
335 | return (start1 <= start2) && (end1 >= end2); | |
336 | } | |
337 | ||
338 | /* | |
339 | * is l1 and l2 intersecting? | |
340 | * start1 end1 | |
341 | * [----------------------------------) | |
342 | * start2 end2 | |
343 | * [----------------) | |
344 | */ | |
345 | static inline int | |
346 | lo_seg_intersecting(struct pnfs_layout_range *l1, | |
347 | struct pnfs_layout_range *l2) | |
348 | { | |
349 | u64 start1 = l1->offset; | |
350 | u64 end1 = end_offset(start1, l1->length); | |
351 | u64 start2 = l2->offset; | |
352 | u64 end2 = end_offset(start2, l2->length); | |
353 | ||
354 | return (end1 == NFS4_MAX_UINT64 || end1 > start2) && | |
355 | (end2 == NFS4_MAX_UINT64 || end2 > start1); | |
356 | } | |
357 | ||
4541d16c | 358 | static bool |
778b5502 BH |
359 | should_free_lseg(struct pnfs_layout_range *lseg_range, |
360 | struct pnfs_layout_range *recall_range) | |
4541d16c | 361 | { |
778b5502 BH |
362 | return (recall_range->iomode == IOMODE_ANY || |
363 | lseg_range->iomode == recall_range->iomode) && | |
364 | lo_seg_intersecting(lseg_range, recall_range); | |
974cec8c AA |
365 | } |
366 | ||
4541d16c FI |
367 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
368 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
369 | struct list_head *tmp_list) | |
370 | { | |
371 | int rv = 0; | |
372 | ||
373 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
374 | /* Remove the reference keeping the lseg in the | |
375 | * list. It will now be removed when all | |
376 | * outstanding io is finished. | |
377 | */ | |
d684d2ae FI |
378 | dprintk("%s: lseg %p ref %d\n", __func__, lseg, |
379 | atomic_read(&lseg->pls_refcount)); | |
380 | if (atomic_dec_and_test(&lseg->pls_refcount)) { | |
9369a431 | 381 | pnfs_put_lseg_common(lseg); |
d684d2ae FI |
382 | list_add(&lseg->pls_list, tmp_list); |
383 | rv = 1; | |
384 | } | |
4541d16c FI |
385 | } |
386 | return rv; | |
387 | } | |
388 | ||
389 | /* Returns count of number of matching invalid lsegs remaining in list | |
390 | * after call. | |
391 | */ | |
43f1b3da | 392 | int |
49a85061 | 393 | pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, |
4541d16c | 394 | struct list_head *tmp_list, |
778b5502 | 395 | struct pnfs_layout_range *recall_range) |
974cec8c AA |
396 | { |
397 | struct pnfs_layout_segment *lseg, *next; | |
4541d16c | 398 | int invalid = 0, removed = 0; |
974cec8c AA |
399 | |
400 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
401 | ||
38511722 | 402 | if (list_empty(&lo->plh_segs)) { |
2701d086 AA |
403 | /* Reset MDS Threshold I/O counters */ |
404 | NFS_I(lo->plh_inode)->write_io = 0; | |
405 | NFS_I(lo->plh_inode)->read_io = 0; | |
38511722 | 406 | if (!test_and_set_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags)) |
70c3bd2b | 407 | pnfs_put_layout_hdr_locked(lo); |
38511722 FI |
408 | return 0; |
409 | } | |
4541d16c | 410 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
778b5502 BH |
411 | if (!recall_range || |
412 | should_free_lseg(&lseg->pls_range, recall_range)) { | |
4541d16c FI |
413 | dprintk("%s: freeing lseg %p iomode %d " |
414 | "offset %llu length %llu\n", __func__, | |
415 | lseg, lseg->pls_range.iomode, lseg->pls_range.offset, | |
416 | lseg->pls_range.length); | |
417 | invalid++; | |
418 | removed += mark_lseg_invalid(lseg, tmp_list); | |
419 | } | |
420 | dprintk("%s:Return %i\n", __func__, invalid - removed); | |
421 | return invalid - removed; | |
974cec8c AA |
422 | } |
423 | ||
f49f9baa | 424 | /* note free_me must contain lsegs from a single layout_hdr */ |
43f1b3da | 425 | void |
4541d16c | 426 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 427 | { |
4541d16c | 428 | struct pnfs_layout_segment *lseg, *tmp; |
f49f9baa FI |
429 | struct pnfs_layout_hdr *lo; |
430 | ||
431 | if (list_empty(free_me)) | |
432 | return; | |
433 | ||
434 | lo = list_first_entry(free_me, struct pnfs_layout_segment, | |
435 | pls_list)->pls_layout; | |
974cec8c | 436 | |
f49f9baa FI |
437 | if (test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags)) { |
438 | struct nfs_client *clp; | |
439 | ||
440 | clp = NFS_SERVER(lo->plh_inode)->nfs_client; | |
441 | spin_lock(&clp->cl_lock); | |
442 | list_del_init(&lo->plh_layouts); | |
443 | spin_unlock(&clp->cl_lock); | |
444 | } | |
4541d16c | 445 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 446 | list_del(&lseg->pls_list); |
4541d16c | 447 | free_lseg(lseg); |
974cec8c AA |
448 | } |
449 | } | |
450 | ||
e5e94017 BH |
451 | void |
452 | pnfs_destroy_layout(struct nfs_inode *nfsi) | |
453 | { | |
454 | struct pnfs_layout_hdr *lo; | |
974cec8c | 455 | LIST_HEAD(tmp_list); |
e5e94017 BH |
456 | |
457 | spin_lock(&nfsi->vfs_inode.i_lock); | |
458 | lo = nfsi->layout; | |
459 | if (lo) { | |
38511722 | 460 | lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ |
49a85061 | 461 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
e5e94017 BH |
462 | } |
463 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
974cec8c AA |
464 | pnfs_free_lseg_list(&tmp_list); |
465 | } | |
041245c8 | 466 | EXPORT_SYMBOL_GPL(pnfs_destroy_layout); |
974cec8c AA |
467 | |
468 | /* | |
469 | * Called by the state manger to remove all layouts established under an | |
470 | * expired lease. | |
471 | */ | |
472 | void | |
473 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
474 | { | |
6382a441 | 475 | struct nfs_server *server; |
974cec8c AA |
476 | struct pnfs_layout_hdr *lo; |
477 | LIST_HEAD(tmp_list); | |
478 | ||
c47abcf8 AA |
479 | nfs4_deviceid_mark_client_invalid(clp); |
480 | nfs4_deviceid_purge_client(clp); | |
481 | ||
974cec8c | 482 | spin_lock(&clp->cl_lock); |
6382a441 WAA |
483 | rcu_read_lock(); |
484 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { | |
485 | if (!list_empty(&server->layouts)) | |
486 | list_splice_init(&server->layouts, &tmp_list); | |
487 | } | |
488 | rcu_read_unlock(); | |
974cec8c AA |
489 | spin_unlock(&clp->cl_lock); |
490 | ||
491 | while (!list_empty(&tmp_list)) { | |
492 | lo = list_entry(tmp_list.next, struct pnfs_layout_hdr, | |
b7edfaa1 | 493 | plh_layouts); |
974cec8c | 494 | dprintk("%s freeing layout for inode %lu\n", __func__, |
b7edfaa1 | 495 | lo->plh_inode->i_ino); |
2887fe45 | 496 | list_del_init(&lo->plh_layouts); |
b7edfaa1 | 497 | pnfs_destroy_layout(NFS_I(lo->plh_inode)); |
974cec8c | 498 | } |
e5e94017 BH |
499 | } |
500 | ||
fd6002e9 | 501 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
502 | void |
503 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
504 | bool update_barrier) | |
b1f69b75 | 505 | { |
fd6002e9 | 506 | u32 oldseq, newseq; |
b1f69b75 | 507 | |
2d2f24ad TM |
508 | oldseq = be32_to_cpu(lo->plh_stateid.seqid); |
509 | newseq = be32_to_cpu(new->seqid); | |
43f1b3da | 510 | if ((int)(newseq - oldseq) > 0) { |
f597c537 | 511 | nfs4_stateid_copy(&lo->plh_stateid, new); |
43f1b3da | 512 | if (update_barrier) { |
2d2f24ad | 513 | u32 new_barrier = be32_to_cpu(new->seqid); |
