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 | |
18 | * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf | |
19 | * | |
20 | * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
21 | * CA 95054 USA or visit www.sun.com if you need additional information or | |
22 | * have any questions. | |
23 | * | |
24 | * GPL HEADER END | |
25 | */ | |
26 | /* | |
27 | * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. | |
28 | * Use is subject to license terms. | |
29 | * | |
30 | * Copyright (c) 2011, 2012, Intel Corporation. | |
31 | */ | |
32 | /* | |
33 | * This file is part of Lustre, http://www.lustre.org/ | |
34 | * Lustre is a trademark of Sun Microsystems, Inc. | |
35 | * | |
36 | * lustre/llite/file.c | |
37 | * | |
38 | * Author: Peter Braam <braam@clusterfs.com> | |
39 | * Author: Phil Schwan <phil@clusterfs.com> | |
40 | * Author: Andreas Dilger <adilger@clusterfs.com> | |
41 | */ | |
42 | ||
43 | #define DEBUG_SUBSYSTEM S_LLITE | |
44 | #include <lustre_dlm.h> | |
45 | #include <lustre_lite.h> | |
46 | #include <linux/pagemap.h> | |
47 | #include <linux/file.h> | |
48 | #include "llite_internal.h" | |
49 | #include <lustre/ll_fiemap.h> | |
50 | ||
51 | #include "cl_object.h" | |
52 | ||
53 | struct ll_file_data *ll_file_data_get(void) | |
54 | { | |
55 | struct ll_file_data *fd; | |
56 | ||
57 | OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, __GFP_IO); | |
58 | fd->fd_write_failed = false; | |
59 | return fd; | |
60 | } | |
61 | ||
62 | static void ll_file_data_put(struct ll_file_data *fd) | |
63 | { | |
64 | if (fd != NULL) | |
65 | OBD_SLAB_FREE_PTR(fd, ll_file_data_slab); | |
66 | } | |
67 | ||
68 | void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data, | |
69 | struct lustre_handle *fh) | |
70 | { | |
71 | op_data->op_fid1 = ll_i2info(inode)->lli_fid; | |
72 | op_data->op_attr.ia_mode = inode->i_mode; | |
73 | op_data->op_attr.ia_atime = inode->i_atime; | |
74 | op_data->op_attr.ia_mtime = inode->i_mtime; | |
75 | op_data->op_attr.ia_ctime = inode->i_ctime; | |
76 | op_data->op_attr.ia_size = i_size_read(inode); | |
77 | op_data->op_attr_blocks = inode->i_blocks; | |
78 | ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = | |
79 | ll_inode_to_ext_flags(inode->i_flags); | |
80 | op_data->op_ioepoch = ll_i2info(inode)->lli_ioepoch; | |
81 | if (fh) | |
82 | op_data->op_handle = *fh; | |
83 | op_data->op_capa1 = ll_mdscapa_get(inode); | |
84 | ||
85 | if (LLIF_DATA_MODIFIED & ll_i2info(inode)->lli_flags) | |
86 | op_data->op_bias |= MDS_DATA_MODIFIED; | |
87 | } | |
88 | ||
89 | /** | |
90 | * Closes the IO epoch and packs all the attributes into @op_data for | |
91 | * the CLOSE rpc. | |
92 | */ | |
93 | static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data, | |
94 | struct obd_client_handle *och) | |
95 | { | |
96 | ENTRY; | |
97 | ||
98 | op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME_SET | | |
99 | ATTR_MTIME_SET | ATTR_CTIME_SET; | |
100 | ||
101 | if (!(och->och_flags & FMODE_WRITE)) | |
102 | goto out; | |
103 | ||
104 | if (!exp_connect_som(ll_i2mdexp(inode)) || !S_ISREG(inode->i_mode)) | |
105 | op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS; | |
106 | else | |
107 | ll_ioepoch_close(inode, op_data, &och, 0); | |
108 | ||
109 | out: | |
110 | ll_pack_inode2opdata(inode, op_data, &och->och_fh); | |
111 | ll_prep_md_op_data(op_data, inode, NULL, NULL, | |
112 | 0, 0, LUSTRE_OPC_ANY, NULL); | |
113 | EXIT; | |
114 | } | |
115 | ||
116 | static int ll_close_inode_openhandle(struct obd_export *md_exp, | |
117 | struct inode *inode, | |
118 | struct obd_client_handle *och) | |
119 | { | |
120 | struct obd_export *exp = ll_i2mdexp(inode); | |
121 | struct md_op_data *op_data; | |
122 | struct ptlrpc_request *req = NULL; | |
123 | struct obd_device *obd = class_exp2obd(exp); | |
124 | int epoch_close = 1; | |
125 | int rc; | |
126 | ENTRY; | |
127 | ||
128 | if (obd == NULL) { | |
129 | /* | |
130 | * XXX: in case of LMV, is this correct to access | |
131 | * ->exp_handle? | |
132 | */ | |
133 | CERROR("Invalid MDC connection handle "LPX64"\n", | |
134 | ll_i2mdexp(inode)->exp_handle.h_cookie); | |
135 | GOTO(out, rc = 0); | |
136 | } | |
137 | ||
138 | OBD_ALLOC_PTR(op_data); | |
139 | if (op_data == NULL) | |
140 | GOTO(out, rc = -ENOMEM); // XXX We leak openhandle and request here. | |
141 | ||
142 | ll_prepare_close(inode, op_data, och); | |
143 | epoch_close = (op_data->op_flags & MF_EPOCH_CLOSE); | |
144 | rc = md_close(md_exp, op_data, och->och_mod, &req); | |
145 | if (rc == -EAGAIN) { | |
146 | /* This close must have the epoch closed. */ | |
147 | LASSERT(epoch_close); | |
148 | /* MDS has instructed us to obtain Size-on-MDS attribute from | |
149 | * OSTs and send setattr to back to MDS. */ | |
150 | rc = ll_som_update(inode, op_data); | |
151 | if (rc) { | |
152 | CERROR("inode %lu mdc Size-on-MDS update failed: " | |
153 | "rc = %d\n", inode->i_ino, rc); | |
154 | rc = 0; | |
155 | } | |
156 | } else if (rc) { | |
157 | CERROR("inode %lu mdc close failed: rc = %d\n", | |
158 | inode->i_ino, rc); | |
159 | } | |
160 | ||
161 | /* DATA_MODIFIED flag was successfully sent on close, cancel data | |
162 | * modification flag. */ | |
163 | if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) { | |
164 | struct ll_inode_info *lli = ll_i2info(inode); | |
165 | ||
166 | spin_lock(&lli->lli_lock); | |
167 | lli->lli_flags &= ~LLIF_DATA_MODIFIED; | |
168 | spin_unlock(&lli->lli_lock); | |
169 | } | |
170 | ||
171 | ll_finish_md_op_data(op_data); | |
172 | ||
173 | if (rc == 0) { | |
174 | rc = ll_objects_destroy(req, inode); | |
175 | if (rc) | |
176 | CERROR("inode %lu ll_objects destroy: rc = %d\n", | |
177 | inode->i_ino, rc); | |
178 | } | |
179 | ||
180 | EXIT; | |
181 | out: | |
182 | ||
183 | if (exp_connect_som(exp) && !epoch_close && | |
184 | S_ISREG(inode->i_mode) && (och->och_flags & FMODE_WRITE)) { | |
185 | ll_queue_done_writing(inode, LLIF_DONE_WRITING); | |
186 | } else { | |
187 | md_clear_open_replay_data(md_exp, och); | |
188 | /* Free @och if it is not waiting for DONE_WRITING. */ | |
189 | och->och_fh.cookie = DEAD_HANDLE_MAGIC; | |
190 | OBD_FREE_PTR(och); | |
191 | } | |
192 | if (req) /* This is close request */ | |
193 | ptlrpc_req_finished(req); | |
194 | return rc; | |
195 | } | |
196 | ||
197 | int ll_md_real_close(struct inode *inode, int flags) | |
198 | { | |
199 | struct ll_inode_info *lli = ll_i2info(inode); | |
200 | struct obd_client_handle **och_p; | |
201 | struct obd_client_handle *och; | |
202 | __u64 *och_usecount; | |
203 | int rc = 0; | |
204 | ENTRY; | |
205 | ||
206 | if (flags & FMODE_WRITE) { | |
207 | och_p = &lli->lli_mds_write_och; | |
208 | och_usecount = &lli->lli_open_fd_write_count; | |
209 | } else if (flags & FMODE_EXEC) { | |
210 | och_p = &lli->lli_mds_exec_och; | |
211 | och_usecount = &lli->lli_open_fd_exec_count; | |
212 | } else { | |
213 | LASSERT(flags & FMODE_READ); | |
214 | och_p = &lli->lli_mds_read_och; | |
215 | och_usecount = &lli->lli_open_fd_read_count; | |
216 | } | |
217 | ||
218 | mutex_lock(&lli->lli_och_mutex); | |
219 | if (*och_usecount) { /* There are still users of this handle, so | |
220 | skip freeing it. */ | |
221 | mutex_unlock(&lli->lli_och_mutex); | |
222 | RETURN(0); | |
223 | } | |
224 | och=*och_p; | |
225 | *och_p = NULL; | |
226 | mutex_unlock(&lli->lli_och_mutex); | |
227 | ||
228 | if (och) { /* There might be a race and somebody have freed this och | |
229 | already */ | |
230 | rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, | |
231 | inode, och); | |
232 | } | |
233 | ||
234 | RETURN(rc); | |
235 | } | |
236 | ||
237 | int ll_md_close(struct obd_export *md_exp, struct inode *inode, | |
238 | struct file *file) | |
239 | { | |
240 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
241 | struct ll_inode_info *lli = ll_i2info(inode); | |
242 | int rc = 0; | |
243 | ENTRY; | |
244 | ||
245 | /* clear group lock, if present */ | |
246 | if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED)) | |
247 | ll_put_grouplock(inode, file, fd->fd_grouplock.cg_gid); | |
248 | ||
249 | /* Let's see if we have good enough OPEN lock on the file and if | |
250 | we can skip talking to MDS */ | |
251 | if (file->f_dentry->d_inode) { /* Can this ever be false? */ | |
252 | int lockmode; | |
253 | int flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK; | |
254 | struct lustre_handle lockh; | |
255 | struct inode *inode = file->f_dentry->d_inode; | |
256 | ldlm_policy_data_t policy = {.l_inodebits={MDS_INODELOCK_OPEN}}; | |
257 | ||
258 | mutex_lock(&lli->lli_och_mutex); | |
259 | if (fd->fd_omode & FMODE_WRITE) { | |
260 | lockmode = LCK_CW; | |
261 | LASSERT(lli->lli_open_fd_write_count); | |
262 | lli->lli_open_fd_write_count--; | |
263 | } else if (fd->fd_omode & FMODE_EXEC) { | |
264 | lockmode = LCK_PR; | |
265 | LASSERT(lli->lli_open_fd_exec_count); | |
266 | lli->lli_open_fd_exec_count--; | |
267 | } else { | |
268 | lockmode = LCK_CR; | |
269 | LASSERT(lli->lli_open_fd_read_count); | |
270 | lli->lli_open_fd_read_count--; | |
271 | } | |
272 | mutex_unlock(&lli->lli_och_mutex); | |
273 | ||
274 | if (!md_lock_match(md_exp, flags, ll_inode2fid(inode), | |
275 | LDLM_IBITS, &policy, lockmode, | |
276 | &lockh)) { | |
277 | rc = ll_md_real_close(file->f_dentry->d_inode, | |
278 | fd->fd_omode); | |
279 | } | |
280 | } else { | |
281 | CERROR("Releasing a file %p with negative dentry %p. Name %s", | |
282 | file, file->f_dentry, file->f_dentry->d_name.name); | |
283 | } | |
284 | ||
285 | LUSTRE_FPRIVATE(file) = NULL; | |
286 | ll_file_data_put(fd); | |
287 | ll_capa_close(inode); | |
288 | ||
289 | RETURN(rc); | |
290 | } | |
291 | ||
292 | /* While this returns an error code, fput() the caller does not, so we need | |
293 | * to make every effort to clean up all of our state here. Also, applications | |
294 | * rarely check close errors and even if an error is returned they will not | |
295 | * re-try the close call. | |
296 | */ | |
297 | int ll_file_release(struct inode *inode, struct file *file) | |
298 | { | |
299 | struct ll_file_data *fd; | |
300 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
301 | struct ll_inode_info *lli = ll_i2info(inode); | |
302 | int rc; | |
303 | ENTRY; | |
304 | ||
305 | CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino, | |
306 | inode->i_generation, inode); | |
307 | ||
308 | #ifdef CONFIG_FS_POSIX_ACL | |
309 | if (sbi->ll_flags & LL_SBI_RMT_CLIENT && | |
310 | inode == inode->i_sb->s_root->d_inode) { | |
311 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
312 | ||
313 | LASSERT(fd != NULL); | |
314 | if (unlikely(fd->fd_flags & LL_FILE_RMTACL)) { | |
315 | fd->fd_flags &= ~LL_FILE_RMTACL; | |
316 | rct_del(&sbi->ll_rct, current_pid()); | |
317 | et_search_free(&sbi->ll_et, current_pid()); | |
318 | } | |
319 | } | |
320 | #endif | |
321 | ||
322 | if (inode->i_sb->s_root != file->f_dentry) | |
323 | ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1); | |
324 | fd = LUSTRE_FPRIVATE(file); | |
325 | LASSERT(fd != NULL); | |
326 | ||
327 | /* The last ref on @file, maybe not the the owner pid of statahead. | |
328 | * Different processes can open the same dir, "ll_opendir_key" means: | |
329 | * it is me that should stop the statahead thread. */ | |
330 | if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd && | |
331 | lli->lli_opendir_pid != 0) | |
332 | ll_stop_statahead(inode, lli->lli_opendir_key); | |
333 | ||
334 | if (inode->i_sb->s_root == file->f_dentry) { | |
335 | LUSTRE_FPRIVATE(file) = NULL; | |
336 | ll_file_data_put(fd); | |
337 | RETURN(0); | |
338 | } | |
339 | ||
340 | if (!S_ISDIR(inode->i_mode)) { | |
341 | lov_read_and_clear_async_rc(lli->lli_clob); | |
342 | lli->lli_async_rc = 0; | |
343 | } | |
344 | ||
345 | rc = ll_md_close(sbi->ll_md_exp, inode, file); | |
346 | ||
347 | if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val)) | |
348 | libcfs_debug_dumplog(); | |
349 | ||
350 | RETURN(rc); | |
351 | } | |
352 | ||
353 | static int ll_intent_file_open(struct file *file, void *lmm, | |
354 | int lmmsize, struct lookup_intent *itp) | |
355 | { | |
356 | struct ll_sb_info *sbi = ll_i2sbi(file->f_dentry->d_inode); | |
357 | struct dentry *parent = file->f_dentry->d_parent; | |
358 | const char *name = file->f_dentry->d_name.name; | |
359 | const int len = file->f_dentry->d_name.len; | |
360 | struct md_op_data *op_data; | |
361 | struct ptlrpc_request *req; | |
362 | __u32 opc = LUSTRE_OPC_ANY; | |
363 | int rc; | |
364 | ENTRY; | |
365 | ||
366 | if (!parent) | |
367 | RETURN(-ENOENT); | |
368 | ||
369 | /* Usually we come here only for NFSD, and we want open lock. | |
370 | But we can also get here with pre 2.6.15 patchless kernels, and in | |
371 | that case that lock is also ok */ | |
372 | /* We can also get here if there was cached open handle in revalidate_it | |
373 | * but it disappeared while we were getting from there to ll_file_open. | |
374 | * But this means this file was closed and immediatelly opened which | |
375 | * makes a good candidate for using OPEN lock */ | |
376 | /* If lmmsize & lmm are not 0, we are just setting stripe info | |
377 | * parameters. No need for the open lock */ | |
378 | if (lmm == NULL && lmmsize == 0) { | |
379 | itp->it_flags |= MDS_OPEN_LOCK; | |
380 | if (itp->it_flags & FMODE_WRITE) | |
381 | opc = LUSTRE_OPC_CREATE; | |
382 | } | |
383 | ||
384 | op_data = ll_prep_md_op_data(NULL, parent->d_inode, | |
385 | file->f_dentry->d_inode, name, len, | |
386 | O_RDWR, opc, NULL); | |
387 | if (IS_ERR(op_data)) | |
388 | RETURN(PTR_ERR(op_data)); | |
389 | ||
390 | itp->it_flags |= MDS_OPEN_BY_FID; | |
391 | rc = md_intent_lock(sbi->ll_md_exp, op_data, lmm, lmmsize, itp, | |
392 | 0 /*unused */, &req, ll_md_blocking_ast, 0); | |
393 | ll_finish_md_op_data(op_data); | |
394 | if (rc == -ESTALE) { | |
395 | /* reason for keep own exit path - don`t flood log | |
396 | * with messages with -ESTALE errors. | |
397 | */ | |
398 | if (!it_disposition(itp, DISP_OPEN_OPEN) || | |
399 | it_open_error(DISP_OPEN_OPEN, itp)) | |
400 | GOTO(out, rc); | |
401 | ll_release_openhandle(file->f_dentry, itp); | |
