Commit Graph

1217 Commits

Author SHA1 Message Date
Takeshi Nakatani 839a33de49
Fixed not to call Flush even if the file size is increased (#1887)
Changed s3fs_truncate function.
This change reduces the number of file uploads if the file size is changed.

On macOS, I have found that the truncate call when "size=0" cannot reflect the file size.(This reason is not understood...)
To avoid this, only when "size=0", the flush method is called as before.

Other than that, I found a bug in FdEntity::Open() and fixed it.

Fixes #1875.
2022-02-15 21:29:07 +09:00
Petr Vaněk 4dfe2bfdd7 Include climits to support musl libc
PATH_MAX constant is not visible from any of currently included header
files in system with musl libc, where compilation fails with an error
below. The constant is defined in limits.h which is directly include via
climits header file.

fdcache.cpp: In static member function 'static FILE* FdManager::MakeTempFile()':
fdcache.cpp:381:14: error: 'PATH_MAX' was not declared in this scope
  381 |     char cfn[PATH_MAX];
      |              ^~~~~~~~

Fixes: d67b83e671 ("Allow configuration for temporary files directory")
2022-02-14 09:19:30 +09:00
Takeshi Nakatani 1678803566 Added S3fsCred class and moved Credential related processing in it 2022-02-13 21:38:30 +09:00
Takeshi Nakatani d7e929e0a8
Fixed some Github Actions errors. (#1886)
- Fix knownConditionTrueFalse cppcheck(2.7) error on MacOS
- Fixed package installing failure of appstream download on centos8
2022-02-13 14:23:35 +09:00
Andrew Gaul 9224f792f0
Use CLOCK_REALTIME for UTIME_NOW (#1881)
Previously s3fs_utimens used CLOCK_MONOTONIC_COARSE which was not
1970-based.  Found via pjdfstest.  References #1589.
2022-01-30 22:19:15 +09:00
Andrew Gaul f6ed972926
Always flush open files with O_CREAT flag (#1879)
Previously s3fs only created files that had dirty data and not those
with zero-bytes.  Regression from
771bbfeac5.  References #1013.  Found
via pjdfstest.  References #1589.
2022-01-30 22:02:37 +09:00
Andrew Gaul 0c75a63184
Preserve sub-second precision with utimens (#1880)
Found via pjdfstest.  References #1589.
2022-01-30 21:45:51 +09:00
Takeshi Nakatani e452ef3940 Fixed the fault tolerance when time stamp getting fails 2022-01-30 18:31:36 +09:00
Andrew Gaul cd5a69b9eb
Handle UTIME_NOW and UTIME_OMIT special values (#1868)
FUSE 3 will require this behavior.  References #1159.
2022-01-29 11:35:37 +09:00
Andrew Gaul 662882d2f0
Always call clock_gettime(2) (#1871)
e01ded9e27 introduced this compatibility
shim but macOS 10.12 (2016) added this:
https://stackoverflow.com/a/39801564 .  Also remove fallback to
time(3) which loses precision.
2022-01-25 08:36:27 +09:00
Andrew Gaul de0c87c801
Convert S3FS_LOW_LOGPRN from a macro to a function (#1869)
This shrinks the binary size from 770 to 540 KB and reduces compile
times.
2022-01-23 23:10:09 +09:00
Naoki Ikeguchi 41aaa4184f Avoid double setting values in statfs 2022-01-23 21:49:51 +09:00
Naoki Ikeguchi 581f5c0356 Move strptime polyfill to string_util 2022-01-23 21:49:51 +09:00
Naoki Ikeguchi e5f6f112db Fix typo 2022-01-23 21:49:51 +09:00
Naoki Ikeguchi b3cef944b2 Fix test_page_list_SOURCES has no if MSYS clause 2022-01-23 21:49:51 +09:00
Naoki Ikeguchi 6edb6067f3 Remove strcasestr polyfill 2022-01-23 21:49:51 +09:00
Naoki Ikeguchi b2c659c0a6 Disable compiling polyfills in not MSYS2 env 2022-01-23 21:49:51 +09:00
Naoki Ikeguchi 3ac9f571f5 Use std::get_time instead in strptime polyfill 2022-01-23 21:49:51 +09:00
Naoki Ikeguchi 19303a546e Fix the statfs issue, using f_frsize instead 2022-01-23 21:49:51 +09:00
Naoki Ikeguchi 6370e150dd Disable features that causes problems on Windows 2022-01-23 21:49:51 +09:00
Naoki Ikeguchi b14e39815b Use polyfills in MSYS2 environment 2022-01-23 21:49:51 +09:00
Naoki Ikeguchi 6aaf9433a5 Add polyfills for MSYS2 environment 2022-01-23 21:49:51 +09:00
Andrew Gaul 93d1c30d4d
Use XML parsing with PUT HTTP 200 responses (#1858)
This works around the missing strcasestr on win32.  References #728.
2022-01-14 16:10:22 +09:00
Andrew Gaul 25012f3839 Fix typo in -o enable_unsigned_payload 2022-01-12 22:50:49 +09:00
Andrew Gaul e734763002
Remove createbucket option (#1841)
AWS CLI can do this.  Fixes #1840.
2022-01-05 01:59:31 +09:00
LiuBingrun 85ca2a3e45
fix mixupload return EntityTooSmall while a copypart is less than 5MB after split (#1809)
* fix  mixupload return EntityTooSmall while a copypart is less than 5MB after split
* fix possible part exceeds 5GB when multipart_copy_size is set to 5120MB
* Update curl.cpp
Co-authored-by: liubingrun <liubr1@chinatelecom.cn>
2021-11-27 16:53:26 +09:00
Andrew Gaul 3cf00626a2
Add option to allow unsigned payloads (#1801)
This reduces CPU usage of sigv4.  This reduces test run-time by 7
seconds per flag.
2021-11-01 23:33:55 +09:00
Takeshi Nakatani 7273d561f5 Added exclusive control of statc variables in s3fs xml parser 2021-10-27 08:18:19 +09:00
Takeshi Nakatani 78126aea0b Added exclusive control of statc variables in s3fs xml parser 2021-10-27 08:18:19 +09:00
Takeshi Nakatani 7892eee207 Fixed a bug that copied without considering the length of xmlChar 2021-10-27 08:18:19 +09:00
Takeshi Nakatani 023aaf7dff Fixed wrong stat of cache after new creation file
And added a test for stat of cache after new creation file
2021-10-17 16:10:14 +09:00
Takeshi Nakatani 2f412804e2 Fixed forgetting to clear the dirty flag for meta information
Addressed an error in macos cpp check
2021-10-15 22:54:55 +09:00
Carsten Grohmann be0b17329a
Fix wrong function name in log message (#1774) 2021-10-10 11:08:32 +09:00
Kamil Jakrzewski b4edad86d6 remove Expect: 100-continue header when requesting an IMDSv2 access token 2021-09-09 08:12:36 +09:00
Noah Meyerhans 9d1552a54e fix IAM role retrieval from IMDSv2
AWS IMDSv2 support was added in #1462, but the implementation did not
cover the addional IMDS access that occurs with the iam_role=auto
configuration.  This change implements IMDSv2 support for the IMDS
call to determine the instance's role name.

