]>
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> | |
ad7a60de | 10 | #include <linux/falloc.h> |
124e68e7 SW |
11 | |
12 | #include "super.h" | |
13 | #include "mds_client.h" | |
99ccbd22 | 14 | #include "cache.h" |
124e68e7 SW |
15 | |
16 | /* | |
17 | * Ceph file operations | |
18 | * | |
19 | * Implement basic open/close functionality, and implement | |
20 | * read/write. | |
21 | * | |
22 | * We implement three modes of file I/O: | |
23 | * - buffered uses the generic_file_aio_{read,write} helpers | |
24 | * | |
25 | * - synchronous is used when there is multi-client read/write | |
26 | * sharing, avoids the page cache, and synchronously waits for an | |
27 | * ack from the OSD. | |
28 | * | |
29 | * - direct io takes the variant of the sync path that references | |
30 | * user pages directly. | |
31 | * | |
32 | * fsync() flushes and waits on dirty pages, but just queues metadata | |
33 | * for writeback: since the MDS can recover size and mtime there is no | |
34 | * need to wait for MDS acknowledgement. | |
35 | */ | |
36 | ||
b5b98989 ZC |
37 | /* |
38 | * Calculate the length sum of direct io vectors that can | |
39 | * be combined into one page vector. | |
40 | */ | |
41 | static size_t dio_get_pagev_size(const struct iov_iter *it) | |
42 | { | |
43 | const struct iovec *iov = it->iov; | |
44 | const struct iovec *iovend = iov + it->nr_segs; | |
45 | size_t size; | |
46 | ||
47 | size = iov->iov_len - it->iov_offset; | |
48 | /* | |
49 | * An iov can be page vectored when both the current tail | |
50 | * and the next base are page aligned. | |
51 | */ | |
52 | while (PAGE_ALIGNED((iov->iov_base + iov->iov_len)) && | |
53 | (++iov < iovend && PAGE_ALIGNED((iov->iov_base)))) { | |
54 | size += iov->iov_len; | |
55 | } | |
56 | dout("dio_get_pagevlen len = %zu\n", size); | |
57 | return size; | |
58 | } | |
59 | ||
60 | /* | |
61 | * Allocate a page vector based on (@it, @nbytes). | |
62 | * The return value is the tuple describing a page vector, | |
63 | * that is (@pages, @page_align, @num_pages). | |
64 | */ | |
65 | static struct page ** | |
66 | dio_get_pages_alloc(const struct iov_iter *it, size_t nbytes, | |
67 | size_t *page_align, int *num_pages) | |
68 | { | |
69 | struct iov_iter tmp_it = *it; | |
70 | size_t align; | |
71 | struct page **pages; | |
72 | int ret = 0, idx, npages; | |
73 | ||
74 | align = (unsigned long)(it->iov->iov_base + it->iov_offset) & | |
75 | (PAGE_SIZE - 1); | |
76 | npages = calc_pages_for(align, nbytes); | |
77 | pages = kmalloc(sizeof(*pages) * npages, GFP_KERNEL); | |
78 | if (!pages) { | |
79 | pages = vmalloc(sizeof(*pages) * npages); | |
80 | if (!pages) | |
81 | return ERR_PTR(-ENOMEM); | |
82 | } | |
83 | ||
84 | for (idx = 0; idx < npages; ) { | |
85 | size_t start; | |
86 | ret = iov_iter_get_pages(&tmp_it, pages + idx, nbytes, | |
87 | npages - idx, &start); | |
88 | if (ret < 0) | |
89 | goto fail; | |
90 | ||
91 | iov_iter_advance(&tmp_it, ret); | |
92 | nbytes -= ret; | |
93 | idx += (ret + start + PAGE_SIZE - 1) / PAGE_SIZE; | |
94 | } | |
95 | ||
96 | BUG_ON(nbytes != 0); | |
97 | *num_pages = npages; | |
98 | *page_align = align; | |
99 | dout("dio_get_pages_alloc: got %d pages align %zu\n", npages, align); | |
100 | return pages; | |
101 | fail: | |
102 | ceph_put_page_vector(pages, idx, false); | |
103 | return ERR_PTR(ret); | |
104 | } | |
124e68e7 SW |
105 | |
106 | /* | |
107 | * Prepare an open request. Preallocate ceph_cap to avoid an | |
108 | * inopportune ENOMEM later. | |
109 | */ | |
110 | static struct ceph_mds_request * | |
111 | prepare_open_request(struct super_block *sb, int flags, int create_mode) | |
112 | { | |
3d14c5d2 YS |
113 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); |
114 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
115 | struct ceph_mds_request *req; |
116 | int want_auth = USE_ANY_MDS; | |
117 | int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN; | |
118 | ||
119 | if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC)) | |
120 | want_auth = USE_AUTH_MDS; | |
121 | ||
122 | req = ceph_mdsc_create_request(mdsc, op, want_auth); | |
123 | if (IS_ERR(req)) | |
124 | goto out; | |
125 | req->r_fmode = ceph_flags_to_mode(flags); | |
126 | req->r_args.open.flags = cpu_to_le32(flags); | |
127 | req->r_args.open.mode = cpu_to_le32(create_mode); | |
124e68e7 SW |
128 | out: |
129 | return req; | |
130 | } | |
131 | ||
132 | /* | |
133 | * initialize private struct file data. | |
134 | * if we fail, clean up by dropping fmode reference on the ceph_inode | |
135 | */ | |
136 | static int ceph_init_file(struct inode *inode, struct file *file, int fmode) | |
137 | { | |
138 | struct ceph_file_info *cf; | |
139 | int ret = 0; | |
99ccbd22 MT |
140 | struct ceph_inode_info *ci = ceph_inode(inode); |
141 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); | |
142 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
143 | |
144 | switch (inode->i_mode & S_IFMT) { | |
145 | case S_IFREG: | |
99ccbd22 MT |
146 | /* First file open request creates the cookie, we want to keep |
147 | * this cookie around for the filetime of the inode as not to | |
148 | * have to worry about fscache register / revoke / operation | |
149 | * races. | |
150 | * | |
151 | * Also, if we know the operation is going to invalidate data | |
152 | * (non readonly) just nuke the cache right away. | |
153 | */ | |
154 | ceph_fscache_register_inode_cookie(mdsc->fsc, ci); | |
155 | if ((fmode & CEPH_FILE_MODE_WR)) | |
156 | ceph_fscache_invalidate(inode); | |
124e68e7 SW |
157 | case S_IFDIR: |
158 | dout("init_file %p %p 0%o (regular)\n", inode, file, | |
159 | inode->i_mode); | |
687265e5 | 160 | cf = kmem_cache_alloc(ceph_file_cachep, GFP_KERNEL | __GFP_ZERO); |
124e68e7 SW |
161 | if (cf == NULL) { |
162 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
163 | return -ENOMEM; | |
164 | } | |
165 | cf->fmode = fmode; | |
166 | cf->next_offset = 2; | |
fdd4e158 | 167 | cf->readdir_cache_idx = -1; |
124e68e7 SW |
168 | file->private_data = cf; |
169 | BUG_ON(inode->i_fop->release != ceph_release); | |
170 | break; | |
171 | ||
172 | case S_IFLNK: | |
173 | dout("init_file %p %p 0%o (symlink)\n", inode, file, | |
174 | inode->i_mode); | |
175 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
176 | break; | |
177 | ||
178 | default: | |
179 | dout("init_file %p %p 0%o (special)\n", inode, file, | |
180 | inode->i_mode); | |
181 | /* | |
182 | * we need to drop the open ref now, since we don't | |
183 | * have .release set to ceph_release. | |
184 | */ | |
185 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ | |
186 | BUG_ON(inode->i_fop->release == ceph_release); | |
187 | ||
188 | /* call the proper open fop */ | |
