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> | |
10 | ||
11 | #include "super.h" | |
12 | #include "mds_client.h" | |
13 | ||
14 | /* | |
15 | * Ceph file operations | |
16 | * | |
17 | * Implement basic open/close functionality, and implement | |
18 | * read/write. | |
19 | * | |
20 | * We implement three modes of file I/O: | |
21 | * - buffered uses the generic_file_aio_{read,write} helpers | |
22 | * | |
23 | * - synchronous is used when there is multi-client read/write | |
24 | * sharing, avoids the page cache, and synchronously waits for an | |
25 | * ack from the OSD. | |
26 | * | |
27 | * - direct io takes the variant of the sync path that references | |
28 | * user pages directly. | |
29 | * | |
30 | * fsync() flushes and waits on dirty pages, but just queues metadata | |
31 | * for writeback: since the MDS can recover size and mtime there is no | |
32 | * need to wait for MDS acknowledgement. | |
33 | */ | |
34 | ||
35 | ||
36 | /* | |
37 | * Prepare an open request. Preallocate ceph_cap to avoid an | |
38 | * inopportune ENOMEM later. | |
39 | */ | |
40 | static struct ceph_mds_request * | |
41 | prepare_open_request(struct super_block *sb, int flags, int create_mode) | |
42 | { | |
3d14c5d2 YS |
43 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); |
44 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
45 | struct ceph_mds_request *req; |
46 | int want_auth = USE_ANY_MDS; | |
47 | int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN; | |
48 | ||
49 | if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC)) | |
50 | want_auth = USE_AUTH_MDS; | |
51 | ||
52 | req = ceph_mdsc_create_request(mdsc, op, want_auth); | |
53 | if (IS_ERR(req)) | |
54 | goto out; | |
55 | req->r_fmode = ceph_flags_to_mode(flags); | |
56 | req->r_args.open.flags = cpu_to_le32(flags); | |
57 | req->r_args.open.mode = cpu_to_le32(create_mode); | |
124e68e7 SW |
58 | out: |
59 | return req; | |
60 | } | |
61 | ||
62 | /* | |
63 | * initialize private struct file data. | |
64 | * if we fail, clean up by dropping fmode reference on the ceph_inode | |
65 | */ | |
66 | static int ceph_init_file(struct inode *inode, struct file *file, int fmode) | |
67 | { | |
68 | struct ceph_file_info *cf; | |
69 | int ret = 0; | |
70 | ||
71 | switch (inode->i_mode & S_IFMT) { | |
72 | case S_IFREG: | |
73 | case S_IFDIR: | |
74 | dout("init_file %p %p 0%o (regular)\n", inode, file, | |
75 | inode->i_mode); | |
76 | cf = kmem_cache_alloc(ceph_file_cachep, GFP_NOFS | __GFP_ZERO); | |
77 | if (cf == NULL) { | |
78 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
79 | return -ENOMEM; | |
80 | } | |
81 | cf->fmode = fmode; | |
82 | cf->next_offset = 2; | |
83 | file->private_data = cf; | |
84 | BUG_ON(inode->i_fop->release != ceph_release); | |
85 | break; | |
86 | ||
87 | case S_IFLNK: | |
88 | dout("init_file %p %p 0%o (symlink)\n", inode, file, | |
89 | inode->i_mode); | |
90 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
91 | break; | |
92 | ||
93 | default: | |
94 | dout("init_file %p %p 0%o (special)\n", inode, file, | |
95 | inode->i_mode); | |
96 | /* | |
97 | * we need to drop the open ref now, since we don't | |
98 | * have .release set to ceph_release. | |
99 | */ | |
100 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
101 | BUG_ON(inode->i_fop->release == ceph_release); | |
102 | ||
103 | /* call the proper open fop */ | |
104 | ret = inode->i_fop->open(inode, file); | |
105 | } | |
106 | return ret; | |
107 | } | |
108 | ||
109 | /* | |
124e68e7 SW |
110 | * If we already have the requisite capabilities, we can satisfy |
111 | * the open request locally (no need to request new caps from the | |
112 | * MDS). We do, however, need to inform the MDS (asynchronously) | |
113 | * if our wanted caps set expands. | |
114 | */ | |
115 | int ceph_open(struct inode *inode, struct file *file) | |
116 | { | |
117 | struct ceph_inode_info *ci = ceph_inode(inode); | |
3d14c5d2 YS |
118 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
119 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
120 | struct ceph_mds_request *req; |
121 | struct ceph_file_info *cf = file->private_data; | |
5f21c96d | 122 | struct inode *parent_inode = NULL; |
