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> |
ca440c38 | 33 | #include <linux/sort.h> |
974cec8c | 34 | #include "internal.h" |
85e174ba | 35 | #include "pnfs.h" |
64419a9b | 36 | #include "iostat.h" |
cc668ab3 | 37 | #include "nfs4trace.h" |
40dd4b7a | 38 | #include "delegation.h" |
8733408d | 39 | #include "nfs42.h" |
1b146fcf | 40 | #include "nfs4_fs.h" |
85e174ba RL |
41 | |
42 | #define NFSDBG_FACILITY NFSDBG_PNFS | |
25c75333 | 43 | #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ) |
85e174ba | 44 | |
02c35fca FI |
45 | /* Locking: |
46 | * | |
47 | * pnfs_spinlock: | |
48 | * protects pnfs_modules_tbl. | |
49 | */ | |
50 | static DEFINE_SPINLOCK(pnfs_spinlock); | |
51 | ||
52 | /* | |
53 | * pnfs_modules_tbl holds all pnfs modules | |
54 | */ | |
55 | static LIST_HEAD(pnfs_modules_tbl); | |
56 | ||
13c13a6a | 57 | static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo); |
68f74479 TM |
58 | static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo, |
59 | struct list_head *free_me, | |
60 | const struct pnfs_layout_range *range, | |
61 | u32 seq); | |
fe1cf946 TM |
62 | static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg, |
63 | struct list_head *tmp_list); | |
aa1e0e3a | 64 | |
02c35fca FI |
65 | /* Return the registered pnfs layout driver module matching given id */ |
66 | static struct pnfs_layoutdriver_type * | |
67 | find_pnfs_driver_locked(u32 id) | |
68 | { | |
69 | struct pnfs_layoutdriver_type *local; | |
70 | ||
71 | list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid) | |
72 | if (local->id == id) | |
73 | goto out; | |
74 | local = NULL; | |
75 | out: | |
76 | dprintk("%s: Searching for id %u, found %p\n", __func__, id, local); | |
77 | return local; | |
78 | } | |
79 | ||
85e174ba RL |
80 | static struct pnfs_layoutdriver_type * |
81 | find_pnfs_driver(u32 id) | |
82 | { | |
02c35fca FI |
83 | struct pnfs_layoutdriver_type *local; |
84 | ||
85 | spin_lock(&pnfs_spinlock); | |
86 | local = find_pnfs_driver_locked(id); | |
0a9c63fa TM |
87 | if (local != NULL && !try_module_get(local->owner)) { |
88 | dprintk("%s: Could not grab reference on module\n", __func__); | |
89 | local = NULL; | |
90 | } | |
02c35fca FI |
91 | spin_unlock(&pnfs_spinlock); |
92 | return local; | |
85e174ba RL |
93 | } |
94 | ||
7c9d845f TM |
95 | const struct pnfs_layoutdriver_type *pnfs_find_layoutdriver(u32 id) |
96 | { | |
97 | return find_pnfs_driver(id); | |
98 | } | |
99 | ||
100 | void pnfs_put_layoutdriver(const struct pnfs_layoutdriver_type *ld) | |
101 | { | |
102 | if (ld) | |
103 | module_put(ld->owner); | |
104 | } | |
105 | ||
85e174ba RL |
106 | void |
107 | unset_pnfs_layoutdriver(struct nfs_server *nfss) | |
108 | { | |
738fd0f3 BH |
109 | if (nfss->pnfs_curr_ld) { |
110 | if (nfss->pnfs_curr_ld->clear_layoutdriver) | |
111 | nfss->pnfs_curr_ld->clear_layoutdriver(nfss); | |
2a4c8994 TM |
112 | /* Decrement the MDS count. Purge the deviceid cache if zero */ |
113 | if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count)) | |
114 | nfs4_deviceid_purge_client(nfss->nfs_client); | |
02c35fca | 115 | module_put(nfss->pnfs_curr_ld->owner); |
738fd0f3 | 116 | } |
85e174ba RL |
117 | nfss->pnfs_curr_ld = NULL; |
118 | } | |
119 | ||
ca440c38 JL |
120 | /* |
121 | * When the server sends a list of layout types, we choose one in the order | |
122 | * given in the list below. | |
123 | * | |
124 | * FIXME: should this list be configurable in some fashion? module param? | |
125 | * mount option? something else? | |
126 | */ | |
127 | static const u32 ld_prefs[] = { | |
128 | LAYOUT_SCSI, | |
129 | LAYOUT_BLOCK_VOLUME, | |
130 | LAYOUT_OSD2_OBJECTS, | |
131 | LAYOUT_FLEX_FILES, | |
132 | LAYOUT_NFSV4_1_FILES, | |
133 | 0 | |
134 | }; | |
135 | ||
136 | static int | |
137 | ld_cmp(const void *e1, const void *e2) | |
138 | { | |
139 | u32 ld1 = *((u32 *)e1); | |
140 | u32 ld2 = *((u32 *)e2); | |
141 | int i; | |
142 | ||
143 | for (i = 0; ld_prefs[i] != 0; i++) { | |
144 | if (ld1 == ld_prefs[i]) | |
145 | return -1; | |
146 | ||
147 | if (ld2 == ld_prefs[i]) | |
148 | return 1; | |
149 | } | |
150 | return 0; | |
151 | } | |
152 | ||
85e174ba RL |
153 | /* |
154 | * Try to set the server's pnfs module to the pnfs layout type specified by id. | |
155 | * Currently only one pNFS layout driver per filesystem is supported. | |
156 | * | |
3132e49e | 157 | * @ids array of layout types supported by MDS. |
85e174ba RL |
158 | */ |
159 | void | |
738fd0f3 | 160 | set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh, |
ca440c38 | 161 | struct nfs_fsinfo *fsinfo) |
85e174ba RL |
162 | { |
163 | struct pnfs_layoutdriver_type *ld_type = NULL; | |
3132e49e | 164 | u32 id; |
ca440c38 | 165 | int i; |
85e174ba | 166 | |
19274716 AS |
167 | if (fsinfo->nlayouttypes == 0) |
168 | goto out_no_driver; | |
85e174ba RL |
169 | if (!(server->nfs_client->cl_exchange_flags & |
170 | (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) { | |
3132e49e JL |
171 | printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n", |
172 | __func__, server->nfs_client->cl_exchange_flags); | |
85e174ba RL |
173 | goto out_no_driver; |
174 | } | |
3132e49e | 175 | |
ca440c38 JL |
176 | sort(fsinfo->layouttype, fsinfo->nlayouttypes, |
177 | sizeof(*fsinfo->layouttype), ld_cmp, NULL); | |
3132e49e | 178 | |
ca440c38 JL |
179 | for (i = 0; i < fsinfo->nlayouttypes; i++) { |
180 | id = fsinfo->layouttype[i]; | |
85e174ba | 181 | ld_type = find_pnfs_driver(id); |
ca440c38 JL |
182 | if (!ld_type) { |
183 | request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, | |
184 | id); | |
185 | ld_type = find_pnfs_driver(id); | |
186 | } | |
187 | if (ld_type) | |
188 | break; | |
85e174ba | 189 | } |
3132e49e JL |
190 | |
191 | if (!ld_type) { | |
ca440c38 | 192 | dprintk("%s: No pNFS module found!\n", __func__); |
3132e49e JL |
193 | goto out_no_driver; |
194 | } | |
195 | ||
85e174ba | 196 | server->pnfs_curr_ld = ld_type; |
738fd0f3 BH |
197 | if (ld_type->set_layoutdriver |
198 | && ld_type->set_layoutdriver(server, mntfh)) { | |
a030889a WAA |
199 | printk(KERN_ERR "NFS: %s: Error initializing pNFS layout " |
200 | "driver %u.\n", __func__, id); | |
738fd0f3 BH |
201 | module_put(ld_type->owner); |
202 | goto out_no_driver; | |
203 | } | |
2a4c8994 TM |
204 | /* Bump the MDS count */ |
205 | atomic_inc(&server->nfs_client->cl_mds_count); | |
ea8eecdd | 206 | |
85e174ba RL |
207 | dprintk("%s: pNFS module for %u set\n", __func__, id); |
208 | return; | |
209 | ||
210 | out_no_driver: | |
211 | dprintk("%s: Using NFSv4 I/O\n", __func__); | |
212 | server->pnfs_curr_ld = NULL; | |
213 | } | |
02c35fca FI |
214 | |
215 | int | |
216 | pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
217 | { | |
218 | int status = -EINVAL; | |
219 | struct pnfs_layoutdriver_type *tmp; | |
220 | ||
221 | if (ld_type->id == 0) { | |
a030889a | 222 | printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__); |
02c35fca FI |
223 | return status; |
224 | } | |
b1f69b75 | 225 | if (!ld_type->alloc_lseg || !ld_type->free_lseg) { |
a030889a | 226 | printk(KERN_ERR "NFS: %s Layout driver must provide " |
b1f69b75 AA |
227 | "alloc_lseg and free_lseg.\n", __func__); |
228 | return status; | |
229 | } | |
02c35fca FI |
230 | |
231 | spin_lock(&pnfs_spinlock); | |
232 | tmp = find_pnfs_driver_locked(ld_type->id); | |
233 | if (!tmp) { | |
234 | list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl); | |
235 | status = 0; | |
236 | dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id, | |
237 | ld_type->name); | |
238 | } else { | |
a030889a | 239 | printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n", |
02c35fca FI |
240 | __func__, ld_type->id); |
241 | } | |
242 | spin_unlock(&pnfs_spinlock); | |
243 | ||
244 | return status; | |
245 | } | |
246 | EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver); | |
247 | ||
248 | void | |
249 | pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type) | |
250 | { | |
251 | dprintk("%s Deregistering id:%u\n", __func__, ld_type->id); | |
252 | spin_lock(&pnfs_spinlock); | |
253 | list_del(&ld_type->pnfs_tblid); | |
254 | spin_unlock(&pnfs_spinlock); | |
255 | } | |
256 | EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver); | |
e5e94017 | 257 | |
b1f69b75 AA |
258 | /* |
259 | * pNFS client layout cache | |
260 | */ | |
261 | ||
cc6e5340 | 262 | /* Need to hold i_lock if caller does not already hold reference */ |
43f1b3da | 263 | void |
70c3bd2b | 264 | pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 265 | { |
2b28a7be | 266 | refcount_inc(&lo->plh_refcount); |
e5e94017 BH |
267 | } |
268 | ||
636fb9c8 BH |
269 | static struct pnfs_layout_hdr * |
270 | pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags) | |
271 | { | |
272 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld; | |
57934278 | 273 | return ld->alloc_layout_hdr(ino, gfp_flags); |
636fb9c8 BH |
274 | } |
275 | ||
276 | static void | |
277 | pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo) | |
278 | { | |
9c626381 TM |
279 | struct nfs_server *server = NFS_SERVER(lo->plh_inode); |
280 | struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld; | |
281 | ||
cf6605d1 | 282 | if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) { |
9c626381 TM |
283 | struct nfs_client *clp = server->nfs_client; |
284 | ||
285 | spin_lock(&clp->cl_lock); | |
cf6605d1 | 286 | list_del_rcu(&lo->plh_layouts); |
9c626381 TM |
287 | spin_unlock(&clp->cl_lock); |
288 | } | |
a52458b4 | 289 | put_cred(lo->plh_lc_cred); |
57934278 | 290 | return ld->free_layout_hdr(lo); |
636fb9c8 BH |
291 | } |
292 | ||
e5e94017 | 293 | static void |
6622c3ea | 294 | pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo) |
e5e94017 | 295 | { |
bb346f63 | 296 | struct nfs_inode *nfsi = NFS_I(lo->plh_inode); |
cc6e5340 | 297 | dprintk("%s: freeing layout cache %p\n", __func__, lo); |
bb346f63 TM |
298 | nfsi->layout = NULL; |
299 | /* Reset MDS Threshold I/O counters */ | |
300 | nfsi->write_io = 0; | |
301 | nfsi->read_io = 0; | |
e5e94017 BH |
302 | } |
303 | ||
b1f69b75 | 304 | void |
70c3bd2b | 305 | pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo) |
974cec8c | 306 | { |
9c6376eb | 307 | struct inode *inode; |
b6d49ecd | 308 | unsigned long i_state; |
cc6e5340 | 309 | |
9c6376eb TM |
310 | if (!lo) |
311 | return; | |
312 | inode = lo->plh_inode; | |
13c13a6a TM |
313 | pnfs_layoutreturn_before_put_layout_hdr(lo); |
314 | ||
2b28a7be | 315 | if (refcount_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) { |
566f8737 PT |
316 | if (!list_empty(&lo->plh_segs)) |
317 | WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n"); | |
6622c3ea | 318 | pnfs_detach_layout_hdr(lo); |
b6d49ecd | 319 | i_state = inode->i_state; |
cc6e5340 | 320 | spin_unlock(&inode->i_lock); |
6622c3ea | 321 | pnfs_free_layout_hdr(lo); |
b6d49ecd TM |
322 | /* Notify pnfs_destroy_layout_final() that we're done */ |
323 | if (i_state & (I_FREEING | I_CLEAR)) | |
324 | wake_up_var(lo); | |
cc6e5340 | 325 | } |
974cec8c AA |
326 | } |
327 | ||
b5fdf841 TM |
328 | static struct inode * |
329 | pnfs_grab_inode_layout_hdr(struct pnfs_layout_hdr *lo) | |
330 | { | |
331 | struct inode *inode = igrab(lo->plh_inode); | |
332 | if (inode) | |
333 | return inode; | |
334 | set_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags); | |
335 | return NULL; | |
336 | } | |
337 | ||
1bcf34fd TM |
338 | /* |
339 | * Compare 2 layout stateid sequence ids, to see which is newer, | |
340 | * taking into account wraparound issues. | |
341 | */ | |
342 | static bool pnfs_seqid_is_newer(u32 s1, u32 s2) | |
343 | { | |
344 | return (s32)(s1 - s2) > 0; | |
345 | } | |
346 | ||
347 | static void pnfs_barrier_update(struct pnfs_layout_hdr *lo, u32 newseq) | |
348 | { | |
45baadaa | 349 | if (pnfs_seqid_is_newer(newseq, lo->plh_barrier) || !lo->plh_barrier) |
1bcf34fd TM |
350 | lo->plh_barrier = newseq; |
351 | } | |
352 | ||
4aab9732 TM |
353 | static void |
354 | pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode, | |
355 | u32 seq) | |
356 | { | |
357 | if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode) | |
358 | iomode = IOMODE_ANY; | |
359 | lo->plh_return_iomode = iomode; | |
360 | set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags); | |
e20772cb TM |
361 | /* |
362 | * We must set lo->plh_return_seq to avoid livelocks with | |
363 | * pnfs_layout_need_return() | |
364 | */ | |
365 | if (seq == 0) | |
366 | seq = be32_to_cpu(lo->plh_stateid.seqid); | |
367 | if (!lo->plh_return_seq || pnfs_seqid_is_newer(seq, lo->plh_return_seq)) | |
4aab9732 | 368 | lo->plh_return_seq = seq; |
e20772cb | 369 | pnfs_barrier_update(lo, seq); |
4aab9732 TM |
370 | } |
371 | ||
ae5a459d TM |
372 | static void |
373 | pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo) | |
374 | { | |
5466d214 | 375 | struct pnfs_layout_segment *lseg; |
ae5a459d TM |
376 | lo->plh_return_iomode = 0; |
377 | lo->plh_return_seq = 0; | |
378 | clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags); | |
5466d214 TM |
379 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { |
380 | if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)) | |
381 | continue; | |
382 | pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0); | |
383 | } | |
ae5a459d TM |
384 | } |
385 | ||
362fb578 TM |
386 | static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo) |
387 | { | |
388 | clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags); | |
389 | clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags); | |
390 | smp_mb__after_atomic(); | |
391 | wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN); | |
392 | rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq); | |
393 | } | |
394 | ||
fe1cf946 TM |
395 | static void |
396 | pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg, | |
397 | struct list_head *free_me) | |
398 | { | |
399 | clear_bit(NFS_LSEG_ROC, &lseg->pls_flags); | |
400 | clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags); | |
401 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) | |
402 | pnfs_lseg_dec_and_remove_zero(lseg, free_me); | |
403 | if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
404 | pnfs_lseg_dec_and_remove_zero(lseg, free_me); | |
405 | } | |
406 | ||
7380020e | 407 | /* |
30cb3ee2 TM |
408 | * Update the seqid of a layout stateid after receiving |
409 | * NFS4ERR_OLD_STATEID | |
7380020e | 410 | */ |
30cb3ee2 | 411 | bool nfs4_layout_refresh_old_stateid(nfs4_stateid *dst, |
ecf84026 TM |
412 | struct pnfs_layout_range *dst_range, |
413 | struct inode *inode) | |
7380020e TM |
414 | { |
415 | struct pnfs_layout_hdr *lo; | |
c16467dc TM |
416 | struct pnfs_layout_range range = { |
417 | .iomode = IOMODE_ANY, | |
418 | .offset = 0, | |
419 | .length = NFS4_MAX_UINT64, | |
420 | }; | |
7380020e | 421 | bool ret = false; |
c16467dc TM |
422 | LIST_HEAD(head); |
423 | int err; | |
7380020e TM |
424 | |
425 | spin_lock(&inode->i_lock); | |
426 | lo = NFS_I(inode)->layout; | |
30cb3ee2 TM |
427 | if (lo && pnfs_layout_is_valid(lo) && |
428 | nfs4_stateid_match_other(dst, &lo->plh_stateid)) { | |
429 | /* Is our call using the most recent seqid? If so, bump it */ | |
430 | if (!nfs4_stateid_is_newer(&lo->plh_stateid, dst)) { | |
431 | nfs4_stateid_seqid_inc(dst); | |
432 | ret = true; | |
433 | goto out; | |
434 | } | |
435 | /* Try to update the seqid to the most recent */ | |
c16467dc TM |
436 | err = pnfs_mark_matching_lsegs_return(lo, &head, &range, 0); |
437 | if (err != -EBUSY) { | |
438 | dst->seqid = lo->plh_stateid.seqid; | |
ecf84026 | 439 | *dst_range = range; |
c16467dc TM |
440 | ret = true; |
441 | } | |
7380020e | 442 | } |
30cb3ee2 | 443 | out: |
7380020e | 444 | spin_unlock(&inode->i_lock); |
c16467dc | 445 | pnfs_free_lseg_list(&head); |
7380020e TM |
446 | return ret; |
447 | } | |
448 | ||
2454dfea TM |
449 | /* |
450 | * Mark a pnfs_layout_hdr and all associated layout segments as invalid | |
451 | * | |
452 | * In order to continue using the pnfs_layout_hdr, a full recovery | |
453 | * is required. | |
454 | * Note that caller must hold inode->i_lock. | |
455 | */ | |
5f46be04 | 456 | int |
2454dfea TM |
457 | pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo, |
458 | struct list_head *lseg_list) | |
459 | { | |
460 | struct pnfs_layout_range range = { | |
461 | .iomode = IOMODE_ANY, | |
462 | .offset = 0, | |
463 | .length = NFS4_MAX_UINT64, | |
464 | }; | |
fe1cf946 | 465 | struct pnfs_layout_segment *lseg, *next; |
2454dfea TM |
466 | |
467 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