43f1b3da FI |
514 | |
515 | if ((int)(new_barrier - lo->plh_barrier)) | |
516 | lo->plh_barrier = new_barrier; | |
517 | } else { | |
518 | /* Because of wraparound, we want to keep the barrier | |
519 | * "close" to the current seqids. It needs to be | |
520 | * within 2**31 to count as "behind", so if it | |
521 | * gets too near that limit, give us a litle leeway | |
522 | * and bring it to within 2**30. | |
523 | * NOTE - and yes, this is all unsigned arithmetic. | |
524 | */ | |
525 | if (unlikely((newseq - lo->plh_barrier) > (3 << 29))) | |
526 | lo->plh_barrier = newseq - (1 << 30); | |
527 | } | |
528 | } | |
b1f69b75 AA |
529 | } |
530 | ||
cf7d63f1 FI |
531 | /* lget is set to 1 if called from inside send_layoutget call chain */ |
532 | static bool | |
43f1b3da FI |
533 | pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid, |
534 | int lget) | |
535 | { | |
536 | if ((stateid) && | |
2d2f24ad | 537 | (int)(lo->plh_barrier - be32_to_cpu(stateid->seqid)) >= 0) |
43f1b3da | 538 | return true; |
f7e8917a | 539 | return lo->plh_block_lgets || |
38511722 | 540 | test_bit(NFS_LAYOUT_DESTROYED, &lo->plh_flags) || |
f7e8917a | 541 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) || |
43f1b3da | 542 | (list_empty(&lo->plh_segs) && |
cf7d63f1 FI |
543 | (atomic_read(&lo->plh_outstanding) > lget)); |
544 | } | |
545 | ||
fd6002e9 FI |
546 | int |
547 | pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, | |
548 | struct nfs4_state *open_state) | |
b1f69b75 | 549 | { |
fd6002e9 | 550 | int status = 0; |
974cec8c | 551 | |
b1f69b75 | 552 | dprintk("--> %s\n", __func__); |
fd6002e9 | 553 | spin_lock(&lo->plh_inode->i_lock); |
43f1b3da | 554 | if (pnfs_layoutgets_blocked(lo, NULL, 1)) { |
cf7d63f1 FI |
555 | status = -EAGAIN; |
556 | } else if (list_empty(&lo->plh_segs)) { | |
fd6002e9 FI |
557 | int seq; |
558 | ||
559 | do { | |
560 | seq = read_seqbegin(&open_state->seqlock); | |
f597c537 | 561 | nfs4_stateid_copy(dst, &open_state->stateid); |
fd6002e9 FI |
562 | } while (read_seqretry(&open_state->seqlock, seq)); |
563 | } else | |
f597c537 | 564 | nfs4_stateid_copy(dst, &lo->plh_stateid); |
fd6002e9 | 565 | spin_unlock(&lo->plh_inode->i_lock); |
b1f69b75 | 566 | dprintk("<-- %s\n", __func__); |
fd6002e9 | 567 | return status; |
b1f69b75 AA |
568 | } |
569 | ||
570 | /* | |
571 | * Get layout from server. | |
572 | * for now, assume that whole file layouts are requested. | |
573 | * arg->offset: 0 | |
574 | * arg->length: all ones | |
575 | */ | |
e5e94017 BH |
576 | static struct pnfs_layout_segment * |
577 | send_layoutget(struct pnfs_layout_hdr *lo, | |
578 | struct nfs_open_context *ctx, | |
fb3296eb | 579 | struct pnfs_layout_range *range, |
a75b9df9 | 580 | gfp_t gfp_flags) |
e5e94017 | 581 | { |
b7edfaa1 | 582 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
583 | struct nfs_server *server = NFS_SERVER(ino); |
584 | struct nfs4_layoutget *lgp; | |
a0b0a6e3 | 585 | struct pnfs_layout_segment *lseg; |
b1f69b75 AA |
586 | |
587 | dprintk("--> %s\n", __func__); | |
e5e94017 | 588 | |
b1f69b75 | 589 | BUG_ON(ctx == NULL); |
a75b9df9 | 590 | lgp = kzalloc(sizeof(*lgp), gfp_flags); |
cf7d63f1 | 591 | if (lgp == NULL) |
b1f69b75 | 592 | return NULL; |
35124a09 | 593 | |
fb3296eb BH |
594 | lgp->args.minlength = PAGE_CACHE_SIZE; |
595 | if (lgp->args.minlength > range->length) | |
596 | lgp->args.minlength = range->length; | |
b1f69b75 | 597 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; |
fb3296eb | 598 | lgp->args.range = *range; |
b1f69b75 AA |
599 | lgp->args.type = server->pnfs_curr_ld->id; |
600 | lgp->args.inode = ino; | |
601 | lgp->args.ctx = get_nfs_open_context(ctx); | |
a75b9df9 | 602 | lgp->gfp_flags = gfp_flags; |
b1f69b75 AA |
603 | |
604 | /* Synchronously retrieve layout information from server and | |
605 | * store in lseg. | |
606 | */ | |
a0b0a6e3 TM |
607 | lseg = nfs4_proc_layoutget(lgp, gfp_flags); |
608 | if (IS_ERR(lseg)) { | |
609 | switch (PTR_ERR(lseg)) { | |
610 | case -ENOMEM: | |
611 | case -ERESTARTSYS: | |
612 | break; | |
613 | default: | |
614 | /* remember that LAYOUTGET failed and suspend trying */ | |
615 | set_bit(lo_fail_bit(range->iomode), &lo->plh_flags); | |
616 | } | |
617 | return NULL; | |
974cec8c | 618 | } |
35124a09 | 619 | |
974cec8c AA |
620 | return lseg; |
621 | } | |
622 | ||
293b3b06 AA |
623 | /* |
624 | * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr | |
625 | * when the layout segment list is empty. | |
626 | * | |
627 | * Note that a pnfs_layout_hdr can exist with an empty layout segment | |
628 | * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the | |
629 | * deviceid is marked invalid. | |
630 | */ | |
cbe82603 BH |
631 | int |
632 | _pnfs_return_layout(struct inode *ino) | |
633 | { | |
634 | struct pnfs_layout_hdr *lo = NULL; | |
635 | struct nfs_inode *nfsi = NFS_I(ino); | |
636 | LIST_HEAD(tmp_list); | |
637 | struct nfs4_layoutreturn *lrp; | |
638 | nfs4_stateid stateid; | |
293b3b06 | 639 | int status = 0, empty; |
cbe82603 | 640 | |
366d5052 | 641 | dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); |
cbe82603 BH |
642 | |
643 | spin_lock(&ino->i_lock); | |
644 | lo = nfsi->layout; | |
366d5052 | 645 | if (!lo || pnfs_test_layout_returned(lo)) { |
cbe82603 | 646 | spin_unlock(&ino->i_lock); |
293b3b06 AA |
647 | dprintk("NFS: %s no layout to return\n", __func__); |
648 | goto out; | |
cbe82603 BH |
649 | } |
650 | stateid = nfsi->layout->plh_stateid; | |
651 | /* Reference matched in nfs4_layoutreturn_release */ | |
70c3bd2b | 652 | pnfs_get_layout_hdr(lo); |
293b3b06 | 653 | empty = list_empty(&lo->plh_segs); |
49a85061 | 654 | pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); |
293b3b06 AA |
655 | /* Don't send a LAYOUTRETURN if list was initially empty */ |
656 | if (empty) { | |
657 | spin_unlock(&ino->i_lock); | |
70c3bd2b | 658 | pnfs_put_layout_hdr(lo); |
293b3b06 AA |
659 | dprintk("NFS: %s no layout segments to return\n", __func__); |
660 | goto out; | |
661 | } | |
ea0ded74 | 662 | lo->plh_block_lgets++; |
366d5052 | 663 | pnfs_mark_layout_returned(lo); |
cbe82603 BH |
664 | spin_unlock(&ino->i_lock); |
665 | pnfs_free_lseg_list(&tmp_list); | |
666 | ||
667 | WARN_ON(test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)); | |
668 | ||
669 | lrp = kzalloc(sizeof(*lrp), GFP_KERNEL); | |
670 | if (unlikely(lrp == NULL)) { | |
671 | status = -ENOMEM; | |
9e2dfdb3 FI |
672 | set_bit(NFS_LAYOUT_RW_FAILED, &lo->plh_flags); |
673 | set_bit(NFS_LAYOUT_RO_FAILED, &lo->plh_flags); | |
366d5052 | 674 | pnfs_clear_layout_returned(lo); |
70c3bd2b | 675 | pnfs_put_layout_hdr(lo); |
cbe82603 BH |
676 | goto out; |
677 | } | |
678 | ||
679 | lrp->args.stateid = stateid; | |
680 | lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id; | |
681 | lrp->args.inode = ino; | |
a56aaa02 | 682 | lrp->args.layout = lo; |
cbe82603 BH |