124e68e7 SW |
123 | int err; |
124 | int flags, fmode, wanted; | |
125 | ||
126 | if (cf) { | |
127 | dout("open file %p is already opened\n", file); | |
128 | return 0; | |
129 | } | |
130 | ||
131 | /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */ | |
132 | flags = file->f_flags & ~(O_CREAT|O_EXCL); | |
133 | if (S_ISDIR(inode->i_mode)) | |
134 | flags = O_DIRECTORY; /* mds likes to know */ | |
135 | ||
136 | dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode, | |
137 | ceph_vinop(inode), file, flags, file->f_flags); | |
138 | fmode = ceph_flags_to_mode(flags); | |
139 | wanted = ceph_caps_for_mode(fmode); | |
140 | ||
141 | /* snapped files are read-only */ | |
142 | if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE)) | |
143 | return -EROFS; | |
144 | ||
145 | /* trivially open snapdir */ | |
146 | if (ceph_snap(inode) == CEPH_SNAPDIR) { | |
be655596 | 147 | spin_lock(&ci->i_ceph_lock); |
124e68e7 | 148 | __ceph_get_fmode(ci, fmode); |
be655596 | 149 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
150 | return ceph_init_file(inode, file, fmode); |
151 | } | |
152 | ||
153 | /* | |
7421ab80 SW |
154 | * No need to block if we have caps on the auth MDS (for |
155 | * write) or any MDS (for read). Update wanted set | |
124e68e7 SW |
156 | * asynchronously. |
157 | */ | |
be655596 | 158 | spin_lock(&ci->i_ceph_lock); |
7421ab80 SW |
159 | if (__ceph_is_any_real_caps(ci) && |
160 | (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) { | |
124e68e7 SW |
161 | int mds_wanted = __ceph_caps_mds_wanted(ci); |
162 | int issued = __ceph_caps_issued(ci, NULL); | |
163 | ||
164 | dout("open %p fmode %d want %s issued %s using existing\n", | |
165 | inode, fmode, ceph_cap_string(wanted), | |
166 | ceph_cap_string(issued)); | |
167 | __ceph_get_fmode(ci, fmode); | |
be655596 | 168 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
169 | |
170 | /* adjust wanted? */ | |
171 | if ((issued & wanted) != wanted && | |
172 | (mds_wanted & wanted) != wanted && | |
173 | ceph_snap(inode) != CEPH_SNAPDIR) | |
174 | ceph_check_caps(ci, 0, NULL); | |
175 | ||
176 | return ceph_init_file(inode, file, fmode); | |
177 | } else if (ceph_snap(inode) != CEPH_NOSNAP && | |
178 | (ci->i_snap_caps & wanted) == wanted) { | |
179 | __ceph_get_fmode(ci, fmode); | |
be655596 | 180 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
181 | return ceph_init_file(inode, file, fmode); |
182 | } | |
be655596 | 183 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
184 | |
185 | dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted)); | |
186 | req = prepare_open_request(inode->i_sb, flags, 0); | |
187 | if (IS_ERR(req)) { | |
188 | err = PTR_ERR(req); | |
189 | goto out; | |
190 | } | |
70b666c3 SW |
191 | req->r_inode = inode; |
192 | ihold(inode); | |
124e68e7 | 193 | req->r_num_caps = 1; |
5f21c96d SW |
194 | if (flags & (O_CREAT|O_TRUNC)) |
195 | parent_inode = ceph_get_dentry_parent_inode(file->f_dentry); | |
124e68e7 | 196 | err = ceph_mdsc_do_request(mdsc, parent_inode, req); |
5f21c96d | 197 | iput(parent_inode); |
124e68e7 SW |
198 | if (!err) |
199 | err = ceph_init_file(inode, file, req->r_fmode); | |
200 | ceph_mdsc_put_request(req); | |
201 | dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode)); | |
202 | out: | |
203 | return err; | |
204 | } | |
205 | ||
206 | ||
207 | /* | |
5ef50c3b SW |
208 | * Do a lookup + open with a single request. If we get a non-existent |
209 | * file or symlink, return 1 so the VFS can retry. | |
124e68e7 | 210 | */ |
5ef50c3b | 211 | int ceph_atomic_open(struct inode *dir, struct dentry *dentry, |
30d90494 | 212 | struct file *file, unsigned flags, umode_t mode, |
d9585277 | 213 | int *opened) |
124e68e7 | 214 | { |
3d14c5d2 YS |
215 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
216 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 | 217 | struct ceph_mds_request *req; |
5ef50c3b | 218 | struct dentry *dn; |
124e68e7 | 219 | int err; |
124e68e7 | 220 | |
5ef50c3b SW |
221 | dout("atomic_open %p dentry %p '%.*s' %s flags %d mode 0%o\n", |
222 | dir, dentry, dentry->d_name.len, dentry->d_name.name, | |
223 | d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode); | |
224 | ||
