Commit | Line | Data |
---|---|---|
d7e09d03 PT |
1 | /* |
2 | * GPL HEADER START | |
3 | * | |
4 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License version 2 only, | |
8 | * as published by the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, but | |
11 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
13 | * General Public License version 2 for more details (a copy is included | |
14 | * in the LICENSE file that accompanied this code). | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * version 2 along with this program; If not, see | |
6a5b99a4 | 18 | * http://www.gnu.org/licenses/gpl-2.0.html |
d7e09d03 | 19 | * |
d7e09d03 PT |
20 | * GPL HEADER END |
21 | */ | |
22 | /* | |
23 | * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. | |
24 | * Use is subject to license terms. | |
25 | * | |
1dc563a6 | 26 | * Copyright (c) 2011, 2015, Intel Corporation. |
d7e09d03 PT |
27 | */ |
28 | /* | |
29 | * This file is part of Lustre, http://www.lustre.org/ | |
30 | * Lustre is a trademark of Sun Microsystems, Inc. | |
31 | * | |
32 | * lustre/llite/file.c | |
33 | * | |
34 | * Author: Peter Braam <braam@clusterfs.com> | |
35 | * Author: Phil Schwan <phil@clusterfs.com> | |
36 | * Author: Andreas Dilger <adilger@clusterfs.com> | |
37 | */ | |
38 | ||
39 | #define DEBUG_SUBSYSTEM S_LLITE | |
67a235f5 GKH |
40 | #include "../include/lustre_dlm.h" |
41 | #include "../include/lustre_lite.h" | |
d7e09d03 PT |
42 | #include <linux/pagemap.h> |
43 | #include <linux/file.h> | |
bb41292b | 44 | #include <linux/mount.h> |
d7e09d03 | 45 | #include "llite_internal.h" |
67a235f5 | 46 | #include "../include/lustre/ll_fiemap.h" |
d7e09d03 | 47 | |
67a235f5 | 48 | #include "../include/cl_object.h" |
d7e09d03 | 49 | |
2d95f10e JH |
50 | static int |
51 | ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg); | |
52 | ||
53 | static int ll_lease_close(struct obd_client_handle *och, struct inode *inode, | |
54 | bool *lease_broken); | |
55 | ||
56 | static enum llioc_iter | |
57 | ll_iocontrol_call(struct inode *inode, struct file *file, | |
58 | unsigned int cmd, unsigned long arg, int *rcp); | |
59 | ||
60 | static struct ll_file_data *ll_file_data_get(void) | |
d7e09d03 PT |
61 | { |
62 | struct ll_file_data *fd; | |
63 | ||
21068c46 | 64 | fd = kmem_cache_zalloc(ll_file_data_slab, GFP_NOFS); |
6e16818b | 65 | if (!fd) |
73863d83 | 66 | return NULL; |
d7e09d03 PT |
67 | fd->fd_write_failed = false; |
68 | return fd; | |
69 | } | |
70 | ||
71 | static void ll_file_data_put(struct ll_file_data *fd) | |
72 | { | |
6e16818b | 73 | if (fd) |
50d30362 | 74 | kmem_cache_free(ll_file_data_slab, fd); |
d7e09d03 PT |
75 | } |
76 | ||
77 | void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data, | |
78 | struct lustre_handle *fh) | |
79 | { | |
80 | op_data->op_fid1 = ll_i2info(inode)->lli_fid; | |
81 | op_data->op_attr.ia_mode = inode->i_mode; | |
82 | op_data->op_attr.ia_atime = inode->i_atime; | |
83 | op_data->op_attr.ia_mtime = inode->i_mtime; | |
84 | op_data->op_attr.ia_ctime = inode->i_ctime; | |
85 | op_data->op_attr.ia_size = i_size_read(inode); | |
86 | op_data->op_attr_blocks = inode->i_blocks; | |
bb41292b | 87 | op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags); |
d7e09d03 PT |
88 | op_data->op_ioepoch = ll_i2info(inode)->lli_ioepoch; |
89 | if (fh) | |
90 | op_data->op_handle = *fh; | |
d7e09d03 | 91 | |
1f6eaf83 | 92 | if (ll_i2info(inode)->lli_flags & LLIF_DATA_MODIFIED) |
d7e09d03 PT |
93 | op_data->op_bias |= MDS_DATA_MODIFIED; |
94 | } | |
95 | ||
96 | /** | |
97 | * Closes the IO epoch and packs all the attributes into @op_data for | |
98 | * the CLOSE rpc. | |
99 | */ | |
100 | static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data, | |
101 | struct obd_client_handle *och) | |
102 | { | |
f57d9a72 EL |
103 | op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET | |
104 | ATTR_MTIME | ATTR_MTIME_SET | | |
105 | ATTR_CTIME | ATTR_CTIME_SET; | |
d7e09d03 PT |
106 | |
107 | if (!(och->och_flags & FMODE_WRITE)) | |
108 | goto out; | |
109 | ||
110 | if (!exp_connect_som(ll_i2mdexp(inode)) || !S_ISREG(inode->i_mode)) | |
111 | op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS; | |
112 | else | |
113 | ll_ioepoch_close(inode, op_data, &och, 0); | |
114 | ||
115 | out: | |
116 | ll_pack_inode2opdata(inode, op_data, &och->och_fh); | |
117 | ll_prep_md_op_data(op_data, inode, NULL, NULL, | |
118 | 0, 0, LUSTRE_OPC_ANY, NULL); | |
d7e09d03 PT |
119 | } |
120 | ||
121 | static int ll_close_inode_openhandle(struct obd_export *md_exp, | |
122 | struct inode *inode, | |
48d23e61 JX |
123 | struct obd_client_handle *och, |
124 | const __u64 *data_version) | |
d7e09d03 PT |
125 | { |
126 | struct obd_export *exp = ll_i2mdexp(inode); | |
127 | struct md_op_data *op_data; | |
128 | struct ptlrpc_request *req = NULL; | |
129 | struct obd_device *obd = class_exp2obd(exp); | |
130 | int epoch_close = 1; | |
131 | int rc; | |
d7e09d03 | 132 | |
6e16818b | 133 | if (!obd) { |
d7e09d03 PT |
134 | /* |
135 | * XXX: in case of LMV, is this correct to access | |
136 | * ->exp_handle? | |
137 | */ | |
55f5a824 | 138 | CERROR("Invalid MDC connection handle %#llx\n", |
d7e09d03 | 139 | ll_i2mdexp(inode)->exp_handle.h_cookie); |
34e1f2bb JL |
140 | rc = 0; |
141 | goto out; | |
d7e09d03 PT |
142 | } |
143 | ||
496a51bd JL |
144 | op_data = kzalloc(sizeof(*op_data), GFP_NOFS); |
145 | if (!op_data) { | |
34e1f2bb JL |
146 | /* XXX We leak openhandle and request here. */ |
147 | rc = -ENOMEM; | |
148 | goto out; | |
149 | } | |
d7e09d03 PT |
150 | |
151 | ll_prepare_close(inode, op_data, och); | |
6e16818b | 152 | if (data_version) { |
48d23e61 JX |
153 | /* Pass in data_version implies release. */ |
154 | op_data->op_bias |= MDS_HSM_RELEASE; | |
155 | op_data->op_data_version = *data_version; | |
156 | op_data->op_lease_handle = och->och_lease_handle; | |
157 | op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS; | |
158 | } | |
b6ee3824 | 159 | epoch_close = op_data->op_flags & MF_EPOCH_CLOSE; |
d7e09d03 PT |
160 | rc = md_close(md_exp, op_data, och->och_mod, &req); |
161 | if (rc == -EAGAIN) { | |
162 | /* This close must have the epoch closed. */ | |
163 | LASSERT(epoch_close); | |
164 | /* MDS has instructed us to obtain Size-on-MDS attribute from | |
c0894c6c OD |
165 | * OSTs and send setattr to back to MDS. |
166 | */ | |
d7e09d03 PT |
167 | rc = ll_som_update(inode, op_data); |
168 | if (rc) { | |
97a075cd JN |
169 | CERROR("%s: inode "DFID" mdc Size-on-MDS update failed: rc = %d\n", |
170 | ll_i2mdexp(inode)->exp_obd->obd_name, | |
171 | PFID(ll_inode2fid(inode)), rc); | |
d7e09d03 PT |
172 | rc = 0; |
173 | } | |
174 | } else if (rc) { | |
97a075cd JN |
175 | CERROR("%s: inode "DFID" mdc close failed: rc = %d\n", |
176 | ll_i2mdexp(inode)->exp_obd->obd_name, | |
177 | PFID(ll_inode2fid(inode)), rc); | |
d7e09d03 PT |
178 | } |
179 | ||
180 | /* DATA_MODIFIED flag was successfully sent on close, cancel data | |
c0894c6c OD |
181 | * modification flag. |
182 | */ | |
d7e09d03 PT |
183 | if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) { |
184 | struct ll_inode_info *lli = ll_i2info(inode); | |
185 | ||
186 | spin_lock(&lli->lli_lock); | |
187 | lli->lli_flags &= ~LLIF_DATA_MODIFIED; | |
188 | spin_unlock(&lli->lli_lock); | |
189 | } | |
190 | ||
d7e09d03 PT |
191 | if (rc == 0) { |
192 | rc = ll_objects_destroy(req, inode); | |
193 | if (rc) | |
194 | CERROR("inode %lu ll_objects destroy: rc = %d\n", | |
195 | inode->i_ino, rc); | |
196 | } | |
48d23e61 JX |
197 | if (rc == 0 && op_data->op_bias & MDS_HSM_RELEASE) { |
198 | struct mdt_body *body; | |
cea812cd | 199 | |
48d23e61 JX |
200 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); |
201 | if (!(body->valid & OBD_MD_FLRELEASED)) | |
202 | rc = -EBUSY; | |
203 | } | |
204 | ||
205 | ll_finish_md_op_data(op_data); | |
d7e09d03 | 206 | |
d7e09d03 | 207 | out: |
d7e09d03 PT |
208 | if (exp_connect_som(exp) && !epoch_close && |
209 | S_ISREG(inode->i_mode) && (och->och_flags & FMODE_WRITE)) { | |
210 | ll_queue_done_writing(inode, LLIF_DONE_WRITING); | |
211 | } else { | |
212 | md_clear_open_replay_data(md_exp, och); | |
213 | /* Free @och if it is not waiting for DONE_WRITING. */ | |
214 | och->och_fh.cookie = DEAD_HANDLE_MAGIC; | |
97903a26 | 215 | kfree(och); |
d7e09d03 PT |
216 | } |
217 | if (req) /* This is close request */ | |
218 | ptlrpc_req_finished(req); | |
219 | return rc; | |
220 | } | |
221 | ||
45b2a010 | 222 | int ll_md_real_close(struct inode *inode, fmode_t fmode) |
d7e09d03 PT |
223 | { |
224 | struct ll_inode_info *lli = ll_i2info(inode); | |
225 | struct obd_client_handle **och_p; | |
226 | struct obd_client_handle *och; | |
227 | __u64 *och_usecount; | |
228 | int rc = 0; | |
d7e09d03 | 229 | |
45b2a010 | 230 | if (fmode & FMODE_WRITE) { |
d7e09d03 PT |
231 | och_p = &lli->lli_mds_write_och; |
232 | och_usecount = &lli->lli_open_fd_write_count; | |
45b2a010 | 233 | } else if (fmode & FMODE_EXEC) { |
d7e09d03 PT |
234 | och_p = &lli->lli_mds_exec_och; |
235 | och_usecount = &lli->lli_open_fd_exec_count; | |
236 | } else { | |
45b2a010 | 237 | LASSERT(fmode & FMODE_READ); |
d7e09d03 PT |
238 | och_p = &lli->lli_mds_read_och; |
239 | och_usecount = &lli->lli_open_fd_read_count; | |
240 | } | |
241 | ||
242 | mutex_lock(&lli->lli_och_mutex); | |
45b2a010 JH |
243 | if (*och_usecount > 0) { |
244 | /* There are still users of this handle, so skip | |
c0894c6c OD |
245 | * freeing it. |
246 | */ | |
d7e09d03 | 247 | mutex_unlock(&lli->lli_och_mutex); |
0a3bdb00 | 248 | return 0; |
d7e09d03 | 249 | } |
45b2a010 | 250 | |
57303e76 | 251 | och = *och_p; |
d7e09d03 PT |
252 | *och_p = NULL; |
253 | mutex_unlock(&lli->lli_och_mutex); | |
254 | ||
6e16818b | 255 | if (och) { |
45b2a010 | 256 | /* There might be a race and this handle may already |
c0894c6c OD |
257 | * be closed. |
258 | */ | |
d7e09d03 | 259 | rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, |
48d23e61 | 260 | inode, och, NULL); |
d7e09d03 PT |
261 | } |
262 | ||
0a3bdb00 | 263 | return rc; |
d7e09d03 PT |
264 | } |
265 | ||
2d95f10e JH |
266 | static int ll_md_close(struct obd_export *md_exp, struct inode *inode, |
267 | struct file *file) | |
d7e09d03 PT |
268 | { |
269 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
270 | struct ll_inode_info *lli = ll_i2info(inode); | |
74d01958 AV |
271 | int lockmode; |
272 | __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK; | |
273 | struct lustre_handle lockh; | |
8369cfff | 274 | ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_OPEN} }; |
d7e09d03 | 275 | int rc = 0; |
d7e09d03 PT |
276 | |
277 | /* clear group lock, if present */ | |
278 | if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED)) | |
98eae5e7 | 279 | ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid); |
d7e09d03 | 280 | |
6e16818b | 281 | if (fd->fd_lease_och) { |
d3a8a4e2 JX |
282 | bool lease_broken; |
283 | ||
284 | /* Usually the lease is not released when the | |
c0894c6c OD |
285 | * application crashed, we need to release here. |
286 | */ | |
d3a8a4e2 | 287 | rc = ll_lease_close(fd->fd_lease_och, inode, &lease_broken); |
e15ba45d OD |
288 | CDEBUG(rc ? D_ERROR : D_INODE, |
289 | "Clean up lease " DFID " %d/%d\n", | |
290 | PFID(&lli->lli_fid), rc, lease_broken); | |
d3a8a4e2 JX |
291 | |
292 | fd->fd_lease_och = NULL; | |
293 | } | |
294 | ||
6e16818b | 295 | if (fd->fd_och) { |
48d23e61 | 296 | rc = ll_close_inode_openhandle(md_exp, inode, fd->fd_och, NULL); |
d3a8a4e2 | 297 | fd->fd_och = NULL; |
34e1f2bb | 298 | goto out; |
d3a8a4e2 JX |
299 | } |
300 | ||
d7e09d03 | 301 | /* Let's see if we have good enough OPEN lock on the file and if |
c0894c6c OD |
302 | * we can skip talking to MDS |
303 | */ | |
d7e09d03 | 304 | |
74d01958 AV |
305 | mutex_lock(&lli->lli_och_mutex); |
306 | if (fd->fd_omode & FMODE_WRITE) { | |
307 | lockmode = LCK_CW; | |
308 | LASSERT(lli->lli_open_fd_write_count); | |
309 | lli->lli_open_fd_write_count--; | |
310 | } else if (fd->fd_omode & FMODE_EXEC) { | |
311 | lockmode = LCK_PR; | |
312 | LASSERT(lli->lli_open_fd_exec_count); | |
313 | lli->lli_open_fd_exec_count--; | |
d7e09d03 | 314 | } else { |
74d01958 AV |
315 | lockmode = LCK_CR; |
316 | LASSERT(lli->lli_open_fd_read_count); | |
317 | lli->lli_open_fd_read_count--; | |
d7e09d03 | 318 | } |
74d01958 AV |
319 | mutex_unlock(&lli->lli_och_mutex); |
320 | ||
321 | if (!md_lock_match(md_exp, flags, ll_inode2fid(inode), | |
322 | LDLM_IBITS, &policy, lockmode, &lockh)) | |
323 | rc = ll_md_real_close(inode, fd->fd_omode); | |
d7e09d03 | 324 | |
d3a8a4e2 | 325 | out: |
d7e09d03 PT |
326 | LUSTRE_FPRIVATE(file) = NULL; |
327 | ll_file_data_put(fd); | |
d7e09d03 | 328 | |
0a3bdb00 | 329 | return rc; |
d7e09d03 PT |
330 | } |
331 | ||
332 | /* While this returns an error code, fput() the caller does not, so we need | |
333 | * to make every effort to clean up all of our state here. Also, applications | |
334 | * rarely check close errors and even if an error is returned they will not | |
335 | * re-try the close call. | |
336 | */ | |
337 | int ll_file_release(struct inode *inode, struct file *file) | |
338 | { | |
339 | struct ll_file_data *fd; | |
340 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
341 | struct ll_inode_info *lli = ll_i2info(inode); | |
342 | int rc; | |
d7e09d03 | 343 | |
97a075cd JN |
344 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n", |
345 | PFID(ll_inode2fid(inode)), inode); | |
d7e09d03 | 346 | |
f76c23da | 347 | if (!is_root_inode(inode)) |
d7e09d03 PT |
348 | ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1); |
349 | fd = LUSTRE_FPRIVATE(file); | |
6e16818b | 350 | LASSERT(fd); |
d7e09d03 | 351 | |
c0894c6c | 352 | /* The last ref on @file, maybe not be the owner pid of statahead. |
d7e09d03 | 353 | * Different processes can open the same dir, "ll_opendir_key" means: |
c0894c6c OD |
354 | * it is me that should stop the statahead thread. |
355 | */ | |
d7e09d03 PT |
356 | if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd && |
357 | lli->lli_opendir_pid != 0) | |
358 | ll_stop_statahead(inode, lli->lli_opendir_key); | |
359 | ||
f76c23da | 360 | if (is_root_inode(inode)) { |
d7e09d03 PT |
361 | LUSTRE_FPRIVATE(file) = NULL; |
362 | ll_file_data_put(fd); | |
0a3bdb00 | 363 | return 0; |
d7e09d03 PT |
364 | } |
365 | ||
366 | if (!S_ISDIR(inode->i_mode)) { | |
367 | lov_read_and_clear_async_rc(lli->lli_clob); | |
368 | lli->lli_async_rc = 0; | |
369 | } | |
370 | ||
371 | rc = ll_md_close(sbi->ll_md_exp, inode, file); | |
372 | ||
373 | if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val)) | |
374 | libcfs_debug_dumplog(); | |
375 | ||
0a3bdb00 | 376 | return rc; |
d7e09d03 PT |
377 | } |
378 | ||
48eddfd5 | 379 | static int ll_intent_file_open(struct dentry *dentry, void *lmm, |
d7e09d03 PT |
380 | int lmmsize, struct lookup_intent *itp) |
381 | { | |
2b0143b5 | 382 | struct inode *inode = d_inode(dentry); |
48eddfd5 AV |
383 | struct ll_sb_info *sbi = ll_i2sbi(inode); |
384 | struct dentry *parent = dentry->d_parent; | |
385 | const char *name = dentry->d_name.name; | |
386 | const int len = dentry->d_name.len; | |
d7e09d03 PT |
387 | struct md_op_data *op_data; |
388 | struct ptlrpc_request *req; | |
389 | __u32 opc = LUSTRE_OPC_ANY; | |
390 | int rc; | |
d7e09d03 | 391 | |
c0894c6c | 392 | /* Usually we come here only for NFSD, and we want open lock. */ |
d7e09d03 PT |
393 | /* We can also get here if there was cached open handle in revalidate_it |
394 | * but it disappeared while we were getting from there to ll_file_open. | |
bef31c78 | 395 | * But this means this file was closed and immediately opened which |
c0894c6c OD |
396 | * makes a good candidate for using OPEN lock |
397 | */ | |
d7e09d03 | 398 | /* If lmmsize & lmm are not 0, we are just setting stripe info |
c0894c6c OD |
399 | * parameters. No need for the open lock |
400 | */ | |
6e16818b | 401 | if (!lmm && lmmsize == 0) { |
6dad4d89 OD |
402 | struct ll_dentry_data *ldd = ll_d2d(dentry); |
403 | /* | |
404 | * If we came via ll_iget_for_nfs, then we need to request | |
405 | * struct ll_dentry_data *ldd = ll_d2d(file->f_dentry); | |
406 | * | |
407 | * NB: when ldd is NULL, it must have come via normal | |
408 | * lookup path only, since ll_iget_for_nfs always calls | |
409 | * ll_d_init(). | |
410 | */ | |
411 | if (ldd && ldd->lld_nfs_dentry) { | |
412 | ldd->lld_nfs_dentry = 0; | |
413 | itp->it_flags |= MDS_OPEN_LOCK; | |
414 | } | |
d7e09d03 PT |
415 | if (itp->it_flags & FMODE_WRITE) |
416 | opc = LUSTRE_OPC_CREATE; | |
417 | } | |
418 | ||
2b0143b5 | 419 | op_data = ll_prep_md_op_data(NULL, d_inode(parent), |
48eddfd5 | 420 | inode, name, len, |
d7e09d03 PT |
421 | O_RDWR, opc, NULL); |
422 | if (IS_ERR(op_data)) | |
0a3bdb00 | 423 | return PTR_ERR(op_data); |
d7e09d03 PT |
424 | |
425 | itp->it_flags |= MDS_OPEN_BY_FID; | |
426 | rc = md_intent_lock(sbi->ll_md_exp, op_data, lmm, lmmsize, itp, | |
427 | 0 /*unused */, &req, ll_md_blocking_ast, 0); | |
428 | ll_finish_md_op_data(op_data); | |
429 | if (rc == -ESTALE) { | |
430 | /* reason for keep own exit path - don`t flood log | |
431 | * with messages with -ESTALE errors. | |
432 | */ | |
433 | if (!it_disposition(itp, DISP_OPEN_OPEN) || | |
e15ba45d | 434 | it_open_error(DISP_OPEN_OPEN, itp)) |
34e1f2bb | 435 | goto out; |
e22fdcc8 | 436 | ll_release_openhandle(inode, itp); |
34e1f2bb | 437 | goto out; |
d7e09d03 PT |
438 | } |
439 | ||
34e1f2bb JL |
440 | if (it_disposition(itp, DISP_LOOKUP_NEG)) { |
441 | rc = -ENOENT; | |
442 | goto out; | |
443 | } | |
d7e09d03 PT |
444 | |
445 | if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) { | |
446 | rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp); | |
