]>
Commit | Line | Data |
---|---|---|
3d14c5d2 | 1 | #include <linux/ceph/ceph_debug.h> |
124e68e7 | 2 | |
3d14c5d2 | 3 | #include <linux/module.h> |
124e68e7 | 4 | #include <linux/sched.h> |
5a0e3ad6 | 5 | #include <linux/slab.h> |
124e68e7 | 6 | #include <linux/file.h> |
5ef50c3b | 7 | #include <linux/mount.h> |
124e68e7 SW |
8 | #include <linux/namei.h> |
9 | #include <linux/writeback.h> | |
a27bb332 | 10 | #include <linux/aio.h> |
ad7a60de | 11 | #include <linux/falloc.h> |
124e68e7 SW |
12 | |
13 | #include "super.h" | |
14 | #include "mds_client.h" | |
15 | ||
16 | /* | |
17 | * Ceph file operations | |
18 | * | |
19 | * Implement basic open/close functionality, and implement | |
20 | * read/write. | |
21 | * | |
22 | * We implement three modes of file I/O: | |
23 | * - buffered uses the generic_file_aio_{read,write} helpers | |
24 | * | |
25 | * - synchronous is used when there is multi-client read/write | |
26 | * sharing, avoids the page cache, and synchronously waits for an | |
27 | * ack from the OSD. | |
28 | * | |
29 | * - direct io takes the variant of the sync path that references | |
30 | * user pages directly. | |
31 | * | |
32 | * fsync() flushes and waits on dirty pages, but just queues metadata | |
33 | * for writeback: since the MDS can recover size and mtime there is no | |
34 | * need to wait for MDS acknowledgement. | |
35 | */ | |
36 | ||
37 | ||
38 | /* | |
39 | * Prepare an open request. Preallocate ceph_cap to avoid an | |
40 | * inopportune ENOMEM later. | |
41 | */ | |
42 | static struct ceph_mds_request * | |
43 | prepare_open_request(struct super_block *sb, int flags, int create_mode) | |
44 | { | |
3d14c5d2 YS |
45 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); |
46 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
47 | struct ceph_mds_request *req; |
48 | int want_auth = USE_ANY_MDS; | |
49 | int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN; | |
50 | ||
51 | if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC)) | |
52 | want_auth = USE_AUTH_MDS; | |
53 | ||
54 | req = ceph_mdsc_create_request(mdsc, op, want_auth); | |
55 | if (IS_ERR(req)) | |
56 | goto out; | |
57 | req->r_fmode = ceph_flags_to_mode(flags); | |
58 | req->r_args.open.flags = cpu_to_le32(flags); | |
59 | req->r_args.open.mode = cpu_to_le32(create_mode); | |
124e68e7 SW |
60 | out: |
61 | return req; | |
62 | } | |
63 | ||
64 | /* | |
65 | * initialize private struct file data. | |
66 | * if we fail, clean up by dropping fmode reference on the ceph_inode | |
67 | */ | |
68 | static int ceph_init_file(struct inode *inode, struct file *file, int fmode) | |
69 | { | |
70 | struct ceph_file_info *cf; | |
71 | int ret = 0; | |
72 | ||
73 | switch (inode->i_mode & S_IFMT) { | |
74 | case S_IFREG: | |
75 | case S_IFDIR: | |
76 | dout("init_file %p %p 0%o (regular)\n", inode, file, | |
77 | inode->i_mode); | |
78 | cf = kmem_cache_alloc(ceph_file_cachep, GFP_NOFS | __GFP_ZERO); | |
79 | if (cf == NULL) { | |
80 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
81 | return -ENOMEM; | |
82 | } | |
83 | cf->fmode = fmode; | |
84 | cf->next_offset = 2; | |
85 | file->private_data = cf; | |
86 | BUG_ON(inode->i_fop->release != ceph_release); | |
87 | break; | |
88 | ||
89 | case S_IFLNK: | |
90 | dout("init_file %p %p 0%o (symlink)\n", inode, file, | |
91 | inode->i_mode); | |
92 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
93 | break; | |
94 | ||
95 | default: | |
96 | dout("init_file %p %p 0%o (special)\n", inode, file, | |
97 | inode->i_mode); | |
98 | /* | |
99 | * we need to drop the open ref now, since we don't | |
100 | * have .release set to ceph_release. | |
101 | */ | |
102 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
103 | BUG_ON(inode->i_fop->release == ceph_release); | |
104 | ||
105 | /* call the proper open fop */ | |
106 | ret = inode->i_fop->open(inode, file); | |
107 | } | |
108 | return ret; | |
109 | } | |
110 | ||
111 | /* | |
124e68e7 SW |
112 | * If we already have the requisite capabilities, we can satisfy |
113 | * the open request locally (no need to request new caps from the | |
114 | * MDS). We do, however, need to inform the MDS (asynchronously) | |
115 | * if our wanted caps set expands. | |
116 | */ | |
117 | int ceph_open(struct inode *inode, struct file *file) | |
118 | { | |
119 | struct ceph_inode_info *ci = ceph_inode(inode); | |
3d14c5d2 YS |
120 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
121 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
122 | struct ceph_mds_request *req; |
123 | struct ceph_file_info *cf = file->private_data; | |
5f21c96d | 124 | struct inode *parent_inode = NULL; |
124e68e7 SW |
125 | int err; |
126 | int flags, fmode, wanted; | |
127 | ||
128 | if (cf) { | |
129 | dout("open file %p is already opened\n", file); | |
130 | return 0; | |
131 | } | |
132 | ||
133 | /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */ | |
134 | flags = file->f_flags & ~(O_CREAT|O_EXCL); | |
135 | if (S_ISDIR(inode->i_mode)) | |
136 | flags = O_DIRECTORY; /* mds likes to know */ | |
137 | ||
138 | dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode, | |
139 | ceph_vinop(inode), file, flags, file->f_flags); | |
