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" | |
99ccbd22 | 15 | #include "cache.h" |
124e68e7 SW |
16 | |
17 | /* | |
18 | * Ceph file operations | |
19 | * | |
20 | * Implement basic open/close functionality, and implement | |
21 | * read/write. | |
22 | * | |
23 | * We implement three modes of file I/O: | |
24 | * - buffered uses the generic_file_aio_{read,write} helpers | |
25 | * | |
26 | * - synchronous is used when there is multi-client read/write | |
27 | * sharing, avoids the page cache, and synchronously waits for an | |
28 | * ack from the OSD. | |
29 | * | |
30 | * - direct io takes the variant of the sync path that references | |
31 | * user pages directly. | |
32 | * | |
33 | * fsync() flushes and waits on dirty pages, but just queues metadata | |
34 | * for writeback: since the MDS can recover size and mtime there is no | |
35 | * need to wait for MDS acknowledgement. | |
36 | */ | |
37 | ||
38 | ||
39 | /* | |
40 | * Prepare an open request. Preallocate ceph_cap to avoid an | |
41 | * inopportune ENOMEM later. | |
42 | */ | |
43 | static struct ceph_mds_request * | |
44 | prepare_open_request(struct super_block *sb, int flags, int create_mode) | |
45 | { | |
3d14c5d2 YS |
46 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); |
47 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
48 | struct ceph_mds_request *req; |
49 | int want_auth = USE_ANY_MDS; | |
50 | int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN; | |
51 | ||
52 | if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC)) | |
53 | want_auth = USE_AUTH_MDS; | |
54 | ||
55 | req = ceph_mdsc_create_request(mdsc, op, want_auth); | |
56 | if (IS_ERR(req)) | |
57 | goto out; | |
58 | req->r_fmode = ceph_flags_to_mode(flags); | |
59 | req->r_args.open.flags = cpu_to_le32(flags); | |
60 | req->r_args.open.mode = cpu_to_le32(create_mode); | |
124e68e7 SW |
61 | out: |
62 | return req; | |
63 | } | |
64 | ||
65 | /* | |
66 | * initialize private struct file data. | |
67 | * if we fail, clean up by dropping fmode reference on the ceph_inode | |
68 | */ | |
69 | static int ceph_init_file(struct inode *inode, struct file *file, int fmode) | |
70 | { | |
71 | struct ceph_file_info *cf; | |
72 | int ret = 0; | |
99ccbd22 MT |
73 | struct ceph_inode_info *ci = ceph_inode(inode); |
74 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); | |
75 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
76 | |
77 | switch (inode->i_mode & S_IFMT) { | |
78 | case S_IFREG: | |
99ccbd22 MT |
79 | /* First file open request creates the cookie, we want to keep |
80 | * this cookie around for the filetime of the inode as not to | |
81 | * have to worry about fscache register / revoke / operation | |
82 | * races. | |
83 | * | |
84 | * Also, if we know the operation is going to invalidate data | |
85 | * (non readonly) just nuke the cache right away. | |
86 | */ | |
87 | ceph_fscache_register_inode_cookie(mdsc->fsc, ci); | |
88 | if ((fmode & CEPH_FILE_MODE_WR)) | |
89 | ceph_fscache_invalidate(inode); | |
124e68e7 SW |
90 | case S_IFDIR: |
91 | dout("init_file %p %p 0%o (regular)\n", inode, file, | |
92 | inode->i_mode); | |
93 | cf = kmem_cache_alloc(ceph_file_cachep, GFP_NOFS | __GFP_ZERO); | |
94 | if (cf == NULL) { | |
95 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
96 | return -ENOMEM; | |
97 | } | |
98 | cf->fmode = fmode; | |
99 | cf->next_offset = 2; | |
100 | file->private_data = cf; | |
101 | BUG_ON(inode->i_fop->release != ceph_release); | |
102 | break; | |
103 | ||
104 | case S_IFLNK: | |
105 | dout("init_file %p %p 0%o (symlink)\n", inode, file, | |
106 | inode->i_mode); | |
107 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
108 | break; | |
109 | ||
110 | default: | |
111 | dout("init_file %p %p 0%o (special)\n", inode, file, | |
112 | inode->i_mode); | |
113 | /* | |
114 | * we need to drop the open ref now, since we don't | |
115 | * have .release set to ceph_release. | |
116 | */ | |
117 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
118 | BUG_ON(inode->i_fop->release == ceph_release); | |
119 | ||
120 | /* call the proper open fop */ | |
121 | ret = inode->i_fop->open(inode, file); | |
122 | } | |
123 | return ret; | |
124 | } | |
125 | ||
126 | /* | |
124e68e7 SW |
127 | * If we already have the requisite capabilities, we can satisfy |
128 | * the open request locally (no need to request new caps from the | |
129 | * MDS). We do, however, need to inform the MDS (asynchronously) | |
130 | * if our wanted caps set expands. | |
131 | */ | |
132 | int ceph_open(struct inode *inode, struct file *file) | |
133 | { | |
134 | struct ceph_inode_info *ci = ceph_inode(inode); | |
3d14c5d2 YS |
135 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
136 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
137 | struct ceph_mds_request *req; |
138 | struct ceph_file_info *cf = file->private_data; | |
5f21c96d | 139 | struct inode *parent_inode = NULL; |
124e68e7 SW |
140 | int err; |
141 | int flags, fmode, wanted; | |
142 | ||
143 | if (cf) { | |
144 | dout("open file %p is already opened\n", file); | |
145 | return 0; | |
146 | } | |
147 | ||
148 | /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */ | |
149 | flags = file->f_flags & ~(O_CREAT|O_EXCL); | |
150 | if (S_ISDIR(inode->i_mode)) | |
151 | flags = O_DIRECTORY; /* mds likes to know */ | |
152 | ||
153 | dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode, | |
154 | ceph_vinop(inode), file, flags, file->f_flags); | |
155 | fmode = ceph_flags_to_mode(flags); | |
156 | wanted = ceph_caps_for_mode(fmode); | |
157 | ||
158 | /* snapped files are read-only */ | |
159 | if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE)) | |
160 | return -EROFS; | |
161 | ||
162 | /* trivially open snapdir */ | |
163 | if (ceph_snap(inode) == CEPH_SNAPDIR) { | |