447 | CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc); | |
34e1f2bb | 448 | goto out; |
d7e09d03 PT |
449 | } |
450 | ||
48eddfd5 | 451 | rc = ll_prep_inode(&inode, req, NULL, itp); |
e476f2e5 | 452 | if (!rc && itp->it_lock_mode) |
48eddfd5 | 453 | ll_set_lock_data(sbi->ll_md_exp, inode, itp, NULL); |
d7e09d03 PT |
454 | |
455 | out: | |
f236f69b | 456 | ptlrpc_req_finished(req); |
d7e09d03 PT |
457 | ll_intent_drop_lock(itp); |
458 | ||
0a3bdb00 | 459 | return rc; |
d7e09d03 PT |
460 | } |
461 | ||
462 | /** | |
463 | * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does | |
464 | * not believe attributes if a few ioepoch holders exist. Attributes for | |
465 | * previous ioepoch if new one is opened are also skipped by MDS. | |
466 | */ | |
467 | void ll_ioepoch_open(struct ll_inode_info *lli, __u64 ioepoch) | |
468 | { | |
469 | if (ioepoch && lli->lli_ioepoch != ioepoch) { | |
470 | lli->lli_ioepoch = ioepoch; | |
b0f5aad5 | 471 | CDEBUG(D_INODE, "Epoch %llu opened on "DFID"\n", |
d7e09d03 PT |
472 | ioepoch, PFID(&lli->lli_fid)); |
473 | } | |
474 | } | |
475 | ||
ea1db081 JH |
476 | static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it, |
477 | struct obd_client_handle *och) | |
d7e09d03 | 478 | { |
d7e09d03 PT |
479 | struct mdt_body *body; |
480 | ||
8bf86fd9 | 481 | body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY); |
ea1db081 JH |
482 | och->och_fh = body->handle; |
483 | och->och_fid = body->fid1; | |
e476f2e5 | 484 | och->och_lease_handle.cookie = it->it_lock_handle; |
d7e09d03 | 485 | och->och_magic = OBD_CLIENT_HANDLE_MAGIC; |
d7e09d03 | 486 | och->och_flags = it->it_flags; |
d7e09d03 | 487 | |
63d42578 | 488 | return md_set_open_replay_data(md_exp, och, it); |
d7e09d03 PT |
489 | } |
490 | ||
2d95f10e JH |
491 | static int ll_local_open(struct file *file, struct lookup_intent *it, |
492 | struct ll_file_data *fd, struct obd_client_handle *och) | |
d7e09d03 | 493 | { |
2a8a3597 | 494 | struct inode *inode = file_inode(file); |
d7e09d03 | 495 | struct ll_inode_info *lli = ll_i2info(inode); |
d7e09d03 PT |
496 | |
497 | LASSERT(!LUSTRE_FPRIVATE(file)); | |
498 | ||
6e16818b | 499 | LASSERT(fd); |
d7e09d03 PT |
500 | |
501 | if (och) { | |
d7e09d03 PT |
502 | struct mdt_body *body; |
503 | int rc; | |
504 | ||
ea1db081 JH |
505 | rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och); |
506 | if (rc != 0) | |
0a3bdb00 | 507 | return rc; |
d7e09d03 | 508 | |
8bf86fd9 JH |
509 | body = req_capsule_server_get(&it->it_request->rq_pill, |
510 | &RMF_MDT_BODY); | |
ea1db081 | 511 | ll_ioepoch_open(lli, body->ioepoch); |
d7e09d03 PT |
512 | } |
513 | ||
514 | LUSTRE_FPRIVATE(file) = fd; | |
515 | ll_readahead_init(inode, &fd->fd_ras); | |
d3a8a4e2 | 516 | fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC); |
966c4a8f JX |
517 | |
518 | /* ll_cl_context initialize */ | |
519 | rwlock_init(&fd->fd_lock); | |
520 | INIT_LIST_HEAD(&fd->fd_lccs); | |
521 | ||
0a3bdb00 | 522 | return 0; |
d7e09d03 PT |
523 | } |
524 | ||
525 | /* Open a file, and (for the very first open) create objects on the OSTs at | |
526 | * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object | |
527 | * creation or open until ll_lov_setstripe() ioctl is called. | |
528 | * | |
529 | * If we already have the stripe MD locally then we don't request it in | |
530 | * md_open(), by passing a lmm_size = 0. | |
531 | * | |
532 | * It is up to the application to ensure no other processes open this file | |
533 | * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be | |
534 | * used. We might be able to avoid races of that sort by getting lli_open_sem | |
535 | * before returning in the O_LOV_DELAY_CREATE case and dropping it here | |
536 | * or in ll_file_release(), but I'm not sure that is desirable/necessary. | |
537 | */ | |
538 | int ll_file_open(struct inode *inode, struct file *file) | |
539 | { | |
540 | struct ll_inode_info *lli = ll_i2info(inode); | |
541 | struct lookup_intent *it, oit = { .it_op = IT_OPEN, | |
542 | .it_flags = file->f_flags }; | |
543 | struct obd_client_handle **och_p = NULL; | |
544 | __u64 *och_usecount = NULL; | |
545 | struct ll_file_data *fd; | |
546 | int rc = 0, opendir_set = 0; | |
d7e09d03 | 547 | |
97a075cd JN |
548 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n", |
549 | PFID(ll_inode2fid(inode)), inode, file->f_flags); | |
d7e09d03 PT |
550 | |
551 | it = file->private_data; /* XXX: compat macro */ | |
552 | file->private_data = NULL; /* prevent ll_local_open assertion */ | |
553 | ||
554 | fd = ll_file_data_get(); | |
6e16818b | 555 | if (!fd) { |
34e1f2bb JL |
556 | rc = -ENOMEM; |
557 | goto out_openerr; | |
558 | } | |
d7e09d03 PT |
559 | |
560 | fd->fd_file = file; | |
561 | if (S_ISDIR(inode->i_mode)) { | |
562 | spin_lock(&lli->lli_sa_lock); | |
6e16818b | 563 | if (!lli->lli_opendir_key && !lli->lli_sai && |
d7e09d03 PT |
564 | lli->lli_opendir_pid == 0) { |
565 | lli->lli_opendir_key = fd; | |
566 | lli->lli_opendir_pid = current_pid(); | |
567 | opendir_set = 1; | |
568 | } | |
569 | spin_unlock(&lli->lli_sa_lock); | |
570 | } | |
571 | ||
f76c23da | 572 | if (is_root_inode(inode)) { |
d7e09d03 | 573 | LUSTRE_FPRIVATE(file) = fd; |
0a3bdb00 | 574 | return 0; |
d7e09d03 PT |
575 | } |
576 | ||
e476f2e5 | 577 | if (!it || !it->it_disposition) { |
d7e09d03 PT |
578 | /* Convert f_flags into access mode. We cannot use file->f_mode, |
579 | * because everything but O_ACCMODE mask was stripped from | |
c0894c6c OD |
580 | * there |
581 | */ | |
d7e09d03 PT |
582 | if ((oit.it_flags + 1) & O_ACCMODE) |
583 | oit.it_flags++; | |
584 | if (file->f_flags & O_TRUNC) | |
585 | oit.it_flags |= FMODE_WRITE; | |
586 | ||
587 | /* kernel only call f_op->open in dentry_open. filp_open calls | |
588 | * dentry_open after call to open_namei that checks permissions. | |
589 | * Only nfsd_open call dentry_open directly without checking | |
c0894c6c OD |
590 | * permissions and because of that this code below is safe. |
591 | */ | |
d7e09d03 PT |
592 | if (oit.it_flags & (FMODE_WRITE | FMODE_READ)) |
593 | oit.it_flags |= MDS_OPEN_OWNEROVERRIDE; | |
594 | ||
595 | /* We do not want O_EXCL here, presumably we opened the file | |
c0894c6c OD |
596 | * already? XXX - NFS implications? |
597 | */ | |
d7e09d03 PT |
598 | oit.it_flags &= ~O_EXCL; |
599 | ||
600 | /* bug20584, if "it_flags" contains O_CREAT, the file will be | |
601 | * created if necessary, then "IT_CREAT" should be set to keep | |
c0894c6c OD |
602 | * consistent with it |
603 | */ | |
d7e09d03 PT |
604 | if (oit.it_flags & O_CREAT) |
605 | oit.it_op |= IT_CREAT; | |
606 | ||
607 | it = &oit; | |
608 | } | |
609 | ||
610 | restart: | |
611 | /* Let's see if we have file open on MDS already. */ | |
612 | if (it->it_flags & FMODE_WRITE) { | |
613 | och_p = &lli->lli_mds_write_och; | |
614 | och_usecount = &lli->lli_open_fd_write_count; | |
615 | } else if (it->it_flags & FMODE_EXEC) { | |
616 | och_p = &lli->lli_mds_exec_och; | |
617 | och_usecount = &lli->lli_open_fd_exec_count; | |
618 | } else { | |
619 | och_p = &lli->lli_mds_read_och; | |
620 | och_usecount = &lli->lli_open_fd_read_count; | |
621 | } | |
622 | ||
623 | mutex_lock(&lli->lli_och_mutex); | |
624 | if (*och_p) { /* Open handle is present */ | |
625 | if (it_disposition(it, DISP_OPEN_OPEN)) { | |
626 | /* Well, there's extra open request that we do not need, | |
c0894c6c OD |
627 | * let's close it somehow. This will decref request. |
628 | */ | |
d7e09d03 PT |
629 | rc = it_open_error(DISP_OPEN_OPEN, it); |
630 | if (rc) { | |
631 | mutex_unlock(&lli->lli_och_mutex); | |
34e1f2bb | 632 | goto out_openerr; |
d7e09d03 PT |
633 | } |
634 | ||
e22fdcc8 | 635 | ll_release_openhandle(inode, it); |
d7e09d03 PT |
636 | } |
637 | (*och_usecount)++; | |
638 | ||
639 | rc = ll_local_open(file, it, fd, NULL); | |
640 | if (rc) { | |
641 | (*och_usecount)--; | |
642 | mutex_unlock(&lli->lli_och_mutex); | |
34e1f2bb | 643 | goto out_openerr; |
d7e09d03 PT |
644 | } |
645 | } else { | |
646 | LASSERT(*och_usecount == 0); | |
e476f2e5 | 647 | if (!it->it_disposition) { |
d7e09d03 | 648 | /* We cannot just request lock handle now, new ELC code |
c0894c6c OD |
649 | * means that one of other OPEN locks for this file |
650 | * could be cancelled, and since blocking ast handler | |
651 | * would attempt to grab och_mutex as well, that would | |
652 | * result in a deadlock | |
653 | */ | |
d7e09d03 PT |
654 | mutex_unlock(&lli->lli_och_mutex); |
655 | it->it_create_mode |= M_CHECK_STALE; | |
48eddfd5 | 656 | rc = ll_intent_file_open(file->f_path.dentry, NULL, 0, it); |
d7e09d03 PT |
657 | it->it_create_mode &= ~M_CHECK_STALE; |
658 | if (rc) | |
34e1f2bb | 659 | goto out_openerr; |
d7e09d03 PT |
660 | |
661 | goto restart; | |
662 | } | |
496a51bd | 663 | *och_p = kzalloc(sizeof(struct obd_client_handle), GFP_NOFS); |
34e1f2bb JL |
664 | if (!*och_p) { |
665 | rc = -ENOMEM; | |
666 | goto out_och_free; | |
667 | } | |
d7e09d03 PT |
668 | |
669 | (*och_usecount)++; | |
670 | ||
671 | /* md_intent_lock() didn't get a request ref if there was an | |
672 | * open error, so don't do cleanup on the request here | |
c0894c6c OD |
673 | * (bug 3430) |
674 | */ | |
d7e09d03 | 675 | /* XXX (green): Should not we bail out on any error here, not |
c0894c6c OD |
676 | * just open error? |
677 | */ | |
d7e09d03 PT |
678 | rc = it_open_error(DISP_OPEN_OPEN, it); |
679 | if (rc) | |
34e1f2bb | 680 | goto out_och_free; |
d7e09d03 | 681 | |
5787be94 AD |
682 | LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF), |
683 | "inode %p: disposition %x, status %d\n", inode, | |
e476f2e5 | 684 | it_disposition(it, ~0), it->it_status); |
d7e09d03 PT |
685 | |
686 | rc = ll_local_open(file, it, fd, *och_p); | |
687 | if (rc) | |
34e1f2bb | 688 | goto out_och_free; |
d7e09d03 PT |
689 | } |
690 | mutex_unlock(&lli->lli_och_mutex); | |
691 | fd = NULL; | |
692 | ||
693 | /* Must do this outside lli_och_mutex lock to prevent deadlock where | |
c0894c6c OD |
694 | * different kind of OPEN lock for this same inode gets cancelled |
695 | * by ldlm_cancel_lru | |
696 | */ | |
d7e09d03 | 697 | if (!S_ISREG(inode->i_mode)) |
34e1f2bb | 698 | goto out_och_free; |
d7e09d03 | 699 | |
38585ccc AD |
700 | if (!lli->lli_has_smd && |
701 | (cl_is_lov_delay_create(file->f_flags) || | |
702 | (file->f_mode & FMODE_WRITE) == 0)) { | |
703 | CDEBUG(D_INODE, "object creation was delayed\n"); | |
34e1f2bb | 704 | goto out_och_free; |
d7e09d03 | 705 | } |
38585ccc | 706 | cl_lov_delay_create_clear(&file->f_flags); |
34e1f2bb | 707 | goto out_och_free; |
d7e09d03 PT |
708 | |
709 | out_och_free: | |
710 | if (rc) { | |
711 | if (och_p && *och_p) { | |
97903a26 | 712 | kfree(*och_p); |
c0a2472f | 713 | *och_p = NULL; |
d7e09d03 PT |
714 | (*och_usecount)--; |
715 | } | |
716 | mutex_unlock(&lli->lli_och_mutex); | |
717 | ||
718 | out_openerr: | |
719 | if (opendir_set != 0) | |
720 | ll_stop_statahead(inode, lli->lli_opendir_key); | |
a5cb8880 | 721 | ll_file_data_put(fd); |
d7e09d03 PT |
722 | } else { |
723 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1); | |
724 | } | |
725 | ||
726 | if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) { | |
8bf86fd9 | 727 | ptlrpc_req_finished(it->it_request); |
d7e09d03 PT |
728 | it_clear_disposition(it, DISP_ENQ_OPEN_REF); |
729 | } | |
730 | ||
731 | return rc; | |
732 | } | |
733 | ||
d3a8a4e2 | 734 | static int ll_md_blocking_lease_ast(struct ldlm_lock *lock, |
e15ba45d OD |
735 | struct ldlm_lock_desc *desc, |
736 | void *data, int flag) | |
d3a8a4e2 JX |
737 | { |
738 | int rc; | |
739 | struct lustre_handle lockh; | |
740 | ||
741 | switch (flag) { | |
742 | case LDLM_CB_BLOCKING: | |
743 | ldlm_lock2handle(lock, &lockh); | |
744 | rc = ldlm_cli_cancel(&lockh, LCF_ASYNC); | |
745 | if (rc < 0) { | |
746 | CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc); | |
747 | return rc; | |
748 | } | |
749 | break; | |
750 | case LDLM_CB_CANCELING: | |
751 | /* do nothing */ | |
752 | break; | |
753 | } | |
754 | return 0; | |
755 | } | |
756 | ||
757 | /** | |
758 | * Acquire a lease and open the file. | |
759 | */ | |
2d95f10e JH |
760 | static struct obd_client_handle * |
761 | ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode, | |
762 | __u64 open_flags) | |
d3a8a4e2 JX |
763 | { |
764 | struct lookup_intent it = { .it_op = IT_OPEN }; | |
765 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
766 | struct md_op_data *op_data; | |
767 | struct ptlrpc_request *req; | |
768 | struct lustre_handle old_handle = { 0 }; | |
769 | struct obd_client_handle *och = NULL; | |
770 | int rc; | |
771 | int rc2; | |
772 | ||
773 | if (fmode != FMODE_WRITE && fmode != FMODE_READ) | |
774 | return ERR_PTR(-EINVAL); | |
775 | ||
6e16818b | 776 | if (file) { |
d3a8a4e2 JX |
777 | struct ll_inode_info *lli = ll_i2info(inode); |
778 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
779 | struct obd_client_handle **och_p; | |
780 | __u64 *och_usecount; | |
781 | ||
782 | if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC)) | |
783 | return ERR_PTR(-EPERM); | |
784 | ||
785 | /* Get the openhandle of the file */ | |
786 | rc = -EBUSY; | |
787 | mutex_lock(&lli->lli_och_mutex); | |
6e16818b | 788 | if (fd->fd_lease_och) { |
d3a8a4e2 JX |
789 | mutex_unlock(&lli->lli_och_mutex); |
790 | return ERR_PTR(rc); | |
791 | } | |
792 | ||
6e16818b | 793 | if (!fd->fd_och) { |
d3a8a4e2 | 794 | if (file->f_mode & FMODE_WRITE) { |
6e16818b | 795 | LASSERT(lli->lli_mds_write_och); |
d3a8a4e2 JX |
796 | och_p = &lli->lli_mds_write_och; |
797 | och_usecount = &lli->lli_open_fd_write_count; | |
798 | } else { | |
6e16818b | 799 | LASSERT(lli->lli_mds_read_och); |
d3a8a4e2 JX |
800 | och_p = &lli->lli_mds_read_och; |
801 | och_usecount = &lli->lli_open_fd_read_count; | |
802 | } | |
803 | if (*och_usecount == 1) { | |
804 | fd->fd_och = *och_p; | |
805 | *och_p = NULL; | |
806 | *och_usecount = 0; | |
807 | rc = 0; | |
808 | } | |
809 | } | |
810 | mutex_unlock(&lli->lli_och_mutex); | |
811 | if (rc < 0) /* more than 1 opener */ | |
812 | return ERR_PTR(rc); | |
813 | ||
6e16818b | 814 | LASSERT(fd->fd_och); |
d3a8a4e2 JX |
815 | old_handle = fd->fd_och->och_fh; |
816 | } | |
817 | ||
496a51bd JL |
818 | och = kzalloc(sizeof(*och), GFP_NOFS); |
819 | if (!och) | |
d3a8a4e2 JX |
820 | return ERR_PTR(-ENOMEM); |
821 | ||
822 | op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0, | |
e15ba45d | 823 | LUSTRE_OPC_ANY, NULL); |
34e1f2bb JL |
824 | if (IS_ERR(op_data)) { |
825 | rc = PTR_ERR(op_data); | |
826 | goto out; | |
827 | } | |
d3a8a4e2 JX |
828 | |
829 | /* To tell the MDT this openhandle is from the same owner */ | |
830 | op_data->op_handle = old_handle; | |
831 | ||
48d23e61 JX |
832 | it.it_flags = fmode | open_flags; |
833 | it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE; | |
d3a8a4e2 | 834 | rc = md_intent_lock(sbi->ll_md_exp, op_data, NULL, 0, &it, 0, &req, |
e15ba45d | 835 | ll_md_blocking_lease_ast, |
d3a8a4e2 JX |
836 | /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise |
837 | * it can be cancelled which may mislead applications that the lease is | |
838 | * broken; | |
839 | * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal | |
840 | * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast | |
c0894c6c OD |
841 | * doesn't deal with openhandle, so normal openhandle will be leaked. |
842 | */ | |
d3a8a4e2 JX |
843 | LDLM_FL_NO_LRU | LDLM_FL_EXCL); |
844 | ll_finish_md_op_data(op_data); | |
f236f69b | 845 | ptlrpc_req_finished(req); |
d3a8a4e2 | 846 | if (rc < 0) |
34e1f2bb | 847 | goto out_release_it; |
d3a8a4e2 | 848 | |
34e1f2bb JL |
849 | if (it_disposition(&it, DISP_LOOKUP_NEG)) { |
850 | rc = -ENOENT; | |
851 | goto out_release_it; | |
852 | } | |
d3a8a4e2 JX |
853 | |
854 | rc = it_open_error(DISP_OPEN_OPEN, &it); | |
855 | if (rc) | |
34e1f2bb | 856 | goto out_release_it; |
d3a8a4e2 JX |
857 | |
858 | LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF)); | |
859 | ll_och_fill(sbi->ll_md_exp, &it, och); | |
860 | ||
34e1f2bb JL |
861 | if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */ { |
862 | rc = -EOPNOTSUPP; | |
863 | goto out_close; | |
864 | } | |
d3a8a4e2 JX |
865 | |
866 | /* already get lease, handle lease lock */ | |
867 | ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL); | |
e476f2e5 JH |
868 | if (it.it_lock_mode == 0 || |
869 | it.it_lock_bits != MDS_INODELOCK_OPEN) { | |
d3a8a4e2 JX |
870 | /* open lock must return for lease */ |
871 | CERROR(DFID "lease granted but no open lock, %d/%llu.\n", | |
e476f2e5 JH |
872 | PFID(ll_inode2fid(inode)), it.it_lock_mode, |
873 | it.it_lock_bits); | |
34e1f2bb JL |
874 | rc = -EPROTO; |
875 | goto out_close; | |
d3a8a4e2 JX |
876 | } |
877 | ||
878 | ll_intent_release(&it); | |
879 | return och; | |
880 | ||
881 | out_close: | |
e55a68b6 | 882 | /* Cancel open lock */ |
e476f2e5 | 883 | if (it.it_lock_mode != 0) { |
d3a8a4e2 | 884 | ldlm_lock_decref_and_cancel(&och->och_lease_handle, |
e476f2e5 JH |
885 | it.it_lock_mode); |
886 | it.it_lock_mode = 0; | |
e55a68b6 | 887 | och->och_lease_handle.cookie = 0ULL; |
d3a8a4e2 | 888 | } |
e55a68b6 JX |
889 | rc2 = ll_close_inode_openhandle(sbi->ll_md_exp, inode, och, NULL); |
890 | if (rc2 < 0) | |
891 | CERROR("%s: error closing file "DFID": %d\n", | |
892 | ll_get_fsname(inode->i_sb, NULL, 0), | |
893 | PFID(&ll_i2info(inode)->lli_fid), rc2); | |
894 | och = NULL; /* och has been freed in ll_close_inode_openhandle() */ | |
d3a8a4e2 JX |
895 | out_release_it: |
896 | ll_intent_release(&it); | |
897 | out: | |
97903a26 | 898 | kfree(och); |
d3a8a4e2 JX |
899 | return ERR_PTR(rc); |
900 | } | |
d3a8a4e2 JX |
901 | |
902 | /** | |
903 | * Release lease and close the file. | |