225 | if (dentry->d_name.len > NAME_MAX) | |
226 | return -ENAMETOOLONG; | |
227 | ||
228 | err = ceph_init_dentry(dentry); | |
229 | if (err < 0) | |
230 | return err; | |
124e68e7 SW |
231 | |
232 | /* do the open */ | |
233 | req = prepare_open_request(dir->i_sb, flags, mode); | |
234 | if (IS_ERR(req)) | |
d9585277 | 235 | return PTR_ERR(req); |
124e68e7 SW |
236 | req->r_dentry = dget(dentry); |
237 | req->r_num_caps = 2; | |
238 | if (flags & O_CREAT) { | |
239 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
240 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
241 | } | |
242 | req->r_locked_dir = dir; /* caller holds dir->i_mutex */ | |
acda7657 SW |
243 | err = ceph_mdsc_do_request(mdsc, |
244 | (flags & (O_CREAT|O_TRUNC)) ? dir : NULL, | |
245 | req); | |
79aec984 SL |
246 | if (err) |
247 | goto out_err; | |
248 | ||
468640e3 | 249 | err = ceph_handle_snapdir(req, dentry, err); |
5ef50c3b | 250 | if (err == 0 && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry) |
124e68e7 | 251 | err = ceph_handle_notrace_create(dir, dentry); |
2d83bde9 | 252 | |
5ef50c3b SW |
253 | if (d_unhashed(dentry)) { |
254 | dn = ceph_finish_lookup(req, dentry, err); | |
255 | if (IS_ERR(dn)) | |
256 | err = PTR_ERR(dn); | |
257 | } else { | |
258 | /* we were given a hashed negative dentry */ | |
259 | dn = NULL; | |
260 | } | |
261 | if (err) | |
262 | goto out_err; | |
263 | if (dn || dentry->d_inode == NULL || S_ISLNK(dentry->d_inode->i_mode)) { | |
264 | /* make vfs retry on splice, ENOENT, or symlink */ | |
265 | dout("atomic_open finish_no_open on dn %p\n", dn); | |
266 | err = finish_no_open(file, dn); | |
267 | } else { | |
268 | dout("atomic_open finish_open on dn %p\n", dn); | |
6e8575fa SL |
269 | if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) { |
270 | *opened |= FILE_CREATED; | |
271 | } | |
5ef50c3b SW |
272 | err = finish_open(file, dentry, ceph_open, opened); |
273 | } | |
2d83bde9 | 274 | |
5ef50c3b SW |
275 | out_err: |
276 | ceph_mdsc_put_request(req); | |
277 | dout("atomic_open result=%d\n", err); | |
d9585277 | 278 | return err; |
124e68e7 SW |
279 | } |
280 | ||
281 | int ceph_release(struct inode *inode, struct file *file) | |
282 | { | |
283 | struct ceph_inode_info *ci = ceph_inode(inode); | |
284 | struct ceph_file_info *cf = file->private_data; | |
285 | ||
286 | dout("release inode %p file %p\n", inode, file); | |
287 | ceph_put_fmode(ci, cf->fmode); | |
288 | if (cf->last_readdir) | |
289 | ceph_mdsc_put_request(cf->last_readdir); | |
290 | kfree(cf->last_name); | |
291 | kfree(cf->dir_info); | |
292 | dput(cf->dentry); | |
293 | kmem_cache_free(ceph_file_cachep, cf); | |
195d3ce2 SW |
294 | |
295 | /* wake up anyone waiting for caps on this inode */ | |
03066f23 | 296 | wake_up_all(&ci->i_cap_wq); |
124e68e7 SW |
297 | return 0; |
298 | } | |
299 | ||
124e68e7 SW |
300 | /* |
301 | * Read a range of bytes striped over one or more objects. Iterate over | |
302 | * objects we stripe over. (That's not atomic, but good enough for now.) | |
303 | * | |
304 | * If we get a short result from the OSD, check against i_size; we need to | |
305 | * only return a short read to the caller if we hit EOF. | |
306 | */ | |
307 | static int striped_read(struct inode *inode, | |
308 | u64 off, u64 len, | |
6a026589 | 309 | struct page **pages, int num_pages, |
c3cd6283 | 310 | int *checkeof, bool o_direct, |
ab226e21 | 311 | unsigned long buf_align) |
124e68e7 | 312 | { |
3d14c5d2 | 313 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
124e68e7 SW |
314 | struct ceph_inode_info *ci = ceph_inode(inode); |
315 | u64 pos, this_len; | |
b7495fc2 | 316 | int io_align, page_align; |
124e68e7 SW |
317 | int left, pages_left; |
318 | int read; | |
319 | struct page **page_pos; | |
320 | int ret; | |
321 | bool hit_stripe, was_short; | |
322 | ||
323 | /* | |
324 | * we may need to do multiple reads. not atomic, unfortunately. | |
325 | */ | |
326 | pos = off; | |
327 | left = len; | |
328 | page_pos = pages; | |
329 | pages_left = num_pages; | |
330 | read = 0; | |
b7495fc2 | 331 | io_align = off & ~PAGE_MASK; |
124e68e7 SW |
332 | |
333 | more: | |
c3cd6283 | 334 | if (o_direct) |