See also
https://stackoverflow.com/questions/69031023/how-to-make-s3fs-use-imds-v2-when-mounting-s3-buckets-from-ec2-instance
2021-09-03 20:36:34 +09:00
Andrew Gaul 47ebfcc60a
Consume return value from curl_easy_setopt (#1759)
Found via Coverity.
2021-09-02 08:07:06 +09:00
Andrew Gaul beecf32dff
fclose(FILE*) instead of close(fileno(FILE*)) (#1758)
This is the same thing but confuses Coverity.
2021-09-01 19:41:55 +09:00
Andrew Gaul 57b2e4a4f1
Fix 32-bit compilation issues (#1757) 2021-08-31 19:36:02 +09:00
Andrew Gaul 48817d849f
Require explicit length in s3fs_decode64 (#1755)
This is available from std::string::size in callers.
2021-08-31 09:22:10 +09:00
VVoidV d9f2d17040
1. fix RowFlush can not upload last part smaller than 5MB using NoCacheMultipartPost; (#1753)
2. fix deadlock in UploadPendingMeta
2021-08-31 00:41:47 +09:00
Andrew Gaul cd98afdd7b
Do not NUL terminate base64 decoded output (#1752)
This is binary data and must use the explicit length.
2021-08-31 00:15:47 +09:00
Andrew Gaul dac6885fb0
Don't over-allocate in base64 encoding and decoding (#1751) 2021-08-30 00:03:10 +09:00
VVoidV fcd180891b
fix misuse of IsUploading (#1747)
Co-authored-by: liubingrun <liubr1@chinatelecom.cn>
2021-08-29 23:41:02 +09:00
Takeshi Nakatani a868c0656e Changed etaglist_t from string list to new structure etagpairs list 2021-08-16 09:27:12 +09:00
Andrew Gaul 15e89b78de
Add a partial page_list unit test (#1735) 2021-08-04 07:36:32 +09:00
Andrew Gaul 66006ba48d
Add dedicated upper- and lower-case hex functions (#1734)
This makes the call sites more readable than a boolean parameter.
2021-08-04 07:28:51 +09:00
Andrew Gaul 18e9c62087
Make string constants read-only const (#1733)
This removes some global constructors.  Also use a consistent ALL_CAPS
style.
2021-08-03 00:10:27 +09:00
Takeshi Nakatani 34f89e5936 Added fake_diskfree option to deceive free disk space for test 2021-08-02 11:04:42 +09:00
Carsten Grohmann e1f3b9d8c1
Add support for AWS-style environment variables (#1729)
Support AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY and AWS_SESSION_TOKEN
in addition to the AWSACCESSKEYID, AWSSECRETACCESSKEY and
AWSSESSIONTOKEN.

The old environment variables are still supported, but they are
deprecated and no longer documented.

Close #1708
2021-07-26 23:29:45 +09:00
Andrew Gaul d3278f4886
Loosen CheckBucket to check only the bucket (#1728)
Previously it checked if there was an object present, assuming that
this was a directory object normally created for s3fs directories.
However most S3 clients do not create this object for virtual folders.
Fixes #1460.  Fixes #1687.
2021-07-25 12:29:00 +09:00
Andrew Gaul 77f0b75d2f
Add jitter to avoid thundering herd (#1723) 2021-07-25 10:22:19 +09:00
Takeshi Nakatani 199b3d4709 Fixed a bug in disk free space calculation 2021-07-18 14:52:11 +09:00
Andrew Gaul 7890989cbb
Address cppcheck 2.5 warnings (#1720)
This slightly simplifies memory management.
2021-07-14 22:18:09 +09:00
Takeshi Nakatani 945cc2ac54 Added UntreatedParts class instead of untreated upload info members in PseudoFdInfo 2021-07-12 21:45:29 +09:00
Takeshi Nakatani c30acbbf90 Splited some methods of FdEntity class by uploading mode
- Splited FdEntity::Write method by uploading mode
- Splited FdEntity::RowFlush method by uplading mode
2021-07-10 17:24:34 +09:00
Martynov Maxim 881025cc9e
Add initiator pid, uid and gid to a debug log (#1716) 2021-07-10 16:23:20 +09:00
Takeshi Nakatani 4cc210c5ab Prevents the etag string buffer from being destroyed during use
and absorbed MPPART_INFO structure into filepart structure
2021-07-03 22:42:18 +09:00
Andrew Gaul fe44355d25
Report s3fs space as 18.4 EB (#1709)
256 TB probably seemed large in 2007 but in 2021 it is small enough to
confuse some users.
2021-07-02 07:56:05 +09:00
Andrew Gaul 487df27008
Return EREMOTE when reading Glacier objects (#1705)
Previously s3fs returned EIO which was not helpful to the caller.
Returning a more specific error code allows automatically restoring
the object via RestoreObject in a subsequent commit.
References #1466.
2021-06-30 09:25:36 +09:00
Andrew Gaul 1965916f7a
Convert storage class values to uppercase (#1704)
AWS requires uppercase values.  Fixes a regression from
bbcccd6e98.  References #1613.
2021-06-30 09:03:31 +09:00
Andrew Gaul 8948eded09
Address clang-tidy warnings (#1703) 2021-06-30 08:42:44 +09:00
Andrew Gaul 2f59cb5a0a
Remove unneeded volatile qualifiers (#1702)
We should use proper locking instead.  Specifically for
is_meta_pending this does not do what was intended due to the
read-modify-write of the member.
2021-06-30 08:20:44 +09:00
Andrew Gaul f505c8224e
Explicitly specify lock type (#1701)
This makes it more clear and type-safe if the caller already has the
lock.  Follows on to 84174c560d.
2021-06-27 15:15:48 +09:00
Andrew Gaul 2154e898bc
Fix typos (#1700) 2021-06-27 11:22:33 +09:00
Takeshi Nakatani f9e80f995d Fixed a bug about rename existing file of different sizes by mpcopy 2021-06-26 00:04:58 +09:00
Jan Stastny a5c1915772
Set CURLOPT_UNRESTRICTED_AUTH when authenticating (#1681)
This is necessary for authentication to work with AWS when it responds with 307 to the check bucket request when mounting. This happens to newly created buckets according to https://aws.amazon.com/premiumsupport/knowledge-center/s3-http-307-response/.

