Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3 | * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved. | |
4 | * | |
5 | * This copyrighted material is made available to anyone wishing to use, | |
6 | * modify, copy, or redistribute it subject to the terms and conditions | |
7 | * of the GNU General Public License v.2. | |
8 | */ | |
9 | ||
10 | #include <linux/sched.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/spinlock.h> | |
13 | #include <linux/completion.h> | |
14 | #include <linux/buffer_head.h> | |
15 | #include <linux/pagemap.h> | |
16 | #include <linux/uio.h> | |
17 | #include <linux/blkdev.h> | |
18 | #include <linux/mm.h> | |
19 | #include <linux/smp_lock.h> | |
18ec7d5c | 20 | #include <linux/fs.h> |
5c676f6d | 21 | #include <linux/gfs2_ondisk.h> |
b3b94faa DT |
22 | #include <asm/semaphore.h> |
23 | #include <asm/uaccess.h> | |
24 | ||
25 | #include "gfs2.h" | |
5c676f6d SW |
26 | #include "lm_interface.h" |
27 | #include "incore.h" | |
b3b94faa DT |
28 | #include "bmap.h" |
29 | #include "dir.h" | |
30 | #include "glock.h" | |
31 | #include "glops.h" | |
32 | #include "inode.h" | |
b3b94faa DT |
33 | #include "lm.h" |
34 | #include "log.h" | |
35 | #include "meta_io.h" | |
36 | #include "ops_file.h" | |
37 | #include "ops_vm.h" | |
38 | #include "quota.h" | |
39 | #include "rgrp.h" | |
40 | #include "trans.h" | |
5c676f6d | 41 | #include "util.h" |
b3b94faa DT |
42 | |
43 | /* "bad" is for NFS support */ | |
44 | struct filldir_bad_entry { | |
45 | char *fbe_name; | |
46 | unsigned int fbe_length; | |
47 | uint64_t fbe_offset; | |
48 | struct gfs2_inum fbe_inum; | |
49 | unsigned int fbe_type; | |
50 | }; | |
51 | ||
52 | struct filldir_bad { | |
53 | struct gfs2_sbd *fdb_sbd; | |
54 | ||
55 | struct filldir_bad_entry *fdb_entry; | |
56 | unsigned int fdb_entry_num; | |
57 | unsigned int fdb_entry_off; | |
58 | ||
59 | char *fdb_name; | |
60 | unsigned int fdb_name_size; | |
61 | unsigned int fdb_name_off; | |
62 | }; | |
63 | ||
64 | /* For regular, non-NFS */ | |
65 | struct filldir_reg { | |
66 | struct gfs2_sbd *fdr_sbd; | |
67 | int fdr_prefetch; | |
68 | ||
69 | filldir_t fdr_filldir; | |
70 | void *fdr_opaque; | |
71 | }; | |
72 | ||
61a30dcb SW |
73 | /* |
74 | * Most fields left uninitialised to catch anybody who tries to | |
75 | * use them. f_flags set to prevent file_accessed() from touching | |
76 | * any other part of this. Its use is purely as a flag so that we | |
77 | * know (in readpage()) whether or not do to locking. | |
78 | */ | |
79 | struct file gfs2_internal_file_sentinal = { | |
80 | .f_flags = O_NOATIME|O_RDONLY, | |
81 | }; | |
82 | ||
18ec7d5c SW |
83 | static int gfs2_read_actor(read_descriptor_t *desc, struct page *page, |
84 | unsigned long offset, unsigned long size) | |
85 | { | |
86 | char *kaddr; | |
87 | unsigned long count = desc->count; | |
88 | ||
89 | if (size > count) | |
90 | size = count; | |
91 | ||
92 | kaddr = kmap(page); | |
93 | memcpy(desc->arg.buf, kaddr + offset, size); | |
94 | kunmap(page); | |
95 | ||
96 | desc->count = count - size; | |
97 | desc->written += size; | |
98 | desc->arg.buf += size; | |
99 | return size; | |
100 | } | |
101 | ||
102 | int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state, | |
103 | char *buf, loff_t *pos, unsigned size) | |
104 | { | |
105 | struct inode *inode = ip->i_vnode; | |
106 | read_descriptor_t desc; | |
107 | desc.written = 0; | |
108 | desc.arg.buf = buf; | |
109 | desc.count = size; | |
110 | desc.error = 0; | |
61a30dcb SW |
111 | do_generic_mapping_read(inode->i_mapping, ra_state, |