fe1cf946 TM |
468 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
469 | pnfs_clear_lseg_state(lseg, lseg_list); | |
5466d214 | 470 | pnfs_clear_layoutreturn_info(lo); |
68f74479 | 471 | pnfs_free_returned_lsegs(lo, lseg_list, &range, 0); |
880265c7 | 472 | set_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags); |
362fb578 TM |
473 | if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) && |
474 | !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) | |
475 | pnfs_clear_layoutreturn_waitbit(lo); | |
fe1cf946 | 476 | return !list_empty(&lo->plh_segs); |
2454dfea TM |
477 | } |
478 | ||
b9e028fd TM |
479 | static int |
480 | pnfs_iomode_to_fail_bit(u32 iomode) | |
481 | { | |
482 | return iomode == IOMODE_RW ? | |
483 | NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED; | |
484 | } | |
485 | ||
486 | static void | |
3e621214 | 487 | pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) |
b9e028fd | 488 | { |
25c75333 | 489 | lo->plh_retry_timestamp = jiffies; |
39e88fcf | 490 | if (!test_and_set_bit(fail_bit, &lo->plh_flags)) |
2b28a7be | 491 | refcount_inc(&lo->plh_refcount); |
3e621214 TM |
492 | } |
493 | ||
494 | static void | |
495 | pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) | |
496 | { | |
497 | if (test_and_clear_bit(fail_bit, &lo->plh_flags)) | |
2b28a7be | 498 | refcount_dec(&lo->plh_refcount); |
3e621214 TM |
499 | } |
500 | ||
501 | static void | |
502 | pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
503 | { | |
504 | struct inode *inode = lo->plh_inode; | |
115ce575 TM |
505 | struct pnfs_layout_range range = { |
506 | .iomode = iomode, | |
507 | .offset = 0, | |
508 | .length = NFS4_MAX_UINT64, | |
509 | }; | |
510 | LIST_HEAD(head); | |
3e621214 TM |
511 | |
512 | spin_lock(&inode->i_lock); | |
513 | pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
28d4411f | 514 | pnfs_mark_matching_lsegs_return(lo, &head, &range, 0); |
3e621214 | 515 | spin_unlock(&inode->i_lock); |
115ce575 | 516 | pnfs_free_lseg_list(&head); |
b9e028fd TM |
517 | dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__, |
518 | iomode == IOMODE_RW ? "RW" : "READ"); | |
519 | } | |
520 | ||
521 | static bool | |
522 | pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode) | |
523 | { | |
25c75333 | 524 | unsigned long start, end; |
3e621214 TM |
525 | int fail_bit = pnfs_iomode_to_fail_bit(iomode); |
526 | ||
527 | if (test_bit(fail_bit, &lo->plh_flags) == 0) | |
25c75333 TM |
528 | return false; |
529 | end = jiffies; | |
530 | start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT; | |
531 | if (!time_in_range(lo->plh_retry_timestamp, start, end)) { | |
532 | /* It is time to retry the failed layoutgets */ | |
3e621214 | 533 | pnfs_layout_clear_fail_bit(lo, fail_bit); |
25c75333 TM |
534 | return false; |
535 | } | |
536 | return true; | |
b9e028fd TM |
537 | } |
538 | ||
974cec8c | 539 | static void |
119cef97 TM |
540 | pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg, |
541 | const struct pnfs_layout_range *range, | |
542 | const nfs4_stateid *stateid) | |
974cec8c | 543 | { |
566052c5 | 544 | INIT_LIST_HEAD(&lseg->pls_list); |
a9bae566 | 545 | INIT_LIST_HEAD(&lseg->pls_lc_list); |
a9901899 | 546 | INIT_LIST_HEAD(&lseg->pls_commits); |
eba6dd69 | 547 | refcount_set(&lseg->pls_refcount, 1); |
4541d16c | 548 | set_bit(NFS_LSEG_VALID, &lseg->pls_flags); |
566052c5 | 549 | lseg->pls_layout = lo; |
119cef97 TM |
550 | lseg->pls_range = *range; |
551 | lseg->pls_seq = be32_to_cpu(stateid->seqid); | |
974cec8c AA |
552 | } |
553 | ||
905ca191 | 554 | static void pnfs_free_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 555 | { |
68f74479 TM |
556 | if (lseg != NULL) { |
557 | struct inode *inode = lseg->pls_layout->plh_inode; | |
558 | NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg); | |
559 | } | |
974cec8c AA |
560 | } |
561 | ||
d684d2ae | 562 | static void |
57036a37 TM |
563 | pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo, |
564 | struct pnfs_layout_segment *lseg) | |
d684d2ae | 565 | { |
d20581aa | 566 | WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); |
d684d2ae | 567 | list_del_init(&lseg->pls_list); |
8f0d27dc | 568 | /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */ |
2b28a7be | 569 | refcount_dec(&lo->plh_refcount); |
abb3e1c8 TM |
570 | if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)) |
571 | return; | |
7b650994 TM |
572 | if (list_empty(&lo->plh_segs) && |
573 | !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) && | |
574 | !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) { | |
334a8f37 TM |
575 | if (atomic_read(&lo->plh_outstanding) == 0) |
576 | set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
173f77e9 | 577 | clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); |
2d148c7e | 578 | } |
d684d2ae FI |
579 | } |
580 | ||
68f74479 TM |
581 | static bool |
582 | pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo, | |
583 | struct pnfs_layout_segment *lseg) | |
584 | { | |
585 | if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) && | |
586 | pnfs_layout_is_valid(lo)) { | |
4aab9732 | 587 | pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0); |
68f74479 TM |
588 | list_move_tail(&lseg->pls_list, &lo->plh_return_segs); |
589 | return true; | |
590 | } | |
591 | return false; | |
592 | } | |
593 | ||
bae724ef | 594 | void |
9369a431 | 595 | pnfs_put_lseg(struct pnfs_layout_segment *lseg) |
974cec8c | 596 | { |
57036a37 | 597 | struct pnfs_layout_hdr *lo; |
d684d2ae FI |
598 | struct inode *inode; |
599 | ||
600 | if (!lseg) | |
601 | return; | |
602 | ||
4541d16c | 603 | dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, |
eba6dd69 | 604 | refcount_read(&lseg->pls_refcount), |
4541d16c | 605 | test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); |
4ef2e4f8 | 606 | |
57036a37 TM |
607 | lo = lseg->pls_layout; |
608 | inode = lo->plh_inode; | |
4ef2e4f8 | 609 | |
eba6dd69 | 610 | if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) { |
8f0d27dc | 611 | pnfs_get_layout_hdr(lo); |
4ef2e4f8 | 612 | pnfs_layout_remove_lseg(lo, lseg); |
68f74479 TM |
613 | if (pnfs_cache_lseg_for_layoutreturn(lo, lseg)) |
614 | lseg = NULL; | |
4ef2e4f8 TM |
615 | spin_unlock(&inode->i_lock); |
616 | pnfs_free_lseg(lseg); | |
617 | pnfs_put_layout_hdr(lo); | |
4541d16c | 618 | } |
4541d16c | 619 | } |
9369a431 | 620 | EXPORT_SYMBOL_GPL(pnfs_put_lseg); |
974cec8c | 621 | |
fb3296eb BH |
622 | /* |
623 | * is l2 fully contained in l1? | |
624 | * start1 end1 | |
625 | * [----------------------------------) | |
626 | * start2 end2 | |
627 | * [----------------) | |
628 | */ | |
3cb2df17 | 629 | static bool |
7dc0ac70 | 630 | pnfs_lseg_range_contained(const struct pnfs_layout_range *l1, |
3cb2df17 | 631 | const struct pnfs_layout_range *l2) |
fb3296eb BH |
632 | { |
633 | u64 start1 = l1->offset; | |
17822b20 | 634 | u64 end1 = pnfs_end_offset(start1, l1->length); |
fb3296eb | 635 | u64 start2 = l2->offset; |
17822b20 | 636 | u64 end2 = pnfs_end_offset(start2, l2->length); |
fb3296eb BH |
637 | |
638 | return (start1 <= start2) && (end1 >= end2); | |
639 | } | |
640 | ||
24956804 TM |
641 | static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg, |
642 | struct list_head *tmp_list) | |
643 | { | |
eba6dd69 | 644 | if (!refcount_dec_and_test(&lseg->pls_refcount)) |
24956804 TM |
645 | return false; |
646 | pnfs_layout_remove_lseg(lseg->pls_layout, lseg); | |
647 | list_add(&lseg->pls_list, tmp_list); | |
648 | return true; | |
649 | } | |
650 | ||
4541d16c FI |
651 | /* Returns 1 if lseg is removed from list, 0 otherwise */ |
652 | static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, | |
653 | struct list_head *tmp_list) | |
654 | { | |
655 | int rv = 0; | |
656 | ||
657 | if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) { | |
658 | /* Remove the reference keeping the lseg in the | |
659 | * list. It will now be removed when all | |
660 | * outstanding io is finished. | |
661 | */ | |
d684d2ae | 662 | dprintk("%s: lseg %p ref %d\n", __func__, lseg, |
eba6dd69 | 663 | refcount_read(&lseg->pls_refcount)); |
24956804 | 664 | if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list)) |
d684d2ae | 665 | rv = 1; |
4541d16c FI |
666 | } |
667 | return rv; | |
668 | } | |
669 | ||
e036f464 TM |
670 | static bool |
671 | pnfs_should_free_range(const struct pnfs_layout_range *lseg_range, | |
672 | const struct pnfs_layout_range *recall_range) | |
673 | { | |
674 | return (recall_range->iomode == IOMODE_ANY || | |
675 | lseg_range->iomode == recall_range->iomode) && | |
676 | pnfs_lseg_range_intersecting(lseg_range, recall_range); | |
677 | } | |
678 | ||
679 | static bool | |
680 | pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg, | |
681 | const struct pnfs_layout_range *recall_range, | |
682 | u32 seq) | |
683 | { | |
684 | if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq)) | |
685 | return false; | |
686 | if (recall_range == NULL) | |
687 | return true; | |
688 | return pnfs_should_free_range(&lseg->pls_range, recall_range); | |
689 | } | |
690 | ||
6d597e17 JL |
691 | /** |
692 | * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later | |
693 | * @lo: layout header containing the lsegs | |
694 | * @tmp_list: list head where doomed lsegs should go | |
695 | * @recall_range: optional recall range argument to match (may be NULL) | |
696 | * @seq: only invalidate lsegs obtained prior to this sequence (may be 0) | |
697 | * | |
698 | * Walk the list of lsegs in the layout header, and tear down any that should | |
699 | * be destroyed. If "recall_range" is specified then the segment must match | |
700 | * that range. If "seq" is non-zero, then only match segments that were handed | |
701 | * out at or before that sequence. | |
702 | * | |
703 | * Returns number of matching invalid lsegs remaining in list after scanning | |
704 | * it and purging them. | |
4541d16c | 705 | */ |
43f1b3da | 706 | int |
49a85061 | 707 | pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, |
4541d16c | 708 | struct list_head *tmp_list, |
6d597e17 JL |
709 | const struct pnfs_layout_range *recall_range, |
710 | u32 seq) | |
974cec8c AA |
711 | { |
712 | struct pnfs_layout_segment *lseg, *next; | |
b739a5bd | 713 | struct nfs_server *server = NFS_SERVER(lo->plh_inode); |
71b39854 | 714 | int remaining = 0; |
974cec8c AA |
715 | |
716 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
717 | ||
8006bfba | 718 | if (list_empty(&lo->plh_segs)) |
38511722 | 719 | return 0; |
4541d16c | 720 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
e036f464 | 721 | if (pnfs_match_lseg_recall(lseg, recall_range, seq)) { |
b3dce6a2 | 722 | dprintk("%s: freeing lseg %p iomode %d seq %u " |
4541d16c | 723 | "offset %llu length %llu\n", __func__, |
6d597e17 JL |
724 | lseg, lseg->pls_range.iomode, lseg->pls_seq, |
725 | lseg->pls_range.offset, lseg->pls_range.length); | |
b739a5bd TM |
726 | if (mark_lseg_invalid(lseg, tmp_list)) |
727 | continue; | |
728 | remaining++; | |
729 | pnfs_lseg_cancel_io(server, lseg); | |
4541d16c | 730 | } |
71b39854 TM |
731 | dprintk("%s:Return %i\n", __func__, remaining); |
732 | return remaining; | |
974cec8c AA |
733 | } |
734 | ||
68f74479 TM |
735 | static void |
736 | pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo, | |
737 | struct list_head *free_me, | |
738 | const struct pnfs_layout_range *range, | |
739 | u32 seq) | |
740 | { | |
741 | struct pnfs_layout_segment *lseg, *next; | |
742 | ||
743 | list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) { | |
744 | if (pnfs_match_lseg_recall(lseg, range, seq)) | |
745 | list_move_tail(&lseg->pls_list, free_me); | |
746 | } | |
747 | } | |
748 | ||
f49f9baa | 749 | /* note free_me must contain lsegs from a single layout_hdr */ |
43f1b3da | 750 | void |
4541d16c | 751 | pnfs_free_lseg_list(struct list_head *free_me) |
974cec8c | 752 | { |
4541d16c | 753 | struct pnfs_layout_segment *lseg, *tmp; |
f49f9baa FI |
754 | |
755 | if (list_empty(free_me)) | |
756 | return; | |
757 | ||
4541d16c | 758 | list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { |
566052c5 | 759 | list_del(&lseg->pls_list); |
905ca191 | 760 | pnfs_free_lseg(lseg); |
974cec8c AA |
761 | } |
762 | } | |
763 | ||
b6d49ecd | 764 | static struct pnfs_layout_hdr *__pnfs_destroy_layout(struct nfs_inode *nfsi) |
e5e94017 BH |
765 | { |
766 | struct pnfs_layout_hdr *lo; | |
974cec8c | 767 | LIST_HEAD(tmp_list); |
e5e94017 BH |
768 | |
769 | spin_lock(&nfsi->vfs_inode.i_lock); | |
770 | lo = nfsi->layout; | |
771 | if (lo) { | |
3e621214 | 772 | pnfs_get_layout_hdr(lo); |
2454dfea | 773 | pnfs_mark_layout_stateid_invalid(lo, &tmp_list); |
3e621214 TM |
774 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED); |
775 | pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED); | |
776 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
777 | pnfs_free_lseg_list(&tmp_list); | |
1f18b82c | 778 | nfs_commit_inode(&nfsi->vfs_inode, 0); |
3e621214 TM |
779 | pnfs_put_layout_hdr(lo); |
780 | } else | |
781 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
b6d49ecd TM |
782 | return lo; |
783 | } | |
784 | ||
785 | void pnfs_destroy_layout(struct nfs_inode *nfsi) | |
786 | { | |
787 | __pnfs_destroy_layout(nfsi); | |
974cec8c | 788 | } |
041245c8 | 789 | EXPORT_SYMBOL_GPL(pnfs_destroy_layout); |
974cec8c | 790 | |
b6d49ecd TM |
791 | static bool pnfs_layout_removed(struct nfs_inode *nfsi, |
792 | struct pnfs_layout_hdr *lo) | |
793 | { | |
794 | bool ret; | |
795 | ||
796 | spin_lock(&nfsi->vfs_inode.i_lock); | |
797 | ret = nfsi->layout != lo; | |
798 | spin_unlock(&nfsi->vfs_inode.i_lock); | |
799 | return ret; | |
800 | } | |
801 | ||
802 | void pnfs_destroy_layout_final(struct nfs_inode *nfsi) | |
803 | { | |
804 | struct pnfs_layout_hdr *lo = __pnfs_destroy_layout(nfsi); | |
805 | ||
806 | if (lo) | |
807 | wait_var_event(lo, pnfs_layout_removed(nfsi, lo)); | |
808 | } | |
809 | ||
fd9a8d71 TM |
810 | static bool |
811 | pnfs_layout_add_bulk_destroy_list(struct inode *inode, | |
812 | struct list_head *layout_list) | |
974cec8c AA |
813 | { |
814 | struct pnfs_layout_hdr *lo; | |
fd9a8d71 | 815 | bool ret = false; |
974cec8c | 816 | |
fd9a8d71 TM |
817 | spin_lock(&inode->i_lock); |
818 | lo = NFS_I(inode)->layout; | |
819 | if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) { | |
820 | pnfs_get_layout_hdr(lo); | |
821 | list_add(&lo->plh_bulk_destroy, layout_list); | |
822 | ret = true; | |
823 | } | |
824 | spin_unlock(&inode->i_lock); | |
825 | return ret; | |
826 | } | |
827 | ||
828 | /* Caller must hold rcu_read_lock and clp->cl_lock */ | |
829 | static int | |
830 | pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp, | |
831 | struct nfs_server *server, | |
832 | struct list_head *layout_list) | |
5085607d TM |
833 | __must_hold(&clp->cl_lock) |
834 | __must_hold(RCU) | |
fd9a8d71 TM |
835 | { |
836 | struct pnfs_layout_hdr *lo, *next; | |
837 | struct inode *inode; | |
838 | ||
839 | list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) { | |
5085607d TM |
840 | if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags) || |
841 | test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) || | |
842 | !list_empty(&lo->plh_bulk_destroy)) | |
b85f5620 | 843 | continue; |
5085607d TM |
844 | /* If the sb is being destroyed, just bail */ |
845 | if (!nfs_sb_active(server->super)) | |
846 | break; | |
b5fdf841 | 847 | inode = pnfs_grab_inode_layout_hdr(lo); |
5085607d | 848 | if (inode != NULL) { |
cf6605d1 TM |
849 | if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) |
850 | list_del_rcu(&lo->plh_layouts); | |
5085607d TM |
851 | if (pnfs_layout_add_bulk_destroy_list(inode, |
852 | layout_list)) | |
853 | continue; | |
854 | rcu_read_unlock(); | |
855 | spin_unlock(&clp->cl_lock); | |
856 | iput(inode); | |
857 | } else { | |
858 | rcu_read_unlock(); | |
859 | spin_unlock(&clp->cl_lock); | |
5085607d TM |
860 | } |
861 | nfs_sb_deactive(server->super); | |
fd9a8d71 TM |
862 | spin_lock(&clp->cl_lock); |
863 | rcu_read_lock(); | |
864 | return -EAGAIN; | |
865 | } | |
866 | return 0; | |
867 | } | |
868 | ||
869 | static int | |
870 | pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list, | |
871 | bool is_bulk_recall) | |
872 | { | |
873 | struct pnfs_layout_hdr *lo; | |
874 | struct inode *inode; | |
fd9a8d71 TM |
875 | LIST_HEAD(lseg_list); |
876 | int ret = 0; | |
877 | ||
878 | while (!list_empty(layout_list)) { | |
879 | lo = list_entry(layout_list->next, struct pnfs_layout_hdr, | |