683 | lrp->clp = NFS_SERVER(ino)->nfs_client; |
684 | ||
685 | status = nfs4_proc_layoutreturn(lrp); | |
686 | out: | |
687 | dprintk("<-- %s status: %d\n", __func__, status); | |
688 | return status; | |
689 | } | |
0a57cdac | 690 | EXPORT_SYMBOL_GPL(_pnfs_return_layout); |
cbe82603 | 691 | |
f7e8917a FI |
692 | bool pnfs_roc(struct inode *ino) |
693 | { | |
694 | struct pnfs_layout_hdr *lo; | |
695 | struct pnfs_layout_segment *lseg, *tmp; | |
696 | LIST_HEAD(tmp_list); | |
697 | bool found = false; | |
698 | ||
699 | spin_lock(&ino->i_lock); | |
700 | lo = NFS_I(ino)->layout; | |
701 | if (!lo || !test_and_clear_bit(NFS_LAYOUT_ROC, &lo->plh_flags) || | |
702 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) | |
703 | goto out_nolayout; | |
704 | list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) | |
705 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
706 | mark_lseg_invalid(lseg, &tmp_list); | |
707 | found = true; | |
708 | } | |
709 | if (!found) | |
710 | goto out_nolayout; | |
711 | lo->plh_block_lgets++; | |
70c3bd2b | 712 | pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */ |
f7e8917a FI |
713 | spin_unlock(&ino->i_lock); |
714 | pnfs_free_lseg_list(&tmp_list); | |
715 | return true; | |
716 | ||
717 | out_nolayout: | |
718 | spin_unlock(&ino->i_lock); | |
719 | return false; | |
720 | } | |
721 | ||
722 | void pnfs_roc_release(struct inode *ino) | |
723 | { | |
724 | struct pnfs_layout_hdr *lo; | |
725 | ||
726 | spin_lock(&ino->i_lock); | |
727 | lo = NFS_I(ino)->layout; | |
728 | lo->plh_block_lgets--; | |
70c3bd2b | 729 | pnfs_put_layout_hdr_locked(lo); |
f7e8917a FI |
730 | spin_unlock(&ino->i_lock); |
731 | } | |
732 | ||
733 | void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) | |
734 | { | |
735 | struct pnfs_layout_hdr *lo; | |
736 | ||
737 | spin_lock(&ino->i_lock); | |
738 | lo = NFS_I(ino)->layout; | |
739 | if ((int)(barrier - lo->plh_barrier) > 0) | |
740 | lo->plh_barrier = barrier; | |
741 | spin_unlock(&ino->i_lock); | |
742 | } | |
743 | ||
744 | bool pnfs_roc_drain(struct inode *ino, u32 *barrier) | |
745 | { | |
746 | struct nfs_inode *nfsi = NFS_I(ino); | |
747 | struct pnfs_layout_segment *lseg; | |
748 | bool found = false; | |
749 | ||
750 | spin_lock(&ino->i_lock); | |
751 | list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list) | |
752 | if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { | |
753 | found = true; | |
754 | break; | |
755 | } | |
756 | if (!found) { | |
757 | struct pnfs_layout_hdr *lo = nfsi->layout; | |
2d2f24ad | 758 | u32 current_seqid = be32_to_cpu(lo->plh_stateid.seqid); |
f7e8917a FI |
759 | |
760 | /* Since close does not return a layout stateid for use as | |
761 | * a barrier, we choose the worst-case barrier. | |
762 | */ | |
763 | *barrier = current_seqid + atomic_read(&lo->plh_outstanding); | |
764 | } | |
765 | spin_unlock(&ino->i_lock); | |
766 | return found; | |
767 | } | |
768 | ||
b1f69b75 AA |
769 | /* |
770 | * Compare two layout segments for sorting into layout cache. | |
771 | * We want to preferentially return RW over RO layouts, so ensure those | |
772 | * are seen first. | |
773 | */ | |
774 | static s64 | |
fb3296eb BH |
775 | cmp_layout(struct pnfs_layout_range *l1, |
776 | struct pnfs_layout_range *l2) | |
b1f69b75 | 777 | { |
fb3296eb BH |
778 | s64 d; |
779 | ||
780 | /* high offset > low offset */ | |
781 | d = l1->offset - l2->offset; | |
782 | if (d) | |
783 | return d; | |
784 | ||
785 | /* short length > long length */ | |
786 | d = l2->length - l1->length; | |
787 | if (d) | |
788 | return d; | |
789 | ||
b1f69b75 | 790 | /* read > read/write */ |
fb3296eb | 791 | return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); |
b1f69b75 AA |
792 | } |
793 | ||
974cec8c AA |
794 | static void |
795 | pnfs_insert_layout(struct pnfs_layout_hdr *lo, | |
796 | struct pnfs_layout_segment *lseg) | |
797 | { | |
b1f69b75 | 798 | struct pnfs_layout_segment *lp; |
b1f69b75 | 799 | |
974cec8c AA |
800 | dprintk("%s:Begin\n", __func__); |
801 | ||
b7edfaa1 | 802 | assert_spin_locked(&lo->plh_inode->i_lock); |
b7edfaa1 | 803 | list_for_each_entry(lp, &lo->plh_segs, pls_list) { |
fb3296eb | 804 | if (cmp_layout(&lseg->pls_range, &lp->pls_range) > 0) |
b1f69b75 | 805 | continue; |
566052c5 | 806 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
807 | dprintk("%s: inserted lseg %p " |
808 | "iomode %d offset %llu length %llu before " | |
809 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
810 | __func__, lseg, lseg->pls_range.iomode, |
811 | lseg->pls_range.offset, lseg->pls_range.length, | |
812 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
813 | lp->pls_range.length); | |
fb3296eb | 814 | goto out; |
974cec8c | 815 | } |
fb3296eb BH |
816 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
817 | dprintk("%s: inserted lseg %p " | |
818 | "iomode %d offset %llu length %llu at tail\n", | |
819 | __func__, lseg, lseg->pls_range.iomode, | |
820 | lseg->pls_range.offset, lseg->pls_range.length); | |
821 | out: | |
70c3bd2b | 822 | pnfs_get_layout_hdr(lo); |
974cec8c AA |
823 | |
824 | dprintk("%s:Return\n", __func__); | |
e5e94017 BH |
825 | } |
826 | ||
827 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
828 | alloc_init_layout_hdr(struct inode *ino, |
829 | struct nfs_open_context *ctx, | |
830 | gfp_t gfp_flags) | |
e5e94017 BH |
831 | { |
832 | struct pnfs_layout_hdr *lo; | |
833 | ||
636fb9c8 | 834 | lo = pnfs_alloc_layout_hdr(ino, gfp_flags); |
e5e94017 BH |
835 | if (!lo) |
836 | return NULL; | |
cc6e5340 | 837 | atomic_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
838 | INIT_LIST_HEAD(&lo->plh_layouts); |
839 | INIT_LIST_HEAD(&lo->plh_segs); | |
43f1b3da | 840 | INIT_LIST_HEAD(&lo->plh_bulk_recall); |
b7edfaa1 | 841 | lo->plh_inode = ino; |
9fa40758 | 842 | lo->plh_lc_cred = get_rpccred(ctx->state->owner->so_cred); |
e5e94017 BH |
843 | return lo; |
844 | } | |
845 | ||
846 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
847 | pnfs_find_alloc_layout(struct inode *ino, |
848 | struct nfs_open_context *ctx, | |
849 | gfp_t gfp_flags) | |
e5e94017 BH |
850 | { |
851 | struct nfs_inode *nfsi = NFS_I(ino); | |
852 | struct pnfs_layout_hdr *new = NULL; | |
853 | ||
854 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
855 | ||
856 | assert_spin_locked(&ino->i_lock); | |
4541d16c FI |
857 | if (nfsi->layout) { |
858 | if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags)) | |
859 | return NULL; | |
860 | else | |
861 | return nfsi->layout; | |
862 | } | |
e5e94017 | 863 | spin_unlock(&ino->i_lock); |
9fa40758 | 864 | new = alloc_init_layout_hdr(ino, ctx, gfp_flags); |
e5e94017 BH |
865 | spin_lock(&ino->i_lock); |
866 | ||
867 | if (likely(nfsi->layout == NULL)) /* Won the race? */ | |
868 | nfsi->layout = new; | |
869 | else | |
636fb9c8 | 870 | pnfs_free_layout_hdr(new); |
e5e94017 BH |
871 | return nfsi->layout; |
872 | } | |
873 | ||