be655596 | 164 | spin_lock(&ci->i_ceph_lock); |
124e68e7 | 165 | __ceph_get_fmode(ci, fmode); |
be655596 | 166 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
167 | return ceph_init_file(inode, file, fmode); |
168 | } | |
169 | ||
170 | /* | |
7421ab80 SW |
171 | * No need to block if we have caps on the auth MDS (for |
172 | * write) or any MDS (for read). Update wanted set | |
124e68e7 SW |
173 | * asynchronously. |
174 | */ | |
be655596 | 175 | spin_lock(&ci->i_ceph_lock); |
7421ab80 SW |
176 | if (__ceph_is_any_real_caps(ci) && |
177 | (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) { | |
124e68e7 SW |
178 | int mds_wanted = __ceph_caps_mds_wanted(ci); |
179 | int issued = __ceph_caps_issued(ci, NULL); | |
180 | ||
181 | dout("open %p fmode %d want %s issued %s using existing\n", | |
182 | inode, fmode, ceph_cap_string(wanted), | |
183 | ceph_cap_string(issued)); | |
184 | __ceph_get_fmode(ci, fmode); | |
be655596 | 185 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
186 | |
187 | /* adjust wanted? */ | |
188 | if ((issued & wanted) != wanted && | |
189 | (mds_wanted & wanted) != wanted && | |
190 | ceph_snap(inode) != CEPH_SNAPDIR) | |
191 | ceph_check_caps(ci, 0, NULL); | |
192 | ||
193 | return ceph_init_file(inode, file, fmode); | |
194 | } else if (ceph_snap(inode) != CEPH_NOSNAP && | |
195 | (ci->i_snap_caps & wanted) == wanted) { | |
196 | __ceph_get_fmode(ci, fmode); | |
be655596 | 197 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
198 | return ceph_init_file(inode, file, fmode); |
199 | } | |
99ccbd22 | 200 | |
be655596 | 201 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
202 | |
203 | dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted)); | |
204 | req = prepare_open_request(inode->i_sb, flags, 0); | |
205 | if (IS_ERR(req)) { | |
206 | err = PTR_ERR(req); | |
207 | goto out; | |
208 | } | |
70b666c3 SW |
209 | req->r_inode = inode; |
210 | ihold(inode); | |
99ccbd22 | 211 | |
124e68e7 | 212 | req->r_num_caps = 1; |
5f21c96d SW |
213 | if (flags & (O_CREAT|O_TRUNC)) |
214 | parent_inode = ceph_get_dentry_parent_inode(file->f_dentry); | |
124e68e7 | 215 | err = ceph_mdsc_do_request(mdsc, parent_inode, req); |
5f21c96d | 216 | iput(parent_inode); |
124e68e7 SW |
217 | if (!err) |
218 | err = ceph_init_file(inode, file, req->r_fmode); | |
219 | ceph_mdsc_put_request(req); | |
220 | dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode)); | |
221 | out: | |
222 | return err; | |
223 | } | |
224 | ||
225 | ||
226 | /* | |
5ef50c3b SW |
227 | * Do a lookup + open with a single request. If we get a non-existent |
228 | * file or symlink, return 1 so the VFS can retry. | |
124e68e7 | 229 | */ |
5ef50c3b | 230 | int ceph_atomic_open(struct inode *dir, struct dentry *dentry, |
30d90494 | 231 | struct file *file, unsigned flags, umode_t mode, |
d9585277 | 232 | int *opened) |
124e68e7 | 233 | { |
3d14c5d2 YS |
234 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
235 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 | 236 | struct ceph_mds_request *req; |
5ef50c3b | 237 | struct dentry *dn; |
124e68e7 | 238 | int err; |
124e68e7 | 239 | |
5ef50c3b SW |
240 | dout("atomic_open %p dentry %p '%.*s' %s flags %d mode 0%o\n", |
241 | dir, dentry, dentry->d_name.len, dentry->d_name.name, | |
242 | d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode); | |
243 | ||
244 | if (dentry->d_name.len > NAME_MAX) | |
245 | return -ENAMETOOLONG; | |
246 | ||
247 | err = ceph_init_dentry(dentry); | |
248 | if (err < 0) | |
249 | return err; | |
124e68e7 SW |
250 | |
251 | /* do the open */ | |
252 | req = prepare_open_request(dir->i_sb, flags, mode); | |
253 | if (IS_ERR(req)) | |
d9585277 | 254 | return PTR_ERR(req); |
124e68e7 SW |
255 | req->r_dentry = dget(dentry); |
256 | req->r_num_caps = 2; | |
257 | if (flags & O_CREAT) { | |
258 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
259 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
260 | } | |
261 | req->r_locked_dir = dir; /* caller holds dir->i_mutex */ | |
acda7657 SW |
262 | err = ceph_mdsc_do_request(mdsc, |
263 | (flags & (O_CREAT|O_TRUNC)) ? dir : NULL, | |
264 | req); | |
79aec984 SL |
265 | if (err) |
266 | goto out_err; | |
267 | ||
468640e3 | 268 | err = ceph_handle_snapdir(req, dentry, err); |
5ef50c3b | 269 | if (err == 0 && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry) |
124e68e7 | 270 | err = ceph_handle_notrace_create(dir, dentry); |
2d83bde9 | 271 | |
5ef50c3b SW |
272 | if (d_unhashed(dentry)) { |
273 | dn = ceph_finish_lookup(req, dentry, err); | |
274 | if (IS_ERR(dn)) | |
275 | err = PTR_ERR(dn); | |
276 | } else { | |
277 | /* we were given a hashed negative dentry */ | |
278 | dn = NULL; | |
279 | } | |
280 | if (err) | |
281 | goto out_err; | |
282 | if (dn || dentry->d_inode == NULL || S_ISLNK(dentry->d_inode->i_mode)) { | |
283 | /* make vfs retry on splice, ENOENT, or symlink */ | |
284 | dout("atomic_open finish_no_open on dn %p\n", dn); | |
285 | err = finish_no_open(file, dn); | |
286 | } else { | |
287 | dout("atomic_open finish_open on dn %p\n", dn); | |
6e8575fa SL |
288 | if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) { |
289 | *opened |= FILE_CREATED; | |
290 | } | |
5ef50c3b SW |
291 | err = finish_open(file, dentry, ceph_open, opened); |
292 | } | |
2d83bde9 | 293 | |
5ef50c3b SW |
294 | out_err: |
295 | ceph_mdsc_put_request(req); | |
296 | dout("atomic_open result=%d\n", err); | |
d9585277 | 297 | return err; |
124e68e7 SW |
298 | } |
299 | ||
300 | int ceph_release(struct inode *inode, struct file *file) | |
301 | { | |
302 | struct ceph_inode_info *ci = ceph_inode(inode); | |
303 | struct ceph_file_info *cf = file->private_data; | |
304 | ||
305 | dout("release inode %p file %p\n", inode, file); | |
306 | ceph_put_fmode(ci, cf->fmode); | |
307 | if (cf->last_readdir) | |
