]>
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 | ||
353 | if (ret > 0) { | |
773e9b44 | 354 | int didpages = (page_align + ret) >> PAGE_CACHE_SHIFT; |
124e68e7 SW |
355 | |
356 | if (read < pos - off) { | |
357 | dout(" zero gap %llu to %llu\n", off + read, pos); | |
773e9b44 | 358 | ceph_zero_page_vector_range(page_align + read, |
3d14c5d2 | 359 | pos - off - read, pages); |
124e68e7 SW |
360 | } |
361 | pos += ret; | |
362 | read = pos - off; | |
363 | left -= ret; | |
364 | page_pos += didpages; | |
365 | pages_left -= didpages; | |
366 | ||
367 | /* hit stripe? */ | |
368 | if (left && hit_stripe) | |
369 | goto more; | |
370 | } | |
371 | ||
372 | if (was_short) { | |
c3cd6283 SW |
373 | /* did we bounce off eof? */ |
374 | if (pos + left > inode->i_size) | |
375 | *checkeof = 1; | |
376 | ||
377 | /* zero trailing bytes (inside i_size) */ | |
378 | if (left > 0 && pos < inode->i_size) { | |
379 | if (pos + left > inode->i_size) | |
380 | left = inode->i_size - pos; | |
381 | ||
688bac46 | 382 | dout("zero tail %llu\n", left); |
773e9b44 | 383 | ceph_zero_page_vector_range(page_align + read, left, |
3d14c5d2 | 384 | pages); |
c3cd6283 | 385 | read += left; |
124e68e7 | 386 | } |
124e68e7 SW |
387 | } |
388 | ||
124e68e7 SW |
389 | if (ret >= 0) |
390 | ret = read; | |
391 | dout("striped_read returns %d\n", ret); | |
392 | return ret; | |
393 | } | |
394 | ||
395 | /* | |
396 | * Completely synchronous read and write methods. Direct from __user | |
397 | * buffer to osd, or directly to user pages (if O_DIRECT). | |
398 | * | |
399 | * If the read spans object boundary, just do multiple reads. | |
400 | */ | |
401 | static ssize_t ceph_sync_read(struct file *file, char __user *data, | |
6a026589 | 402 | unsigned len, loff_t *poff, int *checkeof) |
124e68e7 | 403 | { |
496ad9aa | 404 | struct inode *inode = file_inode(file); |
124e68e7 SW |
405 | struct page **pages; |
406 | u64 off = *poff; | |
ab226e21 | 407 | int num_pages, ret; |
124e68e7 SW |
408 | |
409 | dout("sync_read on file %p %llu~%u %s\n", file, off, len, | |
410 | (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); | |
411 | ||
ab226e21 HC |
412 | if (file->f_flags & O_DIRECT) { |
413 | num_pages = calc_pages_for((unsigned long)data, len); | |
b6aa5901 | 414 | pages = ceph_get_direct_page_vector(data, num_pages, true); |
ab226e21 HC |
415 | } else { |
416 | num_pages = calc_pages_for(off, len); | |
34d23762 | 417 | pages = ceph_alloc_page_vector(num_pages, GFP_NOFS); |
ab226e21 | 418 | } |
124e68e7 SW |
419 | if (IS_ERR(pages)) |
420 | return PTR_ERR(pages); | |
421 | ||
e98b6fed SW |
422 | /* |
423 | * flush any page cache pages in this range. this | |
424 | * will make concurrent normal and sync io slow, | |
425 | * but it will at least behave sensibly when they are | |
426 | * in sequence. | |
427 | */ | |
29065a51 YS |
428 | ret = filemap_write_and_wait(inode->i_mapping); |
429 | if (ret < 0) | |
430 | goto done; | |
431 | ||
b7495fc2 | 432 | ret = striped_read(inode, off, len, pages, num_pages, checkeof, |
ab226e21 HC |
433 | file->f_flags & O_DIRECT, |
434 | (unsigned long)data & ~PAGE_MASK); | |
124e68e7 SW |
435 | |
436 | if (ret >= 0 && (file->f_flags & O_DIRECT) == 0) | |
3d14c5d2 | 437 | ret = ceph_copy_page_vector_to_user(pages, data, off, ret); |
124e68e7 SW |
438 | if (ret >= 0) |
439 | *poff = off + ret; | |
440 | ||
29065a51 | 441 | done: |
124e68e7 | 442 | if (file->f_flags & O_DIRECT) |
