]>
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; |
752c8bdc | 213 | if (flags & O_CREAT) |
5f21c96d | 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 | 288 | if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) { |
b20a95a0 | 289 | ceph_init_acl(dentry, dentry->d_inode, dir); |
6e8575fa SL |
290 | *opened |= FILE_CREATED; |
291 | } | |
5ef50c3b SW |
292 | err = finish_open(file, dentry, ceph_open, opened); |
293 | } | |
5ef50c3b | 294 | out_err: |
ab866549 YZ |
295 | if (!req->r_err && req->r_target_inode) |
296 | ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode); | |
5ef50c3b SW |
297 | ceph_mdsc_put_request(req); |
298 | dout("atomic_open result=%d\n", err); | |
d9585277 | 299 | return err; |
124e68e7 SW |
300 | } |
301 | ||
302 | int ceph_release(struct inode *inode, struct file *file) | |
303 | { | |
304 | struct ceph_inode_info *ci = ceph_inode(inode); | |
305 | struct ceph_file_info *cf = file->private_data; | |
306 | ||
307 | dout("release inode %p file %p\n", inode, file); | |
308 | ceph_put_fmode(ci, cf->fmode); | |
309 | if (cf->last_readdir) | |
310 | ceph_mdsc_put_request(cf->last_readdir); | |
311 | kfree(cf->last_name); | |
312 | kfree(cf->dir_info); | |
313 | dput(cf->dentry); | |
314 | kmem_cache_free(ceph_file_cachep, cf); | |
195d3ce2 SW |
315 | |
316 | /* wake up anyone waiting for caps on this inode */ | |
03066f23 | 317 | wake_up_all(&ci->i_cap_wq); |
124e68e7 SW |
318 | return 0; |
319 | } | |
320 | ||
124e68e7 SW |
321 | /* |
322 | * Read a range of bytes striped over one or more objects. Iterate over | |
323 | * objects we stripe over. (That's not atomic, but good enough for now.) | |
324 | * | |
325 | * If we get a short result from the OSD, check against i_size; we need to | |
326 | * only return a short read to the caller if we hit EOF. | |
327 | */ | |
328 | static int striped_read(struct inode *inode, | |
329 | u64 off, u64 len, | |
6a026589 | 330 | struct page **pages, int num_pages, |
c3cd6283 | 331 | int *checkeof, bool o_direct, |
ab226e21 | 332 | unsigned long buf_align) |
124e68e7 | 333 | { |
3d14c5d2 | 334 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
124e68e7 | 335 | struct ceph_inode_info *ci = ceph_inode(inode); |
688bac46 | 336 | u64 pos, this_len, left; |
b7495fc2 | 337 | int io_align, page_align; |
688bac46 | 338 | int pages_left; |
124e68e7 SW |
339 | int read; |
340 | struct page **page_pos; | |
341 | int ret; | |
342 | bool hit_stripe, was_short; | |
343 | ||
344 | /* | |
345 | * we may need to do multiple reads. not atomic, unfortunately. | |
346 | */ | |
347 | pos = off; | |
348 | left = len; | |
349 | page_pos = pages; | |
350 | pages_left = num_pages; | |
351 | read = 0; | |
b7495fc2 | 352 | io_align = off & ~PAGE_MASK; |
124e68e7 SW |
353 | |
354 | more: | |
c3cd6283 | 355 | if (o_direct) |
ab226e21 | 356 | page_align = (pos - io_align + buf_align) & ~PAGE_MASK; |
b7495fc2 SW |
357 | else |
358 | page_align = pos & ~PAGE_MASK; | |
124e68e7 | 359 | this_len = left; |
3d14c5d2 | 360 | ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode), |
124e68e7 SW |
361 | &ci->i_layout, pos, &this_len, |
362 | ci->i_truncate_seq, | |
363 | ci->i_truncate_size, | |
b7495fc2 | 364 | page_pos, pages_left, page_align); |
124e68e7 SW |
365 | if (ret == -ENOENT) |
366 | ret = 0; | |
0e98728f SW |
367 | hit_stripe = this_len < left; |
368 | was_short = ret >= 0 && ret < this_len; | |
688bac46 | 369 | dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read, |
124e68e7 SW |
370 | ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : ""); |
371 | ||
02ae66d8 | 372 | if (ret >= 0) { |
373 | int didpages; | |
374 | if (was_short && (pos + ret < inode->i_size)) { | |
375 | u64 tmp = min(this_len - ret, | |
376 | inode->i_size - pos - ret); | |
377 | dout(" zero gap %llu to %llu\n", | |
378 | pos + ret, pos + ret + tmp); | |
379 | ceph_zero_page_vector_range(page_align + read + ret, | |
380 | tmp, pages); | |
381 | ret += tmp; | |
124e68e7 | 382 | } |
02ae66d8 | 383 | |
384 | didpages = (page_align + ret) >> PAGE_CACHE_SHIFT; | |
124e68e7 SW |
385 | pos += ret; |
386 | read = pos - off; | |
387 | left -= ret; | |
388 | page_pos += didpages; | |
389 | pages_left -= didpages; | |
390 | ||
02ae66d8 | 391 | /* hit stripe and need continue*/ |
392 | if (left && hit_stripe && pos < inode->i_size) | |
124e68e7 SW |
393 | goto more; |
394 | } | |
395 | ||
ee7289bf | 396 | if (read > 0) { |
02ae66d8 | 397 | ret = read; |
c3cd6283 SW |
398 | /* did we bounce off eof? */ |