880 | plh_bulk_destroy); | |
881 | dprintk("%s freeing layout for inode %lu\n", __func__, | |
882 | lo->plh_inode->i_ino); | |
883 | inode = lo->plh_inode; | |
7c5d1875 CH |
884 | |
885 | pnfs_layoutcommit_inode(inode, false); | |
886 | ||
fd9a8d71 TM |
887 | spin_lock(&inode->i_lock); |
888 | list_del_init(&lo->plh_bulk_destroy); | |
9fd4b9fc TM |
889 | if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) { |
890 | if (is_bulk_recall) | |
891 | set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); | |
fd9a8d71 | 892 | ret = -EAGAIN; |
9fd4b9fc | 893 | } |
fd9a8d71 TM |
894 | spin_unlock(&inode->i_lock); |
895 | pnfs_free_lseg_list(&lseg_list); | |
b20135d0 TM |
896 | /* Free all lsegs that are attached to commit buckets */ |
897 | nfs_commit_inode(inode, 0); | |
fd9a8d71 | 898 | pnfs_put_layout_hdr(lo); |
5085607d | 899 | nfs_iput_and_deactive(inode); |
fd9a8d71 TM |
900 | } |
901 | return ret; | |
902 | } | |
903 | ||
904 | int | |
905 | pnfs_destroy_layouts_byfsid(struct nfs_client *clp, | |
906 | struct nfs_fsid *fsid, | |
907 | bool is_recall) | |
908 | { | |
909 | struct nfs_server *server; | |
910 | LIST_HEAD(layout_list); | |
c47abcf8 | 911 | |
974cec8c | 912 | spin_lock(&clp->cl_lock); |
6382a441 | 913 | rcu_read_lock(); |
fd9a8d71 | 914 | restart: |
6382a441 | 915 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { |
fd9a8d71 TM |
916 | if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0) |
917 | continue; | |
918 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
919 | server, | |
920 | &layout_list) != 0) | |
921 | goto restart; | |
6382a441 WAA |
922 | } |
923 | rcu_read_unlock(); | |
974cec8c AA |
924 | spin_unlock(&clp->cl_lock); |
925 | ||
fd9a8d71 TM |
926 | if (list_empty(&layout_list)) |
927 | return 0; | |
928 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
929 | } | |
930 | ||
931 | int | |
932 | pnfs_destroy_layouts_byclid(struct nfs_client *clp, | |
933 | bool is_recall) | |
934 | { | |
935 | struct nfs_server *server; | |
936 | LIST_HEAD(layout_list); | |
937 | ||
938 | spin_lock(&clp->cl_lock); | |
939 | rcu_read_lock(); | |
940 | restart: | |
941 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { | |
942 | if (pnfs_layout_bulk_destroy_byserver_locked(clp, | |
943 | server, | |
944 | &layout_list) != 0) | |
945 | goto restart; | |
974cec8c | 946 | } |
fd9a8d71 TM |
947 | rcu_read_unlock(); |
948 | spin_unlock(&clp->cl_lock); | |
949 | ||
950 | if (list_empty(&layout_list)) | |
951 | return 0; | |
952 | return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); | |
953 | } | |
954 | ||
955 | /* | |
9f266451 | 956 | * Called by the state manager to remove all layouts established under an |
fd9a8d71 TM |
957 | * expired lease. |
958 | */ | |
959 | void | |
960 | pnfs_destroy_all_layouts(struct nfs_client *clp) | |
961 | { | |
962 | nfs4_deviceid_mark_client_invalid(clp); | |
963 | nfs4_deviceid_purge_client(clp); | |
964 | ||
965 | pnfs_destroy_layouts_byclid(clp, false); | |
e5e94017 BH |
966 | } |
967 | ||
59b56394 TM |
968 | static void |
969 | pnfs_set_layout_cred(struct pnfs_layout_hdr *lo, const struct cred *cred) | |
970 | { | |
971 | const struct cred *old; | |
972 | ||
973 | if (cred && cred_fscmp(lo->plh_lc_cred, cred) != 0) { | |
974 | old = xchg(&lo->plh_lc_cred, get_cred(cred)); | |
975 | put_cred(old); | |
976 | } | |
977 | } | |
978 | ||
fd6002e9 | 979 | /* update lo->plh_stateid with new if is more recent */ |
43f1b3da FI |
980 | void |
981 | pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, | |
59b56394 | 982 | const struct cred *cred, bool update_barrier) |
b1f69b75 | 983 | { |
aa95edf3 TM |
984 | u32 oldseq = be32_to_cpu(lo->plh_stateid.seqid); |
985 | u32 newseq = be32_to_cpu(new->seqid); | |
2a59a041 TM |
986 | |
987 | if (!pnfs_layout_is_valid(lo)) { | |
59b56394 | 988 | pnfs_set_layout_cred(lo, cred); |
2a59a041 TM |
989 | nfs4_stateid_copy(&lo->plh_stateid, new); |
990 | lo->plh_barrier = newseq; | |
991 | pnfs_clear_layoutreturn_info(lo); | |
992 | clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags); | |
993 | return; | |
994 | } | |
aa95edf3 TM |
995 | |
996 | if (pnfs_seqid_is_newer(newseq, oldseq)) | |
f597c537 | 997 | nfs4_stateid_copy(&lo->plh_stateid, new); |
aa95edf3 TM |
998 | |
999 | if (update_barrier) { | |
1000 | pnfs_barrier_update(lo, newseq); | |
ecebb80b | 1001 | return; |
aa95edf3 TM |
1002 | } |
1003 | /* | |
1004 | * Because of wraparound, we want to keep the barrier | |
1005 | * "close" to the current seqids. We really only want to | |
1006 | * get here from a layoutget call. | |
1007 | */ | |
1008 | if (atomic_read(&lo->plh_outstanding) == 1) | |
1009 | pnfs_barrier_update(lo, be32_to_cpu(lo->plh_stateid.seqid)); | |
b1f69b75 AA |
1010 | } |
1011 | ||
cf7d63f1 | 1012 | static bool |
19c54aba TM |
1013 | pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo, |
1014 | const nfs4_stateid *stateid) | |
43f1b3da | 1015 | { |
19c54aba | 1016 | u32 seqid = be32_to_cpu(stateid->seqid); |
25a1a621 | 1017 | |
d6236a98 | 1018 | return lo->plh_barrier && pnfs_seqid_is_newer(lo->plh_barrier, seqid); |
19c54aba TM |
1019 | } |
1020 | ||
1021 | /* lget is set to 1 if called from inside send_layoutget call chain */ | |
1022 | static bool | |
e1c06f80 | 1023 | pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo) |
19c54aba | 1024 | { |
f7e8917a | 1025 | return lo->plh_block_lgets || |
e1c06f80 | 1026 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); |
cf7d63f1 FI |
1027 | } |
1028 | ||
5e36e2a9 FI |
1029 | static struct nfs_server * |
1030 | pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx) | |
1031 | { | |
1032 | struct nfs_server *server; | |
1033 | ||
78746a38 | 1034 | if (inode) { |
5e36e2a9 | 1035 | server = NFS_SERVER(inode); |
78746a38 | 1036 | } else { |
5e36e2a9 FI |
1037 | struct dentry *parent_dir = dget_parent(ctx->dentry); |
1038 | server = NFS_SERVER(parent_dir->d_inode); | |
1039 | dput(parent_dir); | |
1040 | } | |
1041 | return server; | |
1042 | } | |
1043 | ||
29a8bfe5 TM |
1044 | static void nfs4_free_pages(struct page **pages, size_t size) |
1045 | { | |
1046 | int i; | |
1047 | ||
1048 | if (!pages) | |
1049 | return; | |
1050 | ||
1051 | for (i = 0; i < size; i++) { | |
1052 | if (!pages[i]) | |
1053 | break; | |
1054 | __free_page(pages[i]); | |
1055 | } | |
1056 | kfree(pages); | |
1057 | } | |
1058 | ||
1059 | static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags) | |
1060 | { | |
1061 | struct page **pages; | |
1062 | int i; | |
1063 | ||
a2791d3a | 1064 | pages = kmalloc_array(size, sizeof(struct page *), gfp_flags); |
29a8bfe5 TM |
1065 | if (!pages) { |
1066 | dprintk("%s: can't alloc array of %zu pages\n", __func__, size); | |
1067 | return NULL; | |
1068 | } | |
1069 | ||
1070 | for (i = 0; i < size; i++) { | |
1071 | pages[i] = alloc_page(gfp_flags); | |
1072 | if (!pages[i]) { | |
1073 | dprintk("%s: failed to allocate page\n", __func__); | |
a2791d3a | 1074 | nfs4_free_pages(pages, i); |
29a8bfe5 TM |
1075 | return NULL; |
1076 | } | |
1077 | } | |
1078 | ||
1079 | return pages; | |
1080 | } | |
1081 | ||
587f03de | 1082 | static struct nfs4_layoutget * |
5e36e2a9 | 1083 | pnfs_alloc_init_layoutget_args(struct inode *ino, |
e5e94017 | 1084 | struct nfs_open_context *ctx, |
2409a976 | 1085 | const nfs4_stateid *stateid, |
e144e539 | 1086 | const struct pnfs_layout_range *range, |
587f03de | 1087 | gfp_t gfp_flags) |
e5e94017 | 1088 | { |
5e36e2a9 | 1089 | struct nfs_server *server = pnfs_find_server(ino, ctx); |
28ced9a8 | 1090 | size_t max_reply_sz = server->pnfs_curr_ld->max_layoutget_response; |
dacb452d | 1091 | size_t max_pages = max_response_pages(server); |
b1f69b75 | 1092 | struct nfs4_layoutget *lgp; |
b1f69b75 AA |
1093 | |
1094 | dprintk("--> %s\n", __func__); | |
e5e94017 | 1095 | |
83026d80 JL |
1096 | lgp = kzalloc(sizeof(*lgp), gfp_flags); |
1097 | if (lgp == NULL) | |
587f03de | 1098 | return NULL; |
83026d80 | 1099 | |
28ced9a8 TM |
1100 | if (max_reply_sz) { |
1101 | size_t npages = (max_reply_sz + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1102 | if (npages < max_pages) | |
1103 | max_pages = npages; | |
1104 | } | |
1105 | ||
dacb452d FI |
1106 | lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags); |
1107 | if (!lgp->args.layout.pages) { | |
1108 | kfree(lgp); | |
1109 | return NULL; | |
1110 | } | |
1111 | lgp->args.layout.pglen = max_pages * PAGE_SIZE; | |
1112 | lgp->res.layoutp = &lgp->args.layout; | |
1113 | ||
d49e0d5b TM |
1114 | /* Don't confuse uninitialised result and success */ |
1115 | lgp->res.status = -NFS4ERR_DELAY; | |
83026d80 JL |
1116 | |
1117 | lgp->args.minlength = PAGE_SIZE; | |
1118 | if (lgp->args.minlength > range->length) | |
1119 | lgp->args.minlength = range->length; | |
5e36e2a9 FI |
1120 | if (ino) { |
1121 | loff_t i_size = i_size_read(ino); | |
1122 | ||
1123 | if (range->iomode == IOMODE_READ) { | |
1124 | if (range->offset >= i_size) | |
1125 | lgp->args.minlength = 0; | |
1126 | else if (i_size - range->offset < lgp->args.minlength) | |
1127 | lgp->args.minlength = i_size - range->offset; | |
1128 | } | |
d03ab29d | 1129 | } |
83026d80 JL |
1130 | lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; |
1131 | pnfs_copy_range(&lgp->args.range, range); | |
1132 | lgp->args.type = server->pnfs_curr_ld->id; | |
1133 | lgp->args.inode = ino; | |
1134 | lgp->args.ctx = get_nfs_open_context(ctx); | |
183d9e7b | 1135 | nfs4_stateid_copy(&lgp->args.stateid, stateid); |
83026d80 | 1136 | lgp->gfp_flags = gfp_flags; |
63ec2b69 | 1137 | lgp->cred = ctx->cred; |
587f03de | 1138 | return lgp; |
974cec8c AA |
1139 | } |
1140 | ||
29a8bfe5 TM |
1141 | void pnfs_layoutget_free(struct nfs4_layoutget *lgp) |
1142 | { | |
1143 | size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE; | |
1144 | ||
1145 | nfs4_free_pages(lgp->args.layout.pages, max_pages); | |
b4e89bcb | 1146 | pnfs_put_layout_hdr(lgp->lo); |
29a8bfe5 TM |
1147 | put_nfs_open_context(lgp->args.ctx); |
1148 | kfree(lgp); | |
1149 | } | |
1150 | ||
24956804 TM |
1151 | static void pnfs_clear_layoutcommit(struct inode *inode, |
1152 | struct list_head *head) | |
1153 | { | |
1154 | struct nfs_inode *nfsi = NFS_I(inode); | |
1155 | struct pnfs_layout_segment *lseg, *tmp; | |
1156 | ||
1157 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) | |
1158 | return; | |
1159 | list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) { | |
1160 | if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) | |
1161 | continue; | |
1162 | pnfs_lseg_dec_and_remove_zero(lseg, head); | |
1163 | } | |
1164 | } | |
1165 | ||
68f74479 | 1166 | void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo, |
2a974425 | 1167 | const nfs4_stateid *arg_stateid, |
68f74479 | 1168 | const struct pnfs_layout_range *range, |
68f74479 TM |
1169 | const nfs4_stateid *stateid) |
1170 | { | |
1171 | struct inode *inode = lo->plh_inode; | |
1172 | LIST_HEAD(freeme); | |
1173 | ||
1174 | spin_lock(&inode->i_lock); | |
c18d1e17 | 1175 | if (!pnfs_layout_is_valid(lo) || |
2a974425 TM |
1176 | !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid)) |
1177 | goto out_unlock; | |
68f74479 | 1178 | if (stateid) { |
2a974425 TM |
1179 | u32 seq = be32_to_cpu(arg_stateid->seqid); |
1180 | ||
68f74479 TM |
1181 | pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq); |
1182 | pnfs_free_returned_lsegs(lo, &freeme, range, seq); | |
59b56394 | 1183 | pnfs_set_layout_stateid(lo, stateid, NULL, true); |
68f74479 TM |
1184 | } else |
1185 | pnfs_mark_layout_stateid_invalid(lo, &freeme); | |
2a974425 | 1186 | out_unlock: |
68f74479 TM |
1187 | pnfs_clear_layoutreturn_waitbit(lo); |
1188 | spin_unlock(&inode->i_lock); | |
1189 | pnfs_free_lseg_list(&freeme); | |
1190 | ||
1191 | } | |
1192 | ||
13c13a6a | 1193 | static bool |
e5fd1904 TM |
1194 | pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo, |
1195 | nfs4_stateid *stateid, | |
44ea8dfc | 1196 | const struct cred **cred, |
e5fd1904 | 1197 | enum pnfs_iomode *iomode) |
13c13a6a | 1198 | { |
bf0291dd TM |
1199 | /* Serialise LAYOUTGET/LAYOUTRETURN */ |
1200 | if (atomic_read(&lo->plh_outstanding) != 0) | |
1201 | return false; | |
6604b203 | 1202 | if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) |
13c13a6a | 1203 | return false; |
6604b203 | 1204 | set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags); |
13c13a6a | 1205 | pnfs_get_layout_hdr(lo); |
1bcf34fd TM |
1206 | nfs4_stateid_copy(stateid, &lo->plh_stateid); |
1207 | *cred = get_cred(lo->plh_lc_cred); | |
e5fd1904 | 1208 | if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) { |
44ea8dfc TM |
1209 | if (lo->plh_return_seq != 0) |
1210 | stateid->seqid = cpu_to_be32(lo->plh_return_seq); | |
e5fd1904 TM |
1211 | if (iomode != NULL) |
1212 | *iomode = lo->plh_return_iomode; | |
1213 | pnfs_clear_layoutreturn_info(lo); | |
1bcf34fd | 1214 | } else if (iomode != NULL) |
e5fd1904 | 1215 | *iomode = IOMODE_ANY; |
1bcf34fd | 1216 | pnfs_barrier_update(lo, be32_to_cpu(stateid->seqid)); |
13c13a6a TM |
1217 | return true; |
1218 | } | |
1219 | ||
828ed9ec TM |
1220 | static void |
1221 | pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args, | |
1222 | struct pnfs_layout_hdr *lo, | |
1223 | const nfs4_stateid *stateid, | |
1224 | enum pnfs_iomode iomode) | |
1225 | { | |
1226 | struct inode *inode = lo->plh_inode; | |
1227 | ||
1228 | args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id; | |
1229 | args->inode = inode; | |
1230 | args->range.iomode = iomode; | |
1231 | args->range.offset = 0; | |
1232 | args->range.length = NFS4_MAX_UINT64; | |
1233 | args->layout = lo; | |
1234 | nfs4_stateid_copy(&args->stateid, stateid); | |
1235 | } | |
1236 | ||
f40eb5d0 | 1237 | static int |
44ea8dfc TM |
1238 | pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, |
1239 | const nfs4_stateid *stateid, | |
1240 | const struct cred **pcred, | |
1241 | enum pnfs_iomode iomode, | |
1242 | bool sync) | |
f40eb5d0 PT |
1243 | { |
1244 | struct inode *ino = lo->plh_inode; | |
287bd3e9 | 1245 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld; |
f40eb5d0 | 1246 | struct nfs4_layoutreturn *lrp; |
44ea8dfc | 1247 | const struct cred *cred = *pcred; |
f40eb5d0 PT |
1248 | int status = 0; |
1249 | ||
44ea8dfc | 1250 | *pcred = NULL; |
63d8a41b | 1251 | lrp = kzalloc(sizeof(*lrp), nfs_io_gfp_mask()); |
f40eb5d0 PT |
1252 | if (unlikely(lrp == NULL)) { |
1253 | status = -ENOMEM; | |
1254 | spin_lock(&ino->i_lock); | |
d67ae825 | 1255 | pnfs_clear_layoutreturn_waitbit(lo); |
f40eb5d0 | 1256 | spin_unlock(&ino->i_lock); |
44ea8dfc | 1257 | put_cred(cred); |
f40eb5d0 PT |
1258 | pnfs_put_layout_hdr(lo); |
1259 | goto out; | |
1260 | } | |
1261 | ||
828ed9ec | 1262 | pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode); |
4d796d75 | 1263 | lrp->args.ld_private = &lrp->ld_private; |
f40eb5d0 | 1264 | lrp->clp = NFS_SERVER(ino)->nfs_client; |
44ea8dfc | 1265 | lrp->cred = cred; |
287bd3e9 TM |
1266 | if (ld->prepare_layoutreturn) |
1267 | ld->prepare_layoutreturn(&lrp->args); | |
f40eb5d0 | 1268 | |
6c16605d | 1269 | status = nfs4_proc_layoutreturn(lrp, sync); |
f40eb5d0 PT |
1270 | out: |
1271 | dprintk("<-- %s status: %d\n", __func__, status); | |
1272 | return status; | |
1273 | } | |
1274 | ||
d474f961 TM |
1275 | static bool |
1276 | pnfs_layout_segments_returnable(struct pnfs_layout_hdr *lo, | |
1277 | enum pnfs_iomode iomode, | |
1278 | u32 seq) | |
1279 | { | |
1280 | struct pnfs_layout_range recall_range = { | |
1281 | .length = NFS4_MAX_UINT64, | |
1282 | .iomode = iomode, | |
1283 | }; | |
1284 | return pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, | |
1285 | &recall_range, seq) != -EBUSY; | |
1286 | } | |
1287 | ||
13c13a6a TM |
1288 | /* Return true if layoutreturn is needed */ |
1289 | static bool | |
1290 | pnfs_layout_need_return(struct pnfs_layout_hdr *lo) | |
1291 | { | |
2370abda | 1292 | if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) |
13c13a6a | 1293 | return false; |
d474f961 TM |
1294 | return pnfs_layout_segments_returnable(lo, lo->plh_return_iomode, |
1295 | lo->plh_return_seq); | |
13c13a6a TM |
1296 | } |
1297 | ||
1298 | static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo) | |
1299 | { | |
1300 | struct inode *inode= lo->plh_inode; | |
1301 | ||
2370abda | 1302 | if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) |
13c13a6a TM |
1303 | return; |
1304 | spin_lock(&inode->i_lock); | |
1305 | if (pnfs_layout_need_return(lo)) { | |
44ea8dfc | 1306 | const struct cred *cred; |
13c13a6a TM |
1307 | nfs4_stateid stateid; |
1308 | enum pnfs_iomode iomode; | |