b6aa5901 | 443 | ceph_put_page_vector(pages, num_pages, true); |
124e68e7 SW |
444 | else |
445 | ceph_release_page_vector(pages, num_pages); | |
446 | dout("sync_read result %d\n", ret); | |
447 | return ret; | |
448 | } | |
449 | ||
450 | /* | |
26be8808 AE |
451 | * Write commit request unsafe callback, called to tell us when a |
452 | * request is unsafe (that is, in flight--has been handed to the | |
453 | * messenger to send to its target osd). It is called again when | |
454 | * we've received a response message indicating the request is | |
455 | * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request | |
456 | * is completed early (and unsuccessfully) due to a timeout or | |
457 | * interrupt. | |
458 | * | |
459 | * This is used if we requested both an ACK and ONDISK commit reply | |
460 | * from the OSD. | |
124e68e7 | 461 | */ |
26be8808 | 462 | static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe) |
124e68e7 SW |
463 | { |
464 | struct ceph_inode_info *ci = ceph_inode(req->r_inode); | |
465 | ||
26be8808 AE |
466 | dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid, |
467 | unsafe ? "un" : ""); | |
468 | if (unsafe) { | |
469 | ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR); | |
470 | spin_lock(&ci->i_unsafe_lock); | |
471 | list_add_tail(&req->r_unsafe_item, | |
472 | &ci->i_unsafe_writes); | |
473 | spin_unlock(&ci->i_unsafe_lock); | |
474 | } else { | |
475 | spin_lock(&ci->i_unsafe_lock); | |
476 | list_del_init(&req->r_unsafe_item); | |
477 | spin_unlock(&ci->i_unsafe_lock); | |
478 | ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR); | |
479 | } | |
124e68e7 SW |
480 | } |
481 | ||
482 | /* | |
483 | * Synchronous write, straight from __user pointer or user pages (if | |
484 | * O_DIRECT). | |
485 | * | |
486 | * If write spans object boundary, just do multiple writes. (For a | |
487 | * correct atomic write, we should e.g. take write locks on all | |
488 | * objects, rollback on failure, etc.) | |
489 | */ | |
490 | static ssize_t ceph_sync_write(struct file *file, const char __user *data, | |
03d254ed | 491 | size_t left, loff_t pos, loff_t *ppos) |
124e68e7 | 492 | { |
496ad9aa | 493 | struct inode *inode = file_inode(file); |
124e68e7 | 494 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 | 495 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
acead002 AE |
496 | struct ceph_snap_context *snapc; |
497 | struct ceph_vino vino; | |
124e68e7 | 498 | struct ceph_osd_request *req; |
acead002 | 499 | int num_ops = 1; |
124e68e7 SW |
500 | struct page **pages; |
501 | int num_pages; | |
124e68e7 SW |
502 | u64 len; |
503 | int written = 0; | |
504 | int flags; | |
124e68e7 | 505 | int check_caps = 0; |
b7495fc2 | 506 | int page_align, io_align; |
ab226e21 | 507 | unsigned long buf_align; |
124e68e7 SW |
508 | int ret; |
509 | struct timespec mtime = CURRENT_TIME; | |
43bfe5de | 510 | bool own_pages = false; |
124e68e7 | 511 | |
496ad9aa | 512 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) |
124e68e7 SW |
513 | return -EROFS; |
514 | ||
03d254ed | 515 | dout("sync_write on file %p %lld~%u %s\n", file, pos, |
124e68e7 SW |
516 | (unsigned)left, (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); |
517 | ||
29065a51 YS |
518 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + left); |
519 | if (ret < 0) | |
520 | return ret; | |
521 | ||
522 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
523 | pos >> PAGE_CACHE_SHIFT, | |
524 | (pos + left) >> PAGE_CACHE_SHIFT); | |