904 | * It will check if the lease has ever broken. | |
905 | */ | |
2d95f10e JH |
906 | static int ll_lease_close(struct obd_client_handle *och, struct inode *inode, |
907 | bool *lease_broken) | |
d3a8a4e2 JX |
908 | { |
909 | struct ldlm_lock *lock; | |
910 | bool cancelled = true; | |
911 | int rc; | |
912 | ||
913 | lock = ldlm_handle2lock(&och->och_lease_handle); | |
6e16818b | 914 | if (lock) { |
d3a8a4e2 JX |
915 | lock_res_and_lock(lock); |
916 | cancelled = ldlm_is_cancel(lock); | |
917 | unlock_res_and_lock(lock); | |
ead02808 | 918 | LDLM_LOCK_PUT(lock); |
d3a8a4e2 JX |
919 | } |
920 | ||
e15ba45d OD |
921 | CDEBUG(D_INODE, "lease for " DFID " broken? %d\n", |
922 | PFID(&ll_i2info(inode)->lli_fid), cancelled); | |
d3a8a4e2 JX |
923 | |
924 | if (!cancelled) | |
925 | ldlm_cli_cancel(&och->och_lease_handle, 0); | |
6e16818b | 926 | if (lease_broken) |
d3a8a4e2 JX |
927 | *lease_broken = cancelled; |
928 | ||
48d23e61 JX |
929 | rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, inode, och, |
930 | NULL); | |
d3a8a4e2 JX |
931 | return rc; |
932 | } | |
d3a8a4e2 | 933 | |
d7e09d03 PT |
934 | /* Fills the obdo with the attributes for the lsm */ |
935 | static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp, | |
e1798006 | 936 | struct obdo *obdo, __u64 ioepoch, int dv_flags) |
d7e09d03 PT |
937 | { |
938 | struct ptlrpc_request_set *set; | |
45efd655 | 939 | struct obd_info oinfo = { }; |
d7e09d03 PT |
940 | int rc; |
941 | ||
6e16818b | 942 | LASSERT(lsm); |
d7e09d03 PT |
943 | |
944 | oinfo.oi_md = lsm; | |
945 | oinfo.oi_oa = obdo; | |
946 | oinfo.oi_oa->o_oi = lsm->lsm_oi; | |
947 | oinfo.oi_oa->o_mode = S_IFREG; | |
948 | oinfo.oi_oa->o_ioepoch = ioepoch; | |
949 | oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | | |
950 | OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | | |
951 | OBD_MD_FLBLKSZ | OBD_MD_FLATIME | | |
952 | OBD_MD_FLMTIME | OBD_MD_FLCTIME | | |
953 | OBD_MD_FLGROUP | OBD_MD_FLEPOCH | | |
954 | OBD_MD_FLDATAVERSION; | |
e1798006 | 955 | if (dv_flags & (LL_DV_WR_FLUSH | LL_DV_RD_FLUSH)) { |
d7e09d03 PT |
956 | oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS; |
957 | oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK; | |
e1798006 JX |
958 | if (dv_flags & LL_DV_WR_FLUSH) |
959 | oinfo.oi_oa->o_flags |= OBD_FL_FLUSH; | |
d7e09d03 PT |
960 | } |
961 | ||
962 | set = ptlrpc_prep_set(); | |
6e16818b | 963 | if (!set) { |
19b2056f | 964 | CERROR("cannot allocate ptlrpc set: rc = %d\n", -ENOMEM); |
d7e09d03 PT |
965 | rc = -ENOMEM; |
966 | } else { | |
967 | rc = obd_getattr_async(exp, &oinfo, set); | |
968 | if (rc == 0) | |
969 | rc = ptlrpc_set_wait(set); | |
970 | ptlrpc_set_destroy(set); | |
971 | } | |
e1798006 | 972 | if (rc == 0) { |
d7e09d03 PT |
973 | oinfo.oi_oa->o_valid &= (OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ | |
974 | OBD_MD_FLATIME | OBD_MD_FLMTIME | | |
975 | OBD_MD_FLCTIME | OBD_MD_FLSIZE | | |
e1798006 JX |
976 | OBD_MD_FLDATAVERSION | OBD_MD_FLFLAGS); |
977 | if (dv_flags & LL_DV_WR_FLUSH && | |
978 | !(oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS && | |
979 | oinfo.oi_oa->o_flags & OBD_FL_FLUSH)) | |
980 | return -ENOTSUPP; | |
981 | } | |
0a3bdb00 | 982 | return rc; |
d7e09d03 PT |
983 | } |
984 | ||
985 | /** | |
986 | * Performs the getattr on the inode and updates its fields. | |
987 | * If @sync != 0, perform the getattr under the server-side lock. | |
988 | */ | |
989 | int ll_inode_getattr(struct inode *inode, struct obdo *obdo, | |
990 | __u64 ioepoch, int sync) | |
991 | { | |
d7e09d03 PT |
992 | struct lov_stripe_md *lsm; |
993 | int rc; | |
d7e09d03 PT |
994 | |
995 | lsm = ccc_inode_lsm_get(inode); | |
996 | rc = ll_lsm_getattr(lsm, ll_i2dtexp(inode), | |
e1798006 | 997 | obdo, ioepoch, sync ? LL_DV_RD_FLUSH : 0); |
d7e09d03 PT |
998 | if (rc == 0) { |
999 | struct ost_id *oi = lsm ? &lsm->lsm_oi : &obdo->o_oi; | |
1000 | ||
1001 | obdo_refresh_inode(inode, obdo, obdo->o_valid); | |
2d00bd17 JP |
1002 | CDEBUG(D_INODE, "objid " DOSTID " size %llu, blocks %llu, blksize %lu\n", |
1003 | POSTID(oi), i_size_read(inode), | |
d7e09d03 | 1004 | (unsigned long long)inode->i_blocks, |
16e0631d | 1005 | 1UL << inode->i_blkbits); |
d7e09d03 PT |
1006 | } |
1007 | ccc_inode_lsm_put(inode, lsm); | |
0a3bdb00 | 1008 | return rc; |
d7e09d03 PT |
1009 | } |
1010 | ||
d2995737 | 1011 | int ll_merge_attr(const struct lu_env *env, struct inode *inode) |
d7e09d03 PT |
1012 | { |
1013 | struct ll_inode_info *lli = ll_i2info(inode); | |
1014 | struct cl_object *obj = lli->lli_clob; | |
9acc4500 | 1015 | struct cl_attr *attr = vvp_env_thread_attr(env); |
d2995737 JH |
1016 | s64 atime; |
1017 | s64 mtime; | |
1018 | s64 ctime; | |
d7e09d03 PT |
1019 | int rc = 0; |
1020 | ||
d7e09d03 | 1021 | ll_inode_size_lock(inode); |
d2995737 | 1022 | |
d7e09d03 | 1023 | /* merge timestamps the most recently obtained from mds with |
c0894c6c OD |
1024 | * timestamps obtained from osts |
1025 | */ | |
d2995737 JH |
1026 | LTIME_S(inode->i_atime) = lli->lli_atime; |
1027 | LTIME_S(inode->i_mtime) = lli->lli_mtime; | |
1028 | LTIME_S(inode->i_ctime) = lli->lli_ctime; | |
376ef86b | 1029 | |
d2995737 JH |
1030 | mtime = LTIME_S(inode->i_mtime); |
1031 | atime = LTIME_S(inode->i_atime); | |
1032 | ctime = LTIME_S(inode->i_ctime); | |
d7e09d03 PT |
1033 | |
1034 | cl_object_attr_lock(obj); | |
1035 | rc = cl_object_attr_get(env, obj, attr); | |
1036 | cl_object_attr_unlock(obj); | |
1037 | ||
d2995737 JH |
1038 | if (rc != 0) |
1039 | goto out_size_unlock; | |
d7e09d03 | 1040 | |
d2995737 JH |
1041 | if (atime < attr->cat_atime) |
1042 | atime = attr->cat_atime; | |
d7e09d03 | 1043 | |
d2995737 JH |
1044 | if (ctime < attr->cat_ctime) |
1045 | ctime = attr->cat_ctime; | |
d7e09d03 | 1046 | |
d2995737 JH |
1047 | if (mtime < attr->cat_mtime) |
1048 | mtime = attr->cat_mtime; | |
1049 | ||
1050 | CDEBUG(D_VFSTRACE, DFID " updating i_size %llu\n", | |
1051 | PFID(&lli->lli_fid), attr->cat_size); | |
1052 | ||
1929c433 | 1053 | i_size_write(inode, attr->cat_size); |
d2995737 JH |
1054 | |
1055 | inode->i_blocks = attr->cat_blocks; | |
1056 | ||
1057 | LTIME_S(inode->i_mtime) = mtime; | |
1058 | LTIME_S(inode->i_atime) = atime; | |
1059 | LTIME_S(inode->i_ctime) = ctime; | |
1060 | ||
1061 | out_size_unlock: | |
d7e09d03 PT |
1062 | ll_inode_size_unlock(inode); |
1063 | ||
0a3bdb00 | 1064 | return rc; |
d7e09d03 PT |
1065 | } |
1066 | ||
1067 | int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm, | |
1068 | lstat_t *st) | |
1069 | { | |
1070 | struct obdo obdo = { 0 }; | |
1071 | int rc; | |
1072 | ||
ef2e0f55 | 1073 | rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, &obdo, 0, 0); |
d7e09d03 PT |
1074 | if (rc == 0) { |
1075 | st->st_size = obdo.o_size; | |
1076 | st->st_blocks = obdo.o_blocks; | |
1077 | st->st_mtime = obdo.o_mtime; | |
1078 | st->st_atime = obdo.o_atime; | |
1079 | st->st_ctime = obdo.o_ctime; | |
1080 | } | |
1081 | return rc; | |
1082 | } | |
1083 | ||
ec9bca9c JH |
1084 | static bool file_is_noatime(const struct file *file) |
1085 | { | |
1086 | const struct vfsmount *mnt = file->f_path.mnt; | |
2a8a3597 | 1087 | const struct inode *inode = file_inode(file); |
ec9bca9c JH |
1088 | |
1089 | /* Adapted from file_accessed() and touch_atime().*/ | |
1090 | if (file->f_flags & O_NOATIME) | |
1091 | return true; | |
1092 | ||
1093 | if (inode->i_flags & S_NOATIME) | |
1094 | return true; | |
1095 | ||
1096 | if (IS_NOATIME(inode)) | |
1097 | return true; | |
1098 | ||
1099 | if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY)) | |
1100 | return true; | |
1101 | ||
1102 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) | |
1103 | return true; | |
1104 | ||
1105 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) | |
1106 | return true; | |
1107 | ||
1108 | return false; | |
1109 | } | |
1110 | ||
d7e09d03 PT |
1111 | void ll_io_init(struct cl_io *io, const struct file *file, int write) |
1112 | { | |
2a8a3597 | 1113 | struct inode *inode = file_inode(file); |
d7e09d03 PT |
1114 | |
1115 | io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK; | |
1116 | if (write) { | |
1117 | io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND); | |
1118 | io->u.ci_wr.wr_sync = file->f_flags & O_SYNC || | |
1119 | file->f_flags & O_DIRECT || | |
1120 | IS_SYNC(inode); | |
1121 | } | |
1122 | io->ci_obj = ll_i2info(inode)->lli_clob; | |
1123 | io->ci_lockreq = CILR_MAYBE; | |
1124 | if (ll_file_nolock(file)) { | |
1125 | io->ci_lockreq = CILR_NEVER; | |
1126 | io->ci_no_srvlock = 1; | |
1127 | } else if (file->f_flags & O_APPEND) { | |
1128 | io->ci_lockreq = CILR_MANDATORY; | |
1129 | } | |
ec9bca9c JH |
1130 | |
1131 | io->ci_noatime = file_is_noatime(file); | |
d7e09d03 PT |
1132 | } |
1133 | ||
1134 | static ssize_t | |
1135 | ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args, | |
1136 | struct file *file, enum cl_io_type iot, | |
1137 | loff_t *ppos, size_t count) | |
1138 | { | |
2a8a3597 | 1139 | struct ll_inode_info *lli = ll_i2info(file_inode(file)); |
d7e09d03 PT |
1140 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); |
1141 | struct cl_io *io; | |
1142 | ssize_t result; | |
d7e09d03 | 1143 | |
47b34458 AV |
1144 | CDEBUG(D_VFSTRACE, "file: %pD, type: %d ppos: %llu, count: %zd\n", |
1145 | file, iot, *ppos, count); | |
77605e41 | 1146 | |
d7e09d03 | 1147 | restart: |
9acc4500 | 1148 | io = vvp_env_thread_io(env); |
d7e09d03 PT |
1149 | ll_io_init(io, file, iot == CIT_WRITE); |
1150 | ||
1151 | if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) { | |
e0a8144b | 1152 | struct vvp_io *vio = vvp_env_io(env); |
d7e09d03 PT |
1153 | int write_mutex_locked = 0; |
1154 | ||
e0a8144b JH |
1155 | vio->vui_fd = LUSTRE_FPRIVATE(file); |
1156 | vio->vui_io_subtype = args->via_io_subtype; | |
d7e09d03 | 1157 | |
e0a8144b | 1158 | switch (vio->vui_io_subtype) { |
d7e09d03 | 1159 | case IO_NORMAL: |
e0a8144b JH |
1160 | vio->vui_iter = args->u.normal.via_iter; |
1161 | vio->vui_iocb = args->u.normal.via_iocb; | |
d7e09d03 | 1162 | if ((iot == CIT_WRITE) && |
e0a8144b | 1163 | !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) { |
d7e09d03 | 1164 | if (mutex_lock_interruptible(&lli-> |
34e1f2bb JL |
1165 | lli_write_mutex)) { |
1166 | result = -ERESTARTSYS; | |
1167 | goto out; | |
1168 | } | |
d7e09d03 | 1169 | write_mutex_locked = 1; |
d7e09d03 | 1170 | } |
77605e41 | 1171 | down_read(&lli->lli_trunc_sem); |
d7e09d03 | 1172 | break; |
d7e09d03 | 1173 | case IO_SPLICE: |
e0a8144b JH |
1174 | vio->u.splice.vui_pipe = args->u.splice.via_pipe; |
1175 | vio->u.splice.vui_flags = args->u.splice.via_flags; | |
d7e09d03 PT |
1176 | break; |
1177 | default: | |
e0a8144b | 1178 | CERROR("Unknown IO type - %u\n", vio->vui_io_subtype); |
d7e09d03 PT |
1179 | LBUG(); |
1180 | } | |
966c4a8f | 1181 | ll_cl_add(file, env, io); |
d7e09d03 | 1182 | result = cl_io_loop(env, io); |
966c4a8f | 1183 | ll_cl_remove(file, env); |
77605e41 JX |
1184 | if (args->via_io_subtype == IO_NORMAL) |
1185 | up_read(&lli->lli_trunc_sem); | |
d7e09d03 PT |
1186 | if (write_mutex_locked) |
1187 | mutex_unlock(&lli->lli_write_mutex); | |
d7e09d03 PT |
1188 | } else { |
1189 | /* cl_io_rw_init() handled IO */ | |
1190 | result = io->ci_result; | |
1191 | } | |
1192 | ||
1193 | if (io->ci_nob > 0) { | |
1194 | result = io->ci_nob; | |
1195 | *ppos = io->u.ci_wr.wr.crw_pos; | |
1196 | } | |
34e1f2bb | 1197 | goto out; |
d7e09d03 PT |
1198 | out: |
1199 | cl_io_fini(env, io); | |
1200 | /* If any bit been read/written (result != 0), we just return | |
c0894c6c OD |
1201 | * short read/write instead of restart io. |
1202 | */ | |
5ea17d6c | 1203 | if ((result == 0 || result == -ENODATA) && io->ci_need_restart) { |
09561a53 | 1204 | CDEBUG(D_VFSTRACE, "Restart %s on %pD from %lld, count:%zd\n", |
d7e09d03 | 1205 | iot == CIT_READ ? "read" : "write", |
09561a53 | 1206 | file, *ppos, count); |
19b2056f | 1207 | LASSERTF(io->ci_nob == 0, "%zd\n", io->ci_nob); |
d7e09d03 PT |
1208 | goto restart; |
1209 | } | |
1210 | ||
1211 | if (iot == CIT_READ) { | |
1212 | if (result >= 0) | |
2a8a3597 | 1213 | ll_stats_ops_tally(ll_i2sbi(file_inode(file)), |
d7e09d03 PT |
1214 | LPROC_LL_READ_BYTES, result); |
1215 | } else if (iot == CIT_WRITE) { | |
1216 | if (result >= 0) { | |
2a8a3597 | 1217 | ll_stats_ops_tally(ll_i2sbi(file_inode(file)), |
d7e09d03 PT |
1218 | LPROC_LL_WRITE_BYTES, result); |
1219 | fd->fd_write_failed = false; | |
1220 | } else if (result != -ERESTARTSYS) { | |
1221 | fd->fd_write_failed = true; | |
1222 | } | |
1223 | } | |
77605e41 | 1224 | CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result); |
d7e09d03 PT |
1225 | |
1226 | return result; | |
1227 | } | |
1228 | ||
b42b15fd | 1229 | static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
d7e09d03 PT |
1230 | { |
1231 | struct lu_env *env; | |
1232 | struct vvp_io_args *args; | |
d7e09d03 PT |
1233 | ssize_t result; |
1234 | int refcheck; | |
d7e09d03 | 1235 | |
d7e09d03 PT |
1236 | env = cl_env_get(&refcheck); |
1237 | if (IS_ERR(env)) | |
0a3bdb00 | 1238 | return PTR_ERR(env); |
d7e09d03 | 1239 | |
9989a58e | 1240 | args = ll_env_args(env, IO_NORMAL); |
b42b15fd | 1241 | args->u.normal.via_iter = to; |
d7e09d03 PT |
1242 | args->u.normal.via_iocb = iocb; |
1243 | ||
1244 | result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ, | |
b42b15fd | 1245 | &iocb->ki_pos, iov_iter_count(to)); |
d7e09d03 | 1246 | cl_env_put(env, &refcheck); |
0a3bdb00 | 1247 | return result; |
d7e09d03 PT |
1248 | } |
1249 | ||
1250 | /* | |
1251 | * Write to a file (through the page cache). | |
1252 | */ | |
b42b15fd | 1253 | static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
d7e09d03 PT |
1254 | { |
1255 | struct lu_env *env; | |
1256 | struct vvp_io_args *args; | |
d7e09d03 PT |
1257 | ssize_t result; |
1258 | int refcheck; | |
d7e09d03 | 1259 | |
d7e09d03 PT |
1260 | env = cl_env_get(&refcheck); |
1261 | if (IS_ERR(env)) | |
0a3bdb00 | 1262 | return PTR_ERR(env); |
d7e09d03 | 1263 | |
9989a58e | 1264 | args = ll_env_args(env, IO_NORMAL); |
b42b15fd | 1265 | args->u.normal.via_iter = from; |
d7e09d03 PT |
1266 | args->u.normal.via_iocb = iocb; |
1267 | ||
1268 | result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE, | |
e15ba45d | 1269 | &iocb->ki_pos, iov_iter_count(from)); |
d7e09d03 | 1270 | cl_env_put(env, &refcheck); |
0a3bdb00 | 1271 | return result; |
d7e09d03 PT |
1272 | } |
1273 | ||
d7e09d03 PT |
1274 | /* |
1275 | * Send file content (through pagecache) somewhere with helper | |
1276 | */ | |
1277 | static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos, | |
1278 | struct pipe_inode_info *pipe, size_t count, | |
1279 | unsigned int flags) | |
1280 | { | |
1281 | struct lu_env *env; | |
1282 | struct vvp_io_args *args; | |
1283 | ssize_t result; | |
1284 | int refcheck; | |
d7e09d03 PT |
1285 | |
1286 | env = cl_env_get(&refcheck); | |
1287 | if (IS_ERR(env)) | |
0a3bdb00 | 1288 | return PTR_ERR(env); |
d7e09d03 | 1289 | |
9989a58e | 1290 | args = ll_env_args(env, IO_SPLICE); |
d7e09d03 PT |
1291 | args->u.splice.via_pipe = pipe; |
1292 | args->u.splice.via_flags = flags; | |
1293 | ||
1294 | result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count); | |
1295 | cl_env_put(env, &refcheck); | |
0a3bdb00 | 1296 | return result; |
d7e09d03 PT |
1297 | } |
1298 | ||
21aef7d9 | 1299 | static int ll_lov_recreate(struct inode *inode, struct ost_id *oi, u32 ost_idx) |
d7e09d03 PT |
1300 | { |
1301 | struct obd_export *exp = ll_i2dtexp(inode); | |
1302 | struct obd_trans_info oti = { 0 }; | |
1303 | struct obdo *oa = NULL; | |
1304 | int lsm_size; | |
1305 | int rc = 0; | |
1306 | struct lov_stripe_md *lsm = NULL, *lsm2; | |
d7e09d03 | 1307 | |
21068c46 | 1308 | oa = kmem_cache_zalloc(obdo_cachep, GFP_NOFS); |
6e16818b | 1309 | if (!oa) |
0a3bdb00 | 1310 | return -ENOMEM; |
d7e09d03 PT |
1311 | |
1312 | lsm = ccc_inode_lsm_get(inode); | |
34e1f2bb JL |
1313 | if (!lsm_has_objects(lsm)) { |
1314 | rc = -ENOENT; | |
1315 | goto out; | |
1316 | } | |
d7e09d03 PT |
1317 | |
1318 | lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) * | |
1319 | (lsm->lsm_stripe_count)); | |
1320 | ||
e958f49b | 1321 | lsm2 = libcfs_kvzalloc(lsm_size, GFP_NOFS); |
6e16818b | 1322 | if (!lsm2) { |
34e1f2bb JL |
1323 | rc = -ENOMEM; |
1324 | goto out; | |
1325 | } | |
d7e09d03 PT |
1326 | |
1327 | oa->o_oi = *oi; | |
1328 | oa->o_nlink = ost_idx; | |
1329 | oa->o_flags |= OBD_FL_RECREATE_OBJS; | |
1330 | oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP; | |
1331 | obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME | | |
1332 | OBD_MD_FLMTIME | OBD_MD_FLCTIME); | |
1333 | obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid); | |
1334 | memcpy(lsm2, lsm, lsm_size); | |
1335 | ll_inode_size_lock(inode); | |
1336 | rc = obd_create(NULL, exp, oa, &lsm2, &oti); | |
1337 | ll_inode_size_unlock(inode); | |
1338 | ||
e958f49b | 1339 | kvfree(lsm2); |
34e1f2bb | 1340 | goto out; |
d7e09d03 PT |
1341 | out: |
1342 | ccc_inode_lsm_put(inode, lsm); | |
2ba262fb | 1343 | kmem_cache_free(obdo_cachep, oa); |
d7e09d03 PT |
1344 | return rc; |
1345 | } | |
1346 | ||
1347 | static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg) | |
1348 | { | |
1349 | struct ll_recreate_obj ucreat; | |
1350 | struct ost_id oi; | |
d7e09d03 | 1351 | |
2eb90a75 | 1352 | if (!capable(CFS_CAP_SYS_ADMIN)) |
0a3bdb00 | 1353 | return -EPERM; |
d7e09d03 | 1354 | |
02f9c12e | 1355 | if (copy_from_user(&ucreat, (struct ll_recreate_obj __user *)arg, |
d7e09d03 | 1356 | sizeof(ucreat))) |
0a3bdb00 | 1357 | return -EFAULT; |
d7e09d03 PT |
1358 | |
1359 | ostid_set_seq_mdt0(&oi); | |
1360 | ostid_set_id(&oi, ucreat.lrc_id); | |