189 | ret = inode->i_fop->open(inode, file); | |
190 | } | |
191 | return ret; | |
192 | } | |
193 | ||
194 | /* | |
124e68e7 SW |
195 | * If we already have the requisite capabilities, we can satisfy |
196 | * the open request locally (no need to request new caps from the | |
197 | * MDS). We do, however, need to inform the MDS (asynchronously) | |
198 | * if our wanted caps set expands. | |
199 | */ | |
200 | int ceph_open(struct inode *inode, struct file *file) | |
201 | { | |
202 | struct ceph_inode_info *ci = ceph_inode(inode); | |
3d14c5d2 YS |
203 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
204 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 SW |
205 | struct ceph_mds_request *req; |
206 | struct ceph_file_info *cf = file->private_data; | |
124e68e7 SW |
207 | int err; |
208 | int flags, fmode, wanted; | |
209 | ||
210 | if (cf) { | |
211 | dout("open file %p is already opened\n", file); | |
212 | return 0; | |
213 | } | |
214 | ||
215 | /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */ | |
216 | flags = file->f_flags & ~(O_CREAT|O_EXCL); | |
217 | if (S_ISDIR(inode->i_mode)) | |
218 | flags = O_DIRECTORY; /* mds likes to know */ | |
219 | ||
220 | dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode, | |
221 | ceph_vinop(inode), file, flags, file->f_flags); | |
222 | fmode = ceph_flags_to_mode(flags); | |
223 | wanted = ceph_caps_for_mode(fmode); | |
224 | ||
225 | /* snapped files are read-only */ | |
226 | if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE)) | |
227 | return -EROFS; | |
228 | ||
229 | /* trivially open snapdir */ | |
230 | if (ceph_snap(inode) == CEPH_SNAPDIR) { | |
be655596 | 231 | spin_lock(&ci->i_ceph_lock); |
124e68e7 | 232 | __ceph_get_fmode(ci, fmode); |
be655596 | 233 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
234 | return ceph_init_file(inode, file, fmode); |
235 | } | |
236 | ||
237 | /* | |
7421ab80 SW |
238 | * No need to block if we have caps on the auth MDS (for |
239 | * write) or any MDS (for read). Update wanted set | |
124e68e7 SW |
240 | * asynchronously. |
241 | */ | |
be655596 | 242 | spin_lock(&ci->i_ceph_lock); |
7421ab80 SW |
243 | if (__ceph_is_any_real_caps(ci) && |
244 | (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) { | |
124e68e7 SW |
245 | int mds_wanted = __ceph_caps_mds_wanted(ci); |
246 | int issued = __ceph_caps_issued(ci, NULL); | |
247 | ||
248 | dout("open %p fmode %d want %s issued %s using existing\n", | |
249 | inode, fmode, ceph_cap_string(wanted), | |
250 | ceph_cap_string(issued)); | |
251 | __ceph_get_fmode(ci, fmode); | |
be655596 | 252 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
253 | |
254 | /* adjust wanted? */ | |
255 | if ((issued & wanted) != wanted && | |
256 | (mds_wanted & wanted) != wanted && | |
257 | ceph_snap(inode) != CEPH_SNAPDIR) | |
258 | ceph_check_caps(ci, 0, NULL); | |
259 | ||
260 | return ceph_init_file(inode, file, fmode); | |
261 | } else if (ceph_snap(inode) != CEPH_NOSNAP && | |
262 | (ci->i_snap_caps & wanted) == wanted) { | |
263 | __ceph_get_fmode(ci, fmode); | |
be655596 | 264 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
265 | return ceph_init_file(inode, file, fmode); |
266 | } | |
99ccbd22 | 267 | |
be655596 | 268 | spin_unlock(&ci->i_ceph_lock); |
124e68e7 SW |
269 | |
270 | dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted)); | |
271 | req = prepare_open_request(inode->i_sb, flags, 0); | |
272 | if (IS_ERR(req)) { | |
273 | err = PTR_ERR(req); | |
274 | goto out; | |
275 | } | |
70b666c3 SW |
276 | req->r_inode = inode; |
277 | ihold(inode); | |
99ccbd22 | 278 | |
124e68e7 | 279 | req->r_num_caps = 1; |
e36d571d | 280 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
124e68e7 SW |
281 | if (!err) |
282 | err = ceph_init_file(inode, file, req->r_fmode); | |
283 | ceph_mdsc_put_request(req); | |
284 | dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode)); | |
285 | out: | |
286 | return err; | |
287 | } | |
288 | ||
289 | ||
290 | /* | |
5ef50c3b SW |
291 | * Do a lookup + open with a single request. If we get a non-existent |
292 | * file or symlink, return 1 so the VFS can retry. | |
124e68e7 | 293 | */ |
5ef50c3b | 294 | int ceph_atomic_open(struct inode *dir, struct dentry *dentry, |
30d90494 | 295 | struct file *file, unsigned flags, umode_t mode, |
d9585277 | 296 | int *opened) |
124e68e7 | 297 | { |
3d14c5d2 YS |
298 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
299 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
124e68e7 | 300 | struct ceph_mds_request *req; |
5ef50c3b | 301 | struct dentry *dn; |
b1ee94aa | 302 | struct ceph_acls_info acls = {}; |
124e68e7 | 303 | int err; |
124e68e7 | 304 | |
a455589f AV |
305 | dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n", |
306 | dir, dentry, dentry, | |
5ef50c3b SW |
307 | d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode); |
308 | ||
309 | if (dentry->d_name.len > NAME_MAX) | |
310 | return -ENAMETOOLONG; | |
311 | ||
312 | err = ceph_init_dentry(dentry); | |
313 | if (err < 0) | |
314 | return err; | |
124e68e7 | 315 | |
b1ee94aa YZ |
316 | if (flags & O_CREAT) { |
317 | err = ceph_pre_init_acls(dir, &mode, &acls); | |
318 | if (err < 0) | |
319 | return err; | |
320 | } | |
321 | ||
124e68e7 SW |
322 | /* do the open */ |
323 | req = prepare_open_request(dir->i_sb, flags, mode); | |
b1ee94aa YZ |
324 | if (IS_ERR(req)) { |
325 | err = PTR_ERR(req); | |
326 | goto out_acl; | |
327 | } | |
124e68e7 SW |
328 | req->r_dentry = dget(dentry); |
329 | req->r_num_caps = 2; | |
330 | if (flags & O_CREAT) { | |
331 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
332 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
b1ee94aa YZ |
333 | if (acls.pagelist) { |
334 | req->r_pagelist = acls.pagelist; | |
335 | acls.pagelist = NULL; | |
336 | } | |
124e68e7 SW |
337 | } |
338 | req->r_locked_dir = dir; /* caller holds dir->i_mutex */ | |
acda7657 SW |
339 | err = ceph_mdsc_do_request(mdsc, |
340 | (flags & (O_CREAT|O_TRUNC)) ? dir : NULL, | |
341 | req); | |
bf91c315 | 342 | err = ceph_handle_snapdir(req, dentry, err); |
79aec984 | 343 | if (err) |
b1ee94aa | 344 | goto out_req; |
79aec984 | 345 | |
a43137f7 | 346 | if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry) |
124e68e7 | 347 | err = ceph_handle_notrace_create(dir, dentry); |
2d83bde9 | 348 | |
5ef50c3b SW |
349 | if (d_unhashed(dentry)) { |
350 | dn = ceph_finish_lookup(req, dentry, err); | |
351 | if (IS_ERR(dn)) | |
352 | err = PTR_ERR(dn); | |
353 | } else { | |
354 | /* we were given a hashed negative dentry */ | |
355 | dn = NULL; | |
356 | } | |
357 | if (err) | |
b1ee94aa | 358 | goto out_req; |