308 | ceph_mdsc_put_request(cf->last_readdir); | |
309 | kfree(cf->last_name); | |
310 | kfree(cf->dir_info); | |
311 | dput(cf->dentry); | |
312 | kmem_cache_free(ceph_file_cachep, cf); | |
195d3ce2 SW |
313 | |
314 | /* wake up anyone waiting for caps on this inode */ | |
03066f23 | 315 | wake_up_all(&ci->i_cap_wq); |
124e68e7 SW |
316 | return 0; |
317 | } | |
318 | ||
124e68e7 SW |
319 | /* |
320 | * Read a range of bytes striped over one or more objects. Iterate over | |
321 | * objects we stripe over. (That's not atomic, but good enough for now.) | |
322 | * | |
323 | * If we get a short result from the OSD, check against i_size; we need to | |
324 | * only return a short read to the caller if we hit EOF. | |
325 | */ | |
326 | static int striped_read(struct inode *inode, | |
327 | u64 off, u64 len, | |
6a026589 | 328 | struct page **pages, int num_pages, |
c3cd6283 | 329 | int *checkeof, bool o_direct, |
ab226e21 | 330 | unsigned long buf_align) |
124e68e7 | 331 | { |
3d14c5d2 | 332 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
124e68e7 | 333 | struct ceph_inode_info *ci = ceph_inode(inode); |
688bac46 | 334 | u64 pos, this_len, left; |
b7495fc2 | 335 | int io_align, page_align; |
688bac46 | 336 | int pages_left; |
124e68e7 SW |
337 | int read; |
338 | struct page **page_pos; | |
339 | int ret; | |
340 | bool hit_stripe, was_short; | |
341 | ||
342 | /* | |
343 | * we may need to do multiple reads. not atomic, unfortunately. | |
344 | */ | |
345 | pos = off; | |
346 | left = len; | |
347 | page_pos = pages; | |
348 | pages_left = num_pages; | |
349 | read = 0; | |
b7495fc2 | 350 | io_align = off & ~PAGE_MASK; |
124e68e7 SW |
351 | |
352 | more: | |
c3cd6283 | 353 | if (o_direct) |
ab226e21 | 354 | page_align = (pos - io_align + buf_align) & ~PAGE_MASK; |
b7495fc2 SW |
355 | else |
356 | page_align = pos & ~PAGE_MASK; | |
124e68e7 | 357 | this_len = left; |
3d14c5d2 | 358 | ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode), |
124e68e7 SW |
359 | &ci->i_layout, pos, &this_len, |
360 | ci->i_truncate_seq, | |
361 | ci->i_truncate_size, | |
b7495fc2 | 362 | page_pos, pages_left, page_align); |
124e68e7 SW |
363 | if (ret == -ENOENT) |
364 | ret = 0; | |
0e98728f SW |
365 | hit_stripe = this_len < left; |
366 | was_short = ret >= 0 && ret < this_len; | |
688bac46 | 367 | dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read, |
124e68e7 SW |
368 | ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : ""); |
369 | ||
02ae66d8 | 370 | if (ret >= 0) { |
371 | int didpages; | |
372 | if (was_short && (pos + ret < inode->i_size)) { | |
373 | u64 tmp = min(this_len - ret, | |
374 | inode->i_size - pos - ret); | |
375 | dout(" zero gap %llu to %llu\n", | |
376 | pos + ret, pos + ret + tmp); | |
377 | ceph_zero_page_vector_range(page_align + read + ret, | |
378 | tmp, pages); | |
379 | ret += tmp; | |
124e68e7 | 380 | } |
02ae66d8 | 381 | |
382 | didpages = (page_align + ret) >> PAGE_CACHE_SHIFT; | |
124e68e7 SW |
383 | pos += ret; |
384 | read = pos - off; | |
385 | left -= ret; | |
386 | page_pos += didpages; | |
387 | pages_left -= didpages; | |
388 | ||
02ae66d8 | 389 | /* hit stripe and need continue*/ |
390 | if (left && hit_stripe && pos < inode->i_size) | |
124e68e7 SW |
391 | goto more; |
392 | } | |
393 | ||
ee7289bf | 394 | if (read > 0) { |
02ae66d8 | 395 | ret = read; |
c3cd6283 SW |
396 | /* did we bounce off eof? */ |
397 | if (pos + left > inode->i_size) | |
398 | *checkeof = 1; | |
124e68e7 SW |
399 | } |
400 | ||
124e68e7 SW |
401 | dout("striped_read returns %d\n", ret); |
402 | return ret; | |
403 | } | |
404 | ||
405 | /* | |
406 | * Completely synchronous read and write methods. Direct from __user | |
407 | * buffer to osd, or directly to user pages (if O_DIRECT). | |
408 | * | |
409 | * If the read spans object boundary, just do multiple reads. | |
410 | */ | |
8eb4efb0 | 411 | static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i, |
412 | int *checkeof) | |
124e68e7 | 413 | { |
8eb4efb0 | 414 | struct file *file = iocb->ki_filp; |
496ad9aa | 415 | struct inode *inode = file_inode(file); |
124e68e7 | 416 | struct page **pages; |
8eb4efb0 | 417 | u64 off = iocb->ki_pos; |
ab226e21 | 418 | int num_pages, ret; |
8eb4efb0 | 419 | size_t len = i->count; |
124e68e7 | 420 | |
8eb4efb0 | 421 | dout("sync_read on file %p %llu~%u %s\n", file, off, |
422 | (unsigned)len, | |
124e68e7 | 423 | (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); |
e98b6fed SW |
424 | /* |
425 | * flush any page cache pages in this range. this | |
426 | * will make concurrent normal and sync io slow, | |
427 | * but it will at least behave sensibly when they are | |
428 | * in sequence. | |
429 | */ | |
8eb4efb0 | 430 | ret = filemap_write_and_wait_range(inode->i_mapping, off, |
431 | off + len); | |
29065a51 | 432 | if (ret < 0) |
8eb4efb0 | 433 | return ret; |
29065a51 | 434 | |
8eb4efb0 | 435 | if (file->f_flags & O_DIRECT) { |
436 | while (iov_iter_count(i)) { | |
437 | void __user *data = i->iov[0].iov_base + i->iov_offset; | |
438 | size_t len = i->iov[0].iov_len - i->iov_offset; | |
439 | ||
440 | num_pages = calc_pages_for((unsigned long)data, len); | |
441 | pages = ceph_get_direct_page_vector(data, | |
442 | num_pages, true); | |
443 | if (IS_ERR(pages)) | |
444 | return PTR_ERR(pages); | |
445 | ||
446 | ret = striped_read(inode, off, len, | |
447 | pages, num_pages, checkeof, | |
448 | 1, (unsigned long)data & ~PAGE_MASK); | |
449 | ceph_put_page_vector(pages, num_pages, true); | |
450 | ||
451 | if (ret <= 0) | |
452 | break; | |
453 | off += ret; | |
454 | iov_iter_advance(i, ret); | |
455 | if (ret < len) | |
456 | break; | |
457 | } | |
458 | } else { | |
459 | num_pages = calc_pages_for(off, len); | |
460 | pages = ceph_alloc_page_vector(num_pages, GFP_NOFS); | |
461 | if (IS_ERR(pages)) | |
462 | return PTR_ERR(pages); | |