112 | &gfs2_internal_file_sentinal, pos, &desc, | |
113 | gfs2_read_actor); | |
18ec7d5c SW |
114 | return desc.written ? desc.written : desc.error; |
115 | } | |
b3b94faa DT |
116 | |
117 | /** | |
118 | * gfs2_llseek - seek to a location in a file | |
119 | * @file: the file | |
120 | * @offset: the offset | |
121 | * @origin: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END) | |
122 | * | |
123 | * SEEK_END requires the glock for the file because it references the | |
124 | * file's size. | |
125 | * | |
126 | * Returns: The new offset, or errno | |
127 | */ | |
128 | ||
129 | static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin) | |
130 | { | |
5c676f6d | 131 | struct gfs2_inode *ip = file->f_mapping->host->u.generic_ip; |
b3b94faa DT |
132 | struct gfs2_holder i_gh; |
133 | loff_t error; | |
134 | ||
b3b94faa DT |
135 | if (origin == 2) { |
136 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, | |
137 | &i_gh); | |
138 | if (!error) { | |
139 | error = remote_llseek(file, offset, origin); | |
140 | gfs2_glock_dq_uninit(&i_gh); | |
141 | } | |
142 | } else | |
143 | error = remote_llseek(file, offset, origin); | |
144 | ||
145 | return error; | |
146 | } | |
147 | ||
b3b94faa | 148 | |
18ec7d5c SW |
149 | static ssize_t gfs2_direct_IO_read(struct kiocb *iocb, const struct iovec *iov, |
150 | loff_t offset, unsigned long nr_segs) | |
b3b94faa | 151 | { |
18ec7d5c SW |
152 | struct file *file = iocb->ki_filp; |
153 | struct address_space *mapping = file->f_mapping; | |
154 | ssize_t retval; | |
b3b94faa | 155 | |
18ec7d5c SW |
156 | retval = filemap_write_and_wait(mapping); |
157 | if (retval == 0) { | |
158 | retval = mapping->a_ops->direct_IO(READ, iocb, iov, offset, | |
159 | nr_segs); | |
b3b94faa | 160 | } |
18ec7d5c | 161 | return retval; |
b3b94faa DT |
162 | } |
163 | ||
164 | /** | |
18ec7d5c SW |
165 | * __gfs2_file_aio_read - The main GFS2 read function |
166 | * | |
167 | * N.B. This is almost, but not quite the same as __generic_file_aio_read() | |
168 | * the important subtle different being that inode->i_size isn't valid | |
169 | * unless we are holding a lock, and we do this _only_ on the O_DIRECT | |
170 | * path since otherwise locking is done entirely at the page cache | |
171 | * layer. | |
b3b94faa | 172 | */ |
18ec7d5c SW |
173 | static ssize_t __gfs2_file_aio_read(struct kiocb *iocb, |
174 | const struct iovec *iov, | |
175 | unsigned long nr_segs, loff_t *ppos) | |
b3b94faa | 176 | { |
18ec7d5c | 177 | struct file *filp = iocb->ki_filp; |
5c676f6d | 178 | struct gfs2_inode *ip = filp->f_mapping->host->u.generic_ip; |
b3b94faa | 179 | struct gfs2_holder gh; |
18ec7d5c SW |
180 | ssize_t retval; |
181 | unsigned long seg; | |
182 | size_t count; | |
183 | ||
184 | count = 0; | |
185 | for (seg = 0; seg < nr_segs; seg++) { | |
186 | const struct iovec *iv = &iov[seg]; | |
187 | ||
188 | /* | |
189 | * If any segment has a negative length, or the cumulative | |
190 | * length ever wraps negative then return -EINVAL. | |
191 | */ | |
d1665e41 SW |
192 | count += iv->iov_len; |
193 | if (unlikely((ssize_t)(count|iv->iov_len) < 0)) | |
194 | return -EINVAL; | |
195 | if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len)) | |
196 | continue; | |
197 | if (seg == 0) | |
198 | return -EFAULT; | |
199 | nr_segs = seg; | |
200 | count -= iv->iov_len; /* This segment is no good */ | |
201 | break; | |
18ec7d5c SW |
202 | } |
203 | ||
204 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
205 | if (filp->f_flags & O_DIRECT) { | |
206 | loff_t pos = *ppos, size; | |