399 | if (pos + left > inode->i_size) | |
400 | *checkeof = 1; | |
124e68e7 SW |
401 | } |
402 | ||
124e68e7 SW |
403 | dout("striped_read returns %d\n", ret); |
404 | return ret; | |
405 | } | |
406 | ||
407 | /* | |
408 | * Completely synchronous read and write methods. Direct from __user | |
409 | * buffer to osd, or directly to user pages (if O_DIRECT). | |
410 | * | |
411 | * If the read spans object boundary, just do multiple reads. | |
412 | */ | |
8eb4efb0 | 413 | static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i, |
414 | int *checkeof) | |
124e68e7 | 415 | { |
8eb4efb0 | 416 | struct file *file = iocb->ki_filp; |
496ad9aa | 417 | struct inode *inode = file_inode(file); |
124e68e7 | 418 | struct page **pages; |
8eb4efb0 | 419 | u64 off = iocb->ki_pos; |
ab226e21 | 420 | int num_pages, ret; |
2b777c9d | 421 | size_t len = iov_iter_count(i); |
124e68e7 | 422 | |
8eb4efb0 | 423 | dout("sync_read on file %p %llu~%u %s\n", file, off, |
424 | (unsigned)len, | |
124e68e7 | 425 | (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); |
d0d0db22 YZ |
426 | |
427 | if (!len) | |
428 | return 0; | |
e98b6fed SW |
429 | /* |
430 | * flush any page cache pages in this range. this | |
431 | * will make concurrent normal and sync io slow, | |
432 | * but it will at least behave sensibly when they are | |
433 | * in sequence. | |
434 | */ | |
8eb4efb0 | 435 | ret = filemap_write_and_wait_range(inode->i_mapping, off, |
436 | off + len); | |
29065a51 | 437 | if (ret < 0) |
8eb4efb0 | 438 | return ret; |
29065a51 | 439 | |
8eb4efb0 | 440 | if (file->f_flags & O_DIRECT) { |
441 | while (iov_iter_count(i)) { | |
2b777c9d AV |
442 | size_t start; |
443 | ssize_t n; | |
8eb4efb0 | 444 | |
2b777c9d AV |
445 | n = iov_iter_get_pages_alloc(i, &pages, INT_MAX, &start); |
446 | if (n < 0) | |
447 | return n; | |
8eb4efb0 | 448 | |
2b777c9d AV |
449 | num_pages = (n + start + PAGE_SIZE - 1) / PAGE_SIZE; |
450 | ||
451 | ret = striped_read(inode, off, n, | |
8eb4efb0 | 452 | pages, num_pages, checkeof, |
2b777c9d AV |
453 | 1, start); |
454 | ||
8eb4efb0 | 455 | ceph_put_page_vector(pages, num_pages, true); |
456 | ||
457 | if (ret <= 0) | |
458 | break; | |
459 | off += ret; | |
460 | iov_iter_advance(i, ret); | |
2b777c9d | 461 | if (ret < n) |
8eb4efb0 | 462 | break; |
463 | } | |
464 | } else { | |
465 | num_pages = calc_pages_for(off, len); | |
466 | pages = ceph_alloc_page_vector(num_pages, GFP_NOFS); | |
467 | if (IS_ERR(pages)) | |
468 | return PTR_ERR(pages); | |
469 | ret = striped_read(inode, off, len, pages, | |
470 | num_pages, checkeof, 0, 0); | |
471 | if (ret > 0) { | |
472 | int l, k = 0; | |
2b777c9d | 473 | size_t left = ret; |
8eb4efb0 | 474 | |
475 | while (left) { | |
5aaa432a YZ |
476 | size_t page_off = off & ~PAGE_MASK; |
477 | size_t copy = min_t(size_t, | |
478 | PAGE_SIZE - page_off, left); | |
479 | l = copy_page_to_iter(pages[k++], page_off, | |
480 | copy, i); | |
2b777c9d AV |
481 | off += l; |
482 | left -= l; | |
483 | if (l < copy) | |
8eb4efb0 | 484 | break; |
485 | } | |
486 | } | |
487 | ceph_release_page_vector(pages, num_pages); | |
488 | } | |
124e68e7 | 489 | |
8eb4efb0 | 490 | if (off > iocb->ki_pos) { |
491 | ret = off - iocb->ki_pos; | |
492 | iocb->ki_pos = off; | |
493 | } | |
124e68e7 | 494 | |
124e68e7 SW |
495 | dout("sync_read result %d\n", ret); |
496 | return ret; | |
497 | } | |
498 | ||
499 | /* | |
26be8808 AE |
500 | * Write commit request unsafe callback, called to tell us when a |
501 | * request is unsafe (that is, in flight--has been handed to the | |
502 | * messenger to send to its target osd). It is called again when | |
503 | * we've received a response message indicating the request is | |
504 | * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request | |
505 | * is completed early (and unsuccessfully) due to a timeout or | |
506 | * interrupt. | |
507 | * | |
508 | * This is used if we requested both an ACK and ONDISK commit reply | |
509 | * from the OSD. | |
124e68e7 | 510 | */ |
26be8808 | 511 | static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe) |
124e68e7 SW |
512 | { |
513 | struct ceph_inode_info *ci = ceph_inode(req->r_inode); | |
514 | ||
26be8808 AE |
515 | dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid, |
516 | unsafe ? "un" : ""); | |
517 | if (unsafe) { | |
518 | ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR); | |
519 | spin_lock(&ci->i_unsafe_lock); | |
520 | list_add_tail(&req->r_unsafe_item, | |
521 | &ci->i_unsafe_writes); | |