0a3bdb00 | 1361 | return ll_lov_recreate(inode, &oi, ucreat.lrc_ost_idx); |
d7e09d03 PT |
1362 | } |
1363 | ||
1364 | static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg) | |
1365 | { | |
1366 | struct lu_fid fid; | |
1367 | struct ost_id oi; | |
21aef7d9 | 1368 | u32 ost_idx; |
d7e09d03 | 1369 | |
2eb90a75 | 1370 | if (!capable(CFS_CAP_SYS_ADMIN)) |
0a3bdb00 | 1371 | return -EPERM; |
d7e09d03 | 1372 | |
02f9c12e | 1373 | if (copy_from_user(&fid, (struct lu_fid __user *)arg, sizeof(fid))) |
0a3bdb00 | 1374 | return -EFAULT; |
d7e09d03 PT |
1375 | |
1376 | fid_to_ostid(&fid, &oi); | |
1377 | ost_idx = (fid_seq(&fid) >> 16) & 0xffff; | |
0a3bdb00 | 1378 | return ll_lov_recreate(inode, &oi, ost_idx); |
d7e09d03 PT |
1379 | } |
1380 | ||
c139f3ce | 1381 | int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry, |
d467220e NY |
1382 | __u64 flags, struct lov_user_md *lum, |
1383 | int lum_size) | |
d7e09d03 PT |
1384 | { |
1385 | struct lov_stripe_md *lsm = NULL; | |
1386 | struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags}; | |
1387 | int rc = 0; | |
d7e09d03 PT |
1388 | |
1389 | lsm = ccc_inode_lsm_get(inode); | |
6e16818b | 1390 | if (lsm) { |
d7e09d03 | 1391 | ccc_inode_lsm_put(inode, lsm); |
97a075cd JN |
1392 | CDEBUG(D_IOCTL, "stripe already exists for inode "DFID"\n", |
1393 | PFID(ll_inode2fid(inode))); | |
34e1f2bb JL |
1394 | rc = -EEXIST; |
1395 | goto out; | |
d7e09d03 PT |
1396 | } |
1397 | ||
1398 | ll_inode_size_lock(inode); | |
c139f3ce | 1399 | rc = ll_intent_file_open(dentry, lum, lum_size, &oit); |
d7e09d03 | 1400 | if (rc) |
34e1f2bb | 1401 | goto out_unlock; |
e476f2e5 | 1402 | rc = oit.it_status; |
d7e09d03 | 1403 | if (rc < 0) |
34e1f2bb | 1404 | goto out_req_free; |
d7e09d03 | 1405 | |
e22fdcc8 | 1406 | ll_release_openhandle(inode, &oit); |
d7e09d03 | 1407 | |
38585ccc | 1408 | out_unlock: |
d7e09d03 PT |
1409 | ll_inode_size_unlock(inode); |
1410 | ll_intent_release(&oit); | |
1411 | ccc_inode_lsm_put(inode, lsm); | |
38585ccc | 1412 | out: |
0a3bdb00 | 1413 | return rc; |
d7e09d03 | 1414 | out_req_free: |
8bf86fd9 | 1415 | ptlrpc_req_finished((struct ptlrpc_request *)oit.it_request); |
d7e09d03 PT |
1416 | goto out; |
1417 | } | |
1418 | ||
1419 | int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename, | |
1420 | struct lov_mds_md **lmmp, int *lmm_size, | |
1421 | struct ptlrpc_request **request) | |
1422 | { | |
1423 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1424 | struct mdt_body *body; | |
1425 | struct lov_mds_md *lmm = NULL; | |
1426 | struct ptlrpc_request *req = NULL; | |
1427 | struct md_op_data *op_data; | |
1428 | int rc, lmmsize; | |
1429 | ||
44779340 | 1430 | rc = ll_get_default_mdsize(sbi, &lmmsize); |
d7e09d03 | 1431 | if (rc) |
0a3bdb00 | 1432 | return rc; |
d7e09d03 PT |
1433 | |
1434 | op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, | |
1435 | strlen(filename), lmmsize, | |
1436 | LUSTRE_OPC_ANY, NULL); | |
1437 | if (IS_ERR(op_data)) | |
0a3bdb00 | 1438 | return PTR_ERR(op_data); |
d7e09d03 PT |
1439 | |
1440 | op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA; | |
1441 | rc = md_getattr_name(sbi->ll_md_exp, op_data, &req); | |
1442 | ll_finish_md_op_data(op_data); | |
1443 | if (rc < 0) { | |
2d00bd17 JP |
1444 | CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n", |
1445 | filename, rc); | |
34e1f2bb | 1446 | goto out; |
d7e09d03 PT |
1447 | } |
1448 | ||
1449 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); | |
d7e09d03 PT |
1450 | |
1451 | lmmsize = body->eadatasize; | |
1452 | ||
1453 | if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) || | |
e15ba45d | 1454 | lmmsize == 0) { |
34e1f2bb JL |
1455 | rc = -ENODATA; |
1456 | goto out; | |
d7e09d03 PT |
1457 | } |
1458 | ||
1459 | lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize); | |
d7e09d03 PT |
1460 | |
1461 | if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) && | |
1462 | (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) { | |
34e1f2bb JL |
1463 | rc = -EPROTO; |
1464 | goto out; | |
d7e09d03 PT |
1465 | } |
1466 | ||
1467 | /* | |
1468 | * This is coming from the MDS, so is probably in | |
1469 | * little endian. We convert it to host endian before | |
1470 | * passing it to userspace. | |
1471 | */ | |
1f6eaf83 | 1472 | if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) { |
5dd16419 JX |
1473 | int stripe_count; |
1474 | ||
1475 | stripe_count = le16_to_cpu(lmm->lmm_stripe_count); | |
1476 | if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED) | |
1477 | stripe_count = 0; | |
1478 | ||
d7e09d03 | 1479 | /* if function called for directory - we should |
c0894c6c OD |
1480 | * avoid swab not existent lsm objects |
1481 | */ | |
d7e09d03 PT |
1482 | if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) { |
1483 | lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm); | |
1484 | if (S_ISREG(body->mode)) | |
1485 | lustre_swab_lov_user_md_objects( | |
1486 | ((struct lov_user_md_v1 *)lmm)->lmm_objects, | |
5dd16419 | 1487 | stripe_count); |
d7e09d03 PT |
1488 | } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) { |
1489 | lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm); | |
1490 | if (S_ISREG(body->mode)) | |
1491 | lustre_swab_lov_user_md_objects( | |
1492 | ((struct lov_user_md_v3 *)lmm)->lmm_objects, | |
5dd16419 | 1493 | stripe_count); |
d7e09d03 PT |
1494 | } |
1495 | } | |
1496 | ||
1497 | out: | |
1498 | *lmmp = lmm; | |
1499 | *lmm_size = lmmsize; | |
1500 | *request = req; | |
1501 | return rc; | |
1502 | } | |
1503 | ||
1504 | static int ll_lov_setea(struct inode *inode, struct file *file, | |
e15ba45d | 1505 | unsigned long arg) |
d7e09d03 | 1506 | { |
d467220e | 1507 | __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE; |
d7e09d03 PT |
1508 | struct lov_user_md *lump; |
1509 | int lum_size = sizeof(struct lov_user_md) + | |
1510 | sizeof(struct lov_user_ost_data); | |
1511 | int rc; | |
d7e09d03 | 1512 | |
2eb90a75 | 1513 | if (!capable(CFS_CAP_SYS_ADMIN)) |
0a3bdb00 | 1514 | return -EPERM; |
d7e09d03 | 1515 | |
e958f49b | 1516 | lump = libcfs_kvzalloc(lum_size, GFP_NOFS); |
6e16818b | 1517 | if (!lump) |
0a3bdb00 | 1518 | return -ENOMEM; |
d7e09d03 | 1519 | |
02f9c12e | 1520 | if (copy_from_user(lump, (struct lov_user_md __user *)arg, lum_size)) { |
e958f49b | 1521 | kvfree(lump); |
0a3bdb00 | 1522 | return -EFAULT; |
d7e09d03 PT |
1523 | } |
1524 | ||
c139f3ce | 1525 | rc = ll_lov_setstripe_ea_info(inode, file->f_path.dentry, flags, lump, |
e15ba45d | 1526 | lum_size); |
c139f3ce | 1527 | cl_lov_delay_create_clear(&file->f_flags); |
d7e09d03 | 1528 | |
e958f49b | 1529 | kvfree(lump); |
0a3bdb00 | 1530 | return rc; |
d7e09d03 PT |
1531 | } |
1532 | ||
1533 | static int ll_lov_setstripe(struct inode *inode, struct file *file, | |
1534 | unsigned long arg) | |
1535 | { | |
02f9c12e OD |
1536 | struct lov_user_md_v3 lumv3; |
1537 | struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3; | |
1538 | struct lov_user_md_v1 __user *lumv1p = (void __user *)arg; | |
1539 | struct lov_user_md_v3 __user *lumv3p = (void __user *)arg; | |
1540 | int lum_size, rc; | |
d467220e | 1541 | __u64 flags = FMODE_WRITE; |
d7e09d03 PT |
1542 | |
1543 | /* first try with v1 which is smaller than v3 */ | |
1544 | lum_size = sizeof(struct lov_user_md_v1); | |
1545 | if (copy_from_user(lumv1, lumv1p, lum_size)) | |
0a3bdb00 | 1546 | return -EFAULT; |
d7e09d03 PT |
1547 | |
1548 | if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) { | |
1549 | lum_size = sizeof(struct lov_user_md_v3); | |
1550 | if (copy_from_user(&lumv3, lumv3p, lum_size)) | |
0a3bdb00 | 1551 | return -EFAULT; |
d7e09d03 PT |
1552 | } |
1553 | ||
c139f3ce AV |
1554 | rc = ll_lov_setstripe_ea_info(inode, file->f_path.dentry, flags, lumv1, |
1555 | lum_size); | |
1556 | cl_lov_delay_create_clear(&file->f_flags); | |
d7e09d03 PT |
1557 | if (rc == 0) { |
1558 | struct lov_stripe_md *lsm; | |
1559 | __u32 gen; | |
1560 | ||
1561 | put_user(0, &lumv1p->lmm_stripe_count); | |
1562 | ||
1563 | ll_layout_refresh(inode, &gen); | |
1564 | lsm = ccc_inode_lsm_get(inode); | |
1565 | rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), | |
e09bee34 | 1566 | 0, lsm, (void __user *)arg); |
d7e09d03 PT |
1567 | ccc_inode_lsm_put(inode, lsm); |
1568 | } | |
0a3bdb00 | 1569 | return rc; |
d7e09d03 PT |
1570 | } |
1571 | ||
1572 | static int ll_lov_getstripe(struct inode *inode, unsigned long arg) | |
1573 | { | |
1574 | struct lov_stripe_md *lsm; | |
1575 | int rc = -ENODATA; | |
d7e09d03 PT |
1576 | |
1577 | lsm = ccc_inode_lsm_get(inode); | |
6e16818b | 1578 | if (lsm) |
d7e09d03 | 1579 | rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0, |
e09bee34 | 1580 | lsm, (void __user *)arg); |
d7e09d03 | 1581 | ccc_inode_lsm_put(inode, lsm); |
0a3bdb00 | 1582 | return rc; |
d7e09d03 PT |
1583 | } |
1584 | ||
2d95f10e JH |
1585 | static int |
1586 | ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg) | |
d7e09d03 PT |
1587 | { |
1588 | struct ll_inode_info *lli = ll_i2info(inode); | |
1589 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
98eae5e7 | 1590 | struct ll_grouplock grouplock; |
d7e09d03 | 1591 | int rc; |
d7e09d03 | 1592 | |
431b5678 PF |
1593 | if (arg == 0) { |
1594 | CWARN("group id for group lock must not be 0\n"); | |
1595 | return -EINVAL; | |
1596 | } | |
1597 | ||
d7e09d03 | 1598 | if (ll_file_nolock(file)) |
0a3bdb00 | 1599 | return -EOPNOTSUPP; |
d7e09d03 PT |
1600 | |
1601 | spin_lock(&lli->lli_lock); | |
1602 | if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { | |
1603 | CWARN("group lock already existed with gid %lu\n", | |
98eae5e7 | 1604 | fd->fd_grouplock.lg_gid); |
d7e09d03 | 1605 | spin_unlock(&lli->lli_lock); |
0a3bdb00 | 1606 | return -EINVAL; |
d7e09d03 | 1607 | } |
98eae5e7 | 1608 | LASSERT(!fd->fd_grouplock.lg_lock); |
d7e09d03 PT |
1609 | spin_unlock(&lli->lli_lock); |
1610 | ||
1929c433 | 1611 | rc = cl_get_grouplock(ll_i2info(inode)->lli_clob, |
d7e09d03 PT |
1612 | arg, (file->f_flags & O_NONBLOCK), &grouplock); |
1613 | if (rc) | |
0a3bdb00 | 1614 | return rc; |
d7e09d03 PT |
1615 | |
1616 | spin_lock(&lli->lli_lock); | |
1617 | if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { | |
1618 | spin_unlock(&lli->lli_lock); | |
1619 | CERROR("another thread just won the race\n"); | |
1620 | cl_put_grouplock(&grouplock); | |
0a3bdb00 | 1621 | return -EINVAL; |
d7e09d03 PT |
1622 | } |
1623 | ||
1624 | fd->fd_flags |= LL_FILE_GROUP_LOCKED; | |
1625 | fd->fd_grouplock = grouplock; | |
1626 | spin_unlock(&lli->lli_lock); | |
1627 | ||
1628 | CDEBUG(D_INFO, "group lock %lu obtained\n", arg); | |
0a3bdb00 | 1629 | return 0; |
d7e09d03 PT |
1630 | } |
1631 | ||
920b4f2e LC |
1632 | static int ll_put_grouplock(struct inode *inode, struct file *file, |
1633 | unsigned long arg) | |
d7e09d03 PT |
1634 | { |
1635 | struct ll_inode_info *lli = ll_i2info(inode); | |
1636 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
98eae5e7 | 1637 | struct ll_grouplock grouplock; |
d7e09d03 PT |
1638 | |
1639 | spin_lock(&lli->lli_lock); | |
1640 | if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) { | |
1641 | spin_unlock(&lli->lli_lock); | |
1642 | CWARN("no group lock held\n"); | |
0a3bdb00 | 1643 | return -EINVAL; |
d7e09d03 | 1644 | } |
98eae5e7 | 1645 | LASSERT(fd->fd_grouplock.lg_lock); |
d7e09d03 | 1646 | |
98eae5e7 | 1647 | if (fd->fd_grouplock.lg_gid != arg) { |
d7e09d03 | 1648 | CWARN("group lock %lu doesn't match current id %lu\n", |
98eae5e7 | 1649 | arg, fd->fd_grouplock.lg_gid); |
d7e09d03 | 1650 | spin_unlock(&lli->lli_lock); |
0a3bdb00 | 1651 | return -EINVAL; |
d7e09d03 PT |
1652 | } |
1653 | ||
1654 | grouplock = fd->fd_grouplock; | |
1655 | memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock)); | |
1656 | fd->fd_flags &= ~LL_FILE_GROUP_LOCKED; | |
1657 | spin_unlock(&lli->lli_lock); | |
1658 | ||
1659 | cl_put_grouplock(&grouplock); | |
1660 | CDEBUG(D_INFO, "group lock %lu released\n", arg); | |
0a3bdb00 | 1661 | return 0; |
d7e09d03 PT |
1662 | } |
1663 | ||
1664 | /** | |
1665 | * Close inode open handle | |
1666 | * | |
e22fdcc8 | 1667 | * \param inode [in] inode in question |
d7e09d03 PT |
1668 | * \param it [in,out] intent which contains open info and result |
1669 | * | |
1670 | * \retval 0 success | |
1671 | * \retval <0 failure | |
1672 | */ | |
e22fdcc8 | 1673 | int ll_release_openhandle(struct inode *inode, struct lookup_intent *it) |
d7e09d03 | 1674 | { |
d7e09d03 PT |
1675 | struct obd_client_handle *och; |
1676 | int rc; | |
d7e09d03 PT |
1677 | |
1678 | LASSERT(inode); | |
1679 | ||
1680 | /* Root ? Do nothing. */ | |
f76c23da | 1681 | if (is_root_inode(inode)) |
0a3bdb00 | 1682 | return 0; |
d7e09d03 PT |
1683 | |
1684 | /* No open handle to close? Move away */ | |
1685 | if (!it_disposition(it, DISP_OPEN_OPEN)) | |
0a3bdb00 | 1686 | return 0; |
d7e09d03 PT |
1687 | |
1688 | LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0); | |
1689 | ||
496a51bd | 1690 | och = kzalloc(sizeof(*och), GFP_NOFS); |
34e1f2bb JL |
1691 | if (!och) { |
1692 | rc = -ENOMEM; | |
1693 | goto out; | |
1694 | } | |
d7e09d03 | 1695 | |
ea1db081 | 1696 | ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och); |
d7e09d03 PT |
1697 | |
1698 | rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, | |
48d23e61 JX |
1699 | inode, och, NULL); |
1700 | out: | |
d7e09d03 PT |
1701 | /* this one is in place of ll_file_open */ |
1702 | if (it_disposition(it, DISP_ENQ_OPEN_REF)) { | |
8bf86fd9 | 1703 | ptlrpc_req_finished(it->it_request); |
d7e09d03 PT |
1704 | it_clear_disposition(it, DISP_ENQ_OPEN_REF); |
1705 | } | |
0a3bdb00 | 1706 | return rc; |
d7e09d03 PT |
1707 | } |
1708 | ||
1709 | /** | |
1710 | * Get size for inode for which FIEMAP mapping is requested. | |
1711 | * Make the FIEMAP get_info call and returns the result. | |
1712 | */ | |
2d95f10e | 1713 | static int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap, |
ebdc4fc5 | 1714 | size_t num_bytes) |
d7e09d03 PT |
1715 | { |
1716 | struct obd_export *exp = ll_i2dtexp(inode); | |
1717 | struct lov_stripe_md *lsm = NULL; | |
1718 | struct ll_fiemap_info_key fm_key = { .name = KEY_FIEMAP, }; | |
ebdc4fc5 | 1719 | __u32 vallen = num_bytes; |
d7e09d03 | 1720 | int rc; |
d7e09d03 PT |
1721 | |
1722 | /* Checks for fiemap flags */ | |
1723 | if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) { | |
1724 | fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT; | |
1725 | return -EBADR; | |
1726 | } | |
1727 | ||
1728 | /* Check for FIEMAP_FLAG_SYNC */ | |
1729 | if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) { | |
1730 | rc = filemap_fdatawrite(inode->i_mapping); | |
1731 | if (rc) | |
1732 | return rc; | |
1733 | } | |
1734 | ||
1735 | lsm = ccc_inode_lsm_get(inode); | |
6e16818b | 1736 | if (!lsm) |
d7e09d03 PT |
1737 | return -ENOENT; |
1738 | ||
1739 | /* If the stripe_count > 1 and the application does not understand | |
1740 | * DEVICE_ORDER flag, then it cannot interpret the extents correctly. | |
1741 | */ | |
1742 | if (lsm->lsm_stripe_count > 1 && | |
34e1f2bb JL |
1743 | !(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER)) { |
1744 | rc = -EOPNOTSUPP; | |
1745 | goto out; | |
1746 | } | |
d7e09d03 PT |
1747 | |
1748 | fm_key.oa.o_oi = lsm->lsm_oi; | |
1749 | fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP; | |
1750 | ||
a915ffda LD |
1751 | if (i_size_read(inode) == 0) { |
1752 | rc = ll_glimpse_size(inode); | |
1753 | if (rc) | |
1754 | goto out; | |
1755 | } | |
1756 | ||
d7e09d03 PT |
1757 | obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLSIZE); |
1758 | obdo_set_parent_fid(&fm_key.oa, &ll_i2info(inode)->lli_fid); | |
1759 | /* If filesize is 0, then there would be no objects for mapping */ | |
1760 | if (fm_key.oa.o_size == 0) { | |
1761 | fiemap->fm_mapped_extents = 0; | |
34e1f2bb JL |
1762 | rc = 0; |
1763 | goto out; | |
d7e09d03 PT |
1764 | } |
1765 | ||
1766 | memcpy(&fm_key.fiemap, fiemap, sizeof(*fiemap)); | |
1767 | ||
1768 | rc = obd_get_info(NULL, exp, sizeof(fm_key), &fm_key, &vallen, | |
1769 | fiemap, lsm); | |
1770 | if (rc) | |
1771 | CERROR("obd_get_info failed: rc = %d\n", rc); | |
1772 | ||
1773 | out: | |
1774 | ccc_inode_lsm_put(inode, lsm); | |
0a3bdb00 | 1775 | return rc; |
d7e09d03 PT |
1776 | } |
1777 | ||
2b358b4e | 1778 | int ll_fid2path(struct inode *inode, void __user *arg) |
d7e09d03 | 1779 | { |
2b358b4e FZ |
1780 | struct obd_export *exp = ll_i2mdexp(inode); |
1781 | const struct getinfo_fid2path __user *gfin = arg; | |
1782 | struct getinfo_fid2path *gfout; | |
1783 | u32 pathlen; | |
1784 | size_t outsize; | |
1785 | int rc; | |
d7e09d03 | 1786 | |
2eb90a75 | 1787 | if (!capable(CFS_CAP_DAC_READ_SEARCH) && |
d7e09d03 | 1788 | !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH)) |
0a3bdb00 | 1789 | return -EPERM; |
d7e09d03 | 1790 | |
2b358b4e FZ |
1791 | /* Only need to get the buflen */ |
1792 | if (get_user(pathlen, &gfin->gf_pathlen)) | |
0a3bdb00 | 1793 | return -EFAULT; |
d7e09d03 | 1794 | |
c7b09efa OD |
1795 | if (pathlen > PATH_MAX) |
1796 | return -EINVAL; | |
1797 | ||
2b358b4e FZ |
1798 | outsize = sizeof(*gfout) + pathlen; |
1799 | ||
496a51bd JL |
1800 | gfout = kzalloc(outsize, GFP_NOFS); |
1801 | if (!gfout) | |
0a3bdb00 | 1802 | return -ENOMEM; |
2b358b4e | 1803 | |
34e1f2bb JL |
1804 | if (copy_from_user(gfout, arg, sizeof(*gfout))) { |
1805 | rc = -EFAULT; | |
1806 | goto gf_free; | |
1807 | } | |
d7e09d03 PT |
1808 | |
1809 | /* Call mdc_iocontrol */ | |
1810 | rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL); | |
2b358b4e | 1811 | if (rc != 0) |
34e1f2bb | 1812 | goto gf_free; |
d7e09d03 PT |
1813 | |
1814 | if (copy_to_user(arg, gfout, outsize)) | |
1815 | rc = -EFAULT; | |
1816 | ||