525 | if (ret < 0) | |
526 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
527 | ||
124e68e7 SW |
528 | flags = CEPH_OSD_FLAG_ORDERSNAP | |
529 | CEPH_OSD_FLAG_ONDISK | | |
530 | CEPH_OSD_FLAG_WRITE; | |
531 | if ((file->f_flags & (O_SYNC|O_DIRECT)) == 0) | |
532 | flags |= CEPH_OSD_FLAG_ACK; | |
533 | else | |
acead002 | 534 | num_ops++; /* Also include a 'startsync' command. */ |
124e68e7 SW |
535 | |
536 | /* | |
537 | * we may need to do multiple writes here if we span an object | |
538 | * boundary. this isn't atomic, unfortunately. :( | |
539 | */ | |
540 | more: | |
d7f124f1 SW |
541 | io_align = pos & ~PAGE_MASK; |
542 | buf_align = (unsigned long)data & ~PAGE_MASK; | |
124e68e7 | 543 | len = left; |
3a42b6c4 | 544 | |
acead002 AE |
545 | snapc = ci->i_snap_realm->cached_context; |
546 | vino = ceph_vino(inode); | |
3d14c5d2 | 547 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, |
79528734 | 548 | vino, pos, &len, num_ops, |
acead002 | 549 | CEPH_OSD_OP_WRITE, flags, snapc, |
124e68e7 | 550 | ci->i_truncate_seq, ci->i_truncate_size, |
acead002 | 551 | false); |
6816282d SW |
552 | if (IS_ERR(req)) |
553 | return PTR_ERR(req); | |
124e68e7 | 554 | |
153e5167 AE |
555 | /* write from beginning of first page, regardless of io alignment */ |
556 | page_align = file->f_flags & O_DIRECT ? buf_align : io_align; | |
557 | num_pages = calc_pages_for(page_align, len); | |
124e68e7 | 558 | if (file->f_flags & O_DIRECT) { |
b6aa5901 | 559 | pages = ceph_get_direct_page_vector(data, num_pages, false); |
124e68e7 SW |
560 | if (IS_ERR(pages)) { |
561 | ret = PTR_ERR(pages); | |
562 | goto out; | |
563 | } | |
564 | ||
565 | /* | |
566 | * throw out any page cache pages in this range. this | |
567 | * may block. | |
568 | */ | |
213c99ee | 569 | truncate_inode_pages_range(inode->i_mapping, pos, |
5c6a2cdb | 570 | (pos+len) | (PAGE_CACHE_SIZE-1)); |
124e68e7 | 571 | } else { |
34d23762 | 572 | pages = ceph_alloc_page_vector(num_pages, GFP_NOFS); |
124e68e7 SW |
573 | if (IS_ERR(pages)) { |
574 | ret = PTR_ERR(pages); | |
575 | goto out; | |
576 | } | |
3d14c5d2 | 577 | ret = ceph_copy_user_to_page_vector(pages, data, pos, len); |
124e68e7 SW |
578 | if (ret < 0) { |
579 | ceph_release_page_vector(pages, num_pages); | |
580 | goto out; | |
581 | } | |
582 | ||
583 | if ((file->f_flags & O_SYNC) == 0) { | |
584 | /* get a second commit callback */ | |
26be8808 AE |
585 | req->r_unsafe_callback = ceph_sync_write_unsafe; |
586 | req->r_inode = inode; | |
43bfe5de | 587 | own_pages = true; |
124e68e7 SW |
588 | } |
589 | } | |
406e2c9f AE |
590 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align, |
591 | false, own_pages); | |
124e68e7 | 592 | |
02ee07d3 | 593 | /* BUG_ON(vino.snap != CEPH_NOSNAP); */ |
79528734 | 594 | ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime); |
02ee07d3 | 595 | |
3d14c5d2 | 596 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
26be8808 | 597 | if (!ret) |
3d14c5d2 | 598 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); |
124e68e7 SW |
599 | |
600 | if (file->f_flags & O_DIRECT) | |
b6aa5901 | 601 | ceph_put_page_vector(pages, num_pages, false); |
124e68e7 SW |
602 | else if (file->f_flags & O_SYNC) |
603 | ceph_release_page_vector(pages, num_pages); | |
604 | ||
605 | out: | |
606 | ceph_osdc_put_request(req); | |
607 | if (ret == 0) { | |
608 | pos += len; | |
609 | written += len; | |
610 | left -= len; | |
022f3e2e | 611 | data += len; |