2b0143b5 | 359 | if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) { |
5ef50c3b SW |
360 | /* make vfs retry on splice, ENOENT, or symlink */ |
361 | dout("atomic_open finish_no_open on dn %p\n", dn); | |
362 | err = finish_no_open(file, dn); | |
363 | } else { | |
364 | dout("atomic_open finish_open on dn %p\n", dn); | |
6e8575fa | 365 | if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) { |
2b0143b5 | 366 | ceph_init_inode_acls(d_inode(dentry), &acls); |
6e8575fa SL |
367 | *opened |= FILE_CREATED; |
368 | } | |
5ef50c3b SW |
369 | err = finish_open(file, dentry, ceph_open, opened); |
370 | } | |
b1ee94aa | 371 | out_req: |
ab866549 YZ |
372 | if (!req->r_err && req->r_target_inode) |
373 | ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode); | |
5ef50c3b | 374 | ceph_mdsc_put_request(req); |
b1ee94aa YZ |
375 | out_acl: |
376 | ceph_release_acls_info(&acls); | |
5ef50c3b | 377 | dout("atomic_open result=%d\n", err); |
d9585277 | 378 | return err; |
124e68e7 SW |
379 | } |
380 | ||
381 | int ceph_release(struct inode *inode, struct file *file) | |
382 | { | |
383 | struct ceph_inode_info *ci = ceph_inode(inode); | |
384 | struct ceph_file_info *cf = file->private_data; | |
385 | ||
386 | dout("release inode %p file %p\n", inode, file); | |
387 | ceph_put_fmode(ci, cf->fmode); | |
388 | if (cf->last_readdir) | |
389 | ceph_mdsc_put_request(cf->last_readdir); | |
390 | kfree(cf->last_name); | |
391 | kfree(cf->dir_info); | |
124e68e7 | 392 | kmem_cache_free(ceph_file_cachep, cf); |
195d3ce2 SW |
393 | |
394 | /* wake up anyone waiting for caps on this inode */ | |
03066f23 | 395 | wake_up_all(&ci->i_cap_wq); |
124e68e7 SW |
396 | return 0; |
397 | } | |
398 | ||
83701246 YZ |
399 | enum { |
400 | CHECK_EOF = 1, | |
401 | READ_INLINE = 2, | |
402 | }; | |
403 | ||
124e68e7 SW |
404 | /* |
405 | * Read a range of bytes striped over one or more objects. Iterate over | |
406 | * objects we stripe over. (That's not atomic, but good enough for now.) | |
407 | * | |
408 | * If we get a short result from the OSD, check against i_size; we need to | |
409 | * only return a short read to the caller if we hit EOF. | |
410 | */ | |
411 | static int striped_read(struct inode *inode, | |
412 | u64 off, u64 len, | |
6a026589 | 413 | struct page **pages, int num_pages, |
c3cd6283 | 414 | int *checkeof, bool o_direct, |
ab226e21 | 415 | unsigned long buf_align) |
124e68e7 | 416 | { |
3d14c5d2 | 417 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
124e68e7 | 418 | struct ceph_inode_info *ci = ceph_inode(inode); |
688bac46 | 419 | u64 pos, this_len, left; |
b7495fc2 | 420 | int io_align, page_align; |
688bac46 | 421 | int pages_left; |
124e68e7 SW |
422 | int read; |
423 | struct page **page_pos; | |
424 | int ret; | |
425 | bool hit_stripe, was_short; | |
426 | ||
427 | /* | |
428 | * we may need to do multiple reads. not atomic, unfortunately. | |
429 | */ | |
430 | pos = off; | |
431 | left = len; | |
432 | page_pos = pages; | |
433 | pages_left = num_pages; | |
434 | read = 0; | |
b7495fc2 | 435 | io_align = off & ~PAGE_MASK; |
124e68e7 SW |
436 | |
437 | more: | |
c3cd6283 | 438 | if (o_direct) |
ab226e21 | 439 | page_align = (pos - io_align + buf_align) & ~PAGE_MASK; |
b7495fc2 SW |
440 | else |
441 | page_align = pos & ~PAGE_MASK; | |
124e68e7 | 442 | this_len = left; |
3d14c5d2 | 443 | ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode), |
124e68e7 SW |
444 | &ci->i_layout, pos, &this_len, |
445 | ci->i_truncate_seq, | |
446 | ci->i_truncate_size, | |
b7495fc2 | 447 | page_pos, pages_left, page_align); |
124e68e7 SW |
448 | if (ret == -ENOENT) |
449 | ret = 0; | |
0e98728f SW |
450 | hit_stripe = this_len < left; |
451 | was_short = ret >= 0 && ret < this_len; | |
688bac46 | 452 | dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read, |
124e68e7 SW |
453 | ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : ""); |
454 | ||
02ae66d8 | 455 | if (ret >= 0) { |
456 | int didpages; | |
457 | if (was_short && (pos + ret < inode->i_size)) { | |
1487a688 YZ |
458 | int zlen = min(this_len - ret, |
459 | inode->i_size - pos - ret); | |
460 | int zoff = (o_direct ? buf_align : io_align) + | |
461 | read + ret; | |
02ae66d8 | 462 | dout(" zero gap %llu to %llu\n", |
1487a688 YZ |
463 | pos + ret, pos + ret + zlen); |
464 | ceph_zero_page_vector_range(zoff, zlen, pages); | |
465 | ret += zlen; | |
124e68e7 | 466 | } |
02ae66d8 | 467 | |
468 | didpages = (page_align + ret) >> PAGE_CACHE_SHIFT; | |
124e68e7 SW |
469 | pos += ret; |
470 | read = pos - off; | |
471 | left -= ret; | |
472 | page_pos += didpages; | |
473 | pages_left -= didpages; | |
474 | ||
02ae66d8 | 475 | /* hit stripe and need continue*/ |
476 | if (left && hit_stripe && pos < inode->i_size) | |
124e68e7 SW |
477 | goto more; |
478 | } | |
479 | ||
ee7289bf | 480 | if (read > 0) { |
02ae66d8 | 481 | ret = read; |
c3cd6283 SW |
482 | /* did we bounce off eof? */ |
483 | if (pos + left > inode->i_size) | |
83701246 | 484 | *checkeof = CHECK_EOF; |
124e68e7 SW |
485 | } |
486 | ||
124e68e7 SW |
487 | dout("striped_read returns %d\n", ret); |
488 | return ret; | |
489 | } | |
490 | ||
491 | /* | |
492 | * Completely synchronous read and write methods. Direct from __user | |
493 | * buffer to osd, or directly to user pages (if O_DIRECT). | |
494 | * | |
495 | * If the read spans object boundary, just do multiple reads. | |
496 | */ | |
8eb4efb0 | 497 | static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i, |
498 | int *checkeof) | |
124e68e7 | 499 | { |
8eb4efb0 | 500 | struct file *file = iocb->ki_filp; |
496ad9aa | 501 | struct inode *inode = file_inode(file); |
124e68e7 | 502 | struct page **pages; |
8eb4efb0 | 503 | u64 off = iocb->ki_pos; |
ab226e21 | 504 | int num_pages, ret; |
2b777c9d | 505 | size_t len = iov_iter_count(i); |
124e68e7 | 506 | |
8eb4efb0 | 507 | dout("sync_read on file %p %llu~%u %s\n", file, off, |
508 | (unsigned)len, | |
124e68e7 | 509 | (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); |
d0d0db22 YZ |
510 | |
511 | if (!len) | |
512 | return 0; | |
e98b6fed SW |
513 | /* |
514 | * flush any page cache pages in this range. this | |
515 | * will make concurrent normal and sync io slow, | |
516 | * but it will at least behave sensibly when they are | |
517 | * in sequence. | |
518 | */ | |
8eb4efb0 | 519 | ret = filemap_write_and_wait_range(inode->i_mapping, off, |
520 | off + len); | |
29065a51 | 521 | if (ret < 0) |
8eb4efb0 | 522 | return ret; |
29065a51 | 523 | |
2ba48ce5 | 524 | if (iocb->ki_flags & IOCB_DIRECT) { |
8eb4efb0 | 525 | while (iov_iter_count(i)) { |
2b777c9d AV |
526 | size_t start; |
527 | ssize_t n; | |
8eb4efb0 | 528 | |
b5b98989 ZC |
529 | n = dio_get_pagev_size(i); |