Prior to this, curl would follow the redirect, but would not include the `Authorization` header which would end up with `404`.

Co-authored-by: Takeshi Nakatani <ggtakec@gmail.com>
2021-06-25 20:26:12 +09:00
Takeshi Nakatani 6edf3d6427 Updated to clearly output message about the file descriptor type(pseudo/physical) 2021-06-25 19:18:08 +09:00
Andrew Gaul 20281737b1
Set explicit Content-Length: 0 when initiating MPU (#1696)
This works around a GCS quirk that incorrectly requires this header.
Fixes #1661.
2021-06-25 18:36:40 +09:00
Takeshi Nakatani 5b6684ca19 Fixed a bug utimens is calling before flush 2021-06-21 17:20:24 +09:00
Carsten Grohmann c4ac923b4c
Ensure NUL-terminated result after strncpy (#1694)
Long symlinks may cause that the result buffer is filled and not proper
terminated with a null byte.
2021-06-21 08:08:56 +09:00
Takeshi Nakatani cc022a68f4 Fixed a bug in s3fs_write 2021-06-20 23:13:44 +09:00
Takeshi Nakatani 858562ed53 Fixed comparison warning for size_t variable 2021-06-20 16:53:25 +09:00
Andrew Gaul b8724425d3
Require explicit base parameter in cvt_strtoofft (#1683)
Also convert most callers of cvt_strtoofft to base 10 which avoid the
magical behavior of interpreting a leading 0 as octal.
References #1682.
2021-06-20 11:00:15 +09:00
Dmitrii Vasilev b9ce0faee2 Set Content-MD5 header even for empty objects
The Content-MD5 header is required for any request to upload an object
with a retention period configured using Amazon S3 Object Lock
2021-06-19 16:10:58 +09:00
Carsten Grohmann 1e0e2752bf Improve warning for terminated threads
with non-zero return code.
2021-06-18 07:54:38 +09:00
Carsten Grohmann 7a488b93d0 Set decimal base for converting subseconds to int
The leading 0 triggers an automatic conversion as an octal value. This
fails because it is a decimal value.

Setting the base to 10 prevents this automatism and treats the value
as a decimal value.

Fixes: [WAN] string_util.cpp:cvt_strtoofft(96): something error is occurred in convert std::string(017080564) to off_t, thus return 0 as default.

Related to #1676
2021-06-17 06:27:02 +09:00
Carsten Grohmann d67b83e671 Allow configuration for temporary files directory 2021-06-16 21:29:58 +09:00
Andrew Gaul a100be9dce
Prefer std::string::empty over length == 0 (#1679)
This is more concise.
2021-06-13 20:03:10 +09:00
Andrew Gaul 4d39ea887e
Protect orgmeta with fdent_lock (#1678)
Fixes a crash seen with the CentOS 7 builder and when running with
AddressSanitizer.  Regression introduced by
ac578d188e.  Fixes #1677.
2021-06-13 16:14:24 +09:00
Andrew Gaul 7638b5b3e3
Prefer std::string::rbegin over operator[] (#1673)
This is more concise and safer due to not repeating the variable name.
We cannot use std::string::back since it is not available in C++03.
2021-06-13 13:26:38 +09:00
Andrew Gaul 600cee118d
Fix implicit narrowing conversions (#1672)
These do not appear to be problematic but rather just clean up warnings.
Found via clang -Wshorten-64-to-32.
2021-06-13 12:50:07 +09:00
Takeshi Nakatani c2c56d0263 Added info object about multipart uploading for each pseudo fd
(and fixed typo about method name)
2021-06-04 22:42:58 +09:00
Takeshi Nakatani ac578d188e Introduced pseudo fd and separated fd for each file opening 2021-06-04 22:42:58 +09:00
Andrew Gaul 53dfd48f59
Correct usage of istringstream (#1670)
Previously this looped one more time than necessary due to the eof
check:

https://isocpp.org/wiki/faq/input-output#istream-and-eof

Remove now redundant empty check.
2021-06-02 07:14:32 +09:00
Andrew Gaul 84174c560d
Fix data races caused by incorrect locking (#1668)
Found via Threadsanitizer.  Fixes #1471.
2021-05-29 00:11:55 +09:00
Carsten Grohmann 9bf525ee7a Ensuring multipart size even when storage is low
When the temporary storage was full, the old implementation started an
upload even if there was not enough data to completely fill the minimum
multipart size or fill the user-selected multipart size.

The new implementation ensures the minimum multipart size by forcing
the user-selected multipart size.