522 | spin_unlock(&ci->i_unsafe_lock); | |
523 | } else { | |
524 | spin_lock(&ci->i_unsafe_lock); | |
525 | list_del_init(&req->r_unsafe_item); | |
526 | spin_unlock(&ci->i_unsafe_lock); | |
527 | ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR); | |
528 | } | |
124e68e7 SW |
529 | } |
530 | ||
e8344e66 | 531 | |
124e68e7 | 532 | /* |
e8344e66 | 533 | * Synchronous write, straight from __user pointer or user pages. |
124e68e7 SW |
534 | * |
535 | * If write spans object boundary, just do multiple writes. (For a | |
536 | * correct atomic write, we should e.g. take write locks on all | |
537 | * objects, rollback on failure, etc.) | |
538 | */ | |
e8344e66 | 539 | static ssize_t |
4908b822 | 540 | ceph_sync_direct_write(struct kiocb *iocb, struct iov_iter *from) |
124e68e7 | 541 | { |
e8344e66 | 542 | struct file *file = iocb->ki_filp; |
496ad9aa | 543 | struct inode *inode = file_inode(file); |
124e68e7 | 544 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 | 545 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
acead002 AE |
546 | struct ceph_snap_context *snapc; |
547 | struct ceph_vino vino; | |
124e68e7 SW |
548 | struct ceph_osd_request *req; |
549 | struct page **pages; | |
550 | int num_pages; | |
124e68e7 SW |
551 | int written = 0; |
552 | int flags; | |
124e68e7 SW |
553 | int check_caps = 0; |
554 | int ret; | |
555 | struct timespec mtime = CURRENT_TIME; | |
e8344e66 | 556 | loff_t pos = iocb->ki_pos; |
4908b822 | 557 | size_t count = iov_iter_count(from); |
124e68e7 | 558 | |
496ad9aa | 559 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) |
124e68e7 SW |
560 | return -EROFS; |
561 | ||
e8344e66 | 562 | dout("sync_direct_write on file %p %lld~%u\n", file, pos, |
563 | (unsigned)count); | |
124e68e7 | 564 | |
e8344e66 | 565 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); |
29065a51 YS |
566 | if (ret < 0) |
567 | return ret; | |
568 | ||
569 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
570 | pos >> PAGE_CACHE_SHIFT, | |
e8344e66 | 571 | (pos + count) >> PAGE_CACHE_SHIFT); |
29065a51 YS |
572 | if (ret < 0) |
573 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
574 | ||
124e68e7 SW |
575 | flags = CEPH_OSD_FLAG_ORDERSNAP | |
576 | CEPH_OSD_FLAG_ONDISK | | |
577 | CEPH_OSD_FLAG_WRITE; | |
124e68e7 | 578 | |
4908b822 AV |
579 | while (iov_iter_count(from) > 0) { |
580 | u64 len = iov_iter_single_seg_count(from); | |
64c31311 AV |
581 | size_t start; |
582 | ssize_t n; | |
e8344e66 | 583 | |
584 | snapc = ci->i_snap_realm->cached_context; | |
585 | vino = ceph_vino(inode); | |
586 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
587 | vino, pos, &len, | |
588 | 2,/*include a 'startsync' command*/ | |
589 | CEPH_OSD_OP_WRITE, flags, snapc, | |
590 | ci->i_truncate_seq, | |
591 | ci->i_truncate_size, | |
592 | false); | |
593 | if (IS_ERR(req)) { | |
594 | ret = PTR_ERR(req); | |
eab87235 | 595 | break; |
e8344e66 | 596 | } |
124e68e7 | 597 | |
4908b822 | 598 | n = iov_iter_get_pages_alloc(from, &pages, len, &start); |
64c31311 AV |
599 | if (unlikely(n < 0)) { |
600 | ret = n; | |
601 | ceph_osdc_put_request(req); | |
602 | break; | |
124e68e7 SW |
603 | } |
604 | ||
64c31311 | 605 | num_pages = (n + start + PAGE_SIZE - 1) / PAGE_SIZE; |
124e68e7 SW |
606 | /* |
607 | * throw out any page cache pages in this range. this | |
608 | * may block. | |
609 | */ | |
213c99ee | 610 | truncate_inode_pages_range(inode->i_mapping, pos, |
64c31311 AV |
611 | (pos+n) | (PAGE_CACHE_SIZE-1)); |
612 | osd_req_op_extent_osd_data_pages(req, 0, pages, n, start, | |
e8344e66 | 613 | false, false); |
614 | ||
615 | /* BUG_ON(vino.snap != CEPH_NOSNAP); */ | |
616 | ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime); | |
617 | ||
618 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); | |
619 | if (!ret) | |
620 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
621 | ||
622 | ceph_put_page_vector(pages, num_pages, false); | |
623 | ||
e8344e66 | 624 | ceph_osdc_put_request(req); |
64c31311 | 625 | if (ret) |
e8344e66 | 626 | break; |
64c31311 AV |
627 | pos += n; |
628 | written += n; | |
4908b822 | 629 | iov_iter_advance(from, n); |
64c31311 AV |
630 | |
631 | if (pos > i_size_read(inode)) { | |
632 | check_caps = ceph_inode_set_size(inode, pos); | |
633 | if (check_caps) | |
634 | ceph_check_caps(ceph_inode(inode), | |
635 | CHECK_CAPS_AUTHONLY, | |
636 | NULL); | |
637 | } | |
e8344e66 | 638 | } |
639 | ||
640 | if (ret != -EOLDSNAPC && written > 0) { | |