530 | pages = dio_get_pages_alloc(i, n, &start, &num_pages); | |
531 | if (IS_ERR(pages)) | |
532 | return PTR_ERR(pages); | |
2b777c9d AV |
533 | |
534 | ret = striped_read(inode, off, n, | |
8eb4efb0 | 535 | pages, num_pages, checkeof, |
2b777c9d AV |
536 | 1, start); |
537 | ||
8eb4efb0 | 538 | ceph_put_page_vector(pages, num_pages, true); |
539 | ||
540 | if (ret <= 0) | |
541 | break; | |
542 | off += ret; | |
543 | iov_iter_advance(i, ret); | |
2b777c9d | 544 | if (ret < n) |
8eb4efb0 | 545 | break; |
546 | } | |
547 | } else { | |
548 | num_pages = calc_pages_for(off, len); | |
687265e5 | 549 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); |
8eb4efb0 | 550 | if (IS_ERR(pages)) |
551 | return PTR_ERR(pages); | |
552 | ret = striped_read(inode, off, len, pages, | |
553 | num_pages, checkeof, 0, 0); | |
554 | if (ret > 0) { | |
555 | int l, k = 0; | |
2b777c9d | 556 | size_t left = ret; |
8eb4efb0 | 557 | |
558 | while (left) { | |
5aaa432a YZ |
559 | size_t page_off = off & ~PAGE_MASK; |
560 | size_t copy = min_t(size_t, | |
561 | PAGE_SIZE - page_off, left); | |
562 | l = copy_page_to_iter(pages[k++], page_off, | |
563 | copy, i); | |
2b777c9d AV |
564 | off += l; |
565 | left -= l; | |
566 | if (l < copy) | |
8eb4efb0 | 567 | break; |
568 | } | |
569 | } | |
570 | ceph_release_page_vector(pages, num_pages); | |
571 | } | |
124e68e7 | 572 | |
8eb4efb0 | 573 | if (off > iocb->ki_pos) { |
574 | ret = off - iocb->ki_pos; | |
575 | iocb->ki_pos = off; | |
576 | } | |
124e68e7 | 577 | |
124e68e7 SW |
578 | dout("sync_read result %d\n", ret); |
579 | return ret; | |
580 | } | |
581 | ||
582 | /* | |
26be8808 AE |
583 | * Write commit request unsafe callback, called to tell us when a |
584 | * request is unsafe (that is, in flight--has been handed to the | |
585 | * messenger to send to its target osd). It is called again when | |
586 | * we've received a response message indicating the request is | |
587 | * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request | |
588 | * is completed early (and unsuccessfully) due to a timeout or | |
589 | * interrupt. | |
590 | * | |
591 | * This is used if we requested both an ACK and ONDISK commit reply | |
592 | * from the OSD. | |
124e68e7 | 593 | */ |
26be8808 | 594 | static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe) |
124e68e7 SW |
595 | { |
596 | struct ceph_inode_info *ci = ceph_inode(req->r_inode); | |
597 | ||
26be8808 AE |
598 | dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid, |
599 | unsafe ? "un" : ""); | |
600 | if (unsafe) { | |
601 | ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR); | |
602 | spin_lock(&ci->i_unsafe_lock); | |
603 | list_add_tail(&req->r_unsafe_item, | |
604 | &ci->i_unsafe_writes); | |
605 | spin_unlock(&ci->i_unsafe_lock); | |
606 | } else { | |
607 | spin_lock(&ci->i_unsafe_lock); | |
608 | list_del_init(&req->r_unsafe_item); | |
609 | spin_unlock(&ci->i_unsafe_lock); | |
610 | ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR); | |
611 | } | |
124e68e7 SW |
612 | } |
613 | ||
e8344e66 | 614 | |
124e68e7 | 615 | /* |
e8344e66 | 616 | * Synchronous write, straight from __user pointer or user pages. |
124e68e7 SW |
617 | * |
618 | * If write spans object boundary, just do multiple writes. (For a | |
619 | * correct atomic write, we should e.g. take write locks on all | |
620 | * objects, rollback on failure, etc.) | |
621 | */ | |
e8344e66 | 622 | static ssize_t |
5dda377c YZ |
623 | ceph_sync_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos, |
624 | struct ceph_snap_context *snapc) | |
124e68e7 | 625 | { |
e8344e66 | 626 | struct file *file = iocb->ki_filp; |
496ad9aa | 627 | struct inode *inode = file_inode(file); |
124e68e7 | 628 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 | 629 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
acead002 | 630 | struct ceph_vino vino; |
124e68e7 SW |
631 | struct ceph_osd_request *req; |
632 | struct page **pages; | |
633 | int num_pages; | |
124e68e7 SW |
634 | int written = 0; |
635 | int flags; | |
124e68e7 SW |
636 | int check_caps = 0; |
637 | int ret; | |
638 | struct timespec mtime = CURRENT_TIME; | |
4908b822 | 639 | size_t count = iov_iter_count(from); |
124e68e7 | 640 | |
496ad9aa | 641 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) |
124e68e7 SW |
642 | return -EROFS; |
643 | ||
e8344e66 | 644 | dout("sync_direct_write on file %p %lld~%u\n", file, pos, |
645 | (unsigned)count); | |
124e68e7 | 646 | |
e8344e66 | 647 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); |
29065a51 YS |
648 | if (ret < 0) |
649 | return ret; | |
650 | ||
651 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
652 | pos >> PAGE_CACHE_SHIFT, | |
e8344e66 | 653 | (pos + count) >> PAGE_CACHE_SHIFT); |
29065a51 YS |
654 | if (ret < 0) |
655 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
656 | ||
124e68e7 SW |
657 | flags = CEPH_OSD_FLAG_ORDERSNAP | |
658 | CEPH_OSD_FLAG_ONDISK | | |
659 | CEPH_OSD_FLAG_WRITE; | |
124e68e7 | 660 | |
4908b822 | 661 | while (iov_iter_count(from) > 0) { |
b5b98989 | 662 | u64 len = dio_get_pagev_size(from); |
64c31311 AV |
663 | size_t start; |
664 | ssize_t n; | |
e8344e66 | 665 | |
e8344e66 | 666 | vino = ceph_vino(inode); |
667 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
715e4cd4 | 668 | vino, pos, &len, 0, |
e8344e66 | 669 | 2,/*include a 'startsync' command*/ |
670 | CEPH_OSD_OP_WRITE, flags, snapc, | |
671 | ci->i_truncate_seq, | |
672 | ci->i_truncate_size, | |
673 | false); | |
674 | if (IS_ERR(req)) { | |
675 | ret = PTR_ERR(req); | |
eab87235 | 676 | break; |
e8344e66 | 677 | } |
124e68e7 | 678 | |
144cba14 | 679 | osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0); |
715e4cd4 | 680 | |
b5b98989 ZC |
681 | n = len; |
682 | pages = dio_get_pages_alloc(from, len, &start, &num_pages); | |
683 | if (IS_ERR(pages)) { | |
64c31311 | 684 | ceph_osdc_put_request(req); |
b5b98989 | 685 | ret = PTR_ERR(pages); |
64c31311 | 686 | break; |
124e68e7 SW |
687 | } |
688 | ||
689 | /* | |
690 | * throw out any page cache pages in this range. this | |
691 | * may block. | |
692 | */ | |
213c99ee | 693 | truncate_inode_pages_range(inode->i_mapping, pos, |
64c31311 AV |
694 | (pos+n) | (PAGE_CACHE_SIZE-1)); |
695 | osd_req_op_extent_osd_data_pages(req, 0, pages, n, start, | |
e8344e66 | 696 | false, false); |
697 | ||
698 | /* BUG_ON(vino.snap != CEPH_NOSNAP); */ | |
699 | ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime); | |
700 | ||
701 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); | |
702 | if (!ret) | |
703 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
704 | ||
705 | ceph_put_page_vector(pages, num_pages, false); | |
706 | ||
e8344e66 | 707 | ceph_osdc_put_request(req); |