ab226e21 | 335 | page_align = (pos - io_align + buf_align) & ~PAGE_MASK; |
b7495fc2 SW |
336 | else |
337 | page_align = pos & ~PAGE_MASK; | |
124e68e7 | 338 | this_len = left; |
3d14c5d2 | 339 | ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode), |
124e68e7 SW |
340 | &ci->i_layout, pos, &this_len, |
341 | ci->i_truncate_seq, | |
342 | ci->i_truncate_size, | |
b7495fc2 | 343 | page_pos, pages_left, page_align); |
124e68e7 SW |
344 | if (ret == -ENOENT) |
345 | ret = 0; | |
0e98728f SW |
346 | hit_stripe = this_len < left; |
347 | was_short = ret >= 0 && ret < this_len; | |
124e68e7 SW |
348 | dout("striped_read %llu~%u (read %u) got %d%s%s\n", pos, left, read, |
349 | ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : ""); | |
350 | ||
351 | if (ret > 0) { | |
773e9b44 | 352 | int didpages = (page_align + ret) >> PAGE_CACHE_SHIFT; |
124e68e7 SW |
353 | |
354 | if (read < pos - off) { | |
355 | dout(" zero gap %llu to %llu\n", off + read, pos); | |
773e9b44 | 356 | ceph_zero_page_vector_range(page_align + read, |
3d14c5d2 | 357 | pos - off - read, pages); |
124e68e7 SW |
358 | } |
359 | pos += ret; | |
360 | read = pos - off; | |
361 | left -= ret; | |
362 | page_pos += didpages; | |
363 | pages_left -= didpages; | |
364 | ||
365 | /* hit stripe? */ | |
366 | if (left && hit_stripe) | |
367 | goto more; | |
368 | } | |
369 | ||
370 | if (was_short) { | |
c3cd6283 SW |
371 | /* did we bounce off eof? */ |
372 | if (pos + left > inode->i_size) | |
373 | *checkeof = 1; | |
374 | ||
375 | /* zero trailing bytes (inside i_size) */ | |
376 | if (left > 0 && pos < inode->i_size) { | |
377 | if (pos + left > inode->i_size) | |
378 | left = inode->i_size - pos; | |
379 | ||
380 | dout("zero tail %d\n", left); | |
773e9b44 | 381 | ceph_zero_page_vector_range(page_align + read, left, |
3d14c5d2 | 382 | pages); |
c3cd6283 | 383 | read += left; |
124e68e7 | 384 | } |
124e68e7 SW |
385 | } |
386 | ||
124e68e7 SW |
387 | if (ret >= 0) |
388 | ret = read; | |
389 | dout("striped_read returns %d\n", ret); | |
390 | return ret; | |
391 | } | |
392 | ||
393 | /* | |
394 | * Completely synchronous read and write methods. Direct from __user | |
395 | * buffer to osd, or directly to user pages (if O_DIRECT). | |
396 | * | |
397 | * If the read spans object boundary, just do multiple reads. | |
398 | */ | |
399 | static ssize_t ceph_sync_read(struct file *file, char __user *data, | |
6a026589 | 400 | unsigned len, loff_t *poff, int *checkeof) |
124e68e7 | 401 | { |
496ad9aa | 402 | struct inode *inode = file_inode(file); |
124e68e7 SW |
403 | struct page **pages; |
404 | u64 off = *poff; | |
ab226e21 | 405 | int num_pages, ret; |
124e68e7 SW |
406 | |
407 | dout("sync_read on file %p %llu~%u %s\n", file, off, len, | |
408 | (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); | |
409 | ||
ab226e21 HC |
410 | if (file->f_flags & O_DIRECT) { |
411 | num_pages = calc_pages_for((unsigned long)data, len); | |
b6aa5901 | 412 | pages = ceph_get_direct_page_vector(data, num_pages, true); |
ab226e21 HC |
413 | } else { |
414 | num_pages = calc_pages_for(off, len); | |
34d23762 | 415 | pages = ceph_alloc_page_vector(num_pages, GFP_NOFS); |
ab226e21 | 416 | } |
124e68e7 SW |
417 | if (IS_ERR(pages)) |
418 | return PTR_ERR(pages); | |
419 | ||
e98b6fed SW |
420 | /* |
421 | * flush any page cache pages in this range. this | |
422 | * will make concurrent normal and sync io slow, | |
423 | * but it will at least behave sensibly when they are | |
424 | * in sequence. | |
425 | */ | |
29065a51 YS |
426 | ret = filemap_write_and_wait(inode->i_mapping); |
427 | if (ret < 0) | |
428 | goto done; | |
429 | ||
b7495fc2 | 430 | ret = striped_read(inode, off, len, pages, num_pages, checkeof, |
ab226e21 HC |
431 | file->f_flags & O_DIRECT, |
432 | (unsigned long)data & ~PAGE_MASK); | |
124e68e7 SW |
433 | |
434 | if (ret >= 0 && (file->f_flags & O_DIRECT) == 0) | |
3d14c5d2 | 435 | ret = ceph_copy_page_vector_to_user(pages, data, off, ret); |
124e68e7 SW |
436 | if (ret >= 0) |
437 | *poff = off + ret; | |
438 | ||
29065a51 | 439 | done: |
124e68e7 | 440 | if (file->f_flags & O_DIRECT) |
b6aa5901 | 441 | ceph_put_page_vector(pages, num_pages, true); |
124e68e7 SW |
442 | else |
443 | ceph_release_page_vector(pages, num_pages); | |