641 | iocb->ki_pos = pos; | |
642 | ret = written; | |
643 | } | |
644 | return ret; | |
645 | } | |
646 | ||
647 | ||
648 | /* | |
649 | * Synchronous write, straight from __user pointer or user pages. | |
650 | * | |
651 | * If write spans object boundary, just do multiple writes. (For a | |
652 | * correct atomic write, we should e.g. take write locks on all | |
653 | * objects, rollback on failure, etc.) | |
654 | */ | |
4908b822 | 655 | static ssize_t ceph_sync_write(struct kiocb *iocb, struct iov_iter *from) |
e8344e66 | 656 | { |
657 | struct file *file = iocb->ki_filp; | |
658 | struct inode *inode = file_inode(file); | |
659 | struct ceph_inode_info *ci = ceph_inode(inode); | |
660 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
661 | struct ceph_snap_context *snapc; | |
662 | struct ceph_vino vino; | |
663 | struct ceph_osd_request *req; | |
664 | struct page **pages; | |
665 | u64 len; | |
666 | int num_pages; | |
667 | int written = 0; | |
668 | int flags; | |
669 | int check_caps = 0; | |
670 | int ret; | |
671 | struct timespec mtime = CURRENT_TIME; | |
672 | loff_t pos = iocb->ki_pos; | |
4908b822 | 673 | size_t count = iov_iter_count(from); |
e8344e66 | 674 | |
675 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) | |
676 | return -EROFS; | |
677 | ||
678 | dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count); | |
679 | ||
680 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); | |
681 | if (ret < 0) | |
682 | return ret; | |
683 | ||
684 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
685 | pos >> PAGE_CACHE_SHIFT, | |
686 | (pos + count) >> PAGE_CACHE_SHIFT); | |
687 | if (ret < 0) | |
688 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
689 | ||
690 | flags = CEPH_OSD_FLAG_ORDERSNAP | | |
691 | CEPH_OSD_FLAG_ONDISK | | |
692 | CEPH_OSD_FLAG_WRITE | | |
693 | CEPH_OSD_FLAG_ACK; | |
694 | ||
4908b822 | 695 | while ((len = iov_iter_count(from)) > 0) { |
e8344e66 | 696 | size_t left; |
697 | int n; | |
698 | ||
699 | snapc = ci->i_snap_realm->cached_context; | |
700 | vino = ceph_vino(inode); | |
701 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
702 | vino, pos, &len, 1, | |
703 | CEPH_OSD_OP_WRITE, flags, snapc, | |
704 | ci->i_truncate_seq, | |
705 | ci->i_truncate_size, | |
706 | false); | |
707 | if (IS_ERR(req)) { | |
708 | ret = PTR_ERR(req); | |
eab87235 | 709 | break; |
e8344e66 | 710 | } |
711 | ||
712 | /* | |
713 | * write from beginning of first page, | |
714 | * regardless of io alignment | |
715 | */ | |
716 | num_pages = (len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
717 | ||
34d23762 | 718 | pages = ceph_alloc_page_vector(num_pages, GFP_NOFS); |
124e68e7 SW |
719 | if (IS_ERR(pages)) { |
720 | ret = PTR_ERR(pages); | |
721 | goto out; | |
722 | } | |
e8344e66 | 723 | |
724 | left = len; | |
725 | for (n = 0; n < num_pages; n++) { | |
125d725c | 726 | size_t plen = min_t(size_t, left, PAGE_SIZE); |
4908b822 | 727 | ret = copy_page_from_iter(pages[n], 0, plen, from); |
e8344e66 | 728 | if (ret != plen) { |
729 | ret = -EFAULT; | |
730 | break; | |
731 | } | |
732 | left -= ret; | |
e8344e66 | 733 | } |
734 | ||
124e68e7 SW |
735 | if (ret < 0) { |
736 | ceph_release_page_vector(pages, num_pages); | |
737 | goto out; | |
738 | } | |
739 | ||
e8344e66 | 740 | /* get a second commit callback */ |
741 | req->r_unsafe_callback = ceph_sync_write_unsafe; | |
742 | req->r_inode = inode; | |
124e68e7 | 743 | |
e8344e66 | 744 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, |
745 | false, true); | |
02ee07d3 | 746 | |
e8344e66 | 747 | /* BUG_ON(vino.snap != CEPH_NOSNAP); */ |
748 | ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime); | |
124e68e7 | 749 | |
e8344e66 | 750 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
751 | if (!ret) | |
752 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
124e68e7 SW |
753 | |
754 | out: | |
e8344e66 | 755 | ceph_osdc_put_request(req); |
756 | if (ret == 0) { | |
757 | pos += len; | |
758 | written += len; | |
759 | ||
760 | if (pos > i_size_read(inode)) { | |
761 | check_caps = ceph_inode_set_size(inode, pos); | |
762 | if (check_caps) | |
763 | ceph_check_caps(ceph_inode(inode), | |
764 | CHECK_CAPS_AUTHONLY, | |
765 | NULL); | |
766 | } | |
767 | } else | |
768 | break; | |
769 | } | |
124e68e7 | 770 | |
e8344e66 | 771 | if (ret != -EOLDSNAPC && written > 0) { |
124e68e7 | 772 | ret = written; |
e8344e66 | 773 | iocb->ki_pos = pos; |
124e68e7 SW |
774 | } |
775 | return ret; | |
776 | } | |