207 | struct address_space *mapping; | |
208 | struct inode *inode; | |
209 | ||
210 | mapping = filp->f_mapping; | |
211 | inode = mapping->host; | |
212 | retval = 0; | |
213 | if (!count) | |
214 | goto out; /* skip atime */ | |
215 | ||
216 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh); | |
217 | retval = gfs2_glock_nq_m_atime(1, &gh); | |
218 | if (retval) | |
219 | goto out; | |
d1665e41 SW |
220 | if (gfs2_is_stuffed(ip)) { |
221 | gfs2_glock_dq_m(1, &gh); | |
222 | gfs2_holder_uninit(&gh); | |
223 | goto fallback_to_normal; | |
224 | } | |
18ec7d5c SW |
225 | size = i_size_read(inode); |
226 | if (pos < size) { | |
d1665e41 | 227 | retval = gfs2_direct_IO_read(iocb, iov, pos, nr_segs); |
18ec7d5c SW |
228 | if (retval > 0 && !is_sync_kiocb(iocb)) |
229 | retval = -EIOCBQUEUED; | |
230 | if (retval > 0) | |
231 | *ppos = pos + retval; | |
b3b94faa | 232 | } |
18ec7d5c SW |
233 | file_accessed(filp); |
234 | gfs2_glock_dq_m(1, &gh); | |
235 | gfs2_holder_uninit(&gh); | |
b3b94faa | 236 | goto out; |
18ec7d5c | 237 | } |
b3b94faa | 238 | |
d1665e41 | 239 | fallback_to_normal: |
18ec7d5c SW |
240 | retval = 0; |
241 | if (count) { | |
242 | for (seg = 0; seg < nr_segs; seg++) { | |
243 | read_descriptor_t desc; | |
244 | ||
245 | desc.written = 0; | |
246 | desc.arg.buf = iov[seg].iov_base; | |
247 | desc.count = iov[seg].iov_len; | |
248 | if (desc.count == 0) | |
249 | continue; | |
250 | desc.error = 0; | |
251 | do_generic_file_read(filp,ppos,&desc,file_read_actor); | |
252 | retval += desc.written; | |
253 | if (desc.error) { | |
254 | retval = retval ?: desc.error; | |
255 | break; | |
256 | } | |
257 | } | |
258 | } | |
259 | out: | |
260 | return retval; | |
b3b94faa DT |
261 | } |
262 | ||
263 | /** | |
264 | * gfs2_read - Read bytes from a file | |
265 | * @file: The file to read from | |
266 | * @buf: The buffer to copy into | |
267 | * @size: The amount of data requested | |
268 | * @offset: The current file offset | |
269 | * | |
270 | * Outputs: Offset - updated according to number of bytes read | |
271 | * | |
272 | * Returns: The number of bytes read, errno on failure | |
273 | */ | |
274 | ||
18ec7d5c | 275 | static ssize_t gfs2_read(struct file *filp, char __user *buf, size_t size, |
b3b94faa DT |
276 | loff_t *offset) |
277 | { | |
b3b94faa | 278 | struct iovec local_iov = { .iov_base = buf, .iov_len = size }; |
18ec7d5c SW |
279 | struct kiocb kiocb; |
280 | ssize_t ret; | |
b3b94faa | 281 | |
18ec7d5c SW |
282 | init_sync_kiocb(&kiocb, filp); |
283 | ret = __gfs2_file_aio_read(&kiocb, &local_iov, 1, offset); | |
284 | if (-EIOCBQUEUED == ret) | |
285 | ret = wait_on_sync_kiocb(&kiocb); | |
286 | return ret; | |
b3b94faa DT |
287 | } |
288 | ||
18ec7d5c SW |
289 | static ssize_t gfs2_file_readv(struct file *filp, const struct iovec *iov, |
290 | unsigned long nr_segs, loff_t *ppos) | |
b3b94faa | 291 | { |
18ec7d5c SW |
292 | struct kiocb kiocb; |
293 | ssize_t ret; | |
b3b94faa | 294 | |
18ec7d5c SW |
295 | init_sync_kiocb(&kiocb, filp); |
296 | ret = __gfs2_file_aio_read(&kiocb, iov, nr_segs, ppos); | |
297 | if (-EIOCBQUEUED == ret) | |
298 | ret = wait_on_sync_kiocb(&kiocb); | |
299 | return ret; | |
b3b94faa DT |
300 | } |
301 | ||
18ec7d5c SW |
302 | static ssize_t gfs2_file_aio_read(struct kiocb *iocb, char __user *buf, |
303 | size_t count, loff_t pos) | |
b3b94faa | 304 | { |
18ec7d5c | 305 | struct iovec local_iov = { .iov_base = buf, .iov_len = count }; |
b3b94faa | 306 | |
18ec7d5c SW |
307 | BUG_ON(iocb->ki_pos != pos); |
308 | return __gfs2_file_aio_read(iocb, &local_iov, 1, &iocb->ki_pos); | |