463 | ret = striped_read(inode, off, len, pages, | |
464 | num_pages, checkeof, 0, 0); | |
465 | if (ret > 0) { | |
466 | int l, k = 0; | |
467 | size_t left = len = ret; | |
468 | ||
469 | while (left) { | |
470 | void __user *data = i->iov[0].iov_base | |
471 | + i->iov_offset; | |
472 | l = min(i->iov[0].iov_len - i->iov_offset, | |
473 | left); | |
474 | ||
475 | ret = ceph_copy_page_vector_to_user(&pages[k], | |
476 | data, off, | |
477 | l); | |
478 | if (ret > 0) { | |
479 | iov_iter_advance(i, ret); | |
480 | left -= ret; | |
481 | off += ret; | |
482 | k = calc_pages_for(iocb->ki_pos, | |
483 | len - left + 1) - 1; | |
484 | BUG_ON(k >= num_pages && left); | |
485 | } else | |
486 | break; | |
487 | } | |
488 | } | |
489 | ceph_release_page_vector(pages, num_pages); | |
490 | } | |
124e68e7 | 491 | |
8eb4efb0 | 492 | if (off > iocb->ki_pos) { |
493 | ret = off - iocb->ki_pos; | |
494 | iocb->ki_pos = off; | |
495 | } | |
124e68e7 | 496 | |
124e68e7 SW |
497 | dout("sync_read result %d\n", ret); |
498 | return ret; | |
499 | } | |
500 | ||
501 | /* | |
26be8808 AE |
502 | * Write commit request unsafe callback, called to tell us when a |
503 | * request is unsafe (that is, in flight--has been handed to the | |
504 | * messenger to send to its target osd). It is called again when | |
505 | * we've received a response message indicating the request is | |
506 | * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request | |
507 | * is completed early (and unsuccessfully) due to a timeout or | |
508 | * interrupt. | |
509 | * | |
510 | * This is used if we requested both an ACK and ONDISK commit reply | |
511 | * from the OSD. | |
124e68e7 | 512 | */ |
26be8808 | 513 | static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe) |
124e68e7 SW |
514 | { |
515 | struct ceph_inode_info *ci = ceph_inode(req->r_inode); | |
516 | ||
26be8808 AE |
517 | dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid, |
518 | unsafe ? "un" : ""); | |
519 | if (unsafe) { | |
520 | ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR); | |
521 | spin_lock(&ci->i_unsafe_lock); | |
522 | list_add_tail(&req->r_unsafe_item, | |
523 | &ci->i_unsafe_writes); | |
524 | spin_unlock(&ci->i_unsafe_lock); | |
525 | } else { | |
526 | spin_lock(&ci->i_unsafe_lock); | |
527 | list_del_init(&req->r_unsafe_item); | |
528 | spin_unlock(&ci->i_unsafe_lock); | |
529 | ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR); | |
530 | } | |
124e68e7 SW |
531 | } |
532 | ||
e8344e66 | 533 | |
124e68e7 | 534 | /* |
e8344e66 | 535 | * Synchronous write, straight from __user pointer or user pages. |
124e68e7 SW |
536 | * |
537 | * If write spans object boundary, just do multiple writes. (For a | |
538 | * correct atomic write, we should e.g. take write locks on all | |
539 | * objects, rollback on failure, etc.) | |
540 | */ | |
e8344e66 | 541 | static ssize_t |
542 | ceph_sync_direct_write(struct kiocb *iocb, const struct iovec *iov, | |
543 | unsigned long nr_segs, size_t count) | |
124e68e7 | 544 | { |
e8344e66 | 545 | struct file *file = iocb->ki_filp; |
496ad9aa | 546 | struct inode *inode = file_inode(file); |
124e68e7 | 547 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 | 548 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
acead002 AE |
549 | struct ceph_snap_context *snapc; |
550 | struct ceph_vino vino; | |
124e68e7 SW |
551 | struct ceph_osd_request *req; |
552 | struct page **pages; | |
553 | int num_pages; | |
124e68e7 SW |
554 | int written = 0; |
555 | int flags; | |
124e68e7 | 556 | int check_caps = 0; |
e8344e66 | 557 | int page_align; |
124e68e7 SW |
558 | int ret; |
559 | struct timespec mtime = CURRENT_TIME; | |
e8344e66 | 560 | loff_t pos = iocb->ki_pos; |
561 | struct iov_iter i; | |
124e68e7 | 562 | |
496ad9aa | 563 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) |
124e68e7 SW |
564 | return -EROFS; |
565 | ||
e8344e66 | 566 | dout("sync_direct_write on file %p %lld~%u\n", file, pos, |
567 | (unsigned)count); | |
124e68e7 | 568 | |
e8344e66 | 569 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); |
29065a51 YS |
570 | if (ret < 0) |
571 | return ret; | |
572 | ||
573 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
574 | pos >> PAGE_CACHE_SHIFT, | |
e8344e66 | 575 | (pos + count) >> PAGE_CACHE_SHIFT); |
29065a51 YS |
576 | if (ret < 0) |
577 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
578 | ||
124e68e7 SW |
579 | flags = CEPH_OSD_FLAG_ORDERSNAP | |
580 | CEPH_OSD_FLAG_ONDISK | | |
581 | CEPH_OSD_FLAG_WRITE; | |
124e68e7 | 582 | |
e8344e66 | 583 | iov_iter_init(&i, iov, nr_segs, count, 0); |
584 | ||
585 | while (iov_iter_count(&i) > 0) { | |
586 | void __user *data = i.iov->iov_base + i.iov_offset; | |
587 | u64 len = i.iov->iov_len - i.iov_offset; | |
588 | ||
589 | page_align = (unsigned long)data & ~PAGE_MASK; | |
590 | ||
591 | snapc = ci->i_snap_realm->cached_context; | |
592 | vino = ceph_vino(inode); | |
593 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
594 | vino, pos, &len, | |
595 | 2,/*include a 'startsync' command*/ | |
596 | CEPH_OSD_OP_WRITE, flags, snapc, | |
597 | ci->i_truncate_seq, | |
598 | ci->i_truncate_size, | |
599 | false); | |
600 | if (IS_ERR(req)) { | |
601 | ret = PTR_ERR(req); | |
602 | goto out; | |
603 | } | |
124e68e7 | 604 | |
e8344e66 | 605 | num_pages = calc_pages_for(page_align, len); |
b6aa5901 | 606 | pages = ceph_get_direct_page_vector(data, num_pages, false); |
124e68e7 SW |
607 | if (IS_ERR(pages)) { |
608 | ret = PTR_ERR(pages); | |
609 | goto out; | |
610 | } | |
611 | ||
612 | /* | |
613 | * throw out any page cache pages in this range. this | |
614 | * may block. | |
615 | */ | |
213c99ee | 616 | truncate_inode_pages_range(inode->i_mapping, pos, |
e8344e66 | 617 | (pos+len) | (PAGE_CACHE_SIZE-1)); |
618 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align, | |