777 | ||
778 | /* | |
779 | * Wrap generic_file_aio_read with checks for cap bits on the inode. | |
780 | * Atomically grab references, so that those bits are not released | |
781 | * back to the MDS mid-read. | |
782 | * | |
783 | * Hmm, the sync read case isn't actually async... should it be? | |
784 | */ | |
3644424d | 785 | static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to) |
124e68e7 SW |
786 | { |
787 | struct file *filp = iocb->ki_filp; | |
2962507c | 788 | struct ceph_file_info *fi = filp->private_data; |
8eb4efb0 | 789 | size_t len = iocb->ki_nbytes; |
496ad9aa | 790 | struct inode *inode = file_inode(filp); |
124e68e7 SW |
791 | struct ceph_inode_info *ci = ceph_inode(inode); |
792 | ssize_t ret; | |
2962507c | 793 | int want, got = 0; |
6a026589 | 794 | int checkeof = 0, read = 0; |
124e68e7 | 795 | |
6a026589 | 796 | again: |
8eb4efb0 | 797 | dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n", |
798 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode); | |
799 | ||
2962507c SW |
800 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
801 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
802 | else | |
803 | want = CEPH_CAP_FILE_CACHE; | |
804 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, &got, -1); | |
124e68e7 | 805 | if (ret < 0) |
8eb4efb0 | 806 | return ret; |
124e68e7 | 807 | |
2962507c | 808 | if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 || |
124e68e7 | 809 | (iocb->ki_filp->f_flags & O_DIRECT) || |
8eb4efb0 | 810 | (fi->flags & CEPH_F_SYNC)) { |
8eb4efb0 | 811 | |
812 | dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n", | |
813 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, | |
814 | ceph_cap_string(got)); | |
815 | ||
124e68e7 | 816 | /* hmm, this isn't really async... */ |
3644424d | 817 | ret = ceph_sync_read(iocb, to, &checkeof); |
8eb4efb0 | 818 | } else { |
8eb4efb0 | 819 | dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n", |
3644424d | 820 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, |
8eb4efb0 | 821 | ceph_cap_string(got)); |
124e68e7 | 822 | |
3644424d | 823 | ret = generic_file_read_iter(iocb, to); |
8eb4efb0 | 824 | } |
124e68e7 SW |
825 | dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n", |
826 | inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret); | |
827 | ceph_put_cap_refs(ci, got); | |
6a026589 SW |
828 | |
829 | if (checkeof && ret >= 0) { | |
8eb4efb0 | 830 | int statret = ceph_do_getattr(inode, |
831 | CEPH_STAT_CAP_SIZE); | |
6a026589 SW |
832 | |
833 | /* hit EOF or hole? */ | |
8eb4efb0 | 834 | if (statret == 0 && iocb->ki_pos < inode->i_size && |
835 | ret < len) { | |
836 | dout("sync_read hit hole, ppos %lld < size %lld" | |
837 | ", reading more\n", iocb->ki_pos, | |
838 | inode->i_size); | |
839 | ||
3644424d | 840 | iov_iter_advance(to, ret); |
6a026589 | 841 | read += ret; |
6a026589 SW |
842 | len -= ret; |
843 | checkeof = 0; | |
844 | goto again; | |
845 | } | |
846 | } | |
8eb4efb0 | 847 | |
6a026589 SW |
848 | if (ret >= 0) |
849 | ret += read; | |
850 | ||
124e68e7 SW |
851 | return ret; |
852 | } | |
853 | ||
854 | /* | |
855 | * Take cap references to avoid releasing caps to MDS mid-write. | |
856 | * | |
857 | * If we are synchronous, and write with an old snap context, the OSD | |
858 | * may return EOLDSNAPC. In that case, retry the write.. _after_ | |
859 | * dropping our cap refs and allowing the pending snap to logically | |
860 | * complete _before_ this write occurs. | |
861 | * | |
862 | * If we are near ENOSPC, write synchronously. | |
863 | */ | |
4908b822 | 864 | static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from) |
124e68e7 SW |
865 | { |
866 | struct file *file = iocb->ki_filp; | |
33caad32 | 867 | struct ceph_file_info *fi = file->private_data; |
496ad9aa | 868 | struct inode *inode = file_inode(file); |
124e68e7 | 869 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 YS |
870 | struct ceph_osd_client *osdc = |
871 | &ceph_sb_to_client(inode->i_sb)->client->osdc; | |
4908b822 | 872 | ssize_t count = iov_iter_count(from), written = 0; |
03d254ed | 873 | int err, want, got; |
4908b822 | 874 | loff_t pos = iocb->ki_pos; |
124e68e7 SW |
875 | |
876 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
877 | return -EROFS; | |
878 | ||
03d254ed | 879 | mutex_lock(&inode->i_mutex); |
03d254ed | 880 | |
03d254ed YZ |
881 | /* We can write back this queue in page reclaim */ |
882 | current->backing_dev_info = file->f_mapping->backing_dev_info; | |
883 | ||