140 | fmode = ceph_flags_to_mode(flags); | |
141 | wanted = ceph_caps_for_mode(fmode); | |
142 | ||
143 | /* snapped files are read-only */ | |
144 | if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE)) | |
145 | return -EROFS; | |
146 | ||
147 | /* trivially open snapdir */ | |
148 | if (ceph_snap(inode) == CEPH_SNAPDIR) { | |
be655596 | 149 | spin_lock(&ci->i_ceph_lock); |
124e68e7 | 150 | __ceph_get_fmode(ci, fmode); |
be655596 | 151 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
152 | return ceph_init_file(inode, file, fmode); |
153 | } | |
154 | ||
155 | /* | |
7421ab80 SW |
156 | * No need to block if we have caps on the auth MDS (for |
157 | * write) or any MDS (for read). Update wanted set | |
124e68e7 SW |
158 | * asynchronously. |
159 | */ | |
be655596 | 160 | spin_lock(&ci->i_ceph_lock); |
7421ab80 SW |
161 | if (__ceph_is_any_real_caps(ci) && |
162 | (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) { | |
124e68e7 SW |
163 | int mds_wanted = __ceph_caps_mds_wanted(ci); |
164 | int issued = __ceph_caps_issued(ci, NULL); | |
165 | ||
166 | dout("open %p fmode %d want %s issued %s using existing\n", | |
167 | inode, fmode, ceph_cap_string(wanted), | |
168 | ceph_cap_string(issued)); | |
169 | __ceph_get_fmode(ci, fmode); | |
be655596 | 170 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
171 | |
172 | /* adjust wanted? */ | |
173 | if ((issued & wanted) != wanted && | |
174 | (mds_wanted & wanted) != wanted && | |
175 | ceph_snap(inode) != CEPH_SNAPDIR) | |
176 | ceph_check_caps(ci, 0, NULL); | |
177 | ||
178 | return ceph_init_file(inode, file, fmode); | |
179 | } else if (ceph_snap(inode) != CEPH_NOSNAP && | |
180 | (ci->i_snap_caps & wanted) == wanted) { | |
181 | __ceph_get_fmode(ci, fmode); | |
be655596 | 182 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
183 | return ceph_init_file(inode, file, fmode); |
184 | } | |
be655596 | 185 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
186 | |
187 | dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted)); | |
188 | req = prepare_open_request(inode->i_sb, flags, 0); | |
189 | if (IS_ERR(req)) { | |
190 | err = PTR_ERR(req); | |
191 | goto out; | |
192 | } | |
70b666c3 SW |
193 | req->r_inode = inode; |
194 | ihold(inode); | |
124e68e7 | 195 | req->r_num_caps = 1; |
5f21c96d SW |
196 | if (flags & (O_CREAT|O_TRUNC)) |
197 | parent_inode = ceph_get_dentry_parent_inode(file->f_dentry); | |
124e68e7 | 198 | err = ceph_mdsc_do_request(mdsc, parent_inode, req); |
5f21c96d | 199 | iput(parent_inode); |
124e68e7 SW |
200 | if (!err) |
201 | err = ceph_init_file(inode, file, req->r_fmode); | |
202 | ceph_mdsc_put_request(req); | |
203 | dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode)); | |
204 | out: | |
205 | return err; | |
206 | } | |
207 | ||
208 | ||
209 | /* | |
5ef50c3b SW |
210 | * Do a lookup + open with a single request. If we get a non-existent |
211 | * file or symlink, return 1 so the VFS can retry. | |
124e68e7 | 212 | */ |
5ef50c3b | 213 | int ceph_atomic_open(struct inode *dir, struct dentry *dentry, |
30d90494 | 214 | struct file *file, unsigned flags, umode_t mode, |
d9585277 | 215 | int *opened) |
124e68e7 | 216 | { |
3d14c5d2 YS |
217 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
218 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 | 219 | struct ceph_mds_request *req; |
5ef50c3b | 220 | struct dentry *dn; |
124e68e7 | 221 | int err; |
124e68e7 | 222 | |
5ef50c3b SW |
223 | dout("atomic_open %p dentry %p '%.*s' %s flags %d mode 0%o\n", |
224 | dir, dentry, dentry->d_name.len, dentry->d_name.name, | |
225 | d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode); | |
226 | ||
227 | if (dentry->d_name.len > NAME_MAX) | |
228 | return -ENAMETOOLONG; | |
229 | ||
230 | err = ceph_init_dentry(dentry); | |
231 | if (err < 0) | |
232 | return err; | |
124e68e7 SW |
233 | |
234 | /* do the open */ | |
235 | req = prepare_open_request(dir->i_sb, flags, mode); | |
236 | if (IS_ERR(req)) | |
d9585277 | 237 | return PTR_ERR(req); |
124e68e7 SW |
238 | req->r_dentry = dget(dentry); |
239 | req->r_num_caps = 2; | |
240 | if (flags & O_CREAT) { | |
241 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
242 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
243 | } | |
244 | req->r_locked_dir = dir; /* caller holds dir->i_mutex */ | |
acda7657 SW |
245 | err = ceph_mdsc_do_request(mdsc, |
246 | (flags & (O_CREAT|O_TRUNC)) ? dir : NULL, | |
247 | req); | |
79aec984 SL |
248 | if (err) |
249 | goto out_err; | |
250 | ||
468640e3 | 251 | err = ceph_handle_snapdir(req, dentry, err); |
5ef50c3b | 252 | if (err == 0 && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry) |
124e68e7 | 253 | err = ceph_handle_notrace_create(dir, dentry); |
2d83bde9 | 254 | |
5ef50c3b SW |
255 | if (d_unhashed(dentry)) { |
256 | dn = ceph_finish_lookup(req, dentry, err); | |
257 | if (IS_ERR(dn)) | |
258 | err = PTR_ERR(dn); | |
259 | } else { | |
260 | /* we were given a hashed negative dentry */ | |
261 | dn = NULL; | |
262 | } | |
263 | if (err) | |
264 | goto out_err; | |
265 | if (dn || dentry->d_inode == NULL || S_ISLNK(dentry->d_inode->i_mode)) { | |
