]>
Commit | Line | Data |
---|---|---|
7c673cae FG |
1 | // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- |
2 | // vim: ts=8 sw=2 smarttab | |
3 | ||
4 | #include "librbd/operation/TrimRequest.h" | |
5 | #include "librbd/AsyncObjectThrottle.h" | |
6 | #include "librbd/ExclusiveLock.h" | |
7 | #include "librbd/ImageCtx.h" | |
8 | #include "librbd/internal.h" | |
9 | #include "librbd/ObjectMap.h" | |
10 | #include "librbd/Utils.h" | |
11 | #include "librbd/io/ObjectRequest.h" | |
12 | #include "common/ContextCompletion.h" | |
13 | #include "common/dout.h" | |
14 | #include "common/errno.h" | |
15 | #include "osdc/Striper.h" | |
16 | ||
17 | #include <boost/bind.hpp> | |
18 | #include <boost/lambda/bind.hpp> | |
19 | #include <boost/lambda/construct.hpp> | |
20 | #include <boost/scope_exit.hpp> | |
21 | ||
22 | #define dout_subsys ceph_subsys_rbd | |
23 | #undef dout_prefix | |
24 | #define dout_prefix *_dout << "librbd::TrimRequest: " | |
25 | ||
26 | namespace librbd { | |
27 | namespace operation { | |
28 | ||
29 | template <typename I> | |
30 | class C_CopyupObject : public C_AsyncObjectThrottle<I> { | |
31 | public: | |
32 | C_CopyupObject(AsyncObjectThrottle<I> &throttle, I *image_ctx, | |
33 | ::SnapContext snapc, uint64_t object_no) | |
34 | : C_AsyncObjectThrottle<I>(throttle, *image_ctx), m_snapc(snapc), | |
35 | m_object_no(object_no) | |
36 | { | |
37 | } | |
38 | ||
39 | int send() override { | |
40 | I &image_ctx = this->m_image_ctx; | |
41 | assert(image_ctx.owner_lock.is_locked()); | |
42 | assert(image_ctx.exclusive_lock == nullptr || | |
43 | image_ctx.exclusive_lock->is_lock_owner()); | |
44 | ||
45 | string oid = image_ctx.get_object_name(m_object_no); | |
46 | ldout(image_ctx.cct, 10) << "removing (with copyup) " << oid << dendl; | |
47 | ||
48 | auto req = new io::ObjectTrimRequest(&image_ctx, oid, m_object_no, | |
31f18b77 | 49 | m_snapc, false, this); |
7c673cae FG |
50 | req->send(); |
51 | return 0; | |
52 | } | |
53 | private: | |
54 | ::SnapContext m_snapc; | |
55 | uint64_t m_object_no; | |
56 | }; | |
57 | ||
58 | template <typename I> | |
59 | class C_RemoveObject : public C_AsyncObjectThrottle<I> { | |
60 | public: | |
61 | C_RemoveObject(AsyncObjectThrottle<I> &throttle, ImageCtx *image_ctx, | |
62 | uint64_t object_no) | |
63 | : C_AsyncObjectThrottle<I>(throttle, *image_ctx), m_object_no(object_no) | |
64 | { | |
65 | } | |
66 | ||
67 | int send() override { | |
68 | I &image_ctx = this->m_image_ctx; | |
69 | assert(image_ctx.owner_lock.is_locked()); | |
70 | assert(image_ctx.exclusive_lock == nullptr || | |
71 | image_ctx.exclusive_lock->is_lock_owner()); | |
72 | ||
73 | { | |
74 | RWLock::RLocker snap_locker(image_ctx.snap_lock); | |
75 | if (image_ctx.object_map != nullptr && | |
76 | !image_ctx.object_map->object_may_exist(m_object_no)) { | |
77 | return 1; | |
78 | } | |
79 | } | |
80 | ||
81 | string oid = image_ctx.get_object_name(m_object_no); | |
82 | ldout(image_ctx.cct, 10) << "removing " << oid << dendl; | |
83 | ||
84 | librados::AioCompletion *rados_completion = | |
85 | util::create_rados_callback(this); | |
86 | int r = image_ctx.data_ctx.aio_remove(oid, rados_completion); | |
87 | assert(r == 0); | |
88 | rados_completion->release(); | |