884 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
885 | if (err) | |
886 | goto out; | |
887 | ||
888 | if (count == 0) | |
889 | goto out; | |
4908b822 | 890 | iov_iter_truncate(from, count); |
03d254ed YZ |
891 | |
892 | err = file_remove_suid(file); | |
893 | if (err) | |
894 | goto out; | |
895 | ||
896 | err = file_update_time(file); | |
897 | if (err) | |
898 | goto out; | |
899 | ||
124e68e7 | 900 | retry_snap: |
6070e0c1 | 901 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) { |
03d254ed | 902 | err = -ENOSPC; |
6070e0c1 YZ |
903 | goto out; |
904 | } | |
03d254ed | 905 | |
ac7f29bf | 906 | dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n", |
03d254ed | 907 | inode, ceph_vinop(inode), pos, count, inode->i_size); |
7971bd92 SW |
908 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
909 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
910 | else | |
911 | want = CEPH_CAP_FILE_BUFFER; | |
03d254ed YZ |
912 | got = 0; |
913 | err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, pos + count); | |
914 | if (err < 0) | |
37505d57 | 915 | goto out; |
124e68e7 | 916 | |
ac7f29bf | 917 | dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n", |
03d254ed | 918 | inode, ceph_vinop(inode), pos, count, ceph_cap_string(got)); |
7971bd92 SW |
919 | |
920 | if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 || | |
e8344e66 | 921 | (file->f_flags & O_DIRECT) || (fi->flags & CEPH_F_SYNC)) { |
4908b822 | 922 | struct iov_iter data; |
37505d57 | 923 | mutex_unlock(&inode->i_mutex); |
4908b822 AV |
924 | /* we might need to revert back to that point */ |
925 | data = *from; | |
e8344e66 | 926 | if (file->f_flags & O_DIRECT) |
4908b822 | 927 | written = ceph_sync_direct_write(iocb, &data); |
e8344e66 | 928 | else |
4908b822 | 929 | written = ceph_sync_write(iocb, &data); |
0e5dd45c | 930 | if (written == -EOLDSNAPC) { |
931 | dout("aio_write %p %llx.%llx %llu~%u" | |
932 | "got EOLDSNAPC, retrying\n", | |
933 | inode, ceph_vinop(inode), | |
4908b822 | 934 | pos, (unsigned)count); |
0e5dd45c | 935 | mutex_lock(&inode->i_mutex); |
0e5dd45c | 936 | goto retry_snap; |
937 | } | |
4908b822 AV |
938 | if (written > 0) |
939 | iov_iter_advance(from, written); | |
7971bd92 | 940 | } else { |
32d3e148 | 941 | loff_t old_size = inode->i_size; |
b0d7c223 YZ |
942 | /* |
943 | * No need to acquire the i_truncate_mutex. Because | |
944 | * the MDS revokes Fwb caps before sending truncate | |
945 | * message to us. We can't get Fwb cap while there | |
946 | * are pending vmtruncate. So write and vmtruncate | |
947 | * can not run at the same time | |
948 | */ | |
4908b822 | 949 | written = generic_perform_write(file, from, pos); |
aec605f4 AV |
950 | if (likely(written >= 0)) |
951 | iocb->ki_pos = pos + written; | |
32d3e148 YW |
952 | if (inode->i_size > old_size) |
953 | ceph_fscache_update_objectsize(inode); | |
6070e0c1 | 954 | mutex_unlock(&inode->i_mutex); |
7971bd92 | 955 | } |
d8de9ab6 | 956 | |
03d254ed | 957 | if (written >= 0) { |
fca65b4a | 958 | int dirty; |
be655596 | 959 | spin_lock(&ci->i_ceph_lock); |
fca65b4a | 960 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR); |
be655596 | 961 | spin_unlock(&ci->i_ceph_lock); |
fca65b4a SW |
962 | if (dirty) |
963 | __mark_inode_dirty(inode, dirty); | |
124e68e7 | 964 | } |
7971bd92 | 965 | |
124e68e7 | 966 | dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n", |
4908b822 | 967 | inode, ceph_vinop(inode), pos, (unsigned)count, |
7971bd92 | 968 | ceph_cap_string(got)); |
124e68e7 | 969 | ceph_put_cap_refs(ci, got); |
7971bd92 | 970 | |
03d254ed | 971 | if (written >= 0 && |
6070e0c1 YZ |
972 | ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) || |
973 | ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) { | |
03d254ed | 974 | err = vfs_fsync_range(file, pos, pos + written - 1, 1); |
6070e0c1 | 975 | if (err < 0) |
03d254ed | 976 | written = err; |
6070e0c1 | 977 | } |
03d254ed | 978 | |
2f75e9e1 SW |
979 | goto out_unlocked; |
980 | ||
03d254ed | 981 | out: |
2f75e9e1 SW |
982 | mutex_unlock(&inode->i_mutex); |
983 | out_unlocked: | |
03d254ed | 984 | current->backing_dev_info = NULL; |
03d254ed | 985 | return written ? written : err; |
124e68e7 SW |
986 | } |
987 | ||
988 | /* | |
989 | * llseek. be sure to verify file size on SEEK_END. | |
990 | */ | |
965c8e59 | 991 | static loff_t ceph_llseek(struct file *file, loff_t offset, int whence) |
124e68e7 SW |
992 | { |
993 | struct inode *inode = file->f_mapping->host; | |
994 | int ret; | |