124e68e7 SW |
612 | if (left) |
613 | goto more; | |
614 | ||
615 | ret = written; | |
03d254ed | 616 | *ppos = pos; |
124e68e7 SW |
617 | if (pos > i_size_read(inode)) |
618 | check_caps = ceph_inode_set_size(inode, pos); | |
619 | if (check_caps) | |
620 | ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, | |
621 | NULL); | |
622 | } | |
623 | return ret; | |
624 | } | |
625 | ||
626 | /* | |
627 | * Wrap generic_file_aio_read with checks for cap bits on the inode. | |
628 | * Atomically grab references, so that those bits are not released | |
629 | * back to the MDS mid-read. | |
630 | * | |
631 | * Hmm, the sync read case isn't actually async... should it be? | |
632 | */ | |
633 | static ssize_t ceph_aio_read(struct kiocb *iocb, const struct iovec *iov, | |
634 | unsigned long nr_segs, loff_t pos) | |
635 | { | |
636 | struct file *filp = iocb->ki_filp; | |
2962507c | 637 | struct ceph_file_info *fi = filp->private_data; |
124e68e7 SW |
638 | loff_t *ppos = &iocb->ki_pos; |
639 | size_t len = iov->iov_len; | |
496ad9aa | 640 | struct inode *inode = file_inode(filp); |
124e68e7 | 641 | struct ceph_inode_info *ci = ceph_inode(inode); |
cd84db6e | 642 | void __user *base = iov->iov_base; |
124e68e7 | 643 | ssize_t ret; |
2962507c | 644 | int want, got = 0; |
6a026589 | 645 | int checkeof = 0, read = 0; |
124e68e7 SW |
646 | |
647 | dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n", | |
648 | inode, ceph_vinop(inode), pos, (unsigned)len, inode); | |
6a026589 | 649 | again: |
2962507c SW |
650 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
651 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
652 | else | |
653 | want = CEPH_CAP_FILE_CACHE; | |
654 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, &got, -1); | |
124e68e7 SW |
655 | if (ret < 0) |
656 | goto out; | |
657 | dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n", | |
658 | inode, ceph_vinop(inode), pos, (unsigned)len, | |
659 | ceph_cap_string(got)); | |
660 | ||
2962507c | 661 | if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 || |
124e68e7 | 662 | (iocb->ki_filp->f_flags & O_DIRECT) || |
4918b6d1 | 663 | (fi->flags & CEPH_F_SYNC)) |
124e68e7 | 664 | /* hmm, this isn't really async... */ |
6a026589 | 665 | ret = ceph_sync_read(filp, base, len, ppos, &checkeof); |
124e68e7 SW |
666 | else |
667 | ret = generic_file_aio_read(iocb, iov, nr_segs, pos); | |
668 | ||
669 | out: | |
670 | dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n", | |
671 | inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret); | |
672 | ceph_put_cap_refs(ci, got); | |
6a026589 SW |
673 | |
674 | if (checkeof && ret >= 0) { | |
675 | int statret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE); | |
676 | ||
677 | /* hit EOF or hole? */ | |
678 | if (statret == 0 && *ppos < inode->i_size) { | |
c3cd6283 | 679 | dout("aio_read sync_read hit hole, ppos %lld < size %lld, reading more\n", *ppos, inode->i_size); |
6a026589 SW |
680 | read += ret; |
681 | base += ret; | |
682 | len -= ret; | |
683 | checkeof = 0; | |
684 | goto again; | |
685 | } | |
686 | } | |
687 | if (ret >= 0) | |
688 | ret += read; | |
689 | ||
124e68e7 SW |
690 | return ret; |
691 | } | |
692 | ||
693 | /* | |
694 | * Take cap references to avoid releasing caps to MDS mid-write. | |
695 | * | |
696 | * If we are synchronous, and write with an old snap context, the OSD | |
697 | * may return EOLDSNAPC. In that case, retry the write.. _after_ | |