89 | return 0; | |
90 | } | |
91 | ||
92 | private: | |
93 | uint64_t m_object_no; | |
94 | }; | |
95 | ||
96 | template <typename I> | |
97 | TrimRequest<I>::TrimRequest(I &image_ctx, Context *on_finish, | |
98 | uint64_t original_size, uint64_t new_size, | |
99 | ProgressContext &prog_ctx) | |
100 | : AsyncRequest<I>(image_ctx, on_finish), m_new_size(new_size), | |
101 | m_prog_ctx(prog_ctx) | |
102 | { | |
103 | uint64_t period = image_ctx.get_stripe_period(); | |
104 | uint64_t new_num_periods = ((m_new_size + period - 1) / period); | |
105 | m_delete_off = MIN(new_num_periods * period, original_size); | |
106 | // first object we can delete free and clear | |
107 | m_delete_start = new_num_periods * image_ctx.get_stripe_count(); | |
108 | m_num_objects = Striper::get_num_objects(image_ctx.layout, original_size); | |
109 | ||
110 | CephContext *cct = image_ctx.cct; | |
111 | ldout(cct, 10) << this << " trim image " << original_size << " -> " | |
112 | << m_new_size << " periods " << new_num_periods | |
113 | << " discard to offset " << m_delete_off | |
114 | << " delete objects " << m_delete_start | |
115 | << " to " << m_num_objects << dendl; | |
116 | } | |
117 | ||
118 | template <typename I> | |
119 | bool TrimRequest<I>::should_complete(int r) | |
120 | { | |
121 | I &image_ctx = this->m_image_ctx; | |
122 | CephContext *cct = image_ctx.cct; | |
123 | ldout(cct, 5) << this << " should_complete: r=" << r << dendl; | |
124 | if (r == -ERESTART) { | |
125 | ldout(cct, 5) << "trim operation interrupted" << dendl; | |
126 | return true; | |
127 | } else if (r < 0) { | |
128 | lderr(cct) << "trim encountered an error: " << cpp_strerror(r) << dendl; | |
129 | return true; | |
130 | } | |
131 | ||
132 | RWLock::RLocker owner_lock(image_ctx.owner_lock); | |
133 | switch (m_state) { | |
134 | case STATE_PRE_COPYUP: | |
135 | ldout(cct, 5) << " PRE_COPYUP" << dendl; | |
136 | send_copyup_objects(); | |
137 | break; | |
138 | ||
139 | case STATE_COPYUP_OBJECTS: | |
140 | ldout(cct, 5) << " COPYUP_OBJECTS" << dendl; | |
141 | send_post_copyup(); | |
142 | break; | |
143 | ||
144 | case STATE_POST_COPYUP: | |
145 | ldout(cct, 5) << " POST_COPYUP" << dendl; | |
146 | send_pre_remove(); | |
147 | break; | |
148 | ||
149 | case STATE_PRE_REMOVE: | |
150 | ldout(cct, 5) << " PRE_REMOVE" << dendl; | |
151 | send_remove_objects(); | |
152 | break; | |
153 | ||
154 | case STATE_REMOVE_OBJECTS: | |
155 | ldout(cct, 5) << " REMOVE_OBJECTS" << dendl; | |
156 | send_post_remove(); | |
157 | break; | |
158 | ||
159 | case STATE_POST_REMOVE: | |
160 | ldout(cct, 5) << " POST_OBJECTS" << dendl; | |
161 | send_clean_boundary(); | |
162 | break; | |
163 | ||
164 | case STATE_CLEAN_BOUNDARY: | |
165 | ldout(cct, 5) << "CLEAN_BOUNDARY" << dendl; | |
166 | send_finish(0); | |
167 | break; | |
168 | ||
169 | case STATE_FINISHED: | |
170 | ldout(cct, 5) << "FINISHED" << dendl; | |
171 | return true; | |
172 | ||
173 | default: | |
174 | lderr(cct) << "invalid state: " << m_state << dendl; | |
175 | assert(false); | |
176 | break; | |
177 | } | |
178 | return false; | |
179 | } | |
180 | ||
181 | template <typename I> | |
182 | void TrimRequest<I>::send() { | |
183 | send_pre_copyup(); | |
184 | } | |
185 | ||