1309 | bool send; | |
1310 | ||
44ea8dfc | 1311 | send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode); |
13c13a6a TM |
1312 | spin_unlock(&inode->i_lock); |
1313 | if (send) { | |
1314 | /* Send an async layoutreturn so we dont deadlock */ | |
44ea8dfc | 1315 | pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false); |
13c13a6a TM |
1316 | } |
1317 | } else | |
1318 | spin_unlock(&inode->i_lock); | |
1319 | } | |
1320 | ||
293b3b06 AA |
1321 | /* |
1322 | * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr | |
1323 | * when the layout segment list is empty. | |
1324 | * | |
1325 | * Note that a pnfs_layout_hdr can exist with an empty layout segment | |
1326 | * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the | |
1327 | * deviceid is marked invalid. | |
1328 | */ | |
cbe82603 BH |
1329 | int |
1330 | _pnfs_return_layout(struct inode *ino) | |
1331 | { | |
1332 | struct pnfs_layout_hdr *lo = NULL; | |
1333 | struct nfs_inode *nfsi = NFS_I(ino); | |
a421d218 AS |
1334 | struct pnfs_layout_range range = { |
1335 | .iomode = IOMODE_ANY, | |
1336 | .offset = 0, | |
1337 | .length = NFS4_MAX_UINT64, | |
1338 | }; | |
cbe82603 | 1339 | LIST_HEAD(tmp_list); |
44ea8dfc | 1340 | const struct cred *cred; |
cbe82603 | 1341 | nfs4_stateid stateid; |
24408f52 | 1342 | int status = 0; |
93b7f7ad | 1343 | bool send, valid_layout; |
cbe82603 | 1344 | |
366d5052 | 1345 | dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); |
cbe82603 BH |
1346 | |
1347 | spin_lock(&ino->i_lock); | |
1348 | lo = nfsi->layout; | |
e5929f3c | 1349 | if (!lo) { |
cbe82603 | 1350 | spin_unlock(&ino->i_lock); |
293b3b06 AA |
1351 | dprintk("NFS: %s no layout to return\n", __func__); |
1352 | goto out; | |
cbe82603 | 1353 | } |
cbe82603 | 1354 | /* Reference matched in nfs4_layoutreturn_release */ |
70c3bd2b | 1355 | pnfs_get_layout_hdr(lo); |
24408f52 TM |
1356 | /* Is there an outstanding layoutreturn ? */ |
1357 | if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) { | |
1358 | spin_unlock(&ino->i_lock); | |
1359 | if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN, | |
1360 | TASK_UNINTERRUPTIBLE)) | |
1361 | goto out_put_layout_hdr; | |
1362 | spin_lock(&ino->i_lock); | |
1363 | } | |
93b7f7ad | 1364 | valid_layout = pnfs_layout_is_valid(lo); |
24956804 | 1365 | pnfs_clear_layoutcommit(ino, &tmp_list); |
a421d218 AS |
1366 | pnfs_mark_matching_lsegs_return(lo, &tmp_list, &range, 0); |
1367 | ||
1368 | if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) | |
c88953d8 | 1369 | NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range); |
c88953d8 | 1370 | |
293b3b06 | 1371 | /* Don't send a LAYOUTRETURN if list was initially empty */ |
93b7f7ad OK |
1372 | if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) || |
1373 | !valid_layout) { | |
293b3b06 | 1374 | spin_unlock(&ino->i_lock); |
293b3b06 | 1375 | dprintk("NFS: %s no layout segments to return\n", __func__); |
7bcc1058 | 1376 | goto out_wait_layoutreturn; |
293b3b06 | 1377 | } |
47abadef | 1378 | |
44ea8dfc | 1379 | send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, NULL); |
cbe82603 | 1380 | spin_unlock(&ino->i_lock); |
7f27392c | 1381 | if (send) |
44ea8dfc | 1382 | status = pnfs_send_layoutreturn(lo, &stateid, &cred, IOMODE_ANY, true); |
7bcc1058 TM |
1383 | out_wait_layoutreturn: |
1384 | wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN, TASK_UNINTERRUPTIBLE); | |
7f27392c | 1385 | out_put_layout_hdr: |
ee6625a9 | 1386 | pnfs_free_lseg_list(&tmp_list); |
7f27392c | 1387 | pnfs_put_layout_hdr(lo); |
cbe82603 BH |
1388 | out: |
1389 | dprintk("<-- %s status: %d\n", __func__, status); | |
1390 | return status; | |
1391 | } | |
1392 | ||
24028672 TM |
1393 | int |
1394 | pnfs_commit_and_return_layout(struct inode *inode) | |
1395 | { | |
1396 | struct pnfs_layout_hdr *lo; | |
1397 | int ret; | |
1398 | ||
1399 | spin_lock(&inode->i_lock); | |
1400 | lo = NFS_I(inode)->layout; | |
1401 | if (lo == NULL) { | |
1402 | spin_unlock(&inode->i_lock); | |
1403 | return 0; | |
1404 | } | |
1405 | pnfs_get_layout_hdr(lo); | |
1406 | /* Block new layoutgets and read/write to ds */ | |
1407 | lo->plh_block_lgets++; | |
1408 | spin_unlock(&inode->i_lock); | |
1409 | filemap_fdatawait(inode->i_mapping); | |
1410 | ret = pnfs_layoutcommit_inode(inode, true); | |
1411 | if (ret == 0) | |
1412 | ret = _pnfs_return_layout(inode); | |
1413 | spin_lock(&inode->i_lock); | |
1414 | lo->plh_block_lgets--; | |
1415 | spin_unlock(&inode->i_lock); | |
1416 | pnfs_put_layout_hdr(lo); | |
1417 | return ret; | |
1418 | } | |
1419 | ||
1c5bd76d TM |
1420 | bool pnfs_roc(struct inode *ino, |
1421 | struct nfs4_layoutreturn_args *args, | |
1422 | struct nfs4_layoutreturn_res *res, | |
a52458b4 | 1423 | const struct cred *cred) |
f7e8917a | 1424 | { |
40dd4b7a TM |
1425 | struct nfs_inode *nfsi = NFS_I(ino); |
1426 | struct nfs_open_context *ctx; | |
1427 | struct nfs4_state *state; | |
f7e8917a | 1428 | struct pnfs_layout_hdr *lo; |
1c5bd76d | 1429 | struct pnfs_layout_segment *lseg, *next; |
44ea8dfc | 1430 | const struct cred *lc_cred; |
193e3aa2 | 1431 | nfs4_stateid stateid; |
1c5bd76d TM |
1432 | enum pnfs_iomode iomode = 0; |
1433 | bool layoutreturn = false, roc = false; | |
e71708d4 | 1434 | bool skip_read = false; |
f7e8917a | 1435 | |
1c5bd76d TM |
1436 | if (!nfs_have_layout(ino)) |
1437 | return false; | |
29ade5db | 1438 | retry: |
0de43976 | 1439 | rcu_read_lock(); |
f7e8917a | 1440 | spin_lock(&ino->i_lock); |
40dd4b7a | 1441 | lo = nfsi->layout; |
0cdc329e | 1442 | if (!lo || !pnfs_layout_is_valid(lo) || |
9c6376eb TM |
1443 | test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
1444 | lo = NULL; | |
40dd4b7a | 1445 | goto out_noroc; |
9c6376eb TM |
1446 | } |
1447 | pnfs_get_layout_hdr(lo); | |
29ade5db | 1448 | if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) { |
29ade5db | 1449 | spin_unlock(&ino->i_lock); |
0de43976 | 1450 | rcu_read_unlock(); |
29ade5db TM |
1451 | wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN, |
1452 | TASK_UNINTERRUPTIBLE); | |
1453 | pnfs_put_layout_hdr(lo); | |
1454 | goto retry; | |
1455 | } | |
40dd4b7a | 1456 | |
e755d638 | 1457 | /* no roc if we hold a delegation */ |
e71708d4 TM |
1458 | if (nfs4_check_delegation(ino, FMODE_READ)) { |
1459 | if (nfs4_check_delegation(ino, FMODE_WRITE)) | |
1460 | goto out_noroc; | |
1461 | skip_read = true; | |
1462 | } | |
40dd4b7a | 1463 | |
0de43976 | 1464 | list_for_each_entry_rcu(ctx, &nfsi->open_files, list) { |
40dd4b7a | 1465 | state = ctx->state; |
e71708d4 TM |
1466 | if (state == NULL) |
1467 | continue; | |
40dd4b7a | 1468 | /* Don't return layout if there is open file state */ |
e71708d4 | 1469 | if (state->state & FMODE_WRITE) |
40dd4b7a | 1470 | goto out_noroc; |
e71708d4 TM |
1471 | if (state->state & FMODE_READ) |
1472 | skip_read = true; | |
40dd4b7a TM |
1473 | } |
1474 | ||
e755d638 | 1475 | |
1c5bd76d | 1476 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) { |
e71708d4 TM |
1477 | if (skip_read && lseg->pls_range.iomode == IOMODE_READ) |
1478 | continue; | |
e755d638 | 1479 | /* If we are sending layoutreturn, invalidate all valid lsegs */ |
1c5bd76d TM |
1480 | if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags)) |
1481 | continue; | |
1482 | /* | |
1483 | * Note: mark lseg for return so pnfs_layout_remove_lseg | |
1484 | * doesn't invalidate the layout for us. | |
1485 | */ | |
1486 | set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags); | |
1487 | if (!mark_lseg_invalid(lseg, &lo->plh_return_segs)) | |
1488 | continue; | |
1489 | pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0); | |
69820d22 TM |
1490 | } |
1491 | ||
1c5bd76d TM |
1492 | if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) |
1493 | goto out_noroc; | |
69820d22 | 1494 | |
500d701f | 1495 | /* ROC in two conditions: |
e755d638 PT |
1496 | * 1. there are ROC lsegs |
1497 | * 2. we don't send layoutreturn | |
e755d638 | 1498 | */ |
1c5bd76d | 1499 | /* lo ref dropped in pnfs_roc_release() */ |
44ea8dfc | 1500 | layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &lc_cred, &iomode); |
1c5bd76d | 1501 | /* If the creds don't match, we can't compound the layoutreturn */ |
4d8948c7 | 1502 | if (!layoutreturn || cred_fscmp(cred, lc_cred) != 0) |
1c5bd76d TM |
1503 | goto out_noroc; |
1504 | ||
1505 | roc = layoutreturn; | |
1506 | pnfs_init_layoutreturn_args(args, lo, &stateid, iomode); | |
1507 | res->lrs_present = 0; | |
1508 | layoutreturn = false; | |
44ea8dfc | 1509 | put_cred(lc_cred); |
4d8948c7 | 1510 | |
40dd4b7a | 1511 | out_noroc: |
f7e8917a | 1512 | spin_unlock(&ino->i_lock); |
0de43976 | 1513 | rcu_read_unlock(); |
e755d638 | 1514 | pnfs_layoutcommit_inode(ino, true); |
287bd3e9 TM |
1515 | if (roc) { |
1516 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld; | |
1517 | if (ld->prepare_layoutreturn) | |
1518 | ld->prepare_layoutreturn(args); | |
9c6376eb | 1519 | pnfs_put_layout_hdr(lo); |
287bd3e9 TM |
1520 | return true; |
1521 | } | |
e755d638 | 1522 | if (layoutreturn) |
44ea8dfc | 1523 | pnfs_send_layoutreturn(lo, &stateid, &lc_cred, iomode, true); |
9c6376eb | 1524 | pnfs_put_layout_hdr(lo); |
287bd3e9 | 1525 | return false; |
f7e8917a FI |
1526 | } |
1527 | ||
078000d0 TM |
1528 | int pnfs_roc_done(struct rpc_task *task, struct nfs4_layoutreturn_args **argpp, |
1529 | struct nfs4_layoutreturn_res **respp, int *ret) | |
287a9c55 TM |
1530 | { |
1531 | struct nfs4_layoutreturn_args *arg = *argpp; | |
1532 | int retval = -EAGAIN; | |
1533 | ||
1534 | if (!arg) | |
1535 | return 0; | |
1536 | /* Handle Layoutreturn errors */ | |
1537 | switch (*ret) { | |
1538 | case 0: | |
1539 | retval = 0; | |
1540 | break; | |
6109bcf7 TM |
1541 | case -NFS4ERR_NOMATCHING_LAYOUT: |
1542 | /* Was there an RPC level error? If not, retry */ | |
1543 | if (task->tk_rpc_status == 0) | |
1544 | break; | |
1545 | /* If the call was not sent, let caller handle it */ | |
1546 | if (!RPC_WAS_SENT(task)) | |
1547 | return 0; | |
1548 | /* | |
1549 | * Otherwise, assume the call succeeded and | |
1550 | * that we need to release the layout | |
1551 | */ | |
1552 | *ret = 0; | |
1553 | (*respp)->lrs_present = 0; | |
1554 | retval = 0; | |
1555 | break; | |
078a432d TM |
1556 | case -NFS4ERR_DELAY: |
1557 | /* Let the caller handle the retry */ | |
1558 | *ret = -NFS4ERR_NOMATCHING_LAYOUT; | |
1559 | return 0; | |
287a9c55 | 1560 | case -NFS4ERR_OLD_STATEID: |
30cb3ee2 | 1561 | if (!nfs4_layout_refresh_old_stateid(&arg->stateid, |
078000d0 | 1562 | &arg->range, arg->inode)) |
287a9c55 TM |
1563 | break; |
1564 | *ret = -NFS4ERR_NOMATCHING_LAYOUT; | |
1565 | return -EAGAIN; | |
1566 | } | |
1567 | *argpp = NULL; | |
1568 | *respp = NULL; | |
1569 | return retval; | |
1570 | } | |
1571 | ||
1c5bd76d TM |
1572 | void pnfs_roc_release(struct nfs4_layoutreturn_args *args, |
1573 | struct nfs4_layoutreturn_res *res, | |
1574 | int ret) | |
f7e8917a | 1575 | { |
1c5bd76d | 1576 | struct pnfs_layout_hdr *lo = args->layout; |
67bbceed | 1577 | struct inode *inode = args->inode; |
1c5bd76d | 1578 | const nfs4_stateid *res_stateid = NULL; |
287bd3e9 | 1579 | struct nfs4_xdr_opaque_data *ld_private = args->ld_private; |
f7e8917a | 1580 | |
9c47b18c TM |
1581 | switch (ret) { |
1582 | case -NFS4ERR_NOMATCHING_LAYOUT: | |
67bbceed TM |
1583 | spin_lock(&inode->i_lock); |
1584 | if (pnfs_layout_is_valid(lo) && | |
1585 | nfs4_stateid_match_other(&args->stateid, &lo->plh_stateid)) | |
1586 | pnfs_set_plh_return_info(lo, args->range.iomode, 0); | |
c18d1e17 | 1587 | pnfs_clear_layoutreturn_waitbit(lo); |
67bbceed | 1588 | spin_unlock(&inode->i_lock); |
9c47b18c TM |
1589 | break; |
1590 | case 0: | |
1c5bd76d TM |
1591 | if (res->lrs_present) |
1592 | res_stateid = &res->stateid; | |
df561f66 | 1593 | fallthrough; |
9c47b18c | 1594 | default: |
c18d1e17 TM |
1595 | pnfs_layoutreturn_free_lsegs(lo, &args->stateid, &args->range, |
1596 | res_stateid); | |
1c5bd76d | 1597 | } |
a19b4785 | 1598 | trace_nfs4_layoutreturn_on_close(args->inode, &args->stateid, ret); |
287bd3e9 TM |
1599 | if (ld_private && ld_private->ops && ld_private->ops->free) |
1600 | ld_private->ops->free(ld_private); | |
1c5bd76d | 1601 | pnfs_put_layout_hdr(lo); |
f7e8917a FI |
1602 | } |
1603 | ||
500d701f PT |
1604 | bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task) |
1605 | { | |
1606 | struct nfs_inode *nfsi = NFS_I(ino); | |
1607 | struct pnfs_layout_hdr *lo; | |
1608 | bool sleep = false; | |
1609 | ||
1610 | /* we might not have grabbed lo reference. so need to check under | |
1611 | * i_lock */ | |
1612 | spin_lock(&ino->i_lock); | |
1613 | lo = nfsi->layout; | |
ee284e35 TM |
1614 | if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) { |
1615 | rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); | |
500d701f | 1616 | sleep = true; |
ee284e35 | 1617 | } |
500d701f | 1618 | spin_unlock(&ino->i_lock); |
500d701f PT |
1619 | return sleep; |
1620 | } | |
1621 | ||
b1f69b75 AA |
1622 | /* |
1623 | * Compare two layout segments for sorting into layout cache. | |
1624 | * We want to preferentially return RW over RO layouts, so ensure those | |
1625 | * are seen first. | |
1626 | */ | |
1627 | static s64 | |
7dc0ac70 | 1628 | pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1, |
3cb2df17 | 1629 | const struct pnfs_layout_range *l2) |
b1f69b75 | 1630 | { |
fb3296eb BH |
1631 | s64 d; |
1632 | ||
1633 | /* high offset > low offset */ | |
1634 | d = l1->offset - l2->offset; | |
1635 | if (d) | |
1636 | return d; | |
1637 | ||
1638 | /* short length > long length */ | |
1639 | d = l2->length - l1->length; | |
1640 | if (d) | |
1641 | return d; | |
1642 | ||
b1f69b75 | 1643 | /* read > read/write */ |
fb3296eb | 1644 | return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); |
b1f69b75 AA |
1645 | } |
1646 | ||
03772d2f TM |
1647 | static bool |
1648 | pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1, | |
1649 | const struct pnfs_layout_range *l2) | |
1650 | { | |
1651 | return pnfs_lseg_range_cmp(l1, l2) > 0; | |
1652 | } | |
1653 | ||
1654 | static bool | |
1655 | pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg, | |
1656 | struct pnfs_layout_segment *old) | |
1657 | { | |
1658 | return false; | |
1659 | } | |
1660 | ||
1661 | void | |
1662 | pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo, | |
1663 | struct pnfs_layout_segment *lseg, | |
1664 | bool (*is_after)(const struct pnfs_layout_range *, | |
1665 | const struct pnfs_layout_range *), | |
1666 | bool (*do_merge)(struct pnfs_layout_segment *, | |
1667 | struct pnfs_layout_segment *), | |
1668 | struct list_head *free_me) | |
974cec8c | 1669 | { |
03772d2f | 1670 | struct pnfs_layout_segment *lp, *tmp; |
b1f69b75 | 1671 | |
974cec8c AA |
1672 | dprintk("%s:Begin\n", __func__); |
1673 | ||
03772d2f TM |
1674 | list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) { |
1675 | if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0) | |
1676 | continue; | |
1677 | if (do_merge(lseg, lp)) { | |
1678 | mark_lseg_invalid(lp, free_me); | |
1679 | continue; | |
1680 | } | |
1681 | if (is_after(&lseg->pls_range, &lp->pls_range)) | |
b1f69b75 | 1682 | continue; |
566052c5 | 1683 | list_add_tail(&lseg->pls_list, &lp->pls_list); |
b1f69b75 AA |
1684 | dprintk("%s: inserted lseg %p " |
1685 | "iomode %d offset %llu length %llu before " | |
1686 | "lp %p iomode %d offset %llu length %llu\n", | |
566052c5 FI |
1687 | __func__, lseg, lseg->pls_range.iomode, |
1688 | lseg->pls_range.offset, lseg->pls_range.length, | |
1689 | lp, lp->pls_range.iomode, lp->pls_range.offset, | |
1690 | lp->pls_range.length); | |
fb3296eb | 1691 | goto out; |
974cec8c | 1692 | } |
fb3296eb BH |
1693 | list_add_tail(&lseg->pls_list, &lo->plh_segs); |
1694 | dprintk("%s: inserted lseg %p " | |
1695 | "iomode %d offset %llu length %llu at tail\n", | |
1696 | __func__, lseg, lseg->pls_range.iomode, | |
1697 | lseg->pls_range.offset, lseg->pls_range.length); | |
1698 | out: | |
70c3bd2b | 1699 | pnfs_get_layout_hdr(lo); |
974cec8c AA |
1700 | |
1701 | dprintk("%s:Return\n", __func__); | |
e5e94017 | 1702 | } |
03772d2f TM |
1703 | EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg); |
1704 | ||
1705 | static void | |
1706 | pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo, | |
1707 | struct pnfs_layout_segment *lseg, | |
1708 | struct list_head *free_me) | |
1709 | { | |
1710 | struct inode *inode = lo->plh_inode; | |
1711 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
1712 | ||
1713 | if (ld->add_lseg != NULL) | |