619 | false, false); | |
620 | ||
621 | /* BUG_ON(vino.snap != CEPH_NOSNAP); */ | |
622 | ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime); | |
623 | ||
624 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); | |
625 | if (!ret) | |
626 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
627 | ||
628 | ceph_put_page_vector(pages, num_pages, false); | |
629 | ||
630 | out: | |
631 | ceph_osdc_put_request(req); | |
632 | if (ret == 0) { | |
633 | pos += len; | |
634 | written += len; | |
635 | iov_iter_advance(&i, (size_t)len); | |
636 | ||
637 | if (pos > i_size_read(inode)) { | |
638 | check_caps = ceph_inode_set_size(inode, pos); | |
639 | if (check_caps) | |
640 | ceph_check_caps(ceph_inode(inode), | |
641 | CHECK_CAPS_AUTHONLY, | |
642 | NULL); | |
643 | } | |
644 | } else | |
645 | break; | |
646 | } | |
647 | ||
648 | if (ret != -EOLDSNAPC && written > 0) { | |
649 | iocb->ki_pos = pos; | |
650 | ret = written; | |
651 | } | |
652 | return ret; | |
653 | } | |
654 | ||
655 | ||
656 | /* | |
657 | * Synchronous write, straight from __user pointer or user pages. | |
658 | * | |
659 | * If write spans object boundary, just do multiple writes. (For a | |
660 | * correct atomic write, we should e.g. take write locks on all | |
661 | * objects, rollback on failure, etc.) | |
662 | */ | |
663 | static ssize_t ceph_sync_write(struct kiocb *iocb, const struct iovec *iov, | |
664 | unsigned long nr_segs, size_t count) | |
665 | { | |
666 | struct file *file = iocb->ki_filp; | |
667 | struct inode *inode = file_inode(file); | |
668 | struct ceph_inode_info *ci = ceph_inode(inode); | |
669 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
670 | struct ceph_snap_context *snapc; | |
671 | struct ceph_vino vino; | |
672 | struct ceph_osd_request *req; | |
673 | struct page **pages; | |
674 | u64 len; | |
675 | int num_pages; | |
676 | int written = 0; | |
677 | int flags; | |
678 | int check_caps = 0; | |
679 | int ret; | |
680 | struct timespec mtime = CURRENT_TIME; | |
681 | loff_t pos = iocb->ki_pos; | |
682 | struct iov_iter i; | |
683 | ||
684 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) | |
685 | return -EROFS; | |
686 | ||
687 | dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count); | |
688 | ||
689 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); | |
690 | if (ret < 0) | |
691 | return ret; | |
692 | ||
693 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
694 | pos >> PAGE_CACHE_SHIFT, | |
695 | (pos + count) >> PAGE_CACHE_SHIFT); | |
696 | if (ret < 0) | |
697 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
698 | ||
699 | flags = CEPH_OSD_FLAG_ORDERSNAP | | |
700 | CEPH_OSD_FLAG_ONDISK | | |
701 | CEPH_OSD_FLAG_WRITE | | |
702 | CEPH_OSD_FLAG_ACK; | |
703 | ||
704 | iov_iter_init(&i, iov, nr_segs, count, 0); | |
705 | ||
706 | while ((len = iov_iter_count(&i)) > 0) { | |
707 | size_t left; | |
708 | int n; | |
709 | ||
710 | snapc = ci->i_snap_realm->cached_context; | |
711 | vino = ceph_vino(inode); | |
712 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
713 | vino, pos, &len, 1, | |
714 | CEPH_OSD_OP_WRITE, flags, snapc, | |
715 | ci->i_truncate_seq, | |
716 | ci->i_truncate_size, | |
717 | false); | |
718 | if (IS_ERR(req)) { | |
719 | ret = PTR_ERR(req); | |
720 | goto out; | |
721 | } | |
722 | ||
723 | /* | |
724 | * write from beginning of first page, | |
725 | * regardless of io alignment | |
726 | */ | |
727 | num_pages = (len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
728 | ||
34d23762 | 729 | pages = ceph_alloc_page_vector(num_pages, GFP_NOFS); |
124e68e7 SW |
730 | if (IS_ERR(pages)) { |
731 | ret = PTR_ERR(pages); | |
732 | goto out; | |
733 | } | |
e8344e66 | 734 | |
735 | left = len; | |
736 | for (n = 0; n < num_pages; n++) { | |
737 | size_t plen = min(left, PAGE_SIZE); | |
738 | ret = iov_iter_copy_from_user(pages[n], &i, 0, plen); | |
739 | if (ret != plen) { | |
740 | ret = -EFAULT; | |
741 | break; | |
742 | } | |
743 | left -= ret; | |
744 | iov_iter_advance(&i, ret); | |
745 | } | |
746 | ||
124e68e7 SW |
747 | if (ret < 0) { |
748 | ceph_release_page_vector(pages, num_pages); | |
749 | goto out; | |
750 | } | |
751 | ||
e8344e66 | 752 | /* get a second commit callback */ |
753 | req->r_unsafe_callback = ceph_sync_write_unsafe; | |
754 | req->r_inode = inode; | |
124e68e7 | 755 | |
e8344e66 | 756 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, |
757 | false, true); | |
02ee07d3 | 758 | |
e8344e66 | 759 | /* BUG_ON(vino.snap != CEPH_NOSNAP); */ |
760 | ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime); | |
124e68e7 | 761 | |
e8344e66 | 762 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
763 | if (!ret) | |
764 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
124e68e7 SW |
765 | |
766 | out: | |
e8344e66 | 767 | ceph_osdc_put_request(req); |
768 | if (ret == 0) { | |
769 | pos += len; | |
770 | written += len; | |
771 | ||
772 | if (pos > i_size_read(inode)) { | |
773 | check_caps = ceph_inode_set_size(inode, pos); | |
774 | if (check_caps) | |
775 | ceph_check_caps(ceph_inode(inode), | |
776 | CHECK_CAPS_AUTHONLY, | |
777 | NULL); | |
778 | } | |
779 | } else | |
780 | break; | |
781 | } | |
124e68e7 | 782 | |
e8344e66 | 783 | if (ret != -EOLDSNAPC && written > 0) { |
124e68e7 | 784 | ret = written; |
e8344e66 | 785 | iocb->ki_pos = pos; |
124e68e7 SW |
786 | } |
787 | return ret; | |
788 | } | |
789 | ||
790 | /* | |
791 | * Wrap generic_file_aio_read with checks for cap bits on the inode. | |
792 | * Atomically grab references, so that those bits are not released | |
793 | * back to the MDS mid-read. | |
794 | * | |
795 | * Hmm, the sync read case isn't actually async... should it be? | |
796 | */ | |
797 | static ssize_t ceph_aio_read(struct kiocb *iocb, const struct iovec *iov, | |
798 | unsigned long nr_segs, loff_t pos) | |