64c31311 | 708 | if (ret) |
e8344e66 | 709 | break; |
64c31311 AV |
710 | pos += n; |
711 | written += n; | |
4908b822 | 712 | iov_iter_advance(from, n); |
64c31311 AV |
713 | |
714 | if (pos > i_size_read(inode)) { | |
715 | check_caps = ceph_inode_set_size(inode, pos); | |
716 | if (check_caps) | |
717 | ceph_check_caps(ceph_inode(inode), | |
718 | CHECK_CAPS_AUTHONLY, | |
719 | NULL); | |
720 | } | |
e8344e66 | 721 | } |
722 | ||
723 | if (ret != -EOLDSNAPC && written > 0) { | |
724 | iocb->ki_pos = pos; | |
725 | ret = written; | |
726 | } | |
727 | return ret; | |
728 | } | |
729 | ||
730 | ||
731 | /* | |
732 | * Synchronous write, straight from __user pointer or user pages. | |
733 | * | |
734 | * If write spans object boundary, just do multiple writes. (For a | |
735 | * correct atomic write, we should e.g. take write locks on all | |
736 | * objects, rollback on failure, etc.) | |
737 | */ | |
06fee30f | 738 | static ssize_t |
5dda377c YZ |
739 | ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos, |
740 | struct ceph_snap_context *snapc) | |
e8344e66 | 741 | { |
742 | struct file *file = iocb->ki_filp; | |
743 | struct inode *inode = file_inode(file); | |
744 | struct ceph_inode_info *ci = ceph_inode(inode); | |
745 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
e8344e66 | 746 | struct ceph_vino vino; |
747 | struct ceph_osd_request *req; | |
748 | struct page **pages; | |
749 | u64 len; | |
750 | int num_pages; | |
751 | int written = 0; | |
752 | int flags; | |
753 | int check_caps = 0; | |
754 | int ret; | |
755 | struct timespec mtime = CURRENT_TIME; | |
4908b822 | 756 | size_t count = iov_iter_count(from); |
e8344e66 | 757 | |
758 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) | |
759 | return -EROFS; | |
760 | ||
761 | dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count); | |
762 | ||
763 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); | |
764 | if (ret < 0) | |
765 | return ret; | |
766 | ||
767 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
768 | pos >> PAGE_CACHE_SHIFT, | |
769 | (pos + count) >> PAGE_CACHE_SHIFT); | |
770 | if (ret < 0) | |
771 | dout("invalidate_inode_pages2_range returned %d\n", ret); | |
772 | ||
773 | flags = CEPH_OSD_FLAG_ORDERSNAP | | |
774 | CEPH_OSD_FLAG_ONDISK | | |
775 | CEPH_OSD_FLAG_WRITE | | |
776 | CEPH_OSD_FLAG_ACK; | |
777 | ||
4908b822 | 778 | while ((len = iov_iter_count(from)) > 0) { |
e8344e66 | 779 | size_t left; |
780 | int n; | |
781 | ||
e8344e66 | 782 | vino = ceph_vino(inode); |
783 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
715e4cd4 | 784 | vino, pos, &len, 0, 1, |
e8344e66 | 785 | CEPH_OSD_OP_WRITE, flags, snapc, |
786 | ci->i_truncate_seq, | |
787 | ci->i_truncate_size, | |
788 | false); | |
789 | if (IS_ERR(req)) { | |
790 | ret = PTR_ERR(req); | |
eab87235 | 791 | break; |
e8344e66 | 792 | } |
793 | ||
794 | /* | |
795 | * write from beginning of first page, | |
796 | * regardless of io alignment | |
797 | */ | |
798 | num_pages = (len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
799 | ||
687265e5 | 800 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); |
124e68e7 SW |
801 | if (IS_ERR(pages)) { |
802 | ret = PTR_ERR(pages); | |
803 | goto out; | |
804 | } | |
e8344e66 | 805 | |
806 | left = len; | |
807 | for (n = 0; n < num_pages; n++) { | |
125d725c | 808 | size_t plen = min_t(size_t, left, PAGE_SIZE); |
4908b822 | 809 | ret = copy_page_from_iter(pages[n], 0, plen, from); |
e8344e66 | 810 | if (ret != plen) { |
811 | ret = -EFAULT; | |
812 | break; | |
813 | } | |
814 | left -= ret; | |
e8344e66 | 815 | } |
816 | ||
124e68e7 SW |
817 | if (ret < 0) { |
818 | ceph_release_page_vector(pages, num_pages); | |
819 | goto out; | |
820 | } | |
821 | ||
e8344e66 | 822 | /* get a second commit callback */ |
823 | req->r_unsafe_callback = ceph_sync_write_unsafe; | |
824 | req->r_inode = inode; | |
124e68e7 | 825 | |
e8344e66 | 826 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, |
827 | false, true); | |
02ee07d3 | 828 | |
e8344e66 | 829 | /* BUG_ON(vino.snap != CEPH_NOSNAP); */ |
830 | ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime); | |
124e68e7 | 831 | |
e8344e66 | 832 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
833 | if (!ret) | |
834 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
124e68e7 SW |
835 | |
836 | out: | |
e8344e66 | 837 | ceph_osdc_put_request(req); |
838 | if (ret == 0) { | |
839 | pos += len; | |
840 | written += len; | |
841 | ||
842 | if (pos > i_size_read(inode)) { | |
843 | check_caps = ceph_inode_set_size(inode, pos); | |
844 | if (check_caps) | |
845 | ceph_check_caps(ceph_inode(inode), | |
846 | CHECK_CAPS_AUTHONLY, | |
847 | NULL); | |
848 | } | |
849 | } else | |
850 | break; | |
851 | } | |
124e68e7 | 852 | |
e8344e66 | 853 | if (ret != -EOLDSNAPC && written > 0) { |
124e68e7 | 854 | ret = written; |
e8344e66 | 855 | iocb->ki_pos = pos; |
124e68e7 SW |
856 | } |
857 | return ret; | |
858 | } | |
859 | ||
860 | /* | |
861 | * Wrap generic_file_aio_read with checks for cap bits on the inode. | |
862 | * Atomically grab references, so that those bits are not released | |
863 | * back to the MDS mid-read. | |
864 | * | |
865 | * Hmm, the sync read case isn't actually async... should it be? | |
866 | */ | |
3644424d | 867 | static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to) |
124e68e7 SW |
868 | { |
869 | struct file *filp = iocb->ki_filp; | |
2962507c | 870 | struct ceph_file_info *fi = filp->private_data; |
66ee59af | 871 | size_t len = iov_iter_count(to); |
496ad9aa | 872 | struct inode *inode = file_inode(filp); |
124e68e7 | 873 | struct ceph_inode_info *ci = ceph_inode(inode); |
3738daa6 | 874 | struct page *pinned_page = NULL; |
124e68e7 | 875 | ssize_t ret; |
2962507c | 876 | int want, got = 0; |
83701246 | 877 | int retry_op = 0, read = 0; |
124e68e7 | 878 | |
6a026589 | 879 | again: |
8eb4efb0 | 880 | dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n", |
881 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode); | |
882 | ||
2962507c SW |
883 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
884 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
885 | else | |
886 | want = CEPH_CAP_FILE_CACHE; | |
3738daa6 | 887 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page); |
124e68e7 | 888 | if (ret < 0) |
8eb4efb0 | 889 | return ret; |
124e68e7 | 890 | |
2962507c | 891 | if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 || |
2ba48ce5 | 892 | (iocb->ki_flags & IOCB_DIRECT) || |
8eb4efb0 | 893 | (fi->flags & CEPH_F_SYNC)) { |
8eb4efb0 | 894 | |
895 | dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n", | |
896 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, | |