444 | dout("sync_read result %d\n", ret); | |
445 | return ret; | |
446 | } | |
447 | ||
448 | /* | |
26be8808 AE |
449 | * Write commit request unsafe callback, called to tell us when a |
450 | * request is unsafe (that is, in flight--has been handed to the | |
451 | * messenger to send to its target osd). It is called again when | |
452 | * we've received a response message indicating the request is | |
453 | * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request | |
454 | * is completed early (and unsuccessfully) due to a timeout or | |
455 | * interrupt. | |
456 | * | |
457 | * This is used if we requested both an ACK and ONDISK commit reply | |
458 | * from the OSD. | |
124e68e7 | 459 | */ |
26be8808 | 460 | static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe) |
124e68e7 SW |
461 | { |
462 | struct ceph_inode_info *ci = ceph_inode(req->r_inode); | |
463 | ||
26be8808 AE |
464 | dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid, |
465 | unsafe ? "un" : ""); | |
466 | if (unsafe) { | |
467 | ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR); | |
468 | spin_lock(&ci->i_unsafe_lock); | |
469 | list_add_tail(&req->r_unsafe_item, | |
470 | &ci->i_unsafe_writes); | |
471 | spin_unlock(&ci->i_unsafe_lock); | |
472 | } else { | |
473 | spin_lock(&ci->i_unsafe_lock); | |
474 | list_del_init(&req->r_unsafe_item); | |
475 | spin_unlock(&ci->i_unsafe_lock); | |
476 | ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR); | |
477 | } | |
124e68e7 SW |
478 | } |
479 | ||
480 | /* | |
481 | * Synchronous write, straight from __user pointer or user pages (if | |
482 | * O_DIRECT). | |
483 | * | |
484 | * If write spans object boundary, just do multiple writes. (For a | |
485 | * correct atomic write, we should e.g. take write locks on all | |
486 | * objects, rollback on failure, etc.) | |
487 | */ | |
488 | static ssize_t ceph_sync_write(struct file *file, const char __user *data, | |
489 | size_t left, loff_t *offset) | |
490 | { | |
496ad9aa | 491 | struct inode *inode = file_inode(file); |
124e68e7 | 492 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 | 493 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
acead002 AE |
494 | struct ceph_snap_context *snapc; |
495 | struct ceph_vino vino; | |
124e68e7 | 496 | struct ceph_osd_request *req; |
acead002 | 497 | int num_ops = 1; |
124e68e7 SW |
498 | struct page **pages; |
499 | int num_pages; | |
500 | long long unsigned pos; | |
501 | u64 len; | |
502 | int written = 0; | |
503 | int flags; | |
124e68e7 | 504 | int check_caps = 0; |
b7495fc2 | 505 | int page_align, io_align; |
ab226e21 | 506 | unsigned long buf_align; |
124e68e7 SW |
507 | int ret; |
508 | struct timespec mtime = CURRENT_TIME; | |
43bfe5de | 509 | bool own_pages = false; |
124e68e7 | 510 | |
496ad9aa | 511 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) |
124e68e7 SW |
512 | return -EROFS; |
513 | ||
514 | dout("sync_write on file %p %lld~%u %s\n", file, *offset, | |
515 | (unsigned)left, (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); | |
516 | ||
517 | if (file->f_flags & O_APPEND) | |
518 | pos = i_size_read(inode); | |
519 | else | |
520 | pos = *offset; | |
521 | ||
29065a51 YS |
522 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + left); |
523 | if (ret < 0) | |
524 | return ret; | |
525 | ||
526 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
527 | pos >> PAGE_CACHE_SHIFT, | |
528 | (pos + left) >> PAGE_CACHE_SHIFT); | |
529 | if (ret < 0) | |
530 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
531 | ||
124e68e7 SW |
532 | flags = CEPH_OSD_FLAG_ORDERSNAP | |
533 | CEPH_OSD_FLAG_ONDISK | | |
534 | CEPH_OSD_FLAG_WRITE; | |
535 | if ((file->f_flags & (O_SYNC|O_DIRECT)) == 0) | |
536 | flags |= CEPH_OSD_FLAG_ACK; | |
537 | else | |
acead002 | 538 | num_ops++; /* Also include a 'startsync' command. */ |
124e68e7 SW |
539 | |
540 | /* | |
541 | * we may need to do multiple writes here if we span an object | |
542 | * boundary. this isn't atomic, unfortunately. :( | |
543 | */ | |
544 | more: | |
d7f124f1 SW |
545 | io_align = pos & ~PAGE_MASK; |
546 | buf_align = (unsigned long)data & ~PAGE_MASK; | |
124e68e7 | 547 | len = left; |