Fixes #1591
2021-05-27 22:55:52 +09:00
Takeshi Nakatani 4b69d4b1bb Fixed a bug when the disk capacity was insufficient in RowFlush 2021-05-22 23:11:46 +09:00
Andrew Gaul bbcccd6e98
Make storage class a string (#1663)
This allows non-standard storage classes like Google Cloud Storage
Nearline.  Fixes #1613.
2021-05-21 23:34:31 +09:00
Takeshi Nakatani b5fef788da Additional bug fixing for not creating zero-byte object 2021-05-09 17:33:53 +09:00
Andrew Gaul 42f5965d8a
Fix off_t to int narrowing in PageList::GetSparseFilePages (#1654)
Found via clang-tidy.
2021-05-09 14:35:00 +09:00
Andrew Gaul d904d91252
Prefer std::map::insert over find and operator[] (#1653)
This avoids a duplicate lookups.
2021-05-09 14:11:35 +09:00
Andrew Gaul 9abe3fa662
Use same time when creating directory objects (#1652)
This avoids mismatched times and is slightly faster.
2021-05-09 13:44:12 +09:00
Andrew Gaul bb6d2b1b74
Replace snprintf with string and ostringstream (#1649)
These uses are probably safe from a buffer overflow perspective but
can cause data race issues in logging due to static buffers.
2021-05-08 02:48:47 +09:00
Andrew Gaul 096a230b70
Allow arbitrary size AWS secret keys (#1648)
Previously s3fs limited these to 123 characters.  Fixes #1626.
2021-05-06 22:24:38 +09:00
Andrew Gaul 8ef01d37a9
Fix a few nits (#1645)
Make some strings more const, initialize members, and abort if lock
initialization fails.  Partially found via clang-tidy.
2021-05-06 19:40:35 +09:00
Andrew Gaul cb9148f6cd
Update for clang-tidy 12 (#1644) 2021-05-03 09:43:18 +09:00
Andrew Gaul 771bbfeac5
Do not create zero-byte object when creating file (#1640)
Previously s3fs created this object to store metadata and overwrote it
when flushing.  This prevented use with object stores which do not
allow overwrites like HDS.  Instead only create an in-memory
representation which reduces the time to create small files.
Fixes #1013.
2021-04-30 19:56:33 +09:00
Andrew Gaul 3694786112
Propagate errno instead of EIO (#1638)
This improves error fidelity.  Follows on to
b70f8db037.  References #1523.
2021-04-30 07:09:00 +09:00
Andrew Gaul fbf3c83019
Consider S3 errors in HTTP PUT 200 responses (#1635)
S3 can emit these in unusual situations.  Fixes #1317.
2021-04-25 13:18:11 +09:00
Andrew Gaul 8a51a26819 Allow truncation of open and modified files
Regression introduced in f5bf41cf11.
Fixes #1575.
2021-04-25 12:35:35 +09:00
Takeshi Nakatani 1838f52e19 Declare undefined symbols for fallocate function 2021-04-25 10:12:25 +09:00
Andrew Gaul e9eb248f2f
Fail CheckBucket when S3 returns PermanentRedirect (#1630)
Previously s3fs allowed mounting but all operations returned EIO.
References #693.

Co-authored-by: Takeshi Nakatani <ggtakec@gmail.com>
2021-04-24 18:46:24 +09:00
Ambika Nair 77581eda59 Change in ibm iam endpoint 2021-04-21 21:42:55 +09:00
Andrew Gaul 7f3e423bbe
Preserve sub-second time precision (#1624)
Found via pjdfstests.  References #897.  References #1589.
2021-04-18 13:11:12 +09:00
Andrew Gaul 706e3cbebd
Update ctime and mtime when opening with O_TRUNC (#1623)
Found via pjdfstests.  References #1589.
2021-04-15 07:08:49 +09:00
Takeshi Nakatani 2effffd8e2 Fixed codes for cppcheck 2.4.1 2021-04-13 21:12:35 +09:00
Takeshi Nakatani 2908878988 Fixed codes for cppcheck 2.4.1 2021-04-13 21:12:35 +09:00
Andrew Gaul 01a26a9011
Do not allow open with O_RDONLY and O_TRUNC (#1620)
Found via pjdfstests.  References #1589.
2021-04-13 07:28:10 +09:00
Andrew Gaul 23e1fbf7b9
Update ctime when truncating file (#1619)
Found via pjdfstests.  References #1589.
2021-04-13 07:08:06 +09:00
Andrew Gaul ed2e877bb6
Return EACESS when lacking required executable bit (#1617)
This makes the check consistent with read and write.  Found via
pjdfstests.  References #1551.  References #1589.
2021-04-11 22:39:27 +09:00
Andrew Gaul 3663082a01
Do not allow renaming to a non-empty directory (#1604)
Found via pjdfstests.  References #1589.
2021-03-28 19:10:52 +09:00
Andrew Gaul 9645d57c05
Translate KeyTooLongError to ENAMETOOLONG (#1602)
AWS does not support keys longer than 1024 characters.  Add special
handling for HeadObject which does not have a response body.
Found via pjdfstests.  References #1589.
2021-03-28 13:17:41 +09:00
Andrew Gaul f6fbd75320
Return EFBIG when file exceeds multipart limit (#1600)
Found via pjdfstests.  References #1589.
2021-03-28 12:33:01 +09:00
Carsten Grohmann ef079f4e94 Don't ignore nomultipart when storage is low
When the temporary storage filled up, the old implementation uploaded
all data with multipart uploads, even if "nomultipart" was set.

The new implementation emits a warning and returns -ENOSPC instead.

Fixes #1595
2021-03-10 08:23:54 +09:00
Takeshi Nakatani b589ebec23 Output version and command parameters at startup 2021-03-06 19:11:57 +09:00
Andrew Gaul 032fcf2a47
Allow optional issuing of ListObjectsV2 (#1583)
This allows use of s3fs on object stores that do not implement
the V1 API.  Fixes #1573.
2021-02-23 09:45:13 +09:00
Takeshi Nakatani 5b5bc3114a Fixed a bug that padded null bytes when changing xattr 2021-02-21 12:23:36 +09:00
Takeshi Nakatani 81e267d421 Fixed the same bug in chmod/chown/utimens and added tests 2021-02-20 23:30:50 +09:00
Takeshi Nakatani 4fc92d59f3 Fixed a bug that filling NULL bytes when changing the attribute after renaming 2021-02-20 23:30:50 +09:00
Takeshi Nakatani a0f347b10f Added no_time_stamp_ms option mainly for testing 2021-02-20 20:24:26 +09:00
Takeshi Nakatani 4da56acdcc Improved processing when HEAD response is 400 2021-02-13 17:00:13 +09:00
Takeshi Nakatani 493802a605 Fixed osx mistype declaration and typo warnings 2021-02-13 12:04:35 +09:00
林千里 22b0ae9d51 set IsExpireTime to true by default
fixes #1563
2021-02-12 12:20:47 +09:00
Andrew Gaul 134a54b32f
Allow configuring the multipart threshold (#1562)
Also change default which improves write performance for files >= 25
MB and <= 5 GB, particularly over lossy networks.

Co-authored-by: Takeshi Nakatani <ggtakec@gmail.com>
2021-02-11 23:35:46 +09:00
Andrew Gaul 7f6fbb0021
Correct singlepart_copy_limit unit (#1561)
This is documented to be in MB not KB.

