Import Upstream version 1.8.5
[hcoop/debian/openafs.git] / src / afs / LINUX / osi_file.c
1 /*
2 * Copyright 2000, International Business Machines Corporation and others.
3 * All Rights Reserved.
4 *
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
8 */
9
10 #include <afsconfig.h>
11 #include "afs/param.h"
12
13
14 #include <linux/module.h> /* early to avoid printf->printk mapping */
15 #include "afs/sysincludes.h" /* Standard vendor system headers */
16 #include "afsincludes.h" /* Afs-based standard headers */
17 #include "afs/afs_stats.h" /* afs statistics */
18 #include <linux/namei.h>
19
20 #if defined(HAVE_LINUX_EXPORTFS_H)
21 #include <linux/exportfs.h>
22 #endif
23 #include "osi_compat.h"
24
25 #ifndef CURRENT_TIME
26 # if defined(HAVE_LINUX_KTIME_GET_COARSE_REAL_TS64)
27 # define AFS_CURRENT_TIME(x) (ktime_get_coarse_real_ts64((x)))
28 # else
29 # ifdef IATTR_TAKES_64BIT_TIME
30 # define AFS_CURRENT_TIME(x) do {*(x) = current_kernel_time64();} while (0)
31 # else
32 # define AFS_CURRENT_TIME(x) do {*(x) = current_kernel_time();} while (0)
33 # endif
34 # endif
35 #else
36 # define AFS_CURRENT_TIME(x) do {*(x) = CURRENT_TIME;} while(0)
37 #endif
38
39 int cache_fh_type = -1;
40 int cache_fh_len = -1;
41
42 extern struct osi_dev cacheDev;
43 extern struct vfsmount *afs_cacheMnt;
44 extern struct super_block *afs_cacheSBp;
45 #if defined(STRUCT_TASK_STRUCT_HAS_CRED)
46 extern struct cred *cache_creds;
47 #endif
48
49 /* Old export ops: decode_fh will call back here. Accept any dentry it suggests */
50 int
51 afs_fh_acceptable(void *context, struct dentry *dp)
52 {
53 return 1;
54 }
55
56 struct file *
57 afs_linux_raw_open(afs_dcache_id_t *ainode)
58 {
59 struct inode *tip = NULL;
60 struct dentry *dp = NULL;
61 struct file* filp;
62
63 dp = afs_get_dentry_from_fh(afs_cacheSBp, ainode, cache_fh_len, cache_fh_type,
64 afs_fh_acceptable);
65 if ((!dp) || IS_ERR(dp))
66 osi_Panic("Can't get dentry\n");
67 tip = dp->d_inode;
68 tip->i_flags |= S_NOATIME; /* Disable updating access times. */
69
70 #if defined(STRUCT_TASK_STRUCT_HAS_CRED)
71 /* Use stashed credentials - prevent selinux/apparmor problems */
72 filp = afs_dentry_open(dp, afs_cacheMnt, O_RDWR, cache_creds);
73 if (IS_ERR(filp))
74 filp = afs_dentry_open(dp, afs_cacheMnt, O_RDWR, current_cred());
75 #else
76 filp = dentry_open(dget(dp), mntget(afs_cacheMnt), O_RDWR);
77 #endif
78 if (IS_ERR(filp)) {
79 afs_warn("afs: Cannot open cache file (code %d). Trying to continue, "
80 "but AFS accesses may return errors or panic the system\n",
81 (int) PTR_ERR(filp));
82 filp = NULL;
83 }
84
85 dput(dp);
86
87 return filp;
88 }
89
90 void *
91 osi_UFSOpen(afs_dcache_id_t *ainode)
92 {
93 struct osi_file *afile = NULL;
94 extern int cacheDiskType;
95
96 AFS_STATCNT(osi_UFSOpen);
97 if (cacheDiskType != AFS_FCACHE_TYPE_UFS) {
98 osi_Panic("UFSOpen called for non-UFS cache\n");
99 }
100 if (!afs_osicred_initialized) {
101 /* valid for alpha_osf, SunOS, Ultrix */
102 memset(&afs_osi_cred, 0, sizeof(afs_ucred_t));
103 crhold(&afs_osi_cred); /* don't let it evaporate, since it is static */
104 afs_osicred_initialized = 1;
105 }
106 AFS_GUNLOCK();
107 afile = kmalloc(sizeof(struct osi_file), GFP_NOFS);
108 if (!afile) {
109 osi_Panic("osi_UFSOpen: Failed to allocate %d bytes for osi_file.\n",
110 (int)sizeof(struct osi_file));
111 }
112 memset(afile, 0, sizeof(struct osi_file));
113
114 afile->filp = afs_linux_raw_open(ainode);
115 if (afile->filp) {
116 afile->size = i_size_read(FILE_INODE(afile->filp));
117 }
118 AFS_GLOCK();
119
120 if (!afile->filp) {
121 osi_FreeLargeSpace(afile);
122 return NULL;
123 }
124
125 afile->offset = 0;
126 afile->proc = (int (*)())0;
127 return (void *)afile;
128 }
129
130 /*
131 * Given a dentry, return the file handle as encoded by the filesystem.