b1f69b75 AA |
874 | /* |
875 | * iomode matching rules: | |
876 | * iomode lseg match | |
877 | * ----- ----- ----- | |
878 | * ANY READ true | |
879 | * ANY RW true | |
880 | * RW READ false | |
881 | * RW RW true | |
882 | * READ READ true | |
883 | * READ RW true | |
884 | */ | |
885 | static int | |
fb3296eb BH |
886 | is_matching_lseg(struct pnfs_layout_range *ls_range, |
887 | struct pnfs_layout_range *range) | |
b1f69b75 | 888 | { |
fb3296eb BH |
889 | struct pnfs_layout_range range1; |
890 | ||
891 | if ((range->iomode == IOMODE_RW && | |
892 | ls_range->iomode != IOMODE_RW) || | |
893 | !lo_seg_intersecting(ls_range, range)) | |
894 | return 0; | |
895 | ||
896 | /* range1 covers only the first byte in the range */ | |
897 | range1 = *range; | |
898 | range1.length = 1; | |
899 | return lo_seg_contained(ls_range, &range1); | |
b1f69b75 AA |
900 | } |
901 | ||
902 | /* | |
903 | * lookup range in layout | |
904 | */ | |
e5e94017 | 905 | static struct pnfs_layout_segment * |
fb3296eb BH |
906 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, |
907 | struct pnfs_layout_range *range) | |
e5e94017 | 908 | { |
b1f69b75 AA |
909 | struct pnfs_layout_segment *lseg, *ret = NULL; |
910 | ||
911 | dprintk("%s:Begin\n", __func__); | |
912 | ||
b7edfaa1 FI |
913 | assert_spin_locked(&lo->plh_inode->i_lock); |
914 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { | |
4541d16c | 915 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
fb3296eb | 916 | is_matching_lseg(&lseg->pls_range, range)) { |
9369a431 | 917 | ret = pnfs_get_lseg(lseg); |
b1f69b75 AA |
918 | break; |
919 | } | |
d771e3a4 | 920 | if (lseg->pls_range.offset > range->offset) |
b1f69b75 AA |
921 | break; |
922 | } | |
923 | ||
924 | dprintk("%s:Return lseg %p ref %d\n", | |
4541d16c | 925 | __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0); |
b1f69b75 | 926 | return ret; |
e5e94017 BH |
927 | } |
928 | ||
d23d61c8 AA |
929 | /* |
930 | * Use mdsthreshold hints set at each OPEN to determine if I/O should go | |
931 | * to the MDS or over pNFS | |
932 | * | |
933 | * The nfs_inode read_io and write_io fields are cumulative counters reset | |
934 | * when there are no layout segments. Note that in pnfs_update_layout iomode | |
935 | * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a | |
936 | * WRITE request. | |
937 | * | |
938 | * A return of true means use MDS I/O. | |
939 | * | |
940 | * From rfc 5661: | |
941 | * If a file's size is smaller than the file size threshold, data accesses | |
942 | * SHOULD be sent to the metadata server. If an I/O request has a length that | |
943 | * is below the I/O size threshold, the I/O SHOULD be sent to the metadata | |
944 | * server. If both file size and I/O size are provided, the client SHOULD | |
945 | * reach or exceed both thresholds before sending its read or write | |
946 | * requests to the data server. | |
947 | */ | |
948 | static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, | |
949 | struct inode *ino, int iomode) | |
950 | { | |
951 | struct nfs4_threshold *t = ctx->mdsthreshold; | |
952 | struct nfs_inode *nfsi = NFS_I(ino); | |
953 | loff_t fsize = i_size_read(ino); | |
954 | bool size = false, size_set = false, io = false, io_set = false, ret = false; | |
955 | ||
956 | if (t == NULL) | |
957 | return ret; | |
958 | ||
959 | dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", | |
960 | __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); | |
961 | ||
962 | switch (iomode) { | |
963 | case IOMODE_READ: | |
964 | if (t->bm & THRESHOLD_RD) { | |
965 | dprintk("%s fsize %llu\n", __func__, fsize); | |
966 | size_set = true; | |
967 | if (fsize < t->rd_sz) | |
968 | size = true; | |
969 | } | |
970 | if (t->bm & THRESHOLD_RD_IO) { | |
971 | dprintk("%s nfsi->read_io %llu\n", __func__, | |
972 | nfsi->read_io); | |
973 | io_set = true; | |
974 | if (nfsi->read_io < t->rd_io_sz) | |
975 | io = true; | |
976 | } | |
977 | break; | |
978 | case IOMODE_RW: | |
979 | if (t->bm & THRESHOLD_WR) { | |
980 | dprintk("%s fsize %llu\n", __func__, fsize); | |
981 | size_set = true; | |
982 | if (fsize < t->wr_sz) | |
983 | size = true; | |
984 | } | |
985 | if (t->bm & THRESHOLD_WR_IO) { | |
986 | dprintk("%s nfsi->write_io %llu\n", __func__, | |
987 | nfsi->write_io); | |
988 | io_set = true; | |
989 | if (nfsi->write_io < t->wr_io_sz) | |
990 | io = true; | |
991 | } | |
992 | break; | |
993 | } | |
994 | if (size_set && io_set) { | |
995 | if (size && io) | |
996 | ret = true; | |
997 | } else if (size || io) | |
998 | ret = true; | |
999 | ||
1000 | dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); | |
1001 | return ret; | |
1002 | } | |
1003 | ||
e5e94017 BH |
1004 | /* |
1005 | * Layout segment is retreived from the server if not cached. | |
1006 | * The appropriate layout segment is referenced and returned to the caller. | |
1007 | */ | |
7c24d948 | 1008 | struct pnfs_layout_segment * |
e5e94017 BH |
1009 | pnfs_update_layout(struct inode *ino, |
1010 | struct nfs_open_context *ctx, | |
fb3296eb BH |
1011 | loff_t pos, |
1012 | u64 count, | |
a75b9df9 TM |
1013 | enum pnfs_iomode iomode, |
1014 | gfp_t gfp_flags) | |
e5e94017 | 1015 | { |
fb3296eb BH |
1016 | struct pnfs_layout_range arg = { |
1017 | .iomode = iomode, | |
1018 | .offset = pos, | |
1019 | .length = count, | |
1020 | }; | |
707ed5fd | 1021 | unsigned pg_offset; |
e5e94017 | 1022 | struct nfs_inode *nfsi = NFS_I(ino); |
6382a441 WAA |
1023 | struct nfs_server *server = NFS_SERVER(ino); |
1024 | struct nfs_client *clp = server->nfs_client; | |
e5e94017 BH |
1025 | struct pnfs_layout_hdr *lo; |
1026 | struct pnfs_layout_segment *lseg = NULL; | |
f49f9baa | 1027 | bool first = false; |
e5e94017 BH |
1028 | |
1029 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) | |
1030 | return NULL; | |
d23d61c8 AA |
1031 | |
1032 | if (pnfs_within_mdsthreshold(ctx, ino, iomode)) | |
1033 | return NULL; | |
1034 | ||
e5e94017 | 1035 | spin_lock(&ino->i_lock); |
9fa40758 | 1036 | lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); |
e5e94017 BH |
1037 | if (lo == NULL) { |
1038 | dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__); | |
1039 | goto out_unlock; | |
1040 | } | |
1041 | ||
43f1b3da | 1042 | /* Do we even need to bother with this? */ |
a59c30ac | 1043 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da | 1044 | dprintk("%s matches recall, use MDS\n", __func__); |
e5e94017 BH |
1045 | goto out_unlock; |
1046 | } | |
1047 | ||
1048 | /* if LAYOUTGET already failed once we don't try again */ | |
566052c5 | 1049 | if (test_bit(lo_fail_bit(iomode), &nfsi->layout->plh_flags)) |
e5e94017 BH |
1050 | goto out_unlock; |
1051 | ||
568e8c49 | 1052 | /* Check to see if the layout for the given range already exists */ |
fb3296eb | 1053 | lseg = pnfs_find_lseg(lo, &arg); |
568e8c49 AA |
1054 | if (lseg) |
1055 | goto out_unlock; | |
1056 | ||