Co-authored-by: Takeshi Nakatani <ggtakec@gmail.com>
2021-02-11 22:53:50 +09:00
Andrew Gaul ec8bae9827
Do not periodically flush with nomultipartupload (#1559)
s3fs periodically flushes data to S3 and hole-punches the written
data, marking it as clean.  Without multipart copy, s3fs would write
the now-zeroed data along with any new data.  s3fs should not attempt
to periodically flush dirty data at all without multipart copy
support.  Fixes #1542.
2021-02-11 22:24:39 +09:00
Andrew Gaul a4d916af13
Allow configuring multipart copy size (#1555)
Also align configuration with mixupload.
2021-02-08 20:32:12 +09:00
Takeshi Nakatani 32f096fa3f Use clock_gettime instead of gettimeofday 2021-02-08 10:19:28 +09:00
Andrew Gaul c692093921
Copy at most 5 GB per multipart copy part request (#1553)
Previously mixupload failed with larger objects due to the AWS
constraint on part sizes.  This symptom was seen more frequently due
to periodic flushing of dirty data.  Fixes #1547.
2021-02-07 23:10:07 +09:00
Andrew Gaul 1a6d0826b5
Allow -1 value to disable max_dirty_data (#1552)
This matches the man page.
2021-02-07 22:14:31 +09:00
Andrew Gaul bf33fe7f55
Increase buffer size to avoid warning (#1550)
This can be up to 54 bytes.

Co-authored-by: Takeshi Nakatani <ggtakec@gmail.com>
2021-02-07 12:26:58 +09:00
Andrew Gaul ff9d6a75c7
Parse more specific AWS error codes (#1549)
Fixes #1545.
2021-02-07 11:29:08 +09:00
Carsten Grohmann 4c6690f5f0
Add timestamps to debug log and cache integrity check (#1540)
* Add UTC timestamp to debug output

for easier correlation with external events

Old output:
$ s3fs mybucket /bucket/  -o curldbg,dbglevel=debug -d -d -f
[CRT] s3fs_logger.cpp:LowSetLogLevel(201): change debug level from...
[INF]     s3fs.cpp:set_mountpoint_attribute(3989): PROC(uid=0, ...

New output:
$ s3fs mybucket /bucket/  -o curldbg,dbglevel=debug -d -d -f
2021-01-28T21:09:16.264Z [CRT] s3fs_logger.cpp:LowSetLogLevel(202):...
2021-01-28T21:09:16.264Z [INF]     s3fs.cpp:set_mountpoint_attrib...

* Add UTC timestamp to cache integrity check

for easier correlation with external events.

$ s3fs mybucket /mybucket  -oset_check_cache_sigusr1=/tmp/check.cache

Old output:
$ kill -s SIGUSR1 $(pgrep s3fs)
$ cat /tmp/check.cache
------------------------------------------------------------
Check cache file and its stats file consistency
------------------------------------------------------------
------------------------------------------------------------
Summary - Total files:                0
          Detected error files:       0
          Detected error directories: 0
------------------------------------------------------------

New output:
$ kill -s SIGUSR1 $(pgrep s3fs)
$ cat /tmp/check.cache
---------------------------------------------------------------------------
Check cache file and its stats file consistency at 2021-01-30T13:04:14.111Z
---------------------------------------------------------------------------
---------------------------------------------------------------------------
Summary - Total files:                0
          Detected error files:       0
          Detected error directories: 0
---------------------------------------------------------------------------

* Fix indentation of S3fsLog::GetCurrentTime()
2021-02-04 10:41:29 +09:00
Andrew Gaul 67b9381825 Remove limit for single-part objects
AWS enforces a 5 GB limit for single-part objects but other S3
implementations like GCS support larger objects.  Fixes #1542.
2021-02-04 09:19:37 +09:00
Andrew Gaul 4b53d4bf6b
Suppress spurious SHA256 error message (#1544) 2021-02-04 07:28:51 +09:00
Carsten Grohmann ed85b72bf5 Add warning for failing metadata updates of large files
Metadata updates fail for large files if "-o nocopyapi" or
"-o nomultipart" is set.

This change is related to #1528
2021-01-28 10:20:32 +09:00
Andrew Gaul 17fda89ae9
Handle s3fs_get_content_md5 and s3fs_sha256_hex_fd errors (#1534)
Follows on to 4d833a4fb9.
2021-01-25 19:08:14 +09:00
Andrew Gaul d019dda4f7
Simplify substr manipulations with erase (#1532)
This avoids creating a new std::string.
2021-01-25 18:02:32 +09:00
Andrew Gaul dc9255bc5f
Prefer static_cast where possible (#1531) 2021-01-25 08:15:17 +09:00
Andrew Gaul b0e8758b63
Use result instead of res for consistency (#1530) 2021-01-25 07:56:10 +09:00
Andrew Gaul b70f8db037
Propagate errno from get_local_fent (#1525)
Follows on to 4d833a4fb9.
References #1523.
2021-01-20 23:21:14 +09:00
Andrew Gaul 4d833a4fb9
Return more specific errno when available (#1520)
Previously s3fs threw away some function return values and returned
EIO instead.  This was due to not trusting the mix of -1 and errno
return codes.  Correct the obviously incorrect ones via visual
inspection.  Stronger typing may find more occurrences.  Fixes #1519.
2021-01-18 18:50:49 +09:00
fly3366 168e588ac7
fix: Add reset offset (#1503) 2021-01-04 22:57:56 +09:00
Andrew Gaul bd0fadbe5f
Remove authorization header when remaking handle (#1505)
This avoids including Authorization in SignedHeaders.  s3fs will
recreate the Authorization header before sending the request.
2021-01-04 21:37:34 +09:00
kontrollanten d1c638ab7a fix 404 error message 2020-12-30 14:10:17 +09:00
fly3366 51f65d7b14 fix: miss header when retry 2020-12-23 22:16:07 +09:00
Andrew Gaul a16d00d673
Simply curl_slist_sort_insert (#1494)
Some good taste from Linus:
https://github.com/mkirchner/linked-list-good-taste
Also avoid an allocation when replacing a value and tighten up tests.
2020-12-23 20:29:33 +09:00
Andrew Gaul d0363b118e
Add tests for curl_util (#1481) 2020-11-24 21:37:09 +09:00
Takeshi Nakatani 533322859d Added comment for nullPointerRedundantCheck/cppcheck2.2 2020-11-21 20:06:14 +09:00
Andrew Gaul f8d5b76edb
Simplify handling of returned ETag (#1479)
This works around lifetime warnings uncovered by cppcheck.
References #1478.
2020-11-21 06:56:05 +09:00
Takeshi Nakatani d9f6469b7b Fixed flushing dirty data and compressed the cache size 2020-11-14 16:45:37 +09:00
Andrew Gaul 543231c9f2
Address warnings from clang-tidy 11 (#1470) 2020-11-09 21:15:20 +09:00
Noah Meyerhans d96a08d4ad fixup! Implement AWS IMDSv2 support 2020-11-07 14:48:52 +09:00
Noah Meyerhans f2f930300a Implement AWS IMDSv2 support
AWS IMDSv2 is a session oriented method for retrieving instance metadata,
including IAM credentials, in Amazon EC2.  It is enabled by default in
non-enforcing mode in AWS (meaning it retains backwards compatibility with
existing IMDSv1 clients), but can be switched to enforcing mode, in which
clients are required to return API tokens with requests.