402 | GOTO(out, rc); | |
403 | } | |
404 | ||
405 | if (it_disposition(itp, DISP_LOOKUP_NEG)) | |
406 | GOTO(out, rc = -ENOENT); | |
407 | ||
408 | if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) { | |
409 | rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp); | |
410 | CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc); | |
411 | GOTO(out, rc); | |
412 | } | |
413 | ||
414 | rc = ll_prep_inode(&file->f_dentry->d_inode, req, NULL, itp); | |
415 | if (!rc && itp->d.lustre.it_lock_mode) | |
416 | ll_set_lock_data(sbi->ll_md_exp, file->f_dentry->d_inode, | |
417 | itp, NULL); | |
418 | ||
419 | out: | |
420 | ptlrpc_req_finished(itp->d.lustre.it_data); | |
421 | it_clear_disposition(itp, DISP_ENQ_COMPLETE); | |
422 | ll_intent_drop_lock(itp); | |
423 | ||
424 | RETURN(rc); | |
425 | } | |
426 | ||
427 | /** | |
428 | * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does | |
429 | * not believe attributes if a few ioepoch holders exist. Attributes for | |
430 | * previous ioepoch if new one is opened are also skipped by MDS. | |
431 | */ | |
432 | void ll_ioepoch_open(struct ll_inode_info *lli, __u64 ioepoch) | |
433 | { | |
434 | if (ioepoch && lli->lli_ioepoch != ioepoch) { | |
435 | lli->lli_ioepoch = ioepoch; | |
436 | CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID"\n", | |
437 | ioepoch, PFID(&lli->lli_fid)); | |
438 | } | |
439 | } | |
440 | ||
441 | static int ll_och_fill(struct obd_export *md_exp, struct ll_inode_info *lli, | |
442 | struct lookup_intent *it, struct obd_client_handle *och) | |
443 | { | |
444 | struct ptlrpc_request *req = it->d.lustre.it_data; | |
445 | struct mdt_body *body; | |
446 | ||
447 | LASSERT(och); | |
448 | ||
449 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); | |
450 | LASSERT(body != NULL); /* reply already checked out */ | |
451 | ||
452 | memcpy(&och->och_fh, &body->handle, sizeof(body->handle)); | |
453 | och->och_magic = OBD_CLIENT_HANDLE_MAGIC; | |
454 | och->och_fid = lli->lli_fid; | |
455 | och->och_flags = it->it_flags; | |
456 | ll_ioepoch_open(lli, body->ioepoch); | |
457 | ||
458 | return md_set_open_replay_data(md_exp, och, req); | |
459 | } | |
460 | ||
461 | int ll_local_open(struct file *file, struct lookup_intent *it, | |
462 | struct ll_file_data *fd, struct obd_client_handle *och) | |
463 | { | |
464 | struct inode *inode = file->f_dentry->d_inode; | |
465 | struct ll_inode_info *lli = ll_i2info(inode); | |
466 | ENTRY; | |
467 | ||
468 | LASSERT(!LUSTRE_FPRIVATE(file)); | |
469 | ||
470 | LASSERT(fd != NULL); | |
471 | ||
472 | if (och) { | |
473 | struct ptlrpc_request *req = it->d.lustre.it_data; | |
474 | struct mdt_body *body; | |
475 | int rc; | |
476 | ||
477 | rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, lli, it, och); | |
478 | if (rc) | |
479 | RETURN(rc); | |
480 | ||
481 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); | |
482 | if ((it->it_flags & FMODE_WRITE) && | |
483 | (body->valid & OBD_MD_FLSIZE)) | |
484 | CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID"\n", | |
485 | lli->lli_ioepoch, PFID(&lli->lli_fid)); | |
486 | } | |
487 | ||
488 | LUSTRE_FPRIVATE(file) = fd; | |
489 | ll_readahead_init(inode, &fd->fd_ras); | |
490 | fd->fd_omode = it->it_flags; | |
491 | RETURN(0); | |
492 | } | |
493 | ||
494 | /* Open a file, and (for the very first open) create objects on the OSTs at | |
495 | * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object | |
496 | * creation or open until ll_lov_setstripe() ioctl is called. | |
497 | * | |
498 | * If we already have the stripe MD locally then we don't request it in | |
499 | * md_open(), by passing a lmm_size = 0. | |
500 | * | |
501 | * It is up to the application to ensure no other processes open this file | |
502 | * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be | |
503 | * used. We might be able to avoid races of that sort by getting lli_open_sem | |
504 | * before returning in the O_LOV_DELAY_CREATE case and dropping it here | |
505 | * or in ll_file_release(), but I'm not sure that is desirable/necessary. | |
506 | */ | |
507 | int ll_file_open(struct inode *inode, struct file *file) | |
508 | { | |
509 | struct ll_inode_info *lli = ll_i2info(inode); | |
510 | struct lookup_intent *it, oit = { .it_op = IT_OPEN, | |
511 | .it_flags = file->f_flags }; | |
512 | struct obd_client_handle **och_p = NULL; | |
513 | __u64 *och_usecount = NULL; | |
514 | struct ll_file_data *fd; | |
515 | int rc = 0, opendir_set = 0; | |
516 | ENTRY; | |
517 | ||
518 | CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), flags %o\n", inode->i_ino, | |
519 | inode->i_generation, inode, file->f_flags); | |
520 | ||
521 | it = file->private_data; /* XXX: compat macro */ | |
522 | file->private_data = NULL; /* prevent ll_local_open assertion */ | |
523 | ||
524 | fd = ll_file_data_get(); | |
525 | if (fd == NULL) | |
526 | GOTO(out_och_free, rc = -ENOMEM); | |
527 | ||
528 | fd->fd_file = file; | |
529 | if (S_ISDIR(inode->i_mode)) { | |
530 | spin_lock(&lli->lli_sa_lock); | |
531 | if (lli->lli_opendir_key == NULL && lli->lli_sai == NULL && | |
532 | lli->lli_opendir_pid == 0) { | |
533 | lli->lli_opendir_key = fd; | |
534 | lli->lli_opendir_pid = current_pid(); | |
535 | opendir_set = 1; | |
536 | } | |
537 | spin_unlock(&lli->lli_sa_lock); | |
538 | } | |
539 | ||
540 | if (inode->i_sb->s_root == file->f_dentry) { | |
541 | LUSTRE_FPRIVATE(file) = fd; | |
542 | RETURN(0); | |
543 | } | |
544 | ||
545 | if (!it || !it->d.lustre.it_disposition) { | |
546 | /* Convert f_flags into access mode. We cannot use file->f_mode, | |
547 | * because everything but O_ACCMODE mask was stripped from | |
548 | * there */ | |
549 | if ((oit.it_flags + 1) & O_ACCMODE) | |
550 | oit.it_flags++; | |
551 | if (file->f_flags & O_TRUNC) | |
552 | oit.it_flags |= FMODE_WRITE; | |
553 | ||
554 | /* kernel only call f_op->open in dentry_open. filp_open calls | |
555 | * dentry_open after call to open_namei that checks permissions. | |
556 | * Only nfsd_open call dentry_open directly without checking | |
557 | * permissions and because of that this code below is safe. */ | |
558 | if (oit.it_flags & (FMODE_WRITE | FMODE_READ)) | |
559 | oit.it_flags |= MDS_OPEN_OWNEROVERRIDE; | |
560 | ||
561 | /* We do not want O_EXCL here, presumably we opened the file | |
562 | * already? XXX - NFS implications? */ | |
563 | oit.it_flags &= ~O_EXCL; | |
564 | ||
565 | /* bug20584, if "it_flags" contains O_CREAT, the file will be | |
566 | * created if necessary, then "IT_CREAT" should be set to keep | |
567 | * consistent with it */ | |
568 | if (oit.it_flags & O_CREAT) | |
569 | oit.it_op |= IT_CREAT; | |
570 | ||
571 | it = &oit; | |
572 | } | |
573 | ||
574 | restart: | |
575 | /* Let's see if we have file open on MDS already. */ | |
576 | if (it->it_flags & FMODE_WRITE) { | |
577 | och_p = &lli->lli_mds_write_och; | |
578 | och_usecount = &lli->lli_open_fd_write_count; | |
579 | } else if (it->it_flags & FMODE_EXEC) { | |
580 | och_p = &lli->lli_mds_exec_och; | |
581 | och_usecount = &lli->lli_open_fd_exec_count; | |
582 | } else { | |
583 | och_p = &lli->lli_mds_read_och; | |
584 | och_usecount = &lli->lli_open_fd_read_count; | |
585 | } | |
586 | ||
587 | mutex_lock(&lli->lli_och_mutex); | |
588 | if (*och_p) { /* Open handle is present */ | |
589 | if (it_disposition(it, DISP_OPEN_OPEN)) { | |
590 | /* Well, there's extra open request that we do not need, | |
591 | let's close it somehow. This will decref request. */ | |
592 | rc = it_open_error(DISP_OPEN_OPEN, it); | |
593 | if (rc) { | |
594 | mutex_unlock(&lli->lli_och_mutex); | |
595 | GOTO(out_openerr, rc); | |
596 | } | |
597 | ||
598 | ll_release_openhandle(file->f_dentry, it); | |
599 | } | |
600 | (*och_usecount)++; | |
601 | ||
602 | rc = ll_local_open(file, it, fd, NULL); | |
603 | if (rc) { | |
604 | (*och_usecount)--; | |
605 | mutex_unlock(&lli->lli_och_mutex); | |
606 | GOTO(out_openerr, rc); | |
607 | } | |
608 | } else { | |
609 | LASSERT(*och_usecount == 0); | |
610 | if (!it->d.lustre.it_disposition) { | |
611 | /* We cannot just request lock handle now, new ELC code | |
612 | means that one of other OPEN locks for this file | |
613 | could be cancelled, and since blocking ast handler | |
614 | would attempt to grab och_mutex as well, that would | |
615 | result in a deadlock */ | |
616 | mutex_unlock(&lli->lli_och_mutex); | |
617 | it->it_create_mode |= M_CHECK_STALE; | |
618 | rc = ll_intent_file_open(file, NULL, 0, it); | |
619 | it->it_create_mode &= ~M_CHECK_STALE; | |
620 | if (rc) | |
621 | GOTO(out_openerr, rc); | |
622 | ||
623 | goto restart; | |
624 | } | |
625 | OBD_ALLOC(*och_p, sizeof (struct obd_client_handle)); | |
626 | if (!*och_p) | |
627 | GOTO(out_och_free, rc = -ENOMEM); | |
628 | ||
629 | (*och_usecount)++; | |
630 | ||
631 | /* md_intent_lock() didn't get a request ref if there was an | |
632 | * open error, so don't do cleanup on the request here | |
633 | * (bug 3430) */ | |
634 | /* XXX (green): Should not we bail out on any error here, not | |
635 | * just open error? */ | |
636 | rc = it_open_error(DISP_OPEN_OPEN, it); | |
637 | if (rc) | |
638 | GOTO(out_och_free, rc); | |
639 | ||
640 | LASSERT(it_disposition(it, DISP_ENQ_OPEN_REF)); | |
641 | ||
642 | rc = ll_local_open(file, it, fd, *och_p); | |
643 | if (rc) | |
644 | GOTO(out_och_free, rc); | |
645 | } | |
646 | mutex_unlock(&lli->lli_och_mutex); | |
647 | fd = NULL; | |
648 | ||
649 | /* Must do this outside lli_och_mutex lock to prevent deadlock where | |
650 | different kind of OPEN lock for this same inode gets cancelled | |
651 | by ldlm_cancel_lru */ | |
652 | if (!S_ISREG(inode->i_mode)) | |
653 | GOTO(out_och_free, rc); | |
654 | ||
655 | ll_capa_open(inode); | |
656 | ||
657 | if (!lli->lli_has_smd) { | |
658 | if (file->f_flags & O_LOV_DELAY_CREATE || | |
659 | !(file->f_mode & FMODE_WRITE)) { | |
660 | CDEBUG(D_INODE, "object creation was delayed\n"); | |
661 | GOTO(out_och_free, rc); | |
662 | } | |
663 | } | |
664 | file->f_flags &= ~O_LOV_DELAY_CREATE; | |
665 | GOTO(out_och_free, rc); | |
666 | ||
667 | out_och_free: | |
668 | if (rc) { | |
669 | if (och_p && *och_p) { | |
670 | OBD_FREE(*och_p, sizeof (struct obd_client_handle)); | |
671 | *och_p = NULL; /* OBD_FREE writes some magic there */ | |
672 | (*och_usecount)--; | |
673 | } | |
674 | mutex_unlock(&lli->lli_och_mutex); | |
675 | ||
676 | out_openerr: | |
677 | if (opendir_set != 0) | |
678 | ll_stop_statahead(inode, lli->lli_opendir_key); | |
679 | if (fd != NULL) | |
680 | ll_file_data_put(fd); | |
681 | } else { | |
682 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1); | |
683 | } | |
684 | ||
685 | if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) { | |
686 | ptlrpc_req_finished(it->d.lustre.it_data); | |
687 | it_clear_disposition(it, DISP_ENQ_OPEN_REF); | |
688 | } | |
689 | ||
690 | return rc; | |
691 | } | |
692 | ||
693 | /* Fills the obdo with the attributes for the lsm */ | |
694 | static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp, | |
695 | struct obd_capa *capa, struct obdo *obdo, | |
696 | __u64 ioepoch, int sync) | |
697 | { | |
698 | struct ptlrpc_request_set *set; | |
699 | struct obd_info oinfo = { { { 0 } } }; | |
700 | int rc; | |
701 | ||
702 | ENTRY; | |
703 | ||
704 | LASSERT(lsm != NULL); | |
705 | ||
706 | oinfo.oi_md = lsm; | |
707 | oinfo.oi_oa = obdo; | |
708 | oinfo.oi_oa->o_oi = lsm->lsm_oi; | |
709 | oinfo.oi_oa->o_mode = S_IFREG; | |
710 | oinfo.oi_oa->o_ioepoch = ioepoch; | |
711 | oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | | |
712 | OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | | |
713 | OBD_MD_FLBLKSZ | OBD_MD_FLATIME | | |
714 | OBD_MD_FLMTIME | OBD_MD_FLCTIME | | |
715 | OBD_MD_FLGROUP | OBD_MD_FLEPOCH | | |
716 | OBD_MD_FLDATAVERSION; | |
717 | oinfo.oi_capa = capa; | |
718 | if (sync) { | |
719 | oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS; | |
720 | oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK; | |
721 | } | |
722 | ||
723 | set = ptlrpc_prep_set(); | |
724 | if (set == NULL) { | |
725 | CERROR("can't allocate ptlrpc set\n"); | |
726 | rc = -ENOMEM; | |
727 | } else { | |
728 | rc = obd_getattr_async(exp, &oinfo, set); | |
729 | if (rc == 0) | |
730 | rc = ptlrpc_set_wait(set); | |
731 | ptlrpc_set_destroy(set); | |
732 | } | |
733 | if (rc == 0) | |
734 | oinfo.oi_oa->o_valid &= (OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ | | |
735 | OBD_MD_FLATIME | OBD_MD_FLMTIME | | |
736 | OBD_MD_FLCTIME | OBD_MD_FLSIZE | | |
737 | OBD_MD_FLDATAVERSION); | |
738 | RETURN(rc); | |
739 | } | |
740 | ||
741 | /** | |
742 | * Performs the getattr on the inode and updates its fields. | |
743 | * If @sync != 0, perform the getattr under the server-side lock. | |
744 | */ | |
745 | int ll_inode_getattr(struct inode *inode, struct obdo *obdo, | |
746 | __u64 ioepoch, int sync) | |
747 | { | |
748 | struct obd_capa *capa = ll_mdscapa_get(inode); | |
749 | struct lov_stripe_md *lsm; | |
750 | int rc; | |
751 | ENTRY; | |
752 | ||
753 | lsm = ccc_inode_lsm_get(inode); | |
754 | rc = ll_lsm_getattr(lsm, ll_i2dtexp(inode), | |
755 | capa, obdo, ioepoch, sync); | |
756 | capa_put(capa); | |
757 | if (rc == 0) { | |
758 | struct ost_id *oi = lsm ? &lsm->lsm_oi : &obdo->o_oi; | |
759 | ||
760 | obdo_refresh_inode(inode, obdo, obdo->o_valid); | |
761 | CDEBUG(D_INODE, "objid "DOSTID" size %llu, blocks %llu," | |
762 | " blksize %lu\n", POSTID(oi), i_size_read(inode), | |
763 | (unsigned long long)inode->i_blocks, | |
764 | (unsigned long)ll_inode_blksize(inode)); | |
765 | } | |
766 | ccc_inode_lsm_put(inode, lsm); | |
767 | RETURN(rc); | |
768 | } | |
769 | ||
770 | int ll_merge_lvb(const struct lu_env *env, struct inode *inode) | |
771 | { | |
772 | struct ll_inode_info *lli = ll_i2info(inode); | |
773 | struct cl_object *obj = lli->lli_clob; | |
774 | struct cl_attr *attr = ccc_env_thread_attr(env); | |
775 | struct ost_lvb lvb; | |
776 | int rc = 0; | |
777 | ||
778 | ENTRY; | |
779 | ||
780 | ll_inode_size_lock(inode); | |
781 | /* merge timestamps the most recently obtained from mds with | |