897 | ceph_cap_string(got)); | |
898 | ||
83701246 YZ |
899 | if (ci->i_inline_version == CEPH_INLINE_NONE) { |
900 | /* hmm, this isn't really async... */ | |
901 | ret = ceph_sync_read(iocb, to, &retry_op); | |
902 | } else { | |
903 | retry_op = READ_INLINE; | |
904 | } | |
8eb4efb0 | 905 | } else { |
8eb4efb0 | 906 | dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n", |
3644424d | 907 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, |
8eb4efb0 | 908 | ceph_cap_string(got)); |
124e68e7 | 909 | |
3644424d | 910 | ret = generic_file_read_iter(iocb, to); |
8eb4efb0 | 911 | } |
124e68e7 SW |
912 | dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n", |
913 | inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret); | |
3738daa6 YZ |
914 | if (pinned_page) { |
915 | page_cache_release(pinned_page); | |
916 | pinned_page = NULL; | |
917 | } | |
124e68e7 | 918 | ceph_put_cap_refs(ci, got); |
83701246 YZ |
919 | if (retry_op && ret >= 0) { |
920 | int statret; | |
921 | struct page *page = NULL; | |
922 | loff_t i_size; | |
923 | if (retry_op == READ_INLINE) { | |
687265e5 | 924 | page = __page_cache_alloc(GFP_KERNEL); |
83701246 YZ |
925 | if (!page) |
926 | return -ENOMEM; | |
927 | } | |
6a026589 | 928 | |
83701246 YZ |
929 | statret = __ceph_do_getattr(inode, page, |
930 | CEPH_STAT_CAP_INLINE_DATA, !!page); | |
931 | if (statret < 0) { | |
932 | __free_page(page); | |
933 | if (statret == -ENODATA) { | |
934 | BUG_ON(retry_op != READ_INLINE); | |
935 | goto again; | |
936 | } | |
937 | return statret; | |
938 | } | |
6a026589 | 939 | |
83701246 YZ |
940 | i_size = i_size_read(inode); |
941 | if (retry_op == READ_INLINE) { | |
fcc02d2a YZ |
942 | BUG_ON(ret > 0 || read > 0); |
943 | if (iocb->ki_pos < i_size && | |
944 | iocb->ki_pos < PAGE_CACHE_SIZE) { | |
83701246 YZ |
945 | loff_t end = min_t(loff_t, i_size, |
946 | iocb->ki_pos + len); | |
fcc02d2a | 947 | end = min_t(loff_t, end, PAGE_CACHE_SIZE); |
83701246 YZ |
948 | if (statret < end) |
949 | zero_user_segment(page, statret, end); | |
950 | ret = copy_page_to_iter(page, | |
951 | iocb->ki_pos & ~PAGE_MASK, | |
952 | end - iocb->ki_pos, to); | |
953 | iocb->ki_pos += ret; | |
fcc02d2a YZ |
954 | read += ret; |
955 | } | |
956 | if (iocb->ki_pos < i_size && read < len) { | |
957 | size_t zlen = min_t(size_t, len - read, | |
958 | i_size - iocb->ki_pos); | |
959 | ret = iov_iter_zero(zlen, to); | |
960 | iocb->ki_pos += ret; | |
961 | read += ret; | |
83701246 YZ |
962 | } |
963 | __free_pages(page, 0); | |
fcc02d2a | 964 | return read; |
83701246 | 965 | } |
6a026589 SW |
966 | |
967 | /* hit EOF or hole? */ | |
83701246 | 968 | if (retry_op == CHECK_EOF && iocb->ki_pos < i_size && |
fcc02d2a | 969 | ret < len) { |
8eb4efb0 | 970 | dout("sync_read hit hole, ppos %lld < size %lld" |
971 | ", reading more\n", iocb->ki_pos, | |
972 | inode->i_size); | |
973 | ||
6a026589 | 974 | read += ret; |
6a026589 | 975 | len -= ret; |
83701246 | 976 | retry_op = 0; |
6a026589 SW |
977 | goto again; |
978 | } | |
979 | } | |
8eb4efb0 | 980 | |
6a026589 SW |
981 | if (ret >= 0) |
982 | ret += read; | |
983 | ||
124e68e7 SW |
984 | return ret; |
985 | } | |
986 | ||
987 | /* | |
988 | * Take cap references to avoid releasing caps to MDS mid-write. | |
989 | * | |
990 | * If we are synchronous, and write with an old snap context, the OSD | |
991 | * may return EOLDSNAPC. In that case, retry the write.. _after_ | |
992 | * dropping our cap refs and allowing the pending snap to logically | |
993 | * complete _before_ this write occurs. | |
994 | * | |
995 | * If we are near ENOSPC, write synchronously. | |
996 | */ | |
4908b822 | 997 | static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from) |
124e68e7 SW |
998 | { |
999 | struct file *file = iocb->ki_filp; | |
33caad32 | 1000 | struct ceph_file_info *fi = file->private_data; |
496ad9aa | 1001 | struct inode *inode = file_inode(file); |
124e68e7 | 1002 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 YS |
1003 | struct ceph_osd_client *osdc = |
1004 | &ceph_sb_to_client(inode->i_sb)->client->osdc; | |
f66fd9f0 | 1005 | struct ceph_cap_flush *prealloc_cf; |
3309dd04 | 1006 | ssize_t count, written = 0; |
03d254ed | 1007 | int err, want, got; |
3309dd04 | 1008 | loff_t pos; |
124e68e7 SW |
1009 | |
1010 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
1011 | return -EROFS; | |
1012 | ||
f66fd9f0 YZ |
1013 | prealloc_cf = ceph_alloc_cap_flush(); |
1014 | if (!prealloc_cf) | |
1015 | return -ENOMEM; | |
1016 | ||
03d254ed | 1017 | mutex_lock(&inode->i_mutex); |
03d254ed | 1018 | |
03d254ed | 1019 | /* We can write back this queue in page reclaim */ |
de1414a6 | 1020 | current->backing_dev_info = inode_to_bdi(inode); |
03d254ed | 1021 | |
55b0b31c YZ |
1022 | if (iocb->ki_flags & IOCB_APPEND) { |
1023 | err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); | |
1024 | if (err < 0) | |
1025 | goto out; | |
1026 | } | |
1027 | ||
3309dd04 AV |
1028 | err = generic_write_checks(iocb, from); |
1029 | if (err <= 0) | |
03d254ed YZ |
1030 | goto out; |
1031 | ||
3309dd04 AV |
1032 | pos = iocb->ki_pos; |
1033 | count = iov_iter_count(from); | |
5fa8e0a1 | 1034 | err = file_remove_privs(file); |
03d254ed YZ |
1035 | if (err) |
1036 | goto out; | |
1037 | ||
1038 | err = file_update_time(file); | |
1039 | if (err) | |
1040 | goto out; | |
1041 | ||
28127bdd YZ |
1042 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
1043 | err = ceph_uninline_data(file, NULL); | |
1044 | if (err < 0) | |
1045 | goto out; | |
1046 | } | |
1047 | ||
124e68e7 | 1048 | retry_snap: |
6070e0c1 | 1049 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) { |
03d254ed | 1050 | err = -ENOSPC; |
6070e0c1 YZ |
1051 | goto out; |
1052 | } | |
03d254ed | 1053 | |
ac7f29bf | 1054 | dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n", |
03d254ed | 1055 | inode, ceph_vinop(inode), pos, count, inode->i_size); |
7971bd92 SW |
1056 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
1057 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1058 | else | |
1059 | want = CEPH_CAP_FILE_BUFFER; | |
03d254ed | 1060 | got = 0; |
3738daa6 YZ |
1061 | err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count, |
1062 | &got, NULL); | |
03d254ed | 1063 | if (err < 0) |
37505d57 | 1064 | goto out; |
124e68e7 | 1065 | |
ac7f29bf | 1066 | dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n", |
03d254ed | 1067 | inode, ceph_vinop(inode), pos, count, ceph_cap_string(got)); |
7971bd92 SW |
1068 | |
1069 | if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 || | |
2ba48ce5 | 1070 | (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC)) { |
5dda377c | 1071 | struct ceph_snap_context *snapc; |