1817 | gf_free: | |
97903a26 | 1818 | kfree(gfout); |
0a3bdb00 | 1819 | return rc; |
d7e09d03 PT |
1820 | } |
1821 | ||
1822 | static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg) | |
1823 | { | |
1824 | struct ll_user_fiemap *fiemap_s; | |
1825 | size_t num_bytes, ret_bytes; | |
1826 | unsigned int extent_count; | |
1827 | int rc = 0; | |
1828 | ||
1829 | /* Get the extent count so we can calculate the size of | |
c0894c6c OD |
1830 | * required fiemap buffer |
1831 | */ | |
d7e09d03 | 1832 | if (get_user(extent_count, |
e15ba45d | 1833 | &((struct ll_user_fiemap __user *)arg)->fm_extent_count)) |
0a3bdb00 | 1834 | return -EFAULT; |
7bc3dfa3 VO |
1835 | |
1836 | if (extent_count >= | |
1837 | (SIZE_MAX - sizeof(*fiemap_s)) / sizeof(struct ll_fiemap_extent)) | |
1838 | return -EINVAL; | |
d7e09d03 PT |
1839 | num_bytes = sizeof(*fiemap_s) + (extent_count * |
1840 | sizeof(struct ll_fiemap_extent)); | |
1841 | ||
e958f49b | 1842 | fiemap_s = libcfs_kvzalloc(num_bytes, GFP_NOFS); |
6e16818b | 1843 | if (!fiemap_s) |
0a3bdb00 | 1844 | return -ENOMEM; |
d7e09d03 PT |
1845 | |
1846 | /* get the fiemap value */ | |
1847 | if (copy_from_user(fiemap_s, (struct ll_user_fiemap __user *)arg, | |
34e1f2bb JL |
1848 | sizeof(*fiemap_s))) { |
1849 | rc = -EFAULT; | |
1850 | goto error; | |
1851 | } | |
d7e09d03 PT |
1852 | |
1853 | /* If fm_extent_count is non-zero, read the first extent since | |
1854 | * it is used to calculate end_offset and device from previous | |
c0894c6c OD |
1855 | * fiemap call. |
1856 | */ | |
d7e09d03 PT |
1857 | if (extent_count) { |
1858 | if (copy_from_user(&fiemap_s->fm_extents[0], | |
e15ba45d OD |
1859 | (char __user *)arg + sizeof(*fiemap_s), |
1860 | sizeof(struct ll_fiemap_extent))) { | |
34e1f2bb JL |
1861 | rc = -EFAULT; |
1862 | goto error; | |
1863 | } | |
d7e09d03 PT |
1864 | } |
1865 | ||
1866 | rc = ll_do_fiemap(inode, fiemap_s, num_bytes); | |
1867 | if (rc) | |
34e1f2bb | 1868 | goto error; |
d7e09d03 PT |
1869 | |
1870 | ret_bytes = sizeof(struct ll_user_fiemap); | |
1871 | ||
1872 | if (extent_count != 0) | |
1873 | ret_bytes += (fiemap_s->fm_mapped_extents * | |
1874 | sizeof(struct ll_fiemap_extent)); | |
1875 | ||
02f9c12e | 1876 | if (copy_to_user((void __user *)arg, fiemap_s, ret_bytes)) |
d7e09d03 PT |
1877 | rc = -EFAULT; |
1878 | ||
1879 | error: | |
e958f49b | 1880 | kvfree(fiemap_s); |
0a3bdb00 | 1881 | return rc; |
d7e09d03 PT |
1882 | } |
1883 | ||
1884 | /* | |
1885 | * Read the data_version for inode. | |
1886 | * | |
1887 | * This value is computed using stripe object version on OST. | |
1888 | * Version is computed using server side locking. | |
1889 | * | |
e1798006 JX |
1890 | * @param sync if do sync on the OST side; |
1891 | * 0: no sync | |
1892 | * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs | |
1893 | * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs | |
d7e09d03 | 1894 | */ |
e1798006 | 1895 | int ll_data_version(struct inode *inode, __u64 *data_version, int flags) |
d7e09d03 PT |
1896 | { |
1897 | struct lov_stripe_md *lsm = NULL; | |
1898 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1899 | struct obdo *obdo = NULL; | |
1900 | int rc; | |
d7e09d03 PT |
1901 | |
1902 | /* If no stripe, we consider version is 0. */ | |
1903 | lsm = ccc_inode_lsm_get(inode); | |
5dd16419 | 1904 | if (!lsm_has_objects(lsm)) { |
d7e09d03 PT |
1905 | *data_version = 0; |
1906 | CDEBUG(D_INODE, "No object for inode\n"); | |
34e1f2bb JL |
1907 | rc = 0; |
1908 | goto out; | |
d7e09d03 PT |
1909 | } |
1910 | ||
496a51bd JL |
1911 | obdo = kzalloc(sizeof(*obdo), GFP_NOFS); |
1912 | if (!obdo) { | |
34e1f2bb JL |
1913 | rc = -ENOMEM; |
1914 | goto out; | |
1915 | } | |
d7e09d03 | 1916 | |
e1798006 | 1917 | rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, obdo, 0, flags); |
5dd16419 | 1918 | if (rc == 0) { |
d7e09d03 PT |
1919 | if (!(obdo->o_valid & OBD_MD_FLDATAVERSION)) |
1920 | rc = -EOPNOTSUPP; | |
1921 | else | |
1922 | *data_version = obdo->o_data_version; | |
1923 | } | |
1924 | ||
97903a26 | 1925 | kfree(obdo); |
5dd16419 | 1926 | out: |
d7e09d03 | 1927 | ccc_inode_lsm_put(inode, lsm); |
0a3bdb00 | 1928 | return rc; |
d7e09d03 PT |
1929 | } |
1930 | ||
48d23e61 JX |
1931 | /* |
1932 | * Trigger a HSM release request for the provided inode. | |
1933 | */ | |
1934 | int ll_hsm_release(struct inode *inode) | |
1935 | { | |
1936 | struct cl_env_nest nest; | |
1937 | struct lu_env *env; | |
1938 | struct obd_client_handle *och = NULL; | |
1939 | __u64 data_version = 0; | |
1940 | int rc; | |
1941 | ||
48d23e61 JX |
1942 | CDEBUG(D_INODE, "%s: Releasing file "DFID".\n", |
1943 | ll_get_fsname(inode->i_sb, NULL, 0), | |
1944 | PFID(&ll_i2info(inode)->lli_fid)); | |
1945 | ||
1946 | och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE); | |
34e1f2bb JL |
1947 | if (IS_ERR(och)) { |
1948 | rc = PTR_ERR(och); | |
1949 | goto out; | |
1950 | } | |
48d23e61 JX |
1951 | |
1952 | /* Grab latest data_version and [am]time values */ | |
e1798006 | 1953 | rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH); |
48d23e61 | 1954 | if (rc != 0) |
34e1f2bb | 1955 | goto out; |
48d23e61 JX |
1956 | |
1957 | env = cl_env_nested_get(&nest); | |
34e1f2bb JL |
1958 | if (IS_ERR(env)) { |
1959 | rc = PTR_ERR(env); | |
1960 | goto out; | |
1961 | } | |
48d23e61 | 1962 | |
d2995737 | 1963 | ll_merge_attr(env, inode); |
48d23e61 JX |
1964 | cl_env_nested_put(&nest, env); |
1965 | ||
1966 | /* Release the file. | |
1967 | * NB: lease lock handle is released in mdc_hsm_release_pack() because | |
c0894c6c OD |
1968 | * we still need it to pack l_remote_handle to MDT. |
1969 | */ | |
48d23e61 JX |
1970 | rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, inode, och, |
1971 | &data_version); | |
1972 | och = NULL; | |
1973 | ||
48d23e61 | 1974 | out: |
6e16818b | 1975 | if (och && !IS_ERR(och)) /* close the file */ |
48d23e61 JX |
1976 | ll_lease_close(och, inode, NULL); |
1977 | ||
1978 | return rc; | |
1979 | } | |
1980 | ||
d7e09d03 PT |
1981 | struct ll_swap_stack { |
1982 | struct iattr ia1, ia2; | |
1983 | __u64 dv1, dv2; | |
1984 | struct inode *inode1, *inode2; | |
1985 | bool check_dv1, check_dv2; | |
1986 | }; | |
1987 | ||
1988 | static int ll_swap_layouts(struct file *file1, struct file *file2, | |
1989 | struct lustre_swap_layouts *lsl) | |
1990 | { | |
1991 | struct mdc_swap_layouts msl; | |
1992 | struct md_op_data *op_data; | |
1993 | __u32 gid; | |
1994 | __u64 dv; | |
1995 | struct ll_swap_stack *llss = NULL; | |
1996 | int rc; | |
1997 | ||
496a51bd JL |
1998 | llss = kzalloc(sizeof(*llss), GFP_NOFS); |
1999 | if (!llss) | |
0a3bdb00 | 2000 | return -ENOMEM; |
d7e09d03 | 2001 | |
2a8a3597 AV |
2002 | llss->inode1 = file_inode(file1); |
2003 | llss->inode2 = file_inode(file2); | |
d7e09d03 | 2004 | |
34e1f2bb JL |
2005 | if (!S_ISREG(llss->inode2->i_mode)) { |
2006 | rc = -EINVAL; | |
2007 | goto free; | |
2008 | } | |
d7e09d03 | 2009 | |
9c5fb72c | 2010 | if (inode_permission(llss->inode1, MAY_WRITE) || |
34e1f2bb JL |
2011 | inode_permission(llss->inode2, MAY_WRITE)) { |
2012 | rc = -EPERM; | |
2013 | goto free; | |
2014 | } | |
d7e09d03 | 2015 | |
34e1f2bb JL |
2016 | if (llss->inode2->i_sb != llss->inode1->i_sb) { |
2017 | rc = -EXDEV; | |
2018 | goto free; | |
2019 | } | |
d7e09d03 PT |
2020 | |
2021 | /* we use 2 bool because it is easier to swap than 2 bits */ | |
2022 | if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1) | |
2023 | llss->check_dv1 = true; | |
2024 | ||
2025 | if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2) | |
2026 | llss->check_dv2 = true; | |
2027 | ||
2028 | /* we cannot use lsl->sl_dvX directly because we may swap them */ | |
2029 | llss->dv1 = lsl->sl_dv1; | |
2030 | llss->dv2 = lsl->sl_dv2; | |
2031 | ||
2032 | rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2)); | |
34e1f2bb JL |
2033 | if (rc == 0) /* same file, done! */ { |
2034 | rc = 0; | |
2035 | goto free; | |
2036 | } | |
d7e09d03 PT |
2037 | |
2038 | if (rc < 0) { /* sequentialize it */ | |
2039 | swap(llss->inode1, llss->inode2); | |
2040 | swap(file1, file2); | |
2041 | swap(llss->dv1, llss->dv2); | |
2042 | swap(llss->check_dv1, llss->check_dv2); | |
2043 | } | |
2044 | ||
2045 | gid = lsl->sl_gid; | |
2046 | if (gid != 0) { /* application asks to flush dirty cache */ | |
2047 | rc = ll_get_grouplock(llss->inode1, file1, gid); | |
2048 | if (rc < 0) | |
34e1f2bb | 2049 | goto free; |
d7e09d03 PT |
2050 | |
2051 | rc = ll_get_grouplock(llss->inode2, file2, gid); | |
2052 | if (rc < 0) { | |
2053 | ll_put_grouplock(llss->inode1, file1, gid); | |
34e1f2bb | 2054 | goto free; |
d7e09d03 PT |
2055 | } |
2056 | } | |
2057 | ||
2058 | /* to be able to restore mtime and atime after swap | |
c0894c6c OD |
2059 | * we need to first save them |
2060 | */ | |
d7e09d03 PT |
2061 | if (lsl->sl_flags & |
2062 | (SWAP_LAYOUTS_KEEP_MTIME | SWAP_LAYOUTS_KEEP_ATIME)) { | |
2063 | llss->ia1.ia_mtime = llss->inode1->i_mtime; | |
2064 | llss->ia1.ia_atime = llss->inode1->i_atime; | |
2065 | llss->ia1.ia_valid = ATTR_MTIME | ATTR_ATIME; | |
2066 | llss->ia2.ia_mtime = llss->inode2->i_mtime; | |
2067 | llss->ia2.ia_atime = llss->inode2->i_atime; | |
2068 | llss->ia2.ia_valid = ATTR_MTIME | ATTR_ATIME; | |
2069 | } | |
2070 | ||
d0a0acc3 | 2071 | /* ultimate check, before swapping the layouts we check if |
c0894c6c OD |
2072 | * dataversion has changed (if requested) |
2073 | */ | |
d7e09d03 PT |
2074 | if (llss->check_dv1) { |
2075 | rc = ll_data_version(llss->inode1, &dv, 0); | |
2076 | if (rc) | |
34e1f2bb JL |
2077 | goto putgl; |
2078 | if (dv != llss->dv1) { | |
2079 | rc = -EAGAIN; | |
2080 | goto putgl; | |
2081 | } | |
d7e09d03 PT |
2082 | } |
2083 | ||
2084 | if (llss->check_dv2) { | |
2085 | rc = ll_data_version(llss->inode2, &dv, 0); | |
2086 | if (rc) | |
34e1f2bb JL |
2087 | goto putgl; |
2088 | if (dv != llss->dv2) { | |
2089 | rc = -EAGAIN; | |
2090 | goto putgl; | |
2091 | } | |
d7e09d03 PT |
2092 | } |
2093 | ||
2094 | /* struct md_op_data is used to send the swap args to the mdt | |
2095 | * only flags is missing, so we use struct mdc_swap_layouts | |
c0894c6c OD |
2096 | * through the md_op_data->op_data |
2097 | */ | |
d7e09d03 | 2098 | /* flags from user space have to be converted before they are send to |
c0894c6c OD |
2099 | * server, no flag is sent today, they are only used on the client |
2100 | */ | |
d7e09d03 PT |
2101 | msl.msl_flags = 0; |
2102 | rc = -ENOMEM; | |
2103 | op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0, | |
2104 | 0, LUSTRE_OPC_ANY, &msl); | |
34e1f2bb JL |
2105 | if (IS_ERR(op_data)) { |
2106 | rc = PTR_ERR(op_data); | |
2107 | goto free; | |
2108 | } | |
79a8726a JH |
2109 | |
2110 | rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1), | |
2111 | sizeof(*op_data), op_data, NULL); | |
2112 | ll_finish_md_op_data(op_data); | |
d7e09d03 PT |
2113 | |
2114 | putgl: | |
2115 | if (gid != 0) { | |
2116 | ll_put_grouplock(llss->inode2, file2, gid); | |
2117 | ll_put_grouplock(llss->inode1, file1, gid); | |
2118 | } | |
2119 | ||
2120 | /* rc can be set from obd_iocontrol() or from a GOTO(putgl, ...) */ | |
2121 | if (rc != 0) | |
34e1f2bb | 2122 | goto free; |
d7e09d03 PT |
2123 | |
2124 | /* clear useless flags */ | |
2125 | if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_MTIME)) { | |
2126 | llss->ia1.ia_valid &= ~ATTR_MTIME; | |
2127 | llss->ia2.ia_valid &= ~ATTR_MTIME; | |
2128 | } | |
2129 | ||
2130 | if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_ATIME)) { | |
2131 | llss->ia1.ia_valid &= ~ATTR_ATIME; | |
2132 | llss->ia2.ia_valid &= ~ATTR_ATIME; | |
2133 | } | |
2134 | ||
2135 | /* update time if requested */ | |
2136 | rc = 0; | |
2137 | if (llss->ia2.ia_valid != 0) { | |
5955102c | 2138 | inode_lock(llss->inode1); |
b583043e | 2139 | rc = ll_setattr(file1->f_path.dentry, &llss->ia2); |
5955102c | 2140 | inode_unlock(llss->inode1); |
d7e09d03 PT |
2141 | } |
2142 | ||
2143 | if (llss->ia1.ia_valid != 0) { | |
2144 | int rc1; | |
2145 | ||
5955102c | 2146 | inode_lock(llss->inode2); |
b583043e | 2147 | rc1 = ll_setattr(file2->f_path.dentry, &llss->ia1); |
5955102c | 2148 | inode_unlock(llss->inode2); |
d7e09d03 PT |
2149 | if (rc == 0) |
2150 | rc = rc1; | |
2151 | } | |
2152 | ||
2153 | free: | |
e6b9a3b2 | 2154 | kfree(llss); |
d7e09d03 | 2155 | |
0a3bdb00 | 2156 | return rc; |
d7e09d03 PT |
2157 | } |
2158 | ||
a720b790 JL |
2159 | static int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss) |
2160 | { | |
2161 | struct md_op_data *op_data; | |
2162 | int rc; | |
2163 | ||
2742c75e BF |
2164 | /* Detect out-of range masks */ |
2165 | if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK) | |
2166 | return -EINVAL; | |
2167 | ||
a720b790 | 2168 | /* Non-root users are forbidden to set or clear flags which are |
c0894c6c OD |
2169 | * NOT defined in HSM_USER_MASK. |
2170 | */ | |
a720b790 | 2171 | if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) && |
2eb90a75 | 2172 | !capable(CFS_CAP_SYS_ADMIN)) |
a720b790 JL |
2173 | return -EPERM; |
2174 | ||
2742c75e BF |
2175 | /* Detect out-of range archive id */ |
2176 | if ((hss->hss_valid & HSS_ARCHIVE_ID) && | |
2177 | (hss->hss_archive_id > LL_HSM_MAX_ARCHIVE)) | |
2178 | return -EINVAL; | |
2179 | ||
a720b790 JL |
2180 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, |
2181 | LUSTRE_OPC_ANY, hss); | |
2182 | if (IS_ERR(op_data)) | |
2183 | return PTR_ERR(op_data); | |
2184 | ||
2185 | rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, ll_i2mdexp(inode), | |
2186 | sizeof(*op_data), op_data, NULL); | |
2187 | ||
2188 | ll_finish_md_op_data(op_data); | |
2189 | ||
2190 | return rc; | |
2191 | } | |
2192 | ||
2193 | static int ll_hsm_import(struct inode *inode, struct file *file, | |
2194 | struct hsm_user_import *hui) | |
2195 | { | |
2196 | struct hsm_state_set *hss = NULL; | |
2197 | struct iattr *attr = NULL; | |
2198 | int rc; | |
2199 | ||
a720b790 JL |
2200 | if (!S_ISREG(inode->i_mode)) |
2201 | return -EINVAL; | |
2202 | ||
2203 | /* set HSM flags */ | |
496a51bd | 2204 | hss = kzalloc(sizeof(*hss), GFP_NOFS); |
e6b9a3b2 JL |
2205 | if (!hss) |
2206 | return -ENOMEM; | |
a720b790 JL |
2207 | |
2208 | hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID; | |
2209 | hss->hss_archive_id = hui->hui_archive_id; | |
2210 | hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED; | |
2211 | rc = ll_hsm_state_set(inode, hss); | |
2212 | if (rc != 0) | |
e6b9a3b2 | 2213 | goto free_hss; |
a720b790 | 2214 | |
496a51bd JL |
2215 | attr = kzalloc(sizeof(*attr), GFP_NOFS); |
2216 | if (!attr) { | |
34e1f2bb | 2217 | rc = -ENOMEM; |
e6b9a3b2 | 2218 | goto free_hss; |
34e1f2bb | 2219 | } |
a720b790 JL |
2220 | |
2221 | attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO); | |
2222 | attr->ia_mode |= S_IFREG; | |
2223 | attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid); | |
2224 | attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid); | |
2225 | attr->ia_size = hui->hui_size; | |
2226 | attr->ia_mtime.tv_sec = hui->hui_mtime; | |
2227 | attr->ia_mtime.tv_nsec = hui->hui_mtime_ns; | |
2228 | attr->ia_atime.tv_sec = hui->hui_atime; | |
2229 | attr->ia_atime.tv_nsec = hui->hui_atime_ns; | |
2230 | ||
2231 | attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE | | |
2232 | ATTR_UID | ATTR_GID | | |
2233 | ATTR_MTIME | ATTR_MTIME_SET | | |
2234 | ATTR_ATIME | ATTR_ATIME_SET; | |
2235 | ||
5955102c | 2236 | inode_lock(inode); |
b6ee56fe | 2237 | |
b583043e | 2238 | rc = ll_setattr_raw(file->f_path.dentry, attr, true); |
a720b790 JL |
2239 | if (rc == -ENODATA) |
2240 | rc = 0; | |
2241 | ||
5955102c | 2242 | inode_unlock(inode); |
b6ee56fe | 2243 | |
e6b9a3b2 JL |
2244 | kfree(attr); |
2245 | free_hss: | |
2246 | kfree(hss); | |
a720b790 JL |
2247 | return rc; |
2248 | } | |
2249 | ||
2d95f10e JH |
2250 | static long |
2251 | ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg) | |
d7e09d03 | 2252 | { |
2a8a3597 | 2253 | struct inode *inode = file_inode(file); |
d7e09d03 PT |
2254 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); |
2255 | int flags, rc; | |
d7e09d03 | 2256 | |
97a075cd JN |
2257 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),cmd=%x\n", |
2258 | PFID(ll_inode2fid(inode)), inode, cmd); | |
d7e09d03 PT |
2259 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1); |
2260 | ||
2261 | /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */ | |
2262 | if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */ | |
0a3bdb00 | 2263 | return -ENOTTY; |
d7e09d03 | 2264 | |
a58a38ac | 2265 | switch (cmd) { |
d7e09d03 PT |
2266 | case LL_IOC_GETFLAGS: |
2267 | /* Get the current value of the file flags */ | |
02f9c12e | 2268 | return put_user(fd->fd_flags, (int __user *)arg); |
d7e09d03 PT |
2269 | case LL_IOC_SETFLAGS: |
2270 | case LL_IOC_CLRFLAGS: | |
2271 | /* Set or clear specific file flags */ | |
2272 | /* XXX This probably needs checks to ensure the flags are | |
2273 | * not abused, and to handle any flag side effects. | |
2274 | */ | |
02f9c12e | 2275 | if (get_user(flags, (int __user *)arg)) |
0a3bdb00 | 2276 | return -EFAULT; |
d7e09d03 PT |
2277 | |
2278 | if (cmd == LL_IOC_SETFLAGS) { | |
2279 | if ((flags & LL_FILE_IGNORE_LOCK) && | |
2280 | !(file->f_flags & O_DIRECT)) { | |
2d00bd17 JP |
2281 | CERROR("%s: unable to disable locking on non-O_DIRECT file\n", |
2282 | current->comm); | |
0a3bdb00 | 2283 | return -EINVAL; |
d7e09d03 PT |
2284 | } |
2285 | ||
2286 | fd->fd_flags |= flags; | |
2287 | } else { | |
2288 | fd->fd_flags &= ~flags; | |
2289 | } | |
0a3bdb00 | 2290 | return 0; |
d7e09d03 | 2291 | case LL_IOC_LOV_SETSTRIPE: |
0a3bdb00 | 2292 | return ll_lov_setstripe(inode, file, arg); |
d7e09d03 | 2293 | case LL_IOC_LOV_SETEA: |