266 | /* make vfs retry on splice, ENOENT, or symlink */ | |
267 | dout("atomic_open finish_no_open on dn %p\n", dn); | |
268 | err = finish_no_open(file, dn); | |
269 | } else { | |
270 | dout("atomic_open finish_open on dn %p\n", dn); | |
6e8575fa SL |
271 | if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) { |
272 | *opened |= FILE_CREATED; | |
273 | } | |
5ef50c3b SW |
274 | err = finish_open(file, dentry, ceph_open, opened); |
275 | } | |
2d83bde9 | 276 | |
5ef50c3b SW |
277 | out_err: |
278 | ceph_mdsc_put_request(req); | |
279 | dout("atomic_open result=%d\n", err); | |
d9585277 | 280 | return err; |
124e68e7 SW |
281 | } |
282 | ||
283 | int ceph_release(struct inode *inode, struct file *file) | |
284 | { | |
285 | struct ceph_inode_info *ci = ceph_inode(inode); | |
286 | struct ceph_file_info *cf = file->private_data; | |
287 | ||
288 | dout("release inode %p file %p\n", inode, file); | |
289 | ceph_put_fmode(ci, cf->fmode); | |
290 | if (cf->last_readdir) | |
291 | ceph_mdsc_put_request(cf->last_readdir); | |
292 | kfree(cf->last_name); | |
293 | kfree(cf->dir_info); | |
294 | dput(cf->dentry); | |
295 | kmem_cache_free(ceph_file_cachep, cf); | |
195d3ce2 SW |
296 | |
297 | /* wake up anyone waiting for caps on this inode */ | |
03066f23 | 298 | wake_up_all(&ci->i_cap_wq); |
124e68e7 SW |
299 | return 0; |
300 | } | |
301 | ||
124e68e7 SW |
302 | /* |
303 | * Read a range of bytes striped over one or more objects. Iterate over | |
304 | * objects we stripe over. (That's not atomic, but good enough for now.) | |
305 | * | |
306 | * If we get a short result from the OSD, check against i_size; we need to | |
307 | * only return a short read to the caller if we hit EOF. | |
308 | */ | |
309 | static int striped_read(struct inode *inode, | |
310 | u64 off, u64 len, | |
6a026589 | 311 | struct page **pages, int num_pages, |
c3cd6283 | 312 | int *checkeof, bool o_direct, |
ab226e21 | 313 | unsigned long buf_align) |
124e68e7 | 314 | { |
3d14c5d2 | 315 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
124e68e7 | 316 | struct ceph_inode_info *ci = ceph_inode(inode); |
688bac46 | 317 | u64 pos, this_len, left; |
b7495fc2 | 318 | int io_align, page_align; |
688bac46 | 319 | int pages_left; |
124e68e7 SW |
320 | int read; |
321 | struct page **page_pos; | |
322 | int ret; | |
323 | bool hit_stripe, was_short; | |
324 | ||
325 | /* | |
326 | * we may need to do multiple reads. not atomic, unfortunately. | |
327 | */ | |
328 | pos = off; | |
329 | left = len; | |
330 | page_pos = pages; | |
331 | pages_left = num_pages; | |
332 | read = 0; | |
b7495fc2 | 333 | io_align = off & ~PAGE_MASK; |
124e68e7 SW |
334 | |
335 | more: | |
c3cd6283 | 336 | if (o_direct) |
ab226e21 | 337 | page_align = (pos - io_align + buf_align) & ~PAGE_MASK; |
b7495fc2 SW |
338 | else |
339 | page_align = pos & ~PAGE_MASK; | |
124e68e7 | 340 | this_len = left; |
3d14c5d2 | 341 | ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode), |
124e68e7 SW |
342 | &ci->i_layout, pos, &this_len, |
343 | ci->i_truncate_seq, | |
344 | ci->i_truncate_size, | |
b7495fc2 | 345 | page_pos, pages_left, page_align); |
124e68e7 SW |
346 | if (ret == -ENOENT) |
347 | ret = 0; | |
0e98728f SW |
348 | hit_stripe = this_len < left; |
349 | was_short = ret >= 0 && ret < this_len; | |
688bac46 | 350 | dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read, |
124e68e7 SW |
351 | ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : ""); |
352 | ||
02ae66d8 | 353 | if (ret >= 0) { |
354 | int didpages; | |
355 | if (was_short && (pos + ret < inode->i_size)) { | |
356 | u64 tmp = min(this_len - ret, | |
357 | inode->i_size - pos - ret); | |
358 | dout(" zero gap %llu to %llu\n", | |
359 | pos + ret, pos + ret + tmp); | |
360 | ceph_zero_page_vector_range(page_align + read + ret, | |
361 | tmp, pages); | |
362 | ret += tmp; | |
124e68e7 | 363 | } |
02ae66d8 | 364 | |
365 | didpages = (page_align + ret) >> PAGE_CACHE_SHIFT; | |
124e68e7 SW |
366 | pos += ret; |
367 | read = pos - off; | |
368 | left -= ret; | |
369 | page_pos += didpages; | |
370 | pages_left -= didpages; | |
371 | ||
02ae66d8 | 372 | /* hit stripe and need continue*/ |
373 | if (left && hit_stripe && pos < inode->i_size) | |
124e68e7 SW |
374 | goto more; |
375 | } | |
376 | ||
ee7289bf | 377 | if (read > 0) { |
02ae66d8 | 378 | ret = read; |
c3cd6283 SW |
379 | /* did we bounce off eof? */ |
380 | if (pos + left > inode->i_size) | |
381 | *checkeof = 1; | |
124e68e7 SW |
382 | } |
383 | ||
124e68e7 SW |
384 | dout("striped_read returns %d\n", ret); |
385 | return ret; | |
386 | } | |
387 | ||
388 | /* | |
389 | * Completely synchronous read and write methods. Direct from __user | |
390 | * buffer to osd, or directly to user pages (if O_DIRECT). | |
391 | * | |
392 | * If the read spans object boundary, just do multiple reads. | |
393 | */ | |
394 | static ssize_t ceph_sync_read(struct file *file, char __user *data, | |
6a026589 | 395 | unsigned len, loff_t *poff, int *checkeof) |
124e68e7 | 396 | { |
496ad9aa | 397 | struct inode *inode = file_inode(file); |
124e68e7 SW |
398 | struct page **pages; |
399 | u64 off = *poff; | |
ab226e21 | 400 | int num_pages, ret; |