782 | timestamps obtained from osts */ | |
783 | LTIME_S(inode->i_atime) = lli->lli_lvb.lvb_atime; | |
784 | LTIME_S(inode->i_mtime) = lli->lli_lvb.lvb_mtime; | |
785 | LTIME_S(inode->i_ctime) = lli->lli_lvb.lvb_ctime; | |
786 | inode_init_lvb(inode, &lvb); | |
787 | ||
788 | cl_object_attr_lock(obj); | |
789 | rc = cl_object_attr_get(env, obj, attr); | |
790 | cl_object_attr_unlock(obj); | |
791 | ||
792 | if (rc == 0) { | |
793 | if (lvb.lvb_atime < attr->cat_atime) | |
794 | lvb.lvb_atime = attr->cat_atime; | |
795 | if (lvb.lvb_ctime < attr->cat_ctime) | |
796 | lvb.lvb_ctime = attr->cat_ctime; | |
797 | if (lvb.lvb_mtime < attr->cat_mtime) | |
798 | lvb.lvb_mtime = attr->cat_mtime; | |
799 | ||
800 | CDEBUG(D_VFSTRACE, DFID" updating i_size "LPU64"\n", | |
801 | PFID(&lli->lli_fid), attr->cat_size); | |
802 | cl_isize_write_nolock(inode, attr->cat_size); | |
803 | ||
804 | inode->i_blocks = attr->cat_blocks; | |
805 | ||
806 | LTIME_S(inode->i_mtime) = lvb.lvb_mtime; | |
807 | LTIME_S(inode->i_atime) = lvb.lvb_atime; | |
808 | LTIME_S(inode->i_ctime) = lvb.lvb_ctime; | |
809 | } | |
810 | ll_inode_size_unlock(inode); | |
811 | ||
812 | RETURN(rc); | |
813 | } | |
814 | ||
815 | int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm, | |
816 | lstat_t *st) | |
817 | { | |
818 | struct obdo obdo = { 0 }; | |
819 | int rc; | |
820 | ||
821 | rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, &obdo, 0, 0); | |
822 | if (rc == 0) { | |
823 | st->st_size = obdo.o_size; | |
824 | st->st_blocks = obdo.o_blocks; | |
825 | st->st_mtime = obdo.o_mtime; | |
826 | st->st_atime = obdo.o_atime; | |
827 | st->st_ctime = obdo.o_ctime; | |
828 | } | |
829 | return rc; | |
830 | } | |
831 | ||
832 | void ll_io_init(struct cl_io *io, const struct file *file, int write) | |
833 | { | |
834 | struct inode *inode = file->f_dentry->d_inode; | |
835 | ||
836 | io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK; | |
837 | if (write) { | |
838 | io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND); | |
839 | io->u.ci_wr.wr_sync = file->f_flags & O_SYNC || | |
840 | file->f_flags & O_DIRECT || | |
841 | IS_SYNC(inode); | |
842 | } | |
843 | io->ci_obj = ll_i2info(inode)->lli_clob; | |
844 | io->ci_lockreq = CILR_MAYBE; | |
845 | if (ll_file_nolock(file)) { | |
846 | io->ci_lockreq = CILR_NEVER; | |
847 | io->ci_no_srvlock = 1; | |
848 | } else if (file->f_flags & O_APPEND) { | |
849 | io->ci_lockreq = CILR_MANDATORY; | |
850 | } | |
851 | } | |
852 | ||
853 | static ssize_t | |
854 | ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args, | |
855 | struct file *file, enum cl_io_type iot, | |
856 | loff_t *ppos, size_t count) | |
857 | { | |
858 | struct ll_inode_info *lli = ll_i2info(file->f_dentry->d_inode); | |
859 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
860 | struct cl_io *io; | |
861 | ssize_t result; | |
862 | ENTRY; | |
863 | ||
864 | restart: | |
865 | io = ccc_env_thread_io(env); | |
866 | ll_io_init(io, file, iot == CIT_WRITE); | |
867 | ||
868 | if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) { | |
869 | struct vvp_io *vio = vvp_env_io(env); | |
870 | struct ccc_io *cio = ccc_env_io(env); | |
871 | int write_mutex_locked = 0; | |
872 | ||
873 | cio->cui_fd = LUSTRE_FPRIVATE(file); | |
874 | vio->cui_io_subtype = args->via_io_subtype; | |
875 | ||
876 | switch (vio->cui_io_subtype) { | |
877 | case IO_NORMAL: | |
878 | cio->cui_iov = args->u.normal.via_iov; | |
879 | cio->cui_nrsegs = args->u.normal.via_nrsegs; | |
880 | cio->cui_tot_nrsegs = cio->cui_nrsegs; | |
881 | cio->cui_iocb = args->u.normal.via_iocb; | |
882 | if ((iot == CIT_WRITE) && | |
883 | !(cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) { | |
884 | if (mutex_lock_interruptible(&lli-> | |
885 | lli_write_mutex)) | |
886 | GOTO(out, result = -ERESTARTSYS); | |
887 | write_mutex_locked = 1; | |
888 | } else if (iot == CIT_READ) { | |
889 | down_read(&lli->lli_trunc_sem); | |
890 | } | |
891 | break; | |
892 | case IO_SENDFILE: | |
893 | vio->u.sendfile.cui_actor = args->u.sendfile.via_actor; | |
894 | vio->u.sendfile.cui_target = args->u.sendfile.via_target; | |
895 | break; | |
896 | case IO_SPLICE: | |
897 | vio->u.splice.cui_pipe = args->u.splice.via_pipe; | |
898 | vio->u.splice.cui_flags = args->u.splice.via_flags; | |
899 | break; | |
900 | default: | |
901 | CERROR("Unknow IO type - %u\n", vio->cui_io_subtype); | |
902 | LBUG(); | |
903 | } | |
904 | result = cl_io_loop(env, io); | |
905 | if (write_mutex_locked) | |
906 | mutex_unlock(&lli->lli_write_mutex); | |
907 | else if (args->via_io_subtype == IO_NORMAL && iot == CIT_READ) | |
908 | up_read(&lli->lli_trunc_sem); | |
909 | } else { | |
910 | /* cl_io_rw_init() handled IO */ | |
911 | result = io->ci_result; | |
912 | } | |
913 | ||
914 | if (io->ci_nob > 0) { | |
915 | result = io->ci_nob; | |
916 | *ppos = io->u.ci_wr.wr.crw_pos; | |
917 | } | |
918 | GOTO(out, result); | |
919 | out: | |
920 | cl_io_fini(env, io); | |
921 | /* If any bit been read/written (result != 0), we just return | |
922 | * short read/write instead of restart io. */ | |
923 | if (result == 0 && io->ci_need_restart) { | |
924 | CDEBUG(D_VFSTRACE, "Restart %s on %s from %lld, count:%zd\n", | |
925 | iot == CIT_READ ? "read" : "write", | |
926 | file->f_dentry->d_name.name, *ppos, count); | |
927 | LASSERTF(io->ci_nob == 0, "%zd", io->ci_nob); | |
928 | goto restart; | |
929 | } | |
930 | ||
931 | if (iot == CIT_READ) { | |
932 | if (result >= 0) | |
933 | ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode), | |
934 | LPROC_LL_READ_BYTES, result); | |
935 | } else if (iot == CIT_WRITE) { | |
936 | if (result >= 0) { | |
937 | ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode), | |
938 | LPROC_LL_WRITE_BYTES, result); | |
939 | fd->fd_write_failed = false; | |
940 | } else if (result != -ERESTARTSYS) { | |
941 | fd->fd_write_failed = true; | |
942 | } | |
943 | } | |
944 | ||
945 | return result; | |
946 | } | |
947 | ||
948 | ||
949 | /* | |
950 | * XXX: exact copy from kernel code (__generic_file_aio_write_nolock) | |
951 | */ | |
952 | static int ll_file_get_iov_count(const struct iovec *iov, | |
953 | unsigned long *nr_segs, size_t *count) | |
954 | { | |
955 | size_t cnt = 0; | |
956 | unsigned long seg; | |
957 | ||
958 | for (seg = 0; seg < *nr_segs; seg++) { | |
959 | const struct iovec *iv = &iov[seg]; | |
960 | ||
961 | /* | |
962 | * If any segment has a negative length, or the cumulative | |
963 | * length ever wraps negative then return -EINVAL. | |
964 | */ | |
965 | cnt += iv->iov_len; | |
966 | if (unlikely((ssize_t)(cnt|iv->iov_len) < 0)) | |
967 | return -EINVAL; | |
968 | if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len)) | |
969 | continue; | |
970 | if (seg == 0) | |
971 | return -EFAULT; | |
972 | *nr_segs = seg; | |
973 | cnt -= iv->iov_len; /* This segment is no good */ | |
974 | break; | |
975 | } | |
976 | *count = cnt; | |
977 | return 0; | |
978 | } | |
979 | ||
980 | static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov, | |
981 | unsigned long nr_segs, loff_t pos) | |
982 | { | |
983 | struct lu_env *env; | |
984 | struct vvp_io_args *args; | |
985 | size_t count; | |
986 | ssize_t result; | |
987 | int refcheck; | |
988 | ENTRY; | |
989 | ||
990 | result = ll_file_get_iov_count(iov, &nr_segs, &count); | |
991 | if (result) | |
992 | RETURN(result); | |
993 | ||
994 | env = cl_env_get(&refcheck); | |
995 | if (IS_ERR(env)) | |
996 | RETURN(PTR_ERR(env)); | |
997 | ||
998 | args = vvp_env_args(env, IO_NORMAL); | |
999 | args->u.normal.via_iov = (struct iovec *)iov; | |
1000 | args->u.normal.via_nrsegs = nr_segs; | |
1001 | args->u.normal.via_iocb = iocb; | |
1002 | ||
1003 | result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ, | |
1004 | &iocb->ki_pos, count); | |
1005 | cl_env_put(env, &refcheck); | |
1006 | RETURN(result); | |
1007 | } | |
1008 | ||
1009 | static ssize_t ll_file_read(struct file *file, char *buf, size_t count, | |
1010 | loff_t *ppos) | |
1011 | { | |
1012 | struct lu_env *env; | |
1013 | struct iovec *local_iov; | |
1014 | struct kiocb *kiocb; | |
1015 | ssize_t result; | |
1016 | int refcheck; | |
1017 | ENTRY; | |
1018 | ||
1019 | env = cl_env_get(&refcheck); | |
1020 | if (IS_ERR(env)) | |
1021 | RETURN(PTR_ERR(env)); | |
1022 | ||
1023 | local_iov = &vvp_env_info(env)->vti_local_iov; | |
1024 | kiocb = &vvp_env_info(env)->vti_kiocb; | |
1025 | local_iov->iov_base = (void __user *)buf; | |
1026 | local_iov->iov_len = count; | |
1027 | init_sync_kiocb(kiocb, file); | |
1028 | kiocb->ki_pos = *ppos; | |
1029 | kiocb->ki_left = count; | |
1030 | ||
1031 | result = ll_file_aio_read(kiocb, local_iov, 1, kiocb->ki_pos); | |
1032 | *ppos = kiocb->ki_pos; | |
1033 | ||
1034 | cl_env_put(env, &refcheck); | |
1035 | RETURN(result); | |
1036 | } | |
1037 | ||
1038 | /* | |
1039 | * Write to a file (through the page cache). | |
1040 | */ | |
1041 | static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
1042 | unsigned long nr_segs, loff_t pos) | |
1043 | { | |
1044 | struct lu_env *env; | |
1045 | struct vvp_io_args *args; | |
1046 | size_t count; | |
1047 | ssize_t result; | |
1048 | int refcheck; | |
1049 | ENTRY; | |
1050 | ||
1051 | result = ll_file_get_iov_count(iov, &nr_segs, &count); | |
1052 | if (result) | |
1053 | RETURN(result); | |
1054 | ||
1055 | env = cl_env_get(&refcheck); | |
1056 | if (IS_ERR(env)) | |
1057 | RETURN(PTR_ERR(env)); | |
1058 | ||
1059 | args = vvp_env_args(env, IO_NORMAL); | |
1060 | args->u.normal.via_iov = (struct iovec *)iov; | |
1061 | args->u.normal.via_nrsegs = nr_segs; | |
1062 | args->u.normal.via_iocb = iocb; | |
1063 | ||
1064 | result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE, | |
1065 | &iocb->ki_pos, count); | |
1066 | cl_env_put(env, &refcheck); | |
1067 | RETURN(result); | |
1068 | } | |
1069 | ||
1070 | static ssize_t ll_file_write(struct file *file, const char *buf, size_t count, | |
1071 | loff_t *ppos) | |
1072 | { | |
1073 | struct lu_env *env; | |
1074 | struct iovec *local_iov; | |
1075 | struct kiocb *kiocb; | |
1076 | ssize_t result; | |
1077 | int refcheck; | |
1078 | ENTRY; | |
1079 | ||
1080 | env = cl_env_get(&refcheck); | |
1081 | if (IS_ERR(env)) | |
1082 | RETURN(PTR_ERR(env)); | |
1083 | ||
1084 | local_iov = &vvp_env_info(env)->vti_local_iov; | |
1085 | kiocb = &vvp_env_info(env)->vti_kiocb; | |
1086 | local_iov->iov_base = (void __user *)buf; | |
1087 | local_iov->iov_len = count; | |
1088 | init_sync_kiocb(kiocb, file); | |
1089 | kiocb->ki_pos = *ppos; | |
1090 | kiocb->ki_left = count; | |
1091 | ||
1092 | result = ll_file_aio_write(kiocb, local_iov, 1, kiocb->ki_pos); | |
1093 | *ppos = kiocb->ki_pos; | |
1094 | ||
1095 | cl_env_put(env, &refcheck); | |
1096 | RETURN(result); | |
1097 | } | |
1098 | ||
1099 | ||
1100 | ||
1101 | /* | |
1102 | * Send file content (through pagecache) somewhere with helper | |
1103 | */ | |
1104 | static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos, | |
1105 | struct pipe_inode_info *pipe, size_t count, | |
1106 | unsigned int flags) | |
1107 | { | |
1108 | struct lu_env *env; | |
1109 | struct vvp_io_args *args; | |
1110 | ssize_t result; | |
1111 | int refcheck; | |
1112 | ENTRY; | |
1113 | ||
1114 | env = cl_env_get(&refcheck); | |
1115 | if (IS_ERR(env)) | |
1116 | RETURN(PTR_ERR(env)); | |
1117 | ||
1118 | args = vvp_env_args(env, IO_SPLICE); | |
1119 | args->u.splice.via_pipe = pipe; | |
1120 | args->u.splice.via_flags = flags; | |
1121 | ||
1122 | result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count); | |
1123 | cl_env_put(env, &refcheck); | |
1124 | RETURN(result); | |
1125 | } | |
1126 | ||
1127 | static int ll_lov_recreate(struct inode *inode, struct ost_id *oi, | |
1128 | obd_count ost_idx) | |
1129 | { | |
1130 | struct obd_export *exp = ll_i2dtexp(inode); | |
1131 | struct obd_trans_info oti = { 0 }; | |
1132 | struct obdo *oa = NULL; | |
1133 | int lsm_size; | |
1134 | int rc = 0; | |
1135 | struct lov_stripe_md *lsm = NULL, *lsm2; | |
1136 | ENTRY; | |
1137 | ||
1138 | OBDO_ALLOC(oa); | |
1139 | if (oa == NULL) | |
1140 | RETURN(-ENOMEM); | |
1141 | ||
1142 | lsm = ccc_inode_lsm_get(inode); | |
1143 | if (lsm == NULL) | |
1144 | GOTO(out, rc = -ENOENT); | |
1145 | ||
1146 | lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) * | |
1147 | (lsm->lsm_stripe_count)); | |
1148 | ||
1149 | OBD_ALLOC_LARGE(lsm2, lsm_size); | |
1150 | if (lsm2 == NULL) | |
1151 | GOTO(out, rc = -ENOMEM); | |
1152 | ||
1153 | oa->o_oi = *oi; | |
1154 | oa->o_nlink = ost_idx; | |
1155 | oa->o_flags |= OBD_FL_RECREATE_OBJS; | |
1156 | oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP; | |
1157 | obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME | | |
1158 | OBD_MD_FLMTIME | OBD_MD_FLCTIME); | |
1159 | obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid); | |
1160 | memcpy(lsm2, lsm, lsm_size); | |
1161 | ll_inode_size_lock(inode); | |
1162 | rc = obd_create(NULL, exp, oa, &lsm2, &oti); | |
1163 | ll_inode_size_unlock(inode); | |
1164 | ||
1165 | OBD_FREE_LARGE(lsm2, lsm_size); | |
1166 | GOTO(out, rc); | |
1167 | out: | |
1168 | ccc_inode_lsm_put(inode, lsm); | |
1169 | OBDO_FREE(oa); | |
1170 | return rc; | |
1171 | } | |
1172 | ||
1173 | static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg) | |
1174 | { | |
1175 | struct ll_recreate_obj ucreat; | |
1176 | struct ost_id oi; | |
1177 | ENTRY; | |
1178 | ||
1179 | if (!cfs_capable(CFS_CAP_SYS_ADMIN)) | |
1180 | RETURN(-EPERM); | |
1181 | ||
1182 | if (copy_from_user(&ucreat, (struct ll_recreate_obj *)arg, | |
1183 | sizeof(ucreat))) | |
1184 | RETURN(-EFAULT); | |
1185 | ||
1186 | ostid_set_seq_mdt0(&oi); | |
1187 | ostid_set_id(&oi, ucreat.lrc_id); | |
1188 | RETURN(ll_lov_recreate(inode, &oi, ucreat.lrc_ost_idx)); | |
1189 | } | |
1190 | ||
1191 | static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg) | |
1192 | { | |
1193 | struct lu_fid fid; | |
1194 | struct ost_id oi; | |
1195 | obd_count ost_idx; | |
1196 | ENTRY; | |
1197 | ||
1198 | if (!cfs_capable(CFS_CAP_SYS_ADMIN)) | |
1199 | RETURN(-EPERM); | |
1200 | ||
1201 | if (copy_from_user(&fid, (struct lu_fid *)arg, sizeof(fid))) | |