43f1b3da | 1057 | if (pnfs_layoutgets_blocked(lo, NULL, 0)) |
cf7d63f1 FI |
1058 | goto out_unlock; |
1059 | atomic_inc(&lo->plh_outstanding); | |
1060 | ||
70c3bd2b | 1061 | pnfs_get_layout_hdr(lo); |
f49f9baa FI |
1062 | if (list_empty(&lo->plh_segs)) |
1063 | first = true; | |
366d5052 AA |
1064 | |
1065 | /* Enable LAYOUTRETURNs */ | |
1066 | pnfs_clear_layout_returned(lo); | |
1067 | ||
f49f9baa FI |
1068 | spin_unlock(&ino->i_lock); |
1069 | if (first) { | |
2130ff66 FI |
1070 | /* The lo must be on the clp list if there is any |
1071 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
1072 | */ | |
1073 | spin_lock(&clp->cl_lock); | |
1074 | BUG_ON(!list_empty(&lo->plh_layouts)); | |
6382a441 | 1075 | list_add_tail(&lo->plh_layouts, &server->layouts); |
2130ff66 FI |
1076 | spin_unlock(&clp->cl_lock); |
1077 | } | |
e5e94017 | 1078 | |
707ed5fd BH |
1079 | pg_offset = arg.offset & ~PAGE_CACHE_MASK; |
1080 | if (pg_offset) { | |
1081 | arg.offset -= pg_offset; | |
1082 | arg.length += pg_offset; | |
1083 | } | |
7c24d948 AA |
1084 | if (arg.length != NFS4_MAX_UINT64) |
1085 | arg.length = PAGE_CACHE_ALIGN(arg.length); | |
707ed5fd | 1086 | |
fb3296eb | 1087 | lseg = send_layoutget(lo, ctx, &arg, gfp_flags); |
f49f9baa FI |
1088 | if (!lseg && first) { |
1089 | spin_lock(&clp->cl_lock); | |
1090 | list_del_init(&lo->plh_layouts); | |
1091 | spin_unlock(&clp->cl_lock); | |
2130ff66 | 1092 | } |
cf7d63f1 | 1093 | atomic_dec(&lo->plh_outstanding); |
70c3bd2b | 1094 | pnfs_put_layout_hdr(lo); |
e5e94017 BH |
1095 | out: |
1096 | dprintk("%s end, state 0x%lx lseg %p\n", __func__, | |
bf9c1387 | 1097 | nfsi->layout ? nfsi->layout->plh_flags : -1, lseg); |
e5e94017 BH |
1098 | return lseg; |
1099 | out_unlock: | |
1100 | spin_unlock(&ino->i_lock); | |
1101 | goto out; | |
1102 | } | |
7c24d948 | 1103 | EXPORT_SYMBOL_GPL(pnfs_update_layout); |
b1f69b75 | 1104 | |
a0b0a6e3 | 1105 | struct pnfs_layout_segment * |
b1f69b75 AA |
1106 | pnfs_layout_process(struct nfs4_layoutget *lgp) |
1107 | { | |
1108 | struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; | |
1109 | struct nfs4_layoutget_res *res = &lgp->res; | |
1110 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 1111 | struct inode *ino = lo->plh_inode; |
b1f69b75 AA |
1112 | int status = 0; |
1113 | ||
1114 | /* Inject layout blob into I/O device driver */ | |
a75b9df9 | 1115 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); |
b1f69b75 AA |
1116 | if (!lseg || IS_ERR(lseg)) { |
1117 | if (!lseg) | |
1118 | status = -ENOMEM; | |
1119 | else | |
1120 | status = PTR_ERR(lseg); | |
1121 | dprintk("%s: Could not allocate layout: error %d\n", | |
1122 | __func__, status); | |
1123 | goto out; | |
1124 | } | |
1125 | ||
1126 | spin_lock(&ino->i_lock); | |
a59c30ac | 1127 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
43f1b3da FI |
1128 | dprintk("%s forget reply due to recall\n", __func__); |
1129 | goto out_forget_reply; | |
1130 | } | |
1131 | ||
1132 | if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) { | |
1133 | dprintk("%s forget reply due to state\n", __func__); | |
1134 | goto out_forget_reply; | |
1135 | } | |
b1f69b75 | 1136 | init_lseg(lo, lseg); |
566052c5 | 1137 | lseg->pls_range = res->range; |
9369a431 | 1138 | pnfs_get_lseg(lseg); |
b1f69b75 AA |
1139 | pnfs_insert_layout(lo, lseg); |
1140 | ||
f7e8917a FI |
1141 | if (res->return_on_close) { |
1142 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); | |
1143 | set_bit(NFS_LAYOUT_ROC, &lo->plh_flags); | |
1144 | } | |
1145 | ||
b1f69b75 | 1146 | /* Done processing layoutget. Set the layout stateid */ |
43f1b3da | 1147 | pnfs_set_layout_stateid(lo, &res->stateid, false); |
b1f69b75 | 1148 | spin_unlock(&ino->i_lock); |
a0b0a6e3 | 1149 | return lseg; |
b1f69b75 | 1150 | out: |
a0b0a6e3 | 1151 | return ERR_PTR(status); |
43f1b3da FI |
1152 | |
1153 | out_forget_reply: | |
1154 | spin_unlock(&ino->i_lock); | |
1155 | lseg->pls_layout = lo; | |
1156 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
1157 | goto out; | |
b1f69b75 AA |
1158 | } |
1159 | ||
d8007d4d TM |
1160 | void |
1161 | pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
1162 | { | |
1163 | BUG_ON(pgio->pg_lseg != NULL); | |
1164 | ||
1825a0d0 FI |
1165 | if (req->wb_offset != req->wb_pgbase) { |
1166 | nfs_pageio_reset_read_mds(pgio); | |
1167 | return; | |
1168 | } | |
d8007d4d TM |
1169 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
1170 | req->wb_context, | |
1171 | req_offset(req), | |
1172 | req->wb_bytes, | |
1173 | IOMODE_READ, | |
1174 | GFP_KERNEL); | |
e885de1a TM |
1175 | /* If no lseg, fall back to read through mds */ |
1176 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1177 | nfs_pageio_reset_read_mds(pgio); |
e885de1a | 1178 | |
d8007d4d TM |
1179 | } |
1180 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); | |
1181 | ||
1182 | void | |
1183 | pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
1184 | { | |
1185 | BUG_ON(pgio->pg_lseg != NULL); | |
1186 | ||
1825a0d0 FI |
1187 | if (req->wb_offset != req->wb_pgbase) { |
1188 | nfs_pageio_reset_write_mds(pgio); | |
1189 | return; | |
1190 | } | |
d8007d4d TM |
1191 | pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, |
1192 | req->wb_context, | |
1193 | req_offset(req), | |
1194 | req->wb_bytes, | |
1195 | IOMODE_RW, | |
1196 | GFP_NOFS); | |
e885de1a TM |
1197 | /* If no lseg, fall back to write through mds */ |
1198 | if (pgio->pg_lseg == NULL) | |
1f945357 | 1199 | nfs_pageio_reset_write_mds(pgio); |
d8007d4d TM |
1200 | } |
1201 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); | |
1202 | ||
1abb5088 | 1203 | void |
061ae2ed FI |
1204 | pnfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, struct inode *inode, |
1205 | const struct nfs_pgio_completion_ops *compl_ops) | |
1751c363 TM |
1206 | { |
1207 | struct nfs_server *server = NFS_SERVER(inode); | |
1208 | struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld; | |
1209 | ||
1210 | if (ld == NULL) | |
1abb5088 BS |
1211 | nfs_pageio_init_read(pgio, inode, compl_ops); |
1212 | else | |
1213 | nfs_pageio_init(pgio, inode, ld->pg_read_ops, compl_ops, server->rsize, 0); | |
1751c363 TM |
1214 | } |
1215 | ||
57208fa7 | 1216 | void |
061ae2ed FI |
1217 | pnfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, struct inode *inode, |
1218 | int ioflags, | |
1219 | const struct nfs_pgio_completion_ops *compl_ops) | |
1751c363 TM |
1220 | { |
1221 | struct nfs_server *server = NFS_SERVER(inode); | |
1222 | struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld; | |
1223 | ||
1224 | if (ld == NULL) | |
57208fa7 BS |
1225 | nfs_pageio_init_write(pgio, inode, ioflags, compl_ops); |
1226 | else | |
1227 | nfs_pageio_init(pgio, inode, ld->pg_write_ops, compl_ops, server->wsize, ioflags); | |
1751c363 TM |
1228 | } |
1229 | ||
18ad0a9f | 1230 | bool |
dfed206b BH |
1231 | pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