3a42b6c4 | 548 | |
acead002 AE |
549 | snapc = ci->i_snap_realm->cached_context; |
550 | vino = ceph_vino(inode); | |
3d14c5d2 | 551 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, |
79528734 | 552 | vino, pos, &len, num_ops, |
acead002 | 553 | CEPH_OSD_OP_WRITE, flags, snapc, |
124e68e7 | 554 | ci->i_truncate_seq, ci->i_truncate_size, |
acead002 | 555 | false); |
6816282d SW |
556 | if (IS_ERR(req)) |
557 | return PTR_ERR(req); | |
124e68e7 | 558 | |
153e5167 AE |
559 | /* write from beginning of first page, regardless of io alignment */ |
560 | page_align = file->f_flags & O_DIRECT ? buf_align : io_align; | |
561 | num_pages = calc_pages_for(page_align, len); | |
124e68e7 | 562 | if (file->f_flags & O_DIRECT) { |
b6aa5901 | 563 | pages = ceph_get_direct_page_vector(data, num_pages, false); |
124e68e7 SW |
564 | if (IS_ERR(pages)) { |
565 | ret = PTR_ERR(pages); | |
566 | goto out; | |
567 | } | |
568 | ||
569 | /* | |
570 | * throw out any page cache pages in this range. this | |
571 | * may block. | |
572 | */ | |
213c99ee | 573 | truncate_inode_pages_range(inode->i_mapping, pos, |
5c6a2cdb | 574 | (pos+len) | (PAGE_CACHE_SIZE-1)); |
124e68e7 | 575 | } else { |
34d23762 | 576 | pages = ceph_alloc_page_vector(num_pages, GFP_NOFS); |
124e68e7 SW |
577 | if (IS_ERR(pages)) { |
578 | ret = PTR_ERR(pages); | |
579 | goto out; | |
580 | } | |
3d14c5d2 | 581 | ret = ceph_copy_user_to_page_vector(pages, data, pos, len); |
124e68e7 SW |
582 | if (ret < 0) { |
583 | ceph_release_page_vector(pages, num_pages); | |
584 | goto out; | |
585 | } | |
586 | ||
587 | if ((file->f_flags & O_SYNC) == 0) { | |
588 | /* get a second commit callback */ | |
26be8808 AE |
589 | req->r_unsafe_callback = ceph_sync_write_unsafe; |
590 | req->r_inode = inode; | |
43bfe5de | 591 | own_pages = true; |
124e68e7 SW |
592 | } |
593 | } | |
a4ce40a9 | 594 | osd_req_op_extent_osd_data_pages(req, 0, true, pages, len, |
8c042b0d | 595 | page_align, false, own_pages); |
124e68e7 | 596 | |
02ee07d3 | 597 | /* BUG_ON(vino.snap != CEPH_NOSNAP); */ |
79528734 | 598 | ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime); |
02ee07d3 | 599 | |
3d14c5d2 | 600 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
26be8808 | 601 | if (!ret) |
3d14c5d2 | 602 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); |
124e68e7 SW |
603 | |
604 | if (file->f_flags & O_DIRECT) | |
b6aa5901 | 605 | ceph_put_page_vector(pages, num_pages, false); |
124e68e7 SW |
606 | else if (file->f_flags & O_SYNC) |
607 | ceph_release_page_vector(pages, num_pages); | |
608 | ||
609 | out: | |
610 | ceph_osdc_put_request(req); | |
611 | if (ret == 0) { | |
612 | pos += len; | |
613 | written += len; | |
614 | left -= len; | |
022f3e2e | 615 | data += len; |
124e68e7 SW |
616 | if (left) |
617 | goto more; | |
618 | ||
619 | ret = written; | |
620 | *offset = pos; | |
621 | if (pos > i_size_read(inode)) | |
622 | check_caps = ceph_inode_set_size(inode, pos); | |
623 | if (check_caps) | |
624 | ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, | |
625 | NULL); | |
626 | } | |
627 | return ret; | |
628 | } | |
629 | ||
630 | /* | |
631 | * Wrap generic_file_aio_read with checks for cap bits on the inode. | |
632 | * Atomically grab references, so that those bits are not released | |
633 | * back to the MDS mid-read. | |
634 | * | |
635 | * Hmm, the sync read case isn't actually async... should it be? | |
636 | */ | |
637 | static ssize_t ceph_aio_read(struct kiocb *iocb, const struct iovec *iov, | |
638 | unsigned long nr_segs, loff_t pos) | |
639 | { | |
640 | struct file *filp = iocb->ki_filp; | |
2962507c | 641 | struct ceph_file_info *fi = filp->private_data; |
124e68e7 SW |
642 | loff_t *ppos = &iocb->ki_pos; |
643 | size_t len = iov->iov_len; | |
496ad9aa | 644 | struct inode *inode = file_inode(filp); |
124e68e7 | 645 | struct ceph_inode_info *ci = ceph_inode(inode); |
cd84db6e | 646 | void __user *base = iov->iov_base; |
124e68e7 | 647 | ssize_t ret; |
2962507c | 648 | int want, got = 0; |
6a026589 | 649 | int checkeof = 0, read = 0; |
124e68e7 SW |
650 | |
651 | dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n", | |
652 | inode, ceph_vinop(inode), pos, (unsigned)len, inode); | |