995 | ||
996 | mutex_lock(&inode->i_mutex); | |
6a82c47a | 997 | |
965c8e59 | 998 | if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) { |
124e68e7 SW |
999 | ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE); |
1000 | if (ret < 0) { | |
1001 | offset = ret; | |
1002 | goto out; | |
1003 | } | |
06222e49 JB |
1004 | } |
1005 | ||
965c8e59 | 1006 | switch (whence) { |
06222e49 | 1007 | case SEEK_END: |
124e68e7 SW |
1008 | offset += inode->i_size; |
1009 | break; | |
1010 | case SEEK_CUR: | |
1011 | /* | |
1012 | * Here we special-case the lseek(fd, 0, SEEK_CUR) | |
1013 | * position-querying operation. Avoid rewriting the "same" | |
1014 | * f_pos value back to the file because a concurrent read(), | |
1015 | * write() or lseek() might have altered it | |
1016 | */ | |
1017 | if (offset == 0) { | |
1018 | offset = file->f_pos; | |
1019 | goto out; | |
1020 | } | |
1021 | offset += file->f_pos; | |
1022 | break; | |
06222e49 JB |
1023 | case SEEK_DATA: |
1024 | if (offset >= inode->i_size) { | |
1025 | ret = -ENXIO; | |
1026 | goto out; | |
1027 | } | |
1028 | break; | |
1029 | case SEEK_HOLE: | |
1030 | if (offset >= inode->i_size) { | |
1031 | ret = -ENXIO; | |
1032 | goto out; | |
1033 | } | |
1034 | offset = inode->i_size; | |
1035 | break; | |
124e68e7 SW |
1036 | } |
1037 | ||
46a1c2c7 | 1038 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
124e68e7 SW |
1039 | |
1040 | out: | |
1041 | mutex_unlock(&inode->i_mutex); | |
1042 | return offset; | |
1043 | } | |
1044 | ||
ad7a60de LW |
1045 | static inline void ceph_zero_partial_page( |
1046 | struct inode *inode, loff_t offset, unsigned size) | |
1047 | { | |
1048 | struct page *page; | |
1049 | pgoff_t index = offset >> PAGE_CACHE_SHIFT; | |
1050 | ||
1051 | page = find_lock_page(inode->i_mapping, index); | |
1052 | if (page) { | |
1053 | wait_on_page_writeback(page); | |
1054 | zero_user(page, offset & (PAGE_CACHE_SIZE - 1), size); | |
1055 | unlock_page(page); | |
1056 | page_cache_release(page); | |
1057 | } | |
1058 | } | |
1059 | ||
1060 | static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset, | |
1061 | loff_t length) | |
1062 | { | |
1063 | loff_t nearly = round_up(offset, PAGE_CACHE_SIZE); | |
1064 | if (offset < nearly) { | |
1065 | loff_t size = nearly - offset; | |
1066 | if (length < size) | |
1067 | size = length; | |
1068 | ceph_zero_partial_page(inode, offset, size); | |
1069 | offset += size; | |
1070 | length -= size; | |
1071 | } | |
1072 | if (length >= PAGE_CACHE_SIZE) { | |
1073 | loff_t size = round_down(length, PAGE_CACHE_SIZE); | |
1074 | truncate_pagecache_range(inode, offset, offset + size - 1); | |
1075 | offset += size; | |
1076 | length -= size; | |
1077 | } | |
1078 | if (length) | |
1079 | ceph_zero_partial_page(inode, offset, length); | |
1080 | } | |
1081 | ||
1082 | static int ceph_zero_partial_object(struct inode *inode, | |
1083 | loff_t offset, loff_t *length) | |
1084 | { | |
1085 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1086 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
1087 | struct ceph_osd_request *req; | |
1088 | int ret = 0; | |
1089 | loff_t zero = 0; | |
1090 | int op; | |
1091 | ||
1092 | if (!length) { | |
1093 | op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE; | |
1094 | length = &zero; | |
1095 | } else { | |
1096 | op = CEPH_OSD_OP_ZERO; | |
1097 | } | |
1098 | ||
1099 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
1100 | ceph_vino(inode), | |
1101 | offset, length, | |
1102 | 1, op, | |
1103 | CEPH_OSD_FLAG_WRITE | | |
1104 | CEPH_OSD_FLAG_ONDISK, | |
1105 | NULL, 0, 0, false); | |
1106 | if (IS_ERR(req)) { | |
1107 | ret = PTR_ERR(req); | |
1108 | goto out; | |
1109 | } | |
1110 | ||
1111 | ceph_osdc_build_request(req, offset, NULL, ceph_vino(inode).snap, | |
1112 | &inode->i_mtime); | |
1113 | ||
1114 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); | |
1115 | if (!ret) { | |
1116 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
1117 | if (ret == -ENOENT) | |
1118 | ret = 0; | |
1119 | } | |
1120 | ceph_osdc_put_request(req); | |
1121 | ||
1122 | out: | |
1123 | return ret; | |
1124 | } | |
1125 | ||
1126 | static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length) | |
1127 | { | |
1128 | int ret = 0; | |
1129 | struct ceph_inode_info *ci = ceph_inode(inode); | |
b314a90d SW |
1130 | s32 stripe_unit = ceph_file_layout_su(ci->i_layout); |
1131 | s32 stripe_count = ceph_file_layout_stripe_count(ci->i_layout); | |