1232 | struct nfs_page *req) | |
94ad1c80 | 1233 | { |
d8007d4d TM |
1234 | if (pgio->pg_lseg == NULL) |
1235 | return nfs_generic_pg_test(pgio, prev, req); | |
94ad1c80 | 1236 | |
19982ba8 TM |
1237 | /* |
1238 | * Test if a nfs_page is fully contained in the pnfs_layout_range. | |
1239 | * Note that this test makes several assumptions: | |
1240 | * - that the previous nfs_page in the struct nfs_pageio_descriptor | |
1241 | * is known to lie within the range. | |
1242 | * - that the nfs_page being tested is known to be contiguous with the | |
1243 | * previous nfs_page. | |
1244 | * - Layout ranges are page aligned, so we only have to test the | |
1245 | * start offset of the request. | |
1246 | * | |
1247 | * Please also note that 'end_offset' is actually the offset of the | |
1248 | * first byte that lies outside the pnfs_layout_range. FIXME? | |
1249 | * | |
1250 | */ | |
1251 | return req_offset(req) < end_offset(pgio->pg_lseg->pls_range.offset, | |
1252 | pgio->pg_lseg->pls_range.length); | |
94ad1c80 | 1253 | } |
89a58e32 | 1254 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); |
94ad1c80 | 1255 | |
e7dd79af | 1256 | int pnfs_write_done_resend_to_mds(struct inode *inode, |
061ae2ed FI |
1257 | struct list_head *head, |
1258 | const struct nfs_pgio_completion_ops *compl_ops) | |
e2fecb21 TM |
1259 | { |
1260 | struct nfs_pageio_descriptor pgio; | |
1261 | LIST_HEAD(failed); | |
1262 | ||
1263 | /* Resend all requests through the MDS */ | |
57208fa7 | 1264 | nfs_pageio_init_write(&pgio, inode, FLUSH_STABLE, compl_ops); |
e2fecb21 TM |
1265 | while (!list_empty(head)) { |
1266 | struct nfs_page *req = nfs_list_entry(head->next); | |
1267 | ||
1268 | nfs_list_remove_request(req); | |
1269 | if (!nfs_pageio_add_request(&pgio, req)) | |
1270 | nfs_list_add_request(req, &failed); | |
1271 | } | |
1272 | nfs_pageio_complete(&pgio); | |
1273 | ||
1274 | if (!list_empty(&failed)) { | |
1275 | /* For some reason our attempt to resend pages. Mark the | |
1276 | * overall send request as having failed, and let | |
1277 | * nfs_writeback_release_full deal with the error. | |
1278 | */ | |
1279 | list_move(&failed, head); | |
1280 | return -EIO; | |
1281 | } | |
1282 | return 0; | |
1283 | } | |
e7dd79af | 1284 | EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); |
e2fecb21 | 1285 | |
1acbbb4e FI |
1286 | static void pnfs_ld_handle_write_error(struct nfs_write_data *data) |
1287 | { | |
cd841605 FI |
1288 | struct nfs_pgio_header *hdr = data->header; |
1289 | ||
1290 | dprintk("pnfs write error = %d\n", hdr->pnfs_error); | |
1291 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1292 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 FI |
1293 | clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags); |
1294 | pnfs_return_layout(hdr->inode); | |
1acbbb4e | 1295 | } |
6c75dc0d FI |
1296 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
1297 | data->task.tk_status = pnfs_write_done_resend_to_mds(hdr->inode, | |
061ae2ed FI |
1298 | &hdr->pages, |
1299 | hdr->completion_ops); | |
1acbbb4e FI |
1300 | } |
1301 | ||
d20581aa BH |
1302 | /* |
1303 | * Called by non rpc-based layout drivers | |
1304 | */ | |
8ce160c5 | 1305 | void pnfs_ld_write_done(struct nfs_write_data *data) |
44b83799 | 1306 | { |
cd841605 FI |
1307 | struct nfs_pgio_header *hdr = data->header; |
1308 | ||
1309 | if (!hdr->pnfs_error) { | |
d20581aa | 1310 | pnfs_set_layoutcommit(data); |
cd841605 | 1311 | hdr->mds_ops->rpc_call_done(&data->task, data); |
1acbbb4e FI |
1312 | } else |
1313 | pnfs_ld_handle_write_error(data); | |
cd841605 | 1314 | hdr->mds_ops->rpc_release(data); |
44b83799 | 1315 | } |
d20581aa | 1316 | EXPORT_SYMBOL_GPL(pnfs_ld_write_done); |
44b83799 | 1317 | |
dce81290 TM |
1318 | static void |
1319 | pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, | |
1320 | struct nfs_write_data *data) | |
1321 | { | |
cd841605 FI |
1322 | struct nfs_pgio_header *hdr = data->header; |
1323 | ||
6c75dc0d FI |
1324 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
1325 | list_splice_tail_init(&hdr->pages, &desc->pg_list); | |
1326 | nfs_pageio_reset_write_mds(desc); | |
1327 | desc->pg_recoalesce = 1; | |
1328 | } | |
dce81290 TM |
1329 | nfs_writedata_release(data); |
1330 | } | |
1331 | ||
1332 | static enum pnfs_try_status | |
0382b744 | 1333 | pnfs_try_to_write_data(struct nfs_write_data *wdata, |
dce81290 TM |
1334 | const struct rpc_call_ops *call_ops, |
1335 | struct pnfs_layout_segment *lseg, | |
1336 | int how) | |
0382b744 | 1337 | { |
cd841605 FI |
1338 | struct nfs_pgio_header *hdr = wdata->header; |
1339 | struct inode *inode = hdr->inode; | |
0382b744 AA |
1340 | enum pnfs_try_status trypnfs; |
1341 | struct nfs_server *nfss = NFS_SERVER(inode); | |
1342 | ||
cd841605 | 1343 | hdr->mds_ops = call_ops; |
0382b744 AA |
1344 | |
1345 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
1346 | inode->i_ino, wdata->args.count, wdata->args.offset, how); | |
0382b744 | 1347 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(wdata, how); |
6c75dc0d | 1348 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
0382b744 | 1349 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); |
0382b744 AA |
1350 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1351 | return trypnfs; | |
1352 | } | |
1353 | ||
dce81290 TM |
1354 | static void |
1355 | pnfs_do_multiple_writes(struct nfs_pageio_descriptor *desc, struct list_head *head, int how) | |
1356 | { | |
1357 | struct nfs_write_data *data; | |
1358 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; | |
1359 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
1360 | ||
1361 | desc->pg_lseg = NULL; | |
1362 | while (!list_empty(head)) { | |
1363 | enum pnfs_try_status trypnfs; | |
1364 | ||
6c75dc0d | 1365 | data = list_first_entry(head, struct nfs_write_data, list); |
dce81290 TM |
1366 | list_del_init(&data->list); |
1367 | ||
1368 | trypnfs = pnfs_try_to_write_data(data, call_ops, lseg, how); | |
1369 | if (trypnfs == PNFS_NOT_ATTEMPTED) | |
1370 | pnfs_write_through_mds(desc, data); | |
1371 | } | |
9369a431 | 1372 | pnfs_put_lseg(lseg); |
dce81290 TM |
1373 | } |
1374 | ||
6c75dc0d FI |
1375 | static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) |
1376 | { | |
9369a431 | 1377 | pnfs_put_lseg(hdr->lseg); |
6c75dc0d FI |
1378 | nfs_writehdr_free(hdr); |
1379 | } | |
89d77c8f | 1380 | EXPORT_SYMBOL_GPL(pnfs_writehdr_free); |
6c75dc0d | 1381 | |
dce81290 TM |
1382 | int |
1383 | pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1384 | { | |
6c75dc0d FI |
1385 | struct nfs_write_header *whdr; |
1386 | struct nfs_pgio_header *hdr; | |
dce81290 TM |
1387 | int ret; |
1388 | ||
6c75dc0d FI |
1389 | whdr = nfs_writehdr_alloc(); |
1390 | if (!whdr) { | |
9b5415b5 | 1391 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
9369a431 | 1392 | pnfs_put_lseg(desc->pg_lseg); |
dce81290 | 1393 | desc->pg_lseg = NULL; |
6c75dc0d | 1394 | return -ENOMEM; |