124e68e7 SW |
401 | |
402 | dout("sync_read on file %p %llu~%u %s\n", file, off, len, | |
403 | (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); | |
404 | ||
ab226e21 HC |
405 | if (file->f_flags & O_DIRECT) { |
406 | num_pages = calc_pages_for((unsigned long)data, len); | |
b6aa5901 | 407 | pages = ceph_get_direct_page_vector(data, num_pages, true); |
ab226e21 HC |
408 | } else { |
409 | num_pages = calc_pages_for(off, len); | |
34d23762 | 410 | pages = ceph_alloc_page_vector(num_pages, GFP_NOFS); |
ab226e21 | 411 | } |
124e68e7 SW |
412 | if (IS_ERR(pages)) |
413 | return PTR_ERR(pages); | |
414 | ||
e98b6fed SW |
415 | /* |
416 | * flush any page cache pages in this range. this | |
417 | * will make concurrent normal and sync io slow, | |
418 | * but it will at least behave sensibly when they are | |
419 | * in sequence. | |
420 | */ | |
29065a51 YS |
421 | ret = filemap_write_and_wait(inode->i_mapping); |
422 | if (ret < 0) | |
423 | goto done; | |
424 | ||
b7495fc2 | 425 | ret = striped_read(inode, off, len, pages, num_pages, checkeof, |
ab226e21 HC |
426 | file->f_flags & O_DIRECT, |
427 | (unsigned long)data & ~PAGE_MASK); | |
124e68e7 SW |
428 | |
429 | if (ret >= 0 && (file->f_flags & O_DIRECT) == 0) | |
3d14c5d2 | 430 | ret = ceph_copy_page_vector_to_user(pages, data, off, ret); |
124e68e7 SW |
431 | if (ret >= 0) |
432 | *poff = off + ret; | |
433 | ||
29065a51 | 434 | done: |
124e68e7 | 435 | if (file->f_flags & O_DIRECT) |
b6aa5901 | 436 | ceph_put_page_vector(pages, num_pages, true); |
124e68e7 SW |
437 | else |
438 | ceph_release_page_vector(pages, num_pages); | |
439 | dout("sync_read result %d\n", ret); | |
440 | return ret; | |
441 | } | |
442 | ||
443 | /* | |
26be8808 AE |
444 | * Write commit request unsafe callback, called to tell us when a |
445 | * request is unsafe (that is, in flight--has been handed to the | |
446 | * messenger to send to its target osd). It is called again when | |
447 | * we've received a response message indicating the request is | |
448 | * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request | |
449 | * is completed early (and unsuccessfully) due to a timeout or | |
450 | * interrupt. | |
451 | * | |
452 | * This is used if we requested both an ACK and ONDISK commit reply | |
453 | * from the OSD. | |
124e68e7 | 454 | */ |
26be8808 | 455 | static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe) |
124e68e7 SW |
456 | { |
457 | struct ceph_inode_info *ci = ceph_inode(req->r_inode); | |
458 | ||
26be8808 AE |
459 | dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid, |
460 | unsafe ? "un" : ""); | |
461 | if (unsafe) { | |
462 | ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR); | |
463 | spin_lock(&ci->i_unsafe_lock); | |
464 | list_add_tail(&req->r_unsafe_item, | |
465 | &ci->i_unsafe_writes); | |
466 | spin_unlock(&ci->i_unsafe_lock); | |
467 | } else { | |
468 | spin_lock(&ci->i_unsafe_lock); | |
469 | list_del_init(&req->r_unsafe_item); | |
470 | spin_unlock(&ci->i_unsafe_lock); | |
471 | ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR); | |
472 | } | |
124e68e7 SW |
473 | } |
474 | ||
475 | /* | |
476 | * Synchronous write, straight from __user pointer or user pages (if | |
477 | * O_DIRECT). | |
478 | * | |
479 | * If write spans object boundary, just do multiple writes. (For a | |
480 | * correct atomic write, we should e.g. take write locks on all | |
481 | * objects, rollback on failure, etc.) | |
482 | */ | |
483 | static ssize_t ceph_sync_write(struct file *file, const char __user *data, | |
03d254ed | 484 | size_t left, loff_t pos, loff_t *ppos) |
124e68e7 | 485 | { |
496ad9aa | 486 | struct inode *inode = file_inode(file); |
124e68e7 | 487 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 | 488 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
acead002 AE |
489 | struct ceph_snap_context *snapc; |
490 | struct ceph_vino vino; | |
124e68e7 | 491 | struct ceph_osd_request *req; |
acead002 | 492 | int num_ops = 1; |
124e68e7 SW |
493 | struct page **pages; |
494 | int num_pages; | |
124e68e7 SW |
495 | u64 len; |
496 | int written = 0; | |
497 | int flags; | |
124e68e7 | 498 | int check_caps = 0; |
b7495fc2 | 499 | int page_align, io_align; |
ab226e21 | 500 | unsigned long buf_align; |
124e68e7 SW |
501 | int ret; |
502 | struct timespec mtime = CURRENT_TIME; | |
43bfe5de | 503 | bool own_pages = false; |
124e68e7 | 504 | |
496ad9aa | 505 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) |
124e68e7 SW |
506 | return -EROFS; |
507 | ||
03d254ed | 508 | dout("sync_write on file %p %lld~%u %s\n", file, pos, |
124e68e7 SW |
509 | (unsigned)left, (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); |
510 | ||
29065a51 YS |
511 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + left); |
512 | if (ret < 0) | |
513 | return ret; | |
514 | ||
515 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
516 | pos >> PAGE_CACHE_SHIFT, | |
517 | (pos + left) >> PAGE_CACHE_SHIFT); | |
518 | if (ret < 0) | |
519 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
520 | ||
124e68e7 SW |
521 | flags = CEPH_OSD_FLAG_ORDERSNAP | |
522 | CEPH_OSD_FLAG_ONDISK | | |