132 * We can't assume anything about the length (words, not bytes).
133 * The cache has to live on a single filesystem with uniform file
134 * handles, otherwise we panic.
135 */
136 void osi_get_fh(struct dentry *dp, afs_ufs_dcache_id_t *ainode) {
137 int max_len;
138 int type;
139
140 if (cache_fh_len > 0)
141 max_len = cache_fh_len;
142 else
143 max_len = MAX_FH_LEN;
144 type = afs_get_fh_from_dentry(dp, ainode, &max_len);
145 if (type == 255) {
146 osi_Panic("File handle encoding failed\n");
147 }
148 if (cache_fh_type < 0)
149 cache_fh_type = type;
150 if (cache_fh_len < 0) {
151 cache_fh_len = max_len;
152 }
153 if (type != cache_fh_type || max_len != cache_fh_len) {
154 osi_Panic("Inconsistent file handles within cache\n");
155 }
156 }
157
158 int
159 afs_osi_Stat(struct osi_file *afile, struct osi_stat *astat)
160 {
161 AFS_STATCNT(osi_Stat);
162 astat->size = i_size_read(OSIFILE_INODE(afile));
163 astat->mtime = OSIFILE_INODE(afile)->i_mtime.tv_sec;
164 astat->atime = OSIFILE_INODE(afile)->i_atime.tv_sec;
165
166 return 0;
167 }
168
169 int
170 osi_UFSClose(struct osi_file *afile)
171 {
172 AFS_STATCNT(osi_Close);
173 if (afile) {
174 if (OSIFILE_INODE(afile)) {
175 filp_close(afile->filp, NULL);
176 }
177 }
178 kfree(afile);
179 return 0;
180 }
181
182 int
183 osi_UFSTruncate(struct osi_file *afile, afs_int32 asize)
184 {
185 afs_int32 code;
186 struct osi_stat tstat;
187 struct iattr newattrs;
188 struct inode *inode = OSIFILE_INODE(afile);
189 AFS_STATCNT(osi_Truncate);
190
191 /* This routine only shrinks files, and most systems
192 * have very slow truncates, even when the file is already
193 * small enough. Check now and save some time.
194 */
195 code = afs_osi_Stat(afile, &tstat);
196 if (code || tstat.size <= asize)
197 return code;
198 AFS_GUNLOCK();
199 afs_linux_lock_inode(inode);
200 #ifdef STRUCT_INODE_HAS_I_ALLOC_SEM
201 down_write(&inode->i_alloc_sem);
202 #endif
203 newattrs.ia_size = asize;
204 newattrs.ia_valid = ATTR_SIZE | ATTR_CTIME;
205 AFS_CURRENT_TIME(&newattrs.ia_ctime);
206
207 /* avoid notify_change() since it wants to update dentry->d_parent */
208 #ifdef HAVE_LINUX_SETATTR_PREPARE
209 code = setattr_prepare(file_dentry(afile->filp), &newattrs);
210 #else
211 code = inode_change_ok(inode, &newattrs);
212 #endif
213 if (!code)
214 code = afs_inode_setattr(afile, &newattrs);
215 if (!code)
216 truncate_inode_pages(&inode->i_data, asize);
217 code = -code;
218 #ifdef STRUCT_INODE_HAS_I_ALLOC_SEM
219 up_write(&inode->i_alloc_sem);
220 #endif
221 afs_linux_unlock_inode(inode);
222 AFS_GLOCK();
223 return code;
224 }
225
226
227 /* Generic read interface */
228 int
229 afs_osi_Read(struct osi_file *afile, int offset, void *aptr,
230 afs_int32 asize)
231 {
232 struct uio auio;
233 struct iovec iov;
234 afs_int32 code;
235
236 memset(&auio, 0, sizeof(auio));
237 memset(&iov, 0, sizeof(iov));
238
239 AFS_STATCNT(osi_Read);
240
241 /*
242 * If the osi_file passed in is NULL, panic only if AFS is not shutting
243 * down. No point in crashing when we are already shutting down