799 | { | |
800 | struct file *filp = iocb->ki_filp; | |
2962507c | 801 | struct ceph_file_info *fi = filp->private_data; |
8eb4efb0 | 802 | size_t len = iocb->ki_nbytes; |
496ad9aa | 803 | struct inode *inode = file_inode(filp); |
124e68e7 SW |
804 | struct ceph_inode_info *ci = ceph_inode(inode); |
805 | ssize_t ret; | |
2962507c | 806 | int want, got = 0; |
6a026589 | 807 | int checkeof = 0, read = 0; |
124e68e7 | 808 | |
6a026589 | 809 | again: |
8eb4efb0 | 810 | dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n", |
811 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode); | |
812 | ||
2962507c SW |
813 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
814 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
815 | else | |
816 | want = CEPH_CAP_FILE_CACHE; | |
817 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, &got, -1); | |
124e68e7 | 818 | if (ret < 0) |
8eb4efb0 | 819 | return ret; |
124e68e7 | 820 | |
2962507c | 821 | if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 || |
124e68e7 | 822 | (iocb->ki_filp->f_flags & O_DIRECT) || |
8eb4efb0 | 823 | (fi->flags & CEPH_F_SYNC)) { |
824 | struct iov_iter i; | |
825 | ||
826 | dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n", | |
827 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, | |
828 | ceph_cap_string(got)); | |
829 | ||
830 | if (!read) { | |
831 | ret = generic_segment_checks(iov, &nr_segs, | |
832 | &len, VERIFY_WRITE); | |
833 | if (ret) | |
834 | goto out; | |
835 | } | |
836 | ||
837 | iov_iter_init(&i, iov, nr_segs, len, read); | |
838 | ||
124e68e7 | 839 | /* hmm, this isn't really async... */ |
8eb4efb0 | 840 | ret = ceph_sync_read(iocb, &i, &checkeof); |
841 | } else { | |
842 | /* | |
843 | * We can't modify the content of iov, | |
844 | * so we only read from beginning. | |
845 | */ | |
846 | if (read) { | |
847 | iocb->ki_pos = pos; | |
848 | len = iocb->ki_nbytes; | |
849 | read = 0; | |
850 | } | |
851 | dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n", | |
852 | inode, ceph_vinop(inode), pos, (unsigned)len, | |
853 | ceph_cap_string(got)); | |
124e68e7 | 854 | |
8eb4efb0 | 855 | ret = generic_file_aio_read(iocb, iov, nr_segs, pos); |
856 | } | |
124e68e7 SW |
857 | out: |
858 | dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n", | |
859 | inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret); | |
860 | ceph_put_cap_refs(ci, got); | |
6a026589 SW |
861 | |
862 | if (checkeof && ret >= 0) { | |
8eb4efb0 | 863 | int statret = ceph_do_getattr(inode, |
864 | CEPH_STAT_CAP_SIZE); | |
6a026589 SW |
865 | |
866 | /* hit EOF or hole? */ | |
8eb4efb0 | 867 | if (statret == 0 && iocb->ki_pos < inode->i_size && |
868 | ret < len) { | |
869 | dout("sync_read hit hole, ppos %lld < size %lld" | |
870 | ", reading more\n", iocb->ki_pos, | |
871 | inode->i_size); | |
872 | ||
6a026589 | 873 | read += ret; |
6a026589 SW |
874 | len -= ret; |
875 | checkeof = 0; | |
876 | goto again; | |
877 | } | |
878 | } | |
8eb4efb0 | 879 | |
6a026589 SW |
880 | if (ret >= 0) |
881 | ret += read; | |
882 | ||
124e68e7 SW |
883 | return ret; |
884 | } | |
885 | ||
886 | /* | |
887 | * Take cap references to avoid releasing caps to MDS mid-write. | |
888 | * | |
889 | * If we are synchronous, and write with an old snap context, the OSD | |
890 | * may return EOLDSNAPC. In that case, retry the write.. _after_ | |
891 | * dropping our cap refs and allowing the pending snap to logically | |
892 | * complete _before_ this write occurs. | |
893 | * | |
894 | * If we are near ENOSPC, write synchronously. | |
895 | */ | |
896 | static ssize_t ceph_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
897 | unsigned long nr_segs, loff_t pos) | |
898 | { | |
899 | struct file *file = iocb->ki_filp; | |
33caad32 | 900 | struct ceph_file_info *fi = file->private_data; |
496ad9aa | 901 | struct inode *inode = file_inode(file); |
124e68e7 | 902 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 YS |
903 | struct ceph_osd_client *osdc = |
904 | &ceph_sb_to_client(inode->i_sb)->client->osdc; | |
03d254ed YZ |
905 | ssize_t count, written = 0; |
906 | int err, want, got; | |
124e68e7 SW |
907 | |
908 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
909 | return -EROFS; | |
910 | ||
03d254ed | 911 | mutex_lock(&inode->i_mutex); |
03d254ed YZ |
912 | |
913 | err = generic_segment_checks(iov, &nr_segs, &count, VERIFY_READ); | |
914 | if (err) | |
915 | goto out; | |
916 | ||
917 | /* We can write back this queue in page reclaim */ | |
918 | current->backing_dev_info = file->f_mapping->backing_dev_info; | |
919 | ||
920 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
921 | if (err) | |
922 | goto out; | |
923 | ||
924 | if (count == 0) | |
925 | goto out; | |
926 | ||
927 | err = file_remove_suid(file); | |
928 | if (err) | |
929 | goto out; | |
930 | ||
931 | err = file_update_time(file); | |
932 | if (err) | |
933 | goto out; | |
934 | ||
124e68e7 | 935 | retry_snap: |
6070e0c1 | 936 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) { |
03d254ed | 937 | err = -ENOSPC; |
6070e0c1 YZ |
938 | goto out; |
939 | } | |
03d254ed | 940 | |
ac7f29bf | 941 | dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n", |
03d254ed | 942 | inode, ceph_vinop(inode), pos, count, inode->i_size); |
7971bd92 SW |
943 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
944 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
945 | else | |
946 | want = CEPH_CAP_FILE_BUFFER; | |
03d254ed YZ |
947 | got = 0; |
948 | err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, pos + count); | |
949 | if (err < 0) | |
37505d57 | 950 | goto out; |
124e68e7 | 951 | |
ac7f29bf | 952 | dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n", |
03d254ed | 953 | inode, ceph_vinop(inode), pos, count, ceph_cap_string(got)); |
7971bd92 SW |
954 | |