b3b94faa DT |
309 | } |
310 | ||
b3b94faa DT |
311 | |
312 | /** | |
313 | * filldir_reg_func - Report a directory entry to the caller of gfs2_dir_read() | |
314 | * @opaque: opaque data used by the function | |
315 | * @name: the name of the directory entry | |
316 | * @length: the length of the name | |
317 | * @offset: the entry's offset in the directory | |
318 | * @inum: the inode number the entry points to | |
319 | * @type: the type of inode the entry points to | |
320 | * | |
321 | * Returns: 0 on success, 1 if buffer full | |
322 | */ | |
323 | ||
324 | static int filldir_reg_func(void *opaque, const char *name, unsigned int length, | |
325 | uint64_t offset, struct gfs2_inum *inum, | |
326 | unsigned int type) | |
327 | { | |
328 | struct filldir_reg *fdr = (struct filldir_reg *)opaque; | |
329 | struct gfs2_sbd *sdp = fdr->fdr_sbd; | |
330 | int error; | |
331 | ||
332 | error = fdr->fdr_filldir(fdr->fdr_opaque, name, length, offset, | |
333 | inum->no_formal_ino, type); | |
334 | if (error) | |
335 | return 1; | |
336 | ||
337 | if (fdr->fdr_prefetch && !(length == 1 && *name == '.')) { | |
338 | gfs2_glock_prefetch_num(sdp, | |
339 | inum->no_addr, &gfs2_inode_glops, | |
340 | LM_ST_SHARED, LM_FLAG_TRY | LM_FLAG_ANY); | |
341 | gfs2_glock_prefetch_num(sdp, | |
342 | inum->no_addr, &gfs2_iopen_glops, | |
343 | LM_ST_SHARED, LM_FLAG_TRY); | |
344 | } | |
345 | ||
346 | return 0; | |
347 | } | |
348 | ||
349 | /** | |
350 | * readdir_reg - Read directory entries from a directory | |
351 | * @file: The directory to read from | |
352 | * @dirent: Buffer for dirents | |
353 | * @filldir: Function used to do the copying | |
354 | * | |
355 | * Returns: errno | |
356 | */ | |
357 | ||
358 | static int readdir_reg(struct file *file, void *dirent, filldir_t filldir) | |
359 | { | |
5c676f6d | 360 | struct gfs2_inode *dip = file->f_mapping->host->u.generic_ip; |
b3b94faa DT |
361 | struct filldir_reg fdr; |
362 | struct gfs2_holder d_gh; | |
363 | uint64_t offset = file->f_pos; | |
364 | int error; | |
365 | ||
366 | fdr.fdr_sbd = dip->i_sbd; | |
367 | fdr.fdr_prefetch = 1; | |
368 | fdr.fdr_filldir = filldir; | |
369 | fdr.fdr_opaque = dirent; | |
370 | ||
371 | gfs2_holder_init(dip->i_gl, LM_ST_SHARED, GL_ATIME, &d_gh); | |
372 | error = gfs2_glock_nq_atime(&d_gh); | |
373 | if (error) { | |
374 | gfs2_holder_uninit(&d_gh); | |
375 | return error; | |
376 | } | |
377 | ||
378 | error = gfs2_dir_read(dip, &offset, &fdr, filldir_reg_func); | |
379 | ||
380 | gfs2_glock_dq_uninit(&d_gh); | |
381 | ||
382 | file->f_pos = offset; | |
383 | ||
384 | return error; | |
385 | } | |
386 | ||
387 | /** | |
388 | * filldir_bad_func - Report a directory entry to the caller of gfs2_dir_read() | |
389 | * @opaque: opaque data used by the function | |
390 | * @name: the name of the directory entry | |
391 | * @length: the length of the name | |
392 | * @offset: the entry's offset in the directory | |
393 | * @inum: the inode number the entry points to | |
394 | * @type: the type of inode the entry points to | |
395 | * | |
396 | * For supporting NFS. | |
397 | * | |
398 | * Returns: 0 on success, 1 if buffer full | |
399 | */ | |
400 | ||
401 | static int filldir_bad_func(void *opaque, const char *name, unsigned int length, | |
402 | uint64_t offset, struct gfs2_inum *inum, | |
403 | unsigned int type) | |
404 | { | |
405 | struct filldir_bad *fdb = (struct filldir_bad *)opaque; | |
406 | struct gfs2_sbd *sdp = fdb->fdb_sbd; | |
407 | struct filldir_bad_entry *fbe; | |
408 | ||
409 | if (fdb->fdb_entry_off == fdb->fdb_entry_num || | |