698 | * dropping our cap refs and allowing the pending snap to logically | |
699 | * complete _before_ this write occurs. | |
700 | * | |
701 | * If we are near ENOSPC, write synchronously. | |
702 | */ | |
703 | static ssize_t ceph_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
704 | unsigned long nr_segs, loff_t pos) | |
705 | { | |
706 | struct file *file = iocb->ki_filp; | |
33caad32 | 707 | struct ceph_file_info *fi = file->private_data; |
496ad9aa | 708 | struct inode *inode = file_inode(file); |
124e68e7 | 709 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 YS |
710 | struct ceph_osd_client *osdc = |
711 | &ceph_sb_to_client(inode->i_sb)->client->osdc; | |
03d254ed YZ |
712 | ssize_t count, written = 0; |
713 | int err, want, got; | |
124e68e7 SW |
714 | |
715 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
716 | return -EROFS; | |
717 | ||
03d254ed | 718 | mutex_lock(&inode->i_mutex); |
03d254ed YZ |
719 | |
720 | err = generic_segment_checks(iov, &nr_segs, &count, VERIFY_READ); | |
721 | if (err) | |
722 | goto out; | |
723 | ||
724 | /* We can write back this queue in page reclaim */ | |
725 | current->backing_dev_info = file->f_mapping->backing_dev_info; | |
726 | ||
727 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
728 | if (err) | |
729 | goto out; | |
730 | ||
731 | if (count == 0) | |
732 | goto out; | |
733 | ||
734 | err = file_remove_suid(file); | |
735 | if (err) | |
736 | goto out; | |
737 | ||
738 | err = file_update_time(file); | |
739 | if (err) | |
740 | goto out; | |
741 | ||
124e68e7 | 742 | retry_snap: |
6070e0c1 | 743 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) { |
03d254ed | 744 | err = -ENOSPC; |
6070e0c1 YZ |
745 | goto out; |
746 | } | |
03d254ed | 747 | |
ac7f29bf | 748 | dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n", |
03d254ed | 749 | inode, ceph_vinop(inode), pos, count, inode->i_size); |
7971bd92 SW |
750 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
751 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
752 | else | |
753 | want = CEPH_CAP_FILE_BUFFER; | |
03d254ed YZ |
754 | got = 0; |
755 | err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, pos + count); | |
756 | if (err < 0) | |
37505d57 | 757 | goto out; |
124e68e7 | 758 | |
ac7f29bf | 759 | dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n", |
03d254ed | 760 | inode, ceph_vinop(inode), pos, count, ceph_cap_string(got)); |
7971bd92 SW |
761 | |
762 | if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 || | |
763 | (iocb->ki_filp->f_flags & O_DIRECT) || | |
7971bd92 | 764 | (fi->flags & CEPH_F_SYNC)) { |
37505d57 | 765 | mutex_unlock(&inode->i_mutex); |
03d254ed YZ |
766 | written = ceph_sync_write(file, iov->iov_base, count, |
767 | pos, &iocb->ki_pos); | |
0e5dd45c | 768 | if (written == -EOLDSNAPC) { |
769 | dout("aio_write %p %llx.%llx %llu~%u" | |
770 | "got EOLDSNAPC, retrying\n", | |
771 | inode, ceph_vinop(inode), | |
772 | pos, (unsigned)iov->iov_len); | |
773 | mutex_lock(&inode->i_mutex); | |
0e5dd45c | 774 | goto retry_snap; |
775 | } | |
7971bd92 | 776 | } else { |
b0d7c223 YZ |
777 | /* |
778 | * No need to acquire the i_truncate_mutex. Because | |
779 | * the MDS revokes Fwb caps before sending truncate | |
780 | * message to us. We can't get Fwb cap while there | |
781 | * are pending vmtruncate. So write and vmtruncate | |
782 | * can not run at the same time | |
783 | */ | |
03d254ed YZ |
784 | written = generic_file_buffered_write(iocb, iov, nr_segs, |