4908b822 | 1072 | struct iov_iter data; |
37505d57 | 1073 | mutex_unlock(&inode->i_mutex); |
5dda377c YZ |
1074 | |
1075 | spin_lock(&ci->i_ceph_lock); | |
1076 | if (__ceph_have_pending_cap_snap(ci)) { | |
1077 | struct ceph_cap_snap *capsnap = | |
1078 | list_last_entry(&ci->i_cap_snaps, | |
1079 | struct ceph_cap_snap, | |
1080 | ci_item); | |
1081 | snapc = ceph_get_snap_context(capsnap->context); | |
1082 | } else { | |
1083 | BUG_ON(!ci->i_head_snapc); | |
1084 | snapc = ceph_get_snap_context(ci->i_head_snapc); | |
1085 | } | |
1086 | spin_unlock(&ci->i_ceph_lock); | |
1087 | ||
4908b822 AV |
1088 | /* we might need to revert back to that point */ |
1089 | data = *from; | |
2ba48ce5 | 1090 | if (iocb->ki_flags & IOCB_DIRECT) |
5dda377c YZ |
1091 | written = ceph_sync_direct_write(iocb, &data, pos, |
1092 | snapc); | |
e8344e66 | 1093 | else |
5dda377c | 1094 | written = ceph_sync_write(iocb, &data, pos, snapc); |
0e5dd45c | 1095 | if (written == -EOLDSNAPC) { |
1096 | dout("aio_write %p %llx.%llx %llu~%u" | |
1097 | "got EOLDSNAPC, retrying\n", | |
1098 | inode, ceph_vinop(inode), | |
4908b822 | 1099 | pos, (unsigned)count); |
0e5dd45c | 1100 | mutex_lock(&inode->i_mutex); |
0e5dd45c | 1101 | goto retry_snap; |
1102 | } | |
4908b822 AV |
1103 | if (written > 0) |
1104 | iov_iter_advance(from, written); | |
5dda377c | 1105 | ceph_put_snap_context(snapc); |
7971bd92 | 1106 | } else { |
32d3e148 | 1107 | loff_t old_size = inode->i_size; |
b0d7c223 YZ |
1108 | /* |
1109 | * No need to acquire the i_truncate_mutex. Because | |
1110 | * the MDS revokes Fwb caps before sending truncate | |
1111 | * message to us. We can't get Fwb cap while there | |
1112 | * are pending vmtruncate. So write and vmtruncate | |
1113 | * can not run at the same time | |
1114 | */ | |
4908b822 | 1115 | written = generic_perform_write(file, from, pos); |
aec605f4 AV |
1116 | if (likely(written >= 0)) |
1117 | iocb->ki_pos = pos + written; | |
32d3e148 YW |
1118 | if (inode->i_size > old_size) |
1119 | ceph_fscache_update_objectsize(inode); | |
6070e0c1 | 1120 | mutex_unlock(&inode->i_mutex); |
7971bd92 | 1121 | } |
d8de9ab6 | 1122 | |
03d254ed | 1123 | if (written >= 0) { |
fca65b4a | 1124 | int dirty; |
be655596 | 1125 | spin_lock(&ci->i_ceph_lock); |
28127bdd | 1126 | ci->i_inline_version = CEPH_INLINE_NONE; |
f66fd9f0 YZ |
1127 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
1128 | &prealloc_cf); | |
be655596 | 1129 | spin_unlock(&ci->i_ceph_lock); |
fca65b4a SW |
1130 | if (dirty) |
1131 | __mark_inode_dirty(inode, dirty); | |
124e68e7 | 1132 | } |
7971bd92 | 1133 | |
124e68e7 | 1134 | dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n", |
4908b822 | 1135 | inode, ceph_vinop(inode), pos, (unsigned)count, |
7971bd92 | 1136 | ceph_cap_string(got)); |
124e68e7 | 1137 | ceph_put_cap_refs(ci, got); |
7971bd92 | 1138 | |
03d254ed | 1139 | if (written >= 0 && |
6070e0c1 YZ |
1140 | ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) || |
1141 | ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) { | |
03d254ed | 1142 | err = vfs_fsync_range(file, pos, pos + written - 1, 1); |
6070e0c1 | 1143 | if (err < 0) |
03d254ed | 1144 | written = err; |
6070e0c1 | 1145 | } |
03d254ed | 1146 | |
2f75e9e1 SW |
1147 | goto out_unlocked; |
1148 | ||
03d254ed | 1149 | out: |
2f75e9e1 SW |
1150 | mutex_unlock(&inode->i_mutex); |
1151 | out_unlocked: | |
f66fd9f0 | 1152 | ceph_free_cap_flush(prealloc_cf); |
03d254ed | 1153 | current->backing_dev_info = NULL; |
03d254ed | 1154 | return written ? written : err; |
124e68e7 SW |
1155 | } |
1156 | ||
1157 | /* | |
1158 | * llseek. be sure to verify file size on SEEK_END. | |
1159 | */ | |
965c8e59 | 1160 | static loff_t ceph_llseek(struct file *file, loff_t offset, int whence) |
124e68e7 SW |
1161 | { |
1162 | struct inode *inode = file->f_mapping->host; | |
1163 | int ret; | |
1164 | ||
1165 | mutex_lock(&inode->i_mutex); | |
6a82c47a | 1166 | |
965c8e59 | 1167 | if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) { |
508b32d8 | 1168 | ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); |
124e68e7 SW |
1169 | if (ret < 0) { |
1170 | offset = ret; | |
1171 | goto out; | |
1172 | } | |
06222e49 JB |
1173 | } |
1174 | ||
965c8e59 | 1175 | switch (whence) { |
06222e49 | 1176 | case SEEK_END: |
124e68e7 SW |
1177 | offset += inode->i_size; |
1178 | break; | |
1179 | case SEEK_CUR: | |
1180 | /* | |
1181 | * Here we special-case the lseek(fd, 0, SEEK_CUR) | |
1182 | * position-querying operation. Avoid rewriting the "same" | |
1183 | * f_pos value back to the file because a concurrent read(), | |
1184 | * write() or lseek() might have altered it | |
1185 | */ | |
1186 | if (offset == 0) { | |
1187 | offset = file->f_pos; | |
1188 | goto out; | |
1189 | } | |
1190 | offset += file->f_pos; | |
1191 | break; | |
06222e49 JB |
1192 | case SEEK_DATA: |
1193 | if (offset >= inode->i_size) { | |
1194 | ret = -ENXIO; | |
1195 | goto out; | |
1196 | } | |
1197 | break; | |
1198 | case SEEK_HOLE: | |
1199 | if (offset >= inode->i_size) { | |
1200 | ret = -ENXIO; | |
1201 | goto out; | |
1202 | } | |
1203 | offset = inode->i_size; | |
1204 | break; | |
124e68e7 SW |
1205 | } |
1206 | ||
46a1c2c7 | 1207 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
124e68e7 SW |
1208 | |
1209 | out: | |
1210 | mutex_unlock(&inode->i_mutex); | |
1211 | return offset; | |
1212 | } | |
1213 | ||
ad7a60de LW |
1214 | static inline void ceph_zero_partial_page( |
1215 | struct inode *inode, loff_t offset, unsigned size) | |
1216 | { | |
1217 | struct page *page; | |
1218 | pgoff_t index = offset >> PAGE_CACHE_SHIFT; | |
1219 | ||
1220 | page = find_lock_page(inode->i_mapping, index); | |
1221 | if (page) { | |
1222 | wait_on_page_writeback(page); | |
1223 | zero_user(page, offset & (PAGE_CACHE_SIZE - 1), size); | |
1224 | unlock_page(page); | |
1225 | page_cache_release(page); | |
1226 | } | |
1227 | } | |
1228 | ||
1229 | static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset, | |
1230 | loff_t length) | |
1231 | { | |
1232 | loff_t nearly = round_up(offset, PAGE_CACHE_SIZE); | |
1233 | if (offset < nearly) { | |
1234 | loff_t size = nearly - offset; | |
1235 | if (length < size) | |
1236 | size = length; | |
1237 | ceph_zero_partial_page(inode, offset, size); | |
1238 | offset += size; | |
1239 | length -= size; | |
1240 | } | |
1241 | if (length >= PAGE_CACHE_SIZE) { | |
1242 | loff_t size = round_down(length, PAGE_CACHE_SIZE); | |
1243 | truncate_pagecache_range(inode, offset, offset + size - 1); | |
1244 | offset += size; | |
1245 | length -= size; | |
1246 | } | |
1247 | if (length) | |
1248 | ceph_zero_partial_page(inode, offset, length); | |
1249 | } | |
1250 | ||