955 | if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 || | |
e8344e66 | 956 | (file->f_flags & O_DIRECT) || (fi->flags & CEPH_F_SYNC)) { |
37505d57 | 957 | mutex_unlock(&inode->i_mutex); |
e8344e66 | 958 | if (file->f_flags & O_DIRECT) |
959 | written = ceph_sync_direct_write(iocb, iov, | |
960 | nr_segs, count); | |
961 | else | |
962 | written = ceph_sync_write(iocb, iov, nr_segs, count); | |
0e5dd45c | 963 | if (written == -EOLDSNAPC) { |
964 | dout("aio_write %p %llx.%llx %llu~%u" | |
965 | "got EOLDSNAPC, retrying\n", | |
966 | inode, ceph_vinop(inode), | |
967 | pos, (unsigned)iov->iov_len); | |
968 | mutex_lock(&inode->i_mutex); | |
0e5dd45c | 969 | goto retry_snap; |
970 | } | |
7971bd92 | 971 | } else { |
b0d7c223 YZ |
972 | /* |
973 | * No need to acquire the i_truncate_mutex. Because | |
974 | * the MDS revokes Fwb caps before sending truncate | |
975 | * message to us. We can't get Fwb cap while there | |
976 | * are pending vmtruncate. So write and vmtruncate | |
977 | * can not run at the same time | |
978 | */ | |
03d254ed YZ |
979 | written = generic_file_buffered_write(iocb, iov, nr_segs, |
980 | pos, &iocb->ki_pos, | |
981 | count, 0); | |
6070e0c1 | 982 | mutex_unlock(&inode->i_mutex); |
7971bd92 | 983 | } |
d8de9ab6 | 984 | |
03d254ed | 985 | if (written >= 0) { |
fca65b4a | 986 | int dirty; |
be655596 | 987 | spin_lock(&ci->i_ceph_lock); |
fca65b4a | 988 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR); |
be655596 | 989 | spin_unlock(&ci->i_ceph_lock); |
fca65b4a SW |
990 | if (dirty) |
991 | __mark_inode_dirty(inode, dirty); | |
124e68e7 | 992 | } |
7971bd92 | 993 | |
124e68e7 | 994 | dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n", |
7971bd92 SW |
995 | inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len, |
996 | ceph_cap_string(got)); | |
124e68e7 | 997 | ceph_put_cap_refs(ci, got); |
7971bd92 | 998 | |
03d254ed | 999 | if (written >= 0 && |
6070e0c1 YZ |
1000 | ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) || |
1001 | ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) { | |
03d254ed | 1002 | err = vfs_fsync_range(file, pos, pos + written - 1, 1); |
6070e0c1 | 1003 | if (err < 0) |
03d254ed | 1004 | written = err; |
6070e0c1 | 1005 | } |
03d254ed | 1006 | |
2f75e9e1 SW |
1007 | goto out_unlocked; |
1008 | ||
03d254ed | 1009 | out: |
2f75e9e1 SW |
1010 | mutex_unlock(&inode->i_mutex); |
1011 | out_unlocked: | |
03d254ed | 1012 | current->backing_dev_info = NULL; |
03d254ed | 1013 | return written ? written : err; |
124e68e7 SW |
1014 | } |
1015 | ||
1016 | /* | |
1017 | * llseek. be sure to verify file size on SEEK_END. | |
1018 | */ | |
965c8e59 | 1019 | static loff_t ceph_llseek(struct file *file, loff_t offset, int whence) |
124e68e7 SW |
1020 | { |
1021 | struct inode *inode = file->f_mapping->host; | |
1022 | int ret; | |
1023 | ||
1024 | mutex_lock(&inode->i_mutex); | |
6a82c47a | 1025 | |
965c8e59 | 1026 | if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) { |
124e68e7 SW |
1027 | ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE); |
1028 | if (ret < 0) { | |
1029 | offset = ret; | |
1030 | goto out; | |
1031 | } | |
06222e49 JB |
1032 | } |
1033 | ||
965c8e59 | 1034 | switch (whence) { |
06222e49 | 1035 | case SEEK_END: |
124e68e7 SW |
1036 | offset += inode->i_size; |
1037 | break; | |
1038 | case SEEK_CUR: | |
1039 | /* | |
1040 | * Here we special-case the lseek(fd, 0, SEEK_CUR) | |
1041 | * position-querying operation. Avoid rewriting the "same" | |
1042 | * f_pos value back to the file because a concurrent read(), | |
1043 | * write() or lseek() might have altered it | |
1044 | */ | |
1045 | if (offset == 0) { | |
1046 | offset = file->f_pos; | |
1047 | goto out; | |
1048 | } | |
1049 | offset += file->f_pos; | |
1050 | break; | |
06222e49 JB |
1051 | case SEEK_DATA: |
1052 | if (offset >= inode->i_size) { | |
1053 | ret = -ENXIO; | |
1054 | goto out; | |
1055 | } | |
1056 | break; | |
1057 | case SEEK_HOLE: | |
1058 | if (offset >= inode->i_size) { | |
1059 | ret = -ENXIO; | |
1060 | goto out; | |
1061 | } | |
1062 | offset = inode->i_size; | |
1063 | break; | |
124e68e7 SW |
1064 | } |
1065 | ||
46a1c2c7 | 1066 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
124e68e7 SW |
1067 | |
1068 | out: | |
1069 | mutex_unlock(&inode->i_mutex); | |
1070 | return offset; | |
1071 | } | |
1072 | ||
ad7a60de LW |
1073 | static inline void ceph_zero_partial_page( |
1074 | struct inode *inode, loff_t offset, unsigned size) | |
1075 | { | |
1076 | struct page *page; | |
1077 | pgoff_t index = offset >> PAGE_CACHE_SHIFT; | |
1078 | ||
1079 | page = find_lock_page(inode->i_mapping, index); | |
1080 | if (page) { | |
1081 | wait_on_page_writeback(page); | |
1082 | zero_user(page, offset & (PAGE_CACHE_SIZE - 1), size); | |
1083 | unlock_page(page); | |
1084 | page_cache_release(page); | |
1085 | } | |
1086 | } | |
1087 | ||
1088 | static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset, | |
1089 | loff_t length) | |
1090 | { | |
1091 | loff_t nearly = round_up(offset, PAGE_CACHE_SIZE); | |
1092 | if (offset < nearly) { | |
1093 | loff_t size = nearly - offset; | |
1094 | if (length < size) | |
1095 | size = length; | |
1096 | ceph_zero_partial_page(inode, offset, size); | |
1097 | offset += size; | |
1098 | length -= size; | |
1099 | } | |
1100 | if (length >= PAGE_CACHE_SIZE) { | |
1101 | loff_t size = round_down(length, PAGE_CACHE_SIZE); | |
1102 | truncate_pagecache_range(inode, offset, offset + size - 1); | |
1103 | offset += size; | |
1104 | length -= size; | |
1105 | } | |
1106 | if (length) | |
1107 | ceph_zero_partial_page(inode, offset, length); | |
1108 | } | |
1109 | ||
1110 | static int ceph_zero_partial_object(struct inode *inode, | |
1111 | loff_t offset, loff_t *length) | |
1112 | { | |
1113 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1114 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