1714 | ld->add_lseg(lo, lseg, free_me); | |
1715 | else | |
1716 | pnfs_generic_layout_insert_lseg(lo, lseg, | |
1717 | pnfs_lseg_range_is_after, | |
1718 | pnfs_lseg_no_merge, | |
1719 | free_me); | |
1720 | } | |
e5e94017 BH |
1721 | |
1722 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1723 | alloc_init_layout_hdr(struct inode *ino, |
1724 | struct nfs_open_context *ctx, | |
1725 | gfp_t gfp_flags) | |
e5e94017 BH |
1726 | { |
1727 | struct pnfs_layout_hdr *lo; | |
1728 | ||
636fb9c8 | 1729 | lo = pnfs_alloc_layout_hdr(ino, gfp_flags); |
e5e94017 BH |
1730 | if (!lo) |
1731 | return NULL; | |
2b28a7be | 1732 | refcount_set(&lo->plh_refcount, 1); |
b7edfaa1 FI |
1733 | INIT_LIST_HEAD(&lo->plh_layouts); |
1734 | INIT_LIST_HEAD(&lo->plh_segs); | |
68f74479 | 1735 | INIT_LIST_HEAD(&lo->plh_return_segs); |
fd9a8d71 | 1736 | INIT_LIST_HEAD(&lo->plh_bulk_destroy); |
b7edfaa1 | 1737 | lo->plh_inode = ino; |
a52458b4 | 1738 | lo->plh_lc_cred = get_cred(ctx->cred); |
67a3b721 | 1739 | lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID; |
e5e94017 BH |
1740 | return lo; |
1741 | } | |
1742 | ||
1743 | static struct pnfs_layout_hdr * | |
9fa40758 PT |
1744 | pnfs_find_alloc_layout(struct inode *ino, |
1745 | struct nfs_open_context *ctx, | |
1746 | gfp_t gfp_flags) | |
e5241e43 TM |
1747 | __releases(&ino->i_lock) |
1748 | __acquires(&ino->i_lock) | |
e5e94017 BH |
1749 | { |
1750 | struct nfs_inode *nfsi = NFS_I(ino); | |
1751 | struct pnfs_layout_hdr *new = NULL; | |
1752 | ||
1753 | dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); | |
1754 | ||
251ec410 TM |
1755 | if (nfsi->layout != NULL) |
1756 | goto out_existing; | |
e5e94017 | 1757 | spin_unlock(&ino->i_lock); |
9fa40758 | 1758 | new = alloc_init_layout_hdr(ino, ctx, gfp_flags); |
e5e94017 BH |
1759 | spin_lock(&ino->i_lock); |
1760 | ||
251ec410 | 1761 | if (likely(nfsi->layout == NULL)) { /* Won the race? */ |
e5e94017 | 1762 | nfsi->layout = new; |
251ec410 | 1763 | return new; |
7175fe90 YN |
1764 | } else if (new != NULL) |
1765 | pnfs_free_layout_hdr(new); | |
251ec410 TM |
1766 | out_existing: |
1767 | pnfs_get_layout_hdr(nfsi->layout); | |
e5e94017 BH |
1768 | return nfsi->layout; |
1769 | } | |
1770 | ||
b1f69b75 AA |
1771 | /* |
1772 | * iomode matching rules: | |
c7d73af2 TH |
1773 | * iomode lseg strict match |
1774 | * iomode | |
1775 | * ----- ----- ------ ----- | |
1776 | * ANY READ N/A true | |
1777 | * ANY RW N/A true | |
1778 | * RW READ N/A false | |
1779 | * RW RW N/A true | |
1780 | * READ READ N/A true | |
1781 | * READ RW true false | |
1782 | * READ RW false true | |
b1f69b75 | 1783 | */ |
3cb2df17 | 1784 | static bool |
7dc0ac70 | 1785 | pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range, |
c7d73af2 TH |
1786 | const struct pnfs_layout_range *range, |
1787 | bool strict_iomode) | |
b1f69b75 | 1788 | { |
fb3296eb BH |
1789 | struct pnfs_layout_range range1; |
1790 | ||
1791 | if ((range->iomode == IOMODE_RW && | |
1792 | ls_range->iomode != IOMODE_RW) || | |
c7d73af2 | 1793 | (range->iomode != ls_range->iomode && |
6089dd0d | 1794 | strict_iomode) || |
7dc0ac70 | 1795 | !pnfs_lseg_range_intersecting(ls_range, range)) |
10db5b7a | 1796 | return false; |
fb3296eb BH |
1797 | |
1798 | /* range1 covers only the first byte in the range */ | |
1799 | range1 = *range; | |
1800 | range1.length = 1; | |
7dc0ac70 | 1801 | return pnfs_lseg_range_contained(ls_range, &range1); |
b1f69b75 AA |
1802 | } |
1803 | ||
1804 | /* | |
1805 | * lookup range in layout | |
1806 | */ | |
e5e94017 | 1807 | static struct pnfs_layout_segment * |
fb3296eb | 1808 | pnfs_find_lseg(struct pnfs_layout_hdr *lo, |
c7d73af2 TH |
1809 | struct pnfs_layout_range *range, |
1810 | bool strict_iomode) | |
e5e94017 | 1811 | { |
b1f69b75 AA |
1812 | struct pnfs_layout_segment *lseg, *ret = NULL; |
1813 | ||
1814 | dprintk("%s:Begin\n", __func__); | |
1815 | ||
b7edfaa1 | 1816 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { |
4541d16c | 1817 | if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && |
c7d73af2 TH |
1818 | pnfs_lseg_range_match(&lseg->pls_range, range, |
1819 | strict_iomode)) { | |
9369a431 | 1820 | ret = pnfs_get_lseg(lseg); |
b1f69b75 AA |
1821 | break; |
1822 | } | |
b1f69b75 AA |
1823 | } |
1824 | ||
1825 | dprintk("%s:Return lseg %p ref %d\n", | |
eba6dd69 | 1826 | __func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0); |
b1f69b75 | 1827 | return ret; |
e5e94017 BH |
1828 | } |
1829 | ||
d23d61c8 AA |
1830 | /* |
1831 | * Use mdsthreshold hints set at each OPEN to determine if I/O should go | |
1832 | * to the MDS or over pNFS | |
1833 | * | |
1834 | * The nfs_inode read_io and write_io fields are cumulative counters reset | |
1835 | * when there are no layout segments. Note that in pnfs_update_layout iomode | |
1836 | * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a | |
1837 | * WRITE request. | |
1838 | * | |
1839 | * A return of true means use MDS I/O. | |
1840 | * | |
1841 | * From rfc 5661: | |
1842 | * If a file's size is smaller than the file size threshold, data accesses | |
1843 | * SHOULD be sent to the metadata server. If an I/O request has a length that | |
1844 | * is below the I/O size threshold, the I/O SHOULD be sent to the metadata | |
1845 | * server. If both file size and I/O size are provided, the client SHOULD | |
1846 | * reach or exceed both thresholds before sending its read or write | |
1847 | * requests to the data server. | |
1848 | */ | |
1849 | static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, | |
1850 | struct inode *ino, int iomode) | |
1851 | { | |
1852 | struct nfs4_threshold *t = ctx->mdsthreshold; | |
1853 | struct nfs_inode *nfsi = NFS_I(ino); | |
1854 | loff_t fsize = i_size_read(ino); | |
1855 | bool size = false, size_set = false, io = false, io_set = false, ret = false; | |
1856 | ||
1857 | if (t == NULL) | |
1858 | return ret; | |
1859 | ||
1860 | dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", | |
1861 | __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); | |
1862 | ||
1863 | switch (iomode) { | |
1864 | case IOMODE_READ: | |
1865 | if (t->bm & THRESHOLD_RD) { | |
1866 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1867 | size_set = true; | |
1868 | if (fsize < t->rd_sz) | |
1869 | size = true; | |
1870 | } | |
1871 | if (t->bm & THRESHOLD_RD_IO) { | |
1872 | dprintk("%s nfsi->read_io %llu\n", __func__, | |
1873 | nfsi->read_io); | |
1874 | io_set = true; | |
1875 | if (nfsi->read_io < t->rd_io_sz) | |
1876 | io = true; | |
1877 | } | |
1878 | break; | |
1879 | case IOMODE_RW: | |
1880 | if (t->bm & THRESHOLD_WR) { | |
1881 | dprintk("%s fsize %llu\n", __func__, fsize); | |
1882 | size_set = true; | |
1883 | if (fsize < t->wr_sz) | |
1884 | size = true; | |
1885 | } | |
1886 | if (t->bm & THRESHOLD_WR_IO) { | |
1887 | dprintk("%s nfsi->write_io %llu\n", __func__, | |
1888 | nfsi->write_io); | |
1889 | io_set = true; | |
1890 | if (nfsi->write_io < t->wr_io_sz) | |
1891 | io = true; | |
1892 | } | |
1893 | break; | |
1894 | } | |
1895 | if (size_set && io_set) { | |
1896 | if (size && io) | |
1897 | ret = true; | |
1898 | } else if (size || io) | |
1899 | ret = true; | |
1900 | ||
1901 | dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); | |
1902 | return ret; | |
1903 | } | |
1904 | ||
d03360aa | 1905 | static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo) |
aa8a45ee PT |
1906 | { |
1907 | /* | |
1908 | * send layoutcommit as it can hold up layoutreturn due to lseg | |
1909 | * reference | |
1910 | */ | |
1911 | pnfs_layoutcommit_inode(lo->plh_inode, false); | |
d03360aa | 1912 | return wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN, |
2e5b29f0 | 1913 | nfs_wait_bit_killable, |
f5d39b02 | 1914 | TASK_KILLABLE|TASK_FREEZABLE_UNSAFE); |
aa8a45ee PT |
1915 | } |
1916 | ||
411ae722 TM |
1917 | static void nfs_layoutget_begin(struct pnfs_layout_hdr *lo) |
1918 | { | |
1919 | atomic_inc(&lo->plh_outstanding); | |
1920 | } | |
1921 | ||
1922 | static void nfs_layoutget_end(struct pnfs_layout_hdr *lo) | |
1923 | { | |
880265c7 TM |
1924 | if (atomic_dec_and_test(&lo->plh_outstanding) && |
1925 | test_and_clear_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags)) | |
1926 | wake_up_bit(&lo->plh_flags, NFS_LAYOUT_DRAIN); | |
411ae722 TM |
1927 | } |
1928 | ||
d29b468d TM |
1929 | static bool pnfs_is_first_layoutget(struct pnfs_layout_hdr *lo) |
1930 | { | |
1931 | return test_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags); | |
1932 | } | |
1933 | ||
d67ae825 TH |
1934 | static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo) |
1935 | { | |
1936 | unsigned long *bitlock = &lo->plh_flags; | |
1937 | ||
1938 | clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock); | |
1939 | smp_mb__after_atomic(); | |
1940 | wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET); | |
1941 | } | |
1942 | ||
78746a38 FI |
1943 | static void _add_to_server_list(struct pnfs_layout_hdr *lo, |
1944 | struct nfs_server *server) | |
1945 | { | |
cf6605d1 | 1946 | if (!test_and_set_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) { |
78746a38 FI |
1947 | struct nfs_client *clp = server->nfs_client; |
1948 | ||
1949 | /* The lo must be on the clp list if there is any | |
1950 | * chance of a CB_LAYOUTRECALL(FILE) coming in. | |
1951 | */ | |
1952 | spin_lock(&clp->cl_lock); | |
cf6605d1 | 1953 | list_add_tail_rcu(&lo->plh_layouts, &server->layouts); |
78746a38 FI |
1954 | spin_unlock(&clp->cl_lock); |
1955 | } | |
1956 | } | |
1957 | ||
e5e94017 BH |
1958 | /* |
1959 | * Layout segment is retreived from the server if not cached. | |
1960 | * The appropriate layout segment is referenced and returned to the caller. | |
1961 | */ | |
7c24d948 | 1962 | struct pnfs_layout_segment * |
e5e94017 BH |
1963 | pnfs_update_layout(struct inode *ino, |
1964 | struct nfs_open_context *ctx, | |
fb3296eb BH |
1965 | loff_t pos, |
1966 | u64 count, | |
a75b9df9 | 1967 | enum pnfs_iomode iomode, |
c7d73af2 | 1968 | bool strict_iomode, |
a75b9df9 | 1969 | gfp_t gfp_flags) |
e5e94017 | 1970 | { |
fb3296eb BH |
1971 | struct pnfs_layout_range arg = { |
1972 | .iomode = iomode, | |
1973 | .offset = pos, | |
1974 | .length = count, | |
1975 | }; | |
70d2f7b1 | 1976 | unsigned pg_offset; |
6382a441 WAA |
1977 | struct nfs_server *server = NFS_SERVER(ino); |
1978 | struct nfs_client *clp = server->nfs_client; | |
183d9e7b | 1979 | struct pnfs_layout_hdr *lo = NULL; |
e5e94017 | 1980 | struct pnfs_layout_segment *lseg = NULL; |
587f03de | 1981 | struct nfs4_layoutget *lgp; |
183d9e7b JL |
1982 | nfs4_stateid stateid; |
1983 | long timeout = 0; | |
66b53f32 | 1984 | unsigned long giveup = jiffies + (clp->cl_lease_time << 1); |
30005121 | 1985 | bool first; |
e5e94017 | 1986 | |
9a4bf31d | 1987 | if (!pnfs_enabled_sb(NFS_SERVER(ino))) { |
183d9e7b | 1988 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1989 | PNFS_UPDATE_LAYOUT_NO_PNFS); |
f86bbcf8 | 1990 | goto out; |
9a4bf31d | 1991 | } |
d23d61c8 | 1992 | |
9a4bf31d | 1993 | if (pnfs_within_mdsthreshold(ctx, ino, iomode)) { |
183d9e7b | 1994 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 1995 | PNFS_UPDATE_LAYOUT_MDSTHRESH); |
f86bbcf8 | 1996 | goto out; |
9a4bf31d | 1997 | } |
d23d61c8 | 1998 | |
9bf87482 | 1999 | lookup_again: |
d03360aa TM |
2000 | lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp)); |
2001 | if (IS_ERR(lseg)) | |
2002 | goto out; | |
9bf87482 | 2003 | first = false; |
e5e94017 | 2004 | spin_lock(&ino->i_lock); |
9fa40758 | 2005 | lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); |
830ffb56 TM |
2006 | if (lo == NULL) { |
2007 | spin_unlock(&ino->i_lock); | |
3764a17e | 2008 | lseg = ERR_PTR(-ENOMEM); |
183d9e7b | 2009 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 2010 | PNFS_UPDATE_LAYOUT_NOMEM); |
830ffb56 TM |
2011 | goto out; |
2012 | } | |
e5e94017 | 2013 | |
43f1b3da | 2014 | /* Do we even need to bother with this? */ |
a59c30ac | 2015 | if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { |
183d9e7b | 2016 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 2017 | PNFS_UPDATE_LAYOUT_BULK_RECALL); |
43f1b3da | 2018 | dprintk("%s matches recall, use MDS\n", __func__); |
e5e94017 BH |
2019 | goto out_unlock; |
2020 | } | |
2021 | ||
2022 | /* if LAYOUTGET already failed once we don't try again */ | |
2e5b29f0 | 2023 | if (pnfs_layout_io_test_failed(lo, iomode)) { |
183d9e7b | 2024 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 2025 | PNFS_UPDATE_LAYOUT_IO_TEST_FAIL); |
e5e94017 | 2026 | goto out_unlock; |
9a4bf31d | 2027 | } |
e5e94017 | 2028 | |
411ae722 TM |
2029 | /* |
2030 | * If the layout segment list is empty, but there are outstanding | |
2031 | * layoutget calls, then they might be subject to a layoutrecall. | |
2032 | */ | |
880265c7 | 2033 | if (test_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags) && |
411ae722 TM |
2034 | atomic_read(&lo->plh_outstanding) != 0) { |
2035 | spin_unlock(&ino->i_lock); | |
880265c7 TM |
2036 | lseg = ERR_PTR(wait_on_bit(&lo->plh_flags, NFS_LAYOUT_DRAIN, |
2037 | TASK_KILLABLE)); | |
58bbeab4 | 2038 | if (IS_ERR(lseg)) |
411ae722 TM |
2039 | goto out_put_layout_hdr; |
2040 | pnfs_put_layout_hdr(lo); | |
2041 | goto lookup_again; | |
2042 | } | |
2043 | ||
2c8d5fc3 TM |
2044 | /* |
2045 | * Because we free lsegs when sending LAYOUTRETURN, we need to wait | |
2046 | * for LAYOUTRETURN. | |
2047 | */ | |
2048 | if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) { | |
2049 | spin_unlock(&ino->i_lock); | |
2050 | dprintk("%s wait for layoutreturn\n", __func__); | |
2051 | lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo)); | |
2052 | if (!IS_ERR(lseg)) { | |
2053 | pnfs_put_layout_hdr(lo); | |
2054 | dprintk("%s retrying\n", __func__); | |
2055 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, | |
2056 | lseg, | |
2057 | PNFS_UPDATE_LAYOUT_RETRY); | |
2058 | goto lookup_again; | |
2059 | } | |
2060 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, | |
2061 | PNFS_UPDATE_LAYOUT_RETURN); | |
2062 | goto out_put_layout_hdr; | |
2063 | } | |
2064 | ||
c7d73af2 | 2065 | lseg = pnfs_find_lseg(lo, &arg, strict_iomode); |
183d9e7b JL |
2066 | if (lseg) { |
2067 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, | |
2068 | PNFS_UPDATE_LAYOUT_FOUND_CACHED); | |
2069 | goto out_unlock; | |
2070 | } | |
2071 | ||
183d9e7b JL |
2072 | /* |
2073 | * Choose a stateid for the LAYOUTGET. If we don't have a layout | |
2074 | * stateid, or it has been invalidated, then we must use the open | |
2075 | * stateid. | |
2076 | */ | |
67a3b721 | 2077 | if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) { |
d9aba2b4 | 2078 | int status; |
183d9e7b JL |
2079 | |
2080 | /* | |
2081 | * The first layoutget for the file. Need to serialize per | |
9bf87482 PT |
2082 | * RFC 5661 Errata 3208. |
2083 | */ | |
2084 | if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, | |
2085 | &lo->plh_flags)) { | |
2086 | spin_unlock(&ino->i_lock); | |
d03360aa TM |
2087 | lseg = ERR_PTR(wait_on_bit(&lo->plh_flags, |
2088 | NFS_LAYOUT_FIRST_LAYOUTGET, | |
2089 | TASK_KILLABLE)); | |
2090 | if (IS_ERR(lseg)) | |
2091 | goto out_put_layout_hdr; | |
9bf87482 | 2092 | pnfs_put_layout_hdr(lo); |
183d9e7b | 2093 | dprintk("%s retrying\n", __func__); |
9bf87482 PT |
2094 | goto lookup_again; |
2095 | } | |
183d9e7b | 2096 | |
fbf4bcc9 | 2097 | spin_unlock(&ino->i_lock); |
183d9e7b | 2098 | first = true; |
d9aba2b4 | 2099 | status = nfs4_select_rw_stateid(ctx->state, |
70d2f7b1 | 2100 | iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ, |
d9aba2b4 TM |
2101 | NULL, &stateid, NULL); |
2102 | if (status != 0) { | |
731c74dd | 2103 | lseg = ERR_PTR(status); |
70d2f7b1 TM |
2104 | trace_pnfs_update_layout(ino, pos, count, |
2105 | iomode, lo, lseg, | |
2106 | PNFS_UPDATE_LAYOUT_INVALID_OPEN); | |
d9aba2b4 TM |
2107 | nfs4_schedule_stateid_recovery(server, ctx->state); |
2108 | pnfs_clear_first_layoutget(lo); | |
2109 | pnfs_put_layout_hdr(lo); | |
2110 | goto lookup_again; | |
70d2f7b1 | 2111 | } |
fbf4bcc9 | 2112 | spin_lock(&ino->i_lock); |
9bf87482 | 2113 | } else { |
183d9e7b | 2114 | nfs4_stateid_copy(&stateid, &lo->plh_stateid); |
9bf87482 | 2115 | } |
568e8c49 | 2116 | |
9a4bf31d | 2117 | if (pnfs_layoutgets_blocked(lo)) { |
183d9e7b | 2118 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
9a4bf31d | 2119 | PNFS_UPDATE_LAYOUT_BLOCKED); |
cf7d63f1 | 2120 | goto out_unlock; |
9a4bf31d | 2121 | } |
411ae722 | 2122 | nfs_layoutget_begin(lo); |
f49f9baa | 2123 | spin_unlock(&ino->i_lock); |
30005121 | 2124 | |
78746a38 | 2125 | _add_to_server_list(lo, server); |