dce81290 | 1395 | } |
6c75dc0d FI |
1396 | hdr = &whdr->header; |
1397 | nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); | |
9369a431 | 1398 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
6c75dc0d FI |
1399 | atomic_inc(&hdr->refcnt); |
1400 | ret = nfs_generic_flush(desc, hdr); | |
1401 | if (ret != 0) { | |
9369a431 | 1402 | pnfs_put_lseg(desc->pg_lseg); |
6c75dc0d | 1403 | desc->pg_lseg = NULL; |
6c75dc0d FI |
1404 | } else |
1405 | pnfs_do_multiple_writes(desc, &hdr->rpc_list, desc->pg_ioflags); | |
1406 | if (atomic_dec_and_test(&hdr->refcnt)) | |
061ae2ed | 1407 | hdr->completion_ops->completion(hdr); |
6c75dc0d | 1408 | return ret; |
dce81290 TM |
1409 | } |
1410 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); | |
1411 | ||
e7dd79af | 1412 | int pnfs_read_done_resend_to_mds(struct inode *inode, |
061ae2ed FI |
1413 | struct list_head *head, |
1414 | const struct nfs_pgio_completion_ops *compl_ops) | |
62e4a769 TM |
1415 | { |
1416 | struct nfs_pageio_descriptor pgio; | |
1acbbb4e | 1417 | LIST_HEAD(failed); |
62e4a769 | 1418 | |
1acbbb4e | 1419 | /* Resend all requests through the MDS */ |
1abb5088 | 1420 | nfs_pageio_init_read(&pgio, inode, compl_ops); |
1acbbb4e FI |
1421 | while (!list_empty(head)) { |
1422 | struct nfs_page *req = nfs_list_entry(head->next); | |
62e4a769 TM |
1423 | |
1424 | nfs_list_remove_request(req); | |
1acbbb4e FI |
1425 | if (!nfs_pageio_add_request(&pgio, req)) |
1426 | nfs_list_add_request(req, &failed); | |
62e4a769 TM |
1427 | } |
1428 | nfs_pageio_complete(&pgio); | |
1acbbb4e FI |
1429 | |
1430 | if (!list_empty(&failed)) { | |
1431 | list_move(&failed, head); | |
1432 | return -EIO; | |
1433 | } | |
1434 | return 0; | |
1435 | } | |
e7dd79af | 1436 | EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); |
1acbbb4e FI |
1437 | |
1438 | static void pnfs_ld_handle_read_error(struct nfs_read_data *data) | |
1439 | { | |
cd841605 FI |
1440 | struct nfs_pgio_header *hdr = data->header; |
1441 | ||
1442 | dprintk("pnfs read error = %d\n", hdr->pnfs_error); | |
1443 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 1444 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 FI |
1445 | clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags); |
1446 | pnfs_return_layout(hdr->inode); | |
1acbbb4e | 1447 | } |
4db6e0b7 FI |
1448 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
1449 | data->task.tk_status = pnfs_read_done_resend_to_mds(hdr->inode, | |
061ae2ed FI |
1450 | &hdr->pages, |
1451 | hdr->completion_ops); | |
62e4a769 TM |
1452 | } |
1453 | ||
d20581aa BH |
1454 | /* |
1455 | * Called by non rpc-based layout drivers | |
1456 | */ | |
9b7eecdc | 1457 | void pnfs_ld_read_done(struct nfs_read_data *data) |
d20581aa | 1458 | { |
cd841605 FI |
1459 | struct nfs_pgio_header *hdr = data->header; |
1460 | ||
1461 | if (likely(!hdr->pnfs_error)) { | |
d20581aa | 1462 | __nfs4_read_done_cb(data); |
cd841605 | 1463 | hdr->mds_ops->rpc_call_done(&data->task, data); |
62e4a769 TM |
1464 | } else |
1465 | pnfs_ld_handle_read_error(data); | |
cd841605 | 1466 | hdr->mds_ops->rpc_release(data); |
d20581aa BH |
1467 | } |
1468 | EXPORT_SYMBOL_GPL(pnfs_ld_read_done); | |
1469 | ||
493292dd TM |
1470 | static void |
1471 | pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, | |
1472 | struct nfs_read_data *data) | |
1473 | { | |
cd841605 FI |
1474 | struct nfs_pgio_header *hdr = data->header; |
1475 | ||
4db6e0b7 FI |
1476 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
1477 | list_splice_tail_init(&hdr->pages, &desc->pg_list); | |
1478 | nfs_pageio_reset_read_mds(desc); | |
1479 | desc->pg_recoalesce = 1; | |
1480 | } | |
493292dd TM |
1481 | nfs_readdata_release(data); |
1482 | } | |
1483 | ||
64419a9b AA |
1484 | /* |
1485 | * Call the appropriate parallel I/O subsystem read function. | |
1486 | */ | |
493292dd | 1487 | static enum pnfs_try_status |
64419a9b | 1488 | pnfs_try_to_read_data(struct nfs_read_data *rdata, |
493292dd TM |
1489 | const struct rpc_call_ops *call_ops, |
1490 | struct pnfs_layout_segment *lseg) | |
64419a9b | 1491 | { |
cd841605 FI |
1492 | struct nfs_pgio_header *hdr = rdata->header; |
1493 | struct inode *inode = hdr->inode; | |
64419a9b AA |
1494 | struct nfs_server *nfss = NFS_SERVER(inode); |
1495 | enum pnfs_try_status trypnfs; | |
1496 | ||
cd841605 | 1497 | hdr->mds_ops = call_ops; |
64419a9b AA |
1498 | |
1499 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
1500 | __func__, inode->i_ino, rdata->args.count, rdata->args.offset); | |
1501 | ||
1502 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(rdata); | |
4db6e0b7 | 1503 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
64419a9b | 1504 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); |
64419a9b AA |
1505 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
1506 | return trypnfs; | |
1507 | } | |
863a3c6c | 1508 | |
493292dd TM |
1509 | static void |
1510 | pnfs_do_multiple_reads(struct nfs_pageio_descriptor *desc, struct list_head *head) | |
1511 | { | |
1512 | struct nfs_read_data *data; | |
1513 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; | |
1514 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
1515 | ||
1516 | desc->pg_lseg = NULL; | |
1517 | while (!list_empty(head)) { | |
1518 | enum pnfs_try_status trypnfs; | |
1519 | ||
4db6e0b7 | 1520 | data = list_first_entry(head, struct nfs_read_data, list); |
493292dd TM |
1521 | list_del_init(&data->list); |
1522 | ||
1523 | trypnfs = pnfs_try_to_read_data(data, call_ops, lseg); | |
1524 | if (trypnfs == PNFS_NOT_ATTEMPTED) | |
1525 | pnfs_read_through_mds(desc, data); | |
1526 | } | |
9369a431 | 1527 | pnfs_put_lseg(lseg); |
493292dd TM |
1528 | } |
1529 | ||
4db6e0b7 FI |
1530 | static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) |
1531 | { | |
9369a431 | 1532 | pnfs_put_lseg(hdr->lseg); |
4db6e0b7 FI |
1533 | nfs_readhdr_free(hdr); |
1534 | } | |
89d77c8f | 1535 | EXPORT_SYMBOL_GPL(pnfs_readhdr_free); |
4db6e0b7 | 1536 | |
493292dd TM |
1537 | int |
1538 | pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
1539 | { | |
4db6e0b7 FI |
1540 | struct nfs_read_header *rhdr; |
1541 | struct nfs_pgio_header *hdr; | |
493292dd TM |
1542 | int ret; |
1543 | ||
4db6e0b7 FI |
1544 | rhdr = nfs_readhdr_alloc(); |
1545 | if (!rhdr) { | |
061ae2ed | 1546 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
4db6e0b7 | 1547 | ret = -ENOMEM; |
9369a431 | 1548 | pnfs_put_lseg(desc->pg_lseg); |
493292dd TM |
1549 | desc->pg_lseg = NULL; |
1550 | return ret; | |
1551 | } | |
4db6e0b7 FI |
1552 | hdr = &rhdr->header; |
1553 | nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); | |
9369a431 | 1554 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
4db6e0b7 FI |
1555 | atomic_inc(&hdr->refcnt); |
1556 | ret = nfs_generic_pagein(desc, hdr); | |
1557 | if (ret != 0) { | |
9369a431 | 1558 | pnfs_put_lseg(desc->pg_lseg); |