785 | pos, &iocb->ki_pos, | |
786 | count, 0); | |
6070e0c1 | 787 | mutex_unlock(&inode->i_mutex); |
7971bd92 | 788 | } |
d8de9ab6 | 789 | |
03d254ed | 790 | if (written >= 0) { |
fca65b4a | 791 | int dirty; |
be655596 | 792 | spin_lock(&ci->i_ceph_lock); |
fca65b4a | 793 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR); |
be655596 | 794 | spin_unlock(&ci->i_ceph_lock); |
fca65b4a SW |
795 | if (dirty) |
796 | __mark_inode_dirty(inode, dirty); | |
124e68e7 | 797 | } |
7971bd92 | 798 | |
124e68e7 | 799 | dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n", |
7971bd92 SW |
800 | inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len, |
801 | ceph_cap_string(got)); | |
124e68e7 | 802 | ceph_put_cap_refs(ci, got); |
7971bd92 | 803 | |
03d254ed | 804 | if (written >= 0 && |
6070e0c1 YZ |
805 | ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) || |
806 | ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) { | |
03d254ed | 807 | err = vfs_fsync_range(file, pos, pos + written - 1, 1); |
6070e0c1 | 808 | if (err < 0) |
03d254ed | 809 | written = err; |
6070e0c1 | 810 | } |
03d254ed | 811 | |
2f75e9e1 SW |
812 | goto out_unlocked; |
813 | ||
03d254ed | 814 | out: |
2f75e9e1 SW |
815 | mutex_unlock(&inode->i_mutex); |
816 | out_unlocked: | |
03d254ed | 817 | current->backing_dev_info = NULL; |
03d254ed | 818 | return written ? written : err; |
124e68e7 SW |
819 | } |
820 | ||
821 | /* | |
822 | * llseek. be sure to verify file size on SEEK_END. | |
823 | */ | |
965c8e59 | 824 | static loff_t ceph_llseek(struct file *file, loff_t offset, int whence) |
124e68e7 SW |
825 | { |
826 | struct inode *inode = file->f_mapping->host; | |
827 | int ret; | |
828 | ||
829 | mutex_lock(&inode->i_mutex); | |
6a82c47a | 830 | |
965c8e59 | 831 | if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) { |
124e68e7 SW |
832 | ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE); |
833 | if (ret < 0) { | |
834 | offset = ret; | |
835 | goto out; | |
836 | } | |
06222e49 JB |
837 | } |
838 | ||
965c8e59 | 839 | switch (whence) { |
06222e49 | 840 | case SEEK_END: |
124e68e7 SW |
841 | offset += inode->i_size; |
842 | break; | |
843 | case SEEK_CUR: | |
844 | /* | |
845 | * Here we special-case the lseek(fd, 0, SEEK_CUR) | |
846 | * position-querying operation. Avoid rewriting the "same" | |
847 | * f_pos value back to the file because a concurrent read(), | |
848 | * write() or lseek() might have altered it | |
849 | */ | |
850 | if (offset == 0) { | |
851 | offset = file->f_pos; | |
852 | goto out; | |
853 | } | |
854 | offset += file->f_pos; | |
855 | break; | |
06222e49 JB |
856 | case SEEK_DATA: |
857 | if (offset >= inode->i_size) { | |
858 | ret = -ENXIO; | |
859 | goto out; | |
860 | } | |
861 | break; | |
862 | case SEEK_HOLE: | |
863 | if (offset >= inode->i_size) { | |
864 | ret = -ENXIO; | |
865 | goto out; | |
866 | } | |
867 | offset = inode->i_size; | |
868 | break; | |
124e68e7 SW |
869 | } |
870 | ||
46a1c2c7 | 871 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
124e68e7 SW |
872 | |
873 | out: | |
874 | mutex_unlock(&inode->i_mutex); | |
875 | return offset; | |
876 | } | |
877 | ||
ad7a60de LW |
878 | static inline void ceph_zero_partial_page( |
879 | struct inode *inode, loff_t offset, unsigned size) | |
880 | { | |
881 | struct page *page; | |
882 | pgoff_t index = offset >> PAGE_CACHE_SHIFT; | |
883 | ||
884 | page = find_lock_page(inode->i_mapping, index); | |