1202 | RETURN(-EFAULT); | |
1203 | ||
1204 | fid_to_ostid(&fid, &oi); | |
1205 | ost_idx = (fid_seq(&fid) >> 16) & 0xffff; | |
1206 | RETURN(ll_lov_recreate(inode, &oi, ost_idx)); | |
1207 | } | |
1208 | ||
1209 | int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file, | |
1210 | int flags, struct lov_user_md *lum, int lum_size) | |
1211 | { | |
1212 | struct lov_stripe_md *lsm = NULL; | |
1213 | struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags}; | |
1214 | int rc = 0; | |
1215 | ENTRY; | |
1216 | ||
1217 | lsm = ccc_inode_lsm_get(inode); | |
1218 | if (lsm != NULL) { | |
1219 | ccc_inode_lsm_put(inode, lsm); | |
1220 | CDEBUG(D_IOCTL, "stripe already exists for ino %lu\n", | |
1221 | inode->i_ino); | |
1222 | RETURN(-EEXIST); | |
1223 | } | |
1224 | ||
1225 | ll_inode_size_lock(inode); | |
1226 | rc = ll_intent_file_open(file, lum, lum_size, &oit); | |
1227 | if (rc) | |
1228 | GOTO(out, rc); | |
1229 | rc = oit.d.lustre.it_status; | |
1230 | if (rc < 0) | |
1231 | GOTO(out_req_free, rc); | |
1232 | ||
1233 | ll_release_openhandle(file->f_dentry, &oit); | |
1234 | ||
1235 | out: | |
1236 | ll_inode_size_unlock(inode); | |
1237 | ll_intent_release(&oit); | |
1238 | ccc_inode_lsm_put(inode, lsm); | |
1239 | RETURN(rc); | |
1240 | out_req_free: | |
1241 | ptlrpc_req_finished((struct ptlrpc_request *) oit.d.lustre.it_data); | |
1242 | goto out; | |
1243 | } | |
1244 | ||
1245 | int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename, | |
1246 | struct lov_mds_md **lmmp, int *lmm_size, | |
1247 | struct ptlrpc_request **request) | |
1248 | { | |
1249 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1250 | struct mdt_body *body; | |
1251 | struct lov_mds_md *lmm = NULL; | |
1252 | struct ptlrpc_request *req = NULL; | |
1253 | struct md_op_data *op_data; | |
1254 | int rc, lmmsize; | |
1255 | ||
1256 | rc = ll_get_max_mdsize(sbi, &lmmsize); | |
1257 | if (rc) | |
1258 | RETURN(rc); | |
1259 | ||
1260 | op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, | |
1261 | strlen(filename), lmmsize, | |
1262 | LUSTRE_OPC_ANY, NULL); | |
1263 | if (IS_ERR(op_data)) | |
1264 | RETURN(PTR_ERR(op_data)); | |
1265 | ||
1266 | op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA; | |
1267 | rc = md_getattr_name(sbi->ll_md_exp, op_data, &req); | |
1268 | ll_finish_md_op_data(op_data); | |
1269 | if (rc < 0) { | |
1270 | CDEBUG(D_INFO, "md_getattr_name failed " | |
1271 | "on %s: rc %d\n", filename, rc); | |
1272 | GOTO(out, rc); | |
1273 | } | |
1274 | ||
1275 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); | |
1276 | LASSERT(body != NULL); /* checked by mdc_getattr_name */ | |
1277 | ||
1278 | lmmsize = body->eadatasize; | |
1279 | ||
1280 | if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) || | |
1281 | lmmsize == 0) { | |
1282 | GOTO(out, rc = -ENODATA); | |
1283 | } | |
1284 | ||
1285 | lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize); | |
1286 | LASSERT(lmm != NULL); | |
1287 | ||
1288 | if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) && | |
1289 | (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) { | |
1290 | GOTO(out, rc = -EPROTO); | |
1291 | } | |
1292 | ||
1293 | /* | |
1294 | * This is coming from the MDS, so is probably in | |
1295 | * little endian. We convert it to host endian before | |
1296 | * passing it to userspace. | |
1297 | */ | |
1298 | if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) { | |
1299 | /* if function called for directory - we should | |
1300 | * avoid swab not existent lsm objects */ | |
1301 | if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) { | |
1302 | lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm); | |
1303 | if (S_ISREG(body->mode)) | |
1304 | lustre_swab_lov_user_md_objects( | |
1305 | ((struct lov_user_md_v1 *)lmm)->lmm_objects, | |
1306 | ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count); | |
1307 | } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) { | |
1308 | lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm); | |
1309 | if (S_ISREG(body->mode)) | |
1310 | lustre_swab_lov_user_md_objects( | |
1311 | ((struct lov_user_md_v3 *)lmm)->lmm_objects, | |
1312 | ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count); | |
1313 | } | |
1314 | } | |
1315 | ||
1316 | out: | |
1317 | *lmmp = lmm; | |
1318 | *lmm_size = lmmsize; | |
1319 | *request = req; | |
1320 | return rc; | |
1321 | } | |
1322 | ||
1323 | static int ll_lov_setea(struct inode *inode, struct file *file, | |
1324 | unsigned long arg) | |
1325 | { | |
1326 | int flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE; | |
1327 | struct lov_user_md *lump; | |
1328 | int lum_size = sizeof(struct lov_user_md) + | |
1329 | sizeof(struct lov_user_ost_data); | |
1330 | int rc; | |
1331 | ENTRY; | |
1332 | ||
1333 | if (!cfs_capable(CFS_CAP_SYS_ADMIN)) | |
1334 | RETURN(-EPERM); | |
1335 | ||
1336 | OBD_ALLOC_LARGE(lump, lum_size); | |
1337 | if (lump == NULL) | |
1338 | RETURN(-ENOMEM); | |
1339 | ||
1340 | if (copy_from_user(lump, (struct lov_user_md *)arg, lum_size)) { | |
1341 | OBD_FREE_LARGE(lump, lum_size); | |
1342 | RETURN(-EFAULT); | |
1343 | } | |
1344 | ||
1345 | rc = ll_lov_setstripe_ea_info(inode, file, flags, lump, lum_size); | |
1346 | ||
1347 | OBD_FREE_LARGE(lump, lum_size); | |
1348 | RETURN(rc); | |
1349 | } | |
1350 | ||
1351 | static int ll_lov_setstripe(struct inode *inode, struct file *file, | |
1352 | unsigned long arg) | |
1353 | { | |
1354 | struct lov_user_md_v3 lumv3; | |
1355 | struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3; | |
1356 | struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg; | |
1357 | struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg; | |
1358 | int lum_size, rc; | |
1359 | int flags = FMODE_WRITE; | |
1360 | ENTRY; | |
1361 | ||
1362 | /* first try with v1 which is smaller than v3 */ | |
1363 | lum_size = sizeof(struct lov_user_md_v1); | |
1364 | if (copy_from_user(lumv1, lumv1p, lum_size)) | |
1365 | RETURN(-EFAULT); | |
1366 | ||
1367 | if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) { | |
1368 | lum_size = sizeof(struct lov_user_md_v3); | |
1369 | if (copy_from_user(&lumv3, lumv3p, lum_size)) | |
1370 | RETURN(-EFAULT); | |
1371 | } | |
1372 | ||
1373 | rc = ll_lov_setstripe_ea_info(inode, file, flags, lumv1, lum_size); | |
1374 | if (rc == 0) { | |
1375 | struct lov_stripe_md *lsm; | |
1376 | __u32 gen; | |
1377 | ||
1378 | put_user(0, &lumv1p->lmm_stripe_count); | |
1379 | ||
1380 | ll_layout_refresh(inode, &gen); | |
1381 | lsm = ccc_inode_lsm_get(inode); | |
1382 | rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), | |
1383 | 0, lsm, (void *)arg); | |
1384 | ccc_inode_lsm_put(inode, lsm); | |
1385 | } | |
1386 | RETURN(rc); | |
1387 | } | |
1388 | ||
1389 | static int ll_lov_getstripe(struct inode *inode, unsigned long arg) | |
1390 | { | |
1391 | struct lov_stripe_md *lsm; | |
1392 | int rc = -ENODATA; | |
1393 | ENTRY; | |
1394 | ||
1395 | lsm = ccc_inode_lsm_get(inode); | |
1396 | if (lsm != NULL) | |
1397 | rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0, | |
1398 | lsm, (void *)arg); | |
1399 | ccc_inode_lsm_put(inode, lsm); | |
1400 | RETURN(rc); | |
1401 | } | |
1402 | ||
1403 | int ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg) | |
1404 | { | |
1405 | struct ll_inode_info *lli = ll_i2info(inode); | |
1406 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
1407 | struct ccc_grouplock grouplock; | |
1408 | int rc; | |
1409 | ENTRY; | |
1410 | ||
1411 | if (ll_file_nolock(file)) | |
1412 | RETURN(-EOPNOTSUPP); | |
1413 | ||
1414 | spin_lock(&lli->lli_lock); | |
1415 | if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { | |
1416 | CWARN("group lock already existed with gid %lu\n", | |
1417 | fd->fd_grouplock.cg_gid); | |
1418 | spin_unlock(&lli->lli_lock); | |
1419 | RETURN(-EINVAL); | |
1420 | } | |
1421 | LASSERT(fd->fd_grouplock.cg_lock == NULL); | |
1422 | spin_unlock(&lli->lli_lock); | |
1423 | ||
1424 | rc = cl_get_grouplock(cl_i2info(inode)->lli_clob, | |
1425 | arg, (file->f_flags & O_NONBLOCK), &grouplock); | |
1426 | if (rc) | |
1427 | RETURN(rc); | |
1428 | ||
1429 | spin_lock(&lli->lli_lock); | |
1430 | if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { | |
1431 | spin_unlock(&lli->lli_lock); | |
1432 | CERROR("another thread just won the race\n"); | |
1433 | cl_put_grouplock(&grouplock); | |
1434 | RETURN(-EINVAL); | |
1435 | } | |
1436 | ||
1437 | fd->fd_flags |= LL_FILE_GROUP_LOCKED; | |
1438 | fd->fd_grouplock = grouplock; | |
1439 | spin_unlock(&lli->lli_lock); | |
1440 | ||
1441 | CDEBUG(D_INFO, "group lock %lu obtained\n", arg); | |
1442 | RETURN(0); | |
1443 | } | |
1444 | ||
1445 | int ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg) | |
1446 | { | |
1447 | struct ll_inode_info *lli = ll_i2info(inode); | |
1448 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
1449 | struct ccc_grouplock grouplock; | |
1450 | ENTRY; | |
1451 | ||
1452 | spin_lock(&lli->lli_lock); | |
1453 | if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) { | |
1454 | spin_unlock(&lli->lli_lock); | |
1455 | CWARN("no group lock held\n"); | |
1456 | RETURN(-EINVAL); | |
1457 | } | |
1458 | LASSERT(fd->fd_grouplock.cg_lock != NULL); | |
1459 | ||
1460 | if (fd->fd_grouplock.cg_gid != arg) { | |
1461 | CWARN("group lock %lu doesn't match current id %lu\n", | |
1462 | arg, fd->fd_grouplock.cg_gid); | |
1463 | spin_unlock(&lli->lli_lock); | |
1464 | RETURN(-EINVAL); | |
1465 | } | |
1466 | ||
1467 | grouplock = fd->fd_grouplock; | |
1468 | memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock)); | |
1469 | fd->fd_flags &= ~LL_FILE_GROUP_LOCKED; | |
1470 | spin_unlock(&lli->lli_lock); | |
1471 | ||
1472 | cl_put_grouplock(&grouplock); | |
1473 | CDEBUG(D_INFO, "group lock %lu released\n", arg); | |
1474 | RETURN(0); | |
1475 | } | |
1476 | ||
1477 | /** | |
1478 | * Close inode open handle | |
1479 | * | |
1480 | * \param dentry [in] dentry which contains the inode | |
1481 | * \param it [in,out] intent which contains open info and result | |
1482 | * | |
1483 | * \retval 0 success | |
1484 | * \retval <0 failure | |
1485 | */ | |
1486 | int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it) | |
1487 | { | |
1488 | struct inode *inode = dentry->d_inode; | |
1489 | struct obd_client_handle *och; | |
1490 | int rc; | |
1491 | ENTRY; | |
1492 | ||
1493 | LASSERT(inode); | |
1494 | ||
1495 | /* Root ? Do nothing. */ | |
1496 | if (dentry->d_inode->i_sb->s_root == dentry) | |
1497 | RETURN(0); | |
1498 | ||
1499 | /* No open handle to close? Move away */ | |
1500 | if (!it_disposition(it, DISP_OPEN_OPEN)) | |
1501 | RETURN(0); | |
1502 | ||
1503 | LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0); | |
1504 | ||
1505 | OBD_ALLOC(och, sizeof(*och)); | |
1506 | if (!och) | |
1507 | GOTO(out, rc = -ENOMEM); | |
1508 | ||
1509 | ll_och_fill(ll_i2sbi(inode)->ll_md_exp, | |
1510 | ll_i2info(inode), it, och); | |
1511 | ||
1512 | rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, | |
1513 | inode, och); | |
1514 | out: | |
1515 | /* this one is in place of ll_file_open */ | |
1516 | if (it_disposition(it, DISP_ENQ_OPEN_REF)) { | |
1517 | ptlrpc_req_finished(it->d.lustre.it_data); | |
1518 | it_clear_disposition(it, DISP_ENQ_OPEN_REF); | |
1519 | } | |
1520 | RETURN(rc); | |
1521 | } | |
1522 | ||
1523 | /** | |
1524 | * Get size for inode for which FIEMAP mapping is requested. | |
1525 | * Make the FIEMAP get_info call and returns the result. | |
1526 | */ | |
1527 | int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap, | |
1528 | int num_bytes) | |
1529 | { | |
1530 | struct obd_export *exp = ll_i2dtexp(inode); | |
1531 | struct lov_stripe_md *lsm = NULL; | |
1532 | struct ll_fiemap_info_key fm_key = { .name = KEY_FIEMAP, }; | |
1533 | int vallen = num_bytes; | |
1534 | int rc; | |
1535 | ENTRY; | |
1536 | ||
1537 | /* Checks for fiemap flags */ | |
1538 | if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) { | |
1539 | fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT; | |
1540 | return -EBADR; | |
1541 | } | |
1542 | ||
1543 | /* Check for FIEMAP_FLAG_SYNC */ | |
1544 | if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) { | |
1545 | rc = filemap_fdatawrite(inode->i_mapping); | |
1546 | if (rc) | |
1547 | return rc; | |
1548 | } | |
1549 | ||
1550 | lsm = ccc_inode_lsm_get(inode); | |
1551 | if (lsm == NULL) | |
1552 | return -ENOENT; | |
1553 | ||
1554 | /* If the stripe_count > 1 and the application does not understand | |
1555 | * DEVICE_ORDER flag, then it cannot interpret the extents correctly. | |
1556 | */ | |
1557 | if (lsm->lsm_stripe_count > 1 && | |
1558 | !(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER)) | |
1559 | GOTO(out, rc = -EOPNOTSUPP); | |
1560 | ||
1561 | fm_key.oa.o_oi = lsm->lsm_oi; | |
1562 | fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP; | |
1563 | ||
1564 | obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLSIZE); | |
1565 | obdo_set_parent_fid(&fm_key.oa, &ll_i2info(inode)->lli_fid); | |
1566 | /* If filesize is 0, then there would be no objects for mapping */ | |
1567 | if (fm_key.oa.o_size == 0) { | |
1568 | fiemap->fm_mapped_extents = 0; | |
1569 | GOTO(out, rc = 0); | |
1570 | } | |
1571 | ||
1572 | memcpy(&fm_key.fiemap, fiemap, sizeof(*fiemap)); | |
1573 | ||
1574 | rc = obd_get_info(NULL, exp, sizeof(fm_key), &fm_key, &vallen, | |
1575 | fiemap, lsm); | |
1576 | if (rc) | |
1577 | CERROR("obd_get_info failed: rc = %d\n", rc); | |
1578 | ||
1579 | out: | |
1580 | ccc_inode_lsm_put(inode, lsm); | |
1581 | RETURN(rc); | |
1582 | } | |
1583 | ||
1584 | int ll_fid2path(struct inode *inode, void *arg) | |
1585 | { | |
1586 | struct obd_export *exp = ll_i2mdexp(inode); | |
1587 | struct getinfo_fid2path *gfout, *gfin; | |
1588 | int outsize, rc; | |
1589 | ENTRY; | |
1590 | ||
1591 | if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) && | |
1592 | !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH)) | |
1593 | RETURN(-EPERM); | |
1594 | ||
1595 | /* Need to get the buflen */ | |
1596 | OBD_ALLOC_PTR(gfin); | |
1597 | if (gfin == NULL) | |
1598 | RETURN(-ENOMEM); | |