244 */
245 if (!afile) {
246 if (afs_shuttingdown == AFS_RUNNING)
247 osi_Panic("osi_Read called with null param");
248 else
249 return -EIO;
250 }
251
252 if (offset != -1)
253 afile->offset = offset;
254 setup_uio(&auio, &iov, aptr, afile->offset, asize, UIO_READ, AFS_UIOSYS);
255 AFS_GUNLOCK();
256 code = osi_rdwr(afile, &auio, UIO_READ);
257 AFS_GLOCK();
258 if (code == 0) {
259 code = asize - auio.uio_resid;
260 afile->offset += code;
261 } else {
262 afs_Trace2(afs_iclSetp, CM_TRACE_READFAILED, ICL_TYPE_INT32, auio.uio_resid,
263 ICL_TYPE_INT32, code);
264 if (code > 0) {
265 code = -code;
266 }
267 }
268 return code;
269 }
270
271 /* Generic write interface */
272 int
273 afs_osi_Write(struct osi_file *afile, afs_int32 offset, void *aptr,
274 afs_int32 asize)
275 {
276 struct uio auio;
277 struct iovec iov;
278 afs_int32 code;
279
280 memset(&auio, 0, sizeof(auio));
281 memset(&iov, 0, sizeof(iov));
282
283 AFS_STATCNT(osi_Write);
284
285 if (!afile) {
286 if (afs_shuttingdown == AFS_RUNNING)
287 osi_Panic("afs_osi_Write called with null param");
288 else
289 return -EIO;
290 }
291
292 if (offset != -1)
293 afile->offset = offset;
294 setup_uio(&auio, &iov, aptr, afile->offset, asize, UIO_WRITE, AFS_UIOSYS);
295 AFS_GUNLOCK();
296 code = osi_rdwr(afile, &auio, UIO_WRITE);
297 AFS_GLOCK();
298 if (code == 0) {
299 code = asize - auio.uio_resid;
300 afile->offset += code;
301 } else {
302 if (code == ENOSPC)
303 afs_WarnENOSPC();
304 if (code > 0) {
305 code = -code;
306 }
307 }
308
309 if (afile->proc)
310 (*afile->proc)(afile, code);
311
312 return code;
313 }
314
315
316 /* This work should be handled by physstrat in ca/machdep.c.
317 This routine written from the RT NFS port strategy routine.
318 It has been generalized a bit, but should still be pretty clear. */
319 int
320 afs_osi_MapStrategy(int (*aproc) (struct buf * bp), struct buf *bp)
321 {
322 afs_int32 returnCode;
323
324 AFS_STATCNT(osi_MapStrategy);
325 returnCode = (*aproc) (bp);
326
327 return returnCode;
328 }
329
330 void
331 shutdown_osifile(void)
332 {
333 AFS_STATCNT(shutdown_osifile);
334 if (afs_cold_shutdown) {
335 afs_osicred_initialized = 0;
336 }
337 }
338
339 /* Intialize cache device info and fragment size for disk cache partition. */
340 int
341 osi_InitCacheInfo(char *aname)
342 {
343 int code;
344 extern afs_dcache_id_t cacheInode;
345 struct dentry *dp;
346 extern struct osi_dev cacheDev;
347 extern afs_int32 afs_fsfragsize;
348 extern struct super_block *afs_cacheSBp;
349 extern struct vfsmount *afs_cacheMnt;
350 code = osi_lookupname_internal(aname, 1, &afs_cacheMnt, &dp);
351 if (code)
352 return ENOENT;
353
354 osi_get_fh(dp, &cacheInode.ufs);
355 cacheDev.dev = dp->d_inode->i_sb->s_dev;
356 afs_fsfragsize = dp->d_inode->i_sb->s_blocksize - 1;
357 afs_cacheSBp = dp->d_inode->i_sb;
358
359 dput(dp);
360
361 afs_init_sb_export_ops(afs_cacheSBp);
362
363 return 0;
364 }
365
366
367 /* osi_rdwr
368 * seek, then read or write to an open inode. addrp points to data in
369 * kernel space.