4db6e0b7 | 1559 | desc->pg_lseg = NULL; |
4db6e0b7 FI |
1560 | } else |
1561 | pnfs_do_multiple_reads(desc, &hdr->rpc_list); | |
1562 | if (atomic_dec_and_test(&hdr->refcnt)) | |
061ae2ed | 1563 | hdr->completion_ops->completion(hdr); |
4db6e0b7 | 1564 | return ret; |
493292dd TM |
1565 | } |
1566 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); | |
1567 | ||
863a3c6c | 1568 | /* |
a9bae566 | 1569 | * There can be multiple RW segments. |
863a3c6c | 1570 | */ |
a9bae566 | 1571 | static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) |
863a3c6c | 1572 | { |
a9bae566 | 1573 | struct pnfs_layout_segment *lseg; |
863a3c6c | 1574 | |
a9bae566 PT |
1575 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { |
1576 | if (lseg->pls_range.iomode == IOMODE_RW && | |
1577 | test_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
1578 | list_add(&lseg->pls_lc_list, listp); | |
1579 | } | |
863a3c6c AA |
1580 | } |
1581 | ||
1b0ae068 PT |
1582 | void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) |
1583 | { | |
1584 | if (lseg->pls_range.iomode == IOMODE_RW) { | |
1585 | dprintk("%s Setting layout IOMODE_RW fail bit\n", __func__); | |
1586 | set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags); | |
1587 | } else { | |
1588 | dprintk("%s Setting layout IOMODE_READ fail bit\n", __func__); | |
1589 | set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags); | |
1590 | } | |
1591 | } | |
1592 | EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); | |
1593 | ||
863a3c6c AA |
1594 | void |
1595 | pnfs_set_layoutcommit(struct nfs_write_data *wdata) | |
1596 | { | |
cd841605 FI |
1597 | struct nfs_pgio_header *hdr = wdata->header; |
1598 | struct inode *inode = hdr->inode; | |
1599 | struct nfs_inode *nfsi = NFS_I(inode); | |
4b8ee2b8 | 1600 | loff_t end_pos = wdata->mds_offset + wdata->res.count; |
79a48a1f | 1601 | bool mark_as_dirty = false; |
863a3c6c | 1602 | |
cd841605 | 1603 | spin_lock(&inode->i_lock); |
863a3c6c | 1604 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
79a48a1f | 1605 | mark_as_dirty = true; |
863a3c6c | 1606 | dprintk("%s: Set layoutcommit for inode %lu ", |
cd841605 | 1607 | __func__, inode->i_ino); |
863a3c6c | 1608 | } |
cd841605 | 1609 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) { |
a9bae566 | 1610 | /* references matched in nfs4_layoutcommit_release */ |
9369a431 | 1611 | pnfs_get_lseg(hdr->lseg); |
a9bae566 | 1612 | } |
acff5880 PT |
1613 | if (end_pos > nfsi->layout->plh_lwb) |
1614 | nfsi->layout->plh_lwb = end_pos; | |
cd841605 | 1615 | spin_unlock(&inode->i_lock); |
acff5880 | 1616 | dprintk("%s: lseg %p end_pos %llu\n", |
cd841605 | 1617 | __func__, hdr->lseg, nfsi->layout->plh_lwb); |
79a48a1f WAA |
1618 | |
1619 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
1620 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
1621 | if (mark_as_dirty) | |
cd841605 | 1622 | mark_inode_dirty_sync(inode); |
863a3c6c AA |
1623 | } |
1624 | EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); | |
1625 | ||
db29c089 AA |
1626 | void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) |
1627 | { | |
1628 | struct nfs_server *nfss = NFS_SERVER(data->args.inode); | |
1629 | ||
1630 | if (nfss->pnfs_curr_ld->cleanup_layoutcommit) | |
1631 | nfss->pnfs_curr_ld->cleanup_layoutcommit(data); | |
1632 | } | |
1633 | ||
de4b15c7 AA |
1634 | /* |
1635 | * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and | |
1636 | * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough | |
1637 | * data to disk to allow the server to recover the data if it crashes. | |
1638 | * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag | |
1639 | * is off, and a COMMIT is sent to a data server, or | |
1640 | * if WRITEs to a data server return NFS_DATA_SYNC. | |
1641 | */ | |
863a3c6c | 1642 | int |
ef311537 | 1643 | pnfs_layoutcommit_inode(struct inode *inode, bool sync) |
863a3c6c AA |
1644 | { |
1645 | struct nfs4_layoutcommit_data *data; | |
1646 | struct nfs_inode *nfsi = NFS_I(inode); | |
863a3c6c AA |
1647 | loff_t end_pos; |
1648 | int status = 0; | |
1649 | ||
1650 | dprintk("--> %s inode %lu\n", __func__, inode->i_ino); | |
1651 | ||
de4b15c7 AA |
1652 | if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
1653 | return 0; | |
1654 | ||
863a3c6c AA |
1655 | /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ |
1656 | data = kzalloc(sizeof(*data), GFP_NOFS); | |
de4b15c7 | 1657 | if (!data) { |
de4b15c7 AA |
1658 | status = -ENOMEM; |
1659 | goto out; | |
1660 | } | |
863a3c6c | 1661 | |
92407e75 PT |
1662 | if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
1663 | goto out_free; | |
1664 | ||
1665 | if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { | |
1666 | if (!sync) { | |
1667 | status = -EAGAIN; | |
1668 | goto out_free; | |
1669 | } | |
1670 | status = wait_on_bit_lock(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING, | |
1671 | nfs_wait_bit_killable, TASK_KILLABLE); | |
1672 | if (status) | |
1673 | goto out_free; | |
1674 | } | |
1675 | ||
a9bae566 | 1676 | INIT_LIST_HEAD(&data->lseg_list); |
de4b15c7 | 1677 | spin_lock(&inode->i_lock); |
863a3c6c | 1678 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
92407e75 | 1679 | clear_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags); |
863a3c6c | 1680 | spin_unlock(&inode->i_lock); |
92407e75 PT |
1681 | wake_up_bit(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING); |
1682 | goto out_free; | |
863a3c6c | 1683 | } |
a9bae566 PT |
1684 | |
1685 | pnfs_list_write_lseg(inode, &data->lseg_list); | |
863a3c6c | 1686 | |
acff5880 | 1687 | end_pos = nfsi->layout->plh_lwb; |
acff5880 | 1688 | nfsi->layout->plh_lwb = 0; |
863a3c6c | 1689 | |
f597c537 | 1690 | nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); |
863a3c6c AA |
1691 | spin_unlock(&inode->i_lock); |
1692 | ||
1693 | data->args.inode = inode; | |
9fa40758 | 1694 | data->cred = get_rpccred(nfsi->layout->plh_lc_cred); |
863a3c6c AA |
1695 | nfs_fattr_init(&data->fattr); |
1696 | data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; | |
1697 | data->res.fattr = &data->fattr; | |
1698 | data->args.lastbytewritten = end_pos - 1; | |
1699 | data->res.server = NFS_SERVER(inode); | |
1700 | ||
1701 | status = nfs4_proc_layoutcommit(data, sync); | |
1702 | out: | |
92407e75 PT |
1703 | if (status) |
1704 | mark_inode_dirty_sync(inode); | |
863a3c6c AA |
1705 | dprintk("<-- %s status %d\n", __func__, status); |
1706 | return status; | |
92407e75 PT |
1707 | out_free: |
1708 | kfree(data); | |
1709 | goto out; | |
863a3c6c | 1710 | } |
82be417a AA |
1711 | |
1712 | struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) | |
1713 | { | |
1714 | struct nfs4_threshold *thp; | |
1715 | ||
1716 | thp = kzalloc(sizeof(*thp), GFP_NOFS); | |
1717 | if (!thp) { | |
1718 | dprintk("%s mdsthreshold allocation failed\n", __func__); | |
1719 | return NULL; | |
1720 | } | |
1721 | return thp; | |
1722 | } |