6a026589 | 653 | again: |
2962507c SW |
654 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
655 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
656 | else | |
657 | want = CEPH_CAP_FILE_CACHE; | |
658 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, &got, -1); | |
124e68e7 SW |
659 | if (ret < 0) |
660 | goto out; | |
661 | dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n", | |
662 | inode, ceph_vinop(inode), pos, (unsigned)len, | |
663 | ceph_cap_string(got)); | |
664 | ||
2962507c | 665 | if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 || |
124e68e7 | 666 | (iocb->ki_filp->f_flags & O_DIRECT) || |
4918b6d1 SW |
667 | (inode->i_sb->s_flags & MS_SYNCHRONOUS) || |
668 | (fi->flags & CEPH_F_SYNC)) | |
124e68e7 | 669 | /* hmm, this isn't really async... */ |
6a026589 | 670 | ret = ceph_sync_read(filp, base, len, ppos, &checkeof); |
124e68e7 SW |
671 | else |
672 | ret = generic_file_aio_read(iocb, iov, nr_segs, pos); | |
673 | ||
674 | out: | |
675 | dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n", | |
676 | inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret); | |
677 | ceph_put_cap_refs(ci, got); | |
6a026589 SW |
678 | |
679 | if (checkeof && ret >= 0) { | |
680 | int statret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE); | |
681 | ||
682 | /* hit EOF or hole? */ | |
683 | if (statret == 0 && *ppos < inode->i_size) { | |
c3cd6283 | 684 | dout("aio_read sync_read hit hole, ppos %lld < size %lld, reading more\n", *ppos, inode->i_size); |
6a026589 SW |
685 | read += ret; |
686 | base += ret; | |
687 | len -= ret; | |
688 | checkeof = 0; | |
689 | goto again; | |
690 | } | |
691 | } | |
692 | if (ret >= 0) | |
693 | ret += read; | |
694 | ||
124e68e7 SW |
695 | return ret; |
696 | } | |
697 | ||
698 | /* | |
699 | * Take cap references to avoid releasing caps to MDS mid-write. | |
700 | * | |
701 | * If we are synchronous, and write with an old snap context, the OSD | |
702 | * may return EOLDSNAPC. In that case, retry the write.. _after_ | |
703 | * dropping our cap refs and allowing the pending snap to logically | |
704 | * complete _before_ this write occurs. | |
705 | * | |
706 | * If we are near ENOSPC, write synchronously. | |
707 | */ | |
708 | static ssize_t ceph_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
709 | unsigned long nr_segs, loff_t pos) | |
710 | { | |
711 | struct file *file = iocb->ki_filp; | |
33caad32 | 712 | struct ceph_file_info *fi = file->private_data; |
496ad9aa | 713 | struct inode *inode = file_inode(file); |
124e68e7 | 714 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 YS |
715 | struct ceph_osd_client *osdc = |
716 | &ceph_sb_to_client(inode->i_sb)->client->osdc; | |
124e68e7 | 717 | loff_t endoff = pos + iov->iov_len; |
7971bd92 SW |
718 | int want, got = 0; |
719 | int ret, err; | |
124e68e7 SW |
720 | |
721 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
722 | return -EROFS; | |
723 | ||
6070e0c1 | 724 | sb_start_write(inode->i_sb); |
124e68e7 | 725 | retry_snap: |
6070e0c1 YZ |
726 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) { |
727 | ret = -ENOSPC; | |
728 | goto out; | |
729 | } | |
37505d57 | 730 | mutex_lock(&inode->i_mutex); |
7971bd92 SW |
731 | dout("aio_write %p %llx.%llx %llu~%u getting caps. i_size %llu\n", |
732 | inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len, | |
733 | inode->i_size); | |
734 | if (fi->fmode & CEPH_FILE_MODE_LAZY) | |
735 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
736 | else | |
737 | want = CEPH_CAP_FILE_BUFFER; | |
738 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, endoff); | |
37505d57 YZ |
739 | if (ret < 0) { |
740 | mutex_unlock(&inode->i_mutex); | |
741 | goto out; | |
742 | } | |
124e68e7 | 743 | |
7971bd92 SW |
744 | dout("aio_write %p %llx.%llx %llu~%u got cap refs on %s\n", |
745 | inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len, | |
746 | ceph_cap_string(got)); | |
747 | ||
748 | if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 || | |
749 | (iocb->ki_filp->f_flags & O_DIRECT) || | |
750 | (inode->i_sb->s_flags & MS_SYNCHRONOUS) || | |
751 | (fi->flags & CEPH_F_SYNC)) { | |
37505d57 | 752 | mutex_unlock(&inode->i_mutex); |
7971bd92 SW |