370 */
371 int
372 osi_rdwr(struct osi_file *osifile, struct uio *uiop, int rw)
373 {
374 struct file *filp = osifile->filp;
375 #ifdef AFS_FILE_NEEDS_SET_FS
376 mm_segment_t old_fs = {0};
377 #endif /* AFS_FILE_NEEDS_SET_FS */
378 int code = 0;
379 struct iovec *iov;
380 size_t count;
381 unsigned long savelim;
382 loff_t pos;
383
384 savelim = current->TASK_STRUCT_RLIM[RLIMIT_FSIZE].rlim_cur;
385 current->TASK_STRUCT_RLIM[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
386
387 #ifdef AFS_FILE_NEEDS_SET_FS
388 if (uiop->uio_seg == AFS_UIOSYS) {
389 /* Switch into user space */
390 old_fs = get_fs();
391 set_fs(get_ds());
392 }
393 #endif /* AFS_FILE_NEEDS_SET_FS */
394
395 while (code == 0 && uiop->uio_resid > 0 && uiop->uio_iovcnt > 0) {
396 iov = uiop->uio_iov;
397 count = iov->iov_len;
398 if (count == 0) {
399 uiop->uio_iov++;
400 uiop->uio_iovcnt--;
401 continue;
402 }
403
404 pos = uiop->uio_offset;
405 if (rw == UIO_READ)
406 code = afs_file_read(filp, iov->iov_base, count, &pos);
407 else
408 code = afs_file_write(filp, iov->iov_base, count, &pos);
409
410 if (code < 0) {
411 code = -code;
412 break;
413 } else if (code == 0) {
414 /*
415 * This is bad -- we can't read any more data from the
416 * file, but we have no good way of signaling a partial
417 * read either.
418 */
419 code = EIO;
420 break;
421 }
422
423 iov->iov_base += code;
424 iov->iov_len -= code;
425 uiop->uio_resid -= code;
426 uiop->uio_offset += code;
427 code = 0;
428 }
429
430 #ifdef AFS_FILE_NEEDS_SET_FS
431 if (uiop->uio_seg == AFS_UIOSYS) {
432 /* Switch back into kernel space */
433 set_fs(old_fs);
434 }
435 #endif /* AFS_FILE_NEEDS_SET_FS */
436
437 current->TASK_STRUCT_RLIM[RLIMIT_FSIZE].rlim_cur = savelim;
438
439 return code;
440 }
441
442 /* setup_uio
443 * Setup a uio struct.
444 */
445 void
446 setup_uio(struct uio *uiop, struct iovec *iovecp, const char *buf, afs_offs_t pos,
447 int count, uio_flag_t flag, uio_seg_t seg)
448 {
449 iovecp->iov_base = (char *)buf;
450 iovecp->iov_len = count;
451 uiop->uio_iov = iovecp;
452 uiop->uio_iovcnt = 1;
453 uiop->uio_offset = pos;
454 uiop->uio_seg = seg;
455 uiop->uio_resid = count;
456 uiop->uio_flag = flag;
457 }
458
459
460 /* uiomove
461 * UIO_READ : dp -> uio
462 * UIO_WRITE : uio -> dp
463 */
464 int
465 uiomove(char *dp, int length, uio_flag_t rw, struct uio *uiop)
466 {
467 int count;
468 struct iovec *iov;
469 int code;
470
471 while (length > 0 && uiop->uio_resid > 0 && uiop->uio_iovcnt > 0) {
472 iov = uiop->uio_iov;
473 count = iov->iov_len;
474
475 if (!count) {
476 uiop->uio_iov++;
477 uiop->uio_iovcnt--;
478 continue;
479 }
480
481 if (count > length)
482 count = length;
483
484 switch (uiop->uio_seg) {
485 case AFS_UIOSYS:
486 switch (rw) {
487 case UIO_READ:
488 memcpy(iov->iov_base, dp, count);
489 break;
490 case UIO_WRITE:
491 memcpy(dp, iov->iov_base, count);
492 break;
493 default:
494 printf("uiomove: Bad rw = %d\n", rw);
495 return -EINVAL;
496 }
497 break;
498 case AFS_UIOUSER:
499 switch (rw) {
500 case UIO_READ:
501 AFS_COPYOUT(dp, iov->iov_base, count, code);
502 break;
503 case UIO_WRITE:
504 AFS_COPYIN(iov->iov_base, dp, count, code);
505 break;
506 default:
507 printf("uiomove: Bad rw = %d\n", rw);
508 return -EINVAL;
509 }
510 break;
511 default:
512 printf("uiomove: Bad seg = %d\n", uiop->uio_seg);
513 return -EINVAL;
514 }
515
516 dp += count;
517 length -= count;
518 iov->iov_base += count;
519 iov->iov_len -= count;
520 uiop->uio_offset += count;
521 uiop->uio_resid -= count;
522 }
523 return 0;
524 }
525