885 | if (page) { | |
886 | wait_on_page_writeback(page); | |
887 | zero_user(page, offset & (PAGE_CACHE_SIZE - 1), size); | |
888 | unlock_page(page); | |
889 | page_cache_release(page); | |
890 | } | |
891 | } | |
892 | ||
893 | static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset, | |
894 | loff_t length) | |
895 | { | |
896 | loff_t nearly = round_up(offset, PAGE_CACHE_SIZE); | |
897 | if (offset < nearly) { | |
898 | loff_t size = nearly - offset; | |
899 | if (length < size) | |
900 | size = length; | |
901 | ceph_zero_partial_page(inode, offset, size); | |
902 | offset += size; | |
903 | length -= size; | |
904 | } | |
905 | if (length >= PAGE_CACHE_SIZE) { | |
906 | loff_t size = round_down(length, PAGE_CACHE_SIZE); | |
907 | truncate_pagecache_range(inode, offset, offset + size - 1); | |
908 | offset += size; | |
909 | length -= size; | |
910 | } | |
911 | if (length) | |
912 | ceph_zero_partial_page(inode, offset, length); | |
913 | } | |
914 | ||
915 | static int ceph_zero_partial_object(struct inode *inode, | |
916 | loff_t offset, loff_t *length) | |
917 | { | |
918 | struct ceph_inode_info *ci = ceph_inode(inode); | |
919 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
920 | struct ceph_osd_request *req; | |
921 | int ret = 0; | |
922 | loff_t zero = 0; | |
923 | int op; | |
924 | ||
925 | if (!length) { | |
926 | op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE; | |
927 | length = &zero; | |
928 | } else { | |
929 | op = CEPH_OSD_OP_ZERO; | |
930 | } | |
931 | ||
932 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
933 | ceph_vino(inode), | |
934 | offset, length, | |
935 | 1, op, | |
936 | CEPH_OSD_FLAG_WRITE | | |
937 | CEPH_OSD_FLAG_ONDISK, | |
938 | NULL, 0, 0, false); | |
939 | if (IS_ERR(req)) { | |
940 | ret = PTR_ERR(req); | |
941 | goto out; | |
942 | } | |
943 | ||
944 | ceph_osdc_build_request(req, offset, NULL, ceph_vino(inode).snap, | |
945 | &inode->i_mtime); | |
946 | ||
947 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); | |
948 | if (!ret) { | |
949 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
950 | if (ret == -ENOENT) | |
951 | ret = 0; | |
952 | } | |
953 | ceph_osdc_put_request(req); | |
954 | ||
955 | out: | |
956 | return ret; | |
957 | } | |
958 | ||
959 | static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length) | |
960 | { | |
961 | int ret = 0; | |
962 | struct ceph_inode_info *ci = ceph_inode(inode); | |
b314a90d SW |
963 | s32 stripe_unit = ceph_file_layout_su(ci->i_layout); |
964 | s32 stripe_count = ceph_file_layout_stripe_count(ci->i_layout); | |
965 | s32 object_size = ceph_file_layout_object_size(ci->i_layout); | |
966 | u64 object_set_size = object_size * stripe_count; | |
967 | u64 nearly, t; | |
968 | ||
969 | /* round offset up to next period boundary */ | |
970 | nearly = offset + object_set_size - 1; | |
971 | t = nearly; | |
972 | nearly -= do_div(t, object_set_size); | |
ad7a60de | 973 | |
ad7a60de LW |
974 | while (length && offset < nearly) { |
975 | loff_t size = length; | |
976 | ret = ceph_zero_partial_object(inode, offset, &size); | |
977 | if (ret < 0) | |
978 | return ret; | |
979 | offset += size; | |
980 | length -= size; | |
981 | } | |
982 | while (length >= object_set_size) { | |
983 | int i; | |
984 | loff_t pos = offset; | |
985 | for (i = 0; i < stripe_count; ++i) { | |
986 | ret = ceph_zero_partial_object(inode, pos, NULL); | |
987 | if (ret < 0) | |
988 | return ret; | |