1599 | if (copy_from_user(gfin, arg, sizeof(*gfin))) { | |
1600 | OBD_FREE_PTR(gfin); | |
1601 | RETURN(-EFAULT); | |
1602 | } | |
1603 | ||
1604 | outsize = sizeof(*gfout) + gfin->gf_pathlen; | |
1605 | OBD_ALLOC(gfout, outsize); | |
1606 | if (gfout == NULL) { | |
1607 | OBD_FREE_PTR(gfin); | |
1608 | RETURN(-ENOMEM); | |
1609 | } | |
1610 | memcpy(gfout, gfin, sizeof(*gfout)); | |
1611 | OBD_FREE_PTR(gfin); | |
1612 | ||
1613 | /* Call mdc_iocontrol */ | |
1614 | rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL); | |
1615 | if (rc) | |
1616 | GOTO(gf_free, rc); | |
1617 | ||
1618 | if (copy_to_user(arg, gfout, outsize)) | |
1619 | rc = -EFAULT; | |
1620 | ||
1621 | gf_free: | |
1622 | OBD_FREE(gfout, outsize); | |
1623 | RETURN(rc); | |
1624 | } | |
1625 | ||
1626 | static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg) | |
1627 | { | |
1628 | struct ll_user_fiemap *fiemap_s; | |
1629 | size_t num_bytes, ret_bytes; | |
1630 | unsigned int extent_count; | |
1631 | int rc = 0; | |
1632 | ||
1633 | /* Get the extent count so we can calculate the size of | |
1634 | * required fiemap buffer */ | |
1635 | if (get_user(extent_count, | |
1636 | &((struct ll_user_fiemap __user *)arg)->fm_extent_count)) | |
1637 | RETURN(-EFAULT); | |
1638 | num_bytes = sizeof(*fiemap_s) + (extent_count * | |
1639 | sizeof(struct ll_fiemap_extent)); | |
1640 | ||
1641 | OBD_ALLOC_LARGE(fiemap_s, num_bytes); | |
1642 | if (fiemap_s == NULL) | |
1643 | RETURN(-ENOMEM); | |
1644 | ||
1645 | /* get the fiemap value */ | |
1646 | if (copy_from_user(fiemap_s, (struct ll_user_fiemap __user *)arg, | |
1647 | sizeof(*fiemap_s))) | |
1648 | GOTO(error, rc = -EFAULT); | |
1649 | ||
1650 | /* If fm_extent_count is non-zero, read the first extent since | |
1651 | * it is used to calculate end_offset and device from previous | |
1652 | * fiemap call. */ | |
1653 | if (extent_count) { | |
1654 | if (copy_from_user(&fiemap_s->fm_extents[0], | |
1655 | (char __user *)arg + sizeof(*fiemap_s), | |
1656 | sizeof(struct ll_fiemap_extent))) | |
1657 | GOTO(error, rc = -EFAULT); | |
1658 | } | |
1659 | ||
1660 | rc = ll_do_fiemap(inode, fiemap_s, num_bytes); | |
1661 | if (rc) | |
1662 | GOTO(error, rc); | |
1663 | ||
1664 | ret_bytes = sizeof(struct ll_user_fiemap); | |
1665 | ||
1666 | if (extent_count != 0) | |
1667 | ret_bytes += (fiemap_s->fm_mapped_extents * | |
1668 | sizeof(struct ll_fiemap_extent)); | |
1669 | ||
1670 | if (copy_to_user((void *)arg, fiemap_s, ret_bytes)) | |
1671 | rc = -EFAULT; | |
1672 | ||
1673 | error: | |
1674 | OBD_FREE_LARGE(fiemap_s, num_bytes); | |
1675 | RETURN(rc); | |
1676 | } | |
1677 | ||
1678 | /* | |
1679 | * Read the data_version for inode. | |
1680 | * | |
1681 | * This value is computed using stripe object version on OST. | |
1682 | * Version is computed using server side locking. | |
1683 | * | |
1684 | * @param extent_lock Take extent lock. Not needed if a process is already | |
1685 | * holding the OST object group locks. | |
1686 | */ | |
1687 | int ll_data_version(struct inode *inode, __u64 *data_version, | |
1688 | int extent_lock) | |
1689 | { | |
1690 | struct lov_stripe_md *lsm = NULL; | |
1691 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
1692 | struct obdo *obdo = NULL; | |
1693 | int rc; | |
1694 | ENTRY; | |
1695 | ||
1696 | /* If no stripe, we consider version is 0. */ | |
1697 | lsm = ccc_inode_lsm_get(inode); | |
1698 | if (lsm == NULL) { | |
1699 | *data_version = 0; | |
1700 | CDEBUG(D_INODE, "No object for inode\n"); | |
1701 | RETURN(0); | |
1702 | } | |
1703 | ||
1704 | OBD_ALLOC_PTR(obdo); | |
1705 | if (obdo == NULL) { | |
1706 | ccc_inode_lsm_put(inode, lsm); | |
1707 | RETURN(-ENOMEM); | |
1708 | } | |
1709 | ||
1710 | rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, obdo, 0, extent_lock); | |
1711 | if (!rc) { | |
1712 | if (!(obdo->o_valid & OBD_MD_FLDATAVERSION)) | |
1713 | rc = -EOPNOTSUPP; | |
1714 | else | |
1715 | *data_version = obdo->o_data_version; | |
1716 | } | |
1717 | ||
1718 | OBD_FREE_PTR(obdo); | |
1719 | ccc_inode_lsm_put(inode, lsm); | |
1720 | ||
1721 | RETURN(rc); | |
1722 | } | |
1723 | ||
1724 | struct ll_swap_stack { | |
1725 | struct iattr ia1, ia2; | |
1726 | __u64 dv1, dv2; | |
1727 | struct inode *inode1, *inode2; | |
1728 | bool check_dv1, check_dv2; | |
1729 | }; | |
1730 | ||
1731 | static int ll_swap_layouts(struct file *file1, struct file *file2, | |
1732 | struct lustre_swap_layouts *lsl) | |
1733 | { | |
1734 | struct mdc_swap_layouts msl; | |
1735 | struct md_op_data *op_data; | |
1736 | __u32 gid; | |
1737 | __u64 dv; | |
1738 | struct ll_swap_stack *llss = NULL; | |
1739 | int rc; | |
1740 | ||
1741 | OBD_ALLOC_PTR(llss); | |
1742 | if (llss == NULL) | |
1743 | RETURN(-ENOMEM); | |
1744 | ||
1745 | llss->inode1 = file1->f_dentry->d_inode; | |
1746 | llss->inode2 = file2->f_dentry->d_inode; | |
1747 | ||
1748 | if (!S_ISREG(llss->inode2->i_mode)) | |
1749 | GOTO(free, rc = -EINVAL); | |
1750 | ||
1751 | if (ll_permission(llss->inode1, MAY_WRITE, NULL) || | |
1752 | ll_permission(llss->inode2, MAY_WRITE, NULL)) | |
1753 | GOTO(free, rc = -EPERM); | |
1754 | ||
1755 | if (llss->inode2->i_sb != llss->inode1->i_sb) | |
1756 | GOTO(free, rc = -EXDEV); | |
1757 | ||
1758 | /* we use 2 bool because it is easier to swap than 2 bits */ | |
1759 | if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1) | |
1760 | llss->check_dv1 = true; | |
1761 | ||
1762 | if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2) | |
1763 | llss->check_dv2 = true; | |
1764 | ||
1765 | /* we cannot use lsl->sl_dvX directly because we may swap them */ | |
1766 | llss->dv1 = lsl->sl_dv1; | |
1767 | llss->dv2 = lsl->sl_dv2; | |
1768 | ||
1769 | rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2)); | |
1770 | if (rc == 0) /* same file, done! */ | |
1771 | GOTO(free, rc = 0); | |
1772 | ||
1773 | if (rc < 0) { /* sequentialize it */ | |
1774 | swap(llss->inode1, llss->inode2); | |
1775 | swap(file1, file2); | |
1776 | swap(llss->dv1, llss->dv2); | |
1777 | swap(llss->check_dv1, llss->check_dv2); | |
1778 | } | |
1779 | ||
1780 | gid = lsl->sl_gid; | |
1781 | if (gid != 0) { /* application asks to flush dirty cache */ | |
1782 | rc = ll_get_grouplock(llss->inode1, file1, gid); | |
1783 | if (rc < 0) | |
1784 | GOTO(free, rc); | |
1785 | ||
1786 | rc = ll_get_grouplock(llss->inode2, file2, gid); | |
1787 | if (rc < 0) { | |
1788 | ll_put_grouplock(llss->inode1, file1, gid); | |
1789 | GOTO(free, rc); | |
1790 | } | |
1791 | } | |
1792 | ||
1793 | /* to be able to restore mtime and atime after swap | |
1794 | * we need to first save them */ | |
1795 | if (lsl->sl_flags & | |
1796 | (SWAP_LAYOUTS_KEEP_MTIME | SWAP_LAYOUTS_KEEP_ATIME)) { | |
1797 | llss->ia1.ia_mtime = llss->inode1->i_mtime; | |
1798 | llss->ia1.ia_atime = llss->inode1->i_atime; | |
1799 | llss->ia1.ia_valid = ATTR_MTIME | ATTR_ATIME; | |
1800 | llss->ia2.ia_mtime = llss->inode2->i_mtime; | |
1801 | llss->ia2.ia_atime = llss->inode2->i_atime; | |
1802 | llss->ia2.ia_valid = ATTR_MTIME | ATTR_ATIME; | |
1803 | } | |
1804 | ||
1805 | /* ultimate check, before swaping the layouts we check if | |
1806 | * dataversion has changed (if requested) */ | |
1807 | if (llss->check_dv1) { | |
1808 | rc = ll_data_version(llss->inode1, &dv, 0); | |
1809 | if (rc) | |
1810 | GOTO(putgl, rc); | |
1811 | if (dv != llss->dv1) | |
1812 | GOTO(putgl, rc = -EAGAIN); | |
1813 | } | |
1814 | ||
1815 | if (llss->check_dv2) { | |
1816 | rc = ll_data_version(llss->inode2, &dv, 0); | |
1817 | if (rc) | |
1818 | GOTO(putgl, rc); | |
1819 | if (dv != llss->dv2) | |
1820 | GOTO(putgl, rc = -EAGAIN); | |
1821 | } | |
1822 | ||
1823 | /* struct md_op_data is used to send the swap args to the mdt | |
1824 | * only flags is missing, so we use struct mdc_swap_layouts | |
1825 | * through the md_op_data->op_data */ | |
1826 | /* flags from user space have to be converted before they are send to | |
1827 | * server, no flag is sent today, they are only used on the client */ | |
1828 | msl.msl_flags = 0; | |
1829 | rc = -ENOMEM; | |
1830 | op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0, | |
1831 | 0, LUSTRE_OPC_ANY, &msl); | |
1832 | if (op_data != NULL) { | |
1833 | rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, | |
1834 | ll_i2mdexp(llss->inode1), | |
1835 | sizeof(*op_data), op_data, NULL); | |
1836 | ll_finish_md_op_data(op_data); | |
1837 | } | |
1838 | ||
1839 | putgl: | |
1840 | if (gid != 0) { | |
1841 | ll_put_grouplock(llss->inode2, file2, gid); | |
1842 | ll_put_grouplock(llss->inode1, file1, gid); | |
1843 | } | |
1844 | ||
1845 | /* rc can be set from obd_iocontrol() or from a GOTO(putgl, ...) */ | |
1846 | if (rc != 0) | |
1847 | GOTO(free, rc); | |
1848 | ||
1849 | /* clear useless flags */ | |
1850 | if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_MTIME)) { | |
1851 | llss->ia1.ia_valid &= ~ATTR_MTIME; | |
1852 | llss->ia2.ia_valid &= ~ATTR_MTIME; | |
1853 | } | |
1854 | ||
1855 | if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_ATIME)) { | |
1856 | llss->ia1.ia_valid &= ~ATTR_ATIME; | |
1857 | llss->ia2.ia_valid &= ~ATTR_ATIME; | |
1858 | } | |
1859 | ||
1860 | /* update time if requested */ | |
1861 | rc = 0; | |
1862 | if (llss->ia2.ia_valid != 0) { | |
1863 | mutex_lock(&llss->inode1->i_mutex); | |
1864 | rc = ll_setattr(file1->f_dentry, &llss->ia2); | |
1865 | mutex_unlock(&llss->inode1->i_mutex); | |
1866 | } | |
1867 | ||
1868 | if (llss->ia1.ia_valid != 0) { | |
1869 | int rc1; | |
1870 | ||
1871 | mutex_lock(&llss->inode2->i_mutex); | |
1872 | rc1 = ll_setattr(file2->f_dentry, &llss->ia1); | |
1873 | mutex_unlock(&llss->inode2->i_mutex); | |
1874 | if (rc == 0) | |
1875 | rc = rc1; | |
1876 | } | |
1877 | ||
1878 | free: | |
1879 | if (llss != NULL) | |
1880 | OBD_FREE_PTR(llss); | |
1881 | ||
1882 | RETURN(rc); | |
1883 | } | |
1884 | ||
1885 | long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg) | |
1886 | { | |
1887 | struct inode *inode = file->f_dentry->d_inode; | |
1888 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
1889 | int flags, rc; | |
1890 | ENTRY; | |
1891 | ||
1892 | CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),cmd=%x\n", inode->i_ino, | |
1893 | inode->i_generation, inode, cmd); | |
1894 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1); | |
1895 | ||
1896 | /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */ | |
1897 | if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */ | |
1898 | RETURN(-ENOTTY); | |
1899 | ||
1900 | switch(cmd) { | |
1901 | case LL_IOC_GETFLAGS: | |
1902 | /* Get the current value of the file flags */ | |
1903 | return put_user(fd->fd_flags, (int *)arg); | |
1904 | case LL_IOC_SETFLAGS: | |
1905 | case LL_IOC_CLRFLAGS: | |
1906 | /* Set or clear specific file flags */ | |
1907 | /* XXX This probably needs checks to ensure the flags are | |
1908 | * not abused, and to handle any flag side effects. | |
1909 | */ | |
1910 | if (get_user(flags, (int *) arg)) | |
1911 | RETURN(-EFAULT); | |
1912 | ||
1913 | if (cmd == LL_IOC_SETFLAGS) { | |
1914 | if ((flags & LL_FILE_IGNORE_LOCK) && | |
1915 | !(file->f_flags & O_DIRECT)) { | |
1916 | CERROR("%s: unable to disable locking on " | |
1917 | "non-O_DIRECT file\n", current->comm); | |
1918 | RETURN(-EINVAL); | |
1919 | } | |
1920 | ||
1921 | fd->fd_flags |= flags; | |
1922 | } else { | |
1923 | fd->fd_flags &= ~flags; | |
1924 | } | |
1925 | RETURN(0); | |
1926 | case LL_IOC_LOV_SETSTRIPE: | |
1927 | RETURN(ll_lov_setstripe(inode, file, arg)); | |
1928 | case LL_IOC_LOV_SETEA: | |
1929 | RETURN(ll_lov_setea(inode, file, arg)); | |
1930 | case LL_IOC_LOV_SWAP_LAYOUTS: { | |
1931 | struct file *file2; | |
1932 | struct lustre_swap_layouts lsl; | |
1933 | ||
1934 | if (copy_from_user(&lsl, (char *)arg, | |
1935 | sizeof(struct lustre_swap_layouts))) | |
1936 | RETURN(-EFAULT); | |
1937 | ||
1938 | if ((file->f_flags & O_ACCMODE) == 0) /* O_RDONLY */ | |
1939 | RETURN(-EPERM); | |
1940 | ||
1941 | file2 = fget(lsl.sl_fd); | |
1942 | if (file2 == NULL) | |
1943 | RETURN(-EBADF); | |
1944 | ||
1945 | rc = -EPERM; | |
1946 | if ((file2->f_flags & O_ACCMODE) != 0) /* O_WRONLY or O_RDWR */ | |
1947 | rc = ll_swap_layouts(file, file2, &lsl); | |
1948 | fput(file2); | |
1949 | RETURN(rc); | |
1950 | } | |
1951 | case LL_IOC_LOV_GETSTRIPE: | |
1952 | RETURN(ll_lov_getstripe(inode, arg)); | |
1953 | case LL_IOC_RECREATE_OBJ: | |
1954 | RETURN(ll_lov_recreate_obj(inode, arg)); | |
1955 | case LL_IOC_RECREATE_FID: | |
1956 | RETURN(ll_lov_recreate_fid(inode, arg)); | |
1957 | case FSFILT_IOC_FIEMAP: | |
1958 | RETURN(ll_ioctl_fiemap(inode, arg)); | |
1959 | case FSFILT_IOC_GETFLAGS: | |
1960 | case FSFILT_IOC_SETFLAGS: | |
1961 | RETURN(ll_iocontrol(inode, file, cmd, arg)); | |
1962 | case FSFILT_IOC_GETVERSION_OLD: | |
1963 | case FSFILT_IOC_GETVERSION: | |
1964 | RETURN(put_user(inode->i_generation, (int *)arg)); | |
1965 | case LL_IOC_GROUP_LOCK: | |
1966 | RETURN(ll_get_grouplock(inode, file, arg)); | |
1967 | case LL_IOC_GROUP_UNLOCK: | |
1968 | RETURN(ll_put_grouplock(inode, file, arg)); | |
1969 | case IOC_OBD_STATFS: | |
1970 | RETURN(ll_obd_statfs(inode, (void *)arg)); | |
1971 | ||
1972 | /* We need to special case any other ioctls we want to handle, | |
1973 | * to send them to the MDS/OST as appropriate and to properly | |
1974 | * network encode the arg field. | |
1975 | case FSFILT_IOC_SETVERSION_OLD: | |
1976 | case FSFILT_IOC_SETVERSION: | |
1977 | */ | |
1978 | case LL_IOC_FLUSHCTX: | |
1979 | RETURN(ll_flush_ctx(inode)); | |
1980 | case LL_IOC_PATH2FID: { | |
1981 | if (copy_to_user((void *)arg, ll_inode2fid(inode), | |
1982 | sizeof(struct lu_fid))) | |
1983 | RETURN(-EFAULT); | |
1984 | ||
1985 | RETURN(0); | |
1986 | } | |
1987 | case OBD_IOC_FID2PATH: | |
1988 | RETURN(ll_fid2path(inode, (void *)arg)); | |
1989 | case LL_IOC_DATA_VERSION: { | |
1990 | struct ioc_data_version idv; | |
1991 | int rc; | |
1992 | ||
1993 | if (copy_from_user(&idv, (char *)arg, sizeof(idv))) | |
1994 | RETURN(-EFAULT); | |
1995 | ||
1996 | rc = ll_data_version(inode, &idv.idv_version, | |
1997 | !(idv.idv_flags & LL_DV_NOFLUSH)); | |
1998 | ||