523 | CEPH_OSD_FLAG_WRITE; | |
524 | if ((file->f_flags & (O_SYNC|O_DIRECT)) == 0) | |
525 | flags |= CEPH_OSD_FLAG_ACK; | |
526 | else | |
acead002 | 527 | num_ops++; /* Also include a 'startsync' command. */ |
124e68e7 SW |
528 | |
529 | /* | |
530 | * we may need to do multiple writes here if we span an object | |
531 | * boundary. this isn't atomic, unfortunately. :( | |
532 | */ | |
533 | more: | |
d7f124f1 SW |
534 | io_align = pos & ~PAGE_MASK; |
535 | buf_align = (unsigned long)data & ~PAGE_MASK; | |
124e68e7 | 536 | len = left; |
3a42b6c4 | 537 | |
acead002 AE |
538 | snapc = ci->i_snap_realm->cached_context; |
539 | vino = ceph_vino(inode); | |
3d14c5d2 | 540 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, |
79528734 | 541 | vino, pos, &len, num_ops, |
acead002 | 542 | CEPH_OSD_OP_WRITE, flags, snapc, |
124e68e7 | 543 | ci->i_truncate_seq, ci->i_truncate_size, |
acead002 | 544 | false); |
6816282d SW |
545 | if (IS_ERR(req)) |
546 | return PTR_ERR(req); | |
124e68e7 | 547 | |
153e5167 AE |
548 | /* write from beginning of first page, regardless of io alignment */ |
549 | page_align = file->f_flags & O_DIRECT ? buf_align : io_align; | |
550 | num_pages = calc_pages_for(page_align, len); | |
124e68e7 | 551 | if (file->f_flags & O_DIRECT) { |
b6aa5901 | 552 | pages = ceph_get_direct_page_vector(data, num_pages, false); |
124e68e7 SW |
553 | if (IS_ERR(pages)) { |
554 | ret = PTR_ERR(pages); | |
555 | goto out; | |
556 | } | |
557 | ||
558 | /* | |
559 | * throw out any page cache pages in this range. this | |
560 | * may block. | |
561 | */ | |
213c99ee | 562 | truncate_inode_pages_range(inode->i_mapping, pos, |
5c6a2cdb | 563 | (pos+len) | (PAGE_CACHE_SIZE-1)); |
124e68e7 | 564 | } else { |
34d23762 | 565 | pages = ceph_alloc_page_vector(num_pages, GFP_NOFS); |
124e68e7 SW |
566 | if (IS_ERR(pages)) { |
567 | ret = PTR_ERR(pages); | |
568 | goto out; | |
569 | } | |
3d14c5d2 | 570 | ret = ceph_copy_user_to_page_vector(pages, data, pos, len); |
124e68e7 SW |
571 | if (ret < 0) { |
572 | ceph_release_page_vector(pages, num_pages); | |
573 | goto out; | |
574 | } | |
575 | ||
576 | if ((file->f_flags & O_SYNC) == 0) { | |
577 | /* get a second commit callback */ | |
26be8808 AE |
578 | req->r_unsafe_callback = ceph_sync_write_unsafe; |
579 | req->r_inode = inode; | |
43bfe5de | 580 | own_pages = true; |
124e68e7 SW |
581 | } |
582 | } | |
406e2c9f AE |
583 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align, |
584 | false, own_pages); | |
124e68e7 | 585 | |
02ee07d3 | 586 | /* BUG_ON(vino.snap != CEPH_NOSNAP); */ |
79528734 | 587 | ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime); |
02ee07d3 | 588 | |
3d14c5d2 | 589 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
26be8808 | 590 | if (!ret) |
3d14c5d2 | 591 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); |
124e68e7 SW |
592 | |
593 | if (file->f_flags & O_DIRECT) | |
b6aa5901 | 594 | ceph_put_page_vector(pages, num_pages, false); |
124e68e7 SW |
595 | else if (file->f_flags & O_SYNC) |
596 | ceph_release_page_vector(pages, num_pages); | |
597 | ||
598 | out: | |
599 | ceph_osdc_put_request(req); | |
600 | if (ret == 0) { | |
601 | pos += len; | |
602 | written += len; | |
603 | left -= len; | |
022f3e2e | 604 | data += len; |
124e68e7 SW |
605 | if (left) |
606 | goto more; | |
607 | ||
608 | ret = written; | |
03d254ed | 609 | *ppos = pos; |
124e68e7 SW |
610 | if (pos > i_size_read(inode)) |
611 | check_caps = ceph_inode_set_size(inode, pos); | |
612 | if (check_caps) | |
613 | ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, | |
614 | NULL); | |
ee7289bf | 615 | } else if (ret != -EOLDSNAPC && written > 0) { |
616 | ret = written; | |
124e68e7 SW |
617 | } |
618 | return ret; | |
619 | } | |
620 | ||
621 | /* | |
622 | * Wrap generic_file_aio_read with checks for cap bits on the inode. | |
623 | * Atomically grab references, so that those bits are not released | |
624 | * back to the MDS mid-read. | |
625 | * | |
626 | * Hmm, the sync read case isn't actually async... should it be? | |
627 | */ | |
628 | static ssize_t ceph_aio_read(struct kiocb *iocb, const struct iovec *iov, | |
629 | unsigned long nr_segs, loff_t pos) | |
630 | { | |
631 | struct file *filp = iocb->ki_filp; | |
2962507c | 632 | struct ceph_file_info *fi = filp->private_data; |
124e68e7 SW |
633 | loff_t *ppos = &iocb->ki_pos; |
634 | size_t len = iov->iov_len; | |
496ad9aa | 635 | struct inode *inode = file_inode(filp); |
124e68e7 | 636 | struct ceph_inode_info *ci = ceph_inode(inode); |
cd84db6e | 637 | void __user *base = iov->iov_base; |
124e68e7 | 638 | ssize_t ret; |
2962507c | 639 | int want, got = 0; |
6a026589 | 640 | int checkeof = 0, read = 0; |
124e68e7 SW |
641 | |
642 | dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n", | |
643 | inode, ceph_vinop(inode), pos, (unsigned)len, inode); | |
6a026589 | 644 | again: |
2962507c SW |
645 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
646 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
647 | else | |
648 | want = CEPH_CAP_FILE_CACHE; | |
649 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, &got, -1); | |
124e68e7 SW |
650 | if (ret < 0) |