1115 | struct ceph_osd_request *req; | |
1116 | int ret = 0; | |
1117 | loff_t zero = 0; | |
1118 | int op; | |
1119 | ||
1120 | if (!length) { | |
1121 | op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE; | |
1122 | length = &zero; | |
1123 | } else { | |
1124 | op = CEPH_OSD_OP_ZERO; | |
1125 | } | |
1126 | ||
1127 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
1128 | ceph_vino(inode), | |
1129 | offset, length, | |
1130 | 1, op, | |
1131 | CEPH_OSD_FLAG_WRITE | | |
1132 | CEPH_OSD_FLAG_ONDISK, | |
1133 | NULL, 0, 0, false); | |
1134 | if (IS_ERR(req)) { | |
1135 | ret = PTR_ERR(req); | |
1136 | goto out; | |
1137 | } | |
1138 | ||
1139 | ceph_osdc_build_request(req, offset, NULL, ceph_vino(inode).snap, | |
1140 | &inode->i_mtime); | |
1141 | ||
1142 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); | |
1143 | if (!ret) { | |
1144 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
1145 | if (ret == -ENOENT) | |
1146 | ret = 0; | |
1147 | } | |
1148 | ceph_osdc_put_request(req); | |
1149 | ||
1150 | out: | |
1151 | return ret; | |
1152 | } | |
1153 | ||
1154 | static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length) | |
1155 | { | |
1156 | int ret = 0; | |
1157 | struct ceph_inode_info *ci = ceph_inode(inode); | |
b314a90d SW |
1158 | s32 stripe_unit = ceph_file_layout_su(ci->i_layout); |
1159 | s32 stripe_count = ceph_file_layout_stripe_count(ci->i_layout); | |
1160 | s32 object_size = ceph_file_layout_object_size(ci->i_layout); | |
1161 | u64 object_set_size = object_size * stripe_count; | |
1162 | u64 nearly, t; | |
1163 | ||
1164 | /* round offset up to next period boundary */ | |
1165 | nearly = offset + object_set_size - 1; | |
1166 | t = nearly; | |
1167 | nearly -= do_div(t, object_set_size); | |
ad7a60de | 1168 | |
ad7a60de LW |
1169 | while (length && offset < nearly) { |
1170 | loff_t size = length; | |
1171 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1172 | if (ret < 0) | |
1173 | return ret; | |
1174 | offset += size; | |
1175 | length -= size; | |
1176 | } | |
1177 | while (length >= object_set_size) { | |
1178 | int i; | |
1179 | loff_t pos = offset; | |
1180 | for (i = 0; i < stripe_count; ++i) { | |
1181 | ret = ceph_zero_partial_object(inode, pos, NULL); | |
1182 | if (ret < 0) | |
1183 | return ret; | |
1184 | pos += stripe_unit; | |
1185 | } | |
1186 | offset += object_set_size; | |
1187 | length -= object_set_size; | |
1188 | } | |
1189 | while (length) { | |
1190 | loff_t size = length; | |
1191 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1192 | if (ret < 0) | |
1193 | return ret; | |
1194 | offset += size; | |
1195 | length -= size; | |
1196 | } | |
1197 | return ret; | |
1198 | } | |
1199 | ||
1200 | static long ceph_fallocate(struct file *file, int mode, | |
1201 | loff_t offset, loff_t length) | |
1202 | { | |
1203 | struct ceph_file_info *fi = file->private_data; | |
1204 | struct inode *inode = file->f_dentry->d_inode; | |
1205 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1206 | struct ceph_osd_client *osdc = | |
1207 | &ceph_inode_to_client(inode)->client->osdc; | |
1208 | int want, got = 0; | |
1209 | int dirty; | |
1210 | int ret = 0; | |
1211 | loff_t endoff = 0; | |
1212 | loff_t size; | |
1213 | ||
1214 | if (!S_ISREG(inode->i_mode)) | |
1215 | return -EOPNOTSUPP; | |
1216 | ||
1217 | if (IS_SWAPFILE(inode)) | |
1218 | return -ETXTBSY; | |
1219 | ||
1220 | mutex_lock(&inode->i_mutex); | |
1221 | ||
1222 | if (ceph_snap(inode) != CEPH_NOSNAP) { | |
1223 | ret = -EROFS; | |
1224 | goto unlock; | |
1225 | } | |
1226 | ||
1227 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) && | |
1228 | !(mode & FALLOC_FL_PUNCH_HOLE)) { | |
1229 | ret = -ENOSPC; | |
1230 | goto unlock; | |
1231 | } | |
1232 | ||
1233 | size = i_size_read(inode); | |
1234 | if (!(mode & FALLOC_FL_KEEP_SIZE)) | |
1235 | endoff = offset + length; | |
1236 | ||
1237 | if (fi->fmode & CEPH_FILE_MODE_LAZY) | |
1238 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1239 | else | |
1240 | want = CEPH_CAP_FILE_BUFFER; | |
1241 | ||
1242 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, endoff); | |
1243 | if (ret < 0) | |
1244 | goto unlock; | |
1245 | ||
1246 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
1247 | if (offset < size) | |
1248 | ceph_zero_pagecache_range(inode, offset, length); | |
1249 | ret = ceph_zero_objects(inode, offset, length); | |
1250 | } else if (endoff > size) { | |
1251 | truncate_pagecache_range(inode, size, -1); | |
1252 | if (ceph_inode_set_size(inode, endoff)) | |
1253 | ceph_check_caps(ceph_inode(inode), | |
1254 | CHECK_CAPS_AUTHONLY, NULL); | |
1255 | } | |
1256 | ||
1257 | if (!ret) { | |
1258 | spin_lock(&ci->i_ceph_lock); | |
1259 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR); | |
1260 | spin_unlock(&ci->i_ceph_lock); | |
1261 | if (dirty) | |
1262 | __mark_inode_dirty(inode, dirty); | |
1263 | } | |
1264 | ||
1265 | ceph_put_cap_refs(ci, got); | |
1266 | unlock: | |
1267 | mutex_unlock(&inode->i_mutex); | |
1268 | return ret; | |
1269 | } | |
1270 | ||
124e68e7 SW |
1271 | const struct file_operations ceph_file_fops = { |
1272 | .open = ceph_open, | |
1273 | .release = ceph_release, | |
1274 | .llseek = ceph_llseek, | |
1275 | .read = do_sync_read, | |
1276 | .write = do_sync_write, | |
1277 | .aio_read = ceph_aio_read, | |
1278 | .aio_write = ceph_aio_write, | |
1279 | .mmap = ceph_mmap, | |
1280 | .fsync = ceph_fsync, | |
40819f6f GF |
1281 | .lock = ceph_lock, |
1282 | .flock = ceph_flock, | |
124e68e7 SW |
1283 | .splice_read = generic_file_splice_read, |
1284 | .splice_write = generic_file_splice_write, | |
1285 | .unlocked_ioctl = ceph_ioctl, | |
1286 | .compat_ioctl = ceph_ioctl, | |
ad7a60de | 1287 | .fallocate = ceph_fallocate, |
124e68e7 SW |
1288 | }; |
1289 |