1999 | if (rc == 0 && copy_to_user((char *) arg, &idv, sizeof(idv))) | |
2000 | RETURN(-EFAULT); | |
2001 | ||
2002 | RETURN(rc); | |
2003 | } | |
2004 | ||
2005 | case LL_IOC_GET_MDTIDX: { | |
2006 | int mdtidx; | |
2007 | ||
2008 | mdtidx = ll_get_mdt_idx(inode); | |
2009 | if (mdtidx < 0) | |
2010 | RETURN(mdtidx); | |
2011 | ||
2012 | if (put_user((int)mdtidx, (int*)arg)) | |
2013 | RETURN(-EFAULT); | |
2014 | ||
2015 | RETURN(0); | |
2016 | } | |
2017 | case OBD_IOC_GETDTNAME: | |
2018 | case OBD_IOC_GETMDNAME: | |
2019 | RETURN(ll_get_obd_name(inode, cmd, arg)); | |
2020 | case LL_IOC_HSM_STATE_GET: { | |
2021 | struct md_op_data *op_data; | |
2022 | struct hsm_user_state *hus; | |
2023 | int rc; | |
2024 | ||
2025 | OBD_ALLOC_PTR(hus); | |
2026 | if (hus == NULL) | |
2027 | RETURN(-ENOMEM); | |
2028 | ||
2029 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, | |
2030 | LUSTRE_OPC_ANY, hus); | |
2031 | if (op_data == NULL) { | |
2032 | OBD_FREE_PTR(hus); | |
2033 | RETURN(-ENOMEM); | |
2034 | } | |
2035 | ||
2036 | rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data), | |
2037 | op_data, NULL); | |
2038 | ||
2039 | if (copy_to_user((void *)arg, hus, sizeof(*hus))) | |
2040 | rc = -EFAULT; | |
2041 | ||
2042 | ll_finish_md_op_data(op_data); | |
2043 | OBD_FREE_PTR(hus); | |
2044 | RETURN(rc); | |
2045 | } | |
2046 | case LL_IOC_HSM_STATE_SET: { | |
2047 | struct md_op_data *op_data; | |
2048 | struct hsm_state_set *hss; | |
2049 | int rc; | |
2050 | ||
2051 | OBD_ALLOC_PTR(hss); | |
2052 | if (hss == NULL) | |
2053 | RETURN(-ENOMEM); | |
2054 | if (copy_from_user(hss, (char *)arg, sizeof(*hss))) { | |
2055 | OBD_FREE_PTR(hss); | |
2056 | RETURN(-EFAULT); | |
2057 | } | |
2058 | ||
2059 | /* Non-root users are forbidden to set or clear flags which are | |
2060 | * NOT defined in HSM_USER_MASK. */ | |
2061 | if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) | |
2062 | && !cfs_capable(CFS_CAP_SYS_ADMIN)) { | |
2063 | OBD_FREE_PTR(hss); | |
2064 | RETURN(-EPERM); | |
2065 | } | |
2066 | ||
2067 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, | |
2068 | LUSTRE_OPC_ANY, hss); | |
2069 | if (op_data == NULL) { | |
2070 | OBD_FREE_PTR(hss); | |
2071 | RETURN(-ENOMEM); | |
2072 | } | |
2073 | ||
2074 | rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data), | |
2075 | op_data, NULL); | |
2076 | ||
2077 | ll_finish_md_op_data(op_data); | |
2078 | ||
2079 | OBD_FREE_PTR(hss); | |
2080 | RETURN(rc); | |
2081 | } | |
2082 | case LL_IOC_HSM_ACTION: { | |
2083 | struct md_op_data *op_data; | |
2084 | struct hsm_current_action *hca; | |
2085 | int rc; | |
2086 | ||
2087 | OBD_ALLOC_PTR(hca); | |
2088 | if (hca == NULL) | |
2089 | RETURN(-ENOMEM); | |
2090 | ||
2091 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, | |
2092 | LUSTRE_OPC_ANY, hca); | |
2093 | if (op_data == NULL) { | |
2094 | OBD_FREE_PTR(hca); | |
2095 | RETURN(-ENOMEM); | |
2096 | } | |
2097 | ||
2098 | rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data), | |
2099 | op_data, NULL); | |
2100 | ||
2101 | if (copy_to_user((char *)arg, hca, sizeof(*hca))) | |
2102 | rc = -EFAULT; | |
2103 | ||
2104 | ll_finish_md_op_data(op_data); | |
2105 | OBD_FREE_PTR(hca); | |
2106 | RETURN(rc); | |
2107 | } | |
2108 | default: { | |
2109 | int err; | |
2110 | ||
2111 | if (LLIOC_STOP == | |
2112 | ll_iocontrol_call(inode, file, cmd, arg, &err)) | |
2113 | RETURN(err); | |
2114 | ||
2115 | RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL, | |
2116 | (void *)arg)); | |
2117 | } | |
2118 | } | |
2119 | } | |
2120 | ||
2121 | ||
2122 | loff_t ll_file_seek(struct file *file, loff_t offset, int origin) | |
2123 | { | |
2124 | struct inode *inode = file->f_dentry->d_inode; | |
2125 | loff_t retval, eof = 0; | |
2126 | ||
2127 | ENTRY; | |
2128 | retval = offset + ((origin == SEEK_END) ? i_size_read(inode) : | |
2129 | (origin == SEEK_CUR) ? file->f_pos : 0); | |
2130 | CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), to=%llu=%#llx(%d)\n", | |
2131 | inode->i_ino, inode->i_generation, inode, retval, retval, | |
2132 | origin); | |
2133 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1); | |
2134 | ||
2135 | if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) { | |
2136 | retval = ll_glimpse_size(inode); | |
2137 | if (retval != 0) | |
2138 | RETURN(retval); | |
2139 | eof = i_size_read(inode); | |
2140 | } | |
2141 | ||
2142 | retval = ll_generic_file_llseek_size(file, offset, origin, | |
2143 | ll_file_maxbytes(inode), eof); | |
2144 | RETURN(retval); | |
2145 | } | |
2146 | ||
2147 | int ll_flush(struct file *file, fl_owner_t id) | |
2148 | { | |
2149 | struct inode *inode = file->f_dentry->d_inode; | |
2150 | struct ll_inode_info *lli = ll_i2info(inode); | |
2151 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
2152 | int rc, err; | |
2153 | ||
2154 | LASSERT(!S_ISDIR(inode->i_mode)); | |
2155 | ||
2156 | /* catch async errors that were recorded back when async writeback | |
2157 | * failed for pages in this mapping. */ | |
2158 | rc = lli->lli_async_rc; | |
2159 | lli->lli_async_rc = 0; | |
2160 | err = lov_read_and_clear_async_rc(lli->lli_clob); | |
2161 | if (rc == 0) | |
2162 | rc = err; | |
2163 | ||
2164 | /* The application has been told write failure already. | |
2165 | * Do not report failure again. */ | |
2166 | if (fd->fd_write_failed) | |
2167 | return 0; | |
2168 | return rc ? -EIO : 0; | |
2169 | } | |
2170 | ||
2171 | /** | |
2172 | * Called to make sure a portion of file has been written out. | |
2173 | * if @local_only is not true, it will send OST_SYNC RPCs to ost. | |
2174 | * | |
2175 | * Return how many pages have been written. | |
2176 | */ | |
2177 | int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end, | |
2178 | enum cl_fsync_mode mode) | |
2179 | { | |
2180 | struct cl_env_nest nest; | |
2181 | struct lu_env *env; | |
2182 | struct cl_io *io; | |
2183 | struct obd_capa *capa = NULL; | |
2184 | struct cl_fsync_io *fio; | |
2185 | int result; | |
2186 | ENTRY; | |
2187 | ||
2188 | if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL && | |
2189 | mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL) | |
2190 | RETURN(-EINVAL); | |
2191 | ||
2192 | env = cl_env_nested_get(&nest); | |
2193 | if (IS_ERR(env)) | |
2194 | RETURN(PTR_ERR(env)); | |
2195 | ||
2196 | capa = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE); | |
2197 | ||
2198 | io = ccc_env_thread_io(env); | |
2199 | io->ci_obj = cl_i2info(inode)->lli_clob; | |
2200 | io->ci_ignore_layout = 1; | |
2201 | ||
2202 | /* initialize parameters for sync */ | |
2203 | fio = &io->u.ci_fsync; | |
2204 | fio->fi_capa = capa; | |
2205 | fio->fi_start = start; | |
2206 | fio->fi_end = end; | |
2207 | fio->fi_fid = ll_inode2fid(inode); | |
2208 | fio->fi_mode = mode; | |
2209 | fio->fi_nr_written = 0; | |
2210 | ||
2211 | if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0) | |
2212 | result = cl_io_loop(env, io); | |
2213 | else | |
2214 | result = io->ci_result; | |
2215 | if (result == 0) | |
2216 | result = fio->fi_nr_written; | |
2217 | cl_io_fini(env, io); | |
2218 | cl_env_nested_put(&nest, env); | |
2219 | ||
2220 | capa_put(capa); | |
2221 | ||
2222 | RETURN(result); | |
2223 | } | |
2224 | ||
2225 | /* | |
2226 | * When dentry is provided (the 'else' case), *file->f_dentry may be | |
2227 | * null and dentry must be used directly rather than pulled from | |
2228 | * *file->f_dentry as is done otherwise. | |
2229 | */ | |
2230 | ||
2231 | int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync) | |
2232 | { | |
2233 | struct dentry *dentry = file->f_dentry; | |
2234 | struct inode *inode = dentry->d_inode; | |
2235 | struct ll_inode_info *lli = ll_i2info(inode); | |
2236 | struct ptlrpc_request *req; | |
2237 | struct obd_capa *oc; | |
2238 | int rc, err; | |
2239 | ENTRY; | |
2240 | ||
2241 | CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino, | |
2242 | inode->i_generation, inode); | |
2243 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1); | |
2244 | ||
2245 | rc = filemap_write_and_wait_range(inode->i_mapping, start, end); | |
2246 | mutex_lock(&inode->i_mutex); | |
2247 | ||
2248 | /* catch async errors that were recorded back when async writeback | |
2249 | * failed for pages in this mapping. */ | |
2250 | if (!S_ISDIR(inode->i_mode)) { | |
2251 | err = lli->lli_async_rc; | |
2252 | lli->lli_async_rc = 0; | |
2253 | if (rc == 0) | |
2254 | rc = err; | |
2255 | err = lov_read_and_clear_async_rc(lli->lli_clob); | |
2256 | if (rc == 0) | |
2257 | rc = err; | |
2258 | } | |
2259 | ||
2260 | oc = ll_mdscapa_get(inode); | |
2261 | err = md_sync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc, | |
2262 | &req); | |
2263 | capa_put(oc); | |
2264 | if (!rc) | |
2265 | rc = err; | |
2266 | if (!err) | |
2267 | ptlrpc_req_finished(req); | |
2268 | ||
2269 | if (datasync && S_ISREG(inode->i_mode)) { | |
2270 | struct ll_file_data *fd = LUSTRE_FPRIVATE(file); | |
2271 | ||
2272 | err = cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, | |
2273 | CL_FSYNC_ALL); | |
2274 | if (rc == 0 && err < 0) | |
2275 | rc = err; | |
2276 | if (rc < 0) | |
2277 | fd->fd_write_failed = true; | |
2278 | else | |
2279 | fd->fd_write_failed = false; | |
2280 | } | |
2281 | ||
2282 | mutex_unlock(&inode->i_mutex); | |
2283 | RETURN(rc); | |
2284 | } | |
2285 | ||
2286 | int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) | |
2287 | { | |
2288 | struct inode *inode = file->f_dentry->d_inode; | |
2289 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
2290 | struct ldlm_enqueue_info einfo = { .ei_type = LDLM_FLOCK, | |
2291 | .ei_cb_cp =ldlm_flock_completion_ast, | |
2292 | .ei_cbdata = file_lock }; | |
2293 | struct md_op_data *op_data; | |
2294 | struct lustre_handle lockh = {0}; | |
2295 | ldlm_policy_data_t flock = {{0}}; | |
2296 | int flags = 0; | |
2297 | int rc; | |
2298 | int rc2 = 0; | |
2299 | ENTRY; | |
2300 | ||
2301 | CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu file_lock=%p\n", | |
2302 | inode->i_ino, file_lock); | |
2303 | ||
2304 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1); | |
2305 | ||
2306 | if (file_lock->fl_flags & FL_FLOCK) { | |
2307 | LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK)); | |
2308 | /* flocks are whole-file locks */ | |
2309 | flock.l_flock.end = OFFSET_MAX; | |
2310 | /* For flocks owner is determined by the local file desctiptor*/ | |
2311 | flock.l_flock.owner = (unsigned long)file_lock->fl_file; | |
2312 | } else if (file_lock->fl_flags & FL_POSIX) { | |
2313 | flock.l_flock.owner = (unsigned long)file_lock->fl_owner; | |
2314 | flock.l_flock.start = file_lock->fl_start; | |
2315 | flock.l_flock.end = file_lock->fl_end; | |
2316 | } else { | |
2317 | RETURN(-EINVAL); | |
2318 | } | |
2319 | flock.l_flock.pid = file_lock->fl_pid; | |
2320 | ||
2321 | /* Somewhat ugly workaround for svc lockd. | |
2322 | * lockd installs custom fl_lmops->lm_compare_owner that checks | |
2323 | * for the fl_owner to be the same (which it always is on local node | |
2324 | * I guess between lockd processes) and then compares pid. | |
2325 | * As such we assign pid to the owner field to make it all work, | |
2326 | * conflict with normal locks is unlikely since pid space and | |
2327 | * pointer space for current->files are not intersecting */ | |
2328 | if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner) | |
2329 | flock.l_flock.owner = (unsigned long)file_lock->fl_pid; | |
2330 | ||
2331 | switch (file_lock->fl_type) { | |
2332 | case F_RDLCK: | |
2333 | einfo.ei_mode = LCK_PR; | |
2334 | break; | |
2335 | case F_UNLCK: | |
2336 | /* An unlock request may or may not have any relation to | |
2337 | * existing locks so we may not be able to pass a lock handle | |
2338 | * via a normal ldlm_lock_cancel() request. The request may even | |
2339 | * unlock a byte range in the middle of an existing lock. In | |
2340 | * order to process an unlock request we need all of the same | |
2341 | * information that is given with a normal read or write record | |
2342 | * lock request. To avoid creating another ldlm unlock (cancel) | |
2343 | * message we'll treat a LCK_NL flock request as an unlock. */ | |
2344 | einfo.ei_mode = LCK_NL; | |
2345 | break; | |
2346 | case F_WRLCK: | |
2347 | einfo.ei_mode = LCK_PW; | |
2348 | break; | |
2349 | default: | |
2350 | CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", | |
2351 | file_lock->fl_type); | |
2352 | RETURN (-ENOTSUPP); | |
2353 | } | |
2354 | ||
2355 | switch (cmd) { | |
2356 | case F_SETLKW: | |
2357 | #ifdef F_SETLKW64 | |
2358 | case F_SETLKW64: | |
2359 | #endif | |
2360 | flags = 0; | |
2361 | break; | |
2362 | case F_SETLK: | |
2363 | #ifdef F_SETLK64 | |
2364 | case F_SETLK64: | |
2365 | #endif | |
2366 | flags = LDLM_FL_BLOCK_NOWAIT; | |
2367 | break; | |
2368 | case F_GETLK: | |
2369 | #ifdef F_GETLK64 | |
2370 | case F_GETLK64: | |
2371 | #endif | |
2372 | flags = LDLM_FL_TEST_LOCK; | |
2373 | /* Save the old mode so that if the mode in the lock changes we | |
2374 | * can decrement the appropriate reader or writer refcount. */ | |
2375 | file_lock->fl_type = einfo.ei_mode; | |
2376 | break; | |
2377 | default: | |
2378 | CERROR("unknown fcntl lock command: %d\n", cmd); | |
2379 | RETURN (-EINVAL); | |
2380 | } | |
2381 | ||
2382 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, | |
2383 | LUSTRE_OPC_ANY, NULL); | |
2384 | if (IS_ERR(op_data)) | |
2385 | RETURN(PTR_ERR(op_data)); | |
2386 | ||
2387 | CDEBUG(D_DLMTRACE, "inode=%lu, pid=%u, flags=%#x, mode=%u, " | |
2388 | "start="LPU64", end="LPU64"\n", inode->i_ino, flock.l_flock.pid, | |
2389 | flags, einfo.ei_mode, flock.l_flock.start, flock.l_flock.end); | |
2390 | ||
2391 | rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL, | |
2392 | op_data, &lockh, &flock, 0, NULL /* req */, flags); | |
2393 | ||
2394 | if ((file_lock->fl_flags & FL_FLOCK) && | |
2395 | (rc == 0 || file_lock->fl_type == F_UNLCK)) | |
2396 | rc2 = flock_lock_file_wait(file, file_lock); | |
2397 | if ((file_lock->fl_flags & FL_POSIX) && | |
2398 | (rc == 0 || file_lock->fl_type == F_UNLCK) && | |
2399 | !(flags & LDLM_FL_TEST_LOCK)) | |