186 | template<typename I> | |
187 | void TrimRequest<I>::send_copyup_objects() { | |
188 | I &image_ctx = this->m_image_ctx; | |
189 | assert(image_ctx.owner_lock.is_locked()); | |
190 | ||
191 | ldout(image_ctx.cct, 5) << this << " send_copyup_objects: " | |
192 | << " start object=" << m_copyup_start << ", " | |
193 | << " end object=" << m_copyup_end << dendl; | |
194 | m_state = STATE_COPYUP_OBJECTS; | |
195 | ||
196 | ::SnapContext snapc; | |
197 | { | |
198 | RWLock::RLocker snap_locker(image_ctx.snap_lock); | |
199 | RWLock::RLocker parent_locker(image_ctx.parent_lock); | |
200 | snapc = image_ctx.snapc; | |
201 | } | |
202 | ||
203 | Context *ctx = this->create_callback_context(); | |
204 | typename AsyncObjectThrottle<I>::ContextFactory context_factory( | |
205 | boost::lambda::bind(boost::lambda::new_ptr<C_CopyupObject<I> >(), | |
206 | boost::lambda::_1, &image_ctx, snapc, boost::lambda::_2)); | |
207 | AsyncObjectThrottle<I> *throttle = new AsyncObjectThrottle<I>( | |
208 | this, image_ctx, context_factory, ctx, &m_prog_ctx, m_copyup_start, | |
209 | m_copyup_end); | |
210 | throttle->start_ops(image_ctx.concurrent_management_ops); | |
211 | } | |
212 | ||
213 | template <typename I> | |
214 | void TrimRequest<I>::send_remove_objects() { | |
215 | I &image_ctx = this->m_image_ctx; | |
216 | assert(image_ctx.owner_lock.is_locked()); | |
217 | ||
218 | ldout(image_ctx.cct, 5) << this << " send_remove_objects: " | |
219 | << " delete_start=" << m_delete_start | |
220 | << " num_objects=" << m_num_objects << dendl; | |
221 | m_state = STATE_REMOVE_OBJECTS; | |
222 | ||
223 | Context *ctx = this->create_callback_context(); | |
224 | typename AsyncObjectThrottle<I>::ContextFactory context_factory( | |
225 | boost::lambda::bind(boost::lambda::new_ptr<C_RemoveObject<I> >(), | |
226 | boost::lambda::_1, &image_ctx, boost::lambda::_2)); | |
227 | AsyncObjectThrottle<I> *throttle = new AsyncObjectThrottle<I>( | |
228 | this, image_ctx, context_factory, ctx, &m_prog_ctx, m_delete_start, | |
229 | m_num_objects); | |
230 | throttle->start_ops(image_ctx.concurrent_management_ops); | |
231 | } | |
232 | ||
233 | template<typename I> | |
234 | void TrimRequest<I>::send_pre_copyup() { | |
235 | I &image_ctx = this->m_image_ctx; | |
236 | assert(image_ctx.owner_lock.is_locked()); | |
237 | ||
238 | if (m_delete_start >= m_num_objects) { | |
239 | send_clean_boundary(); | |
240 | return; | |
241 | } | |
242 | ||
243 | bool has_snapshots; | |
244 | uint64_t parent_overlap; | |
245 | { | |
246 | RWLock::RLocker snap_locker(image_ctx.snap_lock); | |
247 | RWLock::RLocker parent_locker(image_ctx.parent_lock); | |
248 | ||
249 | has_snapshots = !image_ctx.snaps.empty(); | |
250 | int r = image_ctx.get_parent_overlap(CEPH_NOSNAP, &parent_overlap); | |
251 | assert(r == 0); | |
252 | } | |
253 | ||
254 | // copyup is only required for portion of image that overlaps parent | |
255 | m_copyup_end = Striper::get_num_objects(image_ctx.layout, parent_overlap); | |
256 | ||
257 | // TODO: protect against concurrent shrink and snap create? | |
258 | // skip to remove if no copyup is required. | |
259 | if (m_copyup_end <= m_delete_start || !has_snapshots) { | |
260 | send_pre_remove(); | |
261 | return; | |
262 | } | |