410 | fdb->fdb_name_off + length > fdb->fdb_name_size) | |
411 | return 1; | |
412 | ||
413 | fbe = &fdb->fdb_entry[fdb->fdb_entry_off]; | |
414 | fbe->fbe_name = fdb->fdb_name + fdb->fdb_name_off; | |
415 | memcpy(fbe->fbe_name, name, length); | |
416 | fbe->fbe_length = length; | |
417 | fbe->fbe_offset = offset; | |
418 | fbe->fbe_inum = *inum; | |
419 | fbe->fbe_type = type; | |
420 | ||
421 | fdb->fdb_entry_off++; | |
422 | fdb->fdb_name_off += length; | |
423 | ||
424 | if (!(length == 1 && *name == '.')) { | |
425 | gfs2_glock_prefetch_num(sdp, | |
426 | inum->no_addr, &gfs2_inode_glops, | |
427 | LM_ST_SHARED, LM_FLAG_TRY | LM_FLAG_ANY); | |
428 | gfs2_glock_prefetch_num(sdp, | |
429 | inum->no_addr, &gfs2_iopen_glops, | |
430 | LM_ST_SHARED, LM_FLAG_TRY); | |
431 | } | |
432 | ||
433 | return 0; | |
434 | } | |
435 | ||
436 | /** | |
437 | * readdir_bad - Read directory entries from a directory | |
438 | * @file: The directory to read from | |
439 | * @dirent: Buffer for dirents | |
440 | * @filldir: Function used to do the copying | |
441 | * | |
442 | * For supporting NFS. | |
443 | * | |
444 | * Returns: errno | |
445 | */ | |
446 | ||
447 | static int readdir_bad(struct file *file, void *dirent, filldir_t filldir) | |
448 | { | |
5c676f6d | 449 | struct gfs2_inode *dip = file->f_mapping->host->u.generic_ip; |
b3b94faa DT |
450 | struct gfs2_sbd *sdp = dip->i_sbd; |
451 | struct filldir_reg fdr; | |
452 | unsigned int entries, size; | |
453 | struct filldir_bad *fdb; | |
454 | struct gfs2_holder d_gh; | |
455 | uint64_t offset = file->f_pos; | |
456 | unsigned int x; | |
457 | struct filldir_bad_entry *fbe; | |
458 | int error; | |
459 | ||
460 | entries = gfs2_tune_get(sdp, gt_entries_per_readdir); | |
461 | size = sizeof(struct filldir_bad) + | |
462 | entries * (sizeof(struct filldir_bad_entry) + GFS2_FAST_NAME_SIZE); | |
463 | ||
464 | fdb = kzalloc(size, GFP_KERNEL); | |
465 | if (!fdb) | |
466 | return -ENOMEM; | |
467 | ||
468 | fdb->fdb_sbd = sdp; | |
469 | fdb->fdb_entry = (struct filldir_bad_entry *)(fdb + 1); | |
470 | fdb->fdb_entry_num = entries; | |
471 | fdb->fdb_name = ((char *)fdb) + sizeof(struct filldir_bad) + | |
472 | entries * sizeof(struct filldir_bad_entry); | |
473 | fdb->fdb_name_size = entries * GFS2_FAST_NAME_SIZE; | |
474 | ||
475 | gfs2_holder_init(dip->i_gl, LM_ST_SHARED, GL_ATIME, &d_gh); | |
476 | error = gfs2_glock_nq_atime(&d_gh); | |
477 | if (error) { | |
478 | gfs2_holder_uninit(&d_gh); | |
479 | goto out; | |
480 | } | |
481 | ||
482 | error = gfs2_dir_read(dip, &offset, fdb, filldir_bad_func); | |
483 | ||
484 | gfs2_glock_dq_uninit(&d_gh); | |
485 | ||
486 | fdr.fdr_sbd = sdp; | |
487 | fdr.fdr_prefetch = 0; | |
488 | fdr.fdr_filldir = filldir; | |
489 | fdr.fdr_opaque = dirent; | |
490 | ||
491 | for (x = 0; x < fdb->fdb_entry_off; x++) { | |
492 | fbe = &fdb->fdb_entry[x]; | |
493 | ||
494 | error = filldir_reg_func(&fdr, | |
495 | fbe->fbe_name, fbe->fbe_length, | |
496 | fbe->fbe_offset, | |
497 | &fbe->fbe_inum, fbe->fbe_type); | |
498 | if (error) { | |
499 | file->f_pos = fbe->fbe_offset; | |
500 | error = 0; | |
501 | goto out; | |
502 | } | |
503 | } | |
504 | ||
505 | file->f_pos = offset; | |
506 | ||
507 | out: | |
508 | kfree(fdb); | |
509 | ||
510 | return error; | |
511 | } | |
512 | ||
513 | /** | |
514 | * gfs2_readdir - Read directory entries from a directory | |
515 | * @file: The directory to read from | |
516 | * @dirent: Buffer for dirents | |
517 | * @filldir: Function used to do the copying | |
518 | * | |
519 | * Returns: errno | |