With this change, we implement support for IMDSv2 and enable it by default when
IAM roles are our source for authentication credentials.  In the event that
s3fs is running in cloud environment offering an IMDSv1-compatible API, we
support graceful fallback to that mode.  It can also be selected explicitly via
the imdsv1only mount option.

More details on IMDSv2 are available at
https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/instancedata-data-retrieval.html
and
https://aws.amazon.com/blogs/security/defense-in-depth-open-firewalls-reverse-proxies-ssrf-vulnerabilities-ec2-instance-metadata-service/

Signed-off-by: Noah Meyerhans <nmeyerha@amazon.com>
2020-11-07 14:48:52 +09:00
yoshihiko 81ad3ce0ae Fix directory authority 2020-11-07 14:47:53 +09:00
Takeshi Nakatani 38e1eaa8a3 Added logfile option for non-syslog logging 2020-10-19 20:10:19 +09:00
Andrew Gaul 6aa786b886
Flush dirty data after a writing a number of bytes (#1448)
This allows s3fs to write large files without consuming a large amount
of temporary local storage but can slow uploads due to server-side
copies.  References #617.  Fixed #1056.  Fixes #1257.
2020-10-13 22:30:42 +09:00
Takeshi Nakatani 58750cc441 Fixed a bug that symlink could not be read after restarting s3fs 2020-10-11 22:14:36 +09:00
Andrew Gaul 910255745e Allow relative path with use_path
Fixes #1161.
2020-10-10 18:13:23 +09:00
Takeshi Nakatani cf86fa51b0
Merge pull request #1441 from gaul/coverity
Suppress false positive from Coverity
2020-10-04 23:44:51 +09:00
Takeshi Nakatani 0492f75197
Merge pull request #1439 from gaul/run-once
Protect curl_warnings_once with a mutex
2020-10-04 22:43:56 +09:00
Takeshi Nakatani 493cf20f95
Merge pull request #1435 from gaul/no-exceptions
Remove exceptions from s3fs_strtoofft
2020-10-04 22:19:25 +09:00
Andrew Gaul 3553fb65a0 Suppress false positive from Coverity 2020-10-04 14:58:13 +09:00
Takeshi Nakatani 059cc57ba6 Added atime and Corrected atime/mtime/ctime operations 2020-10-04 13:54:01 +09:00
Andrew Gaul 4df4ffe06f Protect curl_warnings_once with a mutex
This allows ThreadSanitizer to run without warnings.
2020-10-03 21:09:35 +09:00
Takeshi Nakatani 462347256d Changed the message level in the AutoFdEntity::GetFdEntity 2020-10-03 20:41:17 +09:00
Takeshi Nakatani 133feb67c3
Merge pull request #1434 from gaul/const-string
Use const std::string& where possible
2020-10-03 10:02:03 +09:00
Takeshi Nakatani 2cf195741c Fixed signature error due to case of hex string 2020-10-02 18:09:13 +00:00
Takeshi Nakatani f61baada46
Merge pull request #1433 from gaul/hex
Further simplify and centralize hex conversion
2020-10-03 00:34:02 +09:00
Takeshi Nakatani 6e1e0d1d31
Merge pull request #1432 from gaul/v2-or-v4
Add sigv4 flag and fix v4 regression
2020-10-03 00:09:55 +09:00
Takeshi Nakatani 66419e7292
Merge pull request #1431 from gaul/string/find_first_of
Prefer string::find_first_of(char) where possible
2020-10-02 23:42:17 +09:00
Takeshi Nakatani 9e998877e9
Merge pull request #1430 from gaul/ostringstream
Simplify use of ostringstream
2020-10-02 23:26:41 +09:00
Andrew Gaul 3628b9d1e2 Remove exceptions from s3fs_strtoofft
Explicit return codes are simpler and safer.
2020-10-02 22:53:12 +09:00
Andrew Gaul 8d2bd874d7 Use const std::string& where possible
This hints to callers on usage.
2020-10-02 08:48:43 +09:00
Andrew Gaul 7e27c6cf7d Further simplify and centralize hex conversion 2020-10-01 23:31:06 +09:00
Andrew Gaul b8ff4ede49 Sign GET requests with V4 signature correctly
This bug was previously masked by v2 fallback.  Fixes regression from
81805715bd.
2020-10-01 20:03:14 +09:00
Andrew Gaul 081d6c1245 Allow setting signature V4-only
Default to allowing V2 or V4.
2020-10-01 20:03:14 +09:00
Andrew Gaul eb8004c355 Prefer string::find_first_of(char) where possible 2020-10-01 18:24:45 +09:00
Andrew Gaul b3bf9f8f54 Simplify use of ostringstream
Resetting this does not re-use the underlying buffer so creating a new
ostringstream has similar efficiency.
2020-10-01 17:55:34 +09:00
Andrew Gaul 503c86bb8a Call is_prefix instead of compare and substr 2020-09-27 22:19:54 +09:00
Takeshi Nakatani 757f4caee8
Merge pull request #1425 from gaul/create-bucket/sse
Do not send SSE headers during bucket creation
2020-09-27 11:34:15 +09:00
Takeshi Nakatani 95fabd1f3a
Merge pull request #1423 from gaul/nettle
Fix MD5 when using gnutls with nettle
2020-09-27 11:17:55 +09:00
Takeshi Nakatani 0b42e08636
Merge pull request #1422 from gaul/c++/append-assign-at
Remove calls to append, assign, and at
2020-09-27 10:59:48 +09:00
Takeshi Nakatani d5e4f99e72
Merge pull request #1413 from liuyongqing/master
not call put headers if not exist pending meta
2020-09-27 10:37:22 +09:00
Andrew Gaul f35fe850c0 Do not send SSE headers during bucket creation
This allows integration tests to pass with use_sse.
2020-09-26 13:25:18 +09:00
Andrew Gaul 9a55c9fd9f Fix MD5 when using gnutls with nettle
This matches the non-nettle definition of s3fs_md5_fd.
2020-09-25 23:07:50 +09:00
Neeraj Kumar Kashyap 041b4ec05c enforce exact key match in IBM IAM response 2020-09-25 22:41:19 +09:00
Andrew Gaul 2438066d52 Remove calls to append, assign, and at
operator+, operator=, and operator[] are more idiomatic and consistent
with the code base.
2020-09-25 12:19:51 +09:00
yongqingliu e98827ec6f not call put headers if not exist pending meta 2020-09-21 14:12:07 +08:00
Andrew Gaul 05863a3178 Fix clang-tidy warnings 2020-09-21 07:03:42 +09:00
Takeshi Nakatani 075d161bb1
Merge pull request #1408 from gaul/env
Ensure environment variable is set when using ECS
2020-09-20 13:01:17 +09:00
Takeshi Nakatani 6f40503328
Merge pull request #1406 from gaul/strcmp
Remove unnecessary calls to strcmp(str.c_str())
2020-09-20 11:28:41 +09:00
Takeshi Nakatani 3440c3348c
Merge pull request #1405 from gaul/is-prefix
Rename STR2NCMP to is_prefix for clarity
2020-09-20 10:03:53 +09:00
Takeshi Nakatani 853be26612
Merge pull request #1404 from liuyongqing/master
fix dead lock in disk insufficient and optimize code
2020-09-20 01:34:02 +09:00
Andrew Gaul 44d2cc15f7 Ensure environment variable is set when using ECS
Previously s3fs crashed when AWS_CONTAINER_CREDENTIALS_RELATIVE_URI
was not set.  References #1162.
2020-09-18 08:14:58 +09:00
Andrew Gaul 81805715bd Remove unnecessary calls to strcmp(str.c_str()) 2020-09-17 20:00:45 +09:00
Andrew Gaul ebcbb02d16 Rename STR2NCMP to is_prefix for clarity
This may prevent confusion seen in
39ec8e242e.
2020-09-17 19:45:54 +09:00
Takeshi Nakatani 93cf67c65d
Merge pull request #1399 from gaul/md5-sha256
Use more descriptive names for digest functions
2020-09-17 17:11:19 +09:00
yongqingliu 211cc0f5f2 fix dead lock in disk insufficient and optimize code 2020-09-16 22:45:28 +08:00
Takeshi Nakatani c58c91fc4f
Merge pull request #1398 from gaul/memset
Remove unneeded memset calls
2020-09-15 22:09:54 +09:00
Andrew Gaul b7187352e1 Use more descriptive names for digest functions
Previously they used hex to refer to binary data which is confusing
since other std::string return types are actually hex.  Also remove
unused s3fs_md5sum.
2020-09-15 21:18:58 +09:00
Takeshi Nakatani 1520ca6220
Merge pull request #1397 from gaul/hex
Centralize hex conversion
2020-09-15 21:14:23 +09:00
Takeshi Nakatani f7a63d5c97
Merge pull request #1396 from gaul/32-bit-read
Use off_t for read sizes
2020-09-15 20:05:19 +09:00
Takeshi Nakatani 3958450c05 Added a class for automating fdentity reference counts 2020-09-14 22:02:28 +09:00
Andrew Gaul 5121c73ed1 Remove unneeded memset calls
Also use constant for array lengths.
2020-09-14 19:51:36 +09:00
Andrew Gaul 44eaac8471 Centralize hex conversion 2020-09-14 19:28:59 +09:00
Andrew Gaul 77501c3600 Use off_t for read sizes
This fixes an issue with large multipart_size on 32-bit platforms like
Raspberry Pi.
2020-09-14 18:12:23 +09:00
Takeshi Nakatani 8205607716 Fixed incorrect Signal related message at s3fs initialization and termination 2020-09-13 18:54:10 +09:00
Andrew Gaul 1043e08dfa Remove uses of implicit namespace std
Fixed via:

sed -i '/using namespace std/{N;d}' src/*.cpp
sed -i 's/ string/ std::string/g' src/*.cpp
sed -i 's/(string/(std::string/g' src/*.cpp
sed -i 's/\[string/\[std::string/g' src/*.cpp
sed -i 's/^string/std::string/g' src/*.cpp
sed -i 's/ ifstream/ std::ifstream/g' src/*.cpp
sed -i 's/ istringstream/ std::istringstream/g' src/*.cpp
sed -i 's/ ostringstream/ std::ostringstream/g' src/*.cpp
sed -i 's/ max(/ std::max(/g' src/*.cpp
sed -i 's/ min(/ std::min(/g' src/*.cpp
sed -i 's/ endl/ std::endl/g' src/*.cpp
2020-09-13 11:57:20 +09:00
Takeshi Nakatani e5f4f9b69e
Merge pull request #1393 from gaul/32-bit-casting
Fix casting warnings on 32-bit platforms
2020-09-13 10:40:56 +09:00
Takeshi Nakatani 1e3c10d803
Merge pull request #1392 from gaul/large-singlepart-32-bit
Use 64-bit off_t when computing digests
2020-09-13 10:39:34 +09:00
Andrew Gaul 6d1e704e34 Fix casting warnings on 32-bit platforms 2020-09-12 18:22:32 +09:00
Andrew Gaul ffc33a447f Use 64-bit off_t when computing digests
This allows 32-bit platforms like Raspberry Pi to upload single-part
objects larger than 2 GB.
2020-09-12 15:00:23 +09:00
Andrew Gaul 39ec8e242e Remove incorrect string prefix comparisons
Previously flags like -o nouafoobar were parsed as -o noua.  Found
via:

grep STR2NCMP src/s3fs.cpp | grep -v '="'
2020-09-11 17:38:20 +09:00
Ben Mares 0d4e39ad1c
return 0 after parsing "use_session_token" arg (#1388)
return 0 after parsing "use_session_token" arg

I hope this might fix #651.