0a3bdb00 | 2294 | return ll_lov_setea(inode, file, arg); |
d7e09d03 PT |
2295 | case LL_IOC_LOV_SWAP_LAYOUTS: { |
2296 | struct file *file2; | |
2297 | struct lustre_swap_layouts lsl; | |
2298 | ||
02f9c12e OD |
2299 | if (copy_from_user(&lsl, (char __user *)arg, |
2300 | sizeof(struct lustre_swap_layouts))) | |
0a3bdb00 | 2301 | return -EFAULT; |
d7e09d03 PT |
2302 | |
2303 | if ((file->f_flags & O_ACCMODE) == 0) /* O_RDONLY */ | |
0a3bdb00 | 2304 | return -EPERM; |
d7e09d03 PT |
2305 | |
2306 | file2 = fget(lsl.sl_fd); | |
6e16818b | 2307 | if (!file2) |
0a3bdb00 | 2308 | return -EBADF; |
d7e09d03 PT |
2309 | |
2310 | rc = -EPERM; | |
2311 | if ((file2->f_flags & O_ACCMODE) != 0) /* O_WRONLY or O_RDWR */ | |
2312 | rc = ll_swap_layouts(file, file2, &lsl); | |
2313 | fput(file2); | |
0a3bdb00 | 2314 | return rc; |
d7e09d03 PT |
2315 | } |
2316 | case LL_IOC_LOV_GETSTRIPE: | |
0a3bdb00 | 2317 | return ll_lov_getstripe(inode, arg); |
d7e09d03 | 2318 | case LL_IOC_RECREATE_OBJ: |
0a3bdb00 | 2319 | return ll_lov_recreate_obj(inode, arg); |
d7e09d03 | 2320 | case LL_IOC_RECREATE_FID: |
0a3bdb00 | 2321 | return ll_lov_recreate_fid(inode, arg); |
d7e09d03 | 2322 | case FSFILT_IOC_FIEMAP: |
0a3bdb00 | 2323 | return ll_ioctl_fiemap(inode, arg); |
d7e09d03 PT |
2324 | case FSFILT_IOC_GETFLAGS: |
2325 | case FSFILT_IOC_SETFLAGS: | |
0a3bdb00 | 2326 | return ll_iocontrol(inode, file, cmd, arg); |
d7e09d03 PT |
2327 | case FSFILT_IOC_GETVERSION_OLD: |
2328 | case FSFILT_IOC_GETVERSION: | |
02f9c12e | 2329 | return put_user(inode->i_generation, (int __user *)arg); |
d7e09d03 | 2330 | case LL_IOC_GROUP_LOCK: |
0a3bdb00 | 2331 | return ll_get_grouplock(inode, file, arg); |
d7e09d03 | 2332 | case LL_IOC_GROUP_UNLOCK: |
0a3bdb00 | 2333 | return ll_put_grouplock(inode, file, arg); |
d7e09d03 | 2334 | case IOC_OBD_STATFS: |
4c6243ec | 2335 | return ll_obd_statfs(inode, (void __user *)arg); |
d7e09d03 PT |
2336 | |
2337 | /* We need to special case any other ioctls we want to handle, | |
2338 | * to send them to the MDS/OST as appropriate and to properly | |
2339 | * network encode the arg field. | |
2340 | case FSFILT_IOC_SETVERSION_OLD: | |
2341 | case FSFILT_IOC_SETVERSION: | |
2342 | */ | |
2343 | case LL_IOC_FLUSHCTX: | |
0a3bdb00 | 2344 | return ll_flush_ctx(inode); |
d7e09d03 | 2345 | case LL_IOC_PATH2FID: { |
02f9c12e | 2346 | if (copy_to_user((void __user *)arg, ll_inode2fid(inode), |
d7e09d03 | 2347 | sizeof(struct lu_fid))) |
0a3bdb00 | 2348 | return -EFAULT; |
d7e09d03 | 2349 | |
0a3bdb00 | 2350 | return 0; |
d7e09d03 PT |
2351 | } |
2352 | case OBD_IOC_FID2PATH: | |
61dad0ba | 2353 | return ll_fid2path(inode, (void __user *)arg); |
d7e09d03 PT |
2354 | case LL_IOC_DATA_VERSION: { |
2355 | struct ioc_data_version idv; | |
2356 | int rc; | |
2357 | ||
02f9c12e | 2358 | if (copy_from_user(&idv, (char __user *)arg, sizeof(idv))) |
0a3bdb00 | 2359 | return -EFAULT; |
d7e09d03 | 2360 | |
e1798006 JX |
2361 | idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH; |
2362 | rc = ll_data_version(inode, &idv.idv_version, idv.idv_flags); | |
02f9c12e OD |
2363 | if (rc == 0 && copy_to_user((char __user *)arg, &idv, |
2364 | sizeof(idv))) | |
0a3bdb00 | 2365 | return -EFAULT; |
d7e09d03 | 2366 | |
0a3bdb00 | 2367 | return rc; |
d7e09d03 PT |
2368 | } |
2369 | ||
2370 | case LL_IOC_GET_MDTIDX: { | |
2371 | int mdtidx; | |
2372 | ||
2373 | mdtidx = ll_get_mdt_idx(inode); | |
2374 | if (mdtidx < 0) | |
0a3bdb00 | 2375 | return mdtidx; |
d7e09d03 | 2376 | |
02f9c12e | 2377 | if (put_user(mdtidx, (int __user *)arg)) |
0a3bdb00 | 2378 | return -EFAULT; |
d7e09d03 | 2379 | |
0a3bdb00 | 2380 | return 0; |
d7e09d03 PT |
2381 | } |
2382 | case OBD_IOC_GETDTNAME: | |
2383 | case OBD_IOC_GETMDNAME: | |
0a3bdb00 | 2384 | return ll_get_obd_name(inode, cmd, arg); |
d7e09d03 PT |
2385 | case LL_IOC_HSM_STATE_GET: { |
2386 | struct md_op_data *op_data; | |
2387 | struct hsm_user_state *hus; | |
2388 | int rc; | |
2389 | ||
496a51bd JL |
2390 | hus = kzalloc(sizeof(*hus), GFP_NOFS); |
2391 | if (!hus) | |
0a3bdb00 | 2392 | return -ENOMEM; |
d7e09d03 PT |
2393 | |
2394 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, | |
2395 | LUSTRE_OPC_ANY, hus); | |
79a8726a | 2396 | if (IS_ERR(op_data)) { |
97903a26 | 2397 | kfree(hus); |
0a3bdb00 | 2398 | return PTR_ERR(op_data); |
d7e09d03 PT |
2399 | } |
2400 | ||
2401 | rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data), | |
2402 | op_data, NULL); | |
2403 | ||
02f9c12e | 2404 | if (copy_to_user((void __user *)arg, hus, sizeof(*hus))) |
d7e09d03 PT |
2405 | rc = -EFAULT; |
2406 | ||
2407 | ll_finish_md_op_data(op_data); | |
97903a26 | 2408 | kfree(hus); |
0a3bdb00 | 2409 | return rc; |
d7e09d03 PT |
2410 | } |
2411 | case LL_IOC_HSM_STATE_SET: { | |
d7e09d03 PT |
2412 | struct hsm_state_set *hss; |
2413 | int rc; | |
2414 | ||
02f9c12e | 2415 | hss = memdup_user((char __user *)arg, sizeof(*hss)); |
0c027bc3 AH |
2416 | if (IS_ERR(hss)) |
2417 | return PTR_ERR(hss); | |
d7e09d03 | 2418 | |
a720b790 | 2419 | rc = ll_hsm_state_set(inode, hss); |
d7e09d03 | 2420 | |
97903a26 | 2421 | kfree(hss); |
0a3bdb00 | 2422 | return rc; |
d7e09d03 PT |
2423 | } |
2424 | case LL_IOC_HSM_ACTION: { | |
2425 | struct md_op_data *op_data; | |
2426 | struct hsm_current_action *hca; | |
2427 | int rc; | |
2428 | ||
496a51bd JL |
2429 | hca = kzalloc(sizeof(*hca), GFP_NOFS); |
2430 | if (!hca) | |
0a3bdb00 | 2431 | return -ENOMEM; |
d7e09d03 PT |
2432 | |
2433 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, | |
2434 | LUSTRE_OPC_ANY, hca); | |
79a8726a | 2435 | if (IS_ERR(op_data)) { |
97903a26 | 2436 | kfree(hca); |
0a3bdb00 | 2437 | return PTR_ERR(op_data); |
d7e09d03 PT |
2438 | } |
2439 | ||
2440 | rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data), | |
2441 | op_data, NULL); | |
2442 | ||
02f9c12e | 2443 | if (copy_to_user((char __user *)arg, hca, sizeof(*hca))) |
d7e09d03 PT |
2444 | rc = -EFAULT; |
2445 | ||
2446 | ll_finish_md_op_data(op_data); | |
97903a26 | 2447 | kfree(hca); |
0a3bdb00 | 2448 | return rc; |
d7e09d03 | 2449 | } |
d3a8a4e2 JX |
2450 | case LL_IOC_SET_LEASE: { |
2451 | struct ll_inode_info *lli = ll_i2info(inode); | |
2452 | struct obd_client_handle *och = NULL; | |
2453 | bool lease_broken; | |
2454 | fmode_t mode = 0; | |
2455 | ||
2456 | switch (arg) { | |
2457 | case F_WRLCK: | |
2458 | if (!(file->f_mode & FMODE_WRITE)) | |
2459 | return -EPERM; | |
2460 | mode = FMODE_WRITE; | |
2461 | break; | |
2462 | case F_RDLCK: | |
2463 | if (!(file->f_mode & FMODE_READ)) | |
2464 | return -EPERM; | |
2465 | mode = FMODE_READ; | |
2466 | break; | |
2467 | case F_UNLCK: | |
2468 | mutex_lock(&lli->lli_och_mutex); | |
6e16818b | 2469 | if (fd->fd_lease_och) { |
d3a8a4e2 JX |
2470 | och = fd->fd_lease_och; |
2471 | fd->fd_lease_och = NULL; | |
2472 | } | |
2473 | mutex_unlock(&lli->lli_och_mutex); | |
2474 | ||
6e16818b | 2475 | if (och) { |
d3a8a4e2 JX |
2476 | mode = och->och_flags & |
2477 | (FMODE_READ|FMODE_WRITE); | |
2478 | rc = ll_lease_close(och, inode, &lease_broken); | |
2479 | if (rc == 0 && lease_broken) | |
2480 | mode = 0; | |
2481 | } else { | |
2482 | rc = -ENOLCK; | |
2483 | } | |
2484 | ||
2485 | /* return the type of lease or error */ | |
2486 | return rc < 0 ? rc : (int)mode; | |
2487 | default: | |
2488 | return -EINVAL; | |
2489 | } | |
2490 | ||
2491 | CDEBUG(D_INODE, "Set lease with mode %d\n", mode); | |
2492 | ||
2493 | /* apply for lease */ | |
48d23e61 | 2494 | och = ll_lease_open(inode, file, mode, 0); |
d3a8a4e2 JX |
2495 | if (IS_ERR(och)) |
2496 | return PTR_ERR(och); | |
2497 | ||
2498 | rc = 0; | |
2499 | mutex_lock(&lli->lli_och_mutex); | |
6e16818b | 2500 | if (!fd->fd_lease_och) { |
d3a8a4e2 JX |
2501 | fd->fd_lease_och = och; |
2502 | och = NULL; | |
2503 | } | |
2504 | mutex_unlock(&lli->lli_och_mutex); | |
6e16818b | 2505 | if (och) { |
d3a8a4e2 JX |
2506 | /* impossible now that only excl is supported for now */ |
2507 | ll_lease_close(och, inode, &lease_broken); | |
2508 | rc = -EBUSY; | |
2509 | } | |
2510 | return rc; | |
2511 | } | |
2512 | case LL_IOC_GET_LEASE: { | |
2513 | struct ll_inode_info *lli = ll_i2info(inode); | |
2514 | struct ldlm_lock *lock = NULL; | |
2515 | ||
2516 | rc = 0; | |
2517 | mutex_lock(&lli->lli_och_mutex); | |
6e16818b | 2518 | if (fd->fd_lease_och) { |
d3a8a4e2 JX |
2519 | struct obd_client_handle *och = fd->fd_lease_och; |
2520 | ||
2521 | lock = ldlm_handle2lock(&och->och_lease_handle); | |
6e16818b | 2522 | if (lock) { |
d3a8a4e2 JX |
2523 | lock_res_and_lock(lock); |
2524 | if (!ldlm_is_cancel(lock)) | |
2525 | rc = och->och_flags & | |
2526 | (FMODE_READ | FMODE_WRITE); | |
2527 | unlock_res_and_lock(lock); | |
ead02808 | 2528 | LDLM_LOCK_PUT(lock); |
d3a8a4e2 JX |
2529 | } |
2530 | } | |
2531 | mutex_unlock(&lli->lli_och_mutex); | |
a720b790 JL |
2532 | return rc; |
2533 | } | |
2534 | case LL_IOC_HSM_IMPORT: { | |
2535 | struct hsm_user_import *hui; | |
2536 | ||
02f9c12e | 2537 | hui = memdup_user((void __user *)arg, sizeof(*hui)); |
0c027bc3 AH |
2538 | if (IS_ERR(hui)) |
2539 | return PTR_ERR(hui); | |
a720b790 JL |
2540 | |
2541 | rc = ll_hsm_import(inode, file, hui); | |
d3a8a4e2 | 2542 | |
97903a26 | 2543 | kfree(hui); |
d3a8a4e2 JX |
2544 | return rc; |
2545 | } | |
d7e09d03 PT |
2546 | default: { |
2547 | int err; | |
2548 | ||
1f6eaf83 JL |
2549 | if (ll_iocontrol_call(inode, file, cmd, arg, &err) == |
2550 | LLIOC_STOP) | |
0a3bdb00 | 2551 | return err; |
d7e09d03 | 2552 | |
0a3bdb00 | 2553 | return obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL, |
e09bee34 | 2554 | (void __user *)arg); |
d7e09d03 PT |
2555 | } |
2556 | } | |
2557 | } | |
2558 | ||
2d95f10e | 2559 | static loff_t ll_file_seek(struct file *file, loff_t offset, int origin) |
d7e09d03 | 2560 | { |
2a8a3597 | 2561 | struct inode *inode = file_inode(file); |
d7e09d03 PT |
2562 | loff_t retval, eof = 0; |
2563 | ||
d7e09d03 PT |
2564 | retval = offset + ((origin == SEEK_END) ? i_size_read(inode) : |
2565 | (origin == SEEK_CUR) ? file->f_pos : 0); | |
97a075cd JN |
2566 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n", |
2567 | PFID(ll_inode2fid(inode)), inode, retval, retval, origin); | |
d7e09d03 PT |
2568 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1); |
2569 | ||
2570 | if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) { | |
2571 | retval = ll_glimpse_size(inode); | |
2572 | if (retval != 0) | |
0a3bdb00 | 2573 | return retval; |
d7e09d03 PT |
2574 | eof = i_size_read(inode); |
2575 | } | |
2576 | ||
6f014339 | 2577 | retval = generic_file_llseek_size(file, offset, origin, |
d7e09d03 | 2578 | ll_file_maxbytes(inode), eof); |
0a3bdb00 | 2579 | return retval; |
d7e09d03 PT |
2580 | } |
2581 | ||
2d95f10e | 2582 | static int ll_flush(struct file *file, fl_owner_t id) |
d7e09d03 | 2583 | { |
2a8a3597 | 2584 | struct inode *inode = file_inode(file); |
d7e09d03 PT |
2585 | struct ll_inode_info *lli = ll_i2info(inode); |
2586 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
2587 | int rc, err; | |
2588 | ||
2589 | LASSERT(!S_ISDIR(inode->i_mode)); | |
2590 | ||
2591 | /* catch async errors that were recorded back when async writeback | |
c0894c6c OD |
2592 | * failed for pages in this mapping. |
2593 | */ | |
d7e09d03 PT |
2594 | rc = lli->lli_async_rc; |
2595 | lli->lli_async_rc = 0; | |
2596 | err = lov_read_and_clear_async_rc(lli->lli_clob); | |
2597 | if (rc == 0) | |
2598 | rc = err; | |
2599 | ||
c0894c6c OD |
2600 | /* The application has been told about write failure already. |
2601 | * Do not report failure again. | |
2602 | */ | |
d7e09d03 PT |
2603 | if (fd->fd_write_failed) |
2604 | return 0; | |
2605 | return rc ? -EIO : 0; | |
2606 | } | |
2607 | ||
2608 | /** | |
2609 | * Called to make sure a portion of file has been written out. | |
05289927 | 2610 | * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST. |
d7e09d03 PT |
2611 | * |
2612 | * Return how many pages have been written. | |
2613 | */ | |
2614 | int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end, | |
65fb55d1 | 2615 | enum cl_fsync_mode mode, int ignore_layout) |
d7e09d03 PT |
2616 | { |
2617 | struct cl_env_nest nest; | |
2618 | struct lu_env *env; | |
2619 | struct cl_io *io; | |
d7e09d03 PT |
2620 | struct cl_fsync_io *fio; |
2621 | int result; | |
d7e09d03 PT |
2622 | |
2623 | if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL && | |
2624 | mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL) | |
0a3bdb00 | 2625 | return -EINVAL; |
d7e09d03 PT |
2626 | |
2627 | env = cl_env_nested_get(&nest); | |
2628 | if (IS_ERR(env)) | |
0a3bdb00 | 2629 | return PTR_ERR(env); |
d7e09d03 | 2630 | |
9acc4500 | 2631 | io = vvp_env_thread_io(env); |
1929c433 | 2632 | io->ci_obj = ll_i2info(inode)->lli_clob; |
65fb55d1 | 2633 | io->ci_ignore_layout = ignore_layout; |
d7e09d03 PT |
2634 | |
2635 | /* initialize parameters for sync */ | |
2636 | fio = &io->u.ci_fsync; | |
d7e09d03 PT |
2637 | fio->fi_start = start; |
2638 | fio->fi_end = end; | |
2639 | fio->fi_fid = ll_inode2fid(inode); | |
2640 | fio->fi_mode = mode; | |
2641 | fio->fi_nr_written = 0; | |
2642 | ||
2643 | if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0) | |
2644 | result = cl_io_loop(env, io); | |
2645 | else | |
2646 | result = io->ci_result; | |
2647 | if (result == 0) | |
2648 | result = fio->fi_nr_written; | |
2649 | cl_io_fini(env, io); | |
2650 | cl_env_nested_put(&nest, env); | |
2651 | ||
0a3bdb00 | 2652 | return result; |
d7e09d03 PT |
2653 | } |
2654 | ||
d7e09d03 PT |
2655 | int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync) |
2656 | { | |
2a8a3597 | 2657 | struct inode *inode = file_inode(file); |
d7e09d03 PT |
2658 | struct ll_inode_info *lli = ll_i2info(inode); |
2659 | struct ptlrpc_request *req; | |
d7e09d03 | 2660 | int rc, err; |
d7e09d03 | 2661 | |
97a075cd JN |
2662 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n", |
2663 | PFID(ll_inode2fid(inode)), inode); | |
d7e09d03 PT |
2664 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1); |
2665 | ||
2666 | rc = filemap_write_and_wait_range(inode->i_mapping, start, end); | |
5955102c | 2667 | inode_lock(inode); |
d7e09d03 PT |
2668 | |
2669 | /* catch async errors that were recorded back when async writeback | |
c0894c6c OD |
2670 | * failed for pages in this mapping. |
2671 | */ | |
d7e09d03 PT |
2672 | if (!S_ISDIR(inode->i_mode)) { |
2673 | err = lli->lli_async_rc; | |
2674 | lli->lli_async_rc = 0; | |
2675 | if (rc == 0) | |
2676 | rc = err; | |
2677 | err = lov_read_and_clear_async_rc(lli->lli_clob); | |
2678 | if (rc == 0) | |
2679 | rc = err; | |
2680 | } | |
2681 | ||
ef2e0f55 | 2682 | err = md_sync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req); |
d7e09d03 PT |
2683 | if (!rc) |
2684 | rc = err; | |
2685 | if (!err) | |
2686 | ptlrpc_req_finished(req); | |
2687 | ||
8d97deb9 | 2688 | if (S_ISREG(inode->i_mode)) { |
d7e09d03 PT |
2689 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); |
2690 | ||
05289927 | 2691 | err = cl_sync_file_range(inode, start, end, CL_FSYNC_ALL, 0); |
d7e09d03 PT |
2692 | if (rc == 0 && err < 0) |
2693 | rc = err; | |
2694 | if (rc < 0) | |
2695 | fd->fd_write_failed = true; | |
2696 | else | |
2697 | fd->fd_write_failed = false; | |
2698 | } | |
2699 | ||
5955102c | 2700 | inode_unlock(inode); |
0a3bdb00 | 2701 | return rc; |
d7e09d03 PT |
2702 | } |
2703 | ||
2d95f10e JH |
2704 | static int |
2705 | ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) | |
d7e09d03 | 2706 | { |
2a8a3597 | 2707 | struct inode *inode = file_inode(file); |
d7e09d03 | 2708 | struct ll_sb_info *sbi = ll_i2sbi(inode); |
f2145eae BK |
2709 | struct ldlm_enqueue_info einfo = { |
2710 | .ei_type = LDLM_FLOCK, | |
2711 | .ei_cb_cp = ldlm_flock_completion_ast, | |
2712 | .ei_cbdata = file_lock, | |
2713 | }; | |
d7e09d03 PT |
2714 | struct md_op_data *op_data; |
2715 | struct lustre_handle lockh = {0}; | |
8369cfff | 2716 | ldlm_policy_data_t flock = { {0} }; |
875332d4 | 2717 | __u64 flags = 0; |
d7e09d03 PT |
2718 | int rc; |
2719 | int rc2 = 0; | |
d7e09d03 | 2720 | |
97a075cd JN |
2721 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n", |
2722 | PFID(ll_inode2fid(inode)), file_lock); | |
d7e09d03 PT |
2723 | |
2724 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1); | |
2725 | ||
130d1f95 | 2726 | if (file_lock->fl_flags & FL_FLOCK) |
d7e09d03 | 2727 | LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK)); |
130d1f95 | 2728 | else if (!(file_lock->fl_flags & FL_POSIX)) |
0a3bdb00 | 2729 | return -EINVAL; |
130d1f95 JL |
2730 | |
2731 | flock.l_flock.owner = (unsigned long)file_lock->fl_owner; | |
d7e09d03 | 2732 | flock.l_flock.pid = file_lock->fl_pid; |
130d1f95 JL |
2733 | flock.l_flock.start = file_lock->fl_start; |
2734 | flock.l_flock.end = file_lock->fl_end; | |
d7e09d03 PT |
2735 | |
2736 | /* Somewhat ugly workaround for svc lockd. | |
2737 | * lockd installs custom fl_lmops->lm_compare_owner that checks | |
2738 | * for the fl_owner to be the same (which it always is on local node | |
2739 | * I guess between lockd processes) and then compares pid. | |
2740 | * As such we assign pid to the owner field to make it all work, | |
2741 | * conflict with normal locks is unlikely since pid space and | |
c0894c6c OD |
2742 | * pointer space for current->files are not intersecting |
2743 | */ | |
d7e09d03 PT |
2744 | if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner) |
2745 | flock.l_flock.owner = (unsigned long)file_lock->fl_pid; | |
2746 | ||
2747 | switch (file_lock->fl_type) { | |
2748 | case F_RDLCK: | |
2749 | einfo.ei_mode = LCK_PR; | |
2750 | break; | |
2751 | case F_UNLCK: | |
2752 | /* An unlock request may or may not have any relation to | |