2400 | rc2 = posix_lock_file_wait(file, file_lock); | |
2401 | ||
2402 | if (rc2 && file_lock->fl_type != F_UNLCK) { | |
2403 | einfo.ei_mode = LCK_NL; | |
2404 | md_enqueue(sbi->ll_md_exp, &einfo, NULL, | |
2405 | op_data, &lockh, &flock, 0, NULL /* req */, flags); | |
2406 | rc = rc2; | |
2407 | } | |
2408 | ||
2409 | ll_finish_md_op_data(op_data); | |
2410 | ||
2411 | RETURN(rc); | |
2412 | } | |
2413 | ||
2414 | int ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock) | |
2415 | { | |
2416 | ENTRY; | |
2417 | ||
2418 | RETURN(-ENOSYS); | |
2419 | } | |
2420 | ||
2421 | /** | |
2422 | * test if some locks matching bits and l_req_mode are acquired | |
2423 | * - bits can be in different locks | |
2424 | * - if found clear the common lock bits in *bits | |
2425 | * - the bits not found, are kept in *bits | |
2426 | * \param inode [IN] | |
2427 | * \param bits [IN] searched lock bits [IN] | |
2428 | * \param l_req_mode [IN] searched lock mode | |
2429 | * \retval boolean, true iff all bits are found | |
2430 | */ | |
2431 | int ll_have_md_lock(struct inode *inode, __u64 *bits, ldlm_mode_t l_req_mode) | |
2432 | { | |
2433 | struct lustre_handle lockh; | |
2434 | ldlm_policy_data_t policy; | |
2435 | ldlm_mode_t mode = (l_req_mode == LCK_MINMODE) ? | |
2436 | (LCK_CR|LCK_CW|LCK_PR|LCK_PW) : l_req_mode; | |
2437 | struct lu_fid *fid; | |
2438 | __u64 flags; | |
2439 | int i; | |
2440 | ENTRY; | |
2441 | ||
2442 | if (!inode) | |
2443 | RETURN(0); | |
2444 | ||
2445 | fid = &ll_i2info(inode)->lli_fid; | |
2446 | CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid), | |
2447 | ldlm_lockname[mode]); | |
2448 | ||
2449 | flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK; | |
2450 | for (i = 0; i < MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) { | |
2451 | policy.l_inodebits.bits = *bits & (1 << i); | |
2452 | if (policy.l_inodebits.bits == 0) | |
2453 | continue; | |
2454 | ||
2455 | if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS, | |
2456 | &policy, mode, &lockh)) { | |
2457 | struct ldlm_lock *lock; | |
2458 | ||
2459 | lock = ldlm_handle2lock(&lockh); | |
2460 | if (lock) { | |
2461 | *bits &= | |
2462 | ~(lock->l_policy_data.l_inodebits.bits); | |
2463 | LDLM_LOCK_PUT(lock); | |
2464 | } else { | |
2465 | *bits &= ~policy.l_inodebits.bits; | |
2466 | } | |
2467 | } | |
2468 | } | |
2469 | RETURN(*bits == 0); | |
2470 | } | |
2471 | ||
2472 | ldlm_mode_t ll_take_md_lock(struct inode *inode, __u64 bits, | |
2473 | struct lustre_handle *lockh, __u64 flags) | |
2474 | { | |
2475 | ldlm_policy_data_t policy = { .l_inodebits = {bits}}; | |
2476 | struct lu_fid *fid; | |
2477 | ldlm_mode_t rc; | |
2478 | ENTRY; | |
2479 | ||
2480 | fid = &ll_i2info(inode)->lli_fid; | |
2481 | CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid)); | |
2482 | ||
2483 | rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags, | |
2484 | fid, LDLM_IBITS, &policy, | |
2485 | LCK_CR|LCK_CW|LCK_PR|LCK_PW, lockh); | |
2486 | RETURN(rc); | |
2487 | } | |
2488 | ||
2489 | static int ll_inode_revalidate_fini(struct inode *inode, int rc) | |
2490 | { | |
2491 | /* Already unlinked. Just update nlink and return success */ | |
2492 | if (rc == -ENOENT) { | |
2493 | clear_nlink(inode); | |
2494 | /* This path cannot be hit for regular files unless in | |
2495 | * case of obscure races, so no need to to validate | |
2496 | * size. */ | |
2497 | if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode)) | |
2498 | return 0; | |
2499 | } else if (rc != 0) { | |
2500 | CERROR("%s: revalidate FID "DFID" error: rc = %d\n", | |
2501 | ll_get_fsname(inode->i_sb, NULL, 0), | |
2502 | PFID(ll_inode2fid(inode)), rc); | |
2503 | } | |
2504 | ||
2505 | return rc; | |
2506 | } | |
2507 | ||
2508 | int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, | |
2509 | __u64 ibits) | |
2510 | { | |
2511 | struct inode *inode = dentry->d_inode; | |
2512 | struct ptlrpc_request *req = NULL; | |
2513 | struct obd_export *exp; | |
2514 | int rc = 0; | |
2515 | ENTRY; | |
2516 | ||
2517 | LASSERT(inode != NULL); | |
2518 | ||
2519 | CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),name=%s\n", | |
2520 | inode->i_ino, inode->i_generation, inode, dentry->d_name.name); | |
2521 | ||
2522 | exp = ll_i2mdexp(inode); | |
2523 | ||
2524 | /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC. | |
2525 | * But under CMD case, it caused some lock issues, should be fixed | |
2526 | * with new CMD ibits lock. See bug 12718 */ | |
2527 | if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) { | |
2528 | struct lookup_intent oit = { .it_op = IT_GETATTR }; | |
2529 | struct md_op_data *op_data; | |
2530 | ||
2531 | if (ibits == MDS_INODELOCK_LOOKUP) | |
2532 | oit.it_op = IT_LOOKUP; | |
2533 | ||
2534 | /* Call getattr by fid, so do not provide name at all. */ | |
2535 | op_data = ll_prep_md_op_data(NULL, dentry->d_parent->d_inode, | |
2536 | dentry->d_inode, NULL, 0, 0, | |
2537 | LUSTRE_OPC_ANY, NULL); | |
2538 | if (IS_ERR(op_data)) | |
2539 | RETURN(PTR_ERR(op_data)); | |
2540 | ||
2541 | oit.it_create_mode |= M_CHECK_STALE; | |
2542 | rc = md_intent_lock(exp, op_data, NULL, 0, | |
2543 | /* we are not interested in name | |
2544 | based lookup */ | |
2545 | &oit, 0, &req, | |
2546 | ll_md_blocking_ast, 0); | |
2547 | ll_finish_md_op_data(op_data); | |
2548 | oit.it_create_mode &= ~M_CHECK_STALE; | |
2549 | if (rc < 0) { | |
2550 | rc = ll_inode_revalidate_fini(inode, rc); | |
2551 | GOTO (out, rc); | |
2552 | } | |
2553 | ||
2554 | rc = ll_revalidate_it_finish(req, &oit, dentry); | |
2555 | if (rc != 0) { | |
2556 | ll_intent_release(&oit); | |
2557 | GOTO(out, rc); | |
2558 | } | |
2559 | ||
2560 | /* Unlinked? Unhash dentry, so it is not picked up later by | |
2561 | do_lookup() -> ll_revalidate_it(). We cannot use d_drop | |
2562 | here to preserve get_cwd functionality on 2.6. | |
2563 | Bug 10503 */ | |
2564 | if (!dentry->d_inode->i_nlink) | |
2565 | d_lustre_invalidate(dentry); | |
2566 | ||
2567 | ll_lookup_finish_locks(&oit, dentry); | |
2568 | } else if (!ll_have_md_lock(dentry->d_inode, &ibits, LCK_MINMODE)) { | |
2569 | struct ll_sb_info *sbi = ll_i2sbi(dentry->d_inode); | |
2570 | obd_valid valid = OBD_MD_FLGETATTR; | |
2571 | struct md_op_data *op_data; | |
2572 | int ealen = 0; | |
2573 | ||
2574 | if (S_ISREG(inode->i_mode)) { | |
2575 | rc = ll_get_max_mdsize(sbi, &ealen); | |
2576 | if (rc) | |
2577 | RETURN(rc); | |
2578 | valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE; | |
2579 | } | |
2580 | ||
2581 | op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, | |
2582 | 0, ealen, LUSTRE_OPC_ANY, | |
2583 | NULL); | |
2584 | if (IS_ERR(op_data)) | |
2585 | RETURN(PTR_ERR(op_data)); | |
2586 | ||
2587 | op_data->op_valid = valid; | |
2588 | /* Once OBD_CONNECT_ATTRFID is not supported, we can't find one | |
2589 | * capa for this inode. Because we only keep capas of dirs | |
2590 | * fresh. */ | |
2591 | rc = md_getattr(sbi->ll_md_exp, op_data, &req); | |
2592 | ll_finish_md_op_data(op_data); | |
2593 | if (rc) { | |
2594 | rc = ll_inode_revalidate_fini(inode, rc); | |
2595 | RETURN(rc); | |
2596 | } | |
2597 | ||
2598 | rc = ll_prep_inode(&inode, req, NULL, NULL); | |
2599 | } | |
2600 | out: | |
2601 | ptlrpc_req_finished(req); | |
2602 | return rc; | |
2603 | } | |
2604 | ||
2605 | int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, | |
2606 | __u64 ibits) | |
2607 | { | |
2608 | struct inode *inode = dentry->d_inode; | |
2609 | int rc; | |
2610 | ENTRY; | |
2611 | ||
2612 | rc = __ll_inode_revalidate_it(dentry, it, ibits); | |
2613 | if (rc != 0) | |
2614 | RETURN(rc); | |
2615 | ||
2616 | /* if object isn't regular file, don't validate size */ | |
2617 | if (!S_ISREG(inode->i_mode)) { | |
2618 | LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_lvb.lvb_atime; | |
2619 | LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_lvb.lvb_mtime; | |
2620 | LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_lvb.lvb_ctime; | |
2621 | } else { | |
2622 | rc = ll_glimpse_size(inode); | |
2623 | } | |
2624 | RETURN(rc); | |
2625 | } | |
2626 | ||
2627 | int ll_getattr_it(struct vfsmount *mnt, struct dentry *de, | |
2628 | struct lookup_intent *it, struct kstat *stat) | |
2629 | { | |
2630 | struct inode *inode = de->d_inode; | |
2631 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
2632 | struct ll_inode_info *lli = ll_i2info(inode); | |
2633 | int res = 0; | |
2634 | ||
2635 | res = ll_inode_revalidate_it(de, it, MDS_INODELOCK_UPDATE | | |
2636 | MDS_INODELOCK_LOOKUP); | |
2637 | ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1); | |
2638 | ||
2639 | if (res) | |
2640 | return res; | |
2641 | ||
2642 | stat->dev = inode->i_sb->s_dev; | |
2643 | if (ll_need_32bit_api(sbi)) | |
2644 | stat->ino = cl_fid_build_ino(&lli->lli_fid, 1); | |
2645 | else | |
2646 | stat->ino = inode->i_ino; | |
2647 | stat->mode = inode->i_mode; | |
2648 | stat->nlink = inode->i_nlink; | |
2649 | stat->uid = inode->i_uid; | |
2650 | stat->gid = inode->i_gid; | |
2651 | stat->rdev = inode->i_rdev; | |
2652 | stat->atime = inode->i_atime; | |
2653 | stat->mtime = inode->i_mtime; | |
2654 | stat->ctime = inode->i_ctime; | |
2655 | stat->blksize = 1 << inode->i_blkbits; | |
2656 | ||
2657 | stat->size = i_size_read(inode); | |
2658 | stat->blocks = inode->i_blocks; | |
2659 | ||
2660 | return 0; | |
2661 | } | |
2662 | int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat) | |
2663 | { | |
2664 | struct lookup_intent it = { .it_op = IT_GETATTR }; | |
2665 | ||
2666 | return ll_getattr_it(mnt, de, &it, stat); | |
2667 | } | |
2668 | ||
2669 | ||
2670 | struct posix_acl * ll_get_acl(struct inode *inode, int type) | |
2671 | { | |
2672 | struct ll_inode_info *lli = ll_i2info(inode); | |
2673 | struct posix_acl *acl = NULL; | |
2674 | ENTRY; | |
2675 | ||
2676 | spin_lock(&lli->lli_lock); | |
2677 | /* VFS' acl_permission_check->check_acl will release the refcount */ | |
2678 | acl = posix_acl_dup(lli->lli_posix_acl); | |
2679 | spin_unlock(&lli->lli_lock); | |
2680 | ||
2681 | RETURN(acl); | |
2682 | } | |
2683 | ||
2684 | ||
2685 | int ll_inode_permission(struct inode *inode, int mask) | |
2686 | { | |
2687 | int rc = 0; | |
2688 | ENTRY; | |
2689 | ||
2690 | #ifdef MAY_NOT_BLOCK | |
2691 | if (mask & MAY_NOT_BLOCK) | |
2692 | return -ECHILD; | |
2693 | #endif | |
2694 | ||
2695 | /* as root inode are NOT getting validated in lookup operation, | |
2696 | * need to do it before permission check. */ | |
2697 | ||
2698 | if (inode == inode->i_sb->s_root->d_inode) { | |
2699 | struct lookup_intent it = { .it_op = IT_LOOKUP }; | |
2700 | ||
2701 | rc = __ll_inode_revalidate_it(inode->i_sb->s_root, &it, | |
2702 | MDS_INODELOCK_LOOKUP); | |
2703 | if (rc) | |
2704 | RETURN(rc); | |
2705 | } | |
2706 | ||
2707 | CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), inode mode %x mask %o\n", | |
2708 | inode->i_ino, inode->i_generation, inode, inode->i_mode, mask); | |
2709 | ||
2710 | if (ll_i2sbi(inode)->ll_flags & LL_SBI_RMT_CLIENT) | |
2711 | return lustre_check_remote_perm(inode, mask); | |
2712 | ||
2713 | ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1); | |
2714 | rc = ll_generic_permission(inode, mask, flags, ll_check_acl); | |
2715 | ||
2716 | RETURN(rc); | |
2717 | } | |
2718 | ||
2719 | #define READ_METHOD aio_read | |
2720 | #define READ_FUNCTION ll_file_aio_read | |
2721 | #define WRITE_METHOD aio_write | |
2722 | #define WRITE_FUNCTION ll_file_aio_write | |
2723 | ||
2724 | /* -o localflock - only provides locally consistent flock locks */ | |
2725 | struct file_operations ll_file_operations = { | |
2726 | .read = ll_file_read, | |
2727 | .READ_METHOD = READ_FUNCTION, | |
2728 | .write = ll_file_write, | |
2729 | .WRITE_METHOD = WRITE_FUNCTION, | |
2730 | .unlocked_ioctl = ll_file_ioctl, | |
2731 | .open = ll_file_open, | |
2732 | .release = ll_file_release, | |
2733 | .mmap = ll_file_mmap, | |
2734 | .llseek = ll_file_seek, | |
2735 | .splice_read = ll_file_splice_read, | |
2736 | .fsync = ll_fsync, | |
2737 | .flush = ll_flush | |
2738 | }; | |
2739 | ||
2740 | struct file_operations ll_file_operations_flock = { | |
2741 | .read = ll_file_read, | |
2742 | .READ_METHOD = READ_FUNCTION, | |
2743 | .write = ll_file_write, | |
2744 | .WRITE_METHOD = WRITE_FUNCTION, | |
2745 | .unlocked_ioctl = ll_file_ioctl, | |
2746 | .open = ll_file_open, | |
2747 | .release = ll_file_release, | |
2748 | .mmap = ll_file_mmap, | |
2749 | .llseek = ll_file_seek, | |
2750 | .splice_read = ll_file_splice_read, | |
2751 | .fsync = ll_fsync, | |
2752 | .flush = ll_flush, | |
2753 | .flock = ll_file_flock, | |
2754 | .lock = ll_file_flock | |
2755 | }; | |
2756 | ||
2757 | /* These are for -o noflock - to return ENOSYS on flock calls */ | |
2758 | struct file_operations ll_file_operations_noflock = { | |
2759 | .read = ll_file_read, | |
2760 | .READ_METHOD = READ_FUNCTION, | |
2761 | .write = ll_file_write, | |
2762 | .WRITE_METHOD = WRITE_FUNCTION, | |
2763 | .unlocked_ioctl = ll_file_ioctl, | |
2764 | .open = ll_file_open, | |
2765 | .release = ll_file_release, | |
2766 | .mmap = ll_file_mmap, | |
2767 | .llseek = ll_file_seek, | |
2768 | .splice_read = ll_file_splice_read, | |
2769 | .fsync = ll_fsync, | |
2770 | .flush = ll_flush, | |
2771 | .flock = ll_file_noflock, | |
2772 | .lock = ll_file_noflock | |
2773 | }; | |
2774 | ||
2775 | struct inode_operations ll_file_inode_operations = { | |
2776 | .setattr = ll_setattr, | |
2777 | .getattr = ll_getattr, | |
2778 | .permission = ll_inode_permission, | |
2779 | .setxattr = ll_setxattr, | |
2780 | .getxattr = ll_getxattr, | |
2781 | .listxattr = ll_listxattr, | |
2782 | .removexattr = ll_removexattr, | |
2783 | .get_acl = ll_get_acl, | |
2784 | }; | |
2785 | ||
2786 | /* dynamic ioctl number support routins */ | |
2787 | static struct llioc_ctl_data { | |
2788 | struct rw_semaphore ioc_sem; | |
2789 | struct list_head ioc_head; | |
2790 | } llioc = { | |
2791 | __RWSEM_INITIALIZER(llioc.ioc_sem), | |
2792 | LIST_HEAD_INIT(llioc.ioc_head) | |
2793 | }; | |
2794 | ||
2795 | ||
2796 | struct llioc_data { | |
2797 | struct list_head iocd_list; | |