1132 | s32 object_size = ceph_file_layout_object_size(ci->i_layout); | |
1133 | u64 object_set_size = object_size * stripe_count; | |
1134 | u64 nearly, t; | |
1135 | ||
1136 | /* round offset up to next period boundary */ | |
1137 | nearly = offset + object_set_size - 1; | |
1138 | t = nearly; | |
1139 | nearly -= do_div(t, object_set_size); | |
ad7a60de | 1140 | |
ad7a60de LW |
1141 | while (length && offset < nearly) { |
1142 | loff_t size = length; | |
1143 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1144 | if (ret < 0) | |
1145 | return ret; | |
1146 | offset += size; | |
1147 | length -= size; | |
1148 | } | |
1149 | while (length >= object_set_size) { | |
1150 | int i; | |
1151 | loff_t pos = offset; | |
1152 | for (i = 0; i < stripe_count; ++i) { | |
1153 | ret = ceph_zero_partial_object(inode, pos, NULL); | |
1154 | if (ret < 0) | |
1155 | return ret; | |
1156 | pos += stripe_unit; | |
1157 | } | |
1158 | offset += object_set_size; | |
1159 | length -= object_set_size; | |
1160 | } | |
1161 | while (length) { | |
1162 | loff_t size = length; | |
1163 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1164 | if (ret < 0) | |
1165 | return ret; | |
1166 | offset += size; | |
1167 | length -= size; | |
1168 | } | |
1169 | return ret; | |
1170 | } | |
1171 | ||
1172 | static long ceph_fallocate(struct file *file, int mode, | |
1173 | loff_t offset, loff_t length) | |
1174 | { | |
1175 | struct ceph_file_info *fi = file->private_data; | |
aa8b60e0 | 1176 | struct inode *inode = file_inode(file); |
ad7a60de LW |
1177 | struct ceph_inode_info *ci = ceph_inode(inode); |
1178 | struct ceph_osd_client *osdc = | |
1179 | &ceph_inode_to_client(inode)->client->osdc; | |
1180 | int want, got = 0; | |
1181 | int dirty; | |
1182 | int ret = 0; | |
1183 | loff_t endoff = 0; | |
1184 | loff_t size; | |
1185 | ||
494d77bf YZ |
1186 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
1187 | return -EOPNOTSUPP; | |
1188 | ||
ad7a60de LW |
1189 | if (!S_ISREG(inode->i_mode)) |
1190 | return -EOPNOTSUPP; | |
1191 | ||
ad7a60de LW |
1192 | mutex_lock(&inode->i_mutex); |
1193 | ||
1194 | if (ceph_snap(inode) != CEPH_NOSNAP) { | |
1195 | ret = -EROFS; | |
1196 | goto unlock; | |
1197 | } | |
1198 | ||
1199 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) && | |
1200 | !(mode & FALLOC_FL_PUNCH_HOLE)) { | |
1201 | ret = -ENOSPC; | |
1202 | goto unlock; | |
1203 | } | |
1204 | ||
1205 | size = i_size_read(inode); | |
1206 | if (!(mode & FALLOC_FL_KEEP_SIZE)) | |
1207 | endoff = offset + length; | |
1208 | ||
1209 | if (fi->fmode & CEPH_FILE_MODE_LAZY) | |
1210 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1211 | else | |
1212 | want = CEPH_CAP_FILE_BUFFER; | |
1213 | ||
1214 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, endoff); | |
1215 | if (ret < 0) | |
1216 | goto unlock; | |
1217 | ||
1218 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
1219 | if (offset < size) | |
1220 | ceph_zero_pagecache_range(inode, offset, length); | |
1221 | ret = ceph_zero_objects(inode, offset, length); | |
1222 | } else if (endoff > size) { | |
1223 | truncate_pagecache_range(inode, size, -1); | |
1224 | if (ceph_inode_set_size(inode, endoff)) | |
1225 | ceph_check_caps(ceph_inode(inode), | |
1226 | CHECK_CAPS_AUTHONLY, NULL); | |
1227 | } | |
1228 | ||
1229 | if (!ret) { | |
1230 | spin_lock(&ci->i_ceph_lock); | |
1231 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR); | |
1232 | spin_unlock(&ci->i_ceph_lock); | |
1233 | if (dirty) | |
1234 | __mark_inode_dirty(inode, dirty); | |
1235 | } | |
1236 | ||
1237 | ceph_put_cap_refs(ci, got); | |
1238 | unlock: | |
1239 | mutex_unlock(&inode->i_mutex); | |
1240 | return ret; | |
1241 | } | |
1242 | ||
124e68e7 SW |
1243 | const struct file_operations ceph_file_fops = { |
1244 | .open = ceph_open, | |
1245 | .release = ceph_release, | |
1246 | .llseek = ceph_llseek, | |
3644424d | 1247 | .read = new_sync_read, |
4908b822 | 1248 | .write = new_sync_write, |
3644424d | 1249 | .read_iter = ceph_read_iter, |
4908b822 | 1250 | .write_iter = ceph_write_iter, |
124e68e7 SW |
1251 | .mmap = ceph_mmap, |
1252 | .fsync = ceph_fsync, | |
40819f6f GF |
1253 | .lock = ceph_lock, |
1254 | .flock = ceph_flock, | |
124e68e7 | 1255 | .splice_read = generic_file_splice_read, |
3551dd79 | 1256 | .splice_write = iter_file_splice_write, |
124e68e7 SW |
1257 | .unlocked_ioctl = ceph_ioctl, |
1258 | .compat_ioctl = ceph_ioctl, | |
ad7a60de | 1259 | .fallocate = ceph_fallocate, |
124e68e7 SW |
1260 | }; |
1261 |