520 | */ | |
521 | ||
522 | static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir) | |
523 | { | |
524 | int error; | |
525 | ||
b3b94faa DT |
526 | if (strcmp(current->comm, "nfsd") != 0) |
527 | error = readdir_reg(file, dirent, filldir); | |
528 | else | |
529 | error = readdir_bad(file, dirent, filldir); | |
530 | ||
531 | return error; | |
532 | } | |
533 | ||
b3b94faa DT |
534 | /** |
535 | * gfs2_mmap - | |
536 | * @file: The file to map | |
537 | * @vma: The VMA which described the mapping | |
538 | * | |
539 | * Returns: 0 or error code | |
540 | */ | |
541 | ||
542 | static int gfs2_mmap(struct file *file, struct vm_area_struct *vma) | |
543 | { | |
5c676f6d | 544 | struct gfs2_inode *ip = file->f_mapping->host->u.generic_ip; |
b3b94faa DT |
545 | struct gfs2_holder i_gh; |
546 | int error; | |
547 | ||
b3b94faa DT |
548 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh); |
549 | error = gfs2_glock_nq_atime(&i_gh); | |
550 | if (error) { | |
551 | gfs2_holder_uninit(&i_gh); | |
552 | return error; | |
553 | } | |
554 | ||
18ec7d5c SW |
555 | /* This is VM_MAYWRITE instead of VM_WRITE because a call |
556 | to mprotect() can turn on VM_WRITE later. */ | |
557 | ||
558 | if ((vma->vm_flags & (VM_MAYSHARE | VM_MAYWRITE)) == | |
559 | (VM_MAYSHARE | VM_MAYWRITE)) | |
560 | vma->vm_ops = &gfs2_vm_ops_sharewrite; | |
561 | else | |
562 | vma->vm_ops = &gfs2_vm_ops_private; | |
b3b94faa DT |
563 | |
564 | gfs2_glock_dq_uninit(&i_gh); | |
565 | ||
566 | return error; | |
567 | } | |
568 | ||
569 | /** | |
570 | * gfs2_open - open a file | |
571 | * @inode: the inode to open | |
572 | * @file: the struct file for this opening | |
573 | * | |
574 | * Returns: errno | |
575 | */ | |
576 | ||
577 | static int gfs2_open(struct inode *inode, struct file *file) | |
578 | { | |
5c676f6d | 579 | struct gfs2_inode *ip = inode->u.generic_ip; |
b3b94faa DT |
580 | struct gfs2_holder i_gh; |
581 | struct gfs2_file *fp; | |
582 | int error; | |
583 | ||
b3b94faa DT |
584 | fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL); |
585 | if (!fp) | |
586 | return -ENOMEM; | |
587 | ||
f55ab26a | 588 | mutex_init(&fp->f_fl_mutex); |
b3b94faa DT |
589 | |
590 | fp->f_inode = ip; | |
591 | fp->f_vfile = file; | |
592 | ||
5c676f6d SW |
593 | gfs2_assert_warn(ip->i_sbd, !file->private_data); |
594 | file->private_data = fp; | |
b3b94faa DT |
595 | |
596 | if (S_ISREG(ip->i_di.di_mode)) { | |
597 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, | |
598 | &i_gh); | |
599 | if (error) | |
600 | goto fail; | |
601 | ||
602 | if (!(file->f_flags & O_LARGEFILE) && | |
603 | ip->i_di.di_size > MAX_NON_LFS) { | |
604 | error = -EFBIG; | |
605 | goto fail_gunlock; | |
606 | } | |
607 | ||
608 | /* Listen to the Direct I/O flag */ | |
609 | ||
610 | if (ip->i_di.di_flags & GFS2_DIF_DIRECTIO) | |
611 | file->f_flags |= O_DIRECT; | |
612 | ||
b3b94faa DT |
613 | gfs2_glock_dq_uninit(&i_gh); |
614 | } | |
615 | ||
616 | return 0; | |
617 | ||
618 | fail_gunlock: | |
619 | gfs2_glock_dq_uninit(&i_gh); | |
620 | ||
621 | fail: | |
5c676f6d | 622 | file->private_data = NULL; |
b3b94faa DT |
623 | kfree(fp); |
624 | ||
625 | return error; | |
626 | } | |
627 | ||
628 | /** | |
629 | * gfs2_close - called to close a struct file | |
630 | * @inode: the inode the struct file belongs to | |
631 | * @file: the struct file being closed | |
632 | * | |
633 | * Returns: errno | |
634 | */ | |
635 | ||
636 | static int gfs2_close(struct inode *inode, struct file *file) | |
637 | { | |
5c676f6d | 638 | struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; |