2798 | unsigned int iocd_size; | |
2799 | llioc_callback_t iocd_cb; | |
2800 | unsigned int iocd_count; | |
2801 | unsigned int iocd_cmd[0]; | |
2802 | }; | |
2803 | ||
2804 | void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd) | |
2805 | { | |
2806 | unsigned int size; | |
2807 | struct llioc_data *in_data = NULL; | |
2808 | ENTRY; | |
2809 | ||
2810 | if (cb == NULL || cmd == NULL || | |
2811 | count > LLIOC_MAX_CMD || count < 0) | |
2812 | RETURN(NULL); | |
2813 | ||
2814 | size = sizeof(*in_data) + count * sizeof(unsigned int); | |
2815 | OBD_ALLOC(in_data, size); | |
2816 | if (in_data == NULL) | |
2817 | RETURN(NULL); | |
2818 | ||
2819 | memset(in_data, 0, sizeof(*in_data)); | |
2820 | in_data->iocd_size = size; | |
2821 | in_data->iocd_cb = cb; | |
2822 | in_data->iocd_count = count; | |
2823 | memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count); | |
2824 | ||
2825 | down_write(&llioc.ioc_sem); | |
2826 | list_add_tail(&in_data->iocd_list, &llioc.ioc_head); | |
2827 | up_write(&llioc.ioc_sem); | |
2828 | ||
2829 | RETURN(in_data); | |
2830 | } | |
2831 | ||
2832 | void ll_iocontrol_unregister(void *magic) | |
2833 | { | |
2834 | struct llioc_data *tmp; | |
2835 | ||
2836 | if (magic == NULL) | |
2837 | return; | |
2838 | ||
2839 | down_write(&llioc.ioc_sem); | |
2840 | list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) { | |
2841 | if (tmp == magic) { | |
2842 | unsigned int size = tmp->iocd_size; | |
2843 | ||
2844 | list_del(&tmp->iocd_list); | |
2845 | up_write(&llioc.ioc_sem); | |
2846 | ||
2847 | OBD_FREE(tmp, size); | |
2848 | return; | |
2849 | } | |
2850 | } | |
2851 | up_write(&llioc.ioc_sem); | |
2852 | ||
2853 | CWARN("didn't find iocontrol register block with magic: %p\n", magic); | |
2854 | } | |
2855 | ||
2856 | EXPORT_SYMBOL(ll_iocontrol_register); | |
2857 | EXPORT_SYMBOL(ll_iocontrol_unregister); | |
2858 | ||
2859 | enum llioc_iter ll_iocontrol_call(struct inode *inode, struct file *file, | |
2860 | unsigned int cmd, unsigned long arg, int *rcp) | |
2861 | { | |
2862 | enum llioc_iter ret = LLIOC_CONT; | |
2863 | struct llioc_data *data; | |
2864 | int rc = -EINVAL, i; | |
2865 | ||
2866 | down_read(&llioc.ioc_sem); | |
2867 | list_for_each_entry(data, &llioc.ioc_head, iocd_list) { | |
2868 | for (i = 0; i < data->iocd_count; i++) { | |
2869 | if (cmd != data->iocd_cmd[i]) | |
2870 | continue; | |
2871 | ||
2872 | ret = data->iocd_cb(inode, file, cmd, arg, data, &rc); | |
2873 | break; | |
2874 | } | |
2875 | ||
2876 | if (ret == LLIOC_STOP) | |
2877 | break; | |
2878 | } | |
2879 | up_read(&llioc.ioc_sem); | |
2880 | ||
2881 | if (rcp) | |
2882 | *rcp = rc; | |
2883 | return ret; | |
2884 | } | |
2885 | ||
2886 | int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf) | |
2887 | { | |
2888 | struct ll_inode_info *lli = ll_i2info(inode); | |
2889 | struct cl_env_nest nest; | |
2890 | struct lu_env *env; | |
2891 | int result; | |
2892 | ENTRY; | |
2893 | ||
2894 | if (lli->lli_clob == NULL) | |
2895 | RETURN(0); | |
2896 | ||
2897 | env = cl_env_nested_get(&nest); | |
2898 | if (IS_ERR(env)) | |
2899 | RETURN(PTR_ERR(env)); | |
2900 | ||
2901 | result = cl_conf_set(env, lli->lli_clob, conf); | |
2902 | cl_env_nested_put(&nest, env); | |
2903 | ||
2904 | if (conf->coc_opc == OBJECT_CONF_SET) { | |
2905 | struct ldlm_lock *lock = conf->coc_lock; | |
2906 | ||
2907 | LASSERT(lock != NULL); | |
2908 | LASSERT(ldlm_has_layout(lock)); | |
2909 | if (result == 0) { | |
2910 | /* it can only be allowed to match after layout is | |
2911 | * applied to inode otherwise false layout would be | |
2912 | * seen. Applying layout shoud happen before dropping | |
2913 | * the intent lock. */ | |
2914 | ldlm_lock_allow_match(lock); | |
2915 | } | |
2916 | } | |
2917 | RETURN(result); | |
2918 | } | |
2919 | ||
2920 | /* Fetch layout from MDT with getxattr request, if it's not ready yet */ | |
2921 | static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock) | |
2922 | ||
2923 | { | |
2924 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
2925 | struct obd_capa *oc; | |
2926 | struct ptlrpc_request *req; | |
2927 | struct mdt_body *body; | |
2928 | void *lvbdata; | |
2929 | void *lmm; | |
2930 | int lmmsize; | |
2931 | int rc; | |
2932 | ENTRY; | |
2933 | ||
2934 | if (lock->l_lvb_data != NULL) | |
2935 | RETURN(0); | |
2936 | ||
2937 | /* if layout lock was granted right away, the layout is returned | |
2938 | * within DLM_LVB of dlm reply; otherwise if the lock was ever | |
2939 | * blocked and then granted via completion ast, we have to fetch | |
2940 | * layout here. Please note that we can't use the LVB buffer in | |
2941 | * completion AST because it doesn't have a large enough buffer */ | |
2942 | oc = ll_mdscapa_get(inode); | |
2943 | rc = ll_get_max_mdsize(sbi, &lmmsize); | |
2944 | if (rc == 0) | |
2945 | rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), oc, | |
2946 | OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0, | |
2947 | lmmsize, 0, &req); | |
2948 | capa_put(oc); | |
2949 | if (rc < 0) | |
2950 | RETURN(rc); | |
2951 | ||
2952 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); | |
2953 | if (body == NULL || body->eadatasize > lmmsize) | |
2954 | GOTO(out, rc = -EPROTO); | |
2955 | ||
2956 | lmmsize = body->eadatasize; | |
2957 | if (lmmsize == 0) /* empty layout */ | |
2958 | GOTO(out, rc = 0); | |
2959 | ||
2960 | lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize); | |
2961 | if (lmm == NULL) | |
2962 | GOTO(out, rc = -EFAULT); | |
2963 | ||
2964 | OBD_ALLOC_LARGE(lvbdata, lmmsize); | |
2965 | if (lvbdata == NULL) | |
2966 | GOTO(out, rc = -ENOMEM); | |
2967 | ||
2968 | memcpy(lvbdata, lmm, lmmsize); | |
2969 | lock_res_and_lock(lock); | |
2970 | if (lock->l_lvb_data == NULL) { | |
2971 | lock->l_lvb_data = lvbdata; | |
2972 | lock->l_lvb_len = lmmsize; | |
2973 | lvbdata = NULL; | |
2974 | } | |
2975 | unlock_res_and_lock(lock); | |
2976 | ||
2977 | if (lvbdata != NULL) | |
2978 | OBD_FREE_LARGE(lvbdata, lmmsize); | |
2979 | EXIT; | |
2980 | ||
2981 | out: | |
2982 | ptlrpc_req_finished(req); | |
2983 | return rc; | |
2984 | } | |
2985 | ||
2986 | /** | |
2987 | * Apply the layout to the inode. Layout lock is held and will be released | |
2988 | * in this function. | |
2989 | */ | |
2990 | static int ll_layout_lock_set(struct lustre_handle *lockh, ldlm_mode_t mode, | |
2991 | struct inode *inode, __u32 *gen, bool reconf) | |
2992 | { | |
2993 | struct ll_inode_info *lli = ll_i2info(inode); | |
2994 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
2995 | struct ldlm_lock *lock; | |
2996 | struct lustre_md md = { NULL }; | |
2997 | struct cl_object_conf conf; | |
2998 | int rc = 0; | |
2999 | bool lvb_ready; | |
3000 | bool wait_layout = false; | |
3001 | ENTRY; | |
3002 | ||
3003 | LASSERT(lustre_handle_is_used(lockh)); | |
3004 | ||
3005 | lock = ldlm_handle2lock(lockh); | |
3006 | LASSERT(lock != NULL); | |
3007 | LASSERT(ldlm_has_layout(lock)); | |
3008 | ||
3009 | LDLM_DEBUG(lock, "File %p/"DFID" being reconfigured: %d.\n", | |
3010 | inode, PFID(&lli->lli_fid), reconf); | |
3011 | ||
3012 | lock_res_and_lock(lock); | |
3013 | lvb_ready = !!(lock->l_flags & LDLM_FL_LVB_READY); | |
3014 | unlock_res_and_lock(lock); | |
3015 | /* checking lvb_ready is racy but this is okay. The worst case is | |
3016 | * that multi processes may configure the file on the same time. */ | |
3017 | if (lvb_ready || !reconf) { | |
3018 | rc = -ENODATA; | |
3019 | if (lvb_ready) { | |
3020 | /* layout_gen must be valid if layout lock is not | |
3021 | * cancelled and stripe has already set */ | |
3022 | *gen = lli->lli_layout_gen; | |
3023 | rc = 0; | |
3024 | } | |
3025 | GOTO(out, rc); | |
3026 | } | |
3027 | ||
3028 | rc = ll_layout_fetch(inode, lock); | |
3029 | if (rc < 0) | |
3030 | GOTO(out, rc); | |
3031 | ||
3032 | /* for layout lock, lmm is returned in lock's lvb. | |
3033 | * lvb_data is immutable if the lock is held so it's safe to access it | |
3034 | * without res lock. See the description in ldlm_lock_decref_internal() | |
3035 | * for the condition to free lvb_data of layout lock */ | |
3036 | if (lock->l_lvb_data != NULL) { | |
3037 | rc = obd_unpackmd(sbi->ll_dt_exp, &md.lsm, | |
3038 | lock->l_lvb_data, lock->l_lvb_len); | |
3039 | if (rc >= 0) { | |
3040 | *gen = LL_LAYOUT_GEN_EMPTY; | |
3041 | if (md.lsm != NULL) | |
3042 | *gen = md.lsm->lsm_layout_gen; | |
3043 | rc = 0; | |
3044 | } else { | |
3045 | CERROR("%s: file "DFID" unpackmd error: %d\n", | |
3046 | ll_get_fsname(inode->i_sb, NULL, 0), | |
3047 | PFID(&lli->lli_fid), rc); | |
3048 | } | |
3049 | } | |
3050 | if (rc < 0) | |
3051 | GOTO(out, rc); | |
3052 | ||
3053 | /* set layout to file. Unlikely this will fail as old layout was | |
3054 | * surely eliminated */ | |
3055 | memset(&conf, 0, sizeof conf); | |
3056 | conf.coc_opc = OBJECT_CONF_SET; | |
3057 | conf.coc_inode = inode; | |
3058 | conf.coc_lock = lock; | |
3059 | conf.u.coc_md = &md; | |
3060 | rc = ll_layout_conf(inode, &conf); | |
3061 | ||
3062 | if (md.lsm != NULL) | |
3063 | obd_free_memmd(sbi->ll_dt_exp, &md.lsm); | |
3064 | ||
3065 | /* refresh layout failed, need to wait */ | |
3066 | wait_layout = rc == -EBUSY; | |
3067 | EXIT; | |
3068 | ||
3069 | out: | |
3070 | LDLM_LOCK_PUT(lock); | |
3071 | ldlm_lock_decref(lockh, mode); | |
3072 | ||
3073 | /* wait for IO to complete if it's still being used. */ | |
3074 | if (wait_layout) { | |
3075 | CDEBUG(D_INODE, "%s: %p/"DFID" wait for layout reconf.\n", | |
3076 | ll_get_fsname(inode->i_sb, NULL, 0), | |
3077 | inode, PFID(&lli->lli_fid)); | |
3078 | ||
3079 | memset(&conf, 0, sizeof conf); | |
3080 | conf.coc_opc = OBJECT_CONF_WAIT; | |
3081 | conf.coc_inode = inode; | |
3082 | rc = ll_layout_conf(inode, &conf); | |
3083 | if (rc == 0) | |
3084 | rc = -EAGAIN; | |
3085 | ||
3086 | CDEBUG(D_INODE, "file: "DFID" waiting layout return: %d.\n", | |
3087 | PFID(&lli->lli_fid), rc); | |
3088 | } | |
3089 | RETURN(rc); | |
3090 | } | |
3091 | ||
3092 | /** | |
3093 | * This function checks if there exists a LAYOUT lock on the client side, | |
3094 | * or enqueues it if it doesn't have one in cache. | |
3095 | * | |
3096 | * This function will not hold layout lock so it may be revoked any time after | |
3097 | * this function returns. Any operations depend on layout should be redone | |
3098 | * in that case. | |
3099 | * | |
3100 | * This function should be called before lov_io_init() to get an uptodate | |
3101 | * layout version, the caller should save the version number and after IO | |
3102 | * is finished, this function should be called again to verify that layout | |
3103 | * is not changed during IO time. | |
3104 | */ | |
3105 | int ll_layout_refresh(struct inode *inode, __u32 *gen) | |
3106 | { | |
3107 | struct ll_inode_info *lli = ll_i2info(inode); | |
3108 | struct ll_sb_info *sbi = ll_i2sbi(inode); | |
3109 | struct md_op_data *op_data; | |
3110 | struct lookup_intent it; | |
3111 | struct lustre_handle lockh; | |
3112 | ldlm_mode_t mode; | |
3113 | struct ldlm_enqueue_info einfo = { .ei_type = LDLM_IBITS, | |
3114 | .ei_mode = LCK_CR, | |
3115 | .ei_cb_bl = ll_md_blocking_ast, | |
3116 | .ei_cb_cp = ldlm_completion_ast, | |
3117 | .ei_cbdata = NULL }; | |
3118 | int rc; | |
3119 | ENTRY; | |
3120 | ||
3121 | *gen = lli->lli_layout_gen; | |
3122 | if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK)) | |
3123 | RETURN(0); | |
3124 | ||
3125 | /* sanity checks */ | |
3126 | LASSERT(fid_is_sane(ll_inode2fid(inode))); | |
3127 | LASSERT(S_ISREG(inode->i_mode)); | |
3128 | ||
3129 | /* mostly layout lock is caching on the local side, so try to match | |
3130 | * it before grabbing layout lock mutex. */ | |
3131 | mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0); | |
3132 | if (mode != 0) { /* hit cached lock */ | |
3133 | rc = ll_layout_lock_set(&lockh, mode, inode, gen, false); | |
3134 | if (rc == 0) | |
3135 | RETURN(0); | |
3136 | ||
3137 | /* better hold lli_layout_mutex to try again otherwise | |
3138 | * it will have starvation problem. */ | |
3139 | } | |
3140 | ||
3141 | /* take layout lock mutex to enqueue layout lock exclusively. */ | |
3142 | mutex_lock(&lli->lli_layout_mutex); | |
3143 | ||
3144 | again: | |
3145 | /* try again. Maybe somebody else has done this. */ | |
3146 | mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0); | |
3147 | if (mode != 0) { /* hit cached lock */ | |
3148 | rc = ll_layout_lock_set(&lockh, mode, inode, gen, true); | |
3149 | if (rc == -EAGAIN) | |
3150 | goto again; | |
3151 | ||
3152 | mutex_unlock(&lli->lli_layout_mutex); | |
3153 | RETURN(rc); | |
3154 | } | |
3155 | ||
3156 | op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, | |
3157 | 0, 0, LUSTRE_OPC_ANY, NULL); | |
3158 | if (IS_ERR(op_data)) { | |
3159 | mutex_unlock(&lli->lli_layout_mutex); | |
3160 | RETURN(PTR_ERR(op_data)); | |
3161 | } | |
3162 | ||
3163 | /* have to enqueue one */ | |
3164 | memset(&it, 0, sizeof(it)); | |
3165 | it.it_op = IT_LAYOUT; | |
3166 | lockh.cookie = 0ULL; | |
3167 | ||
3168 | LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file %p/"DFID".\n", | |
3169 | ll_get_fsname(inode->i_sb, NULL, 0), inode, | |
3170 | PFID(&lli->lli_fid)); | |
3171 | ||
3172 | rc = md_enqueue(sbi->ll_md_exp, &einfo, &it, op_data, &lockh, | |
3173 | NULL, 0, NULL, 0); | |
3174 | if (it.d.lustre.it_data != NULL) | |
3175 | ptlrpc_req_finished(it.d.lustre.it_data); | |
3176 | it.d.lustre.it_data = NULL; | |
3177 | ||
3178 | ll_finish_md_op_data(op_data); | |
3179 | ||
3180 | md_set_lock_data(sbi->ll_md_exp, &it.d.lustre.it_lock_handle, inode, NULL); | |
3181 | ||
3182 | mode = it.d.lustre.it_lock_mode; | |
3183 | it.d.lustre.it_lock_mode = 0; | |
3184 | ll_intent_drop_lock(&it); | |
3185 | ||
3186 | if (rc == 0) { | |
3187 | /* set lock data in case this is a new lock */ | |
3188 | ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL); | |
3189 | rc = ll_layout_lock_set(&lockh, mode, inode, gen, true); | |
3190 | if (rc == -EAGAIN) | |
3191 | goto again; | |
3192 | } | |
3193 | mutex_unlock(&lli->lli_layout_mutex); | |
3194 | ||
3195 | RETURN(rc); | |
3196 | } |