263 | ||
264 | m_copyup_start = m_delete_start; | |
265 | m_delete_start = m_copyup_end; | |
266 | ||
267 | { | |
268 | RWLock::RLocker snap_locker(image_ctx.snap_lock); | |
269 | if (image_ctx.object_map != nullptr) { | |
270 | ldout(image_ctx.cct, 5) << this << " send_pre_copyup: " | |
271 | << " copyup_start=" << m_copyup_start | |
272 | << " copyup_end=" << m_copyup_end << dendl; | |
273 | m_state = STATE_PRE_COPYUP; | |
274 | ||
275 | assert(image_ctx.exclusive_lock->is_lock_owner()); | |
276 | ||
277 | RWLock::WLocker object_map_locker(image_ctx.object_map_lock); | |
278 | if (image_ctx.object_map->template aio_update<AsyncRequest<I> >( | |
279 | CEPH_NOSNAP, m_copyup_start, m_copyup_end, OBJECT_PENDING, | |
31f18b77 | 280 | OBJECT_EXISTS, {}, this)) { |
7c673cae FG |
281 | return; |
282 | } | |
283 | } | |
284 | } | |
285 | ||
286 | send_copyup_objects(); | |
287 | } | |
288 | ||
289 | template <typename I> | |
290 | void TrimRequest<I>::send_pre_remove() { | |
291 | I &image_ctx = this->m_image_ctx; | |
292 | assert(image_ctx.owner_lock.is_locked()); | |
293 | if (m_delete_start >= m_num_objects) { | |
294 | send_clean_boundary(); | |
295 | return; | |
296 | } | |
297 | ||
298 | { | |
299 | RWLock::RLocker snap_locker(image_ctx.snap_lock); | |
300 | if (image_ctx.object_map != nullptr) { | |
301 | ldout(image_ctx.cct, 5) << this << " send_pre_remove: " | |
302 | << " delete_start=" << m_delete_start | |
303 | << " num_objects=" << m_num_objects << dendl; | |
304 | m_state = STATE_PRE_REMOVE; | |
305 | ||
306 | assert(image_ctx.exclusive_lock->is_lock_owner()); | |
307 | ||
308 | // flag the objects as pending deletion | |
309 | RWLock::WLocker object_map_locker(image_ctx.object_map_lock); | |
310 | if (image_ctx.object_map->template aio_update<AsyncRequest<I> >( | |
311 | CEPH_NOSNAP, m_delete_start, m_num_objects, OBJECT_PENDING, | |
31f18b77 | 312 | OBJECT_EXISTS, {}, this)) { |
7c673cae FG |
313 | return; |
314 | } | |
315 | } | |
316 | } | |
317 | ||
318 | // no object map update required | |
319 | send_remove_objects(); | |
320 | } | |
321 | ||
322 | template<typename I> | |
323 | void TrimRequest<I>::send_post_copyup() { | |
324 | I &image_ctx = this->m_image_ctx; | |
325 | assert(image_ctx.owner_lock.is_locked()); | |
326 | ||
327 | { | |
328 | RWLock::RLocker snap_locker(image_ctx.snap_lock); | |
329 | if (image_ctx.object_map != nullptr) { | |
330 | ldout(image_ctx.cct, 5) << this << " send_post_copyup:" | |
331 | << " copyup_start=" << m_copyup_start | |
332 | << " copyup_end=" << m_copyup_end << dendl; | |
333 | m_state = STATE_POST_COPYUP; | |
334 | ||
335 | assert(image_ctx.exclusive_lock->is_lock_owner()); | |
336 | ||
337 | RWLock::WLocker object_map_locker(image_ctx.object_map_lock); | |
338 | if (image_ctx.object_map->template aio_update<AsyncRequest<I> >( | |
339 | CEPH_NOSNAP, m_copyup_start, m_copyup_end, OBJECT_NONEXISTENT, | |
31f18b77 | 340 | OBJECT_PENDING, {}, this)) { |
7c673cae FG |
341 | return; |
342 | } | |
343 | } | |
344 | } | |
345 | ||
346 | send_pre_remove(); | |
347 | } | |
348 | ||
349 | template <typename I> | |
350 | void TrimRequest<I>::send_post_remove() { | |
351 | I &image_ctx = this->m_image_ctx; | |
352 | assert(image_ctx.owner_lock.is_locked()); | |