651 | goto out; | |
652 | dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n", | |
653 | inode, ceph_vinop(inode), pos, (unsigned)len, | |
654 | ceph_cap_string(got)); | |
655 | ||
2962507c | 656 | if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 || |
124e68e7 | 657 | (iocb->ki_filp->f_flags & O_DIRECT) || |
4918b6d1 | 658 | (fi->flags & CEPH_F_SYNC)) |
124e68e7 | 659 | /* hmm, this isn't really async... */ |
6a026589 | 660 | ret = ceph_sync_read(filp, base, len, ppos, &checkeof); |
124e68e7 SW |
661 | else |
662 | ret = generic_file_aio_read(iocb, iov, nr_segs, pos); | |
663 | ||
664 | out: | |
665 | dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n", | |
666 | inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret); | |
667 | ceph_put_cap_refs(ci, got); | |
6a026589 SW |
668 | |
669 | if (checkeof && ret >= 0) { | |
670 | int statret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE); | |
671 | ||
672 | /* hit EOF or hole? */ | |
673 | if (statret == 0 && *ppos < inode->i_size) { | |
c3cd6283 | 674 | dout("aio_read sync_read hit hole, ppos %lld < size %lld, reading more\n", *ppos, inode->i_size); |
6a026589 SW |
675 | read += ret; |
676 | base += ret; | |
677 | len -= ret; | |
678 | checkeof = 0; | |
679 | goto again; | |
680 | } | |
681 | } | |
682 | if (ret >= 0) | |
683 | ret += read; | |
684 | ||
124e68e7 SW |
685 | return ret; |
686 | } | |
687 | ||
688 | /* | |
689 | * Take cap references to avoid releasing caps to MDS mid-write. | |
690 | * | |
691 | * If we are synchronous, and write with an old snap context, the OSD | |
692 | * may return EOLDSNAPC. In that case, retry the write.. _after_ | |
693 | * dropping our cap refs and allowing the pending snap to logically | |
694 | * complete _before_ this write occurs. | |
695 | * | |
696 | * If we are near ENOSPC, write synchronously. | |
697 | */ | |
698 | static ssize_t ceph_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
699 | unsigned long nr_segs, loff_t pos) | |
700 | { | |
701 | struct file *file = iocb->ki_filp; | |
33caad32 | 702 | struct ceph_file_info *fi = file->private_data; |
496ad9aa | 703 | struct inode *inode = file_inode(file); |
124e68e7 | 704 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 YS |
705 | struct ceph_osd_client *osdc = |
706 | &ceph_sb_to_client(inode->i_sb)->client->osdc; | |
03d254ed YZ |
707 | ssize_t count, written = 0; |
708 | int err, want, got; | |
124e68e7 SW |
709 | |
710 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
711 | return -EROFS; | |
712 | ||
03d254ed | 713 | mutex_lock(&inode->i_mutex); |
03d254ed YZ |
714 | |
715 | err = generic_segment_checks(iov, &nr_segs, &count, VERIFY_READ); | |
716 | if (err) | |
717 | goto out; | |
718 | ||
719 | /* We can write back this queue in page reclaim */ | |
720 | current->backing_dev_info = file->f_mapping->backing_dev_info; | |
721 | ||
722 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
723 | if (err) | |
724 | goto out; | |
725 | ||
726 | if (count == 0) | |
727 | goto out; | |
728 | ||
729 | err = file_remove_suid(file); | |
730 | if (err) | |
731 | goto out; | |
732 | ||
733 | err = file_update_time(file); | |
734 | if (err) | |
735 | goto out; | |
736 | ||
124e68e7 | 737 | retry_snap: |
6070e0c1 | 738 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) { |
03d254ed | 739 | err = -ENOSPC; |
6070e0c1 YZ |
740 | goto out; |
741 | } | |
03d254ed | 742 | |
ac7f29bf | 743 | dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n", |
03d254ed | 744 | inode, ceph_vinop(inode), pos, count, inode->i_size); |
7971bd92 SW |
745 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
746 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
747 | else | |
748 | want = CEPH_CAP_FILE_BUFFER; | |
03d254ed YZ |
749 | got = 0; |
750 | err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, pos + count); | |
751 | if (err < 0) | |
37505d57 | 752 | goto out; |
124e68e7 | 753 | |
ac7f29bf | 754 | dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n", |
03d254ed | 755 | inode, ceph_vinop(inode), pos, count, ceph_cap_string(got)); |
7971bd92 SW |
756 | |
757 | if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 || | |
758 | (iocb->ki_filp->f_flags & O_DIRECT) || | |
7971bd92 | 759 | (fi->flags & CEPH_F_SYNC)) { |
37505d57 | 760 | mutex_unlock(&inode->i_mutex); |
03d254ed YZ |
761 | written = ceph_sync_write(file, iov->iov_base, count, |
762 | pos, &iocb->ki_pos); | |
0e5dd45c | 763 | if (written == -EOLDSNAPC) { |
764 | dout("aio_write %p %llx.%llx %llu~%u" | |
765 | "got EOLDSNAPC, retrying\n", | |
766 | inode, ceph_vinop(inode), | |
767 | pos, (unsigned)iov->iov_len); | |
768 | mutex_lock(&inode->i_mutex); | |
0e5dd45c | 769 | goto retry_snap; |
770 | } | |
7971bd92 | 771 | } else { |
b0d7c223 YZ |
772 | /* |
773 | * No need to acquire the i_truncate_mutex. Because | |
774 | * the MDS revokes Fwb caps before sending truncate | |
775 | * message to us. We can't get Fwb cap while there | |
776 | * are pending vmtruncate. So write and vmtruncate | |
777 | * can not run at the same time | |
778 | */ | |
03d254ed YZ |
779 | written = generic_file_buffered_write(iocb, iov, nr_segs, |
780 | pos, &iocb->ki_pos, | |