753 | ret = ceph_sync_write(file, iov->iov_base, iov->iov_len, |
754 | &iocb->ki_pos); | |
755 | } else { | |
6070e0c1 YZ |
756 | ret = __generic_file_aio_write(iocb, iov, nr_segs, |
757 | &iocb->ki_pos); | |
758 | mutex_unlock(&inode->i_mutex); | |
7971bd92 | 759 | } |
d8de9ab6 | 760 | |
124e68e7 | 761 | if (ret >= 0) { |
fca65b4a | 762 | int dirty; |
be655596 | 763 | spin_lock(&ci->i_ceph_lock); |
fca65b4a | 764 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR); |
be655596 | 765 | spin_unlock(&ci->i_ceph_lock); |
fca65b4a SW |
766 | if (dirty) |
767 | __mark_inode_dirty(inode, dirty); | |
124e68e7 | 768 | } |
7971bd92 | 769 | |
124e68e7 | 770 | dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n", |
7971bd92 SW |
771 | inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len, |
772 | ceph_cap_string(got)); | |
124e68e7 | 773 | ceph_put_cap_refs(ci, got); |
7971bd92 | 774 | |
6070e0c1 YZ |
775 | if (ret >= 0 && |
776 | ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) || | |
777 | ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) { | |
778 | err = vfs_fsync_range(file, pos, pos + ret - 1, 1); | |
779 | if (err < 0) | |
780 | ret = err; | |
781 | } | |
d8de9ab6 | 782 | out: |
124e68e7 SW |
783 | if (ret == -EOLDSNAPC) { |
784 | dout("aio_write %p %llx.%llx %llu~%u got EOLDSNAPC, retrying\n", | |
785 | inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len); | |
786 | goto retry_snap; | |
787 | } | |
6070e0c1 | 788 | sb_end_write(inode->i_sb); |
124e68e7 SW |
789 | |
790 | return ret; | |
791 | } | |
792 | ||
793 | /* | |
794 | * llseek. be sure to verify file size on SEEK_END. | |
795 | */ | |
965c8e59 | 796 | static loff_t ceph_llseek(struct file *file, loff_t offset, int whence) |
124e68e7 SW |
797 | { |
798 | struct inode *inode = file->f_mapping->host; | |
799 | int ret; | |
800 | ||
801 | mutex_lock(&inode->i_mutex); | |
3f99969f | 802 | __ceph_do_pending_vmtruncate(inode, false); |
6a82c47a | 803 | |
965c8e59 | 804 | if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) { |
124e68e7 SW |
805 | ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE); |
806 | if (ret < 0) { | |
807 | offset = ret; | |
808 | goto out; | |
809 | } | |
06222e49 JB |
810 | } |
811 | ||
965c8e59 | 812 | switch (whence) { |
06222e49 | 813 | case SEEK_END: |
124e68e7 SW |
814 | offset += inode->i_size; |
815 | break; | |
816 | case SEEK_CUR: | |
817 | /* | |
818 | * Here we special-case the lseek(fd, 0, SEEK_CUR) | |
819 | * position-querying operation. Avoid rewriting the "same" | |
820 | * f_pos value back to the file because a concurrent read(), | |
821 | * write() or lseek() might have altered it | |
822 | */ | |
823 | if (offset == 0) { | |
824 | offset = file->f_pos; | |
825 | goto out; | |
826 | } | |
827 | offset += file->f_pos; | |
828 | break; | |
06222e49 JB |
829 | case SEEK_DATA: |
830 | if (offset >= inode->i_size) { | |
831 | ret = -ENXIO; | |
832 | goto out; | |
833 | } | |
834 | break; | |
835 | case SEEK_HOLE: | |
836 | if (offset >= inode->i_size) { | |
837 | ret = -ENXIO; | |
838 | goto out; | |
839 | } | |
840 | offset = inode->i_size; | |
841 | break; | |
124e68e7 SW |
842 | } |
843 | ||
844 | if (offset < 0 || offset > inode->i_sb->s_maxbytes) { | |
845 | offset = -EINVAL; | |
846 | goto out; | |
847 | } | |
848 | ||
849 | /* Special lock needed here? */ | |
850 | if (offset != file->f_pos) { | |
851 | file->f_pos = offset; | |
852 | file->f_version = 0; | |
853 | } | |
854 | ||
855 | out: | |
856 | mutex_unlock(&inode->i_mutex); | |
857 | return offset; | |
858 | } | |
859 | ||
860 | const struct file_operations ceph_file_fops = { | |
861 | .open = ceph_open, | |
862 | .release = ceph_release, | |
863 | .llseek = ceph_llseek, | |
864 | .read = do_sync_read, | |
865 | .write = do_sync_write, | |
866 | .aio_read = ceph_aio_read, | |
867 | .aio_write = ceph_aio_write, | |
868 | .mmap = ceph_mmap, | |
869 | .fsync = ceph_fsync, | |
40819f6f GF |
870 | .lock = ceph_lock, |
871 | .flock = ceph_flock, | |
124e68e7 SW |
872 | .splice_read = generic_file_splice_read, |
873 | .splice_write = generic_file_splice_write, | |
874 | .unlocked_ioctl = ceph_ioctl, | |
875 | .compat_ioctl = ceph_ioctl, | |
876 | }; | |
877 |