There's clearly an open bug in #651 confirmed by several users related to the parsing of the `-o use_session_token`. Looking at the source, I noticed that there was a return value everywhere except here, so I suspect this may be responsible.
2020-09-11 08:09:36 +09:00
yongqingliu 986fab8738 close FdEntity to avoid leakage 2020-09-07 20:48:16 +08:00
Takeshi Nakatani b5ffd419d8 Source file division and set 4 spaces and cleanup 2020-08-26 17:43:50 +09:00
Takeshi Nakatani c6e23212bb
Merge pull request #1375 from gaul/cppcheck
Add const where possible
2020-08-21 07:32:39 +09:00
Andrew Gaul 1ec8528502 Add const where possible
Found via cppcheck 2.1
2020-08-20 23:46:11 +09:00
Andrew Gaul 892e7129c5 Add support for deep archive storage class 2020-08-20 23:42:45 +09:00
Takeshi Nakatani bdea2ee5c8
Merge pull request #1373 from gaul/enum
Use scoped enums for acl_t, sse_type_t, and storage_class_t
2020-08-20 23:06:52 +09:00
Takeshi Nakatani a5186c73c2
Merge pull request #1365 from ggtakec/check_lseek
Dynamically determine whether lseek extended options are supported
2020-08-20 22:36:56 +09:00
Andrew Gaul 60d456a993 Use scoped enums for acl_t, sse_type_t, and storage_class_t
This prevents some kinds of implicit conversions.  Also deduplicate
str/from_str logic.  References #1371.
2020-08-20 18:46:18 +09:00
Takeshi Nakatani bcf6838e86
Merge pull request #1370 from gaul/vim/modeline
Correct vim modeline to 2-space indentation
2020-08-19 22:44:40 +09:00
Andrew Gaul 958ad83a4b Correct vim modeline to 2-space indentation 2020-08-19 21:03:46 +09:00
Andrew Gaul ba61470bae Prefer named locks in curl
This removes the confusing and unsafe pthread_mutex_t array.
2020-08-18 22:20:41 +09:00
Andrew Gaul 8210a1b2f2 Consistently lock curl_times and curl_progress
References #1362.
2020-08-17 09:47:38 +09:00
Takeshi Nakatani 55cb8920d5 Dynamically determine whether lseek extended options are supported 2020-08-16 12:37:11 +00:00
Takeshi Nakatani 46acbf10ba
Merge pull request #1363 from gaul/is-modified-race
Lock fdent_data_lock before accessing pagelist
2020-08-16 20:25:24 +09:00
Takeshi Nakatani 2c0408b95a
Merge pull request #1357 from gaul/readdir
Call readdir instead of readdir_r
2020-08-16 18:50:38 +09:00
Takeshi Nakatani 057da86d87
Merge pull request #1356 from gaul/printf-size_t
Prefer %zd and %zu for ssize_t and size_t
2020-08-16 18:16:59 +09:00
Andrew Gaul 8de6cb3504 Lock fdent_data_lock before accessing pagelist
Found via ThreadSanitizer.  References #1353.  References #1362.
2020-08-16 17:44:03 +09:00
Takeshi Nakatani 2bb745cdd7 Fixed a bug about move file over limit of ensure space 2020-08-16 17:42:20 +09:00
Andrew Gaul 35090ba4d5 Call readdir instead of readdir_r
Only a single thread uses this directory stream.  Further, modern
implementations are thread-safe by default and deprecated this call:

https://man7.org/linux/man-pages/man3/readdir_r.3.html
2020-08-16 13:33:23 +09:00
Andrew Gaul 132a1bebbb Prefer %zd and %zu for ssize_t and size_t
This removes unnecessary casting.
2020-08-15 17:30:23 +09:00
Takeshi Nakatani cbd925c56f Moved the SIGUSR2 handler to S3fsSignals class 2020-07-28 14:54:35 +09:00
Takeshi Nakatani 63bbb47378
Merge pull request #1341 from gaul/stat-cache/default
Change default stat_cache_expire
2020-07-26 23:54:14 +09:00
Andrew Gaul 0fbd0eac80 Change default stat_cache_expire
Previously s3fs cached files forever which confused users with
creating objects using another client.
2020-07-26 23:04:43 +09:00
Takeshi Nakatani ad1961417d Added SIGUSR1 option for cache file integrity test 2020-07-26 21:04:11 +09:00
Takeshi Nakatani 700e288718 Put similar processing together into method GetCacheFileStatTopDir 2020-07-12 08:12:02 +00:00
Takeshi Nakatani dc365b65a0 Fix NULL pointer deference
Found via cppcheck 2.1.
2020-07-09 21:40:23 +09:00
Takeshi Nakatani 1f796d432d Fixed upload error about mixuploading sparse file and truncating file 2020-06-27 22:44:19 +09:00
Takeshi Nakatani 35006e318f Fixed about ParallelMixMultipartUpload 2020-06-24 12:48:55 +09:00
Eric Vantillard 7d0c66e08a Add support for glacier storage class.
Just a copy of what have been done in PR #271.
2020-06-23 11:23:21 +09:00
Takeshi Nakatani f324d8e04f Fixed a bug about serializing from cache file 2020-06-19 12:57:27 +00:00
Takeshi Nakatani f16ee96d7e
Merge pull request #1306 from gaul/http/500
Retry with exponential backoff during 500 error
2020-06-06 15:30:22 +09:00
Andrew Gaul d8766b2051 Retry with exponential backoff during 500 error
Amazon suggests retrying on both 500 and 503:

https://aws.amazon.com/premiumsupport/knowledge-center/http-5xx-errors-s3/

Fixes #1251.
2020-06-05 21:01:30 +09:00
Andrew Gaul 9db70bab63 Ignore case when comparing ETags
This allows multipart upload to work with Alibaba OSS.
References #1297.
2020-06-05 18:17:52 +09:00
Takeshi Nakatani 0cb057dadd
Merge pull request #1303 from gaul/rename/use_cache
Relink cache stats file atomically via rename
2020-06-01 00:10:33 +09:00
Takeshi Nakatani 0f5db0d1bf
Merge pull request #1302 from gaul/rename/nocopy
Fix renames of open files with nocopyapi option
2020-05-31 23:46:46 +09:00