781 | count, 0); | |
6070e0c1 | 782 | mutex_unlock(&inode->i_mutex); |
7971bd92 | 783 | } |
d8de9ab6 | 784 | |
03d254ed | 785 | if (written >= 0) { |
fca65b4a | 786 | int dirty; |
be655596 | 787 | spin_lock(&ci->i_ceph_lock); |
fca65b4a | 788 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR); |
be655596 | 789 | spin_unlock(&ci->i_ceph_lock); |
fca65b4a SW |
790 | if (dirty) |
791 | __mark_inode_dirty(inode, dirty); | |
124e68e7 | 792 | } |
7971bd92 | 793 | |
124e68e7 | 794 | dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n", |
7971bd92 SW |
795 | inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len, |
796 | ceph_cap_string(got)); | |
124e68e7 | 797 | ceph_put_cap_refs(ci, got); |
7971bd92 | 798 | |
03d254ed | 799 | if (written >= 0 && |
6070e0c1 YZ |
800 | ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) || |
801 | ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) { | |
03d254ed | 802 | err = vfs_fsync_range(file, pos, pos + written - 1, 1); |
6070e0c1 | 803 | if (err < 0) |
03d254ed | 804 | written = err; |
6070e0c1 | 805 | } |
03d254ed | 806 | |
2f75e9e1 SW |
807 | goto out_unlocked; |
808 | ||
03d254ed | 809 | out: |
2f75e9e1 SW |
810 | mutex_unlock(&inode->i_mutex); |
811 | out_unlocked: | |
03d254ed | 812 | current->backing_dev_info = NULL; |
03d254ed | 813 | return written ? written : err; |
124e68e7 SW |
814 | } |
815 | ||
816 | /* | |
817 | * llseek. be sure to verify file size on SEEK_END. | |
818 | */ | |
965c8e59 | 819 | static loff_t ceph_llseek(struct file *file, loff_t offset, int whence) |
124e68e7 SW |
820 | { |
821 | struct inode *inode = file->f_mapping->host; | |
822 | int ret; | |
823 | ||
824 | mutex_lock(&inode->i_mutex); | |
6a82c47a | 825 | |
965c8e59 | 826 | if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) { |
124e68e7 SW |
827 | ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE); |
828 | if (ret < 0) { | |
829 | offset = ret; | |
830 | goto out; | |
831 | } | |
06222e49 JB |
832 | } |
833 | ||
965c8e59 | 834 | switch (whence) { |
06222e49 | 835 | case SEEK_END: |
124e68e7 SW |
836 | offset += inode->i_size; |
837 | break; | |
838 | case SEEK_CUR: | |
839 | /* | |
840 | * Here we special-case the lseek(fd, 0, SEEK_CUR) | |
841 | * position-querying operation. Avoid rewriting the "same" | |
842 | * f_pos value back to the file because a concurrent read(), | |
843 | * write() or lseek() might have altered it | |
844 | */ | |
845 | if (offset == 0) { | |
846 | offset = file->f_pos; | |
847 | goto out; | |
848 | } | |
849 | offset += file->f_pos; | |
850 | break; | |
06222e49 JB |
851 | case SEEK_DATA: |
852 | if (offset >= inode->i_size) { | |
853 | ret = -ENXIO; | |
854 | goto out; | |
855 | } | |
856 | break; | |
857 | case SEEK_HOLE: | |
858 | if (offset >= inode->i_size) { | |
859 | ret = -ENXIO; | |
860 | goto out; | |
861 | } | |
862 | offset = inode->i_size; | |
863 | break; | |
124e68e7 SW |
864 | } |
865 | ||
46a1c2c7 | 866 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
124e68e7 SW |
867 | |
868 | out: | |
869 | mutex_unlock(&inode->i_mutex); | |
870 | return offset; | |
871 | } | |
872 | ||
ad7a60de LW |
873 | static inline void ceph_zero_partial_page( |
874 | struct inode *inode, loff_t offset, unsigned size) | |
875 | { | |
876 | struct page *page; | |
877 | pgoff_t index = offset >> PAGE_CACHE_SHIFT; | |
878 | ||
879 | page = find_lock_page(inode->i_mapping, index); | |
880 | if (page) { | |
881 | wait_on_page_writeback(page); | |
882 | zero_user(page, offset & (PAGE_CACHE_SIZE - 1), size); | |
883 | unlock_page(page); | |
884 | page_cache_release(page); | |
885 | } | |
886 | } | |
887 | ||
888 | static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset, | |
889 | loff_t length) | |
890 | { | |
891 | loff_t nearly = round_up(offset, PAGE_CACHE_SIZE); | |
892 | if (offset < nearly) { | |
893 | loff_t size = nearly - offset; | |
894 | if (length < size) | |
895 | size = length; | |
896 | ceph_zero_partial_page(inode, offset, size); | |
897 | offset += size; | |
898 | length -= size; | |
899 | } | |
900 | if (length >= PAGE_CACHE_SIZE) { | |
901 | loff_t size = round_down(length, PAGE_CACHE_SIZE); | |
902 | truncate_pagecache_range(inode, offset, offset + size - 1); | |
903 | offset += size; | |
904 | length -= size; | |
905 | } | |
906 | if (length) | |
907 | ceph_zero_partial_page(inode, offset, length); | |
908 | } | |
909 | ||
910 | static int ceph_zero_partial_object(struct inode *inode, | |
911 | loff_t offset, loff_t *length) | |
912 | { | |
913 | struct ceph_inode_info *ci = ceph_inode(inode); | |
914 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
915 | struct ceph_osd_request *req; | |
916 | int ret = 0; | |
917 | loff_t zero = 0; | |
918 | int op; | |
919 | ||
920 | if (!length) { | |
921 | op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE; | |
922 | length = &zero; | |
923 | } else { | |
924 | op = CEPH_OSD_OP_ZERO; | |
925 | } | |
926 | ||
927 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
928 | ceph_vino(inode), | |
929 | offset, length, | |
930 | 1, op, | |
931 | CEPH_OSD_FLAG_WRITE | | |
932 | CEPH_OSD_FLAG_ONDISK, | |
933 | NULL, 0, 0, false); | |