b3b94faa DT |
639 | struct gfs2_file *fp; |
640 | ||
5c676f6d SW |
641 | fp = file->private_data; |
642 | file->private_data = NULL; | |
b3b94faa DT |
643 | |
644 | if (gfs2_assert_warn(sdp, fp)) | |
645 | return -EIO; | |
646 | ||
647 | kfree(fp); | |
648 | ||
649 | return 0; | |
650 | } | |
651 | ||
652 | /** | |
653 | * gfs2_fsync - sync the dirty data for a file (across the cluster) | |
654 | * @file: the file that points to the dentry (we ignore this) | |
655 | * @dentry: the dentry that points to the inode to sync | |
656 | * | |
657 | * Returns: errno | |
658 | */ | |
659 | ||
660 | static int gfs2_fsync(struct file *file, struct dentry *dentry, int datasync) | |
661 | { | |
5c676f6d | 662 | struct gfs2_inode *ip = dentry->d_inode->u.generic_ip; |
b3b94faa | 663 | |
b3b94faa DT |
664 | gfs2_log_flush_glock(ip->i_gl); |
665 | ||
666 | return 0; | |
667 | } | |
668 | ||
669 | /** | |
670 | * gfs2_lock - acquire/release a posix lock on a file | |
671 | * @file: the file pointer | |
672 | * @cmd: either modify or retrieve lock state, possibly wait | |
673 | * @fl: type and range of lock | |
674 | * | |
675 | * Returns: errno | |
676 | */ | |
677 | ||
678 | static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) | |
679 | { | |
5c676f6d | 680 | struct gfs2_inode *ip = file->f_mapping->host->u.generic_ip; |
b3b94faa DT |
681 | struct gfs2_sbd *sdp = ip->i_sbd; |
682 | struct lm_lockname name = | |
683 | { .ln_number = ip->i_num.no_addr, | |
684 | .ln_type = LM_TYPE_PLOCK }; | |
685 | ||
b3b94faa DT |
686 | if (!(fl->fl_flags & FL_POSIX)) |
687 | return -ENOLCK; | |
688 | if ((ip->i_di.di_mode & (S_ISGID | S_IXGRP)) == S_ISGID) | |
689 | return -ENOLCK; | |
690 | ||
691 | if (sdp->sd_args.ar_localflocks) { | |
692 | if (IS_GETLK(cmd)) { | |
693 | struct file_lock *tmp; | |
694 | lock_kernel(); | |
695 | tmp = posix_test_lock(file, fl); | |
696 | fl->fl_type = F_UNLCK; | |
697 | if (tmp) | |
698 | memcpy(fl, tmp, sizeof(struct file_lock)); | |
699 | unlock_kernel(); | |
700 | return 0; | |
701 | } else { | |
702 | int error; | |
703 | lock_kernel(); | |
704 | error = posix_lock_file_wait(file, fl); | |
705 | unlock_kernel(); | |
706 | return error; | |
707 | } | |
708 | } | |
709 | ||
710 | if (IS_GETLK(cmd)) | |
711 | return gfs2_lm_plock_get(sdp, &name, file, fl); | |
712 | else if (fl->fl_type == F_UNLCK) | |
713 | return gfs2_lm_punlock(sdp, &name, file, fl); | |
714 | else | |
715 | return gfs2_lm_plock(sdp, &name, file, cmd, fl); | |
716 | } | |
717 | ||
718 | /** | |
719 | * gfs2_sendfile - Send bytes to a file or socket | |
720 | * @in_file: The file to read from | |
721 | * @out_file: The file to write to | |
722 | * @count: The amount of data | |
723 | * @offset: The beginning file offset | |
724 | * | |
725 | * Outputs: offset - updated according to number of bytes read | |
726 | * | |
727 | * Returns: The number of bytes sent, errno on failure | |
728 | */ | |
729 | ||
730 | static ssize_t gfs2_sendfile(struct file *in_file, loff_t *offset, size_t count, | |
731 | read_actor_t actor, void *target) | |
732 | { | |
18ec7d5c | 733 | return generic_file_sendfile(in_file, offset, count, actor, target); |
b3b94faa DT |
734 | } |
735 | ||
736 | static int do_flock(struct file *file, int cmd, struct file_lock *fl) | |
737 | { | |
5c676f6d | 738 | struct gfs2_file *fp = file->private_data; |
b3b94faa DT |
739 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; |
740 | struct gfs2_inode *ip = fp->f_inode; | |
741 | struct gfs2_glock *gl; | |
742 | unsigned int state; | |