353 | ||
354 | { | |
355 | RWLock::RLocker snap_locker(image_ctx.snap_lock); | |
356 | if (image_ctx.object_map != nullptr) { | |
357 | ldout(image_ctx.cct, 5) << this << " send_post_remove: " | |
358 | << " delete_start=" << m_delete_start | |
359 | << " num_objects=" << m_num_objects << dendl; | |
360 | m_state = STATE_POST_REMOVE; | |
361 | ||
362 | assert(image_ctx.exclusive_lock->is_lock_owner()); | |
363 | ||
364 | // flag the pending objects as removed | |
365 | RWLock::WLocker object_map_locker(image_ctx.object_map_lock); | |
366 | if (image_ctx.object_map->template aio_update<AsyncRequest<I> >( | |
367 | CEPH_NOSNAP, m_delete_start, m_num_objects, OBJECT_NONEXISTENT, | |
31f18b77 | 368 | OBJECT_PENDING, {}, this)) { |
7c673cae FG |
369 | return; |
370 | } | |
371 | } | |
372 | } | |
373 | ||
374 | // no object map update required | |
375 | send_clean_boundary(); | |
376 | } | |
377 | ||
378 | template <typename I> | |
379 | void TrimRequest<I>::send_clean_boundary() { | |
380 | I &image_ctx = this->m_image_ctx; | |
381 | assert(image_ctx.owner_lock.is_locked()); | |
382 | CephContext *cct = image_ctx.cct; | |
383 | if (m_delete_off <= m_new_size) { | |
384 | send_finish(0); | |
385 | return; | |
386 | } | |
387 | ||
388 | // should have been canceled prior to releasing lock | |
389 | assert(image_ctx.exclusive_lock == nullptr || | |
390 | image_ctx.exclusive_lock->is_lock_owner()); | |
391 | uint64_t delete_len = m_delete_off - m_new_size; | |
392 | ldout(image_ctx.cct, 5) << this << " send_clean_boundary: " | |
393 | << " delete_off=" << m_delete_off | |
394 | << " length=" << delete_len << dendl; | |
395 | m_state = STATE_CLEAN_BOUNDARY; | |
396 | ||
397 | ::SnapContext snapc; | |
398 | { | |
399 | RWLock::RLocker snap_locker(image_ctx.snap_lock); | |
400 | snapc = image_ctx.snapc; | |
401 | } | |
402 | ||
403 | // discard the weird boundary | |
404 | std::vector<ObjectExtent> extents; | |
405 | Striper::file_to_extents(cct, image_ctx.format_string, | |
406 | &image_ctx.layout, m_new_size, delete_len, 0, | |
407 | extents); | |
408 | ||
409 | ContextCompletion *completion = | |
410 | new ContextCompletion(this->create_async_callback_context(), true); | |
411 | for (vector<ObjectExtent>::iterator p = extents.begin(); | |
412 | p != extents.end(); ++p) { | |
413 | ldout(cct, 20) << " ex " << *p << dendl; | |
414 | Context *req_comp = new C_ContextCompletion(*completion); | |
415 | ||
416 | io::ObjectRequest<> *req; | |
417 | if (p->offset == 0) { | |
418 | req = new io::ObjectTrimRequest(&image_ctx, p->oid.name, p->objectno, | |
31f18b77 | 419 | snapc, true, req_comp); |
7c673cae FG |
420 | } else { |
421 | req = new io::ObjectTruncateRequest(&image_ctx, p->oid.name, p->objectno, | |
31f18b77 | 422 | p->offset, snapc, {}, req_comp); |
7c673cae FG |
423 | } |
424 | req->send(); | |
425 | } | |
426 | completion->finish_adding_requests(); | |
427 | } | |
428 | ||
429 | template <typename I> | |
430 | void TrimRequest<I>::send_finish(int r) { | |
431 | m_state = STATE_FINISHED; | |
432 | this->async_complete(r); | |
433 | } | |
434 | ||
435 | } // namespace operation | |
436 | } // namespace librbd | |
437 | ||
438 | template class librbd::operation::TrimRequest<librbd::ImageCtx>; |