2753 | * existing locks so we may not be able to pass a lock handle | |
2754 | * via a normal ldlm_lock_cancel() request. The request may even | |
2755 | * unlock a byte range in the middle of an existing lock. In | |
2756 | * order to process an unlock request we need all of the same | |
2757 | * information that is given with a normal read or write record | |
2758 | * lock request. To avoid creating another ldlm unlock (cancel) | |
c0894c6c OD |
2759 | * message we'll treat a LCK_NL flock request as an unlock. |
2760 | */ | |
d7e09d03 PT |
2761 | einfo.ei_mode = LCK_NL; |
2762 | break; | |
2763 | case F_WRLCK: | |
2764 | einfo.ei_mode = LCK_PW; | |
2765 | break; | |
2766 | default: | |
2767 | CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", | |
e15ba45d | 2768 | file_lock->fl_type); |
0a3bdb00 | 2769 | return -ENOTSUPP; |
d7e09d03 PT |
2770 | } |
2771 | ||
2772 | switch (cmd) { | |
2773 | case F_SETLKW: | |
2faedcd5 | 2774 | #ifdef F_SETLKW64 |
d7e09d03 | 2775 | case F_SETLKW64: |
2faedcd5 | 2776 | #endif |
d7e09d03 PT |
2777 | flags = 0; |
2778 | break; | |
2779 | case F_SETLK: | |
2faedcd5 | 2780 | #ifdef F_SETLK64 |
d7e09d03 | 2781 | case F_SETLK64: |
2faedcd5 | 2782 | #endif |
d7e09d03 PT |
2783 | flags = LDLM_FL_BLOCK_NOWAIT; |
2784 | break; | |
2785 | case F_GETLK: | |
2faedcd5 | 2786 | #ifdef F_GETLK64 |
d7e09d03 | 2787 | case F_GETLK64: |
2faedcd5 | 2788 | #endif |
d7e09d03 PT |
2789 | flags = LDLM_FL_TEST_LOCK; |
2790 | /* Save the old mode so that if the mode in the lock changes we | |
c0894c6c OD |
2791 | * can decrement the appropriate reader or writer refcount. |
2792 | */ | |
d7e09d03 PT |
2793 | file_lock->fl_type = einfo.ei_mode; |
2794 | break; | |
2795 | default: | |
2796 | CERROR("unknown fcntl lock command: %d\n", cmd); | |
0a3bdb00 | 2797 | return -EINVAL; |
d7e09d03 PT |
2798 | } |
2799 | ||
2800 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, | |
2801 | LUSTRE_OPC_ANY, NULL); | |
2802 | if (IS_ERR(op_data)) | |
0a3bdb00 | 2803 | return PTR_ERR(op_data); |
d7e09d03 | 2804 | |
97a075cd JN |
2805 | CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, start=%llu, end=%llu\n", |
2806 | PFID(ll_inode2fid(inode)), flock.l_flock.pid, flags, | |
2807 | einfo.ei_mode, flock.l_flock.start, flock.l_flock.end); | |
d7e09d03 PT |
2808 | |
2809 | rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL, | |
2810 | op_data, &lockh, &flock, 0, NULL /* req */, flags); | |
2811 | ||
4f656367 | 2812 | if ((rc == 0 || file_lock->fl_type == F_UNLCK) && |
d7e09d03 | 2813 | !(flags & LDLM_FL_TEST_LOCK)) |
4f656367 | 2814 | rc2 = locks_lock_file_wait(file, file_lock); |
d7e09d03 PT |
2815 | |
2816 | if (rc2 && file_lock->fl_type != F_UNLCK) { | |
2817 | einfo.ei_mode = LCK_NL; | |
2818 | md_enqueue(sbi->ll_md_exp, &einfo, NULL, | |
e15ba45d | 2819 | op_data, &lockh, &flock, 0, NULL /* req */, flags); |
d7e09d03 PT |
2820 | rc = rc2; |
2821 | } | |
2822 | ||
2823 | ll_finish_md_op_data(op_data); | |
2824 | ||
0a3bdb00 | 2825 | return rc; |
d7e09d03 PT |
2826 | } |
2827 | ||
2d95f10e JH |
2828 | static int |
2829 | ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock) | |
d7e09d03 | 2830 | { |
0a3bdb00 | 2831 | return -ENOSYS; |
d7e09d03 PT |
2832 | } |
2833 | ||
2834 | /** | |
2835 | * test if some locks matching bits and l_req_mode are acquired | |
2836 | * - bits can be in different locks | |
2837 | * - if found clear the common lock bits in *bits | |
2838 | * - the bits not found, are kept in *bits | |
2839 | * \param inode [IN] | |
2840 | * \param bits [IN] searched lock bits [IN] | |
2841 | * \param l_req_mode [IN] searched lock mode | |
2842 | * \retval boolean, true iff all bits are found | |
2843 | */ | |
52ee0d20 OD |
2844 | int ll_have_md_lock(struct inode *inode, __u64 *bits, |
2845 | enum ldlm_mode l_req_mode) | |
d7e09d03 PT |
2846 | { |
2847 | struct lustre_handle lockh; | |
2848 | ldlm_policy_data_t policy; | |
52ee0d20 | 2849 | enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ? |
d7e09d03 PT |
2850 | (LCK_CR|LCK_CW|LCK_PR|LCK_PW) : l_req_mode; |
2851 | struct lu_fid *fid; | |
2852 | __u64 flags; | |
2853 | int i; | |
d7e09d03 PT |
2854 | |
2855 | if (!inode) | |
ef075edc | 2856 | return 0; |
d7e09d03 PT |
2857 | |
2858 | fid = &ll_i2info(inode)->lli_fid; | |
2859 | CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid), | |
2860 | ldlm_lockname[mode]); | |
2861 | ||
2862 | flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK; | |
1253b2e8 | 2863 | for (i = 0; i <= MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) { |
d7e09d03 PT |
2864 | policy.l_inodebits.bits = *bits & (1 << i); |
2865 | if (policy.l_inodebits.bits == 0) | |
2866 | continue; | |
2867 | ||
2868 | if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS, | |
2869 | &policy, mode, &lockh)) { | |
2870 | struct ldlm_lock *lock; | |
2871 | ||
2872 | lock = ldlm_handle2lock(&lockh); | |
2873 | if (lock) { | |
2874 | *bits &= | |
2875 | ~(lock->l_policy_data.l_inodebits.bits); | |
2876 | LDLM_LOCK_PUT(lock); | |
2877 | } else { | |
2878 | *bits &= ~policy.l_inodebits.bits; | |
2879 | } | |
2880 | } | |
2881 | } | |
0a3bdb00 | 2882 | return *bits == 0; |
d7e09d03 PT |
2883 | } |
2884 | ||
52ee0d20 OD |
2885 | enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits, |
2886 | struct lustre_handle *lockh, __u64 flags, | |
2887 | enum ldlm_mode mode) | |
d7e09d03 | 2888 | { |
57303e76 | 2889 | ldlm_policy_data_t policy = { .l_inodebits = {bits} }; |
d7e09d03 | 2890 | struct lu_fid *fid; |
52ee0d20 | 2891 | enum ldlm_mode rc; |
d7e09d03 PT |
2892 | |
2893 | fid = &ll_i2info(inode)->lli_fid; | |
2894 | CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid)); | |
2895 | ||
1f6eaf83 | 2896 | rc = md_lock_match(ll_i2mdexp(inode), flags | LDLM_FL_BLOCK_GRANTED, |
7fc1f831 AP |
2897 | fid, LDLM_IBITS, &policy, mode, lockh); |
2898 | ||
0a3bdb00 | 2899 | return rc; |
d7e09d03 PT |
2900 | } |
2901 | ||
2902 | static int ll_inode_revalidate_fini(struct inode *inode, int rc) | |
2903 | { | |
2904 | /* Already unlinked. Just update nlink and return success */ | |
2905 | if (rc == -ENOENT) { | |
2906 | clear_nlink(inode); | |
2907 | /* This path cannot be hit for regular files unless in | |
bef31c78 MI |
2908 | * case of obscure races, so no need to validate size. |
2909 | */ | |
d7e09d03 PT |
2910 | if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode)) |
2911 | return 0; | |
2912 | } else if (rc != 0) { | |
e49634bb AD |
2913 | CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR, |
2914 | "%s: revalidate FID "DFID" error: rc = %d\n", | |
2915 | ll_get_fsname(inode->i_sb, NULL, 0), | |
2916 | PFID(ll_inode2fid(inode)), rc); | |
d7e09d03 PT |
2917 | } |
2918 | ||
2919 | return rc; | |
2920 | } | |
2921 | ||
2d95f10e | 2922 | static int __ll_inode_revalidate(struct dentry *dentry, __u64 ibits) |
d7e09d03 | 2923 | { |
2b0143b5 | 2924 | struct inode *inode = d_inode(dentry); |
d7e09d03 PT |
2925 | struct ptlrpc_request *req = NULL; |
2926 | struct obd_export *exp; | |
2927 | int rc = 0; | |
d7e09d03 | 2928 | |
97a075cd JN |
2929 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%pd\n", |
2930 | PFID(ll_inode2fid(inode)), inode, dentry); | |
d7e09d03 PT |
2931 | |
2932 | exp = ll_i2mdexp(inode); | |
2933 | ||
2934 | /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC. | |
2935 | * But under CMD case, it caused some lock issues, should be fixed | |
c0894c6c OD |
2936 | * with new CMD ibits lock. See bug 12718 |
2937 | */ | |
d7e09d03 PT |
2938 | if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) { |
2939 | struct lookup_intent oit = { .it_op = IT_GETATTR }; | |
2940 | struct md_op_data *op_data; | |
2941 | ||
2942 | if (ibits == MDS_INODELOCK_LOOKUP) | |
2943 | oit.it_op = IT_LOOKUP; | |
2944 | ||
2945 | /* Call getattr by fid, so do not provide name at all. */ | |
dbca51dd AV |
2946 | op_data = ll_prep_md_op_data(NULL, inode, |
2947 | inode, NULL, 0, 0, | |
d7e09d03 PT |
2948 | LUSTRE_OPC_ANY, NULL); |
2949 | if (IS_ERR(op_data)) | |
0a3bdb00 | 2950 | return PTR_ERR(op_data); |
d7e09d03 PT |
2951 | |
2952 | oit.it_create_mode |= M_CHECK_STALE; | |
2953 | rc = md_intent_lock(exp, op_data, NULL, 0, | |
2954 | /* we are not interested in name | |
c0894c6c OD |
2955 | * based lookup |
2956 | */ | |
d7e09d03 PT |
2957 | &oit, 0, &req, |
2958 | ll_md_blocking_ast, 0); | |
2959 | ll_finish_md_op_data(op_data); | |
2960 | oit.it_create_mode &= ~M_CHECK_STALE; | |
2961 | if (rc < 0) { | |
2962 | rc = ll_inode_revalidate_fini(inode, rc); | |
34e1f2bb | 2963 | goto out; |
d7e09d03 PT |
2964 | } |
2965 | ||
dbca51dd | 2966 | rc = ll_revalidate_it_finish(req, &oit, inode); |
d7e09d03 PT |
2967 | if (rc != 0) { |
2968 | ll_intent_release(&oit); | |
34e1f2bb | 2969 | goto out; |
d7e09d03 PT |
2970 | } |
2971 | ||
2972 | /* Unlinked? Unhash dentry, so it is not picked up later by | |
c0894c6c OD |
2973 | * do_lookup() -> ll_revalidate_it(). We cannot use d_drop |
2974 | * here to preserve get_cwd functionality on 2.6. | |
2975 | * Bug 10503 | |
2976 | */ | |
c9cc8d0f BF |
2977 | if (!d_inode(dentry)->i_nlink) { |
2978 | spin_lock(&inode->i_lock); | |
b1d2a127 | 2979 | d_lustre_invalidate(dentry, 0); |
c9cc8d0f BF |
2980 | spin_unlock(&inode->i_lock); |
2981 | } | |
d7e09d03 | 2982 | |
dbca51dd | 2983 | ll_lookup_finish_locks(&oit, inode); |
2b0143b5 DH |
2984 | } else if (!ll_have_md_lock(d_inode(dentry), &ibits, LCK_MINMODE)) { |
2985 | struct ll_sb_info *sbi = ll_i2sbi(d_inode(dentry)); | |
21aef7d9 | 2986 | u64 valid = OBD_MD_FLGETATTR; |
d7e09d03 PT |
2987 | struct md_op_data *op_data; |
2988 | int ealen = 0; | |
2989 | ||
2990 | if (S_ISREG(inode->i_mode)) { | |
44779340 | 2991 | rc = ll_get_default_mdsize(sbi, &ealen); |
d7e09d03 | 2992 | if (rc) |
0a3bdb00 | 2993 | return rc; |
d7e09d03 PT |
2994 | valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE; |
2995 | } | |
2996 | ||
2997 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, | |
2998 | 0, ealen, LUSTRE_OPC_ANY, | |
2999 | NULL); | |
3000 | if (IS_ERR(op_data)) | |
0a3bdb00 | 3001 | return PTR_ERR(op_data); |
d7e09d03 PT |
3002 | |
3003 | op_data->op_valid = valid; | |
d7e09d03 PT |
3004 | rc = md_getattr(sbi->ll_md_exp, op_data, &req); |
3005 | ll_finish_md_op_data(op_data); | |
3006 | if (rc) { | |
3007 | rc = ll_inode_revalidate_fini(inode, rc); | |
0a3bdb00 | 3008 | return rc; |
d7e09d03 PT |
3009 | } |
3010 | ||
3011 | rc = ll_prep_inode(&inode, req, NULL, NULL); | |
3012 | } | |
3013 | out: | |
3014 | ptlrpc_req_finished(req); | |
3015 | return rc; | |
3016 | } | |
3017 | ||
2d95f10e | 3018 | static int ll_inode_revalidate(struct dentry *dentry, __u64 ibits) |
d7e09d03 | 3019 | { |
2b0143b5 | 3020 | struct inode *inode = d_inode(dentry); |
d7e09d03 | 3021 | int rc; |
d7e09d03 | 3022 | |
2d95f10e | 3023 | rc = __ll_inode_revalidate(dentry, ibits); |
d7e09d03 | 3024 | if (rc != 0) |
0a3bdb00 | 3025 | return rc; |
d7e09d03 PT |
3026 | |
3027 | /* if object isn't regular file, don't validate size */ | |
3028 | if (!S_ISREG(inode->i_mode)) { | |
d2995737 JH |
3029 | LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_atime; |
3030 | LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_mtime; | |
3031 | LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_ctime; | |
d7e09d03 | 3032 | } else { |
5ea17d6c JL |
3033 | /* In case of restore, the MDT has the right size and has |
3034 | * already send it back without granting the layout lock, | |
3035 | * inode is up-to-date so glimpse is useless. | |
3036 | * Also to glimpse we need the layout, in case of a running | |
3037 | * restore the MDT holds the layout lock so the glimpse will | |
3038 | * block up to the end of restore (getattr will block) | |
3039 | */ | |
3040 | if (!(ll_i2info(inode)->lli_flags & LLIF_FILE_RESTORING)) | |
3041 | rc = ll_glimpse_size(inode); | |
d7e09d03 | 3042 | } |
0a3bdb00 | 3043 | return rc; |
d7e09d03 PT |
3044 | } |
3045 | ||
2d95f10e | 3046 | int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat) |
d7e09d03 | 3047 | { |
2b0143b5 | 3048 | struct inode *inode = d_inode(de); |
d7e09d03 PT |
3049 | struct ll_sb_info *sbi = ll_i2sbi(inode); |
3050 | struct ll_inode_info *lli = ll_i2info(inode); | |
f82ced5d | 3051 | int res; |
d7e09d03 | 3052 | |
2d95f10e JH |
3053 | res = ll_inode_revalidate(de, MDS_INODELOCK_UPDATE | |
3054 | MDS_INODELOCK_LOOKUP); | |
d7e09d03 PT |
3055 | ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1); |
3056 | ||
3057 | if (res) | |
3058 | return res; | |
3059 | ||
3060 | stat->dev = inode->i_sb->s_dev; | |
3061 | if (ll_need_32bit_api(sbi)) | |
3062 | stat->ino = cl_fid_build_ino(&lli->lli_fid, 1); | |
3063 | else | |
3064 | stat->ino = inode->i_ino; | |
3065 | stat->mode = inode->i_mode; | |
3066 | stat->nlink = inode->i_nlink; | |
3067 | stat->uid = inode->i_uid; | |
3068 | stat->gid = inode->i_gid; | |
3069 | stat->rdev = inode->i_rdev; | |
3070 | stat->atime = inode->i_atime; | |
3071 | stat->mtime = inode->i_mtime; | |
3072 | stat->ctime = inode->i_ctime; | |
3073 | stat->blksize = 1 << inode->i_blkbits; | |
3074 | ||
3075 | stat->size = i_size_read(inode); | |
3076 | stat->blocks = inode->i_blocks; | |
3077 | ||
3078 | return 0; | |
3079 | } | |
d7e09d03 | 3080 | |
2d95f10e JH |
3081 | static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
3082 | __u64 start, __u64 len) | |
89580e37 PT |
3083 | { |
3084 | int rc; | |
3085 | size_t num_bytes; | |
3086 | struct ll_user_fiemap *fiemap; | |
3087 | unsigned int extent_count = fieinfo->fi_extents_max; | |
3088 | ||
3089 | num_bytes = sizeof(*fiemap) + (extent_count * | |
3090 | sizeof(struct ll_fiemap_extent)); | |
e958f49b | 3091 | fiemap = libcfs_kvzalloc(num_bytes, GFP_NOFS); |
89580e37 | 3092 | |
6e16818b | 3093 | if (!fiemap) |
89580e37 PT |
3094 | return -ENOMEM; |
3095 | ||
3096 | fiemap->fm_flags = fieinfo->fi_flags; | |
3097 | fiemap->fm_extent_count = fieinfo->fi_extents_max; | |
3098 | fiemap->fm_start = start; | |
3099 | fiemap->fm_length = len; | |
97514241 OD |
3100 | if (extent_count > 0 && |
3101 | copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start, | |
3102 | sizeof(struct ll_fiemap_extent)) != 0) { | |
3103 | rc = -EFAULT; | |
3104 | goto out; | |
3105 | } | |
89580e37 PT |
3106 | |
3107 | rc = ll_do_fiemap(inode, fiemap, num_bytes); | |
3108 | ||
3109 | fieinfo->fi_flags = fiemap->fm_flags; | |
3110 | fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents; | |
97514241 OD |
3111 | if (extent_count > 0 && |
3112 | copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0], | |
3113 | fiemap->fm_mapped_extents * | |
3114 | sizeof(struct ll_fiemap_extent)) != 0) { | |
3115 | rc = -EFAULT; | |
3116 | goto out; | |
3117 | } | |
89580e37 | 3118 | |
97514241 | 3119 | out: |
e958f49b | 3120 | kvfree(fiemap); |
89580e37 PT |
3121 | return rc; |
3122 | } | |
d7e09d03 | 3123 | |
2d95f10e | 3124 | struct posix_acl *ll_get_acl(struct inode *inode, int type) |
d7e09d03 PT |
3125 | { |
3126 | struct ll_inode_info *lli = ll_i2info(inode); | |
3127 | struct posix_acl *acl = NULL; | |
d7e09d03 PT |
3128 | |
3129 | spin_lock(&lli->lli_lock); | |
3130 | /* VFS' acl_permission_check->check_acl will release the refcount */ | |
3131 | acl = posix_acl_dup(lli->lli_posix_acl); | |
ed7bdf5c | 3132 | #ifdef CONFIG_FS_POSIX_ACL |
b788dc51 | 3133 | forget_cached_acl(inode, type); |
ed7bdf5c | 3134 | #endif |
d7e09d03 PT |
3135 | spin_unlock(&lli->lli_lock); |
3136 | ||
0a3bdb00 | 3137 | return acl; |
d7e09d03 PT |
3138 | } |
3139 | ||
d7e09d03 PT |
3140 | int ll_inode_permission(struct inode *inode, int mask) |
3141 | { | |
3142 | int rc = 0; | |
d7e09d03 | 3143 | |
d7e09d03 PT |
3144 | if (mask & MAY_NOT_BLOCK) |
3145 | return -ECHILD; | |
d7e09d03 PT |
3146 | |
3147 | /* as root inode are NOT getting validated in lookup operation, | |
c0894c6c OD |
3148 | * need to do it before permission check. |
3149 | */ | |
d7e09d03 | 3150 | |
f76c23da | 3151 | if (is_root_inode(inode)) { |
2d95f10e JH |
3152 | rc = __ll_inode_revalidate(inode->i_sb->s_root, |
3153 | MDS_INODELOCK_LOOKUP); | |
d7e09d03 | 3154 | if (rc) |
0a3bdb00 | 3155 | return rc; |
d7e09d03 PT |
3156 | } |
3157 | ||
97a075cd JN |
3158 | CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n", |
3159 | PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask); | |
d7e09d03 | 3160 | |
d7e09d03 | 3161 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1); |
8707c96e | 3162 | rc = generic_permission(inode, mask); |
d7e09d03 | 3163 | |
0a3bdb00 | 3164 | return rc; |
d7e09d03 PT |
3165 | } |
3166 | ||
d7e09d03 PT |
3167 | /* -o localflock - only provides locally consistent flock locks */ |
3168 | struct file_operations ll_file_operations = { | |
b42b15fd | 3169 | .read_iter = ll_file_read_iter, |
b42b15fd | 3170 | .write_iter = ll_file_write_iter, |
d7e09d03 PT |
3171 | .unlocked_ioctl = ll_file_ioctl, |
3172 | .open = ll_file_open, | |
3173 | .release = ll_file_release, | |
3174 | .mmap = ll_file_mmap, | |
3175 | .llseek = ll_file_seek, | |
3176 | .splice_read = ll_file_splice_read, | |
3177 | .fsync = ll_fsync, | |
3178 | .flush = ll_flush | |
3179 | }; | |
3180 | ||
3181 | struct file_operations ll_file_operations_flock = { | |
b42b15fd | 3182 | .read_iter = ll_file_read_iter, |
b42b15fd | 3183 | .write_iter = ll_file_write_iter, |
d7e09d03 PT |
3184 | .unlocked_ioctl = ll_file_ioctl, |
3185 | .open = ll_file_open, | |
3186 | .release = ll_file_release, | |
3187 | .mmap = ll_file_mmap, | |
3188 | .llseek = ll_file_seek, | |
3189 | .splice_read = ll_file_splice_read, | |
3190 | .fsync = ll_fsync, | |
3191 | .flush = ll_flush, | |
3192 | .flock = ll_file_flock, | |
3193 | .lock = ll_file_flock | |
3194 | }; | |
3195 | ||
3196 | /* These are for -o noflock - to return ENOSYS on flock calls */ | |