743 | int flags; | |
744 | int error = 0; | |
745 | ||
746 | state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; | |
747 | flags = ((IS_SETLKW(cmd)) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE; | |
748 | ||
f55ab26a | 749 | mutex_lock(&fp->f_fl_mutex); |
b3b94faa DT |
750 | |
751 | gl = fl_gh->gh_gl; | |
752 | if (gl) { | |
753 | if (fl_gh->gh_state == state) | |
754 | goto out; | |
755 | gfs2_glock_hold(gl); | |
756 | flock_lock_file_wait(file, | |
757 | &(struct file_lock){.fl_type = F_UNLCK}); | |
758 | gfs2_glock_dq_uninit(fl_gh); | |
759 | } else { | |
760 | error = gfs2_glock_get(ip->i_sbd, | |
761 | ip->i_num.no_addr, &gfs2_flock_glops, | |
762 | CREATE, &gl); | |
763 | if (error) | |
764 | goto out; | |
765 | } | |
766 | ||
767 | gfs2_holder_init(gl, state, flags, fl_gh); | |
768 | gfs2_glock_put(gl); | |
769 | ||
770 | error = gfs2_glock_nq(fl_gh); | |
771 | if (error) { | |
772 | gfs2_holder_uninit(fl_gh); | |
773 | if (error == GLR_TRYFAILED) | |
774 | error = -EAGAIN; | |
775 | } else { | |
776 | error = flock_lock_file_wait(file, fl); | |
777 | gfs2_assert_warn(ip->i_sbd, !error); | |
778 | } | |
779 | ||
780 | out: | |
f55ab26a | 781 | mutex_unlock(&fp->f_fl_mutex); |
b3b94faa DT |
782 | |
783 | return error; | |
784 | } | |
785 | ||
786 | static void do_unflock(struct file *file, struct file_lock *fl) | |
787 | { | |
5c676f6d | 788 | struct gfs2_file *fp = file->private_data; |
b3b94faa DT |
789 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; |
790 | ||
f55ab26a | 791 | mutex_lock(&fp->f_fl_mutex); |
b3b94faa DT |
792 | flock_lock_file_wait(file, fl); |
793 | if (fl_gh->gh_gl) | |
794 | gfs2_glock_dq_uninit(fl_gh); | |
f55ab26a | 795 | mutex_unlock(&fp->f_fl_mutex); |
b3b94faa DT |
796 | } |
797 | ||
798 | /** | |
799 | * gfs2_flock - acquire/release a flock lock on a file | |
800 | * @file: the file pointer | |
801 | * @cmd: either modify or retrieve lock state, possibly wait | |
802 | * @fl: type and range of lock | |
803 | * | |
804 | * Returns: errno | |
805 | */ | |
806 | ||
807 | static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl) | |
808 | { | |
5c676f6d | 809 | struct gfs2_inode *ip = file->f_mapping->host->u.generic_ip; |
b3b94faa DT |
810 | struct gfs2_sbd *sdp = ip->i_sbd; |
811 | ||
b3b94faa DT |
812 | if (!(fl->fl_flags & FL_FLOCK)) |
813 | return -ENOLCK; | |
814 | if ((ip->i_di.di_mode & (S_ISGID | S_IXGRP)) == S_ISGID) | |
815 | return -ENOLCK; | |
816 | ||
817 | if (sdp->sd_args.ar_localflocks) | |
818 | return flock_lock_file_wait(file, fl); | |
819 | ||
820 | if (fl->fl_type == F_UNLCK) { | |
821 | do_unflock(file, fl); | |
822 | return 0; | |
823 | } else | |
824 | return do_flock(file, cmd, fl); | |
825 | } | |
826 | ||
827 | struct file_operations gfs2_file_fops = { | |
828 | .llseek = gfs2_llseek, | |
829 | .read = gfs2_read, | |
18ec7d5c SW |
830 | .readv = gfs2_file_readv, |
831 | .aio_read = gfs2_file_aio_read, | |
832 | .write = generic_file_write, | |
833 | .writev = generic_file_writev, | |
834 | .aio_write = generic_file_aio_write, | |
b3b94faa DT |
835 | .mmap = gfs2_mmap, |
836 | .open = gfs2_open, | |
837 | .release = gfs2_close, | |
838 | .fsync = gfs2_fsync, | |
839 | .lock = gfs2_lock, | |
840 | .sendfile = gfs2_sendfile, | |
841 | .flock = gfs2_flock, | |
842 | }; | |
843 | ||
844 | struct file_operations gfs2_dir_fops = { | |
845 | .readdir = gfs2_readdir, | |
b3b94faa DT |
846 | .open = gfs2_open, |
847 | .release = gfs2_close, | |
848 | .fsync = gfs2_fsync, | |
849 | .lock = gfs2_lock, | |
850 | .flock = gfs2_flock, | |
851 | }; | |
852 |