e5e94017 | 2126 | |
09cbfeaf | 2127 | pg_offset = arg.offset & ~PAGE_MASK; |
707ed5fd BH |
2128 | if (pg_offset) { |
2129 | arg.offset -= pg_offset; | |
2130 | arg.length += pg_offset; | |
2131 | } | |
7c24d948 | 2132 | if (arg.length != NFS4_MAX_UINT64) |
09cbfeaf | 2133 | arg.length = PAGE_ALIGN(arg.length); |
707ed5fd | 2134 | |
5e36e2a9 | 2135 | lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags); |
587f03de | 2136 | if (!lgp) { |
3764a17e | 2137 | lseg = ERR_PTR(-ENOMEM); |
587f03de FI |
2138 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL, |
2139 | PNFS_UPDATE_LAYOUT_NOMEM); | |
411ae722 | 2140 | nfs_layoutget_end(lo); |
587f03de FI |
2141 | goto out_put_layout_hdr; |
2142 | } | |
2143 | ||
b4e89bcb TM |
2144 | lgp->lo = lo; |
2145 | pnfs_get_layout_hdr(lo); | |
2146 | ||
dacb452d | 2147 | lseg = nfs4_proc_layoutget(lgp, &timeout); |
183d9e7b JL |
2148 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
2149 | PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET); | |
411ae722 | 2150 | nfs_layoutget_end(lo); |
83026d80 | 2151 | if (IS_ERR(lseg)) { |
183d9e7b | 2152 | switch(PTR_ERR(lseg)) { |
e85d7ee4 | 2153 | case -EBUSY: |
183d9e7b JL |
2154 | if (time_after(jiffies, giveup)) |
2155 | lseg = NULL; | |
66b53f32 TM |
2156 | break; |
2157 | case -ERECALLCONFLICT: | |
183d9e7b | 2158 | case -EAGAIN: |
56b38a1f | 2159 | break; |
fe44fb23 TM |
2160 | case -ENODATA: |
2161 | /* The server returned NFS4ERR_LAYOUTUNAVAILABLE */ | |
2162 | pnfs_layout_set_fail_bit( | |
2163 | lo, pnfs_iomode_to_fail_bit(iomode)); | |
2164 | lseg = NULL; | |
2165 | goto out_put_layout_hdr; | |
183d9e7b JL |
2166 | default: |
2167 | if (!nfs_error_is_fatal(PTR_ERR(lseg))) { | |
2168 | pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
2169 | lseg = NULL; | |
2170 | } | |
56b38a1f TM |
2171 | goto out_put_layout_hdr; |
2172 | } | |
2173 | if (lseg) { | |
2174 | if (first) | |
2175 | pnfs_clear_first_layoutget(lo); | |
2176 | trace_pnfs_update_layout(ino, pos, count, | |
2177 | iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY); | |
2178 | pnfs_put_layout_hdr(lo); | |
2179 | goto lookup_again; | |
83026d80 JL |
2180 | } |
2181 | } else { | |
2182 | pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
2183 | } | |
2184 | ||
830ffb56 | 2185 | out_put_layout_hdr: |
d67ae825 TH |
2186 | if (first) |
2187 | pnfs_clear_first_layoutget(lo); | |
d5b9216f TM |
2188 | trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, |
2189 | PNFS_UPDATE_LAYOUT_EXIT); | |
70c3bd2b | 2190 | pnfs_put_layout_hdr(lo); |
e5e94017 | 2191 | out: |
f86bbcf8 TM |
2192 | dprintk("%s: inode %s/%llu pNFS layout segment %s for " |
2193 | "(%s, offset: %llu, length: %llu)\n", | |
2194 | __func__, ino->i_sb->s_id, | |
2195 | (unsigned long long)NFS_FILEID(ino), | |
d600ad1f | 2196 | IS_ERR_OR_NULL(lseg) ? "not found" : "found", |
f86bbcf8 TM |
2197 | iomode==IOMODE_RW ? "read/write" : "read-only", |
2198 | (unsigned long long)pos, | |
2199 | (unsigned long long)count); | |
e5e94017 BH |
2200 | return lseg; |
2201 | out_unlock: | |
2202 | spin_unlock(&ino->i_lock); | |
830ffb56 | 2203 | goto out_put_layout_hdr; |
e5e94017 | 2204 | } |
7c24d948 | 2205 | EXPORT_SYMBOL_GPL(pnfs_update_layout); |
b1f69b75 | 2206 | |
540d9864 TM |
2207 | static bool |
2208 | pnfs_sanity_check_layout_range(struct pnfs_layout_range *range) | |
2209 | { | |
2210 | switch (range->iomode) { | |
2211 | case IOMODE_READ: | |
2212 | case IOMODE_RW: | |
2213 | break; | |
2214 | default: | |
2215 | return false; | |
2216 | } | |
2217 | if (range->offset == NFS4_MAX_UINT64) | |
2218 | return false; | |
2219 | if (range->length == 0) | |
2220 | return false; | |
2221 | if (range->length != NFS4_MAX_UINT64 && | |
2222 | range->length > NFS4_MAX_UINT64 - range->offset) | |
2223 | return false; | |
2224 | return true; | |
2225 | } | |
2226 | ||
78746a38 FI |
2227 | static struct pnfs_layout_hdr * |
2228 | _pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx) | |
2229 | { | |
2230 | struct pnfs_layout_hdr *lo; | |
2231 | ||
2232 | spin_lock(&ino->i_lock); | |
63d8a41b | 2233 | lo = pnfs_find_alloc_layout(ino, ctx, nfs_io_gfp_mask()); |
78746a38 FI |
2234 | if (!lo) |
2235 | goto out_unlock; | |
2236 | if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) | |
2237 | goto out_unlock; | |
2238 | if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) | |
2239 | goto out_unlock; | |
2240 | if (pnfs_layoutgets_blocked(lo)) | |
2241 | goto out_unlock; | |
2242 | if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags)) | |
2243 | goto out_unlock; | |
411ae722 | 2244 | nfs_layoutget_begin(lo); |
78746a38 FI |
2245 | spin_unlock(&ino->i_lock); |
2246 | _add_to_server_list(lo, NFS_SERVER(ino)); | |
2247 | return lo; | |
2248 | ||
2249 | out_unlock: | |
2250 | spin_unlock(&ino->i_lock); | |
2251 | pnfs_put_layout_hdr(lo); | |
2252 | return NULL; | |
2253 | } | |
2254 | ||
2409a976 FI |
2255 | static void _lgopen_prepare_attached(struct nfs4_opendata *data, |
2256 | struct nfs_open_context *ctx) | |
2257 | { | |
78746a38 FI |
2258 | struct inode *ino = data->dentry->d_inode; |
2259 | struct pnfs_layout_range rng = { | |
2260 | .iomode = (data->o_arg.fmode & FMODE_WRITE) ? | |
2261 | IOMODE_RW: IOMODE_READ, | |
2262 | .offset = 0, | |
2263 | .length = NFS4_MAX_UINT64, | |
2264 | }; | |
2265 | struct nfs4_layoutget *lgp; | |
2266 | struct pnfs_layout_hdr *lo; | |
2267 | ||
64294b08 TM |
2268 | /* Heuristic: don't send layoutget if we have cached data */ |
2269 | if (rng.iomode == IOMODE_READ && | |
2270 | (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0)) | |
2271 | return; | |
2272 | ||
78746a38 FI |
2273 | lo = _pnfs_grab_empty_layout(ino, ctx); |
2274 | if (!lo) | |
2275 | return; | |
63d8a41b TM |
2276 | lgp = pnfs_alloc_init_layoutget_args(ino, ctx, ¤t_stateid, &rng, |
2277 | nfs_io_gfp_mask()); | |
78746a38 FI |
2278 | if (!lgp) { |
2279 | pnfs_clear_first_layoutget(lo); | |
cb2856c5 | 2280 | nfs_layoutget_end(lo); |
78746a38 FI |
2281 | pnfs_put_layout_hdr(lo); |
2282 | return; | |
2283 | } | |
b4e89bcb | 2284 | lgp->lo = lo; |
78746a38 FI |
2285 | data->lgp = lgp; |
2286 | data->o_arg.lg_args = &lgp->args; | |
2287 | data->o_res.lg_res = &lgp->res; | |
2409a976 FI |
2288 | } |
2289 | ||
2290 | static void _lgopen_prepare_floating(struct nfs4_opendata *data, | |
2291 | struct nfs_open_context *ctx) | |
2292 | { | |
b4e89bcb | 2293 | struct inode *ino = data->dentry->d_inode; |
2409a976 FI |
2294 | struct pnfs_layout_range rng = { |
2295 | .iomode = (data->o_arg.fmode & FMODE_WRITE) ? | |
2296 | IOMODE_RW: IOMODE_READ, | |
2297 | .offset = 0, | |
2298 | .length = NFS4_MAX_UINT64, | |
2299 | }; | |
2300 | struct nfs4_layoutget *lgp; | |
2301 | ||
63d8a41b TM |
2302 | lgp = pnfs_alloc_init_layoutget_args(ino, ctx, ¤t_stateid, &rng, |
2303 | nfs_io_gfp_mask()); | |
2409a976 FI |
2304 | if (!lgp) |
2305 | return; | |
2306 | data->lgp = lgp; | |
2307 | data->o_arg.lg_args = &lgp->args; | |
2308 | data->o_res.lg_res = &lgp->res; | |
2309 | } | |
2310 | ||
2311 | void pnfs_lgopen_prepare(struct nfs4_opendata *data, | |
2312 | struct nfs_open_context *ctx) | |
2313 | { | |
2314 | struct nfs_server *server = NFS_SERVER(data->dir->d_inode); | |
2315 | ||
2316 | if (!(pnfs_enabled_sb(server) && | |
2317 | server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN)) | |
2318 | return; | |
2319 | /* Could check on max_ops, but currently hardcoded high enough */ | |
6e01260c FI |
2320 | if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN)) |
2321 | return; | |
b4e89bcb TM |
2322 | if (data->lgp) |
2323 | return; | |
2409a976 FI |
2324 | if (data->state) |
2325 | _lgopen_prepare_attached(data, ctx); | |
2326 | else | |
2327 | _lgopen_prepare_floating(data, ctx); | |
2328 | } | |
2329 | ||
2330 | void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp, | |
2331 | struct nfs_open_context *ctx) | |
2332 | { | |
2333 | struct pnfs_layout_hdr *lo; | |
2334 | struct pnfs_layout_segment *lseg; | |
c49b5209 | 2335 | struct nfs_server *srv = NFS_SERVER(ino); |
2409a976 FI |
2336 | u32 iomode; |
2337 | ||
6e01260c | 2338 | if (!lgp) |
2409a976 | 2339 | return; |
6e01260c FI |
2340 | dprintk("%s: entered with status %i\n", __func__, lgp->res.status); |
2341 | if (lgp->res.status) { | |
2342 | switch (lgp->res.status) { | |
6e01260c | 2343 | default: |
8dc96566 TM |
2344 | break; |
2345 | /* | |
2346 | * Halt lgopen attempts if the server doesn't recognise | |
2347 | * the "current stateid" value, the layout type, or the | |
2348 | * layoutget operation as being valid. | |
2349 | * Also if it complains about too many ops in the compound | |
2350 | * or of the request/reply being too big. | |
2351 | */ | |
2352 | case -NFS4ERR_BAD_STATEID: | |
2353 | case -NFS4ERR_NOTSUPP: | |
2354 | case -NFS4ERR_REP_TOO_BIG: | |
2355 | case -NFS4ERR_REP_TOO_BIG_TO_CACHE: | |
2356 | case -NFS4ERR_REQ_TOO_BIG: | |
2357 | case -NFS4ERR_TOO_MANY_OPS: | |
2358 | case -NFS4ERR_UNKNOWN_LAYOUTTYPE: | |
c49b5209 | 2359 | srv->caps &= ~NFS_CAP_LGOPEN; |
6e01260c FI |
2360 | } |
2361 | return; | |
2362 | } | |
b4e89bcb | 2363 | if (!lgp->lo) { |
78746a38 FI |
2364 | lo = _pnfs_grab_empty_layout(ino, ctx); |
2365 | if (!lo) | |
2366 | return; | |
b4e89bcb | 2367 | lgp->lo = lo; |
2409a976 | 2368 | } else |
b4e89bcb | 2369 | lo = lgp->lo; |
2409a976 FI |
2370 | |
2371 | lseg = pnfs_layout_process(lgp); | |
32f1c28f | 2372 | if (!IS_ERR(lseg)) { |
2409a976 FI |
2373 | iomode = lgp->args.range.iomode; |
2374 | pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); | |
2375 | pnfs_put_lseg(lseg); | |
2376 | } | |
30ae2412 FI |
2377 | } |
2378 | ||
2379 | void nfs4_lgopen_release(struct nfs4_layoutget *lgp) | |
2380 | { | |
2381 | if (lgp != NULL) { | |
b4e89bcb TM |
2382 | if (lgp->lo) { |
2383 | pnfs_clear_first_layoutget(lgp->lo); | |
2384 | nfs_layoutget_end(lgp->lo); | |
30ae2412 FI |
2385 | } |
2386 | pnfs_layoutget_free(lgp); | |
2387 | } | |
2409a976 FI |
2388 | } |
2389 | ||
a0b0a6e3 | 2390 | struct pnfs_layout_segment * |
b1f69b75 AA |
2391 | pnfs_layout_process(struct nfs4_layoutget *lgp) |
2392 | { | |
b4e89bcb | 2393 | struct pnfs_layout_hdr *lo = lgp->lo; |
b1f69b75 AA |
2394 | struct nfs4_layoutget_res *res = &lgp->res; |
2395 | struct pnfs_layout_segment *lseg; | |
b7edfaa1 | 2396 | struct inode *ino = lo->plh_inode; |
78096cca | 2397 | LIST_HEAD(free_me); |
540d9864 TM |
2398 | |
2399 | if (!pnfs_sanity_check_layout_range(&res->range)) | |
1b3c6d07 | 2400 | return ERR_PTR(-EINVAL); |
b1f69b75 AA |
2401 | |
2402 | /* Inject layout blob into I/O device driver */ | |
a75b9df9 | 2403 | lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); |
1b3c6d07 | 2404 | if (IS_ERR_OR_NULL(lseg)) { |
b1f69b75 | 2405 | if (!lseg) |
1b3c6d07 JL |
2406 | lseg = ERR_PTR(-ENOMEM); |
2407 | ||
2408 | dprintk("%s: Could not allocate layout: error %ld\n", | |
2409 | __func__, PTR_ERR(lseg)); | |
2410 | return lseg; | |
b1f69b75 AA |
2411 | } |
2412 | ||
119cef97 | 2413 | pnfs_init_lseg(lo, lseg, &res->range, &res->stateid); |
1013df61 | 2414 | |
b1f69b75 | 2415 | spin_lock(&ino->i_lock); |
e1c06f80 | 2416 | if (pnfs_layoutgets_blocked(lo)) { |
43f1b3da | 2417 | dprintk("%s forget reply due to state\n", __func__); |
1b3c6d07 | 2418 | goto out_forget; |
43f1b3da | 2419 | } |
038d6493 | 2420 | |
880265c7 TM |
2421 | if (test_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags) && |
2422 | !pnfs_is_first_layoutget(lo)) | |
0b77f97a TM |
2423 | goto out_forget; |
2424 | ||
d29b468d | 2425 | if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) { |
362f7474 CH |
2426 | /* existing state ID, make sure the sequence number matches. */ |
2427 | if (pnfs_layout_stateid_blocked(lo, &res->stateid)) { | |
0b77f97a | 2428 | if (!pnfs_layout_is_valid(lo)) |
d29b468d | 2429 | lo->plh_barrier = 0; |
362f7474 | 2430 | dprintk("%s forget reply due to sequence\n", __func__); |
1b3c6d07 | 2431 | goto out_forget; |
362f7474 | 2432 | } |
59b56394 | 2433 | pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, false); |
d29b468d | 2434 | } else if (pnfs_layout_is_valid(lo)) { |
362f7474 CH |
2435 | /* |
2436 | * We got an entirely new state ID. Mark all segments for the | |
9888d837 | 2437 | * inode invalid, and retry the layoutget |
362f7474 | 2438 | */ |
08bd8dbe TM |
2439 | struct pnfs_layout_range range = { |
2440 | .iomode = IOMODE_ANY, | |
2441 | .length = NFS4_MAX_UINT64, | |
2442 | }; | |
fb700ef0 | 2443 | pnfs_mark_matching_lsegs_return(lo, &free_me, &range, 0); |
9888d837 | 2444 | goto out_forget; |
d29b468d TM |
2445 | } else { |
2446 | /* We have a completely new layout */ | |
d29b468d | 2447 | pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, true); |
362f7474 | 2448 | } |
038d6493 | 2449 | |
9369a431 | 2450 | pnfs_get_lseg(lseg); |
03772d2f | 2451 | pnfs_layout_insert_lseg(lo, lseg, &free_me); |
8e0acf90 | 2452 | |
b1f69b75 | 2453 | |
3976143b | 2454 | if (res->return_on_close) |
f7e8917a | 2455 | set_bit(NFS_LSEG_ROC, &lseg->pls_flags); |
f7e8917a | 2456 | |
b1f69b75 | 2457 | spin_unlock(&ino->i_lock); |
78096cca | 2458 | pnfs_free_lseg_list(&free_me); |
a0b0a6e3 | 2459 | return lseg; |
43f1b3da | 2460 | |
1b3c6d07 | 2461 | out_forget: |
43f1b3da FI |
2462 | spin_unlock(&ino->i_lock); |
2463 | lseg->pls_layout = lo; | |
2464 | NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); | |
1b3c6d07 | 2465 | return ERR_PTR(-EAGAIN); |
b1f69b75 AA |
2466 | } |
2467 | ||
2f215968 TM |
2468 | /** |
2469 | * pnfs_mark_matching_lsegs_return - Free or return matching layout segments | |
2470 | * @lo: pointer to layout header | |
2471 | * @tmp_list: list header to be used with pnfs_free_lseg_list() | |
2472 | * @return_range: describe layout segment ranges to be returned | |
e0b7d420 | 2473 | * @seq: stateid seqid to match |
2f215968 TM |
2474 | * |
2475 | * This function is mainly intended for use by layoutrecall. It attempts | |
2476 | * to free the layout segment immediately, or else to mark it for return | |
2477 | * as soon as its reference count drops to zero. | |
e0b7d420 TM |
2478 | * |
2479 | * Returns | |
2480 | * - 0: a layoutreturn needs to be scheduled. | |
2481 | * - EBUSY: there are layout segment that are still in use. | |
2482 | * - ENOENT: there are no layout segments that need to be returned. | |
2f215968 | 2483 | */ |
10335556 | 2484 | int |
016256df PT |
2485 | pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo, |
2486 | struct list_head *tmp_list, | |
6d597e17 JL |
2487 | const struct pnfs_layout_range *return_range, |
2488 | u32 seq) | |
016256df PT |
2489 | { |
2490 | struct pnfs_layout_segment *lseg, *next; | |
b739a5bd | 2491 | struct nfs_server *server = NFS_SERVER(lo->plh_inode); |
10335556 | 2492 | int remaining = 0; |
016256df PT |
2493 | |
2494 | dprintk("%s:Begin lo %p\n", __func__, lo); | |
2495 | ||
fc7ff367 | 2496 | assert_spin_locked(&lo->plh_inode->i_lock); |
016256df | 2497 | |
39fd0186 TM |
2498 | if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) |
2499 | tmp_list = &lo->plh_return_segs; | |
2500 | ||
016256df | 2501 | list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) |
e036f464 | 2502 | if (pnfs_match_lseg_recall(lseg, return_range, seq)) { |
016256df PT |
2503 | dprintk("%s: marking lseg %p iomode %d " |
2504 | "offset %llu length %llu\n", __func__, | |
2505 | lseg, lseg->pls_range.iomode, | |
2506 | lseg->pls_range.offset, | |
2507 | lseg->pls_range.length); | |
39fd0186 TM |
2508 | if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)) |
2509 | tmp_list = &lo->plh_return_segs; | |
ff041727 | 2510 | if (mark_lseg_invalid(lseg, tmp_list)) |
2f215968 TM |
2511 | continue; |
2512 | remaining++; | |
016256df | 2513 | set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags); |
b739a5bd | 2514 | pnfs_lseg_cancel_io(server, lseg); |
016256df | 2515 | } |
6d597e17 | 2516 | |
e0b7d420 | 2517 | if (remaining) { |
6d597e17 | 2518 | pnfs_set_plh_return_info(lo, return_range->iomode, seq); |
e0b7d420 TM |
2519 | return -EBUSY; |
2520 | } | |
6d597e17 | 2521 | |
e0b7d420 TM |
2522 | if (!list_empty(&lo->plh_return_segs)) { |
2523 | pnfs_set_plh_return_info(lo, return_range->iomode, seq); | |
2524 | return 0; | |
2525 | } | |
2526 | ||
2527 | return -ENOENT; | |
016256df PT |
2528 | } |
2529 | ||
b5fdf841 TM |
2530 | static void |
2531 | pnfs_mark_layout_for_return(struct inode *inode, | |
2532 | const struct pnfs_layout_range *range) | |
016256df | 2533 | { |
b5fdf841 | 2534 | struct pnfs_layout_hdr *lo; |
10335556 | 2535 | bool return_now = false; |
016256df PT |
2536 | |