1251 | static int ceph_zero_partial_object(struct inode *inode, | |
1252 | loff_t offset, loff_t *length) | |
1253 | { | |
1254 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1255 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); | |
1256 | struct ceph_osd_request *req; | |
1257 | int ret = 0; | |
1258 | loff_t zero = 0; | |
1259 | int op; | |
1260 | ||
1261 | if (!length) { | |
1262 | op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE; | |
1263 | length = &zero; | |
1264 | } else { | |
1265 | op = CEPH_OSD_OP_ZERO; | |
1266 | } | |
1267 | ||
1268 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, | |
1269 | ceph_vino(inode), | |
1270 | offset, length, | |
715e4cd4 | 1271 | 0, 1, op, |
ad7a60de LW |
1272 | CEPH_OSD_FLAG_WRITE | |
1273 | CEPH_OSD_FLAG_ONDISK, | |
1274 | NULL, 0, 0, false); | |
1275 | if (IS_ERR(req)) { | |
1276 | ret = PTR_ERR(req); | |
1277 | goto out; | |
1278 | } | |
1279 | ||
1280 | ceph_osdc_build_request(req, offset, NULL, ceph_vino(inode).snap, | |
1281 | &inode->i_mtime); | |
1282 | ||
1283 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); | |
1284 | if (!ret) { | |
1285 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); | |
1286 | if (ret == -ENOENT) | |
1287 | ret = 0; | |
1288 | } | |
1289 | ceph_osdc_put_request(req); | |
1290 | ||
1291 | out: | |
1292 | return ret; | |
1293 | } | |
1294 | ||
1295 | static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length) | |
1296 | { | |
1297 | int ret = 0; | |
1298 | struct ceph_inode_info *ci = ceph_inode(inode); | |
b314a90d SW |
1299 | s32 stripe_unit = ceph_file_layout_su(ci->i_layout); |
1300 | s32 stripe_count = ceph_file_layout_stripe_count(ci->i_layout); | |
1301 | s32 object_size = ceph_file_layout_object_size(ci->i_layout); | |
1302 | u64 object_set_size = object_size * stripe_count; | |
1303 | u64 nearly, t; | |
1304 | ||
1305 | /* round offset up to next period boundary */ | |
1306 | nearly = offset + object_set_size - 1; | |
1307 | t = nearly; | |
1308 | nearly -= do_div(t, object_set_size); | |
ad7a60de | 1309 | |
ad7a60de LW |
1310 | while (length && offset < nearly) { |
1311 | loff_t size = length; | |
1312 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1313 | if (ret < 0) | |
1314 | return ret; | |
1315 | offset += size; | |
1316 | length -= size; | |
1317 | } | |
1318 | while (length >= object_set_size) { | |
1319 | int i; | |
1320 | loff_t pos = offset; | |
1321 | for (i = 0; i < stripe_count; ++i) { | |
1322 | ret = ceph_zero_partial_object(inode, pos, NULL); | |
1323 | if (ret < 0) | |
1324 | return ret; | |
1325 | pos += stripe_unit; | |
1326 | } | |
1327 | offset += object_set_size; | |
1328 | length -= object_set_size; | |
1329 | } | |
1330 | while (length) { | |
1331 | loff_t size = length; | |
1332 | ret = ceph_zero_partial_object(inode, offset, &size); | |
1333 | if (ret < 0) | |
1334 | return ret; | |
1335 | offset += size; | |
1336 | length -= size; | |
1337 | } | |
1338 | return ret; | |
1339 | } | |
1340 | ||
1341 | static long ceph_fallocate(struct file *file, int mode, | |
1342 | loff_t offset, loff_t length) | |
1343 | { | |
1344 | struct ceph_file_info *fi = file->private_data; | |
aa8b60e0 | 1345 | struct inode *inode = file_inode(file); |
ad7a60de LW |
1346 | struct ceph_inode_info *ci = ceph_inode(inode); |
1347 | struct ceph_osd_client *osdc = | |
1348 | &ceph_inode_to_client(inode)->client->osdc; | |
f66fd9f0 | 1349 | struct ceph_cap_flush *prealloc_cf; |
ad7a60de LW |
1350 | int want, got = 0; |
1351 | int dirty; | |
1352 | int ret = 0; | |
1353 | loff_t endoff = 0; | |
1354 | loff_t size; | |
1355 | ||
494d77bf YZ |
1356 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
1357 | return -EOPNOTSUPP; | |
1358 | ||
ad7a60de LW |
1359 | if (!S_ISREG(inode->i_mode)) |
1360 | return -EOPNOTSUPP; | |
1361 | ||
f66fd9f0 YZ |
1362 | prealloc_cf = ceph_alloc_cap_flush(); |
1363 | if (!prealloc_cf) | |
1364 | return -ENOMEM; | |
1365 | ||
ad7a60de LW |
1366 | mutex_lock(&inode->i_mutex); |
1367 | ||
1368 | if (ceph_snap(inode) != CEPH_NOSNAP) { | |
1369 | ret = -EROFS; | |
1370 | goto unlock; | |
1371 | } | |
1372 | ||
1373 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) && | |
1374 | !(mode & FALLOC_FL_PUNCH_HOLE)) { | |
1375 | ret = -ENOSPC; | |
1376 | goto unlock; | |
1377 | } | |
1378 | ||
28127bdd YZ |
1379 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
1380 | ret = ceph_uninline_data(file, NULL); | |
1381 | if (ret < 0) | |
1382 | goto unlock; | |
1383 | } | |
1384 | ||
ad7a60de LW |
1385 | size = i_size_read(inode); |
1386 | if (!(mode & FALLOC_FL_KEEP_SIZE)) | |
1387 | endoff = offset + length; | |
1388 | ||
1389 | if (fi->fmode & CEPH_FILE_MODE_LAZY) | |
1390 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
1391 | else | |
1392 | want = CEPH_CAP_FILE_BUFFER; | |
1393 | ||
3738daa6 | 1394 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL); |
ad7a60de LW |
1395 | if (ret < 0) |
1396 | goto unlock; | |
1397 | ||
1398 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
1399 | if (offset < size) | |
1400 | ceph_zero_pagecache_range(inode, offset, length); | |
1401 | ret = ceph_zero_objects(inode, offset, length); | |
1402 | } else if (endoff > size) { | |
1403 | truncate_pagecache_range(inode, size, -1); | |
1404 | if (ceph_inode_set_size(inode, endoff)) | |
1405 | ceph_check_caps(ceph_inode(inode), | |
1406 | CHECK_CAPS_AUTHONLY, NULL); | |
1407 | } | |
1408 | ||
1409 | if (!ret) { | |
1410 | spin_lock(&ci->i_ceph_lock); | |
28127bdd | 1411 | ci->i_inline_version = CEPH_INLINE_NONE; |
f66fd9f0 YZ |
1412 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
1413 | &prealloc_cf); | |
ad7a60de LW |
1414 | spin_unlock(&ci->i_ceph_lock); |
1415 | if (dirty) | |
1416 | __mark_inode_dirty(inode, dirty); | |
1417 | } | |
1418 | ||
1419 | ceph_put_cap_refs(ci, got); | |
1420 | unlock: | |
1421 | mutex_unlock(&inode->i_mutex); | |
f66fd9f0 | 1422 | ceph_free_cap_flush(prealloc_cf); |
ad7a60de LW |
1423 | return ret; |
1424 | } | |
1425 | ||
124e68e7 SW |
1426 | const struct file_operations ceph_file_fops = { |
1427 | .open = ceph_open, | |
1428 | .release = ceph_release, | |
1429 | .llseek = ceph_llseek, | |
3644424d | 1430 | .read_iter = ceph_read_iter, |
4908b822 | 1431 | .write_iter = ceph_write_iter, |
124e68e7 SW |
1432 | .mmap = ceph_mmap, |
1433 | .fsync = ceph_fsync, | |
40819f6f GF |
1434 | .lock = ceph_lock, |
1435 | .flock = ceph_flock, | |
124e68e7 | 1436 | .splice_read = generic_file_splice_read, |
3551dd79 | 1437 | .splice_write = iter_file_splice_write, |
124e68e7 SW |
1438 | .unlocked_ioctl = ceph_ioctl, |
1439 | .compat_ioctl = ceph_ioctl, | |
ad7a60de | 1440 | .fallocate = ceph_fallocate, |
124e68e7 SW |
1441 | }; |
1442 |