3197 | struct file_operations ll_file_operations_noflock = { | |
b42b15fd | 3198 | .read_iter = ll_file_read_iter, |
b42b15fd | 3199 | .write_iter = ll_file_write_iter, |
d7e09d03 PT |
3200 | .unlocked_ioctl = ll_file_ioctl, |
3201 | .open = ll_file_open, | |
3202 | .release = ll_file_release, | |
3203 | .mmap = ll_file_mmap, | |
3204 | .llseek = ll_file_seek, | |
3205 | .splice_read = ll_file_splice_read, | |
3206 | .fsync = ll_fsync, | |
3207 | .flush = ll_flush, | |
3208 | .flock = ll_file_noflock, | |
3209 | .lock = ll_file_noflock | |
3210 | }; | |
3211 | ||
d2d32738 | 3212 | const struct inode_operations ll_file_inode_operations = { |
d7e09d03 PT |
3213 | .setattr = ll_setattr, |
3214 | .getattr = ll_getattr, | |
3215 | .permission = ll_inode_permission, | |
3216 | .setxattr = ll_setxattr, | |
3217 | .getxattr = ll_getxattr, | |
3218 | .listxattr = ll_listxattr, | |
3219 | .removexattr = ll_removexattr, | |
89580e37 | 3220 | .fiemap = ll_fiemap, |
d7e09d03 PT |
3221 | .get_acl = ll_get_acl, |
3222 | }; | |
3223 | ||
d0a0acc3 | 3224 | /* dynamic ioctl number support routines */ |
d7e09d03 PT |
3225 | static struct llioc_ctl_data { |
3226 | struct rw_semaphore ioc_sem; | |
3227 | struct list_head ioc_head; | |
3228 | } llioc = { | |
3229 | __RWSEM_INITIALIZER(llioc.ioc_sem), | |
3230 | LIST_HEAD_INIT(llioc.ioc_head) | |
3231 | }; | |
3232 | ||
d7e09d03 PT |
3233 | struct llioc_data { |
3234 | struct list_head iocd_list; | |
3235 | unsigned int iocd_size; | |
3236 | llioc_callback_t iocd_cb; | |
3237 | unsigned int iocd_count; | |
3238 | unsigned int iocd_cmd[0]; | |
3239 | }; | |
3240 | ||
3241 | void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd) | |
3242 | { | |
3243 | unsigned int size; | |
3244 | struct llioc_data *in_data = NULL; | |
d7e09d03 | 3245 | |
6e16818b | 3246 | if (!cb || !cmd || count > LLIOC_MAX_CMD || count < 0) |
0a3bdb00 | 3247 | return NULL; |
d7e09d03 PT |
3248 | |
3249 | size = sizeof(*in_data) + count * sizeof(unsigned int); | |
496a51bd JL |
3250 | in_data = kzalloc(size, GFP_NOFS); |
3251 | if (!in_data) | |
0a3bdb00 | 3252 | return NULL; |
d7e09d03 PT |
3253 | |
3254 | memset(in_data, 0, sizeof(*in_data)); | |
3255 | in_data->iocd_size = size; | |
3256 | in_data->iocd_cb = cb; | |
3257 | in_data->iocd_count = count; | |
3258 | memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count); | |
3259 | ||
3260 | down_write(&llioc.ioc_sem); | |
3261 | list_add_tail(&in_data->iocd_list, &llioc.ioc_head); | |
3262 | up_write(&llioc.ioc_sem); | |
3263 | ||
0a3bdb00 | 3264 | return in_data; |
d7e09d03 | 3265 | } |
93133eb4 | 3266 | EXPORT_SYMBOL(ll_iocontrol_register); |
d7e09d03 PT |
3267 | |
3268 | void ll_iocontrol_unregister(void *magic) | |
3269 | { | |
3270 | struct llioc_data *tmp; | |
3271 | ||
6e16818b | 3272 | if (!magic) |
d7e09d03 PT |
3273 | return; |
3274 | ||
3275 | down_write(&llioc.ioc_sem); | |
3276 | list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) { | |
3277 | if (tmp == magic) { | |
d7e09d03 PT |
3278 | list_del(&tmp->iocd_list); |
3279 | up_write(&llioc.ioc_sem); | |
3280 | ||
97903a26 | 3281 | kfree(tmp); |
d7e09d03 PT |
3282 | return; |
3283 | } | |
3284 | } | |
3285 | up_write(&llioc.ioc_sem); | |
3286 | ||
3287 | CWARN("didn't find iocontrol register block with magic: %p\n", magic); | |
3288 | } | |
d7e09d03 PT |
3289 | EXPORT_SYMBOL(ll_iocontrol_unregister); |
3290 | ||
2d95f10e JH |
3291 | static enum llioc_iter |
3292 | ll_iocontrol_call(struct inode *inode, struct file *file, | |
3293 | unsigned int cmd, unsigned long arg, int *rcp) | |
d7e09d03 PT |
3294 | { |
3295 | enum llioc_iter ret = LLIOC_CONT; | |
3296 | struct llioc_data *data; | |
3297 | int rc = -EINVAL, i; | |
3298 | ||
3299 | down_read(&llioc.ioc_sem); | |
3300 | list_for_each_entry(data, &llioc.ioc_head, iocd_list) { | |
3301 | for (i = 0; i < data->iocd_count; i++) { | |
3302 | if (cmd != data->iocd_cmd[i]) | |
3303 | continue; | |
3304 | ||
3305 | ret = data->iocd_cb(inode, file, cmd, arg, data, &rc); | |
3306 | break; | |
3307 | } | |
3308 | ||
3309 | if (ret == LLIOC_STOP) | |
3310 | break; | |
3311 | } | |
3312 | up_read(&llioc.ioc_sem); | |
3313 | ||
3314 | if (rcp) | |
3315 | *rcp = rc; | |
3316 | return ret; | |
3317 | } | |
3318 | ||
3319 | int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf) | |
3320 | { | |
3321 | struct ll_inode_info *lli = ll_i2info(inode); | |
3322 | struct cl_env_nest nest; | |
3323 | struct lu_env *env; | |
3324 | int result; | |
d7e09d03 | 3325 | |
6e16818b | 3326 | if (!lli->lli_clob) |
0a3bdb00 | 3327 | return 0; |
d7e09d03 PT |
3328 | |
3329 | env = cl_env_nested_get(&nest); | |
3330 | if (IS_ERR(env)) | |
0a3bdb00 | 3331 | return PTR_ERR(env); |
d7e09d03 PT |
3332 | |
3333 | result = cl_conf_set(env, lli->lli_clob, conf); | |
3334 | cl_env_nested_put(&nest, env); | |
3335 | ||
3336 | if (conf->coc_opc == OBJECT_CONF_SET) { | |
3337 | struct ldlm_lock *lock = conf->coc_lock; | |
3338 | ||
6e16818b | 3339 | LASSERT(lock); |
d7e09d03 PT |
3340 | LASSERT(ldlm_has_layout(lock)); |
3341 | if (result == 0) { | |
3342 | /* it can only be allowed to match after layout is | |
3343 | * applied to inode otherwise false layout would be | |
d0a0acc3 | 3344 | * seen. Applying layout should happen before dropping |
c0894c6c OD |
3345 | * the intent lock. |
3346 | */ | |
d7e09d03 PT |
3347 | ldlm_lock_allow_match(lock); |
3348 | } | |
3349 | } | |
0a3bdb00 | 3350 | return result; |
d7e09d03 PT |
3351 | } |
3352 | ||
3353 | /* Fetch layout from MDT with getxattr request, if it's not ready yet */ | |
3354 | static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock) | |
3355 | ||
3356 | { | |
3357 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
d7e09d03 PT |
3358 | struct ptlrpc_request *req; |
3359 | struct mdt_body *body; | |
3360 | void *lvbdata; | |
3361 | void *lmm; | |
3362 | int lmmsize; | |
3363 | int rc; | |
d7e09d03 | 3364 | |
e2335e5d | 3365 | CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n", |
5a9a80ba | 3366 | PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock), |
e2335e5d | 3367 | lock->l_lvb_data, lock->l_lvb_len); |
3368 | ||
5a9a80ba | 3369 | if (lock->l_lvb_data && ldlm_is_lvb_ready(lock)) |
0a3bdb00 | 3370 | return 0; |
d7e09d03 PT |
3371 | |
3372 | /* if layout lock was granted right away, the layout is returned | |
3373 | * within DLM_LVB of dlm reply; otherwise if the lock was ever | |
3374 | * blocked and then granted via completion ast, we have to fetch | |
3375 | * layout here. Please note that we can't use the LVB buffer in | |
c0894c6c OD |
3376 | * completion AST because it doesn't have a large enough buffer |
3377 | */ | |
44779340 | 3378 | rc = ll_get_default_mdsize(sbi, &lmmsize); |
d7e09d03 | 3379 | if (rc == 0) |
ef2e0f55 OD |
3380 | rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), |
3381 | OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0, | |
3382 | lmmsize, 0, &req); | |
d7e09d03 | 3383 | if (rc < 0) |
0a3bdb00 | 3384 | return rc; |
d7e09d03 PT |
3385 | |
3386 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); | |
6e16818b | 3387 | if (!body) { |
34e1f2bb JL |
3388 | rc = -EPROTO; |
3389 | goto out; | |
3390 | } | |
d7e09d03 PT |
3391 | |
3392 | lmmsize = body->eadatasize; | |
34e1f2bb JL |
3393 | if (lmmsize == 0) /* empty layout */ { |
3394 | rc = 0; | |
3395 | goto out; | |
3396 | } | |
d7e09d03 PT |
3397 | |
3398 | lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize); | |
6e16818b | 3399 | if (!lmm) { |
34e1f2bb JL |
3400 | rc = -EFAULT; |
3401 | goto out; | |
3402 | } | |
d7e09d03 | 3403 | |
e958f49b | 3404 | lvbdata = libcfs_kvzalloc(lmmsize, GFP_NOFS); |
6e16818b | 3405 | if (!lvbdata) { |
34e1f2bb JL |
3406 | rc = -ENOMEM; |
3407 | goto out; | |
3408 | } | |
d7e09d03 PT |
3409 | |
3410 | memcpy(lvbdata, lmm, lmmsize); | |
3411 | lock_res_and_lock(lock); | |
6e16818b | 3412 | if (lock->l_lvb_data) |
e958f49b | 3413 | kvfree(lock->l_lvb_data); |
e2335e5d | 3414 | |
3415 | lock->l_lvb_data = lvbdata; | |
3416 | lock->l_lvb_len = lmmsize; | |
d7e09d03 PT |
3417 | unlock_res_and_lock(lock); |
3418 | ||
d7e09d03 PT |
3419 | out: |
3420 | ptlrpc_req_finished(req); | |
3421 | return rc; | |
3422 | } | |
3423 | ||
3424 | /** | |
3425 | * Apply the layout to the inode. Layout lock is held and will be released | |
3426 | * in this function. | |
3427 | */ | |
52ee0d20 OD |
3428 | static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode, |
3429 | struct inode *inode, __u32 *gen, bool reconf) | |
d7e09d03 PT |
3430 | { |
3431 | struct ll_inode_info *lli = ll_i2info(inode); | |
3432 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
3433 | struct ldlm_lock *lock; | |
3434 | struct lustre_md md = { NULL }; | |
3435 | struct cl_object_conf conf; | |
3436 | int rc = 0; | |
3437 | bool lvb_ready; | |
3438 | bool wait_layout = false; | |
d7e09d03 PT |
3439 | |
3440 | LASSERT(lustre_handle_is_used(lockh)); | |
3441 | ||
3442 | lock = ldlm_handle2lock(lockh); | |
6e16818b | 3443 | LASSERT(lock); |
d7e09d03 PT |
3444 | LASSERT(ldlm_has_layout(lock)); |
3445 | ||
97a075cd JN |
3446 | LDLM_DEBUG(lock, "File "DFID"(%p) being reconfigured: %d", |
3447 | PFID(&lli->lli_fid), inode, reconf); | |
d7e09d03 | 3448 | |
bc969176 JL |
3449 | /* in case this is a caching lock and reinstate with new inode */ |
3450 | md_set_lock_data(sbi->ll_md_exp, &lockh->cookie, inode, NULL); | |
3451 | ||
d7e09d03 | 3452 | lock_res_and_lock(lock); |
5a9a80ba | 3453 | lvb_ready = ldlm_is_lvb_ready(lock); |
d7e09d03 PT |
3454 | unlock_res_and_lock(lock); |
3455 | /* checking lvb_ready is racy but this is okay. The worst case is | |
c0894c6c OD |
3456 | * that multi processes may configure the file on the same time. |
3457 | */ | |
d7e09d03 PT |
3458 | if (lvb_ready || !reconf) { |
3459 | rc = -ENODATA; | |
3460 | if (lvb_ready) { | |
3461 | /* layout_gen must be valid if layout lock is not | |
c0894c6c OD |
3462 | * cancelled and stripe has already set |
3463 | */ | |
09aed8a5 | 3464 | *gen = ll_layout_version_get(lli); |
d7e09d03 PT |
3465 | rc = 0; |
3466 | } | |
34e1f2bb | 3467 | goto out; |
d7e09d03 PT |
3468 | } |
3469 | ||
3470 | rc = ll_layout_fetch(inode, lock); | |
3471 | if (rc < 0) | |
34e1f2bb | 3472 | goto out; |
d7e09d03 PT |
3473 | |
3474 | /* for layout lock, lmm is returned in lock's lvb. | |
3475 | * lvb_data is immutable if the lock is held so it's safe to access it | |
3476 | * without res lock. See the description in ldlm_lock_decref_internal() | |
c0894c6c OD |
3477 | * for the condition to free lvb_data of layout lock |
3478 | */ | |
6e16818b | 3479 | if (lock->l_lvb_data) { |
d7e09d03 PT |
3480 | rc = obd_unpackmd(sbi->ll_dt_exp, &md.lsm, |
3481 | lock->l_lvb_data, lock->l_lvb_len); | |
3482 | if (rc >= 0) { | |
3483 | *gen = LL_LAYOUT_GEN_EMPTY; | |
6e16818b | 3484 | if (md.lsm) |
d7e09d03 PT |
3485 | *gen = md.lsm->lsm_layout_gen; |
3486 | rc = 0; | |
3487 | } else { | |
e15ba45d OD |
3488 | CERROR("%s: file " DFID " unpackmd error: %d\n", |
3489 | ll_get_fsname(inode->i_sb, NULL, 0), | |
3490 | PFID(&lli->lli_fid), rc); | |
d7e09d03 PT |
3491 | } |
3492 | } | |
3493 | if (rc < 0) | |
34e1f2bb | 3494 | goto out; |
d7e09d03 PT |
3495 | |
3496 | /* set layout to file. Unlikely this will fail as old layout was | |
c0894c6c OD |
3497 | * surely eliminated |
3498 | */ | |
ec83e611 | 3499 | memset(&conf, 0, sizeof(conf)); |
d7e09d03 PT |
3500 | conf.coc_opc = OBJECT_CONF_SET; |
3501 | conf.coc_inode = inode; | |
3502 | conf.coc_lock = lock; | |
3503 | conf.u.coc_md = &md; | |
3504 | rc = ll_layout_conf(inode, &conf); | |
3505 | ||
6e16818b | 3506 | if (md.lsm) |
d7e09d03 PT |
3507 | obd_free_memmd(sbi->ll_dt_exp, &md.lsm); |
3508 | ||
3509 | /* refresh layout failed, need to wait */ | |
3510 | wait_layout = rc == -EBUSY; | |
d7e09d03 PT |
3511 | |
3512 | out: | |
3513 | LDLM_LOCK_PUT(lock); | |
3514 | ldlm_lock_decref(lockh, mode); | |
3515 | ||
3516 | /* wait for IO to complete if it's still being used. */ | |
3517 | if (wait_layout) { | |
97a075cd | 3518 | CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n", |
e15ba45d | 3519 | ll_get_fsname(inode->i_sb, NULL, 0), |
97a075cd | 3520 | PFID(&lli->lli_fid), inode); |
d7e09d03 | 3521 | |
ec83e611 | 3522 | memset(&conf, 0, sizeof(conf)); |
d7e09d03 PT |
3523 | conf.coc_opc = OBJECT_CONF_WAIT; |
3524 | conf.coc_inode = inode; | |
3525 | rc = ll_layout_conf(inode, &conf); | |
3526 | if (rc == 0) | |
3527 | rc = -EAGAIN; | |
3528 | ||
97a075cd JN |
3529 | CDEBUG(D_INODE, "%s: file="DFID" waiting layout return: %d.\n", |
3530 | ll_get_fsname(inode->i_sb, NULL, 0), | |
e15ba45d | 3531 | PFID(&lli->lli_fid), rc); |
d7e09d03 | 3532 | } |
0a3bdb00 | 3533 | return rc; |
d7e09d03 PT |
3534 | } |
3535 | ||
3536 | /** | |
3537 | * This function checks if there exists a LAYOUT lock on the client side, | |
3538 | * or enqueues it if it doesn't have one in cache. | |
3539 | * | |
3540 | * This function will not hold layout lock so it may be revoked any time after | |
3541 | * this function returns. Any operations depend on layout should be redone | |
3542 | * in that case. | |
3543 | * | |
3544 | * This function should be called before lov_io_init() to get an uptodate | |
3545 | * layout version, the caller should save the version number and after IO | |
3546 | * is finished, this function should be called again to verify that layout | |
3547 | * is not changed during IO time. | |
3548 | */ | |
3549 | int ll_layout_refresh(struct inode *inode, __u32 *gen) | |
3550 | { | |
3551 | struct ll_inode_info *lli = ll_i2info(inode); | |
3552 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
3553 | struct md_op_data *op_data; | |
3554 | struct lookup_intent it; | |
3555 | struct lustre_handle lockh; | |
52ee0d20 | 3556 | enum ldlm_mode mode; |
f2145eae BK |
3557 | struct ldlm_enqueue_info einfo = { |
3558 | .ei_type = LDLM_IBITS, | |
3559 | .ei_mode = LCK_CR, | |
3560 | .ei_cb_bl = ll_md_blocking_ast, | |
3561 | .ei_cb_cp = ldlm_completion_ast, | |
3562 | }; | |
d7e09d03 | 3563 | int rc; |
d7e09d03 | 3564 | |
09aed8a5 JX |
3565 | *gen = ll_layout_version_get(lli); |
3566 | if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != LL_LAYOUT_GEN_NONE) | |
0a3bdb00 | 3567 | return 0; |
d7e09d03 PT |
3568 | |
3569 | /* sanity checks */ | |
3570 | LASSERT(fid_is_sane(ll_inode2fid(inode))); | |
3571 | LASSERT(S_ISREG(inode->i_mode)); | |
3572 | ||
d7e09d03 PT |
3573 | /* take layout lock mutex to enqueue layout lock exclusively. */ |
3574 | mutex_lock(&lli->lli_layout_mutex); | |
3575 | ||
3576 | again: | |
09aed8a5 | 3577 | /* mostly layout lock is caching on the local side, so try to match |
c0894c6c OD |
3578 | * it before grabbing layout lock mutex. |
3579 | */ | |
7fc1f831 AP |
3580 | mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0, |
3581 | LCK_CR | LCK_CW | LCK_PR | LCK_PW); | |
d7e09d03 PT |
3582 | if (mode != 0) { /* hit cached lock */ |
3583 | rc = ll_layout_lock_set(&lockh, mode, inode, gen, true); | |
3584 | if (rc == -EAGAIN) | |
3585 | goto again; | |
3586 | ||
3587 | mutex_unlock(&lli->lli_layout_mutex); | |
0a3bdb00 | 3588 | return rc; |
d7e09d03 PT |
3589 | } |
3590 | ||
3591 | op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, | |
e15ba45d | 3592 | 0, 0, LUSTRE_OPC_ANY, NULL); |
d7e09d03 PT |
3593 | if (IS_ERR(op_data)) { |
3594 | mutex_unlock(&lli->lli_layout_mutex); | |
0a3bdb00 | 3595 | return PTR_ERR(op_data); |
d7e09d03 PT |
3596 | } |
3597 | ||
3598 | /* have to enqueue one */ | |
3599 | memset(&it, 0, sizeof(it)); | |
3600 | it.it_op = IT_LAYOUT; | |
3601 | lockh.cookie = 0ULL; | |
3602 | ||
97a075cd JN |
3603 | LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)", |
3604 | ll_get_fsname(inode->i_sb, NULL, 0), | |
3605 | PFID(&lli->lli_fid), inode); | |
d7e09d03 PT |
3606 | |
3607 | rc = md_enqueue(sbi->ll_md_exp, &einfo, &it, op_data, &lockh, | |
3608 | NULL, 0, NULL, 0); | |
8bf86fd9 JH |
3609 | ptlrpc_req_finished(it.it_request); |
3610 | it.it_request = NULL; | |
d7e09d03 PT |
3611 | |
3612 | ll_finish_md_op_data(op_data); | |
3613 | ||
e476f2e5 JH |
3614 | mode = it.it_lock_mode; |
3615 | it.it_lock_mode = 0; | |
d7e09d03 PT |
3616 | ll_intent_drop_lock(&it); |
3617 | ||
3618 | if (rc == 0) { | |
3619 | /* set lock data in case this is a new lock */ | |
3620 | ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL); | |
3621 | rc = ll_layout_lock_set(&lockh, mode, inode, gen, true); | |
3622 | if (rc == -EAGAIN) | |
3623 | goto again; | |
3624 | } | |
3625 | mutex_unlock(&lli->lli_layout_mutex); | |
3626 | ||
0a3bdb00 | 3627 | return rc; |
d7e09d03 | 3628 | } |
5ea17d6c JL |
3629 | |
3630 | /** | |
3631 | * This function send a restore request to the MDT | |
3632 | */ | |
1b1594da | 3633 | int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length) |
5ea17d6c JL |
3634 | { |
3635 | struct hsm_user_request *hur; | |
3636 | int len, rc; | |
3637 | ||
3638 | len = sizeof(struct hsm_user_request) + | |
3639 | sizeof(struct hsm_user_item); | |
496a51bd JL |
3640 | hur = kzalloc(len, GFP_NOFS); |
3641 | if (!hur) | |
5ea17d6c JL |
3642 | return -ENOMEM; |
3643 | ||
3644 | hur->hur_request.hr_action = HUA_RESTORE; | |
3645 | hur->hur_request.hr_archive_id = 0; | |
3646 | hur->hur_request.hr_flags = 0; | |
3647 | memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid, | |
3648 | sizeof(hur->hur_user_item[0].hui_fid)); | |
1b1594da JX |
3649 | hur->hur_user_item[0].hui_extent.offset = offset; |
3650 | hur->hur_user_item[0].hui_extent.length = length; | |
5ea17d6c | 3651 | hur->hur_request.hr_itemcount = 1; |
1929c433 | 3652 | rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp, |
5ea17d6c | 3653 | len, hur, NULL); |
97903a26 | 3654 | kfree(hur); |
5ea17d6c JL |
3655 | return rc; |
3656 | } |