989 | pos += stripe_unit; | |
990 | } | |
991 | offset += object_set_size; | |
992 | length -= object_set_size; | |
993 | } | |
994 | while (length) { | |
995 | loff_t size = length; | |
996 | ret = ceph_zero_partial_object(inode, offset, &size); | |
997 | if (ret < 0) | |
998 | return ret; | |
999 | offset += size; | |
1000 | length -= size; | |
1001 | } | |
1002 | return ret; | |
1003 | } | |
1004 | ||
1005 | static long ceph_fallocate(struct file *file, int mode, | |
1006 | loff_t offset, loff_t length) | |
1007 | { | |
1008 | struct ceph_file_info *fi = file->private_data; | |
1009 | struct inode *inode = file->f_dentry->d_inode; | |
1010 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1011 | struct ceph_osd_client *osdc = | |
1012 | &ceph_inode_to_client(inode)->client->osdc; | |
1013 | int want, got = 0; | |
1014 | int dirty; | |
1015 | int ret = 0; | |
1016 | loff_t endoff = 0; | |
1017 | loff_t size; | |
1018 | ||
1019 | if (!S_ISREG(inode->i_mode)) | |
1020 | return -EOPNOTSUPP; | |
1021 | ||
1022 | if (IS_SWAPFILE(inode)) | |
1023 | return -ETXTBSY; | |
1024 | ||
1025 | mutex_lock(&inode->i_mutex); | |
1026 | ||
1027 | if (ceph_snap(inode) != CEPH_NOSNAP) { | |
1028 | ret = -EROFS; | |
1029 | goto unlock; | |
1030 | } | |
1031 | ||
1032 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) && | |
1033 | !(mode & FALLOC_FL_PUNCH_HOLE)) { | |
1034 | ret = -ENOSPC; | |
1035 | goto unlock; | |
1036 | } | |
1037 | ||
1038 | size = i_size_read(inode); | |
1039 | if (!(mode & FALLOC_FL_KEEP_SIZE)) | |
1040 | endoff = offset + length; | |
1041 | ||
1042 | if (fi->fmode & CEPH_FILE_MODE_LAZY) | |
1043 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1044 | else | |
1045 | want = CEPH_CAP_FILE_BUFFER; | |
1046 | ||
1047 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, endoff); | |
1048 | if (ret < 0) | |
1049 | goto unlock; | |
1050 | ||
1051 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
1052 | if (offset < size) | |
1053 | ceph_zero_pagecache_range(inode, offset, length); | |
1054 | ret = ceph_zero_objects(inode, offset, length); | |
1055 | } else if (endoff > size) { | |
1056 | truncate_pagecache_range(inode, size, -1); | |
1057 | if (ceph_inode_set_size(inode, endoff)) | |
1058 | ceph_check_caps(ceph_inode(inode), | |
1059 | CHECK_CAPS_AUTHONLY, NULL); | |
1060 | } | |
1061 | ||
1062 | if (!ret) { | |
1063 | spin_lock(&ci->i_ceph_lock); | |
1064 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR); | |
1065 | spin_unlock(&ci->i_ceph_lock); | |
1066 | if (dirty) | |
1067 | __mark_inode_dirty(inode, dirty); | |
1068 | } | |
1069 | ||
1070 | ceph_put_cap_refs(ci, got); | |
1071 | unlock: | |
1072 | mutex_unlock(&inode->i_mutex); | |
1073 | return ret; | |
1074 | } | |
1075 | ||
124e68e7 SW |
1076 | const struct file_operations ceph_file_fops = { |
1077 | .open = ceph_open, | |
1078 | .release = ceph_release, | |
1079 | .llseek = ceph_llseek, | |
1080 | .read = do_sync_read, | |
1081 | .write = do_sync_write, | |
1082 | .aio_read = ceph_aio_read, | |
1083 | .aio_write = ceph_aio_write, | |
1084 | .mmap = ceph_mmap, | |
1085 | .fsync = ceph_fsync, | |
40819f6f GF |
1086 | .lock = ceph_lock, |
1087 | .flock = ceph_flock, | |
124e68e7 SW |
1088 | .splice_read = generic_file_splice_read, |
1089 | .splice_write = generic_file_splice_write, | |
1090 | .unlocked_ioctl = ceph_ioctl, | |
1091 | .compat_ioctl = ceph_ioctl, | |
ad7a60de | 1092 | .fallocate = ceph_fallocate, |
124e68e7 SW |
1093 | }; |
1094 |