2537 | spin_lock(&inode->i_lock); | |
b5fdf841 | 2538 | lo = NFS_I(inode)->layout; |
bdebfccd TM |
2539 | if (!pnfs_layout_is_valid(lo)) { |
2540 | spin_unlock(&inode->i_lock); | |
2541 | return; | |
2542 | } | |
b5fdf841 | 2543 | pnfs_set_plh_return_info(lo, range->iomode, 0); |
016256df PT |
2544 | /* |
2545 | * mark all matching lsegs so that we are sure to have no live | |
2546 | * segments at hand when sending layoutreturn. See pnfs_put_lseg() | |
2547 | * for how it works. | |
2548 | */ | |
b5fdf841 | 2549 | if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, range, 0) != -EBUSY) { |
44ea8dfc | 2550 | const struct cred *cred; |
10335556 | 2551 | nfs4_stateid stateid; |
e5fd1904 | 2552 | enum pnfs_iomode iomode; |
10335556 | 2553 | |
44ea8dfc | 2554 | return_now = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode); |
10335556 TM |
2555 | spin_unlock(&inode->i_lock); |
2556 | if (return_now) | |
44ea8dfc | 2557 | pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false); |
10335556 TM |
2558 | } else { |
2559 | spin_unlock(&inode->i_lock); | |
2560 | nfs_commit_inode(inode, 0); | |
2561 | } | |
016256df | 2562 | } |
b5fdf841 TM |
2563 | |
2564 | void pnfs_error_mark_layout_for_return(struct inode *inode, | |
2565 | struct pnfs_layout_segment *lseg) | |
2566 | { | |
2567 | struct pnfs_layout_range range = { | |
2568 | .iomode = lseg->pls_range.iomode, | |
2569 | .offset = 0, | |
2570 | .length = NFS4_MAX_UINT64, | |
2571 | }; | |
2572 | ||
2573 | pnfs_mark_layout_for_return(inode, &range); | |
2574 | } | |
016256df PT |
2575 | EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return); |
2576 | ||
b5fdf841 TM |
2577 | static bool |
2578 | pnfs_layout_can_be_returned(struct pnfs_layout_hdr *lo) | |
2579 | { | |
2580 | return pnfs_layout_is_valid(lo) && | |
2581 | !test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) && | |
2582 | !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags); | |
2583 | } | |
2584 | ||
2585 | static struct pnfs_layout_segment * | |
2586 | pnfs_find_first_lseg(struct pnfs_layout_hdr *lo, | |
2587 | const struct pnfs_layout_range *range, | |
2588 | enum pnfs_iomode iomode) | |
2589 | { | |
2590 | struct pnfs_layout_segment *lseg; | |
2591 | ||
2592 | list_for_each_entry(lseg, &lo->plh_segs, pls_list) { | |
2593 | if (!test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) | |
2594 | continue; | |
2595 | if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)) | |
2596 | continue; | |
2597 | if (lseg->pls_range.iomode != iomode && iomode != IOMODE_ANY) | |
2598 | continue; | |
2599 | if (pnfs_lseg_range_intersecting(&lseg->pls_range, range)) | |
2600 | return lseg; | |
2601 | } | |
2602 | return NULL; | |
2603 | } | |
2604 | ||
2605 | /* Find open file states whose mode matches that of the range */ | |
2606 | static bool | |
2607 | pnfs_should_return_unused_layout(struct pnfs_layout_hdr *lo, | |
2608 | const struct pnfs_layout_range *range) | |
2609 | { | |
2610 | struct list_head *head; | |
2611 | struct nfs_open_context *ctx; | |
2612 | fmode_t mode = 0; | |
2613 | ||
2614 | if (!pnfs_layout_can_be_returned(lo) || | |
2615 | !pnfs_find_first_lseg(lo, range, range->iomode)) | |
2616 | return false; | |
2617 | ||
2618 | head = &NFS_I(lo->plh_inode)->open_files; | |
2619 | list_for_each_entry_rcu(ctx, head, list) { | |
2620 | if (ctx->state) | |
2621 | mode |= ctx->state->state & (FMODE_READ|FMODE_WRITE); | |
2622 | } | |
2623 | ||
2624 | switch (range->iomode) { | |
2625 | default: | |
2626 | break; | |
2627 | case IOMODE_READ: | |
2628 | mode &= ~FMODE_WRITE; | |
2629 | break; | |
2630 | case IOMODE_RW: | |
2631 | if (pnfs_find_first_lseg(lo, range, IOMODE_READ)) | |
2632 | mode &= ~FMODE_READ; | |
2633 | } | |
2634 | return mode == 0; | |
2635 | } | |
2636 | ||
2637 | static int | |
2638 | pnfs_layout_return_unused_byserver(struct nfs_server *server, void *data) | |
2639 | { | |
2640 | const struct pnfs_layout_range *range = data; | |
2641 | struct pnfs_layout_hdr *lo; | |
2642 | struct inode *inode; | |
2643 | restart: | |
2644 | rcu_read_lock(); | |
2645 | list_for_each_entry_rcu(lo, &server->layouts, plh_layouts) { | |
2646 | if (!pnfs_layout_can_be_returned(lo) || | |
2647 | test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) | |
2648 | continue; | |
2649 | inode = lo->plh_inode; | |
2650 | spin_lock(&inode->i_lock); | |
2651 | if (!pnfs_should_return_unused_layout(lo, range)) { | |
2652 | spin_unlock(&inode->i_lock); | |
2653 | continue; | |
2654 | } | |
2655 | spin_unlock(&inode->i_lock); | |
2656 | inode = pnfs_grab_inode_layout_hdr(lo); | |
2657 | if (!inode) | |
2658 | continue; | |
2659 | rcu_read_unlock(); | |
2660 | pnfs_mark_layout_for_return(inode, range); | |
2661 | iput(inode); | |
2662 | cond_resched(); | |
2663 | goto restart; | |
2664 | } | |
2665 | rcu_read_unlock(); | |
2666 | return 0; | |
2667 | } | |
2668 | ||
2669 | void | |
2670 | pnfs_layout_return_unused_byclid(struct nfs_client *clp, | |
2671 | enum pnfs_iomode iomode) | |
2672 | { | |
2673 | struct pnfs_layout_range range = { | |
2674 | .iomode = iomode, | |
2675 | .offset = 0, | |
2676 | .length = NFS4_MAX_UINT64, | |
2677 | }; | |
2678 | ||
2679 | nfs_client_for_each_server(clp, pnfs_layout_return_unused_byserver, | |
2680 | &range); | |
2681 | } | |
2682 | ||
b3230e80 TM |
2683 | void |
2684 | pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio) | |
2685 | { | |
2686 | if (pgio->pg_lseg == NULL || | |
2687 | test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags)) | |
2688 | return; | |
2689 | pnfs_put_lseg(pgio->pg_lseg); | |
2690 | pgio->pg_lseg = NULL; | |
2691 | } | |
2692 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout); | |
2693 | ||
08cb5b0f BC |
2694 | /* |
2695 | * Check for any intersection between the request and the pgio->pg_lseg, | |
2696 | * and if none, put this pgio->pg_lseg away. | |
2697 | */ | |
e1e54ab7 | 2698 | void |
08cb5b0f BC |
2699 | pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) |
2700 | { | |
2701 | if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) { | |
2702 | pnfs_put_lseg(pgio->pg_lseg); | |
2703 | pgio->pg_lseg = NULL; | |
2704 | } | |
2705 | } | |
e1e54ab7 | 2706 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_range); |
08cb5b0f | 2707 | |
d8007d4d TM |
2708 | void |
2709 | pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
2710 | { | |
d1d97395 | 2711 | u64 rd_size; |
1fd937bd | 2712 | |
b3230e80 | 2713 | pnfs_generic_pg_check_layout(pgio); |
08cb5b0f | 2714 | pnfs_generic_pg_check_range(pgio, req); |
cb5d04bc PT |
2715 | if (pgio->pg_lseg == NULL) { |
2716 | if (pgio->pg_dreq == NULL) | |
2717 | rd_size = i_size_read(pgio->pg_inode) - req_offset(req); | |
2718 | else | |
2719 | rd_size = nfs_dreq_bytes_left(pgio->pg_dreq); | |
2720 | ||
63d8a41b TM |
2721 | pgio->pg_lseg = |
2722 | pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req), | |
2723 | req_offset(req), rd_size, | |
2724 | IOMODE_READ, false, | |
2725 | nfs_io_gfp_mask()); | |
d600ad1f PT |
2726 | if (IS_ERR(pgio->pg_lseg)) { |
2727 | pgio->pg_error = PTR_ERR(pgio->pg_lseg); | |
2728 | pgio->pg_lseg = NULL; | |
2729 | return; | |
2730 | } | |
cb5d04bc | 2731 | } |
e885de1a TM |
2732 | /* If no lseg, fall back to read through mds */ |
2733 | if (pgio->pg_lseg == NULL) | |
1f945357 | 2734 | nfs_pageio_reset_read_mds(pgio); |
e885de1a | 2735 | |
d8007d4d TM |
2736 | } |
2737 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); | |
2738 | ||
2739 | void | |
6296556f PT |
2740 | pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, |
2741 | struct nfs_page *req, u64 wb_size) | |
d8007d4d | 2742 | { |
b3230e80 | 2743 | pnfs_generic_pg_check_layout(pgio); |
08cb5b0f | 2744 | pnfs_generic_pg_check_range(pgio, req); |
d600ad1f | 2745 | if (pgio->pg_lseg == NULL) { |
63d8a41b TM |
2746 | pgio->pg_lseg = |
2747 | pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req), | |
2748 | req_offset(req), wb_size, IOMODE_RW, | |
2749 | false, nfs_io_gfp_mask()); | |
d600ad1f PT |
2750 | if (IS_ERR(pgio->pg_lseg)) { |
2751 | pgio->pg_error = PTR_ERR(pgio->pg_lseg); | |
2752 | pgio->pg_lseg = NULL; | |
2753 | return; | |
2754 | } | |
2755 | } | |
e885de1a TM |
2756 | /* If no lseg, fall back to write through mds */ |
2757 | if (pgio->pg_lseg == NULL) | |
1f945357 | 2758 | nfs_pageio_reset_write_mds(pgio); |
d8007d4d TM |
2759 | } |
2760 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); | |
2761 | ||
180bb5ec WAA |
2762 | void |
2763 | pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc) | |
2764 | { | |
2765 | if (desc->pg_lseg) { | |
2766 | pnfs_put_lseg(desc->pg_lseg); | |
2767 | desc->pg_lseg = NULL; | |
2768 | } | |
2769 | } | |
2770 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup); | |
2771 | ||
b4fdac1a WAA |
2772 | /* |
2773 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
2774 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
2775 | */ | |
2776 | size_t | |
a7d42ddb WAA |
2777 | pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, |
2778 | struct nfs_page *prev, struct nfs_page *req) | |
94ad1c80 | 2779 | { |
0f9c429e | 2780 | unsigned int size; |
c5e20cb7 | 2781 | u64 seg_end, req_start, seg_left; |
0f9c429e WAA |
2782 | |
2783 | size = nfs_generic_pg_test(pgio, prev, req); | |
0f9c429e WAA |
2784 | if (!size) |
2785 | return 0; | |
94ad1c80 | 2786 | |
19982ba8 | 2787 | /* |
c5e20cb7 WAA |
2788 | * 'size' contains the number of bytes left in the current page (up |
2789 | * to the original size asked for in @req->wb_bytes). | |
2790 | * | |
2791 | * Calculate how many bytes are left in the layout segment | |
2792 | * and if there are less bytes than 'size', return that instead. | |
19982ba8 TM |
2793 | * |
2794 | * Please also note that 'end_offset' is actually the offset of the | |
2795 | * first byte that lies outside the pnfs_layout_range. FIXME? | |
2796 | * | |
2797 | */ | |
19b54848 | 2798 | if (pgio->pg_lseg) { |
17822b20 | 2799 | seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset, |
c5e20cb7 WAA |
2800 | pgio->pg_lseg->pls_range.length); |
2801 | req_start = req_offset(req); | |
08cb5b0f | 2802 | |
c5e20cb7 | 2803 | /* start of request is past the last byte of this segment */ |
08cb5b0f | 2804 | if (req_start >= seg_end) |
19b54848 | 2805 | return 0; |
c5e20cb7 WAA |
2806 | |
2807 | /* adjust 'size' iff there are fewer bytes left in the | |
2808 | * segment than what nfs_generic_pg_test returned */ | |
2809 | seg_left = seg_end - req_start; | |
2810 | if (seg_left < size) | |
2811 | size = (unsigned int)seg_left; | |
19b54848 | 2812 | } |
0f9c429e | 2813 | |
19b54848 | 2814 | return size; |
94ad1c80 | 2815 | } |
89a58e32 | 2816 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); |
94ad1c80 | 2817 | |
53113ad3 | 2818 | int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr) |
e2fecb21 TM |
2819 | { |
2820 | struct nfs_pageio_descriptor pgio; | |
e2fecb21 TM |
2821 | |
2822 | /* Resend all requests through the MDS */ | |
53113ad3 WAA |
2823 | nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true, |
2824 | hdr->completion_ops); | |
2825 | return nfs_pageio_resend(&pgio, hdr); | |
e2fecb21 | 2826 | } |
e7dd79af | 2827 | EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); |
e2fecb21 | 2828 | |
d45f60c6 | 2829 | static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 2830 | { |
cd841605 FI |
2831 | |
2832 | dprintk("pnfs write error = %d\n", hdr->pnfs_error); | |
2833 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 2834 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 2835 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 2836 | } |
6c75dc0d | 2837 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 2838 | hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr); |
1acbbb4e FI |
2839 | } |
2840 | ||
d20581aa BH |
2841 | /* |
2842 | * Called by non rpc-based layout drivers | |
2843 | */ | |
d45f60c6 | 2844 | void pnfs_ld_write_done(struct nfs_pgio_header *hdr) |
44b83799 | 2845 | { |
f8417b48 | 2846 | if (likely(!hdr->pnfs_error)) { |
67af7611 TM |
2847 | pnfs_set_layoutcommit(hdr->inode, hdr->lseg, |
2848 | hdr->mds_offset + hdr->res.count); | |
d45f60c6 | 2849 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); |
f8417b48 KM |
2850 | } |
2851 | trace_nfs4_pnfs_write(hdr, hdr->pnfs_error); | |
2852 | if (unlikely(hdr->pnfs_error)) | |
d45f60c6 WAA |
2853 | pnfs_ld_handle_write_error(hdr); |
2854 | hdr->mds_ops->rpc_release(hdr); | |
44b83799 | 2855 | } |
d20581aa | 2856 | EXPORT_SYMBOL_GPL(pnfs_ld_write_done); |
44b83799 | 2857 | |
dce81290 TM |
2858 | static void |
2859 | pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 2860 | struct nfs_pgio_header *hdr) |
dce81290 | 2861 | { |
48d635f1 | 2862 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 2863 | |
6c75dc0d | 2864 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 2865 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
6c75dc0d | 2866 | nfs_pageio_reset_write_mds(desc); |
a7d42ddb | 2867 | mirror->pg_recoalesce = 1; |
6c75dc0d | 2868 | } |
ba4a76f7 | 2869 | hdr->completion_ops->completion(hdr); |
dce81290 TM |
2870 | } |
2871 | ||
2872 | static enum pnfs_try_status | |
d45f60c6 | 2873 | pnfs_try_to_write_data(struct nfs_pgio_header *hdr, |
dce81290 TM |
2874 | const struct rpc_call_ops *call_ops, |
2875 | struct pnfs_layout_segment *lseg, | |
2876 | int how) | |
0382b744 | 2877 | { |
cd841605 | 2878 | struct inode *inode = hdr->inode; |
0382b744 AA |
2879 | enum pnfs_try_status trypnfs; |
2880 | struct nfs_server *nfss = NFS_SERVER(inode); | |
2881 | ||
cd841605 | 2882 | hdr->mds_ops = call_ops; |
0382b744 AA |
2883 | |
2884 | dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, | |
d45f60c6 WAA |
2885 | inode->i_ino, hdr->args.count, hdr->args.offset, how); |
2886 | trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how); | |
6c75dc0d | 2887 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
0382b744 | 2888 | nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); |
0382b744 AA |
2889 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
2890 | return trypnfs; | |
2891 | } | |
2892 | ||
dce81290 | 2893 | static void |
7f714720 WAA |
2894 | pnfs_do_write(struct nfs_pageio_descriptor *desc, |
2895 | struct nfs_pgio_header *hdr, int how) | |
dce81290 | 2896 | { |
dce81290 TM |
2897 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
2898 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 2899 | enum pnfs_try_status trypnfs; |
dce81290 | 2900 | |
d45f60c6 | 2901 | trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how); |
37f8aa16 TM |
2902 | switch (trypnfs) { |
2903 | case PNFS_NOT_ATTEMPTED: | |
d45f60c6 | 2904 | pnfs_write_through_mds(desc, hdr); |
ffb81717 | 2905 | break; |
37f8aa16 TM |
2906 | case PNFS_ATTEMPTED: |
2907 | break; | |
2908 | case PNFS_TRY_AGAIN: | |
2909 | /* cleanup hdr and prepare to redo pnfs */ | |
2910 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { | |
2911 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); | |
2912 | list_splice_init(&hdr->pages, &mirror->pg_list); | |
2913 | mirror->pg_recoalesce = 1; | |
2914 | } | |
2915 | hdr->mds_ops->rpc_release(hdr); | |
2916 | } | |
dce81290 TM |
2917 | } |
2918 | ||
6c75dc0d FI |
2919 | static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) |
2920 | { | |
9369a431 | 2921 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 2922 | nfs_pgio_header_free(hdr); |
6c75dc0d FI |
2923 | } |
2924 | ||
dce81290 TM |
2925 | int |
2926 | pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
2927 | { | |
6c75dc0d | 2928 | struct nfs_pgio_header *hdr; |
dce81290 TM |
2929 | int ret; |
2930 | ||
1e7f3a48 WAA |
2931 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
2932 | if (!hdr) { | |
2bff2288 PT |
2933 | desc->pg_error = -ENOMEM; |
2934 | return desc->pg_error; | |
dce81290 | 2935 | } |
6c75dc0d | 2936 | nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); |
180bb5ec | 2937 | |
9369a431 | 2938 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 2939 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 2940 | if (!ret) |
7f714720 | 2941 | pnfs_do_write(desc, hdr, desc->pg_ioflags); |
a7d42ddb | 2942 | |
6c75dc0d | 2943 | return ret; |
dce81290 TM |
2944 | } |
2945 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); | |
2946 | ||
53113ad3 | 2947 | int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr) |
62e4a769 TM |
2948 | { |
2949 | struct nfs_pageio_descriptor pgio; | |
2950 | ||
1acbbb4e | 2951 | /* Resend all requests through the MDS */ |
53113ad3 WAA |
2952 | nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops); |