934 | if (IS_ERR(req)) { | |
935 | ret = PTR_ERR(req); | |
936 | goto out; | |
937 | } | |
938 | ||
939 | ceph_osdc_build_request(req, offset, NULL, ceph_vino(inode).snap, | |
940 | &inode->i_mtime); | |
941 | ||
942 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); | |
943 | if (!ret) { | |
944 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
945 | if (ret == -ENOENT) | |
946 | ret = 0; | |
947 | } | |
948 | ceph_osdc_put_request(req); | |
949 | ||
950 | out: | |
951 | return ret; | |
952 | } | |
953 | ||
954 | static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length) | |
955 | { | |
956 | int ret = 0; | |
957 | struct ceph_inode_info *ci = ceph_inode(inode); | |
b314a90d SW |
958 | s32 stripe_unit = ceph_file_layout_su(ci->i_layout); |
959 | s32 stripe_count = ceph_file_layout_stripe_count(ci->i_layout); | |
960 | s32 object_size = ceph_file_layout_object_size(ci->i_layout); | |
961 | u64 object_set_size = object_size * stripe_count; | |
962 | u64 nearly, t; | |
963 | ||
964 | /* round offset up to next period boundary */ | |
965 | nearly = offset + object_set_size - 1; | |
966 | t = nearly; | |
967 | nearly -= do_div(t, object_set_size); | |
ad7a60de | 968 | |
ad7a60de LW |
969 | while (length && offset < nearly) { |
970 | loff_t size = length; | |
971 | ret = ceph_zero_partial_object(inode, offset, &size); | |
972 | if (ret < 0) | |
973 | return ret; | |
974 | offset += size; | |
975 | length -= size; | |
976 | } | |
977 | while (length >= object_set_size) { | |
978 | int i; | |
979 | loff_t pos = offset; | |
980 | for (i = 0; i < stripe_count; ++i) { | |
981 | ret = ceph_zero_partial_object(inode, pos, NULL); | |
982 | if (ret < 0) | |
983 | return ret; | |
984 | pos += stripe_unit; | |
985 | } | |
986 | offset += object_set_size; | |
987 | length -= object_set_size; | |
988 | } | |
989 | while (length) { | |
990 | loff_t size = length; | |
991 | ret = ceph_zero_partial_object(inode, offset, &size); | |
992 | if (ret < 0) | |
993 | return ret; | |
994 | offset += size; | |
995 | length -= size; | |
996 | } | |
997 | return ret; | |
998 | } | |
999 | ||
1000 | static long ceph_fallocate(struct file *file, int mode, | |
1001 | loff_t offset, loff_t length) | |
1002 | { | |
1003 | struct ceph_file_info *fi = file->private_data; | |
1004 | struct inode *inode = file->f_dentry->d_inode; | |
1005 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1006 | struct ceph_osd_client *osdc = | |
1007 | &ceph_inode_to_client(inode)->client->osdc; | |
1008 | int want, got = 0; | |
1009 | int dirty; | |
1010 | int ret = 0; | |
1011 | loff_t endoff = 0; | |
1012 | loff_t size; | |
1013 | ||
1014 | if (!S_ISREG(inode->i_mode)) | |
1015 | return -EOPNOTSUPP; | |
1016 | ||
1017 | if (IS_SWAPFILE(inode)) | |
1018 | return -ETXTBSY; | |
1019 | ||
1020 | mutex_lock(&inode->i_mutex); | |
1021 | ||
1022 | if (ceph_snap(inode) != CEPH_NOSNAP) { | |
1023 | ret = -EROFS; | |
1024 | goto unlock; | |
1025 | } | |
1026 | ||
1027 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) && | |
1028 | !(mode & FALLOC_FL_PUNCH_HOLE)) { | |
1029 | ret = -ENOSPC; | |
1030 | goto unlock; | |
1031 | } | |
1032 | ||
1033 | size = i_size_read(inode); | |
1034 | if (!(mode & FALLOC_FL_KEEP_SIZE)) | |
1035 | endoff = offset + length; | |
1036 | ||
1037 | if (fi->fmode & CEPH_FILE_MODE_LAZY) | |
1038 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1039 | else | |
1040 | want = CEPH_CAP_FILE_BUFFER; | |
1041 | ||
1042 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, endoff); | |
1043 | if (ret < 0) | |
1044 | goto unlock; | |
1045 | ||
1046 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
1047 | if (offset < size) | |
1048 | ceph_zero_pagecache_range(inode, offset, length); | |
1049 | ret = ceph_zero_objects(inode, offset, length); | |
1050 | } else if (endoff > size) { | |
1051 | truncate_pagecache_range(inode, size, -1); | |
1052 | if (ceph_inode_set_size(inode, endoff)) | |
1053 | ceph_check_caps(ceph_inode(inode), | |
1054 | CHECK_CAPS_AUTHONLY, NULL); | |
1055 | } | |
1056 | ||
1057 | if (!ret) { | |
1058 | spin_lock(&ci->i_ceph_lock); | |
1059 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR); | |
1060 | spin_unlock(&ci->i_ceph_lock); | |
1061 | if (dirty) | |
1062 | __mark_inode_dirty(inode, dirty); | |
1063 | } | |
1064 | ||
1065 | ceph_put_cap_refs(ci, got); | |
1066 | unlock: | |
1067 | mutex_unlock(&inode->i_mutex); | |
1068 | return ret; | |
1069 | } | |
1070 | ||
124e68e7 SW |
1071 | const struct file_operations ceph_file_fops = { |
1072 | .open = ceph_open, | |
1073 | .release = ceph_release, | |
1074 | .llseek = ceph_llseek, | |
1075 | .read = do_sync_read, | |
1076 | .write = do_sync_write, | |
1077 | .aio_read = ceph_aio_read, | |
1078 | .aio_write = ceph_aio_write, | |
1079 | .mmap = ceph_mmap, | |
1080 | .fsync = ceph_fsync, | |
40819f6f GF |
1081 | .lock = ceph_lock, |
1082 | .flock = ceph_flock, | |
124e68e7 SW |
1083 | .splice_read = generic_file_splice_read, |
1084 | .splice_write = generic_file_splice_write, | |
1085 | .unlocked_ioctl = ceph_ioctl, | |
1086 | .compat_ioctl = ceph_ioctl, | |
ad7a60de | 1087 | .fallocate = ceph_fallocate, |
124e68e7 SW |
1088 | }; |
1089 |