2953 | return nfs_pageio_resend(&pgio, hdr); | |
1acbbb4e | 2954 | } |
e7dd79af | 2955 | EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); |
1acbbb4e | 2956 | |
d45f60c6 | 2957 | static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr) |
1acbbb4e | 2958 | { |
cd841605 FI |
2959 | dprintk("pnfs read error = %d\n", hdr->pnfs_error); |
2960 | if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & | |
1acbbb4e | 2961 | PNFS_LAYOUTRET_ON_ERROR) { |
cd841605 | 2962 | pnfs_return_layout(hdr->inode); |
1acbbb4e | 2963 | } |
4db6e0b7 | 2964 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) |
53113ad3 | 2965 | hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr); |
62e4a769 TM |
2966 | } |
2967 | ||
d20581aa BH |
2968 | /* |
2969 | * Called by non rpc-based layout drivers | |
2970 | */ | |
d45f60c6 | 2971 | void pnfs_ld_read_done(struct nfs_pgio_header *hdr) |
d20581aa | 2972 | { |
bfc505de | 2973 | if (likely(!hdr->pnfs_error)) |
d45f60c6 | 2974 | hdr->mds_ops->rpc_call_done(&hdr->task, hdr); |
f8417b48 KM |
2975 | trace_nfs4_pnfs_read(hdr, hdr->pnfs_error); |
2976 | if (unlikely(hdr->pnfs_error)) | |
d45f60c6 WAA |
2977 | pnfs_ld_handle_read_error(hdr); |
2978 | hdr->mds_ops->rpc_release(hdr); | |
d20581aa BH |
2979 | } |
2980 | EXPORT_SYMBOL_GPL(pnfs_ld_read_done); | |
2981 | ||
493292dd TM |
2982 | static void |
2983 | pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, | |
d45f60c6 | 2984 | struct nfs_pgio_header *hdr) |
493292dd | 2985 | { |
48d635f1 | 2986 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); |
a7d42ddb | 2987 | |
4db6e0b7 | 2988 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
a7d42ddb | 2989 | list_splice_tail_init(&hdr->pages, &mirror->pg_list); |
4db6e0b7 | 2990 | nfs_pageio_reset_read_mds(desc); |
a7d42ddb | 2991 | mirror->pg_recoalesce = 1; |
4db6e0b7 | 2992 | } |
ba4a76f7 | 2993 | hdr->completion_ops->completion(hdr); |
493292dd TM |
2994 | } |
2995 | ||
64419a9b AA |
2996 | /* |
2997 | * Call the appropriate parallel I/O subsystem read function. | |
2998 | */ | |
493292dd | 2999 | static enum pnfs_try_status |
d45f60c6 | 3000 | pnfs_try_to_read_data(struct nfs_pgio_header *hdr, |
493292dd TM |
3001 | const struct rpc_call_ops *call_ops, |
3002 | struct pnfs_layout_segment *lseg) | |
64419a9b | 3003 | { |
cd841605 | 3004 | struct inode *inode = hdr->inode; |
64419a9b AA |
3005 | struct nfs_server *nfss = NFS_SERVER(inode); |
3006 | enum pnfs_try_status trypnfs; | |
3007 | ||
cd841605 | 3008 | hdr->mds_ops = call_ops; |
64419a9b AA |
3009 | |
3010 | dprintk("%s: Reading ino:%lu %u@%llu\n", | |
d45f60c6 | 3011 | __func__, inode->i_ino, hdr->args.count, hdr->args.offset); |
64419a9b | 3012 | |
d45f60c6 | 3013 | trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr); |
4db6e0b7 | 3014 | if (trypnfs != PNFS_NOT_ATTEMPTED) |
64419a9b | 3015 | nfs_inc_stats(inode, NFSIOS_PNFS_READ); |
64419a9b AA |
3016 | dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); |
3017 | return trypnfs; | |
3018 | } | |
863a3c6c | 3019 | |
ceb11e13 | 3020 | /* Resend all requests through pnfs. */ |
563c53e7 TM |
3021 | void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr, |
3022 | unsigned int mirror_idx) | |
ceb11e13 PT |
3023 | { |
3024 | struct nfs_pageio_descriptor pgio; | |
3025 | ||
1b1bc66b | 3026 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { |
54e4a0df TM |
3027 | /* Prevent deadlocks with layoutreturn! */ |
3028 | pnfs_put_lseg(hdr->lseg); | |
3029 | hdr->lseg = NULL; | |
3030 | ||
1b1bc66b WAA |
3031 | nfs_pageio_init_read(&pgio, hdr->inode, false, |
3032 | hdr->completion_ops); | |
563c53e7 | 3033 | pgio.pg_mirror_idx = mirror_idx; |
1b1bc66b WAA |
3034 | hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr); |
3035 | } | |
ceb11e13 PT |
3036 | } |
3037 | EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs); | |
3038 | ||
493292dd | 3039 | static void |
7f714720 | 3040 | pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr) |
493292dd | 3041 | { |
493292dd TM |
3042 | const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; |
3043 | struct pnfs_layout_segment *lseg = desc->pg_lseg; | |
7f714720 | 3044 | enum pnfs_try_status trypnfs; |
493292dd | 3045 | |
d45f60c6 | 3046 | trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg); |
6aeafd05 TM |
3047 | switch (trypnfs) { |
3048 | case PNFS_NOT_ATTEMPTED: | |
d45f60c6 | 3049 | pnfs_read_through_mds(desc, hdr); |
ffb81717 | 3050 | break; |
6aeafd05 TM |
3051 | case PNFS_ATTEMPTED: |
3052 | break; | |
3053 | case PNFS_TRY_AGAIN: | |
3054 | /* cleanup hdr and prepare to redo pnfs */ | |
3055 | if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { | |
3056 | struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc); | |
3057 | list_splice_init(&hdr->pages, &mirror->pg_list); | |
3058 | mirror->pg_recoalesce = 1; | |
3059 | } | |
3060 | hdr->mds_ops->rpc_release(hdr); | |
3061 | } | |
493292dd TM |
3062 | } |
3063 | ||
4db6e0b7 FI |
3064 | static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) |
3065 | { | |
9369a431 | 3066 | pnfs_put_lseg(hdr->lseg); |
1e7f3a48 | 3067 | nfs_pgio_header_free(hdr); |
4db6e0b7 FI |
3068 | } |
3069 | ||
493292dd TM |
3070 | int |
3071 | pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) | |
3072 | { | |
4db6e0b7 | 3073 | struct nfs_pgio_header *hdr; |
493292dd TM |
3074 | int ret; |
3075 | ||
1e7f3a48 WAA |
3076 | hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); |
3077 | if (!hdr) { | |
2bff2288 PT |
3078 | desc->pg_error = -ENOMEM; |
3079 | return desc->pg_error; | |
493292dd | 3080 | } |
4db6e0b7 | 3081 | nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); |
9369a431 | 3082 | hdr->lseg = pnfs_get_lseg(desc->pg_lseg); |
ef2c488c | 3083 | ret = nfs_generic_pgio(desc, hdr); |
180bb5ec | 3084 | if (!ret) |
7f714720 | 3085 | pnfs_do_read(desc, hdr); |
4db6e0b7 | 3086 | return ret; |
493292dd TM |
3087 | } |
3088 | EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); | |
3089 | ||
71244d9b TM |
3090 | static void pnfs_clear_layoutcommitting(struct inode *inode) |
3091 | { | |
3092 | unsigned long *bitlock = &NFS_I(inode)->flags; | |
3093 | ||
3094 | clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock); | |
4e857c58 | 3095 | smp_mb__after_atomic(); |
71244d9b TM |
3096 | wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING); |
3097 | } | |
3098 | ||
863a3c6c | 3099 | /* |
a9bae566 | 3100 | * There can be multiple RW segments. |
863a3c6c | 3101 | */ |
a9bae566 | 3102 | static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) |
863a3c6c | 3103 | { |
a9bae566 | 3104 | struct pnfs_layout_segment *lseg; |
863a3c6c | 3105 | |
a9bae566 PT |
3106 | list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { |
3107 | if (lseg->pls_range.iomode == IOMODE_RW && | |
a073dbff | 3108 | test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) |
a9bae566 PT |
3109 | list_add(&lseg->pls_lc_list, listp); |
3110 | } | |
863a3c6c AA |
3111 | } |
3112 | ||
a073dbff TM |
3113 | static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp) |
3114 | { | |
3115 | struct pnfs_layout_segment *lseg, *tmp; | |
a073dbff TM |
3116 | |
3117 | /* Matched by references in pnfs_set_layoutcommit */ | |
3118 | list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) { | |
3119 | list_del_init(&lseg->pls_lc_list); | |
3120 | pnfs_put_lseg(lseg); | |
3121 | } | |
3122 | ||
71244d9b | 3123 | pnfs_clear_layoutcommitting(inode); |
a073dbff TM |
3124 | } |
3125 | ||
1b0ae068 PT |
3126 | void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) |
3127 | { | |
b9e028fd | 3128 | pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode); |
1b0ae068 PT |
3129 | } |
3130 | EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); | |
3131 | ||
863a3c6c | 3132 | void |
67af7611 TM |
3133 | pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg, |
3134 | loff_t end_pos) | |
863a3c6c | 3135 | { |
cd841605 | 3136 | struct nfs_inode *nfsi = NFS_I(inode); |
79a48a1f | 3137 | bool mark_as_dirty = false; |
863a3c6c | 3138 | |
cd841605 | 3139 | spin_lock(&inode->i_lock); |
863a3c6c | 3140 | if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { |
29559b11 | 3141 | nfsi->layout->plh_lwb = end_pos; |
79a48a1f | 3142 | mark_as_dirty = true; |
863a3c6c | 3143 | dprintk("%s: Set layoutcommit for inode %lu ", |
cd841605 | 3144 | __func__, inode->i_ino); |
29559b11 TM |
3145 | } else if (end_pos > nfsi->layout->plh_lwb) |
3146 | nfsi->layout->plh_lwb = end_pos; | |
67af7611 | 3147 | if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) { |
a9bae566 | 3148 | /* references matched in nfs4_layoutcommit_release */ |
67af7611 | 3149 | pnfs_get_lseg(lseg); |
a9bae566 | 3150 | } |
cd841605 | 3151 | spin_unlock(&inode->i_lock); |
acff5880 | 3152 | dprintk("%s: lseg %p end_pos %llu\n", |
67af7611 | 3153 | __func__, lseg, nfsi->layout->plh_lwb); |
79a48a1f WAA |
3154 | |
3155 | /* if pnfs_layoutcommit_inode() runs between inode locks, the next one | |
3156 | * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ | |
3157 | if (mark_as_dirty) | |
cd841605 | 3158 | mark_inode_dirty_sync(inode); |
863a3c6c AA |
3159 | } |
3160 | EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); | |
3161 | ||
db29c089 AA |
3162 | void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) |
3163 | { | |
3164 | struct nfs_server *nfss = NFS_SERVER(data->args.inode); | |
3165 | ||
3166 | if (nfss->pnfs_curr_ld->cleanup_layoutcommit) | |
3167 | nfss->pnfs_curr_ld->cleanup_layoutcommit(data); | |
a073dbff | 3168 | pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list); |
db29c089 AA |
3169 | } |
3170 | ||
de4b15c7 AA |
3171 | /* |
3172 | * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and | |
3173 | * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough | |
3174 | * data to disk to allow the server to recover the data if it crashes. | |
3175 | * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag | |
3176 | * is off, and a COMMIT is sent to a data server, or | |
3177 | * if WRITEs to a data server return NFS_DATA_SYNC. | |
3178 | */ | |
863a3c6c | 3179 | int |
ef311537 | 3180 | pnfs_layoutcommit_inode(struct inode *inode, bool sync) |
863a3c6c | 3181 | { |
5f919c9f | 3182 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; |
863a3c6c AA |
3183 | struct nfs4_layoutcommit_data *data; |
3184 | struct nfs_inode *nfsi = NFS_I(inode); | |
863a3c6c | 3185 | loff_t end_pos; |
71244d9b | 3186 | int status; |
863a3c6c | 3187 | |
71244d9b | 3188 | if (!pnfs_layoutcommit_outstanding(inode)) |
de4b15c7 AA |
3189 | return 0; |
3190 | ||
71244d9b | 3191 | dprintk("--> %s inode %lu\n", __func__, inode->i_ino); |
92407e75 | 3192 | |
71244d9b | 3193 | status = -EAGAIN; |
92407e75 | 3194 | if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { |
71244d9b TM |
3195 | if (!sync) |
3196 | goto out; | |
74316201 | 3197 | status = wait_on_bit_lock_action(&nfsi->flags, |
71244d9b TM |
3198 | NFS_INO_LAYOUTCOMMITTING, |
3199 | nfs_wait_bit_killable, | |
f5d39b02 | 3200 | TASK_KILLABLE|TASK_FREEZABLE_UNSAFE); |
92407e75 | 3201 | if (status) |
71244d9b | 3202 | goto out; |
92407e75 PT |
3203 | } |
3204 | ||
71244d9b TM |
3205 | status = -ENOMEM; |
3206 | /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ | |
63d8a41b | 3207 | data = kzalloc(sizeof(*data), nfs_io_gfp_mask()); |
71244d9b TM |
3208 | if (!data) |
3209 | goto clear_layoutcommitting; | |
3210 | ||
3211 | status = 0; | |
de4b15c7 | 3212 | spin_lock(&inode->i_lock); |
71244d9b TM |
3213 | if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) |
3214 | goto out_unlock; | |
a9bae566 | 3215 | |
71244d9b | 3216 | INIT_LIST_HEAD(&data->lseg_list); |
a9bae566 | 3217 | pnfs_list_write_lseg(inode, &data->lseg_list); |
863a3c6c | 3218 | |
acff5880 | 3219 | end_pos = nfsi->layout->plh_lwb; |
863a3c6c | 3220 | |
f597c537 | 3221 | nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); |
97a728f5 | 3222 | data->cred = get_cred(nfsi->layout->plh_lc_cred); |
863a3c6c AA |
3223 | spin_unlock(&inode->i_lock); |
3224 | ||
3225 | data->args.inode = inode; | |
863a3c6c AA |
3226 | nfs_fattr_init(&data->fattr); |
3227 | data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; | |
3228 | data->res.fattr = &data->fattr; | |
2e18d4d8 TM |
3229 | if (end_pos != 0) |
3230 | data->args.lastbytewritten = end_pos - 1; | |
3231 | else | |
3232 | data->args.lastbytewritten = U64_MAX; | |
863a3c6c AA |
3233 | data->res.server = NFS_SERVER(inode); |
3234 | ||
5f919c9f CH |
3235 | if (ld->prepare_layoutcommit) { |
3236 | status = ld->prepare_layoutcommit(&data->args); | |
3237 | if (status) { | |
a52458b4 | 3238 | put_cred(data->cred); |
5f919c9f | 3239 | spin_lock(&inode->i_lock); |
29559b11 TM |
3240 | set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags); |
3241 | if (end_pos > nfsi->layout->plh_lwb) | |
5f919c9f | 3242 | nfsi->layout->plh_lwb = end_pos; |
3471648a | 3243 | goto out_unlock; |
5f919c9f CH |
3244 | } |
3245 | } | |
3246 | ||
3247 | ||
863a3c6c AA |
3248 | status = nfs4_proc_layoutcommit(data, sync); |
3249 | out: | |
92407e75 PT |
3250 | if (status) |
3251 | mark_inode_dirty_sync(inode); | |
863a3c6c AA |
3252 | dprintk("<-- %s status %d\n", __func__, status); |
3253 | return status; | |
71244d9b TM |
3254 | out_unlock: |
3255 | spin_unlock(&inode->i_lock); | |
92407e75 | 3256 | kfree(data); |
71244d9b TM |
3257 | clear_layoutcommitting: |
3258 | pnfs_clear_layoutcommitting(inode); | |
92407e75 | 3259 | goto out; |
863a3c6c | 3260 | } |
72cff449 | 3261 | EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode); |
82be417a | 3262 | |
5bb89b47 TM |
3263 | int |
3264 | pnfs_generic_sync(struct inode *inode, bool datasync) | |
3265 | { | |
3266 | return pnfs_layoutcommit_inode(inode, true); | |
3267 | } | |
3268 | EXPORT_SYMBOL_GPL(pnfs_generic_sync); | |
3269 | ||
82be417a AA |
3270 | struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) |
3271 | { | |
3272 | struct nfs4_threshold *thp; | |
3273 | ||
63d8a41b | 3274 | thp = kzalloc(sizeof(*thp), nfs_io_gfp_mask()); |
82be417a AA |
3275 | if (!thp) { |
3276 | dprintk("%s mdsthreshold allocation failed\n", __func__); | |
3277 | return NULL; | |
3278 | } | |
3279 | return thp; | |
3280 | } | |
8733408d | 3281 | |
865a7ecb | 3282 | #if IS_ENABLED(CONFIG_NFS_V4_2) |
8733408d | 3283 | int |
c8ad8894 | 3284 | pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags) |
8733408d PT |
3285 | { |
3286 | struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; | |
3287 | struct nfs_server *server = NFS_SERVER(inode); | |
1bfe3b25 | 3288 | struct nfs_inode *nfsi = NFS_I(inode); |
8733408d PT |
3289 | struct nfs42_layoutstat_data *data; |
3290 | struct pnfs_layout_hdr *hdr; | |
3291 | int status = 0; | |
3292 | ||
3293 | if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats) | |
3294 | goto out; | |
3295 | ||
6c5a0d89 TM |
3296 | if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS)) |
3297 | goto out; | |
3298 | ||
1bfe3b25 PT |
3299 | if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags)) |
3300 | goto out; | |
3301 | ||
8733408d PT |
3302 | spin_lock(&inode->i_lock); |
3303 | if (!NFS_I(inode)->layout) { | |
3304 | spin_unlock(&inode->i_lock); | |
f538d0ba | 3305 | goto out_clear_layoutstats; |
8733408d PT |
3306 | } |
3307 | hdr = NFS_I(inode)->layout; | |
3308 | pnfs_get_layout_hdr(hdr); | |
3309 | spin_unlock(&inode->i_lock); | |
3310 | ||
c8ad8894 | 3311 | data = kzalloc(sizeof(*data), gfp_flags); |
8733408d PT |
3312 | if (!data) { |
3313 | status = -ENOMEM; | |
3314 | goto out_put; | |
3315 | } | |
3316 | ||
3317 | data->args.fh = NFS_FH(inode); | |
3318 | data->args.inode = inode; | |
8733408d PT |
3319 | status = ld->prepare_layoutstats(&data->args); |
3320 | if (status) | |
3321 | goto out_free; | |
3322 | ||
3323 | status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data); | |
3324 | ||
3325 | out: | |
3326 | dprintk("%s returns %d\n", __func__, status); | |
3327 | return status; | |
3328 | ||
3329 | out_free: | |
3330 | kfree(data); | |
3331 | out_put: | |
3332 | pnfs_put_layout_hdr(hdr); | |
f538d0ba | 3333 | out_clear_layoutstats: |
1bfe3b25 PT |
3334 | smp_mb__before_atomic(); |
3335 | clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags); | |
3336 | smp_mb__after_atomic(); | |
8733408d PT |
3337 | goto out; |
3338 | } | |
3339 | EXPORT_SYMBOL_GPL(pnfs_report_layoutstat); | |
865a7ecb | 3340 | #endif |
bbf58bf3 TM |
3341 | |
3342 | unsigned int layoutstats_timer; | |
3343 | module_param(layoutstats_timer, uint, 0644); | |
3344 | EXPORT_SYMBOL_GPL(layoutstats_timer); |